Mutex is now local to particular CCGDM, and guarding edge hash which is only
used by a single function only. There is no need to acquire read lock after
edge hash was created.
This is a bit annoying to have per-DM locking, but it's way better (as in, up to
4 times better) for playback speed when having lots of subsurf objects,
This is really similar to what GPU module was already doing. There are number of
possible improvements still:
- Re-use allocated VAOs when requesting new ones instead of going to the trouble
of freeing VAO and then re-creating it again.
- Move VAO handling to GPU module.
Fixes T46589: OpenSubdiv crash with drivers
Make it so CCGDM reports 0 number of geometry when it uses GPU backend for
drawing. This screws up a bit statistics in info header and requires to have
some special handle of CCGDM in the drawing code, but makes it so non of the
areas will try to access non-existing geometry.
This commit contains all the remained parts needed for initial integration of
OpenSubdiv into Blender's subdivision surface code. Includes both GPU and CPU
backends which works in the following way:
- When SubSurf modifier is the last in the modifiers stack then GPU pipeline
of OpenSubdiv is used, making viewport performance as fast as possible.
This also requires graphscard with GLSL 1.5 support. If this requirement is
not met, then no GPU pipeline is used at all.
- If SubSurf is not a last modifier or if DerivesMesh is being evaluated for
rendering then CPU limit evaluation API from OpenSubdiv is used. This only
replaces the legacy evaluation code from CCGSubSurf_legacy, but keeps CCG
structures exactly the same as they used to be for ages now.
This integration is fully covered with ifdef and not enabled by default
because there are several TODOs to be solved first:
- Face varying data interpolation is not really cleanly implemented for GPU
in OpenSubdiv 3.0. It is also not implemented for limit evaluation API.
This basically means we'll have really hard time supporting UVs.
- Limit evaluation only works with adaptivly subdivided meshes so far, which
basically means all the points of CCG are pushed to the limit. This gives
different result from old code.
- There are some serious optimizations possible on the topology refiner
creation, which would speed up initial OpenSubdiv mesh creation.
- There are some hardcoded asumptions in the GPU and DerivedMesh areas which
could be generalized.
That's something where Antony and Campbell can help, making it so the code
is structured in a way which is reusable by all planned viewport projects.
- There are also some workarounds in the dependency graph to make sure OpenGL
buffers are only freed from the main thread.
Those who'll be wanting to make experiments with this code should grab dev
branch (NOT master) from
https://github.com/Nazg-Gul/OpenSubdiv/tree/dev
There are some patches applied in there which we're working on on getting
into upstream.
multires.
Code had special guards for such edges to stop this from happening. I
don't see why this is needed though since code above assigns smoothed
positions for all vertices in the grid.
After removing the guards I saw that this in fact was the only place
where grd adjacency was used, so I completely removed it.
* Add new CCG function ccgSubSurf_setAllocMask(). Similar to to
ccgSubSurf_setCalcVertexNormals(), it sets whether the CCG elements
have a mask layer and what that layer's offset is. Unlike normals
however, it doesn't change any behavior during CCG calculation; it's
there only to give CCGKey information on the mask.
* Add a new flag to _getSubSurf(), CCG_ALLOC_MASK. If set, space for
an extra layer is allocated, but the number of CCG layers is not set
to include it. This is done because GridPaintMasks are absolute,
rather than being relative to the subdivided output (as MDisp
displacements are), so we skip subdividing paint masks here.
* Add a new flag to subsurf_make_derived_from_derived(),
SUBSURF_ALLOC_PAINT_MASK. This controls whether CCG_ALLOC_MASK is
set for _getSubSurf(). Related, masks are never loaded in during
ss_sync_from_derivedmesh(). After subdivision is finished, if the
alloc mask flag is set, the number of CCG layers is increase to 4
with ccgSubSurf_setNumLayers().
* Add a new flag to multires_make_from_derived(),
MULTIRES_ALLOC_PAINT_MASK. Not all multires functions need paint
mask data (e.g. multiresModifier_base_apply.) This flag is always
set in MOD_multires.c so that subdividing a mesh with a mask updates
properly even when not in sculpt mode.
* Update multiresModifier_disp_run() to apply, calculate, and add mask
elements. It's almost the same as the existing operations with xyz
coordinates, but treats masks as absolute rather than displacements
relative to subdivided values.
* Update multires_customdata_delete to free CD_GRID_PAINT_MASK in
addition to CD_MDISPS.
* Update multires_del_higher() to call the new function
multires_grid_paint_mask_downsample(), which allocates a
lower-resolution paint mask grid and copies values over from the
high-resolution grid.
