[PATCH] Add optimization remarks to the loop unroller and vectorizer.
dnovillo at google.com
Wed Apr 23 10:06:27 PDT 2014
On Tue, Apr 22, 2014 at 9:24 PM, Hal Finkel <hfinkel at anl.gov> wrote:
> Thinking about it, it is probably pretty rare that no other pass splits
the preheader. Nevertheless, I'd prefer to be smart about it if practical.
> some way we can choose the "last" debug location?
Sure. I've changed it to anchor on the terminating instruction of the
preheader or header (whichever gets a debug loc). I suspect that we will
not always get a clean spot. This depends a lot on how much optimizations
manage to preserve !dbg metadata. But we can tackle those oddities as we
> I like this better.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits