diff options
Diffstat (limited to 'thirdparty/embree/kernels/common/scene.cpp')
| -rw-r--r-- | thirdparty/embree/kernels/common/scene.cpp | 269 |
1 files changed, 175 insertions, 94 deletions
diff --git a/thirdparty/embree/kernels/common/scene.cpp b/thirdparty/embree/kernels/common/scene.cpp index 65d31d0f81..10cb3c4bec 100644 --- a/thirdparty/embree/kernels/common/scene.cpp +++ b/thirdparty/embree/kernels/common/scene.cpp @@ -3,12 +3,35 @@ #include "scene.h" +#include "../../common/tasking/taskscheduler.h" + #include "../bvh/bvh4_factory.h" #include "../bvh/bvh8_factory.h" + #include "../../common/algorithms/parallel_reduce.h" - + +#if defined(EMBREE_SYCL_SUPPORT) +# include "../sycl/rthwif_embree_builder.h" +#endif + + namespace embree { + + struct TaskGroup { + /*! global lock step task scheduler */ +#if defined(TASKING_INTERNAL) + MutexSys schedulerMutex; + Ref<TaskScheduler> scheduler; +#elif defined(TASKING_TBB) && TASKING_TBB_USE_TASK_ISOLATION + tbb::isolated_task_group group; +#elif defined(TASKING_TBB) + tbb::task_group group; +#elif defined(TASKING_PPL) + concurrency::task_group group; +#endif + }; + /* error raising rtcIntersect and rtcOccluded functions */ void missing_rtcCommit() { throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed"); } void invalid_rtcIntersect1() { throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcIntersect and rtcOccluded not enabled"); } @@ -22,13 +45,20 @@ namespace embree flags_modified(true), enabled_geometry_types(0), scene_flags(RTC_SCENE_FLAG_NONE), quality_flags(RTC_BUILD_QUALITY_MEDIUM), - is_build(false), modified(true), + modified(true), + taskGroup(new TaskGroup()), progressInterface(this), progress_monitor_function(nullptr), progress_monitor_ptr(nullptr), progress_monitor_counter(0) { device->refInc(); intersectors = Accel::Intersectors(missing_rtcCommit); + /* use proper device and context for SYCL allocations */ +#if defined(EMBREE_SYCL_SUPPORT) + if (DeviceGPU* gpu_device = dynamic_cast<DeviceGPU*>(device)) + hwaccel = AccelBuffer(AccelAllocator<char>(device,gpu_device->getGPUDevice(),gpu_device->getGPUContext()),0); +#endif + /* one can overwrite flags through device for debugging */ if (device->quality_flags != -1) quality_flags = (RTCBuildQuality) device->quality_flags; @@ -90,10 +120,11 @@ namespace embree void Scene::createTriangleAccel() { #if defined(EMBREE_GEOMETRY_TRIANGLE) + if (device->tri_accel == "default") { if (quality_flags != RTC_BUILD_QUALITY_LOW) - { + { int mode = 2*(int)isCompactAccel() + 1*(int)isRobustAccel(); switch (mode) { case /*0b00*/ 0: @@ -168,11 +199,13 @@ namespace embree #endif else throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"unknown triangle acceleration structure "+device->tri_accel); #endif + } void Scene::createTriangleMBAccel() { #if defined(EMBREE_GEOMETRY_TRIANGLE) + if (device->tri_accel_mb == "default") { int mode = 2*(int)isCompactAccel() + 1*(int)isRobustAccel(); @@ -211,6 +244,7 @@ namespace embree void Scene::createQuadAccel() { #if defined(EMBREE_GEOMETRY_QUAD) + if (device->quad_accel == "default") { if (quality_flags != RTC_BUILD_QUALITY_LOW) @@ -292,6 +326,7 @@ namespace embree void Scene::createQuadMBAccel() { #if defined(EMBREE_GEOMETRY_QUAD) + if (device->quad_accel_mb == "default") { int mode = 2*(int)isCompactAccel() + 1*(int)isRobustAccel(); @@ -329,6 +364,7 @@ namespace embree void Scene::createHairAccel() { #if defined(EMBREE_GEOMETRY_CURVE) || defined(EMBREE_GEOMETRY_POINT) + if (device->hair_accel == "default") { int mode = 2*(int)isCompactAccel() + 1*(int)isRobustAccel(); @@ -366,6 +402,7 @@ namespace embree void Scene::createHairMBAccel() { #if defined(EMBREE_GEOMETRY_CURVE) || defined(EMBREE_GEOMETRY_POINT) + if (device->hair_accel_mb == "default") { #if defined (EMBREE_TARGET_SIMD8) @@ -416,7 +453,8 @@ namespace embree void Scene::createUserGeometryAccel() { #if defined(EMBREE_GEOMETRY_USER) - if (device->object_accel == "default") + + if (device->object_accel == "default") { #if defined (EMBREE_TARGET_SIMD8) if (device->canUseAVX() && !isCompactAccel()) @@ -448,6 +486,7 @@ namespace embree void Scene::createUserGeometryMBAccel() { #if defined(EMBREE_GEOMETRY_USER) + if (device->object_accel_mb == "default" ) { #if defined (EMBREE_TARGET_SIMD8) if (device->canUseAVX() && !isCompactAccel()) @@ -467,6 +506,7 @@ namespace embree void Scene::createInstanceAccel() { #if defined(EMBREE_GEOMETRY_INSTANCE) + // if (device->object_accel == "default") { #if defined (EMBREE_TARGET_SIMD8) @@ -494,6 +534,7 @@ namespace embree void Scene::createInstanceMBAccel() { #if defined(EMBREE_GEOMETRY_INSTANCE) + //if (device->instance_accel_mb == "default") { #if defined (EMBREE_TARGET_SIMD8) @@ -550,10 +591,58 @@ namespace embree #endif } + void Scene::createInstanceArrayAccel() + { +#if defined(EMBREE_GEOMETRY_INSTANCE_ARRAY) + + // if (device->object_accel == "default") + { +#if defined (EMBREE_TARGET_SIMD8) + if (device->canUseAVX() && !isCompactAccel()) { + if (quality_flags != RTC_BUILD_QUALITY_LOW) { + accels_add(device->bvh8_factory->BVH8InstanceArray(this, BVHFactory::BuildVariant::STATIC)); + } else { + accels_add(device->bvh8_factory->BVH8InstanceArray(this, BVHFactory::BuildVariant::DYNAMIC)); + } + } + else +#endif + { + if (quality_flags != RTC_BUILD_QUALITY_LOW) { + accels_add(device->bvh4_factory->BVH4InstanceArray(this, BVHFactory::BuildVariant::STATIC)); + } else { + accels_add(device->bvh4_factory->BVH4InstanceArray(this, BVHFactory::BuildVariant::DYNAMIC)); + } + } + } + // else throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"unknown instance accel "+device->instance_accel); +#endif + } + + void Scene::createInstanceArrayMBAccel() + { +#if defined(EMBREE_GEOMETRY_INSTANCE_ARRAY) + + //if (device->instance_accel_mb == "default") + { +#if defined (EMBREE_TARGET_SIMD8) + if (device->canUseAVX() && !isCompactAccel()) + accels_add(device->bvh8_factory->BVH8InstanceArrayMB(this)); + else +#endif + accels_add(device->bvh4_factory->BVH4InstanceArrayMB(this)); + } + //else throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"unknown instance mblur accel "+device->instance_accel_mb); +#endif + } + + void Scene::createGridAccel() { - BVHFactory::IntersectVariant ivariant = isRobustAccel() ? BVHFactory::IntersectVariant::ROBUST : BVHFactory::IntersectVariant::FAST; #if defined(EMBREE_GEOMETRY_GRID) + + BVHFactory::IntersectVariant ivariant = isRobustAccel() ? BVHFactory::IntersectVariant::ROBUST : BVHFactory::IntersectVariant::FAST; + if (device->grid_accel == "default") { #if defined (EMBREE_TARGET_SIMD8) @@ -579,6 +668,7 @@ namespace embree void Scene::createGridMBAccel() { #if defined(EMBREE_GEOMETRY_GRID) + if (device->grid_accel_mb == "default") { accels_add(device->bvh4_factory->BVH4GridMB(this,BVHFactory::BuildVariant::STATIC)); @@ -588,13 +678,13 @@ namespace embree #endif } - + void Scene::clear() { } unsigned Scene::bind(unsigned geomID, Ref<Geometry> geometry) { - Lock<SpinLock> lock(geometriesMutex); + Lock<MutexSys> lock(geometriesMutex); if (geomID == RTC_INVALID_GEOMETRY_ID) { geomID = id_pool.allocate(); if (geomID == RTC_INVALID_GEOMETRY_ID) @@ -620,7 +710,7 @@ namespace