On 10/25/2013 10:13 AM, Tom Musta wrote: > On 10/25/2013 11:42 AM, Richard Henderson wrote: >> I believe that a better implementation could use float*_muladd, and check the >> result for float_flag_invalid. If set, compute the intermediate product so >> you >> can figure out the VXISI setting. But we'd expect that to be an unlikely >> path. > > Interesting thought. I think I see a way to re-arrange the code. Thanks, > Richard.
Actually, you don't even have to compute the intermediate product. The only way you can have VXISI for a*b+c is for isinf(c) && (isinf(a) || isinf(b)) since the intermediate product a*b is infinite precision, and thus cannot overflow to inf unless one of the multiplicands is already inf. r~
