Perform delayed freeing of the geometry BVHs similar to OptiX.
Previously BLAS memory was allocated in the device class as part of
device_update, but released in the BVHHIPRT destructor which gets called
when deleting geometry outside of device_update.
To avoid the GPU accessing unmapped memory, do a delayed free of this
memory in the device class as part of either device_update or device
destruction. This ensures it is in sync with other device memory changes.
Fix#148276Fix#139013Fix#138043Fix#140763
Pull Request: https://projects.blender.org/blender/blender/pulls/147247
- "Acion Slot" -> "Action": typo.
- "Specifies the input node used the created zone" -> "by the created
zone": typo.
- "No reference available {!r}, Update..." -> "update": lower case.
- "uninitialized file-list" -> "Uninitialized": sentence case.
Some issues reported by Alexandr Fatih.
Pull Request: https://projects.blender.org/blender/blender/pulls/148265
Since "Add Rest Position" is not a shape key property, move it out of
`draw_shape_key_properties`. That way property will be drawn even when
list is empty.
See PR description for photos
Pull Request: https://projects.blender.org/blender/blender/pulls/147685
Caused by 113d91aba8
Code first gathers a list of all vertexgroups on all objects.
Then it iterates the objects to join, and for each of them, puts their
vertexgroups in an index map.
However (by a typo?), it was doing it for the `active mesh` (instead of
the one currently visited) over and over again, resulting in weights
from other objects written to `def_nr` from groups of the `active mesh`.
Now corrected.
Pull Request: https://projects.blender.org/blender/blender/pulls/148259
This is an alternative to ba0f73d076. I'm not exactly sure why that on
didn't work yet. Seems like there is some hidden state somewhere, not sure.
Now this fix is more similar to what is done in `curves_blend_write`.
Pull Request: https://projects.blender.org/blender/blender/pulls/148267
Caused by some tricky use after-free / stack corruption.
This affect both Cycles and EEVEE.
The `sms` local variable is getting dereferenced by `get_inscattering`
inside the threaded for loop but is passed by copy to the lambda expression.
This makes its lifetime ill-defined in a multithreaded context.
I am not fully sure about the rules at play here so maybe my understanding
is wrong. But removing the call to `get_inscattering` avoids the crash.
Note that `SkyMultipleScattering` is also very big (it contains the whole LUT).
So copying it might have caused stack overflow. But that should trigger a
system interupt.
Passing everything by references fixes the issue.
This seems to be safe since all as the other local variables are `const` anyway.
Also the loop doesn't seem to modify the one that aren't.
Pull Request: https://projects.blender.org/blender/blender/pulls/148260
While more of a theoritical issue currently, this change is a
pre-requisite to support properly compo nodes duplication in 'linked
copy' case (see !148210).
Also add a basic unittest for this case.
Pull Request: https://projects.blender.org/blender/blender/pulls/148222
Since matrices are usually indexed **row first**, displaying matrices in
the UI seems flipped.
They are stored as flat arrays.
When `matrix[1][0]` is accessed via python, the logic from
`pyrna_py_from_array_index` will actually look up index 2 in the flat
array (rightfully so).
But UI code was flipped in that regard.
The second value in the flat array **should** show at row 1 column 0,
but it currently doesnt.
Now (hopefully) handled properly (not only in `ui_item_array` but also
made code in `ui_item_rna_size` detect proper row count for a `height`)
Pull Request: https://projects.blender.org/blender/blender/pulls/148197
Reported for the geometry nodes gizmos, but in general `cage2d`,
`cage3d` and `arrow` were affected (so for example Forcefield arrows, UV
editing transform tool, Lamp angle gizmo, ...)
`gizmo_cage3d_modal`, `gizmo_cage2d_modal` and `gizmo_arrow_modal` add a
`WM_event_add_mousemove` which ... lets the modal run even if the mouse
does **not** move...
This seems to date back to d8f931c9b7 in the arrow gizmo, others
probably just copied from there (not the DialGizmo though), did not do
further git archeology.
Propose to just remove `WM_event_add_mousemove` (even though there
_might_ still be reasons for this -- could not spot any though...)
Pull Request: https://projects.blender.org/blender/blender/pulls/146310
The issue is in the `ED_region_overlap_isect_` family of functions I
believe. Their purpose is to check if a coordinate is in the
"non-transparent" or opaque parts of the overlapping region.
These are getting event coordinates (in window space), converted to
region relative space (substracting region->winrct).
Comparison is in view space (through `UI_view2d_region_to_view_y`)
to account for scrolling and zooming.
The thing where it goes wrong is that we are actually comparing to
`region->v2d.tot` (this can be huge, most of it not visible), where I
think we should be comparing to `region->v2d.cur` (that is the part that
is actually visible...)
