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

CUDA: remove DMMV, consolidate F16 mult mat vec (#10318)

Browse source
This commit is contained in:
Johannes Gäßler 2024-11-17 09:09:55 +01:00 • committed by GitHub
parent 467576b6cc
commit c3ea58aca4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 246 additions and 1000 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml/src/ggml-hip/CMakeLists.txt
View file

@ -75,18 +75,11 @@ target_include_directories(ggml-hip PRIVATE . ..)
target_compile_definitions(ggml PUBLIC GGML_USE_CUDA)
add_compile_definitions(GGML_USE_HIP)
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(K_QUANTS_PER_ITERATION=${GGML_CUDA_KQUANTS_ITER})
if (GGML_HIP_UMA)
add_compile_definitions(GGML_HIP_UMA)
endif()
if (GGML_CUDA_FORCE_DMMV)
add_compile_definitions(GGML_CUDA_FORCE_DMMV)
endif()
if (GGML_CUDA_FORCE_MMQ)
add_compile_definitions(GGML_CUDA_FORCE_MMQ)
endif()