diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a32_interleave6_block1_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a32_interleave6_block1_fp32_fp32.hpp
index 0742999..807511f 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a32_interleave6_block1_fp32_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a32_interleave6_block1_fp32_fp32.hpp
@@ -22,9 +22,6 @@
  * SOFTWARE.
  */
 #pragma once
-#if (defined(__GNUC__) && (__GNUC__ >= 7))
-#pragma GCC diagnostic ignored "-Wimplicit-fallthrough"
-#endif
 
 #ifdef __arm__
 
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 8054c2b..6a8caf6 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -58,19 +58,19 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q19, [x22], #0x10\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
+      "subs %x[width], %x[width], #0x10\n"
       "ldr q18, [x21], #0x10\n"
+      "cmp %x[width], #0x10\n"
       "ldr q17, [x20], #0x10\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "ldr q16, [x19], #0x10\n"
+      "prfm pldl1keep, [x22, #0x70]\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"
-      "prfm pldl1keep, [x19, #0x70]\n"
       "str q17, [%x[out_ptr], #0x20]\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "cmp %x[width], #0x10\n"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "bge 2b\n"
       "3:"  // Main loop skip
@@ -171,7 +171,7 @@
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "12:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 1650916..954a866 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -84,24 +84,24 @@
       "mov x22, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q19, [x23], #0x10\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q18, [x21], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "ldr q16, [x19], #0x10\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q19, [%x[out_ptr], #0x0]\n"
-      "sadalp v28.8h, v19.16b\n"
-      "prfm pldl1keep, [x19, #0x70]\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
-      "sadalp v27.8h, v18.16b\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"
+      "ldr q18, [x21], #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"
+      "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"
       "bge 3b\n"
       "5:"  // Main loop skip
@@ -215,7 +215,7 @@
       "add v24.4s, v24.4s, v20.4s\n"
       "str q24, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 af3efb2..c811462 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -84,24 +84,24 @@
       "mov x22, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q19, [x23], #0x10\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q18, [x21], #0x10\n"
-      "ldr q17, [x20], #0x10\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "ldr q16, [x19], #0x10\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "str q19, [%x[out_ptr], #0x0]\n"
-      "uadalp v28.8h, v19.16b\n"
-      "prfm pldl1keep, [x19, #0x70]\n"
-      "str q18, [%x[out_ptr], #0x10]\n"
-      "uadalp v27.8h, v18.16b\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"
+      "ldr q18, [x21], #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"
+      "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"
       "bge 3b\n"
       "5:"  // Main loop skip
@@ -215,7 +215,7 @@
       "add v24.4s, v24.4s, v20.4s\n"
       "str q24, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 34d25f2..4257429 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -31,7 +31,7 @@
 )
 {
   __asm__ __volatile__(
-      "movi v29.8h, #0x0\n"
+      "movi v30.8h, #0x0\n"
       "ldr x27, [%x[in], #0x0]\n"
       "cmp %x[height], #0x8\n"
       "ldr x26, [%x[in], #0x8]\n"
@@ -80,53 +80,53 @@
       "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d28, [x27], #0x8\n"
-      "zip1 v28.8h, v29.8h, v28.8h\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d27, [x26], #0x8\n"
-      "zip1 v27.8h, v29.8h, v27.8h\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d26, [x25], #0x8\n"
-      "zip1 v26.8h, v29.8h, v26.8h\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d25, [x24], #0x8\n"
-      "zip1 v20.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "zip1 v25.8h, v29.8h, v25.8h\n"
-      "ldr d24, [x23], #0x8\n"
-      "zip1 v19.4s, v27.4s, v25.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "zip1 v24.8h, v29.8h, v24.8h\n"
-      "ldr d23, [x22], #0x8\n"
-      "zip1 v16.4s, v20.4s, v19.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip1 v23.8h, v29.8h, v23.8h\n"
-      "ldr d22, [x21], #0x8\n"
-      "zip2 v19.4s, v20.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "zip1 v22.8h, v29.8h, v22.8h\n"
-      "ldr d21, [x20], #0x8\n"
-      "zip1 v18.4s, v24.4s, v22.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v21.8h, v29.8h, v21.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v17.4s, v23.4s, v21.4s\n"
+      "ldr d29, [x27], #0x8\n"
+      "zip1 v29.8h, v30.8h, v29.8h\n"
+      "ldr d28, [x26], #0x8\n"
       "subs %x[width], %x[width], #0x4\n"
-      "zip2 v20.4s, v28.4s, v26.4s\n"
+      "zip1 v28.8h, v30.8h, v28.8h\n"
+      "ldr d24, [x25], #0x8\n"
       "cmp %x[width], #0x4\n"
-      "zip1 v16.4s, v18.4s, v17.4s\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip1 v20.4s, v28.4s, v27.4s\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"
+      "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, v18.4s, v17.4s\n"
-      "str q19, [%x[out_ptr], #0x20]\n"
-      "zip2 v19.4s, v27.4s, v25.4s\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, v20.4s, v19.4s\n"
+      "zip1 v16.4s, v24.4s, v19.4s\n"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v18.4s, v24.4s, v22.4s\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], #0x50]\n"
-      "zip2 v16.4s, v20.4s, v19.4s\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], #0x70]\n"
@@ -135,77 +135,77 @@
       "3:"  // Main loop skip
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
-      "ldr s28, [x27], #0x4\n"
-      "ldr s27, [x26], #0x4\n"
-      "ldr s26, [x25], #0x4\n"
-      "ldr s25, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\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"
-      "mov x19, #0x2\n"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v28.h }[2], [x27]\n"
-      "ld1 { v27.h }[2], [x26]\n"
-      "ld1 { v26.h }[2], [x25]\n"
-      "ld1 { v25.h }[2], [x24]\n"
-      "ld1 { v24.h }[2], [x23]\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"
-      "mov x19, #0x3\n"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr h28, [x27, #0x0]\n"
-      "ldr h27, [x26, #0x0]\n"
-      "ldr h26, [x25, #0x0]\n"
-      "ldr h25, [x24, #0x0]\n"
-      "ldr h24, [x23, #0x0]\n"
+      "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"
-      "mov x19, #0x1\n"
       "5:"  // Odd load end
-      "zip1 v28.8h, v29.8h, v28.8h\n"
+      "zip1 v29.8h, v30.8h, v29.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v27.8h, v29.8h, v27.8h\n"
-      "zip1 v26.8h, v29.8h, v26.8h\n"
-      "zip1 v25.8h, v29.8h, v25.8h\n"
-      "zip1 v24.8h, v29.8h, v24.8h\n"
-      "zip1 v23.8h, v29.8h, v23.8h\n"
-      "zip1 v22.8h, v29.8h, v22.8h\n"
-      "zip1 v21.8h, v29.8h, v21.8h\n"
-      "zip1 v20.4s, v28.4s, v26.4s\n"
-      "zip1 v19.4s, v27.4s, v25.4s\n"
-      "zip1 v16.4s, v20.4s, v19.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v24.4s, v22.4s\n"
-      "zip1 v17.4s, v23.4s, v21.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\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 v19.4s, v20.4s, v19.4s\n"
-      "zip2 v16.4s, v18.4s, v17.4s\n"
-      "str q19, [%x[out_ptr], #0x0]\n"
-      "str q16, [%x[out_ptr], #0x10]\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 v20.4s, v28.4s, v26.4s\n"
-      "zip2 v19.4s, v27.4s, v25.4s\n"
-      "zip1 v16.4s, v20.4s, v19.4s\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"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.4s, v24.4s, v22.4s\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"
       "6:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 d547957..62d1657 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,51 +80,51 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q30, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q29, [x26], #0x10\n"
-      "ldr q28, [x25], #0x10\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q27, [x24], #0x10\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q24, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q23, [x21], #0x10\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
+      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v20.8h, v17.8h\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v21.8h, v29.8h, v21.8h\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"
+      "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"
+      "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"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x20]\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v20.8h, v19.8h\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, v24.8h, v21.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\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"
@@ -138,129 +138,129 @@
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
-      "ldr d23, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "ldr d23, [x20], #0x8\n"
       "tbz %x[width], #1, 4f\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v24.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
-      "ld1 { v23.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v24.h }[6], [x23]\n"
-      "ld1 { v25.h }[6], [x22]\n"
-      "ld1 { v23.h }[6], [x21]\n"
-      "ld1 { v22.h }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 7f\n"
       "4:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[4], [x27]\n"
       "ld1 { v29.h }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v24.h }[4], [x23]\n"
-      "ld1 { v25.h }[4], [x22]\n"
-      "ld1 { v23.h }[4], [x21]\n"
-      "ld1 { v22.h }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
+      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
-      "ldr s23, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
-      "mov x19, #0x2\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
+      "ldr s23, [x20], #0x4\n"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v24.h }[2], [x23]\n"
-      "ld1 { v25.h }[2], [x22]\n"
-      "ld1 { v23.h }[2], [x21]\n"
-      "ld1 { v22.h }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 7f\n"
       "6:"  // odd_loads_1_0
       "ldr h30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h24, [x23, #0x0]\n"
-      "ldr h25, [x22, #0x0]\n"
-      "ldr h23, [x21, #0x0]\n"
-      "ldr h22, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h21, [x22, #0x0]\n"
+      "ldr h24, [x21, #0x0]\n"
+      "ldr h23, [x20, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v24.8h\n"
+      "zip1 v26.8h, v30.8h, v25.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "zip1 v19.8h, v27.8h, v22.8h\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"
+      "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"
+      "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"
+      "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, v20.8h, v17.8h\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"
       "beq 8f\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
+      "zip2 v18.8h, v25.8h, v20.8h\n"
       "zip2 v17.8h, v21.8h, v19.8h\n"
-      "subs x19, x19, #0x1\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"
-      "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\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 %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v24.8h, 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 %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 b45e622..b67840b 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -79,132 +79,132 @@
       "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d29, [x27], #0x8\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d28, [x26], #0x8\n"
-      "ldr d27, [x25], #0x8\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d26, [x24], #0x8\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d24, [x22], #0x8\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr d23, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "fcvtl v29.4s, v29.4h\n"
-      "fcvtl v28.4s, v28.4h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "fcvtl v27.4s, v27.4h\n"
-      "zip1 v20.4s, v29.4s, v27.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "fcvtl v26.4s, v26.4h\n"
-      "zip2 v18.4s, v29.4s, v27.4s\n"
-      "fcvtl v25.4s, v25.4h\n"
-      "fcvtl v24.4s, v24.4h\n"
-      "zip1 v19.4s, v28.4s, v26.4s\n"
-      "fcvtl v23.4s, v23.4h\n"
-      "zip2 v17.4s, v28.4s, v26.4s\n"
-      "fcvtl v22.4s, v22.4h\n"
-      "zip1 v16.4s, v20.4s, v19.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v21.4s, v20.4s, v19.4s\n"
+      "ldr d30, [x27], #0x8\n"
       "subs %x[width], %x[width], #0x4\n"
-      "zip1 v20.4s, v18.4s, v17.4s\n"
+      "ldr d29, [x26], #0x8\n"
       "cmp %x[width], #0x4\n"
-      "zip2 v19.4s, v18.4s, v17.4s\n"
-      "zip1 v18.4s, v25.4s, v23.4s\n"
-      "zip1 v17.4s, v24.4s, v22.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\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"
+      "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"
+      "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"
+      "prfm pldl1keep, [x25, #0x70]\n"
+      "fcvtl v24.4s, v24.4h\n"
+      "zip2 v16.4s, v29.4s, v21.4s\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"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.4s, v18.4s, v17.4s\n"
-      "str q21, [%x[out_ptr], #0x20]\n"
-      "zip2 v18.4s, v25.4s, v23.4s\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v17.4s, v24.4s, v22.4s\n"
-      "str q20, [%x[out_ptr], #0x40]\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v16.4s, v18.4s, v17.4s\n"
-      "str q19, [%x[out_ptr], #0x60]\n"
+      "zip2 v17.4s, v19.4s, v17.4s\n"
+      "str q22, [%x[out_ptr], #0x20]\n"
+      "zip2 v16.4s, v26.4s, v24.4s\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"
       "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"
-      "ldr s27, [x25], #0x4\n"
-      "ldr s26, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s24, [x22], #0x4\n"
-      "ldr s23, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\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"
       "tbz %x[width], #0, 5f\n"
