tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

metal : enable shader debugging (cmake option) (#4705)

Browse Source 
* ggml : disable fast-math for Metal (cmake build only)

ggml-ci

* metal : fix Metal API debug warnings

* cmake : add -fno-inline for Metal build (#4545)

* metal : fix API debug warnings

* metal : fix compile warnings

* metal : use uint64_t for strides

* cmake : rename option to LLAMA_METAL_SHADER_DEBUG

* metal : fix mat-vec Q8_0 kernel for BS > 1

* metal : normalize mat-vec kernel signatures

* cmake : respect LLAMA_QKK_64 option

* metal : fix mat-vec Q4_K kernel for QK_K == 64

ggml-ci
This commit is contained in:
Georgi Gerganov
2024-01-02 10:57:44 +02:00
committed by GitHub
parent edd1ab7bc3
commit 58ba655af0
5 changed files with 329 additions and 230 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
CMakeLists.txt
View File

@ -95,6 +95,7 @@ option(LLAMA_HIP_UMA "llama: use HIP unified memory arch
option(LLAMA_CLBLAST "llama: use CLBlast" OFF)
option(LLAMA_METAL "llama: use Metal" ${LLAMA_METAL_DEFAULT})
option(LLAMA_METAL_NDEBUG "llama: disable Metal debugging" OFF)
option(LLAMA_METAL_SHADER_DEBUG "llama: compile Metal with -fno-fast-math" OFF)
option(LLAMA_MPI "llama: use MPI" OFF)
option(LLAMA_QKK_64 "llama: use super-block size of 64 for k-quants" OFF)
@ -154,9 +155,9 @@ if (APPLE AND LLAMA_ACCELERATE)
endif()
if (LLAMA_METAL)
find_library(FOUNDATION_LIBRARY Foundation REQUIRED)
find_library(METAL_FRAMEWORK Metal REQUIRED)
find_library(METALKIT_FRAMEWORK MetalKit REQUIRED)
find_library(FOUNDATION_LIBRARY Foundation REQUIRED)
find_library(METAL_FRAMEWORK Metal REQUIRED)
find_library(METALKIT_FRAMEWORK MetalKit REQUIRED)
message(STATUS "Metal framework found")
set(GGML_HEADERS_METAL ggml-metal.h)
@ -173,6 +174,33 @@ if (LLAMA_METAL)
# copy ggml-metal.metal to bin directory
configure_file(ggml-metal.metal ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal COPYONLY)
if (LLAMA_METAL_SHADER_DEBUG)
# custom command to do the following:
# xcrun -sdk macosx metal -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
# xcrun -sdk macosx metallib ggml-metal.air -o ggml.metallib
#
# note: this is the only way I found to disable fast-math in Metal. it's ugly, but at least it works
# 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
set(XC_FLAGS -fno-fast-math -fno-inline -g)
if (LLAMA_QKK_64)
set(XC_FLAGS ${XC_FLAGS} -DQK_K=64)
endif()
add_custom_command(
OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml.metallib
COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
COMMAND xcrun -sdk macosx metallib ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml.metallib
DEPENDS ggml-metal.metal
COMMENT "Compiling Metal kernels"
)
add_custom_target(
ggml-metal ALL
DEPENDS ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml.metallib
)
endif()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS}
${FOUNDATION_LIBRARY}
${METAL_FRAMEWORK}