ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-05 20:55:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Cosmetics

Browse Source
This commit is contained in:
myamlak
2022-05-16 11:18:28 +00:00
parent 55927aaf7e
commit ffe12e2ea6
1 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
include/ck/tensor_operation/gpu/device/device_cgemm_4gemm_xdl_cshuffle.hpp
View File

@@ -510,7 +510,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_real - aux here
// c_real = c_real - aux needed here!!!
launch_kernel(kernel,
dim3(grid_size),
@@ -542,7 +542,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_imag + aux here
// c_imag = c_imag + aux needed here!!!
}
else
{
@@ -580,7 +580,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_real - aux here
// // c_real = c_real - aux needed here!!!
ave_time +=
launch_and_time_kernel(kernel,
@@ -616,7 +616,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_imag + aux here
// c_imag = c_imag + aux needed here!!!
}
}
else
@@ -666,7 +666,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_real - aux here
// // c_real = c_real - aux needed here!!!
launch_kernel(kernel,
dim3(grid_size),
@@ -698,7 +698,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_imag + aux here
// c_imag = c_imag + aux needed here!!!
}
else
{
@@ -736,7 +736,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_real - aux here
// c_real = c_real - aux needed here!!!
ave_time +=
launch_and_time_kernel(kernel,
@@ -772,7 +772,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
arg.block_2_ctile_map_);
// c_imag + aux here
// c_imag = c_imag + aux needed here!!!
}
}