[Bug(CMake 3.17)] nvcc fatal : Unknown option '-mf16c'

[fumiama]

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• I am running the latest code. Development is very rapid so there are no tagged versions as of now.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new bug or useful enhancement to share.

Expected Behavior

Successfully compiled.

cmake3 --build . --config Release

Current Behavior

Got error as title.

$ cmake3 --build . --config Release
-- cuBLAS found
-- CMAKE_SYSTEM_PROCESSOR: x86_64
-- x86 detected
-- GGML CUDA sources found, configuring CUDA architecture
-- Configuring done
-- Generating done
-- Build files have been written to: /users/fumiama/src/llama.cpp/build
[  4%] Building CUDA object CMakeFiles/ggml.dir/ggml-cuda.cu.o
nvcc fatal   : Unknown option '-mf16c'
gmake[2]: *** [CMakeFiles/ggml.dir/ggml-cuda.cu.o] 错误 1
gmake[1]: *** [CMakeFiles/ggml.dir/all] 错误 2
gmake: *** [all] 错误 2

Environment and Context

• Physical (or virtual) hardware you are using: Centos7.6

• Operating System, e.g. for Linux:

$ uname -a
Linux localhost.localdomain 3.10.0-1160.88.1.el7.x86_64 #1 SMP Tue Mar 7 15:41:52 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

• SDK version, e.g. for Linux:

$ cmake3 --version
cmake3 version 3.17.5

CMake suite maintained and supported by Kitware (kitware.com/cmake).

$ gcc --version
gcc (GCC) 6.3.1 20170216 (Red Hat 6.3.1-3)
Copyright (C) 2016 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Fri_Jan__6_16:45:21_PST_2023
Cuda compilation tools, release 12.0, V12.0.140
Build cuda_12.0.r12.0/compiler.32267302_0

[slaren]

I guess that the add_compile_options need to be filtered to not apply to cuda, similar to what was done for MSVC.

[fumiama]

I guess that the add_compile_options need to be filtered to not apply to cuda, similar to what was done for MSVC.

Sure and according to my latest test, cmake 3.26.3 does not have this issue. Maybe it only happens under cmake 3.17.x.

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.