[clang] [clang][test] Add `%clang_cc1_arm64_neon` substitution (PR #188547)

Andrzej Warzyński via cfe-commits cfe-commits at lists.llvm.org
Tue Apr 7 12:12:53 PDT 2026


banach-space wrote:

> > > So there is no Windows-specific testing needed?
> > 
> > 
> > There is currently much less Windows-specific coverage in this area, so I think it is reasonable to keep those tests separate for now.
> 
> Good to know, thanks! I suppose that's fine, but I do worry that we'll then end up with three more substitutions later for Windows environments, too.
> 

I agree that we should try to avoid an explosion of substitutions. TBH, I wasn't planning (nor expecting) Windows to be an issue - AFAIK, there are no plans to expand those tests to Windows.

Also, we can also avoid creating new aliases by e.g. tweaking the definition of `%clang_cc1_cg_arm64_neon` depending on sub-directory. As in, we could have:
* clang/test/CodeGen/AArch64/linux/lit.local.cfg,
* clang/test/CodeGen/AArch64/windows/lit.local.cfg.

Or, similarly (note - we are creating these sub-dirs anyway):
* clang/test/CodeGen/AArch64/neon/lit.local.cfg,
* clang/test/CodeGen/AArch64/sve/lit.local.cfg,
* clang/test/CodeGen/AArch64/sme/lit.local.cfg.

In fact, it will be much easier to tweak to overall set-up once the tests start using substitutions.

> FWIW, `-Wall` flags are not purely sema related but are mostly sema related. e.g., we do have diagnostics which we emit only at codegen time, but I'd expect those tests to pass `-verify` in those cases, and I wouldn't expect `-Werror`. `-flax-vector-conversions` is handled mostly in Sema but changes which implicit conversions happen and so I can sort of see it as a codegen option but I think we could test that without requiring codegen tests (e.g., we could write tests which match the AST as opposed to codegen, if we wanted).
> 
> I'm not convinced those flags are so problematic in the RUN line that it's better to hide them from people reading the test.

👍🏻 

Let me just remove `%sema_flags` entirely, that won't impact the expected output (i.e. code-gen).

https://github.com/llvm/llvm-project/pull/188547


More information about the cfe-commits mailing list