File embree.changes of Package embree
-------------------------------------------------------------------
Sat Apr 05 21:41:12 UTC 2025 - Marcus Rueckert <mrueckert@suse.de>
- Update to version 4.4.0:
* renamed test-package job, needed for creating zip with test data
* update doc
* ci: disable signing on windows for non-release build jobs
* ci: fix integrate workflow
* ci: remove unused test-package job from release workflow
* Remove unused .gitlab-ci.yml file
* removed dpcpp-version from release-linux-sycl-test job
* using libXXX.so* semantics
* updated dpcpp v6.0.0 runtime in package.cmake
* Disabled memory compression for acceleration buffer
* Ensuring that instances point to acceleration structures in device memory, not host memory.
* Removed vscode files, not part of this PR
* update compiler to intel llvm v6.0.0 for all release jobs also CPU
* update compiler to intel llvm v6.0.0 for all release jobs
* fixing host_device tutorial race condition
* fixing embree_info and embree_minimal sycl tutorials missing DL_LIBS
* disabled internal driver tests
* Moved buffer releases in host/device tutorial
* disabled internal driver tests for release
* removed bvh data host device test, not needed any more
* updated windows internal driver version
* updated windows public driver, upgraded dg2 tests to ubuntu24.04 fixed coverity issue update gfx-ubuntu24-internal/public.env for CI update linux intenal drivers fix linux release CI fix integration CI
* Fixed mistake in async buffer copies Fixed forest tutorial Skip pvc tests in nightly.yml fixed unititialized empty geometry commit segfault
* Removed redundant copies in host/device tutorial
* skip pvc tests
* removed EMBREE_SYCL_BVH_DATA_HOST_DEVICE cmake option and made it enabled by default
* little cleanup
* Fixed uninitialized memory
* Device buffers are now automatically updated on Geometry::commit()
* documentation: rtcTraversable invalidation on rtcCommit
* removed internal rtas builder, made EMBREE_SYCL_L0_RTAS_BUILDER the default
* fix issues with host/device memory API on iGPU fix blender segfault with geometry == NULL inside geometries array
* remove timing output from dynamic scene tutorial
* update internal Linux and Window driver in CI
* changelog, typos
* add CI test for BVH data in host device memory
* add export filters for commit with queue API functions
* more general documentation
* documentation for ray queries using RTCTraversable
* updating general intro documentation
* make rtcCommitSceneWithQueue work with out-of-order SYCL queues, too
* change rtcCommit???WithQueue interface to return sycl::event
* docu fixes
* revert accidental changes in ray flag settings
* rename memory types and account for host unified memory in buffer constructor
* update rtcCommitScene(|WithQueue) documentation
* update rtcSet(New|Shared)GeometryBuffer(|HostDevice) documentation
* fix wrong deallocation in multi_instance_geometry tutorial.
* update documentation for RTCBuffer
* add nightly Windows multilevel test.
* refactor alloc. remove global (tls) SYCL context and device.
* rename host_device_memory tutorial and add CI tests
* clear host scratch space for static scenes
* add cmake option for explicit host/device BVH data
* add sycl events for sychronization
* less alloc/free when creating device representation of scenen/geometry
* small fix for validation API
* add padding to tree vertex and color data buffers for safe SIMD access.
* create CPP from ISPC tutorials
* add ISPC API using RTCTraversable
* change API naming for buffers form 'Ex' to 'HostDevice'
* port forest tutorial to new host/device memory support
* port of multi_instanced_geometry tutorial to RTCTraversable interface was missing
* do not use ALIGNED_STRUCT_ macro for Scene and Geometry
* cleanups
* test address sanitizer issue
* rework buffer interface
* memory passing interface
* scene is now in device memory, too. USM shared memory is now only used in tutorials.
* add debug tutorial and restore state of other tutorials
* BVH buffer in explicitly managed host/device memory on systems without unified memory (e.g. dGPUs) geometry data works
* fix wrong assert about maxTimeSegments. fix missing sign-extension when using device memory for BVH data.
* Suppressed warnings with latest clang/sycl compiler
* Version update to 4.4.0
* internal ci: update dependencies
* remove requirement for CPU feature RDRAND
* refactor scene array for (multi-level) instancing into struct
* simplify intersect_instance and intersect_instance_array
* update dpcpp runtime dependencies for installation
* update driver versions for CI tests
* update sycl nightly version for CI tests
* Removed usage of no longer supported SYCL subgroup shuffles.
* fix public ci (#2732)
* update documentation
* update version in testing package
* update reusable workflows.
* bump version and update changelog
* Compute hwaccel adresses on the fly to prevent back-migration of BVH data.
* Fix include order causing unwanted dllexport symbols
* disable unneccessary stripping of files on Windows to WA binary signing issue.
* Update API documentation and changelog.
* Add RTC_ERROR_LEVEL_ZERO_RAYTRACING_SUPPORT_MISSING and the API functions rtcGetDeviceLastErrorMessage and rtcGetErrorString.
* fix unused variable warning
* fix sycl header path warning
* test new public Windows driver.
* update to sycl nightly-2024-06-12
* Cleaned up whitespace
* Split array of Scene* into root and instances
* Fixed duplicate SYCL context
* revert pipeline renaming
* basic public ci (#2725)
* fix typo in max_triangles_per_leaf option.
* Fix triangle splitting crash where FP error causes binning differences
* Provide a configuration setting for limiting leaf size
* use composite mode in release job for pvc testing due to switch to older dpcpp.
* clean up continouus tests and move some to nightly.
* use same compiler version for release as last time.
* remove ICX-RC tests. ICX 2024.1 is out.
* turn L0RTAS builder off for nightly validation test.
* add known issue about ISPC and macOS universal binary in changelog.
* use UMD migration for PVC for now because KMD migration is buggy.
* move function for USM prefetch into embree4.dll
* add test function dummy
* turn level zero raytracing acceleration structure build on by default.
* update tests to ICX 2024.1 compiler.
* add instruction to install intel-level-zero-gpu-raytracing to docu and changelog.
* fix misleading "USM allocation not enabled" exception thrown Tutorial destructor when a SYCL device could not be created successfully.
* update linux internal driver.
* add embree_info and embree_minimal to CI tests for debugging purposes.
* check for driver support for building RTAS properly.
* update Linux drivers.
* update Windows drivers.
* fix output operator, issue #486
* add scene flag to trigger USM shared prefetch to GPU at the end of rtcCommitScene
* update PVC tests.
* remove references to obsolete CMake variables
* more tests with sycl rc.
* use public driver for nightly PVC tests, too.
* regress to older SYCL version due to issues with getting profiling information.
* handle exception in device info querying code.
* fix dpcpp download script for new versions of the compiler
* build Windows DG2 tests with sycl-nightly, too
* fix release CI job.
* Documentation update
* version and changelog
* use sycl-5-2-0 for release binaries.
* bump cmake_minimum_required verion to 3.5
* Fix tbb macOS port name
* add static assert for proper RTCHitNt<N> alignment in verify.
* Added alignment definitions to RTCHitNt and RTCRayNt
* Fix for barycentric coordinate verification in
* remove unused image code.
* use public drivers for PVC, too.
* fixes for current sycl nightly and some warning fixes
* add tests for syclos release candidate.
* add macOS universal binary build test.
* update to TBB 2021.12.0
* Build for ARM on Apple silicon machines by default when CMAKE_OSX_ARCHITECTURES is not set.
* Enabled support for Universal Binaries on macOS
* update drivers enable l0rtas in continuous and nigtly tests
* Do/upgrade outdated os tests (#2711)
* add token permissions for github workflows.
* add buildbench to performance CI pipeline
* add SYCL version of buildbench
* fine tuning google benchmark
* more precise perf-CI measurements
* Do/scripts cleanup (#2701)
* better output of which ISPC executable is found/used
* test ispc 1.23.0
* Do/performance regression tests (#2699)
* remove printf
* test AVX512 compilation on x86 macOS in continuous test.
* Run Embree in AVX512 mode on AVX512 capable x86_64 machines running macOS.
* compile embree_info also for ARM.
* print warning about "flush to zero" and "denormals are zero" only on x86 systems
* fix missing backslash for overwriting EMBREE_TESTING_INTENSITY
* fix path for macos-arm binary analysis and antivirus scan (#2692)
* allow overriding of EMBREE_TESTING_INTENSITY for tests
* update documentation
* apply internally needed patch to glfw only when explicitly stated.
* use same self-compiled tbb on macOS for x86_64 and arm64 versions.
* update drivers
* some issues with nightly tests
* add bdba and virus scan for arm binaries, too
* use TBB compiled for NAS for ARM release package
* fix find command in package test check
* docu
* increase test epsilon for Apple Clang 14 on macOS ARM
* add missing tests of release packages
* some fixes.
