This patch replaces / redoes the entire MeshExtractors system.
Although they were useful and facilitated the addition of new buffers, they made it difficult to control the threads and added a lot of threading overhead.
Part of the problem was in traversing the same loop type in different threads. The concurrent access of the BMesh Elements slowed the reading.
This patch simplifies the use of threads by merging all the old callbacks from the extracts into a single series of iteration functions.
The type of extraction can be chosen using flags.
This optimized the process by around 34%.
Initial idea and implementation By @mano-wii.
Fine-tuning, cleanup by @atmind.
MASTER:
large_mesh_editing:
- rdata 9ms iter 50ms (frame 155ms)
- Average: 6.462874 FPS
PATCH:
large_mesh_editing:
- rdata 9ms iter 34ms (frame 136ms)
- Average: 7.379491 FPS
Differential Revision: https://developer.blender.org/D11425
Reuse loose geometry during selection (and other operations) from
previous calculation. Loose geometry stays the same, but was
recalculated to determine the size of GPU buffers. This patch would
reuse the previous loose geometry when geometry wasn't changed.
Although not the main bottleneck during selection it is measurable.
Master.
`rdata 46ms iter 55ms (frame 410ms)`
This patch.
`rdata 5ms iter 52ms (frame 342ms)`
Reviewed By: mano-wii
Differential Revision: https://developer.blender.org/D11339
Some discard of vertbuf were not correctly followed by discards of
the GPUBatches that were using them. This lead to a use-after-free
situation where GPUBatches would reuse old VBO information. This did
not crash immediatly because the VBO indices were cached by our VAO
caching system. It kept working on some implementation because VBO
reference in the VAO (probably) preventing the VBO from being freed
by reference counting.
This fixes T85977 NVidia: Random crashes in 'DrvPresentBuffers'
Before this change, you could have the new sculpt symmetry code and the
older weight paint symmetry code active at the same time. This would
lead to users easily trashing their weigh paint data if they were not
careful when switching between modes.
Now the specific weight paint symmetry code is an exclusive toggle so
the user can't accidentally mirror strokes and vertex groups at the same
time. This also paves the way of supporting Y and Z symmetry in the
future for weight groups mirroring if we decide to add it in the future.
Reviewed By: Sybren
Differential Revision: http://developer.blender.org/D10426
Weight Paint Multi-Paint definitely depends on the weight specific
flag, and vertex group locking also involves group name symmetry
via BKE_object_defgroup_mirror_selection. These two are also
features implemented by me so I feel confident.
The rest of object_vgroup.c possibly should be changed too, but
that requires more consideration than these obvious cases.
The root cause of this bug is that the function that updates the PBVH
normals is drw_sculpt_generate_calls. As now both the overlays and
mesh can be drawn without using pbvh drawing, the normals were not
updating. This patch forces a normals updates also in the no PBVH
drawing code path of the overlays. This was affecting both shading and
sculpt surface sampling in both flat and smooth shading modes.
Having the sculpt normals being updated by the drawing code is a wrong
design which also causes other issues like:
Brushes that sample the surface and do multiple stroke steps between
redraws will sample invalid normals, creating artifacts during the
stroke clearly visible in some brushes.
Brushes that do not need to sample the surface update the normals on
each redraw. This affects performance a lot as in some cases, updating the
normals takes more time than doing the brush deformation. If flat shading
is being used, this is only necessary to do once after the stroke ends.
Reviewed By: fclem
Differential Revision: https://developer.blender.org/D9535
Issue caused by {9582797d4b50} in b2.90. The surface per material used
an index buffer owned by the batch. These index buffers are created at
the same time the surface tris index buffer was created. When a material
per batch buffer was invalidated it used the surface tris index buffer
rendering all materials on all surfaces making the last draw command
render succeed.
This patch stores the surface tris per material in the cache so they can
be reused. There is also no need to use the `saved_elem_ranges` anymore as they are
now part of the cache.
The ugly bit of the implementation is that in `extract_tris_finish` the
MeshBufferCache is retrieved. But as this part was already documented as
a hack and it is something that is only used for final meshes. Other
solutions would impact performance or made the fix not condensed
(passing parameters that shouldn't be used).
