diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2020-09-18 09:48:16 +0100 |
|---|---|---|
| committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2020-09-21 13:06:03 +0000 |
| commit | 34654b2d8dcaf268a9d1bf9e0cdb5ba548ced2b7 (patch) | |
| tree | 7aa2868b7d3f281b0397cb94bd6d0c80f7fedaf3 | |
| parent | 6f9313477f6a158210479996523c210452d4f07a (diff) | |
| download | ComputeLibrary-34654b2d8dcaf268a9d1bf9e0cdb5ba548ced2b7.tar.gz | |
COMPMID-3787: Update heuristic for m = 1 on CLGEMMReshapedOnlyRHSKernel
- The change affects Mali-G71 GPUs and should improve the performance of
GEMM in case of m = 1
Change-Id: I6b0e217e93fe468ec1325a5da74684811519c42f
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4002
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
| -rw-r--r-- | src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp | 8 |
1 files changed, 3 insertions, 5 deletions
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp index f9b65dc931..dcb0e0be96 100644 --- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp +++ b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp @@ -122,15 +122,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi if(m == 1) { - if(n > 2048) + if ( n <= 2548 ) { - const unsigned int h0 = std::max(n / 4, 1U); - return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, h0, false, true, false, true); + return configure_lhs_rhs_info(m,n,1,2,16,1,4,false,true,false,true,false); } else { - const unsigned int h0 = std::max(n / 2, 1U); - return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, h0, false, true, false, true); + return configure_lhs_rhs_info(m,n,1,4,16,1,8,false,true,false,true,false); } } else |