[PATCH][LoopVectorizer] Restrict the unroll factor of reductions in loops

Gerolf Hoflehner ghoflehner at apple.com
Wed Aug 13 21:25:43 PDT 2014


There are a number of regression (lowercase -40%, link pack ~-5%) and gains (puzzle? >30%, sgefa > 13%). I’ll do some analysis in the next couple of days and follow up with bug filings as necessary. So go for the gains for now and trust that we’ll recover the losses.

I haven’t done any compile-time measurements for the patch. Did you cover it with high confidence?

Thanks
Gerolf



On Aug 12, 2014, at 10:19 AM, Arnold Schwaighofer <aschwaighofer at apple.com> wrote:

> LGTM.
> 
> I did see some regressions running the whole test-suite on the two patches (+ "ST->isCyclone()") we (on our end) might want to look into - I’ll send my results to Gerolf.
> 
> 
> Thanks,
> Arnold
> 
> 
>> On Aug 11, 2014, at 9:17 AM, James Molloy <james at jamesmolloy.co.uk> wrote:
>> 
>> Hi Arnold, Gerolf, Hal,
>> 
>> Good idea about a tuning option. Attached is a patch that implements that tuning option. Is it OK to commit?
>> 
>> Gerolf, did you want me to add "|| isCyclone()" to AArch64TargetTransformInfo?
>> 
>> Cheers,
>> 
>> James
>> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140813/a64e0fc0/attachment.html>


More information about the llvm-commits mailing list