diff options
Diffstat (limited to 'src/core')
-rw-r--r-- | src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationValhall.cpp | 31 |
1 files changed, 24 insertions, 7 deletions
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationValhall.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationValhall.cpp index b44b297722..951447e1a0 100644 --- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationValhall.cpp +++ b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationValhall.cpp @@ -75,17 +75,26 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi if(m == 1) { - return configure_lhs_rhs_info(m, n, 1, 4, 8, 1, 8, false, true, false, true); + if(n > 2048) + { + return configure_lhs_rhs_info(m, n, 1, 8, 2, 1, 256, false, true, false, true); + } + else + { + return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 256, false, true, false, true); + } } else { - if(m % 5) + if(m > 300) { - return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 8, false, true, false, true); + const int v0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1)); + return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, v0, false, true, false, true); } else { - return configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 8, false, true, false, true); + const int v0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1)); + return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, v0, false, true, false, true); } } } @@ -97,12 +106,20 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi if(m == 1) { - const unsigned int h0 = std::max(n / 2, 1U); - return configure_lhs_rhs_info(m, n, 1, 4, 8, 1, h0, false, true, false, true); + const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1)); + return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, h0, false, true, false, true); } else { - return configure_lhs_rhs_info(m, n, 4, 8, 8, 1, 4, false, true, false, true); + const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1)); + if(k > 512) + { + return configure_lhs_rhs_info(m, n, 2, 4, 16, 1, h0, false, true, false, false); + } + else + { + return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, h0, false, true, false, false); + } } } |