* different names for x86 and arm macOS packages
* update to TBB 2021.11 and ISPC 1.22
* bump embree version to 4.3.1
* add NEON and NEON2X tests to continuous jobs again.
* update sycl-nightly and ICX RC
* Do/macos notarization update (#2684)
* removed AOEUAOEU string from integration test exe (#2683)
* Do/devel (#2682)
* add dependloadflag also for msvc
* Fail when no tests are executed.
* Do/fix macos test execution (#2677)
* updating PVC deviceID check
* fix artifact name
* move Windows DG2 ICX tests to continuous again.
* update docu regarding supported compilers.
* update Windows public driver version again.
* remove Windows release driver version. We use official version now.
* drop ICX versions from 2023
* remove EMBREE_LEVEL_ZERO CMake variable from presets
* disable ARM emultation tests for now because runner seems to be broken
* cleanups
* fixes for nightly tests
* revert public Windows driver
* add DG2 Windows tests with public and release drivers in continous CI, too
* added back notarization with TODO comment to mac tests
* make notarize.sh executable
* update to correct icx version
* add missing feature flag checks
* Do/release macos extern notarize (#2670)
* Run Windows tests in CI again
* Do/driver update (#2668)
* Do/threadsanitizer issue (#2667)
* fix dll dependency location for ICX 2024
* add dependent load flag to executables and shared libraries.
* test intel-llvm/nightly-2023-12-18
* update continouus ICX tests to 2024.0 and remove RC tests.
* catching exceptions in rthwif tests
* adding compare threshold to cornell_box test
* windows testing ICX 2023.2.1 (#2663)
* using force-delete for windows CI (#2662)
* fix nightly L0RTAS typo error (#2661)
* Sh/simd sse2neon 1.6.0 (#2660)
* fix coverity, upgrade all CI cmakepresets, accelerate continuous (#2659)
* add nightly test for validation api
* EMBREE_SYCL_RT_VALIDATION_API=ON and EMBREE_SYCL_IMPLICIT_DISPATCH_GLOBALS=OFF works again
* add ze_api header to repository directly.
* WA compiler issues under Windows with -D_DEBUG.
* make rthwif tests compatible with TBB 2023
* fix static linking issues with ze_wrapper library
* user-defined thread count precedence
* moving most Windows DG2 tests to nightly testing
* enabling usage of L0RTAS builder in CI under Windows
* requiring RTAS extension when enabling EMBREE_SYCL_L0_RTAS_BUILDER
* updating gfx-driver
* Add missing EMBREE_GEOMETRY_??? to embree-config.cmake
* updates
* add old include/lib directory structure for ICX 2024, too.
* add also weird libsycl.so version (libsycl.so.7.0.0-8 ?!?!?)
* add pi_win_proxy_loader.dll to release package
* update documentation
* use rockylinux for release and split linux sycl
* update compilation docu and version number
* add instance array geometry type.
* disable nightly bezier_round furball tests on 2 windows configs for now
* add tests for sycl-nightly RK version
* Update changelog.md
* add ICX release candidate tests
* WA for changed sycl include headers in oneAPI DPC++ 2024.0
* validating that rtasFormat is not ZE_RTAS_FORMAT_EXP_INVALID
* added argument to set resolution support for cornell_box test
* integration of coverity
* disabling L0 RTAS builder by default
* disabling L0 rtas builder only for release
* added cmake option to enable usage of L0 rtas builder
* properly checking if rtas extension can get loaded
* not using rtas_builder extension by default in rthwif tests
* not using rtas_builder extension by default yet
* removed unused file
* fixed deviceID check for PVC
* cherry pick from carlocab (#2623)
* do not sign the zip package after build for windows any more (#2624)
* update to intel-llvm 20230830
* remove using sycl::fmax and using sycl::fmin
* Do/update compiler cmakepreset (#2621)
* use ubuntu 20.04 for release again and update release driver
* do not source dpcpp environment manually Please enter the commit message for your changes. Lines starting
* consistently using main version of reusable workflow
* Revert "use DPCPP_SETUP_SCRIPT environment variable in test.py script"
* updated graphics drivers in CI
* use DPCPP_SETUP_SCRIPT environment variable in test.py script
* CI-perf sycl build on Ubuntu 22.04 (#2617)
* docker GPU image for perf CI to embree/ubuntu:22.04 (#2615)
* cleanups to sycl namespace usage
* fixed sub_group related compile warning
* - updating to dpcpp compiler sycl-nightly/20230724 - fixing compile issues in rthwif tests
* rthwif tests use RTAS build extension only when available
* using L0 RTAS build extension only when available
* implemented varying version of rtcGetGeometryTransform for ISPC
* matching rcp math between rthwif_test and BVH builder
* Add support for ARM64 windows platform cmake
* added configuration to run rtas builder tests using internal or level zero implementation
* Do/update driver (#2607)
* add memory monitor to GPU BVH build
* fix imgui empty label complaint on Windows
* adding check if rtas extension present and properly initialized
* checking ze_rtas_builder_procedural_geometry_info_exp_t reserved member to be 0
* Do/release (#2601)
* Do/release (#2597)
* fix typo (#2592)
* removed extra comma (#2591)
* setting version to 4.2.0
* SYCL version of Embree with GPU support is no longer in beta phase.
* updated changelog
* Added PVC runner to perf CI (#2588)
* clarified that rtcOccluded and occlusion filter cannot get used to gather next n hits
* Added rtcGetGeometryTransformFromScene API function that can get used inside SYCL kernels.
* Added CI feature to store results in database (#2584)
* Do/fix release package name sycl (#2583)
* Do/release test package (#2579)
* passing dispatch globals for through debug RTAS build API extension
* fix to get dispatch globals allocation working again for debugging purposes
* New model repository address (#2575)
* Improved BVH build performance on many core machines by avoiding spin locks.
* using zeDeviceGetProperties to query RTAS device properties
* - using new rtas API header - properly returning ZE_RESULT_EXP_RTAS_BUILD_DEFERRED and ZE_RESULT_ERROR_HANDLE_OBJECT_IN_USE error codes
* adding RTASBuilder support to ze_wrapper
* moving ze_rtas.h to level_zero folder
* using decltype to get function types in ze_wrapper
* init ZeWrapper in rthwif tests.
* fix rthwif tests with new level zero runtime loading change.
* adding MTL support
* fixing experimental disable of deviceID check
* fix rthwif tests for test package (#2567)
* load ze_loader library during runtime with dlopen/LoadLibrary. Remove EMBREE_LEVEL_ZERO CMake variable. It is used now by default.
* Do/integration (#2562)
* adding Impl to RTAS builder to have separate symbols to API function names
* added filter files for test execution (#2541)
* Do/fix fopen (#2558)
* replace insecure fopen with secure std::ifstream (#2557)
* using ZE_RESULT_EXP_ERROR_RETRY_RTAS_BUILD error code to re-try rtas build operation
* removed zeRTASInitExp/zeRTASExitExp functions
* checking that number of geometries is in range
* prepared L0 rtas builder to support multiple rtas formats
* limiting triangle pair primID delta to 5 bits
* L0 ray tracing support tests can use rtcore SW simulation
* Do/release (#2554)
* rebase release branch into devel (#2553)
* Do/release (#2551)
* changed package filenames (#2550)
* adding always_inline for Embree SYCL API functions
* make EMBREE_NO_SPLASH a proper CMake option.