-      "ld1 { v29.h }[2], [x27]\n"
-      "ld1 { v28.h }[2], [x26]\n"
-      "ld1 { v27.h }[2], [x25]\n"
-      "ld1 { v26.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v24.h }[2], [x22]\n"
-      "ld1 { v23.h }[2], [x21]\n"
-      "ld1 { v22.h }[2], [x20]\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"
       "b 5f\n"
       "4:"  // odd_loads_1_0
-      "ldr h29, [x27, #0x0]\n"
-      "ldr h28, [x26, #0x0]\n"
-      "ldr h27, [x25, #0x0]\n"
-      "ldr h26, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h24, [x22, #0x0]\n"
-      "ldr h23, [x21, #0x0]\n"
-      "ldr h22, [x20, #0x0]\n"
+      "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"
       "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 v20.4s, v29.4s, v27.4s\n"
+      "zip1 v18.4s, v29.4s, v21.4s\n"
       "fcvtl v26.4s, v26.4h\n"
       "fcvtl v25.4s, v25.4h\n"
-      "zip1 v19.4s, v28.4s, v26.4s\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"
-      "fcvtl v23.4s, v23.4h\n"
-      "zip1 v16.4s, v20.4s, v19.4s\n"
-      "fcvtl v22.4s, v22.4h\n"
-      "zip1 v18.4s, v25.4s, v23.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v17.4s, v24.4s, v22.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
+      "zip1 v17.4s, v26.4s, v24.4s\n"
+      "zip1 v16.4s, v19.4s, v17.4s\n"
       "str q16, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v21.4s, v20.4s, v19.4s\n"
-      "zip2 v16.4s, v18.4s, v17.4s\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"
+      "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"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "subs x19, x19, #0x1\n"
-      "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      "beq 6f\n"
-      "zip2 v18.4s, v29.4s, v27.4s\n"
-      "zip2 v17.4s, v28.4s, v26.4s\n"
-      "zip1 v20.4s, v18.4s, v17.4s\n"
-      "str q20, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.4s, v25.4s, v23.4s\n"
-      "zip2 v17.4s, v24.4s, v22.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "6:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 3f38859..eefb854 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,45 +80,45 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q28, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q27, [x26], #0x10\n"
-      "ldr q26, [x25], #0x10\n"
-      "zip1 v23.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q22, [x24], #0x10\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q24, [x22], #0x10\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
-      "ldr q21, [x20], #0x10\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v16.4s, v24.4s, v21.4s\n"
       "subs %x[width], %x[width], #0x4\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x4\n"
-      "zip2 v16.4s, v18.4s, v16.4s\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"
+      "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"
+      "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"
+      "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 v19.4s, v25.4s, v19.4s\n"
-      "str q23, [%x[out_ptr], #0x20]\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q20, [%x[out_ptr], #0x40]\n"
-      "zip2 v17.4s, v26.4s, v22.4s\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v16.4s, v19.4s, v18.4s\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"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -127,68 +127,68 @@
       "cbz %x[width], 6f\n"
       "tbz %x[width], #1, 4f\n"
       "ldr d28, [x27], #0x8\n"
-      "ldr d27, [x26], #0x8\n"
-      "ldr d26, [x25], #0x8\n"
-      "ldr d22, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d24, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
-      "ldr d21, [x20], #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"
       "tbz %x[width], #0, 5f\n"
       "ld1 { v28.s }[2], [x27]\n"
-      "ld1 { v27.s }[2], [x26]\n"
-      "ld1 { v26.s }[2], [x25]\n"
-      "ld1 { v22.s }[2], [x24]\n"
-      "ld1 { v25.s }[2], [x23]\n"
-      "ld1 { v24.s }[2], [x22]\n"
-      "ld1 { v19.s }[2], [x21]\n"
-      "ld1 { v21.s }[2], [x20]\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"
       "b 5f\n"
       "4:"  // odd_loads_1_0
       "ldr s28, [x27, #0x0]\n"
-      "ldr s27, [x26, #0x0]\n"
-      "ldr s26, [x25, #0x0]\n"
-      "ldr s22, [x24, #0x0]\n"
-      "ldr s25, [x23, #0x0]\n"
-      "ldr s24, [x22, #0x0]\n"
-      "ldr s19, [x21, #0x0]\n"
-      "ldr s21, [x20, #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"
       "5:"  // Odd load end
-      "zip1 v23.4s, v28.4s, v26.4s\n"
+      "zip1 v22.4s, v28.4s, v25.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "zip1 v16.4s, v24.4s, v21.4s\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "zip2 v16.4s, v18.4s, v16.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 6f\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "str q20, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v25.4s, v19.4s\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "6:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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", "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", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 03f552a..b0523b9 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,51 +80,51 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q30, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q29, [x26], #0x10\n"
-      "ldr q28, [x25], #0x10\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q27, [x24], #0x10\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q24, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q23, [x21], #0x10\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
+      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v20.8h, v17.8h\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v21.8h, v29.8h, v21.8h\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"
+      "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"
+      "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"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x20]\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v20.8h, v19.8h\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, v24.8h, v21.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\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"
@@ -138,129 +138,129 @@
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
-      "ldr d23, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "ldr d23, [x20], #0x8\n"
       "tbz %x[width], #1, 4f\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v24.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
-      "ld1 { v23.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v24.h }[6], [x23]\n"
-      "ld1 { v25.h }[6], [x22]\n"
-      "ld1 { v23.h }[6], [x21]\n"
-      "ld1 { v22.h }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 7f\n"
       "4:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[4], [x27]\n"
       "ld1 { v29.h }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v24.h }[4], [x23]\n"
-      "ld1 { v25.h }[4], [x22]\n"
-      "ld1 { v23.h }[4], [x21]\n"
-      "ld1 { v22.h }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
+      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
-      "ldr s23, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
-      "mov x19, #0x2\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
+      "ldr s23, [x20], #0x4\n"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v30.h }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v24.h }[2], [x23]\n"
-      "ld1 { v25.h }[2], [x22]\n"
-      "ld1 { v23.h }[2], [x21]\n"
-      "ld1 { v22.h }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 7f\n"
       "6:"  // odd_loads_1_0
       "ldr h30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h24, [x23, #0x0]\n"
-      "ldr h25, [x22, #0x0]\n"
-      "ldr h23, [x21, #0x0]\n"
-      "ldr h22, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h21, [x22, #0x0]\n"
+      "ldr h24, [x21, #0x0]\n"
+      "ldr h23, [x20, #0x0]\n"
       "7:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v24.8h\n"
+      "zip1 v26.8h, v30.8h, v25.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "zip1 v19.8h, v27.8h, v22.8h\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"
+      "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"
+      "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"
+      "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, v20.8h, v17.8h\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"
       "beq 8f\n"
-      "zip2 v16.8h, v20.8h, v17.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
+      "zip2 v18.8h, v25.8h, v20.8h\n"
       "zip2 v17.8h, v21.8h, v19.8h\n"
-      "subs x19, x19, #0x1\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"
-      "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\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 %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
-      "zip2 v18.8h, v24.8h, 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 %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 35c7719..292a38f 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -96,58 +96,58 @@
       "movi v1.8h, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q30, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q29, [x26], #0x10\n"
-      "ldr q28, [x25], #0x10\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q27, [x24], #0x10\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q24, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q23, [x21], #0x10\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "add x19, x19, #0x1\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
+      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "ldr q28, [x25], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v20.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.8h, v21.8h, v19.8h\n"
-      "add v1.8h, v1.8h, v17.8h\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v21.8h, v29.8h, v21.8h\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"
+      "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"
+      "prfm pldl1keep, [x22, #0x70]\n"
       "zip1 v17.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "add v1.8h, v1.8h, v17.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "zip1 v18.8h, v24.8h, v21.8h\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 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "zip2 v17.8h, v22.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "add v1.8h, v1.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "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 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
       "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v24.8h, v21.8h\n"
+      "zip2 v18.8h, v25.8h, v20.8h\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "zip2 v17.8h, v21.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x60]\n"
       "add v1.8h, v1.8h, v16.8h\n"
@@ -163,140 +163,140 @@
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
-      "ldr d23, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "ldr d23, [x20], #0x8\n"
       "tbz %x[width], #1, 6f\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v24.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
-      "ld1 { v23.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v24.h }[6], [x23]\n"
-      "ld1 { v25.h }[6], [x22]\n"
-      "ld1 { v23.h }[6], [x21]\n"
-      "ld1 { v22.h }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 9f\n"
       "6:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[4], [x27]\n"
       "ld1 { v29.h }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v24.h }[4], [x23]\n"
-      "ld1 { v25.h }[4], [x22]\n"
-      "ld1 { v23.h }[4], [x21]\n"
-      "ld1 { v22.h }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
+      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
-      "ldr s23, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
-      "mov x19, #0x2\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
+      "ldr s23, [x20], #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v24.h }[2], [x23]\n"
-      "ld1 { v25.h }[2], [x22]\n"
-      "ld1 { v23.h }[2], [x21]\n"
-      "ld1 { v22.h }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 9f\n"
       "8:"  // odd_loads_1_0
       "ldr h30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h24, [x23, #0x0]\n"
-      "ldr h25, [x22, #0x0]\n"
-      "ldr h23, [x21, #0x0]\n"
-      "ldr h22, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h21, [x22, #0x0]\n"
+      "ldr h24, [x21, #0x0]\n"
+      "ldr h23, [x20, #0x0]\n"
       "9:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v24.8h\n"
+      "zip1 v26.8h, v30.8h, v25.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v17.8h, v20.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip2 v16.8h, v21.8h, v19.8h\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 v1.8h, v1.8h, v17.8h\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, v16.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
+      "zip2 v18.8h, v26.8h, v20.8h\n"
+      "zip2 v17.8h, v22.8h, v19.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\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 v1.8h, v1.8h, 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"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\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"
+      "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"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, 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"
+      "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"
       "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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 582836f..6cfed8f 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -79,206 +79,206 @@
       "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d30, [x27], #0x8\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d28, [x25], #0x8\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d27, [x24], #0x8\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr d26, [x21], #0x8\n"
-      "ldr d25, [x20], #0x8\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "sshll v30.8h, v30.8b, #0x0\n"
-      "sshll v29.8h, v29.8b, #0x0\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "sshll v28.8h, v28.8b, #0x0\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "sshll v27.8h, v27.8b, #0x0\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "sshll v21.8h, v21.8b, #0x0\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "sshll v26.8h, v26.8b, #0x0\n"
