🔒 Locked

sin/cos in HLSL vs_2_0

Started by jaafit Mar 23, 2007 at 7:30 PM 6 replies 13k views

Original Post

Mar 23, 2007 at 7:30 PM

How exactly are sin/cos supposed to work in HLSL? I'm getting two very different results from what looks like the same code:


	float theta = 0;
	windUV.x = cos(theta) * worldPos.x - sin(theta) * worldPos.y;
	windUV.y = sin(theta) * worldPos.x + cos(theta) * worldPos.y;

The above results in the same value for windUV for each vertex.


	windUV.x = cos(0) * worldPos.x - sin(0) * worldPos.y;
	windUV.y = sin(0) * worldPos.x + cos(0) * worldPos.y;

The above works as expected. ?

Asesh

Mar 23, 2007 at 10:31 PM

The argument to sin/cos is the angle in radians

Sc4Freak

Mar 24, 2007 at 12:08 AM

Quote:
Original post by Asesh
The argument to sin/cos is the angle in radians

Shouldn't really make a difference.

Keep in mind, the shader compiler probably optimises out the entire second example and just sets the variables to 0 instead of going through the sin/cos.

jaafit

Mar 24, 2007 at 12:13 AM

So there's a bug in the optimizer?

Sc4Freak

Mar 24, 2007 at 12:24 AM

Whoops, sorry, misread the example. Didn't see the cosines in there.

ET3D

Mar 24, 2007 at 8:19 AM

I'd suggest looking at the resulting assembly code, and seeing what the difference is.

TheAdmiral

Mar 24, 2007 at 2:53 PM

Indeed, the only difference between the two samples is that one will invoke the sin and cos functions each frame and the other compiles them out to static code. Neglecting floating-point error, they are equivalent. Are you sure that the error can't lie elsewhere?

If not, then perhaps your problems are floating-inaccuracy-based. I'm not sure if video hardware treats sin and cos of zero as special cases, but if it doesn't, and your camera happens to be zoomed in on a patch of land that is very far away from the origin, the minute imprecision may be scaled up arbitrarily.

Admiral

ET3D

Mar 25, 2007 at 3:04 AM

Another way to check this further would be to change the first example to assign cos(theta) and sin(theta) to variables and look at them.

Also, when you say that you get the same value for windUV, knowing what value might provide a clue.

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.