next »

[Sigma]

Hi all,
 Is it the right thing to update uniform variables in a pixel shader every frame ? if not what's the best approach to update variable in a pixel shader every frame. I'm seeing poor fps after i changed my code to update uniform variables every frame. Also the updating variables are an object of a struct.

All helps are highly appreciated

[epcc]

Setting uniform variables isn't very expensive, they can be updated every frame or even multiple times per frame without much of a performance hit.
Try profiling your code to see what slows it down.

[Boreal]

Best would be to use a buffer-backed uniform block and update with GL_DYNAMIC_DRAW and glBufferSubData().

[Sigma]

Best would be to use a buffer-backed uniform block and update with GL_DYNAMIC_DRAW and glBufferSubData().

i have no idea how to achieve this using cocos2dx and i'm new to openglES as well, better u can help me understand.

Setting uniform variables isn't very expensive, they can be updated every frame or even multiple times per frame without much of a performance hit.
Try profiling your code to see what slows it down.

actually i'm running a pixel shader on a full screen size image (960x640), i have distance calculation as well that eats up most of the fps. i could see this by commenting out the distance calculation

without shader FPS - 60
With shader FPS - 15

i have pasted my code here plz have a look and let me know whats wrong in this code

Code:

#ifdef GL_ES
precision mediump float;
#endif

struct Light
{
  vec3 position;
  vec3 color;
  float intensity;
  int active;
};

uniform Light lights[5];

varying mediump vec2 v_texCoord;
uniform float u_lightInnerFallOff;
uniform vec2 u_resolution;
uniform float u_fragAlpha;

void main()
{
    vec4 fragColor = texture2D(CC_Texture0, v_texCoord);
    vec3 v_fragmentIntensity = fragColor.rgb;
    float fragAlpha = u_fragAlpha;
    for (int index = 0; index < 5; index++)
    {
        if (lights[index].active == 1)
	{
	    mediump float lightDist = distance(vec3((v_texCoord * u_resolution), 0.0), lights[index].position);
	    
	    lowp float lightValue;

	    if (lightDist < u_lightInnerFallOff)
	    {
		lightValue = 1.0;
		v_fragmentIntensity = lights[index].color * vec3(1.0, 1.0, 1.0);
	    }
	    else
	    {
		lightValue = u_lightInnerFallOff/lightDist;
		v_fragmentIntensity = lights[index].color * v_fragmentIntensity;
	    }
	    fragAlpha = fragAlpha + (lightValue * lights[index].intensity);
	}
   }
    gl_FragColor = vec4(v_fragmentIntensity, fragAlpha);
}

Things i have tried so far
    1. changed the precision to lowp - no performance boost
    2. commented out the v_Coord manipulation as per apple dev doc - no performance boost
    3. after commenting distance calculation in shader FPS increased to 30;
   
still something is eating up 30 fps, dont know where, all helps are highly appreciated.


Thanks
 

[rosholger]

your problem is probably all of that branching, branching in shaders kill performance sadly

[Sigma]

your problem is probably all of that branching, branching in shaders kill performance sadly

You saved my life, i never knew that. Thanks a lot... Just removing the for loop and if-else statement shoots up the fps to 30 from 12. Is there any other things i can improve in this code.

[rosholger]

your problem is probably all of that branching, branching in shaders kill performance sadly

You saved my life, i never knew that. Thanks a lot... Just removing the for loop and if-else statement shoots up the fps to 30 from 12. Is there any other things i can improve in this code.

Your welcome! don't know of anything else, im actually pretty bad at shaders 

[Sigma]

your problem is probably all of that branching, branching in shaders kill performance sadly

You saved my life, i never knew that. Thanks a lot... Just removing the for loop and if-else statement shoots up the fps to 30 from 12. Is there any other things i can improve in this code.

Your welcome! don't know of anything else, im actually pretty bad at shaders 

anyway thank u

[Sigma]

only in older GeForce GPU its slow, in Adreno GPU's its more than 60FPS always

[manabreak]

Like already said, branching is bad. And so are loops. If you need loops but get a performance bump, one thing to do is to unroll the loops.

Code:

// Loop
for(int i = 0; i < 5; ++i) {
    nippleColors[i] = calculateNippleColor(i);
}

// Loop unrolled
nippleColors[0] = calculateNippleColor(0);
nippleColors[1] = calculateNippleColor(1);
nippleColors[2] = calculateNippleColor(2);
nippleColors[3] = calculateNippleColor(3);
nippleColors[4] = calculateNippleColor(4);

This speeds the code up a bit because it doesn't have to do the conditional check (i < 5) every time the loop is finished. Of course, it increases the binary size of the shader program, but that isn't usually a big problem - the size doesn't get dramatically larger in such simple cases.

[BorisTheBrave]

Most drivers will be happy to unroll that loop for you.