asi-sc wrote:

> Thoughts on making this a function attribute, instead of trying to directly 
> pass it to the linker?

I'd say this PR is a quick fix for a common issue users face when trying to use 
`falign-loops` option -- it should not introduce (I hope) any additional 
issues. It doesn't fix all the problems though: for example, in two-step 
compilation with LTO enabled, users must not forget to pass this option at the 
linking step; the option passed to the first step (.c -> .o) is silently 
ignored.

I agree that the proper solution would likely require the use of a function 
attribute or some metadata (I think `[[clang::code_align(N)]]` use it). That 
would allow us to provide diagnostics if users tried to apply different 
alignments accross compilation steps, handle alignments merging, or something 
else. Unfortunately, I don’t have the bandwidth to take on that larger redesign 
right now.

https://github.com/llvm/llvm-project/pull/181702
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to