blender
/
blender
132
398
Fork 577
Code Issues 6.2k Pull Requests 748 Projects 19 Wiki Activity

Compositor: Enable GPU denoising for OpenImageDenoise #115242

Open
Stefan Werner wants to merge 1 commits from Stefan_Werner/blender:compositor_gpu_denoising into main
pull from: Stefan_Werner/blender:compositor_gpu_denoising
merge into: blender:main

When changing the target branch, be careful to rebase the branch in your fork to match. See documentation.
Conversation 5 Commits 1 Files Changed 1 +33 -2
Stefan Werner commented 2023-11-21 16:27:36 +01:00
Member
Copy Link

When using OpenImageDenoise 2, the denoiser now gets to decide what device to use. Device buffers will be created where necessary. If the device supports system RAM access, it will be used directly with zero copying.

Ref #115045

When using OpenImageDenoise 2, the denoiser now gets to decide what device to use. Device buffers will be created where necessary. If the device supports system RAM access, it will be used directly with zero copying. Ref #115045
Stefan Werner added 1 commit 2023-11-21 16:27:48 +01:00
5c2b7cb01a Compositor: Enabled GPU denoising for OpenImageDenoise 
When using OpenImageDenoise 2, the denoiser now gets to decide
what device to use. Device buffers will be created where necessary.
If the device supports system RAM access, it will be used directly
with zero copying.
Stefan Werner added this to the 4.1 milestone 2023-11-21 16:28:06 +01:00
Stefan Werner added the
Interest
Compositing
 label 2023-11-21 16:28:30 +01:00
Stefan Werner added this to the Compositing project 2023-11-21 16:28:38 +01:00
Brecht Van Lommel requested review from Brecht Van Lommel 2023-11-22 18:43:47 +01:00
Brecht Van Lommel requested review from Sergey Sharybin 2023-11-22 18:43:47 +01:00
Brecht Van Lommel requested review from Omar Emara 2023-11-22 18:43:47 +01:00
Brecht Van Lommel commented 2023-11-22 18:47:16 +01:00
Owner
Copy Link

I don't think that we should be using GPU devices for compositing by default, when everything else runs on the CPU? To me this seems like it should be tied to using the realtime GPU compositor.

I don't think that we should be using GPU devices for compositing by default, when everything else runs on the CPU? To me this seems like it should be tied to using the realtime GPU compositor.
Brecht Van Lommel commented 2023-11-22 18:53:44 +01:00
Owner
Copy Link

When systemMemorySupported is available, does that mean this will use little GPU memory, so that there is very little chance for this to fail? Or does it mean it may still need to copy and run out of GPU memory?

When `systemMemorySupported` is available, does that mean this will use little GPU memory, so that there is very little chance for this to fail? Or does it mean it may still need to copy and run out of GPU memory?
Stefan Werner commented 2023-11-23 11:58:19 +01:00
Author
Member
Copy Link

systemMemorySupported means that the selected device can use input/output buffers in host memory, allocated by malloc, directly, without copying. Looking at the OIDN implementation, this is currently only true for the CPU device, I could imaging that in the future this could be also return true on systems with unified memory architecture.

For what it's worth, OIDN's automatic selection is intended to use the best device it can find - so on systems with a fast CPU and a slow integrated GPU, it should pick the iGPU. CPU denoising still takes 100s of ms, which can add up when running the compositor on a sequence of already rendered frames.

If there is a way to detect and connect to the real time compositor, sure, I'll be happy to look into that.

`systemMemorySupported` means that the selected device can use input/output buffers in host memory, allocated by `malloc`, directly, without copying. Looking at the OIDN implementation, this is currently only true for the CPU device, I could imaging that in the future this could be also return true on systems with unified memory architecture. For what it's worth, OIDN's automatic selection is intended to use the best device it can find - so on systems with a fast CPU and a slow integrated GPU, it should pick the iGPU. CPU denoising still takes 100s of ms, which can add up when running the compositor on a sequence of already rendered frames. If there is a way to detect and connect to the real time compositor, sure, I'll be happy to look into that.
Sergey Sharybin requested changes 2023-11-23 12:31:37 +01:00
Sergey Sharybin left a comment
Owner
Copy Link

The automatic detection of the device is something I am not sure about:

  • The render farms might not be happy with GPU selected without explicit "permission"
  • Render farms might as well run multiple Blender instances in parallel. If denoiser picks up same GPU for all instances this is less than ideal.
  • Artists might have a secondary (non-display) GPU and be rendering shot in background, while tweaking compositor setup meanwhile. That is another example when automatic selection of the fastest device will not be desired.

