1) Who locked it and why is this thread locked now ? 2) This guy don't like our coding conventions. Why not, but we don't want to change it now. So we keep "myVariable" and will not use "my_variable".
3) His patch compute gradients another way. What is wrong is that it only check for 4 directions instead of 8. So I guess it will help speed, but globule would not be able to walk trough a diagonal tight path. What is bad is that the resources is computed less often. So yes, it's faster, but some units can wait longer. If we add any "hard drop" policy, then it has to be a flag, and dis-activated by default. 4) He suggested some smart optimization in the pre-gradient computation. I knew about it but I where too lazy to do it until now. He provided a patch with other "improvements" (see 3), and his coding way. So I just added it cleanly my own way. He said it would improve about 30% of the speed. I would have expected some good improvement, but... after serious tests I found it to be 0% to 1% slower on my computer. I am really surprised. Did you see any changes on your computer? 5) Some statements he did like "it's memory optimized instead of speed optimized" are ... wrong ? Well which of these two lines is faster: (y << wDec | x) (y*w + x) He says the second is faster, and I don't see why, because I think it's either the same or the first is faster. Can someone explain me why ? Computer never do it the way you expect it.... Nuage Stéphane Magnenat wrote: > On Sunday 18 December 2005 04:34, Bradley Arsenault wrote: > >>There is a person on the forums that has a patch that, as he says, can >>make a great improvement on the gradient calculation speed. > > > It's there: > http://globulation2.org/forums/viewtopic.php?t=10 > > Nuage, could you have a look plz ? > Your mediawiki account wotks flor forums > > Steph > > > _______________________________________________ > glob2-devel mailing list > [email protected] > http://lists.nongnu.org/mailman/listinfo/glob2-devel > _______________________________________________ glob2-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/glob2-devel
