Age | Commit message (Collapse) | Author |
|
This implements an optimization pass for multi-function procedures.
It optimizes memory reuse by moving destruct instructions up.
For more details see the in-code comment.
In very large fields with many short lived intermediate values, this change
can improve performance 3-4x. Furthermore, in such cases, peak memory
consumption is reduced significantly (e.g. 100x lower peak memory usage).
Differential Revision: https://developer.blender.org/D13548
|
|
Previously, there was a fixed grain size for all multi-functions. That was
not sufficient because some functions could benefit a lot from smaller
grain sizes.
This refactors adds a new `MultiFunction::call_auto` method which has the
same effect as just calling `MultiFunction::call` but additionally figures
out how to execute the specific multi-function efficiently. It determines
a good grain size and decides whether the mask indices should be shifted
or not.
Most multi-function evaluations benefit from this, but medium sized work
loads (1000 - 50000 elements) benefit from it the most. Especially when
expensive multi-functions (e.g. noise) is involved. This is because for
smaller work loads, threading is rarely used and for larger work loads
threading worked fine before already.
With this patch, multi-functions can specify execution hints, that allow
the caller to execute it most efficiently. These execution hints still
have to be added to more functions.
Some performance measurements of a field evaluation involving noise and
math nodes, ordered by the number of elements being evaluated:
```
1,000,000: 133 ms -> 120 ms
100,000: 30 ms -> 18 ms
10,000: 20 ms -> 2.7 ms
1,000: 4 ms -> 0.5 ms
100: 0.5 ms -> 0.4 ms
```
|
|
|
|
|
|
Sometimes it's useful to pass around a set of values with a generic
type. The virtual array data structures allow this, but they don't
have logical ownership. My initial use case for this is as a return
type for the functions that interpolate curve attributes to evaluated
points, but a need for this data structure has come up in a few other
places as well. It also reduced the need for templates.
Differential Revision: https://developer.blender.org/D11103
|
|
|
|
This adds a new `ParallelMultiFunction` which wraps another multi-function
and evaluates it with multiple threads. The speeds up field evaluation
quite a bit (the effect is most noticeable when the number of evaluations
and the field is large).
There are still other single-threaded performance bottlenecks in field
evaluation that will need to be solved separately. Most notably here
is the process of copying the computed data into the position attribute
in the Set Position node.
Differential Revision: https://developer.blender.org/D12457
|
|
This implements the initial core framework for fields and anonymous
attributes (also see T91274).
The new functionality is hidden behind the "Geometry Nodes Fields"
feature flag. When enabled in the user preferences, the following
new nodes become available: `Position`, `Index`, `Normal`,
`Set Position` and `Attribute Capture`.
Socket inspection has not been updated to work with fields yet.
Besides these changes at the user level, this patch contains the
ground work for:
* building and evaluating fields at run-time (`FN_fields.hh`) and
* creating and accessing anonymous attributes on geometry
(`BKE_anonymous_attribute.h`).
For evaluating fields we use a new so called multi-function procedure
(`FN_multi_function_procedure.hh`). It allows composing multi-functions
in arbitrary ways and supports efficient evaluation as is required by
fields. See `FN_multi_function_procedure.hh` for more details on how
this evaluation mechanism can be used.
A new `AttributeIDRef` has been added which allows handling named
and anonymous attributes in the same way in many places.
Hans and I worked on this patch together.
Differential Revision: https://developer.blender.org/D12414
|
|
The multi-function network system was able to compose multiple
multi-functions into a new one and to evaluate that efficiently.
This functionality was heavily used by the particle nodes prototype
a year ago. However, since then we only used multi-functions
without the need to compose them in geometry nodes.
The upcoming "fields" in geometry nodes will need a way to
compose multi-functions again. Unfortunately, the code removed
in this commit was not ideal for this different kind of function
composition. I've been working on an alternative that will be added
separately when it becomes needed.
I've had to update all the function nodes, because their interface
depended on the multi-function network data structure a bit.
The actual multi-function implementations are still the same though.
|
|
When a function is executed for many elements (e.g. per point) it is often the case
that some parameters are different for every element and other parameters are
the same (there are some more less common cases). To simplify writing such
functions one can use a "virtual array". This is a data structure that has a value
for every index, but might not be stored as an actual array internally. Instead, it
might be just a single value or is computed on the fly. There are various tradeoffs
involved when using this data structure which are mentioned in `BLI_virtual_array.hh`.
