blender
/
blender
130
398
Fork 573
Code Issues 6.2k Pull Requests 745 Projects 19 Wiki Activity

WIP: Fix: Respect Blender Cycles setting for GPU denoising #118841

Draft
Nikita Sirgienko wants to merge 1 commits from Sirgienko/blender:address_gpu_denoising_settings_respect into main
pull from: Sirgienko/blender:address_gpu_denoising_settings_respect
merge into: blender:main

When changing the target branch, be careful to rebase the branch in your fork to match. See documentation.
Conversation 6 Commits 1 Files Changed 14 +266 -119
Nikita Sirgienko commented 2024-02-28 10:20:23 +01:00
Member
Copy Link

Previously, GPU denoisers were ignoring settings about render
configuration and were using any available GPU. With these changes,
GPU denoisers will use the device selected in Blender Cycles
settings.
This allows any GPU denoiser to be used with CPU rendering.

Previously, GPU denoisers were ignoring settings about render configuration and were using any available GPU. With these changes, GPU denoisers will use the device selected in Blender Cycles settings. This allows any GPU denoiser to be used with CPU rendering.
Nikita Sirgienko added the
Module
Render & Cycles
 label 2024-02-28 10:20:51 +01:00
Nikita Sirgienko added this to the 4.1 milestone 2024-02-28 10:20:58 +01:00
Nikita Sirgienko added this to the Render & Cycles project 2024-02-28 10:21:04 +01:00
Xavier Hallade was assigned by Nikita Sirgienko 2024-02-28 10:21:12 +01:00
Nikita Sirgienko requested review from Brecht Van Lommel 2024-02-28 10:21:26 +01:00
Nikita Sirgienko commented 2024-02-28 10:32:50 +01:00
Author
Member
Copy Link

@brecht Can you please take a look on this solution? It is about respecting Blender setting for GPU denoising device selection (both OptiX and OIDN). I think it make sense to get this fix to clarify UX for the users, but the patch is not small, so I would understand if you would want this to be postponed for 4.2.
Also, I could actually address this behavior with more easy to implement approach that "Device" UI option, which is defining device used for render, should affect denoising device as well - but then it would be not possible to use GPU denoising during CPU render, which may be not very popular configuration, but I would expected that some people may use it so it wouldn't be nice to remove this functionality even if wasn't intended initially.

@brecht Can you please take a look on this solution? It is about respecting Blender setting for GPU denoising device selection (both OptiX and OIDN). I think it make sense to get this fix to clarify UX for the users, but the patch is not small, so I would understand if you would want this to be postponed for 4.2. Also, I could actually address this behavior with more easy to implement approach that "Device" UI option, which is defining device used for render, should affect denoising device as well - but then it would be not possible to use GPU denoising during CPU render, which may be not very popular configuration, but I would expected that some people may use it so it wouldn't be nice to remove this functionality even if wasn't intended initially.
Nikita Sirgienko commented 2024-02-28 11:00:03 +01:00
Author
Member
Copy Link

A clarification about how it is working right now and how it suppose to work with this change.
Current behavior - despite Cycles preference (red arrows), denoising is happening (yellow arrow) on first found GPU device:
image
Behaviour with this changes - if None or CPU devices or GPU devices, which are not supporting GPU denoising are selected in the Cycles preferences, then GPU denoising options (Optix denoiser and "Use GPU" OIDN option) will be grayout out and won't be applied, as denoising will happen on CPU:
image
image

And when GPU devices, which is supporting GPU denoising, is selected in the Cycles preferences, then this GPU denoising options won't be grayed out and will apply according to the selection:
image

A clarification about how it is working right now and how it suppose to work with this change. Current behavior - despite Cycles preference (red arrows), denoising is happening (yellow arrow) on first found GPU device: ![image](/attachments/810001e7-b133-468a-ab78-72b2092860ec) Behaviour with this changes - if None or CPU devices or GPU devices, which are not supporting GPU denoising are selected in the Cycles preferences, then GPU denoising options (Optix denoiser and "Use GPU" OIDN option) will be grayout out and won't be applied, as denoising will happen on CPU: ![image](/attachments/c25e17b5-8c4c-4313-b8cd-bbae044e6769) ![image](/attachments/c1d395ef-1f77-4451-aec5-fbef9b02d676) And when GPU devices, which is supporting GPU denoising, is selected in the Cycles preferences, then this GPU denoising options won't be grayed out and will apply according to the selection: ![image](/attachments/240e91c1-9fd2-4da2-8614-77f39ada26e6)
image.png
1.1 MiB
image.png
1.1 MiB
image.png
562 KiB
image.png
867 KiB
Nikita Sirgienko commented 2024-02-28 11:06:09 +01:00
Author
Member
Copy Link

