On Fri, Aug 5, 2016 at 10:43 PM, Jeff Law <law@redhat.com> wrote
On 08/05/2016 01:45 AM, Richard Biener wrote:
There is no need to avoid threading to a loop header, the threading
code can cope with this just fine. Noticed when working on PR72772.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2016-08-05 Richard Biener <rguenther@suse.de>
* tree-ssa-threadupdate.c (thread_block_1): Remove unnecessary
restriction on threading to a loop header.
This restriction exists to help prevent threading through a loop header into
the loop which can easily create an irreducible region.
In fact what you've done is revert the fix for 58343. Now there is of
course a test for 58343 in the regression suite.
It's entirely possible that 58343 is just latent now or other changes in
tree-ssa-threadupdate.c prevent the problems that we saw with 58343.
Well, I removed the restriction that the forward threading path ends in a loop
header. I did this to restore threadings we've done when there was a forwarder
before the loop header we threaded to -- a forwarder with a degenerate PHI
that was later removed and merged with the header anyway. So I basically
changed nothing here. I also don't see how forward threading to a loop header
(not across it) can cause an irreducible region to appear - the other part of
the condition which I retained certainly guards the case of crossing a
loop header.
But maybe I am missing something.