Use the shared cache system introduced in e8f4010611 for the
"looptris" triangulation cache. This avoids recalculation when meshes
are copied but the positions or topology don't change. The most obvious
improvement is for cases like a large meshes being adjusted slightly
with a simple geometry nodes modifier. In a basic test with a transform
node with a 1 million point grid I observed an improvement of 13%, from
9.75 to 11 FPS, which shows that we avoid spending 6ms recalculating
the triangulation of every update.
This also makes the thread safety for the triangulation data use a
more standard double-checked lock pattern, which is nice because we
can avoid holding a lock whenever the cached data is retrieved.
Split from https://developer.blender.org/D16530
As part of T95966, this patch moves loose edge information out of the
flag on each edge and into a new lazily calculated cache in mesh
runtime data. The number of loose edges is also cached, so further
processing can be skipped completely when there are no loose edges.
Previously the `ME_LOOSEEDGE` flag was updated on a "best effort"
basis. In order to be sure that it was correct, you had to be sure
to call `BKE_mesh_calc_edges_loose` first. Now the loose edge tag
is always correct. It also doesn't have to be calculated eagerly
in various places like the screw modifier where the complexity
wasn't worth the theoretical performance benefit.
The patch also adds a function to eagerly set the number of loose
edges to zero to avoid building the cache. This is used by various
primitive nodes, with the goal of improving drawing performance.
This results in a few ms shaved off extracting draw data for some
large meshes in my tests.
In the Python API, `MeshEdge.is_loose` is no longer editable.
No built-in addons set the value anyway. The upside is that
addons can be sure the data is correct based on the mesh.
**Tests**
There is one test failure in the Python OBJ exporter: `export_obj_cube`
that happens because of existing incorrect versioning. Opening the
file in master, all the edges were set to "loose", which is fixed
by this patch.
Differential Revision: https://developer.blender.org/D16504
This separates the UV reverse sampling and the barycentric mixing of
the mesh attribute into separate multi-functions. This separates
concerns and allows for future de-duplication of the UV sampling
function if that is implemented as an optimization pass. That would
be helpful since it's the much more expensive operation.
This was simplified by returning the triangle index in the reverse
UV sampler rather than a pointer to the triangle, which required
passing a span of triangles separately in a few places.
(Probably requires ASan for a reliable crash.)
Steps to reproduce were:
* Enter Geometry Nodes Workspace
* Press "New" button in the geometry nodes editor header
* Right-click the data-block selector -> "Mark as Asset"
* Change 3D View to Asset Browser
* Create a catalog
* Drag new Geometry Nodes asset into the catalog
* Save the file
* Press Shift+A in the geometry nodes editor
There was a general issue here with keeping catalog pointers around
during the add menu building. The way it does things, catalogs may be
reloaded in between.
Since the Current File asset library isn't loaded in a separate thread,
the use-after-free would always happen in between. For other libraries
it could still happen, but apparently didn't by chance.
* This patch just moves runtime data to the runtime struct to cleanup
the dna struct. Arguably, some of this data should not even be there
because it's very use case specific. This can be cleaned up separately.
* `miniwidth` was removed completely, because it was not used anywhere.
The corresponding rna property `width_hidden` is kept to avoid
script breakage, but does not do anything (e.g. node wrangler sets it).
* Since rna is in C, some helper functions where added to access the
C++ runtime data from rna.
* This size of `bNode` decreases from 432 to 368 bytes.
A `using FooPtr = std::unique_ptr<Foo>` isn't that useful usually, just
saves a few character stokes. It obfuscates the underlying type, which
is usually relevant information. Plus, `Ptr` for a unique pointer is
misleading (should be `UPtr` or similar).
Move "using" declarations and member variables to the top of the class.
See https://wiki.blender.org/wiki/Style_Guide/C_Cpp#Class_Layout.
Changes access specifiers of some variables from public/protected to
private, there was no point in not having them private.
- Move main comment on class to header comment where it's more visible.
- Improve comment.
- Move stdlib includes first, like we do it usually
- Separate includes my code module
- Remove unnecessary forward declarations
This allows for optimizations because one does not have to iterate
over all nodes anymore to find all nodes within a frame.
Differential Revision: https://developer.blender.org/D16106
Add a new flag value `CUMA_REMOVE` to explicitly tag duplicate points
for removal. This prevents a bug where all curve points with vector
handles were deleted, when removing duplicate curve points while
updating the widget. This happened, because the flag value used to tag
points for removal was the same as the value of `CUMA_HANDLE_VECTOR`
used to store the handle type of the curve point.
