diff options
author | Pablo Tello <pablo.tello@arm.com> | 2018-03-20 16:46:55 +0000 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:49:16 +0000 |
commit | 99ef8407cd5b27fdec6f8dfaf8b55f820b6dea71 (patch) | |
tree | 7d7448ebc71d20c15611076375eb0cbb22f83f5a /src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp | |
parent | 2d9de0a3fa6ad858e70040124f362799a962bb6a (diff) | |
download | ComputeLibrary-99ef8407cd5b27fdec6f8dfaf8b55f820b6dea71.tar.gz |
COMPMID-881: Updated arm_gemm to the lastest
Change-Id: Iba2664f33320e79bd15ca9c1399e65e4cc165be6
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125265
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@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 | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp index 1baa21fd1b..a5b41cac2f 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp @@ -49,16 +49,16 @@ UniqueGemmCommon<float, float> gemm<float, float>(const CPUInfo &ci, const unsig return UniqueGemmCommon<float, float>(new GemvPretransposed<sgemv_pretransposed, float, float>(&ci, N, K, trB, beta)); } - /* GemvNativeTransposed: requires M=1, no trA or trB, doesn't handle beta */ - if(M == 1 && beta == 1.0f && !trA && !trB) + /* GemvNativeTransposed: requires M=1, no trA or trB, doesn't handle alpha */ + if(M == 1 && alpha == 1.0f && !trA && !trB) { - return UniqueGemmCommon<float, float>(new GemvNativeTransposed<sgemv_trans, float, float>(&ci, N, K, alpha)); + return UniqueGemmCommon<float, float>(new GemvNativeTransposed<sgemv_trans, float, float>(&ci, N, K, beta)); } - /* Native GEMM: requires M to be a multiple of 4, K a multiple of 4, N a + /* Native GEMM: requires M to be a multiple of 4, K at least 4, N a * multiple of 16, doesn't handle alpha and only makes sense for small * sizes. */ - if(N <= 128 && K <= 128 && ((M % 4) == 0) && ((K % 4) == 0) && ((N % 16) == 0) && alpha == 1.0f) + if(N <= 128 && K <= 128 && ((M % 4) == 0) && (K >= 4) && ((N % 16) == 0) && alpha == 1.0f) { return UniqueGemmCommon<float, float>(new GemmNative<sgemm_native_16x4, float, float>(&ci, M, N, K, beta)); } |