"David Powell" <david@xxxxxxxxxxxxxxxxx> wrote: > 2008/7/29 Axel Dörfler <axeld@xxxxxxxxxxxxxxxx>: > > Spacing of the function calls is wrong - it's always fprintf("") > > not > > fprintf ("") nor fprintf( "" ) :-) > > Also, looking at the end of the patch, the output looks wrong, too, > > as > > there is an additional tab before the last '};'. > > Other than that, the patch looks good to me. > Do you want me to fix those issues and re-submit the patch, or should > I fix them later? I would prefer to fix them before committing the patch, at least. Looks like later anyway ;-) > > Grayscale is probably enough for the moment. You could do a simple > > palette replacement algorithm instead of a real color conversion, > > though -- dunno if that would look better, though. > > If we just do a simple conversion like this, I would actually > > prefer to > > do it "live" in the boot loader, though; at least it should consume > > less space, and should still be pretty much instant. > Since posting the patch, I've been working on the > generate_boot_screen > program, and I've managed to add "Gervautz-Purgathofer octree > quantization" > to the program, which means that it can now generate an optimized 256 > colour > palette. (It's a open source implementation that I've tweaked, I > wouldn't want to take credit for stuff that isn't mine) > > I've also added a nearest-colour mapping, which gives pretty good > result > for an indexed colour image. I'm currently working on applying a > dither algorithm to the image > to make it look even better, though I may submit a patch without the > dither, as > it's a significant improvement on the greyscale image anyhow. > > The upshot of all of that is that the colour image in 8-bit mode > looks > really good, > however the genereate_boot_screen can take a second or two to run > with the > colour quantization, and probably longer once I get dithering to > work, > so I don't think it would be a good idea to include those algorithms > in the bootloader to > do it "live" as there would be a noticable speed penalty. Okay, but at least it would then only hit those people that actually use it rather than everyone ;-) Anyway, that sounds like a worthwhile approach. Would it be faster if the palette was created by the generate_boot_screen command, and if the boot loader would only do the color lookup? Bye, Axel.