When running compositor on CPU with an actual node graph even the current OIDN node is not that often a bottleneck. So I'd follow the same rule as for the Cycles rendering: when compositing on GPU use GPU denoiser, when compositing on CPU use CPU denoiser.

The automatic detection of the device is something I am not sure about: - The render farms might not be happy with GPU selected without explicit "permission" - Render farms might as well run multiple Blender instances in parallel. If denoiser picks up same GPU for all instances this is less than ideal. - Artists might have a secondary (non-display) GPU and be rendering shot in background, while tweaking compositor setup meanwhile. That is another example when automatic selection of the fastest device will not be desired. When running compositor on CPU with an actual node graph even the current OIDN node is not that often a bottleneck. So I'd follow the same rule as for the Cycles rendering: when compositing on GPU use GPU denoiser, when compositing on CPU use CPU denoiser.
source/blender/compositor/operations/COM_DenoiseOperation.cc
@ -33,6 +33,10 @@ class DenoiseFilter {
oidn::DeviceRef device_;
oidn::FilterRef filter_;
bool initialized_ = false;
bool system_memory_supported_ = true;
Sergey Sharybin commented 2023-11-23 12:20:18 +01:00
Owner
Copy Link

can_use_host_memory_ seems to be a better name which indicates the intent more clearly.

`can_use_host_memory_ ` seems to be a better name which indicates the intent more clearly.
Attila Áfra commented 2023-11-24 17:25:23 +01:00
Contributor
Copy Link

I don't think so. "Host memory" in a GPU compute context often means host pinned memory which is always supported. That's not what this flag indicates. CUDA calls it "system allocated memory", which is where the naming of this flag comes from. Some exotic CUDA systems (with HMM) also support this, not just CPUs, but I haven't encountered such a system so far.

I don't think so. "Host memory" in a GPU compute context often means host *pinned* memory which is always supported. That's not what this flag indicates. CUDA calls it "system allocated memory", which is where the naming of this flag comes from. Some exotic CUDA systems (with HMM) also support this, not just CPUs, but I haven't encountered such a system so far.
Sergey Sharybin commented 2023-11-27 11:24:31 +01:00
Owner
Copy Link

This is a bit different from the conventions we follow in Cycles, but I see your point.

How about can_use_system_memory_ with a comment explaining that when true OIDN will use Blender-side allocated memory as-is, without copy to a temporary buffer? Or, if you prefer the current naming, add a comment nevertheless.

This is a bit different from the conventions we follow in Cycles, but I see your point. How about `can_use_system_memory_` with a comment explaining that when true OIDN will use Blender-side allocated memory as-is, without copy to a temporary buffer? Or, if you prefer the current naming, add a comment nevertheless.
Attila Áfra commented 2023-11-27 14:41:23 +01:00
Contributor
Copy Link

Sure! I agree that such a comment should be added.

Sure! I agree that such a comment should be added.
source/blender/compositor/operations/COM_DenoiseOperation.cc
@ -50,3 +54,3 @@
BLI_mutex_lock(&oidn_lock);
device_ = oidn::newDevice(oidn::DeviceType::CPU);
device_ = oidn::newDevice();
Sergey Sharybin commented 2023-11-23 12:16:30 +01:00
Owner
Copy Link

Does the OIDN 1 always use CPU device?

Does the OIDN 1 always use CPU device?
Stefan Werner commented 2023-11-23 13:57:40 +01:00
Author
Member
Copy Link

Yes. OIDN version 1 does not support any other devices. Explicitly requesting the CPU device here was only added when the library was upgraded to 2.0.0 to prevent automatic selection of other devices.