-      "sshll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr d31, [x27], #0x8\n"
+      "sshll v31.8h, v31.8b, #0x0\n"
+      "ldr d30, [x26], #0x8\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "sshll v30.8h, v30.8b, #0x0\n"
+      "ldr d29, [x25], #0x8\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
+      "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 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"
+      "sshll v26.8h, v26.8b, #0x0\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"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v22.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v21.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v20.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v19.8h, v17.8h\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 s30, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "ldr s28, [x25], #0x4\n"
-      "ldr s27, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s26, [x21], #0x4\n"
-      "ldr s25, [x20], #0x4\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"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v30.h }[2], [x27], #0x2\n"
-      "ld1 { v29.h }[2], [x26], #0x2\n"
-      "ld1 { v28.h }[2], [x25], #0x2\n"
-      "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v23.h }[2], [x23], #0x2\n"
-      "ld1 { v21.h }[2], [x22], #0x2\n"
-      "ld1 { v26.h }[2], [x21], #0x2\n"
-      "ld1 { v25.h }[2], [x20], #0x2\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"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[6], [x27]\n"
-      "ld1 { v29.b }[6], [x26]\n"
-      "ld1 { v28.b }[6], [x25]\n"
-      "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v23.b }[6], [x23]\n"
-      "ld1 { v21.b }[6], [x22]\n"
-      "ld1 { v26.b }[6], [x21]\n"
-      "ld1 { v25.b }[6], [x20]\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"
       "b 7f\n"
       "4:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[4], [x27]\n"
-      "ld1 { v29.b }[4], [x26]\n"
-      "ld1 { v28.b }[4], [x25]\n"
-      "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v23.b }[4], [x23]\n"
-      "ld1 { v21.b }[4], [x22]\n"
-      "ld1 { v26.b }[4], [x21]\n"
-      "ld1 { v25.b }[4], [x20]\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"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr h30, [x27], #0x2\n"
-      "ldr h29, [x26], #0x2\n"
-      "ldr h28, [x25], #0x2\n"
-      "ldr h27, [x24], #0x2\n"
-      "ldr h23, [x23], #0x2\n"
-      "ldr h21, [x22], #0x2\n"
-      "ldr h26, [x21], #0x2\n"
-      "ldr h25, [x20], #0x2\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"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[2], [x27]\n"
-      "ld1 { v29.b }[2], [x26]\n"
-      "ld1 { v28.b }[2], [x25]\n"
-      "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v23.b }[2], [x23]\n"
-      "ld1 { v21.b }[2], [x22]\n"
-      "ld1 { v26.b }[2], [x21]\n"
-      "ld1 { v25.b }[2], [x20]\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"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr b30, [x27, #0x0]\n"
-      "ldr b29, [x26, #0x0]\n"
-      "ldr b28, [x25, #0x0]\n"
-      "ldr b27, [x24, #0x0]\n"
-      "ldr b23, [x23, #0x0]\n"
-      "ldr b21, [x22, #0x0]\n"
-      "ldr b26, [x21, #0x0]\n"
-      "ldr b25, [x20, #0x0]\n"
+      "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"
       "7:"  // Odd load end
+      "sshll v31.8h, v31.8b, #0x0\n"
+      "subs x19, x19, #0x1\n"
       "sshll v30.8h, v30.8b, #0x0\n"
       "sshll v29.8h, v29.8b, #0x0\n"
       "sshll v28.8h, v28.8b, #0x0\n"
-      "sshll v27.8h, v27.8b, #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 v26.8h, v26.8b, #0x0\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
-      "sshll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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"
+      "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"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v21.8h, 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 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v25.8h, v31.8h, v25.8h\n"
+      "zip2 v19.8h, v29.8h, v27.8h\n"
       "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
+      "zip1 v20.8h, v25.8h, v19.8h\n"
       "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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"
-      "subs x19, x19, #0x1\n"
+      "zip2 v16.8h, v20.8h, 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, v23.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 35dc3dc..b710861 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -96,72 +96,72 @@
       "movi v1.8h, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr d30, [x27], #0x8\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d28, [x25], #0x8\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d27, [x24], #0x8\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr d26, [x21], #0x8\n"
-      "ldr d25, [x20], #0x8\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
       "sshll v30.8h, v30.8b, #0x0\n"
-      "sshll v29.8h, v29.8b, #0x0\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "sshll v28.8h, v28.8b, #0x0\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "sshll v27.8h, v27.8b, #0x0\n"
-      "sshll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "sshll v21.8h, v21.8b, #0x0\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "sshll v26.8h, v26.8b, #0x0\n"
-      "sshll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr d29, [x26], #0x8\n"
       "add x19, x19, #0x1\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "sshll v29.8h, v29.8b, #0x0\n"
+      "ldr d28, [x25], #0x8\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
+      "sshll v28.8h, v28.8b, #0x0\n"
+      "ldr d27, [x24], #0x8\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v18.8h, v24.8h, v20.8h\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"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v26.8h\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 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v16.8h, v21.8h, v17.8h\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, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v23.8h, v20.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x70]\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
@@ -171,148 +171,148 @@
       "ldr s29, [x26], #0x4\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s26, [x21], #0x4\n"
-      "ldr s25, [x20], #0x4\n"
+      "ldr s24, [x23], #0x4\n"
+      "ldr s23, [x22], #0x4\n"
+      "ldr s21, [x21], #0x4\n"
+      "ldr s26, [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
       "ld1 { v30.h }[2], [x27], #0x2\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.h }[2], [x26], #0x2\n"
       "ld1 { v28.h }[2], [x25], #0x2\n"
       "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v23.h }[2], [x23], #0x2\n"
-      "ld1 { v21.h }[2], [x22], #0x2\n"
-      "ld1 { v26.h }[2], [x21], #0x2\n"
-      "ld1 { v25.h }[2], [x20], #0x2\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.b }[6], [x26]\n"
       "ld1 { v28.b }[6], [x25]\n"
       "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v23.b }[6], [x23]\n"
-      "ld1 { v21.b }[6], [x22]\n"
-      "ld1 { v26.b }[6], [x21]\n"
-      "ld1 { v25.b }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 9f\n"
       "6:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[4], [x27]\n"
       "ld1 { v29.b }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.b }[4], [x25]\n"
       "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v23.b }[4], [x23]\n"
-      "ld1 { v21.b }[4], [x22]\n"
-      "ld1 { v26.b }[4], [x21]\n"
-      "ld1 { v25.b }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
       "ldr h30, [x27], #0x2\n"
       "ldr h29, [x26], #0x2\n"
+      "mov x19, #0x2\n"
       "ldr h28, [x25], #0x2\n"
       "ldr h27, [x24], #0x2\n"
-      "ldr h23, [x23], #0x2\n"
-      "ldr h21, [x22], #0x2\n"
-      "ldr h26, [x21], #0x2\n"
-      "ldr h25, [x20], #0x2\n"
-      "mov x19, #0x2\n"
+      "ldr h24, [x23], #0x2\n"
+      "ldr h23, [x22], #0x2\n"
+      "ldr h21, [x21], #0x2\n"
+      "ldr h26, [x20], #0x2\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.b }[2], [x26]\n"
       "ld1 { v28.b }[2], [x25]\n"
       "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v23.b }[2], [x23]\n"
-      "ld1 { v21.b }[2], [x22]\n"
-      "ld1 { v26.b }[2], [x21]\n"
-      "ld1 { v25.b }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 9f\n"
       "8:"  // odd_loads_1_0
       "ldr b30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr b29, [x26, #0x0]\n"
       "ldr b28, [x25, #0x0]\n"
       "ldr b27, [x24, #0x0]\n"
-      "ldr b23, [x23, #0x0]\n"
-      "ldr b21, [x22, #0x0]\n"
-      "ldr b26, [x21, #0x0]\n"
-      "ldr b25, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr b24, [x23, #0x0]\n"
+      "ldr b23, [x22, #0x0]\n"
+      "ldr b21, [x21, #0x0]\n"
+      "ldr b26, [x20, #0x0]\n"
       "9:"  // Odd load end
       "sshll v30.8h, v30.8b, #0x0\n"
+      "subs x19, x19, #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 v24.8h, v30.8h, v23.8h\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 v20.8h, v28.8h, v26.8h\n"
-      "sshll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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 %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"
+      "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"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v24.8h, v30.8h, v24.8h\n"
+      "zip2 v20.8h, v28.8h, v21.8h\n"
       "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\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"
       "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 %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v23.8h, 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"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "add v1.8h, v1.8h, v17.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 bfa8989..24ece9a 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -96,58 +96,58 @@
       "movi v1.8h, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q30, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q29, [x26], #0x10\n"
-      "ldr q28, [x25], #0x10\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q27, [x24], #0x10\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q24, [x23], #0x10\n"
-      "zip1 v26.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q23, [x21], #0x10\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "add x19, x19, #0x1\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
+      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
+      "ldr q28, [x25], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v20.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x10]\n"
-      "zip2 v16.8h, v21.8h, v19.8h\n"
-      "add v1.8h, v1.8h, v17.8h\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v21.8h, v29.8h, v21.8h\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"
+      "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"
+      "prfm pldl1keep, [x22, #0x70]\n"
       "zip1 v17.8h, v18.8h, v16.8h\n"
-      "str q17, [%x[out_ptr], #0x20]\n"
-      "zip2 v16.8h, v18.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "add v1.8h, v1.8h, v17.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
-      "zip1 v18.8h, v24.8h, v21.8h\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 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v20.8h, v19.8h\n"
+      "zip2 v17.8h, v22.8h, v19.8h\n"
+      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "str q16, [%x[out_ptr], #0x20]\n"
+      "add v1.8h, v1.8h, v16.8h\n"
+      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "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 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x40]\n"
       "add v1.8h, v1.8h, v16.8h\n"
       "zip2 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v24.8h, v21.8h\n"
+      "zip2 v18.8h, v25.8h, v20.8h\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\n"
+      "zip2 v17.8h, v21.8h, v19.8h\n"
       "zip1 v16.8h, v18.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x60]\n"
       "add v1.8h, v1.8h, v16.8h\n"
@@ -163,140 +163,140 @@
       "ldr d29, [x26], #0x8\n"
       "ldr d28, [x25], #0x8\n"
       "ldr d27, [x24], #0x8\n"
-      "ldr d24, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
-      "ldr d23, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d25, [x23], #0x8\n"
+      "ldr d21, [x22], #0x8\n"
+      "ldr d24, [x21], #0x8\n"
+      "ldr d23, [x20], #0x8\n"
       "tbz %x[width], #1, 6f\n"
       "ld1 { v30.s }[2], [x27], #0x4\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.s }[2], [x26], #0x4\n"
       "ld1 { v28.s }[2], [x25], #0x4\n"
       "ld1 { v27.s }[2], [x24], #0x4\n"
-      "ld1 { v24.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
-      "ld1 { v23.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.h }[6], [x26]\n"
       "ld1 { v28.h }[6], [x25]\n"
       "ld1 { v27.h }[6], [x24]\n"
-      "ld1 { v24.h }[6], [x23]\n"
-      "ld1 { v25.h }[6], [x22]\n"
-      "ld1 { v23.h }[6], [x21]\n"
-      "ld1 { v22.h }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 9f\n"
       "6:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[4], [x27]\n"
       "ld1 { v29.h }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.h }[4], [x25]\n"
       "ld1 { v27.h }[4], [x24]\n"
-      "ld1 { v24.h }[4], [x23]\n"
-      "ld1 { v25.h }[4], [x22]\n"
-      "ld1 { v23.h }[4], [x21]\n"
-      "ld1 { v22.h }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
       "ldr s30, [x27], #0x4\n"
       "ldr s29, [x26], #0x4\n"
+      "mov x19, #0x2\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s24, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
-      "ldr s23, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
-      "mov x19, #0x2\n"
+      "ldr s25, [x23], #0x4\n"
+      "ldr s21, [x22], #0x4\n"
+      "ldr s24, [x21], #0x4\n"
+      "ldr s23, [x20], #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.h }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.h }[2], [x26]\n"
       "ld1 { v28.h }[2], [x25]\n"
       "ld1 { v27.h }[2], [x24]\n"
-      "ld1 { v24.h }[2], [x23]\n"
-      "ld1 { v25.h }[2], [x22]\n"
-      "ld1 { v23.h }[2], [x21]\n"
