This node is a bit of a weird case, because it uses the value stored in an
output socket as an input. So when we want to determine if the Dot
changed, we also have to check if the Normal output changed.
A cleaner solution would be to refactor this by either storing the normal
on the node directly (instead of in an output socket), or by exposing it
by a separate input. This refactor should be done separately though.
The animation playback did not take into account individual stereoscopic views.
This patch fixes this by playing back the active view render.
Reviewed By: campbellbarton
Maniphest Tasks: T91423
Differential Revision: https://developer.blender.org/D14070
Workaround for a compilation issue preventing kernels compiling for AMD GPUs: Avoid problematic use of templates on Metal by making `gpu_parallel_active_index_array` a wrapper macro, and moving `blocksize` to be a macro parameter.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D14081
* Apple Silicon support enabled on macOS 12.2+
* AMD support enabled on macOS 12.3+
This patch also fixes a device enumeration crash on certain AMD configs which
was caused by over-release of MTLDevice objects.
Differential Revision: https://developer.blender.org/D14090
The rbit instruction is only available starting with ARMv6T2 and
the register prefix is different from what AARCH64 uses.
Separate the 32 and 64 bit ARM branches, add missing ISA checks.
Made sure the code works as intended on macMini with Apple silicon,
and on Raspberry Pi 4 B running 32bit Raspbian OS.
Differential Revision: https://developer.blender.org/D14056
Instead of creating and destroying threads when starting and stopping renders,
keep a single thread alive for the duration of the session. This makes it so all
display driver OpenGL resource allocation and destruction can happen in the same
thread.
This was implemented as part of trying to solve another bug, but it did not
help. Still I prefer this behavior, to eliminate potential future issues wit
graphics drivers or with future Cycles display driver implementations.
Differential Revision: https://developer.blender.org/D14086
For reasons unclear, destroying and then recreating a vertex buffer in the
render OpenGL context is affecting the immediate mode vertex buffer in the
draw manager OpenGL context.
Instead just create a single vertex buffer and use it for the lifetime of
the render OpenGL context. There's not really any need to have a separate
one per tile as far as I can tell.
Differential Revision: https://developer.blender.org/D14084
Due to the freeing and re-creation of textures performed when binding
offscreen viewports, VR viewport textures would be needlessly
re-created every drawing iteration, leading to a negative impact on VR
frame rate.
This was brought to light by 6738ecb64e, which introduced an
additional texture clear operation on initialization and was
prohibitively costly on some systems when performed every frame.
Now, the textures for VR viewports will not be always re-created
during offscreen binding, but only when necessary using a pre-drawing
step (`wm_xr_session_surface_offscreen_ensure()`).
Reviewed By: jbakker, fclem
Differential Revision: https://developer.blender.org/D14059
When using a RGBA16 (`GL_RGBA16`, `DXGI_FORMAT_R16G16B16A16_UNORM`)
swapchain format with Quest 2, no image is presented to the headset.
This can occur when using the SteamVR runtime with an AMD graphics card
(ex. T95374).
Workaround is to move this format after the Quest 2-compatible RGBA16F
formats in the candidates list so that the RGBA16F formats are chosen
instead.
Reviewed By: Severin
Differential Revision: https://developer.blender.org/D14024
Crash was caused since the function pointers
`s_xrGetOpenGLGraphicsRequirementsKHR_fn`/
`s_xrGetD3D11GraphicsRequirementsKHR_fn` were static and were not
updated with the correct proc address after being set the first time.
As stated in the OpenXR spec: "function pointers returned by
xrGetInstanceProcAddr using one XrInstance may not be valid when used
with objects related to a different XrInstance".
Although it would seem reasonable that the proc address would not
change if the instance was the same (hence the `static XrInstance s_instance;`),
in testing, repeated calls to `xrGetInstanceProcAddress()`
with the same instance still can result in changes (at least for the
SteamVR runtime) so the workaround is to simply set the function pointers
every time, essentially trivializing their `static` designations.
Reviewed By: Severin
Maniphest Tasks: T94268
Differential Revision: https://developer.blender.org/D14023
For the majority of node groups created in Blender 3.0 the behavior does not change.
So far we only found a single file where this setting has an effect.
Differential Revision: https://developer.blender.org/D14078
Complex Solidify creates edge bevel weights on the rim if the
according vertex has some vertex bevel weight. If there are no
edge bevel weights, they were left disabled even if vertex bevel
weights are used.
Some curve objects don't have an evaluated mesh at all, but line art
currently assumes that all curve objects have one before converting
it to a mesh internally. Fix this by checking if the curve object has an
evaluated mesh before skipping it.
The remaining problem is that evalauted from non-mesh objects or
evaluated curves from non-curve objects, etc. will be ignored if
"Allow Duplicates" is off. That's a different problem though.
Differential Revision: https://developer.blender.org/D14036
For curve-heavy scenes, memory consumption regressed when we switched from MetalRT to bvh2. Allow users to opt in to MetalRT to workaround this.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D14071
Disable binary archives on Apple Silicon (issue stems from instancing multiple PSOs from the same binary archive). Pipeline creation still filters through the OS shader cache, mitigating any impact on setup times after the initial render.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D14072
The main issue is that the image and image user is not updated correctly
in `rna_ImageUser_update`. `BKE_image_user_frame_calc` does not set the
correct frame, because the image is null. Also `IMA_GPU_REFRESH` is not
set for the same reason.
