Implicit sharing means attribute ownership is shared between geometry
data-blocks, and the sharing happens automatically. So it's unnecessary
to choose whether to enable it when copying a mesh.
Add the ability to retrieve implicit sharing info directly from the
C++ attribute API, which simplifies memory usage and performance
optimizations making use of it. This commit uses the additions to
the API to avoid copies in a few places:
- The "rest_position" attribute in the mesh modifier stack
- Instance on Points node
- Instances to points node
- Mesh to points node
- Points to vertices node
Many files are affected because in order to include the new information
in the API's returned data, I had to switch a bunch of types from
`VArray` to `AttributeReader`. This generally makes sense anyway, since
it allows retrieving the domain, which wasn't possible before in some
cases. I overloaded the `*` deference operator for some syntactic sugar
to avoid the (very ugly) `.varray` that would be necessary otherwise.
Pull Request: https://projects.blender.org/blender/blender/pulls/107059
Implements #95966, as the final step of #95965.
This commit changes the storage of mesh edge vertex indices from the
`MEdge` type to the generic `int2` attribute type. This follows the
general design for geometry and the attribute system, where the data
storage type and the usage semantics are separated.
The main benefit of the change is reduced memory usage-- the
requirements of storing mesh edges is reduced by 1/3. For example,
this saves 8MB on a 1 million vertex grid. This also gives performance
benefits to any memory-bound mesh processing algorithm that uses edges.
Another benefit is that all of the edge's vertex indices are
contiguous. In a few cases, it's helpful to process all of them as
`Span<int>` rather than `Span<int2>`. Similarly, the type is more
likely to match a generic format used by a library, or code that
shouldn't know about specific Blender `Mesh` types.
Various Notes:
- The `.edge_verts` name is used to reflect a mapping between domains,
similar to `.corner_verts`, etc. The period means that it the data
shouldn't change arbitrarily by the user or procedural operations.
- `edge[0]` is now used instead of `edge.v1`
- Signed integers are used instead of unsigned to reduce the mixing
of signed-ness, which can be error prone.
- All of the previously used core mesh data types (`MVert`, `MEdge`,
`MLoop`, `MPoly` are now deprecated. Only generic types are used).
- The `vec2i` DNA type is used in the few C files where necessary.
Pull Request: https://projects.blender.org/blender/blender/pulls/106638
This integrates the new implicit-sharing system (from fbcddfcd68)
with `CustomData`. Now the potentially long arrays referenced by custom
data layers can be shared between different systems but most importantly
between different geometries. This makes e.g. copying a mesh much cheaper
because none of the attributes has to be copied. Only when an attribute
is modified does it have to be copied.
Also see the original design task: #95845.
This reduces memory and improves performance by avoiding unnecessary
data copies. For example, the used memory after loading a highly
subdivided mesh is reduced from 2.4GB to 1.79GB. This is about 25%
less which is the expected amount because in `main` there are 4 copies
of the data:
1. The original data which is allocated when the file is loaded.
2. The copy for the depsgraph allocated during depsgraph evaluation.
3. The copy for the undo system allocated when the first undo step is
created right after loading the file.
4. GPU buffers allocated for drawing.
This patch only gets rid of copy number 2 for the depsgraph. In theory
the other copies can be removed as part of follow up PRs as well though.
-----
The patch has three main components:
* Slightly modified `CustomData` API to make it work better with implicit
sharing:
* `CD_REFERENCE` and `CD_DUPLICATE` have been removed because they are
meaningless when implicit-sharing is used.
* `CD_ASSIGN` has been removed as well because it's not an allocation
type anyway. The functionality of using existing arrays as custom
data layers has not been removed though.
* This can still be done with `CustomData_add_layer_with_data` which
also has a new argument that allows passing in information about
whether the array is shared.
* `CD_FLAG_NOFREE` has been removed because it's no longer necessary. It
only existed because of `CD_REFERENCE`.
* `CustomData_copy` and `CustomData_merge` have been split up into a
functions that do copy the actual attribute values and those that do
not. The latter functions now have the `_layout` suffix
(e.g. `CustomData_copy_layout`).
