Fix conversion warnings in MSVC

NVIDIA · Aug 16, 2024 · 26a40a2 · 26a40a2
1 parent cc6aae8
commit 26a40a2
Showing 1 changed file with 6 additions and 5 deletions.
diff --git a/cub/cub/device/dispatch/dispatch_transform.cuh b/cub/cub/device/dispatch/dispatch_transform.cuh
@@ -749,7 +749,7 @@ struct dispatch_t<RequiresStableAddress,
     assert(chosen_tile_size > 0);
     assert((sizeof...(RandomAccessIteratorsIn) == 0) != (chosen_smem_size != 0)); // logical xor
 
-    const Offset grid_dim = ::cuda::ceil_div(num_items, Offset{chosen_tile_size});
+    const auto grid_dim = static_cast<unsigned int>(::cuda::ceil_div(num_items, Offset{chosen_tile_size}));
     return ::cuda::std::make_tuple(
       THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(grid_dim, block_dim, chosen_smem_size, stream),
       kernel,
@@ -811,7 +811,7 @@ struct dispatch_t<RequiresStableAddress,
     assert(chosen_tile_size > 0);
     assert((sizeof...(RandomAccessIteratorsIn) == 0) != (chosen_smem_size != 0)); // logical xor
 
-    const Offset grid_dim = ::cuda::ceil_div(num_items, Offset{chosen_tile_size});
+    const auto grid_dim = static_cast<unsigned int>(::cuda::ceil_div(num_items, Offset{chosen_tile_size}));
     return ::cuda::std::make_tuple(
       THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(grid_dim, block_dim, chosen_smem_size, stream),
       kernel,
@@ -864,8 +864,9 @@ struct dispatch_t<RequiresStableAddress,
   CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t invoke_algorithm(
     cuda::std::index_sequence<Is...>, ::cuda::std::integral_constant<Algorithm, Algorithm::unrolled_staged>)
   {
-    using policy_t        = typename ActivePolicy::algo_policy;
-    const Offset grid_dim = ::cuda::ceil_div(num_items, Offset{policy_t::BLOCK_THREADS * policy_t::ITEMS_PER_THREAD});
+    using policy_t      = typename ActivePolicy::algo_policy;
+    const auto grid_dim = static_cast<unsigned int>(
+      ::cuda::ceil_div(num_items, Offset{policy_t::BLOCK_THREADS * policy_t::ITEMS_PER_THREAD}));
     return THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(grid_dim, policy_t::BLOCK_THREADS, 0, stream)
       .doit(kernel,
             num_items,
@@ -896,7 +897,7 @@ struct dispatch_t<RequiresStableAddress,
     const int items_per_thread_clamped =
       ::cuda::std::clamp(items_per_thread, +policy_t::MIN_ITEMS_PER_THREAD, +policy_t::MAX_ITEMS_PER_THREAD);
     const Offset tile_size = block_dim * items_per_thread_clamped;
-    const Offset grid_dim  = ::cuda::ceil_div(num_items, tile_size);
+    const auto grid_dim    = static_cast<unsigned int>(::cuda::ceil_div(num_items, tile_size));
 
     return THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(grid_dim, block_dim, 0, stream)
       .doit(kernel,