Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : add GGML_CUDA_USE_GRAPHS option, restore GGML_CUDA_FORCE_CUBLAS (cmake) (#8140)

Browse Source
This commit is contained in:
slaren 2024-06-26 21:34:14 +02:00 committed by GitHub
parent c7ab7b612c
commit 31ec3993f6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CMakeLists.txt
View File

@ -80,6 +80,7 @@ set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED})
set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS}) set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS})
set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS}) set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS})
set(GGML_LLAMAFILE ON) set(GGML_LLAMAFILE ON)
set(GGML_CUDA_USE_GRAPHS ON)
# transition helpers # transition helpers
function (llama_option_depr TYPE OLD NEW) function (llama_option_depr TYPE OLD NEW)

2
ggml/CMakeLists.txt
View File

@ -109,6 +109,7 @@ option(GGML_LLAMAFILE "ggml: use ggml SGEMM"
option(GGML_CUDA "ggml: use CUDA" OFF) option(GGML_CUDA "ggml: use CUDA" OFF)
option(GGML_CUDA_FORCE_DMMV "ggml: use dmmv instead of mmvq CUDA kernels" OFF) option(GGML_CUDA_FORCE_DMMV "ggml: use dmmv instead of mmvq CUDA kernels" OFF)
option(GGML_CUDA_FORCE_MMQ "ggml: use mmq kernels instead of cuBLAS" OFF) option(GGML_CUDA_FORCE_MMQ "ggml: use mmq kernels instead of cuBLAS" OFF)
option(GGML_CUDA_FORCE_CUBLAS "ggml: always use cuBLAS instead of mmq kernels" OFF)
set (GGML_CUDA_DMMV_X "32" CACHE STRING "ggml: x stride for dmmv CUDA kernels") set (GGML_CUDA_DMMV_X "32" CACHE STRING "ggml: x stride for dmmv CUDA kernels")
set (GGML_CUDA_MMV_Y "1" CACHE STRING "ggml: y block size for mmv CUDA kernels") set (GGML_CUDA_MMV_Y "1" CACHE STRING "ggml: y block size for mmv CUDA kernels")
option(GGML_CUDA_F16 "ggml: use 16 bit floats for some calculations" OFF) option(GGML_CUDA_F16 "ggml: use 16 bit floats for some calculations" OFF)
@ -119,6 +120,7 @@ set (GGML_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
option(GGML_CUDA_NO_PEER_COPY "ggml: do not use peer to peer copies" OFF) option(GGML_CUDA_NO_PEER_COPY "ggml: do not use peer to peer copies" OFF)
option(GGML_CUDA_NO_VMM "ggml: do not try to use CUDA VMM" OFF) option(GGML_CUDA_NO_VMM "ggml: do not try to use CUDA VMM" OFF)
option(GGML_CUDA_FA_ALL_QUANTS "ggml: compile all quants for FlashAttention" OFF) option(GGML_CUDA_FA_ALL_QUANTS "ggml: compile all quants for FlashAttention" OFF)
option(GGML_CUDA_USE_GRAPHS "ggml: use CUDA graphs (llama.cpp only)" OFF)
option(GGML_CURL "ggml: use libcurl to download model from an URL" OFF) option(GGML_CURL "ggml: use libcurl to download model from an URL" OFF)
option(GGML_HIPBLAS "ggml: use hipBLAS" OFF) option(GGML_HIPBLAS "ggml: use hipBLAS" OFF)

5
ggml/src/CMakeLists.txt
View File

@ -295,12 +295,15 @@ if (GGML_CUDA)
list(APPEND GGML_CDEF_PUBLIC GGML_USE_CUDA) list(APPEND GGML_CDEF_PUBLIC GGML_USE_CUDA)
add_compile_definitions(GGML_CUDA_USE_GRAPHS)
add_compile_definitions(GGML_CUDA_DMMV_X=${GGML_CUDA_DMMV_X}) add_compile_definitions(GGML_CUDA_DMMV_X=${GGML_CUDA_DMMV_X})
add_compile_definitions(GGML_CUDA_MMV_Y=${GGML_CUDA_MMV_Y}) add_compile_definitions(GGML_CUDA_MMV_Y=${GGML_CUDA_MMV_Y})
add_compile_definitions(K_QUANTS_PER_ITERATION=${GGML_CUDA_KQUANTS_ITER}) add_compile_definitions(K_QUANTS_PER_ITERATION=${GGML_CUDA_KQUANTS_ITER})
add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${GGML_CUDA_PEER_MAX_BATCH_SIZE}) add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${GGML_CUDA_PEER_MAX_BATCH_SIZE})
if (GGML_CUDA_USE_GRAPHS)
add_compile_definitions(GGML_CUDA_USE_GRAPHS)
endif()
if (GGML_CUDA_FORCE_DMMV) if (GGML_CUDA_FORCE_DMMV)
add_compile_definitions(GGML_CUDA_FORCE_DMMV) add_compile_definitions(GGML_CUDA_FORCE_DMMV)
endif() endif()