[BUG]: Race reported in cub::BlockReduceRaking

[alliepiper]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else

Component

CUB

Describe the bug

https://github.com/NVIDIA/cccl/actions/runs/9606424817/job/26497170643?pr=1879

  	compute-sanitizer --tool racecheck --check-device-heap yes --leak-check full --padding 512 --track-stream-ordered-races all --check-warpgroup-mma yes --require-cuda-init no --check-exit-code yes --error-exitcode 1 --nvtx true /home/coder/cccl/build/cuda12.4-gcc13/cub-cpp17/bin/cub.cpp17.test.block_reduce.dimx_65.dimyz_1
  ========= COMPUTE-SANITIZER
  ========= Warning: Race reported between Write access at void cub::CUB_200500_600_700_800_NS::ThreadStoreVolatilePtr<unsigned char>(T1 *, T1, cub::CUB_200500_600_700_800_NS::Int2Type<(int)1>)+0x250 in /home/coder/cccl/cub/cub/thread/thread_store.cuh:284
  =========     and Read access at unsigned char cub::CUB_200500_600_700_800_NS::BlockReduceRaking<unsigned char, (int)65, (int)1, (int)1, (int)0>::Reduce<(bool)1, cuda::std::__4::plus<void>>(unsigned char, int, T2)+0x1b0 in /home/coder/cccl/cub/cub/block/specializations/block_reduce_raking.cuh:222 [7 hazards]
  =========     and Read access at unsigned char cub::CUB_200500_600_700_800_NS::BlockReduceRaking<unsigned char, (int)65, (int)1, (int)1, (int)0>::RakingReduction<(bool)1, cuda::std::__4::plus<void>, (int)1>(T2, unsigned char *, unsigned char, int, cub::CUB_200500_600_700_800_NS::Int2Type<T3>)+0x1c0 in /home/coder/cccl/cub/cub/block/specializations/block_reduce_raking.cuh:161 [7 hazards]
  =========     and Read access at unsigned char cub::CUB_200500_600_700_800_NS::BlockReduceRaking<unsigned char, (int)65, (int)1, (int)1, (int)0>::RakingReduction<(bool)1, cuda::std::__4::plus<void>, (int)1>(T2, unsigned char *, unsigned char, int, cub::CUB_200500_600_700_800_NS::Int2Type<T3>)+0x1d0 in /home/coder/cccl/cub/cub/block/specializations/block_reduce_raking.cuh:161 [7 hazards]

How to Reproduce

compute-sanitizer --tool racecheck --check-device-heap yes --leak-check full --padding 512 --track-stream-ordered-races all --check-warpgroup-mma yes --require-cuda-init no --check-exit-code yes --error-exitcode 1 --nvtx true /home/coder/cccl/build/cuda12.4-gcc13/cub-cpp17/bin/cub.cpp17.test.block_reduce.dimx_65.dimyz_1

Expected behavior

No diagnostics emitted.

Reproduction link

No response

Operating System

No response

nvidia-smi output

No response

NVCC version

No response