diff options
author | Francesco Petrogalli <francesco.petrogalli@arm.com> | 2022-06-30 10:22:01 +0000 |
---|---|---|
committer | Francesco Petrogalli <francesco.petrogalli@arm.com> | 2022-07-19 09:26:27 +0000 |
commit | 553f6953fe3bdfad53c11c25f305a16d79d83b24 (patch) | |
tree | 73642b948b79662096f593458c6138d2f7f48ec6 /src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp | |
parent | 99c46475daf277aa53e6747f9e41209f418fed33 (diff) | |
download | ComputeLibrary-553f6953fe3bdfad53c11c25f305a16d79d83b24.tar.gz |
[ONCPUML-951] Variable weight support for Convolution.
API changes for NEGEMMConvolutionLayer and CpuGemmConv2d
Built with:
scons neon=1 opencl=0 os=linux arch=armv8.2-a multi_isa=1 \
build=native -j32 Werror=false validation_tests=1 build_dir=opt \
standalone=1 asserts=1 experimental_fixed_format_kernels=1 .
Tested with:
./build/opt/tests/arm_compute_validation
Hardware where the test executable was run:
Neoverse N1
Test coverage:
* NEGEMMConvolutionLayer, CpuGemmConv2d
* NHWC (the only one supported by the fixed-format kernels)
* F16, F32
* Shapes: RunSmall
Change-Id: I4fd3e495a7cbf61210ea02d37440ba9652934e99
Signed-off-by: Francesco Petrogalli <francesco.petrogalli@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7632
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp | 18 |
1 files changed, 9 insertions, 9 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp index 4f7e191fb3..0fc9e8b912 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp @@ -31,12 +31,12 @@ #include "gemv_pretransposed.hpp" #include "kernels/a32_sgemm_8x6.hpp" -#ifdef ENABLE_FIXED_FORMAT_KERNELS +#ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #include "kernels/a64_ffhybrid_fp32_mla_6x16.hpp" #include "kernels/a64_ffhybrid_fp32bf16fp32_mmla_4x24.hpp" #include "kernels/a64_ffinterleaved_bf16fp32_mmla_8x12.hpp" #include "kernels/a64_ffinterleaved_fp32_mla_8x12.hpp" -#endif // ENABLE_FIXED_FORMAT_KERNELS +#endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #include "kernels/a64_hybrid_fp32bf16fp32_mmla_4x24.hpp" #include "kernels/a64_hybrid_fp32bf16fp32_mmla_6x16.hpp" #include "kernels/a64_hybrid_fp32_mla_4x24.hpp" @@ -48,12 +48,12 @@ #include "kernels/a64_smallK_hybrid_fp32_mla_6x4.hpp" #include "kernels/a64_smallK_hybrid_fp32_mla_8x4.hpp" -#ifdef ENABLE_FIXED_FORMAT_KERNELS +#ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #include "kernels/sve_ffhybrid_fp32_mla_6x4VL.hpp" #include "kernels/sve_ffhybrid_fp32bf16fp32_mmla_4x6VL.hpp" #include "kernels/sve_ffinterleaved_fp32_mla_8x3VL.hpp" #include "kernels/sve_ffinterleaved_bf16fp32_mmla_8x3VL.hpp" -#endif // ENABLE_FIXED_FORMAT_KERNELS +#endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #include "kernels/sve_hybrid_fp32bf16fp32_mmla_4x6VL.hpp" #include "kernels/sve_hybrid_fp32bf16fp32_mmla_6x4VL.hpp" #include "kernels/sve_hybrid_fp32_mla_6x4VL.hpp" @@ -165,7 +165,7 @@ GemmImplementation<float, float>::with_estimate( [](const GemmArgs &args) { return GemmInterleaved<cls_sve_interleaved_fp32_mla_8x3VL, float, float>::estimate_cycles<float>(args); }, [](const GemmArgs &args) { return new GemmInterleaved<cls_sve_interleaved_fp32_mla_8x3VL, float, float>(args); } ), - #ifdef ENABLE_FIXED_FORMAT_KERNELS + #ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #ifdef ARM_COMPUTE_ENABLE_BF16 GemmImplementation<float, float>::with_estimate( GemmMethod::GEMM_INTERLEAVED, @@ -200,7 +200,7 @@ GemmImplementation<float, float>::with_estimate( [](const GemmArgs &args) { return GemmHybridIndirectFixedFormat<cls_sve_ffhybrid_fp32_mla_6x4VL, float, float>::estimate_cycles<float>(args); }, [](const GemmArgs &args) { return new GemmHybridIndirectFixedFormat<cls_sve_ffhybrid_fp32_mla_6x4VL, float, float>(args); } ), -#endif // ENABLE_FIXED_FORMAT_KERNELS +#endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #endif // ARM_COMPUTE_ENABLE_SVE // Cortex-A35 specific kernel - use for any problem on A35, and never in any other cases. { @@ -253,7 +253,7 @@ GemmImplementation<float, float>::with_estimate( [](const GemmArgs &args) { return GemmInterleaved<cls_a64_sgemm_8x12, float, float>::estimate_cycles<float>(args); }, [](const GemmArgs &args) { return new GemmInterleaved<cls_a64_sgemm_8x12, float, float>(args); } ), -#ifdef ENABLE_FIXED_FORMAT_KERNELS +#ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #ifdef ARM_COMPUTE_ENABLE_BF16 // "fast mode" (BF16) kernels GemmImplementation<float, float>::with_estimate( @@ -289,7 +289,7 @@ GemmImplementation<float, float>::with_estimate( [](const GemmArgs &args) { return GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32_mla_6x16, float, float>::estimate_cycles<float>(args); }, [](const GemmArgs &args) { return new GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32_mla_6x16, float, float>(args); } ), -#endif // ENABLE_FIXED_FORMAT_KERNELS +#endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS #endif // __aarch64__ #ifdef __arm__ @@ -318,7 +318,7 @@ const GemmImplementation<float, float> *gemm_implementation_list<float, float>() /* Explicitly instantiate the external functions for these types. */ template UniqueGemmCommon<float, float> gemm<float, float, Nothing>(const GemmArgs &args, const Nothing &); -template bool has_opt_gemm<float, float, Nothing>(const GemmArgs &args, const Nothing &); +template bool has_opt_gemm<float, float, Nothing>(WeightFormat &weight_format, const GemmArgs &args, const Nothing &); template KernelDescription get_gemm_method<float, float, Nothing>(const GemmArgs &args, const Nothing &); template std::vector<KernelDescription> get_compatible_kernels<float, float, Nothing> (const GemmArgs &args, const Nothing &); |