* Changes in `customdata.cc` to make it actually use implicit-sharing.
* Changes in various other files to adapt to the changes in `BKE_customdata.h`.
Pull Request: https://projects.blender.org/blender/blender/pulls/106228
Since the positions of the final mesh are modified rather than a
separate array after d20f992322, the final mesh has
to become the deformed mesh after the deform modifiers, rather
than the input mesh. The arrays can't be shared anymore, but that
performance loss will be solved by implicit sharing shortly.
Implements #95967.
Currently the `MPoly` struct is 12 bytes, and stores the index of a
face's first corner and the number of corners/verts/edges. Polygons
and corners are always created in order by Blender, meaning each
face's corners will be after the previous face's corners. We can take
advantage of this fact and eliminate the redundancy in mesh face
storage by only storing a single integer corner offset for each face.
The size of the face is then encoded by the offset of the next face.
The size of a single integer is 4 bytes, so this reduces memory
usage by 3 times.
The same method is used for `CurvesGeometry`, so Blender already has
an abstraction to simplify using these offsets called `OffsetIndices`.
This class is used to easily retrieve a range of corner indices for
each face. This also gives the opportunity for sharing some logic with
curves.
Another benefit of the change is that the offsets and sizes stored in
`MPoly` can no longer disagree with each other. Storing faces in the
order of their corners can simplify some code too.
Face/polygon variables now use the `IndexRange` type, which comes with
quite a few utilities that can simplify code.
Some:
- The offset integer array has to be one longer than the face count to
avoid a branch for every face, which means the data is no longer part
of the mesh's `CustomData`.
- We lose the ability to "reference" an original mesh's offset array
until more reusable CoW from #104478 is committed. That will be added
in a separate commit.
- Since they aren't part of `CustomData`, poly offsets often have to be
copied manually.
- To simplify using `OffsetIndices` in many places, some functions and
structs in headers were moved to only compile in C++.
- All meshes created by Blender use the same order for faces and face
corners, but just in case, meshes with mismatched order are fixed by
versioning code.
- `MeshPolygon.totloop` is no longer editable in RNA. This API break is
necessary here unfortunately. It should be worth it in 3.6, since
that's the best way to allow loading meshes from 4.0, which is
important for an LTS version.
Pull Request: https://projects.blender.org/blender/blender/pulls/105938
This was only used in a single place, multires baking, where it's simple
to just retrieve the looptris from the temporary `Mesh` that we already
create to simplify accessing normals. This allows removing a bunch of
complexity from `DerivedMesh`, to help progress towards its complete
removal at some point in the future.
Also use `Span` more instead of raw pointers in multires baking,
and pass around the spans instead of the low-resolution DerivedMesh.
The goal is to solve confusion of the "All rights reserved" for licensing
code under an open-source license.
The phrase "All rights reserved" comes from a historical convention that
required this phrase for the copyright protection to apply. This convention
is no longer relevant.
However, even though the phrase has no meaning in establishing the copyright
it has not lost meaning in terms of licensing.
This change makes it so code under the Blender Foundation copyright does
not use "all rights reserved". This is also how the GPL license itself
states how to apply it to the source code:
<one line to give the program's name and a brief idea of what it does.>
Copyright (C) <year> <name of author>
This program is free software ...
This change does not change copyright notice in cases when the copyright
is dual (BF and an author), or just an author of the code. It also does
mot change copyright which is inherited from NaN Holding BV as it needs
some further investigation about what is the proper way to handle it.
Implements #102359.
Split the `MLoop` struct into two separate integer arrays called
`corner_verts` and `corner_edges`, referring to the vertex each corner
is attached to and the next edge around the face at each corner. These
arrays can be sliced to give access to the edges or vertices in a face.
Then they are often referred to as "poly_verts" or "poly_edges".
The main benefits are halving the necessary memory bandwidth when only
one array is used and simplifications from using regular integer indices
instead of a special-purpose struct.
The commit also starts a renaming from "loop" to "corner" in mesh code.
