ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-31 03:29:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Native build ooption for CUDA when GGML_NATIVE is set (#280)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-03-22 18:17:51 +01:00
committed by GitHub
parent 13ecc5332e
commit 6028362ef6
1 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/CMakeLists.txt
View File

@@ -297,10 +297,12 @@ if (GGML_CUDA)
# 60 == FP16 CUDA intrinsics
# 61 == integer CUDA intrinsics
# 70 == compute capability at which unrolling a loop in mul_mat_q kernels is faster
if (GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75")
if (GGML_NATIVE AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.6" AND CMAKE_VERSION VERSION_GREATER_EQUAL "3.24")
set(CMAKE_CUDA_ARCHITECTURES "native")
elseif (GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75;80")
else()
set(CMAKE_CUDA_ARCHITECTURES "52;61;70;75")
set(CMAKE_CUDA_ARCHITECTURES "50;61;70;75;80")
#set(CMAKE_CUDA_ARCHITECTURES "OFF") # use this to compile much faster, but only F16 models work
endif()
endif()