From c14f1cd2c7667304f65822e219d12edfcacffd9b Mon Sep 17 00:00:00 2001 From: Benson Ma Date: Fri, 24 Jul 2026 11:02:38 -0700 Subject: [PATCH] Drop redundant uint32_t narrowing at cap_grid_dim_x in repeat_arange/get_source_mask Summary: cap_grid_dim_x takes int64_t; the static_cast(batch_size) pre-narrowing is redundant. Drops it in repeat_arange and get_source_mask. Reviewed By: cthi Differential Revision: D112381247 --- fbgemm_gpu/src/jagged_tensor_ops/get_source_mask.cu | 4 +--- fbgemm_gpu/src/jagged_tensor_ops/repeat_arange.cu | 4 +--- 2 files changed, 2 insertions(+), 6 deletions(-) diff --git a/fbgemm_gpu/src/jagged_tensor_ops/get_source_mask.cu b/fbgemm_gpu/src/jagged_tensor_ops/get_source_mask.cu index f3a6ab1a39..a205e1d2fe 100644 --- a/fbgemm_gpu/src/jagged_tensor_ops/get_source_mask.cu +++ b/fbgemm_gpu/src/jagged_tensor_ops/get_source_mask.cu @@ -86,9 +86,7 @@ Tensor get_source_mask_cuda( // correctness-preserving. // See: https://github.com/ROCm/hip/issues/2253 const auto num_blocks = utils::cuda::cap_grid_dim_x( - static_cast(batch_size), - kMaxThreads, - at::cuda::getCurrentCUDAStream()); + batch_size, kMaxThreads, at::cuda::getCurrentCUDAStream()); AT_DISPATCH_INDEX_TYPES( num_sources.scalar_type(), "get_source_mask_kernel", ([&] { diff --git a/fbgemm_gpu/src/jagged_tensor_ops/repeat_arange.cu b/fbgemm_gpu/src/jagged_tensor_ops/repeat_arange.cu index 2bd9bb4bf0..63dbe859d5 100644 --- a/fbgemm_gpu/src/jagged_tensor_ops/repeat_arange.cu +++ b/fbgemm_gpu/src/jagged_tensor_ops/repeat_arange.cu @@ -109,9 +109,7 @@ Tensor repeat_arange_cuda(const Tensor& lengths) { // correctness-preserving. // See: https://github.com/ROCm/hip/issues/2253 const auto num_blocks = utils::cuda::cap_grid_dim_x( - static_cast(batch_size), - kMaxThreads, - at::cuda::getCurrentCUDAStream()); + batch_size, kMaxThreads, at::cuda::getCurrentCUDAStream()); AT_DISPATCH_INDEX_TYPES( lengths.scalar_type(), "repeat_arange_kernel", ([&] {