refactor: move model-specific operators and fusions into backend extensions (#2253)

This commit is contained in:
leejet authored and GitHub committed 2026-11-01 21:18:37 +08:00
1 parent c6dfa9def1
commit 45a63ac460
68 files changed
+12173 -151

No files matched your search

+42
View File
@@ -225,6 +225,8 @@ file(GLOB SD_LIB_SOURCES CONFIGURE_DEPENDS
"src/core/*.h"
"src/core/*.cpp"
"src/core/*.hpp"
"src/core/cpu/*.h"
"src/core/cpu/*.cpp"
"src/core/probe/*.h"
"src/core/probe/*.cpp"
"src/extensions/*.h"
@@ -343,6 +345,46 @@ if (SD_CUDA)
# Keep the driver stub on downstream link lines when no driver is installed.
target_link_libraries(${SD_LIB} PUBLIC CUDA::cuda_driver)
set_property(SOURCE src/core/ggml_extend_backend.cpp APPEND PROPERTY COMPILE_DEFINITIONS SD_USE_CUDA)
if (NOT SD_USE_UPSTREAM_GGML)
include(cmake/cutlass.cmake)
enable_language(CUDA)
set_target_properties(${SD_LIB} PROPERTIES CUDA_STANDARD 17 CUDA_STANDARD_REQUIRED ON)
if (SD_CUDA_CUTLASS_PATH)
target_sources(${SD_LIB} PRIVATE src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu)
set_source_files_properties(src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu PROPERTIES
INCLUDE_DIRECTORIES "${SD_CUDA_CUTLASS_PATH}/include"
COMPILE_OPTIONS "--use_fast_math;--extended-lambda;--expt-relaxed-constexpr")
if (MSVC)
set_property(SOURCE src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu APPEND PROPERTY COMPILE_OPTIONS -Xcompiler=/Zc:preprocessor,/Zc:__cplusplus)
endif()
if (GGML_CUDA_FORCE_CUBLAS)
set_property(SOURCE src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_FORCE_CUBLAS)
endif()
endif()
endif()
endif()
set(SD_BACKEND_REGISTRATIONS)
if (NOT SD_USE_UPSTREAM_GGML AND (SD_CUDA OR SD_HIPBLAS OR SD_MUSA))
if(SD_HIPBLAS)
include(cmake/hip.cmake)
endif()
add_subdirectory(src/core/cuda)
foreach(VENDOR CUDA HIP MUSA)
if(SD_${VENDOR} OR (VENDOR STREQUAL "HIP" AND SD_HIPBLAS))
string(TOLOWER "${VENDOR}" NAMESPACE)
list(APPEND SD_BACKEND_REGISTRATIONS "sd_${NAMESPACE}::register_extensions")
endif()
endforeach()
endif()
if (SD_VULKAN AND NOT SD_USE_UPSTREAM_GGML)
add_subdirectory(src/core/vulkan)
list(APPEND SD_BACKEND_REGISTRATIONS sd_ggml_register_vulkan_rope sd_ggml_register_vulkan_i8)
endif()
if(SD_BACKEND_REGISTRATIONS)
string(REPLACE ";" ",\n " SD_BACKEND_REGISTRATIONS "${SD_BACKEND_REGISTRATIONS}")
configure_file(cmake/backend_extensions.h.in generated/sd_backend_extensions.h @ONLY)
set_property(SOURCE src/core/ggml_extensions.cpp APPEND PROPERTY INCLUDE_DIRECTORIES "${CMAKE_CURRENT_BINARY_DIR}/generated")
set_property(SOURCE src/core/ggml_extensions.cpp APPEND PROPERTY COMPILE_DEFINITIONS SD_BACKEND_EXTENSIONS)
endif()
if (SD_HIPBLAS)
find_package(hip REQUIRED CONFIG HINTS "$ENV{ROCM_PATH}" /opt/rocm)