Re:Optimizing matrix multiplication in C
by sickboy on Sunday August 28, 2016 @06:52PM
(#12378921)
fma may help to double performance，if you control memoryrw wrll
[ Parent
| on HN ]
Right, but the price is slightly different product because incompatible rounding.
Also, too few CPUs support those, FMA only available since Intel Haswell (2013) and AMD Bulldozer (2011).
Finally, there’s also incompatibility between 3 and 4, but that’s the least of the problems because simple to workaround in runtime.