x86_64: FMA primop generates wrong result

Summary

On x86, FMA primop generates wrong result with -mfma -O1.

Steps to reproduce

Compile the following code with ghc -fforce-recomp -O1 -mfma TwoProdFMA.hs:

{-# LANGUAGE MagicHash #-}
{-# LANGUAGE UnboxedTuples #-}
import GHC.Exts

twoProductFloat# :: Float# -> Float# -> (# Float#, Float# #)
twoProductFloat# x y = let !r = x `timesFloat#` y
                       in (# r, fmsubFloat# x y r #)
{-# NOINLINE twoProductFloat# #-}

twoProductDouble# :: Double# -> Double# -> (# Double#, Double# #)
twoProductDouble# x y = let !r = x *## y
                        in (# r, fmsubDouble# x y r #)
{-# NOINLINE twoProductDouble# #-}

main :: IO ()
main = do
    print $ case twoProductFloat# 2.0# 3.0# of (# r, s #) -> (F# r, F# s)
    print $ case twoProductDouble# 2.0## 3.0## of (# r, s #) -> (D# r, D# s)

The result is wrong:

$ ghc -fforce-recomp -mfma -O1 TwoProdFMA.hs
$ ./TwoProdFMA
(6.0,-2.0)
(6.0,-2.0)

Expected behavior

The expected result is:

(6.0,0.0)
(6.0,0.0)

Environment

• GHC version used: 9.8.2, master (b4cae4ec)
• Operating System: Linux
• System Architecture: x86-64