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

ROCm Port (#1087)

Browse source 
* use hipblas based on cublas
* Update Makefile for the Cuda kernels
* Expand arch list and make it overrideable
* Fix multi GPU on multiple amd architectures with rocblas_initialize() (#5)
* add hipBLAS to README
* new build arg LLAMA_CUDA_MMQ_Y
* fix half2 decomposition
* Add intrinsics polyfills for AMD
* AMD assembly optimized __dp4a
* Allow overriding CC_TURING
* use "ROCm" instead of "CUDA"
* ignore all build dirs
* Add Dockerfiles
* fix llama-bench
* fix -nommq help for non CUDA/HIP

---------

Co-authored-by: YellowRoseCx <80486540+YellowRoseCx@users.noreply.github.com>
Co-authored-by: ardfork <134447697+ardfork@users.noreply.github.com>
Co-authored-by: funnbot <22226942+funnbot@users.noreply.github.com>
Co-authored-by: Engininja2 <139037756+Engininja2@users.noreply.github.com>
Co-authored-by: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
Co-authored-by: jammm <2500920+jammm@users.noreply.github.com>
Co-authored-by: jdecourval <7315817+jdecourval@users.noreply.github.com>
This commit is contained in:
Henri Vasserman 2023-08-25 12:09:42 +03:00 • committed by GitHub
parent 3f460a2b72
commit 6bbc598a63
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
12 changed files with 335 additions and 59 deletions
Download patch file Download diff file Expand all files Collapse all files

38
CMakeLists.txt
View file

@ -74,6 +74,7 @@ set(LLAMA_CUDA_DMMV_X "32" CACHE STRING "llama: x stride for dmmv CUDA kern
set(LLAMA_CUDA_MMV_Y "1" CACHE STRING "llama: y block size for mmv CUDA kernels")
option(LLAMA_CUDA_F16 "llama: use 16 bit floats for some calculations" OFF)
set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for Q2_K/Q6_K")
option(LLAMA_HIPBLAS "llama: use hipBLAS" OFF)
option(LLAMA_CLBLAST "llama: use CLBlast" OFF)
option(LLAMA_METAL "llama: use Metal" OFF)
option(LLAMA_MPI "llama: use MPI" OFF)
@ -352,6 +353,43 @@ if (LLAMA_CLBLAST)
endif()
endif()
if (LLAMA_HIPBLAS)
list(APPEND CMAKE_PREFIX_PATH /opt/rocm)
if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for HIP, hint: CC=/opt/rocm/llvm/bin/clang")
endif()
if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for HIP, hint: CXX=/opt/rocm/llvm/bin/clang++")
endif()
find_package(hip)
find_package(hipblas)
find_package(rocblas)
if (${hipblas_FOUND} AND ${hip_FOUND})
message(STATUS "HIP and hipBLAS found")
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS)
add_library(ggml-rocm OBJECT ggml-cuda.cu ggml-cuda.h)
if (LLAMA_CUDA_FORCE_DMMV)
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_FORCE_DMMV)
endif()
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_DMMV_X=${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})
target_compile_definitions(ggml-rocm PRIVATE CC_TURING=1000000000)
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)
message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
endif()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ggml-rocm)
else()
message(WARNING "hipBLAS or HIP not found. Try setting CMAKE_PREFIX_PATH=/opt/rocm")
endif()
endif()
if (LLAMA_ALL_WARNINGS)
if (NOT MSVC)
set(c_flags