* making ze_rtas_builder_geometry_exp_flag_t opaque by default
* updated to latest L0 rtas build API spec
* checking if rtas allocation is in 48 bit address range
* using proper ZE_STRUCTURE_TYPE_RAYTRACING_MEM_ALLOC_EXT_DESC stype for rtas allocations
* - changed version to 4.1.0 - added support for Intel(R) Data Center GPU Max series to changelog - removed limitation of support to specific Arc/Flex version, all products are supported now. - updated DPC++, driver, and compute-runtime versions to use
* returning ZE_RESULT_ERROR_INVALID_ENUMERATION error when stype is not as expected
* properly checking for stype of builder properties and descriptor
* adding Level Zero rtas header from spec
* updates to compile Level Zero ray tracing builder API to match spec
* using argument structure in bounds callback for Level Zero ray tracing build API
* using standard Level Zero results in Level Zero rtas build API
* removing ze_raytracing CI tests
* Do/remove macosx embree tag (#2543)
* cleanups to cornell_box rtas build example
* only including rttrace.h
* renamed rtbuild and rttrace API header files
* returning proper error codes in rtas builder
* adding flags to rtas builder properties
* adjusted Level Zero ray tracing Build API struct member names to match spec
* adjusted Level Zero ray tracing Build API function names to match spec
* implemented rtas device properties query for L0 ray tracing build API
* adjusted property query for parallel operation object in L0 ray tracing build API
* removed not required update of build arguments
* adopted build and size query functions of L0 ray tracing build API
* introduced rtas builder concept to L0 ray tracing API
* using L0 ray tracing API to build acceleration structure for cornell_box test
* * Added CMake options for sphere and curve backface culling * Moved sphere backface culling documentation to the right place
* Added EMBREE_BACKFACE_CULLING_SPHERES option
* do not use exception mechanism for BVH build retry due to TBB 2020.3
* update dpcpp to 20230417 and fix deprecation warnings
* fix config error in generated embree-addtest.cmake (#2536)
* fix tarball package unpacks into single folder (#2535)
* using proper RT allocation function to allocate RTDispatchGlobals
* fixed issue that caused allocation of RTDispatchGlobals by default
* build a deliverable testing package (#2532)
* fixed success job
* updates to compile ze_raytracing builder in isolation in separate repository
* properly destroying parallel build operations
* ci: update gfx-driver internal to ci-neo-025905
* all names of ze_raytracing build API implementation match spec
* fixed interface of some ze_raytracing API functions
* implemented zeRaytracingAccelFormatCompatibilityExt in ze_raytracing build API
* implemented build flag to skip duplicate any hit shader invokation
* added stype and pNext to ze_raytracing build API implementation descriptors
* removed rthwifGetSupportedFeatures from ze_raytracing build API
* introduced triangle format, quad format, and vertex format in ze_raytracing build API
* added arm support to changelog and doc (#2524)
* added EMBREE_ISA_NEON2X to embree-config.cmake (#2526)
* Minor fixes to performance CI (#2525)
* Use all models from repo
* added arm emulation testing to nightly (#2520)
* turning EMBREE_LEVEL_ZERO on in one CI test
* fixed success CI task
* update doc and fix intelstyle repo url
* disable furball hermite round test for now.
* use cmake --build instead of make in package scripts
* don't use -no-intel-lib along with -static-intel on Linux with SYCL enabled
* add imf-use-svml=false even though it seems to not work
* set next Embree version to 4.0.1
* make EMBREE_LEVEL_ZERO a proper CMake option
* add frexp implementation for DPC++ with -no-intel-lib
* changelog
* try no-intel-libs again along with static-intel flag
* - rthwif can get build standalone - creating libze_raytracing packages in CI
* Intel(R) Data Center GPU Flex Series is supported now
* update drivers
* add EMBREE_NO_SPLASH enviroment to disable beta splash screen (#2504)
* update sycl-nightly to 20230304
* Increase compare threshold for known issue with Apple AVX2 linear curves (#2508)
* Revert "disabling spatial splits in rthwif builder"
* use cmake --build instead of make to be able to use Ninja
* use Ninja for DPC++ builds
* fixed issue with EMBREE_BUILD_GOOGLE_BENCHMARK_FROM_SOURCE
* fix renamed benchmark target name for ispc tutorials
* fix struct/class redeclaration warning
* clean up google benchmark CMake cache variables
* download and build google benchmark from source
* rthwif builder no longer throws RTCError exception
* introduced separate priminfo_mb.h file and moved primref into builders folder
* disabling spatial splits in rthwif builder
* separated rtbuild and rttrace in rthwif
* removed Embree specific code out of rthwif
* Do/improvetesting (#2499)
* workaround for accidentally installed glfw (setting directory properties only works for CMake >= 3.22)
* don't install embree_info source code to src folder in release package
* prevent creation of default sycl device to make CPU versions of tutorials runnable on systems without available GPU drivers
* - added usage example to L0 rt build API - added zeRaytracingAccelCompatibilityExt function to L0 rt build API - added device input to ze_raytracing_build_accel_ext_desc_t descriptor
* ci: add ATS-M testing job
* fix release sycl linux build
* Fixed a bug that limited the size of meshes supported in RT hardware BVH builder.
* changelog
* add continuous sycl test with TBB 2020.3
* fix rthwif build retry for TBB 2020.3
* added support for RT software simulation for validation purposes
* Frei/devel (#2489)
* Updates to L0 ray tracing build API - properly naming flag enum and introducing flags uint - renaming ze_raytracing_accel_size_ext_desc_t to ze_raytracing_accel_size_ext_properties_t - introducing ze_raytracing_format_ext_t and specifying format for triangle, quad, and vertex buffers - removed ZE_RAYTRACING_BUILD_EXT_FLAG_DYNAMIC
* fix line.gather with time interpolation causing inconsistent sphere cap intersections
* calculating improved epsilon for triangles and quads in pathtracer
* fix small errors in test.py
* fix set but unused warning in tinyexr
* fix sycl deprecation warning
* update release jobs, too
* reintroduce setup_dpcpp and setup_gfx stages again
* update to sycl-nightly/20230209 and clean up .ci-env
* initialize debug data with time if --time is passed
* fix linux-PVC-2T job to use docker_gpu
* Remove core.autocrlf from CMake patch command
* set autocrlf to fix glfw patch issues in new reusable workflows (thanks @Krzysztof!)
* use Trevors GHA for installing drivers on-the-fly.
* move 2 continuous mac tests to nightly
* throwing error when build hardware BVH is too deep
* further updates to L0 ray tracing build extension
* fixed issue with winding order in perfect watertight ray/triangle intersector
* - Using proper naming convention for L0 ray tracing build extension - Adding Doxygen support to documentation of L0 ray tracing build extension - Updating documentation of L0 ray tracing build extension
* added back missing check_arm_neon.cpp file
* Improved performance for Tiger Lake, Comet Lake, Cannon Lake, Kaby Lake, and Skylake client CPUs by using 256 bit SIMD instructions by default.
* rthwif already uses resize, which allocates exact for static scenes, and may over allocate for dynamic scenes that increase in size
* Resize of empty embree vector allocates exact size to match std::vector behavior.
* ci: switch internal driver to ci-neo-master-025593
* add missing .so files for Linux CPU only version.
* allocating proper BVH buffer size
* using ZE_RELAXED_ALLOCATION_LIMITS_EXP_FLAG_MAX_SIZE mode to allocate large BVH buffers with L0
* throwing Embree error when BVH memory allocation failed
* properly initalizing pNext member of ze_raytracing_mem_alloc_ext_desc_t
* Improved performance for Tiger Lake CPUs.
* setting version to v4.1.0
* without -no-intel-lib, svml_dispmd.dll is also needed for CPU only version.
* turn off no-intel-lib for now. this causes ICX compiler bugs.
* Fix build error in rtcore_common.isph
* update GLFW to latest version
* some fixes in embree-config.cmake for static Embree build.
* add CMake variable to control building GLFW from source
* patch DLL loading path in ImGUI.
* download GLFW from source and patch DLL loading path.
* add vcruntime and msvcp DLLs when packaging Embree.
* clean up search for SYCL dependencies.
* also add tbbmalloc.dll/.so when packaging Embree
* WA on Windows when using -no-intel-lib.
* Windows debug mode WA for std::vector operator[] access in SYCL kernel.
* cleanup of source_sycl_env.sh
* disable statically linked TBB for rthwif builder by default.
* Updated beta statement and regenerated documentation.
* fixed interchanged windows/linux build artifacts in release.yaml
* regenerated documentation
* Added messages what Embree SYCL version is still in beta phase.
* Fix docu with regard to setting embree_DIR path correctly
* remove unused device parameter from test.py invokations
* remove iGPU enabled problem from known issues
* add "threads per EU" as criteria to filter out non RTHW cards and add max num of EUs as device selection score
* add embree_info tutorial and printSYCLDevice functionality
* add creation of startup.bat script in download_dpcpp.sh
* Spelling compilation.md
* Perf-CI compiler change to ICX (#2451)
* small changes in release package names
* add note about deprecated Intel Compiler being not officially supported anymore
* Fix alignment problem in ray mask tutorial
* Turn ispc support off by default
* Added third-party-programs file for Intel(R) oneAPI DPC++/C++ compiler
* Add reference to minimal CMake version in compile docu.
* update supported compilers and remove MAX_ISA from cmake config example
* rename RTC_FEATURE_FLAG_RAY_MASK to RTC_FEATURE_FLAG_32_BIT_RAY_MASK
* modify minimal_sycl tutorial to match documentation wrt specialization constants
* set maximum glibc version to 2.28 on Linux CPU release
* Add documenation for compilation with the Intel(R) oneAPI DPC++/C++ compiler.
* change kernel.rc encoding to UTF-8
* fix memleak in ray mask tutorial
* add explanation of the RTC_FEATURE_FLAG_RAY_MASK flag
* add test for ray mask tutorial
* modify ray tutorial to pass klockwork scan
* fix SYCL code when ray masks are turned off.
* execute trav loop when ray masks are used.
* add ray mask tutorial
* libmmd is needed on Windows with ICX in CPU only mode
* don't link against DPC++ intel libs when SYCL support is off.
