mirror of
https://github.com/leejet/stable-diffusion.cpp.git
synced 2026-11-11 11:26:34 +00:00
perf: accelerate packed INT4 CUDA inference with CUTLASS (#2278)
This commit is contained in:
1 parent
7952693004
commit
f5df5f8b84
6 files changed
+247
-26
No files matched your search
@@ -27,6 +27,12 @@ if(SD_CUDA)
|
||||
set_source_files_properties(sage_attention.cu sol_attention.cu PROPERTIES COMPILE_OPTIONS "--use_fast_math")
|
||||
set_source_files_properties(${SD_CUDA_SOURCES} PROPERTIES COMPILE_OPTIONS "--use_fast_math")
|
||||
if(SD_CUDA_CUTLASS_PATH)
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/w4_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_W4_CUTLASS)
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/w4_cuda.cu" APPEND PROPERTY INCLUDE_DIRECTORIES "${SD_CUDA_CUTLASS_PATH}/include")
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/w4_cuda.cu" APPEND PROPERTY COMPILE_OPTIONS --expt-relaxed-constexpr)
|
||||
if(MSVC)
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/w4_cuda.cu" APPEND PROPERTY COMPILE_OPTIONS -Xcompiler=/Zc:preprocessor,/Zc:__cplusplus)
|
||||
endif()
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/bf16_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_BF16_CUTLASS)
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/int8_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_I8_CUTLASS)
|
||||
set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/registration_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_CUTE_ATTENTION)
|
||||
|
||||
Reference in new issue
Block a user