The main issue is that the image and image user is not updated correctly
in `rna_ImageUser_update`. `BKE_image_user_frame_calc` does not set the
correct frame, because the image is null. Also `IMA_GPU_REFRESH` is not
set for the same reason.
When gpu materials are first created, it is expected that the frame is set
correctly, and the flag is set if necessary. Therefore, somewhere during
depsgraph evaluation, those have to be updated. The depsgraph node
to do the update existed already. Now there is a new relation so that it is
executed when the node tree changed, not only when the frame changed.
Use the ID.recalc flag to detect when updates after frame-change is
needed. Since comparing the last calculated frame doesn't take undo into
account (see code-comment for details).
`ID_RECALC_AUDIO_SEEK` has been renamed to `ID_RECALC_FRAME_CHANGE`
since this is not only related to audio however internally this flag is
still categorized in `NodeType::AUDIO`.
Reviewed By: sergey
Ref D13942
The issue was happening with a specific file where the ID management
code was not fully copying all modifiers because of the extra check
in the `BKE_object_support_modifier_type_check()`.
While it is arguable that copy-on-write should be a 1:1 copy there is
no real need to maintain the per-modifier pointer to its original.
Use its SessionUUID to perform lookup in the original datablock.
Downside of this approach is that it is a linear lookup instead of
direct pointer access, but the upside is that there is less pointers
to manage and that the file with unsupported modifiers does behave
correct without any asserts.
Differential Revision: https://developer.blender.org/D13993
The modifiers are mapped between original and evaluated objects based on
their session IDs. The pointer to original modifier is no longer needed
for the backup: it remained from the initial implementation which was
rewritten at some point.
This is a preparation for removal of the pointer to original modifier.
Blender would have crashed when renaming bone in Edit Mode, Saving, and
than selecting/deselecting.
Caused by a mistake in the 0f89bcdbeb: can not "short-circuit" the
CoW update if it was explicitly requested.
Safest for now solution seems to be to store whether the CoW component
has been explicitly tagged, so that the following configuration can be
supported:
DEG_id_tag_update(id, ID_RECALC_GEOMETRY);
DEG_id_tag_update(id, ID_RECALC_COPY_ON_WRITE);
Differential Revision: https://developer.blender.org/D13966
The evaluated mesh is a result of evaluated modifiers, and referencing
other evaluated IDs such as materials.
It can not be stored in the EditMesh structure which is intended to be
re-used by many areas. Such sharing was causing ownership errors causing
bugs like
T93855: Cycles crash with edit mode and simultaneous viewport and final render
The proposed solution is to store the evaluated edit mesh and its cage in
the object's runtime field. The motivation goes as following:
- It allows to avoid ownership problems like the ones in the linked report.
- Object level is chosen over mesh level is because the evaluated mesh
is affected by modifiers, which are on the object level.
This patch allows to have modifier stack of an object which shares mesh with
an object which is in edit mode to be properly taken into account (before
the change the modifier stack from the active object will be used for all
objects which share the mesh).
There is a change in the way how copy-on-write is handled in the edit mode to
allow proper state update when changing active scene (or having two windows
with different scenes). Previously, the copt-on-write would have been ignored
by skipping tagging CoW component. Now it is ignored from within the CoW
operation callback. This allows to update edit pointers for objects which are
not from the current depsgraph and where the edit_mesh was never assigned in
the case when the depsgraph was evaluated prior the active depsgraph.
There is no user level changes changes expected with the CoW handling changes:
should not affect on neither performance, nor memory consumption.
Tested scenarios:
- Various modifiers configurations of objects sharing mesh and be part of the
same scene.
- Steps from the reports: T93855, T82952, T77359
This also fixes T76609, T72733 and perhaps other reports.
Differential Revision: https://developer.blender.org/D13824
The previous optimization did not work in general yet, unfortunately.
This change makes the code more correct, but also brings back
some unnecessary updates (e.g. when creating a node group).
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
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
The issue was caused by rBd09b1d2759861aa012ab2e7e4ce2ffa2.
Since this commit, the image users in gpu materials were updated
during depsgraph evaluation as well. However, there was a race
condition when one thread is deleting gpu materials in `BKE_material_eval`
while another thread is updating the image users at the same time.
The solution is to make sure that deleting gpu materials is done before
iterating over all gpu materials, by adding a new depsgraph relation.
