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: blender/blender#104424
Since normals are derived data, it's always a change to something
else that will cause them to change, like the winding order of a face
or vertex positions. So it's clearer to use tags for those things
directly. It's correct to remove the tag in one place since dirty is
the default state of a new mesh.
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: blender/blender#105416
Consistent with naming from 1af62cb3bf. Keep the "coord"
naming in the "vert_coords_alloc" set of functions since they should be
removed (see #103789).
When there are no loose edges, the loose edge bitmap shouldn't be used.
That was already documented in the loose edge storage struct, but the
bit vector wasn't actually cleared.
As described in #95966, replace the `ME_EDGEDRAW` flag with a bit
vector in mesh runtime data. Currently the the flag is only ever set
to false for the "optimal display" feature of the subdivision surface
modifier. When creating an "original" mesh in the main data-base,
the flag is always supposed to be true.
The bit vector is now created by the modifier only as necessary, and
is cleared for topology-changing operations. This fixes incorrect
interpolation of the flag as noted in #104376. Generally it isn't
possible to interpolate it through topology-changing operations.
After this, only the seam status needs to be removed from edges before
we can replace them with the generic `int2` type (or something similar)
and reduce memory usage by 1/3.
Related:
- 10131a6f62
- 145839aa42
In the future `BM_ELEM_DRAW` could be removed as well. Currently it is
used and aliased by other defines in some non-obvious ways though.
Pull Request #104417
**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
Instead of creating a new mesh from scratch, modify an existing mesh.
This allows us to keep derived caches for triangulation and bounds
alive more easily, and allows keeping materials and non-generic
attributes like vertex groups alive on the mesh.
It also has other performance benefits, since face and face corner
attributes aren't affected at all, and because of reduced overhead
from not allocating a new mesh.
Updating edge attributes is a bit more complicated now, since we
have to completely replace the arrays but keep the existing attribute
IDs around. The new mesh update tag is also slightly too specific IMO.
But I think both of those things will improve in the future because
of existing plans for further refactoring these areas:
- New attribute storage that gives pointer stability
- Further use and granularity of mesh update tagging that will
make the correct API more clear
Fixes T102711
Differential Revision: https://developer.blender.org/D16615
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 only applies to procedural operations rather than edit mode
operations, but it might save some recalculations of these caches
for the transform geometry node in some cases.
Separate freeing and clearing mesh runtime data in a more obvious way.
This makes it easier to see what data is meant to be cleared on certain
changes, rather than conflating it with freeing all of the runtime
caches.
Also comment and reduce the surface area of the "mesh runtime" API.
The redundancy in some functions made it confusing which one should
be used, resulting in subtle bugs or unnecessary boilerplate code.
Also, now bke::MeshRuntime is able to free all the data it owns by
itself, which makes this area easier to reason about. That required
changing the interface of a few functions to avoid passing Mesh when
they really just dealt with some runtime struct.
With more RAII semantics in the future, more of this manual freeing
will become unnecessary.
Bounding box calculation can be a large in some situations, especially
instancing. This patch caches the min and max of the bounding box in
runtime data of meshes, point clouds, and curves, implementing part of
T96968.
Bounds are now calculated lazily-- only after they are tagged dirty.
Also, cached bounds are also shared when copying geometry data-blocks
that have equivalent data. When bounds are calculated on an evaluated
data-block, they are also accessible on the original, and the next
evaluated ID will also share them. A geometry will stop sharing bounds
as soon as its positions (or radii) are changed.
Just caching the bounds gave a 2-3x speedup with thousands of mesh
geometry instances in the viewport. Sharing the bounds can eliminate
recalculations entirely in cases like copying meshes in geometry nodes
or the selection paint brush in curves sculpt mode, which causes a
reevaluation but doesn't change the positions.
**Implementation**
The sharing is achieved with a `shared_ptr` that points to a cache mutex
(from D16419) and the cached bounds data. When geometries are copied,
the bounds are shared by default, and only "un-shared" when the bounds
are tagged dirty.
Point clouds have a new runtime struct to store this data. Functions
for tagging the data dirty are improved for added for point clouds
and improved for curves. A missing tag has also been fixed for mesh
sculpt mode.
**Future**
There are further improvements which can be worked on next
- Apply changes to volume objects and other types where it makes sense
- Continue cleanup changes described in T96968
- Apply shared cache design to more expensive data like triangulation
or normals
Differential Revision: https://developer.blender.org/D16204
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 the conventional way of dealing with unused arguments in C++,
since it works on all compilers.
Regex find and replace: `UNUSED\((\w+)\)` -> `/*$1*/`
The mesh's triangulation cache is often created for other operations
besides the drawing code, but during the mesh draw cache extraction
it is recalculated on every single time. It is simpler and faster to use
the existing MLoopTri array. It can also save memory if the cache
already exists by avoiding allocating a duplicate array. For a 4 million
face quad mesh, that is already 128 MB.
Also use face normals for mesh triangulation if they aren't dirty,
which should provide a general speedup when they're both necessary.
Recently 54182e4925 made this more reliable, since the triangulation
cache is invalidated properly when the mesh is deformed.
Fixes T98073
Differential Revision: https://developer.blender.org/D15550
In `BKE_mesh_tag_coords_changed_uniformly` the checks for dirty vertex
and dirty poly normals were swapped around, causing an assert to be
triggered.
Differential Revision: https://developer.blender.org/D16002
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
We store various lazily calculated caches on meshes, some of which
depend on the vertex positions staying the same. The current API to
invalidate these caches is a bit confusing. With an explicit set of
functions modeled after the functions in `BKE_node_tree_update.h`,
it becomes clear which function to call. This may become more
important if more lazy caches are added in the future.
Differential Revision: https://developer.blender.org/D14760
Continuing the refactors described in T93602, this commit moves
the face dot tag set by the subdivision surface modifier out of
`MVert` to `MeshRuntime`. This clarifies its status as runtime data
and allows further refactoring of mesh positions in the future.
Before, `BKE_modifiers_uses_subsurf_facedots` was used to check
whether subsurf face dots should be drawn, but now we can just check
if the tags exist on the mesh. Modifiers that create new new geometry
or modify topology will already remove the array by clearing mesh
runtime data.
Differential Revision: https://developer.blender.org/D14680