* use oneTBB 2021.7.0 for mac release
* remove WA for old sycl standard of ICX
* windows and linux release use ispc 1.18.1 and tbb 2021.7.0
* cleanup unused configuration options
* remove unneccessary --include-intel-llvm
* turn off vsync in tutorial
* clean up continuous test names
* test tbb 2021.7.0 and ispc 1.18.1
* update package_linux.sh script (changes cherry-picked from Raszkowski, Krzysztof)
* fix Linux SYCL packaging for ICX 2023.0.0
* fix test script invokation on ubuntu 22.04
* update glibc requirements for Linux CPU build (ICX2023.0.0 requirements)
* update continuous and nightly tests and replace ICC with ICX
* use Rocky Linux 8.7 for Linux CPU release (req. of ICX 2023.0.0)
* fix artifact names in SYCL releases
* also package svml_dispmd.dll and libmmd.dll form the DPC++ redistributables on Windows
* remove dpcpp specific package script for Windows
* do not copy ze_loader library/dll. it is in the driver.
* Windows now uses ICX and Ninja for release build package
* update failing nightly tests to use ICX 2023.0.0
* use correct stack protection flag on Linux for ICX
* update release builds to use ICX 2023.0.0 and TBB 2021.5.0
* move debug linux ICX test into nightly (takes too long) and change build type to release
* new perf CI runners, added asia_dragon workload (#2440)
* add tests for oneAPI DPC++ 2023.0.0 compiler
* Added simple performance measurements to CI (#2435)
* added cmake option EMBREE_RTHWIF_TBB_GIT_REPOSITORY to set git repository location of TBB
* added cmake option EMBREE_RTHWIF_TBB_URL to set URL to TBB version to use
* bugfix linei missing sphere cap use select for linei leftExists filter
* added rudimentary fur ball to scenegraph/xml_loader
* read/write images with tinyexr and stb_image
* fix imgui event handling
* compiling DPC++ release builds with all ISAs enabled
* added list of supported Arc GPUs
* updated documentation to DPC++ compiler 2022-12-14
* increasing testing intensity of DPC++ release builds to 4
* added SYCL windows release build with JIT mode
* fixing path to SYCL libraries when creating package under Windows
* fixing imgui compile issues under macOS with ICC
* bugfix of SYCL compilation when AVX enabled
* referencing embree_rthwif-targets.cmake in embree-config only when Embree compiled to static library
* updates to device ID detection
* make oneAPI DPC++ compiler (ICX) work with EMBREE_SYCL_SUPPORT on Windows
* add CMake option to disable direct usage of level zero
* some fixes for oneAPI dpcpp compiler. still WIP.
* ICX works again (without EMBREE_SYCL_SUPPORT) under Windows
* clean up dpcpp.cmake on Windos and fix compile warnings
* prevent warnings on Windows caused by TBB, and use of getenv and fopen
* add stack protector option to dpcpp on Windows
* removed RTC_ERROR_UNSUPPORTED_GPU error code as not required
* update sycl-nightly to 20221214
* update ImGUI
* Revert "switching VS studio tests <= 2017 to release mode due to long path problem."
* using proper SYCL device and context to allocate ray tracing hardware BVH
* bugfix to rtcSYCLDeviceSelector API function
* building Embree SYCL release on Ubuntu 20.04
* removing level=2 workaround in intersection_filter_sycl, seems no longer required
* matching intel_ray_flags_t layout for RTCRayQueryFlags
* - renamed RTCIntersectContext to RTCRayQueryContext - renamed RTCIntersectContextFlags to RTCRayQueryFlags
* updates to API documentation
* Upgrading to gfx-driver-ci-master-12953, ci-neo-master-025103 in CI
* switching VS studio tests <= 2017 to release mode due to long path problem.
* prefix all executables with embree_
* setting version to v4.0.0
* updated Embree version from 3 to 4 in man pages
* embree_rthwif is static library now to not have additional shared libraries in release package
* renamed embree_sycl.a to embree4_sycl.a
* renamed RTC_FEATURE_FLAGS_ to RTC_FEATURE_FLAG_ to be consistent with other API flag namings
* fixed download link, properly installing rthwif
* - passing DPC++ include path is not required for -fsycl compilations - cleanups to minimal tutorial CMakeList.txt
* using gfx-driver-ci-master-12917 in CI and disabling implicit scaling workaround
* updates to documentation
* added documentation of rtcGetSYCLDeviceFunctionPointer
* added documentation of rtcSetGeometryEnableFilterFunctionFromArguments API function
* documented rtcNewSYCLDevice, rtcIsSYCLDeviceSupported, rtcSYCLDeviceSelector, and rtcSetDeviceSYCLDevice API functions.
* enabling rthwif testing also under Windows
* updated docu link to compute runtime, documenting need to pass path to DPC++ headers
* removed need of using special linker and -L on Windows from documentation
* documentation updates
* introduced RTC_INTERSECT_CONTEXT_FLAG_INVOKE_ARGUMENT_FILTER to enforce argument filter invokation for each geometry
* use lld-link which is now shipped with dpcpp.
* turn EMBREE_SYCL_SUPPORT off on Windows when DPC++ compiler is not used.
* using releases_22ww49-ci-master-12886 windows driver in CI
* enabling curves in DPCPP release builds
* removed some workaround that disable CI tests
* using neo-releases-22.43.24595.35 NEO release in CI
* removing rt_validation_api:OFF from CI, validation API is OFF by default
* upgrading CI to gfx-driver-ci-master-12886 and disabling ZEBinary workarounds
* upgrading CI to ci-neo-master-025028 and removing ZEBinary workarounds
* disabing benchmark output for rthwif_test_builder_instances CI tests
* marking most SYCL cmake options as advanced
* removed EMBREE_SYCL_MBLUR cmake option
* changes to upgrading from Embree 3 to Embree 4 section as will as changelog
* updates to section Embree SYCL known issues
* documented that JIT compilation is preferred
* updates to CMake configuration docu section
* updates to Building Embree SYCL Applications section
* added optimization section on generic pointers
* documented RTCFeatureFlags
* updated documentation of rtcIntersect1/4/8/16 and rtcOccluded1/4/8/16
* updates to Embree SYCL API documentation
* enabling pre-splits in rthwif BVH builder only in high quality mode
* checking if ray tracing extension supported by L0
* enabled geometry callbacks for some CPP tutorials for testing purposes
* added rtcSetGeometryEnableFilterFunctionFromArguments API function
* multiscene_geometry tutorial uses proper feature mask
* regenerated CPP tutorials from ISPC version
* updates to fix ISPC to CPP tutorial conversion
* - updating to ci-neo-master-024996 and gfx-driver-ci-master-12868 in CI - enabling production ray tracing API in CI tests - enabling production ray tracing API by default in cmake - switching Windows SYCL CI tests to AOT mode
* introduced RTCOccludedArguments to API
* optimizations to intersection_filter tutorial
* removed _IN_GEOMETRY and _IN_ARGUMENTS cmake options as no longer required
* fixing issues with using Embree install package
* disabling SYCL by default
* updating to neo-releases-22.43.24595.34 in CI
* rtcSetDeviceSYCLDevice does not check if device is supported by Embree as user may not necessarily run on that device
* update sign file location
* added missing dependency from embree_rthwif to embree_rthwif_sycl
* added rtcSetDeviceSYCLDevice API call
* disabling geometry callbacks by default in SYCL only
* using ci-neo-master-024982 in CI
* adding rthwif_production into embree_rthwif target
* avoiding dispatch globals allocation when implicit dispatch globals are used
* marking TBB cmake variables as advanced
* some DPC++ to SYCL replacements in docu
* updating Linux DPC++ compilation and driver install instructions
* added L0 code to query if RT hardware supported, still kept disabled due to issues
* printing render frame rate and total frame rate in window
* setting device revision again for DG2 in AOT mode as a temporary workaround
* measuring frame rate displayed in the application window on the device
* added benchmark mode for dynamic_scene tutorial
* no longer passing arrays through TutorialArguments for quaternion_motion_blur tutorial
* no longer passing arrays through TutorialArguments for motion_blur_geometry tutorial
* No longer passing arrays through TutorialArguments for user_geometry and instanced_geometry tutorials. Using USM allocations for these arrays significantly improved performance.
* making accumulation buffer read/write
* - using SYCL JIT compilation by default - using EMBREE_CI_DEVICE to determine GPU test is run on - removing EMBREE_SYCL_AOT_DEVICE_REVISION cmake variable, no longer required
* Setting default frequency level to SIMD256 for ARM on all platforms.