Swapping `region->v2d.tot` with `region->v2d.cur` is what this PR does.
Pull Request: https://projects.blender.org/blender/blender/pulls/146032
Oversight in 4bede1b555
`transformApply` / `transformEnd` already gives us a `NC_MASK |
NA_EDITED` notifier (via `viewRedrawForce`), so listen to that in
`image_main_region_listener`.
NOTE: alternatively, we could send an appropriate notifier in
`special_aftertrans_update__mask` (or remove the condition from the one
that is already there), with the difference that the gizmo would only
update after the transform is confirmed -- which may or may not be
desired
Pull Request: https://projects.blender.org/blender/blender/pulls/146292
The recently introduced size, strength, and jitter pressure curves
affect most paint code. This exposed further odd behavior inside
`paint_brush_update` where the size pressure curve was being evaluated
even if the brush's size did not vary with pressure.
To fix this issue, this commit clarifies a few comments and updates the
code flow such that cached input values and evaluated pressure values
are used more consistently.
Pull Request: https://projects.blender.org/blender/blender/pulls/148077
The Cycles cpu/device.cpp `device_cpu_capabilities()` function used to
fill out a string of supported CPU capabilities separated with spaces,
with some trailing-space cleaning logic at the end of the function.
However, if no check succeeded, and especially after commit 2bf6d0fd71
which left only one check and removed the need for removing trailing
spaces, the check would run against an empty string, resulting in an
unsigned 0 - 1 operation which would then cause an out of bound access
catched by ASan.
Fixed by removing the now superflous trailing space cleaning logic and
simplifying to a direct return.
Pull Request: https://projects.blender.org/blender/blender/pulls/148227
This was caused by 3dfec1ff73
which introduce the new behavior. This was to fix workflows
using a lot of semi-transparent objects which made nagivation
difficult.
This patch first roll back to the previous behavior: The
unselectable object will affect depth-aware operators.
This patch introduces a new visibility property to remove
the influence of objects in all depth picking operations
and selection operations. However the object is still
selectable through non-drawing selection operators
(e.g. select by material) and through the outliner.
This is to adress the aforementionned navigation issues.
Pull Request: https://projects.blender.org/blender/blender/pulls/146706
644fb2b679 fixed a long standing issue
that offscreen example showed the wrong colors. However the fix assumes
that input texture color space is always sRGB.
This adds a shader variation that draws textures that are stored in scene referred
linear color space (like all of our Image data-block).
Co-authored-by: Clément Foucault <foucault.clem@gmail.com>
Pull Request: https://projects.blender.org/blender/blender/pulls/147788
Adding panel toggles in nodegroups have somewhat of a UX antipattern. When
running the operator, it checks for conditions that indicate it should not run,
and if those are hit, it cancels execution and mentions the invalid condition in
the footer bar.
This is not ideal, the user should not have to call the operator to find out
whether it can be called.
Why it got implemented like this is likely a consequence of all interface items
being the same "New Item" operator. Poll functions cannot use operator
properties, so variants of the same operator cannot check for different
conditions for execution.
This is a problem for panel toggles, as they have more restrictions to when they
can be added that don't apply to other interface items.
This patch creates a separate operator for adding panel toggles. This allows the
condition checks to be implemented in the poll function, which enables greying
out the operator buttons and showing on tooltips what condition is invalid.
Pull Request: https://projects.blender.org/blender/blender/pulls/146379
The implementation of this was broken and not actually used for the EEVEE
tests, as there is currently no separate reference directory for AMD.
Move it to the render report module so it can be reused by different tests.
Pull Request: https://projects.blender.org/blender/blender/pulls/148148
Only use this feature when building for 1 or 2 CUDA architectures.
Otherwise CMake will build the binaries in parallel, and NVCC will then
also launch multiple threads for each binary.
We could add more manual control for this, but the main use case for
this is local builds and an automatic heuristic seems more likely to
help than an option that developers or users might not discover.
For minimal memory usage WITH_CYCLES_CUDA_BUILD_SERIAL still exists
to use only 1 thread for CUDA compilation.
Pull Request: https://projects.blender.org/blender/blender/pulls/147303
It's not very obvious to some people what the effect will be from
looking at the thumbnail alone. Also, we plan to make the thumbnails
smaller (48px vs 64px) by default to match other asset shelves size.
This PR updates the thumbnails by making the effects stronger. See PR
for example images.
Pull Request: https://projects.blender.org/blender/blender/pulls/148192
The issue was that the data-blocks of two different undo steps were detected to
be identical, even if the attributes changed. That's because even if the
implicitly-shared data was different, they were turned into the same pointer by
cadb3fe5c5 on write.
