2022-02-11 09:07:11 +11:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0-or-later */
|
2020-12-02 13:25:25 +01:00
|
|
|
|
|
|
|
|
#pragma once
|
|
|
|
|
|
|
|
|
|
#include <mutex>
|
|
|
|
|
|
2021-09-09 12:54:20 +02:00
|
|
|
#include "BKE_anonymous_attribute.hh"
|
2020-12-02 13:25:25 +01:00
|
|
|
#include "BKE_attribute.h"
|
|
|
|
|
|
2020-12-03 16:25:48 +01:00
|
|
|
#include "BLI_color.hh"
|
2021-05-14 15:53:43 -05:00
|
|
|
#include "BLI_function_ref.hh"
|
2022-03-19 08:26:29 +01:00
|
|
|
#include "BLI_generic_span.hh"
|
|
|
|
|
#include "BLI_generic_virtual_array.hh"
|
BLI: Refactor vector types & functions to use templates
This patch implements the vector types (i.e:`float2`) by making heavy
usage of templating. All vector functions are now outside of the vector
classes (inside the `blender::math` namespace) and are not vector size
dependent for the most part.
In the ongoing effort to make shaders less GL centric, we are aiming
to share more code between GLSL and C++ to avoid code duplication.
####Motivations:
- We are aiming to share UBO and SSBO structures between GLSL and C++.
This means we will use many of the existing vector types and others
we currently don't have (uintX, intX). All these variations were
asking for many more code duplication.
- Deduplicate existing code which is duplicated for each vector size.
- We also want to share small functions. Which means that vector
functions should be static and not in the class namespace.
- Reduce friction to use these types in new projects due to their
incompleteness.
- The current state of the `BLI_(float|double|mpq)(2|3|4).hh` is a
bit of a let down. Most clases are incomplete, out of sync with each
others with different codestyles, and some functions that should be
static are not (i.e: `float3::reflect()`).
####Upsides:
- Still support `.x, .y, .z, .w` for readability.
- Compact, readable and easilly extendable.
- All of the vector functions are available for all the vectors types
and can be restricted to certain types. Also template specialization
let us define exception for special class (like mpq).
- With optimization ON, the compiler unroll the loops and performance
is the same.
####Downsides:
- Might impact debugability. Though I would arge that the bugs are
rarelly caused by the vector class itself (since the operations are
quite trivial) but by the type conversions.
- Might impact compile time. I did not saw a significant impact since
the usage is not really widespread.
- Functions needs to be rewritten to support arbitrary vector length.
For instance, one can't call `len_squared_v3v3` in
`math::length_squared()` and call it a day.
- Type cast does not work with the template version of the `math::`
vector functions. Meaning you need to manually cast `float *` and
`(float *)[3]` to `float3` for the function calls.
i.e: `math::distance_squared(float3(nearest.co), positions[i]);`
- Some parts might loose in readability:
`float3::dot(v1.normalized(), v2.normalized())`
becoming
`math::dot(math::normalize(v1), math::normalize(v2))`
But I propose, when appropriate, to use
`using namespace blender::math;` on function local or file scope to
increase readability.
`dot(normalize(v1), normalize(v2))`
####Consideration:
- Include back `.length()` method. It is quite handy and is more C++
oriented.
- I considered the GLM library as a candidate for replacement. It felt
like too much for what we need and would be difficult to extend / modify
to our needs.
- I used Macros to reduce code in operators declaration and potential
copy paste bugs. This could reduce debugability and could be reverted.
- This touches `delaunay_2d.cc` and the intersection code. I would like
to know @howardt opinion on the matter.
- The `noexcept` on the copy constructor of `mpq(2|3)` is being removed.
But according to @JacquesLucke it is not a real problem for now.
I would like to give a huge thanks to @JacquesLucke who helped during this
and pushed me to reduce the duplication further.
