The input data to the OptiX denoiser was clamped to 0..10000 as required, but it could easily
exceed that range with a high number of samples (since the data contains the overall sum). To
fix that, divide by the number of samples first and multiply it back in after the denoiser ran.
This change makes it so vertices of edge are only stored when edge
has non-zero crease. This allows to lower memory footprint of 1.5M
faces from 78 MiB to 54 MiB in the case all creases are zero.
Meshes with crease are more hard to predict due to array-based
storage, so it all depends on index of edge with crease. Worst case
(all edges are creased) still stays at 78 MiB.
Makes it possible to set adjacent vertices after edge sharpness.
Initially it seemed like useful sanity check, but with time it
became rather a burden.
Avoid per-face pointer and allocation: store everything as continuous
arrays.
Memory footprint for 1.5M faces:
- Theoretical worst case (all vertices and edges have crease) memory
goes down from 114 MiB to 96 MiB (15% improvement).
This case is not currently achievable since Blender does not expose
vertex crease yet.
- Current real life worst case (all edges have crease) memory goes
down from 108 MiB to 90 MiB (17% improvement).
- Best case (no creases at all) memory goes down from 96 MiB to 78 MiB
(19% improvement).
While this looks trivial it already allowed to catch issues in one
of previous attempt to optimize memory usage. It will totally be
useful for an upcoming refactor of face topology storage.
Allows to perform comparison by doing linear comparison of indices.
Before cyclic match was used to deal with possibly changed winding from
OpenSubdiv side.
Speeds up comparison (and hence improves FPS), makes code more reliable
nut uses more memory.
This change makes it so topology refiner comparison will check vertices
of all existing/provided edges.
The initial claim that due to manifold nature of mesh there is no need
in "deep" edges check was wrong: some areas might only provide edges
with non-zero creases. So if crease of one edge goes changes from 1.0
to 0.0 and crease of other edge goes from 0.0 to 1.0 the old comparison
code would not have caught it.
Similar to previous change in vertex sharpness, explicitly store value
provided by the converter.
Allows to avoid rather fragile check for boundary edges.
Also allows to avoid need in constructing edge map. This lowers memory
footprint of the comparison process and avoids memory allocations
during the comparison (which is an extra benefit from the performance
point of view).
This change starts the transition of topology refiner comparison
to compare actual values given by the converter, which will not be
affected by the refinement or face winding synchronization steps.
Currently is only implemented for vertex sharpness, but will be
extended further as followup development.
Fixes T71908: Subdiv: Incorrect topology comparison, leading to poor performance
The idea is to use this explicit storage for topology comparison rather
than using base level. While this will have memory overhead it allows
to simplify comparison of such things as:
- Vertex sharpness (where base level from topology refiner will have it
refined, meaning it will be different from what application requested
for non-manifold and corner vertices).
- It will allow to simplify face-vertices comparison, where currently
O(N^2) algorithm is used due to possible difference in face winding.
- It will also allow to avoid comparison-time allocation of edge map.
Currently no functional changes, just preparing for development which
will happen next.
Previously it was enabled for debug builds, now it is to be enabled
explicitly.
The reason for this is to reduce overhead when debugging other areas
which might involve subdivision surface. When conversion is to be
debugged set this manually in the code.
Consolidate it inside of the topology refiner implementation class,
which would allow to store extra data acquired during construction
of the OpenSubdiv's object.
Only use OBJECT_GUARDED_{NEW. DELETE} for structures which are part of
public C-API (and hence can not have new/delete operators overloaded).
Could try being brave and override new/delete from under C++ ifdef.
The problem was that Cycles would store a pointer to an object in
`DEGObjectIterData->templ_dupli_object`. This pointer was then accessed
when the iterator was already freed.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D7797
Previously this would be enabled when threads were used, but threads are now
basically always in use so there is no point. Further, this is only needed for
guarded allocation with --debug-memory which is not performance critical.
All parts of drawing (shaders, GL mesh descriptor, material partitioner
and so on) needs to be redone for the draw manager and new OpenSubdiv
library.
Removing untested code which is doomed to be replaced to make localized
refactoring easier.
The code was trying to make winding consistent and manifold, same as
OpenSubdiv expects it to.
Unfortunately, the code was having some issues in corner cases so the
winding wasn't really correct.
Fortunately, the latter (compared to when this code was originally
written) supports orientation on OpenSubdiv side.
Removing code which is currently unused in Blender and which had
known issues. Is simple enough to bring the code from Git history
if the functionality is needed in the future.
The previous naming scheme for the "selected to active" baking options
lead to confusion and they were not describing what they actually did.
To remedy this, I've added a new settings that does what the older setting implied it did.
Reviewed By: Brecht, Dalai, Andy Davies
Differential Revision: http://developer.blender.org/D7733
For some reason was only visible with gcc-10 in release builds.
Kind of makes sense since there is no CMake code which removes strict
compiler flag, so deal with strict flags in the code itself.
A parenthesized type followed by an initializer list is a non-standard explicit type conversion syntax.
Removing the cast seems fine for both MSVC and GCC
There should be no user visible change from this, except that tile size
now affects performance. The goal here is to simplify bake denoising in
D3099, letting it reuse more denoising tiles and pass code.
