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 1 Packages Projects Releases Wiki Activity

[CUDA] Always use batched matmul (#2404)

Browse Source 
* cuda batched mm

* addmm as well

* comment
This commit is contained in:
Awni Hannun
2025-07-24 20:46:02 -07:00
committed by GitHub
parent 7d9d6ef456
commit 9acec364c2
8 changed files with 694 additions and 322 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
mlx/backend/cuda/CMakeLists.txt
View File

@@ -21,7 +21,8 @@ target_sources(
${CMAKE_CURRENT_SOURCE_DIR}/eval.cpp
${CMAKE_CURRENT_SOURCE_DIR}/event.cu
${CMAKE_CURRENT_SOURCE_DIR}/fence.cpp
${CMAKE_CURRENT_SOURCE_DIR}/gemv.cu
${CMAKE_CURRENT_SOURCE_DIR}/gemms/gemv.cu
${CMAKE_CURRENT_SOURCE_DIR}/gemms/cublas_gemm.cpp
${CMAKE_CURRENT_SOURCE_DIR}/jit_module.cpp
${CMAKE_CURRENT_SOURCE_DIR}/indexing.cpp
${CMAKE_CURRENT_SOURCE_DIR}/kernel_utils.cu
@@ -47,6 +48,14 @@ target_sources(
${CMAKE_CURRENT_SOURCE_DIR}/quantized.cu
${CMAKE_CURRENT_SOURCE_DIR}/worker.cpp)
if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.9.0)
target_sources(
mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/gemms/cublas_batched_gemm_12_9.cu)
else()
target_sources(
mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/gemms/cublas_batched_gemm_12_0.cpp)
endif()
target_compile_definitions(mlx PRIVATE MLX_USE_CUDA)
# Embed kernel sources in binary for JIT compilation.