* Refactoring of CI testing: - mainly testing JIT in continuous tests - testing internal an public release driver in continuous tests - moved some less important tests to nighty testing
* - building releases with public gfx driver - building releases in JIT mode - DEFAULT gfx version renamed to INTERNAL
* using neo-releases-22.43.24595.29 as public gfx driver in CI
* using gfx-driver-ci-master-12830 in CI
* removed geometry filter invokation from quaternion_motion_blur tutorial
* - renamed rtcFilterIntersection to rtcInvokeIntersectFilterFromGeometry - renamed rtcFilterOcclusion to rtcInvokeOccludedFilterFromGeometry
* rtcFilterIntersection and rtcFilterOcclusion only invokes geometry filter function
* adding instID argument to rtcForwardIntersect/Occluded functions
* fixed compile warning
* rtcForwardIntersect/Occluded do not need to handle NULL intersect arguments
* using default RTCIntersectContext whenever possible
* moving RTCIntersectContext to RTCIntersectArguments
* update Ubuntu docker to 22.04 for all SYCL/GPU builds
* implemented argument filter function support in pathtracer, keeping filter functions disabled
* - disabling filter function in pathtracer - disabling pathtracer tests that require filter function - enabling pathtracer testing in JIT mode
* removed OBJ intersection filter from pathtracer to simplify filter callbacks
* removed unused intersectionFilterReject function from pathtracer
* motion blur tutorial supports argument intersect function
* quaternion motion blur tutorial uses argument intersect function
* updated overview and platforms documentation sections
* optimization to next_hit tutorial to allow inlining of filter callback
* - hair_geometry tutorial uses argument version of filter function - improved performance of hair_geometry tutorial by stopping shadow rays when transparency close to 0
* not using open-linux driver for release builds for now
* also using open-linux-driver-ci-dev_igc-12785 in CI
* enabling ray masking by default in cmake
* bugfix on ray mask test in verify
* updating CI to gfx-driver-ci-master-12767
* bugfix on ray mask test in verify
* renamed IN_CONTEXT -> IN_ARGUMENTS for cmake variables and feature flags
* documented that an Embree SYCL device can only get used on the GPU
* moved invokation of CPU and GPU builder to separate functions in scene.cpp
* removed unused is_build variable from Embree scene
* Removing stream tracing API
* rtcFilterIntersection and rtcFilterOcclusion invokes geometry and context filter function
* using faster copy of org and dir for rtcForwardIntersect/Occluded
* - documented rtcForwardIntersect1/4/8/16 and rtcForwardOccluded1/4/8/16 - ray forwarding only forwards ray origin and direction
* passing SYCL context and device objects directly to rtcNewSYCLDevice
* renamed RTC_FEATURE_ to RTC_FEATURE_FLAGS_ in Embree API
* using gfx-driver-ci-master-12725 in CI
* Changes required for Windows ARM64
* bugfix to user_geometry tutorial when compiled without context user geometry callbacks
* using gfx-driver-ci-master-12700 in CI
* update mac jobs to at least TBB 2021.1.1
* macOS jobs use the "build" tag.
* fixed in export map files due to slightly changed name mangling in clang when using Embree API namespace
* passing context and device to rtcNewSYCLDevice
* removed RTC_SYCL_KERNEL makro as no longer required
* rtcForwardIntersect pushes instance ID
* user_geometry tutorial pushes instance geomID
* renamed rtcGetSYCLFunctionPointer to rtcGetSYCLDeviceFunctionPointer
* added rtcSYCLDeviceSelector API function
* renamed RTC_ERROR_UNSUPPORTED_DEVICE to RTC_ERROR_UNSUPPORTED_GPU
* removed workaround that avoided passing function pointer types as kernel argument
* removed unnecessary casting to user geometry and filter callback types
* fixed compile issue when user geometries were disabled
* using proper argument types for API callback type definitions
* - passing RTCIntersectContext to rtcIntersect and rtcOccluded functions - removing rtcIntersectEx1 function - moving all RTCIntersectContext members to RTCIntersectArguments, except instance stack
* added -cl-intel-greater-than-4GB-buffer-required flag to rthwif tests as bug workaround
* disabling some IGC workarounds
* using gfx-driver-ci-master-12316 in CI
* geometry user callback and geometry filter callback can get properly disabled using feature mask
* added support to use production version of ray tracing API
* removed ARM detection using TRY_COMPILE, no longer required
* add rthwif CMake export
* Section title format fix for "Embree Support and Contact"
* fixed compile warning
* Fixed compilation issues for ARM AArch64 processor under Linux.
* Set `EXCLUDE_FROM_ALL` on the tbb_static build dir to not install it
* enabling intensive JIT testing for DG2
* embree_rthwif can get compiled stand alone
* using new allocator in rthwif builder implementation
* - adding --frames <int> command line option and testing triangle_geometry_sycl with many frames - enabling triangle_geometry_sycl tutorial in PVC CI again
* disabling windows ICX test temporarily
* Frei/devel (#2309)
* disabling windows ICX test temporarily
* fix cmake_minimum_required position
* running release CI job only nightly
* Fix spelling mistake
* added ZE_RESULT_RAYTRACING_OPERATION_DEFERRED result to rthwif_builder_spec
* first draft of L0 version of rthwif_builder API
* removing SYCL dependency from rthwif_builder API
* making TBB content fetching verbose
* fixed compile warning related to specialization constant usage
* made rthwif_production implementation independent to embree specific headers
* Improvements and bugfixes to scene based feature mask selection - scene hierarchy is traversed completely - curves and point s sub-types are detected properly - ISPC crash fixed - pathtracers properly disables specialization constant usage in AOT mode
* Viewer and Pathtracer are now determining their specialization constants based on the scene
* renamed EMBREE_DPCPP cmake variables to EMBREE_SYCL
* rthwif_test: adding scratch and accel buffer bounds checking
* rthwif_test: adding multi-threading
* rthwif_test: adding benchmark mode
* rthwif_build: returning new size estimate from failed build
* rthwif_build: returning acceleration structure size
* rthwif_builder: using estimateSize function from SAH builder
* rthwif_test: adding random NULL geometries for testing
* rthwif_test: adding test of worst case accel bytes
* removing EMBREE_DPCPP_IMPLICIT_DISPATCH_GLOBALS define and added EMBREE_DPCPP_ALLOC_DISPATCH_GLOBALS define
* using intel_ray_query_abandon
* Using ray tracing allocation desciptor (ze_raytracing_mem_alloc_ext_desc_t) to allocate acceleration structure data.
* fixed to rtcore_scene.h
* rthwif_build: builder uses scratch buffer
* rthwif_build: matching implementation header with specification
* rthwif_test: using device_read_only USM allocations and increased test size
* rthwif_test: added mixed geometry type BVH build test
* rthwif_test: added instance geometry BVH test
* rthwif_test: added procedural geometries BVH build test
* rthwif_test: generalized calculation of expected output
* rthwif_test: testing also meshes that cannot get quadified
* rthwif_test: random triangle selection
* rthwif_test: storing triangle ID as 4th triangle index
* rthwif_test: adding first BVH build test
* rthwif_test: moved expected output generation to separate function
* introduced instance transformation formats in rthwif build API
* rthwif build API uses bounds pointer for instances to avoid read from hardware accel in top-level build
* rthwif builder performs check of geometry descriptors
* Add --compare argument to rthwif_cornell_box test
* Fix download gfx_driver & dpcpp scripts
* added EMBREE_DISABLE_DEVICEID_CHECK env option to disable deviceID check
* enabling spill compression disable workaround again
* upgrading CI to DPC++ compiler 20220914
* using sycl::gpu_selector_v for newer DPC++ compiler versions
* updates to rthwif_builder_spec, improved transformation layout spec and naming
* running release CI workflow only in nightly build
* building separate mblur BVHs over static and mblur geometry
* removing internal rthw BVH builder that was directly connected to Embree scene
* using statically build TBB in rthwif_builder implementation
* implemented rthwif builder thread join interface
* enabling binary rthw BVH build interface by default
* Disabling Windows DG2 CI test for internal tasking system to get faster CI turn arounds.
* fixed bug with invalid instance bounds in rthw BVH builder
* enabling implicit dispatch globals usage by default
* using default sub group size for kernels
* running continuous workflow no longer on pull requests
* Extensions to rthwif build API: - adding format for transformations - adding list of acceleration structures to be intersected sequentially - adding parallel build operation for parallelization
* enabled CI test for binary HW BVH build interface
* introduced EmbreeHWAccel struct to access list of HWAccels
* upgrading to CI to gfx-driver-ci-master-12316
* fixed include paths to parallel_for.h
* removing spill compression disable workaround
* ci: leave only benchmark jobs in gitlab (#2270)
* remove inplace version of RTHWIF_GEOMETRY_TYPE_INSTANCE
* updates to rthwif_test
* rthwif_builder uses new procedural callback that returns array of bounds
* renames to make rthwif_builder match spec
* documented accelBufferBytesOut semantics when build succeeds/fails
* further changes to rthwif_builder_spec
* added improved and documented rthwif_builder_spec
* Make SYCL tutorials work when iGPU is present:
* adding vcpkg install instructions also to docu source files
* Add vcpkg installation instructions
* fixed bug in hardware BVH builder when geometries where NULL
* using internal gfx-driver for DPC++ release build for now
* running nightly release build in github at 00:00 UTC
* setting JIT and AOT compile options
* increased SYCL test timeout to 50s
* disabling public gfx driver tests
* removed no longer required disabling of EU fusion workaround
* running nightly github CI at 00:00 UTC
* added cmake option to generate dbg info in dpcpp kernels
* ci: add github actions workflows
* using high instance mask bit in instance_geometry tutorial
* fixed instance mask bug with high mask bits were used
* Enabled trav_loop always when instances are enabled to fix issues with high instance mask bits set.
