I was wondering... if the idea of storing color information inside vertices was replaced with storing an index and supplying an array of colors/material on a model basis, would there be any significant performance difference or reprecussions?
The reason I ask is this system seems a lot more flexible and I think you would be able to use much more material information on a vertex level inside the shader.
0
Palette Based Rendering?
Started by TheAnti, Apr 05 2012 04:45 PM
9 replies to this topic
Sponsor:
#2 Members - Reputation: 951
Posted 05 April 2012 - 08:05 PM
How would you interpolate said values across a face?
clb: At the end of 2012, the positions of jupiter, saturn, mercury, and deimos are aligned so as to cause a denormalized flush-to-zero bug when computing earth's gravitational force, slinging it to the sun.
#3 Members - Reputation: 1409
Posted 05 April 2012 - 10:42 PM
Well to interpolate properly you would need to do the index -> color transformation in the vertex shader.
Of course any kind of lookup table (an array shader variable, or a texture lookup in the vertex shader) is going to have a performance impact, but without trying it out in your scenario it's hard to say what it would be.
Of course any kind of lookup table (an array shader variable, or a texture lookup in the vertex shader) is going to have a performance impact, but without trying it out in your scenario it's hard to say what it would be.
#4 Members - Reputation: 790
Posted 05 April 2012 - 11:23 PM
Why do you think this is even necessary? Do all your meshes have color data? Is it severely impacting the size of your meshes? Is vertex size having a large impact on your performance? Is there nothing better you could spend your time on improving?
Seems like a case of premature optimization, and I should know, I admit to having wasted much of my own time on it before.
Seems like a case of premature optimization, and I should know, I admit to having wasted much of my own time on it before.
#5 GDNet+ - Reputation: 1813
Posted 06 April 2012 - 01:17 AM
On GeForce6? I hardly believe so. On some Intel like GMA950? Possible. On D3D10 silicon. No way.I was wondering... if the idea of storing color information inside vertices was replaced with storing an index and supplying an array of colors/material on a model basis, would there be any significant performance difference or reprecussions?
Will it be significant? I would expect no difference.
Flexible? I'm not on it. Indexing has some uses, but definitely not that. More information to the VS? I don't see the point.The reason I ask is this system seems a lot more flexible and I think you would be able to use much more material information on a vertex level inside the shader.
Chris++
#6 Members - Reputation: 251
Posted 06 April 2012 - 04:48 AM
I'm wasn't so much thinking about a performance boost, I was just wondering if the theory is sound.
I tend to use a lot of vertex coloring and it seemed to make sense in the event that I wanted to swap out the certain colors of something while rendering the same model instead of copying the entire model over or changing the color of everything.
I've never seen it done before so I assumed there was a a reason not to do it.
I tend to use a lot of vertex coloring and it seemed to make sense in the event that I wanted to swap out the certain colors of something while rendering the same model instead of copying the entire model over or changing the color of everything.
I've never seen it done before so I assumed there was a a reason not to do it.
#7 Members - Reputation: 951
Posted 06 April 2012 - 06:37 AM
I'm wasn't so much thinking about a performance boost, I was just wondering if the theory is sound.
I tend to use a lot of vertex coloring and it seemed to make sense in the event that I wanted to swap out the certain colors of something while rendering the same model instead of copying the entire model over or changing the color of everything.
I've never seen it done before so I assumed there was a a reason not to do it.
Oh, OK, that makes more sense. Look into vertex data streams/input layouts, you can instruct the input assembler to pull the vertex color from a separate chunk of memory from the rest of your data. You'd just bind a different vertex buffer to your 'vertex color data' slot and leave the rest of it untouched.
clb: At the end of 2012, the positions of jupiter, saturn, mercury, and deimos are aligned so as to cause a denormalized flush-to-zero bug when computing earth's gravitational force, slinging it to the sun.
#9 Members - Reputation: 589
Posted 12 April 2012 - 04:34 PM
Yeah, you can do that, the only thing is that you would need to perform the interpolation the pixel shader end by passing the two colors from the vertex shader to the pixel shader as varying variables. That way it will interpolate the two values.
#10 Members - Reputation: 4750
Posted 12 April 2012 - 11:51 PM
I think that this skipped, because at times where you use vertex color, you often needed the register for other tasks (i.e. skeleton animation data). The most recent approach in this direction is to use gradient mapping at the pixel shader level to recolor a model. Take a look at Shading a Bigger, Better Sequel: Techniques in Left 4 Dead 2, there's the gradient technique of L4D2 explained.I've never seen it done before so I assumed there was a a reason not to do it.
Ashaman
My game: Gnoblins
Developer journal about Gnoblins
Small goodies: Simple alpha transparency in deferred shader
My game: Gnoblins
Developer journal about Gnoblins
Small goodies: Simple alpha transparency in deferred shader
