4.1 Regression: Cycles: Area lamp artifacts #120119

New Issue

Carlos · 2024-03-31T21:27:19+02:00

Carlos commented

2024-03-31 21:27:19 +02:00

System Information
Operating system: Linux-6.1.77-2-MANJARO-x86_64-with-glibc2.39 64 Bits, X11 UI
Graphics card: NVIDIA GeForce RTX 3070/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 545.29.06

Blender Version
Broken: version: 4.2.0 Alpha, branch: main, commit date: 2024-03-29 00:22, hash: af7a34dee7d4
Worked: 4.0

Short description of error
It just happens in Cycles
Using area lamps, whether square or disck, produces a reflection with an edge on the object
Things that eliminate it:
Lamp
-use a size smaller than .47 or greater than 1.51 m
Material
-disconect or reduce the strenght of the normal map
- reduce the emission less than .5 or remove it
-change emission sampling to none
Cycles
-uncheck light tree.

Exact steps for others to reproduce the error
I couldn't reproduce it in a fresh new file with other objets, it only happens to this object whether I copy it or append it to a new file.

Thank you

**System Information** Operating system: Linux-6.1.77-2-MANJARO-x86_64-with-glibc2.39 64 Bits, X11 UI Graphics card: NVIDIA GeForce RTX 3070/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 545.29.06 **Blender Version** Broken: version: 4.2.0 Alpha, branch: main, commit date: 2024-03-29 00:22, hash: `af7a34dee7d4` Worked: 4.0 **Short description of error** It just happens in Cycles Using area lamps, whether square or disck, produces a reflection with an edge on the object Things that eliminate it: Lamp -use a size smaller than .47 or greater than 1.51 m Material -disconect or reduce the strenght of the normal map - reduce the emission less than .5 or remove it -change emission sampling to none Cycles -uncheck light tree. ![imagen](/attachments/577dd2b8-6c85-42e9-a67e-8063224fd9d1) ![imagen](/attachments/c79e6064-08e1-479a-9c17-d91c58fe6171) **Exact steps for others to reproduce the error** I couldn't reproduce it in a fresh new file with other objets, it only happens to this object whether I copy it or append it to a new file. Thank you

imagen.png

616 KiB

lightartifacts.blend

48 MiB

imagen.png

451 KiB

Carlos added the

labels 2024-03-31 21:27:20 +02:00

Alaska commented

2024-04-01 02:09:04 +02:00

I can confirm this issue.

It occurs because the forward and NEE parts of the light rendering deviate from each other. Here's images showing the issue:

Forward only, but still applying MIS weighting	NEE Only, but still applying MIS weighting

The fact they're different brightnesses, and have patches on them, is expected. The issue is that the patches aren't the same shape, and the artifacts occur wherever the patches don't align.

I'll run a quick bisect to see which commit triggered this issue.

Marking as high as this is a regression in 4.1

I can confirm this issue. It occurs because the forward and NEE parts of the light rendering deviate from each other. Here's images showing the issue: |Forward only, but still applying MIS weighting|NEE Only, but still applying MIS weighting| |-|-| |![Forward Only.png](/attachments/7f35aced-5d4e-4794-896d-19e7c3ea0cdb)|![NEE Only.png](/attachments/beab873c-f44e-4c28-95fa-b155ddf17b11)| The fact they're different brightnesses, and have patches on them, is expected. The issue is that the patches aren't the same shape, and the artifacts occur wherever the patches don't align. I'll run a quick bisect to see which commit triggered this issue. Marking as high as this is a regression in 4.1

NEE Only.png

449 KiB

Forward Only.png

530 KiB

Alaska added

and removed

labels 2024-04-01 02:09:26 +02:00

Alaska changed title from ~~Cycles: Area lamp artifacts~~ to 4.1 Regression: Cycles: Area lamp artifacts

2024-04-01 02:09:55 +02:00

Alaska commented

2024-04-01 03:52:17 +02:00

Bisecting points to 36e603c430 being the culprit commit. Specifically this change in intern/cycles/kernel/integrator/shade_surface.h:

-    INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = sd->N;
+    INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = sc->N;

Reverting that change fixes the issue. This code change also explains why this issue is only present when using a normal map.

CC @Sergey can you take a look into this? I'm not sure why you made the change to sc->N and I'm not sure if reverting back to sd->N has a negative impact on the feature you implemented with that commit.

Bisecting points to 36e603c430ca90a4a19574a5071db54a7cebcb39 being the culprit commit. Specifically this change in `intern/cycles/kernel/integrator/shade_surface.h`: ``` - INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = sd->N; + INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = sc->N; ``` Reverting that change fixes the issue. This code change also explains why this issue is only present when using a normal map. CC @Sergey can you take a look into this? I'm not sure why you made the change to `sc->N` and I'm not sure if reverting back to `sd->N` has a negative impact on the feature you implemented with that commit.

Sergey Sharybin commented

2024-04-02 13:23:17 +02:00

@Alaska Unfortunately, I do not quickly recall details, so would need to dig into the code again to get exact details. The vague idea i recall is that we wanted to make it so if an artist plugs normal to a shader it is used as-is.

On a quick note I can only confirm that reverting back to sd->N does indeed have an affect on the feature. I've attached a simplified .blend file which uses this functionality. If the code is rolled back to sd->N and you render the file, you will see "fireflies" in the shadow areas of the rock.

@Alaska Unfortunately, I do not quickly recall details, so would need to dig into the code again to get exact details. The vague idea i recall is that we wanted to make it so if an artist plugs normal to a shader it is used as-is. On a quick note I can only confirm that reverting back to `sd->N` does indeed have an affect on the feature. I've attached a simplified .blend file which uses this functionality. If the code is rolled back to `sd->N` and you render the file, you will see "fireflies" in the shadow areas of the rock.

reef_shot3.blend

4.5 MiB

👍 1

Sergey Sharybin self-assigned this 2024-04-03 13:50:24 +02:00

Sergey Sharybin referenced a pull request that will close this issue

2024-04-03 17:17:44 +02:00

Fix #120119: Area lamp artifacts in Cycles with light trees #120216

Sergey Sharybin referenced this issue

2024-04-03 17:25:06 +02:00

Fix #120119: Area lamp artifacts in Cycles with light trees #120216

Sergey Sharybin referenced this issue from a commit

2024-04-05 10:11:46 +02:00