-      "ld1 { v22.h }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 9f\n"
       "8:"  // odd_loads_1_0
       "ldr h30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr h29, [x26, #0x0]\n"
       "ldr h28, [x25, #0x0]\n"
       "ldr h27, [x24, #0x0]\n"
-      "ldr h24, [x23, #0x0]\n"
-      "ldr h25, [x22, #0x0]\n"
-      "ldr h23, [x21, #0x0]\n"
-      "ldr h22, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr h25, [x23, #0x0]\n"
+      "ldr h21, [x22, #0x0]\n"
+      "ldr h24, [x21, #0x0]\n"
+      "ldr h23, [x20, #0x0]\n"
       "9:"  // Odd load end
-      "zip1 v26.8h, v30.8h, v24.8h\n"
+      "zip1 v26.8h, v30.8h, v25.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v28.8h, v23.8h\n"
-      "zip1 v20.8h, v26.8h, v18.8h\n"
-      "zip1 v21.8h, v29.8h, v25.8h\n"
-      "zip1 v19.8h, v27.8h, v22.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v20.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v17.8h, v20.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v17.8h\n"
-      "str q17, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v26.8h, v18.8h\n"
-      "zip2 v16.8h, v21.8h, v19.8h\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 v1.8h, v1.8h, v17.8h\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, v16.8h\n"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v24.8h, v30.8h, v24.8h\n"
-      "zip2 v21.8h, v28.8h, v23.8h\n"
+      "zip2 v18.8h, v26.8h, v20.8h\n"
+      "zip2 v17.8h, v22.8h, v19.8h\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v20.8h, v29.8h, v25.8h\n"
-      "zip2 v19.8h, v27.8h, v22.8h\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 v1.8h, v1.8h, 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"
+      "str q16, [%x[out_ptr], #0x0]\n"
       "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v24.8h, v21.8h\n"
-      "zip2 v17.8h, v20.8h, v19.8h\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"
+      "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"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, 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"
+      "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"
       "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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 86b90f1..0db2f7f 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -79,206 +79,206 @@
       "prfm pldl1keep, [x20, #0x40]\n"
       "blt 3f\n"
       "2:"  // Main loop head
-      "ldr d30, [x27], #0x8\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d28, [x25], #0x8\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d27, [x24], #0x8\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr d26, [x21], #0x8\n"
-      "ldr d25, [x20], #0x8\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ushll v30.8h, v30.8b, #0x0\n"
-      "ushll v29.8h, v29.8b, #0x0\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "ushll v28.8h, v28.8b, #0x0\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "ushll v27.8h, v27.8b, #0x0\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "ushll v21.8h, v21.8b, #0x0\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "ushll v26.8h, v26.8b, #0x0\n"
-      "ushll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr d31, [x27], #0x8\n"
+      "ushll v31.8h, v31.8b, #0x0\n"
+      "ldr d30, [x26], #0x8\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "ushll v30.8h, v30.8b, #0x0\n"
+      "ldr d29, [x25], #0x8\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
+      "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 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"
+      "ushll v26.8h, v26.8b, #0x0\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"
       "str q16, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v22.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v21.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x40]\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v20.8h, v17.8h\n"
       "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v19.8h, v17.8h\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 s30, [x27], #0x4\n"
-      "ldr s29, [x26], #0x4\n"
-      "ldr s28, [x25], #0x4\n"
-      "ldr s27, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s26, [x21], #0x4\n"
-      "ldr s25, [x20], #0x4\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"
       "tbz %x[width], #1, 4f\n"
-      "ld1 { v30.h }[2], [x27], #0x2\n"
-      "ld1 { v29.h }[2], [x26], #0x2\n"
-      "ld1 { v28.h }[2], [x25], #0x2\n"
-      "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v23.h }[2], [x23], #0x2\n"
-      "ld1 { v21.h }[2], [x22], #0x2\n"
-      "ld1 { v26.h }[2], [x21], #0x2\n"
-      "ld1 { v25.h }[2], [x20], #0x2\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"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[6], [x27]\n"
-      "ld1 { v29.b }[6], [x26]\n"
-      "ld1 { v28.b }[6], [x25]\n"
-      "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v23.b }[6], [x23]\n"
-      "ld1 { v21.b }[6], [x22]\n"
-      "ld1 { v26.b }[6], [x21]\n"
-      "ld1 { v25.b }[6], [x20]\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"
       "b 7f\n"
       "4:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[4], [x27]\n"
-      "ld1 { v29.b }[4], [x26]\n"
-      "ld1 { v28.b }[4], [x25]\n"
-      "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v23.b }[4], [x23]\n"
-      "ld1 { v21.b }[4], [x22]\n"
-      "ld1 { v26.b }[4], [x21]\n"
-      "ld1 { v25.b }[4], [x20]\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"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
-      "ldr h30, [x27], #0x2\n"
-      "ldr h29, [x26], #0x2\n"
-      "ldr h28, [x25], #0x2\n"
-      "ldr h27, [x24], #0x2\n"
-      "ldr h23, [x23], #0x2\n"
-      "ldr h21, [x22], #0x2\n"
-      "ldr h26, [x21], #0x2\n"
-      "ldr h25, [x20], #0x2\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"
       "tbz %x[width], #0, 7f\n"
-      "ld1 { v30.b }[2], [x27]\n"
-      "ld1 { v29.b }[2], [x26]\n"
-      "ld1 { v28.b }[2], [x25]\n"
-      "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v23.b }[2], [x23]\n"
-      "ld1 { v21.b }[2], [x22]\n"
-      "ld1 { v26.b }[2], [x21]\n"
-      "ld1 { v25.b }[2], [x20]\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"
       "b 7f\n"
       "6:"  // odd_loads_1_0
-      "ldr b30, [x27, #0x0]\n"
-      "ldr b29, [x26, #0x0]\n"
-      "ldr b28, [x25, #0x0]\n"
-      "ldr b27, [x24, #0x0]\n"
-      "ldr b23, [x23, #0x0]\n"
-      "ldr b21, [x22, #0x0]\n"
-      "ldr b26, [x21, #0x0]\n"
-      "ldr b25, [x20, #0x0]\n"
+      "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"
       "7:"  // Odd load end
+      "ushll v31.8h, v31.8b, #0x0\n"
+      "subs x19, x19, #0x1\n"
       "ushll v30.8h, v30.8b, #0x0\n"
       "ushll v29.8h, v29.8b, #0x0\n"
       "ushll v28.8h, v28.8b, #0x0\n"
-      "ushll v27.8h, v27.8b, #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 v26.8h, v26.8b, #0x0\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
-      "ushll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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"
+      "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"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 8f\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v16.8h, v21.8h, 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 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v25.8h, v31.8h, v25.8h\n"
+      "zip2 v19.8h, v29.8h, v27.8h\n"
       "subs x19, x19, #0x1\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 8f\n"
+      "zip1 v20.8h, v25.8h, v19.8h\n"
       "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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"
-      "subs x19, x19, #0x1\n"
+      "zip2 v16.8h, v20.8h, 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, v23.8h, v20.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "8:"  // Odds skip
 
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 cefb70c..7c7d774 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -96,72 +96,72 @@
       "movi v1.8h, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr d30, [x27], #0x8\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr d29, [x26], #0x8\n"
-      "ldr d28, [x25], #0x8\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr d27, [x24], #0x8\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr d23, [x23], #0x8\n"
-      "ldr d21, [x22], #0x8\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr d26, [x21], #0x8\n"
-      "ldr d25, [x20], #0x8\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
       "ushll v30.8h, v30.8b, #0x0\n"
-      "ushll v29.8h, v29.8b, #0x0\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "ushll v28.8h, v28.8b, #0x0\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "ushll v27.8h, v27.8b, #0x0\n"
-      "ushll v23.8h, v23.8b, #0x0\n"
-      "zip1 v24.8h, v30.8h, v23.8h\n"
-      "ushll v21.8h, v21.8b, #0x0\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "ushll v26.8h, v26.8b, #0x0\n"
-      "ushll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
+      "ldr d29, [x26], #0x8\n"
       "add x19, x19, #0x1\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
+      "ushll v29.8h, v29.8b, #0x0\n"
+      "ldr d28, [x25], #0x8\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v20.8h, v28.8h, v26.8h\n"
+      "ushll v28.8h, v28.8b, #0x0\n"
+      "ldr d27, [x24], #0x8\n"
       "cmp %x[width], #0x8\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x0]\n"
       "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v18.8h, v24.8h, v20.8h\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"
-      "str q16, [%x[out_ptr], #0x20]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x30]\n"
-      "zip2 v20.8h, v28.8h, v26.8h\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 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
+      "zip1 v16.8h, v21.8h, v17.8h\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, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v18.8h, v23.8h, v20.8h\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x60]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "zip2 v16.8h, v18.8h, v17.8h\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"
       "str q16, [%x[out_ptr], #0x70]\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
@@ -171,148 +171,148 @@
       "ldr s29, [x26], #0x4\n"
       "ldr s28, [x25], #0x4\n"
       "ldr s27, [x24], #0x4\n"
-      "ldr s23, [x23], #0x4\n"
-      "ldr s21, [x22], #0x4\n"
-      "ldr s26, [x21], #0x4\n"
-      "ldr s25, [x20], #0x4\n"
+      "ldr s24, [x23], #0x4\n"
+      "ldr s23, [x22], #0x4\n"
+      "ldr s21, [x21], #0x4\n"
+      "ldr s26, [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
       "ld1 { v30.h }[2], [x27], #0x2\n"
+      "mov x19, #0x6\n"
       "ld1 { v29.h }[2], [x26], #0x2\n"
       "ld1 { v28.h }[2], [x25], #0x2\n"
       "ld1 { v27.h }[2], [x24], #0x2\n"
-      "ld1 { v23.h }[2], [x23], #0x2\n"
-      "ld1 { v21.h }[2], [x22], #0x2\n"
-      "ld1 { v26.h }[2], [x21], #0x2\n"
-      "ld1 { v25.h }[2], [x20], #0x2\n"
-      "mov x19, #0x6\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"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[6], [x27]\n"
+      "mov x19, #0x7\n"
       "ld1 { v29.b }[6], [x26]\n"
       "ld1 { v28.b }[6], [x25]\n"
       "ld1 { v27.b }[6], [x24]\n"
-      "ld1 { v23.b }[6], [x23]\n"
-      "ld1 { v21.b }[6], [x22]\n"
-      "ld1 { v26.b }[6], [x21]\n"
-      "ld1 { v25.b }[6], [x20]\n"
-      "mov x19, #0x7\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"
       "b 9f\n"
       "6:"  // odd_loads_1_4
       "mov x19, #0x4\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[4], [x27]\n"
       "ld1 { v29.b }[4], [x26]\n"
+      "mov x19, #0x5\n"
       "ld1 { v28.b }[4], [x25]\n"
       "ld1 { v27.b }[4], [x24]\n"
-      "ld1 { v23.b }[4], [x23]\n"
-      "ld1 { v21.b }[4], [x22]\n"
-      "ld1 { v26.b }[4], [x21]\n"
-      "ld1 { v25.b }[4], [x20]\n"
-      "mov x19, #0x5\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"
       "b 9f\n"
       "7:"  // odd_loads_2_0
       "tbz %x[width], #1, 8f\n"
       "ldr h30, [x27], #0x2\n"
       "ldr h29, [x26], #0x2\n"
+      "mov x19, #0x2\n"
       "ldr h28, [x25], #0x2\n"
       "ldr h27, [x24], #0x2\n"
-      "ldr h23, [x23], #0x2\n"
-      "ldr h21, [x22], #0x2\n"
-      "ldr h26, [x21], #0x2\n"
-      "ldr h25, [x20], #0x2\n"
-      "mov x19, #0x2\n"
+      "ldr h24, [x23], #0x2\n"
+      "ldr h23, [x22], #0x2\n"
+      "ldr h21, [x21], #0x2\n"
+      "ldr h26, [x20], #0x2\n"
       "tbz %x[width], #0, 9f\n"
       "ld1 { v30.b }[2], [x27]\n"
+      "mov x19, #0x3\n"
       "ld1 { v29.b }[2], [x26]\n"
       "ld1 { v28.b }[2], [x25]\n"
       "ld1 { v27.b }[2], [x24]\n"
-      "ld1 { v23.b }[2], [x23]\n"
-      "ld1 { v21.b }[2], [x22]\n"
-      "ld1 { v26.b }[2], [x21]\n"
-      "ld1 { v25.b }[2], [x20]\n"
-      "mov x19, #0x3\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"
       "b 9f\n"
       "8:"  // odd_loads_1_0
       "ldr b30, [x27, #0x0]\n"
+      "mov x19, #0x1\n"
       "ldr b29, [x26, #0x0]\n"
       "ldr b28, [x25, #0x0]\n"
       "ldr b27, [x24, #0x0]\n"
-      "ldr b23, [x23, #0x0]\n"
-      "ldr b21, [x22, #0x0]\n"
-      "ldr b26, [x21, #0x0]\n"
-      "ldr b25, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
+      "ldr b24, [x23, #0x0]\n"
+      "ldr b23, [x22, #0x0]\n"
+      "ldr b21, [x21, #0x0]\n"
+      "ldr b26, [x20, #0x0]\n"
       "9:"  // Odd load end
       "ushll v30.8h, v30.8b, #0x0\n"
+      "subs x19, x19, #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 v24.8h, v30.8h, v23.8h\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 v20.8h, v28.8h, v26.8h\n"
-      "ushll v25.8h, v25.8b, #0x0\n"
-      "zip1 v22.8h, v29.8h, v21.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v24.8h, v20.8h\n"
-      "zip1 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v22.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\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"
       "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 %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"
+      "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"
+      "str q16, [%x[out_ptr], #0x0]\n"
+      "subs x19, x19, #0x1\n"
       "add v1.8h, v1.8h, v16.8h\n"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
+      "zip2 v24.8h, v30.8h, v24.8h\n"
+      "zip2 v20.8h, v28.8h, v21.8h\n"
