[llvm] r200621 - LoopVectorizer: Enable unrolling of conditional stores and the load/store

[Nadav Rotem]

The E5430 is a Penryn. We haven’t tuned the vectorizer for anything older than Sandybridge. 

On Feb 3, 2014, at 2:42 PM, Tobias Grosser <tobias at grosser.es> wrote:

> On 02/03/2014 07:14 PM, Arnold Schwaighofer wrote:
>> I cannot reproduce this regression on my side.
>> 
>> On a Sandy-bridge machine I don’t see a regression for -O3 (without -mavx). The only difference between before/after in my test is that we unroll one vectorized loop by two:
> 
> It was run on an:
> 
> Intel(R) Xeon(R) CPU E5430  @ 2.66GHz
> 
> using plain 'clang -O3'.
> 
> I just reproduced the regression though it is rather close and there is some variance, but after several runs it becomes clear.
> 
> I attached the LLVM-IR before and after the commit. It is just a single change (and seems to be the one you have also seen).
> 
> I did not run the -mavx numbers and personally don't worry about this slight performance change (posted numbers just for information). In case you want to investigate yourself, I can give you SSH access on the machine.
> 
> Tobias
> <old.ll><new.ll>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits