[PATCH] D82091: [ARM][CostModel] Select instruction costs.

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

I think this should be reasonable. Can you explain where the cost of 2 is coming form? I would expect IT to be roughly free in a lot of places. ARM has conditional statements for pretty much everything. And armv8.1-m has csel :) (But, we don't actually use that at the moment..)

Do you have benchmark results? Not that you have to share them here, but do they look, um, reliably better?



================
Comment at: llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp:513
+  if (CostKind != TTI::TCK_RecipThroughput || !ValTy->isVectorTy()) {
+    // Simplift and assume more (scalar) instructions reduces throughput.
+    if (ISD == ISD::SELECT) {
----------------
Simplift ?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82091/new/

https://reviews.llvm.org/D82091