Release Release 0.2.0 · NVIDIA-Omniverse/ovrtx

Support for Gaussian Splats and other particle field primitive types using the UsdVol.ParticleField schema.
Operation status query API and logging callback for monitoring renderer operations.
Dedicated functions for creating supported config values.
GPU selection by CUDA device index at renderer creation using ovrtx_config_entry_active_cuda_gpus() and per render product using uint[] deviceIds.
ovrtx_get_version() Python binding with version compatibility check between the Python package and native library.
Python bindings for enable_profiling, read_gpu_transforms config entries.
Async data access in Python bindings (write_attribute, write_array_attribute) matching the C API.

Upgraded DLPack from 0.8 to 1.3 in both C and Python APIs, allowing creating boolean tensors.
Attribute writes now accept any memory-compatible tensors as input. This means that a N-element, 4x4 matrix can now be
written as shape=[N, 4, 4] instead of shape[N] with 16 lanes as was previously required, making interop with numpy
much simpler and removing the need for extra helper functions to correct tensor shapes.
C API headers are now pure C compatible (removed C++ constructs from ovrtx_attributes.h and pathdictionary_helper.h)
Transform attributes now use the omni:xform alias; direct writing of localMatrix/worldMatrix is no longer
supported and transforms must be written to omni:xform instead. If those transforms are in world space as opposed to
local space then bool resetXformStack=true must also be set on the same prim.

Removed spurious IRenderSettings::getRenderSettings failed warning when no global RenderSettings prim is present in the USD stage.
Removed nvidia-smi printout on startup.
Fixed material-related memory leaks.
Fixed a visual glitch when calling ovrtx_reset_stage().

Provide feedback