Yesterday we tested some new srqt functions. First I tried a quick
hack after the suggested newton-approximation algorithm and it was three
times as fast as the builtin function.
Next we tested the function Linus suggested. The result was really astonishing:
It seems to be thirthy (in numbers: 30) times as fast as the bultin function.
Next we tried to recompile POVray, which's performance on teh ALPHA was
slightly disappointing in comparison to intel-based machines.
And, _WOW_, it's twice as fast on the JENSEN as with the bulting sqrt.
(fish13.pov 320x200 -q9 ~141s usertime on JENSEN with 150MHz 21064)
Now it's approx. twice as fast as my 486DX4-100.
POVray seem to use sqrt _really_ heavily.
BTW.: We used POVray with the upcoming SVGALIB for JENSEN by Stephan Kanthak.
-- |/ |\ai
-- Kai Schuetz kais@i6.informatik.rwth-aachen.de