This patch makes it so that for undo steps, implicitly shared data does not use
the pointer stability feature (in a sense, implicit-sharing itself provides
pointer stability for undo steps already).
The main tricky aspect is that we need to know if a pointer is implicitly shared
in `writestruct_at_address_nr` and oftentimes that's called before the
corresponding shared data is actually written with `BLO_write_shared`. The
solution is to enforce that the blend-write code has to know what pointers are
implicitly-shared before they are written the first time. The simplest way to
ensure that is to call `BLO_write_shared` first. However, that's not always
possible, especially when the pointer is directly embedded in an ID. Therefore,
there is a new `BLO_write_shared_tag` function that can be used in such cases.
The undo performance for the file in #141262 is still fixed with this change.
Pull Request: https://projects.blender.org/blender/blender/pulls/148144
This is an alternate solution to !146889 to improve labels in the
camera UI, while being much less invasive. It doesn't take custom
labels into account, but it simply uses the parameter names with title
case.
Pull Request: https://projects.blender.org/blender/blender/pulls/148141
When saving asset catalogs to an already-existing file, only perform
that save when there were local changes to the catalogs. This prevents
unnecessary writes to the catalog definition file, as that can cause
conflicts when multiple Blender instances share the same asset library
(either directly or via Syncthing/Dropbox/etc.)
Pull Request: https://projects.blender.org/blender/blender/pulls/148205
Swapchains can be destroyed safely after a new image
is presented in a new swapchain or when all submission
and presentation fences are signaled, this last ones
where missing causing issues with some drivers when
recreating swapchains.
Pull Request: https://projects.blender.org/blender/blender/pulls/147226
This has ripple effects by making the "hide value" setting ineffective and
always hiding socket values as well as graying out the "hide value" node group
option.
Pull Request: https://projects.blender.org/blender/blender/pulls/148188
Follow on from PR #141891. The `MTLAccelerationStructureUsagePreferFastIntersection` flag didn't exist until Xcode 26.0, so we ensure that it is defined for forward-compatibility. The runtime `if (@available(macos 26.0, *))` checks still remain.
Pull Request: https://projects.blender.org/blender/blender/pulls/147561
The bone collection operator was not updated to handle the new flag
which is now on the `bPoseChannel` instead of the `Bone`.
For this to work, the operator now needs the `bArmature` as well as the `Object`
and they need to be in sync. Additional code was added to the poll function
to ensure this is the case.
As a bonus, when working with multiple armatures this now works as expected
where only the bones of the active armature are selected even if the armature is
shared. The active object is determined by the last bone clicked.
Pull Request: https://projects.blender.org/blender/blender/pulls/148185
Instead of always resetting the 'standard' transform, only reset those
array elements that were actually animated (and whose animation has
thus been transfered to the corresponding 'delta' transform).
This approach also has the advantage of using the defaults from RNA,
rather than hard-coding defaults based on the property name.
Pull Request: https://projects.blender.org/blender/blender/pulls/147982
When connecting node sockets, a search menu opens. Its items need to
be added, and translated manually. This commits adds a few missing
items from recent nodes.
It does the same for sockets coming from node group assets.
Reported by Ye Gui in #43295.
Pull Request: https://projects.blender.org/blender/blender/pulls/148079
This PR slightly raises the accepted noise level for the following EEVEE
tests that were likely incorrectly marked as failed on Intel GPUs:
- render_layer: aov_transparency
- light: triangle_light_sampling
- bsdf: metallic_thinfilm_physical, metallic_thinfilm_f82
- principled_bsdf: principled_bsdf_transmission
Brush directions are dynamic, the enum can have different options
depending on brush type. This commit manually extracts the options.
The labels use the "Brush" translation context, to disambiguate
"Deflate", the operation, with the compression algorithm.
Also `smooth_direction_items` is moved inside
`rna_Brush_direction_itemf()`, since it is used only there, same as
the other brush direction items.
Reported by Ye Gui in #43295.
Pull Request: https://projects.blender.org/blender/blender/pulls/148081
When select-sync was used with both edge & face modes enabled,
vertex selection logic was used which resulted in no visible selection.
Now edge selection is used when both edge and face modes are enabled.
Ref !148181
Implement pinned with select-sync (technically not a bug),
more an oversight in !138197.
Some subtle functional changes have been made.
- Select pinned now only works in vertex select mode
since previously it was possible to select vertices in edge/face modes
where the selection wasn't displayed.
- The island selection option is ignored when selecting pinned.
- The select pinned operator wasn't working with select sync edge/face
modes. Exits with an error instead.
Ref !148181