The core issue is that flushing dependencies are created from an object
to a node tree when it contains e.g. a Texture Coordinate node.
That is an issue because the evaluation of the node tree itself does not
depend on the object (node tree evaluation is essentially a no-op).
Only other systems that parse and evaluate the node tree in a specific
context actually depend on e.g. the position of the referenced object.
It can even be the case that the node tree depends on objects that
the actual evaluator (geometry nodes modifier/material) does not depend
on, because a node is not connected to the output.
Geometry nodes makes the distinction between dependencies to the
node tree and to the evaluator already. Shader nodes do not.
Therefore, shader nodes need a flushing relation from node groups
to their parent node groups.
This brings back some unnecessary updates from rB7e712b2d6a0d
(e.g. when creating a node group from nodes that are not connected
to the output). This is a bit unfortunate, but refactoring how
dependencies work with shader nodes is a out of scope for this fix.
`DEG_OBJECT_ITER_FOR_RENDER_ENGINE_BEGIN` creates temporary objects that
correspond to duplicates or instances.
These temporary objects can share same pointers with the original object
as in the case of Bounding Box.
Bound Box of temporary objects is marked dirty in
`BKE_object_replace_data_on_shallow_copy` since `ob->data` is different.
This causes the original Bounding Box, calculated for the evaluated
geometry, to be lost.
The solution in this commit is to change the boundbox reference of the
temporary objects, so the boundbox of the non-temporary object (with
the data curve) is not marked dirty.
Differential Revision: https://developer.blender.org/D13581
This relation is intended to ensure that the properties of the IK
constraint are ready by the time the IK solver tree is built. This
however can cause spurious dependency cycles, because there is only
one init tree node for the whole armature, and the relation actually
implies dependency on all properties of the bone.
This patch reduces spurious dependencies by only creating the relation
if any properties of the IK constraint specifically are animated.
Differential Revision: https://developer.blender.org/D13714
If multiple bones have a custom property with the same name,
depsgraph didn't distinguish between them, potentially leading
to spurious cycles.
This patch moves ID_PROPERTY operation nodes for bone custom
properties from the parameters component to individual bone
components, thus decoupling them.
Differential Revision: https://developer.blender.org/D13729
Continuation of the D13404 which finished the design of not having
geometry-level nodes dependent on object-level.
Differential Revision: https://developer.blender.org/D13405
In the past that worked because the `GPUMaterial` referenced the
`ImageUser` from the image node. However, that design was incompatible
with the recent node tree update refactor (rB7e712b2d6a0d257d272e).
Also, in general it is a bad idea to have references between data that is
owned by two different data blocks.
This incompatibility was resolved by copying the image user from the node
to the `GPUMaterial` (rB28df0107d4a8). Unfortunately, eevee depended
on this reference, because the image user on the node was update when the
frame changed. Because the image user was copied, the image user in the
`GPUMaterial` did not receive the frame update anymore.
This frame update is added back by this commit. The main change is that
the image user iterator now also iterates over image users in `GPUMaterial`s
on material and world data blocks. An issue is that these materials don't
exist on the original data blocks and that caused the check in
`build_animation_images` in the depsgraph to give the wrong answer.
Therefore the check is extended.
Right now the check is not optimal, because it results in more depsgraph
nodes than are necessary. This can be improved when it becomes cheaper
to check if a node tree contains any references to a video texture.
The node tree update refactor mentioned before makes it much easier
to construct this kind of run-time data from the bottom up, instead of
scanning the entire node tree recursively every time some information
is needed.
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
- Added space below non doc-string comments to make it clear
these aren't comments for the symbols directly below them.
- Use doxy sections for some headers.
Ref T92709
The are few things in the dependency graph which lead to the issue:
- IDs are only built once.
- Object-data level (Armature, i,e,) builder dependent on the object
visibility.
This caused issues when an armature is first built as not directly
visible (via driver, i.e.) and then was built as a directly visible.
This did not update visibility flag on the node for the custom shape
object.
The idea behind the fix is to go away form passing object visibility
flag to the geometry-level builders and instead rely on the common
visibility flush post-processing to make sure certain objects are
fully visible when needed.
This is the safest minimal part of the change for 3.0 release which
acts as an additional way to ensure visibility. This means that it
might not be a complete fix (if some configuration was overseen) but
it should not make currently working cases to not work.
The fix should also make modifiers used on rigify widgets to work.
