On Mon, May 31, 2010 at 10:02 AM, Matthias Bentrup <matthias.bent...@googlemail.com> wrote: > the "fast" approximate sqrt may not be so fast on current CPUs. The > fast sqrt has to move the value from an FPU or SSE register into an integer > register and back, and these instructions have a high latency, while the > built-in sqrt (or rsqrt) functions have been optimized down to a few cycles.
That may explain why I do not see a noticeable improvement when benchmarking it when replacing the function with sqrt(). I imagine that it did have a very noticeable impact on the old days (or let's hope it had) and it would correspondingly show also on newer hardware if qualitatively there wasn't a change. _______________________________________________ ioquake3 mailing list ioquake3@lists.ioquake.org http://lists.ioquake.org/listinfo.cgi/ioquake3-ioquake.org By sending this message I agree to love ioquake3 and libsdl.
