Like other runtime structs, it doesn't make sense to write this
data to files. And moving it out of DNA to an allocated C++ struct
means we can use other C++ features in it, like the new Mutex
type which I switched to in this commit.
Pull Request: https://projects.blender.org/blender/blender/pulls/138551
The interactive compositor crashes if the scene resolution is changed
after rendering. That's because the compositor queries its render size
from the scene directly, which might be different from the size that was
already rendered.
To fix this, the compositor should use the size of the existing render
result if available, while it should fallback to the scene resolution.
Pull Request: https://projects.blender.org/blender/blender/pulls/137674
This replaces the deprecated DrawData mechanism by the
usage of the update timestamp `last_update`.
The compositor keeps the `last_update` value of the cached ID
and compares it with the value on the ID at the time of evaluation.
Rel #134690
Pull Request: https://projects.blender.org/blender/blender/pulls/134878
The main issue of 'type-less' standard C allocations is that there is no check on
allocated type possible.
This is a serious source of annoyance (and crashes) when making some
low-level structs non-trivial, as tracking down all usages of these
structs in higher-level other structs and their allocation is... really
painful.
MEM_[cm]allocN<T> templates on the other hand do check that the
given type is trivial, at build time (static assert), which makes such issue...
trivial to catch.
NOTE: New code should strive to use MEM_new (i.e. allocation and
construction) as much as possible, even for trivial PoD types.
Pull Request: https://projects.blender.org/blender/blender/pulls/135813
The compositor previously overloaded the vector type to represent
multiple dimensions that are always stored in a 4D float vector. This
patch introduce a dedicated type for float4, leaving the vector type to
always represent a 3D vector, which will be done in a later commit.
This is not exposed to the user as a separate socket type with a
different color, it is only an internal type that uses the same vector
socket shape and color.
Since the vector socket represents both 4D and 3D vectors, code
generally assumes that such sockets represents 3D vectors, and the
developer is expected to set it to a 4D vector if needed in the node
operation constructor, or use the newly added skip_type_conversion flag
for nodes that do not care about types, like the File Output node.
Though this should be redundant once we add a dimension property for
vector sockets.
Pull Request: https://projects.blender.org/blender/blender/pulls/134486
This patch removes the compositor texture pool implementation which
relies on the DRW texture pool, and replaces it with the new texture
pool implementation from the GPU module.
Since the GPU module texture pool does not rely on the global DST, we
can use it for both the viewport compositor engine and the GPU
compositor, so the virtual texture pool implementation is removed and
the GPU texture pool is used directly.
The viewport compositor engine does not need to reset the pool because
that is done by the draw manager. But the GPU compositor needs to reset
the pool every evaluation. The pool is deleted directly after rendering
using the render pipeline or through RE_FreeUnusedGPUResources for the
interactive compositor.
Pull Request: https://projects.blender.org/blender/blender/pulls/134437
This patch refactors the Result class in the compositor to use
GMutableSpan and std::variant to wrap the result's data. This reduces
the complexity of the code and slightly optimizes performance. This will
also make it easier to add new types and interface with other code like
multi-function procedures.
Pull Request: https://projects.blender.org/blender/blender/pulls/134112
This patch allows the compositor context to specify exactly which
outputs it needs, selecting from: Composite, Viewer, File Output, and
Previews. Previously, the compositor fully executed if any of those were
needed, without granular control on which outputs are needed exactly.
For the viewport compositor engine, it requests Composite and Viewer,
with no Previews or File Outputs.
For the render pipeline, it requests Composite and File Output, with
node Viewer or Previews.
For the interactive compositor, it requests Viewer if the backdrop is
visible or an image editor with the viewer image is visible, it requests
Compositor if an image editor with the render result is visible, it
requests Previews if a node editor has previews overlay enabled. File
outputs are never requested.
Pull Request: https://projects.blender.org/blender/blender/pulls/133960
There is a special case in the compositor code where viewer nodes are
treated as composite nodes. This patch renames relevant methods and
updates comments to clarify this use case.
Pull Request: https://projects.blender.org/blender/blender/pulls/133811
This commit exposes the "Quality" option of the Open Image Denoiser
to the user for the denoise node in the compositor.
There are a few quality modes:
- High - Highest quality, but takes the longest to process.
- Balanced - Slightly lower quality, but usually halves
the processing time compared to High.
- Fast - Further reduce the quality, for a small increase in
speed over Balanced.
Along with that there is a `Follow Scene` option which will use the
quality set in the scene settings.
This allows users that have multiple denoise nodes
(E.g. For multi-pass denoising), to quickly switch all nodes between
different quality modes.
Performance (denoising time):
High: 13 seconds
Balanced: 6 seconds
Fast: 5 seconds
Test setup:
CPU: AMD Ryzen 9 5950X
Denoising a 3840x2160 render
---
Follow ups:
Ideally the "Denoise Nodes" UI panel in the render properties panel
would be hidden if the compositor setup does not contain any
denoise nodes.
