While compiling ffmpeg with --enable-cuda-llvm on macos, it shows ERROR: cuda_llvm requested but not found

[Aniark]

ffmpeg version: latest git from master branch

llvm-mingw version : nightly and 20.1.0 final has been tried

error information from ffbuild/config.log:

BEGIN /var/folders/0j/bwqcs4y508s2n4ck4dhf3rpc0000gn/T//ffconf.cGW7qct2/test.c
    1   #include <stdbit.h>
    2   #include <assert.h>
    3   int main(void) { static_assert(__STDC_VERSION_STDBIT_H__ >= 202311L, "Compiler lacks stdbit.h"); return 0; }
END /var/folders/0j/bwqcs4y508s2n4ck4dhf3rpc0000gn/T//ffconf.cGW7qct2/test.c
x86_64-w64-mingw32-gcc -D_ISOC11_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -DWIN32_LEAN_AND_MEAN -U__STRICT_ANSI__ -D__USE_MINGW_ANSI_STDIO=1 -D_POSIX_C_SOURCE=200112 -D_XOPEN_SOURCE=600 -DPIC -std=c17 -fomit-frame-pointer -I/usr/local/include -Wdeclaration-after-statement -Wall -Wdisabled-optimization -Wpointer-arith -Wredundant-decls -Wwrite-strings -Wtype-limits -Wundef -Wmissing-prototypes -Wstrict-prototypes -Wempty-body -Wno-parentheses -Wno-switch -Wno-format-zero-length -Wno-pointer-sign -Wno-unused-const-variable -Wno-bool-operation -Wno-char-subscripts -Wno-implicit-const-int-float-conversion -O3 -fno-math-errno -fno-signed-zeros -mstack-alignment=16 -Qunused-arguments -Werror=implicit-function-declaration -Werror=missing-prototypes -Werror=return-type -c -o /var/folders/0j/bwqcs4y508s2n4ck4dhf3rpc0000gn/T//ffconf.cGW7qct2/test.o /var/folders/0j/bwqcs4y508s2n4ck4dhf3rpc0000gn/T//ffconf.cGW7qct2/test.c
/var/folders/0j/bwqcs4y508s2n4ck4dhf3rpc0000gn/T//ffconf.cGW7qct2/test.c:1:10: fatal error: 'stdbit.h' file not found
    1 | #include <stdbit.h>
      |          ^~~~~~~~~~
1 error generated.
ERROR: cuda_llvm requested but not found

windows and linux do not have this issue

[mstorsjo]

Please provide exact instructions for how to reproduce this. I tried configuring with --cross-prefix=x86_64-w64-mingw32- --arch=x86_64 --target-os=mingw32 --enable-cuda-llvm on macOS, with the latest llvm-mingw release added in $PATH, and it configured and built fine. Does one need to have some CUDA SDK installed as well, in order to hit this?

[Andarwinux]

As far as I know cuda-llvm only needs clang with nvtpx supported and ffnvcodec headers.

The errors shown in your config.log are not directly related to cuda-llvm, but that's understandable, ffmpeg's config.log often has misleading, irrelevant output, and you'd need to read the whole config.log to know what's really going on.

[Aniark]

@mstorsjo I'm tested on github actions, macos-13 macos-14 macos-15 all os above didn't work.

To reproduce, just try to compile on those runner os with latest or nightly llvm-mingw

git clone https://github.com/FFmpeg/FFmpeg
cd FFmpeg
./configure --cross-prefix=x86_64-w64-mingw32- --arch=x86_64 --target-os=mingw32 --enable-cuda-llvm