vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

first update for migration

Browse source
This commit is contained in:
jianyuzh 2023-12-27 11:19:46 +08:00 • committed by Meng, Hengyu
parent 011e8ec577
commit 7a4343df61
4 changed files with 12461 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

76
CMakeLists.txt
View file

@ -1,5 +1,6 @@
cmake_minimum_required(VERSION 3.14) # for add_link_options and implicit target directories. cmake_minimum_required(VERSION 3.14) # for add_link_options and implicit target directories.
project("llama.cpp" C CXX) project("llama.cpp" C CXX)
include(CheckIncludeFileCXX)
set(CMAKE_EXPORT_COMPILE_COMMANDS ON) set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
@ -96,11 +97,11 @@ set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for
set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
"llama: max. batch size for using peer access") "llama: max. batch size for using peer access")
option(LLAMA_HIPBLAS "llama: use hipBLAS" OFF) 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_CLBLAST "llama: use CLBlast" OFF)
option(LLAMA_METAL "llama: use Metal" ${LLAMA_METAL_DEFAULT}) option(LLAMA_METAL "llama: use Metal" ${LLAMA_METAL_DEFAULT})
option(LLAMA_METAL_NDEBUG "llama: disable Metal debugging" OFF) option(LLAMA_METAL_NDEBUG "llama: disable Metal debugging" OFF)
option(LLAMA_METAL_SHADER_DEBUG "llama: compile Metal with -fno-fast-math" OFF) option(LLAMA_METAL_SHADER_DEBUG "llama: compile Metal with -fno-fast-math" OFF)
option(LLAMA_SYCL "llama: use SYCL" OFF)
option(LLAMA_MPI "llama: use MPI" OFF) option(LLAMA_MPI "llama: use MPI" OFF)
option(LLAMA_QKK_64 "llama: use super-block size of 64 for k-quants" OFF) option(LLAMA_QKK_64 "llama: use super-block size of 64 for k-quants" OFF)
@ -122,7 +123,7 @@ include(${CMAKE_CURRENT_SOURCE_DIR}/scripts/build-info.cmake)
# Compile flags # Compile flags
# #
set(CMAKE_CXX_STANDARD 11) set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED true) set(CMAKE_CXX_STANDARD_REQUIRED true)
set(CMAKE_C_STANDARD 11) set(CMAKE_C_STANDARD 11)
set(CMAKE_C_STANDARD_REQUIRED true) set(CMAKE_C_STANDARD_REQUIRED true)
@ -338,18 +339,11 @@ if (LLAMA_CUBLAS)
add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${LLAMA_CUDA_PEER_MAX_BATCH_SIZE}) add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${LLAMA_CUDA_PEER_MAX_BATCH_SIZE})
if (LLAMA_STATIC) if (LLAMA_STATIC)
if (WIN32) set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
# 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() else()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt) set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt)
endif() endif()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cuda_driver)
if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES) if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# 52 == lowest CUDA 12 standard # 52 == lowest CUDA 12 standard
# 60 == f16 CUDA intrinsics # 60 == f16 CUDA intrinsics
@ -426,9 +420,6 @@ if (LLAMA_HIPBLAS)
if (${hipblas_FOUND} AND ${hip_FOUND}) if (${hipblas_FOUND} AND ${hip_FOUND})
message(STATUS "HIP and hipBLAS found") message(STATUS "HIP and hipBLAS found")
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS) 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) add_library(ggml-rocm OBJECT ggml-cuda.cu ggml-cuda.h)
if (BUILD_SHARED_LIBS) if (BUILD_SHARED_LIBS)
set_target_properties(ggml-rocm PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(ggml-rocm PROPERTIES POSITION_INDEPENDENT_CODE ON)
