The spline `length` function assumed that the curve always had evaluated
edges. That is clearly false. This commit adds a check to `length` and a
special case for a single point in the curve resample node.
This commit uses two changes to improve the performance of the point
instance node.
**Prevent Reallocations**
At 64 bytes, the transform matrix for every instance is rather large,
so reallocating the vector as it grows can become a performance bottle-
neck. This commit reserves memory for the instances that will be added
to prevent unecessary reallocations as the instance vector grows.
In a test with 4 million instances of 3 objects in a collection, the
node was about 40% faster, from 370ms to 270ms for the node.
**Parallelization**
Currently the instances are added by appending to a vector. By changing
this slightly to fill indices instead, we can parallelize the operation
so that multiple threads can fill data at the same time. Tested on a
Ryzen 3700x, this reduced the runtime from the above 270ms to 44ms
average, bringing the total speedup to ~8x.
Note that displaying the instances in the viewport is still much slower
than the calculations in node, this change doesn't affect that.
This patch refactors the instance component to make use of the earlier
refactoring in rB4599cea15dcf. Now we don't have to build an array of
instance references the size of the point domain, and we can gather the
possible instances only once and use the same vector for all component
types. Generally the node should be a bit faster and use less memory.
The logic is moved around a bit, especially the hashing of the ID
attribute to pick from the instance list, but the result is unchanged.
Differential Revision: https://developer.blender.org/D11203
Those were mostly just left over from previous work on particle nodes.
They solved the problem of keeping a reference to an object over
multiple frames and in a cache. Currently, we do not have this problem
in geometry nodes, so we can also remove this layer of complexity
for now.
This node generates a naturally parametarized (even length edge) poly
spline version of every spline in the input. There are two modes,
"Count", and "Length". These are similar to the same options for the
line primitive node in end points mode.
I implemented this instead of a "Sample Points" node, because for this
operation it's trivial to keep the result as a curve, which is nice
since it increases flexibility, and because it can make instancing
simpler, i.e. using the transforms of each evaluated point rather than
requiring the construction of a "rotation" attribute.
Differential Revision: https://developer.blender.org/D11173
This adds `parallel_for` to the Attribute Curve Map node to improve performance.
Grain size set to 512.
Reviewed By: HooglyBoogly
Differential Revision: https://developer.blender.org/D11194
This node has the same functionality as the color and vector curve
mapping nodes in the shader editor. Here is works on every value for
the selected attribute, and it can also output a float value. Other
than that, the implementation is quite straightforward-- almost
completely boilerplate code.
Differential Revision: https://developer.blender.org/D10921
The main goal of this refactor is to not store Object/Collection
pointers for every individual instance. Instead instances now
store a handle for the referenced data. The actual Object/Collection
pointers are stored in a new `InstanceReference` class.
This refactor also allows for some better optimizations further down
the line, because one does not have to search through all instances
anymore to find what data is instanced.
Furthermore, this refactor makes it easier to support instancing
`GeometrySet` or any other data that has to be owned by the
`InstancesComponent`.
Differential Revision: https://developer.blender.org/D11125
This patch adds initial curve support to geometry nodes. Currently
there is only one node available, the "Curve to Mesh" node, T87428.
However, the aim of the changes here is larger than just supporting
curve data in nodes-- it also uses the opportunity to add better spline
data structures, intended to replace the existing curve evaluation code.
The curve code in Blender is quite old, and it's generally regarded as
some of the messiest, hardest-to-understand code as well. The classes
in `BKE_spline.hh` aim to be faster, more extensible, and much more
easily understandable. Further explanation can be found in comments in
that file.
Initial builtin spline attributes are supported-- reading and writing
from the `cyclic` and `resolution` attributes works with any of the
attribute nodes. Also, only Z-up normal calculation is implemented
at the moment, and tilts do not apply yet.
**Limitations**
- For now, you must bring curves into the node tree with an "Object
Info" node. Changes to the curve modifier stack will come later.
- Converting to a mesh is necessary to visualize the curve data.
Further progress can be tracked in: T87245
Higher level design document: https://wiki.blender.org/wiki/Modules/Physics_Nodes/Projects/EverythingNodes/CurveNodes
Differential Revision: https://developer.blender.org/D11091
This commit significantly speeds up many of the attribute nodes when
multiple threads are available in linear situations when parallelism
cannot be achieved elsewhere.
