Based on testing by Intel, rendering on Iris GPUs and upcoming Xe GPUs
should work. This is enabled on Windows and Linux.
More testing is needed to verify correctness and performance in production
scenes, but our basic benchmark files seem to give correct results.
Scale Mac trackpad scrolling changes by pixel size of output device.
Differential Revision: https://developer.blender.org/D9723
Reviewed by Brecht Van Lommel
When we had to get special user directories, we'd usually do it in varying,
rather ad-hoc ways. It would be done with a bunch of `#ifdef`s for the
different operating systems. Also, some of the used Win32 functions were legacy
ones and the API docs recommend using newer ones.
Further, seems `BKE_appdir_folder_default()` used `XDG_DOCUMENTS_DIR` wrong.
It's not supposed to be an environment variable but a value inside a config
file.
This adds the platform dependent logic to Ghost, so we can abstract it away
nicely using the `GHOST_ISystemPaths` interface. Getting the desktop directory
for example can now easily be done with:
`GHOST_getUserSpecialDir(GHOST_kUserSpecialDirDesktop).`
For now I added the logic for desktop, documents, downloads, videos, images and
music directories, even though we only use the Documents one. We can extend/
change this as needed, it's easy to do now.
On Windows and macOS, it uses pretty much the same way to access the
directories as elsewhere already. On Linux, it uses the `xdg-user-dir` command
that seems to be available by default on most Linux systems.
No functional changes. The new queries are not actually used yet.
Differential Revision: https://developer.blender.org/D9800
Reviewed by: Brecht Van Lommel
Adds support for building multiple BVH types in order to support using both CPU and OptiX
devices for rendering simultaneously. Primitive packing for Embree and OptiX is now
standalone, so it only needs to be run once and can be shared between the two. Additionally,
BVH building was made a device call, so that each device backend can decide how to
perform the building. The multi-device for instance creates a special multi-BVH that holds
references to several sub-BVHs, one for each sub-device.
Reviewed By: brecht, kevindietrich
Differential Revision: https://developer.blender.org/D9718
The issue is caused by stale data on the Mesh Node which is not cleared
during synchronizing since the socket API refactor so that we can detect
changes. However, synchronization only updates the sockets of the Mesh,
so other properties were left with outdated values.
This caused an underflow when computing attribute size for undisplaced
coordinates as it was using the current number of vertices minus the
previous count of subdivision vertices, which at this point should be 0.
Added a simple method to clear non socket data. Also modified
Mesh.add_undisplaced to always use an ATTR_PRIM_GEOMETRY as the data is
not subdivided yet and it avoids any further issues regarding computing
attribute sizes.
The SVM AO node calls "scene_intersect_local" with a NULL pointer for the intersection
information, which caused a crash with OptiX since it was not checking for this case and
always dereferencing this pointer. This fixes that by checking whether any hit information
was requested first (like is done in the BVH2 intersection routines).
OptiX support is not in fact experimental anymore, so it is time for that notice to go.
All Cycles features that are currently supported on the GPU do work now when OptiX is selected.
This enables support for baking when OptiX is active, but uses CUDA for that behind the scenes, since
the way baking is currently implemented does not work well with OptiX.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D9784
Blender has now the place to store the Cryptomatte settings. This patch
migrates Cycles to use the new settings.
Reviewed By: Brecht van Lommel
Differential Revision: https://developer.blender.org/D9746
I could use a 16 bit atomic fetch + AND for D9719. The alternative would be to
turn a `short` into a `int` in DNA, which isn't a nice workaround.
Also adds tests for the new functions.
Support for the AO and bevel shader nodes requires calling "optixTrace" from within the shading
VM, which is only allowed from inlined functions to the raygen program or callables. This patch
therefore converts the shading VM to use direct callables to make it work. To prevent performance
regressions a separate kernel module is compiled and used for this purpose.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D9733
NanoVDB includes "assert.h" and makes use of "assert" in several places and since the compile
pipeline for CUDA/OptiX kernels does not define "NDEBUG" for release builds, those debug
checks were always added. This is not intended, so this patch disables "assert" for CUDA/OptiX
by defining "NDEBUG" before including NanoVDB headers.
This also fixes a warning about unknown pragmas in NanoVDB thrown by the CUDA compiler.
This infinite loop is caused by a conflict between the volume mesh
creation which unintentionally clears the shaders before early exiting
when no grid is found, and the Blender exporter which adds back the
shaders causing us to reupdate as the shaders changed.
To fix this simply preserve the shaders on the Volume node.
Log to verbosity level 1 rather than INFO severity.
Avoids a lot of overhead coming from construction of the INFO stream
and improves performance and threadability of code which uses logging.
This makes tracking of 250 frames of a track of default settings to
drop down from 0.6sec to 0.4sec.
The event queue can contain events from before pointer warping, ignore those
now. This is an old issue, but became more common now that we disabled event
coalescing and started using the event mouse location rather than the current
mouse location.
Thanks to Yevgeny Makarov and Nicholas Rishel for helping solve this.
Ref D9662
Allows to use mutex, scoped_lock, and conditional_variable from within
the libmv namespace.
Implementation is coming from C++11. Other configurations are easy to
implement, but currently C++11 is the way to go.
This separates out PugiXML that was previously
bundled by OIIO.
As this linux/mac libs are not available
this commit only contains the builder and windows
changes, and the option to enable pugixml is
guarded by a platform if, this can be removed
once all platforms have committed the svn libs.
