zmodem wrote:

Thanks for working on this!

> This patch introduces size-based preferred alignment for x86-64 global 
> variables, similar to the existing LargeArrayAlign mechanism for arrays.

I wasn't aware of the `LargeArrayAlign` mechanism, but apparently that was 
motivated by a requirement in the x86-64 psABI: 
https://github.com/llvm/llvm-project/issues/5971

I agree with Eli that it seems better to do this optimization in the backend. 
It should probably be easier to do it there too: you can use 
`GlobalObject::canIncreaseAlignment()` rather than replicate it, you could 
check if the variable is actually used by memcpy etc., and it will also benefit 
Rust code which is a slowly growing part of Chromium and presumably Edge too.

https://github.com/llvm/llvm-project/pull/185487
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to