The more complete fix will have `is_object_visible` argument removed
from the geometry-level builder functions.
Differential Revision: https://developer.blender.org/D13404
Not sure why this bug was only discovered by such an elaborate steps
and why it took so long to be discovered. The root of the issue is
that in the 956c539e59 the typical flow of tag+flush+evaluate was
violated and tagging for visibility change happened after flush.
Copy-on-write data blocks could be referenced from python but were not
properly managing python reference counting.
This would leak memory for any evaluated data-blocks accessed by Python.
Reviewed By: sergey
Ref D12850
Seems to be residue from an early 2.80 days: the placeholder code path
is no longer used.
Remove all the tricky code for this, and make it clear that the
`deg_expand_copy_on_write_datablock` is used on an non-expanded
datablock.
Should be no functional changes. And should help simplify D12850.
Differential Revision: https://developer.blender.org/D12852
This reverts commit 0558907ae6.
Based on discussion with Sergey, having Python references to un-expanded
data should not happen - this change needs to be reconsidered.
This commit adds a node that generates a text paragraph as curve
instances. The inputs on the node control the overall shape of the
paragraph, and other nodes can be used to move the individual instances
afterwards. To output more than one line, the "Special Characters" node
can be used.
The node outputs instances instead of real geometry so that it doesn't
have to duplicate work for every character afterwards. This is much
more efficient, because all of the curve evaluation and nodes like fill
curve don't have to repeat the same calculation for every instance of
the same character.
In the future, the instances component will support attributes, and the
node can output attribute fields like "Word Index" and "Line Index".
Differential Revision: https://developer.blender.org/D11522
With this commit, curve objects support the geometry nodes modifier.
Curves objects now evaluate to `CurveEval` unless there was a previous
implicit conversion (tessellating modifiers, mesh modifiers, or the
settings in the curve "Geometry" panel). In the new code, curves are
only considered to be the wire edges-- any generated surface is a mesh
instead, stored in the evaluated geometry set.
The consolidation of concepts mentioned above allows remove a lot of
code that had to do with maintaining the `DispList` type temporarily
for modifiers and rendering. Instead, render engines see a separate
object for the mesh from the mesh geometry component, and when the
curve object evaluates to a curve, the `CurveEval` is always used for
drawing wire edges.
However, currently the `DispList` type is still maintained and used as
an intermediate step in implicit mesh conversion. In the future, more
uses of it could be changed to use `CurveEval` and `Mesh` instead.
This is mostly not changed behavior, it is just a formalization of
existing logic after recent fixes for 2.8 versions last year and two
years ago. Also, in the future more functionality can be converted
to nodes, removing cases of implicit conversions. For more discussion
on that topic, see T89676.
The `use_fill_deform` option is removed. It has not worked properly
since 2.62, and the choice for filling a curve before or after
deformation will work much better and be clearer with a node system.
Applying the geometry nodes modifier to generate a curve is not
implemented with this commit, so applying the modifier won't work
at all. This is a separate technical challenge, and should be solved
in a separate step.
Differential Revision: https://developer.blender.org/D11597
Evaluating the dependency graph potentially executes Python code when
evaluating drivers. In specific situations (see T91046) this could
deadlock Blender entirely. Temporarily releasing the GIL when evaluating
the depsgraph resolves this.
This is an improved version of
rBfc460351170478e712740ae1917a2e24803eba3b, thanks @brecht for the diff!
Manifest task: T91046
It is causing crashes in rendering, when releasing the GIL in render threads
while the main thread is holding it.
Ref T91046
This reverts commit fc46035117.
Evaluating the dependency graph potentially executes Python code when
evaluating drivers. In specific situations (see T91046) this could deadlock
Blender entirely. Temporarily releasing the GIL when evaluating the depsgraph
resolves this.
Calling the `BPy_BEGIN_ALLOW_THREADS` macro is relatively safe, as it's
a no-op when the current thread does not have the GIL.
Developed in collaboration with @sergey
Manifest task: T91046
Previously, the Point Instance node in geometry nodes could only instance
existing objects or collections. The reason was that large parts of Blender
worked under the assumption that objects are the main unit of instancing.
Now we also want to instance geometry within an object, so a slightly larger
refactor was necessary.
This should not affect files that do not use the new kind of instances.
The main change is a redefinition of what "instanced data" is. Now, an
instances is a cow-object + object-data (the geometry). This can be nicely
seen in `struct DupliObject`. This allows the same object to generate
multiple geometries of different types which can be instanced individually.
