Merge commit '31f27758fa' into ceb/nomic-vulkan
This commit is contained in:
Jared Van Bortel
commit
3959283eed
62 changed files with 10619 additions and 2566 deletions
160
CMakeLists.txt
160
CMakeLists.txt
|
|
@ -91,6 +91,7 @@ set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for
|
|||
set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
|
||||
"llama: max. batch size for using peer access")
|
||||
option(LLAMA_HIPBLAS "llama: use hipBLAS" OFF)
|
||||
option(LLAMA_HIP_UMA "llama: use HIP unified memory architecture" OFF)
|
||||
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)
|
||||
|
|
@ -98,9 +99,9 @@ option(LLAMA_KOMPUTE "llama: use Kompute"
|
|||
option(LLAMA_MPI "llama: use MPI" OFF)
|
||||
option(LLAMA_QKK_64 "llama: use super-block size of 64 for k-quants" OFF)
|
||||
|
||||
option(LLAMA_BUILD_TESTS "llama: build tests" ${LLAMA_STANDALONE})
|
||||
option(LLAMA_BUILD_EXAMPLES "llama: build examples" ${LLAMA_STANDALONE})
|
||||
option(LLAMA_BUILD_SERVER "llama: build server example" ON)
|
||||
option(LLAMA_BUILD_TESTS "llama: build tests" ${LLAMA_STANDALONE})
|
||||
option(LLAMA_BUILD_EXAMPLES "llama: build examples" ${LLAMA_STANDALONE})
|
||||
option(LLAMA_BUILD_SERVER "llama: build server example" ON)
|
||||
|
||||
# Required for relocatable CMake package
|
||||
include(${CMAKE_CURRENT_SOURCE_DIR}/scripts/build-info.cmake)
|
||||
|
|
@ -292,7 +293,12 @@ if (LLAMA_CUBLAS)
|
|||
add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${LLAMA_CUDA_PEER_MAX_BATCH_SIZE})
|
||||
|
||||
if (LLAMA_STATIC)
|
||||
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
|
||||
if (WIN32)
|
||||
# As of 12.3.1 CUDA Tookit for Windows does not offer a static cublas library
|
||||
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas CUDA::cublasLt)
|
||||
else ()
|
||||
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
|
||||
endif()
|
||||
else()
|
||||
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt)
|
||||
endif()
|
||||
|
|
@ -373,6 +379,9 @@ if (LLAMA_HIPBLAS)
|
|||
if (${hipblas_FOUND} AND ${hip_FOUND})
|
||||
message(STATUS "HIP and hipBLAS found")
|
||||
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS)
|
||||
if (LLAMA_HIP_UMA)
|
||||
add_compile_definitions(GGML_HIP_UMA)
|
||||
endif()
|
||||
add_library(ggml-rocm OBJECT ggml-cuda.cu ggml-cuda.h)
|
||||
if (BUILD_SHARED_LIBS)
|
||||
set_target_properties(ggml-rocm PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
||||
|
|
@ -537,57 +546,102 @@ if (LLAMA_KOMPUTE)
|
|||
endif()
|
||||
endif()
|
||||
|
||||
if (LLAMA_ALL_WARNINGS)
|
||||
if (NOT MSVC)
|
||||
set(warning_flags -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function)
|
||||
set(c_flags -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration)
|
||||
set(cxx_flags -Wmissing-declarations -Wmissing-noreturn)
|
||||
set(host_cxx_flags "")
|
||||
function(get_flags CCID CCVER)
|
||||
set(C_FLAGS "")
|
||||
set(CXX_FLAGS "")
|
||||
|
||||
if (CMAKE_C_COMPILER_ID MATCHES "Clang")
|
||||
set(warning_flags ${warning_flags} -Wunreachable-code-break -Wunreachable-code-return)
|
||||
