Eevee: Bloom artifacts when triangle has colinear vertices #51979

New Issue

Dalai Felinto · 2017-07-06T14:58:04+02:00

Dalai Felinto commented

2017-07-06 14:58:04 +02:00

Blender Version
Broken: c217d518ef

Short description of error
Big black squares when using bloom with bad triangle.

Exact steps for others to reproduce the error
Open this file and rotate around, you should see black squares taking over the screen:
triangle-bloom.blend

If you want to create the file from scratch:

Enable Bloom
Make a triangle with its three vertices aligned
(e.g., add a plane, delete one of the vertices, make a face from the other three vertices, move one of the vertices to the midpoint between the other two vertices).

**Blender Version** Broken: c217d518ef **Short description of error** Big black squares when using bloom with bad triangle. **Exact steps for others to reproduce the error** Open this file and rotate around, you should see black squares taking over the screen: [triangle-bloom.blend](https://archive.blender.org/developer/F653761/triangle-bloom.blend) If you want to create the file from scratch: * Enable Bloom * Make a triangle with its three vertices aligned (e.g., add a plane, delete one of the vertices, make a face from the other three vertices, move one of the vertices to the midpoint between the other two vertices).

Dalai Felinto commented

2017-07-06 14:58:04 +02:00

Changed status to: 'Open'

Dalai Felinto commented

2017-07-06 14:58:04 +02:00

Added subscriber: @dfelinto

Clément Foucault was assigned by Dalai Felinto

2017-07-06 14:58:32 +02:00

Clément Foucault commented

2017-07-06 16:04:29 +02:00

Fixed in d02711ed88 but the real problem must come from somwhere else like bad encoding or interpolation.

Fixed in d02711ed883e but the real problem must come from somwhere else like bad encoding or interpolation.

Clément Foucault removed their assignment 2017-07-06 16:22:18 +02:00

Campbell Barton was assigned by Clément Foucault

2017-07-06 16:22:18 +02:00

Clément Foucault commented

2017-07-06 16:22:18 +02:00

Added subscriber: @fclem

Clément Foucault commented

2017-07-06 16:22:18 +02:00

This is a problem between the vbo creation and the fragment shader.
Putting a vec3(0.0) does not reproduce the same artifact, and normals components are set to 0 in the Gawain packet format, so problem is inbetween. Revert d02711ed88 and enable bloom to see if it fails.

This is a problem between the vbo creation and the fragment shader. Putting a vec3(0.0) does not reproduce the same artifact, and normals components are set to 0 in the Gawain packet format, so problem is inbetween. Revert d02711ed883e and enable bloom to see if it fails.

Campbell Barton commented

2017-07-13 17:57:39 +02:00

Looking into this and eevee_surface_lit is getting a non-finite normal input.

Since this functions input is setup externally GPU_link(..) , its not so simple to sanitize each arg thats passed in.
And not sure we should be adding zero-length checks to generic normalization functions either.

This is a workaround (similar to your fix) but checks for nan instead of using an offset.

diff --git a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
index c8fd57412ef..cae330bac96 100644
--- a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
@@ -30,7 +30,17 @@ vec3 eevee_surface_lit(vec3 N, vec3 albedo, vec3 f0, float roughness, float ao)
 	float roughnessSquared = roughness * roughness;
 
 	vec3 V = cameraVec;
-	N = normalize(N);
+#if 0
+    N = normalize(N);
+#else
+    {
+        float l = length(N);
+        if (isnan(l)) {
+            return vec3(0);
+        }
+        N = N / l;
+    }
+#endif
 
 	vec4 rand = texture(utilTex, vec3(gl_FragCoord.xy / LUT_SIZE, 2.0));

Looking into this and `eevee_surface_lit` is getting a non-finite normal input. Since this functions input is setup externally `GPU_link(..)` , its not so simple to sanitize each arg thats passed in. And not sure we should be adding zero-length checks to generic normalization functions either. This is a workaround (similar to your fix) but checks for nan instead of using an offset. ``` diff --git a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl index c8fd57412ef..cae330bac96 100644 --- a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl +++ b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl @@ -30,7 +30,17 @@ vec3 eevee_surface_lit(vec3 N, vec3 albedo, vec3 f0, float roughness, float ao) float roughnessSquared = roughness * roughness; vec3 V = cameraVec; - N = normalize(N); +#if 0 + N = normalize(N); +#else + { + float l = length(N); + if (isnan(l)) { + return vec3(0); + } + N = N / l; + } +#endif vec4 rand = texture(utilTex, vec3(gl_FragCoord.xy / LUT_SIZE, 2.0)); ```

Clément Foucault commented

2017-07-14 00:25:09 +02:00

Ha I did not know about isnan ! sure that makes it a more elegant solution. Thanks!

Ha I did not know about `isnan` ! sure that makes it a more elegant solution. Thanks!

blender-admin commented

2017-07-14 07:27:08 +02:00

This issue was referenced by 886ea37572

This issue was referenced by 886ea375723df23a219af38688fbad4b7766fb3a

Campbell Barton commented

2017-07-14 07:40:23 +02:00

Changed status from 'Open' to: 'Resolved'

Campbell Barton closed this issue

2017-07-14 07:40:23 +02:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Eevee: Bloom artifacts when triangle has colinear vertices #51979