diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8.hpp
index 6a8caf6..4dfe464 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,44 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x22, [%x[in], #0x0]\n"
+      "ldr x23, [%x[in], #0x0]\n"
+      "ldr x22, [%x[in], #0x8]\n"
       "cmp %x[height], #0x4\n"
-      "ldr x21, [%x[in], #0x8]\n"
+      "add x23, x23, %x[row_offset]\n"
+      "ldr x21, [%x[in], #0x10]\n"
+      "ldr x20, [%x[in], #0x18]\n"
       "add x22, x22, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x10]\n"
-      "ldr x19, [%x[in], #0x18]\n"
       "add x21, x21, %x[row_offset]\n"
       "add x20, x20, %x[row_offset]\n"
-      "add x19, x19, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x19, x22\n"
       "cmp %x[height], #0x2\n"
-      "csel x21, x21, x22, GE\n"
-      "csel x20, x20, x22, GT\n"
+      "mov x20, x23\n"
+      "csel x22, x22, x23, GE\n"
+      "csel x21, x21, x23, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x22, #0x0]\n"
       "cmp %x[width], #0x10\n"
+      "prfm pldl1keep, [x23, #0x0]\n"
+      "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
       "prfm pldl1keep, [x20, #0x0]\n"
-      "prfm pldl1keep, [x19, #0x0]\n"
+      "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
       "prfm pldl1keep, [x20, #0x40]\n"
-      "prfm pldl1keep, [x19, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q19, [x22], #0x10\n"
+      "ldr q19, [x23], #0x10\n"
+      "ldr q18, [x22], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q18, [x21], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
-      "ldr q16, [x19], #0x10\n"
+      "ldr q17, [x21], #0x10\n"
+      "ldr q16, [x20], #0x10\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
       "prfm pldl1keep, [x22, #0x70]\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
-      "prfm pldl1keep, [x19, #0x70]\n"
-      "str q19, [%x[out_ptr], #0x0]\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
       "str q17, [%x[out_ptr], #0x20]\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
@@ -76,93 +76,93 @@
       "3:"  // Main loop skip
       "cbz %x[width], 12f\n"
       "tbz %x[width], #3, 7f\n"
-      "ldr d19, [x22], #0x8\n"
-      "ldr d18, [x21], #0x8\n"
-      "ldr d17, [x20], #0x8\n"
-      "ldr d16, [x19], #0x8\n"
+      "ldr d19, [x23], #0x8\n"
+      "ldr d18, [x22], #0x8\n"
+      "ldr d17, [x21], #0x8\n"
+      "ldr d16, [x20], #0x8\n"
       "tbz %x[width], #2, 5f\n"
-      "ld1 { v19.s }[2], [x22], #0x4\n"
-      "ld1 { v18.s }[2], [x21], #0x4\n"
-      "ld1 { v17.s }[2], [x20], #0x4\n"
-      "ld1 { v16.s }[2], [x19], #0x4\n"
+      "ld1 { v19.s }[2], [x23], #0x4\n"
+      "ld1 { v18.s }[2], [x22], #0x4\n"
+      "ld1 { v17.s }[2], [x21], #0x4\n"
+      "ld1 { v16.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v19.h }[6], [x22], #0x2\n"
-      "ld1 { v18.h }[6], [x21], #0x2\n"
-      "ld1 { v17.h }[6], [x20], #0x2\n"
-      "ld1 { v16.h }[6], [x19], #0x2\n"
+      "ld1 { v19.h }[6], [x23], #0x2\n"
+      "ld1 { v18.h }[6], [x22], #0x2\n"
+      "ld1 { v17.h }[6], [x21], #0x2\n"
+      "ld1 { v16.h }[6], [x20], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[14], [x22]\n"
-      "ld1 { v18.b }[14], [x21]\n"
-      "ld1 { v17.b }[14], [x20]\n"
-      "ld1 { v16.b }[14], [x19]\n"
+      "ld1 { v19.b }[14], [x23]\n"
+      "ld1 { v18.b }[14], [x22]\n"
+      "ld1 { v17.b }[14], [x21]\n"
+      "ld1 { v16.b }[14], [x20]\n"
       "b 11f\n"
       "4:"  // odd_loads_1_12
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[12], [x22]\n"
-      "ld1 { v18.b }[12], [x21]\n"
-      "ld1 { v17.b }[12], [x20]\n"
-      "ld1 { v16.b }[12], [x19]\n"
+      "ld1 { v19.b }[12], [x23]\n"
+      "ld1 { v18.b }[12], [x22]\n"
+      "ld1 { v17.b }[12], [x21]\n"
+      "ld1 { v16.b }[12], [x20]\n"
       "b 11f\n"
       "5:"  // odd_loads_2_8
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v19.h }[4], [x22], #0x2\n"
-      "ld1 { v18.h }[4], [x21], #0x2\n"
-      "ld1 { v17.h }[4], [x20], #0x2\n"
-      "ld1 { v16.h }[4], [x19], #0x2\n"
+      "ld1 { v19.h }[4], [x23], #0x2\n"
+      "ld1 { v18.h }[4], [x22], #0x2\n"
+      "ld1 { v17.h }[4], [x21], #0x2\n"
+      "ld1 { v16.h }[4], [x20], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[10], [x22]\n"
-      "ld1 { v18.b }[10], [x21]\n"
-      "ld1 { v17.b }[10], [x20]\n"
-      "ld1 { v16.b }[10], [x19]\n"
+      "ld1 { v19.b }[10], [x23]\n"
+      "ld1 { v18.b }[10], [x22]\n"
+      "ld1 { v17.b }[10], [x21]\n"
+      "ld1 { v16.b }[10], [x20]\n"
       "b 11f\n"
       "6:"  // odd_loads_1_8
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[8], [x22]\n"
-      "ld1 { v18.b }[8], [x21]\n"
-      "ld1 { v17.b }[8], [x20]\n"
-      "ld1 { v16.b }[8], [x19]\n"
+      "ld1 { v19.b }[8], [x23]\n"
+      "ld1 { v18.b }[8], [x22]\n"
+      "ld1 { v17.b }[8], [x21]\n"
+      "ld1 { v16.b }[8], [x20]\n"
       "b 11f\n"
       "7:"  // odd_loads_4_0
       "tbz %x[width], #2, 9f\n"
-      "ldr s19, [x22], #0x4\n"
-      "ldr s18, [x21], #0x4\n"
-      "ldr s17, [x20], #0x4\n"
-      "ldr s16, [x19], #0x4\n"
+      "ldr s19, [x23], #0x4\n"
+      "ldr s18, [x22], #0x4\n"
+      "ldr s17, [x21], #0x4\n"
+      "ldr s16, [x20], #0x4\n"
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v19.h }[2], [x22], #0x2\n"
-      "ld1 { v18.h }[2], [x21], #0x2\n"
-      "ld1 { v17.h }[2], [x20], #0x2\n"
-      "ld1 { v16.h }[2], [x19], #0x2\n"
+      "ld1 { v19.h }[2], [x23], #0x2\n"
+      "ld1 { v18.h }[2], [x22], #0x2\n"
+      "ld1 { v17.h }[2], [x21], #0x2\n"
+      "ld1 { v16.h }[2], [x20], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[6], [x22]\n"
-      "ld1 { v18.b }[6], [x21]\n"
-      "ld1 { v17.b }[6], [x20]\n"
-      "ld1 { v16.b }[6], [x19]\n"
+      "ld1 { v19.b }[6], [x23]\n"
+      "ld1 { v18.b }[6], [x22]\n"
+      "ld1 { v17.b }[6], [x21]\n"
+      "ld1 { v16.b }[6], [x20]\n"
       "b 11f\n"
       "8:"  // odd_loads_1_4
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[4], [x22]\n"
-      "ld1 { v18.b }[4], [x21]\n"
-      "ld1 { v17.b }[4], [x20]\n"
-      "ld1 { v16.b }[4], [x19]\n"
+      "ld1 { v19.b }[4], [x23]\n"
+      "ld1 { v18.b }[4], [x22]\n"
+      "ld1 { v17.b }[4], [x21]\n"
+      "ld1 { v16.b }[4], [x20]\n"
       "b 11f\n"
       "9:"  // odd_loads_2_0
       "tbz %x[width], #1, 10f\n"
-      "ldr h19, [x22], #0x2\n"
-      "ldr h18, [x21], #0x2\n"
-      "ldr h17, [x20], #0x2\n"
-      "ldr h16, [x19], #0x2\n"
+      "ldr h19, [x23], #0x2\n"
+      "ldr h18, [x22], #0x2\n"
+      "ldr h17, [x21], #0x2\n"
+      "ldr h16, [x20], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v19.b }[2], [x22]\n"
-      "ld1 { v18.b }[2], [x21]\n"
-      "ld1 { v17.b }[2], [x20]\n"
-      "ld1 { v16.b }[2], [x19]\n"
+      "ld1 { v19.b }[2], [x23]\n"
+      "ld1 { v18.b }[2], [x22]\n"
+      "ld1 { v17.b }[2], [x21]\n"
+      "ld1 { v16.b }[2], [x20]\n"
       "b 11f\n"
       "10:"  // odd_loads_1_0
-      "ldr b19, [x22, #0x0]\n"
-      "ldr b18, [x21, #0x0]\n"
-      "ldr b17, [x20, #0x0]\n"
-      "ldr b16, [x19, #0x0]\n"
+      "ldr b19, [x23, #0x0]\n"
+      "ldr b18, [x22, #0x0]\n"
+      "ldr b17, [x21, #0x0]\n"
+      "ldr b16, [x20, #0x0]\n"
       "11:"  // Odd load end
       "str q19, [%x[out_ptr], #0x0]\n"
       "str q18, [%x[out_ptr], #0x10]\n"
@@ -173,7 +173,7 @@
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "x19", "x20", "x21", "x22"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "x20", "x21", "x22", "x23"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp
index 954a866..56ca49a 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,39 @@
 )
 {
   __asm__ __volatile__(
-      "movi v28.8h, #0x0\n"
-      "ldr x23, [%x[in], #0x0]\n"
-      "mov x22, #0x0\n"
-      "movi v27.8h, #0x0\n"
-      "ldr x21, [%x[in], #0x8]\n"
+      "ldr x24, [%x[in], #0x0]\n"
+      "ldr x23, [%x[in], #0x8]\n"
       "cmp %x[height], #0x4\n"
+      "mov x22, #0x0\n"
+      "ldr x21, [%x[in], #0x10]\n"
+      "ldr x20, [%x[in], #0x18]\n"
+      "movi v28.8h, #0x0\n"
+      "movi v27.8h, #0x0\n"
       "movi v26.8h, #0x0\n"
-      "ldr x20, [%x[in], #0x10]\n"
-      "add x23, x23, %x[row_offset]\n"
       "movi v25.8h, #0x0\n"
-      "ldr x19, [%x[in], #0x18]\n"
+      "add x24, x24, %x[row_offset]\n"
+      "add x23, x23, %x[row_offset]\n"
       "movi v24.4s, #0x0\n"
-      "add x21, x21, %x[row_offset]\n"
       "movi v23.4s, #0x0\n"
+      "add x21, x21, %x[row_offset]\n"
       "add x20, x20, %x[row_offset]\n"
       "movi v22.4s, #0x0\n"
-      "add x19, x19, %x[row_offset]\n"
       "movi v21.4s, #0x0\n"
       "beq 1f\n"
-      "mov x19, x23\n"
       "cmp %x[height], #0x2\n"
-      "csel x21, x21, x23, GE\n"
-      "csel x20, x20, x23, GT\n"
+      "mov x20, x24\n"
+      "csel x23, x23, x24, GE\n"
+      "csel x21, x21, x24, GT\n"
       "1:"  // no_pointer_adj
-      "movi v20.4s, #0x0\n"
+      "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
+      "movi v20.4s, #0x0\n"
       "prfm pldl1keep, [x21, #0x0]\n"
       "prfm pldl1keep, [x20, #0x0]\n"
-      "prfm pldl1keep, [x19, #0x0]\n"
+      "prfm pldl1keep, [x24, #0x40]\n"
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
       "prfm pldl1keep, [x20, #0x40]\n"
-      "prfm pldl1keep, [x19, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x10\n"
       "ld1 { v20.4s }, [%x[out_ptr]]\n"
@@ -75,141 +75,141 @@
       "ble 4f\n"
       "sadalp v24.4s, v28.8h\n"
       "movi v28.8h, #0x0\n"
+      "mov x22, #0x0\n"
       "sadalp v23.4s, v27.8h\n"
       "movi v27.8h, #0x0\n"
       "sadalp v22.4s, v26.8h\n"
       "movi v26.8h, #0x0\n"
       "sadalp v21.4s, v25.8h\n"
       "movi v25.8h, #0x0\n"
-      "mov x22, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q19, [x23], #0x10\n"
-      "add x22, x22, #0x1\n"
-      "ldr q18, [x21], #0x10\n"
+      "ldr q19, [x24], #0x10\n"
+      "ldr q18, [x23], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "ldr q16, [x19], #0x10\n"
-      "sadalp v28.8h, v19.16b\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "sadalp v27.8h, v18.16b\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "sadalp v26.8h, v17.16b\n"
-      "prfm pldl1keep, [x19, #0x70]\n"
-      "sadalp v25.8h, v16.16b\n"
+      "ldr q17, [x21], #0x10\n"
+      "ldr q16, [x20], #0x10\n"
       "str q19, [%x[out_ptr], #0x0]\n"
+      "sadalp v28.8h, v19.16b\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
       "str q18, [%x[out_ptr], #0x10]\n"
+      "sadalp v27.8h, v18.16b\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "prfm pldl1keep, [x20, #0x70]\n"
       "str q17, [%x[out_ptr], #0x20]\n"
+      "sadalp v26.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x30]\n"
+      "sadalp v25.8h, v16.16b\n"
+      "add x22, x22, #0x1\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d19, [x23], #0x8\n"
-      "ldr d18, [x21], #0x8\n"
-      "ldr d17, [x20], #0x8\n"
-      "ldr d16, [x19], #0x8\n"
+      "ldr d19, [x24], #0x8\n"
+      "ldr d18, [x23], #0x8\n"
+      "ldr d17, [x21], #0x8\n"
+      "ldr d16, [x20], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v19.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x21], #0x4\n"
-      "ld1 { v17.s }[2], [x20], #0x4\n"
-      "ld1 { v16.s }[2], [x19], #0x4\n"
+      "ld1 { v19.s }[2], [x24], #0x4\n"
+      "ld1 { v18.s }[2], [x23], #0x4\n"
+      "ld1 { v17.s }[2], [x21], #0x4\n"
+      "ld1 { v16.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v19.h }[6], [x23], #0x2\n"
-      "ld1 { v18.h }[6], [x21], #0x2\n"
-      "ld1 { v17.h }[6], [x20], #0x2\n"
-      "ld1 { v16.h }[6], [x19], #0x2\n"
+      "ld1 { v19.h }[6], [x24], #0x2\n"
+      "ld1 { v18.h }[6], [x23], #0x2\n"
+      "ld1 { v17.h }[6], [x21], #0x2\n"
+      "ld1 { v16.h }[6], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[14], [x23]\n"
-      "ld1 { v18.b }[14], [x21]\n"
-      "ld1 { v17.b }[14], [x20]\n"
-      "ld1 { v16.b }[14], [x19]\n"
+      "ld1 { v19.b }[14], [x24]\n"
+      "ld1 { v18.b }[14], [x23]\n"
+      "ld1 { v17.b }[14], [x21]\n"
+      "ld1 { v16.b }[14], [x20]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[12], [x23]\n"
-      "ld1 { v18.b }[12], [x21]\n"
-      "ld1 { v17.b }[12], [x20]\n"
-      "ld1 { v16.b }[12], [x19]\n"
+      "ld1 { v19.b }[12], [x24]\n"
+      "ld1 { v18.b }[12], [x23]\n"
+      "ld1 { v17.b }[12], [x21]\n"
+      "ld1 { v16.b }[12], [x20]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v19.h }[4], [x23], #0x2\n"
-      "ld1 { v18.h }[4], [x21], #0x2\n"
-      "ld1 { v17.h }[4], [x20], #0x2\n"
-      "ld1 { v16.h }[4], [x19], #0x2\n"
+      "ld1 { v19.h }[4], [x24], #0x2\n"
+      "ld1 { v18.h }[4], [x23], #0x2\n"
+      "ld1 { v17.h }[4], [x21], #0x2\n"
+      "ld1 { v16.h }[4], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[10], [x23]\n"
-      "ld1 { v18.b }[10], [x21]\n"
-      "ld1 { v17.b }[10], [x20]\n"
-      "ld1 { v16.b }[10], [x19]\n"
+      "ld1 { v19.b }[10], [x24]\n"
+      "ld1 { v18.b }[10], [x23]\n"
+      "ld1 { v17.b }[10], [x21]\n"
+      "ld1 { v16.b }[10], [x20]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[8], [x23]\n"
-      "ld1 { v18.b }[8], [x21]\n"
-      "ld1 { v17.b }[8], [x20]\n"
-      "ld1 { v16.b }[8], [x19]\n"
+      "ld1 { v19.b }[8], [x24]\n"
+      "ld1 { v18.b }[8], [x23]\n"
+      "ld1 { v17.b }[8], [x21]\n"
+      "ld1 { v16.b }[8], [x20]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s19, [x23], #0x4\n"
-      "ldr s18, [x21], #0x4\n"
-      "ldr s17, [x20], #0x4\n"
-      "ldr s16, [x19], #0x4\n"
+      "ldr s19, [x24], #0x4\n"
+      "ldr s18, [x23], #0x4\n"
+      "ldr s17, [x21], #0x4\n"
+      "ldr s16, [x20], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v19.h }[2], [x23], #0x2\n"
-      "ld1 { v18.h }[2], [x21], #0x2\n"
-      "ld1 { v17.h }[2], [x20], #0x2\n"
-      "ld1 { v16.h }[2], [x19], #0x2\n"
+      "ld1 { v19.h }[2], [x24], #0x2\n"
+      "ld1 { v18.h }[2], [x23], #0x2\n"
+      "ld1 { v17.h }[2], [x21], #0x2\n"
+      "ld1 { v16.h }[2], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[6], [x23]\n"
-      "ld1 { v18.b }[6], [x21]\n"
-      "ld1 { v17.b }[6], [x20]\n"
-      "ld1 { v16.b }[6], [x19]\n"
+      "ld1 { v19.b }[6], [x24]\n"
+      "ld1 { v18.b }[6], [x23]\n"
+      "ld1 { v17.b }[6], [x21]\n"
+      "ld1 { v16.b }[6], [x20]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[4], [x23]\n"
-      "ld1 { v18.b }[4], [x21]\n"
-      "ld1 { v17.b }[4], [x20]\n"
-      "ld1 { v16.b }[4], [x19]\n"
+      "ld1 { v19.b }[4], [x24]\n"
+      "ld1 { v18.b }[4], [x23]\n"
+      "ld1 { v17.b }[4], [x21]\n"
+      "ld1 { v16.b }[4], [x20]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h19, [x23], #0x2\n"
-      "ldr h18, [x21], #0x2\n"
-      "ldr h17, [x20], #0x2\n"
-      "ldr h16, [x19], #0x2\n"
+      "ldr h19, [x24], #0x2\n"
+      "ldr h18, [x23], #0x2\n"
+      "ldr h17, [x21], #0x2\n"
+      "ldr h16, [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[2], [x23]\n"
-      "ld1 { v18.b }[2], [x21]\n"
-      "ld1 { v17.b }[2], [x20]\n"
-      "ld1 { v16.b }[2], [x19]\n"
+      "ld1 { v19.b }[2], [x24]\n"
+      "ld1 { v18.b }[2], [x23]\n"
+      "ld1 { v17.b }[2], [x21]\n"
+      "ld1 { v16.b }[2], [x20]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b19, [x23, #0x0]\n"
-      "ldr b18, [x21, #0x0]\n"
-      "ldr b17, [x20, #0x0]\n"
-      "ldr b16, [x19, #0x0]\n"
+      "ldr b19, [x24, #0x0]\n"
+      "ldr b18, [x23, #0x0]\n"
+      "ldr b17, [x21, #0x0]\n"
+      "ldr b16, [x20, #0x0]\n"
       "13:"  // Odd load end
       "str q19, [%x[out_ptr], #0x0]\n"
       "sadalp v28.8h, v19.16b\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
       "sadalp v27.8h, v18.16b\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
       "sadalp v26.8h, v17.16b\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
       "sadalp v25.8h, v16.16b\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "14:"  // Odds skip
       "sadalp v24.4s, v28.8h\n"
       "sadalp v23.4s, v27.8h\n"
-      "addp v24.4s, v24.4s, v23.4s\n"
       "sadalp v22.4s, v26.8h\n"
       "sadalp v21.4s, v25.8h\n"
+      "addp v24.4s, v24.4s, v23.4s\n"
       "addp v23.4s, v22.4s, v21.4s\n"
       "addp v24.4s, v24.4s, v23.4s\n"
       "add v24.4s, v24.4s, v20.4s\n"
@@ -217,7 +217,7 @@
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x19", "x20", "x21", "x22", "x23"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_u8_u8_summing.hpp
index c811462..4c7bb71 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,39 @@
 )
 {
   __asm__ __volatile__(
-      "movi v28.8h, #0x0\n"
-      "ldr x23, [%x[in], #0x0]\n"
-      "mov x22, #0x0\n"
-      "movi v27.8h, #0x0\n"
-      "ldr x21, [%x[in], #0x8]\n"
+      "ldr x24, [%x[in], #0x0]\n"
+      "ldr x23, [%x[in], #0x8]\n"
       "cmp %x[height], #0x4\n"
+      "mov x22, #0x0\n"
+      "ldr x21, [%x[in], #0x10]\n"
+      "ldr x20, [%x[in], #0x18]\n"
+      "movi v28.8h, #0x0\n"
+      "movi v27.8h, #0x0\n"
       "movi v26.8h, #0x0\n"
-      "ldr x20, [%x[in], #0x10]\n"
-      "add x23, x23, %x[row_offset]\n"
       "movi v25.8h, #0x0\n"
-      "ldr x19, [%x[in], #0x18]\n"
+      "add x24, x24, %x[row_offset]\n"
+      "add x23, x23, %x[row_offset]\n"
       "movi v24.4s, #0x0\n"
-      "add x21, x21, %x[row_offset]\n"
       "movi v23.4s, #0x0\n"
+      "add x21, x21, %x[row_offset]\n"
       "add x20, x20, %x[row_offset]\n"
       "movi v22.4s, #0x0\n"
-      "add x19, x19, %x[row_offset]\n"
       "movi v21.4s, #0x0\n"
       "beq 1f\n"
-      "mov x19, x23\n"
       "cmp %x[height], #0x2\n"
-      "csel x21, x21, x23, GE\n"
-      "csel x20, x20, x23, GT\n"
+      "mov x20, x24\n"
+      "csel x23, x23, x24, GE\n"
+      "csel x21, x21, x24, GT\n"
       "1:"  // no_pointer_adj
-      "movi v20.4s, #0x0\n"
+      "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
+      "movi v20.4s, #0x0\n"
       "prfm pldl1keep, [x21, #0x0]\n"
       "prfm pldl1keep, [x20, #0x0]\n"
-      "prfm pldl1keep, [x19, #0x0]\n"
+      "prfm pldl1keep, [x24, #0x40]\n"
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
       "prfm pldl1keep, [x20, #0x40]\n"
-      "prfm pldl1keep, [x19, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x10\n"
       "ld1 { v20.4s }, [%x[out_ptr]]\n"
@@ -75,141 +75,141 @@
       "ble 4f\n"
       "uadalp v24.4s, v28.8h\n"
       "movi v28.8h, #0x0\n"
+      "mov x22, #0x0\n"
       "uadalp v23.4s, v27.8h\n"
       "movi v27.8h, #0x0\n"
       "uadalp v22.4s, v26.8h\n"
       "movi v26.8h, #0x0\n"
       "uadalp v21.4s, v25.8h\n"
       "movi v25.8h, #0x0\n"
-      "mov x22, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q19, [x23], #0x10\n"
-      "add x22, x22, #0x1\n"
-      "ldr q18, [x21], #0x10\n"
+      "ldr q19, [x24], #0x10\n"
+      "ldr q18, [x23], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "ldr q16, [x19], #0x10\n"
-      "uadalp v28.8h, v19.16b\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "uadalp v27.8h, v18.16b\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "uadalp v26.8h, v17.16b\n"
-      "prfm pldl1keep, [x19, #0x70]\n"
-      "uadalp v25.8h, v16.16b\n"
+      "ldr q17, [x21], #0x10\n"
+      "ldr q16, [x20], #0x10\n"
       "str q19, [%x[out_ptr], #0x0]\n"
+      "uadalp v28.8h, v19.16b\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
       "str q18, [%x[out_ptr], #0x10]\n"
+      "uadalp v27.8h, v18.16b\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "prfm pldl1keep, [x20, #0x70]\n"
       "str q17, [%x[out_ptr], #0x20]\n"
+      "uadalp v26.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x30]\n"
+      "uadalp v25.8h, v16.16b\n"
+      "add x22, x22, #0x1\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d19, [x23], #0x8\n"
-      "ldr d18, [x21], #0x8\n"
-      "ldr d17, [x20], #0x8\n"
-      "ldr d16, [x19], #0x8\n"
+      "ldr d19, [x24], #0x8\n"
+      "ldr d18, [x23], #0x8\n"
+      "ldr d17, [x21], #0x8\n"
+      "ldr d16, [x20], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v19.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x21], #0x4\n"
-      "ld1 { v17.s }[2], [x20], #0x4\n"
-      "ld1 { v16.s }[2], [x19], #0x4\n"
+      "ld1 { v19.s }[2], [x24], #0x4\n"
+      "ld1 { v18.s }[2], [x23], #0x4\n"
+      "ld1 { v17.s }[2], [x21], #0x4\n"
+      "ld1 { v16.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v19.h }[6], [x23], #0x2\n"
-      "ld1 { v18.h }[6], [x21], #0x2\n"
-      "ld1 { v17.h }[6], [x20], #0x2\n"
-      "ld1 { v16.h }[6], [x19], #0x2\n"
+      "ld1 { v19.h }[6], [x24], #0x2\n"
+      "ld1 { v18.h }[6], [x23], #0x2\n"
+      "ld1 { v17.h }[6], [x21], #0x2\n"
+      "ld1 { v16.h }[6], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[14], [x23]\n"
-      "ld1 { v18.b }[14], [x21]\n"
-      "ld1 { v17.b }[14], [x20]\n"
-      "ld1 { v16.b }[14], [x19]\n"
+      "ld1 { v19.b }[14], [x24]\n"
+      "ld1 { v18.b }[14], [x23]\n"
+      "ld1 { v17.b }[14], [x21]\n"
+      "ld1 { v16.b }[14], [x20]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[12], [x23]\n"
-      "ld1 { v18.b }[12], [x21]\n"
-      "ld1 { v17.b }[12], [x20]\n"
-      "ld1 { v16.b }[12], [x19]\n"
+      "ld1 { v19.b }[12], [x24]\n"
+      "ld1 { v18.b }[12], [x23]\n"
+      "ld1 { v17.b }[12], [x21]\n"
+      "ld1 { v16.b }[12], [x20]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v19.h }[4], [x23], #0x2\n"
-      "ld1 { v18.h }[4], [x21], #0x2\n"
-      "ld1 { v17.h }[4], [x20], #0x2\n"
-      "ld1 { v16.h }[4], [x19], #0x2\n"
+      "ld1 { v19.h }[4], [x24], #0x2\n"
+      "ld1 { v18.h }[4], [x23], #0x2\n"
+      "ld1 { v17.h }[4], [x21], #0x2\n"
+      "ld1 { v16.h }[4], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[10], [x23]\n"
-      "ld1 { v18.b }[10], [x21]\n"
-      "ld1 { v17.b }[10], [x20]\n"
-      "ld1 { v16.b }[10], [x19]\n"
+      "ld1 { v19.b }[10], [x24]\n"
+      "ld1 { v18.b }[10], [x23]\n"
+      "ld1 { v17.b }[10], [x21]\n"
+      "ld1 { v16.b }[10], [x20]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[8], [x23]\n"
-      "ld1 { v18.b }[8], [x21]\n"
-      "ld1 { v17.b }[8], [x20]\n"
-      "ld1 { v16.b }[8], [x19]\n"
+      "ld1 { v19.b }[8], [x24]\n"
+      "ld1 { v18.b }[8], [x23]\n"
+      "ld1 { v17.b }[8], [x21]\n"
+      "ld1 { v16.b }[8], [x20]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s19, [x23], #0x4\n"
-      "ldr s18, [x21], #0x4\n"
-      "ldr s17, [x20], #0x4\n"
-      "ldr s16, [x19], #0x4\n"
+      "ldr s19, [x24], #0x4\n"
+      "ldr s18, [x23], #0x4\n"
+      "ldr s17, [x21], #0x4\n"
+      "ldr s16, [x20], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v19.h }[2], [x23], #0x2\n"
-      "ld1 { v18.h }[2], [x21], #0x2\n"
-      "ld1 { v17.h }[2], [x20], #0x2\n"
-      "ld1 { v16.h }[2], [x19], #0x2\n"
+      "ld1 { v19.h }[2], [x24], #0x2\n"
+      "ld1 { v18.h }[2], [x23], #0x2\n"
+      "ld1 { v17.h }[2], [x21], #0x2\n"
+      "ld1 { v16.h }[2], [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[6], [x23]\n"
-      "ld1 { v18.b }[6], [x21]\n"
-      "ld1 { v17.b }[6], [x20]\n"
-      "ld1 { v16.b }[6], [x19]\n"
+      "ld1 { v19.b }[6], [x24]\n"
+      "ld1 { v18.b }[6], [x23]\n"
+      "ld1 { v17.b }[6], [x21]\n"
+      "ld1 { v16.b }[6], [x20]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[4], [x23]\n"
-      "ld1 { v18.b }[4], [x21]\n"
-      "ld1 { v17.b }[4], [x20]\n"
-      "ld1 { v16.b }[4], [x19]\n"
+      "ld1 { v19.b }[4], [x24]\n"
+      "ld1 { v18.b }[4], [x23]\n"
+      "ld1 { v17.b }[4], [x21]\n"
+      "ld1 { v16.b }[4], [x20]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h19, [x23], #0x2\n"
-      "ldr h18, [x21], #0x2\n"
-      "ldr h17, [x20], #0x2\n"
-      "ldr h16, [x19], #0x2\n"
+      "ldr h19, [x24], #0x2\n"
+      "ldr h18, [x23], #0x2\n"
+      "ldr h17, [x21], #0x2\n"
+      "ldr h16, [x20], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v19.b }[2], [x23]\n"
-      "ld1 { v18.b }[2], [x21]\n"
-      "ld1 { v17.b }[2], [x20]\n"
-      "ld1 { v16.b }[2], [x19]\n"
+      "ld1 { v19.b }[2], [x24]\n"
+      "ld1 { v18.b }[2], [x23]\n"
+      "ld1 { v17.b }[2], [x21]\n"
+      "ld1 { v16.b }[2], [x20]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b19, [x23, #0x0]\n"
-      "ldr b18, [x21, #0x0]\n"
-      "ldr b17, [x20, #0x0]\n"
-      "ldr b16, [x19, #0x0]\n"
+      "ldr b19, [x24, #0x0]\n"
+      "ldr b18, [x23, #0x0]\n"
+      "ldr b17, [x21, #0x0]\n"
+      "ldr b16, [x20, #0x0]\n"
       "13:"  // Odd load end
       "str q19, [%x[out_ptr], #0x0]\n"
       "uadalp v28.8h, v19.16b\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
       "uadalp v27.8h, v18.16b\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
       "uadalp v26.8h, v17.16b\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
       "uadalp v25.8h, v16.16b\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "14:"  // Odds skip
       "uadalp v24.4s, v28.8h\n"
       "uadalp v23.4s, v27.8h\n"
-      "addp v24.4s, v24.4s, v23.4s\n"
       "uadalp v22.4s, v26.8h\n"
       "uadalp v21.4s, v25.8h\n"
+      "addp v24.4s, v24.4s, v23.4s\n"
       "addp v23.4s, v22.4s, v21.4s\n"
       "addp v24.4s, v24.4s, v23.4s\n"
       "add v24.4s, v24.4s, v20.4s\n"
@@ -217,7 +217,7 @@
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x19", "x20", "x21", "x22", "x23"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
index 4257429..2ba2aa8 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,45 +31,46 @@
 )
 {
   __asm__ __volatile__(
-      "movi v30.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "movi v16.8h, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x4\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -77,135 +78,134 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d29, [x27], #0x8\n"
-      "zip1 v29.8h, v30.8h, v29.8h\n"
-      "ldr d28, [x26], #0x8\n"
+      "ldr d28, [x28], #0x8\n"
+      "ldr d27, [x27], #0x8\n"
+      "shll v28.4s, v28.4h, #0x10\n"
+      "shll v27.4s, v27.4h, #0x10\n"
+      "ldr d22, [x26], #0x8\n"
+      "ldr d21, [x25], #0x8\n"
+      "shll v22.4s, v22.4h, #0x10\n"
+      "shll v21.4s, v21.4h, #0x10\n"
+      "ldr d26, [x24], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
+      "shll v26.4s, v26.4h, #0x10\n"
+      "shll v25.4s, v25.4h, #0x10\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d19, [x21], #0x8\n"
+      "shll v20.4s, v20.4h, #0x10\n"
+      "shll v19.4s, v19.4h, #0x10\n"
+      "zip1 v24.4s, v28.4s, v22.4s\n"
+      "zip1 v23.4s, v27.4s, v21.4s\n"
       "subs %x[width], %x[width], #0x4\n"
-      "zip1 v28.8h, v30.8h, v28.8h\n"
-      "ldr d24, [x25], #0x8\n"
       "cmp %x[width], #0x4\n"
-      "zip1 v24.8h, v30.8h, v24.8h\n"
-      "ldr d27, [x24], #0x8\n"
-      "ldr d26, [x23], #0x8\n"
-      "zip1 v25.4s, v29.4s, v24.4s\n"
-      "zip2 v24.4s, v29.4s, v24.4s\n"
-      "ldr d23, [x22], #0x8\n"
-      "ldr d22, [x21], #0x8\n"
-      "zip1 v27.8h, v30.8h, v27.8h\n"
-      "ldr d21, [x20], #0x8\n"
-      "zip1 v26.8h, v30.8h, v26.8h\n"
+      "zip1 v18.4s, v26.4s, v20.4s\n"
+      "zip1 v17.4s, v25.4s, v19.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip1 v20.4s, v28.4s, v27.4s\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v23.8h, v30.8h, v23.8h\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v22.8h, v30.8h, v22.8h\n"
+      "zip2 v20.4s, v26.4s, v20.4s\n"
+      "zip2 v19.4s, v25.4s, v19.4s\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v21.8h, v30.8h, v21.8h\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v17.4s, v25.4s, v20.4s\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v19.4s, v26.4s, v22.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip1 v18.4s, v23.4s, v21.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v17.4s, v25.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.4s, v19.4s, v18.4s\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
-      "zip2 v19.4s, v28.4s, v27.4s\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
-      "zip1 v16.4s, v24.4s, v19.4s\n"
-      "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v18.4s, v26.4s, v22.4s\n"
-      "zip2 v17.4s, v23.4s, v21.4s\n"
+      "zip1 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "zip1 v16.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v16.4s, v24.4s, v19.4s\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "zip2 v16.4s, v18.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
+      "zip2 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.4s, v18.4s, v17.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q17, [%x[out_ptr], #0x30]\n"
+      "zip1 v18.4s, v20.4s, v19.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x40]\n"
+      "zip2 v16.4s, v20.4s, v19.4s\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr s29, [x27], #0x4\n"
-      "ldr s28, [x26], #0x4\n"
-      "mov x19, #0x2\n"
-      "ldr s24, [x25], #0x4\n"
-      "ldr s27, [x24], #0x4\n"
-      "ldr s26, [x23], #0x4\n"
-      "ldr s23, [x22], #0x4\n"
-      "ldr s22, [x21], #0x4\n"
-      "ldr s21, [x20], #0x4\n"
+      "ldr s28, [x28], #0x4\n"
+      "ldr s27, [x27], #0x4\n"
+      "mov x20, #0x2\n"
+      "ldr s22, [x26], #0x4\n"
+      "ldr s21, [x25], #0x4\n"
+      "ldr s26, [x24], #0x4\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s19, [x21], #0x4\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v29.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v28.h }[2], [x26]\n"
-      "ld1 { v24.h }[2], [x25]\n"
-      "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v26.h }[2], [x23]\n"
-      "ld1 { v23.h }[2], [x22]\n"
-      "ld1 { v22.h }[2], [x21]\n"
-      "ld1 { v21.h }[2], [x20]\n"
+      "ld1 { v28.h }[2], [x28]\n"
+      "ld1 { v27.h }[2], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.h }[2], [x26]\n"
+      "ld1 { v21.h }[2], [x25]\n"
+      "ld1 { v26.h }[2], [x24]\n"
+      "ld1 { v25.h }[2], [x23]\n"
+      "ld1 { v20.h }[2], [x22]\n"
+      "ld1 { v19.h }[2], [x21]\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr h29, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr h28, [x26, #0x0]\n"
-      "ldr h24, [x25, #0x0]\n"
-      "ldr h27, [x24, #0x0]\n"
-      "ldr h26, [x23, #0x0]\n"
-      "ldr h23, [x22, #0x0]\n"
-      "ldr h22, [x21, #0x0]\n"
-      "ldr h21, [x20, #0x0]\n"
+      "ldr h28, [x28, #0x0]\n"
+      "ldr h27, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr h22, [x26, #0x0]\n"
+      "ldr h21, [x25, #0x0]\n"
+      "ldr h26, [x24, #0x0]\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h20, [x22, #0x0]\n"
+      "ldr h19, [x21, #0x0]\n"
       "5:"  // Odd load end
-      "zip1 v29.8h, v30.8h, v29.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v28.8h, v30.8h, v28.8h\n"
-      "zip1 v24.8h, v30.8h, v24.8h\n"
-      "zip1 v27.8h, v30.8h, v27.8h\n"
-      "zip1 v26.8h, v30.8h, v26.8h\n"
-      "zip1 v23.8h, v30.8h, v23.8h\n"
-      "zip1 v22.8h, v30.8h, v22.8h\n"
-      "zip1 v21.8h, v30.8h, v21.8h\n"
-      "zip1 v25.4s, v29.4s, v24.4s\n"
-      "zip1 v20.4s, v28.4s, v27.4s\n"
-      "zip1 v17.4s, v25.4s, v20.4s\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip1 v19.4s, v26.4s, v22.4s\n"
-      "zip1 v18.4s, v23.4s, v21.4s\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "beq 6f\n"
-      "zip2 v17.4s, v25.4s, v20.4s\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.4s, v19.4s, v18.4s\n"
-      "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "beq 6f\n"
-      "zip2 v24.4s, v29.4s, v24.4s\n"
-      "zip2 v19.4s, v28.4s, v27.4s\n"
-      "zip1 v16.4s, v24.4s, v19.4s\n"
+      "shll v28.4s, v28.4h, #0x10\n"
+      "shll v27.4s, v27.4h, #0x10\n"
+      "subs x20, x20, #0x1\n"
+      "shll v22.4s, v22.4h, #0x10\n"
+      "shll v21.4s, v21.4h, #0x10\n"
+      "shll v26.4s, v26.4h, #0x10\n"
+      "shll v25.4s, v25.4h, #0x10\n"
+      "shll v20.4s, v20.4h, #0x10\n"
+      "shll v19.4s, v19.4h, #0x10\n"
+      "zip1 v24.4s, v28.4s, v22.4s\n"
+      "zip1 v23.4s, v27.4s, v21.4s\n"
+      "zip1 v18.4s, v26.4s, v20.4s\n"
+      "zip1 v17.4s, v25.4s, v19.4s\n"
+      "zip1 v16.4s, v24.4s, v23.4s\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.4s, v26.4s, v22.4s\n"
-      "zip2 v17.4s, v23.4s, v21.4s\n"
       "zip1 v16.4s, v18.4s, v17.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "beq 6f\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v17.4s, v18.4s, v17.4s\n"
+      "str q17, [%x[out_ptr], #0x10]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "beq 6f\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "zip2 v20.4s, v26.4s, v20.4s\n"
+      "zip2 v19.4s, v25.4s, v19.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v18.4s, v20.4s, v19.4s\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "6:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp16.hpp
index 62d1657..f55c2be 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,193 +77,192 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
+      "ldr q25, [x28], #0x10\n"
       "ldr q30, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
+      "ldr q29, [x26], #0x10\n"
       "ldr q28, [x25], #0x10\n"
-      "ldr q27, [x24], #0x10\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "ldr q21, [x22], #0x10\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "ldr q24, [x21], #0x10\n"
-      "ldr q23, [x20], #0x10\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr q21, [x24], #0x10\n"
+      "ldr q27, [x23], #0x10\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v26.8h, v30.8h, v27.8h\n"
+      "ldr q20, [x22], #0x10\n"
+      "ldr q22, [x21], #0x10\n"
+      "zip1 v19.8h, v29.8h, v20.8h\n"
+      "zip1 v18.8h, v28.8h, v22.8h\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v22.8h\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
+      "zip1 v24.8h, v23.8h, v19.8h\n"
+      "zip1 v17.8h, v26.8h, v18.8h\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
+      "zip2 v23.8h, v23.8h, v19.8h\n"
+      "zip2 v19.8h, v26.8h, v18.8h\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip2 v16.8h, v21.8h, v20.8h\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
+      "ldr d25, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
-      "ldr d27, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "ldr d24, [x21], #0x8\n"
-      "ldr d23, [x20], #0x8\n"
+      "ldr d21, [x24], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d22, [x21], #0x8\n"
       "tbz %x[width], #1, 4f\n"
+      "ld1 { v25.s }[2], [x28], #0x4\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
-      "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v21.s }[2], [x22], #0x4\n"
-      "ld1 { v24.s }[2], [x21], #0x4\n"
-      "ld1 { v23.s }[2], [x20], #0x4\n"
+      "ld1 { v21.s }[2], [x24], #0x4\n"
+      "ld1 { v27.s }[2], [x23], #0x4\n"
+      "ld1 { v20.s }[2], [x22], #0x4\n"
+      "ld1 { v22.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[6], [x28]\n"
       "ld1 { v30.h }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
-      "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v25.h }[6], [x23]\n"
-      "ld1 { v21.h }[6], [x22]\n"
-      "ld1 { v24.h }[6], [x21]\n"
-      "ld1 { v23.h }[6], [x20]\n"
+      "ld1 { v21.h }[6], [x24]\n"
+      "ld1 { v27.h }[6], [x23]\n"
+      "ld1 { v20.h }[6], [x22]\n"
+      "ld1 { v22.h }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[4], [x28]\n"
       "ld1 { v30.h }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.h }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
-      "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v25.h }[4], [x23]\n"
-      "ld1 { v21.h }[4], [x22]\n"
-      "ld1 { v24.h }[4], [x21]\n"
-      "ld1 { v23.h }[4], [x20]\n"
+      "ld1 { v21.h }[4], [x24]\n"
+      "ld1 { v27.h }[4], [x23]\n"
+      "ld1 { v20.h }[4], [x22]\n"
+      "ld1 { v22.h }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
+      "ldr s25, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
+      "mov x20, #0x2\n"
       "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
-      "ldr s27, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s24, [x21], #0x4\n"
-      "ldr s23, [x20], #0x4\n"
+      "ldr s21, [x24], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s22, [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[2], [x28]\n"
       "ld1 { v30.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
-      "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v21.h }[2], [x22]\n"
-      "ld1 { v24.h }[2], [x21]\n"
-      "ld1 { v23.h }[2], [x20]\n"
+      "ld1 { v21.h }[2], [x24]\n"
+      "ld1 { v27.h }[2], [x23]\n"
+      "ld1 { v20.h }[2], [x22]\n"
+      "ld1 { v22.h }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
+      "ldr h25, [x28, #0x0]\n"
       "ldr h30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
-      "ldr h27, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h21, [x22, #0x0]\n"
-      "ldr h24, [x21, #0x0]\n"
-      "ldr h23, [x20, #0x0]\n"
+      "ldr h21, [x24, #0x0]\n"
+      "ldr h27, [x23, #0x0]\n"
+      "ldr h20, [x22, #0x0]\n"
+      "ldr h22, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v19.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v26.8h, v30.8h, v27.8h\n"
+      "zip1 v18.8h, v28.8h, v22.8h\n"
+      "zip1 v24.8h, v23.8h, v19.8h\n"
+      "zip1 v17.8h, v26.8h, v18.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "zip2 v23.8h, v23.8h, v19.8h\n"
+      "zip2 v19.8h, v26.8h, v18.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
       "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v22.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
+      "str q18, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp32.hpp
index b67840b..f64db0b 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp16_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x4\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,135 +77,134 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d30, [x27], #0x8\n"
-      "subs %x[width], %x[width], #0x4\n"
-      "ldr d29, [x26], #0x8\n"
-      "cmp %x[width], #0x4\n"
-      "ldr d28, [x25], #0x8\n"
-      "fcvtl v30.4s, v30.4h\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "fcvtl v29.4s, v29.4h\n"
-      "ldr d26, [x22], #0x8\n"
+      "ldr d28, [x28], #0x8\n"
+      "ldr d27, [x27], #0x8\n"
       "fcvtl v28.4s, v28.4h\n"
-      "zip1 v20.4s, v30.4s, v28.4s\n"
-      "ldr d25, [x21], #0x8\n"
-      "fcvtl v21.4s, v21.4h\n"
-      "zip2 v17.4s, v30.4s, v28.4s\n"
-      "ldr d24, [x20], #0x8\n"
       "fcvtl v27.4s, v27.4h\n"
-      "zip1 v18.4s, v29.4s, v21.4s\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "ldr d22, [x26], #0x8\n"
+      "ldr d21, [x25], #0x8\n"
+      "fcvtl v22.4s, v22.4h\n"
+      "fcvtl v21.4s, v21.4h\n"
+      "ldr d26, [x24], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
       "fcvtl v26.4s, v26.4h\n"
-      "zip1 v23.4s, v20.4s, v18.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "fcvtl v25.4s, v25.4h\n"
-      "zip2 v22.4s, v20.4s, v18.4s\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d19, [x21], #0x8\n"
+      "fcvtl v20.4s, v20.4h\n"
+      "fcvtl v19.4s, v19.4h\n"
+      "zip1 v24.4s, v28.4s, v22.4s\n"
+      "zip1 v23.4s, v27.4s, v21.4s\n"
+      "subs %x[width], %x[width], #0x4\n"
+      "cmp %x[width], #0x4\n"
+      "zip1 v18.4s, v26.4s, v20.4s\n"
+      "zip1 v17.4s, v25.4s, v19.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "prfm pldl1keep, [x26, #0x70]\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "fcvtl v24.4s, v24.4h\n"
-      "zip2 v16.4s, v29.4s, v21.4s\n"
+      "zip2 v20.4s, v26.4s, v20.4s\n"
+      "zip2 v19.4s, v25.4s, v19.4s\n"
       "prfm pldl1keep, [x24, #0x70]\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v21.4s, v17.4s, v16.4s\n"
-      "zip2 v20.4s, v17.4s, v16.4s\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip1 v19.4s, v27.4s, v25.4s\n"
-      "zip2 v18.4s, v27.4s, v25.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v17.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v16.4s, v19.4s, v17.4s\n"
+      "zip1 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v18.4s, v17.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.4s, v19.4s, v17.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
+      "zip2 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.4s, v18.4s, v17.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
       "str q17, [%x[out_ptr], #0x30]\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x50]\n"
-      "str q20, [%x[out_ptr], #0x60]\n"
+      "zip1 v18.4s, v20.4s, v19.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x40]\n"
+      "zip2 v16.4s, v20.4s, v19.4s\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr s30, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x2\n"
-      "ldr s28, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s27, [x23], #0x4\n"
-      "ldr s26, [x22], #0x4\n"
-      "ldr s25, [x21], #0x4\n"
-      "ldr s24, [x20], #0x4\n"
+      "ldr s28, [x28], #0x4\n"
+      "ldr s27, [x27], #0x4\n"
+      "mov x20, #0x2\n"
+      "ldr s22, [x26], #0x4\n"
+      "ldr s21, [x25], #0x4\n"
+      "ldr s26, [x24], #0x4\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s19, [x21], #0x4\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v30.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v29.h }[2], [x26]\n"
-      "ld1 { v28.h }[2], [x25]\n"
-      "ld1 { v21.h }[2], [x24]\n"
-      "ld1 { v27.h }[2], [x23]\n"
-      "ld1 { v26.h }[2], [x22]\n"
-      "ld1 { v25.h }[2], [x21]\n"
-      "ld1 { v24.h }[2], [x20]\n"
+      "ld1 { v28.h }[2], [x28]\n"
+      "ld1 { v27.h }[2], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.h }[2], [x26]\n"
+      "ld1 { v21.h }[2], [x25]\n"
+      "ld1 { v26.h }[2], [x24]\n"
+      "ld1 { v25.h }[2], [x23]\n"
+      "ld1 { v20.h }[2], [x22]\n"
+      "ld1 { v19.h }[2], [x21]\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr h30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr h29, [x26, #0x0]\n"
-      "ldr h28, [x25, #0x0]\n"
-      "ldr h21, [x24, #0x0]\n"
-      "ldr h27, [x23, #0x0]\n"
-      "ldr h26, [x22, #0x0]\n"
-      "ldr h25, [x21, #0x0]\n"
-      "ldr h24, [x20, #0x0]\n"
+      "ldr h28, [x28, #0x0]\n"
+      "ldr h27, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr h22, [x26, #0x0]\n"
+      "ldr h21, [x25, #0x0]\n"
+      "ldr h26, [x24, #0x0]\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h20, [x22, #0x0]\n"
+      "ldr h19, [x21, #0x0]\n"
       "5:"  // Odd load end
-      "fcvtl v30.4s, v30.4h\n"
-      "fcvtl v29.4s, v29.4h\n"
       "fcvtl v28.4s, v28.4h\n"
-      "zip1 v20.4s, v30.4s, v28.4s\n"
-      "fcvtl v21.4s, v21.4h\n"
       "fcvtl v27.4s, v27.4h\n"
-      "zip1 v18.4s, v29.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "fcvtl v22.4s, v22.4h\n"
+      "fcvtl v21.4s, v21.4h\n"
       "fcvtl v26.4s, v26.4h\n"
       "fcvtl v25.4s, v25.4h\n"
-      "zip1 v23.4s, v20.4s, v18.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v19.4s, v27.4s, v25.4s\n"
-      "fcvtl v24.4s, v24.4h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v17.4s, v26.4s, v24.4s\n"
-      "zip1 v16.4s, v19.4s, v17.4s\n"
+      "fcvtl v20.4s, v20.4h\n"
+      "fcvtl v19.4s, v19.4h\n"
+      "zip1 v24.4s, v28.4s, v22.4s\n"
+      "zip1 v23.4s, v27.4s, v21.4s\n"
+      "zip1 v18.4s, v26.4s, v20.4s\n"
+      "zip1 v17.4s, v25.4s, v19.4s\n"
+      "zip1 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v18.4s, v17.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v22.4s, v20.4s, v18.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v17.4s, v19.4s, v17.4s\n"
-      "subs x19, x19, #0x1\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.4s, v24.4s, v23.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v17.4s, v18.4s, v17.4s\n"
       "str q17, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v17.4s, v30.4s, v28.4s\n"
-      "zip2 v16.4s, v29.4s, v21.4s\n"
-      "zip1 v21.4s, v17.4s, v16.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.4s, v27.4s, v25.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "zip2 v20.4s, v26.4s, v20.4s\n"
+      "zip2 v19.4s, v25.4s, v19.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v18.4s, v20.4s, v19.4s\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "6:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp32_fp32.hpp
index eefb854..6c009b3 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_fp32_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #2\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #2\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #2\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #2\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #2\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #2\n"
       "add x22, x22, %x[row_offset], LSL #2\n"
       "add x21, x21, %x[row_offset], LSL #2\n"
-      "add x20, x20, %x[row_offset], LSL #2\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x4\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,49 +77,48 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q28, [x27], #0x10\n"
+      "ldr q28, [x28], #0x10\n"
+      "ldr q27, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x4\n"
-      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x4\n"
-      "ldr q25, [x25], #0x10\n"
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "ldr q21, [x24], #0x10\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "ldr q27, [x23], #0x10\n"
-      "ldr q26, [x22], #0x10\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "ldr q24, [x20], #0x10\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "ldr q22, [x26], #0x10\n"
+      "ldr q21, [x25], #0x10\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "ldr q24, [x24], #0x10\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "ldr q19, [x22], #0x10\n"
+      "ldr q18, [x21], #0x10\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x40]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v19.4s, v18.4s\n"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -126,69 +126,69 @@
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr d28, [x27], #0x8\n"
-      "ldr d29, [x26], #0x8\n"
-      "mov x19, #0x2\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "ldr d26, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d24, [x20], #0x8\n"
+      "ldr d28, [x28], #0x8\n"
+      "ldr d27, [x27], #0x8\n"
+      "mov x20, #0x2\n"
+      "ldr d22, [x26], #0x8\n"
+      "ldr d21, [x25], #0x8\n"
+      "ldr d24, [x24], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
+      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x21], #0x8\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v28.s }[2], [x27]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v29.s }[2], [x26]\n"
-      "ld1 { v25.s }[2], [x25]\n"
-      "ld1 { v21.s }[2], [x24]\n"
-      "ld1 { v27.s }[2], [x23]\n"
-      "ld1 { v26.s }[2], [x22]\n"
-      "ld1 { v19.s }[2], [x21]\n"
-      "ld1 { v24.s }[2], [x20]\n"
+      "ld1 { v28.s }[2], [x28]\n"
+      "ld1 { v27.s }[2], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.s }[2], [x26]\n"
+      "ld1 { v21.s }[2], [x25]\n"
+      "ld1 { v24.s }[2], [x24]\n"
+      "ld1 { v23.s }[2], [x23]\n"
+      "ld1 { v19.s }[2], [x22]\n"
+      "ld1 { v18.s }[2], [x21]\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr s28, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr s29, [x26, #0x0]\n"
-      "ldr s25, [x25, #0x0]\n"
-      "ldr s21, [x24, #0x0]\n"
-      "ldr s27, [x23, #0x0]\n"
-      "ldr s26, [x22, #0x0]\n"
-      "ldr s19, [x21, #0x0]\n"
-      "ldr s24, [x20, #0x0]\n"
+      "ldr s28, [x28, #0x0]\n"
+      "ldr s27, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr s22, [x26, #0x0]\n"
+      "ldr s21, [x25, #0x0]\n"
+      "ldr s24, [x24, #0x0]\n"
+      "ldr s23, [x23, #0x0]\n"
+      "ldr s19, [x22, #0x0]\n"
+      "ldr s18, [x21, #0x0]\n"
       "5:"  // Odd load end
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
-      "subs x19, x19, #0x1\n"
-      "str q20, [%x[out_ptr], #0x10]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "6:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16.hpp
index b0523b9..767d468 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,193 +77,192 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
+      "ldr q25, [x28], #0x10\n"
       "ldr q30, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
+      "ldr q29, [x26], #0x10\n"
       "ldr q28, [x25], #0x10\n"
-      "ldr q27, [x24], #0x10\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "ldr q21, [x22], #0x10\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "ldr q24, [x21], #0x10\n"
-      "ldr q23, [x20], #0x10\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr q21, [x24], #0x10\n"
+      "ldr q27, [x23], #0x10\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v26.8h, v30.8h, v27.8h\n"
+      "ldr q20, [x22], #0x10\n"
+      "ldr q22, [x21], #0x10\n"
+      "zip1 v19.8h, v29.8h, v20.8h\n"
+      "zip1 v18.8h, v28.8h, v22.8h\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v22.8h\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
+      "zip1 v24.8h, v23.8h, v19.8h\n"
+      "zip1 v17.8h, v26.8h, v18.8h\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
+      "zip2 v23.8h, v23.8h, v19.8h\n"
+      "zip2 v19.8h, v26.8h, v18.8h\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip2 v16.8h, v21.8h, v20.8h\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
+      "ldr d25, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
-      "ldr d27, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "ldr d24, [x21], #0x8\n"
-      "ldr d23, [x20], #0x8\n"
+      "ldr d21, [x24], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d22, [x21], #0x8\n"
       "tbz %x[width], #1, 4f\n"
+      "ld1 { v25.s }[2], [x28], #0x4\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
-      "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v21.s }[2], [x22], #0x4\n"
-      "ld1 { v24.s }[2], [x21], #0x4\n"
-      "ld1 { v23.s }[2], [x20], #0x4\n"
+      "ld1 { v21.s }[2], [x24], #0x4\n"
+      "ld1 { v27.s }[2], [x23], #0x4\n"
+      "ld1 { v20.s }[2], [x22], #0x4\n"
+      "ld1 { v22.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[6], [x28]\n"
       "ld1 { v30.h }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
-      "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v25.h }[6], [x23]\n"
-      "ld1 { v21.h }[6], [x22]\n"
-      "ld1 { v24.h }[6], [x21]\n"
-      "ld1 { v23.h }[6], [x20]\n"
+      "ld1 { v21.h }[6], [x24]\n"
+      "ld1 { v27.h }[6], [x23]\n"
+      "ld1 { v20.h }[6], [x22]\n"
+      "ld1 { v22.h }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[4], [x28]\n"
       "ld1 { v30.h }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.h }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
-      "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v25.h }[4], [x23]\n"
-      "ld1 { v21.h }[4], [x22]\n"
-      "ld1 { v24.h }[4], [x21]\n"
-      "ld1 { v23.h }[4], [x20]\n"
+      "ld1 { v21.h }[4], [x24]\n"
+      "ld1 { v27.h }[4], [x23]\n"
+      "ld1 { v20.h }[4], [x22]\n"
+      "ld1 { v22.h }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
+      "ldr s25, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
+      "mov x20, #0x2\n"
       "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
-      "ldr s27, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s24, [x21], #0x4\n"
-      "ldr s23, [x20], #0x4\n"
+      "ldr s21, [x24], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s22, [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
+      "ld1 { v25.h }[2], [x28]\n"
       "ld1 { v30.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
-      "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v21.h }[2], [x22]\n"
-      "ld1 { v24.h }[2], [x21]\n"
-      "ld1 { v23.h }[2], [x20]\n"
+      "ld1 { v21.h }[2], [x24]\n"
+      "ld1 { v27.h }[2], [x23]\n"
+      "ld1 { v20.h }[2], [x22]\n"
+      "ld1 { v22.h }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
