Reuse cuda_calc_xblock_count and remove duplicated copy (pytorch#2607)

Summary: Pull Request resolved: pytorch#2607 Just reuse cuda_calc_xblock_count and remove duplicated copy. Reviewed By: jiawenliu64 Differential Revision: D57533730 fbshipit-source-id: 659a6a77fadbbef0becc5363cef10ff6e72a6d12
q10 · May 22, 2024 · 8f19539 · 8f19539
1 parent d7a5500
commit 8f19539
Show file tree

Hide file tree

Showing 2 changed files with 1 addition and 151 deletions.
diff --git a/fbgemm_gpu/experimental/gen_ai/src/quantize/cuda_dispatch_utils.h b/fbgemm_gpu/experimental/gen_ai/src/quantize/cuda_dispatch_utils.h
diff --git a/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cu b/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cu
@@ -16,6 +16,7 @@
 #include <ATen/cuda/Atomic.cuh>
 #include <algorithm>
 
+#include <fbgemm_gpu/sparse_ops_utils.h>
 #include "fbgemm_gpu/fbgemm_cuda_utils.cuh"
 
 #if !(                                                  \
@@ -33,7 +34,6 @@
 #include <mma.h>
 #endif
 #include <cub/cub.cuh>
-#include "cuda_dispatch_utils.h"
 
 #include <torch/torch.h>