ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-18 09:38:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Small script improvements.

Browse Source
This commit is contained in:
Ville Pietilä
2025-10-21 06:38:49 +00:00
parent 2efd174b33
commit 159bcf6750
2 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp
View File

@@ -37,7 +37,7 @@ using tile_grouped_conv_bwd_weight_bf16_instances_opt = std::tuple<
GroupedConvolutionBackwardWeightInvoker<NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough, 2, 128, 128, 64, 2, 2, 1, 32, 32, 16, 8, 8, 8, false, CK_TILE_PIPELINE_COMPUTE_V3>,
GroupedConvolutionBackwardWeightInvoker<NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough, 2, 128, 128, 64, 2, 2, 1, 32, 32, 16, 8, 8, 8, true, CK_TILE_PIPELINE_COMPUTE_V4>,
// Skinny kernels.
// Skinny GEMM-K kernels.
GroupedConvolutionBackwardWeightInvoker<NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough, 2, 128, 128, 16, 2, 2, 1, 32, 32, 16, 8, 8, 4, false, CK_TILE_PIPELINE_COMPUTE_V3>,
GroupedConvolutionBackwardWeightInvoker<NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough, 2, 128, 128, 16, 2, 2, 1, 32, 32, 16, 8, 8, 4, true, CK_TILE_PIPELINE_COMPUTE_V4>,
GroupedConvolutionBackwardWeightInvoker<NDimSpatial, ALayout, BLayout, ELayout, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough, 1, 128, 128, 16, 2, 2, 1, 32, 32, 16, 8, 8, 4, false, CK_TILE_PIPELINE_COMPUTE_V3>,

3
script/benchmark_ck_vs_ck_tile.py
View File

@@ -234,6 +234,9 @@ def main():
profiler_commands = get_profiler_commands(args.input_file)
print(f"Got {len(profiler_commands)} unique commands to run.")
if not os.path.exists(args.results_path):
os.makedirs(args.results_path)
results_file = os.path.join(args.results_path, f"ck_vs_ck_tile_results_{os.getpid()}.txt")
for i, cmd in enumerate(profiler_commands):