Reviewed By: brecht, sergey, JacquesLucke
Differential Revision: https://developer.blender.org/D13791
2022-01-12 12:57:07 +01:00
|
|
|
#include "BLI_math_vec_types.hh"
|
2021-05-14 15:53:43 -05:00
|
|
|
|
2021-12-25 15:08:38 -06:00
|
|
|
/**
|
|
|
|
|
* This file defines classes that help to provide access to attribute data on a #GeometryComponent.
|
|
|
|
|
* The API for retrieving attributes is defined in `BKE_geometry_set.hh`, but this comment has some
|
|
|
|
|
* general comments about the system.
|
|
|
|
|
*
|
|
|
|
|
* Attributes are stored in geometry data, though they can also be stored in instances. Their
|
|
|
|
|
* storage is often tied to `CustomData`, which is a system to store "layers" of data with specific
|
|
|
|
|
* types and names. However, since `CustomData` was added to Blender before attributes were
|
|
|
|
|
* conceptualized, it combines the "legacy" style of task-specific attribute types with generic
|
|
|
|
|
* types like "Float". The attribute API here only provides access to generic types.
|
|
|
|
|
*
|
|
|
|
|
* Attributes are retrieved from geometry components by providing an "id" (#AttributeIDRef). This
|
|
|
|
|
* is most commonly just an attribute name. The attribute API on geometry components has some more
|
|
|
|
|
* advanced capabilities:
|
|
|
|
|
* 1. Read-only access: With a `const` geometry component, an attribute on the geometry cannot be
|
|
|
|
|
* modified, so the `for_write` and `for_output` versions of the API are not available. This is
|
|
|
|
|
* extremely important for writing coherent bug-free code. When an attribute is retrieved with
|
|
|
|
|
* write access, via #WriteAttributeLookup or #OutputAttribute, the geometry component must be
|
|
|
|
|
* tagged to clear caches that depend on the changed data.
|
2022-06-01 14:38:06 +10:00
|
|
|
* 2. Domain interpolation: When retrieving an attribute, a domain (#eAttrDomain) can be
|
2021-12-25 15:08:38 -06:00
|
|
|
* provided. If the attribute is stored on a different domain and conversion is possible, a
|
|
|
|
|
* version of the data interpolated to the requested domain will be provided. These conversions
|
|
|
|
|
* are implemented in each #GeometryComponent by `attribute_try_adapt_domain_impl`.
|
|
|
|
|
* 3. Implicit type conversion: In addition to interpolating domains, attribute types can be
|
|
|
|
|
* converted, using the conversions in `BKE_type_conversions.hh`. The #VArray / #GVArray system
|
|
|
|
|
* makes it possible to only convert necessary indices on-demand.
|
|
|
|
|
* 4. Anonymous attributes: The "id" used to look up an attribute can also be an anonymous
|
|
|
|
|
* attribute reference. Currently anonymous attributes are only used in geometry nodes.
|
|
|
|
|
* 5. Abstracted storage: Since the data returned from the API is usually a virtual array,
|
|
|
|
|
* it doesn't have to be stored contiguously (even though that is generally preferred). This
|
|
|
|
|
* allows accessing "legacy" attributes like `material_index`, which is stored in `MPoly`.
|
|
|
|
|
*/
|
|
|
|
|
|
2021-09-09 12:54:20 +02:00
|
|
|
namespace blender::bke {
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Identifies an attribute that is either named or anonymous.
|
|
|
|
|
* It does not own the identifier, so it is just a reference.
