在 glsl 中处理未定义、NaN 或 Inf 值。数学错误的思考
我仍然没有摆脱 glsl 着色器(GL 2.0)中的一些恼人的错误...我已经寻找这个问题的解决方案超过 3 周了,但我没有得到任何结果。谷歌上有很多例子和解释,但没有一个真正有效。
首先,这是我的fragmentshader代码。之后我将解释到底问题是什么:
//fragmentshader
const int MAX_POINT_LIGHTS = LL.MAX_POINT_LIGHTS;
const int MAX_DIR_LIGHTS = LL.MAX_DIR_LIGHTS;
const int MAX_TEXTURES = LL.MAX_TEXTURES;
//float-precision
precision mediump float;
//varyings
varying vec2 texcoord;
varying vec3 normal;
varying vec3 position;
varying float depth;
//pointlights
uniform float pl_range[MAX_POINT_LIGHTS];
uniform vec3 pl_position[MAX_POINT_LIGHTS];
uniform vec3 pl_color[MAX_POINT_LIGHTS];
uniform vec3 pl_specular[MAX_POINT_LIGHTS];
//dirlights
uniform vec3 dl_direction[MAX_DIR_LIGHTS];
uniform vec3 dl_position[MAX_DIR_LIGHTS];
uniform vec3 dl_color[MAX_DIR_LIGHTS];
uniform vec3 dl_specular[MAX_POINT_LIGHTS];
//cameras
uniform vec3 camera;
//fog
uniform vec4 uVecFog;
uniform float uFogNear;
uniform float uFogFar;
//material
uniform vec3 mat_diff;
uniform vec3 mat_spec;
uniform float mat_shine;
uniform float mat_elum;
//texture
uniform sampler2D uTexture;
//ambientlight
vec4 ambient = vec4(0.0,0.0,0.0,0.0);
void main() {
vec3 N = normalize(normal);
vec3 V = normalize(camera - position);
vec4 diff = vec4(0.0,0.0,0.0,0.0);
vec4 spec = vec4(0.0,0.0,0.0,0.0);
//dirlights
for(int i=0;i<MAX_DIR_LIGHTS;i++) {
vec3 L = -normalize(dl_direction[i]);
vec3 R = -normalize(reflect(L,N));
float RV = max(dot(R,V),0.0);
float S = pow(max(dot(R,V),0.0),mat_shine);
float F = clamp(dot(L,N),0.0,1.0);
diff += vec4(F * mat_diff * dl_color[i],1.0);
spec = vec4(S * mat_spec * dl_specular[i],1.0);
}
//pointlights
for(int i=0;i<MAX_POINT_LIGHTS;i++) {
vec3 L = normalize(pl_position[i] - position);
vec3 R = -normalize(reflect(L,N));
float D = distance(pl_position[i],position);
float DF = clamp(1.0 - D/pl_range[i],0.0,1.0);
float S = pow(max(dot(R,V),0.0),mat_shine);
float F = clamp(DF * max(dot(L,N),0.0),0.0,1.0);
diff += vec4(F * mat_diff * pl_color[i],1.0);
spec += vec4(DF * S * mat_spec * pl_specular[i],1.0);
}
//fog
float fogDensity = 0.0;
if (depth > uFogFar) {
fogDensity = 1.0;
} else if(depth > uFogNear) {
float newDepth = depth-uFogNear;
fogDensity = newDepth/(uFogFar-uFogNear);
} else if (depth < uFogNear) {
fogDensity = 0.0;
}
vec4 vecFog = uVecFog * fogDensity;
//texturecolor
vec4 colortex = texture2D(uTexture,vec2(texcoord.x,-texcoord.y));
//final fragmentcolor
gl_FragColor = mix((ambient + colortex * (diff + vec4(mat_diff * mat_elum,1.0)) + spec),vecFog,fogDensity);
}
给出以下变量值,其中:
- pl_ 是点光源值,
- dl_ 是方向光值,
- mat_diff 是材质的漫反射值,
- 还有 mat_spec、mat_elum、mat_shine 用于材质镜面反射、发光和光泽度,
- N = 法线,
- V = 到顶点的视图矢量,
- L = 到顶点的光矢量,
- R = 反射矢量,
- D =光距离,
- DF = 混合光的距离因子,
- S = 计算的镜面反射因子,
- F = L、N
- 和 diff 的点积,计算的漫反射/规格矢量的规格矢量和。
