This required to apply a small fix in the Metal texture uploader.
Without synchronization pixels of a wrong pass can be uploaded to
a wrong texture. This is because this code path is heavily reusing
temporary allocations, and at some point the allocation is not
considered as still in use, unless the command buffer used by the
texture uploader is submitted.
Ref #118919
Pull Request: https://projects.blender.org/blender/blender/pulls/118920
No functional changes.
This PR just replaces the `KEYFRAME_OK_CHECKS` macro with a function
`keyframe_ok_checks`.
The motivation for this is to remove the confusing use of nested macros.
Since the actual check has to be passed in, this new function takes a function as an argument.
Pull Request: https://projects.blender.org/blender/blender/pulls/118914
Blender crashes when using the compositor debugger. That's because it
assumes the names of the operation classes will be demangled and will
proceed to slice them causing crashes. The standard makes no guarantees
about the form of the returned type_info::name, and it is implementation
defined.
While it is possible to demangle those names and return something that
is more human readable, it doesn't seem worth it as this is only used by
developers who should have no problem identifying mangled names. So we
just display the name as is.
Pull Request: https://projects.blender.org/blender/blender/pulls/118904
Remove uses for `uiDefButS`. -1 was passed as a2 still in a few
places, but that was most likely just because of copy & paste,
it doesn't look like that's actually used anywhere.
The color luminosity was written to this value, but never read
anywhere. uiBut::a2 is only read in three more places, none of
which could reasonably use this value.
Sort the incoming Alembic objects by name before processing them. This
mirrors what is already done in USD to partially side-step Blender's
quadratic behavior when checking object names.
The file contained in issue #118839 contains over 365000 individual
objects. This change allows loading to be ~4x faster; from 2.2 hours
down to 28 minutes.
Additionally, change the progress bar to report in finer-grained steps
to better indicate that progress is still being made.
Pull Request: https://projects.blender.org/blender/blender/pulls/118869
Note that this still isn't entirely thread safe since the catalogs of
the asset libraries may still be edited in various ways while building
the all asset library. But at least this avoids a data race when
assigning the catalog service once done building it.
Some of these members are not expected to change throughout the lifetime
of the class instances, so make them constant and only set them via the
constructor. Making them immutable this way helps making clear which
data needs extra attention for thread safety.
This adds implicit sharing support for the `MemFile` undo-step. This decreases memory
usage and increases performance.
Implicit sharing allows the undo system to take (shared) ownership of some data.
Previously, the data would always be serialized and compared to the previous undo-step.
So this turns an O(n) operation into O(1) (in terms of memory usage and time).
Read/write code that wants to make use of this has to use the new `BLO_read_shared`
and `BLO_write_shared` functions respectively. Those either make use of implicit-sharing
internally or do the "full" read/write based on a passed-in function. It seems possible to
use the same API in the future to store shared data to .blend files.
Improvements:
* Much faster undo step creation in many cases by avoiding the majority data copies
and equality checks. This fixes#98574. I found undo step creation and undo step
decoding to be 2-5 times faster in some demo files from the blender website and in
some production files from the Heist project.
* Reduced memory usage when there is large data in `bmain`. For example, when
loading the same highly subdivided mesh that I used in #106228 the memory usage
is 1.03 GB now (compared to 1.62 GB in `main` currently). The main remaining copy
of the data now is done by rendering code.
* Some significant performance improvements were also measured for the new grease
pencil type (#105540).
There is one main downside of using implicit-sharing as implemented here: `MemFile`
undo steps can't be written as .blend files anymore. This has a few consequences:
* Auto-save becomes slower (up to 3x), because it can't just save the previous undo step
anymore and does a normal save instead. This has been discussed in more detail here:
https://devtalk.blender.org/t/remove-support-for-saving-memfile-undo-steps-as-blend-files-proposal/33544
It would be nice to work towards asynchronous auto-save to alleviate this problem.
Some previous work has been done to reduce the impact of this change in 41b10424c7
and f0f304e240. This has been committed separately in efb511a76d.
* Writing `quit.blend` has to do a normal file save now. So it's a bit slower too, but it's
less of a problem in practice.
* The `USE_WRITE_CRASH_BLEND` functionality does not work anymore. It doesn't seem
to be used by anyone (removed in e90f5d03c4)
There are also benefits to not writing `MemFile` from undo steps to disk. It allows us to
more safely do undo-specific optimizations without risking corrupted .blend files. This
is especially useful when we want to preserve forward compatibility in some cases.
This requires converting data before writing the .blend files, but this conversion is not
necessary for undo steps. Trying to implement this kind of optimization in the past has
often lead to bugs (e.g. 43b37fbc93).
Another new problem is that it is harder to know the size of each undo step. Currently, a
heuristic is used to approximate the memory usage, but better solutions could be found
if necessary.
