[llvm] r204699 - SLP vectorizer: Don't hoist vector extracts of phis.
atrick at apple.com
Wed Mar 26 14:47:57 PDT 2014
On Mar 26, 2014, at 2:15 PM, Tobias Grosser <tobias at grosser.es> wrote:
> On 03/25/2014 03:18 AM, Andrew Trick wrote:
>> Author: atrick
>> Date: Mon Mar 24 21:18:47 2014
>> New Revision: 204699
>> URL: http://llvm.org/viewvc/llvm-project?rev=204699&view=rev
>> SLP vectorizer: Don't hoist vector extracts of phis.
>> Extracts coming from phis were being hoisted, while all others were
>> sunk to their uses. This was inconsistent and didn't seem to serve a
>> purpose. Changing all extracts to be sunk to uses is a prerequisite
>> for adding block frequency to the SLP vectorizer's cost model.
>> I benchmarked the change in isolation (without block frequency). I
>> only saw noise on x86 and some potentially significant improvements on
>> ARM. No major regressions is good enough for me.
> Hi Andrew,
> my performance testers spotted a 37% performance regression for fp-convert. In the set of five relevant commits there are only debug info changes, lldb changes and your commit:
> I am not insisting to investigate this, but wanted to give you at least this information.
Thanks. I filed PR19258. When I benchmarked this I saw a slight improvement with fp-convert on sandybridge (2.6s from 2.8s). I don’t see any obvious difference with your builder other than the platform/machine. So I don’t know if this is just a flaky test. It would be great if someone could reproduce and explain what happened.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits