Embree

- Workaround for compile issues with Visual Studio 2017
- Fixed bug in subdiv code for static scenes when using tessellation
levels larger than 50.
- Fixed low performance when adding many geometries to a scene.
- Fixed high memory consumption issue when using instances in
dynamic scene (by disabling two level builder for user geometries
and instances).

- Improved multi-segment motion blur support for scenes with
different number of time steps per mesh.
- New top level BVH builder that improves build times and BVH quality
of two-level BVHs.
- Added support to enable only a single ISA. Previously code was
always compiled for SSE2.
- Improved single ray tracing performance for incoherent rays on
AVX-512 architectures by 5-10%.
- Improved packet/hybrid ray tracing performance for incoherent rays
on AVX-512 architectures by 10-30%.
- Improved stream ray tracing performance for coherent rays in
structure-of-pointers layout by 40-70%.
- BVH builder for compact scenes of triangles and quads needs
essentially no temporary memory anymore. This doubles the
maximal scene size that can be rendered in compact mode.
- Triangles no longer store the geometry normal in fast/default mode
which reduces memory consumption by up to 20%.
- Compact mode uses BVH4 now consistently which reduces memory
consumption by up to 10%.
- Reduced memory consumption for small scenes (of 10k-100k primitives)
and dynamic scenes.
- Improved performance of user geometries and instances through BVH8
support.
- The API supports now specifying the geometry ID of a geometry at
construction time. This way matching the geometry ID used by
Embree and the application is simplified.
- Fixed a bug that would have caused a failure of the BVH builder
for dynamic scenes when run on a machine with more then 1000 threads.
- Fixed a bug that could have been triggered when reaching the maximal
number of mappings under Linux (`vm.max_map_count`). This could have
happened when creating a large number of small static scenes.
- Added huge page support for Windows and MacOSX (experimental).
- Added support for Visual Studio 2017.
- Removed support for Visual Studio 2012.
- Precompiled binaries now require a CPU supporting at least the
SSE4.2 ISA.
- We no longer provide precompiled binaries for 32-bit on Windows.
- Under Windows one now has to use the platform toolset option in
CMake to switch to Clang or the Intel® Compiler.
- Fixed a bug for subdivision meshes when using the incoherent scene
flag.
- Fixed a bug in the line geometry intersection, that caused reporting
an invalid line segment intersection with primID -1.
- Buffer stride for vertex buffers of different time steps of triangle
and quad meshes have to be identical now.
- Fixed a bug in the curve geometry intersection code when passed a
perfect cylinder.

- Added `rtcCommitJoin` mode that allows thread to join a build
operation. When using the internal tasking system this allows
Embree to solely use the threads that called `rtcCommitJoin` to
build the scene, while previously also normal worker threads
participated in the build. You should no longer use `rtcCommit`
to join a build.
- Added `rtcDeviceSetErrorFunction2` API call, which sets an error
callback function which additionally gets passed a user provided
pointer (`rtcDeviceSetErrorFunction` is now deprecated).
- Added `rtcDeviceSetMemoryMonitorFunction2` API call, which sets a
memory monitor callback function which additionally get passed a
user provided pointer. (`rtcDeviceSetMemoryMonitorFunction` is now
deprecated).
- Build performance for hair geometry improved by up to 2×.
- Standard BVH build performance increased by 5%.
- Added API extension to use internal Morton-code based builder, the
standard binned-SAH builder, and the spatial split-based SAH builder.
- Added support for BSpline hair and curves. Embree uses
either the Bézier or BSpline basis internally, and converts other
curves, which requires more memory during rendering. For reduced
memory consumption set the `EMBREE_NATIVE_SPLINE_BASIS` to the basis
your application uses (which is set to `BEZIER` by default).
- Setting the number of threads through `tbb::taskscheduler_init`
object on the application side is now working properly.
- Windows and Linux releases are build using AVX-512 support.
- Implemented hybrid traversal for hair and line segments for
improved ray packet performance.
- AVX-512 code compiles with Clang 4.0.0
- Fixed crash when ray packets were disabled in CMake.

- Added `ignore_config_files` option to init flags that allows the
application to ignore Embree configuration files.
- Face-varying interpolation is now supported for subdivision surfaces.
- Up to 16 user vertex buffers are supported for vertex
attribute interpolation.
- Deprecated `rtcSetBoundaryMode` function, please use the new
`rtcSetSubdivisionMode` function.
- Added `RTC_SUBDIV_PIN_BOUNDARY` mode for handling boundaries of
subdivision meshes.
- Added `RTC_SUBDIV_PIN_ALL` mode to enforce linear interpolation
for subdivision meshes.
- Optimized object generation performance for dynamic scenes.
- Reduced memory consumption when using lots of small dynamic objects.
- Fixed bug for subdivision surfaces using low tessellation rates.
- Hair geometry now uses a new ribbon intersector that intersects with
ray-facing quads. The new intersector also returns the v-coordinate
of the hair intersection, and fixes artefacts at junction points
between segments, at the cost of a small performance hit.
- Added `rtcSetBuffer2` function, that additionally gets the number of
elements of a buffer. In dynamic scenes, this function allows to
quickly change buffer sizes, making it possible to change the number
of primitives of a mesh or the number of crease features for
subdivision surfaces.
- Added simple 'viewer_anim' tutorial for rendering key
frame animations and 'buildbench' for measuring BVH (re-)build
performance for static and dynamic scenes.
- Added more AVX-512 optimizations for future architectures.

- Improved performance for compact (but not robust) scenes.
- Added robust mode for motion blurred triangles and quads.
- Added fast dynamic mode for user geometries.
- Up to 20% faster BVH build performance on the second generation
Intel® Xeon Phi™ processor codenamed Knights Landing.
- Improved quality of the spatial split builder.
- Improved performance for coherent streams of ray packets (SOA
layout), e.g. for fast primary visibility.
- Various bug fixes in tessellation cache, quad-based spatial
split builder, etc.

- Added support for multi-segment motion blur for all primitive types.
- API support for stream of pointers to single rays (`rtcIntersect1Mp`
and `rtcOccluded1Mp`)
- Improved BVH refitting performance for dynamic scenes.
- Improved high-quality mode for quads (added spatial split builder
for quads)
- Faster dynamic scenes for triangle and quad-based meshes on AVX2
enabled machines.
- Performance and correctness bugfix in optimization for streams of
coherent (single) rays.
- Fixed large memory consumption (issue introduced in Embree v2.11.0).
If you use Embree v2.11.0 please upgrade to Embree v2.12.0.
- Reduced memory consumption for dynamic scenes containing small
meshes.
- Added support to start and affinitize Intel® TBB worker threads by passing
"`start_threads=1,set_affinity=1`" to `rtcNewDevice`. These settings
are recommended on systems with a high thread count.
- `rtcInterpolate2` can now be called within a displacement shader.
- Added initial support for Microsoft's Parallel Pattern Library (PPL)
as tasking system alternative (for optimal performance Intel® TBB is
highly recommended).
- Updated to Intel® TBB 2017 which is released under the Apache v2.0 license.
- Dropped support for Visual Studio 2012 Win32 compiler. Visual Studio
2012 x64 is still supported.