View Full Version : gl_FragCoord.z issue

06-24-2006, 01:55 PM
Hi there!

I'm currently implementing some fire effects using GLSL's fragment shaders & following fragment shaders is fine so far:

uniform float pscale; // common pertubation scale
uniform vec2 bounds; // texture bounds
uniform float lod; // some level of detail
uniform sampler2D flames; // flames texture
uniform sampler2D heat; // pertubation map (s offs, t offs, 0, offs scale)
uniform sampler2D scene; // scene texture

void main(void) {

// read pertubation value
vec4 p = texture2D(heat, gl_TexCoord[0].st / lod);

// scale pertubation by fragment's depth
// p.a *= (1.0 - gl_FragCoord.z);

// compute pertubated texture coords
vec2 uv = gl_TexCoord[0].st + (p.xy - vec2(0.5, 0.5)) * p.a * pscale;

// clamp pertubated texture coords to bounds
vec2 st = clamp(uv, vec2(0, 0), bounds);

// read flames & scene colors
vec4 fcol = texture2D(flames, st / lod);
vec4 scol = texture2D(scene, st);

// output mixed flames & scene colors
gl_FragColor = mix(scol, fcol, fcol.a);
}However, when I uncomment the scale-pertubation-by-gl_FragCoord.z line, OpenGL drops to software rendering:

- no matter if GL_DEPTH_TEST is disabled or (enabled && glDepthFunc(GL_ALWAYS))
- default glDepthRange (0..1, I think)
- glPolygonOffset untouched

Env: ATI Radeon 9800 Pro w/ Catalyst 6.2, Win2k

Anybody any clue ?

06-24-2006, 03:08 PM
ATI GPU's do not have gl_FragCoord register. If you use gl_FragCoord it will be emulated using varying variable (interpolated). I'm using gl_FragCoord.xy but I've never tried the z component.
A simple workaround would be to calculate depth in vertex shader and pass it as varying variable.

06-24-2006, 08:02 PM
zimerman, I experienced the exact same issues a while ago, unfortunately I don't EXACTLY know what the issue was or how I fixed it. I am pretty sure it was related to either texcoord array or bound texture sloppiness (on my part). I would suggest you take a look through your texture related code (esp binding and enabling the texcoord array), try things like binding and using only one texture. Other than that, I really don't know what to suggest.

06-25-2006, 03:27 AM
Thanks a lot!

I think I found another work-around: since I render my heat particles as billboards into my pertubation map, I will use per-particle distance-dependend alpha blending there...

@k_szczech: *already knew my next card will be from nvidia*

Cheers & best regards!