set(host_cxx_flags ${host_cxx_flags} -Wmissing-prototypes -Wextra-semi)
|
||||
if (CCID MATCHES "Clang")
|
||||
set(C_FLAGS -Wunreachable-code-break -Wunreachable-code-return)
|
||||
set(CXX_FLAGS -Wunreachable-code-break -Wunreachable-code-return -Wmissing-prototypes -Wextra-semi)
|
||||
|
||||
if (
|
||||
(CMAKE_C_COMPILER_ID STREQUAL "Clang" AND CMAKE_C_COMPILER_VERSION VERSION_GREATER_EQUAL 3.8.0) OR
|
||||
(CMAKE_C_COMPILER_ID STREQUAL "AppleClang" AND CMAKE_C_COMPILER_VERSION VERSION_GREATER_EQUAL 7.3.0)
|
||||
)
|
||||
set(c_flags ${c_flags} -Wdouble-promotion)
|
||||
endif()
|
||||
elseif (CMAKE_C_COMPILER_ID STREQUAL "GNU")
|
||||
set(c_flags ${c_flags} -Wdouble-promotion)
|
||||
set(host_cxx_flags ${host_cxx_flags} -Wno-array-bounds)
|
||||
|
||||
if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 7.1.0)
|
||||
set(host_cxx_flags ${host_cxx_flags} -Wno-format-truncation)
|
||||
endif()
|
||||
if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 8.1.0)
|
||||
set(host_cxx_flags ${host_cxx_flags} -Wextra-semi)
|
||||
endif()
|
||||
if (
|
||||
(CCID STREQUAL "Clang" AND CCVER VERSION_GREATER_EQUAL 3.8.0) OR
|
||||
(CCID STREQUAL "AppleClang" AND CCVER VERSION_GREATER_EQUAL 7.3.0)
|
||||
)
|
||||
set(C_FLAGS ${C_FLAGS} -Wdouble-promotion)
|
||||
endif()
|
||||
elseif (CCID STREQUAL "GNU")
|
||||
set(C_FLAGS -Wdouble-promotion)
|
||||
set(CXX_FLAGS -Wno-array-bounds)
|
||||
|
||||
if (CCVER VERSION_GREATER_EQUAL 7.1.0)
|
||||
set(CXX_FLAGS ${CXX_FLAGS} -Wno-format-truncation)
|
||||
endif()
|
||||
if (CCVER VERSION_GREATER_EQUAL 8.1.0)
|
||||
set(CXX_FLAGS ${CXX_FLAGS} -Wextra-semi)
|
||||
endif()
|
||||
else()
|
||||
# todo : msvc
|
||||
endif()
|
||||
|
||||
set(c_flags ${c_flags} ${warning_flags})
|
||||
set(cxx_flags ${cxx_flags} ${warning_flags})
|
||||
add_compile_options("$<$<COMPILE_LANGUAGE:C>:${c_flags}>"
|
||||
"$<$<COMPILE_LANGUAGE:CXX>:${cxx_flags}>"
|
||||
"$<$<COMPILE_LANGUAGE:CXX>:${host_cxx_flags}>")
|
||||
set(GF_C_FLAGS ${C_FLAGS} PARENT_SCOPE)
|
||||
set(GF_CXX_FLAGS ${CXX_FLAGS} PARENT_SCOPE)
|
||||
endfunction()
|
||||
|
||||
if (LLAMA_ALL_WARNINGS)
|
||||
if (NOT MSVC)
|
||||
set(WARNING_FLAGS -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function)
|
||||
set(C_FLAGS -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes
|
||||
-Werror=implicit-int -Werror=implicit-function-declaration)
|
||||
set(CXX_FLAGS -Wmissing-declarations -Wmissing-noreturn)
|
||||
|
||||
set(C_FLAGS ${WARNING_FLAGS} ${C_FLAGS})
|
||||
set(CXX_FLAGS ${WARNING_FLAGS} ${CXX_FLAGS})
|
||||
|
||||
get_flags(${CMAKE_CXX_COMPILER_ID} ${CMAKE_CXX_COMPILER_VERSION})
|
||||
|
||||
add_compile_options("$<$<COMPILE_LANGUAGE:C>:${C_FLAGS};${GF_C_FLAGS}>"
|
||||
"$<$<COMPILE_LANGUAGE:CXX>:${CXX_FLAGS};${GF_CXX_FLAGS}>")
|
||||
else()
|
||||
# todo : msvc
|
||||
set(C_FLAGS "")
|
||||
set(CXX_FLAGS "")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if (NOT MSVC)
|
||||
set(cuda_flags -Wno-pedantic)
|
||||
endif()
|
||||
set(cuda_flags ${cxx_flags} -use_fast_math ${cuda_flags})
|
||||
if (LLAMA_CUBLAS)
|
||||
set(CUDA_FLAGS ${CXX_FLAGS} -use_fast_math)
|
||||