So in a nutshell: Right now there are a per-scene denoisers configuration options, which are not aligned with user's Cycles preferences and not respecting them. With this change, it will be addressed and GPU denoising device selection will work in the same approach as "Device: GPU compute" is working right now - preferences will determine which device will be used and if compatible device is not selected, then the UI option will be grayed out and CPU fallback will be used.

So in a nutshell: Right now there are a per-scene denoisers configuration options, which are not aligned with user's Cycles preferences and not respecting them. With this change, it will be addressed and GPU denoising device selection will work in the same approach as "Device: GPU compute" is working right now - preferences will determine which device will be used and if compatible device is not selected, then the UI option will be grayed out and CPU fallback will be used.
Nikita Sirgienko force-pushed address_gpu_denoising_settings_respect from ffb4f5bdf6 to 07968d7e0e 2024-02-28 14:51:41 +01:00 Compare
Nikita Sirgienko force-pushed address_gpu_denoising_settings_respect from 07968d7e0e to 83b084ab3d 2024-02-28 14:53:31 +01:00 Compare
Brecht Van Lommel commented 2024-02-28 18:36:17 +01:00
Owner
Copy Link

This seems too risky this close to the release, it's hard to get this properly user tested in such a short time. Particularly the support for CPU rendering I would rather postpone. If there is a bugfix we can consider it, but hopefully the change is simpler then.

This adds a denoise device in Session, but there already exists a local_denoiser_device_ in Denoiser. Isn't it possible to use that instead?

The way I understand the code in Denoiser::ensure_denoiser_device now, is that it will already try to use one of the path tracing devices for denoising. So in that sense it respects the choice of GPUs, except if you happen to be path tracing with a GPU that doesn't support OIDN and have another GPU that does. We could change the logic so it doesn't do that, but it seems a pretty rare corner case.

This seems too risky this close to the release, it's hard to get this properly user tested in such a short time. Particularly the support for CPU rendering I would rather postpone. If there is a bugfix we can consider it, but hopefully the change is simpler then. This adds a denoise device in `Session`, but there already exists a `local_denoiser_device_` in `Denoiser`. Isn't it possible to use that instead? The way I understand the code in `Denoiser::ensure_denoiser_device` now, is that it will already try to use one of the path tracing devices for denoising. So in that sense it respects the choice of GPUs, except if you happen to be path tracing with a GPU that doesn't support OIDN and have another GPU that does. We could change the logic so it doesn't do that, but it seems a pretty rare corner case.
Nikita Sirgienko commented 2024-02-29 23:41:00 +01:00
Author
Member
Copy Link

This seems too risky this close to the release, it's hard to get this properly user tested in such a short time. Particularly the support for CPU rendering I would rather postpone. If there is a bugfix we can consider it, but hopefully the change is simpler then.

Understandable, and unfortunately this is the easiest solution which I have found so far, so I suppose it will be indeed change for the 4.2 LST, I will update milestone information accordingly.

This adds a denoise device in Session, but there already exists a local_denoiser_device_ in Denoiser. Isn't it possible to use that instead?

No, I don't think that is possible to use only local_denoiser_device_, because Denoiser class is not taking into account Blender preferences and Scene settings when determining the local_denoiser_device ("Denoiser device created to perform denoising when none of the rendering devices are capable of denoising."). This is because the Denoiser class does not have access to these settings, as only few Blender Cycles classes and functions are having it - intern/cycles/blender/session.cpp and intern/cycles/blender/python.cpp, in particular.

However, the Session class does have indirect access to preference information, as this data is passed in the form of a SessionParams struct to the Session constructor. And the class responsible for creating the Session object is expected to populate the SessionParams structure with the appropriate information based on the Preferences and Scene settings. This for example how it is done in the BlenderSession::create_session() method and in the BlenderSync::get_session_params method, which constructs the SessionParams structure.