Like the other mesh struct of array refactors, forward compatibility is
kept by writing files with the older format. This will be done until 4.0
to ease the transition process.
Looking at a small portion of the patch should give a good impression
for the rest of the changes. I tried to make the changes as small as
possible so it's easy to tell the correctness from the diff. Though I
found Blender developers have been very inventive over the last decade
when finding different ways to loop over the corners in a face.
For performance, nearly every piece of code that deals with `Mesh` is
slightly impacted. Any algorithm that is memory bottle-necked should
see an improvement. For example, here is a comparison of interpolating
a vertex float attribute to face corners (Ryzen 3700x):
**Before** (Average: 3.7 ms, Min: 3.4 ms)
```
threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) {
for (const int64_t i : range) {
dst[i] = src[loops[i].v];
}
});
```
**After** (Average: 2.9 ms, Min: 2.6 ms)
```
array_utils::gather(src, corner_verts, dst);
```
That's an improvement of 28% to the average timings, and it's also a
simplification, since an index-based routine can be used instead.
For more examples using the new arrays, see the design task.
Pull Request: https://projects.blender.org/blender/blender/pulls/104424
This simplifies the usage of the API and is preparation for #104478.
The `CustomData_add_layer` and `CustomData_add_layer_named` now have corresponding
`*_with_data` functions that should be used when creating the layer from existing data.
Pull Request: https://projects.blender.org/blender/blender/pulls/105708
Refactoring mesh code, it has become clear that local cleanups and
simplifications are limited by the need to keep a C public API for
mesh functions. This change makes code more obvious and makes further
refactoring much easier.
- Add a new `BKE_mesh.hh` header for a C++ only mesh API
- Introduce a new `blender::bke::mesh` namespace, documented here:
https://wiki.blender.org/wiki/Source/Objects/Mesh#Namespaces
- Move some functions to the new namespace, cleaning up their arguments
- Move code to `Array` and `float3` where necessary to use the new API
- Define existing inline mesh data access functions to the new header
- Keep some C API functions where necessary because of RNA
- Move all C++ files to use the new header, which includes the old one
In the future it may make sense to split up `BKE_mesh.hh` more, but for
now keeping the same name as the existing header keeps things simple.
Pull Request: https://projects.blender.org/blender/blender/pulls/105416
With the goal of clearly differentiating between arrays and single
elements, improving consistency across Blender, and using wording
that's easier to read and say, change variable names for Mesh edges
and polygons/faces.
Common renames are the following, with some extra prefixes, etc.
- `mpoly` -> `polys`
- `mpoly`/`mp`/`p` -> `poly`
- `medge` -> `edges`
- `med`/`ed`/`e` -> `edge`
`MLoop` variables aren't affected because they will be replaced
when they're split up into to arrays in #104424.
Remove the use of a separate contiguous positions array now that
they are stored that way in the first place. This allows removing the
complexity of tracking whether it is allocated and deformed in the
mesh modifier stack.
Instead of deferring the creation of the final mesh until after the
positions have been copied and deformed, create the final mesh
first and then deform its positions.
Since vertex and face normals are calculated lazily, we can rely on
individual modifiers to calculate them as necessary and simplify
the modifier stack. This was hard to change before because of the
separate array of deformed positions.
Differential Revision: https://developer.blender.org/D16971
The goal is to give technical artists the ability to optimize modifier usage
and/or geometry node groups for performance. In the long term, it
would be useful if Blender could provide its own UI to display profiling
information to users. However, right now, there are too many open
design questions making it infeasible to tackle this in the short term.
This commit uses a simpler approach: Instead of adding new ui for
profiling data, it exposes the execution-time of modifiers in the Python
API. This allows technical artists to access the information and to build
their own UI to display the relevant information. In the long term this
will hopefully also help us to integrate a native ui for this in Blender
by observing how users use this information.
Note: The execution time of a modifier highly depends on what other
things the CPU is doing at the same time. For example, in many more
complex files, many objects and therefore modifiers are evaluated at
the same time by multiple threads which makes the measurement
much less reliable. For best results, make sure that only one object
is evaluated at a time (e.g. by changing it in isolation) and that no
other process on the system keeps the CPU busy.
