Unify kernel dispatch paths for device reduce between CUB and c.paral…

…lel. (NVIDIA#2591)
pciolkosz · Oct 25, 2024 · 6ba0c8f · 6ba0c8f
1 parent 3848b74
commit 6ba0c8f
Show file tree

Hide file tree

Showing 11 changed files with 488 additions and 301 deletions.
diff --git a/c/parallel/CMakeLists.txt b/c/parallel/CMakeLists.txt
@@ -37,6 +37,8 @@ target_link_libraries(cccl.c.parallel PRIVATE
   CUDA::nvJitLink
   CUDA::cuda_driver
   cccl.compiler_interface_cpp20
+  CUB::CUB
+  Thrust::Thrust
 )
 target_compile_definitions(cccl.c.parallel PUBLIC CCCL_C_EXPERIMENTAL=1)
 target_compile_definitions(cccl.c.parallel PRIVATE NVRTC_GET_TYPE_NAME=1)

diff --git a/c/parallel/include/cccl/c/reduce.h b/c/parallel/include/cccl/c/reduce.h
@@ -24,6 +24,7 @@ struct cccl_device_reduce_build_result_t
   void* cubin;
   size_t cubin_size;
   CUlibrary library;
+  unsigned long long accumulator_size;
   CUkernel single_tile_kernel;
   CUkernel single_tile_second_kernel;
   CUkernel reduction_kernel;