This change reduces the GPU context switches when drawing keyframes.
In the previous situation the keyframe blocks and keyframe keys were
drawn per channel. With this patch first all the keyframe blocks are
drawn for all channels and after that the keyframe keys are collected
for all channels and send to the GPU in a single draw call.
This was left over from when these scripts were loaded as modules,
where their names needed to be compatible with Pythons module naming.
Version patch existing files so text with register enabled
without a `.py` extension wont start executing on startup.
Resolves T89532.
This supported calculating normals for MPoly array which was copied to
an MFace aligned array.
Remove the functions entirely since MFace use is being phased out and
these function isn't used anywhere.
Use BKE_mesh_calc_normals instead of
BKE_mesh_calc_normals_mapping_simple for curve modifier calculation.
This only made sense for derived-mesh which is no longer used.
Remove the 'only_face_normals' argument.
- BKE_mesh_calc_normals_poly for polygon normals.
- BKE_mesh_calc_normals_poly_and_vertex for poly and vertex normals.
Order arguments logically:
- Pair array and length arguments.
- Position normal array arguments (to be filled) last.
Optimize mesh normal calculation.
- Remove the intermediate `lnors_weighted` array, accumulate directly
into the normal array using a spin-lock for thread safety.
- Remove single threaded iteration over loops
(normal calculation is now fully multi-threaded).
- Remove stack array (alloca) for pre-calculating edge-directions.
Summary of Performance Characteristics:
- The largest gains are for single high poly meshes, with isolated
normal-calculation benchmarks of meshes over ~1.5 million showing
2x+ speedup, ~25 million polygons are ~2.85x faster.
- Single lower poly meshes (250k polys) can be ~2x slower.
Since these meshes aren't normally a bottleneck,
and this problem isn't noticeable on large scenes,
we considered the performance trade-off reasonable.
- The performance difference reduces with larger scenes,
tests with production files from "Sprite Fight" showing
the same or slightly better overall performance.
NOTE: tested on a AMD Ryzen TR 3970X 32-Core.
For more details & benchmarking scripts, see the patch description.
Reviewed By: mont29
Ref D11993
The poll for unlinking calls `nla_panel_context` without providing an
adt pointer, and there is a check for this pointer in
`nla_panel_context` leading to never returning true if it is not
provided. (this is fine if there are tracks already, poll would succeed
in this case, `nla_panel_context` goes a different code path then)
Same call to `nla_panel_context` is also done in the beginning of the
corresponding unlink exec function (but this time providing the pointer
because it is used later), so it makes sense to do the same thing in the
poll function. Equal check is also done in the panel poll function, so
now these are all in sync.
Part of T87681.
Maniphest Tasks: T87681
Differential Revision: https://developer.blender.org/D11041
Was reported for a file which does not have an active track set in
AnimData even though it was in strip twek mode (but this was accessed in
is_nlatrack_evaluatable()).
Root cause for this is not totally clear, but I assume the situation is
described as part T87681 (and is fixed in D11052).
This patch here just prevents the crash for files that are already in the
borked state.
Reviewers: sybren
Maniphest Tasks: T89805
Differential Revision: https://developer.blender.org/D12085
Clearing the parent from the UI using the X (or from python) clears the
`parsubstr` and set `partype` back to `PAROBJECT`.
Using the Clear Parent operator would leave the `parsubstr` (and thus
`parent_bone`) untouched even though this operator claims to "clear
parenting relationship completely" (it also removes parent deform
modifiers for example).
So now, also clear `parsubstr` and set back to `PAROBJECT` [which is
default].
Maniphest Tasks: T88498
Differential Revision: https://developer.blender.org/D11503
Add Apply Constraint, Duplicate Constraint, and Copy To Selected
operators, and include them in a menu similar to the menu for modifiers.
The shortcuts in the extras menu are also matched to modifiers.
All the here added operators are intended to work exactly like the
analogous ones for modifiers. That means the apply operator should apply
a constraint as if it was first in the list, just like modifiers do. I
have added the same warning message as for modifiers when that happens.
The decision to use this approach of appling the constraint as if it was
first, was made for consistency with modifiers. People are already used
to how it works there. Is also provides more intricate control over the
applied transforms, then just applying all constraints up to that one.
Apply all constraints is already kinda implemented in Bake Animation.
Reviewed By: HooglyBoogly, sybren, #user_interface
Differential Revision: https://developer.blender.org/D10914
During the processing of a continuous drag event, other mouse move
events may be in the queue waiting to be processed.
But when a mouse wrapping happens, these waiting mouse move events
become out of date as they report a mouse position prior to wrapping.
The current code ignores these events by comparing their `timestamp` to
the time recorded in the last mouse wrapping.