       "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\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"
       "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 %x[out_ptr], %x[out_ptr], #0x10\n"
       "beq 10f\n"
-      "zip2 v18.8h, v24.8h, v20.8h\n"
-      "zip2 v17.8h, v22.8h, v19.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v23.8h, v30.8h, v23.8h\n"
-      "zip2 v20.8h, v28.8h, v26.8h\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v18.8h, v23.8h, v20.8h\n"
-      "zip2 v21.8h, v29.8h, v21.8h\n"
-      "zip2 v19.8h, v27.8h, v25.8h\n"
-      "zip1 v17.8h, v21.8h, v19.8h\n"
-      "zip1 v16.8h, v18.8h, v17.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v16.8h, v18.8h, v17.8h\n"
-      "subs x19, x19, #0x1\n"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
-      "beq 10f\n"
-      "zip2 v18.8h, v23.8h, 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"
-      "add v1.8h, v1.8h, v16.8h\n"
-      "add %x[out_ptr], %x[out_ptr], #0x10\n"
+      "add v1.8h, v1.8h, v17.8h\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 5377edc..1e5d395 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,45 +80,45 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q28, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q27, [x26], #0x10\n"
-      "ldr q26, [x25], #0x10\n"
-      "zip1 v23.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q22, [x24], #0x10\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q24, [x22], #0x10\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
-      "ldr q21, [x20], #0x10\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v16.4s, v24.4s, v21.4s\n"
       "subs %x[width], %x[width], #0x8\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x8\n"
-      "zip2 v16.4s, v18.4s, v16.4s\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"
+      "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"
+      "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"
+      "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 v19.4s, v25.4s, v19.4s\n"
-      "str q23, [%x[out_ptr], #0x20]\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q20, [%x[out_ptr], #0x40]\n"
-      "zip2 v17.4s, v26.4s, v22.4s\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v16.4s, v19.4s, v18.4s\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"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -127,119 +127,119 @@
       "cbz %x[width], 8f\n"
       "tbz %x[width], #2, 5f\n"
       "ldr d28, [x27], #0x8\n"
-      "ldr d27, [x26], #0x8\n"
-      "ldr d26, [x25], #0x8\n"
-      "ldr d22, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d24, [x22], #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 d21, [x20], #0x8\n"
+      "ldr d24, [x20], #0x8\n"
       "tbz %x[width], #1, 4f\n"
       "ld1 { v28.s }[2], [x27], #0x4\n"
-      "ld1 { v27.s }[2], [x26], #0x4\n"
-      "ld1 { v26.s }[2], [x25], #0x4\n"
-      "ld1 { v22.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v24.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v21.s }[2], [x20], #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"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v28.h }[6], [x27]\n"
-      "ld1 { v27.h }[6], [x26]\n"
-      "ld1 { v26.h }[6], [x25]\n"
-      "ld1 { v22.h }[6], [x24]\n"
-      "ld1 { v25.h }[6], [x23]\n"
-      "ld1 { v24.h }[6], [x22]\n"
-      "ld1 { v19.h }[6], [x21]\n"
-      "ld1 { v21.h }[6], [x20]\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"
       "b 7f\n"
       "4:"  // odd_loads_1_4
       "mov x19, #0x2\n"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v28.h }[4], [x27]\n"
-      "ld1 { v27.h }[4], [x26]\n"
-      "ld1 { v26.h }[4], [x25]\n"
-      "ld1 { v22.h }[4], [x24]\n"
-      "ld1 { v25.h }[4], [x23]\n"
-      "ld1 { v24.h }[4], [x22]\n"
-      "ld1 { v19.h }[4], [x21]\n"
-      "ld1 { v21.h }[4], [x20]\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"
       "b 7f\n"
       "5:"  // odd_loads_2_0
       "tbz %x[width], #1, 6f\n"
       "ldr s28, [x27], #0x4\n"
-      "ldr s27, [x26], #0x4\n"
-      "ldr s26, [x25], #0x4\n"
-      "ldr s22, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s24, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
-      "ldr s21, [x20], #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"
       "tbz %x[width], #0, 7f\n"
       "ld1 { v28.h }[2], [x27]\n"
-      "ld1 { v27.h }[2], [x26]\n"
-      "ld1 { v26.h }[2], [x25]\n"
-      "ld1 { v22.h }[2], [x24]\n"
-      "ld1 { v25.h }[2], [x23]\n"
-      "ld1 { v24.h }[2], [x22]\n"
-      "ld1 { v19.h }[2], [x21]\n"
-      "ld1 { v21.h }[2], [x20]\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"
       "b 7f\n"
       "6:"  // odd_loads_1_0
       "ldr h28, [x27, #0x0]\n"
-      "ldr h27, [x26, #0x0]\n"
-      "ldr h26, [x25, #0x0]\n"
-      "ldr h22, [x24, #0x0]\n"
-      "ldr h25, [x23, #0x0]\n"
-      "ldr h24, [x22, #0x0]\n"
-      "ldr h19, [x21, #0x0]\n"
-      "ldr h21, [x20, #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"
       "7:"  // Odd load end
-      "zip1 v23.4s, v28.4s, v26.4s\n"
+      "zip1 v22.4s, v28.4s, v25.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "zip1 v16.4s, v24.4s, v21.4s\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "zip2 v16.4s, v18.4s, v16.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
+      "zip2 v28.4s, v28.4s, v25.4s\n"
+      "zip2 v25.4s, v29.4s, v21.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "str q20, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v25.4s, v19.4s\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 8f\n"
-      "zip2 v17.4s, v26.4s, v22.4s\n"
-      "zip2 v16.4s, v19.4s, v18.4s\n"
+      "zip2 v17.4s, v28.4s, v25.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v19.4s, v16.4s\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)
+      : [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", "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", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 3aea6a8..064207c 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,39 +80,39 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q27, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "subs %x[width], %x[width], #0x4\n"
       "ldr q24, [x26], #0x10\n"
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "ldr q25, [x25], #0x10\n"
+      "cmp %x[width], #0x4\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
       "ldr q21, [x24], #0x10\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q22, [x23], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip1 v22.2d, v25.2d, v21.2d\n"
       "ldr q18, [x22], #0x10\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v21.2d, v25.2d, v21.2d\n"
+      "ldr q20, [x21], #0x10\n"
       "ldr q16, [x20], #0x10\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
+      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v18.2d, v23.2d, v18.2d\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"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "str q20, [%x[out_ptr], #0x20]\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 q16, [%x[out_ptr], #0x70]\n"
-      "subs %x[width], %x[width], #0x4\n"
-      "cmp %x[width], #0x4\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
@@ -120,58 +120,58 @@
       "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 d22, [x23], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
       "ldr d18, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
+      "ldr d20, [x21], #0x8\n"
       "ldr d16, [x20], #0x8\n"
-      "mov x19, #0x1\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 { v22.s }[2], [x23]\n"
+      "ld1 { v23.s }[2], [x23]\n"
       "ld1 { v18.s }[2], [x22]\n"
-      "ld1 { v19.s }[2], [x21]\n"
+      "ld1 { v20.s }[2], [x21]\n"
       "ld1 { v16.s }[2], [x20]\n"
-      "mov x19, #0x2\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 s22, [x23, #0x0]\n"
+      "ldr s23, [x23, #0x0]\n"
       "ldr s18, [x22, #0x0]\n"
-      "ldr s19, [x21, #0x0]\n"
+      "ldr s20, [x21, #0x0]\n"
       "ldr s16, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
       "5:"  // Odd load end
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
-      "str q20, [%x[out_ptr], #0x20]\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"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 6f\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
       "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.2d, v22.2d, v18.2d\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, v19.2d, v16.2d\n"
+      "zip2 v16.2d, v20.2d, v16.2d\n"
       "str q18, [%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)
+      : [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"
     );
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 4780b77..1f86722 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,39 +80,39 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q27, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "subs %x[width], %x[width], #0x8\n"
       "ldr q24, [x26], #0x10\n"
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "ldr q25, [x25], #0x10\n"
+      "cmp %x[width], #0x8\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
       "ldr q21, [x24], #0x10\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q22, [x23], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip1 v22.2d, v25.2d, v21.2d\n"
       "ldr q18, [x22], #0x10\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v21.2d, v25.2d, v21.2d\n"
+      "ldr q20, [x21], #0x10\n"
       "ldr q16, [x20], #0x10\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
+      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v18.2d, v23.2d, v18.2d\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"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "str q20, [%x[out_ptr], #0x20]\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 q16, [%x[out_ptr], #0x70]\n"
-      "subs %x[width], %x[width], #0x8\n"
-      "cmp %x[width], #0x8\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
@@ -122,28 +122,28 @@
       "ldr d24, [x26], #0x8\n"
       "ldr d25, [x25], #0x8\n"
       "ldr d21, [x24], #0x8\n"
-      "ldr d22, [x23], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
       "ldr d18, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
+      "ldr d20, [x21], #0x8\n"
       "ldr d16, [x20], #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 { v22.s }[2], [x23], #0x4\n"
+      "ld1 { v23.s }[2], [x23], #0x4\n"
       "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
+      "ld1 { v20.s }[2], [x21], #0x4\n"
       "ld1 { v16.s }[2], [x20], #0x4\n"
-      "mov x19, #0x2\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 { v22.h }[6], [x23]\n"
+      "ld1 { v23.h }[6], [x23]\n"
       "ld1 { v18.h }[6], [x22]\n"
-      "ld1 { v19.h }[6], [x21]\n"
+      "ld1 { v20.h }[6], [x21]\n"
       "ld1 { v16.h }[6], [x20]\n"
       "b 7f\n"
       "4:"  // odd_loads_1_4
@@ -151,69 +151,69 @@
       "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 { v22.h }[4], [x23]\n"
+      "ld1 { v23.h }[4], [x23]\n"
       "ld1 { v18.h }[4], [x22]\n"
-      "ld1 { v19.h }[4], [x21]\n"
+      "ld1 { v20.h }[4], [x21]\n"
       "ld1 { v16.h }[4], [x20]\n"
-      "mov x19, #0x2\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 s22, [x23], #0x4\n"
+      "ldr s23, [x23], #0x4\n"
       "ldr s18, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
+      "ldr s20, [x21], #0x4\n"
       "ldr s16, [x20], #0x4\n"
-      "mov x19, #0x1\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 { v22.h }[2], [x23]\n"
+      "ld1 { v23.h }[2], [x23]\n"
       "ld1 { v18.h }[2], [x22]\n"
-      "ld1 { v19.h }[2], [x21]\n"
+      "ld1 { v20.h }[2], [x21]\n"
       "ld1 { v16.h }[2], [x20]\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 h22, [x23, #0x0]\n"
+      "ldr h23, [x23, #0x0]\n"
       "ldr h18, [x22, #0x0]\n"
-      "ldr h19, [x21, #0x0]\n"
+      "ldr h20, [x21, #0x0]\n"
       "ldr h16, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
       "7:"  // Odd load end
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
-      "str q20, [%x[out_ptr], #0x20]\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"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 8f\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
       "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.2d, v22.2d, v18.2d\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, v19.2d, v16.2d\n"
+      "zip2 v16.2d, v20.2d, v16.2d\n"
       "str q18, [%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)
+      : [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"
     );
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 a9034f5..659d994 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,45 +80,45 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q28, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q27, [x26], #0x10\n"
-      "ldr q26, [x25], #0x10\n"
-      "zip1 v23.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q22, [x24], #0x10\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q25, [x23], #0x10\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q24, [x22], #0x10\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
-      "ldr q21, [x20], #0x10\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v16.4s, v24.4s, v21.4s\n"
       "subs %x[width], %x[width], #0x10\n"
-      "zip1 v17.4s, v18.4s, v16.4s\n"
+      "ldr q29, [x26], #0x10\n"
       "cmp %x[width], #0x10\n"
-      "zip2 v16.4s, v18.4s, v16.4s\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"
+      "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"
+      "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"
+      "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 v19.4s, v25.4s, v19.4s\n"
-      "str q23, [%x[out_ptr], #0x20]\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "str q16, [%x[out_ptr], #0x30]\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q20, [%x[out_ptr], #0x40]\n"
-      "zip2 v17.4s, v26.4s, v22.4s\n"
-      "str q16, [%x[out_ptr], #0x50]\n"
-      "zip2 v16.4s, v19.4s, v18.4s\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"
       "str q17, [%x[out_ptr], #0x60]\n"
       "str q16, [%x[out_ptr], #0x70]\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
@@ -127,203 +127,203 @@
       "cbz %x[width], 12f\n"
       "tbz %x[width], #3, 7f\n"
       "ldr d28, [x27], #0x8\n"
-      "ldr d27, [x26], #0x8\n"
-      "ldr d26, [x25], #0x8\n"
-      "ldr d22, [x24], #0x8\n"
-      "ldr d25, [x23], #0x8\n"