|
|
|
|
|
*/
|
|
|
|
|
class AttributeIDRef {
|
|
|
|
|
private:
|
|
|
|
|
StringRef name_;
|
|
|
|
|
const AnonymousAttributeID *anonymous_id_ = nullptr;
|
|
|
|
|
|
|
|
|
|
public:
|
2021-10-03 14:39:17 +02:00
|
|
|
AttributeIDRef();
|
|
|
|
|
AttributeIDRef(StringRef name);
|
|
|
|
|
AttributeIDRef(StringRefNull name);
|
|
|
|
|
AttributeIDRef(const char *name);
|
|
|
|
|
AttributeIDRef(const std::string &name);
|
|
|
|
|
AttributeIDRef(const AnonymousAttributeID *anonymous_id);
|
|
|
|
|
|
|
|
|
|
operator bool() const;
|
|
|
|
|
uint64_t hash() const;
|
|
|
|
|
bool is_named() const;
|
|
|
|
|
bool is_anonymous() const;
|
|
|
|
|
StringRef name() const;
|
|
|
|
|
const AnonymousAttributeID &anonymous_id() const;
|
2021-10-20 09:57:54 -05:00
|
|
|
bool should_be_kept() const;
|
2021-10-03 14:39:17 +02:00
|
|
|
|
|
|
|
|
friend bool operator==(const AttributeIDRef &a, const AttributeIDRef &b);
|
2021-09-23 13:46:13 +02:00
|
|
|
friend std::ostream &operator<<(std::ostream &stream, const AttributeIDRef &attribute_id);
|
2021-09-09 12:54:20 +02:00
|
|
|
};
|
|
|
|
|
|
2022-05-31 13:20:16 +02:00
|
|
|
bool allow_procedural_attribute_access(StringRef attribute_name);
|
|
|
|
|
extern const char *no_procedural_access_message;
|
|
|
|
|
|
2021-09-09 12:54:20 +02:00
|
|
|
} // namespace blender::bke
|
|
|
|
|
|
2021-05-14 15:53:43 -05:00
|
|
|
/**
|
|
|
|
|
* Contains information about an attribute in a geometry component.
|
|
|
|
|
* More information can be added in the future. E.g. whether the attribute is builtin and how it is
|
|
|
|
|
* stored (uv map, vertex group, ...).
|
|
|
|
|
*/
|
|
|
|
|
struct AttributeMetaData {
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain;
|
|
|
|
|
eCustomDataType data_type;
|
2021-05-19 13:22:09 -04:00
|
|
|
|
|
|
|
|
constexpr friend bool operator==(AttributeMetaData a, AttributeMetaData b)
|
|
|
|
|
{
|
|
|
|
|
return (a.domain == b.domain) && (a.data_type == b.data_type);
|
|
|
|
|
}
|
2021-05-14 15:53:43 -05:00
|
|
|
};
|
|
|
|
|
|
2021-09-23 17:50:33 +02:00
|
|
|
struct AttributeKind {
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain;
|
|
|
|
|
eCustomDataType data_type;
|
2021-09-23 17:50:33 +02:00
|
|
|
};
|
|
|
|
|
|
2021-05-14 15:53:43 -05:00
|
|
|
/**
|
2021-05-20 17:55:35 +10:00
|
|
|
* Base class for the attribute initializer types described below.
|
2021-05-14 15:53:43 -05:00
|
|
|
*/
|
|
|
|
|
struct AttributeInit {
|
|
|
|
|
enum class Type {
|
|
|
|
|
Default,
|
|
|
|
|
VArray,
|
|
|
|
|
MoveArray,
|
|
|
|
|
};
|
|
|
|
|
Type type;
|
|
|
|
|
AttributeInit(const Type type) : type(type)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Create an attribute using the default value for the data type.
|
|
|
|
|
* The default values may depend on the attribute provider implementation.
|
|
|
|
|
*/
|
|
|
|
|
struct AttributeInitDefault : public AttributeInit {
|
|
|
|
|
AttributeInitDefault() : AttributeInit(Type::Default)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Create an attribute by copying data from an existing virtual array. The virtual array
|
|
|
|
|
* must have the same type as the newly created attribute.
|
|
|
|
|
*
|
|
|
|
|
* Note that this can be used to fill the new attribute with the default
|
|
|
|
|
*/
|
|
|
|
|
struct AttributeInitVArray : public AttributeInit {
|
2022-03-19 08:26:29 +01:00
|
|
|
blender::GVArray varray;
|
2021-05-14 15:53:43 -05:00
|
|
|
|
2022-03-19 08:26:29 +01:00
|
|
|
AttributeInitVArray(blender::GVArray varray)
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
: AttributeInit(Type::VArray), varray(std::move(varray))
|
2021-05-14 15:53:43 -05:00
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Create an attribute with a by passing ownership of a pre-allocated contiguous array of data.
|
|
|
|
|
* Sometimes data is created before a geometry component is available. In that case, it's
|
|
|
|
|
* preferable to move data directly to the created attribute to avoid a new allocation and a copy.
|
|
|
|
|
*
|
|
|
|
|
* Note that this will only have a benefit for attributes that are stored directly as contiguous
|
|
|
|
|
* arrays, so not for some built-in attributes.