See the differential for a table of timing comparisons tested on a
Ryzen 3700x. For an attribute with 4 million elements, the nodes were
about 3 to 9 times faster.
The changes are not exhaustive, other nodes could still be parallelized
in the future. Also, it would be possible to further optimize the grain
size in `parallel_for`, but I'd rather make sure it isn't too small.
I tested some different values, but also relied on intuition--
increasing grain size for less complex operations and vice versa.
Differential Revision: https://developer.blender.org/D11139
Currently there is an "Auto" option for the domain, this commit adds a
similar option for "Auto" data type, that uses the data type from the
target attribute or the source attribute (in that order).
Ref T87347
Differential Revision: https://developer.blender.org/D10932
This adds a callback to bNodeTreeType to check which socket types are
valid for the tree type. Function has been implemented for the normal
tree types, and can be implemented for custom node trees with python,
by adding a `classmethod` to the tree. However, only builtin socket
types are supported.
This is relevant for T87049, but it also has the advantage that it is
now clear which node trees support which sockets. Previously this
was assumed to be known by all developers.
Differential Revision: https://developer.blender.org/D10938
This is a first step towards T87620.
It should not have any functional changes.
Goals of this refactor:
* Move the evaluator out of `MOD_nodes.cc`. That makes it easier to
improve it in isolation.
* Extract core input/out parameter management out of `GeoNodeExecParams`.
Managing this is the responsibility of the evaluator. This separation of
concerns will be useful once we have lazy evaluation of certain inputs/outputs.
Differential Revision: https://developer.blender.org/D11085
This is a first version of an Attribute Transfer node. It only supports two
modes for mapping attributes from one geometry to another for now.
More options are planned for the future.
Ref T87421.
Differential Revision: https://developer.blender.org/D11037
Counts of less than one weren't allowed in end points mode mostly to
avoid a division by zero when calculating the delta. It's trivial to
allow a count of one, so this commit does that, with the point placed
at the start location.
Because we use virtual classes (and for other reasons), we had to do a
small allocation when simply retrieving the data type and domain of an
existing attribute. This happened quite a lot actually-- to determine
these values for result attributes.
This patch adds a simple function to retrieve this meta data without
building the virtual array. This should lower the overhead of every
attribute node, though the difference probably won't be noticible
unless a tree has very many nodes.
Differential Revision: https://developer.blender.org/D11047
This is a first iteration of a switch node. It can only switch between
two inputs values based on a boolean. A more sophisticated switch
node that has an integer selector will probably come later.
Currently, the geometry nodes evaluator does not support lazy evaluation
of individual inputs. Therefore, all inputs will be computed currently.
An improvement to the evaluator will be worked on separately.
Ref: T85374
Differential Revision: https://developer.blender.org/D10460
A virtual array is a data structure that is similar to a normal array
in that its elements can be accessed by an index. However, a virtual
array does not have to be a contiguous array internally. Instead, its
elements can be layed out arbitrarily while element access happens
through a virtual function call. However, the virtual array data
structures are designed so that the virtual function call can be avoided
in cases where it could become a bottleneck.
Most commonly, a virtual array is backed by an actual array/span or
is a single value internally, that is the same for every index.
Besides those, there are many more specialized virtual arrays like the
ones that provides vertex positions based on the `MVert` struct or
vertex group weights.
Not all attributes used by geometry nodes are stored in simple contiguous
arrays. To provide uniform access to all kinds of attributes, the attribute
API has to provide virtual array functionality that hides the implementation
details of attributes.
Before this refactor, the attribute API provided its own virtual array
implementation as part of the `ReadAttribute` and `WriteAttribute` types.
That resulted in unnecessary code duplication with the virtual array system.
Even worse, it bound many algorithms used by geometry nodes to the specifics
of the attribute API, even though they could also use different data sources
(such as data from sockets, default values, later results of expressions, ...).
This refactor removes the `ReadAttribute` and `WriteAttribute` types and
replaces them with `GVArray` and `GVMutableArray` respectively. The `GV`
stands for "generic virtual". The "generic" means that the data type contained
in those virtual arrays is only known at run-time. There are the corresponding
statically typed types `VArray<T>` and `VMutableArray<T>` as well.