As shown below, the execution time has to be accessed on the
evaluated object, not the original object.
```lang=python
import bpy
depsgraph = bpy.context.view_layer.depsgraph
ob = bpy.context.active_object
ob_eval = ob.evaluated_get(depsgraph)
modifier_eval = ob_eval.modifiers[0]
print(modifier_eval.execution_time, "s")
```
Differential Revision: https://developer.blender.org/D17185
Currently you can retrieve a mutable array from a const CustomData.
That makes code unsafe since the compiler can't check for correctness
itself. Fix that by introducing a separate function to retrieve mutable
arrays from CustomData. The new functions have the `_for_write`
suffix that make the code's intention clearer.
Because it makes retrieving write access an explicit step, this change
also makes proper copy-on-write possible for attributes.
Notes:
- The previous "duplicate referenced layer" functions are redundant
with retrieving layers with write access
- The custom data functions that give a specific index only have
`for_write` to simplify the API
Differential Revision: https://developer.blender.org/D14140
Currently the `MLoopUV` struct stores UV coordinates and flags related
to editing UV maps in the UV editor. This patch changes the coordinates
to use the generic 2D vector type, and moves the flags into three
separate boolean attributes. This follows the design in T95965, with
the ultimate intention of simplifying code and improving performance.
Importantly, the change allows exporters and renderers to use UVs
"touched" by geometry nodes, which only creates generic attributes.
It also allows geometry nodes to create "proper" UV maps from scratch,
though only with the Store Named Attribute node for now.
The new design considers any 2D vector attribute on the corner domain
to be a UV map. In the future, they might be distinguished from regular
2D vectors with attribute metadata, which may be helpful because they
are often interpolated differently.
Most of the code changes deal with passing around UV BMesh custom data
offsets and tracking the boolean "sublayers". The boolean layers are
use the following prefixes for attribute names: vert selection: `.vs.`,
edge selection: `.es.`, pinning: `.pn.`. Currently these are short to
avoid using up the maximum length of attribute names. To accommodate
for these 4 extra characters, the name length limit is enlarged to 68
bytes, while the maximum user settable name length is still 64 bytes.
Unfortunately Python/RNA API access to the UV flag data becomes slower.
Accessing the boolean layers directly is be better for performance in
general.
Like the other mesh SoA refactors, backward and forward compatibility
aren't affected, and won't be changed until 4.0. We pay for that by
making mesh reading and writing more expensive with conversions.
Resolves T85962
Differential Revision: https://developer.blender.org/D14365
**Changes**
As described in T93602, this patch removes all use of the `MVert`
struct, replacing it with a generic named attribute with the name
`"position"`, consistent with other geometry types.
Variable names have been changed from `verts` to `positions`, to align
with the attribute name and the more generic design (positions are not
vertices, they are just an attribute stored on the point domain).
This change is made possible by previous commits that moved all other
data out of `MVert` to runtime data or other generic attributes. What
remains is mostly a simple type change. Though, the type still shows up
859 times, so the patch is quite large.
One compromise is that now `CD_MASK_BAREMESH` now contains
`CD_PROP_FLOAT3`. With the general move towards generic attributes
over custom data types, we are removing use of these type masks anyway.
**Benefits**
The most obvious benefit is reduced memory usage and the benefits
that brings in memory-bound situations. `float3` is only 3 bytes, in
comparison to `MVert` which was 4. When there are millions of vertices
this starts to matter more.
The other benefits come from using a more generic type. Instead of
writing algorithms specifically for `MVert`, code can just use arrays
of vectors. This will allow eliminating many temporary arrays or
wrappers used to extract positions.
Many possible improvements aren't implemented in this patch, though
I did switch simplify or remove the process of creating temporary
position arrays in a few places.
The design clarity that "positions are just another attribute" brings
allows removing explicit copying of vertices in some procedural
operations-- they are just processed like most other attributes.
**Performance**
This touches so many areas that it's hard to benchmark exhaustively,
but I observed some areas as examples.
