[PATCH] D16569: Emit line 0 line information for interesting 'orphan' instructions

Frederic Riss via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 26 15:15:55 PST 2016

friss added a comment.

In http://reviews.llvm.org/D16569#336264, @probinson wrote:

> Smooth.
>  I've asked Wolfgang to run a size experiment.


> Can we really have a case where BB will change but there's no label?

In my experiments, yes. Maybe it's about fall through-only BBs, or something more subtle in the MC semantics, but I definitely needed to add a specific case for entry in a new BB.

> Post-call cases...  not as important as top-of-block cases, I think.  Also, harder to detect reliably. Local-value instructions might not be immediately following the call instruction, if the call returns a value that has to be moved somewhere else or there's other post-call state to restore

>  I think it's not horrible if post-call local-value instructions end up implicitly attached to the call's source location.

To be honest, this wasn't in my first version of the patch, but when I looked back at the other thread, I saw that it was addressing this issue too. I do not feel strongly about that, I definitely could go without that part of the patch.


More information about the llvm-commits mailing list