* Changes to DerivedMesh interface: DMGridData has been removed,
getGridData() now returns an array of CCGElem pointers. Also added
getGridKey() to initialize a CCGKey (implemented only by
CCGDerivedMesh.)
* PBVH: added BLI_pbvh_get_grid_key().
* A lot of code is affected, but mainly is just replacing
DMGridData.co, DMGridData.no, and sizeof(DMGridData) with the
CCG_*_elem functions, removing the reliance on grid elements of
exactly six floats.
Changed the create_vert_poly_map function to return a more compact
structure. Memory saved will vary depending on the mesh, but typically
it should be about one third of the old size.
Renamed the multiresModifier_update() function to
multires_modifier_update_mdisps() and made it visible to subsurf_ccg.c
so it can be called directly. No functional change, just a bit simpler.
Previously, the shading and material was set once per PBVHNode when
drawing. This is still the case, but PBVHNodes are now built to
contain only one material and shading mode.
This is done with an extra partitioning step; once the number of
primitives in the node falls below the PBVH leaf limit, it's
primitives are checked for matching materials. If more than one
material or shading mode is present in the node, it is split and
partitioned (partitioned by material rather than 3D location.)
Given a sufficiently 'annoying' input, like a dense mesh with
thousands of materials randomly scattered across it, this could
greatly increase PBVH build time (since nodes might end up containing
a single primitive), but in general this shouldn't come up.
In order to support materials for grids, the CCGDM is building another
grid array (of DMFaceMat structs). This could be used to replace
CCGDM.faceFlag for some small memory savings (TODO).
Previously this was just a character array that stored face/poly flags
and mat_nr. Since mat_nr should be a short, this would have truncated
for higher values.
Replaced with a new DMFaceFlags struct.
Used a crazyspace approach (like in edit mode), but only modifiers with
deformMatricies are allowed atm (currently shapekeys and armature modifiers only).
All the rest modifiers had an warning message that they aren't applied because
of sculpt mode. Deformation of multires is also unsupported.
With all this restictions users will always see the actual "layer" (or maybe
mesh state would be more correct word) they are sculpting on.
Internal changes:
- All modifiers could have deformMatricies callback (the same as deformMatriciesEM but
for non-edit mode usage)
- Added function to build crazyspace for sculpting (sculpt_get_deform_matrices), but it
could be generalized for usage in other painting modes (particle edit mode, i.e)
Todo:
- Implement crazyspace correction to support all kinds of deformation modifiers
- Maybe deformation of multires isn't so difficult?
- And maybe we could avoid extra bad-level-stub for ED_sculpt_modifiers_changed
without code duplicating?
Fixes various crashes and redraw problems, most noticeable new feature
is that you can now sculpt on a multires mesh with deforming modifiers
preceding it.
I've left out support for sculpting on multires with enabled modifiers
following it, in this case only the base mesh can be sculpted now. The
code changes needed to do this are just too ugly in my opinion, would
need a more torough redesign which I don't think we should try now. In
my opinion this is also not really an important case, since it's going
to be incredibly slow anyway to run a modifier on a high res mesh while
sculpting.
So, to summarize current state:
* Fastest sculpting: base mesh with no modifiers or multires with only
modifiers preceding it.
* Slower sculpting: base mesh with modifiers, depends on the speed of
the modifiers.
* Not supported: multires mesh with modifiers following it.
Most likely will not compile for others, I'd appreciate any build errors
and missing files reports (I can never seem to get everything committed
and all the build systems working without help).
Porting over the sculpt/multires tools was a breeze,
thanks goes to brecht for a design that didn't exclude
ngons and was easy to port.
Note that I've not tested externally-backed multires
file support yet. Also, I still need to write version
patch code for some cases.
Some notes:
* Like trunk, topological changes don't update multires right,
so e.g. subdivide will duplicate multires data on the new faces,
instead of subdividing it.
* If you set the debug value (ctrl-alt-d) to 1 it'll turn on
my experiments in speeding up sculpting on higher-res multires
meshes (but note it makes partial redraw not completely accurate).
* There's a bug where you have to go through editmode to get out
of sculpt mode, not sure if I inherited or created this myself.
* Multithread parts of multires and subsurf. Only loops working on
face grid data and do no memory allocation have been multithreaded,
others would be more complicated.
* Force some CCGSubsurf functions to be inlined, gives a small overall
speedup in subsurf code.
* Fix sculpting not working correct with transformed objects.
* Fix a few cases of "spikes" on lower level multires levels. There's
still cases where it happens, usually on boundary cornders. The
problem is that in such cases the limit surfaces can be very different
from the low res surface, so the tangent space is very different too..
* Fix crash deleting multires higher levels with level set to 0.
* Fix crashes that happened sometimes when adding faces in editmode.