Essentially, once denoising is allowed on devices that differ from the path tracing devices, we should employ the same method used for path trace devices to ensure reasonable UI/UX behavior. The denoising devices should be determined based on the Cycles preferences and Scene settings by a class that has access to both. Then this information in a form of DeviceInfo should be passed through various Cycles classes, until it will reach the relevant class (in this case a class above Denoiser due to memory management approach). The relevant class then will create a Device object from this information and then Denoiser will utilize this Device object for computations.

The way I understand the code in Denoiser::ensure_denoiser_device now, is that it will already try to use one of the path tracing devices for denoising. So in that sense it respects the choice of GPUs, except if you happen to be path tracing with a GPU that doesn't support OIDN and have another GPU that does. We could change the logic so it doesn't do that, but it seems a pretty rare corner case.

Approach it will already try to use one of the path tracing devices for denoising is leading to the situation, that we could easily use GPU denoising with GPU path tracing and CPU denoising with CPU path tracing. But as soon as we want to have CPU render and GPU denoising - ensure_denoiser_device will work in quite strange way, in my opinion.

For example, for OIDN, ensure_denoiser_device will never able to find GPU device for denoising when path tracing is happening on CPU - because path tracing CPU device will be checked first in ensure_denoiser_device, and CPU device (almost always) is supporting OIDN denoiser, so no futher search will be conducted.

For the OptiX denoiser, this is not an issue because OptiX is not supported on CPU. However, there is still an uncontrollable aspect regarding which NVIDIA device will be used for GPU denoising when path tracing is happening on CPU. As I see, the code in "find_best_device" prefers an NVIDIA GPU which is connected to a display. This may be acceptable for the Blender's user, but if they prefer otherwise, then there is currently no way to control this behavior. Because the selection of an NVIDIA GPU is hardcoded in the source code.

And, as I have mentioned above, Denoiser class do not have access to the Cycles preference and Scene settings, so this information can't be used in ensure_denoiser_device - as I see, this is a reason why user have so little control over ensure_denoiser_device behaviour.

Another issue, in my view, is that we never actually use the Denoiser::ensure_denoiser_device method directly; instead, it is called from its subclasses, such as OptiXDenoiser, OIDNDenoiser, and OIDNDenoiserGPU. In these instances, we have already partially determined the device we intend to use for denoising—if we create an OptiXDenoiser object, for example, it implies we want OptiX denoising on an NVIDIA GPU. Therefore, I believe that ideally, we should not have the ensure_denoiser_device method at all. The decision on which device to use for denoising should be made at the same level where we decide which denoiser subclass to instantiate.