embree void Scene::detachGeometry(size_t geomID) { - Lock<SpinLock> lock(geometriesMutex); + Lock<MutexSys> lock(geometriesMutex); if (geomID >= geometries.size()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"invalid geometry ID"); @@ -637,45 +727,11 @@ namespace embree geometryModCounters_[geomID] = 0; } - void Scene::updateInterface() - { - is_build = true; - } - - void Scene::commit_task () + void Scene::build_cpu_accels() { - checkIfModifiedAndSet (); - if (!isModified()) { - return; - } - - /* print scene statistics */ - if (device->verbosity(2)) - printStatistics(); - - progress_monitor_counter = 0; - - /* gather scene stats and call preCommit function of each geometry */ - this->world = parallel_reduce (size_t(0), geometries.size(), GeometryCounts (), - [this](const range<size_t>& r)->GeometryCounts - { - GeometryCounts c; - for (auto i=r.begin(); i<r.end(); ++i) - { - if (geometries[i] && geometries[i]->isEnabled()) - { - geometries[i]->preCommit(); - geometries[i]->addElementsToCount (c); - c.numFilterFunctions += (int) geometries[i]->hasFilterFunctions(); - } - } - return c; - }, - std::plus<GeometryCounts>() - ); - /* select acceleration structures to build */ unsigned int new_enabled_geometry_types = world.enabledGeometryTypesMask(); + if (flags_modified || new_enabled_geometry_types != enabled_geometry_types) { accels_init(); @@ -685,7 +741,7 @@ namespace embree parallel_for(geometryModCounters_.size(), [&] ( const size_t i ) { geometryModCounters_[i] = 0; }); - + if (getNumPrimitives(TriangleMesh::geom_type,false)) createTriangleAccel(); if (getNumPrimitives(TriangleMesh::geom_type,true)) createTriangleMBAccel(); if (getNumPrimitives(QuadMesh::geom_type,false)) createQuadAccel(); @@ -702,7 +758,9 @@ namespace embree if (getNumPrimitives(Geometry::MTY_INSTANCE_CHEAP,true)) createInstanceMBAccel(); if (getNumPrimitives(Geometry::MTY_INSTANCE_EXPENSIVE,false)) createInstanceExpensiveAccel(); if (getNumPrimitives(Geometry::MTY_INSTANCE_EXPENSIVE,true)) createInstanceExpensiveMBAccel(); - + if (getNumPrimitives(Geometry::MTY_INSTANCE_ARRAY,false)) createInstanceArrayAccel(); + if (getNumPrimitives(Geometry::MTY_INSTANCE_ARRAY,true)) createInstanceArrayMBAccel(); + flags_modified = false; enabled_geometry_types = new_enabled_geometry_types; } @@ -719,6 +777,61 @@ namespace embree flags_modified = true; // in non-dynamic mode we have to re-create accels } + if (device->verbosity(2)) { + std::cout << "created scene intersector" << std::endl; + accels_print(2); + std::cout << "selected scene intersector" << std::endl; + intersectors.print(2); + } + } + + void Scene::build_gpu_accels() + { +#if defined(EMBREE_SYCL_SUPPORT) + const BBox3f aabb = rthwifBuild(this,hwaccel); + bounds = LBBox<embree::Vec3fa>(aabb); + hwaccel_bounds = aabb; +#endif + } + + void Scene::commit_task () + { + checkIfModifiedAndSet(); + if (!isModified()) return; + + /* print scene statistics */ + if (device->verbosity(2)) + printStatistics(); + + progress_monitor_counter = 0; + + /* gather scene stats and call preCommit function of each geometry */ + this->world = parallel_reduce (size_t(0), geometries.size(), GeometryCounts (), + [this](const range<size_t>& r)->GeometryCounts + { + GeometryCounts c; + for (auto i=r.begin(); i<r.end(); ++i) + { + if (geometries[i] && geometries[i]->isEnabled()) + { + geometries[i]->preCommit(); + geometries[i]->addElementsToCount (c); + c.numFilterFunctions += (int) geometries[i]->hasArgumentFilterFunctions(); + c.numFilterFunctions += (int) geometries[i]->hasGeometryFilterFunctions(); + } + } + return