This repository has been archived on 2023-10-09. You can view files and clone it, but cannot push or open issues or pull requests.
Files
blender-archive/source/blender/nodes/NOD_math_functions.hh

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

473 lines
18 KiB
C++
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
#pragma once
#include "DNA_node_types.h"
#include "BLI_math_base_safe.h"
#include "BLI_math_rotation.h"
#include "BLI_math_vector.hh"
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
#include "BLI_string_ref.hh"
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
#include "FN_multi_function_builder.hh"
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
namespace blender::nodes {
struct FloatMathOperationInfo {
StringRefNull title_case_name;
StringRefNull shader_name;
FloatMathOperationInfo() = delete;
FloatMathOperationInfo(StringRefNull title_case_name, StringRefNull shader_name)
: title_case_name(title_case_name), shader_name(shader_name)
{
}
};
const FloatMathOperationInfo *get_float_math_operation_info(int operation);
const FloatMathOperationInfo *get_float3_math_operation_info(int operation);
const FloatMathOperationInfo *get_float_compare_operation_info(int operation);
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
/**
* This calls the `callback` with two arguments:
* 1. The math function that takes a float as input and outputs a new float.
* 2. A #FloatMathOperationInfo struct reference.
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
* Returns true when the callback has been called, otherwise false.
*
* The math function that is passed to the callback is actually a lambda function that is different
* for every operation. Therefore, if the callback is templated on the math function, it will get
* instantiated for every operation separately. This has two benefits:
* - The compiler can optimize the callback for every operation separately.
* - A static variable declared in the callback will be generated for every operation separately.
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
*
* If separate instantiations are not desired, the callback can also take a function pointer with
* the following signature as input instead: float (*math_function)(float a).
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl_to_fl(const int operation, Callback &&callback)
{
const FloatMathOperationInfo *info = get_float_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
/* This is just an utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
return true;
};
switch (operation) {
case NODE_MATH_EXPONENT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return expf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SQRT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return safe_sqrtf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_INV_SQRT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return safe_inverse_sqrtf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ABSOLUTE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return fabs(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_RADIANS:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return (float)DEG2RAD(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_DEGREES:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return (float)RAD2DEG(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SIGN:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return compatible_signf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ROUND:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return floorf(a + 0.5f); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_FLOOR:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return floorf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_CEIL:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return ceilf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_FRACTION:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return a - floorf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_TRUNC:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a) { return a >= 0.0f ? floorf(a) : ceilf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return sinf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_COSINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return cosf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_TANGENT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return tanf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SINH:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return sinhf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_COSH:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return coshf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_TANH:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return tanhf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ARCSINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return safe_asinf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ARCCOSINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return safe_acosf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ARCTANGENT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a) { return atanf(a); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl_fl_to_fl(const int operation, Callback &&callback)
{
const FloatMathOperationInfo *info = get_float_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
/* This is just an utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
return true;
};
switch (operation) {
case NODE_MATH_ADD:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return a + b; });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SUBTRACT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return a - b; });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_MULTIPLY:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return a * b; });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_DIVIDE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return safe_divide(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_POWER:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a, float b) { return safe_powf(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_LOGARITHM:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a, float b) { return safe_logf(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_MINIMUM:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return std::min(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_MAXIMUM:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return std::max(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_LESS_THAN:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return (float)(a < b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_GREATER_THAN:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return (float)(a > b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_MODULO:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return safe_modf(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SNAP:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float a, float b) { return floorf(safe_divide(a, b)) * b; });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_ARCTAN2:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float a, float b) { return atan2f(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_PINGPONG:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float a, float b) { return pingpongf(a, b); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl_fl_fl_to_fl(const int operation, Callback &&callback)
{
const FloatMathOperationInfo *info = get_float_math_operation_info(operation);
if (info == nullptr) {
return false;
}
/* This is just an utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
return true;
};
switch (operation) {
case NODE_MATH_MULTIPLY_ADD:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(fn::CustomMF_presets::AllSpanOrSingle(),
[](float a, float b, float c) { return a * b + c; });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_COMPARE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(fn::CustomMF_presets::SomeSpanOrSingle<0, 1>(),
[](float a, float b, float c) -> float {
return ((a == b) || (fabsf(a - b) <= fmaxf(c, FLT_EPSILON))) ? 1.0f : 0.0f;
});
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SMOOTH_MIN:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(fn::CustomMF_presets::SomeSpanOrSingle<0, 1>(),
[](float a, float b, float c) { return smoothminf(a, b, c); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_SMOOTH_MAX:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(fn::CustomMF_presets::SomeSpanOrSingle<0, 1>(),
[](float a, float b, float c) { return -smoothminf(-a, -b, c); });
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
case NODE_MATH_WRAP:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(fn::CustomMF_presets::SomeSpanOrSingle<0>(),
[](float a, float b, float c) { return wrapf(a, b, c); });
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_fl3_to_fl3(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_ADD:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return a + b; });
case NODE_VECTOR_MATH_SUBTRACT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return a - b; });
case NODE_VECTOR_MATH_MULTIPLY:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return a * b; });
case NODE_VECTOR_MATH_DIVIDE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return safe_divide(a, b); });
case NODE_VECTOR_MATH_CROSS_PRODUCT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float3 a, float3 b) { return cross_high_precision(a, b); });
case NODE_VECTOR_MATH_PROJECT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return project(a, b); });
case NODE_VECTOR_MATH_REFLECT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float3 a, float3 b) { return reflect(a, normalize(b)); });
case NODE_VECTOR_MATH_SNAP:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float3 a, float3 b) { return floor(safe_divide(a, b)) * b; });
case NODE_VECTOR_MATH_MODULO:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float3 a, float3 b) { return mod(a, b); });
case NODE_VECTOR_MATH_MINIMUM:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return min(a, b); });
case NODE_VECTOR_MATH_MAXIMUM:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return max(a, b); });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_fl3_to_fl(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_DOT_PRODUCT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return dot(a, b); });
case NODE_VECTOR_MATH_DISTANCE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b) { return distance(a, b); });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_fl3_fl3_to_fl3(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_MULTIPLY_ADD:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float3 b, float3 c) { return a * b + c; });
case NODE_VECTOR_MATH_WRAP:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow, [](float3 a, float3 b, float3 c) {
return float3(wrapf(a.x, b.x, c.x), wrapf(a.y, b.y, c.y), wrapf(a.z, b.z, c.z));
});
case NODE_VECTOR_MATH_FACEFORWARD:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float3 a, float3 b, float3 c) { return faceforward(a, b, c); });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_fl3_fl_to_fl3(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_REFRACT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow,
[](float3 a, float3 b, float c) { return refract(a, normalize(b), c); });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_to_fl(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_LENGTH:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 in) { return length(in); });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_fl_to_fl3(const NodeVectorMathOperation operation,
Callback &&callback)
{
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_SCALE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 a, float b) { return a * b; });
default:
return false;
}
return false;
}
/**
* This is similar to try_dispatch_float_math_fl_to_fl, just with a different callback signature.