No regressions are expected from this refactor. It does come with one
improvement for users. The attribute API can convert the data type
on write now. This is especially useful when writing to builtin attributes
like `material_index` with e.g. the Attribute Math node (which usually
just writes to float attributes, while `material_index` is an integer attribute).
Differential Revision: https://developer.blender.org/D10994
This is especially useful when trying to add a node group instance, e.g. via
drag & drop from the Outliner or Asset Browser.
Previously this would just silently fail, with no information why. This is a
source of confusion, e.g. earlier, it took me a moment to realize I was
dragging a node group into itself, which failed of course.
Blender should always try to help the user with useful error messages.
Adds error messages like: "Nesting a node group inside of itself is not
allowed", "Not a compositor node tree", etc.
Adds a disabled hint return argument to node and node tree polling functions.
On error the hint is reported, or could even be shown in advance (e.g. if
checked via an operator poll option).
Differential Revision: https://developer.blender.org/D10422
Reviewed by: Jacques Lucke
This is a minor change to add some plumbing code
to support custom geo nodes. This is working the
same way as the custom cycles and compositor nodes.
An example add-in is attached to D10784
Reviewed By: JacquesLucke
Differential Revision: http://developer.blender.org/D10784
Where possible, nodes in the "Geometry" category should support all
geometry component types. This adds support for volumes in the
recently added bounding box node, based on functions added in the
previous two commits.
Differential Revision: https://developer.blender.org/D10906
The problem was that you could getting write access to a grid from a
`const Volume *` without breaking const correctness. I encountered this
when working on support for volumes in the bounding box node. For
geometry nodes there is an important distinction between getting data
"for read" and "for write", with the former returning a `const` version
of the data.
Also, for volumes it was necessary to cast away const, since all of
the relevant functions in `volume.cc` didn't have const versions. This
patch adds `const` in these places, distinguising between "for read"
and "for write" versions of functions where necessary.
The downside is that loading and unloading in the global volume cache
needs const write-access to some member variables. I see that as an
inherent problem that comes up with caching that never has a beautiful
solution anyway.
Some of the const-ness could probably be propogated futher in EEVEE
code, but I'll leave that out, since there is another level of caching.
Differential Revision: https://developer.blender.org/D10916
The function name was not very specific, this makes it clearer that it
works on instances rather than only real geometry. Also use `r_`
prefix for the return argument.
During review of D10730 it was discovered that the "uv" name causes
issues for cycles, which uses it as a default internal data name. While
that could be fixed in the future, there was no particular reason to use
"uv" instead of "uv_map", so we use the latter instead here, which
is consistent with the lowercase naming scheme chosen for attributes.
Some of the BMesh primitive operators have a lot of overhead due to the
fact that they use other operators like extrusion, removing doubles,
and rotation, to build each shape rather than filling arrays directly.
Implementing the primitives directly with the Mesh data structure is
much more efficient, since it's simple to fill the result data based
on the known inputs. It also allows also skip the conversion from BMesh
to Mesh after the calculations.
Speed matters more for procedural operations than for the existing
operators accessible from the add menu, since they will be executed
every evaluation rather than once before a destructive workflow.
| Shape | Before (ms) | After (ms) | Speedup (x times faster) |
| -------- | -------------| ------------| -------------------------|
| Cylinder | 1.1676 | 0.31327 | 3.72 |
| Cone | 4.5890 | 0.17762 | 25.8 |
| Sphere | 64213.3 | 13.595 | 4720 |
The downside of this change is that there will be two implementations
for these three primitives, in these nodes and in `bmo_primitive.c`.
One option would be re-implementing the BMesh primitives in terms of
this code, but that would require `BMesh` to depend on `Mesh`, which
is currently avoided. On the other hand, it will be easier to add new
features to these nodes like different fill types for the top and the
bottom of a cylinder, rings for a cylinder, and tagging the output with
boolean attributes. Another factor to consider is that the add mesh
object operator could be implemented with these nodes, just providing
more benefit for a faster implementation.
As a future cleanup, there is room to share code that generates the
"rings" topology between different nodes that generate meshes.
Differential Revision: https://developer.blender.org/D10730
The default insert link callback for nodes was trying to move the
existing link away, since it didn't properly handle multi-input sockets
(though the problem wasn't exposed for the join node). We can basically
skip all of this "moving existing links" for multi-input sockets, unless
we are at the link limit.