c; + }, + std::plus<GeometryCounts>() + ); + +#if defined(EMBREE_SYCL_SUPPORT) + if (DeviceGPU* gpu_device = dynamic_cast<DeviceGPU*>(device)) + build_gpu_accels(); + else +#endif + build_cpu_accels(); + /* call postCommit function of each geometry */ parallel_for(geometries.size(), [&] ( const size_t i ) { if (geometries[i] && geometries[i]->isEnabled()) { @@ -727,16 +840,7 @@ namespace embree geometryModCounters_[i] = geometries[i]->getModCounter(); } }); - - updateInterface(); - if (device->verbosity(2)) { - std::cout << "created scene intersector" << std::endl; - accels_print(2); - std::cout << "selected scene intersector" << std::endl; - intersectors.print(2); - } - setModified(false); } @@ -771,11 +875,11 @@ namespace embree /* allocates own taskscheduler for each build */ Ref<TaskScheduler> scheduler = nullptr; { - Lock<MutexSys> lock(schedulerMutex); - scheduler = this->scheduler; + Lock<MutexSys> lock(taskGroup->schedulerMutex); + scheduler = taskGroup->scheduler; if (scheduler == null) { buildLock.lock(); - this->scheduler = scheduler = new TaskScheduler; + taskGroup->scheduler = scheduler = new TaskScheduler; } } @@ -792,13 +896,13 @@ namespace embree /* initiate build */ // -- GODOT start -- // try { - scheduler->spawn_root([&]() { commit_task(); Lock<MutexSys> lock(schedulerMutex); this->scheduler = nullptr; }, 1, !join); + TaskScheduler::TaskGroupContext context; + scheduler->spawn_root([&]() { commit_task(); Lock<MutexSys> lock(taskGroup->schedulerMutex); taskGroup->scheduler = nullptr; }, &context, 1, !join); // } // catch (...) { // accels_clear(); - // updateInterface(); - // Lock<MutexSys> lock(schedulerMutex); - // this->scheduler = nullptr; + // Lock<MutexSys> lock(taskGroup->schedulerMutex); + // taskGroup->scheduler = nullptr; // throw; // } // -- GODOT end -- @@ -809,7 +913,7 @@ namespace embree #if defined(TASKING_TBB) void Scene::commit (bool join) - { + { #if defined(TASKING_TBB) && (TBB_INTERFACE_VERSION_MAJOR < 8) if (join) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcJoinCommitScene not supported with this TBB version"); @@ -827,16 +931,7 @@ namespace embree #endif do { - -#if USE_TASK_ARENA - if (join) { - device->arena->execute([&]{ group.wait(); }); - } - else -#endif - { - group.wait(); - } + device->execute(join, [&](){ taskGroup->group.wait(); }); pause_cpu(); yield(); @@ -857,26 +952,14 @@ namespace embree tbb::task_group_context ctx( tbb::task_group_context::isolated, tbb::task_group_context::default_traits | tbb::task_group_context::fp_settings ); #endif //ctx.set_priority(tbb::priority_high); - -#if USE_TASK_ARENA - if (join) - { - device->arena->execute([&]{ - group.run([&]{ - tbb::parallel_for (size_t(0), size_t(1), size_t(1), [&] (size_t) { commit_task(); }, ctx); - }); - group.wait(); - }); - } - else -#endif + device->execute(join, [&]() { - group.run([&]{ + taskGroup->group.run([&]{ tbb::parallel_for (size_t(0), size_t(1), size_t(1), [&] (size_t) { commit_task(); }, ctx); }); - group.wait(); - } - + taskGroup->group.wait(); + }); + /* reset MXCSR register again */ _mm_setcsr(mxcsr); } @@ -886,7 +969,6 @@ namespace embree _mm_setcsr(mxcsr); accels_clear(); - updateInterface(); throw; } } @@ -915,10 +997,10 @@ namespace embree try { - group.run([&]{ + taskGroup->group.run([&]{ concurrency::parallel_for(size_t(0), size_t(1), size_t(1), [&](size_t) { commit_task(); }); }); - group.wait(); + taskGroup->group.wait(); /* reset MXCSR register again */ _mm_setcsr(mxcsr); @@ -929,7 +1011,6 @@ namespace embree _mm_setcsr(mxcsr); accels_clear(); - updateInterface(); throw; } } |