COMPMID-3003: Integrate assembly kernels utilizing MMLA instruction.

MMLA is a matrix-multiply instruction introduced on armv8.6-A

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I572a54981d48f5a1e0e9e51102cb7ae28ad87806
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2663
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 20 insertions, 0 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp b/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
index aead814d7e..430d35e06d 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
@@ -34,10 +34,12 @@
 #include "kernels/a64_gemm_u8_12x8.hpp"
 #include "kernels/a64_gemm_u8_4x4.hpp"
 #include "kernels/a64_hybrid_u8u32_dot_16x4.hpp"
+#include "kernels/a64_interleaved_u8u32_mmla_12x8.hpp"
 #include "kernels/a64_smallK_hybrid_u8u32_dot_4x6.hpp"
 #include "kernels/a64_smallK_hybrid_u8u32_dot_4x8.hpp"
 #include "kernels/sve_hybrid_u8u32_dot_4VLx4.hpp"
 #include "kernels/sve_interleaved_u8u32_dot_3VLx8.hpp"
+#include "kernels/sve_interleaved_u8u32_mmla_3VLx8.hpp"
 #include "kernels/sve_native_u8u32_dot_4VLx4.hpp"
 #include "kernels/sve_smallK_hybrid_u8u32_dot_1VLx8.hpp"
 
@@ -45,6 +47,15 @@ namespace arm_gemm {
 
 static const GemmImplementation<uint8_t, uint32_t> gemm_u8_methods[] = {
 #ifdef __ARM_FEATURE_SVE
+#ifdef V8P6
+{
+    GemmMethod::GEMM_INTERLEAVED,
+    "interleaved_u8u32_mmla_3VLx8",
+    [](const GemmArgs &args) { return (args._Ksize>8); },
+    nullptr,
+    [](const GemmArgs &args) { return new GemmInterleaved<interleaved_u8u32_mmla_3VLx8, uint8_t, uint32_t>(args); }
+},
+#endif
 {
     GemmMethod::GEMM_HYBRID,
     "smallK_hybrid_u8u32_dot_1VLx8",
@@ -74,6 +85,15 @@ static const GemmImplementation<uint8_t, uint32_t> gemm_u8_methods[] = {
     [](const GemmArgs &args) { return new GemmInterleaved<interleaved_u8u32_dot_3VLx8, uint8_t, uint32_t>(args); }
 },
 #endif
+#ifdef V8P6
+{
+    GemmMethod::GEMM_INTERLEAVED,
+    "interleaved_u8u32_mmla_12x8",
+    [](const GemmArgs &args) { return (args._Ksize>8); },
+    nullptr,
+    [](const GemmArgs &args) { return new GemmInterleaved<interleaved_u8u32_mmla_12x8, uint8_t, uint32_t>(args); }
+},
+#endif
 {
     GemmMethod::GEMM_HYBRID,
     "smallK_hybrid_u8u32_dot_4x8",