[PATCH] D15559: [SCEVExpander] Make findExistingExpansion smarter

[Junmo Park via llvm-commits]

flyingforyou marked an inline comment as done.
flyingforyou added a comment.

I ran test-suite, spec on Galaxy S6 Target(CPU 2.1Ghz, MIF 1.5Ghz, Of course it's developtment board. Not exactly same with Galaxy S6). 
I also can test on Juno. But it's very slow and I want to test on real world environment.

In my test, I got different data.

> Charlie's test result

>  spec on a Cortex-A57 (A64):

>  spec.cpu2000.ref.253_perlbmk 	4.71%

>  spec.cpu2000.ref.256_bzip2 	2.07%

>  spec.cpu2006.ref.433_milc 	1.24%

>  spec.cpu2000.ref.254_gap 	-1.76%




> (I'm focussing on SPEC because it's less noisy than the benchmarks in test-suite, although I do note a 23% improvement in lnt.MultiSource/Benchmarks/MiBench/automotive-susan/automotive-susan and a 7.76% regression in lnt.SingleSource/Benchmarks/Shootout-C++/ackermann)




  My test result about Spec (Opt/Ori)
  spec.cpu2000.ref.253_perlbmk 	-> I can't get result. I think without modifying source code, we can't compile it.
  spec.cpu2000.ref.256_bzip2 	99.93% 
  spec.cpu2006.ref.433_milc 	99.52%
  spec.cpu2000.ref.254_gap 	99.18%



  My test result about test-suite (Opt/Ori)
  I got totally different result on this.
  SingleSource/Benchmarks/Shootout-C++/ackermann	81.35%
  MultiSource/Benchmarks/MiBench/automotive-susan/automotive-susan -> execution time is 0. It's same.

I think both of data also can be worth. And This data is not meaningful any more. As you know, the patch was completely changed. I will share the result data base on recent patch soon.

PS. Could you share how to build "spec.cpu2000.ref.253_perlbmk", please?


http://reviews.llvm.org/D15559