[llvm] [VPlan] CSE ScalarIVSteps recipes (PR #191307)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

> Do we not need to also hash in the InductionOpcode member somehow?

Ah yes, I initially thought that inferring the scalar type would be sufficient, but both Instruction::FAdd and Instruction::FSub can be InductionOpcodes, which would produce different results. Will fix, thanks.

https://github.com/llvm/llvm-project/pull/191307