diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-10-14 19:03:09 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-10-23 12:08:12 +0000 |
commit | 48b3ef89de5f21a0169d8416e3d54081f82c7bf8 (patch) | |
tree | f857d733ccf446c704823dc7ac796a96eb55095e /src/core/NEON/kernels/arm_gemm/gemm_int8.cpp | |
parent | 1dce3101ef8d77c8cf0af7dfd4af6595a0136b91 (diff) | |
download | ComputeLibrary-48b3ef89de5f21a0169d8416e3d54081f82c7bf8.tar.gz |
COMPMID-2577: Fuse bias addition and activation in gemm assembly kernels
Change-Id: I7f52112d2d05b1ea3d3f3d4b19b8eafab05d6c44
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2141
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_int8.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_int8.cpp | 46 |
1 files changed, 23 insertions, 23 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_int8.cpp b/src/core/NEON/kernels/arm_gemm/gemm_int8.cpp index 49241a7129..a3446b9ddc 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_int8.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_int8.cpp @@ -48,66 +48,66 @@ static const GemmImplementation<int8_t, int32_t> gemm_s8_methods[] = { { GemmMethod::GEMM_HYBRID, "smallK_hybrid_s8s32_dot_1VLx8", - [](const GemmArgs<int32_t> &args) { return args._Ksize<=64 && args._alpha==1 && args._beta==0 && !args._trA && args._pretransposed_hint; }, + [](const GemmArgs &args) { return args._Ksize<=64 && !args._trA && args._pretransposed_hint; }, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_1VLx8, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_1VLx8, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_HYBRID, "hybrid_s8s32_dot_4VLx4", - [](const GemmArgs<int32_t> &args) { return args._Ksize>=16 && args._alpha==1 && !args._trA && args._pretransposed_hint; }, - [](const GemmArgs<int32_t> &args) { return ((args._Ksize <= 128) && (args._Nsize <= 128)) || ((args._nmulti > 1) && ((args._Msize / args._maxthreads) < 8)); }, - [](const GemmArgs<int32_t> &args) { return new GemmHybrid<hybrid_s8s32_dot_4VLx4, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return args._Ksize>=16 && !args._trA && args._pretransposed_hint; }, + [](const GemmArgs &args) { return ((args._Ksize <= 128) && (args._Nsize <= 128)) || ((args._nmulti > 1) && ((args._Msize / args._maxthreads) < 8)); }, + [](const GemmArgs &args) { return new GemmHybrid<hybrid_s8s32_dot_4VLx4, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_NATIVE, "native_s8s32_dot_4VLx4", - [](const GemmArgs<int32_t> &args) { return (args._Ksize>=16 && args._alpha==1 && !args._trA && !args._trB); }, - [](const GemmArgs<int32_t> &args) { return ((args._Ksize <= 128) && (args._Nsize <= 128)); }, - [](const GemmArgs<int32_t> &args) { return new GemmNative<native_s8s32_dot_4VLx4, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return (args._Ksize>=16 && !args._trA && !args._trB); }, + [](const GemmArgs &args) { return ((args._Ksize <= 128) && (args._Nsize <= 128)); }, + [](const GemmArgs &args) { return new GemmNative<native_s8s32_dot_4VLx4, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_INTERLEAVED, "interleaved_s8s32_dot_3VLx8", - [](const GemmArgs<int32_t> &args) { return (args._Ksize>4) && args._alpha==1 && (args._beta == 0 || args._beta==1); }, + [](const GemmArgs &args) { return (args._Ksize>4); }, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmInterleaved<interleaved_s8s32_dot_3VLx8, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmInterleaved<interleaved_s8s32_dot_3VLx8, int8_t, int32_t>(args); } }, #endif { GemmMethod::GEMM_HYBRID, "smallK_hybrid_s8s32_dot_4x8", - [](const GemmArgs<int32_t> &args) { return args._ci->has_dotprod() && (args._Nsize % 4 == 0) && (args._Ksize<=32) && args._alpha==1 && args._beta==0 && !args._trA && args._pretransposed_hint; }, + [](const GemmArgs &args) { return args._ci->has_dotprod() && (args._Nsize % 4 == 0) && (args._Ksize<=32) && !args._trA && args._pretransposed_hint; }, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_4x8, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_4x8, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_HYBRID, "smallK_hybrid_s8s32_dot_4x6", - [](const GemmArgs<int32_t> &args) { return args._ci->has_dotprod() && (args._Nsize % 4 == 0) && (args._Ksize>32) && (args._Ksize<=64) && args._alpha==1 && args._beta==0 && !args._trA && args._pretransposed_hint; }, + [](const GemmArgs &args) { return args._ci->has_dotprod() && (args._Nsize % 4 == 0) && (args._Ksize>32) && (args._Ksize<=64) && !args._trA && args._pretransposed_hint; }, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_4x6, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmHybrid<smallK_hybrid_s8s32_dot_4x6, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_HYBRID, "hybrid_s8s32_dot_16x4", - [](const GemmArgs<int32_t> &args) { return args._ci->has_dotprod() && args._Ksize>=16 && args._alpha==1 && !args._trA && !args._trB && args._pretransposed_hint; }, - [](const GemmArgs<int32_t> &args) { return args._Nsize<=256 && args._Ksize>128; }, - [](const GemmArgs<int32_t> &args) { return new GemmHybrid<hybrid_s8s32_dot_16x4, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return args._ci->has_dotprod() && args._Ksize>=16 && !args._trA && !args._trB && args._pretransposed_hint; }, + [](const GemmArgs &args) { return args._Nsize<=256 && args._Ksize>128; }, + [](const GemmArgs &args) { return new GemmHybrid<hybrid_s8s32_dot_16x4, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_INTERLEAVED, "gemm_s8_12x8", - [](const GemmArgs<int32_t> &args) { return args._ci->has_dotprod() && args._alpha==1 && (args._beta==0 || args._beta==1); }, + [](const GemmArgs &args) { return args._ci->has_dotprod(); }, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmInterleaved<gemm_s8_12x8, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmInterleaved<gemm_s8_12x8, int8_t, int32_t>(args); } }, { GemmMethod::GEMM_INTERLEAVED, "gemm_s8_4x4", nullptr, nullptr, - [](const GemmArgs<int32_t> &args) { return new GemmInterleaved<gemm_s8_4x4, int8_t, int32_t>(args); } + [](const GemmArgs &args) { return new GemmInterleaved<gemm_s8_4x4, int8_t, int32_t>(args); } }, { GemmMethod::DEFAULT, @@ -124,9 +124,9 @@ const GemmImplementation<int8_t, int32_t> *gemm_implementation_list<int8_t, int3 } /* Explicitly instantiate the external functions for these types. */ -template UniqueGemmCommon<int8_t, int32_t> gemm<int8_t, int32_t, Nothing>(const GemmArgs<int32_t> &args, const Nothing &); -template KernelDescription get_gemm_method<int8_t, int32_t, Nothing>(const GemmArgs<int32_t> &args, const Nothing &); -template std::vector<KernelDescription> get_compatible_kernels<int8_t, int32_t, Nothing> (const GemmArgs<int32_t> &args, const Nothing &); +template UniqueGemmCommon<int8_t, int32_t> gemm<int8_t, int32_t, Nothing>(const GemmArgs &args, const Nothing &); +template KernelDescription get_gemm_method<int8_t, int32_t, Nothing>(const GemmArgs &args, const Nothing &); +template std::vector<KernelDescription> get_compatible_kernels<int8_t, int32_t, Nothing> (const GemmArgs &args, const Nothing &); } // namespace arm_gemm |