lenary wrote:

> Since the Clang frontend is emitting the correct vector arithmetic and ABI 
> coercion IR, I think it might be better to address these codegen improvements 
> in subsequent backend patches. What are your thoughts on this?

Sounds about right to me.

One thing you could do is open a PR with more testcases. As you point out, the 
first case is covered by tests (though they need a TODO to say they should be 
using a different instruction, if possible). For the second case, adding `llc` 
tests along with a fixme that we need to get rid of the shifts would help too. 
:)

I briefly mentioned to craig that we might be able to use `cross-project-tests` 
to check that the intrinsics are producing the right instructions (when they 
eventually do so). This is a directory in LLVM where we can check all the way 
from clang through to codegen - as is done for both AMDGPU builtins and wasm 
intrinsics.

https://github.com/llvm/llvm-project/pull/181115
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to