ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 17:26:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

clean

Browse Source
This commit is contained in:
Chao Liu
2019-08-23 09:59:23 -05:00
parent 6ff3fe5d05
commit cb6475c77d
2 changed files with 5 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
composable_kernel/include/kernel_algorithm/gridwise_convolution_implicit_gemm_v1r3_chwn_cyxk_khwn_padded.hpp
View File

@@ -138,7 +138,7 @@ struct GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_padded
constexpr auto out_k_h_w_n_thread_desc = make_ConstantTensorDescriptor_packed(
Sequence<KPerThread, HoPerThread, WoPerThread, NPerThread>{});
#if 0
#if 1
// blockwise input copy
// format is [C, Hi, Wi, N]
auto blockwise_in_copy =
@@ -180,7 +180,7 @@ struct GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_padded
{0, 0, 0, 0});
#endif
#if 0
#if 1
// blockwise wei copy
// format is [CPerBlock, KPerBlock]
const auto blockwise_wei_copy =
@@ -278,7 +278,7 @@ struct GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_padded
// set threadwise output tensor to 0
threadwise_matrix_set_zero(c_k_wn_thread_mtx_desc, p_out_thread);
#if 0
#if 1
for(index_t y = 0; y < Y; ++y)
{
for(index_t x = 0; x < X; ++x)
@@ -318,10 +318,8 @@ struct GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_padded
for(index_t c_block_data_begin = 0; c_block_data_begin < C;
c_block_data_begin += CPerBlock)
{
#if 1 // debug
blockwise_in_copy.Run();
blockwise_wei_copy.Run();
#endif
__syncthreads();

4
driver/src/driver.cpp
View File

@@ -74,7 +74,7 @@ int main(int argc, char* argv[])
#if 1
constexpr index_t N = 64;
constexpr index_t C = 8;
constexpr index_t C = 1536;
constexpr index_t HI = 8;
constexpr index_t WI = 8;
constexpr index_t K = 256;
@@ -368,7 +368,7 @@ int main(int argc, char* argv[])
#if 0
device_convolution_direct_v2_nchw_kcyx_nkhw
(in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
#elif 0
#elif 1
device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
#elif 1