Embree

- Compile fix for Intel® ISPC v1.13.0
- Adding RPATH to libembree.so in releases
- Increased required CMake version to 3.1.0
- Made instID member for array of pointers ray stream layout optional again.

- Added support for Intel® Data Center GPU Max Series.
- Added ARM64 Linux support.
- Added EMBREE_BACKFACE_CULLING_SPHERES cmake option. The new cmake option defaults to OFF.

- Improved performance for Tiger Lake, Comet Lake, Cannon Lake, Kaby Lake,
and Skylake client CPUs by using 256 bit SIMD instructions by default.
- Fixed broken motion blur of RTC_GEOMETRY_TYPE_ROUND_LINEAR_CURVE geometry type.
- Fixed bvh build retry issue for TBB 2020.3
- Added support for Intel® Data Center GPU Flex Series
- Fixed issue on systems without a SYCL platform.

- This Embree release adds support for Intel® Arc™ GPUs through SYCL.
- The SYCL support of Embree is in beta phase. Current functionality, quality,
and GPU performance may not reflect that of the final product. Please read the
documentation section "Embree SYCL Known Issues" for known limitations.
- Embree CPU support in this release as at Gold level, incorporating the same quality
and performance as previous releases.
- A small number of API changes were required to get optimal experience and
performance on the CPU and GPU. See documentation section "Upgrading from Embree 3 to
Embree 4" for details.
- rtcIntersect and rtcOccluded function arguments changed slightly.
- RTCIntersectContext is renamed to RTCRayQuery context and most members moved to
new RTCIntersectArguments and RTCOccludedArguments structures.
- rtcFilterIntersection and rtcFilterOcclusion API calls got replaced by
rtcInvokeIntersectFilterFromGeometry and rtcInvokeOccludedFilterFromGeometry API calls.
- rtcSetGeometryEnableFilterFunctionFromArguments enables argument filter functions for some geometry.
- RTC_RAY_QUERY_FLAG_INVOKE_ARGUMENT_FILTER ray query flag enables argument filter functions for each geometry.
- User geometry callbacks have to return if a valid hit was found.
- Ray masking is enabled by default now as required by most users.
- The default ray mask for geometries got changed from 0xFFFFFFFF to 0x1.
- Removed ray stream API as rarely used with minimal performance benefits over packet tracing.
- Introduced rtcForwardIntersect/rtcForwardOccluded API calls to trace tail recursive rays from user geometry callback.
- The rtcGetGeometryUserDataFromScene API call got added to be used in SYCL code.
- Added support for user geometry callback function pointer passed through ray query context
- Feature flags enable reducing code complexity for optimal performance on the GPU.
- Fixed compilation issues for ARM AArch64 processor under Linux.
- Setting default frequency level to SIMD256 for ARM on all platforms.
This allows using double pumped NEON execution by enabling EMBREE_ISA_NEON2X in cmake under Linux.
- Fixed missing end caps of motion blurred line segments.
- EMBREE_ISPC_SUPPORT is turned OFF by default.
- Embree drops support of the deprecated Intel(R) Compiler. It is replaced by
the Intel(R) oneAPI DPC++/C++ Compiler on Windows and Linux and the
Intel(R) C++ Classic Compiler on MacOS (latest tested versions is 2023.0.0).

- Fixed bug in bounding flat Catmull Rom curves of subdivision level 4.
- Improved self intersection avoidance for
RTC_GEOMETRY_TYPE_DISC_POINT geometry type. Intersections are
skipped if the ray origin lies inside the sphere defined by the
point primitive. Self intersection avoidance can get disabled at compile time
using the EMBREE_DISC_POINT_SELF_INTERSECTION_AVOIDANCE cmake option.
- Fixed spatial splitting for non-planar quads.

- Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost.
- Fixed binning related crash in SAH BVH builder.
- Added EMBREE_TBB_COMPONENT cmake option to define the component/library name of Intel® TBB (default: tbb).
- Embree supports now Intel® oneAPI DPC++/C++ Compiler 2022.0.0