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D9136
The alloc was done in the init, but the free was done in the discard.
As discarding can happen more often there were some unneeded if
statements to work around use after free.
This patch moves the free to the `mesh_batch_cache_clear` and removes
the if statements to check for use after free.
This adds XYZ symmetry as a property of meshes and updates all modes to
use the mesh symmetry by default to have a consistent tool behavior
between all modes and when switching objects.
Reviewed By: brecht, mano-wii, campbellbarton
Maniphest Tasks: T79785
Differential Revision: https://developer.blender.org/D8587
This removes the limitation of the sculpt overlays not being visible
with modifiers active.
Reviewed By: fclem
Maniphest Tasks: T68900
Differential Revision: https://developer.blender.org/D8673
This should be using the mesh_cd_ldata_get_from_mesh function in
order to get ldata from BMesh in edit mode.
Reviewed By: sergey
Maniphest Tasks: T78225
Differential Revision: https://developer.blender.org/D8818
This prevents UV layer mix up in
MeshBatchCache.cd_used/cd_needed/cd_used_over_time which depends on the
extraction method.
One object's mesh can be accessed with MR_EXTRACT_MESH, another object
that uses the same mesh can use MR_EXTRACT_BMESH based on
(Object.mode & OB_MODE_EDIT), this causes a problem as the edit-mesh
and the mesh aren't always in sync, the custom data layers wont
necessarily match up, causing T77359.
Reviewed by @jbakker, @brecht
Ref D8645
This workaround addresses T79533 and T79038 for Blender 2.90. The
solution isn't clear and needs more research and work. In order to
continue with blender 2.90 release it was proposed to add this work
around.
It has been tested with the test files provided in the reports.
Reviewed By: Dalai Felinto, Clément Foucault
Differential Revision: https://developer.blender.org/D8695
This workaround addresses T79533 and T79038 for Blender 2.90. The
solution isn't clear and needs more research and work. In order to
continue with blender 2.90 release it was proposed to add this work
around.
It has been tested with the test files provided in the reports.
Reviewed By: Dalai Felinto, Clément Foucault
Differential Revision: https://developer.blender.org/D8695
This addresses warnings from Clang-Tidy's `readability-else-after-return`
rule in the `source/blender/draw` module. Not all warnings are addressed
in this commit.
No functional changes.
When using link duplicated objects it could happen that one object is
calculating the GPUBuffers and the second object is marking these
buffers invalid. This introduces threading issues.
This patch fixes this by combining the surface and surface per material
batches. Most likely the surface per material batches are used and when
requested you will most likely need the surface batch for the depth
tests and overlays.
During tests it slightly improves performance as batches aren't thrown
away without using it.
After this patch we can add a quick path for meshes with one material
and two materials.
Alternative approaches that have been checked:
- sync extraction per object: reduced performance to much (-15%)
({D8292})
- post checks: reduced the threading issues, but didn't solve it.
- separating preparation and execution of the extraction ({D8312})
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D8329
The vertex colors node was using the M_COL attribute type but Sculpt
Vertex Colors use CD_PROP_COLOR
Now the Vertex Color node also fallbacks to legacy vertex colors if
Scultp Vertex Colors are not enabled as experimental.
Reviewed By: brecht
Maniphest Tasks: T78369
Differential Revision: https://developer.blender.org/D8185
Sculpt vertex colors changed the `DRW_MeshCDMask` from 4 bytes to 8 bytes, but
the functions assumed it still was 4 bytes. This patch updates the functions and
adds a compile time check.
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D8215
This disables all Sculpt Vertex Colors tools, operators, panels and rendering capabilities and puts them under the "Use Sculpt Vertex Colors" experimental option.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D8239
Sculpt Vertex Colors is a painting system that runs inside sculpt mode, reusing all its tools and optimizations. This provides much better performance, easier to maintain code and more advanced features (new brush engine, filters, symmetry options, masks and face sets compatibility...). This is also the initial step for future features like vertex painting in Multires and brushes that can sculpt and paint at the same time.
