Skip to content

GitLab

Commit cfff66cd authored by Chao Liu's avatar Chao Liu
Browse files

refactor

parent ecd3240b
Hide whitespace changes
Inline Side-by-side
Showing with 33 additions and 34 deletions
driver/include/device_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw.hpp
View file @ cfff66cd
...@@ -3,7 +3,7 @@ ...@@ -3,7 +3,7 @@
#include "device.hpp" #include "device.hpp"
#include "tensor.hpp" #include "tensor.hpp"
#include "gridwise_operation_wrapper.hpp" #include "gridwise_operation_wrapper.hpp"
#include "gridwise_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw_lds_double_buffer.hpp" #include "gridwise_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw.hpp"
template <typename T, template <typename T,
typename InDesc, typename InDesc,
...@@ -84,38 +84,37 @@ void device_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw(InDesc i ...@@ -84,38 +84,37 @@ void device_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw(InDesc i
printf("%s: BlockSize %u, GridSize %u \n", __func__, BlockSize, GridSize); printf("%s: BlockSize %u, GridSize %u \n", __func__, BlockSize, GridSize);
constexpr auto gridwise_conv = constexpr auto gridwise_conv = GridwiseConvolutionBackwardDataImplicitGemm_v1r1_nchw_kcyx_nkhw<
GridwiseConvolutionBackwardDataImplicitGemm_v1r1_nchw_kcyx_nkhw_lds_double_buffer< GridSize,
GridSize, BlockSize,
BlockSize, T,
T, T,
T, decltype(in_nchw_desc),
decltype(in_nchw_desc), decltype(wei_kcyx_desc),
decltype(wei_kcyx_desc), decltype(out_nkhw_desc),
decltype(out_nkhw_desc), ConvStrides,
ConvStrides, ConvDilations,
ConvDilations, LeftPads,
LeftPads, RightPads,
RightPads, BPerBlock,
BPerBlock, EPerBlock,
EPerBlock, KPerBlock,
KPerBlock, GemmMPerThreadSubC,
GemmMPerThreadSubC, GemmNPerThreadSubC,
GemmNPerThreadSubC, GemmMLevel0Cluster,
GemmMLevel0Cluster, GemmNLevel0Cluster,
GemmNLevel0Cluster, GemmMLevel1Cluster,
GemmMLevel1Cluster, GemmNLevel1Cluster,
GemmNLevel1Cluster, GemmKPerThreadLoop,
GemmKPerThreadLoop, GemmDataPerReadA,
GemmDataPerReadA, GemmDataPerReadB,
GemmDataPerReadB, OutBlockCopySubLengths_K_B,
OutBlockCopySubLengths_K_B, OutBlockCopyClusterLengths_K_B,
OutBlockCopyClusterLengths_K_B, OutBlockCopyDataPerAccess_B,
OutBlockCopyDataPerAccess_B, WeiBlockCopySubLengths_K_E,
WeiBlockCopySubLengths_K_E, WeiBlockCopyClusterLengths_K_E,
WeiBlockCopyClusterLengths_K_E, WeiBlockCopyDataPerAccess_E,
WeiBlockCopyDataPerAccess_E, InThreadCopyDataPerAccess_B>{};
InThreadCopyDataPerAccess_B>{};
for(index_t i = 0; i < nrepeat; ++i) for(index_t i = 0; i < nrepeat; ++i)
{ {
......
driver/src/conv_bwd_data_driver.cpp
View file @ cfff66cd
...@@ -345,7 +345,7 @@ int main(int argc, char* argv[]) ...@@ -345,7 +345,7 @@ int main(int argc, char* argv[])
#endif #endif
} }
#if 0 #if 1
device_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw device_convolution_backward_data_implicit_gemm_v1r1_nchw_kcyx_nkhw
#else #else
device_convolution_backward_data_implicit_gemm_v1r2_nchw_kcyx_nkhw device_convolution_backward_data_implicit_gemm_v1r2_nchw_kcyx_nkhw
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment