Age | Commit message (Collapse) | Author |
|
As described in T91186, this commit moves mesh vertex normals into a
contiguous array of float vectors in a custom data layer, how face
normals are currently stored.
The main interface is documented in `BKE_mesh.h`. Vertex and face
normals are now calculated on-demand and cached, retrieved with an
"ensure" function. Since the logical state of a mesh is now "has
normals when necessary", they can be retrieved from a `const` mesh.
The goal is to use on-demand calculation for all derived data, but
leave room for eager calculation for performance purposes (modifier
evaluation is threaded, but viewport data generation is not).
**Benefits**
This moves us closer to a SoA approach rather than the current AoS
paradigm. Accessing a contiguous `float3` is much more efficient than
retrieving data from a larger struct. The memory requirements for
accessing only normals or vertex locations are smaller, and at the
cost of more memory usage for just normals, they now don't have to
be converted between float and short, which also simplifies code
In the future, the remaining items can be removed from `MVert`,
leaving only `float3`, which has similar benefits (see T93602).
Removing the combination of derived and original data makes it
conceptually simpler to only calculate normals when necessary.
This is especially important now that we have more opportunities
for temporary meshes in geometry nodes.
**Performance**
In addition to the theoretical future performance improvements by
making `MVert == float3`, I've done some basic performance testing
on this patch directly. The data is fairly rough, but it gives an idea
about where things stand generally.
- Mesh line primitive 4m Verts: 1.16x faster (36 -> 31 ms),
showing that accessing just `MVert` is now more efficient.
- Spring Splash Screen: 1.03-1.06 -> 1.06-1.11 FPS, a very slight
change that at least shows there is no regression.
- Sprite Fright Snail Smoosh: 3.30-3.40 -> 3.42-3.50 FPS, a small
but observable speedup.
- Set Position Node with Scaled Normal: 1.36x faster (53 -> 39 ms),
shows that using normals in geometry nodes is faster.
- Normal Calculation 1.6m Vert Cube: 1.19x faster (25 -> 21 ms),
shows that calculating normals is slightly faster now.
- File Size of 1.6m Vert Cube: 1.03x smaller (214.7 -> 208.4 MB),
Normals are not saved in files, which can help with large meshes.
As for memory usage, it may be slightly more in some cases, but
I didn't observe any difference in the production files I tested.
**Tests**
Some modifiers and cycles test results need to be updated with this
commit, for two reasons:
- The subdivision surface modifier is not responsible for calculating
normals anymore. In master, the modifier creates different normals
than the result of the `Mesh` normal calculation, so this is a bug
fix.
- There are small differences in the results of some modifiers that
use normals because they are not converted to and from `short`
anymore.
**Future improvements**
- Remove `ModifierTypeInfo::dependsOnNormals`. Code in each modifier
already retrieves normals if they are needed anyway.
- Copy normals as part of a better CoW system for attributes.
- Make more areas use lazy instead of eager normal calculation.
- Remove `BKE_mesh_normals_tag_dirty` in more places since that is
now the default state of a new mesh.
- Possibly apply a similar change to derived face corner normals.
Differential Revision: https://developer.blender.org/D12770
|
|
There were a couple of function name collisions which were caused
by sharing code with the mask modifier. I just removed the dependence
on the mask modifier now. The code that I duplicated for that purpose
is only in a legacy node, so it can be expected to be removed soonish.
|
|
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
|
|
Includes unwanted changes
This reverts commit 46e049d0ce2bce2f53ddc41a0dbbea2969d00a5d.
|
|
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
|
|
Reverted because the commit removes a lot of commits.
This reverts commit a2c1c368af48644fa8995ecbe7138cc0d7900c30.
|
|
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 @Howard Trickey (howardt) opinion on the matter.
- The noexcept on the copy constructor of mpq(2|3) is being removed.
But according to @Jacques Lucke (JacquesLucke) it is not a real problem
for now.
I would like to give a huge thanks to @Jacques Lucke (JacquesLucke) who
helped during this and pushed me to reduce the duplication further.
Reviewed By: brecht, sergey, JacquesLucke
Differential Revision: http://developer.blender.org/D13791
|
|
MSVC used to warn about const mismatch for arguments passed by value.
Remove these as newer versions of MSVC no longer show this warning.
|
|
|
|
|
|
Currently, a node either supports lazyness during execution (like the Switch
node), or it doesn't. If it does support lazyness, then every input is computed
lazily. However, usually not all inputs actually have to be computed lazily.
E.g. the boolean switch input is always required, while the other inputs
should be computed lazily.