A lot of code is now shared with regular rendering, with the two main
differences being that we read some render result passes from the bake API
when starting to render a tile, and call the bake kernel instead of the
path trace kernel.
With this kind of design where Cycles asks for tiles from the bake API,
it should eventually be easier to reduce memory usage, show tiles as
they are baked, or bake multiple passes at once, though there's still
quite some work needed for that.
Reviewers: #cycles
Subscribers: monio, wmatyjewicz, lukasstockner97, michaelknubben
Differential Revision: https://developer.blender.org/D3108
Don't use linear array with frame as an index since it has the
following disadvantages:
- Requires every application to take care of frame remapping, which
could be way more annoying than it sounds.
- Inefficient from memory point of view when solving part of a footage
which is closer to the end of frame range.
Using map technically is slower from performance point of view, but
could not feel any difference as the actual computation is way more
complex than access of camera on individual frames.
Solves crash aspect of T72009
Since the sampling and evaluation functions handle both cases anyways,
there's not really a point for keeping the distinction in the kernel,
so we might as well cut down the number of CLOSURE_BSDF_MICROFACETs a bit.
Differential Revision: https://developer.blender.org/D7736
Remove old code that added extra updates for shaders that have a dependency on
objects. The dependency graph can now tell Cycles when a material is affected by
an object transform.
The issue was the usage of the global `__main__` Python module.
When running scripts in the text editor, Blender would overwrite
the `__main__` module.
Reviewers: sebbas
Differential Revision: https://developer.blender.org/D7690
With the Oculus runtime, the VR session would freeze when taking off the HMD
and putting it back on. This was caused by the deletion of graphics resources
too early in the OpenXR state machine, at least for Oculus.
The resources will now only be freed once the session is actually destroyed.
Also fixes an issue where it wasn't possible to stop the session via the UI
when the HMD was taken off.
Reviewed By: Julian Eisel
Differential Revision: https://developer.blender.org/D7635
Recent changes assumed OpenCL 2.0 platform. This adds a check to see if
we are compiling on an OpenCL 2.0 platform.
Patch was tested on:
* AMD Radeon Pro WX 7100 with amdgpu-pro-19.50-1011208-ubuntu-18.04 drivers
* AMD Vega 64 with amdgpu-pro-20.10-1048554-ubuntu-18.04 drivers
* AMD RX 5700 with amdgpu-pro-20.10-1048554-ubuntu-18.04 drivers
Reviewed By: Brecht van Lommel
Differential Revision: https://developer.blender.org/D7637
With the Oculus runtime, the VR session would freeze when taking off the HMD
and putting it back on. This was caused by the deletion of graphics resources
too early in the OpenXR state machine, at least for Oculus.
The resources will now only be freed once the session is actually destroyed.
Also fixes an issue where it wasn't possible to stop the session via the UI
when the HMD was taken off.
Reviewed By: Julian Eisel
Differential Revision: https://developer.blender.org/D7635
Got added in 1a30e52142 (and tweaked in follow-ups) but shouldn't be needed
anymore with the newer popup based quit dialog.
It prevents Blender from quitting properly in case macOS closed all Blender
windows. This may happen in some corner-cases unfortunately (e.g. T74101) which
would be nice to have addressed at some point. Until then, users shouldn't have
to force-kill Blender to shut it down if they run into this.
Use the automatic property split layout (hence, change to the new 40/60% split
ratio) and add decorator buttons for animatable properties.
This actually applies to all node input buttons in the properties, e.g. world shading,
light shading, texture nodes.
Doing this makes the layout more consistent with other layouts in the
properties. But the decorators are also a useful hint for users that these
options can be animated. Previously using decorators and the automatic split
layout wasn't possible, I've done a number of changes now to have it supported.
Before I moved the socket icons to the left side, the decorators also looked
weird (two circle icons next to each other).
{F8497704} With nested items: {F8497708}
Reviewed By: William Reynish, Pablo Vazquez
Differential Revision: https://developer.blender.org/D7544
Looping over all primitives for every object is really slow, so this patch avoids that by moving
the necessary assignments inline with the primitive merging done for every geometry.
This patch will add some compiler hints to break unrolling in the
nestled for loops of the voronoi node.
Reviewed by: Brecht van Lommel
Differential Revision: https://developer.blender.org/D7574
Changing the order of include changes broke GHOST_DEBUG,
however it was using defines in a fragile way.
Fix by removing 'GHOST_DEBUG' and use 'WITH_GHOST_DEBUG' which
was already defined by CMake.
Usable with the CMake option 'WITH_GHOST_WAYLAND'
The following functionality is working:
- Building with X11 and Wayland at the same time,
wayland is used when available.
- Keyboard, pointer handling.
- Cursor handling.
- Dedicated off-screen windows.
- Drag & drop.
- Copy & paste.
- Pointer grabbing.
See D6567 for further details.
Neither Nuke nor Natron support OpenCV's radial distortion model
which makes it impossible to have any kind of interoperability.
The new model is available under the distortion model menu in Lens
settings.
Differential Revision: https://developer.blender.org/D7484