ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 01:36:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Gemm+Bilinear (#316)

Browse Source 
* refactor

* update example

* update example

* gemm bilinear

* clean

* update
This commit is contained in:
Chao Liu
2022-07-02 09:15:38 -05:00
committed by GitHub
parent 8e374781d5
commit 9e4429f9c3
75 changed files with 1485 additions and 4658 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
profiler/src/profiler.cpp
View File

@@ -5,12 +5,10 @@
int profile_gemm(int, char*[]);
int profile_gemm_splitk(int, char*[]);
int profile_gemm_bias_2d(int, char*[]);
int profile_gemm_bias_relu(int, char*[]);
int profile_gemm_bias_relu_add(int, char*[]);
int profile_gemm_bias_add_reduce(int, char*[]);
int profile_gemm_bilinear(int, char*[]);
int profile_gemm_add_add_fastgelu(int, char*[]);
int profile_gemm_reduce(int, char*[]);
int profile_gemm_bias_add_reduce(int, char*[]);
int profile_batched_gemm(int, char*[]);
int profile_batched_gemm_reduce(int, char*[]);
int profile_grouped_gemm(int, char*[]);
@@ -28,12 +26,12 @@ static void print_helper_message()
// clang-format off
printf("arg1: tensor operation (gemm: GEMM\n"
" gemm_splitk: Split-K GEMM\n"
" gemm_bias_2d: GEMM+Bias(2D)\n"
" gemm_bias_relu: GEMM+Bias+ReLU\n"
" gemm_bias_relu_add: GEMM+Bias+ReLU+Add\n"
" gemm_bilinear: GEMM+Bilinear\n"
" gemm_add_add_fastgelu: GEMM+Add+Add+FastGeLU\n"
" gemm_reduce: GEMM+Reduce\n"
" gemm_bias_add_reduce: GEMM+Bias+Add+Reduce\n"
" batched_gemm: Batched GEMM\n"
" batched_gemm_reduce: Batched GEMM+Reduce\n"
" grouped_gemm: Grouped GEMM\n"
" conv_fwd: ForwardConvolution\n"
" conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n"
@@ -63,17 +61,13 @@ int main(int argc, char* argv[])
{
return profile_gemm_splitk(argc, argv);
}
else if(strcmp(argv[1], "gemm_bias_2d") == 0)
else if(strcmp(argv[1], "gemm_bilinear") == 0)
{
return profile_gemm_bias_2d(argc, argv);
return profile_gemm_bilinear(argc, argv);
}
else if(strcmp(argv[1], "gemm_bias_relu") == 0)
else if(strcmp(argv[1], "gemm_add_add_fastgelu") == 0)
{
return profile_gemm_bias_relu(argc, argv);
}
else if(strcmp(argv[1], "gemm_bias_relu_add") == 0)
{
return profile_gemm_bias_relu_add(argc, argv);
return profile_gemm_add_add_fastgelu(argc, argv);
}
else if(strcmp(argv[1], "gemm_reduce") == 0)
{
@@ -119,17 +113,13 @@ int main(int argc, char* argv[])
{
return profile_convnd_bwd_data(argc, argv, 3);
}
else if(strcmp(argv[1], "reduce") == 0)
{
return profile_reduce(argc, argv);
}
else if(strcmp(argv[1], "conv2d_bwd_weight") == 0)
{
return profile_conv_bwd_weight(argc, argv);
}
else if(strcmp(argv[1], "gemm_add_add_fastgelu") == 0)
else if(strcmp(argv[1], "reduce") == 0)
{
return profile_gemm_add_add_fastgelu(argc, argv);
return profile_reduce(argc, argv);
}
else if(strcmp(argv[1], "batchnorm") == 0 || strcmp(argv[1], "layernorm") == 0 ||
strcmp(argv[1], "softmax") == 0)