*/
template<typename Callback>
inline bool try_dispatch_float_math_fl3_to_fl3(const NodeVectorMathOperation operation,
Callback &&callback)
{
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
using namespace blender::math;
const FloatMathOperationInfo *info = get_float3_math_operation_info(operation);
if (info == nullptr) {
return false;
}
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
static auto exec_preset_fast = fn::CustomMF_presets::AllSpanOrSingle();
static auto exec_preset_slow = fn::CustomMF_presets::Materialized();
/* This is just a utility function to keep the individual cases smaller. */
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
auto dispatch = [&](auto exec_preset, auto math_function) -> bool {
callback(exec_preset, math_function, *info);
return true;
};
switch (operation) {
case NODE_VECTOR_MATH_NORMALIZE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast,
[](float3 in) { return normalize(in); }); /* Should be safe. */
case NODE_VECTOR_MATH_FLOOR:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 in) { return floor(in); });
case NODE_VECTOR_MATH_CEIL:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 in) { return ceil(in); });
case NODE_VECTOR_MATH_FRACTION:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 in) { return fract(in); });
case NODE_VECTOR_MATH_ABSOLUTE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_fast, [](float3 in) { return abs(in); });
case NODE_VECTOR_MATH_SINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow,
[](float3 in) { return float3(sinf(in.x), sinf(in.y), sinf(in.z)); });
case NODE_VECTOR_MATH_COSINE:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow,
[](float3 in) { return float3(cosf(in.x), cosf(in.y), cosf(in.z)); });
case NODE_VECTOR_MATH_TANGENT:
Geometry Nodes: refactor array devirtualization Goals: * Better high level control over where devirtualization occurs. There is always a trade-off between performance and compile-time/binary-size. * Simplify using array devirtualization. * Better performance for cases where devirtualization wasn't used before. Many geometry nodes accept fields as inputs. Internally, that means that the execution functions have to accept so called "virtual arrays" as inputs. Those can be e.g. actual arrays, just single values, or lazily computed arrays. Due to these different possible virtual arrays implementations, access to individual elements is slower than it would be if everything was just a normal array (access does through a virtual function call). For more complex execution functions, this overhead does not matter, but for small functions (like a simple addition) it very much does. The virtual function call also prevents the compiler from doing some optimizations (e.g. loop unrolling and inserting simd instructions). The solution is to "devirtualize" the virtual arrays for small functions where the overhead is measurable. Essentially, the function is generated many times with different array types as input. Then there is a run-time dispatch that calls the best implementation. We have been doing devirtualization in e.g. math nodes for a long time already. This patch just generalizes the concept and makes it easier to control. It also makes it easier to investigate the different trade-offs when it comes to devirtualization. Nodes that we've optimized using devirtualization before didn't get a speedup. However, a couple of nodes are using devirtualization now, that didn't before. Those got a 2-4x speedup in common cases. * Map Range * Random Value * Switch * Combine XYZ Differential Revision: https://developer.blender.org/D14628
2022-04-26 17:12:34 +02:00
return dispatch(exec_preset_slow,
[](float3 in) { return float3(tanf(in.x), tanf(in.y), tanf(in.z)); });
default:
return false;
}
return false;
}
Geometry Nodes: initial scattering and geometry nodes This is the initial merge from the geometry-nodes branch. Nodes: * Attribute Math * Boolean * Edge Split * Float Compare * Object Info * Point Distribute * Point Instance * Random Attribute * Random Float * Subdivision Surface * Transform * Triangulate It includes the initial evaluation of geometry node groups in the Geometry Nodes modifier. Notes on the Generic attribute access API The API adds an indirection for attribute access. That has the following benefits: * Most code does not have to care about how an attribute is stored internally. This is mainly necessary, because we have to deal with "legacy" attributes such as vertex weights and attributes that are embedded into other structs such as vertex positions. * When reading from an attribute, we generally don't care what domain the attribute is stored on. So we want to abstract away the interpolation that that adapts attributes from one domain to another domain (this is not actually implemented yet). Other possible improvements for later iterations include: * Actually implement interpolation between domains. * Don't use inheritance for the different attribute types. A single class for read access and one for write access might be enough, because we know all the ways in which attributes are stored internally. We don't want more different internal structures in the future. On the contrary, ideally we can consolidate the different storage formats in the future to reduce the need for this indirection. * Remove the need for heap allocations when creating attribute accessors. It includes commits from: * Dalai Felinto * Hans Goudey * Jacques Lucke * Léo Depoix
2020-12-02 13:25:25 +01:00
} // namespace blender::nodes