tltao wrote:
Hmm looks like the testing reveals another problem. We have no way of
invalidating a platform specific constraint that also uses `{}`.
The AMDGCN tests expect `{exec}a` to fail to verify, but it does not fail in
the generic `validateHardRegisterAsmConstraint` since we want to supp
tltao wrote:
I've made the changes as discussed to support multiple hard register
constraints.
https://github.com/llvm/llvm-project/pull/85846
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe
tltao wrote:
> I'm not sure I follow what you mean here; does this mean that if you have a
> register asm variable, it might not end up in that register? That seems like
> a regression for existing code.
Ah, good point. I was thinking more in line of mismatching register classes,
but you're r
tltao wrote:
Thanks for all the comments. I think from @stefan-sf-ibm's response, it sounds
like the GCC side is not fully decided on what to do with multiple constraints
involving hard registers, so we can wait for the details to be hashed out
further there.
In the meantime, I plan on makin
tltao wrote:
> If you do want to reject, please make sure we reject gracefully at both the
> clang level and the LLVM IR level.
Is there a reason we also need to check it in the LLVM IR level if we are
already checking for it in clang? This PR only affects the Clang output so I
don't know for
tltao wrote:
Ping. Restarting this topic as GCC have implemented the same feature.
https://gcc.gnu.org/onlinedocs/gcc/Hard-Register-Constraints.html
https://github.com/llvm/llvm-project/pull/85846
___
cfe-commits mailing list
[email protected]
https://github.com/tltao edited https://github.com/llvm/llvm-project/pull/85846
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
https://github.com/tltao edited https://github.com/llvm/llvm-project/pull/85846
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