Pull Request: https://projects.blender.org/blender/blender/pulls/106903
These shouldn't need to be accessed from outside the asset system
itself, so they should not be exposed with its API. Move them to an own
asset system private headers. This is a further step towards making
asset system classes more encapuslated, so behavior and data flow can be
controlled better (which helps addressing threat safety issues).
Personally I've found it quite confusing to work on higher level issues
of the asset catalog system, because I got lost in the multiple classes.
Hopefully separating them more clearly helps with that too.
Making the member private (or at least protected) makes threat safety
more tangible, and we don't need to expose locking in the API. Generally
we need to make data more encapsulated, so we can make edits more
controlled and threat safe.
Asset libraries also always have a catalog-service, so it can accessed
by reference, rather than pointer that would have to be null-checked.
In many modes, Blender uses the `MemFile` undo step, which serializes all DNA
data in RAM almost as if writing a .blend file. For auto-save, Blender used to
write the last `MemFile` undo step to disk because that was faster serializing
all of DNA again. Furthermore, saving the `quit.blend` file when closing Blender
also used this.
This functionality is now removed in preparate for supporting implicit sharing
in the undo system (#106903). Auto-save and saving the quit.blend file now use
regular file saving.
The removal of this feature and its implications have also been discussed here:
https://devtalk.blender.org/t/remove-support-for-saving-memfile-undo-steps-as-blend-files-proposal/33544
This was a debug feature for developers that allowed writing the last MemFile
undo step as a .blend file when Blender crashed. This was rarely/never used
nowadays and is now removed.
This is in preparation for #106903 which removes support for writing `MemFile`
undo steps as .blend files.
Alternative to #118820 and #118815.
The current optimization to skip work when the offsets are zero
or the positions are the original positions is error prone and overly
specific. Such optimizations should be applied to field evaluation
and attribute capturing in general.
This PR significantly simplifies the set position node to compose
fields with the addition function instead of implementing it directly
in the node. Then the position is saved to the geometry with the
generic utility for capturing attributes on geometry.
The most significant lost optimization is for when the positions are
original and the offset is zero. That could still be added back here
directly if we wanted. Or it could be done later in a more general
way that would also help in other places.
Pull Request: https://projects.blender.org/blender/blender/pulls/118857
This implements a new internal data structure for reverse uv sampling.
Being better than the previous one was not particularly hard, because it
was never really optimized and used a very simple approach. I found the
new implementation to be about 10-20x faster in my tests. Simon was
able to reproduce an comparable speedups in production files.
The speed-up is mainly achieved by a better memory layout and better
multi-threading during construction. The lookup performance is mostly the
same as before.
Like the old data structure, the new one also splits the uv space into
uniformly sized cells. The size of the cells is based on the number of
triangles. Then it sorts all triangles into the rows that they touch. Finally,
it creates a flat array for each row that contains the triangle indices
contained in each cell.
There are still ways to optimize this further, but for now this is a good
improvement already.
Pull Request: https://projects.blender.org/blender/blender/pulls/118772
When using the area light gizmo in the 3d viewport to resize the
area light, the changes where only made visible when EEVEE-Next would
rerender. EEVEE and Cycles did refresh accordingly.
The reason is that EEVEE and Cycles uses a more general approach of
detecting changes. EEVEE-Next listens to exact changes made in the
dependency graph.
This change fixes this by tagging the lamp without any flags similar
to what RNA does. EEVEE and Cycles don't need the general flag for
correct working.
Pull Request: https://projects.blender.org/blender/blender/pulls/118888
Add percentage closer filtering to shadowmap sampling and a
`shadow_filter_radius` property to lights to control it.
Notes:
* This adds PCF to `eevee_shadow_tracing_lib`, but not to
`eevee_shadow_lib`, which is used by volumes (not required) and
thickness.
* PCF is computed based on the LOD0 size. This assumes that higher
LODs are only used when the shadowmap resolution is actually good
enough to match the render resolution.
Pull Request: https://projects.blender.org/blender/blender/pulls/118220
Tiles tagged for update in eevee_shadow_tag_update_comp can be
untagged in eevee_shadow_tilemap_init_comp, since those tiles
might be tagged as rendered.
Regression from 9e015f703c
The `std::all_of` function wasn't used correcty.
Also, the last value of `dst_offsets` was always 0
(because the offsets were still counts at this point)
so the fix also makes sure the last value
is never included in the check.
Auto-save currently only really works in modes that use the `MemFile` undo step,
that excludes things like mesh edit and sculpt mode. Previously, Blender would
attempt to auto-save in those modes, but it would only save the last state from
before the mode was entered, which is useless when staying in the mode for longer.
This problem is *not* fixed here. However, the code now explicitly skips auto-saving
in order to avoid unnecessary short freezes in these modes when Blender auto-saves.
Furthermore, the auto-save will now happen when changing modes.
This reduces the impact of save-time-regressions with #106903.
Pull Request: https://projects.blender.org/blender/blender/pulls/118892