@ -454,6 +445,64 @@ if (LLAMA_HIPBLAS)
endif() endif()
endif() endif()
if (LLAMA_SYCL)
set(ENABLE_AOT ats)
if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for SYCL")
endif()
if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for SYCL")
endif()
#find_package(SYCL REQUIRED)
find_package(IntelSYCL REQUIRED)
# Check SYCL support by the compiler
check_cxx_compiler_flag("-fsycl" _fsycl_option)
if (_fsycl_option)
#set (CMAKE_REQUIRED_INCLUDES ${CMAKE_REQUIRED_INCLUDES} "/opt/intel/oneapi/compiler/2024.0/include")
CHECK_INCLUDE_FILE_CXX("sycl/sycl.hpp" _sycl_header "-fsycl")
set (_sycl_header "/opt/intel/oneapi/compiler/2024.0/include/sycl/sycl.hpp")
if (NOT _sycl_header)
CHECK_INCLUDE_FILE_CXX("CL/sycl.hpp" _sycl_header_old "-fsycl")
endif()
if (_sycl_header OR _sycl_header_old)
set(_sycl_support TRUE)
endif()
endif()
if (_sycl_support)
add_compile_definitions(GGML_USE_CUBLAS)
#add_compile_options(-std=c++17 -O3 -fsycl)
add_compile_options(-I/opt/intel/oneapi/compiler/2024.0/include)
add_compile_options(-I/opt/intel/oneapi/compiler/2024.0/include/sycl)
add_compile_options(-I/opt/intel/oneapi/dpcpp-ct/2024.0/include)
add_compile_options(-I/opt/intel/oneapi/2024.0/include)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++17")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -L${MKLROOT}/lib")
set(GGML_HEADERS_SYCL ggml-cuda.h ggml.h ggml-sycl.hpp)
set(GGML_SOURCES_SYCL ggml-sycl.cpp)
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} sycl OpenCL mkl_core pthread m dl mkl_sycl_blas mkl_sycl_lapack mkl_sycl_dft mkl_sycl_sparse mkl_sycl_vm mkl_sycl_rng mkl_sycl_stats mkl_sycl_data_fitting mkl_intel_ilp64 mkl_tbb_thread)
#add_library(ggml-sycl OBJECT ${GGML_SOURCES_SYCL} ${GGML_HEADERS_SYCL})
#add_executable(${PROJECT_NAME} ${GGML_SOURCES_SYCL} ${GGML_HEADERS_SYCL})
#target_link_libraries(ggml-sycl PRIVATE sycl)
#target_compile_options(${PROJECT_NAME} PRIVATE ${CMAKE_CXX_FLAGS})
#set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} sycl)
#add_sycl_to_target({})
else()
message(FATAL_ERROR "SYCL Support is not present")
endif()
endif()
function(get_flags CCID CCVER) function(get_flags CCID CCVER)
set(C_FLAGS "") set(C_FLAGS "")
set(CXX_FLAGS "") set(CXX_FLAGS "")
@ -790,6 +839,7 @@ add_library(ggml OBJECT
${GGML_SOURCES_METAL} ${GGML_HEADERS_METAL} ${GGML_SOURCES_METAL} ${GGML_HEADERS_METAL}
${GGML_SOURCES_MPI} ${GGML_HEADERS_MPI} ${GGML_SOURCES_MPI} ${GGML_HEADERS_MPI}
${GGML_SOURCES_EXTRA} ${GGML_HEADERS_EXTRA} ${GGML_SOURCES_EXTRA} ${GGML_HEADERS_EXTRA}
${GGML_SOURCES_SYCL} ${GGML_HEADERS_SYCL}
) )
target_include_directories(ggml PUBLIC . ${LLAMA_EXTRA_INCLUDES}) target_include_directories(ggml PUBLIC . ${LLAMA_EXTRA_INCLUDES})

12393
ggml-sycl.cpp Normal file
View file

File diff suppressed because it is too large Load diff

4
ggml-sycl.hpp Normal file
View file

@ -0,0 +1,4 @@
#include <sycl/sycl.hpp>
#include <dpct/dpct.hpp>
typedef half ggml_fp16_t;

2
ggml.h
View file

@ -2283,7 +2283,7 @@ extern "C" {
typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int k); typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int k);
typedef void (*ggml_vec_dot_t) (const int n, float * GGML_RESTRICT s, const void * GGML_RESTRICT x, const void * GGML_RESTRICT y); typedef void (*ggml_vec_dot_t) (const int n, float * GGML_RESTRICT s, const void * GGML_RESTRICT x, const void * GGML_RESTRICT y);
typedef struct { typedef struct dpct_type_994041 {
const char * type_name; const char * type_name;
int blck_size; int blck_size;
size_t type_size; size_t type_size;