commit be13cead34e566bdd561ad3ffc3f645b460e482e
author Michael Tyler <michael.tyler@arm.com> Tue Jan 17 11:04:14 2023 +0000
committer Gian Marco Iodice <gianmarco.iodice@arm.com> Wed Jan 18 09:43:38 2023 +0000
tree cdc086de205d5a07fdd816afa6333d0b2f38d4e9
parent 13bab71a76096985752a9e12711507021e25858d

Revert "Update CPU kernels to remove x19"

This reverts commit 3c59f01c209d2732a15d97d65565ead964787a8b.

Resolves: COMPMID-5817
Change-Id: Ie2443a21854a95db1e3d0cafa2121c0187a5e237
Signed-off-by: Michael Tyler <michael.tyler@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8974
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

Android.bp

diff --git a/Android.bp b/Android.bp
index e81d264..8867d5b 100644
--- a/Android.bp
+++ b/Android.bp
@@ -1054,14 +1054,17 @@
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32bf16fp32_planar_5x5_s1_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32bf16fp32_planar_5x5_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_s8q_planar_3x3_s1_4rows_dot_za/generic.cpp",
+                "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_s8q_planar_3x3_s2_2rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_s8q_planar_3x3_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_s8q_planar_5x5_s1_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_s8q_planar_5x5_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8q_planar_3x3_s1_4rows_dot_za/generic.cpp",
+                "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8q_planar_3x3_s2_2rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8q_planar_3x3_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8q_planar_5x5_s1_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8q_planar_5x5_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8s8u8q_planar_3x3_s1_4rows_dot_za/generic.cpp",
+                "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8s8u8q_planar_3x3_s2_2rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8s8u8q_planar_3x3_s2_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8s8u8q_planar_5x5_s1_4rows_dot_za/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_u8s8u8q_planar_5x5_s2_4rows_dot_za/generic.cpp",
@@ -1075,8 +1078,10 @@
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_3x3_s2_output2x2_mla_depthfirst/generic_indirect.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_direct.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp",
+                "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst/generic_direct.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst/generic_indirect.cpp",
+                "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst_strided/generic.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_indirect.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp",
@@ -1298,6 +1303,9 @@
                 "src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/a64fx.cpp",
                 "src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp",
                 "src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_mmla_8x3VL/generic.cpp",
+                "src/core/NEON/kernels/arm_gemm/kernels/sve_smallK_hybrid_fp32_mla_8x1VL/generic.cpp",
+                "src/core/NEON/kernels/arm_gemm/kernels/sve_smallK_hybrid_s8s32_dot_8x1VL/generic.cpp",
+                "src/core/NEON/kernels/arm_gemm/kernels/sve_smallK_hybrid_u8u32_dot_8x1VL/generic.cpp",
                 "src/core/NEON/kernels/convolution/winograd/input_transforms/a64_fp16_6x6.cpp",
                 "src/core/NEON/kernels/convolution/winograd/input_transforms/a64_fp32_6x6.cpp",
                 "src/core/NEON/kernels/convolution/winograd/input_transforms/sme_fp32_mla_6x6.cpp",