armv7-sony-vita-newlibeabihf should be named thumbv7a-sony-vita-newlibeabihf

[ChrisDenton]

Looking through our target names I notice armv7-sony-vita-newlibeabihf is actually a thumb target (it uses the llvm target thumbv7a-vita-eabihf). So I think it should be name thumbv7a-sony-vita-newlibeabihf to match our other thumb targets?

cc target maintainers: @nikarh @pheki @zetanumbers

[zetanumbers]

I mean you are kinda right. Ideally you could make thumbv7a-sony-vita-newlibeabihf and armv7a-sony-vita-newlibeabihf with thumb instructions disabled. Buuut why would we even need a vita target without thumb instructions anyway?

[zetanumbers]

While renaming a target would break the old code, which I think is not justified just to make target name consistent.

[ChrisDenton]

I'm not so interested in consistency just for consistencies sake. However, while targets triples are very far from well specified, we can (at least for Rust targets) rely on the the arch being right. So anything using the target triple to infer target properties has to know to make an exception for this one target.

[pheki]

I personally think it would be good to get this fixed, the earlier the better. The reason the target name doesn't match the arch is because the arch was originally set as armv7a.

I do agree with @zetanumbers' though, and would like to avoid breakage, especially around tooling (not too worried about code target_os is generally used for conditional compilation). I wonder if we could do something like what was done for wasm32-wasi:

• duplicate the target into thumbv7a-sony-vita-newlibeabihf and deprecate the old target
• fix tooling, docs, ci that use the old target name
  • for cargo vita, can we detect the nightly version and use the appropriate target?
• remove the old target after a few months

[VelocityRa]

(it uses the llvm target thumbv7a-vita-eabihf)

There's no upstream LLVM target called thumbv7a-vita-eabihf so I'm wondering what this is referring to?

Sorry if out of topic, trying to find information on what ABI commercial PS Vita games abide by (how it differs from standard ARMv7 EABI) and run across this.

[pheki]

There's no upstream LLVM target called thumbv7a-vita-eabihf so I'm wondering what this is referring to?

We're in the process to change it, probably to thumbv7a-sony-vita-eabihf at #138426

You can read more about how target triples work in Clang/LLVM here: https://clang.llvm.org/docs/CrossCompilation.html#target-triple

So thumbv7a-sony-vita-eabihf would be:

• arch: thumb
• sub: v7a
• vendor: sony
• sys: vita
• env: eabihf

sony and vita do not currently exist in upstream, so it gets parsed as unknown.

Sorry if out of topic, trying to find information on what ABI commercial PS Vita games abide by (how it differs from standard ARMv7 EABI) and run across this.

AFAIK there are some differences in definitions for C types, like:

• char default to signed by default, while in Arm char defaults to unsigned
• wide chars (wchar_t) are 16-bit

Not sure if there are other differences, you can find the target config we use for Rust here: https://github.com/rust-lang/rust/blob/2ea33b591050c4ca1a3752830b29112638faecf6/compiler/rustc_target/src/spec/targets/armv7_sony_vita_newlibeabihf.rs

[VelocityRa]

@pheki Thanks, I got some help for this from @CreepNT in the HENKaku Discord.

Besides what you said:

hard-float AAPCS (or rather, ATPCS)
r9 as callee-saved GPR