Superminaren/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-04 02:37:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

metal : add metallib build support for xcframework (#28163)

Browse Source
This commit is contained in:
Jhen-Jie Hong
2026-09-02 07:45:56 +08:00
committed by GitHub
parent 3466812d1f
commit b96806d960
3 changed files with 66 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/CMakeLists.txt
+2
View File
@@ -242,6 +242,8 @@ option(GGML_METAL_EMBED_LIBRARY "ggml: embed Metal library"
set (GGML_METAL_MACOSX_VERSION_MIN "" CACHE STRING
"ggml: metal minimum macOS version")
set (GGML_METAL_STD "" CACHE STRING "ggml: metal standard version (-std flag)")
set (GGML_METAL_TARGET_OS "macos" CACHE STRING
"ggml: metal -mtargetos OS name (macos, ios, xros, tvos)")
option(GGML_OPENMP "ggml: use OpenMP" ON)
option(GGML_OPENMP_FETCH "ggml: fetch LLVM OpenMP" OFF)
option(GGML_RPC "ggml: use RPC" OFF)
ggml/src/ggml-metal/CMakeLists.txt
+49 -21
View File
@@ -127,6 +127,18 @@ else()
configure_file(${src} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${src} COPYONLY)
endforeach()
# CMAKE_OSX_SYSROOT is an SDK name or path - xcrun accepts both
set(METAL_SDK ${CMAKE_OSX_SYSROOT})
if (NOT METAL_SDK)
set(METAL_SDK macosx)
endif()
if (CMAKE_OSX_SYSROOT MATCHES "[Ss]imulator")
set(METAL_TARGET_SIM "-simulator")
else()
set(METAL_TARGET_SIM "")
endif()
if (GGML_METAL_SHADER_DEBUG)
# note: disabling fast math is needed in order to pass tests/test-backend-ops
# note: adding -fno-inline fixes the tests when using MTL_SHADER_VALIDATION=1
@@ -138,9 +150,19 @@ else()
set(XC_FLAGS -O3)
endif()
execute_process(COMMAND xcrun -sdk ${METAL_SDK} --show-sdk-version OUTPUT_VARIABLE METAL_SDK_VERSION OUTPUT_STRIP_TRAILING_WHITESPACE)
if (METAL_SDK_VERSION VERSION_GREATER_EQUAL 26.0)
set(GGML_METAL_HAS_TENSOR_LIB ON)
else()
message(STATUS "Metal SDK ${METAL_SDK_VERSION} does not support the tensor API, skipping ggml-tensor.metallib")
endif()
if (GGML_METAL_MACOSX_VERSION_MIN)
message(STATUS "Adding -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN} flag to metal compilation")
list (APPEND XC_FLAGS -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN})
elseif (NOT GGML_METAL_TARGET_OS STREQUAL "macos" AND CMAKE_OSX_DEPLOYMENT_TARGET)
message(STATUS "Adding -mtargetos=${GGML_METAL_TARGET_OS}${CMAKE_OSX_DEPLOYMENT_TARGET}${METAL_TARGET_SIM} flag to metal compilation")
list (APPEND XC_FLAGS -mtargetos=${GGML_METAL_TARGET_OS}${CMAKE_OSX_DEPLOYMENT_TARGET}${METAL_TARGET_SIM})
endif()
if (GGML_METAL_STD)
@@ -156,33 +178,41 @@ else()
list(APPEND AIR_FILES ${AIR})
add_custom_command(
OUTPUT ${AIR}
COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -I ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${src} -o ${AIR}
COMMAND xcrun -sdk ${METAL_SDK} metal ${XC_FLAGS} -I ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${src} -o ${AIR}
DEPENDS ${src} kernels/common.h kernels/dequantize.h kernels/quantize.h ${METALLIB_COMMON} ggml-metal-impl.h
COMMENT "Compiling ${src}"
VERBATIM
)
endforeach()
# the tensor API kernels go in a separate metallib, loaded only where supported
set(AIR_MM_TENSOR "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/mul_mm_tensor.air")
add_custom_command(
OUTPUT ${AIR_MM_TENSOR}
COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -DGGML_METAL_HAS_TENSOR -I ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels/mul_mm.metal -o ${AIR_MM_TENSOR}
DEPENDS kernels/mul_mm.metal kernels/common.h kernels/dequantize.h ${METALLIB_COMMON} ggml-metal-impl.h
COMMENT "Compiling kernels/mul_mm.metal (tensor API)"
VERBATIM
)
set(METALLIB_FILES ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib)
add_custom_command(
OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
COMMAND xcrun -sdk macosx metallib ${AIR_MM_TENSOR} -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
DEPENDS ${AIR_MM_TENSOR}
COMMENT "Linking tensor API Metal kernels into ggml-tensor.metallib"
)
# the tensor API kernels go in a separate metallib, loaded only where supported
if (GGML_METAL_HAS_TENSOR_LIB)
set(AIR_MM_TENSOR "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/mul_mm_tensor.air")
# the tensor API needs OS 26+
set(XC_FLAGS_TENSOR ${XC_FLAGS} -mtargetos=${GGML_METAL_TARGET_OS}26.0${METAL_TARGET_SIM})
add_custom_command(
OUTPUT ${AIR_MM_TENSOR}
COMMAND xcrun -sdk ${METAL_SDK} metal ${XC_FLAGS_TENSOR} -DGGML_METAL_HAS_TENSOR -I ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels/mul_mm.metal -o ${AIR_MM_TENSOR}
DEPENDS kernels/mul_mm.metal kernels/common.h kernels/dequantize.h ${METALLIB_COMMON} ggml-metal-impl.h
COMMENT "Compiling kernels/mul_mm.metal (tensor API)"
VERBATIM
)
add_custom_command(
OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
COMMAND xcrun -sdk ${METAL_SDK} metallib ${AIR_MM_TENSOR} -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
DEPENDS ${AIR_MM_TENSOR}
COMMENT "Linking tensor API Metal kernels into ggml-tensor.metallib"
)
list(APPEND METALLIB_FILES ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib)
endif()
add_custom_command(
OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
COMMAND xcrun -sdk macosx metallib ${AIR_FILES} -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
COMMAND xcrun -sdk ${METAL_SDK} metallib ${AIR_FILES} -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal-impl.h
COMMAND rm -rf ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels
@@ -192,8 +222,7 @@ else()
add_custom_target(
ggml-metal-lib ALL
DEPENDS ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
DEPENDS ${METALLIB_FILES}
)
endif() # GGML_METAL_EMBED_LIBRARY
@@ -205,8 +234,7 @@ if (NOT GGML_METAL_EMBED_LIBRARY)
)
install(
FILES ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-tensor.metallib
FILES ${METALLIB_FILES}
DESTINATION ${CMAKE_INSTALL_BINDIR}
)
endif()