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

cuda : rename build flag to LLAMA_CUDA (#6299)

Browse source
This commit is contained in:
slaren 2024-03-26 01:16:01 +01:00 • committed by GitHub
parent b06c16ef9f
commit 280345968d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
28 changed files with 129 additions and 115 deletions
Download patch file Download diff file Expand all files Collapse all files

21
CMakeLists.txt
View file

@ -89,8 +89,8 @@ endif()
option(LLAMA_ACCELERATE "llama: enable Accelerate framework" ON)
option(LLAMA_BLAS "llama: use BLAS" OFF)
set(LLAMA_BLAS_VENDOR "Generic" CACHE STRING "llama: BLAS library vendor")
option(LLAMA_CUBLAS "llama: use CUDA" OFF)
#option(LLAMA_CUDA_CUBLAS "llama: use cuBLAS for prompt processing" OFF)
option(LLAMA_CUDA "llama: use CUDA" OFF)
option(LLAMA_CUBLAS "llama: use CUDA (deprecated, use LLAMA_CUDA)" OFF)
option(LLAMA_CUDA_FORCE_DMMV "llama: use dmmv instead of mmvq CUDA kernels" OFF)
option(LLAMA_CUDA_FORCE_MMQ "llama: use mmq kernels instead of cuBLAS" OFF)
set(LLAMA_CUDA_DMMV_X "32" CACHE STRING "llama: x stride for dmmv CUDA kernels")
@ -360,11 +360,16 @@ if (LLAMA_QKK_64)
endif()
if (LLAMA_CUBLAS)
message(WARNING "LLAMA_CUBLAS is deprecated and will be removed in the future.\nUse LLAMA_CUDA instead")
set(LLAMA_CUDA ON)
endif()
if (LLAMA_CUDA)
cmake_minimum_required(VERSION 3.17)
find_package(CUDAToolkit)
if (CUDAToolkit_FOUND)
message(STATUS "cuBLAS found")
message(STATUS "CUDA found")
enable_language(CUDA)
@ -373,7 +378,7 @@ if (LLAMA_CUBLAS)
file(GLOB GGML_SOURCES_CUDA "ggml-cuda/*.cu")
list(APPEND GGML_SOURCES_CUDA "ggml-cuda.cu")
add_compile_definitions(GGML_USE_CUBLAS)
add_compile_definitions(GGML_USE_CUDA)
if (LLAMA_CUDA_FORCE_DMMV)
add_compile_definitions(GGML_CUDA_FORCE_DMMV)
endif()
@ -422,7 +427,7 @@ if (LLAMA_CUBLAS)
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
else()
message(WARNING "cuBLAS not found")
message(WARNING "CUDA not found")
endif()
endif()
@ -525,7 +530,7 @@ if (LLAMA_HIPBLAS)
file(GLOB GGML_SOURCES_ROCM "ggml-cuda/*.cu")
list(APPEND GGML_SOURCES_ROCM "ggml-cuda.cu")
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS)
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUDA)
if (LLAMA_HIP_UMA)
add_compile_definitions(GGML_HIP_UMA)
@ -830,7 +835,7 @@ endif()
set(CUDA_CXX_FLAGS "")
if (LLAMA_CUBLAS)
if (LLAMA_CUDA)
set(CUDA_FLAGS -use_fast_math)
if (LLAMA_FATAL_WARNINGS)
@ -1055,7 +1060,7 @@ endif()
add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:${ARCH_FLAGS}>")
add_compile_options("$<$<COMPILE_LANGUAGE:C>:${ARCH_FLAGS}>")
if (LLAMA_CUBLAS)
if (LLAMA_CUDA)
list(APPEND CUDA_CXX_FLAGS ${ARCH_FLAGS})
list(JOIN CUDA_CXX_FLAGS " " CUDA_CXX_FLAGS_JOINED) # pass host compiler flags as a single argument
if (NOT CUDA_CXX_FLAGS_JOINED STREQUAL "")