However implementing this efficiently can be difficult and so it was
decided this task was outside the scope of this commit.
Pull Request: https://projects.blender.org/blender/blender/pulls/130252
NOTE: This also required some changes to Cycles code itself, who is now
directly including `BKE_image.hh` instead of declaring a few prototypes
of these functions in its `blender/utils.h` header (due to C++ functions
names mangling, this was not working anymore).
Pull Request: https://projects.blender.org/blender/blender/pulls/130174
Meta-data are missing on Cryptomatte layers in the GPU compositor, so
they do not get saved using the File Output node. This is due to a use
after free error where a temporary string is used in the meta-data
population logic. This is fixed by assigning the string to a temporary
variable instead.
Thanks to Jorn Visser for finding the cause of the issue.
Blender crashes when changing the compositor execution device. That's
because cached resources that were originally computed for CPU are now
being used for GPU and vice versa, which can be unexpected in code that
uses them.
To fix this, we free and recreate the entire compositor context when the
execution device or precision change, because it is much easier and
safer to recreate everything as opposed to trying to update the
necessary resources.
This patch adds support for passes in the new CPU compositor. This
involves rewriting the get_input_texture method into a get_pass methods
that returns a result as opposed to a texture. The result wraps the
cached GPU texture or image buffer depending on the execution device.
The Render Layers node was implemented for CPU execution and a new
utility constructor for the result class was added to determine type and
precision based on GPU texture format. The fallback depth pass that was
retrieved from the viewport frame buffer was removed, as it was a hack
that can no longer be supported due to the use of stencil format.
Pull Request: https://projects.blender.org/blender/blender/pulls/129154
Blender crashes when using the GPU compositor sometimes. This is because
compositor render data was accessed before it was updated in the
realtime compositor when detecting compositing device. So fix that by
first updating compositor data before calling any context methods.
The Legacy Cryptomatte node doesn't work in GPU execution mode if
Precision is set to Auto. That's because the colors picked from the Pick
layer might be in half precision and thus will not match the colors in
the Cryptomatte layers. This is due to the compositor using the
context's precision for Viewer outputs as opposed to the precision of
the image that actually needs to be viewed in the Viewer node.
To fix this, we set the Viewer node precision to be the precision of its
input, that way, the Cryptomatte pick layer will be output in full
precision as intended.
Pull Request: https://projects.blender.org/blender/blender/pulls/128495
This patch supports the viewer node in the new CPU compositor. To do
that, the context viewer output mechanism was refactored to allow CPU
storage by utilizing the result class as opposed to a GPU texture.
This patch introduces a new experimental option for the new CPU
compositor under development. This is to make development easier such
that it happens directly in main, but the compositor is not expected to
work and will probably crash.
Pull Request: https://projects.blender.org/blender/blender/pulls/125960
The viewport compositor slows down complex scenes even if it has very
simple setups. That's because it internally computes previews which
involves a fair bit of CPU computation, however, those previews are
actually never written to the original tree, so previewers weren't
really visible so it is effectively redundantly computations.
To fix this, we double down on disabling previews for the viewport
compositor and avoid any redundant computations in that case.
The vector pass and potentially other vectors that store 4 values are
stored wrongly, in particular, the last channel is ignored. To fix this
we identify if a vector pass is 4D and store the information in the
result meta data, then use this information to either save a 3D or a 4D
pass in the File Output node.
This is a partial fix for the GPU compositor only. The complete fix for
the CPU compositor will be submitted separately as it is not
straightforward and will likely require a refactor.
Pull Request: https://projects.blender.org/blender/blender/pulls/124522
This patch adds support for meta data in the GPU compositor much like
the mechanism that already exist in the CPU compositor. Only Cryptomatte
meta data is handled at the moment because that is the only meta data
that the compositor supports.
The is_data member of the result was moved to the meta data structure for
consistency with the CPU compositor.
Fixes#124222.
Pull Request: https://projects.blender.org/blender/blender/pulls/124460
Blender crashes when rendering a scene strip that references a scene
with a GPU compositor active. This is because when rendering a scene
strip, a new render with a nullptr system GPU context is created for the
scene it references, which is then used for compositing.
Ideally, the strip scene would have its own context, but we can't ensure
its context because we are not in the main thread. The alternative is to
then identify scenes that will be rendered before hand and set their
renders before starting the job, which doesn't seem like a great
solution. So for now, we just use the DST context in those cases.
Pull Request: https://projects.blender.org/blender/blender/pulls/123057
The system context is expected to be bound prior to the Blender
can be properly initialized. Otherwise GHOST will be doing OpenGL
calls without system context bound.
This follows code from DRW_render_context_enable().
Pull Request: https://projects.blender.org/blender/blender/pulls/122708
This patch refactors the GPU context handing in the GPU compositor.