-      "ldr d24, [x22], #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 d21, [x20], #0x8\n"
+      "ldr d24, [x20], #0x8\n"
       "tbz %x[width], #2, 5f\n"
       "ld1 { v28.s }[2], [x27], #0x4\n"
-      "ld1 { v27.s }[2], [x26], #0x4\n"
-      "ld1 { v26.s }[2], [x25], #0x4\n"
-      "ld1 { v22.s }[2], [x24], #0x4\n"
-      "ld1 { v25.s }[2], [x23], #0x4\n"
-      "ld1 { v24.s }[2], [x22], #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 { v21.s }[2], [x20], #0x4\n"
+      "ld1 { v24.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 4f\n"
       "ld1 { v28.h }[6], [x27], #0x2\n"
-      "ld1 { v27.h }[6], [x26], #0x2\n"
-      "ld1 { v26.h }[6], [x25], #0x2\n"
-      "ld1 { v22.h }[6], [x24], #0x2\n"
-      "ld1 { v25.h }[6], [x23], #0x2\n"
-      "ld1 { v24.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v21.h }[6], [x20], #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"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[14], [x27]\n"
-      "ld1 { v27.b }[14], [x26]\n"
-      "ld1 { v26.b }[14], [x25]\n"
-      "ld1 { v22.b }[14], [x24]\n"
-      "ld1 { v25.b }[14], [x23]\n"
-      "ld1 { v24.b }[14], [x22]\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 { v21.b }[14], [x20]\n"
+      "ld1 { v24.b }[14], [x20]\n"
       "b 11f\n"
       "4:"  // odd_loads_1_12
       "mov x19, #0x3\n"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[12], [x27]\n"
-      "ld1 { v27.b }[12], [x26]\n"
-      "ld1 { v26.b }[12], [x25]\n"
-      "ld1 { v22.b }[12], [x24]\n"
-      "ld1 { v25.b }[12], [x23]\n"
-      "ld1 { v24.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v21.b }[12], [x20]\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"
       "b 11f\n"
       "5:"  // odd_loads_2_8
       "tbz %x[width], #1, 6f\n"
       "ld1 { v28.h }[4], [x27], #0x2\n"
-      "ld1 { v27.h }[4], [x26], #0x2\n"
-      "ld1 { v26.h }[4], [x25], #0x2\n"
-      "ld1 { v22.h }[4], [x24], #0x2\n"
-      "ld1 { v25.h }[4], [x23], #0x2\n"
-      "ld1 { v24.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v21.h }[4], [x20], #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"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[10], [x27]\n"
-      "ld1 { v27.b }[10], [x26]\n"
-      "ld1 { v26.b }[10], [x25]\n"
-      "ld1 { v22.b }[10], [x24]\n"
-      "ld1 { v25.b }[10], [x23]\n"
-      "ld1 { v24.b }[10], [x22]\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 { v21.b }[10], [x20]\n"
+      "ld1 { v24.b }[10], [x20]\n"
       "b 11f\n"
       "6:"  // odd_loads_1_8
       "mov x19, #0x2\n"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[8], [x27]\n"
-      "ld1 { v27.b }[8], [x26]\n"
-      "ld1 { v26.b }[8], [x25]\n"
-      "ld1 { v22.b }[8], [x24]\n"
-      "ld1 { v25.b }[8], [x23]\n"
-      "ld1 { v24.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v21.b }[8], [x20]\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"
       "b 11f\n"
       "7:"  // odd_loads_4_0
       "tbz %x[width], #2, 9f\n"
       "ldr s28, [x27], #0x4\n"
-      "ldr s27, [x26], #0x4\n"
-      "ldr s26, [x25], #0x4\n"
-      "ldr s22, [x24], #0x4\n"
-      "ldr s25, [x23], #0x4\n"
-      "ldr s24, [x22], #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 s21, [x20], #0x4\n"
+      "ldr s24, [x20], #0x4\n"
       "tbz %x[width], #1, 8f\n"
       "ld1 { v28.h }[2], [x27], #0x2\n"
-      "ld1 { v27.h }[2], [x26], #0x2\n"
-      "ld1 { v26.h }[2], [x25], #0x2\n"
-      "ld1 { v22.h }[2], [x24], #0x2\n"
-      "ld1 { v25.h }[2], [x23], #0x2\n"
-      "ld1 { v24.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v21.h }[2], [x20], #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"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[6], [x27]\n"
-      "ld1 { v27.b }[6], [x26]\n"
-      "ld1 { v26.b }[6], [x25]\n"
-      "ld1 { v22.b }[6], [x24]\n"
-      "ld1 { v25.b }[6], [x23]\n"
-      "ld1 { v24.b }[6], [x22]\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 { v21.b }[6], [x20]\n"
+      "ld1 { v24.b }[6], [x20]\n"
       "b 11f\n"
       "8:"  // odd_loads_1_4
       "mov x19, #0x1\n"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[4], [x27]\n"
-      "ld1 { v27.b }[4], [x26]\n"
-      "ld1 { v26.b }[4], [x25]\n"
-      "ld1 { v22.b }[4], [x24]\n"
-      "ld1 { v25.b }[4], [x23]\n"
-      "ld1 { v24.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v21.b }[4], [x20]\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"
       "b 11f\n"
       "9:"  // odd_loads_2_0
       "tbz %x[width], #1, 10f\n"
       "ldr h28, [x27], #0x2\n"
-      "ldr h27, [x26], #0x2\n"
-      "ldr h26, [x25], #0x2\n"
-      "ldr h22, [x24], #0x2\n"
-      "ldr h25, [x23], #0x2\n"
-      "ldr h24, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h21, [x20], #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"
       "tbz %x[width], #0, 11f\n"
       "ld1 { v28.b }[2], [x27]\n"
-      "ld1 { v27.b }[2], [x26]\n"
-      "ld1 { v26.b }[2], [x25]\n"
-      "ld1 { v22.b }[2], [x24]\n"
-      "ld1 { v25.b }[2], [x23]\n"
-      "ld1 { v24.b }[2], [x22]\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 { v21.b }[2], [x20]\n"
+      "ld1 { v24.b }[2], [x20]\n"
       "b 11f\n"
       "10:"  // odd_loads_1_0
       "ldr b28, [x27, #0x0]\n"
-      "ldr b27, [x26, #0x0]\n"
-      "ldr b26, [x25, #0x0]\n"
-      "ldr b22, [x24, #0x0]\n"
-      "ldr b25, [x23, #0x0]\n"
-      "ldr b24, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b21, [x20, #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"
       "11:"  // Odd load end
-      "zip1 v23.4s, v28.4s, v26.4s\n"
+      "zip1 v22.4s, v28.4s, v25.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v27.4s, v22.4s\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v25.4s, v19.4s\n"
-      "zip1 v16.4s, v24.4s, v21.4s\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v23.4s, v23.4s, v20.4s\n"
-      "zip2 v16.4s, v18.4s, v16.4s\n"
-      "str q23, [%x[out_ptr], #0x0]\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v26.4s, v28.4s, v26.4s\n"
-      "zip2 v22.4s, v27.4s, v22.4s\n"
+      "zip2 v28.4s, v28.4s, v25.4s\n"
+      "zip2 v25.4s, v29.4s, v21.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v26.4s, v22.4s\n"
-      "str q20, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v25.4s, v19.4s\n"
-      "zip2 v18.4s, v24.4s, v21.4s\n"
-      "zip1 v16.4s, v19.4s, v18.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\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"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 12f\n"
-      "zip2 v17.4s, v26.4s, v22.4s\n"
-      "zip2 v16.4s, v19.4s, v18.4s\n"
+      "zip2 v17.4s, v28.4s, v25.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v19.4s, v16.4s\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)
+      : [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", "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", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
     );
 }
 
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 2831cb7..dfec94c 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -97,270 +97,270 @@
       "movi v0.8h, #0x0\n"
       "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q29, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q28, [x26], #0x10\n"
-      "ldr q27, [x25], #0x10\n"
-      "zip1 v23.4s, v29.4s, v27.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q21, [x24], #0x10\n"
-      "zip2 v27.4s, v29.4s, v27.4s\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q26, [x23], #0x10\n"
-      "zip1 v20.4s, v28.4s, v21.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v24.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v23.4s, v28.4s, v21.4s\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.4s, v26.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v21.4s, v27.4s, v23.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v17.4s, v25.4s, v22.4s\n"
-      "sadalp v1.8h, v16.16b\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
+      "ldr q28, [x27], #0x10\n"
       "add x19, x19, #0x1\n"
-      "zip2 v20.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v19.4s, v26.4s, v19.4s\n"
-      "sadalp v0.8h, v16.16b\n"
-      "zip2 v16.4s, v25.4s, v22.4s\n"
-      "str q24, [%x[out_ptr], #0x20]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "sadalp v1.8h, v24.16b\n"
-      "zip2 v17.4s, v27.4s, v23.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
-      "sadalp v0.8h, v20.16b\n"
-      "str q17, [%x[out_ptr], #0x60]\n"
-      "sadalp v1.8h, v21.16b\n"
-      "str q16, [%x[out_ptr], #0x70]\n"
+      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "sadalp v0.8h, v18.16b\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "sadalp v1.8h, v23.16b\n"
+      "zip2 v22.4s, v22.4s, v20.4s\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"
+      "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"
+      "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"
-      "add %x[out_ptr], %x[out_ptr], #0x80\n"
+      "str q21, [%x[out_ptr], #0x40]\n"
       "sadalp v0.8h, v16.16b\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
+      "str q16, [%x[out_ptr], #0x70]\n"
+      "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 d29, [x27], #0x8\n"
-      "ldr d28, [x26], #0x8\n"
-      "ldr d27, [x25], #0x8\n"
+      "ldr d28, [x27], #0x8\n"
+      "ldr d29, [x26], #0x8\n"
+      "ldr d25, [x25], #0x8\n"
       "ldr d21, [x24], #0x8\n"
-      "ldr d26, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "ldr d26, [x22], #0x8\n"
       "ldr d19, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d24, [x20], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v29.s }[2], [x27], #0x4\n"
-      "ld1 { v28.s }[2], [x26], #0x4\n"
-      "ld1 { v27.s }[2], [x25], #0x4\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 { v26.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
+      "ld1 { v27.s }[2], [x23], #0x4\n"
+      "ld1 { v26.s }[2], [x22], #0x4\n"
       "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
+      "ld1 { v24.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v29.h }[6], [x27], #0x2\n"
-      "ld1 { v28.h }[6], [x26], #0x2\n"
-      "ld1 { v27.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v26.h }[6], [x23], #0x2\n"
-      "ld1 { v25.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v22.h }[6], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[14], [x27]\n"
-      "ld1 { v28.b }[14], [x26]\n"
-      "ld1 { v27.b }[14], [x25]\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 { v26.b }[14], [x23]\n"
-      "ld1 { v25.b }[14], [x22]\n"
+      "ld1 { v27.b }[14], [x23]\n"
+      "ld1 { v26.b }[14], [x22]\n"
       "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v22.b }[14], [x20]\n"
+      "ld1 { v24.b }[14], [x20]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
       "mov x19, #0x3\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[12], [x27]\n"
-      "ld1 { v28.b }[12], [x26]\n"
-      "ld1 { v27.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v26.b }[12], [x23]\n"
-      "ld1 { v25.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v22.b }[12], [x20]\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"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v29.h }[4], [x27], #0x2\n"
-      "ld1 { v28.h }[4], [x26], #0x2\n"
-      "ld1 { v27.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v26.h }[4], [x23], #0x2\n"
-      "ld1 { v25.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v22.h }[4], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[10], [x27]\n"
-      "ld1 { v28.b }[10], [x26]\n"
-      "ld1 { v27.b }[10], [x25]\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 { v26.b }[10], [x23]\n"
-      "ld1 { v25.b }[10], [x22]\n"
+      "ld1 { v27.b }[10], [x23]\n"
+      "ld1 { v26.b }[10], [x22]\n"
       "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v22.b }[10], [x20]\n"
+      "ld1 { v24.b }[10], [x20]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
       "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[8], [x27]\n"
-      "ld1 { v28.b }[8], [x26]\n"
-      "ld1 { v27.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v26.b }[8], [x23]\n"
-      "ld1 { v25.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v22.b }[8], [x20]\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"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s29, [x27], #0x4\n"
-      "ldr s28, [x26], #0x4\n"
-      "ldr s27, [x25], #0x4\n"
+      "ldr s28, [x27], #0x4\n"
+      "ldr s29, [x26], #0x4\n"
+      "ldr s25, [x25], #0x4\n"
       "ldr s21, [x24], #0x4\n"
-      "ldr s26, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s26, [x22], #0x4\n"
       "ldr s19, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
+      "ldr s24, [x20], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v29.h }[2], [x27], #0x2\n"
-      "ld1 { v28.h }[2], [x26], #0x2\n"
-      "ld1 { v27.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v26.h }[2], [x23], #0x2\n"
-      "ld1 { v25.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v22.h }[2], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[6], [x27]\n"
-      "ld1 { v28.b }[6], [x26]\n"
-      "ld1 { v27.b }[6], [x25]\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 { v26.b }[6], [x23]\n"
-      "ld1 { v25.b }[6], [x22]\n"
+      "ld1 { v27.b }[6], [x23]\n"
+      "ld1 { v26.b }[6], [x22]\n"
       "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v22.b }[6], [x20]\n"
+      "ld1 { v24.b }[6], [x20]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
       "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[4], [x27]\n"
-      "ld1 { v28.b }[4], [x26]\n"
-      "ld1 { v27.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v26.b }[4], [x23]\n"
-      "ld1 { v25.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v22.b }[4], [x20]\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"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h29, [x27], #0x2\n"
-      "ldr h28, [x26], #0x2\n"
-      "ldr h27, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h26, [x23], #0x2\n"
-      "ldr h25, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h22, [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[2], [x27]\n"
-      "ld1 { v28.b }[2], [x26]\n"
-      "ld1 { v27.b }[2], [x25]\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 { v26.b }[2], [x23]\n"
-      "ld1 { v25.b }[2], [x22]\n"
+      "ld1 { v27.b }[2], [x23]\n"
+      "ld1 { v26.b }[2], [x22]\n"
       "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v22.b }[2], [x20]\n"