+      "ldr h25, [x28, #0x0]\n"
       "ldr h30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
-      "ldr h27, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h21, [x22, #0x0]\n"
-      "ldr h24, [x21, #0x0]\n"
-      "ldr h23, [x20, #0x0]\n"
+      "ldr h21, [x24, #0x0]\n"
+      "ldr h27, [x23, #0x0]\n"
+      "ldr h20, [x22, #0x0]\n"
+      "ldr h22, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v19.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v26.8h, v30.8h, v27.8h\n"
+      "zip1 v18.8h, v28.8h, v22.8h\n"
+      "zip1 v24.8h, v23.8h, v19.8h\n"
+      "zip1 v17.8h, v26.8h, v18.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "zip2 v23.8h, v23.8h, v19.8h\n"
+      "zip2 v19.8h, v26.8h, v18.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
       "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v22.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
+      "str q18, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16_summing.hpp
index 292a38f..a737920 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s16_s16_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,40 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.4s, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.4s, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -71,7 +72,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -79,226 +80,225 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v0.4s }, [%x[out_ptr]]\n"
-      "ldr q31, [%x[out_ptr], #0x10]\n"
+      "ld1 { v1.4s }, [%x[out_ptr]]\n"
+      "ldr q0, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x8\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0xe\n"
+      "cmp x20, #0xe\n"
       "ble 4f\n"
-      "saddw v0.4s, v0.4s, v1.4h\n"
-      "saddw2 v31.4s, v31.4s, v1.8h\n"
-      "mov x19, #0x0\n"
-      "movi v1.8h, #0x0\n"
+      "saddw v1.4s, v1.4s, v2.4h\n"
+      "saddw2 v0.4s, v0.4s, v2.8h\n"
+      "mov x20, #0x0\n"
+      "movi v2.8h, #0x0\n"
       "4:"  // no_accumulate_16
+      "ldr q31, [x28], #0x10\n"
       "ldr q30, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q28, [x25], #0x10\n"
       "cmp %x[width], #0x8\n"
+      "ldr q29, [x26], #0x10\n"
+      "ldr q28, [x25], #0x10\n"
+      "add x20, x20, #0x1\n"
       "ldr q27, [x24], #0x10\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "ldr q21, [x22], #0x10\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "ldr q24, [x21], #0x10\n"
-      "ldr q23, [x20], #0x10\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr q26, [x23], #0x10\n"
+      "zip1 v25.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v30.8h, v26.8h\n"
+      "ldr q24, [x22], #0x10\n"
+      "ldr q23, [x21], #0x10\n"
+      "zip1 v18.8h, v29.8h, v24.8h\n"
+      "zip1 v21.8h, v28.8h, v23.8h\n"
+      "zip1 v17.8h, v25.8h, v18.8h\n"
+      "zip1 v16.8h, v22.8h, v21.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "zip1 v20.8h, v17.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v20.8h\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
+      "zip2 v19.8h, v17.8h, v16.8h\n"
+      "zip2 v18.8h, v25.8h, v18.8h\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
+      "zip2 v17.8h, v22.8h, v21.8h\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "str q20, [%x[out_ptr], #0x0]\n"
+      "zip2 v21.8h, v29.8h, v24.8h\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "str q19, [%x[out_ptr], #0x10]\n"
+      "zip2 v19.8h, v28.8h, v23.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x20]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x70]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 10f\n"
       "tbz %x[width], #2, 7f\n"
+      "ldr d31, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "ldr d24, [x21], #0x8\n"
-      "ldr d23, [x20], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
+      "ldr d24, [x22], #0x8\n"
+      "ldr d23, [x21], #0x8\n"
       "tbz %x[width], #1, 6f\n"
+      "ld1 { v31.s }[2], [x28], #0x4\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v21.s }[2], [x22], #0x4\n"
-      "ld1 { v24.s }[2], [x21], #0x4\n"
-      "ld1 { v23.s }[2], [x20], #0x4\n"
+      "ld1 { v26.s }[2], [x23], #0x4\n"
+      "ld1 { v24.s }[2], [x22], #0x4\n"
+      "ld1 { v23.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[6], [x28]\n"
       "ld1 { v30.h }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v25.h }[6], [x23]\n"
-      "ld1 { v21.h }[6], [x22]\n"
-      "ld1 { v24.h }[6], [x21]\n"
-      "ld1 { v23.h }[6], [x20]\n"
+      "ld1 { v26.h }[6], [x23]\n"
+      "ld1 { v24.h }[6], [x22]\n"
+      "ld1 { v23.h }[6], [x21]\n"
       "b 9f\n"
       "6:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[4], [x28]\n"
       "ld1 { v30.h }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.h }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v25.h }[4], [x23]\n"
-      "ld1 { v21.h }[4], [x22]\n"
-      "ld1 { v24.h }[4], [x21]\n"
-      "ld1 { v23.h }[4], [x20]\n"
+      "ld1 { v26.h }[4], [x23]\n"
+      "ld1 { v24.h }[4], [x22]\n"
+      "ld1 { v23.h }[4], [x21]\n"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
+      "ldr s31, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
+      "mov x20, #0x2\n"
       "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s24, [x21], #0x4\n"
-      "ldr s23, [x20], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s24, [x22], #0x4\n"
+      "ldr s23, [x21], #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[2], [x28]\n"
       "ld1 { v30.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v21.h }[2], [x22]\n"
-      "ld1 { v24.h }[2], [x21]\n"
-      "ld1 { v23.h }[2], [x20]\n"
+      "ld1 { v26.h }[2], [x23]\n"
+      "ld1 { v24.h }[2], [x22]\n"
+      "ld1 { v23.h }[2], [x21]\n"
       "b 9f\n"
       "8:"  // odd_loads_1_0
+      "ldr h31, [x28, #0x0]\n"
       "ldr h30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h21, [x22, #0x0]\n"
-      "ldr h24, [x21, #0x0]\n"
-      "ldr h23, [x20, #0x0]\n"
+      "ldr h26, [x23, #0x0]\n"
+      "ldr h24, [x22, #0x0]\n"
+      "ldr h23, [x21, #0x0]\n"
       "9:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "zip1 v25.8h, v31.8h, v27.8h\n"
+      "zip1 v18.8h, v29.8h, v24.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v22.8h, v30.8h, v26.8h\n"
+      "zip1 v21.8h, v28.8h, v23.8h\n"
+      "zip1 v17.8h, v25.8h, v18.8h\n"
+      "zip1 v16.8h, v22.8h, v21.8h\n"
+      "zip1 v20.8h, v17.8h, v16.8h\n"
+      "str q20, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v20.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v19.8h, v17.8h, v16.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v19.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 10f\n"
+      "zip2 v18.8h, v25.8h, v18.8h\n"
+      "zip2 v17.8h, v22.8h, v21.8h\n"
+      "subs x20, x20, #0x1\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "beq 10f\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "zip2 v21.8h, v29.8h, v24.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "zip2 v19.8h, v28.8h, v23.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "beq 10f\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "10:"  // Odds skip
-      "saddw v0.4s, v0.4s, v1.4h\n"
-      "str q0, [%x[out_ptr], #0x0]\n"
-      "saddw2 v31.4s, v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x10]\n"
+      "saddw v1.4s, v1.4s, v2.4h\n"
+      "saddw2 v0.4s, v0.4s, v2.8h\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
+      "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16.hpp
index 6cfed8f..4a38187 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,209 +77,208 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d31, [x27], #0x8\n"
-      "sshll v31.8h, v31.8b, #0x0\n"
-      "ldr d30, [x26], #0x8\n"
-      "subs %x[width], %x[width], #0x8\n"
-      "sshll v30.8h, v30.8b, #0x0\n"
-      "ldr d29, [x25], #0x8\n"
-      "cmp %x[width], #0x8\n"
-      "sshll v29.8h, v29.8b, #0x0\n"
-      "ldr d28, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "sshll v28.8h, v28.8b, #0x0\n"
-      "ldr d23, [x22], #0x8\n"
+      "ldr d25, [x28], #0x8\n"
+      "ldr d30, [x27], #0x8\n"
       "sshll v25.8h, v25.8b, #0x0\n"
-      "ldr d27, [x21], #0x8\n"
-      "zip1 v20.8h, v31.8h, v25.8h\n"
-      "ldr d26, [x20], #0x8\n"
-      "zip2 v25.8h, v31.8h, v25.8h\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
+      "sshll v30.8h, v30.8b, #0x0\n"
+      "ldr d29, [x26], #0x8\n"
+      "ldr d28, [x25], #0x8\n"
+      "sshll v29.8h, v29.8b, #0x0\n"
+      "sshll v28.8h, v28.8b, #0x0\n"
+      "ldr d21, [x24], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "sshll v21.8h, v21.8b, #0x0\n"
       "sshll v27.8h, v27.8b, #0x0\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v19.8h, v29.8h, v27.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v21.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v19.8h, v29.8h, v27.8h\n"
-      "zip1 v20.8h, v25.8h, v19.8h\n"
-      "zip2 v19.8h, v25.8h, v19.8h\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d26, [x21], #0x8\n"
+      "sshll v20.8h, v20.8b, #0x0\n"
       "sshll v26.8h, v26.8b, #0x0\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v22.8h, v29.8h, v20.8h\n"
+      "subs %x[width], %x[width], #0x8\n"
+      "cmp %x[width], #0x8\n"
+      "zip1 v19.8h, v30.8h, v27.8h\n"
       "zip1 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "prfm pldl1keep, [x26, #0x70]\n"
+      "prfm pldl1keep, [x25, #0x70]\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v26.8h\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "zip1 v24.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v19.8h, v18.8h\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v23.8h, v23.8h, v22.8h\n"
+      "zip2 v19.8h, v19.8h, v18.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v22.8h, v17.8h\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v21.8h, v17.8h\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v19.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "zip2 v16.8h, v19.8h, v17.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip2 v16.8h, v21.8h, v20.8h\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
-      "ldr s31, [x27], #0x4\n"
-      "ldr s30, [x26], #0x4\n"
-      "ldr s29, [x25], #0x4\n"
-      "ldr s28, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s23, [x22], #0x4\n"
-      "ldr s27, [x21], #0x4\n"
-      "ldr s26, [x20], #0x4\n"
+      "ldr s25, [x28], #0x4\n"
+      "ldr s30, [x27], #0x4\n"
+      "ldr s29, [x26], #0x4\n"
+      "ldr s28, [x25], #0x4\n"
+      "ldr s21, [x24], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s26, [x21], #0x4\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v31.h }[2], [x27], #0x2\n"
-      "mov x19, #0x6\n"
-      "ld1 { v30.h }[2], [x26], #0x2\n"
-      "ld1 { v29.h }[2], [x25], #0x2\n"
-      "ld1 { v28.h }[2], [x24], #0x2\n"
-      "ld1 { v25.h }[2], [x23], #0x2\n"
-      "ld1 { v23.h }[2], [x22], #0x2\n"
-      "ld1 { v27.h }[2], [x21], #0x2\n"
-      "ld1 { v26.h }[2], [x20], #0x2\n"
+      "ld1 { v25.h }[2], [x28], #0x2\n"
+      "ld1 { v30.h }[2], [x27], #0x2\n"
+      "mov x20, #0x6\n"
+      "ld1 { v29.h }[2], [x26], #0x2\n"
+      "ld1 { v28.h }[2], [x25], #0x2\n"
+      "ld1 { v21.h }[2], [x24], #0x2\n"
+      "ld1 { v27.h }[2], [x23], #0x2\n"
+      "ld1 { v20.h }[2], [x22], #0x2\n"
+      "ld1 { v26.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[6], [x27]\n"
-      "mov x19, #0x7\n"
-      "ld1 { v30.b }[6], [x26]\n"
-      "ld1 { v29.b }[6], [x25]\n"
-      "ld1 { v28.b }[6], [x24]\n"
-      "ld1 { v25.b }[6], [x23]\n"
-      "ld1 { v23.b }[6], [x22]\n"
-      "ld1 { v27.b }[6], [x21]\n"
-      "ld1 { v26.b }[6], [x20]\n"
+      "ld1 { v25.b }[6], [x28]\n"
+      "ld1 { v30.b }[6], [x27]\n"
+      "mov x20, #0x7\n"
+      "ld1 { v29.b }[6], [x26]\n"
+      "ld1 { v28.b }[6], [x25]\n"
+      "ld1 { v21.b }[6], [x24]\n"
+      "ld1 { v27.b }[6], [x23]\n"
+      "ld1 { v20.b }[6], [x22]\n"
+      "ld1 { v26.b }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[4], [x27]\n"
-      "ld1 { v30.b }[4], [x26]\n"
-      "mov x19, #0x5\n"
-      "ld1 { v29.b }[4], [x25]\n"
-      "ld1 { v28.b }[4], [x24]\n"
-      "ld1 { v25.b }[4], [x23]\n"
-      "ld1 { v23.b }[4], [x22]\n"
-      "ld1 { v27.b }[4], [x21]\n"
-      "ld1 { v26.b }[4], [x20]\n"
+      "ld1 { v25.b }[4], [x28]\n"
+      "ld1 { v30.b }[4], [x27]\n"
+      "mov x20, #0x5\n"
+      "ld1 { v29.b }[4], [x26]\n"
+      "ld1 { v28.b }[4], [x25]\n"
+      "ld1 { v21.b }[4], [x24]\n"
+      "ld1 { v27.b }[4], [x23]\n"
+      "ld1 { v20.b }[4], [x22]\n"
+      "ld1 { v26.b }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr h31, [x27], #0x2\n"
-      "ldr h30, [x26], #0x2\n"
-      "mov x19, #0x2\n"
-      "ldr h29, [x25], #0x2\n"
-      "ldr h28, [x24], #0x2\n"
-      "ldr h25, [x23], #0x2\n"
-      "ldr h23, [x22], #0x2\n"
-      "ldr h27, [x21], #0x2\n"
-      "ldr h26, [x20], #0x2\n"
+      "ldr h25, [x28], #0x2\n"
+      "ldr h30, [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ldr h29, [x26], #0x2\n"
+      "ldr h28, [x25], #0x2\n"
+      "ldr h21, [x24], #0x2\n"
+      "ldr h27, [x23], #0x2\n"
+      "ldr h20, [x22], #0x2\n"
+      "ldr h26, [x21], #0x2\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[2], [x27]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v30.b }[2], [x26]\n"
-      "ld1 { v29.b }[2], [x25]\n"
-      "ld1 { v28.b }[2], [x24]\n"
-      "ld1 { v25.b }[2], [x23]\n"
-      "ld1 { v23.b }[2], [x22]\n"
-      "ld1 { v27.b }[2], [x21]\n"
-      "ld1 { v26.b }[2], [x20]\n"
+      "ld1 { v25.b }[2], [x28]\n"
+      "ld1 { v30.b }[2], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v29.b }[2], [x26]\n"
+      "ld1 { v28.b }[2], [x25]\n"
+      "ld1 { v21.b }[2], [x24]\n"
+      "ld1 { v27.b }[2], [x23]\n"
+      "ld1 { v20.b }[2], [x22]\n"
+      "ld1 { v26.b }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr b31, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b30, [x26, #0x0]\n"
-      "ldr b29, [x25, #0x0]\n"
-      "ldr b28, [x24, #0x0]\n"
-      "ldr b25, [x23, #0x0]\n"
-      "ldr b23, [x22, #0x0]\n"
-      "ldr b27, [x21, #0x0]\n"
-      "ldr b26, [x20, #0x0]\n"
+      "ldr b25, [x28, #0x0]\n"
+      "ldr b30, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b29, [x26, #0x0]\n"
+      "ldr b28, [x25, #0x0]\n"
+      "ldr b21, [x24, #0x0]\n"
+      "ldr b27, [x23, #0x0]\n"
+      "ldr b20, [x22, #0x0]\n"
+      "ldr b26, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "sshll v31.8h, v31.8b, #0x0\n"
-      "subs x19, x19, #0x1\n"
+      "sshll v25.8h, v25.8b, #0x0\n"
       "sshll v30.8h, v30.8b, #0x0\n"
+      "subs x20, x20, #0x1\n"
       "sshll v29.8h, v29.8b, #0x0\n"
       "sshll v28.8h, v28.8b, #0x0\n"
-      "sshll v25.8h, v25.8b, #0x0\n"
-      "zip1 v20.8h, v31.8h, v25.8h\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
+      "sshll v21.8h, v21.8b, #0x0\n"
       "sshll v27.8h, v27.8b, #0x0\n"
-      "zip1 v19.8h, v29.8h, v27.8h\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
+      "sshll v20.8h, v20.8b, #0x0\n"
       "sshll v26.8h, v26.8b, #0x0\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v22.8h, v29.8h, v20.8h\n"
+      "zip1 v19.8h, v30.8h, v27.8h\n"
       "zip1 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "zip1 v24.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v19.8h, v18.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v22.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v21.8h, v20.8h, v19.8h\n"
-      "zip2 v17.8h, v24.8h, v18.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v21.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v23.8h, v23.8h, v22.8h\n"
+      "zip2 v19.8h, v19.8h, v18.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v25.8h, v31.8h, v25.8h\n"
-      "zip2 v19.8h, v29.8h, v27.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v25.8h, v19.8h\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v26.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v19.8h, v25.8h, v19.8h\n"
-      "zip2 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v19.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
+      "str q18, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16_summing.hpp
index b710861..3ad103c 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_s8_s16_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,40 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.4s, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.4s, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -71,7 +72,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -79,242 +80,241 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v0.4s }, [%x[out_ptr]]\n"
-      "ldr q31, [%x[out_ptr], #0x10]\n"
+      "ld1 { v1.4s }, [%x[out_ptr]]\n"
+      "ldr q0, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x8\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0xe\n"
+      "cmp x20, #0xe\n"
       "ble 4f\n"
-      "saddw v0.4s, v0.4s, v1.4h\n"
-      "saddw2 v31.4s, v31.4s, v1.8h\n"
-      "mov x19, #0x0\n"
-      "movi v1.8h, #0x0\n"
+      "saddw v1.4s, v1.4s, v2.4h\n"
+      "saddw2 v0.4s, v0.4s, v2.8h\n"
+      "mov x20, #0x0\n"
+      "movi v2.8h, #0x0\n"
       "4:"  // no_accumulate_16
+      "ldr d31, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
+      "sshll v31.8h, v31.8b, #0x0\n"
       "sshll v30.8h, v30.8b, #0x0\n"
       "ldr d29, [x26], #0x8\n"
-      "add x19, x19, #0x1\n"
-      "sshll v29.8h, v29.8b, #0x0\n"
       "ldr d28, [x25], #0x8\n"
-      "subs %x[width], %x[width], #0x8\n"
+      "sshll v29.8h, v29.8b, #0x0\n"
       "sshll v28.8h, v28.8b, #0x0\n"
       "ldr d27, [x24], #0x8\n"
-      "cmp %x[width], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
       "sshll v27.8h, v27.8b, #0x0\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d23, [x22], #0x8\n"
-      "sshll v24.8h, v24.8b, #0x0\n"
-      "ldr d21, [x21], #0x8\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "ldr d26, [x20], #0x8\n"
-      "zip1 v20.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "zip1 v25.8h, v29.8h, v23.8h\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "zip2 v23.8h, v29.8h, v23.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "sshll v21.8h, v21.8b, #0x0\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v19.8h, v28.8h, v21.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v19.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v20.8h, v28.8h, v21.8h\n"
-      "zip1 v21.8h, v24.8h, v20.8h\n"
-      "zip2 v20.8h, v24.8h, v20.8h\n"
       "sshll v26.8h, v26.8b, #0x0\n"
-      "zip1 v18.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v25.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "ldr d25, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "sshll v25.8h, v25.8b, #0x0\n"
+      "sshll v24.8h, v24.8b, #0x0\n"
+      "zip1 v23.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v29.8h, v25.8h\n"
+      "subs %x[width], %x[width], #0x8\n"
+      "cmp %x[width], #0x8\n"
+      "zip1 v21.8h, v30.8h, v26.8h\n"
+      "zip1 v20.8h, v28.8h, v24.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip1 v18.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "prfm pldl1keep, [x26, #0x70]\n"
+      "prfm pldl1keep, [x25, #0x70]\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "zip2 v19.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v23.8h, v22.8h\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v17.8h, v21.8h, v20.8h\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v22.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.8h, v25.8h, v18.8h\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "zip1 v17.8h, v19.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v19.8h, v16.8h\n"
+      "add x20, x20, #0x1\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "str q19, [%x[out_ptr], #0x10]\n"
+      "zip2 v21.8h, v29.8h, v25.8h\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v19.8h, v28.8h, v24.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "zip2 v19.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v19.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v18.8h, v21.8h, v17.8h\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v23.8h, v19.8h\n"
-      "zip1 v17.8h, v20.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v18.8h\n"
-      "zip2 v16.8h, v20.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "str q16, [%x[out_ptr], #0x50]\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "str q16, [%x[out_ptr], #0x60]\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x70]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 10f\n"
       "tbz %x[width], #2, 7f\n"
+      "ldr s31, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s23, [x22], #0x4\n"
-      "ldr s21, [x21], #0x4\n"
-      "ldr s26, [x20], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s25, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
+      "ld1 { v31.h }[2], [x28], #0x2\n"
       "ld1 { v30.h }[2], [x27], #0x2\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.h }[2], [x26], #0x2\n"
       "ld1 { v28.h }[2], [x25], #0x2\n"
       "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v24.h }[2], [x23], #0x2\n"
-      "ld1 { v23.h }[2], [x22], #0x2\n"
-      "ld1 { v21.h }[2], [x21], #0x2\n"
-      "ld1 { v26.h }[2], [x20], #0x2\n"
+      "ld1 { v26.h }[2], [x23], #0x2\n"
+      "ld1 { v25.h }[2], [x22], #0x2\n"
+      "ld1 { v24.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[6], [x28]\n"
       "ld1 { v30.b }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.b }[6], [x26]\n"
       "ld1 { v28.b }[6], [x25]\n"
       "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v24.b }[6], [x23]\n"
-      "ld1 { v23.b }[6], [x22]\n"
-      "ld1 { v21.b }[6], [x21]\n"
-      "ld1 { v26.b }[6], [x20]\n"
+      "ld1 { v26.b }[6], [x23]\n"
+      "ld1 { v25.b }[6], [x22]\n"
+      "ld1 { v24.b }[6], [x21]\n"
       "b 9f\n"
       "6:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[4], [x28]\n"
       "ld1 { v30.b }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.b }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.b }[4], [x25]\n"
       "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v24.b }[4], [x23]\n"
-      "ld1 { v23.b }[4], [x22]\n"
-      "ld1 { v21.b }[4], [x21]\n"
-      "ld1 { v26.b }[4], [x20]\n"
+      "ld1 { v26.b }[4], [x23]\n"
+      "ld1 { v25.b }[4], [x22]\n"
+      "ld1 { v24.b }[4], [x21]\n"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
+      "ldr h31, [x28], #0x2\n"
       "ldr h30, [x27], #0x2\n"
+      "mov x20, #0x2\n"
       "ldr h29, [x26], #0x2\n"
-      "mov x19, #0x2\n"
       "ldr h28, [x25], #0x2\n"
       "ldr h27, [x24], #0x2\n"
-      "ldr h24, [x23], #0x2\n"
-      "ldr h23, [x22], #0x2\n"
-      "ldr h21, [x21], #0x2\n"
-      "ldr h26, [x20], #0x2\n"
+      "ldr h26, [x23], #0x2\n"
+      "ldr h25, [x22], #0x2\n"
+      "ldr h24, [x21], #0x2\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[2], [x28]\n"
       "ld1 { v30.b }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.b }[2], [x26]\n"
       "ld1 { v28.b }[2], [x25]\n"
       "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v24.b }[2], [x23]\n"
-      "ld1 { v23.b }[2], [x22]\n"
-      "ld1 { v21.b }[2], [x21]\n"
-      "ld1 { v26.b }[2], [x20]\n"
+      "ld1 { v26.b }[2], [x23]\n"
+      "ld1 { v25.b }[2], [x22]\n"
+      "ld1 { v24.b }[2], [x21]\n"
       "b 9f\n"
       "8:"  // odd_loads_1_0
+      "ldr b31, [x28, #0x0]\n"
       "ldr b30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr b29, [x26, #0x0]\n"
       "ldr b28, [x25, #0x0]\n"
       "ldr b27, [x24, #0x0]\n"
-      "ldr b24, [x23, #0x0]\n"
-      "ldr b23, [x22, #0x0]\n"
-      "ldr b21, [x21, #0x0]\n"
-      "ldr b26, [x20, #0x0]\n"
+      "ldr b26, [x23, #0x0]\n"
+      "ldr b25, [x22, #0x0]\n"
+      "ldr b24, [x21, #0x0]\n"
       "9:"  // Odd load end
+      "sshll v31.8h, v31.8b, #0x0\n"
       "sshll v30.8h, v30.8b, #0x0\n"
-      "subs x19, x19, #0x1\n"
+      "subs x20, x20, #0x1\n"
       "sshll v29.8h, v29.8b, #0x0\n"
       "sshll v28.8h, v28.8b, #0x0\n"
       "sshll v27.8h, v27.8b, #0x0\n"
-      "sshll v24.8h, v24.8b, #0x0\n"
-      "zip1 v20.8h, v30.8h, v24.8h\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "zip1 v25.8h, v29.8h, v23.8h\n"
-      "sshll v21.8h, v21.8b, #0x0\n"
-      "zip1 v19.8h, v28.8h, v21.8h\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
       "sshll v26.8h, v26.8b, #0x0\n"
-      "zip1 v18.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v25.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "sshll v25.8h, v25.8b, #0x0\n"
+      "sshll v24.8h, v24.8b, #0x0\n"
+      "zip1 v23.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v29.8h, v25.8h\n"
+      "zip1 v21.8h, v30.8h, v26.8h\n"
+      "zip1 v20.8h, v28.8h, v24.8h\n"
+      "zip1 v18.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "beq 10f\n"
-      "zip2 v17.8h, v22.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v19.8h, v20.8h, v19.8h\n"
-      "zip2 v16.8h, v25.8h, v18.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v17.8h, v19.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v19.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
       "beq 10f\n"
-      "zip2 v16.8h, v19.8h, v16.8h\n"
+      "zip2 v18.8h, v23.8h, v22.8h\n"
+      "zip2 v17.8h, v21.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v20.8h, v28.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.8h, v24.8h, v20.8h\n"
-      "zip2 v23.8h, v29.8h, v23.8h\n"
-      "zip2 v19.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v19.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v21.8h, v17.8h\n"
-      "str q18, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v18.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v20.8h, v24.8h, v20.8h\n"
-      "zip2 v16.8h, v23.8h, v19.8h\n"
-      "zip1 v17.8h, v20.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "zip2 v21.8h, v29.8h, v25.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "zip2 v19.8h, v28.8h, v24.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
+      "beq 10f\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 10f\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "10:"  // Odds skip
-      "saddw v0.4s, v0.4s, v1.4h\n"
-      "str q0, [%x[out_ptr], #0x0]\n"
-      "saddw2 v31.4s, v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x10]\n"
+      "saddw v1.4s, v1.4s, v2.4h\n"
+      "saddw2 v0.4s, v0.4s, v2.8h\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
+      "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u16_u16_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u16_u16_summing.hpp
index 24ece9a..de29d77 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u16_u16_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u16_u16_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,40 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.4s, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.4s, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -71,7 +72,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -79,226 +80,225 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v0.4s }, [%x[out_ptr]]\n"
-      "ldr q31, [%x[out_ptr], #0x10]\n"
+      "ld1 { v1.4s }, [%x[out_ptr]]\n"
+      "ldr q0, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x8\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0xe\n"
+      "cmp x20, #0xe\n"
       "ble 4f\n"
-      "uaddw v0.4s, v0.4s, v1.4h\n"
-      "uaddw2 v31.4s, v31.4s, v1.8h\n"
-      "mov x19, #0x0\n"
-      "movi v1.8h, #0x0\n"
+      "uaddw v1.4s, v1.4s, v2.4h\n"
+      "uaddw2 v0.4s, v0.4s, v2.8h\n"
+      "mov x20, #0x0\n"
+      "movi v2.8h, #0x0\n"
       "4:"  // no_accumulate_16
+      "ldr q31, [x28], #0x10\n"
       "ldr q30, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q28, [x25], #0x10\n"
       "cmp %x[width], #0x8\n"
+      "ldr q29, [x26], #0x10\n"
+      "ldr q28, [x25], #0x10\n"
+      "add x20, x20, #0x1\n"
       "ldr q27, [x24], #0x10\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "ldr q21, [x22], #0x10\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "ldr q24, [x21], #0x10\n"
-      "ldr q23, [x20], #0x10\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr q26, [x23], #0x10\n"
+      "zip1 v25.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v30.8h, v26.8h\n"
+      "ldr q24, [x22], #0x10\n"
+      "ldr q23, [x21], #0x10\n"
+      "zip1 v18.8h, v29.8h, v24.8h\n"
+      "zip1 v21.8h, v28.8h, v23.8h\n"
+      "zip1 v17.8h, v25.8h, v18.8h\n"
+      "zip1 v16.8h, v22.8h, v21.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "zip1 v20.8h, v17.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v20.8h\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
+      "zip2 v19.8h, v17.8h, v16.8h\n"
+      "zip2 v18.8h, v25.8h, v18.8h\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
+      "zip2 v17.8h, v22.8h, v21.8h\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "str q20, [%x[out_ptr], #0x0]\n"
+      "zip2 v21.8h, v29.8h, v24.8h\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "str q19, [%x[out_ptr], #0x10]\n"
+      "zip2 v19.8h, v28.8h, v23.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x20]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x70]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 10f\n"
       "tbz %x[width], #2, 7f\n"
+      "ldr d31, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "ldr d24, [x21], #0x8\n"
-      "ldr d23, [x20], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
+      "ldr d24, [x22], #0x8\n"
+      "ldr d23, [x21], #0x8\n"
       "tbz %x[width], #1, 6f\n"
+      "ld1 { v31.s }[2], [x28], #0x4\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v21.s }[2], [x22], #0x4\n"
-      "ld1 { v24.s }[2], [x21], #0x4\n"
-      "ld1 { v23.s }[2], [x20], #0x4\n"
+      "ld1 { v26.s }[2], [x23], #0x4\n"
+      "ld1 { v24.s }[2], [x22], #0x4\n"
+      "ld1 { v23.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[6], [x28]\n"
       "ld1 { v30.h }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v25.h }[6], [x23]\n"
-      "ld1 { v21.h }[6], [x22]\n"
-      "ld1 { v24.h }[6], [x21]\n"
-      "ld1 { v23.h }[6], [x20]\n"
+      "ld1 { v26.h }[6], [x23]\n"
+      "ld1 { v24.h }[6], [x22]\n"
+      "ld1 { v23.h }[6], [x21]\n"
       "b 9f\n"
       "6:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[4], [x28]\n"
       "ld1 { v30.h }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.h }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v25.h }[4], [x23]\n"
-      "ld1 { v21.h }[4], [x22]\n"
-      "ld1 { v24.h }[4], [x21]\n"
-      "ld1 { v23.h }[4], [x20]\n"
+      "ld1 { v26.h }[4], [x23]\n"
+      "ld1 { v24.h }[4], [x22]\n"
+      "ld1 { v23.h }[4], [x21]\n"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
+      "ldr s31, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
+      "mov x20, #0x2\n"
       "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s24, [x21], #0x4\n"
-      "ldr s23, [x20], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s24, [x22], #0x4\n"
+      "ldr s23, [x21], #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.h }[2], [x28]\n"
       "ld1 { v30.h }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v21.h }[2], [x22]\n"
-      "ld1 { v24.h }[2], [x21]\n"
-      "ld1 { v23.h }[2], [x20]\n"
+      "ld1 { v26.h }[2], [x23]\n"
+      "ld1 { v24.h }[2], [x22]\n"
+      "ld1 { v23.h }[2], [x21]\n"
       "b 9f\n"
       "8:"  // odd_loads_1_0
+      "ldr h31, [x28, #0x0]\n"
       "ldr h30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h21, [x22, #0x0]\n"
-      "ldr h24, [x21, #0x0]\n"
-      "ldr h23, [x20, #0x0]\n"
+      "ldr h26, [x23, #0x0]\n"
+      "ldr h24, [x22, #0x0]\n"
+      "ldr h23, [x21, #0x0]\n"
       "9:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v25.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v28.8h, v24.8h\n"
-      "zip1 v18.8h, v26.8h, v20.8h\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "zip1 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v16.8h, v22.8h, v19.8h\n"
-      "zip1 v17.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "zip1 v25.8h, v31.8h, v27.8h\n"
+      "zip1 v18.8h, v29.8h, v24.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v22.8h, v30.8h, v26.8h\n"
+      "zip1 v21.8h, v28.8h, v23.8h\n"
+      "zip1 v17.8h, v25.8h, v18.8h\n"
+      "zip1 v16.8h, v22.8h, v21.8h\n"
+      "zip1 v20.8h, v17.8h, v16.8h\n"
+      "str q20, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v20.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v26.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v19.8h, v17.8h, v16.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v19.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 10f\n"
+      "zip2 v18.8h, v25.8h, v18.8h\n"
+      "zip2 v17.8h, v22.8h, v21.8h\n"
+      "subs x20, x20, #0x1\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "beq 10f\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v25.8h, v30.8h, v25.8h\n"
-      "zip2 v20.8h, v28.8h, v24.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v23.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "zip2 v21.8h, v29.8h, v24.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "zip2 v19.8h, v28.8h, v23.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "beq 10f\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v25.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "10:"  // Odds skip
-      "uaddw v0.4s, v0.4s, v1.4h\n"
-      "str q0, [%x[out_ptr], #0x0]\n"
-      "uaddw2 v31.4s, v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x10]\n"
+      "uaddw v1.4s, v1.4s, v2.4h\n"
+      "uaddw2 v0.4s, v0.4s, v2.8h\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
+      "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16.hpp
index 0db2f7f..43a3a46 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,209 +77,208 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d31, [x27], #0x8\n"
-      "ushll v31.8h, v31.8b, #0x0\n"
-      "ldr d30, [x26], #0x8\n"
-      "subs %x[width], %x[width], #0x8\n"
-      "ushll v30.8h, v30.8b, #0x0\n"
-      "ldr d29, [x25], #0x8\n"
-      "cmp %x[width], #0x8\n"
-      "ushll v29.8h, v29.8b, #0x0\n"
-      "ldr d28, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ushll v28.8h, v28.8b, #0x0\n"
-      "ldr d23, [x22], #0x8\n"
+      "ldr d25, [x28], #0x8\n"
+      "ldr d30, [x27], #0x8\n"
       "ushll v25.8h, v25.8b, #0x0\n"
-      "ldr d27, [x21], #0x8\n"
-      "zip1 v20.8h, v31.8h, v25.8h\n"
-      "ldr d26, [x20], #0x8\n"
-      "zip2 v25.8h, v31.8h, v25.8h\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
+      "ushll v30.8h, v30.8b, #0x0\n"
+      "ldr d29, [x26], #0x8\n"
+      "ldr d28, [x25], #0x8\n"
+      "ushll v29.8h, v29.8b, #0x0\n"
+      "ushll v28.8h, v28.8b, #0x0\n"
+      "ldr d21, [x24], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "ushll v21.8h, v21.8b, #0x0\n"
       "ushll v27.8h, v27.8b, #0x0\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v19.8h, v29.8h, v27.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v21.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v19.8h, v29.8h, v27.8h\n"
-      "zip1 v20.8h, v25.8h, v19.8h\n"
-      "zip2 v19.8h, v25.8h, v19.8h\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d26, [x21], #0x8\n"
+      "ushll v20.8h, v20.8b, #0x0\n"
       "ushll v26.8h, v26.8b, #0x0\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v22.8h, v29.8h, v20.8h\n"
+      "subs %x[width], %x[width], #0x8\n"
+      "cmp %x[width], #0x8\n"
+      "zip1 v19.8h, v30.8h, v27.8h\n"
       "zip1 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "prfm pldl1keep, [x26, #0x70]\n"
+      "prfm pldl1keep, [x25, #0x70]\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v26.8h\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "zip1 v24.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v19.8h, v18.8h\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v23.8h, v23.8h, v22.8h\n"
+      "zip2 v19.8h, v19.8h, v18.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v22.8h, v17.8h\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v21.8h, v17.8h\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v19.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "zip2 v16.8h, v19.8h, v17.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip2 v16.8h, v21.8h, v20.8h\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
-      "ldr s31, [x27], #0x4\n"
-      "ldr s30, [x26], #0x4\n"
-      "ldr s29, [x25], #0x4\n"
-      "ldr s28, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s23, [x22], #0x4\n"
-      "ldr s27, [x21], #0x4\n"
-      "ldr s26, [x20], #0x4\n"
+      "ldr s25, [x28], #0x4\n"
+      "ldr s30, [x27], #0x4\n"
+      "ldr s29, [x26], #0x4\n"
+      "ldr s28, [x25], #0x4\n"
+      "ldr s21, [x24], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s26, [x21], #0x4\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v31.h }[2], [x27], #0x2\n"
-      "mov x19, #0x6\n"
-      "ld1 { v30.h }[2], [x26], #0x2\n"
-      "ld1 { v29.h }[2], [x25], #0x2\n"
-      "ld1 { v28.h }[2], [x24], #0x2\n"
-      "ld1 { v25.h }[2], [x23], #0x2\n"
-      "ld1 { v23.h }[2], [x22], #0x2\n"
-      "ld1 { v27.h }[2], [x21], #0x2\n"
-      "ld1 { v26.h }[2], [x20], #0x2\n"
+      "ld1 { v25.h }[2], [x28], #0x2\n"
+      "ld1 { v30.h }[2], [x27], #0x2\n"
+      "mov x20, #0x6\n"
+      "ld1 { v29.h }[2], [x26], #0x2\n"
+      "ld1 { v28.h }[2], [x25], #0x2\n"
+      "ld1 { v21.h }[2], [x24], #0x2\n"
+      "ld1 { v27.h }[2], [x23], #0x2\n"
+      "ld1 { v20.h }[2], [x22], #0x2\n"
+      "ld1 { v26.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[6], [x27]\n"
-      "mov x19, #0x7\n"
-      "ld1 { v30.b }[6], [x26]\n"
-      "ld1 { v29.b }[6], [x25]\n"
-      "ld1 { v28.b }[6], [x24]\n"
-      "ld1 { v25.b }[6], [x23]\n"
-      "ld1 { v23.b }[6], [x22]\n"
-      "ld1 { v27.b }[6], [x21]\n"
-      "ld1 { v26.b }[6], [x20]\n"
+      "ld1 { v25.b }[6], [x28]\n"
+      "ld1 { v30.b }[6], [x27]\n"
+      "mov x20, #0x7\n"
+      "ld1 { v29.b }[6], [x26]\n"
+      "ld1 { v28.b }[6], [x25]\n"
+      "ld1 { v21.b }[6], [x24]\n"
+      "ld1 { v27.b }[6], [x23]\n"
+      "ld1 { v20.b }[6], [x22]\n"
+      "ld1 { v26.b }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[4], [x27]\n"
-      "ld1 { v30.b }[4], [x26]\n"
-      "mov x19, #0x5\n"
-      "ld1 { v29.b }[4], [x25]\n"
-      "ld1 { v28.b }[4], [x24]\n"
-      "ld1 { v25.b }[4], [x23]\n"
-      "ld1 { v23.b }[4], [x22]\n"
-      "ld1 { v27.b }[4], [x21]\n"
-      "ld1 { v26.b }[4], [x20]\n"
+      "ld1 { v25.b }[4], [x28]\n"
+      "ld1 { v30.b }[4], [x27]\n"
+      "mov x20, #0x5\n"
+      "ld1 { v29.b }[4], [x26]\n"
+      "ld1 { v28.b }[4], [x25]\n"
+      "ld1 { v21.b }[4], [x24]\n"
+      "ld1 { v27.b }[4], [x23]\n"
+      "ld1 { v20.b }[4], [x22]\n"
+      "ld1 { v26.b }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr h31, [x27], #0x2\n"
-      "ldr h30, [x26], #0x2\n"
-      "mov x19, #0x2\n"
-      "ldr h29, [x25], #0x2\n"
-      "ldr h28, [x24], #0x2\n"
-      "ldr h25, [x23], #0x2\n"
-      "ldr h23, [x22], #0x2\n"
-      "ldr h27, [x21], #0x2\n"
-      "ldr h26, [x20], #0x2\n"
+      "ldr h25, [x28], #0x2\n"
+      "ldr h30, [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ldr h29, [x26], #0x2\n"
+      "ldr h28, [x25], #0x2\n"
+      "ldr h21, [x24], #0x2\n"
+      "ldr h27, [x23], #0x2\n"
+      "ldr h20, [x22], #0x2\n"
+      "ldr h26, [x21], #0x2\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v31.b }[2], [x27]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v30.b }[2], [x26]\n"
-      "ld1 { v29.b }[2], [x25]\n"
-      "ld1 { v28.b }[2], [x24]\n"
-      "ld1 { v25.b }[2], [x23]\n"
-      "ld1 { v23.b }[2], [x22]\n"
-      "ld1 { v27.b }[2], [x21]\n"
-      "ld1 { v26.b }[2], [x20]\n"
+      "ld1 { v25.b }[2], [x28]\n"
+      "ld1 { v30.b }[2], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v29.b }[2], [x26]\n"
+      "ld1 { v28.b }[2], [x25]\n"
+      "ld1 { v21.b }[2], [x24]\n"
+      "ld1 { v27.b }[2], [x23]\n"
+      "ld1 { v20.b }[2], [x22]\n"
+      "ld1 { v26.b }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr b31, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b30, [x26, #0x0]\n"
-      "ldr b29, [x25, #0x0]\n"
-      "ldr b28, [x24, #0x0]\n"
-      "ldr b25, [x23, #0x0]\n"
-      "ldr b23, [x22, #0x0]\n"
-      "ldr b27, [x21, #0x0]\n"
-      "ldr b26, [x20, #0x0]\n"
+      "ldr b25, [x28, #0x0]\n"
+      "ldr b30, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b29, [x26, #0x0]\n"
+      "ldr b28, [x25, #0x0]\n"
+      "ldr b21, [x24, #0x0]\n"
+      "ldr b27, [x23, #0x0]\n"
+      "ldr b20, [x22, #0x0]\n"
+      "ldr b26, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "ushll v31.8h, v31.8b, #0x0\n"
-      "subs x19, x19, #0x1\n"
+      "ushll v25.8h, v25.8b, #0x0\n"
       "ushll v30.8h, v30.8b, #0x0\n"
+      "subs x20, x20, #0x1\n"
       "ushll v29.8h, v29.8b, #0x0\n"
       "ushll v28.8h, v28.8b, #0x0\n"
-      "ushll v25.8h, v25.8b, #0x0\n"
-      "zip1 v20.8h, v31.8h, v25.8h\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
+      "ushll v21.8h, v21.8b, #0x0\n"
       "ushll v27.8h, v27.8b, #0x0\n"
-      "zip1 v19.8h, v29.8h, v27.8h\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
+      "ushll v20.8h, v20.8b, #0x0\n"
       "ushll v26.8h, v26.8b, #0x0\n"
+      "zip1 v23.8h, v25.8h, v21.8h\n"
+      "zip1 v22.8h, v29.8h, v20.8h\n"
+      "zip1 v19.8h, v30.8h, v27.8h\n"
       "zip1 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v24.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "zip1 v24.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v19.8h, v18.8h\n"
+      "zip1 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v22.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v21.8h, v20.8h, v19.8h\n"
-      "zip2 v17.8h, v24.8h, v18.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v24.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v21.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v23.8h, v23.8h, v22.8h\n"
+      "zip2 v19.8h, v19.8h, v18.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.8h, v23.8h, v19.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v25.8h, v31.8h, v25.8h\n"
-      "zip2 v19.8h, v29.8h, v27.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.8h, v25.8h, v19.8h\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v18.8h, v28.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.8h, v23.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
+      "zip2 v25.8h, v25.8h, v21.8h\n"
+      "zip2 v21.8h, v29.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v27.8h\n"
+      "zip2 v16.8h, v28.8h, v26.8h\n"
+      "zip1 v22.8h, v25.8h, v21.8h\n"
+      "zip1 v18.8h, v20.8h, v16.8h\n"
+      "zip1 v19.8h, v22.8h, v18.8h\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v19.8h, v25.8h, v19.8h\n"
-      "zip2 v17.8h, v23.8h, v18.8h\n"
-      "zip1 v16.8h, v19.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v18.8h, v22.8h, v18.8h\n"
+      "str q18, [%x[out_ptr], #0x0]\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 8f\n"
+      "zip2 v21.8h, v25.8h, v21.8h\n"
+      "zip2 v20.8h, v20.8h, v16.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "str q17, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16_summing.hpp
index 7c7d774..3ab2436 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_u8_u16_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,39 +31,40 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.4s, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.4s, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -71,7 +72,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -79,242 +80,241 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v0.4s }, [%x[out_ptr]]\n"
-      "ldr q31, [%x[out_ptr], #0x10]\n"
+      "ld1 { v1.4s }, [%x[out_ptr]]\n"
+      "ldr q0, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x8\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0xe\n"
+      "cmp x20, #0xe\n"
       "ble 4f\n"
-      "uaddw v0.4s, v0.4s, v1.4h\n"
-      "uaddw2 v31.4s, v31.4s, v1.8h\n"
-      "mov x19, #0x0\n"
-      "movi v1.8h, #0x0\n"
+      "uaddw v1.4s, v1.4s, v2.4h\n"
+      "uaddw2 v0.4s, v0.4s, v2.8h\n"
+      "mov x20, #0x0\n"
+      "movi v2.8h, #0x0\n"
       "4:"  // no_accumulate_16
+      "ldr d31, [x28], #0x8\n"
       "ldr d30, [x27], #0x8\n"
+      "ushll v31.8h, v31.8b, #0x0\n"
       "ushll v30.8h, v30.8b, #0x0\n"
       "ldr d29, [x26], #0x8\n"
-      "add x19, x19, #0x1\n"
-      "ushll v29.8h, v29.8b, #0x0\n"
       "ldr d28, [x25], #0x8\n"
-      "subs %x[width], %x[width], #0x8\n"
+      "ushll v29.8h, v29.8b, #0x0\n"
       "ushll v28.8h, v28.8b, #0x0\n"
       "ldr d27, [x24], #0x8\n"
-      "cmp %x[width], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
       "ushll v27.8h, v27.8b, #0x0\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d23, [x22], #0x8\n"
-      "ushll v24.8h, v24.8b, #0x0\n"
-      "ldr d21, [x21], #0x8\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "ldr d26, [x20], #0x8\n"
-      "zip1 v20.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "zip1 v25.8h, v29.8h, v23.8h\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "zip2 v23.8h, v29.8h, v23.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ushll v21.8h, v21.8b, #0x0\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v19.8h, v28.8h, v21.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v19.8h, v20.8h, v19.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v20.8h, v28.8h, v21.8h\n"
-      "zip1 v21.8h, v24.8h, v20.8h\n"
-      "zip2 v20.8h, v24.8h, v20.8h\n"
       "ushll v26.8h, v26.8b, #0x0\n"
-      "zip1 v18.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v25.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "ldr d25, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "ushll v25.8h, v25.8b, #0x0\n"
+      "ushll v24.8h, v24.8b, #0x0\n"
+      "zip1 v23.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v29.8h, v25.8h\n"
+      "subs %x[width], %x[width], #0x8\n"
+      "cmp %x[width], #0x8\n"
+      "zip1 v21.8h, v30.8h, v26.8h\n"
+      "zip1 v20.8h, v28.8h, v24.8h\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip1 v18.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "prfm pldl1keep, [x26, #0x70]\n"
+      "prfm pldl1keep, [x25, #0x70]\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "zip2 v19.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v23.8h, v22.8h\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v17.8h, v21.8h, v20.8h\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v22.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.8h, v25.8h, v18.8h\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "zip1 v17.8h, v19.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v19.8h, v16.8h\n"
+      "add x20, x20, #0x1\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "str q19, [%x[out_ptr], #0x10]\n"
+      "zip2 v21.8h, v29.8h, v25.8h\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v19.8h, v28.8h, v24.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "zip2 v19.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v19.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v18.8h, v21.8h, v17.8h\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v23.8h, v19.8h\n"
-      "zip1 v17.8h, v20.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v18.8h\n"
-      "zip2 v16.8h, v20.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "str q16, [%x[out_ptr], #0x50]\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "str q16, [%x[out_ptr], #0x60]\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x70]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 10f\n"
       "tbz %x[width], #2, 7f\n"
+      "ldr s31, [x28], #0x4\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s23, [x22], #0x4\n"
-      "ldr s21, [x21], #0x4\n"
-      "ldr s26, [x20], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s25, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
+      "ld1 { v31.h }[2], [x28], #0x2\n"
       "ld1 { v30.h }[2], [x27], #0x2\n"
-      "mov x19, #0x6\n"
+      "mov x20, #0x6\n"
       "ld1 { v29.h }[2], [x26], #0x2\n"
       "ld1 { v28.h }[2], [x25], #0x2\n"
       "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v24.h }[2], [x23], #0x2\n"
-      "ld1 { v23.h }[2], [x22], #0x2\n"
-      "ld1 { v21.h }[2], [x21], #0x2\n"
-      "ld1 { v26.h }[2], [x20], #0x2\n"
+      "ld1 { v26.h }[2], [x23], #0x2\n"
+      "ld1 { v25.h }[2], [x22], #0x2\n"
+      "ld1 { v24.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[6], [x28]\n"
       "ld1 { v30.b }[6], [x27]\n"
-      "mov x19, #0x7\n"
+      "mov x20, #0x7\n"
       "ld1 { v29.b }[6], [x26]\n"
       "ld1 { v28.b }[6], [x25]\n"
       "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v24.b }[6], [x23]\n"
-      "ld1 { v23.b }[6], [x22]\n"
-      "ld1 { v21.b }[6], [x21]\n"
-      "ld1 { v26.b }[6], [x20]\n"
+      "ld1 { v26.b }[6], [x23]\n"
+      "ld1 { v25.b }[6], [x22]\n"
+      "ld1 { v24.b }[6], [x21]\n"
       "b 9f\n"
       "6:"  // odd_loads_1_4
-      "mov x19, #0x4\n"
+      "mov x20, #0x4\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[4], [x28]\n"
       "ld1 { v30.b }[4], [x27]\n"
+      "mov x20, #0x5\n"
       "ld1 { v29.b }[4], [x26]\n"
-      "mov x19, #0x5\n"
       "ld1 { v28.b }[4], [x25]\n"
       "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v24.b }[4], [x23]\n"
-      "ld1 { v23.b }[4], [x22]\n"
-      "ld1 { v21.b }[4], [x21]\n"
-      "ld1 { v26.b }[4], [x20]\n"
+      "ld1 { v26.b }[4], [x23]\n"
+      "ld1 { v25.b }[4], [x22]\n"
+      "ld1 { v24.b }[4], [x21]\n"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
+      "ldr h31, [x28], #0x2\n"
       "ldr h30, [x27], #0x2\n"
+      "mov x20, #0x2\n"
       "ldr h29, [x26], #0x2\n"
-      "mov x19, #0x2\n"
       "ldr h28, [x25], #0x2\n"
       "ldr h27, [x24], #0x2\n"
-      "ldr h24, [x23], #0x2\n"
-      "ldr h23, [x22], #0x2\n"
-      "ldr h21, [x21], #0x2\n"
-      "ldr h26, [x20], #0x2\n"
+      "ldr h26, [x23], #0x2\n"
+      "ldr h25, [x22], #0x2\n"
+      "ldr h24, [x21], #0x2\n"
       "tbz %x[width], #0, 9f\n"
+      "ld1 { v31.b }[2], [x28]\n"
       "ld1 { v30.b }[2], [x27]\n"
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "ld1 { v29.b }[2], [x26]\n"
       "ld1 { v28.b }[2], [x25]\n"
       "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v24.b }[2], [x23]\n"
-      "ld1 { v23.b }[2], [x22]\n"
-      "ld1 { v21.b }[2], [x21]\n"
-      "ld1 { v26.b }[2], [x20]\n"
+      "ld1 { v26.b }[2], [x23]\n"
+      "ld1 { v25.b }[2], [x22]\n"
+      "ld1 { v24.b }[2], [x21]\n"
       "b 9f\n"
       "8:"  // odd_loads_1_0
+      "ldr b31, [x28, #0x0]\n"
       "ldr b30, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "ldr b29, [x26, #0x0]\n"
       "ldr b28, [x25, #0x0]\n"
       "ldr b27, [x24, #0x0]\n"
-      "ldr b24, [x23, #0x0]\n"
-      "ldr b23, [x22, #0x0]\n"
-      "ldr b21, [x21, #0x0]\n"
-      "ldr b26, [x20, #0x0]\n"
+      "ldr b26, [x23, #0x0]\n"
+      "ldr b25, [x22, #0x0]\n"
+      "ldr b24, [x21, #0x0]\n"
       "9:"  // Odd load end
+      "ushll v31.8h, v31.8b, #0x0\n"
       "ushll v30.8h, v30.8b, #0x0\n"
-      "subs x19, x19, #0x1\n"
+      "subs x20, x20, #0x1\n"
       "ushll v29.8h, v29.8b, #0x0\n"
       "ushll v28.8h, v28.8b, #0x0\n"
       "ushll v27.8h, v27.8b, #0x0\n"
-      "ushll v24.8h, v24.8b, #0x0\n"
-      "zip1 v20.8h, v30.8h, v24.8h\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "zip1 v25.8h, v29.8h, v23.8h\n"
-      "ushll v21.8h, v21.8b, #0x0\n"
-      "zip1 v19.8h, v28.8h, v21.8h\n"
-      "zip1 v22.8h, v20.8h, v19.8h\n"
       "ushll v26.8h, v26.8b, #0x0\n"
-      "zip1 v18.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v25.8h, v18.8h\n"
