Do proper tagging for material changes, and avoid grease-pencil hack
in the relations builder.
The dependency graph code was forcing animation to be run for any
copy-on-write change of material. However, animation is not supposed
to be run on copy-on-write changes.
This allows to have drivers on node trees.
Probably now it will be better to simply add parameters for any
ID node, to avoid such amount of manual work.
Initial idea was to not have nodes if it's not needed to avoid
any possible overhead. Having more robust system is probably more
valuable. We can always optimize overhead in one way or another.
This removes special dependency graph code that was intended to avoid GPU
shader recompiles by preserving the node tree and GPU material in specific
cases. This is no longer needed now that we have a general shader pass
cache that compares the generated shader code. The GPU material is already
being freed in material and world eval as well, so there's no point.
Note also that GPU materials are now safe to free from threads, actual
OpenGL buffer freeing happens delayed.
Convention was not to but after discussion on 918941483f we agree its
best to change the convention.
Names now mostly follow RNA.
Some exceptions:
- Use 'nodetrees' instead of 'nodegroups'
since the struct is called NodeTree.
- Use 'gpencils' instead of 'grease_pencil'
since 'gpencil' is a common abbreviation in the C code.
Other exceptions:
- Leave 'wm' as it's a list of one.
- Leave 'ipo' as is for versioning.
We already have different storages for cddata of verts, edges etc.,
'simply' do the same for the mask flags we use all around Blender code
to request some data, or limit some operation to some layers, etc.
Reason we need this is that some cddata types (like Normals) are
actually shared between verts/polys/loops, and we don’t want to generate
clnors everytime we request vnors!
As a side note, this also does final fix to T59338, which was the
trigger for this patch (need to request computed loop normals for
another mesh than evaluated one).
Reviewers: brecht, campbellbarton, sergey
Differential Revision: https://developer.blender.org/D4407
The issue was discovered only after recent changes, but roots back
to much older changes.
What was happening is scene's ID recalc flags where never cleared,
which caused ensure_view_layer() to always run copy-on-write on the
scene. This resulted in certain runtime data being cleared, without
proper flag stored in the dependency graph.
This was caused by ID recalc clear function checking whether any ID
was tagged for recalc in that graph or not. This was happening due
to all areas using DEG_id_type_tag() which can only set flags on the
graph from viewport scenes, and could not inform render dependency
graph.
Now ID tyoe tagging is happening on per-graph level, which avoids
possibility of flags running out of sync.
In a bit longer term we also need to get rid of two functions which
are clearing flags: DEG_id_type_tag() and deg_graph_clear_tags().
Object of evaluated base is not yet copied, so we can not know whether
it has animation on visibility or not.
This issue was reported in T56635#630383.
Original optimization idea was wrong: it is possible that some other
ID would reference an object which is also used by a base.
Rolled back to a bit more fragile solution.
In the future would be nice to make it somewhat less duplicated with
the builder itself.
Fixes assert failure (and possibly crashes) when adding grease pencil
object and switching to a draw mode.
While it's kind of common to use camel case in C++ this is not
currently agreed style for C++ in Blender.
Got confused by working on other areas with 3rd party libraries.
This allows dependency graph to evaluate drivers of those objects
and put them to a correct state. It will increase memory usage
since now we can no longer save it by skipping copy-on-write for
such objects. It will also currently make things slower, because
we do not have granular enough visibility update of components in
the dependency graph. Can do it later when the rest of the changes
are finished.
This commit does not update restriction flags on the base, since
that is somewhat tricky to do currently: need to somehow see whether
object is disabled due to flags on collection or due to own flags.
Differential Revision: https://developer.blender.org/D4419
The idea is to keep bases which are known for sure to be in the
dependency graph. Previously, this code was duplicating logic
around checking restriction flags, which becomes more hard to
maintain once we are moving towards to more comprehensive checks
about whether base is a part of evaluated scene or not.
No need to have iterator loop in the view layer evaluation,
this only makes it more difficult to have base flags covered
by the dependency graph.
Other good thing is that we don't need to worry about whether
base has been removed from the evaluated view layer or not.
Reviewers: brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D4414
Was once again caused by an ambiguity of the entry/exit operations.
Only did for objects since those are the only one who needs this.
The rest types of IDs needs to be checked and only added extra
operations if needed (adding operations and relations causes some
overhead for evaluation, so need to be careful).
Committing this since it does fix broken logic (previously in that
condition obdata would always be set to NULL, since
`BKE_object_runtime_reset()` is called before).
However, this has presumably been broken that way since 05/2018, so
maybe that whole condition is not needed anymore? Or NULL pointer was
working as well here?
@sergey eyes are required here I guess ;)
Caused by rBae2b677dcb5a70f5, Object.runtime has lot of weird specific
handlings in depsgraph...
For now modified `deg_backup_object_runtime()` and
`deg_restore_object_runtime()` to mimic previous behavior regarding
Object bbox (i.e. pass it around, instead of wiping it clean).
Reported in T61660.
The dependency graph now handles updating image users to point to the current
frame, and tags images to be refreshed on the GPU. The image editor user is
still updated outside of the dependency graph.
We still do not support multiple image users using a different current frame
in the same image, same as 2.7. This may require adding a GPU image texture
cache to keep memory usage under control. Things like rendering an animation
while the viewport stays fixed at the current frame works though.
The initial idea of using char pointer was to save some
memory since the dependency graph was kind of the one
with the main database.
Nowadays dependency graph should be separatable from the
main database and being self-sustainable.
Other issue which was caused by this pointer is the
re-tagging of operations during relations update: it is
possible to have node which as tagged for update but had
the owner of the name removed (i.e. driver or bone was
removed).
Is mainly used by drivers variables. The slow part was about
iterating over all pose channels to find the one which has a
given constraint.
Now we build a look up table, so this operation is way cheaper,
Brings down relations update time from 0.7sec to 0.4 with Spring
production file.