现在出现如下问题: 点光源本身运行完美。定向灯本身也可以完美运行。 但它们在一起不会很好地工作。我来具体说明一下问题: 我最多有 4 个点光源和 4 个方向光。为了进行测试,我使用了 4 个点光源和 1 个方向光。
- 如果我注释掉定向灯的 for 循环计算,我得到以下结果:
- 如果我注释掉点光源的 for 循环,则会得到以下结果:
- 如果我启用了两个 for 循环,则会发生以下情况:
定向光似乎完全被忽略。汽车仅由点光源照亮。
所以,我发现: 如果我改变 for 循环的顺序,点光源 ->定向光,而不是暗光 ->点光源,效果交换:忽略点光源,计算方向光。
我也发现,这个问题一定是由:pow()函数或除以零引起的,但是: 如果我注释掉以下代码行:
spec = vec4(S * mat_spec * dl_specular[i],1.0);
spec += vec4(DF * S * mat_spec * pl_specular[i],1.0);
我可以获得以下结果:
->通常为点光和定向光计算漫反射颜色!
所以我想到了如何消除这个问题,因为我认为0乘以0会导致数学错误。除以零也是如此...... 如何避免未定义的值?
android-glsl 中似乎不存在 isnan() 或 isinf() 函数。
安迪解决方案或提示我可以做得更好,这会起作用吗? 如果任何答案能有所帮助,我会很高兴。三周以来这让我很生气。非常感谢各位帮助者! :)
有关代码的更多信息: https://github.com/Chrise55/Llama3D
I still didn't get rid of some annoying bugs in my glsl shader (GL 2.0)... I searched for a solution of this problem for over 3 weeks now, but I didn't get any. There are many examples and explanations on google, but none of them really works.
First of all, here is my fragmentshader code. After that I will explain what exactly the problem is:
//fragmentshader
const int MAX_POINT_LIGHTS = LL.MAX_POINT_LIGHTS;
const int MAX_DIR_LIGHTS = LL.MAX_DIR_LIGHTS;
const int MAX_TEXTURES = LL.MAX_TEXTURES;
//float-precision
precision mediump float;
//varyings
varying vec2 texcoord;
varying vec3 normal;
varying vec3 position;
varying float depth;
//pointlights
uniform float pl_range[MAX_POINT_LIGHTS];
uniform vec3 pl_position[MAX_POINT_LIGHTS];
uniform vec3 pl_color[MAX_POINT_LIGHTS];
uniform vec3 pl_specular[MAX_POINT_LIGHTS];
//dirlights
uniform vec3 dl_direction[MAX_DIR_LIGHTS];
uniform vec3 dl_position[MAX_DIR_LIGHTS];
uniform vec3 dl_color[MAX_DIR_LIGHTS];
uniform vec3 dl_specular[MAX_POINT_LIGHTS];
//cameras
uniform vec3 camera;
//fog
uniform vec4 uVecFog;
uniform float uFogNear;
uniform float uFogFar;
//material
uniform vec3 mat_diff;
uniform vec3 mat_spec;
uniform float mat_shine;
uniform float mat_elum;
//texture
uniform sampler2D uTexture;
//ambientlight
vec4 ambient = vec4(0.0,0.0,0.0,0.0);
void main() {
vec3 N = normalize(normal);
vec3 V = normalize(camera - position);
vec4 diff = vec4(0.0,0.0,0.0,0.0);
vec4 spec = vec4(0.0,0.0,0.0,0.0);
//dirlights
for(int i=0;i<MAX_DIR_LIGHTS;i++) {
vec3 L = -normalize(dl_direction[i]);
vec3 R = -normalize(reflect(L,N));
float RV = max(dot(R,V),0.0);
float S = pow(max(dot(R,V),0.0),mat_shine);
float F = clamp(dot(L,N),0.0,1.0);
diff += vec4(F * mat_diff * dl_color[i],1.0);
spec = vec4(S * mat_spec * dl_specular[i],1.0);
}
//pointlights
for(int i=0;i<MAX_POINT_LIGHTS;i++) {
vec3 L = normalize(pl_position[i] - position);
vec3 R = -normalize(reflect(L,N));
float D = distance(pl_position[i],position);
float DF = clamp(1.0 - D/pl_range[i],0.0,1.0);
float S = pow(max(dot(R,V),0.0),mat_shine);
float F = clamp(DF * max(dot(L,N),0.0),0.0,1.0);
diff += vec4(F * mat_diff * pl_color[i],1.0);
