[PATCH] D81813: [ARM] MVE FP16 cost adjustments

[Sam Parker via Phabricator via llvm-commits]

samparker added a comment.

There's a lot of changes here... would it be worth committing them as separate patches: MVE, NEON and generic, or splitting of memory and cast?



================
Comment at: llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp:501
+    // and cost 1 vcvt for each lane, so long as the instruction is available.
+    // If not it will become a series of function calls.
+    const int CallCost = CostKind == TTI::TCK_RecipThroughput ? 10 : 3;
----------------
Maybe a comment on your magic number choices, or how about using getIntrinsicInstrCost for the call cost?


================
Comment at: llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp:985
+  if (ST->hasMVEFloatOps() && isa<FixedVectorType>(Src) && I &&
+      ((Opcode == Instruction::Load && I->hasOneUse() &&
+        isa<FPExtInst>(*I->user_begin())) ||
----------------
I thought you stance was that we shouldn't be looking at the context?!


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81813/new/

https://reviews.llvm.org/D81813