Make Subsampling 3x3 filter twice faster (on 4K UHD resolution,
Windows/VS2022/Ryzen5950X: 52.7ms -> 28.3ms), by reformulating how it works:
Conceptually Subsampling filter is a box filter: it sums up N source image
pixels, computes their average and outputs the result. Critical thing is,
that should be done in premultiplied space so that colors from fully or
mostly transparent regions do not "override" opaque colors.
Previously, when operating on byte images, the code achieved this by always
working on byte values, doing "progressively smaller" lerp into byte color
result, taking care of premultiplication and again storing the "straight"
alpha for each sample being processed. This meant that for each sample, there
are 3 divisions involved! This also led to some precision loss, since for all
9 samples all the intermediate results would only be stored at byte precision.
Reformulate that by simply accumulating the premultiplied color as a float.
This gets rid of all divisions, except the last step when said float needs to
be written back into a byte color.
The unit test results have a tiny difference, since now it is arguably better
(as per above, previously it was having some precision loss).
Pull Request: https://projects.blender.org/blender/blender/pulls/117125
This allows code outside of the render pipeline to make proper
decisions about how the imbuf of render passes are to be handled.
For example, IMB_create_gpu_texture() will now properly select
single channel grayscale texture format for depth pass coming from
multilayer EXR, additionally solving assert in the GPU compositor
code which verifies expected and actual imbuf texture format.
Pull Request: https://projects.blender.org/blender/blender/pulls/117184
Set the number of planes based on the number of pass channels.
If the pass contains 2 passes or more than 4 passes set the number
of planes to the previously used value of 32.
This is needed because quite some areas check for the number of
planes for various optimizations. For example, this is one of the
factors which make IMB_create_gpu_texture() to choose the texture
format. If the number of planes for the depth pass is set to the
previously used this function will never consider using single
channel GPU texture.
Unfortunately, this change is not enough to make the GPU texture
to use single channel format as the color space of the image
buffer is also checked, and that is nullptr which means scene linear.
Using screen changing functions with screens used for full screen areas
isn't supported & caused corrupt screen data.
Add checks that the current and overriding screen support switching,
raising an error when they don't. Also add a check when restoring the
context not to change any full screen areas.
CurvesGeometry has no ".selection" attribute when all control points
are selected. The earlier code assumed that the attribute always exists.
Also Python tests are added for the "extrude" operator.
Pull Request: https://projects.blender.org/blender/blender/pulls/117095
There were some abstractions inside `IMB_transform` implementation
that were kinda getting in the way. I want to speed up Subsampled3x3
filtering by reformulating the math being done, but that needs to
sample byte images while storing intermediate result as a float -- but
the whole `Pixel`, `PixelPointer` et al. machinery was not quite prepared
to deal with that. Feels like those helper classes do not achieve much,
just make the code longer.
So remove them (`Pixel`, `PixelPointer`, `ChannelConverter` etc.) and
use simple functions. Now the code is more straightforward (IMHO) and
270 lines shorter (40% of the whole file!), and allows me to do the
follow-up optimization easier.
No functional changes.
Pull Request: https://projects.blender.org/blender/blender/pulls/117182
Using layout panels in this context is not yet supported. Therefor, the fix
is to raise an exception instead of crashing.
The problem here is that layout panels need an a parent panel which
has a persistent state. The popover currently does not only has a very
short-lived panel currently that is freed directly after drawing.
Also see `ui_paneltype_draw_impl`.
It's likely that support for this can be added in the future but it's currently
unclear how many changes this would require.
Opacity modifier implementation based on GP2.
Functionality is largely unchanged.
_Color Mode_ is either `Stroke` or `Fill` for modifying color opacity or
`Hardness`.
_Uniform Opacity_ does two things at once (!):
- Sets the same opacity value for every point in a stroke.
- Sets opacity as an absolute value rather than a factor.
_Weight as Factor_ (button to the right of Opacity Factor): Use the
vertex group as opacity __factor__ rather than an overall __influence__.
This is confusing and hard to convey, but copies behavior from GP2.
The _Influence_ panel contains the same filter settings as the GP2
modifier, with some small changes:
- _Layer_ selects only strokes in the respective layer (with an _Invert_
option)
- _Material_ selects only points with the respective material (with an
_Invert_ option)
- _Layer Pass_ and _Material Pass_ select only strokes/points which are
rendered in the respective pass.
_Note 1: Layers don't have UI for setting a pass yet, this will be a
generic layer attribute. This can be set through the API for testing._
_Note 2: In GP2 a pass value of zero was used to disable pass filters.
Since zero is a valid pass ID an explicit flag has been added for the
purpose of turning pass filters on and off._
- _Vertex Group_: This can be used as an additional influence filter on
points. If _Weight as Factor_ is enable the vertex group instead
replaces the opacity factor. In _Fill_ mode the vertex group weight of
the stroke's first point is used as influence for the entire stroke.
- _Custom Curve_ is another possible influence factor per point. The
curve input value is the relative position of a point along its
stroke.
The Influence settings have been moved into a separate DNA struct, which
should help with reusability in various modifiers. Various utility
functions can be found int `MOD_grease_pencil_util.hh` for handling
influence settings and generating `IndexMasks` for modernized C++ code.
Pull Request: https://projects.blender.org/blender/blender/pulls/116946
Loading pre-4.0.20 node groups with sockets using subtypes causes
broken socket identifiers. This is because the node tree interface now
expects sockets to use the base identifiers ("NodeSocketFloat" instead
of "NodeSocketFloatFactor" etc.).
To correct this the conversion code now replaces socket idnames that
include a subtype suffix with their base names. This fix is also applied
to files between versions 4.0.20 and 4.1.10, where the socket types
may have been converted incorrectly.
