"You may be sure that when a man begins to call himself a realist, he is preparing to do something he is secretly ashamed of doing."
- Sydney Harris
More pages: 1 ... 11 ... 16 17 18 19 20 21 22 23 24 25 26 ... 31 ... 41 ... 51 ... 61 ... 71 ... 81 ... 91 ... 101 ... 111 ... 121 ... 131 ... 141 ... 151 ... 161 ... 171 ... 181 ... 191 ... 201 ... 211 ... 221 ... 231 ... 241 ... 251 ... 261 ... 271 ... 281 ... 291 ... 301 ... 311 ... 321 ... 331 ... 341 ... 351 ... 361 ... 365
Query FailedHumus
Wednesday, August 22, 2012

Sean B, we did it on all platforms. I don't know for sure if it really helps on for instance modern GPUs on PC, but they usually have a vast abundance of vertex shading performance vs. say the PS3, so if it wasn't faster at least it probably wasn't measurably slower. I believe it was faster on both consoles though.
Bart Kuipers
Monday, August 20, 2012

@Sean B: From the notes in the slides: "Further we packed the remaining two vectors into floats. This does not reduce the storage space, but takes away another vertex attribute. Trading a small amount of ALU for one less vertex attribute turned out a performance gain in practice."
Sean B
Monday, August 20, 2012

What's the advantage of the RG32F encoding of the tangents versus two ubyte4s? I can imagine that paying the extra shader decode cost was worth it compared to the cost of of the extra attribute on some platform, but did you do this on all platforms?

(If there are written notes accompanying the slides explaining this, they aren't visible in MS PowerPoint Viewer.)
Bart Kuipers
Monday, August 20, 2012

Thanks a lot! I really love these articles! You kept me busy all morning reading those. Especially the bonus sheet (nr 40) in the 'creating vast gameworlds' kept me busy for a while. I was trying to find out how you simplified C.x back to f/n I did exactly what I should do and came to (fn*f^2)/(fn*n^2), but couldn't wrap my mind around how to solve that into f/n. Took me 2 hours, but I did it nonetheless, for some reason my math skills were below average! (I blame the heat). Nevertheless, multiplying by (1/fn)/(1/fn) gave me (1+f/n)/(1+n/f) which is (1+var)/(1+1/var), multiply by var = (var+var^2)/(var+1) = (var+1)var/(var+1) = var = f/n Dunno why I just had to type that out, but it took me a while to figure it out, so maybe if someone else reads this, he/she might get it as well
Jasmine
Thursday, August 16, 2012

That is great news! Please post the links to the recordings when they go up if you are able Thanks for keeping on keeping on, I always love reading your articles!
Joe
Monday, August 13, 2012

Great site, great framework.

I noticed you mentioned playing around with compute shaders. I'm a big fan of your simple interface design, would be great to have compute shaders included in the framework.
Joe
Monday, August 13, 2012

Great site, great framework.

I noticed you mentioned playing around with compute shaders. I'm a big fan of your simple interface design, would be great to have compute shaders included in the framework.
nuninho1980
Wednesday, August 8, 2012

OOH! Sorry because my double post is bad. But I pressed "update" button using browser = your site is bug.
More pages: 1 ... 11 ... 16 17 18 19 20 21 22 23 24 25 26 ... 31 ... 41 ... 51 ... 61 ... 71 ... 81 ... 91 ... 101 ... 111 ... 121 ... 131 ... 141 ... 151 ... 161 ... 171 ... 181 ... 191 ... 201 ... 211 ... 221 ... 231 ... 241 ... 251 ... 261 ... 271 ... 281 ... 291 ... 301 ... 311 ... 321 ... 331 ... 341 ... 351 ... 361 ... 365