First, GPU context handling in the GPU compositor constructor was
removed, that's because the constructor does nothing GPU related.
Second, the destructor and the execute methods were unified to use the
global DST context for main thread execution and a dedicated system GPU
context for threaded execution. The former is the case for blocking
rendering as well as background mode, so the blocking due to the global
DST is not an issue, but it makes the code more robust to implementation
errors like #122070.
Pull Request: https://projects.blender.org/blender/blender/pulls/122389
This patch adds support for timing GPU compositor executions. This was
previously not possible since there was no mechanism to measure GPU
calls, which is still the case. However, since 2cf8b5c4e1, we now flush
GPU calls immediately for interactive editing, so we can now measure the
GPU evaluation on the host, which is not a very accurate method, but it
is better than having no timing information. Therefore, timing is only
implemented for interactive editing.
This is different from the CPU implementation in that it measures the
total evaluation time, including any preprocessing of the inputs like
implicit type conversion as well as things like previews.
The profiling implementation was moved to the realtime compositor since
the compositor module is optional.
Pull Request: https://projects.blender.org/blender/blender/pulls/122230
The issue originates to the change in default view transform from Filmic
to AgX, which does slightly different clipping, and clips color to black
if there is any negative values.
This change implements an idea of skipping view transform for viewer
node when it is connected to the Pick output of the cryptomatte node.
It actually goes a bit deeper than this and any operation can tag its
result as a non-color data, and the viewer node will respect that.
It is achieved by passing some extra meta-data along the evaluation
pipeline. For the CPU compositor it is done via MetaData, and for the
GPU compositor it is done as part of Result.
Connecting any other node in-between of viewer and Cryptomatte's Pick
will treat the result as color values, and apply color management.
Connecting Pick to the Composite output will also consider it as color,
since there is no concept of non-color-managed render result.
An alternative approaches were tested, including:
- Doing negative value clamping at the viewer node.
It does not work for legacy cryptomatte node, as it needs to have
access to original non-modified Pick result.
- Change the order of components, and store ID in another channel.
Using one of other of Green or Blue channels might work for some view
transforms, but it does not work for AgX.
Using Alpha channel seemingly works better, but it is has different
issues caused by the fact that display transform de-associates alpha,
leading to over-exposed regions which are hard to see in the file from
the report. And might lead to the similar issues as the initial report
with other objects or view transforms.
- Use positive values in the Pick channel.
It does make things visible, but they are all white due to the nature
of how AgX works, making it not so useful as a result.
Pull Request: https://projects.blender.org/blender/blender/pulls/122177
Blender crashes when calling the python render operator when GPU
compositor execution is enabled. This is due to a missing system GPU
context, which is not initialized for blocking rendering. So this patch
ensures the system GPU context before compositing. Additionally, it
removes the assert that ensures a non main thread execution, since the
assumption apparently does not really hold.
Pull Request: https://projects.blender.org/blender/blender/pulls/122176
This allows to expose these settings in the Performance panel in the
render buttons. Also moves compositor-specific options away from the
generic node tree structure.
For the backwards-compatibility the options are still present in the
DNA for the bNodeTree. This is to minimize the impact on the Studio
which has used the GPU compositor for a while now. They can be
removed in a future release.
There is no functional changes expected.
Pull Request: https://projects.blender.org/blender/blender/pulls/121583
The compositor execute functions have a `rendering` argument to specify
if the compositor is executing as part of the render pipeline. But the
render context argument is null if we are not rendering, so the
`rendering` arguement is redundant and can be removed.
Additionally, we no longer use use_file_output as a hack to detect
rendering.
Pull Request: https://projects.blender.org/blender/blender/pulls/120659
This patch improves the interactivity of the GPU compositor for
interactive node tree edits by waiting on GPU work to finish to support
more granular canceling.
This does have a performance penalty, but it does not affect final
rendering and it seems worth it because even though it is slower it will
feel faster for users.
Pull Request: https://projects.blender.org/blender/blender/pulls/120656
This patch refactors the backdrop offset to be stored as a float instead
of an int and to be stored in the image runtime structure instead of the
image itself.
Pull Request: https://projects.blender.org/blender/blender/pulls/119877
This required to apply a small fix in the Metal texture uploader.
Without synchronization pixels of a wrong pass can be uploaded to
a wrong texture. This is because this code path is heavily reusing
temporary allocations, and at some point the allocation is not
considered as still in use, unless the command buffer used by the
texture uploader is submitted.
Ref #118919
Pull Request: https://projects.blender.org/blender/blender/pulls/118920
The non-locking context activation expects the render engine to
have its own GPU contexts, which is not necessarily the case when
compositing happens from the command line.
Ensure that the contexts exist prior to attempt of activating them.
Pull Request: https://projects.blender.org/blender/blender/pulls/118496