+      "ld1 { v24.b }[2], [x20]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b29, [x27, #0x0]\n"
-      "ldr b28, [x26, #0x0]\n"
-      "ldr b27, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b26, [x23, #0x0]\n"
-      "ldr b25, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b22, [x20, #0x0]\n"
+      "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"
       "13:"  // Odd load end
-      "zip1 v23.4s, v29.4s, v27.4s\n"
+      "zip1 v22.4s, v28.4s, v25.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v28.4s, v21.4s\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v26.4s, v19.4s\n"
-      "sadalp v1.8h, v16.16b\n"
-      "zip1 v17.4s, v25.4s, v22.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v16.16b\n"
+      "zip1 v20.4s, v29.4s, v21.4s\n"
+      "zip1 v23.4s, v22.4s, v20.4s\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 v24.4s, v23.4s, v20.4s\n"
-      "zip2 v20.4s, v18.4s, v17.4s\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "sadalp v1.8h, v24.16b\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 v1.8h, v22.16b\n"
       "str q20, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v20.16b\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 v27.4s, v29.4s, v27.4s\n"
-      "zip2 v23.4s, v28.4s, v21.4s\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, v27.4s, v23.4s\n"
+      "zip1 v21.4s, v28.4s, v25.4s\n"
       "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v26.4s, v19.4s\n"
       "sadalp v1.8h, v21.16b\n"
-      "zip2 v16.4s, v25.4s, v22.4s\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"
       "sadalp v0.8h, v18.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 14f\n"
-      "zip2 v17.4s, v27.4s, v23.4s\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
+      "zip2 v17.4s, v28.4s, v25.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v19.4s, v16.4s\n"
       "sadalp v1.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "sadalp v0.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "sadalp v0.8h, v16.16b\n"
       "14:"  // Odds skip
       "sadalp v31.4s, v1.8h\n"
-      "sadalp v30.4s, v0.8h\n"
       "str q31, [%x[out_ptr], #0x0]\n"
+      "sadalp v30.4s, v0.8h\n"
       "str q30, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 7c7857b..1b94c7f 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -97,270 +97,270 @@
       "movi v0.8h, #0x0\n"
       "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
-      "ldr q29, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
-      "ldr q28, [x26], #0x10\n"
-      "ldr q27, [x25], #0x10\n"
-      "zip1 v23.4s, v29.4s, v27.4s\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
-      "ldr q21, [x24], #0x10\n"
-      "zip2 v27.4s, v29.4s, v27.4s\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
-      "ldr q26, [x23], #0x10\n"
-      "zip1 v20.4s, v28.4s, v21.4s\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q25, [x22], #0x10\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v24.4s, v23.4s, v20.4s\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "zip2 v23.4s, v28.4s, v21.4s\n"
-      "ldr q22, [x20], #0x10\n"
-      "zip1 v18.4s, v26.4s, v19.4s\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v21.4s, v27.4s, v23.4s\n"
-      "prfm pldl1keep, [x20, #0x70]\n"
-      "zip1 v17.4s, v25.4s, v22.4s\n"
-      "uadalp v1.8h, v16.16b\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
+      "ldr q28, [x27], #0x10\n"
       "add x19, x19, #0x1\n"
-      "zip2 v20.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "zip2 v19.4s, v26.4s, v19.4s\n"
-      "uadalp v0.8h, v16.16b\n"
-      "zip2 v16.4s, v25.4s, v22.4s\n"
-      "str q24, [%x[out_ptr], #0x20]\n"
-      "zip1 v18.4s, v19.4s, v16.4s\n"
-      "uadalp v1.8h, v24.16b\n"
-      "zip2 v17.4s, v27.4s, v23.4s\n"
-      "str q20, [%x[out_ptr], #0x30]\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
-      "str q21, [%x[out_ptr], #0x40]\n"
-      "str q18, [%x[out_ptr], #0x50]\n"
-      "uadalp v0.8h, v20.16b\n"
-      "str q17, [%x[out_ptr], #0x60]\n"
-      "uadalp v1.8h, v21.16b\n"
-      "str q16, [%x[out_ptr], #0x70]\n"
+      "ldr q29, [x26], #0x10\n"
       "subs %x[width], %x[width], #0x10\n"
-      "uadalp v0.8h, v18.16b\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"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "uadalp v1.8h, v23.16b\n"
+      "zip2 v22.4s, v22.4s, v20.4s\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"
+      "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"
+      "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"
-      "add %x[out_ptr], %x[out_ptr], #0x80\n"
+      "str q21, [%x[out_ptr], #0x40]\n"
       "uadalp v0.8h, v16.16b\n"
+      "str q18, [%x[out_ptr], #0x50]\n"
+      "str q17, [%x[out_ptr], #0x60]\n"
+      "str q16, [%x[out_ptr], #0x70]\n"
+      "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 d29, [x27], #0x8\n"
-      "ldr d28, [x26], #0x8\n"
-      "ldr d27, [x25], #0x8\n"
+      "ldr d28, [x27], #0x8\n"
+      "ldr d29, [x26], #0x8\n"
+      "ldr d25, [x25], #0x8\n"
       "ldr d21, [x24], #0x8\n"
-      "ldr d26, [x23], #0x8\n"
-      "ldr d25, [x22], #0x8\n"
+      "ldr d27, [x23], #0x8\n"
+      "ldr d26, [x22], #0x8\n"
       "ldr d19, [x21], #0x8\n"
-      "ldr d22, [x20], #0x8\n"
+      "ldr d24, [x20], #0x8\n"
       "tbz %x[width], #2, 7f\n"
-      "ld1 { v29.s }[2], [x27], #0x4\n"
-      "ld1 { v28.s }[2], [x26], #0x4\n"
-      "ld1 { v27.s }[2], [x25], #0x4\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 { v26.s }[2], [x23], #0x4\n"
-      "ld1 { v25.s }[2], [x22], #0x4\n"
+      "ld1 { v27.s }[2], [x23], #0x4\n"
+      "ld1 { v26.s }[2], [x22], #0x4\n"
       "ld1 { v19.s }[2], [x21], #0x4\n"
-      "ld1 { v22.s }[2], [x20], #0x4\n"
+      "ld1 { v24.s }[2], [x20], #0x4\n"
       "tbz %x[width], #1, 6f\n"
-      "ld1 { v29.h }[6], [x27], #0x2\n"
-      "ld1 { v28.h }[6], [x26], #0x2\n"
-      "ld1 { v27.h }[6], [x25], #0x2\n"
-      "ld1 { v21.h }[6], [x24], #0x2\n"
-      "ld1 { v26.h }[6], [x23], #0x2\n"
-      "ld1 { v25.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
-      "ld1 { v22.h }[6], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[14], [x27]\n"
-      "ld1 { v28.b }[14], [x26]\n"
-      "ld1 { v27.b }[14], [x25]\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 { v26.b }[14], [x23]\n"
-      "ld1 { v25.b }[14], [x22]\n"
+      "ld1 { v27.b }[14], [x23]\n"
+      "ld1 { v26.b }[14], [x22]\n"
       "ld1 { v19.b }[14], [x21]\n"
-      "ld1 { v22.b }[14], [x20]\n"
+      "ld1 { v24.b }[14], [x20]\n"
       "b 13f\n"
       "6:"  // odd_loads_1_12
       "mov x19, #0x3\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[12], [x27]\n"
-      "ld1 { v28.b }[12], [x26]\n"
-      "ld1 { v27.b }[12], [x25]\n"
-      "ld1 { v21.b }[12], [x24]\n"
-      "ld1 { v26.b }[12], [x23]\n"
-      "ld1 { v25.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
-      "ld1 { v22.b }[12], [x20]\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"
       "b 13f\n"
       "7:"  // odd_loads_2_8
       "tbz %x[width], #1, 8f\n"
-      "ld1 { v29.h }[4], [x27], #0x2\n"
-      "ld1 { v28.h }[4], [x26], #0x2\n"
-      "ld1 { v27.h }[4], [x25], #0x2\n"
-      "ld1 { v21.h }[4], [x24], #0x2\n"
-      "ld1 { v26.h }[4], [x23], #0x2\n"
-      "ld1 { v25.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
-      "ld1 { v22.h }[4], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[10], [x27]\n"
-      "ld1 { v28.b }[10], [x26]\n"
-      "ld1 { v27.b }[10], [x25]\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 { v26.b }[10], [x23]\n"
-      "ld1 { v25.b }[10], [x22]\n"
+      "ld1 { v27.b }[10], [x23]\n"
+      "ld1 { v26.b }[10], [x22]\n"
       "ld1 { v19.b }[10], [x21]\n"
-      "ld1 { v22.b }[10], [x20]\n"
+      "ld1 { v24.b }[10], [x20]\n"
       "b 13f\n"
       "8:"  // odd_loads_1_8
       "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[8], [x27]\n"
-      "ld1 { v28.b }[8], [x26]\n"
-      "ld1 { v27.b }[8], [x25]\n"
-      "ld1 { v21.b }[8], [x24]\n"
-      "ld1 { v26.b }[8], [x23]\n"
-      "ld1 { v25.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
-      "ld1 { v22.b }[8], [x20]\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"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
-      "ldr s29, [x27], #0x4\n"
-      "ldr s28, [x26], #0x4\n"
-      "ldr s27, [x25], #0x4\n"
+      "ldr s28, [x27], #0x4\n"
+      "ldr s29, [x26], #0x4\n"
+      "ldr s25, [x25], #0x4\n"
       "ldr s21, [x24], #0x4\n"
-      "ldr s26, [x23], #0x4\n"
-      "ldr s25, [x22], #0x4\n"
+      "ldr s27, [x23], #0x4\n"
+      "ldr s26, [x22], #0x4\n"
       "ldr s19, [x21], #0x4\n"
-      "ldr s22, [x20], #0x4\n"
+      "ldr s24, [x20], #0x4\n"
       "tbz %x[width], #1, 10f\n"
-      "ld1 { v29.h }[2], [x27], #0x2\n"
-      "ld1 { v28.h }[2], [x26], #0x2\n"
-      "ld1 { v27.h }[2], [x25], #0x2\n"
-      "ld1 { v21.h }[2], [x24], #0x2\n"
-      "ld1 { v26.h }[2], [x23], #0x2\n"
-      "ld1 { v25.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
-      "ld1 { v22.h }[2], [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[6], [x27]\n"
-      "ld1 { v28.b }[6], [x26]\n"
-      "ld1 { v27.b }[6], [x25]\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 { v26.b }[6], [x23]\n"
-      "ld1 { v25.b }[6], [x22]\n"
+      "ld1 { v27.b }[6], [x23]\n"
+      "ld1 { v26.b }[6], [x22]\n"
       "ld1 { v19.b }[6], [x21]\n"
-      "ld1 { v22.b }[6], [x20]\n"
+      "ld1 { v24.b }[6], [x20]\n"
       "b 13f\n"
       "10:"  // odd_loads_1_4
       "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[4], [x27]\n"
-      "ld1 { v28.b }[4], [x26]\n"
-      "ld1 { v27.b }[4], [x25]\n"
-      "ld1 { v21.b }[4], [x24]\n"
-      "ld1 { v26.b }[4], [x23]\n"
-      "ld1 { v25.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
-      "ld1 { v22.b }[4], [x20]\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"
       "b 13f\n"
       "11:"  // odd_loads_2_0
       "tbz %x[width], #1, 12f\n"
-      "ldr h29, [x27], #0x2\n"
-      "ldr h28, [x26], #0x2\n"
-      "ldr h27, [x25], #0x2\n"
-      "ldr h21, [x24], #0x2\n"
-      "ldr h26, [x23], #0x2\n"
-      "ldr h25, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
-      "ldr h22, [x20], #0x2\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"
       "tbz %x[width], #0, 13f\n"
-      "ld1 { v29.b }[2], [x27]\n"
-      "ld1 { v28.b }[2], [x26]\n"
-      "ld1 { v27.b }[2], [x25]\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 { v26.b }[2], [x23]\n"
-      "ld1 { v25.b }[2], [x22]\n"
+      "ld1 { v27.b }[2], [x23]\n"
+      "ld1 { v26.b }[2], [x22]\n"
       "ld1 { v19.b }[2], [x21]\n"
-      "ld1 { v22.b }[2], [x20]\n"
+      "ld1 { v24.b }[2], [x20]\n"
       "b 13f\n"
       "12:"  // odd_loads_1_0
-      "ldr b29, [x27, #0x0]\n"
-      "ldr b28, [x26, #0x0]\n"
-      "ldr b27, [x25, #0x0]\n"
-      "ldr b21, [x24, #0x0]\n"
-      "ldr b26, [x23, #0x0]\n"
-      "ldr b25, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
-      "ldr b22, [x20, #0x0]\n"
+      "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"
       "13:"  // Odd load end
-      "zip1 v23.4s, v29.4s, v27.4s\n"
+      "zip1 v22.4s, v28.4s, v25.4s\n"
       "subs x19, x19, #0x1\n"
-      "zip1 v20.4s, v28.4s, v21.4s\n"
-      "zip1 v16.4s, v23.4s, v20.4s\n"
-      "str q16, [%x[out_ptr], #0x0]\n"
-      "zip1 v18.4s, v26.4s, v19.4s\n"
-      "uadalp v1.8h, v16.16b\n"
-      "zip1 v17.4s, v25.4s, v22.4s\n"
-      "zip1 v16.4s, v18.4s, v17.4s\n"
-      "str q16, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v16.16b\n"
+      "zip1 v20.4s, v29.4s, v21.4s\n"
+      "zip1 v23.4s, v22.4s, v20.4s\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 v24.4s, v23.4s, v20.4s\n"
-      "zip2 v20.4s, v18.4s, v17.4s\n"
-      "str q24, [%x[out_ptr], #0x0]\n"
-      "uadalp v1.8h, v24.16b\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 v1.8h, v22.16b\n"
       "str q20, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v20.16b\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 v27.4s, v29.4s, v27.4s\n"
-      "zip2 v23.4s, v28.4s, v21.4s\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, v27.4s, v23.4s\n"
+      "zip1 v21.4s, v28.4s, v25.4s\n"
       "str q21, [%x[out_ptr], #0x0]\n"
-      "zip2 v19.4s, v26.4s, v19.4s\n"
       "uadalp v1.8h, v21.16b\n"
-      "zip2 v16.4s, v25.4s, v22.4s\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"
       "uadalp v0.8h, v18.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
       "beq 14f\n"
-      "zip2 v17.4s, v27.4s, v23.4s\n"
-      "zip2 v16.4s, v19.4s, v16.4s\n"
+      "zip2 v17.4s, v28.4s, v25.4s\n"
       "str q17, [%x[out_ptr], #0x0]\n"
+      "zip2 v16.4s, v19.4s, v16.4s\n"
       "uadalp v1.8h, v17.16b\n"
       "str q16, [%x[out_ptr], #0x10]\n"
-      "uadalp v0.8h, v16.16b\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
+      "uadalp v0.8h, v16.16b\n"
       "14:"  // Odds skip
       "uadalp v31.4s, v1.8h\n"
-      "uadalp v30.4s, v0.8h\n"
       "str q31, [%x[out_ptr], #0x0]\n"
+      "uadalp v30.4s, v0.8h\n"
       "str q30, [%x[out_ptr], #0x10]\n"
       "add %x[out_ptr], %x[out_ptr], #0x20\n"
-      : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+      : [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"
     );
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 704a4c9..1330593 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -80,39 +80,39 @@
       "blt 3f\n"
       "2:"  // Main loop head
       "ldr q27, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "subs %x[width], %x[width], #0x10\n"
       "ldr q24, [x26], #0x10\n"
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "ldr q25, [x25], #0x10\n"
+      "cmp %x[width], #0x10\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
       "ldr q21, [x24], #0x10\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
-      "ldr q22, [x23], #0x10\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
+      "ldr q23, [x23], #0x10\n"
+      "zip1 v22.2d, v25.2d, v21.2d\n"
       "ldr q18, [x22], #0x10\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
-      "ldr q19, [x21], #0x10\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
+      "zip2 v21.2d, v25.2d, v21.2d\n"
+      "ldr q20, [x21], #0x10\n"
       "ldr q16, [x20], #0x10\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
+      "zip1 v19.2d, v23.2d, v18.2d\n"
+      "prfm pldl1keep, [x27, #0x70]\n"