spec += vec4(DF * S * mat_spec * pl_specular[i],1.0);
}
//fog
float fogDensity = 0.0;
if (depth > uFogFar) {
fogDensity = 1.0;
} else if(depth > uFogNear) {
float newDepth = depth-uFogNear;
fogDensity = newDepth/(uFogFar-uFogNear);
} else if (depth < uFogNear) {
fogDensity = 0.0;
}
vec4 vecFog = uVecFog * fogDensity;
//texturecolor
vec4 colortex = texture2D(uTexture,vec2(texcoord.x,-texcoord.y));
//final fragmentcolor
gl_FragColor = mix((ambient + colortex * (diff + vec4(mat_diff * mat_elum,1.0)) + spec),vecFog,fogDensity);
}
There are give following variable-values, where:
- pl_ are pointlightvalues,
- dl_ are directionallightvalues,
- mat_diff is the diffuse-value for the material,
- also mat_spec, mat_elum, mat_shine for material specular, elumination and shininess,
- N = Normal,
- V = View-Vector to Vertex,
- L = Light-Vector to Vertex,
- R = Reflection Vector,
- D = Lightdistance,
- DF = Distancefactor for blending light,
- S = Calculated Specularityfactor,
- F = Dotproduct from L,N
- and diff, spec vectorsum of the calculated diffuse/spec-vectors.
Now there occures following problem:
Pointlights themself run perfect. Directionallights themself run perfect, too.
But together they won't work very well. I'll precise the problem:
I've got 4 Pointlights and 4 Directionallights for maximum. For testing i've used 4 Pointlights and 1 Directionallight.
- If i commented out the for-loop calculation the directionallights, i got following result:
- If i commented out the for-loop for pointlights, i got following result:
- If i enabled both for-loops, following happens:
The directionallight seems to be ignored completly. The car is lighted only by the pointlights instead.
So, what I found out:
If I change the order of the for-loops, pointlights -> directionallights, instead of dirlights -> pointlights, the effect swaps: pointlights are ignored, directionallight is computed.
What I found out too, this problem must be up to: pow() function or dividing by zero, but:
If I comment out follwing lines of code:
spec = vec4(S * mat_spec * dl_specular[i],1.0);
spec += vec4(DF * S * mat_spec * pl_specular[i],1.0);
i can get following result:
-> The diffuse color is computed normally for point AND directionallights!
So I thought of how to eliminate this problem, because I thought about 0 powered by 0 leads to a math error. Dividing by zero, too....
How can I avoid values that are undefined?
isnan() or isinf() functions seem not to exist in android-glsl.
Andy solutions or tipps what I could do better, that this will work?
I would be glad about any answer if it could help a little bit. This makes me mad since 3 weeks. Thanks so much for helpers! :)
More information about the code:
https://github.com/Chrise55/Llama3D
Chrise
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论