Use on matrix math in shader?

Math and Physics Programming

Started by reaperrar April 13, 2012 01:30 AM

1 comment, last by Bacterius 12 years ago

136

Author

April 13, 2012 01:30 AM

"Use own matrix math in shader?"*
Was just curious, does anyone think it's worth using your own matrix math in say a hlsl shader?

I'm assuming:



float3x3 g_mWorld;

float3x3 g_mViewProj;



mul( g_mWorld, g_mViewProj);

Results in:



g_mModel[0][0] * g_mViewProj[0][0] + g_mModel[1][0] * g_mViewProj[0][1] + g_mModel[2][0] * g_mViewProj[0][2]

g_mModel[0][1] * g_mViewProj[0][0] + g_mModel[1][1] * g_mViewProj[0][1] + g_mModel[2][1] * g_mViewProj[0][2]

g_mModel[0][2] * g_mViewProj[0][0] + g_mModel[1][2] * g_mViewProj[0][1] + g_mModel[2][2] * g_mViewProj[0][2]

g_mModel[0][0] * g_mViewProj[1][0] + g_mModel[1][0] * g_mViewProj[1][1] + g_mModel[2][0] * g_mViewProj[1][2]

g_mModel[0][1] * g_mViewProj[1][0] + g_mModel[1][1] * g_mViewProj[1][1] + g_mModel[2][1] * g_mViewProj[1][2]

g_mModel[0][2] * g_mViewProj[1][0] + g_mModel[1][2] * g_mViewProj[1][1] + g_mModel[2][2] * g_mViewProj[1][2]

g_mModel[0][0] * g_mViewProj[2][0] + g_mModel[1][0] * g_mViewProj[2][1] + g_mModel[2][0] * g_mViewProj[2][2]

g_mModel[0][1] * g_mViewProj[2][0] + g_mModel[1][1] * g_mViewProj[2][1] + g_mModel[2][1] * g_mViewProj[2][2]

g_mModel[0][2] * g_mViewProj[2][0] + g_mModel[1][2] * g_mViewProj[2][1] + g_mModel[2][2] * g_mViewProj[2][2]

But couldn't we save time by using our own math?



g_mModel[0][0] * g_mViewProj[0][0] + g_mModel[1][0] * g_mViewProj[0][1]

g_mModel[0][1] * g_mViewProj[0][0] + g_mModel[1][1] * g_mViewProj[0][1]

0.0f

g_mModel[0][0] * g_mViewProj[1][0] + g_mModel[1][0] * g_mViewProj[1][1]

g_mModel[0][1] * g_mViewProj[1][0] + g_mModel[1][1] * g_mViewProj[1][1]

0.0f

g_mModel[0][0] * g_mViewProj[2][0] + g_mModel[1][0] * g_mViewProj[2][1] + g_mModel[2][0]

g_mModel[0][1] * g_mViewProj[2][0] + g_mModel[1][1] * g_mViewProj[2][1] + g_mModel[2][1]

1.0f

That would save 8 addition and 10 multiplications per matrix * matrix operation per object. The same could be done for vertex transforms too. Is it a worthy optimization?

Postie

1,559

April 14, 2012 12:31 AM

You'd have to profile it to be sure, but my gut feeling is that things like Matrix Math would already be optimized in the shader since its a VERY common GPU operation.

[size="2"]Currently working on an open world survival RPG - For info check out my Development blog:[size="2"] ByteWrangler

Bacterius

13,181

April 14, 2012 05:33 AM

I'd think the mul operation in HLSL would be hardware-accelerated with special SIMD units for matrix multiplication since that's the single most common operation in a vertex shader. Try and look at the GPU assembly output and see what "mul" actually does, and then time them to see if it's faster? It could be but I doubt it.

“If I understand the standard right it is legal and safe to do this but the resulting value could be anything.”

Use on matrix math in shader?

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Use on matrix math in shader?

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Reticulating splines