* The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster.
* The Spring splash screen went from ~4.3 to ~4.5 fps.
* The subdivision surface modifier/node was slightly faster
RNA access through Python may be slightly slower, since now we need
a name lookup instead of just a custom data type lookup for each index.
**Future Improvements**
* Remove uses of "vert_coords" functions:
* `BKE_mesh_vert_coords_alloc`
* `BKE_mesh_vert_coords_get`
* `BKE_mesh_vert_coords_apply{_with_mat4}`
* Remove more hidden copying of positions
* General simplification now possible in many areas
* Convert more code to C++ to use `float3` instead of `float[3]`
* Currently `reinterpret_cast` is used for those C-API functions
Differential Revision: https://developer.blender.org/D15982
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.
The regression is caused by D13824 0f89bcdbeb.
This fix follows the code from Sybren (D7785) to make object-mode
drivers from shapekey value to work. This intuitively makes sense
since the D13824 made the edit mode evaluation and ownership follow
the object mode more closely.
Differential Revision: https://developer.blender.org/D16380
This commit replaces the `Mesh_Runtime` struct embedded in `Mesh`
with `blender::bke::MeshRuntime`. This has quite a few benefits:
- It's possible to use C++ types like `std::mutex`, `Array`,
`BitVector`, etc. more easily
- Meshes saved in files are slightly smaller
- Copying and writing meshes is a bit more obvious without
clearing of runtime data, etc.
The first is by far the most important. It will allows us to avoid a
bunch of manual memory management boilerplate that is error-prone and
annoying. It should also simplify future CoW improvements for runtime
data.
This patch doesn't change anything besides changing `mesh.runtime.data`
to `mesh.runtime->data`. The cleanups above will happen separately.
Differential Revision: https://developer.blender.org/D16180
This is very similar to D14077. There are two differences though.
First is that vertex creases are already stored in a separate layer,
and second is that we can now completely remove use of `Mesh.cd_flag`,
since that information is now inherent to whether the layers exist.
There are two functional differences here:
* Operators are used to add and remove layers instead of a property.
* The "crease" attribute can be created and removed by geometry nodes.
The second change should make various geometry nodes slightly faster,
since the "crease" attribute was always processed before. Creases are
now interpolated generically in the CustomData API too, which should
help maintain the values across edits better.
Meshes get an `edge_creases` RNA property like the existing vertex
property, to provide more efficient access to the data in Cycles.
One test failure is expected, where different rounding between float
the old char storage means that 5 additional points are scattered in
a geometry nodes test.
Differential Revision: https://developer.blender.org/D15927
When a change happens which invalidates view layers the syncing will be postponed until the first usage.
This will improve importing or adding many objects in a single operation/script.
`BKE_view_layer_need_resync_tag` is used to tag the view layer to be out of sync. Before accessing
`BKE_view_layer_active_base_get`, `BKE_view_layer_active_object_get`, `BKE_view_layer_active_collection`
or `BKE_view_layer_object_bases` the caller should call `BKE_view_layer_synced_ensure`.
Having two functions ensures that partial syncing could be added as smaller patches in the future. Tagging a
view layer out of sync could be replaced with a partial sync. Eventually the number of full resyncs could be
reduced. After all tagging has been replaced with partial syncs the ensure_sync could be phased out.
This patch has been added to discuss the details and consequences of the current approach. For clarity
the call to BKE_view_layer_ensure_sync is placed close to the getters.
In the future this could be placed in more strategical places to reduce the number of calls or improve
performance. Finding those strategical places isn't that clear. When multiple operations are grouped
in a single script you might want to always check for resync.
Some areas found that can be improved. This list isn't complete.
These areas aren't addressed by this patch as these changes would be hard to detect to the reviewer.
The idea is to add changes to these areas as a separate patch. It might be that the initial commit would reduce
performance compared to master, but will be fixed by the additional patches.
**Object duplication**
During object duplication the syncing is temporarily disabled. With this patch this isn't useful as when disabled
the view_layer is accessed to locate bases. This can be improved by first locating the source bases, then duplicate
and sync and locate the new bases. Will be solved in a separate patch for clarity reasons ({D15886}).