Better support for such sockets can avoid unnecessary round trips through
the node execution function.
|
|
At the time of allocating the buffer with vertices in context, we don't
know exactly how many vertices are affected, but we do know that it is
less than or equal to twice the number of vertices killed.
|
|
The array modifier does not necessarily tag normals dirty.
If it doesnt, normals are recalculated "internally" using the offset ob
transform. This was happening for the array items, but not for the caps.
Now do the same thing for caps.
Maniphest Tasks: T94422
Differential Revision: https://developer.blender.org/D13681
|
|
I had assumed that the span's size was the same as the length variable.
In the future, separate lengths could be removed in favor of using
lengths directly from spans.
|
|
This improves code readability.
Take the opportunity and improve the comments too.
|
|
|
|
This evaluator is used in order to evaluate subdivision at render time, allowing for
faster renders of meshes with a subdivision surface modifier placed at the last
position in the modifier list.
When evaluating the subsurf modifier, we detect whether we can delegate evaluation
to the draw code. If so, the subdivision is first evaluated on the GPU using our own
custom evaluator (only the coarse data needs to be initially sent to the GPU), then,
buffers for the final `MeshBufferCache` are filled on the GPU using a set of
compute shaders. However, some buffers are still filled on the CPU side, if doing so
on the GPU is impractical (e.g. the line adjacency buffer used for x-ray, whose
logic is hardly GPU compatible).
This is done at the mesh buffer extraction level so that the result can be readily used
in the various OpenGL engines, without having to write custom geometry or tesselation
shaders.
We use our own subdivision evaluation shaders, instead of OpenSubDiv's vanilla one, in
order to control the data layout, and interpolation. For example, we store vertex colors
as compressed 16-bit integers, while OpenSubDiv's default evaluator only work for float
types.
In order to still access the modified geometry on the CPU side, for use in modifiers
or transform operators, a dedicated wrapper type is added `MESH_WRAPPER_TYPE_SUBD`.
Subdivision will be lazily evaluated via `BKE_object_get_evaluated_mesh` which will
create such a wrapper if possible. If the final subdivision surface is not needed on
the CPU side, `BKE_object_get_evaluated_mesh_no_subsurf` should be used.
Enabling or disabling GPU subdivision can be done through the user preferences (under
Viewport -> Subdivision).
See patch description for benchmarks.
Reviewed By: campbellbarton, jbakker, fclem, brecht, #eevee_viewport
Differential Revision: https://developer.blender.org/D12406
|
|
This adds `blender::is_same_any_v` which is the almost the same as
`std::is_same_v`. The difference is that it allows for checking multiple
types at the same time.
Differential Revision: https://developer.blender.org/D13673
|
|
The crash happens when opening a panel (added in rB43f5e761a66e87fed664a199cda867639f8daf3e)
when no CacheFile is set in the modifier.
To fix this, check that the CacheFile pointer is not null before attempting to draw anything.
|
|
This adds interface panels to organize the Cache File UI parameters for
modifiers and constraints into related components: velocity, time, and
render procedural.
Properties relating to the three aforementioned components are separated
from `uiTemplateCacheFile` into their own functions (e.g.
`uiTemplateCacheFileVelocity` for the velocity one), which are in turn
called from the specific panel creation routines of the modifiers and
constraints (for constraints, the functions are exposed to the RNA).
`uiTemplateCacheFile` now only shows the properties for the file path,
and in the case of constraints, the scale property.
The properties that are only defined per modifier (like the velocity
scale), are shown in the proper modifier layout panel if applicable.
Reviewed By: sybren
Differential Revision: https://developer.blender.org/D13652
|
|
Previously, the values passed to a multi-input socket were stored
in the order that they arrived in. Then, when the values are accessed,
they are sorted depending on the link order.
Now, the ordering is determined in the beginning before execution starts.
Every value is assigned to the right index directly, avoiding the sort
in the end. This makes the ordering more explicit.
|
|
Goals of this refactor:
* More unified approach to updating everything that needs to be updated
after a change in a node tree.
* The updates should happen in the correct order and quadratic or worse
algorithms should be avoided.
* Improve detection of changes to the output to avoid tagging the depsgraph
when it's not necessary.
* Move towards a more declarative style of defining nodes by having a
more centralized update procedure.
The refactor consists of two main parts:
* Node tree tagging and update refactor.
* Generally, when changes are done to a node tree, it is tagged dirty
until a global update function is called that updates everything in
the correct order.
* The tagging is more fine-grained compared to before, to allow for more
precise depsgraph update tagging.
* Depsgraph changes.
* The shading specific depsgraph node for node trees as been removed.