* getting rt_production API in sync with spec
* removed bvh_id from ray query and ray forward intrinsics
* removed bvh_level from intel_ray_query_forward_ray intrinsic
* properly using intel_ray_flags_t type in rt_production API
* Using some more descriptive names for some rt_production API functions: - intel_get_hit_barys -> intel_get_hit_barycentrics - intel_get_hit_geomID -> intel_get_hit_geometry_id - intel_get_hit_primID -> intel_get_hit_primitive_id - intel_get_hit_primID_triangle -> intel_get_hit_triangle_primitive_id - intel_get_hit_primID_procedural -> intel_get_hit_procedural_primitive_id - intel_get_hit_instID -> intel_get_hit_instance_id - intel_get_hit_instUserID -> intel_get_hit_instance_user_id - intel_get_hit_triangle_verts -> intel_get_hit_triangle_vertices
* renaming intel_ray_desc_t members O/D to origin/direction in rt_production API
* removed bvh_level from intel_ray_query_init instrinsic in rt_production API
* changed enumerants in rt_production API to be lower case
* Renamed structs in rt_production RT API: - RayFlagsINTEL -> intel_ray_flags_t - RayDescINTEL -> intel_ray_desc_t - ray_query_INTEL -> intel_ray_query_t - raytracing_acceleration_structure_INTEL -> intel_raytracing_acceleration_structure_t - float4x3_INTEL -> intel_float4x3 - HitTypeINTEL -> intel_hit_type - CandidateTypeINTEL -> intel_candidate_type_t
* added cmake option to switch between internal and driver RT hardware BVH builder
* removed old direct connection of HW BVH builder to Embree
* first attempt to unify internal HW builder and API HW builder
* parsing --jit-cache command line option in rthwif_test application
* TBB: Stop task_scheduler_observer when TaskScheduler is destroyed
* upgrading gfx driver to gfx-driver-ci-master-12117
* using sycl namespace instead of cl::sycl namespace
* fixed rtcForwardIntersect/Occluded implementation when driver HW BVH interface is used
* fixed wrong primitive counting in driver HW BVH interface implementation
* fixed wrong instanced accel passed when driver HW BUILD interface was used
* fixed wrong bounds propagation when driver HW BVH build interface was used
* passing ray_query_INTEL* to each rt_production API function
* renamed ray_query_t to ray_query_INTEL, and rtas_t to raytracing_acceleration_structure_INTEL
* added time to RayDescINTEL
* renamed CandidateType to CandidateTypeINTEL and prepended all enumerators with CANDIDATE_TYPE_INTEL_
* HitType -> HitTypeINTEL and prepending HIT_TYPE_INTEL_ to all enumerants
* prepending RAY_FLAGS_INTEL_ to RayFlagsINTEL enumerants
* RayFlagsIntel -> RayFlagsINTEL
* rayquery_t -> ray_query_t
* rt_production test uses randomized transformation for instancing
* rt_production API test checks local2world and world2local output
* moved TraceRayCtrl to rthwif_internal.h
* Include dpcpp compilation target info in CMake config file export
* Fix rounded kernel dispatch size to round up when there's a remainder
* fixed bug in rthwif_test when internal task scheduler is used
* - renamed intel_sync_ray_query to intel_ray_query_sync - using pointers to rayquery_t instead of references - renamed intel_get_ray_tnear to intel_get_ray_tmin - remove intel_get_hit_instanced_accel function
* no longer passing Embree device to binary HW BVH build interface
* using intel_get_hit_instID short form in intel_rt_production API to be consistent with geomID/primID naming
* added testing for intel_get_hit_instanceUserID
* implemented intel_get_hit_instanceUserID, intel_get_hit_world_to_object and intel_get_hit_object_to_world
* cleanups to rthwif_production implementation
* all software instancing rt_production API tests are working
* software instancing rt_production API test works mostly
* procedural triangle rt_production API test works
* skipping rt_production test when using public gfx driver and limiting number of errors printed
* adding manual dispatch globals allocation support to rt_production tests
* first procedural triangle test
* preparing test code to support procedural triangles
* hwinstancing tests work
* rt_production test can create instances
* moved buildAccel to scene object
* started to add instancing support to rt_production API test
* added rthwif_test_triangles_anyhit_shader_commit/reject tests
* rt_production API test checks committed and potential hit
* rt_production API test uses multiple geometries
* enabled user_geometry again when implicit globals are used
* first version of rt_producton API testing
* working around some regression
* bugfix
* fixed compile issue
* added driver HW BVH builder support for mixed mblur and static geometry scenes
* first motion blur support in driver HW BVH builder
* updates
* worked around compilation issue under Windows
* fixed compile issue with disabled DPCPP support
* - making rthwifGetAccelSize future proof - added support of rthw_builder=default/driver to switch between both ray tracing hardware builders
* connected binary interface HW BVH builder to Embree
* first version of binary interface to HW BVH builder
* Fix check for if we're compiling without AOT
* using sycl::ext::oneapi::property::usm::device_read_only() mode and no longer using SYCL_PI_LEVEL_ZERO_USM_ALLOCATOR explicitely
* fixed tagging of wrong functions with RTC_SYCL_INDIRECTLY_CALLABLE in user_geometry tutorial
* upgrading to DPC++ compiler 20220802
* added cornell_box test for production RT API
* using fixed 128 bytes offset to root node in HW accel
* added support to use intel_get_implicit_dispatch_globals intrinsic (disabled by default)
* adding JIT test for PVC
* - upgrading to DPC++ compiler sycl-nightly/20220721 - changing CL/sycl.hpp to sycl/sycl.hpp for new dpcpp compilers
* add tutorial benchmarks
* update models for GPU benchmark tests
* CMake Large GRF option + enabled for fcall testcases
* disabling specialization constant usage in AOT mode due to some DPC++ compiler bug
* switching to sycl-nightly/20220625
* ci: switch testing to PVC B4
* adding unused option -cl-intel-private-memory-minimal-size-per-thread XXX to dpcpp.cmake which can get used to increase private memory size when stack calls are used
* disabling EMBREE_INSTALL_DEPENDENCIES again by default
* run gpu benchmark in regular ubuntu20.04 docker and clone benchmark client with secret token
* merging IGC flag settings for windows and linux to common place
* no longer disabling __noinline for SYCL
* using CMake list for CMAKE_IGC_OPTIONS
* fixed issue where disabling JIT cache did also disable USM chunk allocator
* only enforcing sub-group size when geometry callbacks are enabled
* adding CI test that disables geometry callbacks
* added rtcIsSYCLDeviceSupported API function to check if Embree is supported on some SYCL device
* lights no longer defined as indirectly callable
* upgrading to gfx-driver 11261 under Windows
* removed no longer required environment settings from device.cpp
* No longer including sycl.hpp inside Embree API headers. If SYCL support is required, the user should include sycl.hpp before including the Embree API headers.
* upgrading to gfx-driver-ci-master-11261
* cleanups to rthwif implementation
* intel_get_hit_instanceID now properly returns instanceIndex
* prepending PATH to DPC++ compiler
* added documentation on howto workaround issues when multiple GPUs are present in the system
* using -j 8 in documentation, -j might run out of memory
* enabled JIT caching again when specialization constants are used
* updating Windows driver version in documentation
* added feature mask to pathtracer tutorial
* upgrading public win gfx driver to sdp/1011433 in CI
* building Linux DPC++ release without curve support
* proper options to enable large GRF mode
* some bugfixes to documentation of compiling Embree with DPC++
* setting version to 4.0.0-alpha
* updates to README.md.in file for tutorials
* added DPC++ release build
* added Embree DPC++ known issues docu section
* viewer_sycl disables JIT caching when specialization constants are used
* added --jit-cache command line option to disable JIT caching
* Exporting tasking lib only for static Embree compilation. Configuring tutorial/README.md file.