**Object add**
`BKE_object_add` not only adds a new object, but also selects and activates the new base. This requires the
view_layer to be resynced. Some callers reverse the selection and activation (See `get_new_constraint_target`).
We should make the selection and activation optional. This would make it possible to add multiple objects
without having to resync per object.
**Postpone Activate Base**
Setting the basact is done in many locations. They follow a rule as after an action find the base and set
the basact. Finding the base could require a resync. The idea is to store in the view_layer the object which
base will be set in the basact during the next sync, reducing the times resyncing needs to happen.
Reviewed By: mont29
Maniphest Tasks: T73411
Differential Revision: https://developer.blender.org/D15885
corresponding data layers and using their values for computations.
Avoiding that should increase performance in many operations that
would otherwise have to read, write, or propagate these values.
It also means decreased memory usage-- not just for sculpt mode
but for any mesh that was in sculpt mode. Previously the mask, face set,
and hide status layers were *always* allocated by sculpt mode.
Here are a few basic tests when masking and face sets are not used:
| Test | Before | After |
| Subsurf Modifier | 148 ms | 126 ms |
| Sculpt Overlay Extraction | 24 ms every redraw | 0 ms |
| Memory usage | 252 MB | 236 MB |
I wouldn't expect any difference when they are used though.
The code changes are mostly just making sculpt features safe for when
the layers aren't stored, and some changes to the conversion to and
from the hide layers. Use of the ".hide_poly" attribute replaces testing
whether face sets are negative in many places.
Differential Revision: https://developer.blender.org/D15937
Replace `mesh_attributes`, `mesh_attributes_for_write` and the point
cloud versions with methods on the `Mesh` and `PointCloud` types.
This makes them friendlier to use and improves readability.
Differential Revision: https://developer.blender.org/D15907
Use `verts` instead of `vertices` and `polys` instead of `polygons`
in the API added in 05952aa94d. This aligns better with
existing naming where the shorter names are much more common.
For copy-on-write, we want to share attribute arrays between meshes
where possible. Mutable pointers like `Mesh.mvert` make that difficult
by making ownership vague. They also make code more complex by adding
redundancy.
The simplest solution is just removing them and retrieving layers from
`CustomData` as needed. Similar changes have already been applied to
curves and point clouds (e9f82d3dc7, 410a6efb74). Removing use of
the pointers generally makes code more obvious and more reusable.
Mesh data is now accessed with a C++ API (`Mesh::edges()` or
`Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`).
The CoW changes this commit makes possible are described in T95845
and T95842, and started in D14139 and D14140. The change also simplifies
the ongoing mesh struct-of-array refactors from T95965.
**RNA/Python Access Performance**
Theoretically, accessing mesh elements with the RNA API may become
slower, since the layer needs to be found on every random access.
However, overhead is already high enough that this doesn't make a
noticible differenc, and performance is actually improved in some
cases. Random access can be up to 10% faster, but other situations
might be a bit slower. Generally using `foreach_get/set` are the best
way to improve performance. See the differential revision for more
discussion about Python performance.
Cycles has been updated to use raw pointers and the internal Blender
mesh types, mostly because there is no sense in having this overhead
when it's already compiled with Blender. In my tests this roughly
halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million
face grid).
Differential Revision: https://developer.blender.org/D15488
These few lines making a copy of the final mesh were confusing.
The goal (I'm fairly certain) is to make sure the cage mesh and final
mesh aren't shared when applying the vertex coordinates to the final
mesh. This can be done more simply though, in a way that avoids
duplicating the final mesh if it already isn't shared.
This works well in some basic tests with different modifiers. Though
I doubt it was really a bottleneck anywhere, simplifying the modifier
stack internals is always nice.
Differential Revision: https://developer.blender.org/D15814
In all these cases, it was clear that the layer values were set right
after the layer was created anyway. So there's no point in using
calloc or setting the values to zero first.
See 25237d2625 for more info.
