Re: [AD] Color convertors

[ Thread Index | Date Index | More lists.liballeg.org/allegro-developers Archives ]


Eric Botcazou wrote:
Here's some results: (K6-III 400 Mhz, MMX enabled, Matrox G200)
[snip - results]

Some thoughts:
- the overhead due to the additional code (tests and jumps essentially) is
not too high: between 0 and 11%,

The overhead should be negligeable ( < 2%). The reason why it's so high apparently is because of the K6-3, which is a beefed-up K5 :) The current MMX code isn't well paired, so the K6 and PMMX performance will suffer. The K7 and i686 don't care much about pairning, since the hardware does it alone anyway, and can even split up microcodes (so it's more efficient than manual pairing). The parining issue should be resolved easily, but I don't have the time this week to do it.

I'll see if the 16->24 can be fixed though.


--
- Robert J Ohannessian
"Microsoft code is probably O(n^20)" (my CS prof)
http://pages.infinit.net/voidstar/



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/