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

Make hipBLAS CMake more similar to cuBLAS

Browse source
This commit is contained in:
ardfork 2023-11-10 15:21:58 +00:00
parent a75fa576ab
commit e571faeb6b
2 changed files with 28 additions and 30 deletions
Download patch file Download diff file Expand all files Collapse all files

54
CMakeLists.txt
View file

@ -347,42 +347,39 @@ if (LLAMA_CLBLAST)
endif() endif()
if (LLAMA_HIPBLAS) if (LLAMA_HIPBLAS)
list(APPEND CMAKE_PREFIX_PATH /opt/rocm) cmake_minimum_required(VERSION 3.21) # HIP language support requires 3.21
cmake_policy(VERSION 3.21.3...3.27)
if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang") find_package(HIP)
message(WARNING "Only LLVM is supported for HIP, hint: CC=/opt/rocm/llvm/bin/clang")
endif()
if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for HIP, hint: CXX=/opt/rocm/llvm/bin/clang++")
endif()
find_package(hip)
find_package(hipblas) find_package(hipblas)
find_package(rocblas) find_package(rocblas)
if (${hipblas_FOUND} AND ${hip_FOUND}) if (${HIP_FOUND} AND ${hipblas_FOUND} AND ${rocblas_FOUND})
message(STATUS "HIP and hipBLAS found") message(STATUS "hipBLAS found")
if (NOT DEFINED CMAKE_HIP_ARCHITECTURES)
set(CMAKE_HIP_ARCHITECTURES "native")
endif()
message(STATUS "Using HIP architectures: ${CMAKE_HIP_ARCHITECTURES}")
enable_language(HIP)
set(GGML_HEADERS_HIP ggml-cuda.h)
set(GGML_SOURCES_HIP ggml-cuda.cu)
set_source_files_properties(${GGML_SOURCES_HIP} PROPERTIES LANGUAGE HIP)
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS) add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS)
add_library(ggml-rocm OBJECT ggml-cuda.cu ggml-cuda.h)
if (BUILD_SHARED_LIBS) add_compile_definitions(GGML_CUDA_DMMV_X=${LLAMA_CUDA_DMMV_X})
set_target_properties(ggml-rocm PROPERTIES POSITION_INDEPENDENT_CODE ON) add_compile_definitions(GGML_CUDA_MMV_Y=${LLAMA_CUDA_MMV_Y})
endif() add_compile_definitions(K_QUANTS_PER_ITERATION=${LLAMA_CUDA_KQUANTS_ITER})
if (LLAMA_CUDA_FORCE_DMMV)
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_FORCE_DMMV)
endif()
if (LLAMA_CUDA_FORCE_MMQ)
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_FORCE_MMQ)
endif()
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_DMMV_X=${LLAMA_CUDA_DMMV_X})
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_MMV_Y=${LLAMA_CUDA_MMV_Y})
target_compile_definitions(ggml-rocm PRIVATE K_QUANTS_PER_ITERATION=${LLAMA_CUDA_KQUANTS_ITER})
set_source_files_properties(ggml-cuda.cu PROPERTIES LANGUAGE CXX)
target_link_libraries(ggml-rocm PRIVATE hip::device PUBLIC hip::host roc::rocblas roc::hipblas)
if (LLAMA_STATIC) if (LLAMA_STATIC)
message(FATAL_ERROR "Static linking not supported for HIP/ROCm") message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
endif() endif()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ggml-rocm)
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} roc::hipblas roc::rocblas)
else() else()
message(WARNING "hipBLAS or HIP not found. Try setting CMAKE_PREFIX_PATH=/opt/rocm") message(WARNING "hipBLAS or HIP not found. Try setting CMAKE_PREFIX_PATH=/opt/rocm")
endif() endif()
@ -642,6 +639,7 @@ add_library(ggml OBJECT
ggml-quants.h ggml-quants.h
${GGML_SOURCES_CUDA} ${GGML_HEADERS_CUDA} ${GGML_SOURCES_CUDA} ${GGML_HEADERS_CUDA}
${GGML_SOURCES_OPENCL} ${GGML_HEADERS_OPENCL} ${GGML_SOURCES_OPENCL} ${GGML_HEADERS_OPENCL}
${GGML_SOURCES_HIP} ${GGML_HEADERS_HIP}
${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}
@ -721,7 +719,7 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/LlamaConfig.cmake
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/Llama) DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/Llama)
set(GGML_PUBLIC_HEADERS "ggml.h" set(GGML_PUBLIC_HEADERS "ggml.h"
"${GGML_HEADERS_CUDA}" "${GGML_HEADERS_OPENCL}" "${GGML_HEADERS_CUDA}" "${GGML_HEADERS_OPENCL}" "${GGML_HEADERS_HIP}"
"${GGML_HEADERS_METAL}" "${GGML_HEADERS_MPI}" "${GGML_HEADERS_EXTRA}") "${GGML_HEADERS_METAL}" "${GGML_HEADERS_MPI}" "${GGML_HEADERS_EXTRA}")
set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}") set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}")

4
README.md
View file

@ -419,8 +419,8 @@ Building the program with BLAS support may lead to some performance improvements
```bash ```bash
mkdir build mkdir build
cd build cd build
CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ cmake .. -DLLAMA_HIPBLAS=ON cmake .. -DLLAMA_HIPBLAS=ON
cmake --build . cmake --build . --config Release
``` ```
The environment variable [`HIP_VISIBLE_DEVICES`](https://rocm.docs.amd.com/en/latest/understand/gpu_isolation.html#hip-visible-devices) can be used to specify which GPU(s) will be used. The environment variable [`HIP_VISIBLE_DEVICES`](https://rocm.docs.amd.com/en/latest/understand/gpu_isolation.html#hip-visible-devices) can be used to specify which GPU(s) will be used.