diff options
author | David Mansell <David.Mansell@arm.com> | 2023-10-10 15:20:55 +0100 |
---|---|---|
committer | David Mansell <David.Mansell@arm.com> | 2023-10-12 10:00:24 +0000 |
commit | aeced744b854758768243833bcdf999c0c3c1a5b (patch) | |
tree | d0072699ef925efd0510419e7361f3f47c60c0f4 /src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp | |
parent | 95d477e0662c172c264cf495d196dfce6a791248 (diff) | |
download | ComputeLibrary-aeced744b854758768243833bcdf999c0c3c1a5b.tar.gz |
arm_gemm: Add SME2 FP16 GEMV.
Change-Id: I1f73819c25c66e4d13198e9c79755808d92b343d
Signed-off-by: David Mansell <David.Mansell@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10466
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp b/src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp index efdaeeb170..ff16be36e7 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_fp16.cpp @@ -32,6 +32,7 @@ #include "gemm_hybrid_indirect.hpp" #include "gemm_implementation.hpp" #include "gemm_interleaved.hpp" +#include "gemv_pretransposed.hpp" #include "kernels/a32_sgemm_8x6.hpp" #ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS @@ -42,6 +43,7 @@ #include "kernels/a64_hybrid_fp16_mla_6x32.hpp" #include "kernels/a64_sgemm_8x12.hpp" #ifdef ARM_COMPUTE_ENABLE_SME2 +#include "kernels/sme2_gemv_fp16_mla_16VL.hpp" #include "kernels/sme2_interleaved_nomerge_fp16fp32fp16_mopa_1VLx4VL.hpp" #include "kernels/sme2_interleaved_nomerge_fp16fp32fp16_mopa_2VLx2VL.hpp" #include "kernels/sme2_interleaved_nomerge_fp16fp32fp16_mopa_4VLx1VL.hpp" @@ -59,6 +61,13 @@ static const GemmImplementation<__fp16, __fp16> gemm_fp16_methods[] = { #ifdef ARM_COMPUTE_ENABLE_SVE #ifdef ARM_COMPUTE_ENABLE_SME2 { + GemmMethod::GEMM_HYBRID, + "sme2_gemv_fp16_mla_16VL", + [](const GemmArgs &args) { return args._ci->has_sme2() && args._Msize==1 && args._nbatches==1 && !args._indirect_input; }, + nullptr, + [](const GemmArgs &args) { return new GemvPretransposed<cls_sme2_gemv_fp16_mla_16VL, __fp16, __fp16>(args); } +}, +{ GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_fp16fp32fp16_mopa_4VLx1VL", [](const GemmArgs &args) { return args._ci->has_sme2(); }, |