Jump to content

  • Log In with Google      Sign In   
  • Create Account

Black specs in normal mapping on nvidia


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
9 replies to this topic

#1 frna   Members   -  Reputation: 162

Like
0Likes
Like

Posted 21 April 2012 - 08:45 AM

Hey,

I'm having a problem with my normal mapping on nvidia cards that I just can't seem to fix and hope you guys could help me out.
Here are some pictures showing the problem, the last one shows the same scene on ATI but from a different place in the model.

http://i.imgur.com/unzdx.jpg
http://i.imgur.com/WgiIH.jpg
http://i.imgur.com/A0AAZ.jpg

ati: http://i.imgur.com/5Oa5b.png (works as intended)


I found out that the problem is my shininess float value that I send into the shader. If it's 0.0f it seems like the shader on nvidia thinks its undefined or some other weird value. If I hardcode 0.0 as shininess into the specular component it works just fine.

Any idea how this could happen?

Sponsor:

#2 dpadam450   Members   -  Reputation: 946

Like
0Likes
Like

Posted 21 April 2012 - 06:35 PM

Post your shader most likely looks like the specular part of your shader is actually subtracting.

#3 Hodgman   Moderators   -  Reputation: 31804

Like
2Likes
Like

Posted 21 April 2012 - 08:49 PM

If you look at the pow function in the GLSL specification, you can see that if both parameters are zero (e.g. zero spec mask and zero spec power), then the result is undefined -- each card can generate a different result. It's likely that one of your cards is producing 0.0f as the result, and the other is generating NaN (which propagates through the rest of your shader, causing black speckles).

#4 frna   Members   -  Reputation: 162

Like
0Likes
Like

Posted 22 April 2012 - 06:57 PM

Hey, thanks for your replies.
I was thinking about undefined value, but what is really odd is that if I hardcode this: float spec = pow(max(0.0, dot(r, e)), 0.0) it works just fine. When I send 0.0 to the shader however, black specks everywhere. I cant explain that :S

Here's my shader code:
#version 110

uniform sampler2D tex;
uniform sampler2D bmap;
uniform bool boolBump;
uniform vec4 vecColor;
uniform bool onlyColor;

uniform float fTransparencyThresh;
uniform float fShininess;
uniform float alpha;

varying vec3 vecLight;
varying vec3 vecEye;
varying vec3 vecNormal;
 
vec4 getLighting()
{
//Ambient part
vec4 color = (gl_FrontLightModelProduct.sceneColor * gl_FrontMaterial.ambient) + (gl_LightSource[0].ambient * gl_FrontMaterial.ambient);

//For bump mapping, the normal comes from the bump map texture lookup
vec3 n = normalize(vecNormal);
if(boolBump)
{
  n = normalize(texture2D(bmap, gl_TexCoord[0].st).xyz * 2.0 - 1.0);
}

vec3 l = normalize(vecLight);

//Lambert term
float NdotL = dot(n, l);
color += gl_LightSource[0].diffuse * gl_FrontMaterial.diffuse * max(0.0, NdotL);

//Reflection term
if(NdotL > 0.0)
{ 
  vec3 e = normalize(vecEye);
  vec3 r = normalize(-reflect(l,n)); 
 
  float spec = pow(max(0.0, dot(r, e)), fShininess); 
  color += gl_LightSource[0].specular * gl_FrontMaterial.specular * spec;
}

return color;
}
 
void main(void)
{
vec4 texel = texture2D(tex, gl_TexCoord[0].st);
if(texel.a < fTransparencyThresh)
  discard;

//Get shading
    vec4 color = getLighting();

//Color only mode?
if(onlyColor)
{
  color *= vecColor;
}
else
{
  color *= texel;
}
//Set fragment color, alpha comes from MTL file
gl_FragColor = vec4(color.xyz, alpha);
}


#5 dpadam450   Members   -  Reputation: 946

Like
0Likes
Like

Posted 22 April 2012 - 09:40 PM

what if you send something else to the shader? Could be an issue that its not making it to the shader. Is the shader bound before sending the uniform?

#6 Hodgman   Moderators   -  Reputation: 31804

Like
2Likes
Like

Posted 22 April 2012 - 11:19 PM

but what is really odd is that if I hardcode this [0.0] it works just fine. When I send 0.0 to the shader however, black specks everywhere. I cant explain that :S

When you hard-code a value, the resulting shader assembly will be completely different, largely thanks to the optimising compiler used by your driver. It's possible that with the hard-coded version, the optimiser has said "anything to the power of zero is one, and anything multiplied by one is unchanged, so I'll just remove that code", whereas when you use a uniform, it actually computes 0^0 and gets the undefined result of NaN.

#7 frna   Members   -  Reputation: 162

Like
0Likes
Like

Posted 23 April 2012 - 06:13 AM

I see, that would make more sense to me thanks @Hodgman.

@dpadam450 yes the shader is bound and getUniform returns a location.

What I'm doing now is to check if dot(r,e) > 0.0 and only enter the specular part if that's true. Seems to work fine. Do you reckon that's a valid solution? Any better suggestions?

#8 dpadam450   Members   -  Reputation: 946

Like
0Likes
Like

Posted 23 April 2012 - 09:56 AM

The other suggestion would be to send 1 instead of 0. You shouldnt see much or any difference between those. If you want no specular, you should be setting the glFrontMaterial.specular to 0. If you just want it to have flat specular instead of really "tight" specular, then just use 1.

#9 frna   Members   -  Reputation: 162

Like
0Likes
Like

Posted 23 April 2012 - 11:22 AM

It works fine with any other value appart from 0. That's what really bugs me about this whole problem.

#10 Hodgman   Moderators   -  Reputation: 31804

Like
0Likes
Like

Posted 23 April 2012 - 12:58 PM

A spec power of 1 is equivalent to a perfectly rough/diffuse surface, and numbers lower than 1 aren't very useful. e.g. when decoding spec-power values from a texture, I'd use something like specPow = tex*8192 + 1.
If you want to remove specular from a material, set the spec-mask to 0 instead (gl_FrontMaterial.specular).




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS