Arch Linux User Repository

@npreining You may be correct. '-Xcompiler=-Wl\,-O1' -Wl,-O1 -Wl seems like the possible troublemaker, which is an nvcc flag. It passes these arguments to the compiler, and it's possible Arch is somehow muddling it. It does look like there's an erroneous apostrophe there past the first -O1. That would explain the "Unknown option" error I'm getting, but I'm not sure why this is only happening when building with a PKGBUILD.

That flag above was pulled from the non-truncated version of the linker error I'm getting:
https://paste.cachyos.org/p/49b1846.txt

@Joomzie yes I have the same findings. I have a local branches that use either autoconf or ninja, but in both cases the compilation ends with the same error you mentioned. And yes, the patch should be removed now, and all modules pulled, you are correct.

My only hinch is that Arch injects some arguments along the line that make the build fail, but I faced the same problem setting various flags, too.

Very puzzling.

So, I finally got some time to revisit this, and my original hunch was incorrect. There does seem to be an issue with the PKGBUILD. I was able to clone, and build the master using gcc 16, without any problems. I just followed the build instructions from the Git, so I'm not entirely sure why the PKGBUILD fails.

❯ ./build/nvencc --check-hw
NVEnc (x64) 9.16 (r3587) by rigaya, May 16 2026 12:14:51 (gcc 16.1.1/Linux)
  [NVENC API v13.0, CUDA 13.2]
 reader: raw, y4m, avs, vpy, avsw, avhw [H.264/AVC, H.265/HEVC, MPEG2, VP8, VP9, VC-1, MPEG1, MPEG4, AV1]
 others
  nvml       : yes
  nvrtc      : yes
  nvvfx      : no
  ngx        : no
  nvof fruc  : no
  libass     : yes
  libdovi    : yes
  vulkan     : yes
  libplacebo : yes
  vmaf       : yes
#0: NVIDIA GeForce RTX 3060 (3584 cores, 1807 MHz)[PCIe4x16][595.71]
Avaliable Codec(s)
H.264/AVC
H.265/HEVC

I noticed that there now seems to be a build dep for dtl, which Arch doesn't have package for. There is a submodoule for it, though, so I cheated a little, and modified the PKGBUILD to pull all submodules. I also set it up to use meson, and removed the ldflags patch, as that seemed to be causing problems, as well, but I'm now hitting a problem at the very end with the linker. Below is the diff for the PKGBUILD, as well as the error I'm getting. Functionally speaking, my changes are no different from the instructions from the repo.

Diff

[154/154] Linking target nvencc
FAILED: [code=1] nvencc
...
nvcc fatal   : Unknown option '-Wl,-O1'
ninja: build stopped: subcommand failed.

Been getting a build failure for the past couple of days, but it may be an upstream issue.

NVEncCore.cpp:(.text+0x28065): undefined reference to `NVEncFilterMsharpen::NVEncFilterMsharpen()'
/usr/bin/ld: NVEncCore.cpp:(.text+0x280ba): undefined reference to `vtable for NVEncFilterParamMsharpen'
/usr/bin/ld: NVEncCore.cpp:(.text+0x29e2c): undefined reference to `NVEncFilterMsmooth::NVEncFilterMsmooth()'
/usr/bin/ld: NVEncCore.cpp:(.text+0x29e81): undefined reference to `vtable for NVEncFilterParamMsmooth'
/usr/bin/ld: warning: creating DT_TEXTREL in a PIE
collect2: error: ld returned 1 exit status
make: *** [makefile:18: nvencc] Error 1
==> ERROR: A failure occurred in build().
    Aborting..

I was able to get a successful build the other day, which is what makes me think this might be an upstream issue. Prior to this successful build, it wasn't looking for cuda in /opt, which seemingly corrected itself.

❯ yay -Qi nvenc-git | grep -i build
Build Date      : Fri 17 Apr 2026 02:02:28 PM CDT

Still, figured it'd be worth reporting this current failure here.

Sorry for the belated response, but thanks for putting this up. Figured it'd be more appropriate to leave this here. :)

You should probably build not 'master' but rather 'linux' branch.