-      "zip1 v16.8h, v22.8h, v17.8h\n"
+      "ushll v25.8h, v25.8b, #0x0\n"
+      "ushll v24.8h, v24.8b, #0x0\n"
+      "zip1 v23.8h, v31.8h, v27.8h\n"
+      "zip1 v22.8h, v29.8h, v25.8h\n"
+      "zip1 v21.8h, v30.8h, v26.8h\n"
+      "zip1 v20.8h, v28.8h, v24.8h\n"
+      "zip1 v18.8h, v23.8h, v22.8h\n"
+      "zip1 v17.8h, v21.8h, v20.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "beq 10f\n"
-      "zip2 v17.8h, v22.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v17.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v19.8h, v20.8h, v19.8h\n"
-      "zip2 v16.8h, v25.8h, v18.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v17.8h, v19.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v19.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v19.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
       "beq 10f\n"
-      "zip2 v16.8h, v19.8h, v16.8h\n"
+      "zip2 v18.8h, v23.8h, v22.8h\n"
+      "zip2 v17.8h, v21.8h, v20.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v20.8h, v28.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.8h, v24.8h, v20.8h\n"
-      "zip2 v23.8h, v29.8h, v23.8h\n"
-      "zip2 v19.8h, v27.8h, v26.8h\n"
-      "zip1 v17.8h, v23.8h, v19.8h\n"
-      "zip1 v16.8h, v21.8h, v17.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v21.8h, v17.8h\n"
-      "str q18, [%x[out_ptr], #0x0]\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v18.8h\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v20.8h, v24.8h, v20.8h\n"
-      "zip2 v16.8h, v23.8h, v19.8h\n"
-      "zip1 v17.8h, v20.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v22.8h, v31.8h, v27.8h\n"
+      "zip2 v21.8h, v29.8h, v25.8h\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.8h, v30.8h, v26.8h\n"
+      "zip2 v19.8h, v28.8h, v24.8h\n"
+      "zip1 v18.8h, v22.8h, v21.8h\n"
+      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "add v1.8h, v1.8h, v17.8h\n"
+      "beq 10f\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "subs x20, x20, #0x1\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "beq 10f\n"
+      "zip2 v18.8h, v22.8h, v21.8h\n"
+      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "add v2.8h, v2.8h, v16.8h\n"
+      "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "10:"  // Odds skip
-      "uaddw v0.4s, v0.4s, v1.4h\n"
-      "str q0, [%x[out_ptr], #0x0]\n"
-      "uaddw2 v31.4s, v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x10]\n"
+      "uaddw v1.4s, v1.4s, v2.4h\n"
+      "uaddw2 v0.4s, v0.4s, v2.8h\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
+      "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_bf16_bf16.hpp
index 1e5d395..d4d1504 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,49 +77,48 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q28, [x27], #0x10\n"
+      "ldr q28, [x28], #0x10\n"
+      "ldr q27, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "ldr q25, [x25], #0x10\n"
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "ldr q21, [x24], #0x10\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "ldr q27, [x23], #0x10\n"
-      "ldr q26, [x22], #0x10\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "ldr q24, [x20], #0x10\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "ldr q22, [x26], #0x10\n"
+      "ldr q21, [x25], #0x10\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "ldr q24, [x24], #0x10\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "ldr q19, [x22], #0x10\n"
+      "ldr q18, [x21], #0x10\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x40]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v19.4s, v18.4s\n"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -126,120 +126,120 @@
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
-      "ldr d28, [x27], #0x8\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "ldr d26, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d24, [x20], #0x8\n"
+      "ldr d28, [x28], #0x8\n"
+      "ldr d27, [x27], #0x8\n"
+      "ldr d22, [x26], #0x8\n"
+      "ldr d21, [x25], #0x8\n"
+      "ldr d24, [x24], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
+      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x21], #0x8\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v28.s }[2], [x27], #0x4\n"
-      "mov x19, #0x3\n"
-      "ld1 { v29.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v27.s }[2], [x23], #0x4\n"
-      "ld1 { v26.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v24.s }[2], [x20], #0x4\n"
+      "ld1 { v28.s }[2], [x28], #0x4\n"
+      "ld1 { v27.s }[2], [x27], #0x4\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.s }[2], [x26], #0x4\n"
+      "ld1 { v21.s }[2], [x25], #0x4\n"
+      "ld1 { v24.s }[2], [x24], #0x4\n"
+      "ld1 { v23.s }[2], [x23], #0x4\n"
+      "ld1 { v19.s }[2], [x22], #0x4\n"
+      "ld1 { v18.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v28.h }[6], [x27]\n"
-      "mov x19, #0x4\n"
-      "ld1 { v29.h }[6], [x26]\n"
-      "ld1 { v25.h }[6], [x25]\n"
-      "ld1 { v21.h }[6], [x24]\n"
-      "ld1 { v27.h }[6], [x23]\n"
-      "ld1 { v26.h }[6], [x22]\n"
-      "ld1 { v19.h }[6], [x21]\n"
-      "ld1 { v24.h }[6], [x20]\n"
+      "ld1 { v28.h }[6], [x28]\n"
+      "ld1 { v27.h }[6], [x27]\n"
+      "mov x20, #0x4\n"
+      "ld1 { v22.h }[6], [x26]\n"
+      "ld1 { v21.h }[6], [x25]\n"
+      "ld1 { v24.h }[6], [x24]\n"
+      "ld1 { v23.h }[6], [x23]\n"
+      "ld1 { v19.h }[6], [x22]\n"
+      "ld1 { v18.h }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v28.h }[4], [x27]\n"
-      "ld1 { v29.h }[4], [x26]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.h }[4], [x25]\n"
-      "ld1 { v21.h }[4], [x24]\n"
-      "ld1 { v27.h }[4], [x23]\n"
-      "ld1 { v26.h }[4], [x22]\n"
-      "ld1 { v19.h }[4], [x21]\n"
-      "ld1 { v24.h }[4], [x20]\n"
+      "ld1 { v28.h }[4], [x28]\n"
+      "ld1 { v27.h }[4], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.h }[4], [x26]\n"
+      "ld1 { v21.h }[4], [x25]\n"
+      "ld1 { v24.h }[4], [x24]\n"
+      "ld1 { v23.h }[4], [x23]\n"
+      "ld1 { v19.h }[4], [x22]\n"
+      "ld1 { v18.h }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr s28, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "mov x19, #0x1\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s27, [x23], #0x4\n"
-      "ldr s26, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s24, [x20], #0x4\n"
+      "ldr s28, [x28], #0x4\n"
+      "ldr s27, [x27], #0x4\n"
+      "mov x20, #0x1\n"
+      "ldr s22, [x26], #0x4\n"
+      "ldr s21, [x25], #0x4\n"
+      "ldr s24, [x24], #0x4\n"
+      "ldr s23, [x23], #0x4\n"
+      "ldr s19, [x22], #0x4\n"
+      "ldr s18, [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v28.h }[2], [x27]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v29.h }[2], [x26]\n"
-      "ld1 { v25.h }[2], [x25]\n"
-      "ld1 { v21.h }[2], [x24]\n"
-      "ld1 { v27.h }[2], [x23]\n"
-      "ld1 { v26.h }[2], [x22]\n"
-      "ld1 { v19.h }[2], [x21]\n"
-      "ld1 { v24.h }[2], [x20]\n"
+      "ld1 { v28.h }[2], [x28]\n"
+      "ld1 { v27.h }[2], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v22.h }[2], [x26]\n"
+      "ld1 { v21.h }[2], [x25]\n"
+      "ld1 { v24.h }[2], [x24]\n"
+      "ld1 { v23.h }[2], [x23]\n"
+      "ld1 { v19.h }[2], [x22]\n"
+      "ld1 { v18.h }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr h28, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr h29, [x26, #0x0]\n"
-      "ldr h25, [x25, #0x0]\n"
-      "ldr h21, [x24, #0x0]\n"
-      "ldr h27, [x23, #0x0]\n"
-      "ldr h26, [x22, #0x0]\n"
-      "ldr h19, [x21, #0x0]\n"
-      "ldr h24, [x20, #0x0]\n"
+      "ldr h28, [x28, #0x0]\n"
+      "ldr h27, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr h22, [x26, #0x0]\n"
+      "ldr h21, [x25, #0x0]\n"
+      "ldr h24, [x24, #0x0]\n"
+      "ldr h23, [x23, #0x0]\n"
+      "ldr h19, [x22, #0x0]\n"
+      "ldr h18, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
-      "subs x19, x19, #0x1\n"
-      "str q20, [%x[out_ptr], #0x10]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
+      "zip2 v16.4s, v19.4s, v18.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_fp32_fp32.hpp
index 064207c..358b83a 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block2_fp32_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #2\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #2\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #2\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #2\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #2\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #2\n"
       "add x22, x22, %x[row_offset], LSL #2\n"
       "add x21, x21, %x[row_offset], LSL #2\n"
-      "add x20, x20, %x[row_offset], LSL #2\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x4\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,104 +77,103 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q27, [x27], #0x10\n"
+      "ldr q26, [x28], #0x10\n"
+      "ldr q21, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x4\n"
-      "ldr q24, [x26], #0x10\n"
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "ldr q25, [x25], #0x10\n"
       "cmp %x[width], #0x4\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "ldr q21, [x24], #0x10\n"
-      "ldr q23, [x23], #0x10\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "ldr q18, [x22], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "ldr q20, [x21], #0x10\n"
-      "ldr q16, [x20], #0x10\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "ldr q25, [x26], #0x10\n"
+      "ldr q24, [x25], #0x10\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "ldr q23, [x24], #0x10\n"
+      "ldr q22, [x23], #0x10\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "ldr q20, [x22], #0x10\n"
+      "ldr q19, [x21], #0x10\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
       "prfm pldl1keep, [x24, #0x70]\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "str q24, [%x[out_ptr], #0x40]\n"
-      "str q21, [%x[out_ptr], #0x50]\n"
-      "str q18, [%x[out_ptr], #0x60]\n"
+      "str q21, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr d27, [x27], #0x8\n"
-      "ldr d24, [x26], #0x8\n"
-      "mov x19, #0x1\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d18, [x22], #0x8\n"
-      "ldr d20, [x21], #0x8\n"
-      "ldr d16, [x20], #0x8\n"
+      "ldr d26, [x28], #0x8\n"
+      "ldr d21, [x27], #0x8\n"
+      "mov x20, #0x1\n"
+      "ldr d25, [x26], #0x8\n"
+      "ldr d24, [x25], #0x8\n"
+      "ldr d23, [x24], #0x8\n"
+      "ldr d22, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d19, [x21], #0x8\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v27.s }[2], [x27]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v24.s }[2], [x26]\n"
-      "ld1 { v25.s }[2], [x25]\n"
-      "ld1 { v21.s }[2], [x24]\n"
-      "ld1 { v23.s }[2], [x23]\n"
-      "ld1 { v18.s }[2], [x22]\n"
-      "ld1 { v20.s }[2], [x21]\n"
-      "ld1 { v16.s }[2], [x20]\n"
+      "ld1 { v26.s }[2], [x28]\n"
+      "ld1 { v21.s }[2], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.s }[2], [x26]\n"
+      "ld1 { v24.s }[2], [x25]\n"
+      "ld1 { v23.s }[2], [x24]\n"
+      "ld1 { v22.s }[2], [x23]\n"
+      "ld1 { v20.s }[2], [x22]\n"
+      "ld1 { v19.s }[2], [x21]\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr s27, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr s24, [x26, #0x0]\n"
-      "ldr s25, [x25, #0x0]\n"
-      "ldr s21, [x24, #0x0]\n"
-      "ldr s23, [x23, #0x0]\n"
-      "ldr s18, [x22, #0x0]\n"
-      "ldr s20, [x21, #0x0]\n"
-      "ldr s16, [x20, #0x0]\n"
+      "ldr s26, [x28, #0x0]\n"
+      "ldr s21, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr s25, [x26, #0x0]\n"
+      "ldr s24, [x25, #0x0]\n"
+      "ldr s23, [x24, #0x0]\n"
+      "ldr s22, [x23, #0x0]\n"
+      "ldr s20, [x22, #0x0]\n"
+      "ldr s19, [x21, #0x0]\n"
       "5:"  // Odd load end
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 6f\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "str q21, [%x[out_ptr], #0x0]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "6:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_bf16_bf16.hpp
index 1f86722..d606d5a 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #1\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #1\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #1\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #1\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #1\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #1\n"
       "add x22, x22, %x[row_offset], LSL #1\n"
       "add x21, x21, %x[row_offset], LSL #1\n"
-      "add x20, x20, %x[row_offset], LSL #1\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x8\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,146 +77,145 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q27, [x27], #0x10\n"
+      "ldr q26, [x28], #0x10\n"
+      "ldr q21, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "ldr q24, [x26], #0x10\n"
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "ldr q25, [x25], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "ldr q21, [x24], #0x10\n"
-      "ldr q23, [x23], #0x10\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "ldr q18, [x22], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "ldr q20, [x21], #0x10\n"
-      "ldr q16, [x20], #0x10\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "ldr q25, [x26], #0x10\n"
+      "ldr q24, [x25], #0x10\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "ldr q23, [x24], #0x10\n"
+      "ldr q22, [x23], #0x10\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "ldr q20, [x22], #0x10\n"
+      "ldr q19, [x21], #0x10\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
       "prfm pldl1keep, [x24, #0x70]\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "str q24, [%x[out_ptr], #0x40]\n"
-      "str q21, [%x[out_ptr], #0x50]\n"
-      "str q18, [%x[out_ptr], #0x60]\n"
+      "str q21, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
-      "ldr d27, [x27], #0x8\n"
-      "ldr d24, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d18, [x22], #0x8\n"
-      "ldr d20, [x21], #0x8\n"
-      "ldr d16, [x20], #0x8\n"
+      "ldr d26, [x28], #0x8\n"
+      "ldr d21, [x27], #0x8\n"
+      "ldr d25, [x26], #0x8\n"
+      "ldr d24, [x25], #0x8\n"
+      "ldr d23, [x24], #0x8\n"
+      "ldr d22, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d19, [x21], #0x8\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v27.s }[2], [x27], #0x4\n"
-      "mov x19, #0x2\n"
-      "ld1 { v24.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v23.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v20.s }[2], [x21], #0x4\n"
-      "ld1 { v16.s }[2], [x20], #0x4\n"
+      "ld1 { v26.s }[2], [x28], #0x4\n"
+      "ld1 { v21.s }[2], [x27], #0x4\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.s }[2], [x26], #0x4\n"
+      "ld1 { v24.s }[2], [x25], #0x4\n"
+      "ld1 { v23.s }[2], [x24], #0x4\n"
+      "ld1 { v22.s }[2], [x23], #0x4\n"
+      "ld1 { v20.s }[2], [x22], #0x4\n"
+      "ld1 { v19.s }[2], [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v27.h }[6], [x27]\n"
-      "ld1 { v24.h }[6], [x26]\n"
-      "ld1 { v25.h }[6], [x25]\n"
-      "ld1 { v21.h }[6], [x24]\n"
-      "ld1 { v23.h }[6], [x23]\n"
-      "ld1 { v18.h }[6], [x22]\n"
-      "ld1 { v20.h }[6], [x21]\n"
-      "ld1 { v16.h }[6], [x20]\n"
+      "ld1 { v26.h }[6], [x28]\n"
+      "ld1 { v21.h }[6], [x27]\n"
+      "ld1 { v25.h }[6], [x26]\n"
+      "ld1 { v24.h }[6], [x25]\n"
+      "ld1 { v23.h }[6], [x24]\n"
+      "ld1 { v22.h }[6], [x23]\n"
+      "ld1 { v20.h }[6], [x22]\n"
+      "ld1 { v19.h }[6], [x21]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v27.h }[4], [x27]\n"
-      "ld1 { v24.h }[4], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.h }[4], [x25]\n"
-      "ld1 { v21.h }[4], [x24]\n"
-      "ld1 { v23.h }[4], [x23]\n"
-      "ld1 { v18.h }[4], [x22]\n"
-      "ld1 { v20.h }[4], [x21]\n"
-      "ld1 { v16.h }[4], [x20]\n"
+      "ld1 { v26.h }[4], [x28]\n"
+      "ld1 { v21.h }[4], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[4], [x26]\n"
+      "ld1 { v24.h }[4], [x25]\n"
+      "ld1 { v23.h }[4], [x24]\n"
+      "ld1 { v22.h }[4], [x23]\n"
+      "ld1 { v20.h }[4], [x22]\n"
+      "ld1 { v19.h }[4], [x21]\n"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr s27, [x27], #0x4\n"
-      "ldr s24, [x26], #0x4\n"
-      "mov x19, #0x1\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s18, [x22], #0x4\n"
-      "ldr s20, [x21], #0x4\n"
-      "ldr s16, [x20], #0x4\n"
+      "ldr s26, [x28], #0x4\n"
+      "ldr s21, [x27], #0x4\n"
+      "mov x20, #0x1\n"
+      "ldr s25, [x26], #0x4\n"
+      "ldr s24, [x25], #0x4\n"
+      "ldr s23, [x24], #0x4\n"
+      "ldr s22, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s19, [x21], #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v27.h }[2], [x27]\n"
-      "ld1 { v24.h }[2], [x26]\n"
-      "ld1 { v25.h }[2], [x25]\n"
-      "ld1 { v21.h }[2], [x24]\n"
-      "ld1 { v23.h }[2], [x23]\n"
-      "ld1 { v18.h }[2], [x22]\n"
-      "ld1 { v20.h }[2], [x21]\n"
-      "ld1 { v16.h }[2], [x20]\n"
+      "ld1 { v26.h }[2], [x28]\n"
+      "ld1 { v21.h }[2], [x27]\n"
+      "ld1 { v25.h }[2], [x26]\n"
+      "ld1 { v24.h }[2], [x25]\n"
+      "ld1 { v23.h }[2], [x24]\n"
+      "ld1 { v22.h }[2], [x23]\n"
+      "ld1 { v20.h }[2], [x22]\n"
+      "ld1 { v19.h }[2], [x21]\n"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr h27, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr h24, [x26, #0x0]\n"
-      "ldr h25, [x25, #0x0]\n"
-      "ldr h21, [x24, #0x0]\n"
-      "ldr h23, [x23, #0x0]\n"
-      "ldr h18, [x22, #0x0]\n"
-      "ldr h20, [x21, #0x0]\n"
-      "ldr h16, [x20, #0x0]\n"
+      "ldr h26, [x28, #0x0]\n"
+      "ldr h21, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr h25, [x26, #0x0]\n"
+      "ldr h24, [x25, #0x0]\n"
+      "ldr h23, [x24, #0x0]\n"
+      "ldr h22, [x23, #0x0]\n"
+      "ldr h20, [x22, #0x0]\n"
+      "ldr h19, [x21, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 8f\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "str q21, [%x[out_ptr], #0x0]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "8:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_fp32_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_fp32_bf16.hpp
index 533682c..dfec143 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_fp32_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_fp32_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset], LSL #2\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset], LSL #2\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset], LSL #2\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset], LSL #2\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset], LSL #2\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset], LSL #2\n"
       "add x22, x22, %x[row_offset], LSL #2\n"
       "add x21, x21, %x[row_offset], LSL #2\n"
-      "add x20, x20, %x[row_offset], LSL #2\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x4\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,92 +77,91 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q23, [x27], #0x10\n"
-      ".inst 0x0ea16af7  // bfcvtn v23.4h, v23.4s\n"
+      "ldr q23, [x28], #0x10\n"
       "ldr q22, [x26], #0x10\n"
-      "subs %x[width], %x[width], #0x4\n"
-      ".inst 0x4ea16ad7  // bfcvtn2 v23.8h, v22.4s\n"
-      "ldr q21, [x25], #0x10\n"
-      "cmp %x[width], #0x4\n"
+      ".inst 0x0ea16af7  // bfcvtn v23.4h, v23.4s\n"
+      ".inst 0x0ea16ad6  // bfcvtn v22.4h, v22.4s\n"
+      "ldr q21, [x24], #0x10\n"
+      "ldr q20, [x22], #0x10\n"
       ".inst 0x0ea16ab5  // bfcvtn v21.4h, v21.4s\n"
-      "ldr q20, [x24], #0x10\n"
-      "ldr q18, [x23], #0x10\n"
-      ".inst 0x4ea16a95  // bfcvtn2 v21.8h, v20.4s\n"
-      "ldr q19, [x22], #0x10\n"
-      ".inst 0x0ea16a52  // bfcvtn v18.4h, v18.4s\n"
+      ".inst 0x0ea16a94  // bfcvtn v20.4h, v20.4s\n"
+      "ldr q19, [x27], #0x10\n"
+      "ldr q18, [x25], #0x10\n"
+      "subs %x[width], %x[width], #0x4\n"
+      "cmp %x[width], #0x4\n"
+      "ldr q17, [x23], #0x10\n"
       "ldr q16, [x21], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
-      ".inst 0x4ea16a72  // bfcvtn2 v18.8h, v19.4s\n"
+      ".inst 0x4ea16a77  // bfcvtn2 v23.8h, v19.4s\n"
+      ".inst 0x4ea16a56  // bfcvtn2 v22.8h, v18.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
+      ".inst 0x4ea16a35  // bfcvtn2 v21.8h, v17.4s\n"
+      ".inst 0x4ea16a14  // bfcvtn2 v20.8h, v16.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      ".inst 0x0ea16a10  // bfcvtn v16.4h, v16.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      ".inst 0x4ea16a30  // bfcvtn2 v16.8h, v17.4s\n"
+      "str q23, [%x[out_ptr], #0x0]\n"
       "prfm pldl1keep, [x24, #0x70]\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q22, [%x[out_ptr], #0x10]\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
+      "str q21, [%x[out_ptr], #0x20]\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr d23, [x27], #0x8\n"
+      "ldr d23, [x28], #0x8\n"
+      "ldr d19, [x27], #0x8\n"
+      "mov x20, #0x1\n"
       "ldr d22, [x26], #0x8\n"
-      "mov x19, #0x1\n"
-      "ldr d21, [x25], #0x8\n"
-      "ldr d20, [x24], #0x8\n"
-      "ldr d18, [x23], #0x8\n"
-      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x25], #0x8\n"
+      "ldr d21, [x24], #0x8\n"
+      "ldr d17, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
       "ldr d16, [x21], #0x8\n"
-      "ldr d17, [x20], #0x8\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v23.s }[2], [x27]\n"
+      "ld1 { v23.s }[2], [x28]\n"
+      "ld1 { v19.s }[2], [x27]\n"
       "ld1 { v22.s }[2], [x26]\n"
-      "ld1 { v21.s }[2], [x25]\n"
-      "ld1 { v20.s }[2], [x24]\n"
-      "ld1 { v18.s }[2], [x23]\n"
-      "ld1 { v19.s }[2], [x22]\n"
+      "ld1 { v18.s }[2], [x25]\n"
+      "ld1 { v21.s }[2], [x24]\n"
+      "ld1 { v17.s }[2], [x23]\n"
+      "ld1 { v20.s }[2], [x22]\n"
       "ld1 { v16.s }[2], [x21]\n"
-      "ld1 { v17.s }[2], [x20]\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr s23, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr s23, [x28, #0x0]\n"
+      "ldr s19, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
       "ldr s22, [x26, #0x0]\n"
-      "ldr s21, [x25, #0x0]\n"
-      "ldr s20, [x24, #0x0]\n"
-      "ldr s18, [x23, #0x0]\n"
-      "ldr s19, [x22, #0x0]\n"
+      "ldr s18, [x25, #0x0]\n"
+      "ldr s21, [x24, #0x0]\n"
+      "ldr s17, [x23, #0x0]\n"
+      "ldr s20, [x22, #0x0]\n"
       "ldr s16, [x21, #0x0]\n"
-      "ldr s17, [x20, #0x0]\n"
       "5:"  // Odd load end
       ".inst 0x0ea16af7  // bfcvtn v23.4h, v23.4s\n"
+      ".inst 0x0ea16ad6  // bfcvtn v22.4h, v22.4s\n"
       ".inst 0x0ea16ab5  // bfcvtn v21.4h, v21.4s\n"
-      ".inst 0x0ea16a52  // bfcvtn v18.4h, v18.4s\n"
-      ".inst 0x0ea16a10  // bfcvtn v16.4h, v16.4s\n"
-      ".inst 0x4ea16ad7  // bfcvtn2 v23.8h, v22.4s\n"
+      ".inst 0x0ea16a94  // bfcvtn v20.4h, v20.4s\n"
+      ".inst 0x4ea16a77  // bfcvtn2 v23.8h, v19.4s\n"
+      ".inst 0x4ea16a56  // bfcvtn2 v22.8h, v18.4s\n"
       "str q23, [%x[out_ptr], #0x0]\n"
-      ".inst 0x4ea16a95  // bfcvtn2 v21.8h, v20.4s\n"
-      ".inst 0x4ea16a72  // bfcvtn2 v18.8h, v19.4s\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      ".inst 0x4ea16a30  // bfcvtn2 v16.8h, v17.4s\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
+      ".inst 0x4ea16a35  // bfcvtn2 v21.8h, v17.4s\n"
+      ".inst 0x4ea16a14  // bfcvtn2 v20.8h, v16.4s\n"
+      "str q22, [%x[out_ptr], #0x10]\n"
+      "str q21, [%x[out_ptr], #0x20]\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "6:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8.hpp
index 659d994..54f15f8 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x10\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,49 +77,48 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q28, [x27], #0x10\n"
+      "ldr q28, [x28], #0x10\n"
+      "ldr q27, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "ldr q25, [x25], #0x10\n"
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "ldr q21, [x24], #0x10\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "ldr q27, [x23], #0x10\n"
-      "ldr q26, [x22], #0x10\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "ldr q24, [x20], #0x10\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "ldr q22, [x26], #0x10\n"
+      "ldr q21, [x25], #0x10\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "ldr q24, [x24], #0x10\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "ldr q19, [x22], #0x10\n"
+      "ldr q18, [x21], #0x10\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x40]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x50]\n"
+      "zip2 v16.4s, v19.4s, v18.4s\n"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -126,204 +126,204 @@
       "3:"  // Main loop skip
       "cbz %x[width], 12f\n"
       "tbz %x[width], #3, 7f\n"
-      "ldr d28, [x27], #0x8\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "ldr d26, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d24, [x20], #0x8\n"
+      "ldr d28, [x28], #0x8\n"
+      "ldr d27, [x27], #0x8\n"
+      "ldr d22, [x26], #0x8\n"
+      "ldr d21, [x25], #0x8\n"
+      "ldr d24, [x24], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
+      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x21], #0x8\n"
       "tbz %x[width], #2, 5f\n"
-      "ld1 { v28.s }[2], [x27], #0x4\n"
-      "ld1 { v29.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v27.s }[2], [x23], #0x4\n"
-      "ld1 { v26.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v24.s }[2], [x20], #0x4\n"
+      "ld1 { v28.s }[2], [x28], #0x4\n"
+      "ld1 { v27.s }[2], [x27], #0x4\n"
+      "ld1 { v22.s }[2], [x26], #0x4\n"
+      "ld1 { v21.s }[2], [x25], #0x4\n"
+      "ld1 { v24.s }[2], [x24], #0x4\n"
+      "ld1 { v23.s }[2], [x23], #0x4\n"
+      "ld1 { v19.s }[2], [x22], #0x4\n"
+      "ld1 { v18.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v28.h }[6], [x27], #0x2\n"
-      "mov x19, #0x4\n"
-      "ld1 { v29.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v27.h }[6], [x23], #0x2\n"
-      "ld1 { v26.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v24.h }[6], [x20], #0x2\n"
+      "ld1 { v28.h }[6], [x28], #0x2\n"
+      "ld1 { v27.h }[6], [x27], #0x2\n"
+      "mov x20, #0x4\n"
+      "ld1 { v22.h }[6], [x26], #0x2\n"
+      "ld1 { v21.h }[6], [x25], #0x2\n"
+      "ld1 { v24.h }[6], [x24], #0x2\n"
+      "ld1 { v23.h }[6], [x23], #0x2\n"
+      "ld1 { v19.h }[6], [x22], #0x2\n"
+      "ld1 { v18.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[14], [x27]\n"
-      "ld1 { v29.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v27.b }[14], [x23]\n"
-      "ld1 { v26.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v24.b }[14], [x20]\n"
+      "ld1 { v28.b }[14], [x28]\n"
+      "ld1 { v27.b }[14], [x27]\n"
+      "ld1 { v22.b }[14], [x26]\n"
+      "ld1 { v21.b }[14], [x25]\n"
+      "ld1 { v24.b }[14], [x24]\n"
+      "ld1 { v23.b }[14], [x23]\n"
+      "ld1 { v19.b }[14], [x22]\n"
+      "ld1 { v18.b }[14], [x21]\n"
       "b 11f\n"
       "4:"  // odd_loads_1_12
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[12], [x27]\n"
-      "ld1 { v29.b }[12], [x26]\n"
-      "mov x19, #0x4\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v27.b }[12], [x23]\n"
-      "ld1 { v26.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v24.b }[12], [x20]\n"
+      "ld1 { v28.b }[12], [x28]\n"
+      "ld1 { v27.b }[12], [x27]\n"
+      "mov x20, #0x4\n"
+      "ld1 { v22.b }[12], [x26]\n"
+      "ld1 { v21.b }[12], [x25]\n"
+      "ld1 { v24.b }[12], [x24]\n"
+      "ld1 { v23.b }[12], [x23]\n"
+      "ld1 { v19.b }[12], [x22]\n"
+      "ld1 { v18.b }[12], [x21]\n"
       "b 11f\n"
       "5:"  // odd_loads_2_8
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v28.h }[4], [x27], #0x2\n"
-      "ld1 { v29.h }[4], [x26], #0x2\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v27.h }[4], [x23], #0x2\n"
-      "ld1 { v26.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v24.h }[4], [x20], #0x2\n"
+      "ld1 { v28.h }[4], [x28], #0x2\n"
+      "ld1 { v27.h }[4], [x27], #0x2\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.h }[4], [x26], #0x2\n"
+      "ld1 { v21.h }[4], [x25], #0x2\n"
+      "ld1 { v24.h }[4], [x24], #0x2\n"
+      "ld1 { v23.h }[4], [x23], #0x2\n"
+      "ld1 { v19.h }[4], [x22], #0x2\n"
+      "ld1 { v18.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[10], [x27]\n"
-      "ld1 { v29.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v27.b }[10], [x23]\n"
-      "ld1 { v26.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v24.b }[10], [x20]\n"
+      "ld1 { v28.b }[10], [x28]\n"
+      "ld1 { v27.b }[10], [x27]\n"
+      "ld1 { v22.b }[10], [x26]\n"
+      "ld1 { v21.b }[10], [x25]\n"
+      "ld1 { v24.b }[10], [x24]\n"
+      "ld1 { v23.b }[10], [x23]\n"
+      "ld1 { v19.b }[10], [x22]\n"
+      "ld1 { v18.b }[10], [x21]\n"
       "b 11f\n"
       "6:"  // odd_loads_1_8
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[8], [x27]\n"
-      "ld1 { v29.b }[8], [x26]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v27.b }[8], [x23]\n"
-      "ld1 { v26.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v24.b }[8], [x20]\n"
+      "ld1 { v28.b }[8], [x28]\n"
+      "ld1 { v27.b }[8], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v22.b }[8], [x26]\n"
+      "ld1 { v21.b }[8], [x25]\n"
+      "ld1 { v24.b }[8], [x24]\n"
+      "ld1 { v23.b }[8], [x23]\n"
+      "ld1 { v19.b }[8], [x22]\n"
+      "ld1 { v18.b }[8], [x21]\n"
       "b 11f\n"
       "7:"  // odd_loads_4_0
       "tbz %x[width], #2, 9f\n"
-      "ldr s28, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s27, [x23], #0x4\n"
-      "ldr s26, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s24, [x20], #0x4\n"
+      "ldr s28, [x28], #0x4\n"
+      "ldr s27, [x27], #0x4\n"
+      "ldr s22, [x26], #0x4\n"
+      "ldr s21, [x25], #0x4\n"
+      "ldr s24, [x24], #0x4\n"
+      "ldr s23, [x23], #0x4\n"
+      "ldr s19, [x22], #0x4\n"
+      "ldr s18, [x21], #0x4\n"
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v28.h }[2], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v29.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v27.h }[2], [x23], #0x2\n"
-      "ld1 { v26.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v24.h }[2], [x20], #0x2\n"
+      "ld1 { v28.h }[2], [x28], #0x2\n"
+      "ld1 { v27.h }[2], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v22.h }[2], [x26], #0x2\n"
+      "ld1 { v21.h }[2], [x25], #0x2\n"
+      "ld1 { v24.h }[2], [x24], #0x2\n"
+      "ld1 { v23.h }[2], [x23], #0x2\n"
+      "ld1 { v19.h }[2], [x22], #0x2\n"
+      "ld1 { v18.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[6], [x27]\n"
-      "ld1 { v29.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v27.b }[6], [x23]\n"
-      "ld1 { v26.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v24.b }[6], [x20]\n"
+      "ld1 { v28.b }[6], [x28]\n"
+      "ld1 { v27.b }[6], [x27]\n"
+      "ld1 { v22.b }[6], [x26]\n"
+      "ld1 { v21.b }[6], [x25]\n"
+      "ld1 { v24.b }[6], [x24]\n"
+      "ld1 { v23.b }[6], [x23]\n"
+      "ld1 { v19.b }[6], [x22]\n"
+      "ld1 { v18.b }[6], [x21]\n"
       "b 11f\n"
       "8:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[4], [x27]\n"
-      "ld1 { v29.b }[4], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v27.b }[4], [x23]\n"
-      "ld1 { v26.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v24.b }[4], [x20]\n"
+      "ld1 { v28.b }[4], [x28]\n"
+      "ld1 { v27.b }[4], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v22.b }[4], [x26]\n"
+      "ld1 { v21.b }[4], [x25]\n"
+      "ld1 { v24.b }[4], [x24]\n"
+      "ld1 { v23.b }[4], [x23]\n"
+      "ld1 { v19.b }[4], [x22]\n"
+      "ld1 { v18.b }[4], [x21]\n"
       "b 11f\n"
       "9:"  // odd_loads_2_0
       "tbz %x[width], #1, 10f\n"
-      "ldr h28, [x27], #0x2\n"
-      "ldr h29, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h27, [x23], #0x2\n"
-      "ldr h26, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h24, [x20], #0x2\n"
+      "ldr h28, [x28], #0x2\n"
+      "ldr h27, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h22, [x26], #0x2\n"
+      "ldr h21, [x25], #0x2\n"
+      "ldr h24, [x24], #0x2\n"
+      "ldr h23, [x23], #0x2\n"
+      "ldr h19, [x22], #0x2\n"
+      "ldr h18, [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v28.b }[2], [x27]\n"
-      "ld1 { v29.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v27.b }[2], [x23]\n"
-      "ld1 { v26.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v24.b }[2], [x20]\n"
+      "ld1 { v28.b }[2], [x28]\n"
+      "ld1 { v27.b }[2], [x27]\n"
+      "ld1 { v22.b }[2], [x26]\n"
+      "ld1 { v21.b }[2], [x25]\n"
+      "ld1 { v24.b }[2], [x24]\n"
+      "ld1 { v23.b }[2], [x23]\n"
+      "ld1 { v19.b }[2], [x22]\n"
+      "ld1 { v18.b }[2], [x21]\n"
       "b 11f\n"
       "10:"  // odd_loads_1_0
-      "ldr b28, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b29, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b27, [x23, #0x0]\n"
-      "ldr b26, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b24, [x20, #0x0]\n"
+      "ldr b28, [x28, #0x0]\n"
+      "ldr b27, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b22, [x26, #0x0]\n"
+      "ldr b21, [x25, #0x0]\n"
+      "ldr b24, [x24, #0x0]\n"
+      "ldr b23, [x23, #0x0]\n"
+      "ldr b19, [x22, #0x0]\n"
+      "ldr b18, [x21, #0x0]\n"
       "11:"  // Odd load end
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
+      "zip1 v26.4s, v28.4s, v22.4s\n"
+      "zip1 v25.4s, v27.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v20.4s, v24.4s, v19.4s\n"
+      "zip1 v17.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
-      "subs x19, x19, #0x1\n"
-      "str q20, [%x[out_ptr], #0x10]\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v16.4s, v26.4s, v25.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v20.4s, v17.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v22.4s, v28.4s, v22.4s\n"
+      "zip2 v21.4s, v27.4s, v21.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v19.4s, v24.4s, v19.4s\n"
+      "zip2 v18.4s, v23.4s, v18.4s\n"
+      "zip1 v16.4s, v22.4s, v21.4s\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.4s, v19.4s, v18.4s\n"
+      "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
+      "zip2 v17.4s, v22.4s, v21.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
+      "zip2 v16.4s, v19.4s, v18.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "12:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp
index dfec94c..2db5412 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,40 +31,41 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.8h, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.8h, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
       "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
+      "add x28, x28, %x[row_offset]\n"
       "add x27, x27, %x[row_offset]\n"
-      "movi v30.4s, #0x0\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -72,7 +73,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -80,71 +81,70 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v31.4s }, [%x[out_ptr]]\n"
-      "ldr q30, [%x[out_ptr], #0x10]\n"
+      "ld1 { v0.4s }, [%x[out_ptr]]\n"
+      "ldr q31, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x10\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0x1e\n"
+      "cmp x20, #0x1e\n"
       "ble 4f\n"
+      "sadalp v0.4s, v2.8h\n"
+      "movi v2.8h, #0x0\n"
+      "mov x20, #0x0\n"
       "sadalp v31.4s, v1.8h\n"
       "movi v1.8h, #0x0\n"
-      "sadalp v30.4s, v0.8h\n"
-      "movi v0.8h, #0x0\n"
-      "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q28, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q29, [x26], #0x10\n"
+      "ldr q30, [x28], #0x10\n"
+      "ldr q29, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q25, [x25], #0x10\n"
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "ldr q21, [x24], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "ldr q27, [x23], #0x10\n"
-      "ldr q26, [x22], #0x10\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "ldr q24, [x20], #0x10\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "ldr q28, [x26], #0x10\n"
+      "ldr q27, [x25], #0x10\n"
+      "zip1 v22.4s, v30.4s, v28.4s\n"
+      "zip1 v21.4s, v29.4s, v27.4s\n"
+      "ldr q20, [x24], #0x10\n"
+      "ldr q26, [x23], #0x10\n"
+      "zip1 v25.4s, v22.4s, v21.4s\n"
+      "sadalp v2.8h, v25.16b\n"
+      "ldr q19, [x22], #0x10\n"
+      "ldr q18, [x21], #0x10\n"
+      "zip1 v17.4s, v20.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v24.4s, v17.4s, v16.4s\n"
+      "sadalp v1.8h, v24.16b\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "sadalp v1.8h, v23.16b\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
+      "zip2 v23.4s, v22.4s, v21.4s\n"
+      "zip2 v22.4s, v17.4s, v16.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "sadalp v1.8h, v22.16b\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
+      "zip2 v21.4s, v30.4s, v28.4s\n"
+      "zip2 v17.4s, v29.4s, v27.4s\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "sadalp v1.8h, v21.16b\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "zip2 v20.4s, v20.4s, v19.4s\n"
+      "zip2 v16.4s, v26.4s, v18.4s\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "sadalp v0.8h, v17.16b\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "sadalp v0.8h, v20.16b\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v18.16b\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "sadalp v1.8h, v17.16b\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "sadalp v0.8h, v16.16b\n"
+      "sadalp v2.8h, v23.16b\n"
+      "sadalp v1.8h, v22.16b\n"
+      "str q25, [%x[out_ptr], #0x0]\n"
+      "add x20, x20, #0x1\n"
+      "zip1 v19.4s, v21.4s, v17.4s\n"
+      "zip1 v18.4s, v20.4s, v16.4s\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "sadalp v2.8h, v19.16b\n"
+      "sadalp v1.8h, v18.16b\n"
+      "str q23, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.4s, v21.4s, v17.4s\n"
+      "zip2 v16.4s, v20.4s, v16.4s\n"
+      "str q22, [%x[out_ptr], #0x30]\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "sadalp v2.8h, v17.16b\n"
+      "sadalp v1.8h, v16.16b\n"
       "str q18, [%x[out_ptr], #0x50]\n"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
@@ -153,216 +153,216 @@
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d28, [x27], #0x8\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "ldr d26, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d24, [x20], #0x8\n"
+      "ldr d30, [x28], #0x8\n"
+      "ldr d29, [x27], #0x8\n"
+      "ldr d28, [x26], #0x8\n"
+      "ldr d27, [x25], #0x8\n"
+      "ldr d20, [x24], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
+      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x21], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v28.s }[2], [x27], #0x4\n"
-      "ld1 { v29.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v27.s }[2], [x23], #0x4\n"
-      "ld1 { v26.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v24.s }[2], [x20], #0x4\n"
+      "ld1 { v30.s }[2], [x28], #0x4\n"
+      "ld1 { v29.s }[2], [x27], #0x4\n"
+      "ld1 { v28.s }[2], [x26], #0x4\n"
+      "ld1 { v27.s }[2], [x25], #0x4\n"
+      "ld1 { v20.s }[2], [x24], #0x4\n"
+      "ld1 { v26.s }[2], [x23], #0x4\n"
+      "ld1 { v19.s }[2], [x22], #0x4\n"
+      "ld1 { v18.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v28.h }[6], [x27], #0x2\n"
-      "mov x19, #0x4\n"
-      "ld1 { v29.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v27.h }[6], [x23], #0x2\n"
-      "ld1 { v26.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v24.h }[6], [x20], #0x2\n"
+      "ld1 { v30.h }[6], [x28], #0x2\n"
+      "ld1 { v29.h }[6], [x27], #0x2\n"
+      "mov x20, #0x4\n"
+      "ld1 { v28.h }[6], [x26], #0x2\n"
+      "ld1 { v27.h }[6], [x25], #0x2\n"
+      "ld1 { v20.h }[6], [x24], #0x2\n"
+      "ld1 { v26.h }[6], [x23], #0x2\n"
+      "ld1 { v19.h }[6], [x22], #0x2\n"
+      "ld1 { v18.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[14], [x27]\n"
-      "ld1 { v29.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v27.b }[14], [x23]\n"
-      "ld1 { v26.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v24.b }[14], [x20]\n"
+      "ld1 { v30.b }[14], [x28]\n"
+      "ld1 { v29.b }[14], [x27]\n"
+      "ld1 { v28.b }[14], [x26]\n"
+      "ld1 { v27.b }[14], [x25]\n"
+      "ld1 { v20.b }[14], [x24]\n"
+      "ld1 { v26.b }[14], [x23]\n"
+      "ld1 { v19.b }[14], [x22]\n"
+      "ld1 { v18.b }[14], [x21]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[12], [x27]\n"
-      "ld1 { v29.b }[12], [x26]\n"
-      "mov x19, #0x4\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v27.b }[12], [x23]\n"
-      "ld1 { v26.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v24.b }[12], [x20]\n"
+      "ld1 { v30.b }[12], [x28]\n"
+      "ld1 { v29.b }[12], [x27]\n"
+      "mov x20, #0x4\n"
+      "ld1 { v28.b }[12], [x26]\n"
+      "ld1 { v27.b }[12], [x25]\n"
+      "ld1 { v20.b }[12], [x24]\n"
+      "ld1 { v26.b }[12], [x23]\n"
+      "ld1 { v19.b }[12], [x22]\n"
+      "ld1 { v18.b }[12], [x21]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v28.h }[4], [x27], #0x2\n"
-      "ld1 { v29.h }[4], [x26], #0x2\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v27.h }[4], [x23], #0x2\n"
-      "ld1 { v26.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v24.h }[4], [x20], #0x2\n"
+      "ld1 { v30.h }[4], [x28], #0x2\n"
+      "ld1 { v29.h }[4], [x27], #0x2\n"
+      "mov x20, #0x3\n"
+      "ld1 { v28.h }[4], [x26], #0x2\n"
+      "ld1 { v27.h }[4], [x25], #0x2\n"
+      "ld1 { v20.h }[4], [x24], #0x2\n"
+      "ld1 { v26.h }[4], [x23], #0x2\n"
+      "ld1 { v19.h }[4], [x22], #0x2\n"
+      "ld1 { v18.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[10], [x27]\n"
-      "ld1 { v29.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v27.b }[10], [x23]\n"
-      "ld1 { v26.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v24.b }[10], [x20]\n"
+      "ld1 { v30.b }[10], [x28]\n"
+      "ld1 { v29.b }[10], [x27]\n"
+      "ld1 { v28.b }[10], [x26]\n"
+      "ld1 { v27.b }[10], [x25]\n"
+      "ld1 { v20.b }[10], [x24]\n"
+      "ld1 { v26.b }[10], [x23]\n"
+      "ld1 { v19.b }[10], [x22]\n"
+      "ld1 { v18.b }[10], [x21]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[8], [x27]\n"
-      "ld1 { v29.b }[8], [x26]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v27.b }[8], [x23]\n"
-      "ld1 { v26.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v24.b }[8], [x20]\n"
+      "ld1 { v30.b }[8], [x28]\n"
+      "ld1 { v29.b }[8], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v28.b }[8], [x26]\n"
+      "ld1 { v27.b }[8], [x25]\n"
+      "ld1 { v20.b }[8], [x24]\n"
+      "ld1 { v26.b }[8], [x23]\n"
+      "ld1 { v19.b }[8], [x22]\n"
+      "ld1 { v18.b }[8], [x21]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s28, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s27, [x23], #0x4\n"
-      "ldr s26, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s24, [x20], #0x4\n"
+      "ldr s30, [x28], #0x4\n"
+      "ldr s29, [x27], #0x4\n"
+      "ldr s28, [x26], #0x4\n"
+      "ldr s27, [x25], #0x4\n"
+      "ldr s20, [x24], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s19, [x22], #0x4\n"
+      "ldr s18, [x21], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v28.h }[2], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v29.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v27.h }[2], [x23], #0x2\n"
-      "ld1 { v26.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v24.h }[2], [x20], #0x2\n"
+      "ld1 { v30.h }[2], [x28], #0x2\n"
+      "ld1 { v29.h }[2], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v28.h }[2], [x26], #0x2\n"
+      "ld1 { v27.h }[2], [x25], #0x2\n"
+      "ld1 { v20.h }[2], [x24], #0x2\n"
+      "ld1 { v26.h }[2], [x23], #0x2\n"
+      "ld1 { v19.h }[2], [x22], #0x2\n"
+      "ld1 { v18.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[6], [x27]\n"
-      "ld1 { v29.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v27.b }[6], [x23]\n"
-      "ld1 { v26.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v24.b }[6], [x20]\n"
+      "ld1 { v30.b }[6], [x28]\n"
+      "ld1 { v29.b }[6], [x27]\n"
+      "ld1 { v28.b }[6], [x26]\n"
+      "ld1 { v27.b }[6], [x25]\n"
+      "ld1 { v20.b }[6], [x24]\n"
+      "ld1 { v26.b }[6], [x23]\n"
+      "ld1 { v19.b }[6], [x22]\n"
+      "ld1 { v18.b }[6], [x21]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[4], [x27]\n"
-      "ld1 { v29.b }[4], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v27.b }[4], [x23]\n"
-      "ld1 { v26.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v24.b }[4], [x20]\n"
+      "ld1 { v30.b }[4], [x28]\n"
+      "ld1 { v29.b }[4], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v28.b }[4], [x26]\n"
+      "ld1 { v27.b }[4], [x25]\n"
+      "ld1 { v20.b }[4], [x24]\n"
+      "ld1 { v26.b }[4], [x23]\n"
+      "ld1 { v19.b }[4], [x22]\n"
+      "ld1 { v18.b }[4], [x21]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h28, [x27], #0x2\n"
-      "ldr h29, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h27, [x23], #0x2\n"
-      "ldr h26, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h24, [x20], #0x2\n"
+      "ldr h30, [x28], #0x2\n"
+      "ldr h29, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h28, [x26], #0x2\n"
+      "ldr h27, [x25], #0x2\n"
+      "ldr h20, [x24], #0x2\n"
+      "ldr h26, [x23], #0x2\n"
+      "ldr h19, [x22], #0x2\n"
+      "ldr h18, [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[2], [x27]\n"
-      "ld1 { v29.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v27.b }[2], [x23]\n"
-      "ld1 { v26.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v24.b }[2], [x20]\n"
+      "ld1 { v30.b }[2], [x28]\n"
+      "ld1 { v29.b }[2], [x27]\n"
+      "ld1 { v28.b }[2], [x26]\n"
+      "ld1 { v27.b }[2], [x25]\n"
+      "ld1 { v20.b }[2], [x24]\n"
+      "ld1 { v26.b }[2], [x23]\n"
+      "ld1 { v19.b }[2], [x22]\n"
+      "ld1 { v18.b }[2], [x21]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b28, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b29, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b27, [x23, #0x0]\n"
-      "ldr b26, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b24, [x20, #0x0]\n"
+      "ldr b30, [x28, #0x0]\n"
+      "ldr b29, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b28, [x26, #0x0]\n"
+      "ldr b27, [x25, #0x0]\n"
+      "ldr b20, [x24, #0x0]\n"
+      "ldr b26, [x23, #0x0]\n"
+      "ldr b19, [x22, #0x0]\n"
+      "ldr b18, [x21, #0x0]\n"
       "13:"  // Odd load end
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "zip1 v22.4s, v30.4s, v28.4s\n"
+      "zip1 v21.4s, v29.4s, v27.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.4s, v20.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v25.4s, v22.4s, v21.4s\n"
+      "zip1 v24.4s, v17.4s, v16.4s\n"
+      "str q25, [%x[out_ptr], #0x0]\n"
+      "sadalp v2.8h, v25.16b\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "sadalp v1.8h, v24.16b\n"
+      "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "beq 14f\n"
+      "zip2 v23.4s, v22.4s, v21.4s\n"
+      "zip2 v22.4s, v17.4s, v16.4s\n"
+      "subs x20, x20, #0x1\n"
       "str q23, [%x[out_ptr], #0x0]\n"
-      "sadalp v1.8h, v23.16b\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v17.16b\n"
-      "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "beq 14f\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
+      "sadalp v2.8h, v23.16b\n"
+      "str q22, [%x[out_ptr], #0x10]\n"
       "sadalp v1.8h, v22.16b\n"
-      "str q20, [%x[out_ptr], #0x10]\n"
-      "subs x19, x19, #0x1\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "sadalp v0.8h, v20.16b\n"
       "beq 14f\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "sadalp v1.8h, v21.16b\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
+      "zip2 v21.4s, v30.4s, v28.4s\n"
+      "zip2 v17.4s, v29.4s, v27.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.4s, v20.4s, v19.4s\n"
+      "zip2 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v19.4s, v21.4s, v17.4s\n"
+      "zip1 v18.4s, v20.4s, v16.4s\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "sadalp v2.8h, v19.16b\n"
       "str q18, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v18.16b\n"
+      "sadalp v1.8h, v18.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 14f\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
+      "zip2 v17.4s, v21.4s, v17.4s\n"
+      "zip2 v16.4s, v20.4s, v16.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "sadalp v1.8h, v17.16b\n"
+      "sadalp v2.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x10]\n"
+      "sadalp v1.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "sadalp v0.8h, v16.16b\n"
       "14:"  // Odds skip
+      "sadalp v0.4s, v2.8h\n"
       "sadalp v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x0]\n"
-      "sadalp v30.4s, v0.8h\n"
-      "str q30, [%x[out_ptr], #0x10]\n"
+      "str q0, [%x[out_ptr], #0x0]\n"
+      "str q31, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_u8_u8_summing.hpp
index 1b94c7f..44a79c0 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,40 +31,41 @@
 )
 {
   __asm__ __volatile__(
-      "movi v1.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v0.8h, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v2.8h, #0x0\n"
+      "movi v1.8h, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
+      "movi v0.4s, #0x0\n"
       "movi v31.4s, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
+      "add x28, x28, %x[row_offset]\n"
       "add x27, x27, %x[row_offset]\n"
-      "movi v30.4s, #0x0\n"
-      "ldr x24, [%x[in], #0x18]\n"
-      "ldr x23, [%x[in], #0x20]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -72,7 +73,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -80,71 +81,70 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
-      "ld1 { v31.4s }, [%x[out_ptr]]\n"
-      "ldr q30, [%x[out_ptr], #0x10]\n"
+      "ld1 { v0.4s }, [%x[out_ptr]]\n"
+      "ldr q31, [%x[out_ptr], #0x10]\n"
       "2:"  // first_pass
       "cmp %x[width], #0x10\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0x1e\n"
+      "cmp x20, #0x1e\n"
       "ble 4f\n"
+      "uadalp v0.4s, v2.8h\n"
+      "movi v2.8h, #0x0\n"
+      "mov x20, #0x0\n"
       "uadalp v31.4s, v1.8h\n"
       "movi v1.8h, #0x0\n"
-      "uadalp v30.4s, v0.8h\n"
-      "movi v0.8h, #0x0\n"
-      "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q28, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q29, [x26], #0x10\n"
+      "ldr q30, [x28], #0x10\n"
+      "ldr q29, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q25, [x25], #0x10\n"
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "ldr q21, [x24], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "ldr q27, [x23], #0x10\n"
-      "ldr q26, [x22], #0x10\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "ldr q24, [x20], #0x10\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "ldr q28, [x26], #0x10\n"
+      "ldr q27, [x25], #0x10\n"
+      "zip1 v22.4s, v30.4s, v28.4s\n"
+      "zip1 v21.4s, v29.4s, v27.4s\n"
+      "ldr q20, [x24], #0x10\n"
+      "ldr q26, [x23], #0x10\n"
+      "zip1 v25.4s, v22.4s, v21.4s\n"
+      "uadalp v2.8h, v25.16b\n"
+      "ldr q19, [x22], #0x10\n"
+      "ldr q18, [x21], #0x10\n"
+      "zip1 v17.4s, v20.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v24.4s, v17.4s, v16.4s\n"
