biondizzle/DeepGEMM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove comments

Browse Source
This commit is contained in:
Chenggang Zhao
2025-08-22 17:32:04 +08:00
parent f20256fd50
commit e38c2e3103
1 changed files with 0 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
deep_gemm/include/deep_gemm/impls/sm100_fp8_gemm_1d2d.cuh
View File

@@ -513,13 +513,11 @@ sm100_fp8_gemm_1d2d_impl(float* sfb, int* grouped_layout,
} }
// Flush all stages in the pipeline to make TMA stores visible to the next kernel // Flush all stages in the pipeline to make TMA stores visible to the next kernel
// TODO: do we actually need this?
if (epilogue_thread_idx_in_warpgroup == 0) if (epilogue_thread_idx_in_warpgroup == 0)
cute::tma_store_wait<0>(); cute::tma_store_wait<0>();
// Deallocate tensor memory by warp 1 // Deallocate tensor memory by warp 1
// NOTES: warp 0 is waiting TMA store // NOTES: warp 0 is waiting TMA store
// TODO: do we need 2 SM allocation?
if (epilogue_warp_idx == 1) if (epilogue_warp_idx == 1)
Allocator().free(0, kNumTmemCols); Allocator().free(0, kNumTmemCols);
} }