For details see D8628
This update introduces two improvements from the Mantaflow repository:
(1) Improved particle sampling:
- Liquid and secondary particles are sampled more predictably. With all parameters being equal, baked particles will be computed at the exact same position during every bake.
- Before, this was not guaranteed.
(2) Sparse grid caching:
- While saving grid data to disk, grids will from now on be saved in a sparse structure whenever possible (e.g. density, flame but not levelsets).
- With the sparse optimization grid cells with a value under the 'Empty Space' value (already present in domain settings) will not be cached.
- The main benefits of this optimization are: Smaller cache sizes and faster playback of simulation data in the viewport.
- This optimization works 'out-of-the-box'. There is no option in the UI to enable it.
- For now, only smoke simulation grids will take advantage of this optimization.
Blender uses 64bit atomics to manipulate SessionUUID, and these atomics
were not defined on any of 32bit platforms.
While official support is limited to 64bit platforms only, the code
should not make assumptions about bitness or endianess, in terms that
there should be codepaths and fallback (or provision of them) for 32bit
platforms.
This change makes 64bit atomic functions defined for all platforms.
The atomic_test was compiled and successfully tested on i686 and armv7l
platforms. The rest of compilation process of Blender will be very
tedious, so that was not done.
This change is essential, but not necessarily enough to make Blender
compilable on i686 (ability to compile Blender on 32bit platforms was
lost during the 2.91 development).
This is a functional part of original fix done by Brecht in D9577.
There is a special defines block needed for ARM on Linux. Move it from
public header to an implementation file.
No functional changes.
This is a non-functional part of original fix done by Brecht in D9577.
Recent changes introduced `acc` parameter into the texture read
functions. When nanovdb isn't enabled this leads to compilation errors
as the `acc` variable wasn't defined. OpenCL only compiles needed
features what made it more prominent.
Reviewed By: Patrick Mours
Differential Revision: https://developer.blender.org/D9629
ROCm 3.9 already defined `NULL`. This patch will first check if it was
already defined to remove compilation warnings.
NOTE: This doesn't add official support for ROCm as it still fails to
render correctly (crashes with default cube).
Reviewed By: Brecht van Lommel
Differential Revision: https://developer.blender.org/D9610
The Normal Map node was falling back to (0, 0, 0) when it was missing
the required attributes to calculate a new normal.
(0, 0, 0) is not a valid normal and can lead to NaNs when it is
normalized later in the shader. Instead, we now return sd->N,
the unperturbed surface normal.
Cover all atomic functions with unit tests.
The tests are quite simple, nothing special so far. The goal is to:
- Make sure implementation exists.
- Implementation behaves the same way on all platforms
(We had issue when MSVC and GCC were behaving differently in the
past).
- Proper bitness is used for implementation and non-fixed-size
function implementation.
The tests can be extended further to make sure, for example, that
CAS operations do not cast arguments to a more narrow type for
comparison. Considering it a possible further improvement, as it is
better be done being focused on that specific task.
There is an annoying ifdef around 64bit implementation, which uses
same internal ifdef as the header does. This check is aimed to be
removed, so is easier to simply accept such duplication for now.
The tests seems somewhat duplicate for signed/unsigned variants and
things like this. The reason for that is to keep test code as simple
as possible: attempting to do something smart/tricky in the test code
often causes the test code to be a subject of being covered with its
own unit tests.
Differential Revision: https://developer.blender.org/D9590
Previously the lock-free tests were actually testing guarded allocator
because the main entry point of tests was switching allocator to the
guarded one.
There seems to be no allocations happening between the initialization
sequence and the fixture's SetUp(), so easiest seems to be just to
switch to lockfree implementation in the fixture's SetUp().
The test are passing locally, so the "should work" has high chance
of actually being truth :)
Differential Revision: https://developer.blender.org/D9584
Previously the only way to use lockfree implementation was to start
executable and never switch to guarded allocator.
Surely, it is not possible to switch implementation once any allocation
did happen, but some tests are desired to test lock-free implementation
of the allocator. Those tests did not operate properly because the main
entry point of tests are forcing guarded allocator to help catching
bugs.
This change makes it possible for those tests to ensure they do operate
on lock-free implementation.
There is no functional changes here, preparing boilerplate for an
upcoming work on the allocator tests themselves.
Scaling of forces needs more work. Before making changes to them it would be nice to have a setup, that works physically correct across multiple modifiers (cloth, rigid bodies, fluid).
This will be a to do for 2.92.
Coalescing on macOS overwrites a singular unprocessed mouse event. To
receive all mouse and tablet events coalescing is disabled.
Disabling coalescing for macOS disables coalescing for trackpad
gestures. Repeat trackpad events are unnecessary and found to
negatively impact performance thus are re-coalesced in Window Manager.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D9574
The OpenVDB data structure can store voxel data in leaf nodes or tiles
when all the nodes in a given region have a constant value. However,
Cycles is using the leaf nodes to generate the acceleration structure
for computing volume intersections which did not include constant tiles.
To fix this, we simply voxelize all the active tiles prior to generating
the volume bounding mesh. As we are using a MaskGrid, this will not
allocate actual voxel buffers for each leaf, so the memory usage will be
kept low.
Differential Revision: https://developer.blender.org/D9557
Reviewed by: brecht, JacquesLucke
Now that the Blender sync mechanism deletes nodes from the scene, we need to
ensure scene update is stopped before we do this.
Also add some more early out in scene geometry update to ensure we do not
continue working on incomplete geometry data, though that was not the cause of
this crash.