if (NOT MSVC)
|
||||
set(CUDA_FLAGS ${CUDA_FLAGS} -Wno-pedantic)
|
||||
endif()
|
||||
|
||||
list(JOIN host_cxx_flags " " cuda_host_flags) # pass host compiler flags as a single argument
|
||||
if (NOT cuda_host_flags STREQUAL "")
|
||||
set(cuda_flags ${cuda_flags} -Xcompiler ${cuda_host_flags})
|
||||
endif()
|
||||
if (LLAMA_ALL_WARNINGS AND NOT MSVC)
|
||||
set(NVCC_CMD ${CMAKE_CUDA_COMPILER} .c)
|
||||
if (NOT CMAKE_CUDA_HOST_COMPILER STREQUAL "")
|
||||
set(NVCC_CMD ${NVCC_CMD} -ccbin ${CMAKE_CUDA_HOST_COMPILER})
|
||||
endif()
|
||||
|
||||
add_compile_options("$<$<COMPILE_LANGUAGE:CUDA>:${cuda_flags}>")
|
||||
execute_process(
|
||||
COMMAND ${NVCC_CMD} -Xcompiler --version
|
||||
OUTPUT_VARIABLE CUDA_CCFULLVER
|
||||
ERROR_QUIET
|
||||
)
|
||||
|
||||
if (NOT CUDA_CCFULLVER MATCHES clang)
|
||||
set(CUDA_CCID "GNU")
|
||||
execute_process(
|
||||
COMMAND ${NVCC_CMD} -Xcompiler "-dumpfullversion -dumpversion"
|
||||
OUTPUT_VARIABLE CUDA_CCVER
|
||||
ERROR_QUIET
|
||||
)
|
||||
else()
|
||||
if (CUDA_CCFULLVER MATCHES Apple)
|
||||
set(CUDA_CCID "AppleClang")
|
||||
else()
|
||||
set(CUDA_CCID "Clang")
|
||||
endif()
|
||||
string(REGEX REPLACE "^.* version ([0-9.]*).*$" "\\1" CUDA_CCVER ${CUDA_CCFULLVER})
|
||||
endif()
|
||||
|
||||
message("-- CUDA host compiler is ${CUDA_CCID} ${CUDA_CCVER}")
|
||||
|
||||
get_flags(${CUDA_CCID} ${CUDA_CCVER})
|
||||
list(JOIN GF_CXX_FLAGS " " CUDA_CXX_FLAGS) # pass host compiler flags as a single argument
|
||||
if (NOT CUDA_CXX_FLAGS STREQUAL "")
|
||||
set(CUDA_FLAGS ${CUDA_FLAGS} -Xcompiler ${CUDA_CXX_FLAGS})
|
||||
endif()
|
||||
endif()
|
||||
|
||||
add_compile_options("$<$<COMPILE_LANGUAGE:CUDA>:${CUDA_FLAGS}>")
|
||||
endif()
|
||||
|
||||
if (WIN32)
|
||||
add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
|
||||
|
|
@ -611,6 +665,7 @@ endif()
|
|||
execute_process(
|
||||
COMMAND ${CMAKE_C_COMPILER} ${CMAKE_EXE_LINKER_FLAGS} -Wl,-v
|
||||
ERROR_VARIABLE output
|
||||
OUTPUT_QUIET
|
||||
)
|
||||
if (output MATCHES "dyld-1015\.7")
|
||||
add_compile_definitions(HAVE_BUGGY_APPLE_LINKER)
|
||||
|
|
@ -733,6 +788,11 @@ else()
|
|||
message(STATUS "Unknown architecture")
|
||||
endif()
|
||||
|
||||
if (MINGW)
|
||||
# Target Windows 8 for PrefetchVirtualMemory
|
||||
add_compile_definitions(_WIN32_WINNT=0x602)
|
||||
endif()
|
||||
|
||||
#
|
||||
# POSIX conformance
|
||||
#
|
||||
|
|
@ -802,11 +862,11 @@ add_library(ggml OBJECT
|
|||
ggml-backend.h
|
||||
ggml-quants.c
|
||||
ggml-quants.h
|
||||
${GGML_SOURCES_CUDA} ${GGML_HEADERS_CUDA}
|
||||
${GGML_SOURCES_OPENCL} ${GGML_HEADERS_OPENCL}
|
||||
${GGML_SOURCES_METAL} ${GGML_HEADERS_METAL}
|
||||
${GGML_SOURCES_MPI} ${GGML_HEADERS_MPI}
|
||||
${GGML_SOURCES_EXTRA} ${GGML_HEADERS_EXTRA}
|
||||
${GGML_SOURCES_CUDA} ${GGML_HEADERS_CUDA}
|
||||
${GGML_SOURCES_OPENCL} ${GGML_HEADERS_OPENCL}
|
||||
${GGML_SOURCES_METAL} ${GGML_HEADERS_METAL}
|
||||
${GGML_SOURCES_MPI} ${GGML_HEADERS_MPI}
|
||||
${GGML_SOURCES_EXTRA} ${GGML_HEADERS_EXTRA}
|
||||
${GGML_SOURCES_KOMPUTE} ${GGML_HEADERS_KOMPUTE}
|
||||
)
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue