Talk:ROCm
Before creating a discussion or leaving a comment, please read about using talk pages. To create a new discussion, click here. Comments on an existing discussion should be signed using
~~~~
:
A comment [[User:Larry|Larry]] 13:52, 13 May 2024 (UTC) : A reply [[User:Sally|Sally]] 12:35, 5 November 2024 (UTC) :: Your reply ~~~~
Broken offload
Something is definitely utterly broken with OpenMP offload:
$ /usr/lib/llvm/15/bin/clang++ --rocm-device-lib-path=/usr/lib/amdgcn/bitcode --libomptarget-amdgpu-bc-path="/usr/lib64" -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -x c openmp_offload.c
clang-15: error: unable to execute command: Executable "clang-linker-wrapper" doesn't exist!
clang-15: error: linker command failed with exit code 1 (use -v to see invocation)
Barely documented but found another report here => https://bugs.archlinux.org/task/77814
Reported as Gentoo bug #904143
--Admnd (talk) 17:36, 10 April 2023 (UTC)
The upcoming GCC 13 seems start having AMD (only gfx90a GPUs at this time) capabilities.[1]
--Admnd (talk) 16:26, 18 April 2023 (UTC)
Regarding 'pytorch[rocm]'
The Gentoo repository does not appear to have 'sci-libs/pytorch[rocm]', and while I see it on some overlays, the article itself here doesn't appear to mention overlays being used. Unfortunately, looking at the ebuilds on those overlays, and the CMakeLists of the project itself, it doesn't seem like too much fun to add either. —Chiitoo (talk) 11:12, 29 September 2024 (UTC)
Regarding Supported AMD GPUs
I wonder if the supported GPUs could have some expansion. Looking at the upstream list right now, they seem to only support some of the very latest hardware, and everything older than RDNA3 is "The current ROCm release does not support this hardware. The HIP runtime might continue to run applications for an unsupported GPU, but prebuilt ROCm libraries are not officially supported and will cause runtime errors.". —Chiitoo (talk) 11:14, 29 September 2024 (UTC)