Pull Request: https://projects.blender.org/blender/blender/pulls/117133
The unregistering of already registered classes is expected behavior,
however this was done silently and can be unsafe as two Addons may have
the same names for operators and in case of name collision the addon
registered first will break silently / start behave unexpectedly.
So reporting the unregister step seems reasonable.
During Addon development, this might seem noisy when classes are
frequently updated, however catching the problematic cases instead of
being silent might be beneficial.
Part of #116370
Pull Request: https://projects.blender.org/blender/blender/pulls/116374
In 13fac109 the node panels got support for individual option button
callbacks, but these were not included in the node editor side bar.
Only the older top-level buttons are drawn there.
The panel structure is currently not accessible in python since it is
part of the `NodeDeclaration` system. To draw node input sockets and
buttons in the correct panel order as they appear on the node, a new
template function `uiTemplateNodeInputs` has been added. This iterates
over declared panels and their contents in the appropriate order and
draws the buttons before sockets in the same panel.
Pull Request: https://projects.blender.org/blender/blender/pulls/116936
Double precision pixel coordinate interpolation was added in 3a65d2f591 to
fix an issue of "wobbly" resulting image at very high scale factors. But
the root cause of that was the fact that the scanline loop was repeatedly
adding the floating point step for each pixel, instead of doing multiplication
by pixel index (repeated floating point additions can "drift" due to
imprecision, whereas multiplications are much more accurate).
Change all that math back to use single precision floats. I checked the
original issue the commit was fixing, it is still fine. Also added a gtest
to cover this situation: `imbuf_transform, nearest_very_large_scale`
This makes `IMB_transform` a tiny bit faster, on Windows/VS2022/Ryzen5950X
scaling an image at 4K resolution:
- Bilinear: 5.92 -> 5.83 ms
- Subsampled3x3: 53.6 -> 52.7 ms
Pull Request: https://projects.blender.org/blender/blender/pulls/117160
When using EEVEE on high resolution monitors the light buffers might not
get initialized as there are range checks that pass in the first try.
- number of tiles needed is larger than the max_tile_count_threshold
- total_word_count is smaller than max_word_count_threshold as it is
never set (still initialized to zero.
Solution is to not exit on the first try. In a later stage we might want
to use something that doesn't require any looping.
Fixes: #117128
Pull Request: https://projects.blender.org/blender/blender/pulls/117164
Add `ATTR_WARN_UNUSED_RESULT` to the
`BKE_animsys_eval_context_construct` and
`BKE_animsys_eval_context_construct_at` functions. They are pure
functions, and without using their returned value they are pointless.
No functional changes.
If a scene sequencer strip references a scene with no composite output,
it will crash due to null image buffer output. This patch fixes that by
allocating an empty buffer in those cases.
Pull Request: https://projects.blender.org/blender/blender/pulls/117135
Now that the select engine data uses non-trivial objects in its global
data, storing it at the global scope causes trouble due to arbitrary
construction and destruction order. Instead use the construct on first use
idiom to make the order clear. Though this struct probably shouldn't be
static at all, it does simplify memory management as well, it's nice to
remove the need to manually clear the arrays.
Pull Request: https://projects.blender.org/blender/blender/pulls/117147
Before we always did a float conversion of the colors even if we
wouldn't use the result.
Because the float conversion does alpha premultiplication, we can
use some shortcuts to skip a lot of computations
In a edit with a lot of fully transparent pixels, I observed a reduction
in total render time from 1m 31s to 1m 17s.
Pull Request: https://projects.blender.org/blender/blender/pulls/117134
Sharing of the normals cache between copied meshes was missing from
89e3ba4e25, which under-represented the benefits of the
change. In a simple file where geometry nodes causes a re-evaluation
without changing the normals, this increased FPS from 2.6 to 14.
Recent and ongoing efforts have changed many properties to be stored
contiguously in memory, e.g. mesh attributes. This patch updates
rna_raw_access to make use of this and copy the entire contiguous block
of memory when the property is stored contiguously.
This is faster and scales much better with larger arrays.
Pull Request: https://projects.blender.org/blender/blender/pulls/116015
Part of overall "improve filtering situation" (#116980) task:
* Add Bicubic filtering option to strip Transform "Filter" setting.
Previously this option only existed in Transform Effect "Interpolation"
setting.
- With this addition, it feels like the transform effect could
possibly be marked as legacy/deprecated, since the regular Transform
that is on all strips can do everything that Transform Effect did?
* Speed up bicubic filtering (used now in VSE, but also in CPU Compositor,
image paint, etc.) by slightly simplifying the code and using some SIMD.
Upscaling 96x54 image to 3840x2160 resolution, using Bicubic filtering:
- Windows (VS2022, Ryzen 5950X): 35.5ms -> 15.1ms
- Mac (clang 15, M1 Max): 29.6ms -> 24.4ms
* Add gtest coverage for bicubic functionality.
Pull Request: https://projects.blender.org/blender/blender/pulls/117100
Fix an issue where, after deleting the active bone collection, the
incorrect bone collection became the active one.
The solution was to store the active bone collection index before
deletion, instead of after.
After the replacement of auto smooth with a modifier, sharp edges are
always used, so the "shade smooth", "shade flat", and "smooth by angle"
operators cleared the attribute. However, often users spend significant
time manually tagging edges sharp, and the operators make it too easy to
lose that data.
To keep the old behavior by default, add an option called "Keep Sharp
Edges". Though this can make the operators "ineffective" at their goal
of changing the way the meshes look, or result in redundant data stored
on the mesh, it's a much safer default, especially as users get used to
the new workflow.
Pull Request: https://projects.blender.org/blender/blender/pulls/117069
