fix performance issuse

yadaish · Bernard-Liu · commit bca8c2735266 · 2025-11-11T06:47:52.000Z
diff --git a/fbgemm_gpu/codegen/training/backward/embedding_backward_split_template.cu b/fbgemm_gpu/codegen/training/backward/embedding_backward_split_template.cu
@@ -1425,6 +1425,7 @@ Tensor {{ embedding_cuda_op }}(
                                 32,
                                 false>;
                             blockSize = dim3(32, num_warp_per_row_groups);
+                            // printf("%s:%d warp kernel %d\n", __FILE__, __LINE__, num_warp_per_row_groups);
                         }
                     }
                     {%- endif %}

Original file line number	Diff line number	Diff line change
`@@ -1425,6 +1425,7 @@ Tensor {{ embedding_cuda_op }}(`
`1425`	`1425`	`32,`
`1426`	`1426`	`false>;`
`1427`	`1427`	`blockSize = dim3(32, num_warp_per_row_groups);`
	`1428`	`+ // printf("%s:%d warp kernel %d\n", __FILE__, __LINE__, num_warp_per_row_groups);`
`1428`	`1429`	`}`
`1429`	`1430`	`}`
`1430`	`1431`	`{%- endif %}`