jhuber6 added a comment. In D107668#3464124 <https://reviews.llvm.org/D107668#3464124>, @Naghasan wrote:
> Any rational for not calling `addTempFile` during the binding rather than the > call setup ? I had a brief look there and it seems to me this is the root of > the problem. > > I got bitten by this patch in a downstream project where some of the files in > the offload action are actual inputs and so got deleted. Sorry to hear that. If I recall the main problem was that the bindings always use the `.o` suffix while CUDA requires the `.cubin` suffix to work. There was a similar hack somewhere else so I just copied it here and didn't think too much about it. If you have a solution that solves your problem I could take a look at it. However, we're moving away from this tool anyway, upstream it's no longer used for OpenMP offloading and I'm currently working on making that true for HIP as well. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D107668/new/ https://reviews.llvm.org/D107668 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