Reviewed By: Hans Goudey
Differential Revision: http://developer.blender.org/D16463
The 1a1341c387 made it so that when ID's path changes the ID is
tagged for the source re-evaluation. Another factor here is that
there is a code in the read file which replaces alternative path
slash with the native one.
Typically it is not a problem since IDs are re-evaluated on load,
but the movie clip has the special handling on load to calculate
the image sequence length and initialize principal point.
This change makes it so that the principal point is only reset
when the clip resolution changes. This is something which is
also useful for cases when a non-centered primncipal point is
used and someone accidentally clicks on the clip reload button.
It is not really ideal but covers most of the common cases.
Ideally the principal point will be stored in relative or
normalized space.
The remaining part is that there is now extra image sequence
length calculation after file load. This needs more careful
look.
Steps to reproduce were:
- Open an asset browser
- Open an asset library with assets in it
- Load a different file (e.g. File -> New -> General)
Didn't see a nice way to fix this with the current pre file load handler
callback we use for freeing asset libraries. Using this is cleaner, but
for now, the relationship between UI and asset system is too close
still, so better do explicit freeing at the right point in time.
The launcher is designed to exit as soon as possible
so there's no useless processes idling. Now when steam
launches blender with the launcher, this breaks the
time tracking steam has as the thing it just started
exits within milliseconds.
There already is some code in the launcher that makes
the launcher linger to support background mode. This
patch extends this a bit to also wait if the parent
process is steam.exe
Reviewed by: brecht lichtwerk dingto
Differential Revision: https://developer.blender.org/D16527
- Move code to manage storage to own class in own file, separates
concerns and different levels of abstraction better.
- Store local ID assets separately in the storage class for more
efficient lookups (e.g. for ID remapping).
- Make API function names and comments more complete.
Blender crashes when enabling Use Nodes after the viewport compositor is
already enabled.
This happens because the active viewer key is not yet initialized for the
node tree at this point, which eventually leads to a nullptr.
This patch fixes that by returning the root context in case the active
viewer key is not yet initialized.
Remove unnecessary (and No-op) normal calculation when sculpting on top
of deformed coordinates. Examples are shape keys and deform modifiers.
On a 1 million face mesh, this saved 100ms per stroke update.
This function actually did nothing since cfa53e0fbe,
so that large improvement comes for free.
Conceptually this is correct because when sculpting on deformed
coordinates, we don't change the positions of the base mesh directly.
In the future it might be better to allocate a separate array for
normals when using deformed coordinates, but it's not clear that's
necessary yet.
Ignore difference between source and target tree type. When copying
nodes from clipboard to target tree compatibility is checked. After
pasting nodes only the links between nodes that are existing in the
node tree are added.
See Task T95033.
Differential Revision: https://developer.blender.org/D16349
As described in T100004, add an output socket that returns true if the
attribute accessed by the node was already present in that context.
Initial patch by Edward (@edward88).
Differential Revision: https://developer.blender.org/D16316
10131a6f62 replaced use of the `ME_EDGERENDER` flag with
`ME_EDGEDRAW`. However, left over from previous refactors, code
for leaving edit mode set that flag based on the edge angle. Edge angle
wireframe hiding is currently supposed to be adjustable with the
wireframe overlay settings. This patch restores the previous behavior
from before the cleanup commit.
Differential Revision: https://developer.blender.org/D16451
Currently the positions are retrieved again for every vertex. This is
slow, and will get slower when positions are stored as a named
attribute. Saves around 0.5ms per stroke update when a modifier
is active in my test with a 1 million face mesh.
This reverts commit 676137f043.
This change worked locally with a specific test file and local changes,
but didn't work in general, since we don't reliably retrieve the new
looptris after setting them the first time. This can be improved again
in the future, but probably along with a more general look about ownership
is handled with PBVH.
This avoids recalculation of looptri derived triangulation whenever
switching to sculpt mode or whenever the PBVH is rebuilt, which can
happen after strokes in some situations. In my tests actually building
the PBVH is much more expensive (300ms), but this saves 6ms when
switching to sculpt mode and in other situations.
The cost is the possibility of higher memory usage because the cache
will live in the original main database mesh. However, the impact of
that will be smaller when the shared cache concept from D16204 is
applied to this data too.
This assertion function came from when derived normal data was stored
as custom data layers, which made it harder to keep track of whether
it was allocated and propagated. Nowadays it's all relatively easy to
predict, so there's no point in keeping this function around-- it only
makes code longer and more complex looking.
