From 99ef8407cd5b27fdec6f8dfaf8b55f820b6dea71 Mon Sep 17 00:00:00 2001 From: Pablo Tello Date: Tue, 20 Mar 2018 16:46:55 +0000 Subject: COMPMID-881: Updated arm_gemm to the lastest Change-Id: Iba2664f33320e79bd15ca9c1399e65e4cc165be6 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125265 Tested-by: Jenkins Reviewed-by: Georgios Pinitas --- src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp') 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 gemm(const CPUInfo &ci, const unsig return UniqueGemmCommon(new GemvPretransposed(&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(new GemvNativeTransposed(&ci, N, K, alpha)); + return UniqueGemmCommon(new GemvNativeTransposed(&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(new GemmNative(&ci, M, N, K, beta)); } -- cgit v1.2.1