[llvm] [VectorUtils] Trivially vectorize ldexp, [l]lround (PR #145545)

[David Green via llvm-commits]

davemgreen wrote:

I would expect the costs to be something like 10 for a libcall, and 10*4+scalarization_overhead for the vector. They might need to be overriden on AArch64 as they are marked custom, but I don't see them in BasicTTIImplBase::getTypeBasedIntrinsicInstrCost.

Can you add tests for SLP / scalarizer, as in https://reviews.llvm.org/D124358 or https://github.com/llvm/llvm-project/pull/71416/files?

https://github.com/llvm/llvm-project/pull/145545