<div dir="ltr">I would suggest reverting and discussing more.<div><br></div><div>I'm not really happy with the one-use hack to "fix" this. I think there should be a better way that more directly addresses the problem of extending the liveness of other SSA values, or essentially the canonicalization phase ordering problem that is here.</div>
</div><div class="gmail_extra"><br><br><div class="gmail_quote">On Thu, Mar 27, 2014 at 1:47 PM, Tobias Grosser <span dir="ltr"><<a href="mailto:tobias@grosser.es" target="_blank">tobias@grosser.es</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On 03/27/2014 08:42 PM, Reid Kleckner wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
On Thu, Mar 27, 2014 at 12:30 PM, Benjamin Kramer <<a href="mailto:benny.kra@gmail.com" target="_blank">benny.kra@gmail.com</a>>wrote:<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
On 27.03.2014, at 20:23, Erik Verbruggen <<a href="mailto:erikjv@me.com" target="_blank">erikjv@me.com</a>> wrote:<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Actually, I'm wondering if it's better to revert my patch (and yours).<br>
</blockquote>
The "nightly" test-suite pointed out a regression. If you have a tight for<br>
loop, like:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
for (i = 0; i <= 39; i++)<br>
doSomethingHere;<br>
<br>
then before the patch it would be:<br>
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
%exitcond = icmp eq i64 %indvars.iv.next, 40<br>
<br>
while with the patch it's:<br>
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
%exitcond = icmp eq i64 %indvars.iv, 39<br>
<br>
That keeps the "old" %indvars.iv alive. This causes some performance<br>
</blockquote>
tests to fail, like MultiSource/Benchmarks/<u></u>MiBench/telecomm-gsm.<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
I Cc-ed d0k because he did the review, so what do you guys think?<br>
</blockquote>
<br>
A simple fix would be to check if the add has just one use. I was hoping<br>
that this kind of workaround wouldn't be necessary anymore but it looks<br>
like we still need it in this case :|<br>
</blockquote>
<br>
<br>
I'm not really an optimizer person, so whatever works for you guys, revert<br>
or fix forward. I'm mostly concerned with the miscompile, and that has a<br>
test case now. :)<br>
</blockquote>
<br></div></div>
Yes, I am getting spammed by the LNT buildbots the whole day. Would you guys mind to either commit a fix now or just revert until the issue has been solved.<br>
<br>
Thank you,<br>
Tobias<div class="HOEnZb"><div class="h5"><br>
<br>
______________________________<u></u>_________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu" target="_blank">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/<u></u>mailman/listinfo/llvm-commits</a><br>
</div></div></blockquote></div><br></div>