> This seems too risky this close to the release, it's hard to get this properly user tested in such a short time. Particularly the support for CPU rendering I would rather postpone. If there is a bugfix we can consider it, but hopefully the change is simpler then. Understandable, and unfortunately this is the easiest solution which I have found so far, so I suppose it will be indeed change for the 4.2 LST, I will update milestone information accordingly. >This adds a denoise device in Session, but there already exists a local_denoiser_device_ in Denoiser. Isn't it possible to use that instead? No, I don't think that is possible to use only local_denoiser_device_, because `Denoiser` class is not taking into account Blender preferences and Scene settings when determining the local_denoiser_device ("Denoiser device created to perform denoising when none of the rendering devices are capable of denoising."). This is because the Denoiser class does not have access to these settings, as only few Blender Cycles classes and functions are having it - intern/cycles/blender/session.cpp and intern/cycles/blender/python.cpp, in particular. However, the `Session` class does have indirect access to preference information, as this data is passed in the form of a `SessionParams` struct to the `Session` constructor. And the class responsible for creating the `Session` object is expected to populate the SessionParams structure with the appropriate information based on the Preferences and Scene settings. This for example how it is done in the `BlenderSession::create_session()` method and in the `BlenderSync::get_session_params` method, which constructs the SessionParams structure. Essentially, once denoising is allowed on devices that differ from the path tracing devices, we should employ the same method used for path trace devices to ensure reasonable UI/UX behavior. The denoising devices should be determined based on the Cycles preferences and Scene settings by a class that has access to both. Then this information in a form of `DeviceInfo` should be passed through various Cycles classes, until it will reach the relevant class (in this case a class above `Denoiser` due to memory management approach). The relevant class then will create a Device object from this information and then `Denoiser` will utilize this `Device` object for computations. > The way I understand the code in Denoiser::ensure_denoiser_device now, is that it will already try to use one of the path tracing devices for denoising. So in that sense it respects the choice of GPUs, except if you happen to be path tracing with a GPU that doesn't support OIDN and have another GPU that does. We could change the logic so it doesn't do that, but it seems a pretty rare corner case. Approach `it will already try to use one of the path tracing devices for denoising` is leading to the situation, that we could easily use GPU denoising with GPU path tracing and CPU denoising with CPU path tracing. But as soon as we want to have CPU render and GPU denoising - `ensure_denoiser_device` will work in quite strange way, in my opinion. For example, for OIDN, `ensure_denoiser_device` will never able to find GPU device for denoising when path tracing is happening on CPU - because path tracing CPU device will be checked first in `ensure_denoiser_device`, and CPU device (almost always) is supporting OIDN denoiser, so [no futher search will be conducted](https://projects.blender.org/blender/blender/src/commit/a972988a207051f931725fe964f8eae10818028c/intern/cycles/integrator/denoiser.cpp#L218). For the OptiX denoiser, this is not an issue because OptiX is not supported on CPU. However, there is still an uncontrollable aspect regarding which NVIDIA device will be used for GPU denoising when path tracing is happening on CPU. As I see, the code in "find_best_device" prefers an NVIDIA GPU which is connected to a display. This may be acceptable for the Blender's user, but if they prefer otherwise, then there is currently no way to control this behavior. Because the selection of an NVIDIA GPU is hardcoded in the source code. And, as I have mentioned above, `Denoiser` class do not have access to the Cycles preference and Scene settings, so this information can't be used in `ensure_denoiser_device` - as I see, this is a reason why user have so little control over `ensure_denoiser_device` behaviour. Another issue, in my view, is that we never actually use the `Denoiser::ensure_denoiser_device` method directly; instead, it is called from its subclasses, such as `OptiXDenoiser`, `OIDNDenoiser`, and `OIDNDenoiserGPU`. In these instances, we have already partially determined the device we intend to use for denoising—if we create an `OptiXDenoiser` object, for example, it implies we want `OptiX` denoising on an NVIDIA GPU. Therefore, I believe that ideally, we should not have the `ensure_denoiser_device` method at all. The decision on which device to use for denoising should be made at the same level where we decide which denoiser subclass to instantiate.
Nikita Sirgienko modified the milestone from 4.1 to 4.2 LTS 2024-03-04 21:00:40 +01:00
Sirgienko changed target branch from blender-v4.1-release to main 2024-03-04 21:01:07 +01:00
Brecht Van Lommel commented 2024-03-05 16:39:18 +01:00
Owner
Copy Link

I think we should not have both local_denoiser_device_ in Denoiser, and denoise_device in Session, it should be one or the other. So if this denoiser device is added at the Session level, that should then fully replace the denoiser device detection and creation at the Denoiser level.

It's hard for me to guess which way leads to the simplest implementation. But I'm guessing it's best to remove a bit of abstraction in Denoiser, and make Denoiser::create responsible for creating the denoiser device before the Denoiser instance. That way the logic does not get spread out to the Session, and the confusing flow with subclasses creating the device is avoided.

I think we should not have both `local_denoiser_device_` in `Denoiser`, and `denoise_device` in `Session`, it should be one or the other. So if this denoiser device is added at the Session level, that should then fully replace the denoiser device detection and creation at the Denoiser level. It's hard for me to guess which way leads to the simplest implementation. But I'm guessing it's best to remove a bit of abstraction in Denoiser, and make `Denoiser::create` responsible for creating the denoiser device before the `Denoiser` instance. That way the logic does not get spread out to the Session, and the confusing flow with subclasses creating the device is avoided.
Nikita Sirgienko changed title from Fix: Respect Blender Cycles setting for GPU denoising to WIP: Fix: Respect Blender Cycles setting for GPU denoising 2024-03-11 18:05:56 +01:00
This pull request has changes conflicting with the target branch.
  • intern/cycles/blender/addon/properties.py
  • intern/cycles/blender/addon/ui.py
  • intern/cycles/integrator/denoiser.cpp
View command line instructions.

