[llvm] [WebAssembly] Lower fmuladd to fma (PR #161355)

[Sam Parker via llvm-commits]

sparker-arm wrote:

Performance wise, this looks really good for ML. Running NCNN, via V8 on AArch64, I observe a mean execution time reduction of ~15% over >30 workloads.

https://github.com/llvm/llvm-project/pull/161355