A nice side effect of this refactor is that having multiple geometry components
is not a special case in the depsgraph object iterator anymore, because those
components are integrated with the `DupliObject` system.
Unfortunately, different systems that work with instances in Blender (e.g.
render engines and exporters) often work under the assumption that objects are
the main unit of instancing. So those have to be updated as well to be able to
handle the new instances. This patch updates Cycles, EEVEE and other viewport
engines. Exporters have not been updated yet. Some minimal (not master-ready)
changes to update the obj and alembic exporters can be found in P2336 and P2335.
Different file formats may want to handle these new instances in different ways.
For users, the only thing that changed is that the Point Instance node now
has a geometry mode.
This also fixes T88454.
Differential Revision: https://developer.blender.org/D11841
Clearing the Parent Bone field in relations would result in something
like this:
```
add_relation(Bone Parent) - Could not find op_from
(ComponentKey(OBArmature, BONE))
add_relation(Bone Parent) - Failed, but op_to (ComponentKey(OBEmpty,
TRANSFORM)) was ok
ERROR (bke.object): /source/blender/blenkernel/intern\object.c:3330
ob_parbone: Object Empty with Bone parent: bone doesn't exist
```
Now skip creation of a depsgraph relation if the Parent Bone field is
empty (since this would be invalid anyways).
ref. T91101
Maniphest Tasks: T91101
Differential Revision: https://developer.blender.org/D12389
When the same stroke was used as a driver variable, this could make this
stroke already tagged as built in the course of building driver
variables (via `build_gpencil`), but then important stuff from
`build_object_data_geometry_datablock` could be missed later on (because
both of these funtions use `checkIsBuiltAndTag`). Most importantly,
setting up operations such as GEOMETRY_EVAL would be skipped entirely.
`build_object_data_geometry_datablock` seems to cover greasepencil just
fine (does the same as `build_gpencil` and more). Proposed solution is to
remove `build_gpencil` entirely. In `build_id` it would then also call
`build_object_data_geometry_datablock` for `ID_GD` IDs. Now the covered
types that _call_ `build_object_data_geometry_datablock` match exactly
to what is covered _inside_ `build_object_data_geometry_datablock`.
Think this "duplication" of functionality was just overseen in
rB66da2f537ae8 [`build_gpencil` existed long before and said commit made
greasepencil a real object with geometry and such].
thx @JacquesLucke for additional input!
Maniphest Tasks: T88433
Differential Revision: https://developer.blender.org/D12324
We now use a for_each function with callback to iterate through all sequences in the scene.
This has the benefit that we now only loop over the sequences in the scene once.
Before we would loop over them twice and allocate memory to store temporary data.
The allocation of temporary data lead to unintentional memory leaks if the code used returns to exit out of the iteration loop.
The new for_each callback method doesn't allocate any temporary data and only iterates though all sequences once.
Reviewed By: Richard Antalik, Bastien Montagne
Differential Revision: http://developer.blender.org/D12278
This patch exposes the Cycles Alembic Procedural through the MeshSequenceCache
modifier in order to use and test it from Blender.
To enable it, one has to switch the render feature set to experimental and
activate the Procedural in the modifier. An Alembic Procedural is then
created for each CacheFile from Blender set to use the Procedural, and each
Blender object having a MeshSequenceCache modifier is added to list of objects
of the right procedural.
The procedural's parameters derive from the CacheFile's properties which are
already exposed in the UI through the modifier, although more Cycles specific
options might be added in the future.
As there is currently no cache controls and since we load all the data at the
beginning of the render session, the procedural is only available during
viewport renders at the moment. When an Alembic procedural is rendered, data
from the archive are not read on the Blender side.
If a Cycles render is not active and the CacheFile is set to use the Cycles Procedural,
bounding boxes are used to display the objects in the scene as a signal that the
objects are not processed by Blender anymore. This is standard in other DCCs.
However this does not reduce the memory usage from Blender as the Alembic data
was already loaded either during an import or during a .blend file read.
This is mostly a hack to test the Cycles Alembic procedural until we have a
better Blender side mechanism for letting renderers load their own geometry,
which will be based on import and export settings on Collections (T68933).
Ref T79174, D3089
Reviewed By: brecht, sybren
Maniphest Tasks: T79174
Differential Revision: https://developer.blender.org/D10197
