commit a084b46835d20fdfe6e590b91b7ca64fba3542df
author Aleksandr Nikolaev <aleksandr.nikolaev@arm.com> Thu Jun 25 12:25:52 2020 +0100
committer Georgios Pinitas <georgios.pinitas@arm.com> Thu Jul 16 16:02:50 2020 +0000
tree e4abe49cc10f3406dcaad9d3dfbbc22cc54476a6
parent f738fe6b6e059916294c48b942952c261569df18

[ONCPUML-97]: Implement "int8" support for 2D decomposition at high core counts

Interleaved2d functionality was extended to uint8 and int8 kernels.

Change-Id: If78facbce56e9ec7b2f4c23436af0bd5db7f7b69
Signed-off-by: Aleksandr Nikolaev <aleksandr.nikolaev@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3467
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp

diff --git a/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp b/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
index d5a9e58..caab2e2 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_uint8.cpp
@@ -27,6 +27,7 @@
 #include "gemm_common.hpp"
 #include "gemm_implementation.hpp"
 #include "gemm_interleaved.hpp"
+#include "gemm_interleaved_pretransposed_2d.hpp"
 #include "gemm_hybrid.hpp"
 
 #include "kernels/a64_gemm_u16_12x8.hpp"
@@ -107,15 +108,29 @@
     [](const GemmArgs &args) { return new GemmHybrid<hybrid_u8u32_dot_16x4, uint8_t, uint32_t>(args); }
 },
 {
+    GemmMethod::GEMM_INTERLEAVED_2D,
+    "gemm_u8_12x8_2d",
+    [](const GemmArgs &args) { return args._ci->has_dotprod(); },
+    [](const GemmArgs &args) { return (args._maxthreads >= 8) && (args._Msize >= 8) && (args._Nsize >= 8) ; },
+    [](const GemmArgs &args) { return new GemmInterleavedPretransposed2d<gemm_u8_12x8, uint8_t, uint32_t>(args); }
+},
+{
     GemmMethod::GEMM_INTERLEAVED,
-    "gemm_u8_12x8",
+    "gemm_u8_12x8_1d",
     [](const GemmArgs &args) { return args._ci->has_dotprod(); },
     nullptr,
     [](const GemmArgs &args) { return new GemmInterleaved<gemm_u8_12x8, uint8_t, uint32_t>(args); }
 },
 {
+    GemmMethod::GEMM_INTERLEAVED_2D,
+    "gemm_u8_4x4_2d",
+    nullptr,
+    [](const GemmArgs &args) { return (args._maxthreads >= 8) && (args._Msize >= 8) && (args._Nsize >= 8); },
+    [](const GemmArgs &args) { return new GemmInterleavedPretransposed2d<gemm_u8_4x4, uint8_t, uint32_t>(args); }
+},
+{
     GemmMethod::GEMM_INTERLEAVED,
-    "gemm_u8_4x4",
+    "gemm_u8_4x4_1d",
     nullptr,
     nullptr,
     [](const GemmArgs &args) { return new GemmInterleaved<gemm_u8_4x4, uint8_t, uint32_t>(args); }