ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-02-20 17:12:54 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

CMake: default to -arch=native for CUDA build (llama/10320)

Browse Source
This commit is contained in:
Johannes Gäßler 2024-11-17 09:06:34 +01:00 committed by Georgi Gerganov
parent 24ad19d0e9
commit 3c5c751174
1 changed files with 9 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
ggml/src/ggml-cuda/ggml/CMakeLists.txt
View File

@ -6,15 +6,18 @@ if (CUDAToolkit_FOUND)
message(STATUS "CUDA Toolkit found")
if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# 52 == lowest CUDA 12 standard
# 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)
# native == GPUs available at build time
# 52 == Maxwell, lowest CUDA 12 standard
# 60 == P100, FP16 CUDA intrinsics
# 61 == Pascal, __dp4a instruction (per-byte integer dot product)
# 70 == V100, FP16 tensor cores
# 75 == Turing, int6 tensor cores
if (GGML_NATIVE AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.6")
set(CMAKE_CUDA_ARCHITECTURES "native")
elseif(GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75")
else()
set(CMAKE_CUDA_ARCHITECTURES "52;61;70;75")
#set(CMAKE_CUDA_ARCHITECTURES "OFF") # use this to compile much faster, but only F16 models work
endif()
endif()
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")