zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

rebase + nit (#2260)

Browse Source 
Co-authored-by: Awni Hannun <awni@apple.com>
This commit is contained in:
Cheng
2025-06-11 02:51:51 +09:00
committed by GitHub
parent 62fecf3e13
commit 99c33d011d
10 changed files with 604 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
mlx/backend/cuda/CMakeLists.txt
View File

@@ -7,7 +7,11 @@ target_sources(
mlx
PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/allocator.cpp
${CMAKE_CURRENT_SOURCE_DIR}/binary.cu
${CMAKE_CURRENT_SOURCE_DIR}/copy.cpp
${CMAKE_CURRENT_SOURCE_DIR}/copy.cu
${CMAKE_CURRENT_SOURCE_DIR}/copy/copy_contiguous.cu
${CMAKE_CURRENT_SOURCE_DIR}/copy/copy_general.cu
${CMAKE_CURRENT_SOURCE_DIR}/copy/copy_general_dynamic.cu
${CMAKE_CURRENT_SOURCE_DIR}/copy/copy_general_input.cu
${CMAKE_CURRENT_SOURCE_DIR}/device.cpp
${CMAKE_CURRENT_SOURCE_DIR}/eval.cpp
${CMAKE_CURRENT_SOURCE_DIR}/event.cu
@@ -28,6 +32,15 @@ target_compile_definitions(mlx PRIVATE MLX_USE_CUDA)
target_compile_options(mlx
PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>")
# CUDA 12.8 emits warning #20280-D for copy kernels which is a false positive.
# Explicitly pass this flag to suppress the warning, it is safe to set it to
# true but the warning wouldn't be suppressed.
if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)
target_compile_options(
mlx
PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:--static-global-template-stub=false>")
endif()
# Compute capability 7 is required for synchronization between CPU/GPU with
# managed memory. TODO: Add more architectures for potential performance gain.
set(MLX_CUDA_ARCHITECTURES