This commit includes:
- SCULPT_UNDO_COLOR for undo support in sculpt mode
- SCULPT_UPDATE_COLOR and PBVH flags and rendering
- Sculpt Color API functions
- Sculpt capability for sculpt tools (only enabled in the Paint Brush for now)
- Rendering support in workbench (default to Sculpt Vertex Colors except in Vertex Paint)
- Conversion operator between MPropCol (Sculpt Vertex Colors) and MLoopCol (Vertex Paint)
- Remesher reprojection in the Voxel Remehser
- Paint Brush and Smear Brush with color smoothing in alt-smooth mode
- Parameters for the new brush engine (density, opacity, flow, wet paint mixing, tip scale) implemented in Sculpt Vertex Colors
- Color Filter
- Color picker (uses S shortcut, replaces smooth)
- Color selector in the top bar
Reviewed By: brecht
Maniphest Tasks: T72866
Differential Revision: https://developer.blender.org/D5975
This adds object motion blur vectors for EEVEE as well as better noise
reduction for it.
For TAA reprojection we just compute the motion vector on the fly based on
camera motion and depth buffer. This makes possible to store another motion
vector only for the blurring which is not useful for TAA history fetching.
Motion Data is saved per object & per geometry if using deformation blur.
We support deformation motion blur by saving previous VBO and modifying the
actual GPUBatch for the geometry to include theses VBOs.
We store Previous and Next frame motion in the same motion vector buffer
(RG for prev and BA for next). This makes non linear motion blur (like
rotating objects) less prone to outward/inward blur.
We also improve the motion blur post process to expand outside the objects
border. We use a tile base approach and the max size of the blur is set via
a new render setting.
We use a background reconstruction method that needs another setting
(Background Separation).
Sampling is done using a fixed 8 dithered samples per direction. The final
render samples will clear the noise like other stochastic effects.
One caveat is that hair particles are not yet supported. Support will
come in another patch.
Reviewed By: jbakker
Differential Revision: https://developer.blender.org/D7297
This patch uses a graph flow scheduler for creating all mesh batches.
On a Ryzen 1700 the framerate of Spring/020_02A.anim.blend went from 10 fps to 11.5 fps.
For each mesh where batches needs to be updated a sub-graph will be added to the task_graph.
This sub-graph starts with an extract_render_data_node. This fills/converts the required data
from Mesh.
Small extractions and extractions that can't be multi-threaded are grouped in a single
`extract_single_threaded_task_node`.
Other extractions will create a node for each loop exceeding 4096 items. these nodes are
linked to the `user_data_init_task_node`. the `user_data_init_task_node` prepares the userdata
needed for the extraction based on the data extracted from the mesh.
Note: If the `lines` and `lines_loose` are requested, the `lines_loose` sub-buffer is created
as part of the lines extraction. When the lines_loose is only requested the sub-buffer is
created from the existing `lines` buffer. It is assumed that the lines buffer is always
requested before or together with the lines_loose what is always the case (see
`DRW_batch_requested(cache->batch.loose_edges, GPU_PRIM_LINES)` in `draw_cache_impl_mesh.c`).
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D7618
This resolves a performance regression in 2.8x where every edit-mode
update performed an edit-mesh to mesh conversion.
Now the conversion will be lazily initialized if/when it's required.
New BKE_mesh_wrapper_* functions abstract over mesh data access.
Currently only edit-mesh and regular meshes are supported.
In the future sub-surface meshes may be supported too.
Since rBcf258b02f449, only wires and edges that are mapped to the
original mesh were drawn if the mesh was modified by modifiers.
Above commit was only meant for showing orig wires for paint mask
overlays [where final wireframe is not desired], so now only use
MR_EXTRACT_MAPPED when we are in a paint mode.
Maniphest Tasks: T75343
Differential Revision: https://developer.blender.org/D7333
Note this only changes cases where the variable was declared inside
the for loop. To handle it outside as well is a different challenge.
Differential Revision: https://developer.blender.org/D7320
