[PATCH] D63447: [ARM] Put some of the TTI costmodel behind hasNeon calls.

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

Did you see D63448 <https://reviews.llvm.org/D63448> for getShuffleCost? I'm not early return would work there, but I could try moving the ST->hasNEON() check into each of the if (Kind == TTI::SK_Reverse). Although that's a bit of duplication it would make it work similarly to getCastInstrCost.

The other alternative it to move the MVEDupTbl table in D63448 <https://reviews.llvm.org/D63448> into the same block as NEONDupTbl is here, and switching which array is used on isMVE/isNeon inside the block. I'm not a big fan of that though, as it mangles the two architectures together more than I'd like and I feel like it makes things harder to understand (although would be smaller, the goal should be conceptually simpler, not smaller).


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63447/new/

https://reviews.llvm.org/D63447