|
|
|
|
|
*
|
|
|
|
|
* The array must be allocated with MEM_*, since `attribute_try_create` will free the array if it
|
|
|
|
|
* can't be used directly, and that is generally how Blender expects custom data to be allocated.
|
|
|
|
|
*/
|
|
|
|
|
struct AttributeInitMove : public AttributeInit {
|
|
|
|
|
void *data = nullptr;
|
|
|
|
|
|
|
|
|
|
AttributeInitMove(void *data) : AttributeInit(Type::MoveArray), data(data)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* Returns false when the iteration should be stopped. */
|
2021-09-09 12:54:20 +02:00
|
|
|
using AttributeForeachCallback = blender::FunctionRef<bool(
|
|
|
|
|
const blender::bke::AttributeIDRef &attribute_id, const AttributeMetaData &meta_data)>;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
|
|
|
|
namespace blender::bke {
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
eCustomDataType attribute_data_type_highest_complexity(Span<eCustomDataType> data_types);
|
2021-12-07 17:19:15 +11:00
|
|
|
/**
|
|
|
|
|
* Domains with a higher "information density" have a higher priority,
|
|
|
|
|
* in order to choose a domain that will not lose data through domain conversion.
|
|
|
|
|
*/
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain attribute_domain_highest_priority(Span<eAttrDomain> domains);
|
2020-12-03 16:20:09 +01:00
|
|
|
|
2020-12-02 13:25:25 +01:00
|
|
|
/**
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
* Used when looking up a "plain attribute" based on a name for reading from it.
|
2020-12-02 13:25:25 +01:00
|
|
|
*/
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
struct ReadAttributeLookup {
|
|
|
|
|
/* The virtual array that is used to read from this attribute. */
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
GVArray varray;
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/* Domain the attribute lives on in the geometry. */
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/* Convenience function to check if the attribute has been found. */
|
|
|
|
|
operator bool() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return this->varray;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
};
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/**
|
|
|
|
|
* Used when looking up a "plain attribute" based on a name for reading from it and writing to it.
|
|
|
|
|
*/
|
|
|
|
|
struct WriteAttributeLookup {
|
2021-12-22 16:28:45 -06:00
|
|
|
/** The virtual array that is used to read from and write to the attribute. */
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
GVMutableArray varray;
|
2021-12-22 16:28:45 -06:00
|
|
|
/** Domain the attributes lives on in the geometry component. */
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain;
|
2021-12-22 16:28:45 -06:00
|
|
|
/**
|
|
|
|
|
* Call this after changing the attribute to invalidate caches that depend on this attribute.
|
|
|
|
|
* \note Do not call this after the component the attribute is from has been destructed.
|
|
|
|
|
*/
|
2021-10-29 09:28:31 +02:00
|
|
|
std::function<void()> tag_modified_fn;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/* Convenience function to check if the attribute has been found. */
|
|
|
|
|
operator bool() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return this->varray;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
};
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/**
|
|
|
|
|
* An output attribute allows writing to an attribute (and optionally reading as well). It adds
|
|
|
|
|
* some convenience features on top of `GVMutableArray` that are very commonly used.
|
|
|
|
|
*
|
|
|
|
|
* Supported convenience features:
|
|
|
|
|
* - Implicit type conversion when writing to builtin attributes.
|
|
|
|
|
* - Supports simple access to a span containing the attribute values (that avoids the use of
|
|
|
|
|
* VMutableArray_Span in many cases).
|
|
|
|
|
* - An output attribute can live side by side with an existing attribute with a different domain
|
|
|
|
|
* or data type. The old attribute will only be overwritten when the #save function is called.
|
2021-12-22 16:28:45 -06:00
|
|
|
*
|
2022-02-23 18:24:08 +11:00
|
|
|
* \note The lifetime of an output attribute should not be longer than the lifetime of the
|
2021-12-22 16:28:45 -06:00
|
|
|
* geometry component it comes from, since it can keep a reference to the component for use in
|
|
|
|
|
* the #save method.
