Properly define block size for > sm_100

rapidsai · Jan 23, 2025 · 37c5a64 · 37c5a64
1 parent bd7c69e
commit 37c5a64
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp b/cpp/include/cuml/experimental/fil/detail/gpu_introspection.hpp
@@ -104,7 +104,7 @@ 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
+  __CUDA_ARCH__ == 870 || __CUDA_ARCH__ == 890 || __CUDA_ARCH__ == 1200
 auto constexpr static const MAX_THREADS_PER_SM = index_type{1024};
 #else
 auto constexpr static const MAX_THREADS_PER_SM = index_type{2048};