128x128x128 is actually much more than I need, I'll probably just
need about 40x40x40, which for float32 is 1MB. So video memory
should really be no problem - I'm more concerned about the best
format to use for the 3d texture, the format that will give best
speed.
I really don't know, but I wouldn't be surprised if you need to
round up to power-of-two sizes, so you might need to use 64x64x64.
Which makes it 4MB, just to be clear.
Smaller component size nearly always means faster, it'a just less
data to fetch from memory (and texture caches work better). I
suggest it probably comes down to what you can get away with for
your algorithm, and what hardware you're targeting. Remember that 8
bits can take you -1.0 to 1.0 in less than 0.01 size steps with a
bit of re-scale/biasing.
Thanks for the reminder Paul.
I intent to do color-space transforms: in L*a*b* the precision is
about 0,5% Delta-E, that is OK if only one transform is applied, but
for many I guess the errors could be noticeably.
Do you know what is the best way to speed of a single shader - so I
can experiment how different component sizes affect performance?
Mark
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Mac-opengl mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/mac-opengl/email@hidden
