[llvm] [AArch64] Generate fmls for fneg+fmul (PR #215338)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

Do you have an example of where it increased codesize? When I was testing it actually reduced overall codesize, but that was running at O3 so may be different. I'm surprised vector fmls come up enough in minsize to worry about, I'm not sure what you were testing.

https://github.com/llvm/llvm-project/pull/215338