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
*/
|
|
|
|
|
class OutputAttribute {
|
|
|
|
|
public:
|
|
|
|
|
using SaveFn = std::function<void(OutputAttribute &)>;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
private:
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
GVMutableArray varray_;
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain_ = ATTR_DOMAIN_AUTO;
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
SaveFn save_;
|
2022-03-19 08:26:29 +01:00
|
|
|
std::unique_ptr<GVMutableArray_GSpan> optional_span_varray_;
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
bool ignore_old_values_ = false;
|
2021-04-23 12:44:59 +02:00
|
|
|
bool save_has_been_called_ = false;
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
|
|
|
|
|
public:
|
2021-10-03 14:39:17 +02:00
|
|
|
OutputAttribute();
|
|
|
|
|
OutputAttribute(OutputAttribute &&other);
|
2022-06-01 14:38:06 +10:00
|
|
|
OutputAttribute(GVMutableArray varray, eAttrDomain domain, SaveFn save, bool ignore_old_values);
|
2021-04-23 12:44:59 +02:00
|
|
|
|
|
|
|
|
~OutputAttribute();
|
|
|
|
|
|
2021-10-03 14:39:17 +02:00
|
|
|
operator bool() const;
|
2021-01-13 12:27:16 +01:00
|
|
|
|
2021-10-03 14:39:17 +02:00
|
|
|
GVMutableArray &operator*();
|
2022-03-19 08:26:29 +01:00
|
|
|
GVMutableArray *operator->();
|
2021-10-03 14:39:17 +02:00
|
|
|
GVMutableArray &varray();
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain() const;
|
2021-10-03 14:39:17 +02:00
|
|
|
const CPPType &cpp_type() const;
|
2022-06-01 14:38:06 +10:00
|
|
|
eCustomDataType custom_data_type() const;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
2022-03-19 08:26:29 +01:00
|
|
|
GMutableSpan as_span();
|
2021-10-03 14:39:17 +02:00
|
|
|
template<typename T> MutableSpan<T> as_span();
|
2021-01-13 12:27:16 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
void save();
|
2020-12-02 13:25:25 +01:00
|
|
|
};
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
/**
|
|
|
|
|
* Same as OutputAttribute, but should be used when the data type is known at compile time.
|
|
|
|
|
*/
|
|
|
|
|
template<typename T> class OutputAttribute_Typed {
|
2020-12-02 13:25:25 +01:00
|
|
|
private:
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
OutputAttribute attribute_;
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
VMutableArray<T> varray_;
|
2020-12-02 13:25:25 +01:00
|
|
|
|
|
|
|
|
public:
|
2021-10-03 16:30:14 +02:00
|
|
|
OutputAttribute_Typed();
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
OutputAttribute_Typed(OutputAttribute attribute) : attribute_(std::move(attribute))
|
2021-01-13 10:48:39 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
if (attribute_) {
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
varray_ = attribute_.varray().template typed<T>();
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
}
|
2021-01-13 10:48:39 +01:00
|
|
|
}
|
|
|
|
|
|
2021-10-03 16:30:14 +02:00
|
|
|
OutputAttribute_Typed(OutputAttribute_Typed &&other);
|
|
|
|
|
~OutputAttribute_Typed();
|
2021-10-03 14:49:15 +02:00
|
|
|
|
2021-10-03 15:01:02 +02:00
|
|
|
OutputAttribute_Typed &operator=(OutputAttribute_Typed &&other)
|
|
|
|
|
{
|
|
|
|
|
if (this == &other) {
|
|
|
|
|
return *this;
|
|
|
|
|
}
|
|
|
|
|
this->~OutputAttribute_Typed();
|
|
|
|
|
new (this) OutputAttribute_Typed(std::move(other));
|
|
|
|
|
return *this;
|
|
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
operator bool() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
VMutableArray<T> &operator*()
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
VMutableArray<T> *operator->()
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return &varray_;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
VMutableArray<T> &varray()
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
return attribute_.domain();
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
const CPPType &cpp_type() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
return CPPType::get<T>();
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
eCustomDataType custom_data_type() const
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
return cpp_type_to_custom_data_type(this->cpp_type());
