[llvm-commits] [PATCH] LoopVectorizer

Nadav Rotem nrotem at apple.com
Wed Oct 17 20:48:14 PDT 2012


Yes,  you are correct.  I missed it at first. This is something that we should do. 

On Oct 17, 2012, at 8:47 PM, Hal Finkel <hfinkel at anl.gov> wrote:

> ----- Original Message -----
>> From: "Nadav Rotem" <nrotem at apple.com>
>> To: "Hal Finkel" <hfinkel at anl.gov>
>> Cc: "llvm-commits at cs.uiuc.edu LLVM" <llvm-commits at cs.uiuc.edu>, "Ralf Karrenberg" <karrenberg at cdl.uni-saarland.de>
>> Sent: Wednesday, October 17, 2012 10:32:20 PM
>> Subject: Re: [llvm-commits] [PATCH] LoopVectorizer
>> 
>> 
>> 
>> 
>> On Oct 17, 2012, at 6:24 PM, Hal Finkel < hfinkel at anl.gov > wrote:
>> 
>> 
>> he current pass handles only fixed trip-count (single-exit) loops
>> without inter-iteration dependencies, correct? For this class of
>> loops, I believe that what I proposed is correct. Maybe I'm thinking
>> about this incorrectly, can you please provide a counterexample?
>> 
>> 
>> Yes. If there are no cross-iteration dependencies then x equals to
>> the last SIMD lane. I missed the part which said that there are no
>> cross-iteration deps.
> 
> Right; my point was that this is exactly the class of loops currently handled by the loop vectorizer. If that is correct, then it would be simple to handle these use-of-last-iteration cases as well.
> 
> Thanks again,
> Hal
> 
>> 
> 
> -- 
> Hal Finkel
> Postdoctoral Appointee
> Leadership Computing Facility
> Argonne National Laboratory




More information about the llvm-commits mailing list