Was caused by float imprecision and storage precision
conversion. Use same fix as combined pass but
more selective as the alpha channel can contain
other data for AOVs.
Fix#118494
Previously in
```
edge_y = edge_step(current_vertex_y, edge_y, ¤t_vertex_y);
while (!BM_elem_flag_test(current_vertex_y, BM_ELEM_TAG))
```
With hidden faces this code enters in an infinite loop by returning the same null
edge in edge_step() and the same current_vertex_y causing Blender to crash.
The current fix solves the infinite loop and provides a solution when faces are hidden.
This causes the hidden faces to be merged into one for unsubdivided per original dev's implementation.
At minimum the crash is identified and a solution proposed.
Attached is a video of the fix
Pull Request: https://projects.blender.org/blender/blender/pulls/121086
Added `BLENDER_EEVEE_NEXT` engine ID to polling code called during
"drag search". Main ADD menu routines already had this ID.
Fixes#121026 missing `Shader to RGB` node.
`Specular BSDF` node had similar problem, fixed as well.
Pull Request: https://projects.blender.org/blender/blender/pulls/121231
An earlier commit accidentally removed too much code. This is now
restored, so that when keyframe insertion fails, at least on stdout it's
logged which F-Curve caused this failure.
8dbd167e80 did improve the error
reporting, but even that does not explicitly mention which F-Curves
caused those errors (for good reason, as cluttering the UI would be very
simple then).
Pull Request: https://projects.blender.org/blender/blender/pulls/121527
This affects only transparent materials. Ideally, the shader
compilation should trigger a tag update on the object but
that's bad level call.
Instead, we just discard all shadows after finishing recompiling
a transparent material. This is overly conservative but it fixes
the issue.
Fixes#119624
Displaying an operator in the context menu that would access the asset
via context would fail.
3 issues here in fact:
- 798219225d removed code to attach the asset to the button's context.
Brought this back.
- If the `AssetShelf.get_active_asset()` method (in Python) isn't set,
`AssetViewItem::should_be_active()` would always return false, making
it impossible to lookup the active item. Instead it should return an
empty value so the view's internal active state is used instead.
Mistake in a700c90ec3.
- These kind of context queries can't rely on cursor coordinates, it's
not well defined what state they are in. In fact `wmWindow.eventstate`
is unset in debug builds to help enforce this. Also an issue from
a700c90ec3. With the above two corrections this change can be
undone so the query doesn't depend on cursor coordinates anymore.
This is caused by #121137 as the bug disappear if
`can_use_default_cb` returns false.
The issue is caused by the material binning system.
`ShaderKey` uses the `GPUMaterial` to differentiate between
material. But with #121137, different materials can reuse the
same `GPUMaterial`.
The fix is to not reuse the shader for the prepass (even if opaque)
for transmission enabled shaders.
Fix#121343
This parameter was introduced to mitigate shadow tracing
artifacts. These are no longer present since #121317 and
leaving this parameter breaks PBR rules which opens other
issues with shadow disconnected from shading.
Due to recent changes geometry without any vertices could lead to
a division by zero when copying motion vectors.
This PR fixes this by skipping the copy step when there are no
vertices to copy.
Pull Request: https://projects.blender.org/blender/blender/pulls/121516
Caused by 598af45459
Above commit added respect for the NLA strip offset (rightfully so, by
using `ANIM_nla_mapping_apply_fcurve`).
However, when no closest frame was found, the code moved on to the next
FCurve without restoring to the original FCurve --
`ANIM_nla_mapping_apply_fcurve` needs to be called in pairs.
To resolve, now always restore to the original FCurve.
Pull Request: https://projects.blender.org/blender/blender/pulls/121483
This changes the following keyframing functions to return
`SingleKeyingResult`, which is in turn used for better failure
reporting in the higher-level functions that call them:
- `KeyframeStrip::keyframe_insert()`
- `insert_vert_fcurve()`
As a side effect, this also means that
`rna_KeyframeAnimationStrip_key_insert()` can no longer return an
`FCurve *`, and now instead returns a bool indicating success.
This is part of an ongoing progressive refactor to improve error
messages and failure handling in the keyingframing code.
Pull Request: https://projects.blender.org/blender/blender/pulls/121517
The goal of this PR is to remove the per face, NDC space
tracing for shadow maps.
This requires custom occluder extrapolation but in return
removes quite a lot of complexity in other areas, namely:
- No more per face transform before the tracing
- No more per face jittering (fix#119565)
- No more frustum padding (increased maximum precision)
- Better use of 32bit precision shadow map
- Fix#121343
This improve softness at relatively low step count (default 6
is better) and reduces light leaking at very low sample
count (sharper). It makes it more intuitive now that
higher sample count is smoother.
Pull Request: https://projects.blender.org/blender/blender/pulls/121317
For metric units the depth was being rounded to the nearest meter
(in the range of 10-100 meters), increase the precision from 2 to 4
so it's possible to see centimeter precision.