When gpu materials are first created, it is expected that the frame is set
correctly, and the flag is set if necessary. Therefore, somewhere during
depsgraph evaluation, those have to be updated. The depsgraph node
to do the update existed already. Now there is a new relation so that it is
executed when the node tree changed, not only when the frame changed.
This is partially caused by a stupid mistake in cfa53e0fbe
where I missed initializing the `vert_normals` pointer in
`MResolvePixelData`. It's also caused by questionable assumptions
from DerivedMesh code that vertex normals would be valid.
The fix used here is to create a temporary mesh with the data necessary
to compute vertex normals, and ensure them here. This is used because
normal calculation is only implemented for `Mesh` and edit mesh, not
`DerivedMesh`. While this might not be great for performance, it's
potentially aligned with future refactoring of this code to remove
`DerivedMesh` completely. Since this is one of the last places the data
structure is used, that would be a great improvement.
Differential Revision: https://developer.blender.org/D13960
The crash was happening when the mesh had loose edges.
Loose edges are not part of OpenSubdiv topology and hence should not be
communicated to the refiner. Pass ta boolean flag indicating whether an
edge is loose or not in the mesh foreach routines, which seems to be
the easiest way.
There are two things achieved by this change:
- No possible downcast of size_t to int when calculating motion steps.
- Disambiguate call to `min()` which was for some reason considered
ambiguous on 32bit platforms `min(int, unsigned int)`.
- Do the same for the `max()` call to keep them symmetrical.
On an implementation side the `min()` is defined for a fixed width
integer type to disambiguate uint from size_t on 32bit platforms,
and yet be able to use it for 32bit operands on 64bit platforms without
upcast.
This ended up in a bit bigger change as the conditional compile-in of
functions is easiest if the functions is templated. Making the functions
templated required to remove the other source of ambiguity which is
`algorithm.h` which was pulling min/max from std.
Now it is the `math.h` which is the source of truth for min/max.
It was only one place which was relying on `algorithm.h` for these
functions, hence the choice of `math.h` as the safest and least
intrusive.
Fixes 32bit platforms (such as i386) in Debian package build system.
Differential Revision: https://developer.blender.org/D14062
From a strict language point of view the code required a braces around
`trgba` initialization. But it is easier to rely on the fact that fields
which are not specified are zero-initialized.
Under some circumstances, simply adding a curve object and going
to edit mode would cause a crash. This is because the evaluated
`CurveEval` was accessed but also freed by the dependency graph.
The fix reverts the part of b76918717d that uses the
`CurveEval` for the curve object bounds. While this isn't ideal,
it was the previous behavior, and some unexpected behavior
with object bounds is much better than a crash. Plus, given the plans
of using the new "Curves" data-block for evaluated curves, this
situation will change relatively soon anyway.
There are two things achieved by this change:
- No possible downcast of size_t to int when calculating motion steps.
- Disambiguate call to min() which was for some reason considered
ambiguous on 32bit platforms `min(int, unsigned int)`.
On an implementation side the `min()` is defined for a fixed width
integer type to disambiguate uint from size_t on 32bit platforms,
and yet be able to use it for 32bit operands on 64bit platforms without
upcast.
Fixes 32bit platforms (such as i386) in Debian package build system.
Differential Revision: https://developer.blender.org/D13992
Previously, node selection made no distinction between a frame node and
other nodes. So a frame node would be selected by their whole rect or
center (depending on box/lasso/circle select). As a consequence of this,
box and lasso could not pratically be started inside a frame node (with
the intention to select a subset of contained child nodes) because the
frame would be selected immediately and tweak-transforming started.
Circle selecting would always contain the frame node as well (making
transforming a subset of nodes without also transforming the whole frame
impossible).
Now change selection behavior so that for all selection modes only the
border [the margin area that is automatically added around all nodes,
see note below] of a frame node is considered in selection. This makes
for a much more intuitive experience when arranging nodes inside frames.
note: to make the area of interest for selection/moving more obvious,
the cursor changes when hovering over (as is done for resizing).
note: this also makes the resize margin consistent with other nodes.
note: this also fixes right resize border (was exclusive instead of
inclusive as every other border)
Also fixes T46540.
Account for `CurveEval`, which stores the proper deformed and
procedurally created data, unlike the `nurb` list, which has always
just meant a copy of the original curve.
Also account for the case when the curve is empty by using a -1, 1,
fallback bounding box in that case, just like mesh objects.
The problem was that nullptr was returned which is a valid value for
Mesh * and hence the returned optional was treated as having some value.
There was no check for point clouds so that was fixed as well.
Differential Revision: https://developer.blender.org/D14026
Fixes T95384. New exporter was missing a fix for T94516 that recently got applied to the python exporter.
Also changed the obj export tests code so that when save_failing_test_output is requested and MTL result is different from the golden expectation, it is saved as well, similar to how it's done for the OBJ file result.
