I didn't get an e-mail from you, but got a mail from github with a pull request. Either way, i got the changes.
I already commented on your commit on github, so i'll just copy/paste it here.
Quote:
Awesome! Will check it out in the next few days.
I will most likely kick out the vertex fog and implement it per pixel in the final deferred rendering stage which should be faster and will also look more accurate.
I'm not really convinced about how the mode settings currently work. I think this makes things a bit harder on the user when updating shade-c as the settings will be overwritten. Maybe we can put it in an external file, check if it's there, and then use the values. Otherwise default values will be used. What do you think?
Thanks again! (Especially for fixing the memory leaks!)
[edit]
Quote:
I really like the option to swich to 16/32bit buffers but somehow everything gets brighter and looses contrast. (maybe there's a problem with the conversion between linear rgb and srgb)
edit: also: if you don't use any normal compression in the g-buffer you can use best fit normals to improve the quality (worked well for me)
I'll have a look at it. I think it has something to do with how the lighting is stored (it's some sort of inverted alpha blending which might not work as espected with floating point values). I'll also remove the bitdepth value and add 3 bitdepth values instead. One for g-buffer, one for lighting and one for multi purpose rendertargets. I tried using best fit normals in the past but didn't go for it as i wanted depth and normals in one texture for speed reasons (if you need normals, you also need depth 99% of the time...saved me one tex lookup and some bandwith). Still not sure if this is a good idea though...best fit normals really DO look good and are well worth the extra bandwith and texlookup. Also, after implementing PBR and static shadowmaps, there will still be 2-3 free channels in the g-buffer. More than enough to implement best fit normals and maybe even add some per object motion blur
