[PATCH] AArch64: Use CMP;CCMP sequences for and/or/setcc trees.

[Matthias Braun]

Oh it is not just a performance but even a correctness issue. I will look into this (though as it is only the full gcc benchmark that fails this may take a while to reduce).

- Matthias

> On Jun 16, 2015, at 11:12 AM, Charlie Turner <charlie.turner at arm.com> wrote:
> 
> Hi Matthias,
> 
> | I can't reproduce any regressions on our hardware. I've been using the lnt
> + llvm-testsuite to run the benchmark. I only see 1.7% noise:
> 
> The lnt + llvm-testsuite runs the spec2006 benchmark in only test and train
> modes by default, not ref. It is in ref where the miscompare happens.
> 
> Regards,
> Charlie. 
> 
> 
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits