Can Accel use the ptx-linker?

[bheisler]

Accel already implements some of the features of https://github.com/denzp/rust-ptx-linker, but it seems that the linker handles more cases (eg mangled names containing invalid characters). Could Accel simply use the linker? This would save duplicate work.

[termoshtt]

It will not be simple, but I plan to switch to it.

accel/nvptx crate tried to switch from out-of-tree compile of rustc (using xargo) to in-tree compile of rustc. nvptx also contains patches of libstd (https://github.com/rust-accel/rust/pull/1)

Name mangling issue is resolved in the code generation phase in accel while rust-ptx-linker resolve it on the link phase. I intend to use ptx-linker as a default linker of nvptx64-nvidia-cuda target like Cortex-M team does

https://users.rust-lang.org/t/psa-cortex-m-breakage-lld-as-the-default-linker/19890

Using ptx-linker, we do not need to expand, convert, and link *.rlib to *.bc. It will make nvptx crate much simpler.