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
MutableSpan<T> as_span()
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
return attribute_.as_span<T>();
|
2021-01-12 12:49:34 +01:00
|
|
|
}
|
2020-12-02 13:25:25 +01:00
|
|
|
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
void save()
|
2020-12-02 13:25:25 +01:00
|
|
|
{
|
Geometry Nodes: use virtual arrays in internal attribute api
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
2021-04-17 16:41:03 +02:00
|
|
|
attribute_.save();
|
2020-12-02 13:25:25 +01:00
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
2021-10-03 16:30:14 +02:00
|
|
|
/* These are not defined in the class directly, because when defining them there, the external
|
|
|
|
|
* template instantiation does not work, resulting in longer compile times. */
|
|
|
|
|
template<typename T> inline OutputAttribute_Typed<T>::OutputAttribute_Typed() = default;
|
|
|
|
|
template<typename T>
|
|
|
|
|
inline OutputAttribute_Typed<T>::OutputAttribute_Typed(OutputAttribute_Typed &&other) = default;
|
|
|
|
|
template<typename T> inline OutputAttribute_Typed<T>::~OutputAttribute_Typed() = default;
|
|
|
|
|
|
2021-05-19 13:22:09 -04:00
|
|
|
/**
|
|
|
|
|
* A basic container around DNA CustomData so that its users
|
|
|
|
|
* don't have to implement special copy and move constructors.
|
|
|
|
|
*/
|
|
|
|
|
class CustomDataAttributes {
|
|
|
|
|
/**
|
|
|
|
|
* #CustomData needs a size to be freed, and unfortunately it isn't stored in the struct
|
|
|
|
|
* itself, so keep track of the size here so this class can implement its own destructor.
|
|
|
|
|
* If the implementation of the attribute storage changes, this could be removed.
|
|
|
|
|
*/
|
|
|
|
|
int size_;
|
|
|
|
|
|
|
|
|
|
public:
|
|
|
|
|
CustomData data;
|
|
|
|
|
|
|
|
|
|
CustomDataAttributes();
|
|
|
|
|
~CustomDataAttributes();
|
|
|
|
|
CustomDataAttributes(const CustomDataAttributes &other);
|
|
|
|
|
CustomDataAttributes(CustomDataAttributes &&other);
|
2021-06-09 15:54:26 -05:00
|
|
|
CustomDataAttributes &operator=(const CustomDataAttributes &other);
|
2021-05-19 13:22:09 -04:00
|
|
|
|
2022-01-07 11:38:08 +11:00
|
|
|
void reallocate(int size);
|
2021-05-19 13:22:09 -04:00
|
|
|
|
2021-11-30 10:59:11 -05:00
|
|
|
void clear();
|
|
|
|
|
|
2022-03-19 08:26:29 +01:00
|
|
|
std::optional<blender::GSpan> get_for_read(const AttributeIDRef &attribute_id) const;
|
2021-06-02 08:24:42 -04:00
|
|
|
|
2021-12-07 17:19:15 +11:00
|
|
|
/**
|
|
|
|
|
* Return a virtual array for a stored attribute, or a single value virtual array with the
|
|
|
|
|
* default value if the attribute doesn't exist. If no default value is provided, the default
|
|
|
|
|
* value for the type will be used.
|
|
|
|
|
*/
|
2022-03-19 08:26:29 +01:00
|
|
|
blender::GVArray get_for_read(const AttributeIDRef &attribute_id,
|
2022-06-01 15:06:21 +10:00
|
|
|
eCustomDataType data_type,
|
2022-03-19 08:26:29 +01:00
|
|
|
const void *default_value) const;
|
2021-06-02 08:24:42 -04:00
|
|
|
|
|
|
|
|
template<typename T>
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
blender::VArray<T> get_for_read(const AttributeIDRef &attribute_id, const T &default_value) const
|
2021-06-02 08:24:42 -04:00
|
|
|
{
|
2022-03-18 10:57:45 +01:00
|
|
|
const blender::CPPType &cpp_type = blender::CPPType::get<T>();
|
2022-06-01 14:38:06 +10:00
|
|
|
const eCustomDataType type = blender::bke::cpp_type_to_custom_data_type(cpp_type);
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
GVArray varray = this->get_for_read(attribute_id, type, &default_value);
|
|
|
|
|
return varray.typed<T>();
|
2021-06-02 08:24:42 -04:00
|
|
|
}
|
|
|
|
|
|
2022-03-19 08:26:29 +01:00
|
|
|
std::optional<blender::GMutableSpan> get_for_write(const AttributeIDRef &attribute_id);
|
2022-06-01 15:06:21 +10:00
|
|
|
bool create(const AttributeIDRef &attribute_id, eCustomDataType data_type);
|
|
|
|
|
bool create_by_move(const AttributeIDRef &attribute_id, eCustomDataType data_type, void *buffer);
|
2021-09-09 12:54:20 +02:00
|
|
|
bool remove(const AttributeIDRef &attribute_id);
|
2021-05-19 13:22:09 -04:00
|
|
|
|
2021-11-02 15:16:52 -05:00
|
|
|
/**
|
|
|
|
|
* Change the order of the attributes to match the order of IDs in the argument.
