[llvm] [AArch64][CostModel] Fix stale costs for "ctpop" Intrinsic (PR #191769)

[Yashwant Singh via llvm-commits]

yashssh wrote:

Thanks for the review @davemgreen! Are you suggesting instead of hardcoding scalar costs in table I should add something like below?
```
  if (!RetTy->isVectorTy()) {
    Type *VecTy = VectorType::get(RetTy, 64 / RetTy->getScalarSizeInBits(), false);
    InstructionCost Cost =
        getVectorInstrCost(Instruction::InsertElement, VecTy, CostKind, 0, nullptr, nullptr) +
        getIntrinsicInstrCost({Intrinsic::ctpop, {VecTy}, {}}, CostKind) +
        getVectorInstrCost(Instruction::ExtractElement, VecTy, CostKind, 0, nullptr, nullptr);
    return Cost;
  }
  ```
This ends up giving even bigger costs
  
  >  See https://davemgreen.github.io/codesize.html?from=thru&instr=ctpop for the list of all the instructions.
  
Ah yes, I myself ended up here doing a similar comparison between TTI costs and LLVM-MCA, although I am focused more on Grace and Olympys.

https://github.com/llvm/llvm-project/pull/191769