update address taken: don't drop clobbers
Jeff Law
law@redhat.com
Thu Oct 16 17:23:00 GMT 2014
On 10/16/14 05:20, Richard Biener wrote:
>
> It doesn't even know that it has no uses (the variable still needs to be
> written into SSA form). OTOH it is a missed DSE opportunity before
> update-address-taken?
Perhaps it's a missed DSE prior to update-address-taken, but I suspect
there aren't many of those in general... It wouldn't be hard to shove
in a DSE pass to get some instrumentation across a wider range of code.
>
> As of premature optimization - into-SSA could notice it created SSA
> names with no uses and trigger a fast DCE.
Yea, I'd been pondering that as well. I'm curious how often we have
trivially dead code after into-ssa. If it's most of the time, then
maybe we just schedule the DCE right after and not even bother to check
if we've got SSA_NAMEs without any uses.
>
> Btw, I wonder what this odd folding of variadic builtins is about, and why
> it is not done in the stdarg pass (and only there), which would be earlier.
No clue. I've tried pretty hard not to think about the lowering of
variadic builtins. Perhaps the author of that code was trying to keep
things simple and let the optimizers to the optimization? :-)
jeff
More information about the Gcc-patches
mailing list