I'm currently writing some OpenCL code (in Linux). For code that works fine on a Intel CPU the following error arrises when I try to use the AMD GPU:
AMDGPUMCInstLower::lower - Pseudo instruction doesn't have a target-specific version: 317
This code produces the error for example (but only if lines 33 and 35 are there, if they are commented out the error goes away: https://paste.pound-python.org/show/k4unI4Bf1TVDEiAZj7be/
The error occurs also in other codes in (like it seams) unrelated positions but sometimes with other numbers at the end.
I could not find anything about this type of error. Updating LLVM and Mesa to a newer version did also not change anything.
What exactly does this error mean and how can I get rid of it? It causes the build to fail and makes OpenCL essentially unusable at the moment.
My apologies for this late reply.
The concerned team needs some debugging information. It would be helpful if you could run the program with CLOVER_DEBUG=native,llvm and post the output.