The bug happens because the computed value in
`mach_absolute_time() * 1e-9` for some reason is incompatible with the
value of `[event timestamp]`.
Since macOS 10.6, we have a new way to get the amount of time the
system has been awake. `[[NSProcessInfo processInfo] systemUptime]`.
Using this updated method fixed the problem.
Differential Revision: https://developer.blender.org/D12202
Clearing the window was done in wm_file_read_post which was deferred.
This was needed as it left the context in an invalid state
where the window was set but the screen wasn't.
Crashing when setting up keymaps that attempted to access the
scene from the window in the property update function.
Regression in 497bc4d199
Add RNA_struct_type_find_property_no_base for use in the rare situations
when this isn't desired.
Resolves T90617, where sequence strip sub-types weren't detecting
properties that exist in the base "Sequence" types.
- Remove old comment for editors with weak syntax highlighting.
- Remove disabled code to initialize Blender with a file path.
- Remove file name references to function names since these
were outdated, modern development environments can look up this info.
Resolve order of initialization error reading startup file,
support postponing running wm_file_read_post until Blender
has been initialized.
Deferring updates allows duplicate initialization
to be removed from WM_init.
Reviewed By: mont29
Ref D12184
Makesdna fails to detect issues in 32 bit code that can
only be resolved by adding a padding pointer.
We never noticed since we ourselves no longer build for
32 bit, but debian's 32 bit builds got bitten by this
A rather extensive explanation on why this is alignment
requirement is there can be found in this comment:
https://developer.blender.org/D9389#233034
Differential Revision: https://developer.blender.org/D12188
Reviewed by: sergey, campbellbarton
Some of the dna structs were not properly
aligned for 32 bit builds causing issues
for some of the 32 platforms Debian builds
for.
Reviewed By: sergey, brecht
Differential Revision: https://developer.blender.org/D9389
The loading of a font size or style renders bitmaps of the characters
0-255 and stores them in a cache. But glyphs 128-255 in this cache are
not accessible. What used to be ansi high-bit characters are now multi-
byte UTF-8 sequences.
Therefore this patch reduces the glyph_ascii_table size to 128 and
only caches characters 32-127, the visible portion of ASCII, which
greatly reduces the time to load a font.
See D12189 for more details.
Differential Revision: https://developer.blender.org/D12189
Reviewed by Campbell Barton
Before this commit rendering material icons the floor will is hidden.
This reduces the readability of reflective/refractive materials.
check patch for additional screenshots and notes.
This patch will switch the floor material that uses ray visibility tricks to render a floor for reflective rays.
Eevee uses screen space reflections that makes this a different problem. There is nothing else drawn in
the scene in screen space so we need a different trick. Using math we convert a reflective ray to UV space
and generate a world that projects a checker pattern to infinity.
As now the floor is in the world it is being reflected via
a cubemap. As the film is transparent the background (including the floor isn't rendered)
In the future when Eevee supports vulkan raytracing we can re-evaluate and perhaps remove this
approximation.
We tried lightprobes but that wasn't able to do the trick.
Using the compositor would lead to more memory usage (render layers and intermediate buffers) and slower performance.
Solution has been validated with Simon
Reviewed By: sybren, Severin
Differential Revision: https://developer.blender.org/D11988
To improve the presentation of nodes in the node editor. Recognize the
following metadata from the OSL specification:
* [[ string label = "UI Label" ]]
* [[ string widget = "checkBox" ]]
* [[ string widget = "boolean" ]]
Ref T89741
Differential Revision: https://developer.blender.org/D12074
Idea for 3.0 is to disable all functionality that isn't well polished
and focus on those parts first. Starting with poses.
* Adds a new experimental option "Extended Asset Browser", replacing
"Asset Browser".
* Unlike the previous option, this isn't enabled by default anymore.
This didn't work well in practice and caused plenty of confusion.
* "Mark as Asset" and "Clear Asset" are hidden if the option is
disabled.
* Same for the category selection in the Asset Browser.
* Always show display the "Only Assets" option in the File Browser while
browing inside .blend files. That way you can hide data-blocks that
are not pose assets.
* The Asset Library setup UI in the Preferences is always visible now,
it's needed for pose library access.
Addresses T90181, T90180 and T90300.
Differential Revision: https://developer.blender.org/D12120
ID data-blocks that could be accessed from Python and weren't freed
using BKE_id_free_ex did not release the Python reference count.
Add BKE_libblock_free_data_py function to clear the Python reference
in this case.
Add asserts to ensure no Python reference is held in situations
when ID's are copied for internal use (not exposed through the RNA API),
to ensure these kinds of leaks don't go by unnoticed again.
Using high quality normals for vertex offset when set
for higher precision offsets.
This was only used for calculating even-offset.
Reviewed By: campbellbarton
Ref D12176