diff options
| author | Gunes Bayir <gunes.bayir@arm.com> | 2024-04-26 16:51:54 +0100 |
|---|---|---|
| committer | Gunes Bayir <gunes.bayir@arm.com> | 2024-04-29 09:31:44 +0000 |
| commit | e5ef8c159a14872dda5e36e320f07b0963858d8c (patch) | |
| tree | 2d359eeff83b54b3e7e65484abc37e531aedbdf2 /src/core/NEON | |
| parent | 499b5bca1a897461d4105ba52e4c766ddb5f564a (diff) | |
| download | ComputeLibrary-e5ef8c159a14872dda5e36e320f07b0963858d8c.tar.gz | |
Disable SME2 Gemmlowp s8f32 kernel selection in case results needs to be accumulated
Similar to https://review.mlplatform.org/c/ml/ComputeLibrary/+/11500, s8f32 kernels do not support accumulate mode. This patch modifies the kernel selection and also adds more tests to stress these test cases better.
Partially Resolves: COMPMID-6995
Change-Id: I40e19446c012eb7334e4511e254cce0d635aa234
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11503
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Radu Salavat <radu.salavat@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp b/src/core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp index 782399df8c..38d9b763f6 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp @@ -55,7 +55,7 @@ static const GemmImplementation<int8_t, float, DequantizeFloat> gemm_s8fp32_meth { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8qfp32_mopa_1VLx4VL.hpp", - [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2(); }, + [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2() && !args._accumulate; }, [](const GemmArgs &args, const DequantizeFloat &) { const auto VL = sme::get_vector_length<float>(); return args._Msize <= VL || (2*VL < args._Msize && args._Msize <= 3*VL); }, [](const GemmArgs &args, const DequantizeFloat &dq) { return new GemmInterleavedNoMergeDequantized<cls_sme2_interleaved_nomerge_s8qfp32_mopa_1VLx4VL, int8_t, float>(args, dq); } @@ -63,7 +63,7 @@ static const GemmImplementation<int8_t, float, DequantizeFloat> gemm_s8fp32_meth { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8qfp32_mopa_4Vx1VL.hpp", - [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2(); }, + [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2() && !args._accumulate; }, [](const GemmArgs &args, const DequantizeFloat &) { const auto VL = sme::get_vector_length<float>(); return args._Nsize <= VL || (2*VL < args._Nsize && args._Nsize <= 3*VL); }, [](const GemmArgs &args, const DequantizeFloat &dq) { return new GemmInterleavedNoMergeDequantized<cls_sme2_interleaved_nomerge_s8qfp32_mopa_4VLx1VL, int8_t, float>(args, dq); } @@ -71,7 +71,7 @@ static const GemmImplementation<int8_t, float, DequantizeFloat> gemm_s8fp32_meth { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8qfp32_mopa_2Vx2VL.hpp", - [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2(); }, + [](const GemmArgs &args, const DequantizeFloat &) { return args._ci->has_sme2() && !args._accumulate; }, nullptr, [](const GemmArgs &args, const DequantizeFloat &dq) { return new GemmInterleavedNoMergeDequantized<cls_sme2_interleaved_nomerge_s8qfp32_mopa_2VLx2VL, int8_t, float>(args, dq); } }, |