|
|
|
|
|
*/
|
|
|
|
|
void reorder(Span<AttributeIDRef> new_order);
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
bool foreach_attribute(const AttributeForeachCallback callback, eAttrDomain domain) const;
|
2021-05-19 13:22:09 -04:00
|
|
|
};
|
|
|
|
|
|
2021-10-05 11:10:25 +11:00
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name #AttributeIDRef Inline Methods
|
|
|
|
|
* \{ */
|
2021-10-03 14:39:17 +02:00
|
|
|
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef() = default;
|
|
|
|
|
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef(StringRef name) : name_(name)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef(StringRefNull name) : name_(name)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef(const char *name) : name_(name)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef(const std::string &name) : name_(name)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* The anonymous id is only borrowed, the caller has to keep a reference to it. */
|
|
|
|
|
inline AttributeIDRef::AttributeIDRef(const AnonymousAttributeID *anonymous_id)
|
|
|
|
|
: anonymous_id_(anonymous_id)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline bool operator==(const AttributeIDRef &a, const AttributeIDRef &b)
|
|
|
|
|
{
|
|
|
|
|
return a.anonymous_id_ == b.anonymous_id_ && a.name_ == b.name_;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline AttributeIDRef::operator bool() const
|
|
|
|
|
{
|
|
|
|
|
return this->is_named() || this->is_anonymous();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline uint64_t AttributeIDRef::hash() const
|
|
|
|
|
{
|
|
|
|
|
return get_default_hash_2(name_, anonymous_id_);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline bool AttributeIDRef::is_named() const
|
|
|
|
|
{
|
|
|
|
|
return !name_.is_empty();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline bool AttributeIDRef::is_anonymous() const
|
|
|
|
|
{
|
|
|
|
|
return anonymous_id_ != nullptr;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline StringRef AttributeIDRef::name() const
|
|
|
|
|
{
|
|
|
|
|
BLI_assert(this->is_named());
|
|
|
|
|
return name_;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline const AnonymousAttributeID &AttributeIDRef::anonymous_id() const
|
|
|
|
|
{
|
|
|
|
|
BLI_assert(this->is_anonymous());
|
|
|
|
|
return *anonymous_id_;
|
|
|
|
|
}
|
|
|
|
|
|
2021-10-20 09:57:54 -05:00
|
|
|
/**
|
|
|
|
|
* \return True if the attribute should not be removed automatically as an optimization during
|
|
|
|
|
* processing or copying. Anonymous attributes can be removed when they no longer have any
|
|
|
|
|
* references.
