v0.9.10

v0.9.10

• this version used for pynari 1.2.9
• complete rewritten 'unstructured' data type.
  -> added support for cell centric data
  -> barney's internal formatting of unstructured now same as anari data layout, to avoid banari->native reformatting
  -> banari now uses unstructured according to preliminary ANARI 1.1 spec (except 64-bit index types that are not supported)
• various stability fixes
• experimental support for (non-MPI) ANARI multi-device data parallel using 'tethered' devices

v0.9.8

• new release as used for pynari 1.2.8
• support for anari primID/geomID/instID extension
• various fixes throughput
• major bugfix when data-parallel rendering with island parallelism
• major bugfix re data-parallel rendeirng on multi-gpu machines, where lack of peer access enablement caused core dumps

v0.9.6

• added various missing ANARI parameters (hdri::scale, dirlight::(ir)radiance, ...)
• added more ANARI frame buffer formats, and support for frame buffer mapping on GPU
• various bugfixes throughout, in particular in shading
• experimental new CUDA-native backend for machines that have CUDA but no optix

v0.9.4

• Various bugfixes and stability improvements throughout, mostly around multi-device rendering
• support ANARI 'cone' geometry, including attribute mapping (samples in upcoming pynari) (thx @szellmann )
• support ANARI 'quad' geometry (thx @szellmann )
• support ANARI 'cylinder' geometry, including attribute mapping (samples in upcoming pynari) (thx @szellmann )
• no longer doing automatic tone mapping; default frame buffer now linear, only SRGB format does srgb conversion
• embree backend has embree globals moved in special namespace, no longer clashing with other embree apps

Re-activated MPI device, and stability fixes

Functionality: Re-activated MPI backend(s) after 0.9 changes to rtcore backends

Stability/Bugfixes:

• anari device now finalizes the frame buffer on bnFrameBufferGetPointer()
• barney now properly rebuilds the optix pipeline if programs get loaded after the first frame
• anari device no longer sometimes missses resetting accumulation buffer
• umeshes reactivated
• various smaller bugfixes throughout

v0.9.0

• major rework that allows 'rtcore' abstraction and multiple backends
• support both optix and embree (cpu only) backends
• completely reworked cmake build system (and in particular how stuff
  gets linked)
• reworked install/export/import system, exporting both barney and
  barney_mpi; should always get used through find_package(barney)
  and then linked as import target(s) barney::barney and (if found)
  barney::barney_mpi
• anari device split into anari_library_barney and
  anari_library_barney_mpi
• version used for pynari backend that works on all of
  linux/windows/mac and both cpu/gpu

Known limitations/issues:

• AMR support currently disabled
• umesh support only support cubql sampler

v0.8.0

v0.8.0

• updated to latest anari 0.12 SDK
• changed cuda arch detection to use 'native' by default, but allowing
  to override on cmdline.
• updated cuda arch detection in owl and cubql so as to allow barney
  to tell them which one to use (so it matches across all projects)
• reworked path tracing (and in particular MIS code) to (mostly-)pass
  a furnace test