+      "uadalp v1.8h, v24.16b\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "uadalp v1.8h, v23.16b\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
+      "zip2 v23.4s, v22.4s, v21.4s\n"
+      "zip2 v22.4s, v17.4s, v16.4s\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "uadalp v1.8h, v22.16b\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
+      "zip2 v21.4s, v30.4s, v28.4s\n"
+      "zip2 v17.4s, v29.4s, v27.4s\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
       "prfm pldl1keep, [x23, #0x70]\n"
-      "uadalp v1.8h, v21.16b\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "zip2 v20.4s, v20.4s, v19.4s\n"
+      "zip2 v16.4s, v26.4s, v18.4s\n"
       "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "uadalp v0.8h, v17.16b\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "uadalp v0.8h, v20.16b\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v18.16b\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
-      "str q22, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "uadalp v1.8h, v17.16b\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "uadalp v0.8h, v16.16b\n"
+      "uadalp v2.8h, v23.16b\n"
+      "uadalp v1.8h, v22.16b\n"
+      "str q25, [%x[out_ptr], #0x0]\n"
+      "add x20, x20, #0x1\n"
+      "zip1 v19.4s, v21.4s, v17.4s\n"
+      "zip1 v18.4s, v20.4s, v16.4s\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "uadalp v2.8h, v19.16b\n"
+      "uadalp v1.8h, v18.16b\n"
+      "str q23, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.4s, v21.4s, v17.4s\n"
+      "zip2 v16.4s, v20.4s, v16.4s\n"
+      "str q22, [%x[out_ptr], #0x30]\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "uadalp v2.8h, v17.16b\n"
+      "uadalp v1.8h, v16.16b\n"
       "str q18, [%x[out_ptr], #0x50]\n"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
@@ -153,216 +153,216 @@
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d28, [x27], #0x8\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d27, [x23], #0x8\n"
-      "ldr d26, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d24, [x20], #0x8\n"
+      "ldr d30, [x28], #0x8\n"
+      "ldr d29, [x27], #0x8\n"
+      "ldr d28, [x26], #0x8\n"
+      "ldr d27, [x25], #0x8\n"
+      "ldr d20, [x24], #0x8\n"
+      "ldr d26, [x23], #0x8\n"
+      "ldr d19, [x22], #0x8\n"
+      "ldr d18, [x21], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v28.s }[2], [x27], #0x4\n"
-      "ld1 { v29.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v27.s }[2], [x23], #0x4\n"
-      "ld1 { v26.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v24.s }[2], [x20], #0x4\n"
+      "ld1 { v30.s }[2], [x28], #0x4\n"
+      "ld1 { v29.s }[2], [x27], #0x4\n"
+      "ld1 { v28.s }[2], [x26], #0x4\n"
+      "ld1 { v27.s }[2], [x25], #0x4\n"
+      "ld1 { v20.s }[2], [x24], #0x4\n"
+      "ld1 { v26.s }[2], [x23], #0x4\n"
+      "ld1 { v19.s }[2], [x22], #0x4\n"
+      "ld1 { v18.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v28.h }[6], [x27], #0x2\n"
-      "mov x19, #0x4\n"
-      "ld1 { v29.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v27.h }[6], [x23], #0x2\n"
-      "ld1 { v26.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v24.h }[6], [x20], #0x2\n"
+      "ld1 { v30.h }[6], [x28], #0x2\n"
+      "ld1 { v29.h }[6], [x27], #0x2\n"
+      "mov x20, #0x4\n"
+      "ld1 { v28.h }[6], [x26], #0x2\n"
+      "ld1 { v27.h }[6], [x25], #0x2\n"
+      "ld1 { v20.h }[6], [x24], #0x2\n"
+      "ld1 { v26.h }[6], [x23], #0x2\n"
+      "ld1 { v19.h }[6], [x22], #0x2\n"
+      "ld1 { v18.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[14], [x27]\n"
-      "ld1 { v29.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v27.b }[14], [x23]\n"
-      "ld1 { v26.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v24.b }[14], [x20]\n"
+      "ld1 { v30.b }[14], [x28]\n"
+      "ld1 { v29.b }[14], [x27]\n"
+      "ld1 { v28.b }[14], [x26]\n"
+      "ld1 { v27.b }[14], [x25]\n"
+      "ld1 { v20.b }[14], [x24]\n"
+      "ld1 { v26.b }[14], [x23]\n"
+      "ld1 { v19.b }[14], [x22]\n"
+      "ld1 { v18.b }[14], [x21]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
-      "mov x19, #0x3\n"
+      "mov x20, #0x3\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[12], [x27]\n"
-      "ld1 { v29.b }[12], [x26]\n"
-      "mov x19, #0x4\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v27.b }[12], [x23]\n"
-      "ld1 { v26.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v24.b }[12], [x20]\n"
+      "ld1 { v30.b }[12], [x28]\n"
+      "ld1 { v29.b }[12], [x27]\n"
+      "mov x20, #0x4\n"
+      "ld1 { v28.b }[12], [x26]\n"
+      "ld1 { v27.b }[12], [x25]\n"
+      "ld1 { v20.b }[12], [x24]\n"
+      "ld1 { v26.b }[12], [x23]\n"
+      "ld1 { v19.b }[12], [x22]\n"
+      "ld1 { v18.b }[12], [x21]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v28.h }[4], [x27], #0x2\n"
-      "ld1 { v29.h }[4], [x26], #0x2\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v27.h }[4], [x23], #0x2\n"
-      "ld1 { v26.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v24.h }[4], [x20], #0x2\n"
+      "ld1 { v30.h }[4], [x28], #0x2\n"
+      "ld1 { v29.h }[4], [x27], #0x2\n"
+      "mov x20, #0x3\n"
+      "ld1 { v28.h }[4], [x26], #0x2\n"
+      "ld1 { v27.h }[4], [x25], #0x2\n"
+      "ld1 { v20.h }[4], [x24], #0x2\n"
+      "ld1 { v26.h }[4], [x23], #0x2\n"
+      "ld1 { v19.h }[4], [x22], #0x2\n"
+      "ld1 { v18.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[10], [x27]\n"
-      "ld1 { v29.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v27.b }[10], [x23]\n"
-      "ld1 { v26.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v24.b }[10], [x20]\n"
+      "ld1 { v30.b }[10], [x28]\n"
+      "ld1 { v29.b }[10], [x27]\n"
+      "ld1 { v28.b }[10], [x26]\n"
+      "ld1 { v27.b }[10], [x25]\n"
+      "ld1 { v20.b }[10], [x24]\n"
+      "ld1 { v26.b }[10], [x23]\n"
+      "ld1 { v19.b }[10], [x22]\n"
+      "ld1 { v18.b }[10], [x21]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[8], [x27]\n"
-      "ld1 { v29.b }[8], [x26]\n"
-      "mov x19, #0x3\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v27.b }[8], [x23]\n"
-      "ld1 { v26.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v24.b }[8], [x20]\n"
+      "ld1 { v30.b }[8], [x28]\n"
+      "ld1 { v29.b }[8], [x27]\n"
+      "mov x20, #0x3\n"
+      "ld1 { v28.b }[8], [x26]\n"
+      "ld1 { v27.b }[8], [x25]\n"
+      "ld1 { v20.b }[8], [x24]\n"
+      "ld1 { v26.b }[8], [x23]\n"
+      "ld1 { v19.b }[8], [x22]\n"
+      "ld1 { v18.b }[8], [x21]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s28, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s27, [x23], #0x4\n"
-      "ldr s26, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s24, [x20], #0x4\n"
+      "ldr s30, [x28], #0x4\n"
+      "ldr s29, [x27], #0x4\n"
+      "ldr s28, [x26], #0x4\n"
+      "ldr s27, [x25], #0x4\n"
+      "ldr s20, [x24], #0x4\n"
+      "ldr s26, [x23], #0x4\n"
+      "ldr s19, [x22], #0x4\n"
+      "ldr s18, [x21], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v28.h }[2], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v29.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v27.h }[2], [x23], #0x2\n"
-      "ld1 { v26.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v24.h }[2], [x20], #0x2\n"
+      "ld1 { v30.h }[2], [x28], #0x2\n"
+      "ld1 { v29.h }[2], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v28.h }[2], [x26], #0x2\n"
+      "ld1 { v27.h }[2], [x25], #0x2\n"
+      "ld1 { v20.h }[2], [x24], #0x2\n"
+      "ld1 { v26.h }[2], [x23], #0x2\n"
+      "ld1 { v19.h }[2], [x22], #0x2\n"
+      "ld1 { v18.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[6], [x27]\n"
-      "ld1 { v29.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v27.b }[6], [x23]\n"
-      "ld1 { v26.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v24.b }[6], [x20]\n"
+      "ld1 { v30.b }[6], [x28]\n"
+      "ld1 { v29.b }[6], [x27]\n"
+      "ld1 { v28.b }[6], [x26]\n"
+      "ld1 { v27.b }[6], [x25]\n"
+      "ld1 { v20.b }[6], [x24]\n"
+      "ld1 { v26.b }[6], [x23]\n"
+      "ld1 { v19.b }[6], [x22]\n"
+      "ld1 { v18.b }[6], [x21]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[4], [x27]\n"
-      "ld1 { v29.b }[4], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v27.b }[4], [x23]\n"
-      "ld1 { v26.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v24.b }[4], [x20]\n"
+      "ld1 { v30.b }[4], [x28]\n"
+      "ld1 { v29.b }[4], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v28.b }[4], [x26]\n"
+      "ld1 { v27.b }[4], [x25]\n"
+      "ld1 { v20.b }[4], [x24]\n"
+      "ld1 { v26.b }[4], [x23]\n"
+      "ld1 { v19.b }[4], [x22]\n"
+      "ld1 { v18.b }[4], [x21]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h28, [x27], #0x2\n"
-      "ldr h29, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h27, [x23], #0x2\n"
-      "ldr h26, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h24, [x20], #0x2\n"
+      "ldr h30, [x28], #0x2\n"
+      "ldr h29, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h28, [x26], #0x2\n"
+      "ldr h27, [x25], #0x2\n"
+      "ldr h20, [x24], #0x2\n"
+      "ldr h26, [x23], #0x2\n"
+      "ldr h19, [x22], #0x2\n"
+      "ldr h18, [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v28.b }[2], [x27]\n"
-      "ld1 { v29.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v27.b }[2], [x23]\n"
-      "ld1 { v26.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v24.b }[2], [x20]\n"
+      "ld1 { v30.b }[2], [x28]\n"
+      "ld1 { v29.b }[2], [x27]\n"
+      "ld1 { v28.b }[2], [x26]\n"
+      "ld1 { v27.b }[2], [x25]\n"
+      "ld1 { v20.b }[2], [x24]\n"
+      "ld1 { v26.b }[2], [x23]\n"
+      "ld1 { v19.b }[2], [x22]\n"
+      "ld1 { v18.b }[2], [x21]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b28, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b29, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b27, [x23, #0x0]\n"
-      "ldr b26, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b24, [x20, #0x0]\n"
+      "ldr b30, [x28, #0x0]\n"
+      "ldr b29, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b28, [x26, #0x0]\n"
+      "ldr b27, [x25, #0x0]\n"
+      "ldr b20, [x24, #0x0]\n"
+      "ldr b26, [x23, #0x0]\n"
+      "ldr b19, [x22, #0x0]\n"
+      "ldr b18, [x21, #0x0]\n"
       "13:"  // Odd load end
-      "zip1 v22.4s, v28.4s, v25.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v29.4s, v21.4s\n"
-      "zip1 v23.4s, v22.4s, v20.4s\n"
+      "zip1 v22.4s, v30.4s, v28.4s\n"
+      "zip1 v21.4s, v29.4s, v27.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v17.4s, v20.4s, v19.4s\n"
+      "zip1 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v25.4s, v22.4s, v21.4s\n"
+      "zip1 v24.4s, v17.4s, v16.4s\n"
+      "str q25, [%x[out_ptr], #0x0]\n"
+      "uadalp v2.8h, v25.16b\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "uadalp v1.8h, v24.16b\n"
+      "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "beq 14f\n"
+      "zip2 v23.4s, v22.4s, v21.4s\n"
+      "zip2 v22.4s, v17.4s, v16.4s\n"
+      "subs x20, x20, #0x1\n"
       "str q23, [%x[out_ptr], #0x0]\n"
-      "uadalp v1.8h, v23.16b\n"
-      "zip1 v18.4s, v27.4s, v19.4s\n"
-      "zip1 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v17.16b\n"
-      "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "beq 14f\n"
-      "zip2 v22.4s, v22.4s, v20.4s\n"
-      "str q22, [%x[out_ptr], #0x0]\n"
-      "zip2 v20.4s, v18.4s, v16.4s\n"
+      "uadalp v2.8h, v23.16b\n"
+      "str q22, [%x[out_ptr], #0x10]\n"
       "uadalp v1.8h, v22.16b\n"
-      "str q20, [%x[out_ptr], #0x10]\n"
-      "subs x19, x19, #0x1\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "uadalp v0.8h, v20.16b\n"
       "beq 14f\n"
-      "zip2 v28.4s, v28.4s, v25.4s\n"
-      "zip2 v25.4s, v29.4s, v21.4s\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v21.4s, v28.4s, v25.4s\n"
-      "str q21, [%x[out_ptr], #0x0]\n"
-      "uadalp v1.8h, v21.16b\n"
-      "zip2 v19.4s, v27.4s, v19.4s\n"
-      "zip2 v16.4s, v26.4s, v24.4s\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
+      "zip2 v21.4s, v30.4s, v28.4s\n"
+      "zip2 v17.4s, v29.4s, v27.4s\n"
+      "subs x20, x20, #0x1\n"
+      "zip2 v20.4s, v20.4s, v19.4s\n"
+      "zip2 v16.4s, v26.4s, v18.4s\n"
+      "zip1 v19.4s, v21.4s, v17.4s\n"
+      "zip1 v18.4s, v20.4s, v16.4s\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "uadalp v2.8h, v19.16b\n"
       "str q18, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v18.16b\n"
+      "uadalp v1.8h, v18.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 14f\n"
-      "zip2 v17.4s, v28.4s, v25.4s\n"
+      "zip2 v17.4s, v21.4s, v17.4s\n"
+      "zip2 v16.4s, v20.4s, v16.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "uadalp v1.8h, v17.16b\n"
+      "uadalp v2.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x10]\n"
+      "uadalp v1.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "uadalp v0.8h, v16.16b\n"
       "14:"  // Odds skip
+      "uadalp v0.4s, v2.8h\n"
       "uadalp v31.4s, v1.8h\n"
-      "str q31, [%x[out_ptr], #0x0]\n"
-      "uadalp v30.4s, v0.8h\n"
-      "str q30, [%x[out_ptr], #0x10]\n"
+      "str q0, [%x[out_ptr], #0x0]\n"
+      "str q31, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8.hpp
index 1330593..4bfb360 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,44 +31,45 @@
 )
 {
   __asm__ __volatile__(
-      "ldr x27, [%x[in], #0x0]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
       "add x27, x27, %x[row_offset]\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "ldr x24, [%x[in], #0x18]\n"
       "add x26, x26, %x[row_offset]\n"
-      "ldr x23, [%x[in], #0x20]\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "add x25, x25, %x[row_offset]\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "ldr x21, [%x[in], #0x30]\n"
       "add x24, x24, %x[row_offset]\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "prfm pldl1keep, [x27, #0x0]\n"
       "cmp %x[width], #0x10\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
+      "prfm pldl1keep, [x27, #0x0]\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
       "prfm pldl1keep, [x24, #0x0]\n"
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -76,230 +77,229 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr q27, [x27], #0x10\n"
+      "ldr q26, [x28], #0x10\n"
+      "ldr q21, [x27], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "ldr q24, [x26], #0x10\n"
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "ldr q25, [x25], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "ldr q21, [x24], #0x10\n"
-      "ldr q23, [x23], #0x10\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "ldr q18, [x22], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "ldr q20, [x21], #0x10\n"
-      "ldr q16, [x20], #0x10\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "ldr q25, [x26], #0x10\n"
+      "ldr q24, [x25], #0x10\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "ldr q23, [x24], #0x10\n"
+      "ldr q22, [x23], #0x10\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "ldr q20, [x22], #0x10\n"
+      "ldr q19, [x21], #0x10\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
       "prfm pldl1keep, [x24, #0x70]\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "str q24, [%x[out_ptr], #0x40]\n"
-      "str q21, [%x[out_ptr], #0x50]\n"
-      "str q18, [%x[out_ptr], #0x60]\n"
+      "str q21, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
       "cbz %x[width], 12f\n"
       "tbz %x[width], #3, 7f\n"
-      "ldr d27, [x27], #0x8\n"
-      "ldr d24, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d18, [x22], #0x8\n"
-      "ldr d20, [x21], #0x8\n"
-      "ldr d16, [x20], #0x8\n"
+      "ldr d26, [x28], #0x8\n"
+      "ldr d21, [x27], #0x8\n"
+      "ldr d25, [x26], #0x8\n"
+      "ldr d24, [x25], #0x8\n"
+      "ldr d23, [x24], #0x8\n"
+      "ldr d22, [x23], #0x8\n"
+      "ldr d20, [x22], #0x8\n"
+      "ldr d19, [x21], #0x8\n"
       "tbz %x[width], #2, 5f\n"
-      "ld1 { v27.s }[2], [x27], #0x4\n"
-      "ld1 { v24.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v23.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v20.s }[2], [x21], #0x4\n"
-      "ld1 { v16.s }[2], [x20], #0x4\n"
+      "ld1 { v26.s }[2], [x28], #0x4\n"
+      "ld1 { v21.s }[2], [x27], #0x4\n"
+      "ld1 { v25.s }[2], [x26], #0x4\n"
+      "ld1 { v24.s }[2], [x25], #0x4\n"
+      "ld1 { v23.s }[2], [x24], #0x4\n"
+      "ld1 { v22.s }[2], [x23], #0x4\n"
+      "ld1 { v20.s }[2], [x22], #0x4\n"
+      "ld1 { v19.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v27.h }[6], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v24.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v23.h }[6], [x23], #0x2\n"
-      "ld1 { v18.h }[6], [x22], #0x2\n"
-      "ld1 { v20.h }[6], [x21], #0x2\n"
-      "ld1 { v16.h }[6], [x20], #0x2\n"
+      "ld1 { v26.h }[6], [x28], #0x2\n"
+      "ld1 { v21.h }[6], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[6], [x26], #0x2\n"
+      "ld1 { v24.h }[6], [x25], #0x2\n"
+      "ld1 { v23.h }[6], [x24], #0x2\n"
+      "ld1 { v22.h }[6], [x23], #0x2\n"
+      "ld1 { v20.h }[6], [x22], #0x2\n"
+      "ld1 { v19.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[14], [x27]\n"
-      "ld1 { v24.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v23.b }[14], [x23]\n"
-      "ld1 { v18.b }[14], [x22]\n"
-      "ld1 { v20.b }[14], [x21]\n"
-      "ld1 { v16.b }[14], [x20]\n"
+      "ld1 { v26.b }[14], [x28]\n"
+      "ld1 { v21.b }[14], [x27]\n"
+      "ld1 { v25.b }[14], [x26]\n"
+      "ld1 { v24.b }[14], [x25]\n"
+      "ld1 { v23.b }[14], [x24]\n"
+      "ld1 { v22.b }[14], [x23]\n"
+      "ld1 { v20.b }[14], [x22]\n"
+      "ld1 { v19.b }[14], [x21]\n"
       "b 11f\n"
       "4:"  // odd_loads_1_12
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[12], [x27]\n"
-      "ld1 { v24.b }[12], [x26]\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v23.b }[12], [x23]\n"
-      "ld1 { v18.b }[12], [x22]\n"
-      "ld1 { v20.b }[12], [x21]\n"
-      "ld1 { v16.b }[12], [x20]\n"
+      "ld1 { v26.b }[12], [x28]\n"
+      "ld1 { v21.b }[12], [x27]\n"
+      "ld1 { v25.b }[12], [x26]\n"
+      "ld1 { v24.b }[12], [x25]\n"
+      "ld1 { v23.b }[12], [x24]\n"
+      "ld1 { v22.b }[12], [x23]\n"
+      "ld1 { v20.b }[12], [x22]\n"
+      "ld1 { v19.b }[12], [x21]\n"
       "b 11f\n"
       "5:"  // odd_loads_2_8
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v27.h }[4], [x27], #0x2\n"
-      "ld1 { v24.h }[4], [x26], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v23.h }[4], [x23], #0x2\n"
-      "ld1 { v18.h }[4], [x22], #0x2\n"
-      "ld1 { v20.h }[4], [x21], #0x2\n"
-      "ld1 { v16.h }[4], [x20], #0x2\n"
+      "ld1 { v26.h }[4], [x28], #0x2\n"
+      "ld1 { v21.h }[4], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[4], [x26], #0x2\n"
+      "ld1 { v24.h }[4], [x25], #0x2\n"
+      "ld1 { v23.h }[4], [x24], #0x2\n"
+      "ld1 { v22.h }[4], [x23], #0x2\n"
+      "ld1 { v20.h }[4], [x22], #0x2\n"
+      "ld1 { v19.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[10], [x27]\n"
-      "ld1 { v24.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v23.b }[10], [x23]\n"
-      "ld1 { v18.b }[10], [x22]\n"
-      "ld1 { v20.b }[10], [x21]\n"
-      "ld1 { v16.b }[10], [x20]\n"
+      "ld1 { v26.b }[10], [x28]\n"
+      "ld1 { v21.b }[10], [x27]\n"
+      "ld1 { v25.b }[10], [x26]\n"
+      "ld1 { v24.b }[10], [x25]\n"
+      "ld1 { v23.b }[10], [x24]\n"
+      "ld1 { v22.b }[10], [x23]\n"
+      "ld1 { v20.b }[10], [x22]\n"
+      "ld1 { v19.b }[10], [x21]\n"
       "b 11f\n"
       "6:"  // odd_loads_1_8
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[8], [x27]\n"
-      "ld1 { v24.b }[8], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v23.b }[8], [x23]\n"
-      "ld1 { v18.b }[8], [x22]\n"
-      "ld1 { v20.b }[8], [x21]\n"
-      "ld1 { v16.b }[8], [x20]\n"
+      "ld1 { v26.b }[8], [x28]\n"
+      "ld1 { v21.b }[8], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.b }[8], [x26]\n"
+      "ld1 { v24.b }[8], [x25]\n"
+      "ld1 { v23.b }[8], [x24]\n"
+      "ld1 { v22.b }[8], [x23]\n"
+      "ld1 { v20.b }[8], [x22]\n"
+      "ld1 { v19.b }[8], [x21]\n"
       "b 11f\n"
       "7:"  // odd_loads_4_0
       "tbz %x[width], #2, 9f\n"
-      "ldr s27, [x27], #0x4\n"
-      "ldr s24, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s18, [x22], #0x4\n"
-      "ldr s20, [x21], #0x4\n"
-      "ldr s16, [x20], #0x4\n"
+      "ldr s26, [x28], #0x4\n"
+      "ldr s21, [x27], #0x4\n"
+      "ldr s25, [x26], #0x4\n"
+      "ldr s24, [x25], #0x4\n"
+      "ldr s23, [x24], #0x4\n"
+      "ldr s22, [x23], #0x4\n"
+      "ldr s20, [x22], #0x4\n"
+      "ldr s19, [x21], #0x4\n"
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v27.h }[2], [x27], #0x2\n"
-      "mov x19, #0x1\n"
-      "ld1 { v24.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v23.h }[2], [x23], #0x2\n"
-      "ld1 { v18.h }[2], [x22], #0x2\n"
-      "ld1 { v20.h }[2], [x21], #0x2\n"
-      "ld1 { v16.h }[2], [x20], #0x2\n"
+      "ld1 { v26.h }[2], [x28], #0x2\n"
+      "ld1 { v21.h }[2], [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ld1 { v25.h }[2], [x26], #0x2\n"
+      "ld1 { v24.h }[2], [x25], #0x2\n"
+      "ld1 { v23.h }[2], [x24], #0x2\n"
+      "ld1 { v22.h }[2], [x23], #0x2\n"
+      "ld1 { v20.h }[2], [x22], #0x2\n"
+      "ld1 { v19.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[6], [x27]\n"
-      "ld1 { v24.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v23.b }[6], [x23]\n"
-      "ld1 { v18.b }[6], [x22]\n"
-      "ld1 { v20.b }[6], [x21]\n"
-      "ld1 { v16.b }[6], [x20]\n"
+      "ld1 { v26.b }[6], [x28]\n"
+      "ld1 { v21.b }[6], [x27]\n"
+      "ld1 { v25.b }[6], [x26]\n"
+      "ld1 { v24.b }[6], [x25]\n"
+      "ld1 { v23.b }[6], [x24]\n"
+      "ld1 { v22.b }[6], [x23]\n"
+      "ld1 { v20.b }[6], [x22]\n"
+      "ld1 { v19.b }[6], [x21]\n"
       "b 11f\n"
       "8:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[4], [x27]\n"
-      "ld1 { v24.b }[4], [x26]\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v23.b }[4], [x23]\n"
-      "ld1 { v18.b }[4], [x22]\n"
-      "ld1 { v20.b }[4], [x21]\n"
-      "ld1 { v16.b }[4], [x20]\n"
+      "ld1 { v26.b }[4], [x28]\n"
+      "ld1 { v21.b }[4], [x27]\n"
+      "ld1 { v25.b }[4], [x26]\n"
+      "ld1 { v24.b }[4], [x25]\n"
+      "ld1 { v23.b }[4], [x24]\n"
+      "ld1 { v22.b }[4], [x23]\n"
+      "ld1 { v20.b }[4], [x22]\n"
+      "ld1 { v19.b }[4], [x21]\n"
       "b 11f\n"
       "9:"  // odd_loads_2_0
       "tbz %x[width], #1, 10f\n"
-      "ldr h27, [x27], #0x2\n"
-      "ldr h24, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h23, [x23], #0x2\n"
-      "ldr h18, [x22], #0x2\n"
-      "ldr h20, [x21], #0x2\n"
-      "ldr h16, [x20], #0x2\n"
+      "ldr h26, [x28], #0x2\n"
+      "ldr h21, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h25, [x26], #0x2\n"
+      "ldr h24, [x25], #0x2\n"
+      "ldr h23, [x24], #0x2\n"
+      "ldr h22, [x23], #0x2\n"
+      "ldr h20, [x22], #0x2\n"
+      "ldr h19, [x21], #0x2\n"
       "tbz %x[width], #0, 11f\n"
-      "ld1 { v27.b }[2], [x27]\n"
-      "ld1 { v24.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v23.b }[2], [x23]\n"
-      "ld1 { v18.b }[2], [x22]\n"
-      "ld1 { v20.b }[2], [x21]\n"
-      "ld1 { v16.b }[2], [x20]\n"
+      "ld1 { v26.b }[2], [x28]\n"
+      "ld1 { v21.b }[2], [x27]\n"
+      "ld1 { v25.b }[2], [x26]\n"
+      "ld1 { v24.b }[2], [x25]\n"
+      "ld1 { v23.b }[2], [x24]\n"
+      "ld1 { v22.b }[2], [x23]\n"
+      "ld1 { v20.b }[2], [x22]\n"
+      "ld1 { v19.b }[2], [x21]\n"
       "b 11f\n"
       "10:"  // odd_loads_1_0
-      "ldr b27, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b24, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b23, [x23, #0x0]\n"
-      "ldr b18, [x22, #0x0]\n"
-      "ldr b20, [x21, #0x0]\n"
-      "ldr b16, [x20, #0x0]\n"
+      "ldr b26, [x28, #0x0]\n"
+      "ldr b21, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b25, [x26, #0x0]\n"
+      "ldr b24, [x25, #0x0]\n"
+      "ldr b23, [x24, #0x0]\n"
+      "ldr b22, [x23, #0x0]\n"
+      "ldr b20, [x22, #0x0]\n"
+      "ldr b19, [x21, #0x0]\n"
       "11:"  // Odd load end
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v22.2d, v25.2d, v21.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v19.2d, v23.2d, v18.2d\n"
-      "str q22, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v20.2d, v16.2d\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
+      "subs x20, x20, #0x1\n"
+      "zip1 v16.2d, v26.2d, v21.2d\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "zip1 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip1 v17.2d, v23.2d, v22.2d\n"
+      "zip1 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 12f\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "zip2 v18.2d, v23.2d, v18.2d\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.2d, v20.2d, v16.2d\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
+      "zip2 v21.2d, v26.2d, v21.2d\n"
+      "str q21, [%x[out_ptr], #0x0]\n"
+      "zip2 v18.2d, v25.2d, v24.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "zip2 v17.2d, v23.2d, v22.2d\n"
+      "zip2 v16.2d, v20.2d, v19.2d\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "12:"  // Odds skip
 
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8_summing.hpp
index 3550830..c6ad294 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,46 +31,47 @@
 )
 {
   __asm__ __volatile__(
-      "movi v5.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v4.8h, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v5.8h, #0x0\n"
+      "movi v4.8h, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "movi v3.8h, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "add x27, x27, %x[row_offset]\n"
       "movi v2.8h, #0x0\n"
-      "ldr x24, [%x[in], #0x18]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "movi v1.4s, #0x0\n"
-      "ldr x23, [%x[in], #0x20]\n"
-      "add x26, x26, %x[row_offset]\n"
       "movi v0.4s, #0x0\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "add x25, x25, %x[row_offset]\n"
       "movi v31.4s, #0x0\n"
-      "ldr x21, [%x[in], #0x30]\n"
-      "add x24, x24, %x[row_offset]\n"
       "movi v30.4s, #0x0\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "add x27, x27, %x[row_offset]\n"
+      "add x26, x26, %x[row_offset]\n"
+      "add x25, x25, %x[row_offset]\n"
+      "add x24, x24, %x[row_offset]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "movi v29.4s, #0x0\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
+      "movi v29.4s, #0x0\n"
       "movi v28.4s, #0x0\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -78,7 +79,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -86,7 +87,6 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
       "ld1 { v29.4s }, [%x[out_ptr]]\n"
@@ -95,266 +95,266 @@
       "cmp %x[width], #0x10\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0x3e\n"
+      "cmp x20, #0x3e\n"
       "ble 4f\n"
       "sadalp v1.4s, v5.8h\n"
       "movi v5.8h, #0x0\n"
+      "mov x20, #0x0\n"
       "sadalp v0.4s, v4.8h\n"
       "movi v4.8h, #0x0\n"
       "sadalp v31.4s, v3.8h\n"
       "movi v3.8h, #0x0\n"
       "sadalp v30.4s, v2.8h\n"
       "movi v2.8h, #0x0\n"
-      "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q27, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q24, [x26], #0x10\n"
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "ldr q25, [x25], #0x10\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "ldr q21, [x24], #0x10\n"
+      "ldr q27, [x28], #0x10\n"
+      "ldr q19, [x27], #0x10\n"
+      "zip1 v26.2d, v27.2d, v19.2d\n"
       "sadalp v5.8h, v26.16b\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
-      "ldr q22, [x23], #0x10\n"
+      "ldr q25, [x26], #0x10\n"
+      "ldr q18, [x25], #0x10\n"
+      "zip1 v24.2d, v25.2d, v18.2d\n"
+      "sadalp v4.8h, v24.16b\n"
+      "ldr q23, [x24], #0x10\n"
+      "ldr q17, [x23], #0x10\n"
+      "zip1 v22.2d, v23.2d, v17.2d\n"
+      "sadalp v3.8h, v22.16b\n"
+      "ldr q21, [x22], #0x10\n"
+      "ldr q16, [x21], #0x10\n"
+      "zip1 v20.2d, v21.2d, v16.2d\n"
+      "sadalp v2.8h, v20.16b\n"
+      "zip2 v19.2d, v27.2d, v19.2d\n"
+      "zip2 v18.2d, v25.2d, v18.2d\n"
+      "subs %x[width], %x[width], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "ldr q18, [x22], #0x10\n"
-      "sadalp v4.8h, v23.16b\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "ldr q19, [x21], #0x10\n"
-      "sadalp v5.8h, v24.16b\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "ldr q16, [x20], #0x10\n"
-      "sadalp v3.8h, v20.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
+      "zip2 v17.2d, v23.2d, v17.2d\n"
+      "zip2 v16.2d, v21.2d, v16.2d\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "sadalp v4.8h, v21.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "sadalp v2.8h, v17.16b\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "sadalp v3.8h, v18.16b\n"
+      "str q26, [%x[out_ptr], #0x0]\n"
+      "sadalp v5.8h, v19.16b\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "sadalp v2.8h, v16.16b\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "sadalp v4.8h, v18.16b\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "str q20, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "str q24, [%x[out_ptr], #0x40]\n"
-      "str q21, [%x[out_ptr], #0x50]\n"
-      "str q18, [%x[out_ptr], #0x60]\n"
+      "str q22, [%x[out_ptr], #0x20]\n"
+      "sadalp v3.8h, v17.16b\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
+      "sadalp v2.8h, v16.16b\n"
+      "add x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d27, [x27], #0x8\n"
-      "ldr d24, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d22, [x23], #0x8\n"
-      "ldr d18, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d16, [x20], #0x8\n"
+      "ldr d27, [x28], #0x8\n"
+      "ldr d19, [x27], #0x8\n"
+      "ldr d25, [x26], #0x8\n"
+      "ldr d18, [x25], #0x8\n"
+      "ldr d23, [x24], #0x8\n"
+      "ldr d17, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d16, [x21], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v27.s }[2], [x27], #0x4\n"
-      "ld1 { v24.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v22.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v16.s }[2], [x20], #0x4\n"
+      "ld1 { v27.s }[2], [x28], #0x4\n"
+      "ld1 { v19.s }[2], [x27], #0x4\n"
+      "ld1 { v25.s }[2], [x26], #0x4\n"
+      "ld1 { v18.s }[2], [x25], #0x4\n"
+      "ld1 { v23.s }[2], [x24], #0x4\n"
+      "ld1 { v17.s }[2], [x23], #0x4\n"
+      "ld1 { v21.s }[2], [x22], #0x4\n"
+      "ld1 { v16.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v27.h }[6], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v24.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v22.h }[6], [x23], #0x2\n"
-      "ld1 { v18.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v16.h }[6], [x20], #0x2\n"
+      "ld1 { v27.h }[6], [x28], #0x2\n"
+      "ld1 { v19.h }[6], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[6], [x26], #0x2\n"
+      "ld1 { v18.h }[6], [x25], #0x2\n"
+      "ld1 { v23.h }[6], [x24], #0x2\n"
+      "ld1 { v17.h }[6], [x23], #0x2\n"
+      "ld1 { v21.h }[6], [x22], #0x2\n"
+      "ld1 { v16.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[14], [x27]\n"
-      "ld1 { v24.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v22.b }[14], [x23]\n"
-      "ld1 { v18.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v16.b }[14], [x20]\n"
+      "ld1 { v27.b }[14], [x28]\n"
+      "ld1 { v19.b }[14], [x27]\n"
+      "ld1 { v25.b }[14], [x26]\n"
+      "ld1 { v18.b }[14], [x25]\n"
+      "ld1 { v23.b }[14], [x24]\n"
+      "ld1 { v17.b }[14], [x23]\n"
+      "ld1 { v21.b }[14], [x22]\n"
+      "ld1 { v16.b }[14], [x21]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[12], [x27]\n"
-      "ld1 { v24.b }[12], [x26]\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v22.b }[12], [x23]\n"
-      "ld1 { v18.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v16.b }[12], [x20]\n"
+      "ld1 { v27.b }[12], [x28]\n"
+      "ld1 { v19.b }[12], [x27]\n"
+      "ld1 { v25.b }[12], [x26]\n"
+      "ld1 { v18.b }[12], [x25]\n"
+      "ld1 { v23.b }[12], [x24]\n"
+      "ld1 { v17.b }[12], [x23]\n"
+      "ld1 { v21.b }[12], [x22]\n"
+      "ld1 { v16.b }[12], [x21]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v27.h }[4], [x27], #0x2\n"
-      "ld1 { v24.h }[4], [x26], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v22.h }[4], [x23], #0x2\n"
-      "ld1 { v18.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v16.h }[4], [x20], #0x2\n"
+      "ld1 { v27.h }[4], [x28], #0x2\n"
+      "ld1 { v19.h }[4], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[4], [x26], #0x2\n"
+      "ld1 { v18.h }[4], [x25], #0x2\n"
+      "ld1 { v23.h }[4], [x24], #0x2\n"
+      "ld1 { v17.h }[4], [x23], #0x2\n"
+      "ld1 { v21.h }[4], [x22], #0x2\n"
+      "ld1 { v16.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[10], [x27]\n"
-      "ld1 { v24.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v22.b }[10], [x23]\n"
-      "ld1 { v18.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v16.b }[10], [x20]\n"
+      "ld1 { v27.b }[10], [x28]\n"
+      "ld1 { v19.b }[10], [x27]\n"
+      "ld1 { v25.b }[10], [x26]\n"
+      "ld1 { v18.b }[10], [x25]\n"
+      "ld1 { v23.b }[10], [x24]\n"
+      "ld1 { v17.b }[10], [x23]\n"
+      "ld1 { v21.b }[10], [x22]\n"
+      "ld1 { v16.b }[10], [x21]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[8], [x27]\n"
-      "ld1 { v24.b }[8], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v22.b }[8], [x23]\n"
-      "ld1 { v18.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v16.b }[8], [x20]\n"
+      "ld1 { v27.b }[8], [x28]\n"
+      "ld1 { v19.b }[8], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.b }[8], [x26]\n"
+      "ld1 { v18.b }[8], [x25]\n"
+      "ld1 { v23.b }[8], [x24]\n"
+      "ld1 { v17.b }[8], [x23]\n"
+      "ld1 { v21.b }[8], [x22]\n"
+      "ld1 { v16.b }[8], [x21]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s27, [x27], #0x4\n"
-      "ldr s24, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s22, [x23], #0x4\n"
-      "ldr s18, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s16, [x20], #0x4\n"
+      "ldr s27, [x28], #0x4\n"
+      "ldr s19, [x27], #0x4\n"
+      "ldr s25, [x26], #0x4\n"
+      "ldr s18, [x25], #0x4\n"
+      "ldr s23, [x24], #0x4\n"
+      "ldr s17, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s16, [x21], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v27.h }[2], [x27], #0x2\n"
-      "mov x19, #0x1\n"
-      "ld1 { v24.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v22.h }[2], [x23], #0x2\n"
-      "ld1 { v18.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v16.h }[2], [x20], #0x2\n"
+      "ld1 { v27.h }[2], [x28], #0x2\n"
+      "ld1 { v19.h }[2], [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ld1 { v25.h }[2], [x26], #0x2\n"
+      "ld1 { v18.h }[2], [x25], #0x2\n"
+      "ld1 { v23.h }[2], [x24], #0x2\n"
+      "ld1 { v17.h }[2], [x23], #0x2\n"
+      "ld1 { v21.h }[2], [x22], #0x2\n"
+      "ld1 { v16.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[6], [x27]\n"
-      "ld1 { v24.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v22.b }[6], [x23]\n"
-      "ld1 { v18.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v16.b }[6], [x20]\n"
+      "ld1 { v27.b }[6], [x28]\n"
+      "ld1 { v19.b }[6], [x27]\n"
+      "ld1 { v25.b }[6], [x26]\n"
+      "ld1 { v18.b }[6], [x25]\n"
+      "ld1 { v23.b }[6], [x24]\n"
+      "ld1 { v17.b }[6], [x23]\n"
+      "ld1 { v21.b }[6], [x22]\n"
+      "ld1 { v16.b }[6], [x21]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[4], [x27]\n"
-      "ld1 { v24.b }[4], [x26]\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v22.b }[4], [x23]\n"
-      "ld1 { v18.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v16.b }[4], [x20]\n"
+      "ld1 { v27.b }[4], [x28]\n"
+      "ld1 { v19.b }[4], [x27]\n"
+      "ld1 { v25.b }[4], [x26]\n"
+      "ld1 { v18.b }[4], [x25]\n"
+      "ld1 { v23.b }[4], [x24]\n"
+      "ld1 { v17.b }[4], [x23]\n"
+      "ld1 { v21.b }[4], [x22]\n"
+      "ld1 { v16.b }[4], [x21]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h27, [x27], #0x2\n"
-      "ldr h24, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h22, [x23], #0x2\n"
-      "ldr h18, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h16, [x20], #0x2\n"
+      "ldr h27, [x28], #0x2\n"
+      "ldr h19, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h25, [x26], #0x2\n"
+      "ldr h18, [x25], #0x2\n"
+      "ldr h23, [x24], #0x2\n"
+      "ldr h17, [x23], #0x2\n"
+      "ldr h21, [x22], #0x2\n"
+      "ldr h16, [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[2], [x27]\n"
-      "ld1 { v24.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v22.b }[2], [x23]\n"
-      "ld1 { v18.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v16.b }[2], [x20]\n"
+      "ld1 { v27.b }[2], [x28]\n"
+      "ld1 { v19.b }[2], [x27]\n"
+      "ld1 { v25.b }[2], [x26]\n"
+      "ld1 { v18.b }[2], [x25]\n"
+      "ld1 { v23.b }[2], [x24]\n"
+      "ld1 { v17.b }[2], [x23]\n"
+      "ld1 { v21.b }[2], [x22]\n"
+      "ld1 { v16.b }[2], [x21]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b27, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b24, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b22, [x23, #0x0]\n"
-      "ldr b18, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b16, [x20, #0x0]\n"
+      "ldr b27, [x28, #0x0]\n"
+      "ldr b19, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b25, [x26, #0x0]\n"
+      "ldr b18, [x25, #0x0]\n"
+      "ldr b23, [x24, #0x0]\n"
+      "ldr b17, [x23, #0x0]\n"
+      "ldr b21, [x22, #0x0]\n"
+      "ldr b16, [x21, #0x0]\n"
       "13:"  // Odd load end
-      "zip1 v26.2d, v27.2d, v24.2d\n"
+      "zip1 v26.2d, v27.2d, v19.2d\n"
+      "zip1 v24.2d, v25.2d, v18.2d\n"
+      "subs x20, x20, #0x1\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
+      "zip1 v22.2d, v23.2d, v17.2d\n"
+      "zip1 v20.2d, v21.2d, v16.2d\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
       "sadalp v5.8h, v26.16b\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "sadalp v4.8h, v23.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
-      "str q20, [%x[out_ptr], #0x20]\n"
-      "sadalp v3.8h, v20.16b\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "sadalp v2.8h, v17.16b\n"
-      "subs x19, x19, #0x1\n"
+      "sadalp v4.8h, v24.16b\n"
+      "str q22, [%x[out_ptr], #0x20]\n"
+      "sadalp v3.8h, v22.16b\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
+      "sadalp v2.8h, v20.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 14f\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "sadalp v5.8h, v24.16b\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "sadalp v4.8h, v21.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
-      "sadalp v3.8h, v18.16b\n"
+      "zip2 v19.2d, v27.2d, v19.2d\n"
+      "zip2 v18.2d, v25.2d, v18.2d\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "zip2 v17.2d, v23.2d, v17.2d\n"
+      "zip2 v16.2d, v21.2d, v16.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "sadalp v5.8h, v19.16b\n"
+      "sadalp v4.8h, v18.16b\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "sadalp v3.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "sadalp v2.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "14:"  // Odds skip
       "sadalp v1.4s, v5.8h\n"
       "sadalp v0.4s, v4.8h\n"
-      "addp v1.4s, v1.4s, v0.4s\n"
       "sadalp v31.4s, v3.8h\n"
       "sadalp v30.4s, v2.8h\n"
-      "add v1.4s, v1.4s, v29.4s\n"
-      "str q1, [%x[out_ptr], #0x0]\n"
+      "addp v1.4s, v1.4s, v0.4s\n"
       "addp v0.4s, v31.4s, v30.4s\n"
+      "add v1.4s, v1.4s, v29.4s\n"
       "add v0.4s, v0.4s, v28.4s\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
       "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_u8_u8_summing.hpp
index 454260e..6c4a5fa 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block8_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #ifdef __aarch64__
@@ -31,46 +31,47 @@
 )
 {
   __asm__ __volatile__(
-      "movi v5.8h, #0x0\n"
-      "ldr x27, [%x[in], #0x0]\n"
-      "mov x19, #0x0\n"
-      "movi v4.8h, #0x0\n"
-      "ldr x26, [%x[in], #0x8]\n"
+      "ldr x28, [%x[in], #0x0]\n"
+      "ldr x27, [%x[in], #0x8]\n"
       "cmp %x[height], #0x8\n"
+      "mov x20, #0x0\n"
+      "ldr x26, [%x[in], #0x10]\n"
+      "ldr x25, [%x[in], #0x18]\n"
+      "movi v5.8h, #0x0\n"
+      "movi v4.8h, #0x0\n"
+      "ldr x24, [%x[in], #0x20]\n"
+      "ldr x23, [%x[in], #0x28]\n"
       "movi v3.8h, #0x0\n"
-      "ldr x25, [%x[in], #0x10]\n"
-      "add x27, x27, %x[row_offset]\n"
       "movi v2.8h, #0x0\n"
-      "ldr x24, [%x[in], #0x18]\n"
+      "ldr x22, [%x[in], #0x30]\n"
+      "ldr x21, [%x[in], #0x38]\n"
       "movi v1.4s, #0x0\n"
-      "ldr x23, [%x[in], #0x20]\n"
-      "add x26, x26, %x[row_offset]\n"
       "movi v0.4s, #0x0\n"
-      "ldr x22, [%x[in], #0x28]\n"
-      "add x25, x25, %x[row_offset]\n"
       "movi v31.4s, #0x0\n"
-      "ldr x21, [%x[in], #0x30]\n"
-      "add x24, x24, %x[row_offset]\n"
       "movi v30.4s, #0x0\n"
-      "ldr x20, [%x[in], #0x38]\n"
+      "add x28, x28, %x[row_offset]\n"
+      "add x27, x27, %x[row_offset]\n"
+      "add x26, x26, %x[row_offset]\n"
+      "add x25, x25, %x[row_offset]\n"
+      "add x24, x24, %x[row_offset]\n"
       "add x23, x23, %x[row_offset]\n"
       "add x22, x22, %x[row_offset]\n"
       "add x21, x21, %x[row_offset]\n"
-      "add x20, x20, %x[row_offset]\n"
       "beq 1f\n"
-      "mov x20, x27\n"
       "cmp %x[height], #0x2\n"
-      "csel x26, x26, x27, GE\n"
-      "csel x25, x25, x27, GT\n"
+      "csel x27, x27, x28, GE\n"
+      "csel x26, x26, x28, GT\n"
       "cmp %x[height], #0x4\n"
-      "csel x24, x24, x27, GE\n"
-      "csel x23, x23, x27, GT\n"
+      "csel x25, x25, x28, GE\n"
+      "csel x24, x24, x28, GT\n"
       "cmp %x[height], #0x6\n"
-      "csel x22, x22, x27, GE\n"
-      "csel x21, x21, x27, GT\n"
+      "mov x21, x28\n"
+      "csel x23, x23, x28, GE\n"
+      "csel x22, x22, x28, GT\n"
       "1:"  // no_pointer_adj
-      "movi v29.4s, #0x0\n"
+      "prfm pldl1keep, [x28, #0x0]\n"
       "prfm pldl1keep, [x27, #0x0]\n"
+      "movi v29.4s, #0x0\n"
       "movi v28.4s, #0x0\n"
       "prfm pldl1keep, [x26, #0x0]\n"
       "prfm pldl1keep, [x25, #0x0]\n"
@@ -78,7 +79,7 @@
       "prfm pldl1keep, [x23, #0x0]\n"
       "prfm pldl1keep, [x22, #0x0]\n"
       "prfm pldl1keep, [x21, #0x0]\n"
-      "prfm pldl1keep, [x20, #0x0]\n"
+      "prfm pldl1keep, [x28, #0x40]\n"
       "prfm pldl1keep, [x27, #0x40]\n"
       "prfm pldl1keep, [x26, #0x40]\n"
       "prfm pldl1keep, [x25, #0x40]\n"
@@ -86,7 +87,6 @@
       "prfm pldl1keep, [x23, #0x40]\n"
       "prfm pldl1keep, [x22, #0x40]\n"
       "prfm pldl1keep, [x21, #0x40]\n"
-      "prfm pldl1keep, [x20, #0x40]\n"
       "cbnz %w[first], 2f\n"
       "sub %x[out_ptr], %x[out_ptr], #0x20\n"
       "ld1 { v29.4s }, [%x[out_ptr]]\n"
@@ -95,266 +95,266 @@
       "cmp %x[width], #0x10\n"
       "blt 5f\n"
       "3:"  // Main loop head
-      "cmp x19, #0x3e\n"
+      "cmp x20, #0x3e\n"
       "ble 4f\n"
       "uadalp v1.4s, v5.8h\n"
       "movi v5.8h, #0x0\n"
+      "mov x20, #0x0\n"
       "uadalp v0.4s, v4.8h\n"
       "movi v4.8h, #0x0\n"
       "uadalp v31.4s, v3.8h\n"
       "movi v3.8h, #0x0\n"
       "uadalp v30.4s, v2.8h\n"
       "movi v2.8h, #0x0\n"
-      "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q27, [x27], #0x10\n"
-      "add x19, x19, #0x1\n"
-      "ldr q24, [x26], #0x10\n"
-      "zip1 v26.2d, v27.2d, v24.2d\n"
-      "ldr q25, [x25], #0x10\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "ldr q21, [x24], #0x10\n"
+      "ldr q27, [x28], #0x10\n"
+      "ldr q19, [x27], #0x10\n"
+      "zip1 v26.2d, v27.2d, v19.2d\n"
       "uadalp v5.8h, v26.16b\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
-      "ldr q22, [x23], #0x10\n"
+      "ldr q25, [x26], #0x10\n"
+      "ldr q18, [x25], #0x10\n"
+      "zip1 v24.2d, v25.2d, v18.2d\n"
+      "uadalp v4.8h, v24.16b\n"
+      "ldr q23, [x24], #0x10\n"
+      "ldr q17, [x23], #0x10\n"
+      "zip1 v22.2d, v23.2d, v17.2d\n"
+      "uadalp v3.8h, v22.16b\n"
+      "ldr q21, [x22], #0x10\n"
+      "ldr q16, [x21], #0x10\n"
+      "zip1 v20.2d, v21.2d, v16.2d\n"
+      "uadalp v2.8h, v20.16b\n"
+      "zip2 v19.2d, v27.2d, v19.2d\n"
+      "zip2 v18.2d, v25.2d, v18.2d\n"
+      "subs %x[width], %x[width], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "ldr q18, [x22], #0x10\n"
-      "uadalp v4.8h, v23.16b\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "ldr q19, [x21], #0x10\n"
-      "uadalp v5.8h, v24.16b\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "ldr q16, [x20], #0x10\n"
-      "uadalp v3.8h, v20.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
+      "zip2 v17.2d, v23.2d, v17.2d\n"
+      "zip2 v16.2d, v21.2d, v16.2d\n"
+      "prfm pldl1keep, [x28, #0x70]\n"
       "prfm pldl1keep, [x27, #0x70]\n"
-      "uadalp v4.8h, v21.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
       "prfm pldl1keep, [x26, #0x70]\n"
-      "uadalp v2.8h, v17.16b\n"
       "prfm pldl1keep, [x25, #0x70]\n"
-      "uadalp v3.8h, v18.16b\n"
+      "str q26, [%x[out_ptr], #0x0]\n"
+      "uadalp v5.8h, v19.16b\n"
       "prfm pldl1keep, [x24, #0x70]\n"
-      "uadalp v2.8h, v16.16b\n"
       "prfm pldl1keep, [x23, #0x70]\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
+      "uadalp v4.8h, v18.16b\n"
       "prfm pldl1keep, [x22, #0x70]\n"
       "prfm pldl1keep, [x21, #0x70]\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q26, [%x[out_ptr], #0x0]\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "str q20, [%x[out_ptr], #0x20]\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "str q24, [%x[out_ptr], #0x40]\n"
-      "str q21, [%x[out_ptr], #0x50]\n"
-      "str q18, [%x[out_ptr], #0x60]\n"
+      "str q22, [%x[out_ptr], #0x20]\n"
+      "uadalp v3.8h, v17.16b\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
+      "uadalp v2.8h, v16.16b\n"
+      "add x20, x20, #0x1\n"
+      "str q19, [%x[out_ptr], #0x40]\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 3b\n"
       "5:"  // Main loop skip
       "cbz %x[width], 14f\n"
       "tbz %x[width], #3, 9f\n"
-      "ldr d27, [x27], #0x8\n"
-      "ldr d24, [x26], #0x8\n"
-      "ldr d25, [x25], #0x8\n"
-      "ldr d21, [x24], #0x8\n"
-      "ldr d22, [x23], #0x8\n"
-      "ldr d18, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d16, [x20], #0x8\n"
+      "ldr d27, [x28], #0x8\n"
+      "ldr d19, [x27], #0x8\n"
+      "ldr d25, [x26], #0x8\n"
+      "ldr d18, [x25], #0x8\n"
+      "ldr d23, [x24], #0x8\n"
+      "ldr d17, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d16, [x21], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v27.s }[2], [x27], #0x4\n"
-      "ld1 { v24.s }[2], [x26], #0x4\n"
-      "ld1 { v25.s }[2], [x25], #0x4\n"
-      "ld1 { v21.s }[2], [x24], #0x4\n"
-      "ld1 { v22.s }[2], [x23], #0x4\n"
-      "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v16.s }[2], [x20], #0x4\n"
+      "ld1 { v27.s }[2], [x28], #0x4\n"
+      "ld1 { v19.s }[2], [x27], #0x4\n"
+      "ld1 { v25.s }[2], [x26], #0x4\n"
+      "ld1 { v18.s }[2], [x25], #0x4\n"
+      "ld1 { v23.s }[2], [x24], #0x4\n"
+      "ld1 { v17.s }[2], [x23], #0x4\n"
+      "ld1 { v21.s }[2], [x22], #0x4\n"
+      "ld1 { v16.s }[2], [x21], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v27.h }[6], [x27], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v24.h }[6], [x26], #0x2\n"
-      "ld1 { v25.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v22.h }[6], [x23], #0x2\n"
-      "ld1 { v18.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v16.h }[6], [x20], #0x2\n"
+      "ld1 { v27.h }[6], [x28], #0x2\n"
+      "ld1 { v19.h }[6], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[6], [x26], #0x2\n"
+      "ld1 { v18.h }[6], [x25], #0x2\n"
+      "ld1 { v23.h }[6], [x24], #0x2\n"
+      "ld1 { v17.h }[6], [x23], #0x2\n"
+      "ld1 { v21.h }[6], [x22], #0x2\n"
+      "ld1 { v16.h }[6], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[14], [x27]\n"
-      "ld1 { v24.b }[14], [x26]\n"
-      "ld1 { v25.b }[14], [x25]\n"
-      "ld1 { v21.b }[14], [x24]\n"
-      "ld1 { v22.b }[14], [x23]\n"
-      "ld1 { v18.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v16.b }[14], [x20]\n"
+      "ld1 { v27.b }[14], [x28]\n"
+      "ld1 { v19.b }[14], [x27]\n"
+      "ld1 { v25.b }[14], [x26]\n"
+      "ld1 { v18.b }[14], [x25]\n"
+      "ld1 { v23.b }[14], [x24]\n"
+      "ld1 { v17.b }[14], [x23]\n"
+      "ld1 { v21.b }[14], [x22]\n"
+      "ld1 { v16.b }[14], [x21]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
-      "mov x19, #0x2\n"
+      "mov x20, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[12], [x27]\n"
-      "ld1 { v24.b }[12], [x26]\n"
-      "ld1 { v25.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v22.b }[12], [x23]\n"
-      "ld1 { v18.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v16.b }[12], [x20]\n"
+      "ld1 { v27.b }[12], [x28]\n"
+      "ld1 { v19.b }[12], [x27]\n"
+      "ld1 { v25.b }[12], [x26]\n"
+      "ld1 { v18.b }[12], [x25]\n"
+      "ld1 { v23.b }[12], [x24]\n"
+      "ld1 { v17.b }[12], [x23]\n"
+      "ld1 { v21.b }[12], [x22]\n"
+      "ld1 { v16.b }[12], [x21]\n"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v27.h }[4], [x27], #0x2\n"
-      "ld1 { v24.h }[4], [x26], #0x2\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v22.h }[4], [x23], #0x2\n"
-      "ld1 { v18.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v16.h }[4], [x20], #0x2\n"
+      "ld1 { v27.h }[4], [x28], #0x2\n"
+      "ld1 { v19.h }[4], [x27], #0x2\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.h }[4], [x26], #0x2\n"
+      "ld1 { v18.h }[4], [x25], #0x2\n"
+      "ld1 { v23.h }[4], [x24], #0x2\n"
+      "ld1 { v17.h }[4], [x23], #0x2\n"
+      "ld1 { v21.h }[4], [x22], #0x2\n"
+      "ld1 { v16.h }[4], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[10], [x27]\n"
-      "ld1 { v24.b }[10], [x26]\n"
-      "ld1 { v25.b }[10], [x25]\n"
-      "ld1 { v21.b }[10], [x24]\n"
-      "ld1 { v22.b }[10], [x23]\n"
-      "ld1 { v18.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v16.b }[10], [x20]\n"
+      "ld1 { v27.b }[10], [x28]\n"
+      "ld1 { v19.b }[10], [x27]\n"
+      "ld1 { v25.b }[10], [x26]\n"
+      "ld1 { v18.b }[10], [x25]\n"
+      "ld1 { v23.b }[10], [x24]\n"
+      "ld1 { v17.b }[10], [x23]\n"
+      "ld1 { v21.b }[10], [x22]\n"
+      "ld1 { v16.b }[10], [x21]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[8], [x27]\n"
-      "ld1 { v24.b }[8], [x26]\n"
-      "mov x19, #0x2\n"
-      "ld1 { v25.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v22.b }[8], [x23]\n"
-      "ld1 { v18.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v16.b }[8], [x20]\n"
+      "ld1 { v27.b }[8], [x28]\n"
+      "ld1 { v19.b }[8], [x27]\n"
+      "mov x20, #0x2\n"
+      "ld1 { v25.b }[8], [x26]\n"
+      "ld1 { v18.b }[8], [x25]\n"
+      "ld1 { v23.b }[8], [x24]\n"
+      "ld1 { v17.b }[8], [x23]\n"
+      "ld1 { v21.b }[8], [x22]\n"
+      "ld1 { v16.b }[8], [x21]\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s27, [x27], #0x4\n"
-      "ldr s24, [x26], #0x4\n"
-      "ldr s25, [x25], #0x4\n"
-      "ldr s21, [x24], #0x4\n"
-      "ldr s22, [x23], #0x4\n"
-      "ldr s18, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s16, [x20], #0x4\n"
+      "ldr s27, [x28], #0x4\n"
+      "ldr s19, [x27], #0x4\n"
+      "ldr s25, [x26], #0x4\n"
+      "ldr s18, [x25], #0x4\n"
