bogner wrote:

> I think this can cause issues in the spirv backend. I believe some 
> optimizations will replace short memcopy with a load and store of integers 
> regardless of the actual type of the memory. That does not translate well to 
> spirv.

I guess you're referring to the buggy logic in 
InstCombinerImpl::SimplifyAnyMemTransfer that assumes all integer types are 
appropriate to replace small memcpys with. We had an attempt to (partially) fix 
that in #169966 that was rejected by upstream in preference to wait for the 
[byte type](https://blog.llvm.org/posts/2025-08-29-gsoc-byte-type/) which will 
nominally resolve this. I guess that means we should block the issue here like 
we did with #165753 and wait for the byte type stuff to be finished before 
cleaning this up.

https://github.com/llvm/llvm-project/pull/173093
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to