* Instead, there is a new `NTREE_OUTPUT` depsgrap node, which is only
tagged when the output of the node tree changed (e.g. the Group Output
or Material Output node).
* The copy-on-write relation from node trees to the data block they are
embedded in is now non-flushing. This avoids e.g. triggering a material
update after the shader node tree changed in unrelated ways. Instead
the material has a flushing relation to the new `NTREE_OUTPUT` node now.
* The depsgraph no longer reports data block changes through to cycles
through `Depsgraph.updates` when only the node tree changed in ways
that do not affect the output.
Avoiding unnecessary updates seems to work well for geometry nodes and cycles.
The situation is a bit worse when there are drivers on the node tree, but that
could potentially be improved separately in the future.
Avoiding updates in eevee and the compositor is more tricky, but also less urgent.
* Eevee updates are triggered by calling `DRW_notify_view_update` in
`ED_render_view3d_update` indirectly from `DEG_editors_update`.
* Compositor updates are triggered by `ED_node_composite_job` in `node_area_refresh`.
This is triggered by calling `ED_area_tag_refresh` in `node_area_listener`.
Removing updates always has the risk of breaking some dependency that no
one was aware of. It's not unlikely that this will happen here as well. Adding
back missing updates should be quite a bit easier than getting rid of
unnecessary updates though.
Differential Revision: https://developer.blender.org/D13246
|
|
r7acd3ad7d8e58b913c5 converted a pointer to a reference,
but an assert still compares the variable to a pointer.
|
|
This is a followup to the previous commit.
|
|
In this commit I changed many loops to range-based for loops.
I also removed some of the redundant iterator variables, using
indexing inside the loop instead. Generally an optimizing compiler
should have no problem doing the smartest thing in that situation,
and this way makes it much easier to tell where data is coming from.
I only changed the loops I was confident about, so there is still more
that could be done in the future.
Differential Revision: https://developer.blender.org/D13637
|
|
This moves `MOD_weld.cc` to C++, fixing compiler warnings
coming from the change. It also goes a little bit further and converts
the code to use C++ data structures: `Span`, `Array`, and `Vector`.
This makes the code more shorter and easier to reason about, and
makes memory maneagement more automatic.
Differential Revision: https://developer.blender.org/D13618
|
|
The style guide mentions that unsigned integers shouldn't be used to
show that a value won't be negative. Many places don't follow this
properly yet. The modifier used to cast an array of `uint` to `int` in
order to pass it to `BLI_kdtree_3d_calc_duplicates_fast`. That is no
longer necessary.
Differential Revision: https://developer.blender.org/D13613
|
|
Using the `MEM_*` API from C++ code was a bit annoying:
* When converting C to C++ code, one often has to add a type cast on
returned `void *`. That leads to having the same type name three times
in the same line. This patch reduces the amount to two and removes the
`sizeof(...)` from the line.
* The existing alternative of using `OBJECT_GUARDED_NEW` looks a out
of place compared to other allocation methods. Sometimes
`MEM_CXX_CLASS_ALLOC_FUNCS` can be used when structs are defined
in C++ code. It doesn't look great but it's definitely better. The downside
is that it makes the name of the allocation less useful. That's because
the same name is used for all allocations of a type, independend of
where it is allocated.
This patch introduces three new functions: `MEM_new`, `MEM_cnew` and
`MEM_delete`. These cover the majority of use cases (array allocation is
not covered).
The `OBJECT_GUARDED_*` macros are removed because they are not
needed anymore.
Differential Revision: https://developer.blender.org/D13502
|
|
We cannot depend on node->id being non-null for group nodes.
|
|
Also slightly change naming to avoid camel case.
|
|
This patch refactors the instance-realization code and adds new functionality.
* Named and anonymous attributes are propagated from instances to the
realized geometry. If the same attribute exists on the geometry and on an
instance, the attribute on the geometry has precedence.
* The id attribute has special handling to avoid creating the same id on many
output points. This is necessary to make e.g. the Random Value node work
as expected afterwards.
Realizing instance attributes has an effect on existing files, especially due to the
id attribute. To avoid breaking existing files, the Realize Instances node now has
a legacy option that is enabled for all already existing Realize Instances nodes.
Removing this legacy behavior does affect some existing files (although not many).
We can decide whether it's worth to remove the old behavior as a separate step.
This refactor also improves performance when realizing instances. That is mainly
due to multi-threading. See D13446 to get the file used for benchmarking. The
curve code is not as optimized as it could be yet. That's mainly because the storage
for these attributes might change soonish and it wasn't worth optimizing for the
current storage format right now.