+      "ldr s23, [x24], #0x4\n"
+      "ldr s17, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s16, [x21], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v27.h }[2], [x27], #0x2\n"
-      "mov x19, #0x1\n"
-      "ld1 { v24.h }[2], [x26], #0x2\n"
-      "ld1 { v25.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v22.h }[2], [x23], #0x2\n"
-      "ld1 { v18.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v16.h }[2], [x20], #0x2\n"
+      "ld1 { v27.h }[2], [x28], #0x2\n"
+      "ld1 { v19.h }[2], [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ld1 { v25.h }[2], [x26], #0x2\n"
+      "ld1 { v18.h }[2], [x25], #0x2\n"
+      "ld1 { v23.h }[2], [x24], #0x2\n"
+      "ld1 { v17.h }[2], [x23], #0x2\n"
+      "ld1 { v21.h }[2], [x22], #0x2\n"
+      "ld1 { v16.h }[2], [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[6], [x27]\n"
-      "ld1 { v24.b }[6], [x26]\n"
-      "ld1 { v25.b }[6], [x25]\n"
-      "ld1 { v21.b }[6], [x24]\n"
-      "ld1 { v22.b }[6], [x23]\n"
-      "ld1 { v18.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v16.b }[6], [x20]\n"
+      "ld1 { v27.b }[6], [x28]\n"
+      "ld1 { v19.b }[6], [x27]\n"
+      "ld1 { v25.b }[6], [x26]\n"
+      "ld1 { v18.b }[6], [x25]\n"
+      "ld1 { v23.b }[6], [x24]\n"
+      "ld1 { v17.b }[6], [x23]\n"
+      "ld1 { v21.b }[6], [x22]\n"
+      "ld1 { v16.b }[6], [x21]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
-      "mov x19, #0x1\n"
+      "mov x20, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[4], [x27]\n"
-      "ld1 { v24.b }[4], [x26]\n"
-      "ld1 { v25.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v22.b }[4], [x23]\n"
-      "ld1 { v18.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v16.b }[4], [x20]\n"
+      "ld1 { v27.b }[4], [x28]\n"
+      "ld1 { v19.b }[4], [x27]\n"
+      "ld1 { v25.b }[4], [x26]\n"
+      "ld1 { v18.b }[4], [x25]\n"
+      "ld1 { v23.b }[4], [x24]\n"
+      "ld1 { v17.b }[4], [x23]\n"
+      "ld1 { v21.b }[4], [x22]\n"
+      "ld1 { v16.b }[4], [x21]\n"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h27, [x27], #0x2\n"
-      "ldr h24, [x26], #0x2\n"
-      "mov x19, #0x1\n"
-      "ldr h25, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h22, [x23], #0x2\n"
-      "ldr h18, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h16, [x20], #0x2\n"
+      "ldr h27, [x28], #0x2\n"
+      "ldr h19, [x27], #0x2\n"
+      "mov x20, #0x1\n"
+      "ldr h25, [x26], #0x2\n"
+      "ldr h18, [x25], #0x2\n"
+      "ldr h23, [x24], #0x2\n"
+      "ldr h17, [x23], #0x2\n"
+      "ldr h21, [x22], #0x2\n"
+      "ldr h16, [x21], #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v27.b }[2], [x27]\n"
-      "ld1 { v24.b }[2], [x26]\n"
-      "ld1 { v25.b }[2], [x25]\n"
-      "ld1 { v21.b }[2], [x24]\n"
-      "ld1 { v22.b }[2], [x23]\n"
-      "ld1 { v18.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v16.b }[2], [x20]\n"
+      "ld1 { v27.b }[2], [x28]\n"
+      "ld1 { v19.b }[2], [x27]\n"
+      "ld1 { v25.b }[2], [x26]\n"
+      "ld1 { v18.b }[2], [x25]\n"
+      "ld1 { v23.b }[2], [x24]\n"
+      "ld1 { v17.b }[2], [x23]\n"
+      "ld1 { v21.b }[2], [x22]\n"
+      "ld1 { v16.b }[2], [x21]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b27, [x27, #0x0]\n"
-      "mov x19, #0x1\n"
-      "ldr b24, [x26, #0x0]\n"
-      "ldr b25, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b22, [x23, #0x0]\n"
-      "ldr b18, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b16, [x20, #0x0]\n"
+      "ldr b27, [x28, #0x0]\n"
+      "ldr b19, [x27, #0x0]\n"
+      "mov x20, #0x1\n"
+      "ldr b25, [x26, #0x0]\n"
+      "ldr b18, [x25, #0x0]\n"
+      "ldr b23, [x24, #0x0]\n"
+      "ldr b17, [x23, #0x0]\n"
+      "ldr b21, [x22, #0x0]\n"
+      "ldr b16, [x21, #0x0]\n"
       "13:"  // Odd load end
-      "zip1 v26.2d, v27.2d, v24.2d\n"
+      "zip1 v26.2d, v27.2d, v19.2d\n"
+      "zip1 v24.2d, v25.2d, v18.2d\n"
+      "subs x20, x20, #0x1\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
+      "zip1 v22.2d, v23.2d, v17.2d\n"
+      "zip1 v20.2d, v21.2d, v16.2d\n"
+      "str q24, [%x[out_ptr], #0x10]\n"
       "uadalp v5.8h, v26.16b\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "uadalp v4.8h, v23.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
-      "str q20, [%x[out_ptr], #0x20]\n"
-      "uadalp v3.8h, v20.16b\n"
-      "str q17, [%x[out_ptr], #0x30]\n"
-      "uadalp v2.8h, v17.16b\n"
-      "subs x19, x19, #0x1\n"
+      "uadalp v4.8h, v24.16b\n"
+      "str q22, [%x[out_ptr], #0x20]\n"
+      "uadalp v3.8h, v22.16b\n"
+      "str q20, [%x[out_ptr], #0x30]\n"
+      "uadalp v2.8h, v20.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 14f\n"
-      "zip2 v24.2d, v27.2d, v24.2d\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "uadalp v5.8h, v24.16b\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "str q21, [%x[out_ptr], #0x10]\n"
-      "uadalp v4.8h, v21.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "str q18, [%x[out_ptr], #0x20]\n"
-      "uadalp v3.8h, v18.16b\n"
+      "zip2 v19.2d, v27.2d, v19.2d\n"
+      "zip2 v18.2d, v25.2d, v18.2d\n"
+      "str q19, [%x[out_ptr], #0x0]\n"
+      "zip2 v17.2d, v23.2d, v17.2d\n"
+      "zip2 v16.2d, v21.2d, v16.2d\n"
+      "str q18, [%x[out_ptr], #0x10]\n"
+      "uadalp v5.8h, v19.16b\n"
+      "uadalp v4.8h, v18.16b\n"
+      "str q17, [%x[out_ptr], #0x20]\n"
+      "uadalp v3.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x30]\n"
       "uadalp v2.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "14:"  // Odds skip
       "uadalp v1.4s, v5.8h\n"
       "uadalp v0.4s, v4.8h\n"
-      "addp v1.4s, v1.4s, v0.4s\n"
       "uadalp v31.4s, v3.8h\n"
       "uadalp v30.4s, v2.8h\n"
-      "add v1.4s, v1.4s, v29.4s\n"
-      "str q1, [%x[out_ptr], #0x0]\n"
+      "addp v1.4s, v1.4s, v0.4s\n"
       "addp v0.4s, v31.4s, v30.4s\n"
+      "add v1.4s, v1.4s, v29.4s\n"
       "add v0.4s, v0.4s, v28.4s\n"
+      "str q1, [%x[out_ptr], #0x0]\n"
       "str q0, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
-      : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+      : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave1VL_block2_fp32_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave1VL_block2_fp32_bf16.hpp
index c6ff375..51b91d1 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave1VL_block2_fp32_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave1VL_block2_fp32_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,105 +34,105 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x22, ALL, MUL #2\n"
+      "sub x28, %x[width], #0x1\n"
       "cntw x21, ALL, MUL #2\n"
-      "sub x27, %x[width], #0x1\n"
-      "cntw x20, ALL, MUL #2\n"
-      "sub x19, x21, #0x1\n"
+      "sub x20, x22, #0x1\n"
       "whilelt p10.s, XZR, %x[height]\n"
-      "add x27, x27, x20\n"
-      "ands x26, %x[width], x19\n"
-      "udiv x27, x27, x20\n"
-      "csel x26, x26, x21, NE\n"
-      "mov x25, #0x0\n"
-      "and x24, x27, #0x1\n"
-      "sub x27, x27, #0x1\n"
-      "add x26, x26, #0x1\n"
-      "mov x19, %x[width]\n"
+      "add x28, x28, x21\n"
+      "ands x27, %x[width], x20\n"
+      "udiv x28, x28, x21\n"
+      "csel x27, x27, x22, NE\n"
+      "mov x26, #0x0\n"
+      "and x25, x28, #0x1\n"
+      "sub x28, x28, #0x1\n"
+      "add x27, x27, #0x1\n"
+      "mov x20, %x[width]\n"
       "ptrue p0.b\n"
-      "mov x23, %x[outptr_raw]\n"
-      "mov x22, %x[row_offset]\n"
-      "cntw x21\n"
+      "mov x24, %x[outptr_raw]\n"
+      "mov x23, %x[row_offset]\n"
+      "cntw x22\n"
+      "lsr x28, x28, #0x1\n"
       "lsr x27, x27, #0x1\n"
-      "lsr x26, x26, #0x1\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34731  // whilelt pn9.s, x25, x19, VLx2\n"
-      "mov x20, %x[in]\n"
+      ".inst 0x25b44751  // whilelt pn9.s, x26, x20, VLx2\n"
+      "mov x21, %x[in]\n"
       "1:"  // Width loop: Preamble: Loop
-      "ldr x19, [x20], #0x8\n"
+      "ldr x20, [x21], #0x8\n"
       ".inst 0x25306548  // psel p8.s, p9.s/Z, p10.s[w12]\n"
-      ".inst 0xa0164266  // ld1w { z6.s-z7.s }, pn8.s/Z, [x19, x22, LSL #2]\n"
+      ".inst 0xa0174286  // ld1w { z6.s-z7.s }, pn8.s/Z, [x20, x23, LSL #2]\n"
       ".inst 0xc160e0c6  // bfcvt z6.h, { z6.s-z7.s }\n"
       ".inst 0xc08000c0  // mova za0h.s[x12], p0/M, z6.s\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x21\n"
+      "cmp x12, x22\n"
       "blt 1b\n"
-      "incw x22, ALL, MUL #2\n"
-      "incw x25, ALL, MUL #2\n"
-      "cbz x27, 5f\n"
+      "incw x23, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
+      "cbz x28, 5f\n"
       "2:"  // Width loop
-      "mov x19, %x[width]\n"
+      "mov x20, %x[width]\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34731  // whilelt pn9.s, x25, x19, VLx2\n"
-      "mov x20, %x[in]\n"
+      ".inst 0x25b44751  // whilelt pn9.s, x26, x20, VLx2\n"
+      "mov x21, %x[in]\n"
       "3:"  // Width loop: Odd: Loop
-      "ldr x19, [x20], #0x8\n"
+      "ldr x20, [x21], #0x8\n"
       ".inst 0x25306548  // psel p8.s, p9.s/Z, p10.s[w12]\n"
-      ".inst 0xa016427e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x19, x22, LSL #2]\n"
+      ".inst 0xa017429e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x20, x23, LSL #2]\n"
       ".inst 0xc160e3de  // bfcvt z30.h, { z30.s-z31.s }\n"
       ".inst 0xc08003c8  // mova za2h.s[x12], p0/M, z30.s\n"
       ".inst 0xc082800f  // mova z15.s, p0/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x21\n"
-      "st1w { z15.s }, p0, [x23]\n"
-      "addvl x23, x23, #1\n"
+      "cmp x12, x22\n"
+      "st1w { z15.s }, p0, [x24]\n"
+      "addvl x24, x24, #1\n"
       "blt 3b\n"
-      "incw x25, ALL, MUL #2\n"
-      "mov x19, %x[width]\n"
-      "incw x22, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
+      "mov x20, %x[width]\n"
+      "incw x23, ALL, MUL #2\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34731  // whilelt pn9.s, x25, x19, VLx2\n"
-      "mov x20, %x[in]\n"
+      ".inst 0x25b44751  // whilelt pn9.s, x26, x20, VLx2\n"
+      "mov x21, %x[in]\n"
       "4:"  // Width loop: Even: Loop
-      "ldr x19, [x20], #0x8\n"
+      "ldr x20, [x21], #0x8\n"
       ".inst 0x25306548  // psel p8.s, p9.s/Z, p10.s[w12]\n"
-      ".inst 0xa0164278  // ld1w { z24.s-z25.s }, pn8.s/Z, [x19, x22, LSL #2]\n"
+      ".inst 0xa0174298  // ld1w { z24.s-z25.s }, pn8.s/Z, [x20, x23, LSL #2]\n"
       ".inst 0xc160e318  // bfcvt z24.h, { z24.s-z25.s }\n"
       ".inst 0xc0800300  // mova za0h.s[x12], p0/M, z24.s\n"
       ".inst 0xc0828110  // mova z16.s, p0/M, za2v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x21\n"
-      "st1w { z16.s }, p0, [x23]\n"
-      "addvl x23, x23, #1\n"
+      "cmp x12, x22\n"
+      "st1w { z16.s }, p0, [x24]\n"
+      "addvl x24, x24, #1\n"
       "blt 4b\n"
-      "subs x27, x27, #0x1\n"
-      "incw x22, ALL, MUL #2\n"
-      "incw x25, ALL, MUL #2\n"
+      "subs x28, x28, #0x1\n"
+      "incw x23, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
       "bgt 2b\n"
       "5:"  // Width loop: Tails
-      "cbnz x24, 8f\n"
-      "mov x19, %x[width]\n"
+      "cbnz x25, 8f\n"
+      "mov x20, %x[width]\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34731  // whilelt pn9.s, x25, x19, VLx2\n"
-      "mov x20, %x[in]\n"
+      ".inst 0x25b44751  // whilelt pn9.s, x26, x20, VLx2\n"
+      "mov x21, %x[in]\n"
       "6:"  // Width loop: Tails: Even: Odd: Loop
-      "ldr x19, [x20], #0x8\n"
+      "ldr x20, [x21], #0x8\n"
       ".inst 0x25306548  // psel p8.s, p9.s/Z, p10.s[w12]\n"
-      ".inst 0xa016426e  // ld1w { z14.s-z15.s }, pn8.s/Z, [x19, x22, LSL #2]\n"
+      ".inst 0xa017428e  // ld1w { z14.s-z15.s }, pn8.s/Z, [x20, x23, LSL #2]\n"
       ".inst 0xc160e1ce  // bfcvt z14.h, { z14.s-z15.s }\n"
       ".inst 0xc08001c8  // mova za2h.s[x12], p0/M, z14.s\n"
       ".inst 0xc0828010  // mova z16.s, p0/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x21\n"
-      "st1w { z16.s }, p0, [x23]\n"
-      "addvl x23, x23, #1\n"
+      "cmp x12, x22\n"
+      "st1w { z16.s }, p0, [x24]\n"
+      "addvl x24, x24, #1\n"
       "blt 6b\n"
       "mov x12, #0x0\n"
       "7:"  // Width loop: Tails: Even: Even: Loop
       ".inst 0xc0828110  // mova z16.s, p0/M, za2v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x26\n"
-      "st1w { z16.s }, p0, [x23]\n"
-      "addvl x23, x23, #1\n"
+      "cmp x12, x27\n"
+      "st1w { z16.s }, p0, [x24]\n"
+      "addvl x24, x24, #1\n"
       "blt 7b\n"
       "b 10f\n"
       "8:"  // Width loop: Tails: Odd
@@ -140,16 +140,16 @@
       "9:"  // Width loop: Tails: Odd: Loop
       ".inst 0xc0828010  // mova z16.s, p0/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x26\n"
-      "st1w { z16.s }, p0, [x23]\n"
-      "addvl x23, x23, #1\n"
+      "cmp x12, x27\n"
+      "st1w { z16.s }, p0, [x24]\n"
+      "addvl x24, x24, #1\n"
       "blt 9b\n"
       "10:"  // End
-      "mov %x[outptr_raw], x23\n"
+      "mov %x[outptr_raw], x24\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [outptr_raw] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp
index e712eca..25bfad1 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,62 +34,62 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x22, ALL, MUL #2\n"
+      "cntw x9\n"
+      "sub x28, %x[width], #0x1\n"
       "cntw x21, ALL, MUL #2\n"
-      "cntw x28\n"
-      "sub x27, %x[width], #0x1\n"
-      "cntw x20, ALL, MUL #2\n"
-      "sub x19, x21, #0x1\n"
+      "sub x20, x22, #0x1\n"
       ".inst 0x25207815  // ptrue pn13.b\n"
       "whilelt p12.s, XZR, %x[height]\n"
-      "whilelt p11.s, x28, %x[height]\n"
-      "add x27, x27, x20\n"
-      "ands x26, %x[width], x19\n"
-      "udiv x27, x27, x20\n"
-      "csel x26, x26, x21, NE\n"
-      "mov x25, #0x0\n"
-      "and x24, x27, #0x1\n"
-      "sub x27, x27, #0x1\n"
-      "add x26, x26, #0x1\n"
-      "mov x19, %x[width]\n"
-      "mov x23, %x[in]\n"
+      "whilelt p11.s, x9, %x[height]\n"
+      "add x28, x28, x21\n"
+      "ands x27, %x[width], x20\n"
+      "udiv x28, x28, x21\n"
+      "csel x27, x27, x22, NE\n"
+      "mov x26, #0x0\n"
+      "and x25, x28, #0x1\n"
+      "sub x28, x28, #0x1\n"
+      "add x27, x27, #0x1\n"
+      "mov x20, %x[width]\n"
+      "mov x24, %x[in]\n"
       "ptrue p0.b\n"
-      "mov x22, %x[outptr_raw]\n"
-      "mov x21, %x[row_offset]\n"
+      "mov x23, %x[outptr_raw]\n"
+      "mov x22, %x[row_offset]\n"
+      "lsr x28, x28, #0x1\n"
       "lsr x27, x27, #0x1\n"
-      "lsr x26, x26, #0x1\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34732  // whilelt pn10.s, x25, x19, VLx2\n"
-      "add x20, x23, x28, LSL #3\n"
+      ".inst 0x25b44752  // whilelt pn10.s, x26, x20, VLx2\n"
+      "add x21, x24, x9, LSL #3\n"
       "1:"  // Width loop: Preamble: Loop
-      "ldr x19, [x23], #0x8\n"
+      "ldr x20, [x24], #0x8\n"
       ".inst 0x25306989  // psel p9.s, p10.s/Z, p12.s[w12]\n"
       ".inst 0x25306968  // psel p8.s, p10.s/Z, p11.s[w12]\n"
-      ".inst 0xa0154678  // ld1w { z24.s-z25.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa0154276  // ld1w { z22.s-z23.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa0164698  // ld1w { z24.s-z25.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa0164296  // ld1w { z22.s-z23.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e318  // bfcvt z24.h, { z24.s-z25.s }\n"
       ".inst 0xc160e2d6  // bfcvt z22.h, { z22.s-z23.s }\n"
       ".inst 0xc0800300  // mova za0h.s[x12], p0/M, z24.s\n"
       ".inst 0xc08002c4  // mova za1h.s[x12], p0/M, z22.s\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x28\n"
+      "cmp x12, x9\n"
       "blt 1b\n"
-      "incw x21, ALL, MUL #2\n"
-      "incw x25, ALL, MUL #2\n"
-      "cbz x27, 5f\n"
+      "incw x22, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
+      "cbz x28, 5f\n"
       "2:"  // Width loop
-      "mov x19, %x[width]\n"
-      "mov x23, %x[in]\n"
+      "mov x20, %x[width]\n"
+      "mov x24, %x[in]\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34732  // whilelt pn10.s, x25, x19, VLx2\n"
-      "add x20, x23, x28, LSL #3\n"
+      ".inst 0x25b44752  // whilelt pn10.s, x26, x20, VLx2\n"
+      "add x21, x24, x9, LSL #3\n"
       "3:"  // Width loop: Odd: Loop
-      "ldr x19, [x23], #0x8\n"
+      "ldr x20, [x24], #0x8\n"
       ".inst 0x25306989  // psel p9.s, p10.s/Z, p12.s[w12]\n"
       ".inst 0x25306968  // psel p8.s, p10.s/Z, p11.s[w12]\n"
-      ".inst 0xa0154676  // ld1w { z22.s-z23.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa015426a  // ld1w { z10.s-z11.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa0164696  // ld1w { z22.s-z23.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa016428a  // ld1w { z10.s-z11.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e2d6  // bfcvt z22.h, { z22.s-z23.s }\n"
       ".inst 0xc160e14a  // bfcvt z10.h, { z10.s-z11.s }\n"
       ".inst 0xc08002c8  // mova za2h.s[x12], p0/M, z22.s\n"
@@ -97,24 +97,24 @@
       ".inst 0xc0828008  // mova z8.s, p0/M, za0v.s[x12]\n"
       ".inst 0xc0828089  // mova z9.s, p0/M, za1v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x28\n"
-      ".inst 0xa06056c8  // st1w { z8.s-z9.s }, pn13.b, [x22]\n"
-      "addvl x22, x22, #2\n"
+      "cmp x12, x9\n"
+      ".inst 0xa06056e8  // st1w { z8.s-z9.s }, pn13.b, [x23]\n"
+      "addvl x23, x23, #2\n"
       "blt 3b\n"
-      "incw x25, ALL, MUL #2\n"
-      "mov x19, %x[width]\n"
-      "mov x23, %x[in]\n"
-      "incw x21, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
+      "mov x20, %x[width]\n"
+      "mov x24, %x[in]\n"
+      "incw x22, ALL, MUL #2\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34732  // whilelt pn10.s, x25, x19, VLx2\n"
-      "add x20, x23, x28, LSL #3\n"
+      ".inst 0x25b44752  // whilelt pn10.s, x26, x20, VLx2\n"
+      "add x21, x24, x9, LSL #3\n"
       "4:"  // Width loop: Even: Loop
-      "ldr x19, [x23], #0x8\n"
+      "ldr x20, [x24], #0x8\n"
       ".inst 0x25306989  // psel p9.s, p10.s/Z, p12.s[w12]\n"
       ".inst 0x25306968  // psel p8.s, p10.s/Z, p11.s[w12]\n"
-      ".inst 0xa015467a  // ld1w { z26.s-z27.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa015427e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa016469a  // ld1w { z26.s-z27.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa016429e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e35a  // bfcvt z26.h, { z26.s-z27.s }\n"
       ".inst 0xc160e3de  // bfcvt z30.h, { z30.s-z31.s }\n"
       ".inst 0xc0800340  // mova za0h.s[x12], p0/M, z26.s\n"
@@ -122,28 +122,28 @@
       ".inst 0xc0828106  // mova z6.s, p0/M, za2v.s[x12]\n"
       ".inst 0xc082818e  // mova z14.s, p0/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x28\n"
-      ".inst 0xa16056c6  // st1w { z6.s, z14.s }, pn13.b, [x22]\n"
-      "addvl x22, x22, #2\n"
+      "cmp x12, x9\n"
+      ".inst 0xa16056e6  // st1w { z6.s, z14.s }, pn13.b, [x23]\n"
+      "addvl x23, x23, #2\n"
       "blt 4b\n"
-      "subs x27, x27, #0x1\n"
-      "incw x21, ALL, MUL #2\n"
-      "incw x25, ALL, MUL #2\n"
+      "subs x28, x28, #0x1\n"
+      "incw x22, ALL, MUL #2\n"
+      "incw x26, ALL, MUL #2\n"
       "bgt 2b\n"
       "5:"  // Width loop: Tails
-      "cbnz x24, 8f\n"
-      "mov x19, %x[width]\n"
-      "mov x23, %x[in]\n"
+      "cbnz x25, 8f\n"
+      "mov x20, %x[width]\n"
+      "mov x24, %x[in]\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34732  // whilelt pn10.s, x25, x19, VLx2\n"
-      "add x20, x23, x28, LSL #3\n"
+      ".inst 0x25b44752  // whilelt pn10.s, x26, x20, VLx2\n"
+      "add x21, x24, x9, LSL #3\n"
       "6:"  // Width loop: Tails: Even: Odd: Loop
-      "ldr x19, [x23], #0x8\n"
+      "ldr x20, [x24], #0x8\n"
       ".inst 0x25306989  // psel p9.s, p10.s/Z, p12.s[w12]\n"
       ".inst 0x25306968  // psel p8.s, p10.s/Z, p11.s[w12]\n"
-      ".inst 0xa015466c  // ld1w { z12.s-z13.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa015426e  // ld1w { z14.s-z15.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa016468c  // ld1w { z12.s-z13.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa016428e  // ld1w { z14.s-z15.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e18c  // bfcvt z12.h, { z12.s-z13.s }\n"
       ".inst 0xc160e1ce  // bfcvt z14.h, { z14.s-z15.s }\n"
       ".inst 0xc0800188  // mova za2h.s[x12], p0/M, z12.s\n"
@@ -151,18 +151,18 @@
       ".inst 0xc0828007  // mova z7.s, p0/M, za0v.s[x12]\n"
       ".inst 0xc082808f  // mova z15.s, p0/M, za1v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x28\n"
-      ".inst 0xa16056c7  // st1w { z7.s, z15.s }, pn13.b, [x22]\n"
-      "addvl x22, x22, #2\n"
+      "cmp x12, x9\n"
+      ".inst 0xa16056e7  // st1w { z7.s, z15.s }, pn13.b, [x23]\n"
+      "addvl x23, x23, #2\n"
       "blt 6b\n"
       "mov x12, #0x0\n"
       "7:"  // Width loop: Tails: Even: Even: Loop
       ".inst 0xc082810e  // mova z14.s, p0/M, za2v.s[x12]\n"
       ".inst 0xc082818f  // mova z15.s, p0/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x26\n"
-      ".inst 0xa06056ce  // st1w { z14.s-z15.s }, pn13.b, [x22]\n"
-      "addvl x22, x22, #2\n"
+      "cmp x12, x27\n"
+      ".inst 0xa06056ee  // st1w { z14.s-z15.s }, pn13.b, [x23]\n"
+      "addvl x23, x23, #2\n"
       "blt 7b\n"
       "b 10f\n"
       "8:"  // Width loop: Tails: Odd
@@ -171,16 +171,16 @@
       ".inst 0xc0828014  // mova z20.s, p0/M, za0v.s[x12]\n"
       ".inst 0xc0828095  // mova z21.s, p0/M, za1v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x26\n"
-      ".inst 0xa06056d4  // st1w { z20.s-z21.s }, pn13.b, [x22]\n"
-      "addvl x22, x22, #2\n"
+      "cmp x12, x27\n"
+      ".inst 0xa06056f4  // st1w { z20.s-z21.s }, pn13.b, [x23]\n"
+      "addvl x23, x23, #2\n"
       "blt 9b\n"
       "10:"  // End
-      "mov %x[outptr_raw], x22\n"
+      "mov %x[outptr_raw], x23\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [outptr_raw] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave4VL_block2_fp32_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave4VL_block2_fp32_bf16.hpp
index e08d6d9..9255831 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave4VL_block2_fp32_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave4VL_block2_fp32_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,51 +34,51 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x23, ALL, MUL #2\n"
+      "cntw x10\n"
       "cntw x22, ALL, MUL #2\n"
-      "cntw x9\n"
-      "cntw x21, ALL, MUL #2\n"
-      "cntw x19, ALL, MUL #3\n"
-      "sub x20, x22, #0x1\n"
+      "cntw x20, ALL, MUL #3\n"
+      "sub x21, x23, #0x1\n"
       ".inst 0x25207817  // ptrue pn15.b\n"
       "whilelt p1.s, XZR, %x[height]\n"
-      "whilelt p14.s, x9, %x[height]\n"
-      "whilelt p13.s, x21, %x[height]\n"
-      "whilelt p12.s, x19, %x[height]\n"
-      "sub x28, %x[width], #0x1\n"
-      "cntw x19, ALL, MUL #2\n"
-      "ands x27, %x[width], x20\n"
-      "mov x26, %x[in]\n"
-      "add x28, x28, x19\n"
-      "csel x27, x27, x22, NE\n"
-      "add x25, x26, x9, LSL #3\n"
-      "mov x24, #0x0\n"
-      "udiv x28, x28, x19\n"
-      "add x27, x27, #0x1\n"
-      "mov x19, %x[width]\n"
-      "add x23, x25, x9, LSL #3\n"
+      "whilelt p14.s, x10, %x[height]\n"
+      "whilelt p13.s, x22, %x[height]\n"
+      "whilelt p12.s, x20, %x[height]\n"
+      "sub x9, %x[width], #0x1\n"
+      "cntw x20, ALL, MUL #2\n"
+      "ands x28, %x[width], x21\n"
+      "mov x27, %x[in]\n"
+      "add x9, x9, x20\n"
+      "csel x28, x28, x23, NE\n"
+      "add x26, x27, x10, LSL #3\n"
+      "mov x25, #0x0\n"
+      "udiv x9, x9, x20\n"
+      "add x28, x28, #0x1\n"
+      "mov x20, %x[width]\n"
+      "add x24, x26, x10, LSL #3\n"
       "ptrue p0.b\n"
-      "mov x22, %x[outptr_raw]\n"
-      "mov x21, %x[row_offset]\n"
-      "sub x28, x28, #0x1\n"
-      "lsr x27, x27, #0x1\n"
+      "mov x23, %x[outptr_raw]\n"
+      "mov x22, %x[row_offset]\n"
+      "sub x9, x9, #0x1\n"
+      "lsr x28, x28, #0x1\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34713  // whilelt pn11.s, x24, x19, VLx2\n"
-      "add x20, x23, x9, LSL #3\n"
+      ".inst 0x25b44733  // whilelt pn11.s, x25, x20, VLx2\n"
+      "add x21, x24, x10, LSL #3\n"
       "1:"  // Width loop: Preamble: Loop
-      "ldr x19, [x26], #0x8\n"
+      "ldr x20, [x27], #0x8\n"
       ".inst 0x25306c28  // psel p8.s, p11.s/Z, p1.s[w12]\n"
       ".inst 0x25306dca  // psel p10.s, p11.s/Z, p14.s[w12]\n"
-      ".inst 0xa0154278  // ld1w { z24.s-z25.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x25], #0x8\n"
+      ".inst 0xa0164298  // ld1w { z24.s-z25.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x26], #0x8\n"
       ".inst 0x25306da9  // psel p9.s, p11.s/Z, p13.s[w12]\n"
       ".inst 0x25306d88  // psel p8.s, p11.s/Z, p12.s[w12]\n"
-      ".inst 0xa0154a62  // ld1w { z2.s-z3.s }, pn10.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x23], #0x8\n"
-      ".inst 0xa015466a  // ld1w { z10.s-z11.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa0164a82  // ld1w { z2.s-z3.s }, pn10.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x24], #0x8\n"
+      ".inst 0xa016468a  // ld1w { z10.s-z11.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e318  // bfcvt z24.h, { z24.s-z25.s }\n"
       ".inst 0xc160e042  // bfcvt z2.h, { z2.s-z3.s }\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa015426c  // ld1w { z12.s-z13.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa016428c  // ld1w { z12.s-z13.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e14a  // bfcvt z10.h, { z10.s-z11.s }\n"
       ".inst 0xc160e18c  // bfcvt z12.h, { z12.s-z13.s }\n"
       ".inst 0xc0800300  // mova za0h.s[x12], p0/M, z24.s\n"
@@ -86,11 +86,11 @@
       ".inst 0xc0800148  // mova za2h.s[x12], p0/M, z10.s\n"
       ".inst 0xc080018c  // mova za3h.s[x12], p0/M, z12.s\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
+      "cmp x12, x10\n"
       "blt 1b\n"
-      "incw x21, ALL, MUL #2\n"
-      "incw x24, ALL, MUL #2\n"
-      "cbz x28, 5f\n"
+      "incw x22, ALL, MUL #2\n"
+      "incw x25, ALL, MUL #2\n"
+      "cbz x9, 5f\n"
       "2:"  // Width loop
       "mov x12, #0x0\n"
       "3:"  // Width loop: Store: Loop
@@ -99,32 +99,32 @@
       ".inst 0xc0828119  // mova z25.s, p0/M, za2v.s[x12]\n"
       ".inst 0xc082819d  // mova z29.s, p0/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      ".inst 0xa160ded1  // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x22]\n"
-      "addvl x22, x22, #4\n"
+      "cmp x12, x10\n"
+      ".inst 0xa160def1  // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x23]\n"
+      "addvl x23, x23, #4\n"
       "blt 3b\n"
-      "mov x26, %x[in]\n"
-      "add x25, x26, x9, LSL #3\n"
-      "mov x19, %x[width]\n"
-      "add x23, x25, x9, LSL #3\n"
+      "mov x27, %x[in]\n"
+      "add x26, x27, x10, LSL #3\n"
+      "mov x20, %x[width]\n"
+      "add x24, x26, x10, LSL #3\n"
       "mov x12, #0x0\n"
-      ".inst 0x25b34713  // whilelt pn11.s, x24, x19, VLx2\n"
-      "add x20, x23, x9, LSL #3\n"
+      ".inst 0x25b44733  // whilelt pn11.s, x25, x20, VLx2\n"
+      "add x21, x24, x10, LSL #3\n"
       "4:"  // Width loop: Load: Loop
-      "ldr x19, [x26], #0x8\n"
+      "ldr x20, [x27], #0x8\n"
       ".inst 0x25306c28  // psel p8.s, p11.s/Z, p1.s[w12]\n"
       ".inst 0x25306dca  // psel p10.s, p11.s/Z, p14.s[w12]\n"
-      ".inst 0xa015426c  // ld1w { z12.s-z13.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x25], #0x8\n"
+      ".inst 0xa016428c  // ld1w { z12.s-z13.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x26], #0x8\n"
       ".inst 0x25306da9  // psel p9.s, p11.s/Z, p13.s[w12]\n"
       ".inst 0x25306d88  // psel p8.s, p11.s/Z, p12.s[w12]\n"
-      ".inst 0xa0154a6e  // ld1w { z14.s-z15.s }, pn10.s/Z, [x19, x21, LSL #2]\n"
-      "ldr x19, [x23], #0x8\n"
-      ".inst 0xa0154672  // ld1w { z18.s-z19.s }, pn9.s/Z, [x19, x21, LSL #2]\n"
+      ".inst 0xa0164a8e  // ld1w { z14.s-z15.s }, pn10.s/Z, [x20, x22, LSL #2]\n"
+      "ldr x20, [x24], #0x8\n"
+      ".inst 0xa0164692  // ld1w { z18.s-z19.s }, pn9.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e18c  // bfcvt z12.h, { z12.s-z13.s }\n"
       ".inst 0xc160e1ce  // bfcvt z14.h, { z14.s-z15.s }\n"
-      "ldr x19, [x20], #0x8\n"
-      ".inst 0xa015427e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x19, x21, LSL #2]\n"
+      "ldr x20, [x21], #0x8\n"
+      ".inst 0xa016429e  // ld1w { z30.s-z31.s }, pn8.s/Z, [x20, x22, LSL #2]\n"
       ".inst 0xc160e252  // bfcvt z18.h, { z18.s-z19.s }\n"
       ".inst 0xc160e3de  // bfcvt z30.h, { z30.s-z31.s }\n"
       ".inst 0xc0800180  // mova za0h.s[x12], p0/M, z12.s\n"
@@ -132,11 +132,11 @@
       ".inst 0xc0800248  // mova za2h.s[x12], p0/M, z18.s\n"
       ".inst 0xc08003cc  // mova za3h.s[x12], p0/M, z30.s\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
+      "cmp x12, x10\n"
       "blt 4b\n"
-      "subs x28, x28, #0x1\n"
-      "incw x21, ALL, MUL #2\n"
-      "incw x24, ALL, MUL #2\n"
+      "subs x9, x9, #0x1\n"
+      "incw x22, ALL, MUL #2\n"
+      "incw x25, ALL, MUL #2\n"
       "bgt 2b\n"
       "5:"  // Width loop: Tails
       "mov x12, #0x0\n"
@@ -146,16 +146,16 @@
       ".inst 0xc0828119  // mova z25.s, p0/M, za2v.s[x12]\n"
       ".inst 0xc082819d  // mova z29.s, p0/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
-      ".inst 0xa160ded1  // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x22]\n"
-      "addvl x22, x22, #4\n"
+      "cmp x12, x28\n"
+      ".inst 0xa160def1  // st1w { z17.s, z21.s, z25.s, z29.s }, pn15.b, [x23]\n"
+      "addvl x23, x23, #4\n"
       "blt 6b\n"
       "7:"  // End
-      "mov %x[outptr_raw], x22\n"
+      "mov %x[outptr_raw], x23\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [outptr_raw] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_bf16_bf16.hpp
index 3c8c707..9b66a6f 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,175 +34,175 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "mov x21, %x[width]\n"
+      "inch x21\n"
+      "cnth x11\n"
+      "sub x21, x21, #0x1\n"
+      "udiv x21, x21, x11\n"  // n_passes = ceildiv(width, VL<T>)
       "mov x20, %x[width]\n"
-      "inch x20\n"
-      "cnth x10\n"
-      "sub x20, x20, #0x1\n"
-      "udiv x20, x20, x10\n"  // n_passes = ceildiv(width, VL<T>)
-      "mov x19, %x[width]\n"
-      "sub x9, x10, #0x1\n"
-      "sub x28, x20, #0x1\n"
-      "ands x9, x19, x9\n"
-      "sub x27, x10, #0x2\n"
-      "lsl x19, %x[height], #0x1\n"  // height * 2
-      "mov x26, #0x0\n"
-      "mov x25, %x[in]\n"
-      "lsr x28, x28, #0x1\n"  // n_loops = (n_passes - 1) / 2
-      "ldr x24, [x25, #0x0]\n"
-      "and x23, x20, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
-      "csel x9, x9, x10, NE\n"
-      "ldr x22, [x25, #0x8]\n"
+      "sub x10, x11, #0x1\n"
+      "sub x9, x21, #0x1\n"
+      "ands x10, x20, x10\n"
+      "sub x28, x11, #0x2\n"
+      "lsl x20, %x[height], #0x1\n"  // height * 2
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "lsr x9, x9, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "ldr x25, [x26, #0x0]\n"
+      "and x24, x21, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "csel x10, x10, x11, NE\n"
+      "ldr x23, [x26, #0x8]\n"
       "ptrue p11.h\n"
-      "whilelt p10.h, XZR, x19\n"
-      "mov x21, %x[row_offset]\n"
-      "mov x20, %x[out]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "add x25, x25, #0x10\n"
+      "whilelt p10.h, XZR, x20\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x27, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386140  // psel p0.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
-      ".inst 0xe05502c1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x22, x21, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe05602e1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x23, x22, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
+      "cmp x12, x28\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386140  // psel p0.h, p8.h/Z, p10.h[w12, #1]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe05502c1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x22, x21, LSL #1]\n"
-      "ldr x24, [x25, #0x0]\n"
-      "inch x21\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "inch x26\n"
-      "cbz x28, 8f\n"
-      "mov x19, x28\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe05602e1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x23, x22, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "inch x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "inch x27\n"
+      "cbz x9, 8f\n"
+      "mov x20, x9\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x27, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      ".inst 0xe06a8281  // st1h { za0v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
+      ".inst 0xe06b82a1  // st1h { za0v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "inch x26\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe06a8281  // st1h { za0v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
-      "addvl x20, x20, #2\n"
-      "inch x21\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "inch x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe06b82a1  // st1h { za0v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
+      "addvl x21, x21, #2\n"
+      "inch x22\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x27, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      ".inst 0xe06a8289  // st1h { za1v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
+      ".inst 0xe06b82a9  // st1h { za1v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "subs x19, x19, #0x1\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe06a8289  // st1h { za1v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
-      "addvl x20, x20, #2\n"
-      "inch x26\n"
-      "inch x21\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe06b82a9  // st1h { za1v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
+      "addvl x21, x21, #2\n"
+      "inch x27\n"
+      "inch x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x23, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
-      "add x25, x25, #0x8\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x11\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
       "blt 9b\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 10b\n"
-      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x20\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block2_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block2_bf16_bf16.hpp
index 81b346c..d0375de 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block2_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block2_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,186 +34,186 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cnth x20\n"
+      "cnth x22\n"
+      "mov x21, %x[width]\n"
+      "inch x21\n"
+      "mov x20, %x[width]\n"
+      "sub x11, x22, #0x1\n"
+      "sub x21, x21, #0x1\n"
+      "ands x11, x20, x11\n"
       "cntw x10\n"
-      "mov x19, %x[width]\n"
-      "inch x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x9, x19, #0x1\n"
-      "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x27, x20, #0x1\n"
-      "ands x27, x19, x27\n"
-      "csel x27, x27, x20, NE\n"
-      "add x27, x27, #0x1\n"
-      "lsr x27, x27, #0x1\n"
-      "sub x26, x10, #0x2\n"
+      "udiv x21, x21, x22\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x11, x11, x22, NE\n"
+      "sub x9, x21, #0x1\n"
+      "add x11, x11, #0x1\n"
+      "sub x28, x10, #0x2\n"
+      "lsl x20, %x[height], #0x1\n"  // height * 2
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "lsr x9, x9, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x24, x21, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x23, [x26, #0x8]\n"
+      "lsr x11, x11, #0x1\n"
       "ptrue p11.s\n"
-      "lsl x19, %x[height], #0x1\n" // height * 2
-      "whilelt p10.h, XZR, x19\n"
-      "mov x25, %x[row_offset]\n"
-      "mov x24, %x[out]\n"
-      "mov x23, #0x0\n"
-      "whilelt p9.h, x23, %x[width]\n"
-      "whilelt p8.h, x23, %x[width]\n"
-      "mov x22, %x[in]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      "whilelt p10.h, XZR, x20\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x26, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25286140  // dup p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe05902a0  // ld1h { za0h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25686140  // dup p0.h, p8.h/Z, p10.h[w12, #2]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0590282  // ld1h { za0h.h[x12, #2] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      ".inst 0x25686140  // psel p0.h, p8.h/Z, p10.h[w12, #2]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe05602e2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
       "add x12, x12, #0x4\n"
-      "cmp x12, x26, LSL #1\n"
+      "cmp x12, x28, LSL #1\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25286140  // dup p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe05902a0  // ld1h { za0h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25686140  // dup p0.h, p8.h/Z, p10.h[w12, #2]\n"
-      "mov x22, %x[in]\n"
-      ".inst 0xe0590282  // ld1h { za0h.h[x12, #2] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      "inch x25\n"
-      "inch x23\n"
+      ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      ".inst 0x25686140  // psel p0.h, p8.h/Z, p10.h[w12, #2]\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe05602e2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "inch x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "inch x27\n"
       "cbz x9, 8f\n"
-      "mov x19, x9\n"
+      "mov x20, x9\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.h, x23, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x26, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25396140  // dup p0.h, p8.h/Z, p10.h[w13, #1]\n"
-      ".inst 0xe05922a1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25796140  // dup p0.h, p8.h/Z, p10.h[w13, #3]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0592283  // ld1h { za0h.h[x13, #3] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x4\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
+      ".inst 0x25396140  // psel p0.h, p8.h/Z, p10.h[w13, #1]\n"
+      ".inst 0xe0562321  // ld1h { za0h.h[x13, #1] }, p0/Z, [x25, x22, LSL #1]\n"
+      ".inst 0x25796141  // psel p1.h, p8.h/Z, p10.h[w13, #3]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe05626e3  // ld1h { za0h.h[x13, #3] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0aa82a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x26\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25396140  // dup p0.h, p8.h/Z, p10.h[w13, #1]\n"
-      ".inst 0xe05922a1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25796140  // dup p0.h, p8.h/Z, p10.h[w13, #3]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0592283  // ld1h { za0h.h[x13, #3] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.h, x23, %x[width]\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
-      "inch x23\n"
-      "inch x25\n"
-      "whilelt p8.h, x23, %x[width]\n"
+      ".inst 0x25396140  // psel p0.h, p8.h/Z, p10.h[w13, #1]\n"
+      ".inst 0xe0562321  // ld1h { za0h.h[x13, #1] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25796141  // psel p1.h, p8.h/Z, p10.h[w13, #3]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe05626e3  // ld1h { za0h.h[x13, #3] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "inch x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0aa82a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "inch x22\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x26, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25296140  // dup p0.h, p8.h/Z, p10.h[w13]\n"
-      ".inst 0xe05922a0  // ld1h { za0h.h[x13] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25696140  // dup p0.h, p8.h/Z, p10.h[w13, #2]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0592282  // ld1h { za0h.h[x13, #2] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x4\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
+      ".inst 0x25296140  // psel p0.h, p8.h/Z, p10.h[w13]\n"
+      ".inst 0xe0562320  // ld1h { za0h.h[x13] }, p0/Z, [x25, x22, LSL #1]\n"
+      ".inst 0x25696141  // psel p1.h, p8.h/Z, p10.h[w13, #2]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe05626e2  // ld1h { za0h.h[x13, #2] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0aa82a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x26\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25296140  // dup p0.h, p8.h/Z, p10.h[w13]\n"
-      ".inst 0xe05922a0  // ld1h { za0h.h[x13] }, p0/Z, [x21, x25, LSL #1]\n"
-      ".inst 0x25696140  // dup p0.h, p8.h/Z, p10.h[w13, #2]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0592282  // ld1h { za0h.h[x13, #2] }, p0/Z, [x20, x25, LSL #1]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.h, x23, %x[width]\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
-      "inch x23\n"
-      "inch x25\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25296140  // psel p0.h, p8.h/Z, p10.h[w13]\n"
+      ".inst 0xe0562320  // ld1h { za0h.h[x13] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25696141  // psel p1.h, p8.h/Z, p10.h[w13, #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe05626e2  // ld1h { za0h.h[x13, #2] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0aa82a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "inch x27\n"
+      "inch x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x28, 11f\n"
-      "mov x22, %x[in]\n"
-      "whilelt p8.h, x23, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25396140  // dup p0.h, p8.h/Z, p10.h[w13, #1]\n"
-      "addvl x24, x24, #1\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe05922a1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x21, x25, LSL #1]\n"
-      "add x22, x22, #0x8\n"
-      "add x13, x13, #0x2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      "ldr x25, [x26, #0x0]\n"
       "add x12, x12, #0x1\n"
+      ".inst 0x25396140  // psel p0.h, p8.h/Z, p10.h[w13, #1]\n"
       "cmp x12, x10\n"
+      ".inst 0xe0562321  // ld1h { za0h.h[x13, #1] }, p0/Z, [x25, x22, LSL #1]\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
+      "add x13, x13, #0x2\n"
       "blt 9b\n"
-      "whilelt p9.h, x23, %x[width]\n"
-      "whilelt p8.h, x23, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
-      "add x19, x19, #0x2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x11\n"
+      "addvl x21, x21, #1\n"
+      "add x20, x20, #0x2\n"
       "blt 10b\n"
-      "whilelt p9.h, x23, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x11\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x24\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8.hpp
index bee3cc5..622d9aa 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,189 +34,189 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cntb x20\n"
-      "mov x22, %x[width]\n"
-      "incb x22\n"
-      "mov x19, %x[width]\n"
-      "sub x9, x20, #0x1\n"
-      "cntw x28\n"
-      "sub x22, x22, #0x1\n"
-      "ands x9, x19, x9\n"
-      "udiv x22, x22, x20\n"  // n_passes = ceildiv(width, VL<T>)
-      "csel x9, x9, x20, NE\n"
-      "lsl x21, %x[height], #0x1\n"  // height * 2
-      "lsl x20, x28, #0x1\n"
-      "sub x19, x22, #0x1\n"
-      "add x9, x9, #0x3\n"
-      "sub x27, x28, #0x2\n"
-      "whilelt p9.b, XZR, x21\n"
-      "whilelt p8.b, x20, x21\n"
-      "mov x26, #0x0\n"
-      "mov x25, %x[in]\n"
-      "lsr x19, x19, #0x1\n"  // n_loops = (n_passes - 1) / 2
-      "ldr x24, [x25, #0x0]\n"
-      "and x23, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
-      "lsr x9, x9, #0x2\n"
-      "ldr x22, [x25, #0x8]\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
+      "sub x10, x21, #0x1\n"
+      "cntw x9\n"
+      "sub x23, x23, #0x1\n"
+      "ands x10, x20, x10\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x10, x10, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x9, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x10, x10, #0x3\n"
+      "sub x28, x9, #0x2\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "ldr x25, [x26, #0x0]\n"
+      "and x24, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x10, x10, #0x2\n"
+      "ldr x23, [x26, #0x8]\n"
       "ptrue p11.s\n"
       "zip1 p10.b, p9.b, p8.b\n"
-      "mov x21, %x[row_offset]\n"
-      "mov x20, %x[out]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "add x25, x25, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x27, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe0150300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0160320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
       ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "ldr x24, [x25, #0x0]\n"
-      ".inst 0xe01502c4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x22, x21]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe01602e4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
       "add x12, x12, #0x8\n"
-      "cmp x12, x27, LSL #2\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
+      "cmp x12, x28, LSL #2\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe0150300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0160320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
       ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01502c4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x22, x21]\n"
-      "ldr x24, [x25, #0x0]\n"
-      "incb x21\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "incb x26\n"
-      "cbz x19, 8f\n"
-      "mov x19, x19\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe01602e4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "incb x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "incb x27\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x27, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
       ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0bc8281  // st1w { za0v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
+      ".inst 0xe0a982a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "add x13, x13, #0x8\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "incb x26\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe0bc8281  // st1w { za0v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "addvl x20, x20, #2\n"
-      "incb x21\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "incb x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0a982a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incb x22\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x27, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
       ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe0152300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0162320  // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
       ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0bc8289  // st1w { za2v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
+      ".inst 0xe0a982a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "add x13, x13, #0x8\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
       ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe0152300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x21]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0162320  // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "subs x19, x19, #0x1\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe0bc8289  // st1w { za2v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "addvl x20, x20, #2\n"
-      "incb x26\n"
-      "incb x21\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0a982a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incb x27\n"
+      "incb x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x23, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      "ldr x25, [x26, #0x0]\n"
       "add x12, x12, #0x1\n"
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      "cmp x12, x28\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
-      "add x25, x25, #0x8\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x9\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
       "add x13, x13, #0x4\n"
       "blt 9b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
-      "add x19, x19, #0x4\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
+      "add x20, x20, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x20\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8_summing.hpp
index 3ba1b98..07f0370 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,220 +32,220 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
       "mov z18.b, #0x1\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
       "mov z17.s, #0x0\n"
-      "cntb x20\n"
-      "cntw x10\n"
-      "ptrue p1.b\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x9, x19, #0x1\n"
-      "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x27, x20, #0x1\n"
-      "ands x27, x19, x27\n"
-      "csel x27, x27, x20, NE\n"
-      "add x27, x27, #0x3\n"
-      "lsr x27, x27, #0x2\n"
-      "sub x26, x10, #0x2\n"
+      "sub x10, x21, #0x1\n"
+      "cntw x9\n"
+      "sub x23, x23, #0x1\n"
+      "ands x10, x20, x10\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x10, x10, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x9, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x10, x10, #0x3\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x28, #0x0\n"
+      "ptrue p2.b\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x27, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x10, x10, #0x2\n"
+      "sub x26, x9, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x10, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
       "zip1 p10.b, p9.b, p8.b\n"
       "mov x25, %x[row_offset]\n"
       "mov x24, %x[out]\n"
-      "mov x23, #0x0\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "cbnz %x[first], 1f\n"
       "addvl x24, x24, #-1\n"
-      "ld1w { z17.s }, p1/Z, [x24]\n"
+      "ld1w { z17.s }, p2/Z, [x24]\n"
       "1:"  // K loop: Load row sums: End
-      "mov x22, %x[in]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      "mov x23, %x[in]\n"
+      "ldr x22, [x23, #0x0]\n"
       "mov x12, #0x0\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
       "cbz x26, 3f\n"
       "2:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01902a0  // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25646140  // dup p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0190284  // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01902c0  // ld1b { za0h.b[x12] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01902a4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x21, x25]\n"
       "add x12, x12, #0x8\n"
       "cmp x12, x26, LSL #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
       "blt 2b\n"
       "3:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01902a0  // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25646140  // dup p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "mov x22, %x[in]\n"
-      ".inst 0xe0190284  // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01902c0  // ld1b { za0h.b[x12] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0xe01902a4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x21, x25]\n"
+      "ldr x22, [x23, #0x0]\n"
       "incb x25\n"
-      "incb x23\n"
-      "cbz x9, 9f\n"
-      "mov x19, x9\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
+      "incb x28\n"
+      "cbz x20, 9f\n"
+      "mov x20, x20\n"
       "4:"  // K loop: Main loop
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x26, 6f\n"