Yes. OIDN version 1 does not support any other devices. Explicitly requesting the CPU device here was only added when the library was upgraded to 2.0.0 to prevent automatic selection of other devices.
source/blender/compositor/operations/COM_DenoiseOperation.cc
@ -53,1 +58,4 @@
system_memory_supported_ = device_.get<bool>("systemMemorySupported");
if (device_.get<int>("type") == (int)oidn::DeviceType::CPU)
#endif
device_.set("setAffinity", false);
Sergey Sharybin commented 2023-11-23 12:19:11 +01:00
Owner
Copy Link

I am not really happy with code formatted like this. Such preprocessor in-between of control flow is really not readable.

The more readable code with less functional changes and less changes later on when we completely remove OIDN 1 code paths would be

#  if OIDN_VERSION_MAJOR >= 2
    device_ = oidn::newDevice();
    system_memory_supported_ = device_.get<bool>("systemMemorySupported");
    if (device_.get<int>("type") == (int)oidn::DeviceType::CPU) {
      device_.set("setAffinity", false);
    }
#  else
    device_ = oidn::newDevice(oidn::DeviceType::CPU);
#  endif
I am not really happy with code formatted like this. Such preprocessor in-between of control flow is really not readable. The more readable code with less functional changes and less changes later on when we completely remove OIDN 1 code paths would be ``` # if OIDN_VERSION_MAJOR >= 2 device_ = oidn::newDevice(); system_memory_supported_ = device_.get<bool>("systemMemorySupported"); if (device_.get<int>("type") == (int)oidn::DeviceType::CPU) { device_.set("setAffinity", false); } # else device_ = oidn::newDevice(oidn::DeviceType::CPU); # endif ```
source/blender/compositor/operations/COM_DenoiseOperation.cc
@ -68,2 +76,4 @@
BLI_assert(initialized_);
BLI_assert(!buffer->is_a_single_elem());
oidn::BufferRef oidn_buffer;
size_t buffer_len = buffer->get_elem_bytes_len() * buffer->get_width() * buffer->get_height();
Sergey Sharybin commented 2023-11-23 12:23:46 +01:00
Owner
Copy Link

size_t(buffer->get_elem_bytes_len()) * ...

Otherwise it is only up-casted to int, which will not be enough to composite really high-res images.

`size_t(buffer->get_elem_bytes_len()) * ...` Otherwise it is only up-casted to `int`, which will not be enough to composite really high-res images.
Omar Emara commented 2023-11-23 15:51:07 +01:00
Member
Copy Link

If there is a way to detect and connect to the real time compositor, sure, I'll be happy to look into that.

@Stefan_Werner See the execute method in the node_composite_denoise.cc file. There is no GPU texture interoperability at the moment as far as I can see, so we just denoise on host memory just like your existing implementation.

> If there is a way to detect and connect to the real time compositor, sure, I'll be happy to look into that. @Stefan_Werner See the `execute` method in the `node_composite_denoise.cc` file. There is no GPU texture interoperability at the moment as far as I can see, so we just denoise on host memory just like your existing implementation.
Sergey Sharybin commented 2024-02-05 11:09:43 +01:00
Owner
Copy Link

@Stefan_Werner Hey. What is the status here? Are you working on it, you waiting for some review to happen?

@Stefan_Werner Hey. What is the status here? Are you working on it, you waiting for some review to happen?
👀 1
Brecht Van Lommel modified the milestone from 4.1 to 4.2 LTS 2024-03-14 17:47:27 +01:00
Merge conflict checking is in progress. Try again in few moments.
View command line instructions.

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u compositor_gpu_denoising:Stefan_Werner-compositor_gpu_denoising
git checkout Stefan_Werner-compositor_gpu_denoising
Sign in to join this conversation.
Reviewers
No reviewers
Omar Emara
Brecht Van Lommel
Sergey Sharybin
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
Compositing
Assignees
Clear assignees
No Assignees
5 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#115242
No description provided.
Delete Branch "Stefan_Werner/blender:compositor_gpu_denoising"

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?