diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2020-11-18 14:56:09 +0000 |
---|---|---|
committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2020-11-18 16:42:39 +0000 |
commit | 8b84aaa4db093ac08efa96c2cbf766e800465529 (patch) | |
tree | 656c1de7030dc742fbb84bf9da451f913f97f129 /src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp | |
parent | 75eea338eb232ebdafa2fb84d22e711b5f964785 (diff) | |
download | ComputeLibrary-8b84aaa4db093ac08efa96c2cbf766e800465529.tar.gz |
COMPMID-3981: Fix missing fallback for export to cl_image
- Fix missing fallback in the CLGEMMReshaped heuristic on Mali-G77
Change-Id: I0a243c7ed153216966d0809a3b3348f030a845eb
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4463
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp')
-rw-r--r-- | src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp | 49 |
1 files changed, 42 insertions, 7 deletions
diff --git a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp index 3f82dcab00..4fd446f647 100644 --- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp +++ b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp @@ -95,6 +95,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio const float r_mk = static_cast<float>(m) / static_cast<float>(k); const float r_nk = static_cast<float>(n) / static_cast<float>(k); + GEMMLHSMatrixInfo lhs_info_buf; + GEMMRHSMatrixInfo rhs_info_buf; + GEMMLHSMatrixInfo lhs_info_img; + GEMMRHSMatrixInfo rhs_info_img; + + std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, false, false, true, false, false); + if(r_mk <= 0.11824845522642136) { if(workload <= 880.0) @@ -111,7 +118,11 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio } else { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } } else @@ -135,7 +146,11 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio { if(r_mn <= 2.545312523841858) { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } else { @@ -146,11 +161,19 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio { if(workload <= 2881.199951171875) { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, false, false, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, false, false, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } else { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } } } @@ -160,16 +183,28 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio { if(r_mn <= 6.010416746139526) { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } else { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } } else { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true); + std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true); + + return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img), + std::make_pair(lhs_info_buf, rhs_info_buf), + n, k, b, DataType::F16); } } } |