Proposal
The nvptx-nvidia-cuda (32-bit) target is no longer available in the Rust compiler, and 32-bit ptx is not well supported by Nvidia. The nvptx64-nvidia-cuda is the target used for generating 64-bit ptx which is fully supported by Nvidia.
Unfortunately the code for generating ptx function signatures contains a lot of bugs in the way args are passed to LLVM, this goes for both 32-bit and 64-bit ptx. I have started fixing these bugs for the nvptx64 target but have not yet touched the nvptx target. My proposal is to remove the nvptx codegen module (rustc_target::abi::call::nvptx) entirely instead of fixing the bugs there as well.
Since the nvptx-nvidia-cuda target is removed from rustc it is not trivial to rely on the target. Since the code contains quite a lot of breaking bugs in function arguments codegeneration and 32-bit ptx is not well supported by Nvidia it reinforces the assumption that nobody is relying on this code.
Mentors or Reviewers
Removing this code should be very simple and mentoring should not be needed. @RDambrosio016 will be consulted for looking through the PR as he his working on rust-cuda and is more knowledgeable about this topic than myself. I do not know who will be the most suitable reviewer for this change. @davidtwco is assigned to the bugfix PR and it might make sense to have the same person assigned to several nvptx related things.
Process
The main points of the Major Change Process are as follows:
You can read more about Major Change Proposals on forge.
Comments
This issue is not meant to be used for technical discussion. There is a Zulip stream for that. Use this issue to leave procedural comments, such as volunteering to review, indicating that you second the proposal (or third, etc), or raising a concern that you would like to be addressed.
Proposal
The
nvptx-nvidia-cuda(32-bit) target is no longer available in the Rust compiler, and 32-bit ptx is not well supported by Nvidia. Thenvptx64-nvidia-cudais the target used for generating 64-bit ptx which is fully supported by Nvidia.Unfortunately the code for generating ptx function signatures contains a lot of bugs in the way args are passed to LLVM, this goes for both 32-bit and 64-bit ptx. I have started fixing these bugs for the nvptx64 target but have not yet touched the nvptx target. My proposal is to remove the nvptx codegen module (
rustc_target::abi::call::nvptx) entirely instead of fixing the bugs there as well.Since the
nvptx-nvidia-cudatarget is removed from rustc it is not trivial to rely on the target. Since the code contains quite a lot of breaking bugs in function arguments codegeneration and 32-bit ptx is not well supported by Nvidia it reinforces the assumption that nobody is relying on this code.Mentors or Reviewers
Removing this code should be very simple and mentoring should not be needed. @RDambrosio016 will be consulted for looking through the PR as he his working on rust-cuda and is more knowledgeable about this topic than myself. I do not know who will be the most suitable reviewer for this change. @davidtwco is assigned to the bugfix PR and it might make sense to have the same person assigned to several nvptx related things.
Process
The main points of the Major Change Process are as follows:
@rustbot second.-C flag, then full team check-off is required.@rfcbot fcp mergeon either the MCP or the PR.You can read more about Major Change Proposals on forge.
Comments
This issue is not meant to be used for technical discussion. There is a Zulip stream for that. Use this issue to leave procedural comments, such as volunteering to review, indicating that you second the proposal (or third, etc), or raising a concern that you would like to be addressed.