It is called "virtual", because it uses inheritance and virtual methods.
Furthermore, there is a new virtual vector array data structure, which is an array
of vectors. Both these types have corresponding generic variants, which can be used
when the data type is not known at compile time. This is typically the case when
building a somewhat generic execution system. The function system used these virtual
data structures before, but now they are more versatile.
I've done this refactor in preparation for the attribute processor and other features of
geometry nodes. I moved the typed virtual arrays to blenlib, so that they can be used
independent of the function system.
One open question for me is whether all the generic data structures (and `CPPType`)
should be moved to blenlib as well. They are well isolated and don't really contain
any business logic. That can be done later if necessary.
|
|
This does not need to be included everywhere, because it is only
needed in very few translation units that actually define CPPType's.
|
|
|
|
This is the initial merge from the geometry-nodes branch.
Nodes:
* Attribute Math
* Boolean
* Edge Split
* Float Compare
* Object Info
* Point Distribute
* Point Instance
* Random Attribute
* Random Float
* Subdivision Surface
* Transform
* Triangulate
It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier.
Notes on the Generic attribute access API
The API adds an indirection for attribute access. That has the following benefits:
* Most code does not have to care about how an attribute is stored internally.
This is mainly necessary, because we have to deal with "legacy" attributes
such as vertex weights and attributes that are embedded into other structs
such as vertex positions.
* When reading from an attribute, we generally don't care what domain the
attribute is stored on. So we want to abstract away the interpolation that
that adapts attributes from one domain to another domain (this is not
actually implemented yet).
Other possible improvements for later iterations include:
* Actually implement interpolation between domains.
* Don't use inheritance for the different attribute types. A single class for read
access and one for write access might be enough, because we know all the ways
in which attributes are stored internally. We don't want more different internal
structures in the future. On the contrary, ideally we can consolidate the different
storage formats in the future to reduce the need for this indirection.
* Remove the need for heap allocations when creating attribute accessors.
It includes commits from:
* Dalai Felinto
* Hans Goudey
* Jacques Lucke
* Léo Depoix
|
|
This class represents a pointer whose type is only known at runtime.
|
|
|
|
|
|
Instead of depending on static initialization order of globals use
static variables within functions. Those are initialized on first use.
This is every so slighly less efficient, but avoids a full class of problems.
|
|
|
|
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.
|
|
|
|
Sometimes it is convenient to be able to return a reference to some
dummy function.
|
|
This will be used to reference the content of a CustomData structure
in C++ code, that does not need to know who owns the data but only
works with it.
|
|
A multi-function network is a graph data structure, where nodes are
multi-functions (or dummies) and links represent data flow.
New multi-functions can be derived from such a network. For that
one just has to specify two sets of sockets in the network that
represent the inputs and outputs of the new function.
It is possible to do optimizations like constant folding on this
data structure, but that is not implemented in this patch yet.
In a next step, user generated node trees are converted into a
MFNetwork, so that they can be evaluated efficiently for many particles.
This patch also includes some tests that cover the majority of the code.
However, this seems to be the kind of code that is best tested by some
.blend files. Building graph structures in code is possible, but is
not easy to understand afterwards.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D8049
|
|
typing
|
|
This adds the `MultiFunction` type and some smallish utility types that it uses.
A `MultiFunction` encapsulates a function that is optimized for throughput by
always processing many elements at once.
This is an important part of the new particle system, because it allows us to
execute user generated node trees for many particles efficiently.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D8030
|
|
This adds a new `CPPType` that encapsulates information about how to handle
instances of a specific data type. This is necessary for the function evaluation
system, which will be used to evaluate most of the particle node trees.
Furthermore, this adds an `IndexMask` class which offers a surprisingly useful
abstraction over an array containing unsigned integers. It makes two assumptions
about the underlying integer array:
* The integers are in ascending order.
* There are no duplicates.
`IndexMask` will be used to "select" certain particles that will be
processed in a data-oriented way. Sometimes, operations don't have to
be applied to all particles, but only some, those that are in the indexed by
the `IndexMask`. The two limitations imposed by an `IndexMask` allow for
better performance.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D7957
|