       "5:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25756140  // dup p0.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192286  // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01922a6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       "sdot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828430  // mova z16.s, p1/M, za0v.s[x12, #1]\n"
-      "addvl x24, x24, #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828830  // mova z16.s, p2/M, za0v.s[x12, #1]\n"
+      ".inst 0xe0a98301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x26\n"
       "sdot z17.s, z16.b, z18.b\n"
+      "add x23, x23, #0x10\n"
+      "addvl x24, x24, #2\n"
+      "add x13, x13, #0x8\n"
       "blt 5b\n"
       "6:"  // K loop: Main loop: First: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25756140  // dup p0.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192286  // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0xe01922a6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "sdot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828430  // mova z16.s, p1/M, za0v.s[x12, #1]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828830  // mova z16.s, p2/M, za0v.s[x12, #1]\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8700  // st1w { za0v.s[x12] }, p1/Z, [x24, XZR, LSL #2]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "incb x28\n"
+      "add x23, x23, #0x10\n"
+      ".inst 0xe0a98301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
+      "sdot z17.s, z16.b, z18.b\n"
       "addvl x24, x24, #2\n"
-      "incb x23\n"
       "incb x25\n"
-      "sdot z17.s, z16.b, z18.b\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x26, 8f\n"
       "7:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01922a0  // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25656140  // dup p0.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192284  // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01922a4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       "sdot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828530  // mova z16.s, p1/M, za2v.s[x12, #1]\n"
-      "addvl x24, x24, #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828930  // mova z16.s, p2/M, za2v.s[x12, #1]\n"
+      ".inst 0xe0a98309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x26\n"
       "sdot z17.s, z16.b, z18.b\n"
+      "add x23, x23, #0x10\n"
+      "addvl x24, x24, #2\n"
+      "add x13, x13, #0x8\n"
       "blt 7b\n"
       "8:"  // K loop: Main loop: Second: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01922a0  // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25656140  // dup p0.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192284  // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0xe01922a4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
       "sdot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828530  // mova z16.s, p1/M, za2v.s[x12, #1]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828930  // mova z16.s, p2/M, za2v.s[x12, #1]\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8708  // st1w { za2v.s[x12] }, p1/Z, [x24, XZR, LSL #2]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x23, x23, #0x10\n"
+      ".inst 0xe0a98309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
+      "sdot z17.s, z16.b, z18.b\n"
       "addvl x24, x24, #2\n"
-      "incb x23\n"
+      "incb x28\n"
       "incb x25\n"
-      "sdot z17.s, z16.b, z18.b\n"
-      "subs x19, x19, #0x1\n"
       "bgt 4b\n"
       "9:"  // K loop: Tails
-      "cbnz x28, 12f\n"
-      "mov x22, %x[in]\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "cbnz x27, 12f\n"
+      "mov x23, %x[in]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: First
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      "addvl x24, x24, #1\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      "sdot z17.s, z16.b, z18.b\n"
-      "add x22, x22, #0x8\n"
-      "add x13, x13, #0x4\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      "sdot z17.s, z16.b, z18.b\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      "cmp x12, x9\n"
+      "add x23, x23, #0x8\n"
+      "addvl x24, x24, #1\n"
+      "add x13, x13, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      "whilelt p8.b, x23, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "11:"  // K loop: Tails: Even: Second
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
-      "add x19, x19, #0x4\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
       "add x12, x12, #0x1\n"
+      "cmp x12, x10\n"
       "sdot z17.s, z16.b, z18.b\n"
-      "cmp x12, x27\n"
+      "addvl x24, x24, #1\n"
+      "add x20, x20, #0x4\n"
       "blt 11b\n"
-      "whilelt p9.b, x23, %x[width]\n"
+      "whilelt p9.b, x28, %x[width]\n"
       "b 14f\n"
       "12:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "13:"  // K loop: Tails: Odd: Loop
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x10\n"
       "sdot z17.s, z16.b, z18.b\n"
+      "addvl x24, x24, #1\n"
       "blt 13b\n"
       "14:"  // K loop: End
-      "st1w { z17.s }, p1, [x24]\n"
+      "st1w { z17.s }, p2, [x24]\n"
       "addvl x24, x24, #1\n"
       "mov %x[out], x24\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
index 881dfe1..618570d 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,189 +34,189 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cntb x20\n"
-      "mov x22, %x[width]\n"
-      "incb x22\n"
-      "mov x19, %x[width]\n"
-      "sub x9, x20, #0x1\n"
-      "cntw x28\n"
-      "sub x22, x22, #0x1\n"
-      "ands x9, x19, x9\n"
-      "udiv x22, x22, x20\n"  // n_passes = ceildiv(width, VL<T>)
-      "csel x9, x9, x20, NE\n"
-      "lsl x21, %x[height], #0x1\n"  // height * 2
-      "lsl x20, x28, #0x1\n"
-      "sub x19, x22, #0x1\n"
-      "add x9, x9, #0x3\n"
-      "sub x27, x28, #0x2\n"
-      "whilelt p9.b, XZR, x21\n"
-      "whilelt p8.b, x20, x21\n"
-      "mov x26, #0x0\n"
-      "mov x25, %x[in]\n"
-      "lsr x19, x19, #0x1\n"  // n_loops = (n_passes - 1) / 2
-      "ldr x24, [x25, #0x0]\n"
-      "and x23, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
-      "lsr x9, x9, #0x2\n"
-      "ldr x22, [x25, #0x8]\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
+      "sub x10, x21, #0x1\n"
+      "cntw x9\n"
+      "sub x23, x23, #0x1\n"
+      "ands x10, x20, x10\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x10, x10, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x9, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x10, x10, #0x3\n"
+      "sub x28, x9, #0x2\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "ldr x25, [x26, #0x0]\n"
+      "and x24, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x10, x10, #0x2\n"
+      "ldr x23, [x26, #0x8]\n"
       "ptrue p11.s\n"
       "zip1 p10.b, p9.b, p8.b\n"
-      "mov x21, %x[row_offset]\n"
-      "mov x20, %x[out]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "add x25, x25, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x27, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe0150300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0160320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
       ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "ldr x24, [x25, #0x0]\n"
-      ".inst 0xe01502c4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x22, x21]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe01602e4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
       "add x12, x12, #0x8\n"
-      "cmp x12, x27, LSL #2\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
+      "cmp x12, x28, LSL #2\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe0150300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0160320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
       ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01502c4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x22, x21]\n"
-      "ldr x24, [x25, #0x0]\n"
-      "incb x21\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "incb x26\n"
-      "cbz x19, 8f\n"
-      "mov x19, x19\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe01602e4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "incb x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "incb x27\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x27, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
       ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0bc8281  // st1w { za0v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
+      ".inst 0xe0a982a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "add x13, x13, #0x8\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "incb x26\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe0bc8281  // st1w { za0v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "addvl x20, x20, #2\n"
-      "incb x21\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "incb x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0a982a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incb x22\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x27, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
       ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe0152300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x21]\n"
+      ".inst 0xe0162320  // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
       ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0bc8289  // st1w { za2v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
+      ".inst 0xe0a982a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "add x13, x13, #0x8\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
       ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe0152300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x21]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0162320  // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe01526c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x21]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe01626e4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "subs x19, x19, #0x1\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe0bc8289  // st1w { za2v.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "addvl x20, x20, #2\n"
-      "incb x26\n"
-      "incb x21\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0a982a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incb x27\n"
+      "incb x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x23, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.b, x27, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      "ldr x25, [x26, #0x0]\n"
       "add x12, x12, #0x1\n"
       ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      "cmp x12, x28\n"
-      ".inst 0xe0152302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x21]\n"
-      "add x25, x25, #0x8\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x9\n"
+      ".inst 0xe0162322  // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
       "add x13, x13, #0x4\n"
       "blt 9b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x27, %x[width]\n"
+      "whilelt p8.b, x27, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8288  // st1w { za2v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
-      "add x19, x19, #0x4\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
+      "add x20, x20, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8280  // st1w { za0v.s[x12] }, p0/Z, [x20, XZR, LSL #2]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x20\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp
index 231d7ae..646db0c 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,220 +32,220 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
       "mov z18.b, #0x1\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
       "mov z17.s, #0x0\n"
-      "cntb x20\n"
-      "cntw x10\n"
-      "ptrue p1.b\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x9, x19, #0x1\n"
-      "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x27, x20, #0x1\n"
-      "ands x27, x19, x27\n"
-      "csel x27, x27, x20, NE\n"
-      "add x27, x27, #0x3\n"
-      "lsr x27, x27, #0x2\n"
-      "sub x26, x10, #0x2\n"
+      "sub x10, x21, #0x1\n"
+      "cntw x9\n"
+      "sub x23, x23, #0x1\n"
+      "ands x10, x20, x10\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x10, x10, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x9, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x10, x10, #0x3\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x28, #0x0\n"
+      "ptrue p2.b\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x27, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x10, x10, #0x2\n"
+      "sub x26, x9, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x10, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
       "zip1 p10.b, p9.b, p8.b\n"
       "mov x25, %x[row_offset]\n"
       "mov x24, %x[out]\n"
-      "mov x23, #0x0\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "cbnz %x[first], 1f\n"
       "addvl x24, x24, #-1\n"
-      "ld1w { z17.s }, p1/Z, [x24]\n"
+      "ld1w { z17.s }, p2/Z, [x24]\n"
       "1:"  // K loop: Load row sums: End
-      "mov x22, %x[in]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      "mov x23, %x[in]\n"
+      "ldr x22, [x23, #0x0]\n"
       "mov x12, #0x0\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
       "cbz x26, 3f\n"
       "2:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01902a0  // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25646140  // dup p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0190284  // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01902c0  // ld1b { za0h.b[x12] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01902a4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x21, x25]\n"
       "add x12, x12, #0x8\n"
       "cmp x12, x26, LSL #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
       "blt 2b\n"
       "3:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01902a0  // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25646140  // dup p0.b, p8.b/Z, p10.b[w12, #4]\n"
-      "mov x22, %x[in]\n"
-      ".inst 0xe0190284  // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01902c0  // ld1b { za0h.b[x12] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25646140  // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0xe01902a4  // ld1b { za0h.b[x12, #4] }, p0/Z, [x21, x25]\n"
+      "ldr x22, [x23, #0x0]\n"
       "incb x25\n"
-      "incb x23\n"
-      "cbz x9, 9f\n"
-      "mov x19, x9\n"
+      "ldr x21, [x23, #0x8]\n"
+      "add x23, x23, #0x10\n"
+      "incb x28\n"
+      "cbz x20, 9f\n"
+      "mov x20, x20\n"
       "4:"  // K loop: Main loop
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x26, 6f\n"
       "5:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25756140  // dup p0.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192286  // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01922a6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       "udot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828430  // mova z16.s, p1/M, za0v.s[x12, #1]\n"
-      "addvl x24, x24, #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828830  // mova z16.s, p2/M, za0v.s[x12, #1]\n"
+      ".inst 0xe0a98301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x26\n"
       "udot z17.s, z16.b, z18.b\n"
+      "add x23, x23, #0x10\n"
+      "addvl x24, x24, #2\n"
+      "add x13, x13, #0x8\n"
       "blt 5b\n"
       "6:"  // K loop: Main loop: First: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25756140  // dup p0.b, p8.b/Z, p10.b[w13, #6]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192286  // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0xe01922a6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "udot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828430  // mova z16.s, p1/M, za0v.s[x12, #1]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828830  // mova z16.s, p2/M, za0v.s[x12, #1]\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8700  // st1w { za0v.s[x12] }, p1/Z, [x24, XZR, LSL #2]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "incb x28\n"
+      "add x23, x23, #0x10\n"
+      ".inst 0xe0a98301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
+      "udot z17.s, z16.b, z18.b\n"
       "addvl x24, x24, #2\n"
-      "incb x23\n"
       "incb x25\n"
-      "udot z17.s, z16.b, z18.b\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x26, 8f\n"
       "7:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01922a0  // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25656140  // dup p0.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192284  // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xe01922a4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       "udot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828530  // mova z16.s, p1/M, za2v.s[x12, #1]\n"
-      "addvl x24, x24, #2\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828930  // mova z16.s, p2/M, za2v.s[x12, #1]\n"
+      ".inst 0xe0a98309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x26\n"
       "udot z17.s, z16.b, z18.b\n"
+      "add x23, x23, #0x10\n"
+      "addvl x24, x24, #2\n"
+      "add x13, x13, #0x8\n"
       "blt 7b\n"
       "8:"  // K loop: Main loop: Second: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01922a0  // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n"
-      ".inst 0x25656140  // dup p0.b, p8.b/Z, p10.b[w13, #4]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0192284  // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x25]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0xe01922a4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x21, x25]\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
       "udot z17.s, z16.b, z18.b\n"
-      ".inst 0xc0828530  // mova z16.s, p1/M, za2v.s[x12, #1]\n"
+      "mov x23, %x[in]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0828930  // mova z16.s, p2/M, za2v.s[x12, #1]\n"
+      "ldr x21, [x23, #0x8]\n"
+      ".inst 0xe0bf8708  // st1w { za2v.s[x12] }, p1/Z, [x24, XZR, LSL #2]\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x23, x23, #0x10\n"
+      ".inst 0xe0a98309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x9, LSL #2]\n"
+      "udot z17.s, z16.b, z18.b\n"
       "addvl x24, x24, #2\n"
-      "incb x23\n"
+      "incb x28\n"
       "incb x25\n"
-      "udot z17.s, z16.b, z18.b\n"
-      "subs x19, x19, #0x1\n"
       "bgt 4b\n"
       "9:"  // K loop: Tails
-      "cbnz x28, 12f\n"
-      "mov x22, %x[in]\n"
-      "whilelt p8.b, x23, %x[width]\n"
+      "cbnz x27, 12f\n"
+      "mov x23, %x[in]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: First
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      "addvl x24, x24, #1\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe01922a2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n"
-      "udot z17.s, z16.b, z18.b\n"
-      "add x22, x22, #0x8\n"
-      "add x13, x13, #0x4\n"
+      "ldr x22, [x23, #0x0]\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      "udot z17.s, z16.b, z18.b\n"
+      ".inst 0xe01922c2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x22, x25]\n"
+      "cmp x12, x9\n"
+      "add x23, x23, #0x8\n"
+      "addvl x24, x24, #1\n"
+      "add x13, x13, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x23, %x[width]\n"
-      "whilelt p8.b, x23, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "11:"  // K loop: Tails: Even: Second
-      ".inst 0xc0828510  // mova z16.s, p1/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
-      "add x19, x19, #0x4\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
       "add x12, x12, #0x1\n"
+      "cmp x12, x10\n"
       "udot z17.s, z16.b, z18.b\n"
-      "cmp x12, x27\n"
+      "addvl x24, x24, #1\n"
+      "add x20, x20, #0x4\n"
       "blt 11b\n"
-      "whilelt p9.b, x23, %x[width]\n"
+      "whilelt p9.b, x28, %x[width]\n"
       "b 14f\n"
       "12:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "13:"  // K loop: Tails: Odd: Loop
-      ".inst 0xc0828410  // mova z16.s, p1/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
+      ".inst 0xc0828810  // mova z16.s, p2/M, za0v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x10\n"
       "udot z17.s, z16.b, z18.b\n"
+      "addvl x24, x24, #1\n"
       "blt 13b\n"
       "14:"  // K loop: End
-      "st1w { z17.s }, p1, [x24]\n"
+      "st1w { z17.s }, p2, [x24]\n"
       "addvl x24, x24, #1\n"
       "mov %x[out], x24\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp16_fp16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp16_fp16.hpp
index f80ca64..788c1a2 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp16_fp16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp16_fp16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,175 +34,175 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "mov x21, %x[width]\n"
+      "inch x21\n"
+      "cnth x11\n"
+      "sub x21, x21, #0x1\n"
+      "udiv x21, x21, x11\n"  // n_passes = ceildiv(width, VL<T>)
       "mov x20, %x[width]\n"
-      "inch x20\n"
-      "cnth x10\n"
-      "sub x20, x20, #0x1\n"
-      "udiv x20, x20, x10\n"  // n_passes = ceildiv(width, VL<T>)
-      "mov x19, %x[width]\n"
-      "sub x9, x10, #0x1\n"
-      "sub x28, x20, #0x1\n"
-      "ands x9, x19, x9\n"
-      "sub x27, x10, #0x2\n"
-      "lsl x19, %x[height], #0x1\n"  // height * 2
-      "mov x26, #0x0\n"
-      "mov x25, %x[in]\n"
-      "lsr x28, x28, #0x1\n"  // n_loops = (n_passes - 1) / 2
-      "ldr x24, [x25, #0x0]\n"
-      "and x23, x20, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
-      "csel x9, x9, x10, NE\n"
-      "ldr x22, [x25, #0x8]\n"
+      "sub x10, x11, #0x1\n"
+      "sub x9, x21, #0x1\n"
+      "ands x10, x20, x10\n"
+      "sub x28, x11, #0x2\n"
+      "lsl x20, %x[height], #0x1\n"  // height * 2
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "lsr x9, x9, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "ldr x25, [x26, #0x0]\n"
+      "and x24, x21, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "csel x10, x10, x11, NE\n"
+      "ldr x23, [x26, #0x8]\n"
       "ptrue p11.h\n"
-      "whilelt p10.h, XZR, x19\n"
-      "mov x21, %x[row_offset]\n"
-      "mov x20, %x[out]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "add x25, x25, #0x10\n"
+      "whilelt p10.h, XZR, x20\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x27, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386140  // psel p0.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
-      ".inst 0xe05502c1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x22, x21, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe05602e1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x23, x22, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
+      "cmp x12, x28\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386140  // psel p0.h, p8.h/Z, p10.h[w12, #1]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe05502c1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x22, x21, LSL #1]\n"
-      "ldr x24, [x25, #0x0]\n"
-      "inch x21\n"
-      "ldr x22, [x25, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "inch x26\n"
-      "cbz x28, 8f\n"
-      "mov x19, x28\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe05602e1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x23, x22, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "inch x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "inch x27\n"
+      "cbz x9, 8f\n"
+      "mov x20, x9\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x27, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      ".inst 0xe06a8281  // st1h { za0v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
+      ".inst 0xe06b82a1  // st1h { za0v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e9  // ld1h { za1h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "inch x26\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe06a8281  // st1h { za0v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
-      "addvl x20, x20, #2\n"
-      "inch x21\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "inch x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe06b82a1  // st1h { za0v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
+      "addvl x21, x21, #2\n"
+      "inch x22\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x27, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      ".inst 0xe06a8289  // st1h { za1v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
+      ".inst 0xe06b82a9  // st1h { za1v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27\n"
-      "add x25, x25, #0x10\n"
-      "addvl x20, x20, #2\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550300  // ld1h { za0h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe0560320  // ld1h { za0h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25386141  // psel p1.h, p8.h/Z, p10.h[w12, #1]\n"
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe05506c1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x22, x21, LSL #1]\n"
-      "ldr x22, [x25, #0x8]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe05606e1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       ".inst 0x25386d20  // psel p0.h, p11.h/Z, p9.h[w12, #1]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "subs x19, x19, #0x1\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xe06a8289  // st1h { za1v.h[x12, #1] }, p0/Z, [x20, x10, LSL #1]\n"
-      "addvl x20, x20, #2\n"
-      "inch x26\n"
-      "inch x21\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe06b82a9  // st1h { za1v.h[x12, #1] }, p0/Z, [x21, x11, LSL #1]\n"
+      "addvl x21, x21, #2\n"
+      "inch x27\n"
+      "inch x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x23, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
-      "ldr x24, [x25, #0x0]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
+      "ldr x25, [x26, #0x0]\n"
       ".inst 0x25286140  // psel p0.h, p8.h/Z, p10.h[w12]\n"
-      ".inst 0xe0550308  // ld1h { za1h.h[x12] }, p0/Z, [x24, x21, LSL #1]\n"
+      ".inst 0xe0560328  // ld1h { za1h.h[x12] }, p0/Z, [x25, x22, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
-      "add x25, x25, #0x8\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x11\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
       "blt 9b\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
+      "whilelt p8.h, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8288  // st1h { za1v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe07f82a8  // st1h { za1v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 10b\n"
-      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p9.h, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
       ".inst 0x25286d20  // psel p0.h, p11.h/Z, p9.h[w12]\n"
-      ".inst 0xe07f8280  // st1h { za0v.h[x12] }, p0/Z, [x20, XZR, LSL #1]\n"
+      ".inst 0xe07f82a0  // st1h { za0v.h[x12] }, p0/Z, [x21, XZR, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x9\n"
-      "addvl x20, x20, #1\n"
+      "cmp x12, x10\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x20\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp
index 874fc79..7de8854 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,174 +34,174 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "mov x22, %x[width]\n"
+      "incw x22\n"
       "cntw x10\n"
-      "mov x19, %x[width]\n"
-      "incw x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x10\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x9, x19, #0x1\n"
-      "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x27, x10, #0x1\n"
-      "ands x27, x19, x27\n"
-      "csel x27, x27, x10, NE\n"
-      "sub x26, x10, #0x2\n"
+      "sub x22, x22, #0x1\n"
+      "udiv x22, x22, x10\n"  // n_passes = ceildiv(width, VL<T>)
+      "mov x21, %x[width]\n"
+      "sub x9, x10, #0x1\n"
+      "sub x20, x22, #0x1\n"
+      "ands x9, x21, x9\n"
+      "sub x28, x10, #0x2\n"
+      "mov x27, #0x0\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x24, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x23, [x26, #0x8]\n"
+      "csel x9, x9, x10, NE\n"
       "ptrue p11.s\n"
       "whilelt p10.s, XZR, %x[height]\n"
-      "mov x25, %x[row_offset]\n"
-      "mov x24, %x[out]\n"
-      "mov x23, #0x0\n"
-      "whilelt p9.s, x23, %x[width]\n"
-      "whilelt p8.s, x23, %x[width]\n"
-      "mov x22, %x[in]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.s, x27, %x[width]\n"
+      "whilelt p8.s, x27, %x[width]\n"
+      "add x26, x26, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x26, 2f\n"
+      "cbz x28, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a0  // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0990281  // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      ".inst 0x25706140  // psel p0.s, p8.s/Z, p10.s[w12, #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0xe09602e1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x26\n"
+      "cmp x12, x28\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a0  // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "mov x22, %x[in]\n"
-      ".inst 0xe0990281  // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x21, [x22, #0x0]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      "incw x25\n"
-      "incw x23\n"
-      "cbz x9, 8f\n"
-      "mov x19, x9\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      ".inst 0x25706140  // psel p0.s, p8.s/Z, p10.s[w12, #1]\n"
+      "mov x26, %x[in]\n"
+      ".inst 0xe09602e1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n"
+      "ldr x25, [x26, #0x0]\n"
+      "incw x22\n"
+      "ldr x23, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "incw x27\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.s, x23, %x[width]\n"
+      "whilelt p8.s, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x26, 5f\n"
+      "cbz x28, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a8  // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0990289  // ld1w { za2h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960328  // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      ".inst 0x25706141  // psel p1.s, p8.s/Z, p10.s[w12, #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe09606e9  // ld1w { za2h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0aa82a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x26\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a8  // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0990289  // ld1w { za2h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.s, x23, %x[width]\n"
-      ".inst 0xe0aa8301  // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
-      "incw x23\n"
-      "incw x25\n"
-      "whilelt p8.s, x23, %x[width]\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960328  // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25706141  // psel p1.s, p8.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe09606e9  // ld1w { za2h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "whilelt p9.s, x27, %x[width]\n"
+      "incw x27\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0aa82a1  // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incw x22\n"
+      "whilelt p8.s, x27, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x26, 7f\n"
+      "cbz x28, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a0  // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0990281  // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      ".inst 0x25706141  // psel p1.s, p8.s/Z, p10.s[w12, #1]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe09606e1  // ld1w { za0h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0aa82a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x26\n"
+      "cmp x12, x28\n"
+      "add x26, x26, #0x10\n"
+      "addvl x21, x21, #2\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x22, %x[in]\n"
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      ".inst 0xe09902a0  // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      ".inst 0x25706140  // dup p0.s, p8.s/Z, p10.s[w12, #1]\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe0990281  // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n"
-      "ldr x20, [x22, #0x8]\n"
-      "add x22, x22, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.s, x23, %x[width]\n"
-      ".inst 0xe0aa8309  // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n"
-      "addvl x24, x24, #2\n"
-      "incw x23\n"
-      "incw x25\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
+      "mov x26, %x[in]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25706141  // psel p1.s, p8.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe09606e1  // ld1w { za0h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x26, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "whilelt p9.s, x27, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe0aa82a9  // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n"
+      "addvl x21, x21, #2\n"
+      "incw x27\n"
+      "incw x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x28, 11f\n"
-      "mov x22, %x[in]\n"
-      "whilelt p8.s, x23, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.s, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      ".inst 0x25306140  // dup p0.s, p8.s/Z, p10.s[w12]\n"
-      "addvl x24, x24, #1\n"
-      "ldr x21, [x22, #0x0]\n"
-      ".inst 0xe09902a8  // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n"
-      "add x22, x22, #0x8\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      "ldr x25, [x26, #0x0]\n"
+      ".inst 0x25306140  // psel p0.s, p8.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960328  // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n"
       "add x12, x12, #0x1\n"
       "cmp x12, x10\n"
+      "add x26, x26, #0x8\n"
+      "addvl x21, x21, #1\n"
       "blt 9b\n"
-      "whilelt p9.s, x23, %x[width]\n"
-      "whilelt p8.s, x23, %x[width]\n"
+      "whilelt p9.s, x27, %x[width]\n"
+      "whilelt p8.s, x27, %x[width]\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8308  // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x9\n"
+      "addvl x21, x21, #1\n"
       "blt 10b\n"
-      "whilelt p9.s, x23, %x[width]\n"
+      "whilelt p9.s, x27, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8300  // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n"
-      "addvl x24, x24, #1\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x27\n"
+      "cmp x12, x9\n"
+      "addvl x21, x21, #1\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x24\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_bf16_bf16.hpp
index 61fed43..14ee5d6 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,66 +34,66 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cnth x28\n"
+      "cmp %x[height], x28\n"
       "cnth x27\n"
-      "cmp %x[height], x27\n"
-      "cnth x26\n"
-      "csel x27, %x[height], x27, LT\n"
-      "mov x25, #0x0\n"
+      "csel x28, %x[height], x28, LT\n"
+      "mov x26, #0x0\n"
       "ptrue p13.s\n"
-      "sub x27, x27, #0x1\n"
+      "sub x28, x28, #0x1\n"
       "whilelt p12.h, XZR, %x[height]\n"
-      "whilelt p11.h, x26, %x[height]\n"
-      "mov x24, %x[row_offset]\n"
-      "mov x23, %x[out]\n"
-      "whilelt p10.h, x25, %x[width]\n"
-      "whilelt p9.h, x25, %x[width]\n"
-      "whilelt p8.h, x25, %x[width]\n"
+      "whilelt p11.h, x27, %x[height]\n"
+      "mov x25, %x[row_offset]\n"
+      "mov x24, %x[out]\n"
+      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p8.h, x26, %x[width]\n"
       "1:"  // Width loop
-      "add x22, %x[in], XZR, LSL #3\n"
-      "add x19, %x[in], x26, LSL #3\n"
-      "ldr x21, [x22], #0x8\n"
+      "add x23, %x[in], XZR, LSL #3\n"
+      "add x20, %x[in], x27, LSL #3\n"
+      "ldr x22, [x23], #0x8\n"
       "mov x12, #0x0\n"
-      "ldr x20, [x19], #0x8\n"
-      "cbz x27, 3f\n"
+      "ldr x21, [x20], #0x8\n"
+      "cbz x28, 3f\n"
       "2:"  // Loads: Loop
       ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
       ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05806a0  // ld1h { za0h.h[x12] }, p1/Z, [x21, x24, LSL #1]\n"
-      "ldr x21, [x22], #0x8\n"
-      ".inst 0xe0580288  // ld1h { za1h.h[x12] }, p0/Z, [x20, x24, LSL #1]\n"
+      ".inst 0xe05906c0  // ld1h { za0h.h[x12] }, p1/Z, [x22, x25, LSL #1]\n"
+      "ldr x22, [x23], #0x8\n"
+      ".inst 0xe05902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27, LSL #1\n"
-      "ldr x20, [x19], #0x8\n"
+      "cmp x12, x28, LSL #1\n"
+      "ldr x21, [x20], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
-      "sub x19, %x[width], x25\n"
+      "sub x20, %x[width], x26\n"
       ".inst 0x25286580  // psel p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05802a0  // ld1h { za0h.h[x12] }, p0/Z, [x21, x24, LSL #1]\n"
+      ".inst 0xe05902c0  // ld1h { za0h.h[x12] }, p0/Z, [x22, x25, LSL #1]\n"
       ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
-      "cmp x19, x26\n"
-      ".inst 0xe0580288  // ld1h { za1h.h[x12] }, p0/Z, [x20, x24, LSL #1]\n"
+      "cmp x20, x27\n"
+      ".inst 0xe05902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
       "mov x12, #0x0\n"
-      "csel x19, x19, x26, LT\n"
+      "csel x20, x20, x27, LT\n"
       "4:"  // Stores: Loop
       ".inst 0x25287540  // psel p0.h, p13.h/Z, p10.h[w12]\n"
-      ".inst 0xe07f82e0  // st1h { za0v.h[x12] }, p0/Z, [x23, XZR, LSL #1]\n"
+      ".inst 0xe07f8300  // st1h { za0v.h[x12] }, p0/Z, [x24, XZR, LSL #1]\n"
       ".inst 0x25287540  // psel p0.h, p13.h/Z, p10.h[w12]\n"
-      ".inst 0xe07a82e8  // st1h { za1v.h[x12] }, p0/Z, [x23, x26, LSL #1]\n"
+      ".inst 0xe07b8308  // st1h { za1v.h[x12] }, p0/Z, [x24, x27, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
-      "addvl x23, x23, #4\n"
+      "cmp x12, x20\n"
+      "addvl x24, x24, #4\n"
       "blt 4b\n"
+      "inch x26\n"
+      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p8.h, x26, %x[width]\n"
       "inch x25\n"
-      "whilelt p10.h, x25, %x[width]\n"
-      "whilelt p9.h, x25, %x[width]\n"
-      "whilelt p8.h, x25, %x[width]\n"
-      "inch x24\n"
       "b.any 1b\n"
-      "mov %x[out], x23\n"
+      "mov %x[out], x24\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp
index fc7596e..f648ccf 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,269 +34,269 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cnth x20\n"
+      "cnth x21\n"
+      "mov x22, %x[width]\n"
+      "inch x22\n"
+      "mov x20, %x[width]\n"
+      "sub x17, x21, #0x1\n"
+      "sub x22, x22, #0x1\n"
+      "ands x17, x20, x17\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "mov x19, %x[width]\n"
-      "inch x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x1\n"
-      "lsr x10, x10, #0x1\n"
-      "sub x9, x16, #0x2\n"
+      "udiv x22, x22, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "sub x13, x22, #0x1\n"
+      "add x17, x17, #0x1\n"
+      "sub x15, x16, #0x2\n"
+      "lsl x21, %x[height], #0x1\n"  // height * 2
+      "lsl x20, x16, #0x1\n"
+      "mov x14, #0x0\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      "cntw x28, ALL, MUL #2\n"
+      "cntw x27, ALL, MUL #3\n"
+      "ldr x26, [x10, #0x0]\n"
+      "lsr x13, x13, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x25, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x24, [x11, #0x8]\n"
+      "lsr x17, x17, #0x1\n"
       "ptrue p13.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p12.h, XZR, x20\n"
-      "whilelt p11.h, x19, x20\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "whilelt p12.h, XZR, x21\n"
+      "whilelt p11.h, x20, x21\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x9, 2f\n"
+      "cbz x15, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05c02e0  // ld1h { za0h.h[x12] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05c02c8  // ld1h { za1h.h[x12] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25686580  // dup p0.h, p9.h/Z, p12.h[w12, #2]\n"
-      ".inst 0xe05c02a2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25686160  // dup p0.h, p8.h/Z, p11.h[w12, #2]\n"
-      ".inst 0xe05c028a  // ld1h { za1h.h[x12, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0560520  // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0560348  // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25686581  // psel p1.h, p9.h/Z, p12.h[w12, #2]\n"
+      ".inst 0x25686160  // psel p0.h, p8.h/Z, p11.h[w12, #2]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0560702  // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe05602ea  // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
       "add x12, x12, #0x4\n"
-      "cmp x12, x9, LSL #1\n"
+      "cmp x12, x15, LSL #1\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05c02e0  // ld1h { za0h.h[x12] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05c02c8  // ld1h { za1h.h[x12] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25686580  // dup p0.h, p9.h/Z, p12.h[w12, #2]\n"
-      ".inst 0xe05c02a2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25686160  // dup p0.h, p8.h/Z, p11.h[w12, #2]\n"
-      ".inst 0xe05c028a  // ld1h { za1h.h[x12, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      "inch x28\n"
-      "inch x26\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0560520  // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0560348  // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25686581  // psel p1.h, p9.h/Z, p12.h[w12, #2]\n"
+      ".inst 0x25686160  // psel p0.h, p8.h/Z, p11.h[w12, #2]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0xe0560702  // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe05602ea  // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      "inch x22\n"
+      "inch x14\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "cbz x13, 8f\n"
-      "mov x19, x13\n"
+      "mov x20, x13\n"
       "3:"  // K loop: Main loop
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 5f\n"
+      "cbz x15, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25396580  // dup p0.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe05c22e1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25796580  // dup p0.h, p9.h/Z, p12.h[w13, #3]\n"
-      ".inst 0xe05c22a3  // ld1h { za0h.h[x13, #3] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25796160  // dup p0.h, p8.h/Z, p11.h[w13, #3]\n"
-      ".inst 0xe05c228b  // ld1h { za1h.h[x13, #3] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25796580  // psel p0.h, p9.h/Z, p12.h[w13, #3]\n"
+      ".inst 0x25796162  // psel p2.h, p8.h/Z, p11.h[w13, #3]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562303  // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562aeb  // ld1h { za1h.h[x13, #3] }, p2/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a0  // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x4\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25396580  // dup p0.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe05c22e1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25796580  // dup p0.h, p9.h/Z, p12.h[w13, #3]\n"
-      ".inst 0xe05c22a3  // ld1h { za0h.h[x13, #3] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25796160  // dup p0.h, p8.h/Z, p11.h[w13, #3]\n"
-      ".inst 0xe05c228b  // ld1h { za1h.h[x13, #3] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "inch x26\n"
-      "inch x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25796580  // psel p0.h, p9.h/Z, p12.h[w13, #3]\n"
+      ".inst 0x25796161  // psel p1.h, p8.h/Z, p11.h[w13, #3]\n"
+      ".inst 0xe0562303  // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe05626eb  // ld1h { za1h.h[x13, #3] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aa4  // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "inch x14\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "inch x22\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 7f\n"
+      "cbz x15, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25296580  // dup p0.h, p9.h/Z, p12.h[w13]\n"
-      ".inst 0xe05c22e0  // ld1h { za0h.h[x13] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25296160  // dup p0.h, p8.h/Z, p11.h[w13]\n"
-      ".inst 0xe05c22c8  // ld1h { za1h.h[x13] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25696580  // dup p0.h, p9.h/Z, p12.h[w13, #2]\n"
-      ".inst 0xe05c22a2  // ld1h { za0h.h[x13, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25696160  // dup p0.h, p8.h/Z, p11.h[w13, #2]\n"
-      ".inst 0xe05c228a  // ld1h { za1h.h[x13, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25296581  // psel p1.h, p9.h/Z, p12.h[w13]\n"
+      ".inst 0x25296160  // psel p0.h, p8.h/Z, p11.h[w13]\n"
+      ".inst 0xe0562520  // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0562348  // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25696580  // psel p0.h, p9.h/Z, p12.h[w13, #2]\n"
+      ".inst 0x25696162  // psel p2.h, p8.h/Z, p11.h[w13, #2]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562302  // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562aea  // ld1h { za1h.h[x13, #2] }, p2/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a8  // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x4\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25296580  // dup p0.h, p9.h/Z, p12.h[w13]\n"
-      ".inst 0xe05c22e0  // ld1h { za0h.h[x13] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25296160  // dup p0.h, p8.h/Z, p11.h[w13]\n"
-      ".inst 0xe05c22c8  // ld1h { za1h.h[x13] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25696580  // dup p0.h, p9.h/Z, p12.h[w13, #2]\n"
-      ".inst 0xe05c22a2  // ld1h { za0h.h[x13, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25696160  // dup p0.h, p8.h/Z, p11.h[w13, #2]\n"
-      ".inst 0xe05c228a  // ld1h { za1h.h[x13, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "inch x26\n"
-      "inch x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25296581  // psel p1.h, p9.h/Z, p12.h[w13]\n"
+      ".inst 0x25296160  // psel p0.h, p8.h/Z, p11.h[w13]\n"
+      ".inst 0xe0562520  // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562348  // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25696580  // psel p0.h, p9.h/Z, p12.h[w13, #2]\n"
+      ".inst 0x25696161  // psel p1.h, p8.h/Z, p11.h[w13, #2]\n"
+      ".inst 0xe0562302  // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe05626ea  // ld1h { za1h.h[x13, #2] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aac  // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "inch x14\n"
+      "inch x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x11, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "cbnz x25, 11f\n"
+      "mov x11, %x[in]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25396581  // dup p1.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      "addvl x27, x27, #2\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe05c26e1  // ld1h { za0h.h[x13, #1] }, p1/Z, [x23, x28, LSL #1]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      "ldr x9, [x11, #0x0]\n"
       "add x12, x12, #0x1\n"
-      "add x25, x25, #0x8\n"
-      "add x13, x13, #0x2\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      "ldr x26, [x11, x16, LSL #0x3]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
       "cmp x12, x16\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      "add x11, x11, #0x8\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x2\n"
       "blt 9b\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      "add x19, x19, #0x2\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
+      "add x20, x20, #0x2\n"
       "blt 10b\n"
-      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p10.h, x14, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x27\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "p12", "p13", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_fp16_fp16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_fp16_fp16.hpp
index 67570a1..61536d3 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_fp16_fp16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_fp16_fp16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,269 +34,269 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cnth x20\n"
+      "cnth x21\n"
+      "mov x22, %x[width]\n"
+      "inch x22\n"
+      "mov x20, %x[width]\n"
+      "sub x17, x21, #0x1\n"
+      "sub x22, x22, #0x1\n"
+      "ands x17, x20, x17\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "mov x19, %x[width]\n"
-      "inch x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x1\n"
-      "lsr x10, x10, #0x1\n"
-      "sub x9, x16, #0x2\n"
+      "udiv x22, x22, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "sub x13, x22, #0x1\n"
+      "add x17, x17, #0x1\n"
+      "sub x15, x16, #0x2\n"
+      "lsl x21, %x[height], #0x1\n"  // height * 2
+      "lsl x20, x16, #0x1\n"
+      "mov x14, #0x0\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      "cntw x28, ALL, MUL #2\n"
+      "cntw x27, ALL, MUL #3\n"
+      "ldr x26, [x10, #0x0]\n"
+      "lsr x13, x13, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x25, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x24, [x11, #0x8]\n"
+      "lsr x17, x17, #0x1\n"
       "ptrue p13.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p12.h, XZR, x20\n"
-      "whilelt p11.h, x19, x20\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "whilelt p12.h, XZR, x21\n"
+      "whilelt p11.h, x20, x21\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x9, 2f\n"
+      "cbz x15, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05c02e0  // ld1h { za0h.h[x12] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05c02c8  // ld1h { za1h.h[x12] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25686580  // dup p0.h, p9.h/Z, p12.h[w12, #2]\n"
-      ".inst 0xe05c02a2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25686160  // dup p0.h, p8.h/Z, p11.h[w12, #2]\n"
-      ".inst 0xe05c028a  // ld1h { za1h.h[x12, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0560520  // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0560348  // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25686581  // psel p1.h, p9.h/Z, p12.h[w12, #2]\n"
+      ".inst 0x25686160  // psel p0.h, p8.h/Z, p11.h[w12, #2]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0560702  // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe05602ea  // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
       "add x12, x12, #0x4\n"
-      "cmp x12, x9, LSL #1\n"
+      "cmp x12, x15, LSL #1\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05c02e0  // ld1h { za0h.h[x12] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05c02c8  // ld1h { za1h.h[x12] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25686580  // dup p0.h, p9.h/Z, p12.h[w12, #2]\n"
-      ".inst 0xe05c02a2  // ld1h { za0h.h[x12, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25686160  // dup p0.h, p8.h/Z, p11.h[w12, #2]\n"
-      ".inst 0xe05c028a  // ld1h { za1h.h[x12, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      "inch x28\n"
-      "inch x26\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0560520  // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0560348  // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25686581  // psel p1.h, p9.h/Z, p12.h[w12, #2]\n"
+      ".inst 0x25686160  // psel p0.h, p8.h/Z, p11.h[w12, #2]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0xe0560702  // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe05602ea  // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      "inch x22\n"
+      "inch x14\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "cbz x13, 8f\n"
-      "mov x19, x13\n"
+      "mov x20, x13\n"
       "3:"  // K loop: Main loop