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u address_gpu_denoising_settings_respect:Sirgienko-address_gpu_denoising_settings_respect
git checkout Sirgienko-address_gpu_denoising_settings_respect
Sign in to join this conversation.
Reviewers
No reviewers
Brecht Van Lommel
Labels
Clear labels   
Interest
Alembic

  
Interest
Animation & Rigging

  
Interest
Asset Browser

  
Interest
Asset Browser Project Overview

  
Interest
Audio

  
Interest
Automated Testing

  
Interest
Blender Asset Bundle

  
Interest
BlendFile

  
Interest
Collada

  
Interest
Compatibility
This issue affects/is about backward or forward compatibility

  
Interest
Compositing

  
Interest
Core

  
Interest
Cycles

  
Interest
Dependency Graph

  
Interest
Development Management

  
Interest
EEVEE

  
Interest
EEVEE & Viewport

  
Interest
Freestyle

  
Interest
Geometry Nodes

  
Interest
Grease Pencil

  
Interest
ID Management

  
Interest
Images & Movies

  
Interest
Import Export

  
Interest
Line Art

  
Interest
Masking

  
Interest
Metal

  
Interest
Modeling

  
Interest
Modifiers

  
Interest
Motion Tracking

  
Interest
Nodes & Physics

  
Interest
OpenGL

  
Interest
Overlay

  
Interest
Overrides

  
Interest
Performance

  
Interest
Physics

  
Interest
Pipeline, Assets & IO

  
Interest
Platforms, Builds & Tests

  
Interest
Python API

  
Interest
Render & Cycles

  
Interest
Render Pipeline

  
Interest
Sculpt, Paint & Texture

  
Interest
Text Editor

  
Interest
Translations

  
Interest
Triaging

  
Interest
Undo

  
Interest
USD

  
Interest
User Interface

  
Interest
UV Editing

  
Interest
VFX & Video

  
Interest
Video Sequencer

  
Interest
Virtual Reality

  
Interest
Vulkan

  
Interest
Wayland
Issues relating to Wayland windowing support.

  
Interest
Workbench

  
Interest: X11

Issues relating to Xorg/X11 support.

  
Legacy
Blender 2.8 Project

  
Legacy
Milestone 1: Basic, Local Asset Browser

  
Legacy
OpenGL Error

  
Meta
Good First Issue

  
Meta
Papercut

  
Meta
Retrospective

  
Meta
Security
Issues relating to security: https://wiki.blender.org/wiki/Process/Vulnerability_Reports

  
Module
Animation & Rigging

  
Module
Core

  
Module
Development Management

  
Module
EEVEE & Viewport

  
Module
Grease Pencil

  
Module
Modeling

  
Module
Nodes & Physics

  
Module
Pipeline, Assets & IO

  
Module
Platforms, Builds & Tests

  
Module
Python API

  
Module
Render & Cycles

  
Module
Sculpt, Paint & Texture

  
Module
Triaging

  
Module
User Interface

  
Module
VFX & Video

  
Platform
FreeBSD

  
Platform
Linux

  
Platform
macOS

  
Platform
Windows

  
Priority
High

  
Priority
Low

  
Priority
Normal

  
Priority
Unbreak Now!

  
Status
Archived

  
Status
Confirmed

  
Status
Duplicate

  
Status
Needs Info from Developers

  
Status
Needs Information from User

  
Status
Needs Triage

  
Status
Resolved

  
Type
Bug

  
Type
Design

  
Type
Known Issue

  
Type
Patch

  
Type
Report

  
Type
To Do

No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset Browser
Interest
Asset Browser Project Overview
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
EEVEE & Viewport
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
EEVEE & Viewport
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Priority
High
Priority
Low
Priority
Normal
Priority
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
Milestone
Clear milestone
No items
No Milestone
4.2 LTS
Projects
Clear projects
No project
Render & Cycles
Assignees
Clear assignees
No Assignees
Xavier Hallade
2 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#118841
No description provided.
Delete Branch "Sirgienko/blender:address_gpu_denoising_settings_respect"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?