* setting version to 4.0.0-alpha.0
* increasing required GLIBCXX version to 3.4.14
* removed Intel Compiler 2015 from supported compilers and CI testing as we trigger some compiler bug
* added more shortcuts to specify RTC_FEATURE_LINEAR_CURVES, RTC_FEATURE_BEZIER_CURVES, RTC_FEATURE_BSPLINE_CURVES, RTC_FEATURE_HERMITE_CURVES, and RTC_FEATURE_CURVES
* added feature mask to user_geometry tutorial
* removed specialization constant usage from triangle_geometry_tutorial, we have the viewer for this
* added feature mask to quaternion_motion_blur tutorial
* - moved feature mask for point_geometry tutorial to beginning of file - enabled more testing for point_geometry tutorial
* added feature mask to multi_instanced_geometry tutorial
* added feature mask to motion_blur_geometry tutorial
* added feature mask to minimal_sycl tutorial
* added feature mask to intersection_filter tutorial
* fixed disabling of intersection filter feature through feature mask
* added feature mask to instanced_geometry tutorial
* added feature mask to hair_geometry tutorial
* added feature mask to grid_geometry tutorial
* added feature mask to dynamic_scene tutorial
* added feature mask to curve_geometry tutorial
* improved feature masks for multi-level instancing
* using feature mask also for viewer debug shaders
* moved debug shaders to viewer application only, this significantly improved JIT compile time of all other tutorials
* added specialization constants and command line parsing for feature mask to viewer_sycl
* fixing issue with feature mask and RTC_MAX_INSTANCE_LEVEL_COUNT > 1
* added feature flags also to ISPC API
* fixing compile issue when DPC++ support was disabled
* Added specialization constant to point geometry tutorial
* Added missing feature filter for motion blur
* Added RTC_FEATURE_NONE to common header
* Enabled specialization constants in SYCL triangle tutorial
* Added feature_mask variable to IntersectArguments and condidtionals around features.
* using work group size (4,4) which works on DG2 and PVC
* allowing rendering of arbitrary screen resolutions
* no longer disabling various warnings for SYCL compilation
* removing SYCL_LIBS, seems not required
* removed some workarounds that disabled some tutorials in some configs
* documenting that user geometry bounds callbacks are host side functions always
* printing CMake warning when filter and user geometry callback support in geometries are enabled
* consistently adding -lsycl to link flags under Windows and Linux
* fixes to DPC++ application compilation documentation for Windows
* properly installing SYCL dependency under Windows
* documented how to build an Embree DPC++ application
* enabled EMBREE_ZIP_MODE and EMBREE_INSTALL_DEPENDENCIES by default
* adding minimal tutorial sources to install package
* cleanup
* hide device's task arena behind execute interface
* remove taskingscheduler.h from default.h
* don't link embree_sycl to tasking
* clean up parallel algorithm headers
* move more members in inner structs
* move task group members into inner struct
* move checkIfModifiedAndSet into scene_verify.cpp
* tutorial_ispc and tutorial should not link agains embree_sycl
* fix tasking linking issues and remove rthwif library
* add CI tests that use the internal tasking scheduler for sycl
* make internal tasking scheduler's exception handling consistent with TBB
* adding RTC_SYCL_SIMD_WIDTH to rtcore_config.h
* removed reqd_sub_group_size from API function declarations, enforcing sub group size only works at kernel level
* making dg2 default under Linux and JIT default under Windows
* added DPC++ Memory Pooling section to documentation
* enabling memory pooling and persistent JIT caching in minimal_sycl tutorial
* always detecting dpcpp when EMBREE_DPCPP_SUPPORT is enabled
* updates to Windows build instructions with DPC++ compiler
* introduced RTC_SYCL_KERNEL kernel attribute
* linking embree target to embree_sycl target
* enabled SYCL chunk allocator for shared USM allocations
* removing SYCL allocation workaround
* allocator.clear() was executed too late in qbvh6_builder_sah
* minimal_sycl tutorial properly hits triangle now
* increasing timeout for SYCL CI tests to 30s
* added rtcSetGeometryUserDataFromScene API function to be used during rendering
* removing rtcInterpolate SYCL implementation
* enabled user_geometry_sycl again in JIT test
* fixed some compile warnings for Release build
* updated instructions on compiling Embree with DPC++ under Linux
* using TBB2020.2 for Windows CI tests
* documented that AOT compilation with multiple devices is currently not working
* Added documentation on how to install Windows HPG drivers.
* increasing ctest timeout to 20s to properly run JIT tests
* Revert "running Linux JIT CI test only on dg2-03 machine"
* getting pathtracer to work when geometry filter functions are disabled
* avoiding function pointers for light sampling and evaluation in pathtracer
* running Linux JIT CI test only on dg2-03 machine
* Added Embree DPC++ compilation docu
* Added Embree DPC++ API docu section
* fixed issue with SYCL benchmarking, dg2-b0-512 device config no longer supported, using just dg2 device
* Added documentation for EMBREE_DPCPP_AOT_DEVICES cmake option. Cleanup of EMBREE_DPCPP_AOT_DEVICES cmake option.
* installing SYCL dependency only when EMBREE_DPCPP_SUPPORT is enabled
* documented EMBREE_DPCPP_SUPPORT cmake option
* using newer driver for public CI testing
* adding CI tests that use publicly available drivers
* removed unused build.py and conf.py scripts
* updated third-party-programs files
* adding windows JIT test that uses gfx-driver sdp/1011314
* trying gfx-driver-ci-master release build
* updates to the Embree 3 to Embree 4 porting section in documentation
* added agama CI test
* extending download_gfx script to support agama releases
* passing RTCIntersectArguments through pointer to make API C99 compliant again
* installing only required SYCL libraries
* - Removed Embree 2 to Embree 3 patch script - Removed Embree 2 to Embree 3 upgrade section from documentation
* renamed all *.3embree3 man files to *.4embree4
* Renamed embree3 API include folder to embree4
* updated model hash to devel branch of model repo
* Initial commit of SYCL/DPC++ version of Embree with GPU support
-------------------------------------------------------------------
Thu Aug 3 07:17:55 UTC 2023 - Hans-Peter Jansen <hp@urpla.net>
- Fix library package naming
-------------------------------------------------------------------
Thu Nov 3 07:56:31 UTC 2022 - Hans-Peter Jansen <hpj@urpla.net>
- Obsolete the -devel-static package
-------------------------------------------------------------------
Wed Nov 2 02:29:29 UTC 2022 - Stefan Brüns <stefan.bruens@rwth-aachen.de>
- Build embree as shared library, blender no longer requires a
static library build.
- Reduce memory constraints, use limit_build to avoid build
failures due to OOM.
- Fix build on ARM (gh#embree/embree#383, gh#embree/embree#410)
-------------------------------------------------------------------
Tue Oct 25 10:56:41 UTC 2022 - Hans-Peter Jansen <hpj@urpla.net>
- Remove the cmake config tweaking
-------------------------------------------------------------------
Sat Oct 22 17:44:44 UTC 2022 - hpj@urpla.net
- Update to version 3.13.5:
* regenerated documentation
* setting version to v3.13.5
* ci: add github actions workflows
* FastAllocator: fix the false sharing in SpinLock array
* Fixed spatial splitting for non-planar quads.
* no longer automatically starting thread nor setting affinity
* disabling thread ID affinity re-mapping as it causes issues inside LXC containers
* TBB: Stop task_scheduler_observer when TaskScheduler is destroyed
* Improved self intersection avoidance for RTC_GEOMETRY_TYPE_DISC_POINT geometry type.
* Fix compilation issues on macOS 12.5 and Apple clang version 13.1.6
- Update to version 3.13.4:
* 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
- Switch to _service based package source
- Use simd compiler ispc
-------------------------------------------------------------------
Sat Oct 22 16:56:52 UTC 2022 - Hans-Peter Jansen <hpj@urpla.net>
- Fix cmake devel config to reflect reality
-------------------------------------------------------------------
Tue Apr 26 07:22:16 UTC 2022 - Guillaume GARDET <guillaume.gardet@opensuse.org>
- Update _constraints for aarch64
-------------------------------------------------------------------
Sun Apr 24 17:17:39 UTC 2022 - Hans-Peter Jansen <hpj@urpla.net>
- Update to version v3.13.3
* Invalid multi segment motion blurred normal oriented curves
are properly excluded from BVH build.
* Fixing issue with normal oriented curve construction when
center curve curvature is very large.
* Due to this change normal oriented curve shape changes
slightly.
* Fixed crash caused by disabling a geometry and then detaching
it from the scene.
* Bugfix in emulated ray packet intersection when
EMBREE_RAY_PACKETS is turned off.
* Bugfix for linear quaternion interpolation fallback.
* Fixed issues with spaces in path to Embree build folder.
* Some fixes to compile Embree in SSE mode using WebAssembly.
* Bugfix for occlusion rays with grids and ray packets.
