mirror of
https://github.com/ml-explore/mlx.git
synced 2025-06-24 09:21:16 +08:00
27d70c7d9d
* feature complete metal fft * fix contiguity bug * jit fft * simplify rader/bluestein constant computation * remove kernel/utils.h dep * remove bf16.h dep * format --------- Co-authored-by: Alex Barron <abarron22@apple.com>
157 lines
3.2 KiB
CMake
157 lines
3.2 KiB
CMake
set(
|
|
HEADERS
|
|
bf16.h
|
|
bf16_math.h
|
|
complex.h
|
|
defines.h
|
|
utils.h
|
|
steel/conv/params.h
|
|
)
|
|
|
|
set(
|
|
KERNELS
|
|
"arg_reduce"
|
|
"conv"
|
|
"fft"
|
|
"gemv"
|
|
"quantized"
|
|
"random"
|
|
"rms_norm"
|
|
"layer_norm"
|
|
"rope"
|
|
"scaled_dot_product_attention"
|
|
)
|
|
|
|
if (NOT MLX_METAL_JIT)
|
|
set(
|
|
KERNELS
|
|
${KERNELS}
|
|
"arange"
|
|
"binary"
|
|
"binary_two"
|
|
"unary"
|
|
"ternary"
|
|
"copy"
|
|
"softmax"
|
|
"sort"
|
|
"scan"
|
|
"reduce"
|
|
)
|
|
set(
|
|
HEADERS
|
|
${HEADERS}
|
|
atomic.h
|
|
arange.h
|
|
unary_ops.h
|
|
unary.h
|
|
binary_ops.h
|
|
binary.h
|
|
ternary.h
|
|
copy.h
|
|
fft.h
|
|
fft/radix.h
|
|
fft/readwrite.h
|
|
softmax.h
|
|
sort.h
|
|
scan.h
|
|
reduction/ops.h
|
|
reduction/reduce_init.h
|
|
reduction/reduce_all.h
|
|
reduction/reduce_col.h
|
|
reduction/reduce_row.h
|
|
)
|
|
endif()
|
|
|
|
function(build_kernel_base TARGET SRCFILE DEPS)
|
|
set(METAL_FLAGS -Wall -Wextra -fno-fast-math -D${MLX_METAL_VERSION})
|
|
if(MLX_METAL_DEBUG)
|
|
set(METAL_FLAGS ${METAL_FLAGS}
|
|
-gline-tables-only
|
|
-frecord-sources)
|
|
endif()
|
|
add_custom_command(
|
|
COMMAND xcrun -sdk macosx metal
|
|
${METAL_FLAGS}
|
|
-c ${SRCFILE}
|
|
-I${PROJECT_SOURCE_DIR}
|
|
-o ${TARGET}.air
|
|
DEPENDS ${SRCFILE} ${DEPS}
|
|
OUTPUT ${TARGET}.air
|
|
COMMENT "Building ${TARGET}.air"
|
|
VERBATIM
|
|
)
|
|
endfunction(build_kernel_base)
|
|
|
|
function(build_kernel KERNEL)
|
|
set(SRCFILE ${CMAKE_CURRENT_SOURCE_DIR}/${KERNEL}.metal)
|
|
build_kernel_base(${KERNEL} ${SRCFILE} "${HEADERS}")
|
|
endfunction(build_kernel)
|
|
|
|
foreach(KERNEL ${KERNELS})
|
|
build_kernel(${KERNEL})
|
|
set(KERNEL_AIR ${KERNEL}.air ${KERNEL_AIR})
|
|
endforeach()
|
|
|
|
if (NOT MLX_METAL_JIT)
|
|
set(
|
|
STEEL_KERNELS
|
|
${CMAKE_CURRENT_SOURCE_DIR}/steel/conv/kernels/steel_conv.metal
|
|
${CMAKE_CURRENT_SOURCE_DIR}/steel/conv/kernels/steel_conv_general.metal
|
|
${CMAKE_CURRENT_SOURCE_DIR}/steel/gemm/kernels/steel_gemm_fused.metal
|
|
${CMAKE_CURRENT_SOURCE_DIR}/steel/gemm/kernels/steel_gemm_masked.metal
|
|
${CMAKE_CURRENT_SOURCE_DIR}/steel/gemm/kernels/steel_gemm_splitk.metal
|
|
)
|
|
set(
|
|
STEEL_HEADERS
|
|
steel/defines.h
|
|
steel/utils.h
|
|
steel/conv/conv.h
|
|
steel/conv/loader.h
|
|
steel/conv/loaders/loader_channel_l.h
|
|
steel/conv/loaders/loader_channel_n.h
|
|
steel/conv/loaders/loader_general.h
|
|
steel/conv/kernels/steel_conv.h
|
|
steel/conv/kernels/steel_conv_general.h
|
|
steel/gemm/gemm.h
|
|
steel/gemm/mma.h
|
|
steel/gemm/loader.h
|
|
steel/gemm/transforms.h
|
|
steel/gemm/kernels/steel_gemm_fused.h
|
|
steel/gemm/kernels/steel_gemm_masked.h
|
|
steel/gemm/kernels/steel_gemm_splitk.h
|
|
)
|
|
foreach(KERNEL ${STEEL_KERNELS})
|
|
cmake_path(GET KERNEL STEM TARGET)
|
|
build_kernel_base(${TARGET} ${KERNEL} "${STEEL_HEADERS}")
|
|
set(KERNEL_AIR ${TARGET}.air ${KERNEL_AIR})
|
|
endforeach()
|
|
endif()
|
|
|
|
add_custom_command(
|
|
OUTPUT ${MLX_METAL_PATH}/mlx.metallib
|
|
COMMAND xcrun -sdk macosx metallib ${KERNEL_AIR} -o ${MLX_METAL_PATH}/mlx.metallib
|
|
DEPENDS ${KERNEL_AIR}
|
|
COMMENT "Building mlx.metallib"
|
|
VERBATIM
|
|
)
|
|
|
|
add_custom_target(
|
|
mlx-metallib
|
|
DEPENDS
|
|
${MLX_METAL_PATH}/mlx.metallib
|
|
)
|
|
|
|
add_dependencies(
|
|
mlx
|
|
mlx-metallib
|
|
)
|
|
|
|
# Install metallib
|
|
include(GNUInstallDirs)
|
|
|
|
install(
|
|
FILES ${MLX_METAL_PATH}/mlx.metallib
|
|
DESTINATION ${CMAKE_INSTALL_LIBDIR}
|
|
COMPONENT metallib
|
|
)
|