[llvm-commits] Please review: FMA3 instructions set

[Stephen Canon]

Hi Elena --

"precision" is consistently mis-spelled.  Easily addressed.

Do I understand correctly that this patch lowers fadd + fmul to fma by default?  We want it to be easy for LLVM to generate fma when it is beneficial to performance, but we can't simply naively lower to it everywhere and still conform to language semantics.  Someone else can speak to what policy should be here, but at the very least we will need to have an option to block fma formation.

Thanks,
- Steve

On May 22, 2012, at 7:51 AM, "Demikhovsky, Elena" <elena.demikhovsky at intel.com> wrote:

> I added support for FMA instructions with 3 operands (Intel set)
> -       codegen lowering fadd+fmul to fma (add, sub, neg-add forms vector and scalar)
> -       intrinsics
> -       fixed a bug in encoder (FMA encoding was incorrect)
> -       loads folding for FMA
> -       lit tests
> 
> Thank you.
> 
> - Elena