|
|
|
|
|
*/
|
|
|
|
|
inline bool AttributeIDRef::should_be_kept() const
|
|
|
|
|
{
|
|
|
|
|
return this->is_named() || BKE_anonymous_attribute_id_has_strong_references(anonymous_id_);
|
|
|
|
|
}
|
|
|
|
|
|
2021-10-05 11:10:25 +11:00
|
|
|
/** \} */
|
|
|
|
|
|
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
|
|
|
/** \name #OutputAttribute Inline Methods
|
|
|
|
|
* \{ */
|
2021-10-03 14:39:17 +02:00
|
|
|
|
|
|
|
|
inline OutputAttribute::OutputAttribute() = default;
|
|
|
|
|
inline OutputAttribute::OutputAttribute(OutputAttribute &&other) = default;
|
|
|
|
|
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
inline OutputAttribute::OutputAttribute(GVMutableArray varray,
|
2022-06-01 14:38:06 +10:00
|
|
|
eAttrDomain domain,
|
2021-10-03 14:39:17 +02:00
|
|
|
SaveFn save,
|
|
|
|
|
const bool ignore_old_values)
|
|
|
|
|
: varray_(std::move(varray)),
|
|
|
|
|
domain_(domain),
|
|
|
|
|
save_(std::move(save)),
|
|
|
|
|
ignore_old_values_(ignore_old_values)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline OutputAttribute::operator bool() const
|
|
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2021-10-03 14:39:17 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline GVMutableArray &OutputAttribute::operator*()
|
|
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2021-10-03 14:39:17 +02:00
|
|
|
}
|
|
|
|
|
|
2022-03-19 08:26:29 +01:00
|
|
|
inline GVMutableArray *OutputAttribute::operator->()
|
2021-10-03 14:39:17 +02:00
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return &varray_;
|
2021-10-03 14:39:17 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline GVMutableArray &OutputAttribute::varray()
|
|
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_;
|
2021-10-03 14:39:17 +02:00
|
|
|
}
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
inline eAttrDomain OutputAttribute::domain() const
|
2021-10-03 14:39:17 +02:00
|
|
|
{
|
|
|
|
|
return domain_;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
inline const CPPType &OutputAttribute::cpp_type() const
|
|
|
|
|
{
|
Geometry Nodes: refactor virtual array system
Goals of this refactor:
* Simplify creating virtual arrays.
* Simplify passing virtual arrays around.
* Simplify converting between typed and generic virtual arrays.
* Reduce memory allocations.
As a quick reminder, a virtual arrays is a data structure that behaves like an
array (i.e. it can be accessed using an index). However, it may not actually
be stored as array internally. The two most important implementations
of virtual arrays are those that correspond to an actual plain array and those
that have the same value for every index. However, many more
implementations exist for various reasons (interfacing with legacy attributes,
unified iterator over all points in multiple splines, ...).
With this refactor the core types (`VArray`, `GVArray`, `VMutableArray` and
`GVMutableArray`) can be used like "normal values". They typically live
on the stack. Before, they were usually inside a `std::unique_ptr`. This makes
passing them around much easier. Creation of new virtual arrays is also
much simpler now due to some constructors. Memory allocations are
reduced by making use of small object optimization inside the core types.
Previously, `VArray` was a class with virtual methods that had to be overridden
to change the behavior of a the virtual array. Now,`VArray` has a fixed size
and has no virtual methods. Instead it contains a `VArrayImpl` that is
similar to the old `VArray`. `VArrayImpl` should rarely ever be used directly,
unless a new virtual array implementation is added.
To support the small object optimization for many `VArrayImpl` classes,
a new `blender::Any` type is added. It is similar to `std::any` with two
additional features. It has an adjustable inline buffer size and alignment.
The inline buffer size of `std::any` can't be relied on and is usually too
small for our use case here. Furthermore, `blender::Any` can store
additional user-defined type information without increasing the
stack size.
Differential Revision: https://developer.blender.org/D12986
2021-11-16 10:15:51 +01:00
|
|
|
return varray_.type();
|
2021-10-03 14:39:17 +02:00
|
|
|
}
|
|
|
|
|
|
2022-06-01 14:38:06 +10:00
|
|
|
inline eCustomDataType OutputAttribute::custom_data_type() const
|
2021-10-03 14:39:17 +02:00
|
|
|
{
|
|
|
|
|
return cpp_type_to_custom_data_type(this->cpp_type());
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename T> inline MutableSpan<T> OutputAttribute::as_span()
|
|
|
|
|
{
|
|
|
|
|
return this->as_span().typed<T>();
|
|
|
|
|
}
|
|
|
|
|
|
2021-10-05 11:10:25 +11:00
|
|
|
/** \} */
|
|
|
|
|
|
2020-12-02 13:25:25 +01:00
|
|
|
} // namespace blender::bke
|