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 5f\n"
+      "cbz x15, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25396580  // dup p0.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe05c22e1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25796580  // dup p0.h, p9.h/Z, p12.h[w13, #3]\n"
-      ".inst 0xe05c22a3  // ld1h { za0h.h[x13, #3] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25796160  // dup p0.h, p8.h/Z, p11.h[w13, #3]\n"
-      ".inst 0xe05c228b  // ld1h { za1h.h[x13, #3] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25796580  // psel p0.h, p9.h/Z, p12.h[w13, #3]\n"
+      ".inst 0x25796162  // psel p2.h, p8.h/Z, p11.h[w13, #3]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562303  // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562aeb  // ld1h { za1h.h[x13, #3] }, p2/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a0  // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x4\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25396580  // dup p0.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe05c22e1  // ld1h { za0h.h[x13, #1] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25796580  // dup p0.h, p9.h/Z, p12.h[w13, #3]\n"
-      ".inst 0xe05c22a3  // ld1h { za0h.h[x13, #3] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25796160  // dup p0.h, p8.h/Z, p11.h[w13, #3]\n"
-      ".inst 0xe05c228b  // ld1h { za1h.h[x13, #3] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "inch x26\n"
-      "inch x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25796580  // psel p0.h, p9.h/Z, p12.h[w13, #3]\n"
+      ".inst 0x25796161  // psel p1.h, p8.h/Z, p11.h[w13, #3]\n"
+      ".inst 0xe0562303  // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe05626eb  // ld1h { za1h.h[x13, #3] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aa4  // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "inch x14\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "inch x22\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 7f\n"
+      "cbz x15, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25296580  // dup p0.h, p9.h/Z, p12.h[w13]\n"
-      ".inst 0xe05c22e0  // ld1h { za0h.h[x13] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25296160  // dup p0.h, p8.h/Z, p11.h[w13]\n"
-      ".inst 0xe05c22c8  // ld1h { za1h.h[x13] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25696580  // dup p0.h, p9.h/Z, p12.h[w13, #2]\n"
-      ".inst 0xe05c22a2  // ld1h { za0h.h[x13, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25696160  // dup p0.h, p8.h/Z, p11.h[w13, #2]\n"
-      ".inst 0xe05c228a  // ld1h { za1h.h[x13, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25296581  // psel p1.h, p9.h/Z, p12.h[w13]\n"
+      ".inst 0x25296160  // psel p0.h, p8.h/Z, p11.h[w13]\n"
+      ".inst 0xe0562520  // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0562348  // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n"
+      ".inst 0x25696580  // psel p0.h, p9.h/Z, p12.h[w13, #2]\n"
+      ".inst 0x25696162  // psel p2.h, p8.h/Z, p11.h[w13, #2]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562302  // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0562aea  // ld1h { za1h.h[x13, #2] }, p2/Z, [x23, x22, LSL #1]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a8  // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x4\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25296580  // dup p0.h, p9.h/Z, p12.h[w13]\n"
-      ".inst 0xe05c22e0  // ld1h { za0h.h[x13] }, p0/Z, [x23, x28, LSL #1]\n"
-      ".inst 0x25296160  // dup p0.h, p8.h/Z, p11.h[w13]\n"
-      ".inst 0xe05c22c8  // ld1h { za1h.h[x13] }, p0/Z, [x22, x28, LSL #1]\n"
-      ".inst 0x25696580  // dup p0.h, p9.h/Z, p12.h[w13, #2]\n"
-      ".inst 0xe05c22a2  // ld1h { za0h.h[x13, #2] }, p0/Z, [x21, x28, LSL #1]\n"
-      ".inst 0x25696160  // dup p0.h, p8.h/Z, p11.h[w13, #2]\n"
-      ".inst 0xe05c228a  // ld1h { za1h.h[x13, #2] }, p0/Z, [x20, x28, LSL #1]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "inch x26\n"
-      "inch x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25296581  // psel p1.h, p9.h/Z, p12.h[w13]\n"
+      ".inst 0x25296160  // psel p0.h, p8.h/Z, p11.h[w13]\n"
+      ".inst 0xe0562520  // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562348  // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25696580  // psel p0.h, p9.h/Z, p12.h[w13, #2]\n"
+      ".inst 0x25696161  // psel p1.h, p8.h/Z, p11.h[w13, #2]\n"
+      ".inst 0xe0562302  // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe05626ea  // ld1h { za1h.h[x13, #2] }, p1/Z, [x23, x22, LSL #1]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aac  // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "inch x14\n"
+      "inch x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x11, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
+      "cbnz x25, 11f\n"
+      "mov x11, %x[in]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25396581  // dup p1.h, p9.h/Z, p12.h[w13, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25396160  // dup p0.h, p8.h/Z, p11.h[w13, #1]\n"
-      "addvl x27, x27, #2\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe05c26e1  // ld1h { za0h.h[x13, #1] }, p1/Z, [x23, x28, LSL #1]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      ".inst 0xe05c22c9  // ld1h { za1h.h[x13, #1] }, p0/Z, [x22, x28, LSL #1]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      "ldr x9, [x11, #0x0]\n"
       "add x12, x12, #0x1\n"
-      "add x25, x25, #0x8\n"
-      "add x13, x13, #0x2\n"
+      ".inst 0x25396581  // psel p1.h, p9.h/Z, p12.h[w13, #1]\n"
+      "ldr x26, [x11, x16, LSL #0x3]\n"
+      ".inst 0x25396160  // psel p0.h, p8.h/Z, p11.h[w13, #1]\n"
       "cmp x12, x16\n"
+      ".inst 0xe0562521  // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n"
+      ".inst 0xe0562349  // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n"
+      "add x11, x11, #0x8\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x2\n"
       "blt 9b\n"
-      "whilelt p10.h, x26, %x[width]\n"
-      "whilelt p9.h, x26, %x[width]\n"
-      "whilelt p8.h, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p10.h, x14, %x[width]\n"
+      "whilelt p9.h, x14, %x[width]\n"
+      "whilelt p8.h, x14, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      "add x19, x19, #0x2\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
+      "add x20, x20, #0x2\n"
       "blt 10b\n"
-      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p10.h, x14, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x27\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "p12", "p13", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8.hpp
index 22f0933..4c701cf 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,265 +34,265 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cntb x20\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
+      "sub x17, x21, #0x1\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x3\n"
-      "lsr x10, x10, #0x2\n"
-      "sub x9, x16, #0x2\n"
+      "sub x23, x23, #0x1\n"
+      "ands x17, x20, x17\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x16, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x17, x17, #0x3\n"
+      "sub x15, x16, #0x2\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x14, #0x0\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      "cntw x28, ALL, MUL #2\n"
+      "cntw x27, ALL, MUL #3\n"
+      "ldr x26, [x10, #0x0]\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x25, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x24, [x11, #0x8]\n"
+      "lsr x17, x17, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
+      "ldr x23, [x10, #0x8]\n"
       "zip1 p10.b, p9.b, p8.b\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x9, 2f\n"
+      "cbz x15, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe0160120  // ld1b { za0h.b[x12] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0160341  // ld1b { za0h.b[x12, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0160704  // ld1b { za0h.b[x12, #4] }, p1/Z, [x24, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe01602e5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x23, x22]\n"
       "add x12, x12, #0x8\n"
-      "cmp x12, x9, LSL #2\n"
+      "cmp x12, x15, LSL #2\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      "incb x28\n"
-      "incb x26\n"
-      "cbz x13, 8f\n"
-      "mov x19, x13\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe0160120  // ld1b { za0h.b[x12] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe0160341  // ld1b { za0h.b[x12, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0xe0160704  // ld1b { za0h.b[x12, #4] }, p1/Z, [x24, x22]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe01602e5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x23, x22]\n"
+      "ldr x26, [x10, #0x0]\n"
+      "incb x22\n"
+      "incb x14\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 5f\n"
+      "cbz x15, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6142  // psel p2.b, p8.b/Z, p10.b[w13, #7]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0162306  // ld1b { za0h.b[x13, #6] }, p0/Z, [x24, x22]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0162ae7  // ld1b { za0h.b[x13, #7] }, p2/Z, [x23, x22]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a0  // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x8\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incb x26\n"
-      "incb x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      "mov x11, %x[in]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162306  // ld1b { za0h.b[x13, #6] }, p0/Z, [x24, x22]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0x257d6141  // psel p1.b, p8.b/Z, p10.b[w13, #7]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01626e7  // ld1b { za0h.b[x13, #7] }, p1/Z, [x23, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b08aa4  // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "incb x14\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incb x22\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 7f\n"
+      "cbz x15, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe0162120  // ld1b { za0h.b[x13] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162341  // ld1b { za0h.b[x13, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6142  // psel p2.b, p8.b/Z, p10.b[w13, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0162304  // ld1b { za0h.b[x13, #4] }, p0/Z, [x24, x22]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0162ae5  // ld1b { za0h.b[x13, #5] }, p2/Z, [x23, x22]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a8  // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x8\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incb x26\n"
-      "incb x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe0162120  // ld1b { za0h.b[x13] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      ".inst 0xe0162341  // ld1b { za0h.b[x13, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      "mov x11, %x[in]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162304  // ld1b { za0h.b[x13, #4] }, p0/Z, [x24, x22]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0x256d6141  // psel p1.b, p8.b/Z, p10.b[w13, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01626e5  // ld1b { za0h.b[x13, #5] }, p1/Z, [x23, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b08aac  // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incb x14\n"
+      "incb x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x11, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x25, 11f\n"
+      "mov x11, %x[in]\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      "addvl x27, x27, #2\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26e2  // ld1b { za0h.b[x13, #2] }, p1/Z, [x23, x28]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      "add x25, x25, #0x8\n"
-      "add x13, x13, #0x4\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      "ldr x26, [x11, x16, LSL #0x3]\n"
       "add x12, x12, #0x1\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
       "cmp x12, x16\n"
+      "add x11, x11, #0x8\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x4\n"
       "blt 9b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "add x19, x19, #0x4\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
+      "add x20, x20, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x14, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x27\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8_summing.hpp
index 81cde6c..25262d3 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,321 +32,321 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
       "mov z20.b, #0x1\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
       "mov z19.s, #0x0\n"
-      "cntb x20\n"
       "mov z18.s, #0x0\n"
+      "sub x17, x21, #0x1\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "ptrue p2.b\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x3\n"
-      "lsr x10, x10, #0x2\n"
+      "sub x23, x23, #0x1\n"
+      "ands x17, x20, x17\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x16, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x17, x17, #0x3\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x15, #0x0\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x11, ALL, MUL #3\n"
+      "ptrue p4.b\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x10, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x17, x17, #0x2\n"
       "sub x9, x16, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
       "zip1 p10.b, p9.b, p8.b\n"
       "mov x28, %x[row_offset]\n"
       "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "cbnz %x[first], 1f\n"
       "addvl x27, x27, #-2\n"
-      "ld1w { z19.s }, p2/Z, [x27]\n"
-      "ld1w { z18.s }, p2/Z, [x27, #1, MUL VL]\n"
+      "ld1w { z19.s }, p4/Z, [x27]\n"
+      "ld1w { z18.s }, p4/Z, [x27, #1, MUL VL]\n"
       "1:"  // K loop: Load row sums: End
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
       "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
       "mov x12, #0x0\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
       "cbz x9, 3f\n"
       "2:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c02e1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0xe01c06c4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x22, x28]\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe01c02a5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x21, x28]\n"
       "add x12, x12, #0x8\n"
       "cmp x12, x9, LSL #2\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
       "blt 2b\n"
       "3:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "add x24, %x[in], x16, LSL #3\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe01c02e1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0xe01c06c4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c02a5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x21, x28]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
       "incb x28\n"
-      "incb x26\n"
-      "cbz x13, 9f\n"
-      "mov x19, x13\n"
+      "incb x15\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
+      "cbz x20, 9f\n"
+      "mov x20, x20\n"
       "4:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x9, 6f\n"
       "5:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6142  // psel p2.b, p8.b/Z, p10.b[w13, #7]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0xe01c22c6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x22, x28]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01c2aa7  // ld1b { za0h.b[x13, #7] }, p2/Z, [x21, x28]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
+      ".inst 0xe0bf8760  // st1w { za0v.s[x12] }, p1/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc0829010  // mova z16.s, p4/M, za0v.s[x12]\n"
+      ".inst 0xc0829091  // mova z17.s, p4/M, za1v.s[x12]\n"
+      "sdot z19.s, z16.b, z20.b\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      ".inst 0xe0ae8361  // st1w { za0v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829030  // mova z16.s, p4/M, za0v.s[x12, #1]\n"
+      ".inst 0xe0ab8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xc08290b1  // mova z17.s, p4/M, za1v.s[x12, #1]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x9\n"
+      "add x26, x26, #0x10\n"
+      "add x25, x25, #0x10\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      "addvl x27, x27, #4\n"
+      "add x13, x13, #0x8\n"
       "blt 5b\n"
       "6:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6140  // psel p0.b, p8.b/Z, p10.b[w13, #7]\n"
+      ".inst 0xe01c26c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22a7  // ld1b { za0h.b[x13, #7] }, p0/Z, [x21, x28]\n"
+      ".inst 0xc0829010  // mova z16.s, p4/M, za0v.s[x12]\n"
+      ".inst 0x25306d23  // psel p3.s, p11.s/Z, p9.s[w12]\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      ".inst 0xc0829091  // mova z17.s, p4/M, za1v.s[x12]\n"
+      "sdot z18.s, z17.b, z20.b\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829030  // mova z16.s, p4/M, za0v.s[x12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
+      ".inst 0xe0bf8f60  // st1w { za0v.s[x12] }, p3/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc08290b1  // mova z17.s, p4/M, za1v.s[x12, #1]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      ".inst 0xe0b08b64  // st1w { za1v.s[x12] }, p2/Z, [x27, x16, LSL #2]\n"
+      "incb x15\n"
+      "add x26, x26, #0x10\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      ".inst 0xe0ae8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x14, LSL #2]\n"
       "add x25, x25, #0x10\n"
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "incb x26\n"
-      "sdot z18.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
       "incb x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0xe0ab8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
       "addvl x27, x27, #4\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x9, 8f\n"
       "7:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01c2300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22e1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6142  // psel p2.b, p8.b/Z, p10.b[w13, #5]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0xe01c22c4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x22, x28]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01c2aa5  // ld1b { za0h.b[x13, #5] }, p2/Z, [x21, x28]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
+      ".inst 0xe0bf8768  // st1w { za2v.s[x12] }, p1/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc0829110  // mova z16.s, p4/M, za2v.s[x12]\n"
+      ".inst 0xc0829191  // mova z17.s, p4/M, za3v.s[x12]\n"
+      "sdot z19.s, z16.b, z20.b\n"
       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      ".inst 0xe0ae8369  // st1w { za2v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829130  // mova z16.s, p4/M, za2v.s[x12, #1]\n"
+      ".inst 0xe0ab836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xc08291b1  // mova z17.s, p4/M, za3v.s[x12, #1]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x9\n"
+      "add x26, x26, #0x10\n"
+      "add x25, x25, #0x10\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      "addvl x27, x27, #4\n"
+      "add x13, x13, #0x8\n"
       "blt 7b\n"
       "8:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01c2300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      ".inst 0xe01c22e1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6140  // psel p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0xe01c26c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22a5  // ld1b { za0h.b[x13, #5] }, p0/Z, [x21, x28]\n"
+      ".inst 0xc0829110  // mova z16.s, p4/M, za2v.s[x12]\n"
+      ".inst 0x25306d23  // psel p3.s, p11.s/Z, p9.s[w12]\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      ".inst 0xc0829191  // mova z17.s, p4/M, za3v.s[x12]\n"
+      "sdot z18.s, z17.b, z20.b\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829130  // mova z16.s, p4/M, za2v.s[x12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
+      ".inst 0xe0bf8f68  // st1w { za2v.s[x12] }, p3/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc08291b1  // mova z17.s, p4/M, za3v.s[x12, #1]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      ".inst 0xe0b08b6c  // st1w { za3v.s[x12] }, p2/Z, [x27, x16, LSL #2]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      ".inst 0xe0ae8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x14, LSL #2]\n"
       "add x25, x25, #0x10\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "incb x26\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      "incb x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      "incb x15\n"
+      ".inst 0xe0ab836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
       "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      "incb x28\n"
       "bgt 4b\n"
       "9:"  // K loop: Tails
-      "cbnz x11, 12f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x10, 12f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: First
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829010  // mova z16.s, p4/M, za0v.s[x12]\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "sdot z18.s, z16.b, z20.b\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26e2  // ld1b { za0h.b[x13, #2] }, p1/Z, [x23, x28]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      "addvl x27, x27, #2\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      "add x25, x25, #0x8\n"
-      "add x13, x13, #0x4\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xc0829091  // mova z17.s, p4/M, za1v.s[x12]\n"
+      "ldr x23, [x26, x16, LSL #0x3]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
       "add x12, x12, #0x1\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
       "cmp x12, x16\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      "add x26, x26, #0x8\n"
+      "addvl x27, x27, #2\n"
+      "add x13, x13, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "11:"  // K loop: Tails: Even: Second
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "add x19, x19, #0x4\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829110  // mova z16.s, p4/M, za2v.s[x12]\n"
       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "sdot z18.s, z16.b, z20.b\n"
+      ".inst 0xc0829191  // mova z17.s, p4/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      "addvl x27, x27, #2\n"
+      "add x20, x20, #0x4\n"
       "blt 11b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x15, %x[width]\n"
       "b 14f\n"
       "12:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "13:"  // K loop: Tails: Odd: Loop
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829010  // mova z16.s, p4/M, za0v.s[x12]\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
-      "sdot z19.s, z17.b, z20.b\n"
-      "sdot z18.s, z16.b, z20.b\n"
+      ".inst 0xc0829091  // mova z17.s, p4/M, za1v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "sdot z19.s, z16.b, z20.b\n"
+      "sdot z18.s, z17.b, z20.b\n"
+      "addvl x27, x27, #2\n"
       "blt 13b\n"
       "14:"  // K loop: End
-      "st1w { z19.s }, p2, [x27]\n"
-      "st1w { z18.s }, p2, [x27, #1, MUL VL]\n"
+      "st1w { z19.s }, p4, [x27]\n"
+      "st1w { z18.s }, p4, [x27, #1, MUL VL]\n"
       "addvl x27, x27, #2\n"
       "mov %x[out], x27\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8.hpp
index cd4a766..683a315 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,265 +34,265 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "cntb x20\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
+      "sub x17, x21, #0x1\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x3\n"
-      "lsr x10, x10, #0x2\n"
-      "sub x9, x16, #0x2\n"
+      "sub x23, x23, #0x1\n"
+      "ands x17, x20, x17\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x16, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x17, x17, #0x3\n"
+      "sub x15, x16, #0x2\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x14, #0x0\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      "cntw x28, ALL, MUL #2\n"
+      "cntw x27, ALL, MUL #3\n"
+      "ldr x26, [x10, #0x0]\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x25, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "ldr x24, [x11, #0x8]\n"
+      "lsr x17, x17, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
+      "ldr x23, [x10, #0x8]\n"
       "zip1 p10.b, p9.b, p8.b\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x9, 2f\n"
+      "cbz x15, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe0160120  // ld1b { za0h.b[x12] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0160341  // ld1b { za0h.b[x12, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0160704  // ld1b { za0h.b[x12, #4] }, p1/Z, [x24, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe01602e5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x23, x22]\n"
       "add x12, x12, #0x8\n"
-      "cmp x12, x9, LSL #2\n"
+      "cmp x12, x15, LSL #2\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      "incb x28\n"
-      "incb x26\n"
-      "cbz x13, 8f\n"
-      "mov x19, x13\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe0160120  // ld1b { za0h.b[x12] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe0160341  // ld1b { za0h.b[x12, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0xe0160704  // ld1b { za0h.b[x12, #4] }, p1/Z, [x24, x22]\n"
+      "mov x11, %x[in]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe01602e5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x23, x22]\n"
+      "ldr x26, [x10, #0x0]\n"
+      "incb x22\n"
+      "incb x14\n"
+      "ldr x24, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      "ldr x23, [x10, #0x8]\n"
+      "add x10, x10, #0x10\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 5f\n"
+      "cbz x15, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6142  // psel p2.b, p8.b/Z, p10.b[w13, #7]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0162306  // ld1b { za0h.b[x13, #6] }, p0/Z, [x24, x22]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0162ae7  // ld1b { za0h.b[x13, #7] }, p2/Z, [x23, x22]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a0  // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x8\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incb x26\n"
-      "incb x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      "mov x11, %x[in]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162306  // ld1b { za0h.b[x13, #6] }, p0/Z, [x24, x22]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0x257d6141  // psel p1.b, p8.b/Z, p10.b[w13, #7]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01626e7  // ld1b { za0h.b[x13, #7] }, p1/Z, [x23, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b08aa4  // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "incb x14\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incb x22\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
-      "cbz x9, 7f\n"
+      "cbz x15, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe0162120  // ld1b { za0h.b[x13] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162341  // ld1b { za0h.b[x13, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6142  // psel p2.b, p8.b/Z, p10.b[w13, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0xe0162304  // ld1b { za0h.b[x13, #4] }, p0/Z, [x24, x22]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0162ae5  // ld1b { za0h.b[x13, #5] }, p2/Z, [x23, x22]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf86a8  // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x10, x10, #0x10\n"
       "add x13, x13, #0x8\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incb x26\n"
-      "incb x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe0162120  // ld1b { za0h.b[x13] }, p0/Z, [x9, x22]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      ".inst 0xe0162341  // ld1b { za0h.b[x13, #1] }, p0/Z, [x26, x22]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      "mov x11, %x[in]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0xe0162304  // ld1b { za0h.b[x13, #4] }, p0/Z, [x24, x22]\n"
+      "add x10, %x[in], x16, LSL #3\n"
+      ".inst 0x256d6141  // psel p1.b, p8.b/Z, p10.b[w13, #5]\n"
+      "ldr x26, [x10, #0x0]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01626e5  // ld1b { za0h.b[x13, #5] }, p1/Z, [x23, x22]\n"
+      "ldr x24, [x11, #0x8]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x23, [x10, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xe0b08aac  // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x10, x10, #0x10\n"
+      ".inst 0xe0bb82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incb x14\n"
+      "incb x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x11, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x25, 11f\n"
+      "mov x11, %x[in]\n"
+      "whilelt p8.b, x14, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      "addvl x27, x27, #2\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26e2  // ld1b { za0h.b[x13, #2] }, p1/Z, [x23, x28]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      "add x25, x25, #0x8\n"
-      "add x13, x13, #0x4\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      "ldr x9, [x11, #0x0]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe0162122  // ld1b { za0h.b[x13, #2] }, p0/Z, [x9, x22]\n"
+      "ldr x26, [x11, x16, LSL #0x3]\n"
       "add x12, x12, #0x1\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe0162343  // ld1b { za0h.b[x13, #3] }, p0/Z, [x26, x22]\n"
       "cmp x12, x16\n"
+      "add x11, x11, #0x8\n"
+      "addvl x21, x21, #2\n"
+      "add x13, x13, #0x4\n"
       "blt 9b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x14, %x[width]\n"
+      "whilelt p8.b, x14, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "add x19, x19, #0x4\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
+      "add x20, x20, #0x4\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x14, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x17\n"
+      "addvl x21, x21, #2\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x27\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8_summing.hpp
index 5a71613..e7571f7 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block4_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,321 +32,321 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntb x21\n"
+      "mov x23, %x[width]\n"
       "mov z20.b, #0x1\n"
+      "incb x23\n"
+      "mov x20, %x[width]\n"
       "mov z19.s, #0x0\n"
-      "cntb x20\n"
       "mov z18.s, #0x0\n"
+      "sub x17, x21, #0x1\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "ptrue p2.b\n"
-      "mov x19, %x[width]\n"
-      "incb x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x20, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x20, NE\n"
-      "add x10, x10, #0x3\n"
-      "lsr x10, x10, #0x2\n"
+      "sub x23, x23, #0x1\n"
+      "ands x17, x20, x17\n"
+      "udiv x23, x23, x21\n"  // n_passes = ceildiv(width, VL<T>)
+      "csel x17, x17, x21, NE\n"
+      "lsl x22, %x[height], #0x1\n"  // height * 2
+      "lsl x21, x16, #0x1\n"
+      "sub x20, x23, #0x1\n"
+      "add x17, x17, #0x3\n"
+      "whilelt p9.b, XZR, x22\n"
+      "whilelt p8.b, x21, x22\n"
+      "mov x15, #0x0\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x11, ALL, MUL #3\n"
+      "ptrue p4.b\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "and x10, x23, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "lsr x17, x17, #0x2\n"
       "sub x9, x16, #0x2\n"
       "ptrue p11.s\n"
-      "lsl x20, %x[height], #0x1\n" // height * 2
-      "lsl x19, x16, #0x1\n"
-      "whilelt p9.b, XZR, x20\n"
-      "whilelt p8.b, x19, x20\n"
       "zip1 p10.b, p9.b, p8.b\n"
       "mov x28, %x[row_offset]\n"
       "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "cbnz %x[first], 1f\n"
       "addvl x27, x27, #-2\n"
-      "ld1w { z19.s }, p2/Z, [x27]\n"
-      "ld1w { z18.s }, p2/Z, [x27, #1, MUL VL]\n"
+      "ld1w { z19.s }, p4/Z, [x27]\n"
+      "ld1w { z18.s }, p4/Z, [x27, #1, MUL VL]\n"
       "1:"  // K loop: Load row sums: End
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
       "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
       "mov x12, #0x0\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
       "cbz x9, 3f\n"
       "2:"  // K loop: Charge: Loop
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c02e1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0xe01c06c4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x22, x28]\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      ".inst 0xe01c02a5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x21, x28]\n"
       "add x12, x12, #0x8\n"
       "cmp x12, x9, LSL #2\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
       "blt 2b\n"
       "3:"  // K loop: Charge: End
-      ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
-      ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
-      "mov x25, %x[in]\n"
-      ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
-      "add x24, %x[in], x16, LSL #3\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe01c02e1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25646141  // psel p1.b, p8.b/Z, p10.b[w12, #4]\n"
+      ".inst 0x256c6140  // psel p0.b, p8.b/Z, p10.b[w12, #5]\n"
+      ".inst 0xe01c06c4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c02a5  // ld1b { za0h.b[x12, #5] }, p0/Z, [x21, x28]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
       "incb x28\n"
-      "incb x26\n"
-      "cbz x13, 9f\n"
-      "mov x19, x13\n"
+      "incb x15\n"
+      "ldr x22, [x26, #0x8]\n"
+      "add x26, x26, #0x10\n"
+      "ldr x21, [x25, #0x8]\n"
+      "add x25, x25, #0x10\n"
+      "cbz x20, 9f\n"
+      "mov x20, x20\n"
       "4:"  // K loop: Main loop
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x9, 6f\n"
       "5:"  // K loop: Main loop: First: Loop
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25756140  // psel p0.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6142  // psel p2.b, p8.b/Z, p10.b[w13, #7]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0xe01c22c6  // ld1b { za0h.b[x13, #6] }, p0/Z, [x22, x28]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01c2aa7  // ld1b { za0h.b[x13, #7] }, p2/Z, [x21, x28]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf8760  // st1w { za0v.s[x12] }, p1/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc0829011  // mova z17.s, p4/M, za0v.s[x12]\n"
+      ".inst 0xc0829090  // mova z16.s, p4/M, za1v.s[x12]\n"
       "udot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
-      "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "udot z19.s, z17.b, z20.b\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
       "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0ae8361  // st1w { za0v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829031  // mova z17.s, p4/M, za0v.s[x12, #1]\n"
+      ".inst 0xe0ab8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xc08290b0  // mova z16.s, p4/M, za1v.s[x12, #1]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x9\n"
+      "add x26, x26, #0x10\n"
+      "add x25, x25, #0x10\n"
+      "udot z19.s, z17.b, z20.b\n"
+      "udot z18.s, z16.b, z20.b\n"
+      "addvl x27, x27, #4\n"
+      "add x13, x13, #0x8\n"
       "blt 5b\n"
       "6:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25756141  // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
+      ".inst 0x257d6140  // psel p0.b, p8.b/Z, p10.b[w13, #7]\n"
+      ".inst 0xe01c26c6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22a7  // ld1b { za0h.b[x13, #7] }, p0/Z, [x21, x28]\n"
+      ".inst 0xc0829011  // mova z17.s, p4/M, za0v.s[x12]\n"
+      ".inst 0x25306d23  // psel p3.s, p11.s/Z, p9.s[w12]\n"
       "udot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
+      ".inst 0xc0829090  // mova z16.s, p4/M, za1v.s[x12]\n"
       "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
+      "ldr x23, [x25, #0x0]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829031  // mova z17.s, p4/M, za0v.s[x12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      "ldr x21, [x25, #0x8]\n"
+      ".inst 0xe0bf8f60  // st1w { za0v.s[x12] }, p3/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc08290b0  // mova z16.s, p4/M, za1v.s[x12, #1]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      ".inst 0xe0b08b64  // st1w { za1v.s[x12] }, p2/Z, [x27, x16, LSL #2]\n"
+      "incb x15\n"
+      "add x26, x26, #0x10\n"
       "udot z19.s, z17.b, z20.b\n"
-      "incb x26\n"
+      ".inst 0xe0ae8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x14, LSL #2]\n"
+      "add x25, x25, #0x10\n"
       "udot z18.s, z16.b, z20.b\n"
       "incb x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0xe0ab8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
       "addvl x27, x27, #4\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "cbz x9, 8f\n"
       "7:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01c2300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22e1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25656140  // psel p0.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6142  // psel p2.b, p8.b/Z, p10.b[w13, #5]\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0xe01c22c4  // ld1b { za0h.b[x13, #4] }, p0/Z, [x22, x28]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe01c2aa5  // ld1b { za0h.b[x13, #5] }, p2/Z, [x21, x28]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf8768  // st1w { za2v.s[x12] }, p1/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc0829111  // mova z17.s, p4/M, za2v.s[x12]\n"
+      ".inst 0xc0829190  // mova z16.s, p4/M, za3v.s[x12]\n"
       "udot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
-      "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "add x13, x13, #0x8\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "udot z19.s, z17.b, z20.b\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
       "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0xe0ae8369  // st1w { za2v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829131  // mova z17.s, p4/M, za2v.s[x12, #1]\n"
+      ".inst 0xe0ab836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xc08291b0  // mova z16.s, p4/M, za3v.s[x12, #1]\n"
       "add x12, x12, #0x2\n"
       "cmp x12, x9\n"
+      "add x26, x26, #0x10\n"
+      "add x25, x25, #0x10\n"
+      "udot z19.s, z17.b, z20.b\n"
+      "udot z18.s, z16.b, z20.b\n"
+      "addvl x27, x27, #4\n"
+      "add x13, x13, #0x8\n"
       "blt 7b\n"
       "8:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
-      ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
-      ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0x25256140  // psel p0.b, p8.b/Z, p10.b[w13]\n"
+      ".inst 0xe01c2300  // ld1b { za0h.b[x13] }, p0/Z, [x24, x28]\n"
+      ".inst 0x252d6140  // psel p0.b, p8.b/Z, p10.b[w13, #1]\n"
+      ".inst 0xe01c22e1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x23, x28]\n"
+      ".inst 0x25656141  // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
+      ".inst 0x256d6140  // psel p0.b, p8.b/Z, p10.b[w13, #5]\n"
+      ".inst 0xe01c26c4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x22, x28]\n"
+      "mov x26, %x[in]\n"
+      "add x25, %x[in], x16, LSL #3\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0xe01c22a5  // ld1b { za0h.b[x13, #5] }, p0/Z, [x21, x28]\n"
+      ".inst 0xc0829111  // mova z17.s, p4/M, za2v.s[x12]\n"
+      ".inst 0x25306d23  // psel p3.s, p11.s/Z, p9.s[w12]\n"
+      "udot z19.s, z17.b, z20.b\n"
+      ".inst 0xc0829190  // mova z16.s, p4/M, za3v.s[x12]\n"
+      "udot z18.s, z16.b, z20.b\n"
       "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
-      "ldr x22, [x24, #0x0]\n"
+      ".inst 0x25306d22  // psel p2.s, p11.s/Z, p9.s[w12]\n"
+      "ldr x22, [x26, #0x8]\n"
+      ".inst 0x25706d21  // psel p1.s, p11.s/Z, p9.s[w12, #1]\n"
+      ".inst 0xc0829131  // mova z17.s, p4/M, za2v.s[x12, #1]\n"
+      ".inst 0x25706d20  // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
       "ldr x21, [x25, #0x8]\n"
-      ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
-      "ldr x20, [x24, #0x8]\n"
+      ".inst 0xe0bf8f68  // st1w { za2v.s[x12] }, p3/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xc08291b0  // mova z16.s, p4/M, za3v.s[x12, #1]\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      ".inst 0xe0b08b6c  // st1w { za3v.s[x12] }, p2/Z, [x27, x16, LSL #2]\n"
+      "subs x20, x20, #0x1\n"
+      "add x26, x26, #0x10\n"
+      "udot z19.s, z17.b, z20.b\n"
+      ".inst 0xe0ae8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x14, LSL #2]\n"
       "add x25, x25, #0x10\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "udot z19.s, z17.b, z20.b\n"
-      ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
       "udot z18.s, z16.b, z20.b\n"
-      ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
-      ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "udot z19.s, z17.b, z20.b\n"
-      "incb x26\n"
-      "udot z18.s, z16.b, z20.b\n"
-      "incb x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
+      "incb x15\n"
+      ".inst 0xe0ab836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x11, LSL #2]\n"
       "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      "incb x28\n"
       "bgt 4b\n"
       "9:"  // K loop: Tails
-      "cbnz x11, 12f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p8.b, x26, %x[width]\n"
+      "cbnz x10, 12f\n"
+      "mov x26, %x[in]\n"
+      "whilelt p8.b, x15, %x[width]\n"
       "mov x13, #0x0\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: First
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829011  // mova z17.s, p4/M, za0v.s[x12]\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "ldr x24, [x26, #0x0]\n"
+      ".inst 0x25356140  // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+      ".inst 0xc0829090  // mova z16.s, p4/M, za1v.s[x12]\n"
+      "ldr x23, [x26, x16, LSL #0x3]\n"
+      ".inst 0xe01c2302  // ld1b { za0h.b[x13, #2] }, p0/Z, [x24, x28]\n"
+      "add x12, x12, #0x1\n"
+      ".inst 0x253d6140  // psel p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      "cmp x12, x16\n"
       "udot z19.s, z17.b, z20.b\n"
       "udot z18.s, z16.b, z20.b\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe01c26e2  // ld1b { za0h.b[x13, #2] }, p1/Z, [x23, x28]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
+      ".inst 0xe01c22e3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x23, x28]\n"
+      "add x26, x26, #0x8\n"
       "addvl x27, x27, #2\n"
-      ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
-      "add x25, x25, #0x8\n"
       "add x13, x13, #0x4\n"
-      "add x12, x12, #0x1\n"
-      "cmp x12, x16\n"
       "blt 10b\n"
-      "whilelt p9.b, x26, %x[width]\n"
-      "whilelt p8.b, x26, %x[width]\n"
-      "mov x19, #0x0\n"
+      "whilelt p9.b, x15, %x[width]\n"
+      "whilelt p8.b, x15, %x[width]\n"
+      "mov x20, #0x0\n"
       "mov x12, #0x0\n"
       "11:"  // K loop: Tails: Even: Second
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "add x19, x19, #0x4\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829111  // mova z17.s, p4/M, za2v.s[x12]\n"
       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0xc0829190  // mova z16.s, p4/M, za3v.s[x12]\n"
+      "add x12, x12, #0x1\n"
+      "cmp x12, x17\n"
       "udot z19.s, z17.b, z20.b\n"
       "udot z18.s, z16.b, z20.b\n"
-      "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "addvl x27, x27, #2\n"
+      "add x20, x20, #0x4\n"
       "blt 11b\n"
-      "whilelt p9.b, x26, %x[width]\n"
+      "whilelt p9.b, x15, %x[width]\n"
       "b 14f\n"
       "12:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "13:"  // K loop: Tails: Odd: Loop
-      ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0829011  // mova z17.s, p4/M, za0v.s[x12]\n"
       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0xc0829090  // mova z16.s, p4/M, za1v.s[x12]\n"
+      "add x12, x12, #0x1\n"
+      "cmp x12, x17\n"
       "udot z19.s, z17.b, z20.b\n"
       "udot z18.s, z16.b, z20.b\n"
-      "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "addvl x27, x27, #2\n"
       "blt 13b\n"
       "14:"  // K loop: End
-      "st1w { z19.s }, p2, [x27]\n"
-      "st1w { z18.s }, p2, [x27, #1, MUL VL]\n"
+      "st1w { z19.s }, p4, [x27]\n"
+      "st1w { z18.s }, p4, [x27, #1, MUL VL]\n"
       "addvl x27, x27, #2\n"
       "mov %x[out], x27\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp16_fp16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp16_fp16.hpp
index 3ea616f..522f310 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp16_fp16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp16_fp16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,66 +34,66 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cnth x28\n"
+      "cmp %x[height], x28\n"
       "cnth x27\n"
-      "cmp %x[height], x27\n"
-      "cnth x26\n"
-      "csel x27, %x[height], x27, LT\n"
-      "mov x25, #0x0\n"
+      "csel x28, %x[height], x28, LT\n"
+      "mov x26, #0x0\n"
       "ptrue p13.s\n"
-      "sub x27, x27, #0x1\n"
+      "sub x28, x28, #0x1\n"
       "whilelt p12.h, XZR, %x[height]\n"
-      "whilelt p11.h, x26, %x[height]\n"
-      "mov x24, %x[row_offset]\n"
-      "mov x23, %x[out]\n"
-      "whilelt p10.h, x25, %x[width]\n"
-      "whilelt p9.h, x25, %x[width]\n"
-      "whilelt p8.h, x25, %x[width]\n"
+      "whilelt p11.h, x27, %x[height]\n"
+      "mov x25, %x[row_offset]\n"
+      "mov x24, %x[out]\n"
+      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p8.h, x26, %x[width]\n"
       "1:"  // Width loop
-      "add x22, %x[in], XZR, LSL #3\n"
-      "add x19, %x[in], x26, LSL #3\n"
-      "ldr x21, [x22], #0x8\n"
+      "add x23, %x[in], XZR, LSL #3\n"
+      "add x20, %x[in], x27, LSL #3\n"
+      "ldr x22, [x23], #0x8\n"
       "mov x12, #0x0\n"
-      "ldr x20, [x19], #0x8\n"
-      "cbz x27, 3f\n"
+      "ldr x21, [x20], #0x8\n"
+      "cbz x28, 3f\n"
       "2:"  // Loads: Loop
       ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
       ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe05806a0  // ld1h { za0h.h[x12] }, p1/Z, [x21, x24, LSL #1]\n"
-      "ldr x21, [x22], #0x8\n"
-      ".inst 0xe0580288  // ld1h { za1h.h[x12] }, p0/Z, [x20, x24, LSL #1]\n"
+      ".inst 0xe05906c0  // ld1h { za0h.h[x12] }, p1/Z, [x22, x25, LSL #1]\n"
+      "ldr x22, [x23], #0x8\n"
+      ".inst 0xe05902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x27, LSL #1\n"
-      "ldr x20, [x19], #0x8\n"
+      "cmp x12, x28, LSL #1\n"
+      "ldr x21, [x20], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
-      "sub x19, %x[width], x25\n"
+      "sub x20, %x[width], x26\n"
       ".inst 0x25286580  // psel p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe05802a0  // ld1h { za0h.h[x12] }, p0/Z, [x21, x24, LSL #1]\n"
+      ".inst 0xe05902c0  // ld1h { za0h.h[x12] }, p0/Z, [x22, x25, LSL #1]\n"
       ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
-      "cmp x19, x26\n"
-      ".inst 0xe0580288  // ld1h { za1h.h[x12] }, p0/Z, [x20, x24, LSL #1]\n"
+      "cmp x20, x27\n"
+      ".inst 0xe05902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x25, LSL #1]\n"
       "mov x12, #0x0\n"
-      "csel x19, x19, x26, LT\n"
+      "csel x20, x20, x27, LT\n"
       "4:"  // Stores: Loop
       ".inst 0x25287540  // psel p0.h, p13.h/Z, p10.h[w12]\n"
-      ".inst 0xe07f82e0  // st1h { za0v.h[x12] }, p0/Z, [x23, XZR, LSL #1]\n"
+      ".inst 0xe07f8300  // st1h { za0v.h[x12] }, p0/Z, [x24, XZR, LSL #1]\n"
       ".inst 0x25287540  // psel p0.h, p13.h/Z, p10.h[w12]\n"
-      ".inst 0xe07a82e8  // st1h { za1v.h[x12] }, p0/Z, [x23, x26, LSL #1]\n"
+      ".inst 0xe07b8308  // st1h { za1v.h[x12] }, p0/Z, [x24, x27, LSL #1]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
-      "addvl x23, x23, #4\n"
+      "cmp x12, x20\n"
+      "addvl x24, x24, #4\n"
       "blt 4b\n"
+      "inch x26\n"
+      "whilelt p10.h, x26, %x[width]\n"
+      "whilelt p9.h, x26, %x[width]\n"
+      "whilelt p8.h, x26, %x[width]\n"
       "inch x25\n"
-      "whilelt p10.h, x25, %x[width]\n"
-      "whilelt p9.h, x25, %x[width]\n"
-      "whilelt p8.h, x25, %x[width]\n"
-      "inch x24\n"
       "b.any 1b\n"
-      "mov %x[out], x23\n"
+      "mov %x[out], x24\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp32_fp32.hpp
index d702542..949e003 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_fp32_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,256 +34,256 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "mov x22, %x[width]\n"
+      "incw x22\n"
       "cntw x16\n"
-      "cntw x15, ALL, MUL #2\n"
-      "cntw x14, ALL, MUL #3\n"
-      "mov x19, %x[width]\n"
-      "incw x19\n"
-      "sub x19, x19, #0x1\n"
-      "udiv x19, x19, x16\n" // n_passes = ceildiv(width, VL<T>)
-      "sub x13, x19, #0x1\n"
-      "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
-      "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
-      "mov x19, %x[width]\n"
-      "sub x10, x16, #0x1\n"
-      "ands x10, x19, x10\n"
-      "csel x10, x10, x16, NE\n"
-      "sub x9, x16, #0x2\n"
+      "sub x22, x22, #0x1\n"
+      "udiv x22, x22, x16\n"  // n_passes = ceildiv(width, VL<T>)
+      "mov x21, %x[width]\n"
+      "sub x15, x16, #0x1\n"
+      "sub x20, x22, #0x1\n"
+      "ands x15, x21, x15\n"
+      "sub x14, x16, #0x2\n"
+      "mov x13, #0x0\n"
+      "mov x11, %x[in]\n"
+      "ldr x10, [x11, #0x0]\n"
+      "add x9, %x[in], x16, LSL #3\n"
+      "cntw x28, ALL, MUL #2\n"
+      "ldr x27, [x9, #0x0]\n"
+      "cntw x26, ALL, MUL #3\n"
+      "lsr x20, x20, #0x1\n"  // n_loops = (n_passes - 1) / 2
+      "ldr x25, [x11, #0x8]\n"
+      "and x24, x22, #0x1\n"  // odd_tail = bool(n_passes & 0x1)
+      "csel x15, x15, x16, NE\n"
+      "ldr x23, [x9, #0x8]\n"
       "ptrue p13.s\n"
       "whilelt p12.s, XZR, %x[height]\n"
       "whilelt p11.s, x16, %x[height]\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "mov x26, #0x0\n"
-      "whilelt p10.s, x26, %x[width]\n"
-      "whilelt p9.s, x26, %x[width]\n"
-      "whilelt p8.s, x26, %x[width]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      "mov x22, %x[row_offset]\n"
+      "mov x21, %x[out]\n"
+      "whilelt p10.s, x13, %x[width]\n"
+      "whilelt p9.s, x13, %x[width]\n"
+      "whilelt p8.s, x13, %x[width]\n"
+      "add x11, x11, #0x10\n"
+      "add x9, x9, #0x10\n"
       "mov x12, #0x0\n"
-      "cbz x9, 2f\n"
+      "cbz x14, 2f\n"
       "1:"  // K loop: Charge: Loop
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e0  // ld1w { za0h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02c4  // ld1w { za1h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c0285  // ld1w { za1h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960540  // ld1w { za0h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0xe0960364  // ld1w { za1h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      ".inst 0x25706581  // psel p1.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706160  // psel p0.s, p8.s/Z, p11.s[w12, #1]\n"
+      "ldr x27, [x9, #0x0]\n"
+      ".inst 0xe0960721  // ld1w { za0h.s[x12, #1] }, p1/Z, [x25, x22, LSL #2]\n"
+      "ldr x25, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe09602e5  // ld1w { za1h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x14\n"
+      "ldr x23, [x9, #0x8]\n"
+      "add x9, x9, #0x10\n"
       "blt 1b\n"
       "2:"  // K loop: Charge: End
