From 27beac711a110f8f5c1e619636f004b3cdee3e54 Mon Sep 17 00:00:00 2001 From: Jiaming Yuan Date: Mon, 26 Feb 2024 00:37:27 +0800 Subject: [PATCH 1/2] [BUG] Fix `MAX_THREADS_PER_SM` on sm 89. --- cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp b/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp index ded753d682..60dbb154a8 100644 --- a/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp +++ b/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp @@ -101,7 +101,7 @@ auto constexpr static const MAX_BLOCKS = index_type{65536}; auto constexpr static const WARP_SIZE = index_type{32}; auto constexpr static const MAX_THREADS_PER_BLOCK = index_type{256}; #ifdef __CUDACC__ -#if __CUDA_ARCH__ == 720 || __CUDA_ARCH__ == 750 || __CUDA_ARCH__ == 860 || __CUDA_ARCH__ == 870 +#if __CUDA_ARCH__ == 720 || __CUDA_ARCH__ == 750 || __CUDA_ARCH__ == 860 || __CUDA_ARCH__ == 870 || __CUDA_ARCH__ == 890 auto constexpr static const MAX_THREADS_PER_SM = index_type{1024}; #else auto constexpr static const MAX_THREADS_PER_SM = index_type{2048}; From bb8b551aa154111bf9d7936d0dda32a7bb23bc94 Mon Sep 17 00:00:00 2001 From: Jiaming Yuan Date: Mon, 26 Feb 2024 00:53:23 +0800 Subject: [PATCH 2/2] lint. --- cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp b/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp index 60dbb154a8..1c5c18e857 100644 --- a/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp +++ b/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp @@ -101,7 +101,8 @@ auto constexpr static const MAX_BLOCKS = index_type{65536}; auto constexpr static const WARP_SIZE = index_type{32}; auto constexpr static const MAX_THREADS_PER_BLOCK = index_type{256}; #ifdef __CUDACC__ -#if __CUDA_ARCH__ == 720 || __CUDA_ARCH__ == 750 || __CUDA_ARCH__ == 860 || __CUDA_ARCH__ == 870 || __CUDA_ARCH__ == 890 +#if __CUDA_ARCH__ == 720 || __CUDA_ARCH__ == 750 || __CUDA_ARCH__ == 860 || \ + __CUDA_ARCH__ == 870 || __CUDA_ARCH__ == 890 auto constexpr static const MAX_THREADS_PER_SM = index_type{1024}; #else auto constexpr static const MAX_THREADS_PER_SM = index_type{2048};