Backport fixes for reordering in CUB member initializer lists into 2.…

…0 branch fixes #604
NVIDIA · Apr 20, 2023 · 3f38508 · 3f38508
1 parent 369a2d5
commit 3f38508
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 5 deletions.
diff --git a/cmake/CubBuildCompilerTargets.cmake b/cmake/CubBuildCompilerTargets.cmake
@@ -9,6 +9,7 @@
 function(cub_build_compiler_targets)
   set(cxx_compile_definitions)
   set(cxx_compile_options)
+  set(cuda_compile_options)
 
   if ("MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
     append_option_if_available("/W4" cxx_compile_options)
@@ -38,6 +39,8 @@ function(cub_build_compiler_targets)
     # Some tests require /bigobj to fit everything into their object files:
     append_option_if_available("/bigobj" cxx_compile_options)
   else()
+    append_option_if_available("-Wreorder" cuda_compile_options)
+
     append_option_if_available("-Werror" cxx_compile_options)
     append_option_if_available("-Wall" cxx_compile_options)
     append_option_if_available("-Wextra" cxx_compile_options)
@@ -99,6 +102,12 @@ function(cub_build_compiler_targets)
     ${cxx_compile_definitions}
   )
 
+  foreach (cuda_option IN LISTS cuda_compile_options)
+    target_compile_options(cub.compiler_interface INTERFACE
+      $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:${cuda_option}>
+    )
+  endforeach()
+
   # Promote warnings and display diagnostic numbers for nvcc:
   target_compile_options(cub.compiler_interface INTERFACE
     # If using CUDA w/ NVCC...

diff --git a/cub/block/block_exchange.cuh b/cub/block/block_exchange.cuh
@@ -732,8 +732,8 @@ public:
     :
         temp_storage(PrivateStorage()),
         linear_tid(RowMajorTid(BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z)),
-        warp_id((WARPS == 1) ? 0 : linear_tid / WARP_THREADS),
         lane_id(LaneId()),
+        warp_id((WARPS == 1) ? 0 : linear_tid / WARP_THREADS),
         warp_offset(warp_id * WARP_TIME_SLICED_ITEMS)
     {}
 

diff --git a/cub/device/dispatch/dispatch_radix_sort.cuh b/cub/device/dispatch/dispatch_radix_sort.cuh
@@ -1921,9 +1921,9 @@ struct DispatchSegmentedRadixSort :
         d_end_offsets(d_end_offsets),
         begin_bit(begin_bit),
         end_bit(end_bit),
-        is_overwrite_okay(is_overwrite_okay),
         stream(stream),
-        ptx_version(ptx_version)
+        ptx_version(ptx_version),
+        is_overwrite_okay(is_overwrite_okay)
     {}
 
     CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED
@@ -1954,9 +1954,9 @@ struct DispatchSegmentedRadixSort :
         d_end_offsets(d_end_offsets),
         begin_bit(begin_bit),
         end_bit(end_bit),
-        is_overwrite_okay(is_overwrite_okay),
         stream(stream),
-        ptx_version(ptx_version)
+        ptx_version(ptx_version),
+        is_overwrite_okay(is_overwrite_okay)
     {
       CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
     }