* We do no longer provide installers for Windows and macOS,
please use the ZIP files instead.
* Upgrading to Intel® ISPC 1.17.0 for release build.
* Upgrading to Intel® oneTBB 2021.5.0 for release build.
-------------------------------------------------------------------
Tue Nov 23 21:59:02 UTC 2021 - Ben Greiner <code@bnavigator.de>
- Update to version 3.13.2
* Avoiding spatial split positions that are slightly out of
geometry bounds.
* Introduced rtcGetGeometryThreadSafe function, which is a thread
safe version of rtcGetGeometry.
* Using more accurate rcp implementation.
* Bugfix to rare corner case of high quality BVH builder.
- Embree 3.13.1
* Added support for ISPC+ARM.
* Releases upgrade to TBB 2021.3.0 and ISPC 1.16.1
- Use internal tasking instead of TBB
* gh#embree/embree#348
* boo#1192994
-------------------------------------------------------------------
Tue May 25 22:46:11 UTC 2021 - Ferdinand Thiessen <rpm@fthiessen.de>
- Update to version 3.13.0
* Added support for aarch64 cpu architecture.
* RTC_SUBDIVISION_MODE_NO_BOUNDARY now works properly for
non-manifold edges.
* Embree no longer reads the .embree3 config files, thus all
configuration has to get passed through the config string to
rtcNewDevice.
* Releases upgrade to TBB 2021.2.0 and ISPC 1.15.0
-------------------------------------------------------------------
Sun Apr 25 03:31:24 UTC 2021 - Wang Jun <jgwang@suse.com>
- update to version 3.12.1
* Fixed wrong uv and Ng for grid intersector in robust mode for AVX.
* Removed optimizations for Knights Landing.
* Upgrading release builds to use oneTBB 2021.1.1
-------------------------------------------------------------------
Thu Oct 15 11:06:16 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
- update to version 3.12.1
+ Changed default frequency level to SIMD128 for Skylake, Cannon
Lake, Comet Lake and Tiger Lake CPUs. This change typically
improves performance for renderers that just use SSE by
maintaining higher CPU frequencies. In case your renderer is
AVX optimized you can get higher ray tracing performance
by configuring the frequency level to simd256 through passing
frequency_level=simd256 to rtcNewDevice.
-------------------------------------------------------------------
Fri Sep 25 08:26:06 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
- update to version 3.12.0
+ Added linear cone curve geometry support. In this mode a real
geometric surface for curves
+ with linear basis is rendered using capped cones. They are
discontinuous at edge boundaries.
+ Enabled fast two level builder for instances when low quality
build is requested.
+ Bugfix for BVH build when geometries got disabled.
+ Added EMBREE_BACKFACE_CULLING_CURVES cmake option. This allows
for a cheaper round
+ linear curve intersection when correct internal tracking and
back hits are not required.
+ The new cmake option defaults to OFF.
+ User geometries with invalid bounds with lower>upper in some
dimension will be ignored.
+ Increased robustness for grid interpolation code and fixed
returned out of range u/v
+ coordinates for grid primitive.
+ Fixed handling of motion blur time range for sphere, discs, and
oriented disc geometries.
+ Fixed missing model data in releases.
+ Ensure compatibility to newer versions of oneTBB.
+ Motion blur BVH nodes no longer store NaN values.
-------------------------------------------------------------------
Tue Sep 1 16:22:19 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
- raise limits to 24G, since this version uses up to 22,6G on TW
for build (16,7G on 15.2)
-------------------------------------------------------------------
Thu Aug 20 08:33:47 UTC 2020 - Martin Liška <mliska@suse.cz>
- Use memory limit as the build is dominated by 2 huge gas runs.
-------------------------------------------------------------------
Thu Jul 9 11:07:33 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
- update to version 3.11.0
+ Round linear curves now automatically check for the existence
of left and right
+ connected segments if the flags buffer is empty. Left segments
exist if the
+ segment(id-1) + 1 == segment(id) and similarly for right
segments.
+ Implemented the min-width feature for curves and points, which
allows to increase the
+ radius in a distance dependent way, such that the curve or
points thickness is n pixels wide.
+ Round linear curves are closed now also at their start.
+ Embree no longer supports Visual Studio 2013 starting with this
release.
+ Bugfix in subdivision tessellation level assignment for non-
quad base primitives
+ Small meshes are directly added to top level build phase of
two-level builder to reduce memory consumption.
+ Enabled fast two level builder for user geometries when low
quality build is requested.
- update to version 3.10.0
+ Added EMBREE_COMPACT_POLYS CMake option which enables double
indexed triangle and quad
+ leaves to reduce memory consumption in compact mode by an
additional 40% at about
+ 15% performance impact. This new mode is disabled by default.
+ Compile fix for oneTBB 2021.1-beta05
+ Releases upgrade to TBB 2020.2
+ Compile fix for 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.
- update to version 3.9.0
+ Added round linear curve geometry support. In this mode a real
geometric surface for curves
+ with linear basis is rendered using capped cones with spherical
filling between
+ the curve segments.
+ Added rtcGetSceneDevice API function, that returns the device a
scene got created in.
+ Improved performance of round curve rendering by up to 1.8x.
+ Bugfix to sphere intersection filter invokation for back hit.
+ Fixed wrong assertion that triggered for invalid curves which
anyway get filtered out.
+ RelWithDebInfo mode no longer enables assertions.
+ Fixed an issue in FindTBB.cmake that caused compile error with
Debug build under Linux.
+ Embree releases no longer provide RPMs for Linux. Please use
the RPMs coming with the package
+ manager of your Linux distribution.
-------------------------------------------------------------------
Wed Feb 5 21:11:16 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
- update to version 3.8.0
+ New Features:
* Added collision detection support for user geometries (see
rtcCollide API function)
* Passing geomID to user geometry callbacks.
+ Fixed Issues:
* Bugfix in AVX512VL codepath for rtcIntersect1
* For sphere geometries the intersection filter gets now
invoked for front and back hit.
* Fixed some bugs for quaternion motion blur.
* RTCIntersectContext always non-const in Embree API
* Made RTCHit aligned to 16 bytes in Embree API
- update to version 3.7.0
+ Added quaternion motion blur for correct interpolation of
rotational transformations.
+ Fixed wrong bounding calculations when a motion blurred
instance did instantiate a motion blurred scene.
+ In robust mode the depth test consistently uses tnear <= t <=
tfar now in order to robustly continue traversal at a previous
hit point in a way that guarentees reaching all hits, even hits
at the same place.
+ Fixed depth test in robust mode to be precise at tnear and
tfar.
+ Added next_hit tutorial to demonstrate robustly collecting all
hits along a ray using multiple ray queries.
+ Implemented robust mode for curves. This has a small
performance impact but fixes bounding problems with flat
curves.
+ Improved quality of motion blur BVH by using linear bounds
during binning.
+ Implemented issue with motion blur builder where number of time
segments for SAH heuristic were counted wrong due to some
numerical issues.
+ Fixed an accuracy issue with rendering very short fat curves.
+ rtcCommitScene can now get called during rendering from
multiple threads to lazily build geometry. When TBB is used
this causes a much lower overhead than using
rtcJoinCommitScene.
+ Geometries can now get attached to multiple scenes at the same
time, which simplifies mapping general scene graphs to API.
+ Updated to TBB 2019.9 for release builds.
+ Fixed a bug in the BVH builder for Grid geometries.
+ Added macOS Catalina support to Embree releases.
-------------------------------------------------------------------
Fri Oct 11 18:49:03 UTC 2019 - Adrian Schröter <adrian@suse.de>
- update to version 3.6.1
Added Catmull-Rom curve types.
Added support for multi-level instancing.
Added support for point queries.
Fixed a bug preventing normal oriented curves being used unless timesteps were specified.
Fixed bug in external BVH builder when configured for dynamic build.
Added support for new config flag "user_threads=N" to device initialization
which sets the number of threads used by TBB but created by the user.
Fixed automatic vertex buffer padding when using rtcSetNewGeometry API function.
-------------------------------------------------------------------
Mon Sep 9 17:40:25 UTC 2019 - Jan Engelhardt <jengelh@inai.de>
- Trim marketing from description.
-------------------------------------------------------------------
Sun Sep 8 16:14:54 UTC 2019 - Stefan Brüns <stefan.bruens@rwth-aachen.de>
- Relax _constraints, and use memory-constraints/limit_build instead
- Disable building for every arch but x86_64, as embree strictly requires
at least SSE2.
- Use FAT LTO objects in order to provide proper static library.
- Use tbb-devel BuildRequires instead of cmake(TBB), the latter is not
available in Leap 15.x
-------------------------------------------------------------------
Mon Jul 15 19:54:19 UTC 2019 - Hans-Peter Jansen <hpj@urpla.net>
- version 3.5.2: initial build
