[PATCH] D29660: [OpenMP] Add flag for overwriting default PTX version for OpenMP targets

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

gtbercea added a comment.

Hi Alex,

I have no idea what the errors you mention are. I worked until very late last night to fix the previous error with the compute capability. Maybe fixing that unlocked some other errors.

I would need a lot of help to figure out what's going on since I probably can't reproduce them locally. At the moment the turnaround time for getting feedback for these errors is way too long since I have to wait for the entire test suite to run every time I push something. Could you run everything you reverted locally and check what each failing compile line produces please? Being able to see the output is the first line of attack. I would also need to know what GPUs you have on the test systems that fail: which compute capability they support, what CUDA version you have installed if any etc.

I am extremely pressed for time so I need to land these patches as soon as possible - today at the latest.

Thanks a lot for your help,

--Doru


Repository:
  rL LLVM

https://reviews.llvm.org/D29660