在 2019/12/12 21:22, Liu Hao 写道: > This is wrong. Because FMA shall result the result exactly once. If we
^ round
Apologies for so many typos. I will fix them later.
> roudn the first result to double we get a different result:
^ round
> volatile double c = 0x0.00000000000007FFp0;
>
> assert(a * b + c == 0x1.0000000000001p0);
> assert(xfma(a, b, c) == 0x1.0000000000001p0);
^ fma
--
Best regards,
LH_Mouse
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