BKE_mesh_validate is called by mesh.validate() Python API, as well as optionally
when doing file imports. This PR speeds it up a bit:
- Faster face data sorting by using parallel sort instead of qsort,
- One allocation for all face vertex indices, instead of a separate allocation
for each face,
- (more like a fix) Validation no longer adds a MDeformVert layer when there was
none in the input mesh.
- Small cleanups (more const inputs, etc.)
On my Windows/VS2022/Ryzen5950X machine, import time in seconds (validation on
before this PR -> validation on with this PR, validation off):
- USD (Intel Moore Lane): 9.1 -> 6.7, 4.8.
- OBJ (Blender 3.0 splash): 22.7 -> 18.6, 16.5.
Pull Request: https://projects.blender.org/blender/blender/pulls/121413
This adds the same overlay that already exists for the bake node to the simulation output node.
Without this, it's quite hard to see if the simulation zone is baked or not currently.
Pull Request: https://projects.blender.org/blender/blender/pulls/121497
Bezier handles are recalculated in many places in the
animation code. Threading that code can give a performance
boost all over Blender.
This patch only threads a part of the handle calculation code.
`BKE_nurb_handle_smooth_fcurve` can still be run in parallel,
but that's more complicated, so not done in this PR.
Overall this patch mostly benefits code paths that are not already threaded.
------
Performance delta
| Action | Before | After |
| - | - | - |
| `recalcData_graphedit` moving a single key | 1.06 ms | 1.0 ms |
| `recalcData_graphedit` moving 300 keys of a single FCurve | 1.6 ms | 1.4 ms |
| `recalcData_graphedit` moving 300 keys of multiple FCurves | 60 ms | 55 ms |
| `ANIM_animdata_update` when using the Breakdown operator in the GE | 90 ms | 73 ms |
Test file used
https://download.blender.org/ftp/sybren/animation-rigging/heavy_mocap_test.blend
Pull Request: https://projects.blender.org/blender/blender/pulls/119388
Cleanup to avoid unnecessary copies of VArray. This
requires ref-qualifier overloads of dereference operator
of attribute reader and some move operators and constructor
overloads in the code.
Pull Request: https://projects.blender.org/blender/blender/pulls/118437
- Remove meaningless const for by-value arguments in declarations
- Use std::move for strings (otherwise StringRef should be used)
- Move struct to C++ namespace
- Avoid unnecessary namespace specification
- Use static constexpr instead of macros
- Add const for other arguments in function definitions
- Avoid copying struct in for loop
- Remove redundnat "private" in class
This PR fixes the ternary operation to avoid indexing into a cleared
`BitGroupVector`when trying to create an `IndexMask` for a mesh
with no hidden elements.
Pull Request: https://projects.blender.org/blender/blender/pulls/121461
Remove options that are duplicate and change description of options, so
they describe bit better, how timecodes are actually used.
Timecodes in Blender have pretty much nothing in common with more widely
known term "timecode". This confused users (and developers).
There were 5 options of which 3 were exactly same. This commit leaves
user with 3 options:
- Use timecodes for normal seeking/playback - Record Run
- Ensure, that no frame is duplicated or skipped - "Record Run No Gaps"
- Do not use timecodes - "None"
More verbose description was added to the definition in code.
Naming of these timecode types was kept, even if it is incorrect to not
break scripts and habits.
Pull Request: https://projects.blender.org/blender/blender/pulls/121001
Uses `FileHandlers` to handle file drag-n-drop in the space Sequencer, drop-boxes
still remain since they handle Images ID drag-n-drop.
This also allows to open/drag-n-drop multiple files at once.
When drag-and-drop is performed with a single image/movie/sound
file and only the internal `FileHandler` can be used, is used
the internal drop-box api for drag and drop, since it has a nice
drag-and-drop preview.
Pull Request: https://projects.blender.org/blender/blender/pulls/119927
Mistake in a6ebfb05ad, enabled catalog list shouldn't be free'd, it just
needs to be overridden by the duplicated list correctly. Earlier a clear
(but not a free) was necessary, but now it's overridden anyway so no
clearing is needed.
Windows/Intel UHD 600 iGPUs crash when compiling probe remapping shader.
The cause is that there is a balanced barrier inside a forloop. When removing
the barrier the compilation works. This is a driver bug and most likely not
being solved as the driver maintenance mode will only be updated for critical
fixes.
This PR works around the issue making the loop condition less complex.
Pull Request: https://projects.blender.org/blender/blender/pulls/121105
Detected when testing mr_elephant on an Intel HD520. When copying
the velocity buffer using the copy shader, the number of scheduled
workgroups could be larger than supported by the device.
This PR fixes this by copying multiple vertices per thread when
the work size cannot cover all the pixels.
Pull Request: https://projects.blender.org/blender/blender/pulls/120915