+      "zip2 v18.2d, v23.2d, v18.2d\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"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "str q20, [%x[out_ptr], #0x20]\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 q16, [%x[out_ptr], #0x70]\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "cmp %x[width], #0x10\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 2b\n"
       "3:"  // Main loop skip
@@ -122,37 +122,37 @@
       "ldr d24, [x26], #0x8\n"
       "ldr d25, [x25], #0x8\n"
       "ldr d21, [x24], #0x8\n"
-      "ldr d22, [x23], #0x8\n"
+      "ldr d23, [x23], #0x8\n"
       "ldr d18, [x22], #0x8\n"
-      "ldr d19, [x21], #0x8\n"
+      "ldr d20, [x21], #0x8\n"
       "ldr d16, [x20], #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 { v22.s }[2], [x23], #0x4\n"
+      "ld1 { v23.s }[2], [x23], #0x4\n"
       "ld1 { v18.s }[2], [x22], #0x4\n"
-      "ld1 { v19.s }[2], [x21], #0x4\n"
+      "ld1 { v20.s }[2], [x21], #0x4\n"
       "ld1 { v16.s }[2], [x20], #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 { v22.h }[6], [x23], #0x2\n"
+      "ld1 { v23.h }[6], [x23], #0x2\n"
       "ld1 { v18.h }[6], [x22], #0x2\n"
-      "ld1 { v19.h }[6], [x21], #0x2\n"
+      "ld1 { v20.h }[6], [x21], #0x2\n"
       "ld1 { v16.h }[6], [x20], #0x2\n"
-      "mov x19, #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 { v22.b }[14], [x23]\n"
+      "ld1 { v23.b }[14], [x23]\n"
       "ld1 { v18.b }[14], [x22]\n"
-      "ld1 { v19.b }[14], [x21]\n"
+      "ld1 { v20.b }[14], [x21]\n"
       "ld1 { v16.b }[14], [x20]\n"
       "b 11f\n"
       "4:"  // odd_loads_1_12
@@ -162,30 +162,30 @@
       "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 { v23.b }[12], [x23]\n"
       "ld1 { v18.b }[12], [x22]\n"
-      "ld1 { v19.b }[12], [x21]\n"
+      "ld1 { v20.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 { 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 { v23.h }[4], [x23], #0x2\n"
       "ld1 { v18.h }[4], [x22], #0x2\n"
-      "ld1 { v19.h }[4], [x21], #0x2\n"
+      "ld1 { v20.h }[4], [x21], #0x2\n"
       "ld1 { v16.h }[4], [x20], #0x2\n"
-      "mov x19, #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 { v22.b }[10], [x23]\n"
+      "ld1 { v23.b }[10], [x23]\n"
       "ld1 { v18.b }[10], [x22]\n"
-      "ld1 { v19.b }[10], [x21]\n"
+      "ld1 { v20.b }[10], [x21]\n"
       "ld1 { v16.b }[10], [x20]\n"
       "b 11f\n"
       "6:"  // odd_loads_1_8
@@ -193,13 +193,13 @@
       "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 { v22.b }[8], [x23]\n"
+      "ld1 { v23.b }[8], [x23]\n"
       "ld1 { v18.b }[8], [x22]\n"
-      "ld1 { v19.b }[8], [x21]\n"
+      "ld1 { v20.b }[8], [x21]\n"
       "ld1 { v16.b }[8], [x20]\n"
-      "mov x19, #0x2\n"
       "b 11f\n"
       "7:"  // odd_loads_4_0
       "tbz %x[width], #2, 9f\n"
@@ -207,28 +207,28 @@
       "ldr s24, [x26], #0x4\n"
       "ldr s25, [x25], #0x4\n"
       "ldr s21, [x24], #0x4\n"
-      "ldr s22, [x23], #0x4\n"
+      "ldr s23, [x23], #0x4\n"
       "ldr s18, [x22], #0x4\n"
-      "ldr s19, [x21], #0x4\n"
+      "ldr s20, [x21], #0x4\n"
       "ldr s16, [x20], #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 { v22.h }[2], [x23], #0x2\n"
+      "ld1 { v23.h }[2], [x23], #0x2\n"
       "ld1 { v18.h }[2], [x22], #0x2\n"
-      "ld1 { v19.h }[2], [x21], #0x2\n"
+      "ld1 { v20.h }[2], [x21], #0x2\n"
       "ld1 { v16.h }[2], [x20], #0x2\n"
-      "mov x19, #0x1\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 { v22.b }[6], [x23]\n"
+      "ld1 { v23.b }[6], [x23]\n"
       "ld1 { v18.b }[6], [x22]\n"
-      "ld1 { v19.b }[6], [x21]\n"
+      "ld1 { v20.b }[6], [x21]\n"
       "ld1 { v16.b }[6], [x20]\n"
       "b 11f\n"
       "8:"  // odd_loads_1_4
@@ -238,66 +238,66 @@
       "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 { v23.b }[4], [x23]\n"
       "ld1 { v18.b }[4], [x22]\n"
-      "ld1 { v19.b }[4], [x21]\n"
+      "ld1 { v20.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 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 h23, [x23], #0x2\n"
       "ldr h18, [x22], #0x2\n"
-      "ldr h19, [x21], #0x2\n"
+      "ldr h20, [x21], #0x2\n"
       "ldr h16, [x20], #0x2\n"
-      "mov x19, #0x1\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 { v22.b }[2], [x23]\n"
+      "ld1 { v23.b }[2], [x23]\n"
       "ld1 { v18.b }[2], [x22]\n"
-      "ld1 { v19.b }[2], [x21]\n"
+      "ld1 { v20.b }[2], [x21]\n"
       "ld1 { v16.b }[2], [x20]\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 b22, [x23, #0x0]\n"
+      "ldr b23, [x23, #0x0]\n"
       "ldr b18, [x22, #0x0]\n"
-      "ldr b19, [x21, #0x0]\n"
+      "ldr b20, [x21, #0x0]\n"
       "ldr b16, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
       "11:"  // Odd load end
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
-      "str q23, [%x[out_ptr], #0x10]\n"
-      "zip1 v17.2d, v19.2d, v16.2d\n"
-      "str q20, [%x[out_ptr], #0x20]\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"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 12f\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
       "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.2d, v22.2d, v18.2d\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, v19.2d, v16.2d\n"
+      "zip2 v16.2d, v20.2d, v16.2d\n"
       "str q18, [%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)
+      : [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"
     );
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 2317ece..3550830 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -108,48 +108,48 @@
       "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q27, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "add x19, x19, #0x1\n"
       "ldr q24, [x26], #0x10\n"
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "ldr q25, [x25], #0x10\n"
+      "subs %x[width], %x[width], #0x10\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
       "ldr q21, [x24], #0x10\n"
+      "sadalp v5.8h, v26.16b\n"
       "zip1 v23.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
       "ldr q22, [x23], #0x10\n"
+      "cmp %x[width], #0x10\n"
       "zip2 v21.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
       "ldr q18, [x22], #0x10\n"
+      "sadalp v4.8h, v23.16b\n"
       "zip1 v20.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
       "ldr q19, [x21], #0x10\n"
+      "sadalp v5.8h, v24.16b\n"
       "zip2 v18.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "ldr q16, [x20], #0x10\n"
+      "sadalp v3.8h, v20.16b\n"
       "zip1 v17.2d, v19.2d, v16.2d\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"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "sadalp v2.8h, v16.16b\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "sadalp v5.8h, v26.16b\n"
       "str q23, [%x[out_ptr], #0x10]\n"
-      "sadalp v4.8h, v23.16b\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"
       "str q24, [%x[out_ptr], #0x40]\n"
-      "sadalp v5.8h, v24.16b\n"
       "str q21, [%x[out_ptr], #0x50]\n"
-      "sadalp v4.8h, v21.16b\n"
       "str q18, [%x[out_ptr], #0x60]\n"
-      "sadalp v3.8h, v18.16b\n"
       "str q16, [%x[out_ptr], #0x70]\n"
-      "sadalp v2.8h, v16.16b\n"
-      "add x19, x19, #0x1\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "cmp %x[width], #0x10\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 3b\n"
       "5:"  // Main loop skip
@@ -174,6 +174,7 @@
       "ld1 { v16.s }[2], [x20], #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"
@@ -181,7 +182,6 @@
       "ld1 { v18.h }[6], [x22], #0x2\n"
       "ld1 { v19.h }[6], [x21], #0x2\n"
       "ld1 { v16.h }[6], [x20], #0x2\n"
-      "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[14], [x27]\n"
       "ld1 { v24.b }[14], [x26]\n"
@@ -208,13 +208,13 @@
       "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"
-      "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[10], [x27]\n"
       "ld1 { v24.b }[10], [x26]\n"
@@ -230,13 +230,13 @@
       "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"
-      "mov x19, #0x2\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
@@ -250,6 +250,7 @@
       "ldr s16, [x20], #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"
@@ -257,7 +258,6 @@
       "ld1 { v18.h }[2], [x22], #0x2\n"
       "ld1 { v19.h }[2], [x21], #0x2\n"
       "ld1 { v16.h }[2], [x20], #0x2\n"
-      "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[6], [x27]\n"
       "ld1 { v24.b }[6], [x26]\n"
@@ -284,13 +284,13 @@
       "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"
-      "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[2], [x27]\n"
       "ld1 { v24.b }[2], [x26]\n"
@@ -303,6 +303,7 @@
       "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"
@@ -310,31 +311,30 @@
       "ldr b18, [x22, #0x0]\n"
       "ldr b19, [x21, #0x0]\n"
       "ldr b16, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
       "13:"  // Odd load end
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
+      "zip1 v23.2d, v25.2d, v21.2d\n"
       "sadalp v5.8h, v26.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\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"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 14f\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
       "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
+      "zip2 v21.2d, v25.2d, v21.2d\n"
       "sadalp v5.8h, v24.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\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"
       "str q16, [%x[out_ptr], #0x30]\n"
@@ -352,7 +352,7 @@
       "add v0.4s, v0.4s, v28.4s\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)
+      : [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"
     );
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 07164d6..454260e 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) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -108,48 +108,48 @@
       "mov x19, #0x0\n"
       "4:"  // no_accumulate_16
       "ldr q27, [x27], #0x10\n"
-      "prfm pldl1keep, [x27, #0x70]\n"
+      "add x19, x19, #0x1\n"
       "ldr q24, [x26], #0x10\n"
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x26, #0x70]\n"
       "ldr q25, [x25], #0x10\n"
+      "subs %x[width], %x[width], #0x10\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "prfm pldl1keep, [x25, #0x70]\n"
       "ldr q21, [x24], #0x10\n"
+      "uadalp v5.8h, v26.16b\n"
       "zip1 v23.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x24, #0x70]\n"
       "ldr q22, [x23], #0x10\n"
+      "cmp %x[width], #0x10\n"
       "zip2 v21.2d, v25.2d, v21.2d\n"
-      "prfm pldl1keep, [x23, #0x70]\n"
       "ldr q18, [x22], #0x10\n"
+      "uadalp v4.8h, v23.16b\n"
       "zip1 v20.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x22, #0x70]\n"
       "ldr q19, [x21], #0x10\n"
+      "uadalp v5.8h, v24.16b\n"
       "zip2 v18.2d, v22.2d, v18.2d\n"
-      "prfm pldl1keep, [x21, #0x70]\n"
       "ldr q16, [x20], #0x10\n"
+      "uadalp v3.8h, v20.16b\n"
       "zip1 v17.2d, v19.2d, v16.2d\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"
+      "prfm pldl1keep, [x24, #0x70]\n"
+      "uadalp v2.8h, v16.16b\n"
+      "prfm pldl1keep, [x23, #0x70]\n"
+      "prfm pldl1keep, [x22, #0x70]\n"
+      "prfm pldl1keep, [x21, #0x70]\n"
       "prfm pldl1keep, [x20, #0x70]\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip2 v16.2d, v19.2d, v16.2d\n"
-      "uadalp v5.8h, v26.16b\n"
       "str q23, [%x[out_ptr], #0x10]\n"
-      "uadalp v4.8h, v23.16b\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"
       "str q24, [%x[out_ptr], #0x40]\n"
-      "uadalp v5.8h, v24.16b\n"
       "str q21, [%x[out_ptr], #0x50]\n"
-      "uadalp v4.8h, v21.16b\n"
       "str q18, [%x[out_ptr], #0x60]\n"
-      "uadalp v3.8h, v18.16b\n"
       "str q16, [%x[out_ptr], #0x70]\n"
-      "uadalp v2.8h, v16.16b\n"
-      "add x19, x19, #0x1\n"
-      "subs %x[width], %x[width], #0x10\n"
-      "cmp %x[width], #0x10\n"
       "add %x[out_ptr], %x[out_ptr], #0x80\n"
       "bge 3b\n"
       "5:"  // Main loop skip
@@ -174,6 +174,7 @@
       "ld1 { v16.s }[2], [x20], #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"
@@ -181,7 +182,6 @@
       "ld1 { v18.h }[6], [x22], #0x2\n"
       "ld1 { v19.h }[6], [x21], #0x2\n"
       "ld1 { v16.h }[6], [x20], #0x2\n"
-      "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[14], [x27]\n"
       "ld1 { v24.b }[14], [x26]\n"
@@ -208,13 +208,13 @@
       "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"
-      "mov x19, #0x2\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[10], [x27]\n"
       "ld1 { v24.b }[10], [x26]\n"
@@ -230,13 +230,13 @@
       "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"
-      "mov x19, #0x2\n"
       "b 13f\n"
       "9:"  // odd_loads_4_0
       "tbz %x[width], #2, 11f\n"
@@ -250,6 +250,7 @@
       "ldr s16, [x20], #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"
@@ -257,7 +258,6 @@
       "ld1 { v18.h }[2], [x22], #0x2\n"
       "ld1 { v19.h }[2], [x21], #0x2\n"
       "ld1 { v16.h }[2], [x20], #0x2\n"
-      "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[6], [x27]\n"
       "ld1 { v24.b }[6], [x26]\n"
@@ -284,13 +284,13 @@
       "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"
-      "mov x19, #0x1\n"
       "tbz %x[width], #0, 13f\n"
       "ld1 { v27.b }[2], [x27]\n"
       "ld1 { v24.b }[2], [x26]\n"
@@ -303,6 +303,7 @@
       "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"
@@ -310,31 +311,30 @@
       "ldr b18, [x22, #0x0]\n"
       "ldr b19, [x21, #0x0]\n"
       "ldr b16, [x20, #0x0]\n"
-      "mov x19, #0x1\n"
       "13:"  // Odd load end
       "zip1 v26.2d, v27.2d, v24.2d\n"
-      "subs x19, x19, #0x1\n"
-      "zip1 v23.2d, v25.2d, v21.2d\n"
       "str q26, [%x[out_ptr], #0x0]\n"
-      "zip1 v20.2d, v22.2d, v18.2d\n"
+      "zip1 v23.2d, v25.2d, v21.2d\n"
       "uadalp v5.8h, v26.16b\n"
-      "zip1 v17.2d, v19.2d, v16.2d\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"
       "add %x[out_ptr], %x[out_ptr], #0x40\n"
       "beq 14f\n"
       "zip2 v24.2d, v27.2d, v24.2d\n"
-      "zip2 v21.2d, v25.2d, v21.2d\n"
       "str q24, [%x[out_ptr], #0x0]\n"
-      "zip2 v18.2d, v22.2d, v18.2d\n"
+      "zip2 v21.2d, v25.2d, v21.2d\n"
       "uadalp v5.8h, v24.16b\n"
-      "zip2 v16.2d, v19.2d, v16.2d\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"
       "str q16, [%x[out_ptr], #0x30]\n"
@@ -352,7 +352,7 @@
       "add v0.4s, v0.4s, v28.4s\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)
+      : [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"
     );