The ff_cfhd_init_vlcs() function was using a lot of stack space, which
made linker on macOS unhappy. Using heap allocation allows to silence
the warning without causing other side-effects.
Kept the patch enabled for all platforms to avoid difference in behavior
and performance on different platforms, which could make certain types
of investigation very tricky.
Differential Revision: https://developer.blender.org/D8248
This was the last of the three network optimizations I developed in
the functions branch. Common subnetwork elimination and constant
folding together can get rid of most unnecessary nodes.
This can be used to find separate islands in meshes efficiently (as is
done in cycles already). Furthermore, this helps to implement some
algorithms on node trees more efficiently.
C++17 does not work on 10.12, and Apple extended support ended for 10.12 in
October 2019.
Maniphest Tasks: T76783, T76184
Differential Revision: https://developer.blender.org/D8179
The spelling and capitalization of package name passed to find_package()
and find_package_handle_standard_args() needs to match.
Silences CMake warning about mismatch.
Differential Revision: https://developer.blender.org/D8247
Those optimizations work on the multi-function network level.
Not only will they make the network evaluation faster, but they also
simplify the network a lot. That makes it easier to understand the
exported dot graph.
This can be useful to save the result of a cloth simulation as a
shape key without destroying the simulation, so it's possible to
e.g. re-run it to get other shapes, or simply use the new shape
key to start the simulation already in a draped state.
It also makes sense to allow applying as shape key even when the
mesh is shared, because the operation itself just adds a shape
key. To support this, split the apply operator into Apply and
Apply As Shapekey so that they can have different poll callbacks.
Differential Revision: https://developer.blender.org/D8173
The problem here was numerical precision: The code calculates the angle between
sun and view direction, and the usual acos(dot(a, b)) approach for that has
poor numerical performance for almost parallel angles.
As a result, the generally tiny difference between floating point computation
between CPU and GPU was enough to make the sun vanish at different radii,
causing different results.
The new version fixes the difference by making the computation much more robust
on both platforms.
This adds new callbacks to `bNodeSocketType` and `bNodeType`.
Those are used to generate a multi-function network from a node
tree. Later, this network is evaluated on e.g. particle data.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D8169
The MSVC atomic function is defined for an unsigned type.
Not sure why this became an issue after switch to TBB by default,
maybe some CFLAGS changed to be more strict after that.
The upstream version of nasm does not put version information to the
generated object files, which makes linker to show the following
warning:
building for macOS, but linking in object file
Using own patched version of nasm which puts required information to
the object file, making linker happy.
The plan is to either streamline the patch and provide it to the
upstream, or, it that takes too long, get an independent fix from the
upstream.
Issue is reported on Linux ith Intel HD6xx iGPU. Inside
`gpu_select_sample_query.c` the call to `glGetQueryObjectuiv` froze. After
bisecting this lead to the polyline shader. When using a 3d color shader
in stead of the polyline shader during selection seems to fix the issue.
Other parts of blender might also be effective, but I wasn't able to
freeze blender in these areas. When it does, we might want to add
a similar work-around to button2d, cage2d, cage3d & move3d, navigate.
Backport this patch to 2.83.
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D8217
The knife code currently calls the `BLI_bvhtree_overlap` function that
tests the overlap between the mesh tree and an AABB that encompasses the
points projected in the clip_start, clip_end and or clip_planes of the
view.
This resulted in many false positives since the AABB is very large.
Often all the triangles "overlapped".
The solution was to create a new function that actually tests the
intersection of AABB with a plane.
Even not considering the clip_planes of the view, this solution is more
appropriate than using overlap.
Differential Revision: https://developer.blender.org/D8229
This was due to a bad driver which was not respecting this bit of the
specification:
`If the current primitive does not originate from an instanced draw command, the value of gl_InstanceID is zero.`
This patch adds support for the curve primitive from OptiX to Cycles. It's currently hidden
behind a debug option, since there can be some slight rendering differences still (because no
backface culling is performed and something seems off with endcaps). The curve primitive
was added with the OptiX 7.1 SDK and requires a r450 driver or newer, so this also updates
the codebase to be able to build with the new SDK.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D8223
Don't apply the matrix transform optimization in this case, curve points and
radius can't represent non-uniform scale the way is possible with triangle
meshes and vertices.
This would cause abrupt change if objects had e.g. motion blur in one frame
and not in the next.
Each duplicated (a.k.a. instanced) object has a Persistent ID, which
identifies a dupli within the context of its duplicator. This ID
consists of several numbers when there are nested duplis (for example a
mesh instancing empties on its vertices, where each empty instances a
collection). When exporting to Alembic/USD, these are used to uniquely
name the duplicated objects in the export.
This commit reverses the order of the persistent ID numbers, so that the
first number identifies the first level of recursion. This produces
trees like this:
ABC
`--Triangle
|--Triangle
|--Empty-1
| `--Pole-1-0
| |--Pole
| `--Block-1-1
| `--Block
|--Empty
| `--Pole-0
| |--Pole
| `--Block-1
| `--Block
|--Empty-2
| `--Pole-2-0
| |--Pole
| `--Block-2-1
| `--Block
`--Empty-0
`--Pole-0-0
|--Pole
`--Block-0-1
`--Block
It is now clearer that `Pole-2-0` and `Block-2-1` are instanced by
`Empty-2`. Before this commit, they would have been named `Pole-0-2` and
`Block-1-2`.
Exporting a scene to USD or Alembic would fail when there are multiple
duplicates of parent & child objects, duplicated by the same object. For
example, this happens when such a hierarchy of objects is contained in a
collection, and that collection is instanced multiple times by mesh
vertices. The problem here is that the 'parent' pointer of each
duplicated object points to the real parent; Blender would not figure
out properly which duplicated parent should be used.
This is now resolved by keeping track of the persistent ID of each
duplicated instance, which makes it possible to reconstruct the
parent-child relations of duplicated objects. This does use up some
memory for each dupli, so it could be heavy to export a Spring scene
(with all the pebbles and leaves), but it's only a small addition on top
of the USD/Alembic writer objects that have to be created anyway. At
least with this patch, they're created correctly.
Code-wise, the following changes are made:
- The export graph (that maps export parent to its export children) used
to have as its key (Object, Duplicator). This is insufficient to
correctly distinguish between multiple duplis of the same object by
the same duplicator, so this is now extended to (Object, Duplicator,
Persistent ID). To make this possible, new classes `ObjectIdentifier`
and `PersistentID` are introduced.
- Finding the parent of a duplicated object is done via its persistent
ID. In Python notation, the code first tries to find the parent
instance where `child_persistent_id[1:] == parent_persistent_id[1:]`.
If that fails, the dupli with persistent ID `child_persistent_id[1:]`
is used as parent.
Reviewed By: sergey
Differential Revision: https://developer.blender.org/D8233
We don't need it and it was optionally enabled, causing Blender to fail
to link on certain configuration (when Brotli is installed via Homebrew
for example).
This is just a change in `AbstractHierarchyIterator::debug_print_export_graph()`
to aid in debugging. It'll make it possible to distinguish between
different duplicates of the same object.
No functional changes to Blender itself.
The unit being "pixels".
Before this change the solve errors were unitless in the UI.
With this change in place, the UI is now clear on that the unit of the
reprojection errors is pixels (px).
Differential Revision: https://developer.blender.org/D8000