```
1,000,000 x mesh vertex: 530 ms -> 130 ms
1,000,000 x simple cube: 1290 ms -> 190 ms
1,000,000 x point: 1000 ms -> 150 ms
1,000,000 x curve spiral: 1740 ms -> 330 ms
1,000,000 x curve line: 1110 ms -> 210 ms
10,000 x subdivided cylinder: 170 ms -> 40 ms
10 x subdivided spiral: 180 ms -> 180 ms
```
Differential Revision: https://developer.blender.org/D13446
|
|
Also add groups in some files.
|
|
This node outputs the current scene time in seconds or in frames.
Use of this node eliminates the need to use drivers to control values
in the node tree that are driven by the scene time.
Frame is a float value to provide for subframe rendering for motion
blur.
Differential Revision: https://developer.blender.org/D13455
|
|
Ref T92709
|
|
The type conversions do not depend on other files in the nodes
module. Furthermore we want to use the conversions in the
geometry module without creating a dependency to the
nodes module there.
|
|
Caused by {rB3b6ee8cee708}
Above commit was trying to get the vertexgroup from the mesh that is
passed into `deformVertsEM` (but that can be NULL).
When can it be NULL, when is is non-NULL?
`editbmesh_calc_modifiers` only passes in a non-NULL mesh to
`deformVertsEM` under certain conditions:
- a non-deform-only modifier is handled currently
- a non-deform-only modifier preceeds the current modifier
- a deform-only modifier preceeds the current modifier (and the current
one depends on normals)
So the passed-in mesh cannot be relied on, now get the vertex group from
the context object data (like it was before the culprit commit).
Related commit: rB8f22feefbc20
Maniphest Tasks: T93611
Differential Revision: https://developer.blender.org/D13487
|
|
When enabling or disabling a Mesh Sequence Cache modifier of an Object
with a hair particle system, the hair would switch positions. This is
caused because original coordinates in Blender are expected to be
normalized, and toggling the modifier would cause the usage of different
orco layers: one that is normalized, and the other which isn't.
This bug exposes a few related issues:
- if the Alembic file did not have orco data,
`MOD_deform_mesh_eval_get`, used by the particle system modifier, would
add an orco layer without normalization
- `MOD_deform_mesh_eval_get` would also ignore the presence of an orco
layer (e.g. one that could have been read from Alembic)
- if the Alembic file did have orco data, the data would be read
unnormalized
To fix those various issues, original coordinates are normalized when
read from Alembic and unnormalized when written to Alembic; and a new
utility function `BKE_mesh_orco_ensure` is added to add a normalized
orco layer if none exists on the mesh already, this function derives
from the code used in the particle system.
Reviewed By: brecht
Maniphest Tasks: T92561
Differential Revision: https://developer.blender.org/D13306
|
|
|
|
|
|
|
|
Differential Revision: https://developer.blender.org/D13384
|
|
|
|
The mirror modifiers merge option caused unnecessary re-ordering
to the vertex array with original vertices merging into their copies.
While this wasn't an error, it meant creating a 1:1 mapping from input
vertices to their final output wasn't reliable (when looping over
vertices first to last) as is done in
BKE_editmesh_vert_coords_when_deformed.
As merging in either direction is supported, keep the source meshes
vertices in-order since it allows the vertex coordinates to be extracted.
NOTE: Since this change introduce issues for some cases (e.g. bound
modifiers like SurfaceDeform), this change is only applied to newly
created modifiers, existing ones will still use the old incorrect merge
behavior.
Reviewed By: @brecht
Maniphest Tasks: T93321, T91444
Differential Revision: https://developer.blender.org/D13355
|
|
When a node is executed, it usually schedules other nodes.
Right now, those newly scheduled nodes are added to a
task pool so that another thread can start working on them
immediatly.
However, that leads to the situation where sometimes each
node in a simple chain is executed by another thread. That
leads to additional threading overhead and reduced cache
efficiency (for caches that are not shared between cores).
Now, when a node is executed and schedules other nodes,
the first of those newly scheduled nodes will always be
executed on the same thread once the current node is done.
If it schedules more than one other node, those will be
added to the task pool as before.
The speedup achieved by this is hard to measure. I found it
to be a couple percent faster in some extreme cases, not
much to get excited about. It's nice though that the number
of tasks added to the task pool is commonly reduced by a
factor of 4 or 5.
|
|
This is only meant to be used for development purposes for now,
not to show warnings to the user.
Differential Revision: https://developer.blender.org/D13348
|
|
Previously, there were a couple of cases where nodes were scheduled when
that was not really necessary. This change doesn't seem to have a big impact
on performance, but simplifies the code a bit.
|
|
|
|
|
|
|
|
Differential Revision: https://developer.blender.org/D13340
|