[llvm] [NVPTX] Add IR pass for FMA transformation in the llc pipeline (PR #154735)

[Artem Belevich via llvm-commits]

Artem-B wrote:

> are there any other backends besides NVPTX that currently support FMA vectorisation instructions?

I believe AVX2 on intel has FMA3 instruction:

https://github.com/llvm/llvm-project/blob/b8127cc8d0f5cb48c106199a059442e2719e8656/llvm/test/CodeGen/X86/fmaddsub-combine.ll#L18

https://github.com/llvm/llvm-project/pull/154735