[PATCH] D17288: [CodeGenPrepare] Do select to branch transform when cmp's operand is expensive.

[Chad Rosier via llvm-commits]

mcrosier added a comment.

In http://reviews.llvm.org/D17288#400669, @flyingforyou wrote:

> Hi Chad.
>
>   // Prefer likely predicted branches to selects on out-of-order cores.
>   if (Subtarget->isCortexA57() || Subtarget->isKryo())
>     PredictableSelectIsExpensive = true;
>   
>
> Recently, you changed `PredictableSelectIsExpensive ` flag on for Kryo. How do you think about this change?


Honestly, I was upstreaming on old internal patch and didn't do the analysis for the change.

My understanding is that this flags allows CGP to more aggressively convert selects to a series of branches.  This makes a great deal of sense on out-of-order cores with good branch predictors, which Kryo has..

Unfortunately, I haven't been following this review and I don't completely understand the problem you're trying to solve.  It seems Sanjay and Gerolf are providing good feedback, so I'm going to defer to their judgement.


http://reviews.llvm.org/D17288