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e0  // ld1w { za0h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02c4  // ld1w { za1h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c0285  // ld1w { za1h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      "incw x28\n"
-      "incw x26\n"
-      "cbz x13, 8f\n"
-      "mov x19, x13\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960540  // ld1w { za0h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      ".inst 0xe0960364  // ld1w { za1h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      ".inst 0x25706581  // psel p1.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706160  // psel p0.s, p8.s/Z, p11.s[w12, #1]\n"
+      "mov x11, %x[in]\n"
+      "add x9, %x[in], x16, LSL #3\n"
+      ".inst 0xe0960721  // ld1w { za0h.s[x12, #1] }, p1/Z, [x25, x22, LSL #2]\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0xe09602e5  // ld1w { za1h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n"
+      "ldr x27, [x9, #0x0]\n"
+      "incw x22\n"
+      "incw x13\n"
+      "ldr x25, [x11, #0x8]\n"
+      "add x11, x11, #0x10\n"
+      "ldr x23, [x9, #0x8]\n"
+      "add x9, x9, #0x10\n"
+      "cbz x20, 8f\n"
+      "mov x20, x20\n"
       "3:"  // K loop: Main loop
-      "whilelt p9.s, x26, %x[width]\n"
-      "whilelt p8.s, x26, %x[width]\n"
+      "whilelt p9.s, x13, %x[width]\n"
+      "whilelt p8.s, x13, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x9, 5f\n"
+      "cbz x14, 5f\n"
       "4:"  // K loop: Main loop: First: Loop
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e8  // ld1w { za2h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02cc  // ld1w { za3h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a9  // ld1w { za2h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c028d  // ld1w { za3h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960548  // ld1w { za2h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0xe096036c  // ld1w { za3h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      ".inst 0x25706580  // psel p0.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706162  // psel p2.s, p8.s/Z, p11.s[w12, #1]\n"
+      "ldr x27, [x9, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960329  // ld1w { za2h.s[x12, #1] }, p0/Z, [x25, x22, LSL #2]\n"
+      "ldr x25, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960aed  // ld1w { za3h.s[x12, #1] }, p2/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x9, #0x8]\n"
+      ".inst 0xe0bf86a0  // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x9, x9, #0x10\n"
+      ".inst 0xe0ba82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x26, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x14\n"
+      "addvl x21, x21, #4\n"
       "blt 4b\n"
       "5:"  // K loop: Main loop: First: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e8  // ld1w { za2h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02cc  // ld1w { za3h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a9  // ld1w { za2h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c028d  // ld1w { za3h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.s, x26, %x[width]\n"
-      ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incw x26\n"
-      "incw x28\n"
-      ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "whilelt p9.s, x26, %x[width]\n"
-      "whilelt p8.s, x26, %x[width]\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960548  // ld1w { za2h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      ".inst 0xe096036c  // ld1w { za3h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      "mov x11, %x[in]\n"
+      "add x9, %x[in], x16, LSL #3\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0x25706580  // psel p0.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706161  // psel p1.s, p8.s/Z, p11.s[w12, #1]\n"
+      ".inst 0xe0960329  // ld1w { za2h.s[x12, #1] }, p0/Z, [x25, x22, LSL #2]\n"
+      "ldr x27, [x9, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe09606ed  // ld1w { za3h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x25, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x9, #0x8]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aa4  // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.s, x13, %x[width]\n"
+      "incw x13\n"
+      ".inst 0xe0bc86a1  // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x9, x9, #0x10\n"
+      ".inst 0xe0ba82a5  // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x26, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incw x22\n"
+      "whilelt p9.s, x13, %x[width]\n"
+      "whilelt p8.s, x13, %x[width]\n"
       "mov x12, #0x0\n"
-      "cbz x9, 7f\n"
+      "cbz x14, 7f\n"
       "6:"  // K loop: Main loop: Second: Loop
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e0  // ld1w { za0h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02c4  // ld1w { za1h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c0285  // ld1w { za1h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960540  // ld1w { za0h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0xe0960364  // ld1w { za1h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      ".inst 0x25706580  // psel p0.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706162  // psel p2.s, p8.s/Z, p11.s[w12, #1]\n"
+      "ldr x27, [x9, #0x0]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960321  // ld1w { za0h.s[x12, #1] }, p0/Z, [x25, x22, LSL #2]\n"
+      "ldr x25, [x11, #0x8]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0960ae5  // ld1w { za1h.s[x12, #1] }, p2/Z, [x23, x22, LSL #2]\n"
+      "ldr x23, [x9, #0x8]\n"
+      ".inst 0xe0bf86a8  // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      "add x11, x11, #0x10\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x9, x9, #0x10\n"
+      ".inst 0xe0ba82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x26, LSL #2]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x9\n"
+      "cmp x12, x14\n"
+      "addvl x21, x21, #4\n"
       "blt 6b\n"
       "7:"  // K loop: Main loop: Second: Tail
-      "mov x25, %x[in]\n"
-      "add x24, %x[in], x16, LSL #3\n"
-      ".inst 0x25306580  // dup p0.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe09c02e0  // ld1w { za0h.s[x12] }, p0/Z, [x23, x28, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      ".inst 0xe09c02c4  // ld1w { za1h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
-      ".inst 0x25706580  // dup p0.s, p9.s/Z, p12.s[w12, #1]\n"
-      ".inst 0xe09c02a1  // ld1w { za0h.s[x12, #1] }, p0/Z, [x21, x28, LSL #2]\n"
-      ".inst 0x25706160  // dup p0.s, p8.s/Z, p11.s[w12, #1]\n"
-      ".inst 0xe09c0285  // ld1w { za1h.s[x12, #1] }, p0/Z, [x20, x28, LSL #2]\n"
-      "ldr x23, [x25, #0x0]\n"
-      "ldr x22, [x24, #0x0]\n"
-      "ldr x21, [x25, #0x8]\n"
-      "ldr x20, [x24, #0x8]\n"
-      "add x25, x25, #0x10\n"
-      "add x24, x24, #0x10\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25707541  // dup p1.s, p13.s/Z, p10.s[w12, #1]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25707540  // dup p0.s, p13.s/Z, p10.s[w12, #1]\n"
-      "whilelt p10.s, x26, %x[width]\n"
-      ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
-      "incw x26\n"
-      "incw x28\n"
-      ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
-      "addvl x27, x27, #4\n"
-      "subs x19, x19, #0x1\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      ".inst 0xe0960540  // ld1w { za0h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      ".inst 0xe0960364  // ld1w { za1h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
+      "mov x11, %x[in]\n"
+      "add x9, %x[in], x16, LSL #3\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0x25706580  // psel p0.s, p9.s/Z, p12.s[w12, #1]\n"
+      ".inst 0x25706161  // psel p1.s, p8.s/Z, p11.s[w12, #1]\n"
+      ".inst 0xe0960321  // ld1w { za0h.s[x12, #1] }, p0/Z, [x25, x22, LSL #2]\n"
+      "ldr x27, [x9, #0x0]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe09606e5  // ld1w { za1h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n"
+      "ldr x25, [x11, #0x8]\n"
+      ".inst 0x25307542  // psel p2.s, p13.s/Z, p10.s[w12]\n"
+      "ldr x23, [x9, #0x8]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25707541  // psel p1.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0x25707540  // psel p0.s, p13.s/Z, p10.s[w12, #1]\n"
+      ".inst 0xe0b08aac  // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n"
+      "whilelt p10.s, x13, %x[width]\n"
+      "subs x20, x20, #0x1\n"
+      ".inst 0xe0bc86a9  // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n"
+      "add x11, x11, #0x10\n"
+      "add x9, x9, #0x10\n"
+      ".inst 0xe0ba82ad  // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x26, LSL #2]\n"
+      "addvl x21, x21, #4\n"
+      "incw x13\n"
+      "incw x22\n"
       "bgt 3b\n"
       "8:"  // K loop: Tails
-      "cbnz x11, 11f\n"
-      "mov x25, %x[in]\n"
-      "whilelt p9.s, x26, %x[width]\n"
-      "whilelt p8.s, x26, %x[width]\n"
+      "cbnz x24, 11f\n"
+      "mov x11, %x[in]\n"
+      "whilelt p9.s, x13, %x[width]\n"
+      "whilelt p8.s, x13, %x[width]\n"
       "mov x12, #0x0\n"
       "9:"  // K loop: Tails: Even: First
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25306581  // dup p1.s, p9.s/Z, p12.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      ".inst 0x25306160  // dup p0.s, p8.s/Z, p11.s[w12]\n"
-      "addvl x27, x27, #2\n"
-      "ldr x23, [x25, #0x0]\n"
-      ".inst 0xe09c06e8  // ld1w { za2h.s[x12] }, p1/Z, [x23, x28, LSL #2]\n"
-      "ldr x22, [x25, x16, LSL #0x3]\n"
-      ".inst 0xe09c02cc  // ld1w { za3h.s[x12] }, p0/Z, [x22, x28, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
+      "ldr x10, [x11, #0x0]\n"
+      ".inst 0x25306581  // psel p1.s, p9.s/Z, p12.s[w12]\n"
+      ".inst 0x25306160  // psel p0.s, p8.s/Z, p11.s[w12]\n"
+      "ldr x27, [x11, x16, LSL #0x3]\n"
+      ".inst 0xe0960548  // ld1w { za2h.s[x12] }, p1/Z, [x10, x22, LSL #2]\n"
+      "add x11, x11, #0x8\n"
+      "addvl x21, x21, #2\n"
+      ".inst 0xe096036c  // ld1w { za3h.s[x12] }, p0/Z, [x27, x22, LSL #2]\n"
       "add x12, x12, #0x1\n"
       "cmp x12, x16\n"
-      "add x25, x25, #0x8\n"
       "blt 9b\n"
-      "whilelt p10.s, x26, %x[width]\n"
-      "whilelt p9.s, x26, %x[width]\n"
-      "whilelt p8.s, x26, %x[width]\n"
+      "whilelt p10.s, x13, %x[width]\n"
+      "whilelt p9.s, x13, %x[width]\n"
+      "whilelt p8.s, x13, %x[width]\n"
       "mov x12, #0x0\n"
       "10:"  // K loop: Tails: Even: Second
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a8  // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082ac  // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #2\n"
       "blt 10b\n"
-      "whilelt p10.s, x26, %x[width]\n"
+      "whilelt p10.s, x13, %x[width]\n"
       "b 13f\n"
       "11:"  // K loop: Tails: Odd
       "mov x12, #0x0\n"
       "12:"  // K loop: Tails: Odd: Loop
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
-      "addvl x27, x27, #2\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0bf82a0  // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0xe0b082a4  // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x15\n"
+      "addvl x21, x21, #2\n"
       "blt 12b\n"
       "13:"  // K loop: End
-      "mov %x[out], x27\n"
+      "mov %x[out], x21\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "p12", "p13", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block2_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block2_bf16_bf16.hpp
index 556d148..4cc84d3 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block2_bf16_bf16.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block2_bf16_bf16.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,93 +34,93 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x16\n"
       "cntw x15\n"
       "cntw x14, ALL, MUL #2\n"
       "cntw x13, ALL, MUL #3\n"
-      "cnth x11\n"
-      "ptrue p13.s\n"
-      "cntw x10\n"
-      "cmp %x[height], x10\n"
-      "csel x10, %x[height], x10, LT\n"
-      "sub x10, x10, #0x1\n"
-      "whilelt p10.h, XZR, %x[height]\n"
-      "whilelt p9.h, x15, %x[height]\n"
-      "whilelt p8.h, x14, %x[height]\n"
-      "zip1 p12.h, p10.h, p8.h\n"
+      "cmp %x[height], x16\n"
+      "csel x16, %x[height], x16, LT\n"
+      "whilelt p11.h, XZR, %x[height]\n"
+      "whilelt p10.h, x15, %x[height]\n"
+      "whilelt p9.h, x14, %x[height]\n"
       "whilelt p8.h, x13, %x[height]\n"
-      "zip1 p11.h, p9.h, p8.h\n"
+      "mov x11, #0x0\n"
+      "cnth x10\n"
+      "ptrue p13.s\n"
+      "sub x16, x16, #0x1\n"
+      "zip1 p12.h, p11.h, p9.h\n"
+      "zip1 p11.h, p10.h, p8.h\n"
       "mov x9, %x[row_offset]\n"
       "mov x28, %x[out]\n"
-      "mov x27, #0x0\n"
-      "whilelt p10.h, x27, %x[width]\n"
-      "whilelt p9.h, x27, %x[width]\n"
-      "whilelt p8.h, x27, %x[width]\n"
+      "whilelt p10.h, x11, %x[width]\n"
+      "whilelt p9.h, x11, %x[width]\n"
+      "whilelt p8.h, x11, %x[width]\n"
       "1:"  // Width loop
-      "mov x12, #0x0\n"
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x15, LSL #3\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x15, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
       "add x24, %x[in], x14, LSL #3\n"
-      "add x23, %x[in], x13, LSL #3\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x20, [x24], #0x8\n"
-      "ldr x19, [x23], #0x8\n"
-      "cbz x10, 3f\n"
+      "add x20, %x[in], x13, LSL #3\n"
+      "ldr x23, [x26], #0x8\n"
+      "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
+      "ldr x21, [x20], #0x8\n"
+      "cbz x16, 3f\n"
       "2:"  // Loads: Loop
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe04902c0  // ld1h { za0h.h[x12] }, p0/Z, [x22, x9, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe04902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x9, LSL #1]\n"
-      ".inst 0x25386580  // dup p0.h, p9.h/Z, p12.h[w12, #1]\n"
-      ".inst 0xe0490281  // ld1h { za0h.h[x12, #1] }, p0/Z, [x20, x9, LSL #1]\n"
-      ".inst 0x25386160  // dup p0.h, p8.h/Z, p11.h[w12, #1]\n"
-      ".inst 0xe0490269  // ld1h { za1h.h[x12, #1] }, p0/Z, [x19, x9, LSL #1]\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x20, [x24], #0x8\n"
-      "ldr x19, [x23], #0x8\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0490720  // ld1h { za0h.h[x12] }, p1/Z, [x25, x9, LSL #1]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0xe04902e8  // ld1h { za1h.h[x12] }, p0/Z, [x23, x9, LSL #1]\n"
+      ".inst 0x25386581  // psel p1.h, p9.h/Z, p12.h[w12, #1]\n"
+      ".inst 0x25386160  // psel p0.h, p8.h/Z, p11.h[w12, #1]\n"
+      "ldr x23, [x26], #0x8\n"
+      ".inst 0xe04906c1  // ld1h { za0h.h[x12, #1] }, p1/Z, [x22, x9, LSL #1]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe04902a9  // ld1h { za1h.h[x12, #1] }, p0/Z, [x21, x9, LSL #1]\n"
       "add x12, x12, #0x2\n"
-      "cmp x12, x10, LSL #1\n"
+      "cmp x12, x16, LSL #1\n"
+      "ldr x21, [x20], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
-      ".inst 0x25286580  // dup p0.h, p9.h/Z, p12.h[w12]\n"
-      ".inst 0xe04902c0  // ld1h { za0h.h[x12] }, p0/Z, [x22, x9, LSL #1]\n"
-      ".inst 0x25286160  // dup p0.h, p8.h/Z, p11.h[w12]\n"
-      ".inst 0xe04902a8  // ld1h { za1h.h[x12] }, p0/Z, [x21, x9, LSL #1]\n"
-      ".inst 0x25386580  // dup p0.h, p9.h/Z, p12.h[w12, #1]\n"
-      ".inst 0xe0490281  // ld1h { za0h.h[x12, #1] }, p0/Z, [x20, x9, LSL #1]\n"
-      ".inst 0x25386160  // dup p0.h, p8.h/Z, p11.h[w12, #1]\n"
-      ".inst 0xe0490269  // ld1h { za1h.h[x12, #1] }, p0/Z, [x19, x9, LSL #1]\n"
+      ".inst 0x25286581  // psel p1.h, p9.h/Z, p12.h[w12]\n"
+      ".inst 0x25286160  // psel p0.h, p8.h/Z, p11.h[w12]\n"
+      ".inst 0xe0490720  // ld1h { za0h.h[x12] }, p1/Z, [x25, x9, LSL #1]\n"
+      "sub x20, %x[width], x11\n"
+      ".inst 0xe04902e8  // ld1h { za1h.h[x12] }, p0/Z, [x23, x9, LSL #1]\n"
+      "cmp x20, x10\n"
+      "csel x20, x20, x10, LT\n"
+      ".inst 0x25386580  // psel p0.h, p9.h/Z, p12.h[w12, #1]\n"
+      ".inst 0xe04902c1  // ld1h { za0h.h[x12, #1] }, p0/Z, [x22, x9, LSL #1]\n"
+      ".inst 0x25386160  // psel p0.h, p8.h/Z, p11.h[w12, #1]\n"
+      "add x20, x20, #0x1\n"
+      ".inst 0xe04902a9  // ld1h { za1h.h[x12, #1] }, p0/Z, [x21, x9, LSL #1]\n"
       "mov x12, #0x0\n"
-      "sub x19, %x[width], x27\n"
-      "cmp x19, x11\n"
-      "csel x19, x19, x11, LT\n"
-      "add x19, x19, #0x1\n"
-      "lsr x19, x19, #0x1\n"
+      "lsr x20, x20, #0x1\n"
       "4:"  // Stores: Loop
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
       ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
-      ".inst 0x25307541  // dup p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
       ".inst 0xe0af8384  // st1w { za1v.s[x12] }, p0/Z, [x28, x15, LSL #2]\n"
-      ".inst 0x25307540  // dup p0.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0x25307541  // psel p1.s, p13.s/Z, p10.s[w12]\n"
+      ".inst 0x25307540  // psel p0.s, p13.s/Z, p10.s[w12]\n"
       ".inst 0xe0ae8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x14, LSL #2]\n"
       ".inst 0xe0ad838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x13, LSL #2]\n"
-      "addvl x28, x28, #4\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
+      "cmp x12, x20\n"
+      "addvl x28, x28, #4\n"
       "blt 4b\n"
+      "inch x11\n"
+      "whilelt p10.h, x11, %x[width]\n"
+      "whilelt p9.h, x11, %x[width]\n"
+      "whilelt p8.h, x11, %x[width]\n"
       "inch x9\n"
-      "inch x27\n"
-      "whilelt p10.h, x27, %x[width]\n"
-      "whilelt p9.h, x27, %x[width]\n"
-      "whilelt p8.h, x27, %x[width]\n"
       "b.any 1b\n"
       "mov %x[out], x28\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "p12", "p13", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8.hpp
index 49d1aa5..465939c 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,92 +34,92 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x16\n"
       "cntw x15\n"
-      "cntw x14\n"
-      "cntw x13, ALL, MUL #2\n"
-      "cntw x11, ALL, MUL #3\n"
-      "cmp %x[height], x15\n"
-      "csel x15, %x[height], x15, LT\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x13, ALL, MUL #3\n"
+      "cmp %x[height], x16\n"
+      "csel x16, %x[height], x16, LT\n"
       "whilelt p12.b, XZR, %x[height]\n"
-      "whilelt p10.b, x14, %x[height]\n"
-      "whilelt p9.b, x13, %x[height]\n"
-      "whilelt p8.b, x11, %x[height]\n"
+      "whilelt p10.b, x15, %x[height]\n"
+      "whilelt p9.b, x14, %x[height]\n"
+      "whilelt p8.b, x13, %x[height]\n"
       "zip1 p12.b, p12.b, p9.b\n"
       "zip1 p10.b, p10.b, p8.b\n"
-      "mov x10, #0x0\n"
-      "cntb x9\n"
+      "mov x11, #0x0\n"
+      "cntb x10\n"
       "ptrue p11.s\n"
-      "sub x15, x15, #0x1\n"
+      "sub x16, x16, #0x1\n"
       "zip1 p10.b, p12.b, p10.b\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "whilelt p9.b, x10, %x[width]\n"
-      "whilelt p8.b, x10, %x[width]\n"
+      "mov x9, %x[row_offset]\n"
+      "mov x28, %x[out]\n"
+      "whilelt p9.b, x11, %x[width]\n"
+      "whilelt p8.b, x11, %x[width]\n"
       "1:"  // Width loop
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x14, LSL #3\n"
-      "ldr x24, [x26], #0x8\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x15, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
+      "add x24, %x[in], x14, LSL #3\n"
       "add x23, %x[in], x13, LSL #3\n"
-      "add x22, %x[in], x11, LSL #3\n"
-      "ldr x19, [x25], #0x8\n"
+      "ldr x20, [x26], #0x8\n"
       "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
       "ldr x21, [x23], #0x8\n"
-      "ldr x20, [x22], #0x8\n"
-      "cbz x15, 3f\n"
+      "cbz x16, 3f\n"
       "2:"  // Loads: Loop
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0xe0090320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x9]\n"
       ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      "ldr x24, [x26], #0x8\n"
-      ".inst 0xe01c0261  // ld1b { za0h.b[x12, #1] }, p0/Z, [x19, x28]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0xe0090281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x9]\n"
       ".inst 0x25346141  // psel p1.b, p8.b/Z, p10.b[w12, #2]\n"
       ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
-      "ldr x19, [x25], #0x8\n"
-      ".inst 0xe01c06a2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x21, x28]\n"
-      "ldr x21, [x23], #0x8\n"
-      ".inst 0xe01c0283  // ld1b { za0h.b[x12, #3] }, p0/Z, [x20, x28]\n"
+      "ldr x20, [x26], #0x8\n"
+      ".inst 0xe00906c2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x22, x9]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe00902a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x9]\n"
       "add x12, x12, #0x4\n"
-      "cmp x12, x15, LSL #2\n"
-      "ldr x20, [x22], #0x8\n"
+      "cmp x12, x16, LSL #2\n"
+      "ldr x21, [x23], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0xe0090320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x9]\n"
       ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0xe01c0261  // ld1b { za0h.b[x12, #1] }, p0/Z, [x19, x28]\n"
+      ".inst 0xe0090281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x9]\n"
       ".inst 0x25346140  // psel p0.b, p8.b/Z, p10.b[w12, #2]\n"
-      "sub x19, %x[width], x10\n"
-      ".inst 0xe01c02a2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x21, x28]\n"
-      "cmp x19, x9\n"
-      "csel x19, x19, x9, LT\n"
+      "sub x20, %x[width], x11\n"
+      ".inst 0xe00902c2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x22, x9]\n"
+      "cmp x20, x10\n"
+      "csel x20, x20, x10, LT\n"
       ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
-      "add x19, x19, #0x3\n"
-      ".inst 0xe01c0283  // ld1b { za0h.b[x12, #3] }, p0/Z, [x20, x28]\n"
+      "add x20, x20, #0x3\n"
+      ".inst 0xe00902a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x9]\n"
       "mov x12, #0x0\n"
-      "lsr x19, x19, #0x2\n"
+      "lsr x20, x20, #0x2\n"
       "4:"  // Stores: Loop
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0ae8364  // st1w { za1v.s[x12] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0xe0af8384  // st1w { za1v.s[x12] }, p0/Z, [x28, x15, LSL #2]\n"
       ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0ad8768  // st1w { za2v.s[x12] }, p1/Z, [x27, x13, LSL #2]\n"
-      ".inst 0xe0ab836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xe0ae8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x14, LSL #2]\n"
+      ".inst 0xe0ad838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x13, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
-      "addvl x27, x27, #4\n"
+      "cmp x12, x20\n"
+      "addvl x28, x28, #4\n"
       "blt 4b\n"
-      "incb x10\n"
-      "whilelt p9.b, x10, %x[width]\n"
-      "whilelt p8.b, x10, %x[width]\n"
-      "incb x28\n"
+      "incb x11\n"
+      "whilelt p9.b, x11, %x[width]\n"
+      "whilelt p8.b, x11, %x[width]\n"
+      "incb x9\n"
       "b.any 1b\n"
-      "mov %x[out], x27\n"
+      "mov %x[out], x28\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8_summing.hpp
index 94673d4..ffd9384 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_s8_s8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,118 +32,118 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "mov z24.b, #0x1\n"
-      "mov z23.s, #0x0\n"
-      "ptrue p2.b\n"
-      "mov z22.s, #0x0\n"
       "cntw x16\n"
+      "cntw x15\n"
+      "mov z24.b, #0x1\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x13, ALL, MUL #3\n"
+      "mov z23.s, #0x0\n"
+      "mov z22.s, #0x0\n"
+      "cmp %x[height], x16\n"
+      "csel x16, %x[height], x16, LT\n"
       "mov z21.s, #0x0\n"
-      "cntw x15, ALL, MUL #2\n"
       "mov z20.s, #0x0\n"
-      "cntw x14, ALL, MUL #3\n"
+      "whilelt p12.b, XZR, %x[height]\n"
+      "whilelt p10.b, x15, %x[height]\n"
+      "whilelt p9.b, x14, %x[height]\n"
+      "whilelt p8.b, x13, %x[height]\n"
+      "zip1 p12.b, p12.b, p9.b\n"
+      "zip1 p10.b, p10.b, p8.b\n"
+      "ptrue p2.b\n"
       "cntb x11\n"
       "ptrue p11.s\n"
-      "cntw x10\n"
-      "cmp %x[height], x10\n"
-      "csel x10, %x[height], x10, LT\n"
-      "sub x10, x10, #0x1\n"
-      "whilelt p10.b, XZR, %x[height]\n"
-      "whilelt p9.b, x16, %x[height]\n"
-      "whilelt p8.b, x15, %x[height]\n"
-      "zip1 p10.b, p10.b, p8.b\n"
-      "whilelt p8.b, x14, %x[height]\n"
-      "zip1 p9.b, p9.b, p8.b\n"
-      "mov x9, %x[row_offset]\n"
-      "mov x28, %x[out]\n"
-      "zip1 p10.b, p10.b, p9.b\n"
+      "sub x16, x16, #0x1\n"
+      "zip1 p10.b, p12.b, p10.b\n"
+      "mov x10, %x[row_offset]\n"
+      "mov x9, %x[out]\n"
       "cbnz %x[first], 1f\n"
-      "addvl x28, x28, #-4\n"
-      "ld1w { z23.s }, p2/Z, [x28]\n"
-      "ld1w { z22.s }, p2/Z, [x28, #1, MUL VL]\n"
-      "ld1w { z21.s }, p2/Z, [x28, #2, MUL VL]\n"
-      "ld1w { z20.s }, p2/Z, [x28, #3, MUL VL]\n"
+      "addvl x9, x9, #-4\n"
+      "ld1w { z23.s }, p2/Z, [x9]\n"
+      "ld1w { z22.s }, p2/Z, [x9, #1, MUL VL]\n"
+      "ld1w { z21.s }, p2/Z, [x9, #2, MUL VL]\n"
+      "ld1w { z20.s }, p2/Z, [x9, #3, MUL VL]\n"
       "1:"  // Initialise row sums: End
-      "mov x27, #0x0\n"
-      "whilelt p9.b, x27, %x[width]\n"
-      "whilelt p8.b, x27, %x[width]\n"
+      "mov x28, #0x0\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "2:"  // Width loop
-      "mov x13, #0x0\n"
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x16, LSL #3\n"
-      "add x24, %x[in], x15, LSL #3\n"
-      "add x23, %x[in], x14, LSL #3\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x19, [x24], #0x8\n"
-      "ldr x20, [x23], #0x8\n"
-      "cbz x10, 4f\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x15, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
+      "add x24, %x[in], x14, LSL #3\n"
+      "add x23, %x[in], x13, LSL #3\n"
+      "ldr x20, [x26], #0x8\n"
+      "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
+      "ldr x21, [x23], #0x8\n"
+      "cbz x16, 4f\n"
       "3:"  // Loads: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe00922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x9]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe00922a1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x21, x9]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      "ldr x22, [x26], #0x8\n"
-      ".inst 0xe0092662  // ld1b { za0h.b[x13, #2] }, p1/Z, [x19, x9]\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x19, [x24], #0x8\n"
-      ".inst 0xe0092283  // ld1b { za0h.b[x13, #3] }, p0/Z, [x20, x9]\n"
-      "ldr x20, [x23], #0x8\n"
-      "add x13, x13, #0x4\n"
-      "cmp x13, x10, LSL #2\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe00a0320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0xe00a0281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
+      ".inst 0x25346141  // psel p1.b, p8.b/Z, p10.b[w12, #2]\n"
+      ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
+      "ldr x20, [x26], #0x8\n"
+      ".inst 0xe00a06c2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x22, x10]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe00a02a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
+      "add x12, x12, #0x4\n"
+      "cmp x12, x16, LSL #2\n"
+      "ldr x21, [x23], #0x8\n"
       "blt 3b\n"
       "4:"  // Loads: Tail
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe00922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x9]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe00922a1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x21, x9]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe00a0320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe00a0281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
+      ".inst 0x25346140  // psel p0.b, p8.b/Z, p10.b[w12, #2]\n"
+      "sub x20, %x[width], x28\n"
+      ".inst 0xe00a02c2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x22, x10]\n"
+      "cmp x20, x11\n"
+      "csel x20, x20, x11, LT\n"
+      ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
+      "add x20, x20, #0x3\n"
+      ".inst 0xe00a02a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
       "mov x12, #0x0\n"
-      ".inst 0xe0092662  // ld1b { za0h.b[x13, #2] }, p1/Z, [x19, x9]\n"
-      "sub x19, %x[width], x27\n"
-      "cmp x19, x11\n"
-      ".inst 0xe0092283  // ld1b { za0h.b[x13, #3] }, p0/Z, [x20, x9]\n"
-      "csel x19, x19, x11, LT\n"
-      "add x19, x19, #0x3\n"
-      "lsr x19, x19, #0x2\n"
+      "lsr x20, x20, #0x2\n"
       "5:"  // Stores: Loop
-      ".inst 0xc0828813  // mova z19.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
-      ".inst 0xc0828892  // mova z18.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xe0b08384  // st1w { za1v.s[x12] }, p0/Z, [x28, x16, LSL #2]\n"
-      ".inst 0x25306d21  // dup p1.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
-      "sdot z23.s, z19.b, z24.b\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "sdot z22.s, z18.b, z24.b\n"
-      ".inst 0xe0af8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x15, LSL #2]\n"
-      "sdot z21.s, z17.b, z24.b\n"
-      "sdot z20.s, z16.b, z24.b\n"
-      ".inst 0xe0ae838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x14, LSL #2]\n"
-      "addvl x28, x28, #4\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf8120  // st1w { za0v.s[x12] }, p0/Z, [x9, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0828812  // mova z18.s, p2/M, za0v.s[x12]\n"
+      ".inst 0xe0af8124  // st1w { za1v.s[x12] }, p0/Z, [x9, x15, LSL #2]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0828891  // mova z17.s, p2/M, za1v.s[x12]\n"
+      ".inst 0xe0ae8528  // st1w { za2v.s[x12] }, p1/Z, [x9, x14, LSL #2]\n"
+      ".inst 0xc0828910  // mova z16.s, p2/M, za2v.s[x12]\n"
+      "sdot z23.s, z18.b, z24.b\n"
+      ".inst 0xe0ad812c  // st1w { za3v.s[x12] }, p0/Z, [x9, x13, LSL #2]\n"
+      ".inst 0xc0828993  // mova z19.s, p2/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
+      "cmp x12, x20\n"
+      "sdot z22.s, z17.b, z24.b\n"
+      "sdot z21.s, z16.b, z24.b\n"
+      "addvl x9, x9, #4\n"
+      "sdot z20.s, z19.b, z24.b\n"
       "blt 5b\n"
-      "incb x9\n"
-      "incb x27\n"
-      "whilelt p9.b, x27, %x[width]\n"
-      "whilelt p8.b, x27, %x[width]\n"
+      "incb x28\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
+      "incb x10\n"
       "b.any 2b\n"
-      "st1w { z23.s }, p2, [x28]\n"
-      "st1w { z22.s }, p2, [x28, #1, MUL VL]\n"
-      "st1w { z21.s }, p2, [x28, #2, MUL VL]\n"
-      "st1w { z20.s }, p2, [x28, #3, MUL VL]\n"
-      "addvl x28, x28, #4\n"
-      "mov %x[out], x28\n"
+      "st1w { z23.s }, p2, [x9]\n"
+      "st1w { z22.s }, p2, [x9, #1, MUL VL]\n"
+      "st1w { z21.s }, p2, [x9, #2, MUL VL]\n"
+      "st1w { z20.s }, p2, [x9, #3, MUL VL]\n"
+      "addvl x9, x9, #4\n"
+      "mov %x[out], x9\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8.hpp
index bbdaaa3..9f5db6b 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,92 +34,92 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x16\n"
       "cntw x15\n"
-      "cntw x14\n"
-      "cntw x13, ALL, MUL #2\n"
-      "cntw x11, ALL, MUL #3\n"
-      "cmp %x[height], x15\n"
-      "csel x15, %x[height], x15, LT\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x13, ALL, MUL #3\n"
+      "cmp %x[height], x16\n"
+      "csel x16, %x[height], x16, LT\n"
       "whilelt p12.b, XZR, %x[height]\n"
-      "whilelt p10.b, x14, %x[height]\n"
-      "whilelt p9.b, x13, %x[height]\n"
-      "whilelt p8.b, x11, %x[height]\n"
+      "whilelt p10.b, x15, %x[height]\n"
+      "whilelt p9.b, x14, %x[height]\n"
+      "whilelt p8.b, x13, %x[height]\n"
       "zip1 p12.b, p12.b, p9.b\n"
       "zip1 p10.b, p10.b, p8.b\n"
-      "mov x10, #0x0\n"
-      "cntb x9\n"
+      "mov x11, #0x0\n"
+      "cntb x10\n"
       "ptrue p11.s\n"
-      "sub x15, x15, #0x1\n"
+      "sub x16, x16, #0x1\n"
       "zip1 p10.b, p12.b, p10.b\n"
-      "mov x28, %x[row_offset]\n"
-      "mov x27, %x[out]\n"
-      "whilelt p9.b, x10, %x[width]\n"
-      "whilelt p8.b, x10, %x[width]\n"
+      "mov x9, %x[row_offset]\n"
+      "mov x28, %x[out]\n"
+      "whilelt p9.b, x11, %x[width]\n"
+      "whilelt p8.b, x11, %x[width]\n"
       "1:"  // Width loop
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x14, LSL #3\n"
-      "ldr x24, [x26], #0x8\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x15, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
+      "add x24, %x[in], x14, LSL #3\n"
       "add x23, %x[in], x13, LSL #3\n"
-      "add x22, %x[in], x11, LSL #3\n"
-      "ldr x19, [x25], #0x8\n"
+      "ldr x20, [x26], #0x8\n"
       "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
       "ldr x21, [x23], #0x8\n"
-      "ldr x20, [x22], #0x8\n"
-      "cbz x15, 3f\n"
+      "cbz x16, 3f\n"
       "2:"  // Loads: Loop
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0xe0090320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x9]\n"
       ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      "ldr x24, [x26], #0x8\n"
-      ".inst 0xe01c0261  // ld1b { za0h.b[x12, #1] }, p0/Z, [x19, x28]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0xe0090281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x9]\n"
       ".inst 0x25346141  // psel p1.b, p8.b/Z, p10.b[w12, #2]\n"
       ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
-      "ldr x19, [x25], #0x8\n"
-      ".inst 0xe01c06a2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x21, x28]\n"
-      "ldr x21, [x23], #0x8\n"
-      ".inst 0xe01c0283  // ld1b { za0h.b[x12, #3] }, p0/Z, [x20, x28]\n"
+      "ldr x20, [x26], #0x8\n"
+      ".inst 0xe00906c2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x22, x9]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe00902a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x9]\n"
       "add x12, x12, #0x4\n"
-      "cmp x12, x15, LSL #2\n"
-      "ldr x20, [x22], #0x8\n"
+      "cmp x12, x16, LSL #2\n"
+      "ldr x21, [x23], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
       ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
-      ".inst 0xe01c0300  // ld1b { za0h.b[x12] }, p0/Z, [x24, x28]\n"
+      ".inst 0xe0090320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x9]\n"
       ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
-      ".inst 0xe01c0261  // ld1b { za0h.b[x12, #1] }, p0/Z, [x19, x28]\n"
+      ".inst 0xe0090281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x9]\n"
       ".inst 0x25346140  // psel p0.b, p8.b/Z, p10.b[w12, #2]\n"
-      "sub x19, %x[width], x10\n"
-      ".inst 0xe01c02a2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x21, x28]\n"
-      "cmp x19, x9\n"
-      "csel x19, x19, x9, LT\n"
+      "sub x20, %x[width], x11\n"
+      ".inst 0xe00902c2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x22, x9]\n"
+      "cmp x20, x10\n"
+      "csel x20, x20, x10, LT\n"
       ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
-      "add x19, x19, #0x3\n"
-      ".inst 0xe01c0283  // ld1b { za0h.b[x12, #3] }, p0/Z, [x20, x28]\n"
+      "add x20, x20, #0x3\n"
+      ".inst 0xe00902a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x9]\n"
       "mov x12, #0x0\n"
-      "lsr x19, x19, #0x2\n"
+      "lsr x20, x20, #0x2\n"
       "4:"  // Stores: Loop
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
+      ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0ae8364  // st1w { za1v.s[x12] }, p0/Z, [x27, x14, LSL #2]\n"
+      ".inst 0xe0af8384  // st1w { za1v.s[x12] }, p0/Z, [x28, x15, LSL #2]\n"
       ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0ad8768  // st1w { za2v.s[x12] }, p1/Z, [x27, x13, LSL #2]\n"
-      ".inst 0xe0ab836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x11, LSL #2]\n"
+      ".inst 0xe0ae8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x14, LSL #2]\n"
+      ".inst 0xe0ad838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x13, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
-      "addvl x27, x27, #4\n"
+      "cmp x12, x20\n"
+      "addvl x28, x28, #4\n"
       "blt 4b\n"
-      "incb x10\n"
-      "whilelt p9.b, x10, %x[width]\n"
-      "whilelt p8.b, x10, %x[width]\n"
-      "incb x28\n"
+      "incb x11\n"
+      "whilelt p9.b, x11, %x[width]\n"
+      "whilelt p8.b, x11, %x[width]\n"
+      "incb x9\n"
       "b.any 1b\n"
-      "mov %x[out], x27\n"
+      "mov %x[out], x28\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8_summing.hpp
index 961008a..49d2acf 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8_summing.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_block4_u8_u8_summing.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -32,118 +32,118 @@
 {
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
-      "mov z24.b, #0x1\n"
-      "mov z23.s, #0x0\n"
-      "ptrue p2.b\n"
-      "mov z22.s, #0x0\n"
       "cntw x16\n"
+      "cntw x15\n"
+      "mov z24.b, #0x1\n"
+      "cntw x14, ALL, MUL #2\n"
+      "cntw x13, ALL, MUL #3\n"
+      "mov z23.s, #0x0\n"
+      "mov z22.s, #0x0\n"
+      "cmp %x[height], x16\n"
+      "csel x16, %x[height], x16, LT\n"
       "mov z21.s, #0x0\n"
-      "cntw x15, ALL, MUL #2\n"
       "mov z20.s, #0x0\n"
-      "cntw x14, ALL, MUL #3\n"
+      "whilelt p12.b, XZR, %x[height]\n"
+      "whilelt p10.b, x15, %x[height]\n"
+      "whilelt p9.b, x14, %x[height]\n"
+      "whilelt p8.b, x13, %x[height]\n"
+      "zip1 p12.b, p12.b, p9.b\n"
+      "zip1 p10.b, p10.b, p8.b\n"
+      "ptrue p2.b\n"
       "cntb x11\n"
       "ptrue p11.s\n"
-      "cntw x10\n"
-      "cmp %x[height], x10\n"
-      "csel x10, %x[height], x10, LT\n"
-      "sub x10, x10, #0x1\n"
-      "whilelt p10.b, XZR, %x[height]\n"
-      "whilelt p9.b, x16, %x[height]\n"
-      "whilelt p8.b, x15, %x[height]\n"
-      "zip1 p10.b, p10.b, p8.b\n"
-      "whilelt p8.b, x14, %x[height]\n"
-      "zip1 p9.b, p9.b, p8.b\n"
-      "mov x9, %x[row_offset]\n"
-      "mov x28, %x[out]\n"
-      "zip1 p10.b, p10.b, p9.b\n"
+      "sub x16, x16, #0x1\n"
+      "zip1 p10.b, p12.b, p10.b\n"
+      "mov x10, %x[row_offset]\n"
+      "mov x9, %x[out]\n"
       "cbnz %x[first], 1f\n"
-      "addvl x28, x28, #-4\n"
-      "ld1w { z23.s }, p2/Z, [x28]\n"
-      "ld1w { z22.s }, p2/Z, [x28, #1, MUL VL]\n"
-      "ld1w { z21.s }, p2/Z, [x28, #2, MUL VL]\n"
-      "ld1w { z20.s }, p2/Z, [x28, #3, MUL VL]\n"
+      "addvl x9, x9, #-4\n"
+      "ld1w { z23.s }, p2/Z, [x9]\n"
+      "ld1w { z22.s }, p2/Z, [x9, #1, MUL VL]\n"
+      "ld1w { z21.s }, p2/Z, [x9, #2, MUL VL]\n"
+      "ld1w { z20.s }, p2/Z, [x9, #3, MUL VL]\n"
       "1:"  // Initialise row sums: End
-      "mov x27, #0x0\n"
-      "whilelt p9.b, x27, %x[width]\n"
-      "whilelt p8.b, x27, %x[width]\n"
+      "mov x28, #0x0\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
       "2:"  // Width loop
-      "mov x13, #0x0\n"
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x16, LSL #3\n"
-      "add x24, %x[in], x15, LSL #3\n"
-      "add x23, %x[in], x14, LSL #3\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x19, [x24], #0x8\n"
-      "ldr x20, [x23], #0x8\n"
-      "cbz x10, 4f\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x15, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
+      "add x24, %x[in], x14, LSL #3\n"
+      "add x23, %x[in], x13, LSL #3\n"
+      "ldr x20, [x26], #0x8\n"
+      "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
+      "ldr x21, [x23], #0x8\n"
+      "cbz x16, 4f\n"
       "3:"  // Loads: Loop
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe00922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x9]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe00922a1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x21, x9]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
-      "ldr x22, [x26], #0x8\n"
-      ".inst 0xe0092662  // ld1b { za0h.b[x13, #2] }, p1/Z, [x19, x9]\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x19, [x24], #0x8\n"
-      ".inst 0xe0092283  // ld1b { za0h.b[x13, #3] }, p0/Z, [x20, x9]\n"
-      "ldr x20, [x23], #0x8\n"
-      "add x13, x13, #0x4\n"
-      "cmp x13, x10, LSL #2\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe00a0320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0xe00a0281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
+      ".inst 0x25346141  // psel p1.b, p8.b/Z, p10.b[w12, #2]\n"
+      ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
+      "ldr x20, [x26], #0x8\n"
+      ".inst 0xe00a06c2  // ld1b { za0h.b[x12, #2] }, p1/Z, [x22, x10]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe00a02a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
+      "add x12, x12, #0x4\n"
+      "cmp x12, x16, LSL #2\n"
+      "ldr x21, [x23], #0x8\n"
       "blt 3b\n"
       "4:"  // Loads: Tail
-      ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
-      ".inst 0xe00922c0  // ld1b { za0h.b[x13] }, p0/Z, [x22, x9]\n"
-      ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
-      ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
-      ".inst 0xe00922a1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x21, x9]\n"
-      ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
+      ".inst 0x25246140  // psel p0.b, p8.b/Z, p10.b[w12]\n"
+      ".inst 0xe00a0320  // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
+      ".inst 0x252c6140  // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
+      ".inst 0xe00a0281  // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
+      ".inst 0x25346140  // psel p0.b, p8.b/Z, p10.b[w12, #2]\n"
+      "sub x20, %x[width], x28\n"
+      ".inst 0xe00a02c2  // ld1b { za0h.b[x12, #2] }, p0/Z, [x22, x10]\n"
+      "cmp x20, x11\n"
+      "csel x20, x20, x11, LT\n"
+      ".inst 0x253c6140  // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
+      "add x20, x20, #0x3\n"
+      ".inst 0xe00a02a3  // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
       "mov x12, #0x0\n"
-      ".inst 0xe0092662  // ld1b { za0h.b[x13, #2] }, p1/Z, [x19, x9]\n"
-      "sub x19, %x[width], x27\n"
-      "cmp x19, x11\n"
-      ".inst 0xe0092283  // ld1b { za0h.b[x13, #3] }, p0/Z, [x20, x9]\n"
-      "csel x19, x19, x11, LT\n"
-      "add x19, x19, #0x3\n"
-      "lsr x19, x19, #0x2\n"
+      "lsr x20, x20, #0x2\n"
       "5:"  // Stores: Loop
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xe0bf8120  // st1w { za0v.s[x12] }, p0/Z, [x9, XZR, LSL #2]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
       ".inst 0xc0828813  // mova z19.s, p2/M, za0v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
-      ".inst 0xc0828892  // mova z18.s, p2/M, za1v.s[x12]\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
-      ".inst 0xe0b08384  // st1w { za1v.s[x12] }, p0/Z, [x28, x16, LSL #2]\n"
-      ".inst 0x25306d21  // dup p1.s, p11.s/Z, p9.s[w12]\n"
-      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
+      ".inst 0xe0af8124  // st1w { za1v.s[x12] }, p0/Z, [x9, x15, LSL #2]\n"
+      ".inst 0x25306d21  // psel p1.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0x25306d20  // psel p0.s, p11.s/Z, p9.s[w12]\n"
+      ".inst 0xc0828891  // mova z17.s, p2/M, za1v.s[x12]\n"
+      ".inst 0xe0ae8528  // st1w { za2v.s[x12] }, p1/Z, [x9, x14, LSL #2]\n"
+      ".inst 0xc0828912  // mova z18.s, p2/M, za2v.s[x12]\n"
       "udot z23.s, z19.b, z24.b\n"
-      ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
-      "udot z22.s, z18.b, z24.b\n"
-      ".inst 0xe0af8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x15, LSL #2]\n"
-      "udot z21.s, z17.b, z24.b\n"
-      "udot z20.s, z16.b, z24.b\n"
-      ".inst 0xe0ae838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x14, LSL #2]\n"
-      "addvl x28, x28, #4\n"
+      ".inst 0xe0ad812c  // st1w { za3v.s[x12] }, p0/Z, [x9, x13, LSL #2]\n"
+      ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
+      "cmp x12, x20\n"
+      "udot z22.s, z17.b, z24.b\n"
+      "udot z21.s, z18.b, z24.b\n"
+      "addvl x9, x9, #4\n"
+      "udot z20.s, z16.b, z24.b\n"
       "blt 5b\n"
-      "incb x9\n"
-      "incb x27\n"
-      "whilelt p9.b, x27, %x[width]\n"
-      "whilelt p8.b, x27, %x[width]\n"
+      "incb x28\n"
+      "whilelt p9.b, x28, %x[width]\n"
+      "whilelt p8.b, x28, %x[width]\n"
+      "incb x10\n"
       "b.any 2b\n"
-      "st1w { z23.s }, p2, [x28]\n"
-      "st1w { z22.s }, p2, [x28, #1, MUL VL]\n"
-      "st1w { z21.s }, p2, [x28, #2, MUL VL]\n"
-      "st1w { z20.s }, p2, [x28, #3, MUL VL]\n"
-      "addvl x28, x28, #4\n"
-      "mov %x[out], x28\n"
+      "st1w { z23.s }, p2, [x9]\n"
+      "st1w { z22.s }, p2, [x9, #1, MUL VL]\n"
+      "st1w { z21.s }, p2, [x9, #2, MUL VL]\n"
+      "st1w { z20.s }, p2, [x9, #3, MUL VL]\n"
+      "addvl x9, x9, #4\n"
+      "mov %x[out], x9\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }
 
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_fp32_fp32.hpp
index 141ab00..9579263 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave4VL_fp32_fp32.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,16 +10,16 @@
  * sell copies of the Software, and to permit persons to whom the Software is
  * furnished to do so, subject to the following conditions:
  *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
  */
 
 #if defined(__ARM_FEATURE_SVE)
@@ -34,92 +34,92 @@
 
   __asm__ __volatile__(
       ".inst 0xd503477f  // SMSTART ZA\n"
+      "cntw x15\n"
+      "cmp %x[height], x15\n"
       "cntw x14\n"
       "cntw x13, ALL, MUL #2\n"
       "cntw x11, ALL, MUL #3\n"
-      "ptrue p3.s\n"
-      "cntw x10\n"
-      "cmp %x[height], x10\n"
-      "csel x10, %x[height], x10, LT\n"
-      "sub x10, x10, #0x1\n"
-      "whilelt p2.s, XZR, %x[height]\n"
+      "csel x15, %x[height], x15, LT\n"
+      "mov x10, #0x0\n"
+      "ptrue p4.s\n"
+      "sub x15, x15, #0x1\n"
+      "whilelt p3.s, XZR, %x[height]\n"
       "whilelt p15.s, x14, %x[height]\n"
       "whilelt p14.s, x13, %x[height]\n"
       "whilelt p13.s, x11, %x[height]\n"
       "mov x9, %x[row_offset]\n"
       "mov x28, %x[out]\n"
-      "mov x27, #0x0\n"
-      "whilelt p12.s, x27, %x[width]\n"
-      "whilelt p11.s, x27, %x[width]\n"
-      "whilelt p10.s, x27, %x[width]\n"
-      "whilelt p9.s, x27, %x[width]\n"
-      "whilelt p8.s, x27, %x[width]\n"
+      "whilelt p12.s, x10, %x[width]\n"
+      "whilelt p11.s, x10, %x[width]\n"
+      "whilelt p10.s, x10, %x[width]\n"
+      "whilelt p9.s, x10, %x[width]\n"
+      "whilelt p8.s, x10, %x[width]\n"
       "1:"  // Width loop
-      "mov x12, #0x0\n"
-      "add x26, %x[in], XZR, LSL #3\n"
-      "add x25, %x[in], x14, LSL #3\n"
+      "add x27, %x[in], XZR, LSL #3\n"
+      "add x26, %x[in], x14, LSL #3\n"
+      "ldr x25, [x27], #0x8\n"
       "add x24, %x[in], x13, LSL #3\n"
-      "add x23, %x[in], x11, LSL #3\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x20, [x24], #0x8\n"
-      "ldr x19, [x23], #0x8\n"
-      "cbz x10, 3f\n"
+      "add x20, %x[in], x11, LSL #3\n"
+      "ldr x23, [x26], #0x8\n"
+      "mov x12, #0x0\n"
+      "ldr x22, [x24], #0x8\n"
+      "ldr x21, [x20], #0x8\n"
+      "cbz x15, 3f\n"
       "2:"  // Loads: Loop
-      ".inst 0x25306c40  // dup p0.s, p11.s/Z, p2.s[w12]\n"
-      ".inst 0xe08902c0  // ld1w { za0h.s[x12] }, p0/Z, [x22, x9, LSL #2]\n"
-      ".inst 0x253069e0  // dup p0.s, p10.s/Z, p15.s[w12]\n"
-      ".inst 0xe08902a4  // ld1w { za1h.s[x12] }, p0/Z, [x21, x9, LSL #2]\n"
-      ".inst 0x253065c0  // dup p0.s, p9.s/Z, p14.s[w12]\n"
-      ".inst 0xe0890288  // ld1w { za2h.s[x12] }, p0/Z, [x20, x9, LSL #2]\n"
-      ".inst 0x253061a0  // dup p0.s, p8.s/Z, p13.s[w12]\n"
-      ".inst 0xe089026c  // ld1w { za3h.s[x12] }, p0/Z, [x19, x9, LSL #2]\n"
-      "ldr x22, [x26], #0x8\n"
-      "ldr x21, [x25], #0x8\n"
-      "ldr x20, [x24], #0x8\n"
-      "ldr x19, [x23], #0x8\n"
+      ".inst 0x25306c60  // psel p0.s, p11.s/Z, p3.s[w12]\n"
+      ".inst 0x253069e2  // psel p2.s, p10.s/Z, p15.s[w12]\n"
+      ".inst 0xe0890320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x9, LSL #2]\n"
+      "ldr x25, [x27], #0x8\n"
+      ".inst 0x253065c1  // psel p1.s, p9.s/Z, p14.s[w12]\n"
+      ".inst 0x253061a0  // psel p0.s, p8.s/Z, p13.s[w12]\n"
+      ".inst 0xe0890ae4  // ld1w { za1h.s[x12] }, p2/Z, [x23, x9, LSL #2]\n"
+      "ldr x23, [x26], #0x8\n"
+      ".inst 0xe08906c8  // ld1w { za2h.s[x12] }, p1/Z, [x22, x9, LSL #2]\n"
+      "ldr x22, [x24], #0x8\n"
+      ".inst 0xe08902ac  // ld1w { za3h.s[x12] }, p0/Z, [x21, x9, LSL #2]\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x10\n"
+      "cmp x12, x15\n"
+      "ldr x21, [x20], #0x8\n"
       "blt 2b\n"
       "3:"  // Loads: Tail
-      ".inst 0x25306c40  // dup p0.s, p11.s/Z, p2.s[w12]\n"
-      ".inst 0xe08902c0  // ld1w { za0h.s[x12] }, p0/Z, [x22, x9, LSL #2]\n"
-      ".inst 0x253069e0  // dup p0.s, p10.s/Z, p15.s[w12]\n"
-      ".inst 0xe08902a4  // ld1w { za1h.s[x12] }, p0/Z, [x21, x9, LSL #2]\n"
-      ".inst 0x253065c0  // dup p0.s, p9.s/Z, p14.s[w12]\n"
-      ".inst 0xe0890288  // ld1w { za2h.s[x12] }, p0/Z, [x20, x9, LSL #2]\n"
-      ".inst 0x253061a0  // dup p0.s, p8.s/Z, p13.s[w12]\n"
-      ".inst 0xe089026c  // ld1w { za3h.s[x12] }, p0/Z, [x19, x9, LSL #2]\n"
+      "sub x20, %x[width], x10\n"
+      ".inst 0x25306c60  // psel p0.s, p11.s/Z, p3.s[w12]\n"
+      ".inst 0xe0890320  // ld1w { za0h.s[x12] }, p0/Z, [x25, x9, LSL #2]\n"
+      ".inst 0x253069e0  // psel p0.s, p10.s/Z, p15.s[w12]\n"
+      ".inst 0x253065c1  // psel p1.s, p9.s/Z, p14.s[w12]\n"
+      ".inst 0xe08902e4  // ld1w { za1h.s[x12] }, p0/Z, [x23, x9, LSL #2]\n"
+      ".inst 0x253061a0  // psel p0.s, p8.s/Z, p13.s[w12]\n"
+      "cmp x20, x14\n"
+      ".inst 0xe08906c8  // ld1w { za2h.s[x12] }, p1/Z, [x22, x9, LSL #2]\n"
+      ".inst 0xe08902ac  // ld1w { za3h.s[x12] }, p0/Z, [x21, x9, LSL #2]\n"
       "mov x12, #0x0\n"
-      "sub x19, %x[width], x27\n"
-      "cmp x19, x14\n"
-      "csel x19, x19, x14, LT\n"
+      "csel x20, x20, x14, LT\n"
       "4:"  // Stores: Loop
-      ".inst 0x25304d80  // dup p0.s, p3.s/Z, p12.s[w12]\n"
+      ".inst 0x25305180  // psel p0.s, p4.s/Z, p12.s[w12]\n"
       ".inst 0xe0bf8380  // st1w { za0v.s[x12] }, p0/Z, [x28, XZR, LSL #2]\n"
-      ".inst 0x25304d80  // dup p0.s, p3.s/Z, p12.s[w12]\n"
-      ".inst 0x25304d81  // dup p1.s, p3.s/Z, p12.s[w12]\n"
+      ".inst 0x25305180  // psel p0.s, p4.s/Z, p12.s[w12]\n"
       ".inst 0xe0ae8384  // st1w { za1v.s[x12] }, p0/Z, [x28, x14, LSL #2]\n"
-      ".inst 0x25304d80  // dup p0.s, p3.s/Z, p12.s[w12]\n"
+      ".inst 0x25305181  // psel p1.s, p4.s/Z, p12.s[w12]\n"
+      ".inst 0x25305180  // psel p0.s, p4.s/Z, p12.s[w12]\n"
       ".inst 0xe0ad8788  // st1w { za2v.s[x12] }, p1/Z, [x28, x13, LSL #2]\n"
       ".inst 0xe0ab838c  // st1w { za3v.s[x12] }, p0/Z, [x28, x11, LSL #2]\n"
-      "addvl x28, x28, #4\n"
       "add x12, x12, #0x1\n"
-      "cmp x12, x19\n"
+      "cmp x12, x20\n"
+      "addvl x28, x28, #4\n"
       "blt 4b\n"
+      "incw x10\n"
+      "whilelt p12.s, x10, %x[width]\n"
+      "whilelt p11.s, x10, %x[width]\n"
+      "whilelt p10.s, x10, %x[width]\n"
+      "whilelt p9.s, x10, %x[width]\n"
+      "whilelt p8.s, x10, %x[width]\n"
       "incw x9\n"
-      "incw x27\n"
-      "whilelt p12.s, x27, %x[width]\n"
-      "whilelt p11.s, x27, %x[width]\n"
-      "whilelt p10.s, x27, %x[width]\n"
-      "whilelt p9.s, x27, %x[width]\n"
-      "whilelt p8.s, x27, %x[width]\n"
       "b.any 1b\n"
       "mov %x[out], x28\n"
       ".inst 0xd503467f  // SMSTOP\n"
       : [out] "+&r" (out)
       : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
-      : "cc", "memory", "p0", "p1", "p2", "p3", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+      : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
     );
 }