WIP Bump to Embree v4.4

johguenther · johguenther · commit a9c1b5c3eec4 · 2026-03-17T11:33:27.000+01:00
diff --git a/modules/cpu/CMakeLists.txt b/modules/cpu/CMakeLists.txt
@@ -472,8 +472,8 @@ if (OSPRAY_MODULE_GPU)
     varying=
     unmasked=
     export=
-    rtcIntersectV=rtcIntersect1
-    rtcOccludedV=rtcOccluded1
+    rtcTraversableIntersectV=rtcTraversableIntersect1
+    rtcTraversableOccludedV=rtcTraversableOccluded1
     rtcInterpolateV1=rtcInterpolate1
     vklComputeSampleV=vklComputeSample
     vklComputeGradientV=vklComputeGradient
diff --git a/modules/cpu/camera/Camera.cpp b/modules/cpu/camera/Camera.cpp
@@ -57,15 +57,17 @@ void Camera::commit()
       embreeGeometry = rtcNewGeometry(
           getISPCDevice().getEmbreeDevice(), RTC_GEOMETRY_TYPE_INSTANCE);
       embreeScene = rtcNewScene(getISPCDevice().getEmbreeDevice());
+      rtcSetGeometryInstancedScene(embreeGeometry, embreeScene); // dummy
       rtcAttachGeometryByID(embreeScene, embreeGeometry, 0);
     }
 
     motionTransform.setEmbreeTransform(embreeGeometry);
+    rtcCommitScene(embreeScene);
+    getSh()->traversable = rtcGetSceneTraversable(embreeScene);
 
     if (shutter.lower == shutter.upper || !motionTransform.motionBlur) {
       // directly interpolate to single shutter time
-      rtcGetGeometryTransformFromScene(embreeScene,
-          0,
+      rtcGetGeometryTransform(embreeGeometry,
           shutter.lower,
           RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
           &motionTransform.transform);
@@ -76,8 +78,7 @@ void Camera::commit()
   if (motionTransform.motionBlur) {
     // use main direction at center of shutter time
     affine3f middleTransform;
-    rtcGetGeometryTransformFromScene(embreeScene,
-        0,
+    rtcGetGeometryTransform(embreeGeometry,
         shutter.center(),
         RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
         &middleTransform);
@@ -109,7 +110,6 @@ void Camera::commit()
   getSh()->subImage.upper = imageEnd;
   getSh()->shutter = shutter;
   getSh()->motionBlur = motionTransform.motionBlur;
-  getSh()->scene = embreeScene;
   getSh()->globalShutter = shutterType == OSP_SHUTTER_GLOBAL;
   getSh()->rollingShutterHorizontal = (shutterType == OSP_SHUTTER_ROLLING_RIGHT
       || shutterType == OSP_SHUTTER_ROLLING_LEFT);
diff --git a/modules/cpu/camera/CameraShared.h b/modules/cpu/camera/CameraShared.h
@@ -47,7 +47,7 @@ struct Camera
   bool needTimeSample;
   bool rollingShutterHorizontal;
   float rollingShutterDuration;
-  RTCScene scene; // only to call rtcGetGeometryTransformFromScene
+  RTCTraversable traversable; // only for rtcGetGeometryTransformFromTraversable
 
 #ifdef __cplusplus
   Camera()
@@ -62,7 +62,7 @@ struct Camera
         needTimeSample(false),
         rollingShutterHorizontal(false),
         rollingShutterDuration(0.f),
-        scene(nullptr)
+        traversable(nullptr)
   {}
 };
 } // namespace ispc
diff --git a/modules/cpu/camera/OrthographicCamera.ispc b/modules/cpu/camera/OrthographicCamera.ispc
@@ -28,8 +28,11 @@ SYCL_EXTERNAL void OrthographicCamera_initRay(const Camera *uniform _self,
   const uniform FeatureFlagsOther ffo = getFeatureFlagsOther(ffh);
   if (self->super.motionBlur && (ffo & FFO_CAMERA_MOTION_BLUR)) {
     AffineSpace3f xfm;
-    rtcGetGeometryTransformFromScene(
-        self->super.scene, 0, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
+    rtcGetGeometryTransformFromTraversable(self->super.traversable,
+        0,
+        time,
+        RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
+        &xfm);
 
     // we cannot just transform the final org & dir, because interpolated
     // transforms can scale (even if original transforms are without scale)
diff --git a/modules/cpu/camera/PanoramicCamera.ispc b/modules/cpu/camera/PanoramicCamera.ispc
@@ -65,8 +65,11 @@ SYCL_EXTERNAL void PanoramicCamera_initRay(const Camera *uniform _self,
   const uniform FeatureFlagsOther ffo = getFeatureFlagsOther(ffh);
   if (self->super.motionBlur && (ffo & FFO_CAMERA_MOTION_BLUR)) {
     AffineSpace3f xfm;
-    rtcGetGeometryTransformFromScene(
-        self->super.scene, 0, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
+    rtcGetGeometryTransformFromTraversable(self->super.traversable,
+        0,
+        time,
+        RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
+        &xfm);
 
     // we cannot just transform the final org & dir, because interpolated
     // transforms can scale (even if original transforms are without scale)
diff --git a/modules/cpu/camera/PerspectiveCamera.ispc b/modules/cpu/camera/PerspectiveCamera.ispc
@@ -89,8 +89,8 @@ SYCL_EXTERNAL void PerspectiveCamera_initRayMB(const Camera *uniform _self,
   const float time = Camera_shutterTime(_self, screen, sample.time);
 
   AffineSpace3f xfm;
-  rtcGetGeometryTransformFromScene(
-      self->super.scene, 0, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
+  rtcGetGeometryTransformFromTraversable(
+      self->super.traversable, 0, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
 
   vec3f org = xfmPoint(xfm, self->org);
   vec3f dir = normalize(xfmVector(xfm, self->dir_00));
diff --git a/modules/cpu/common/Instance.cpp b/modules/cpu/common/Instance.cpp
@@ -55,11 +55,9 @@ void Instance::setEmbreeGeom(RTCScene scene, unsigned int geomID)
 {
   RTCGeometry geom = rtcGetGeometry(scene, geomID);
   motionTransform.setEmbreeTransform(geom);
-  getSh()->scene = scene;
   getSh()->geomID = geomID;
   if (getSh()->motionBlur) {
-    rtcGetGeometryTransformFromScene(scene,
-        geomID,
+    rtcGetGeometryTransform(geom,
         .5f,
         RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
         &getSh()->xfm); // for SciVis
diff --git a/modules/cpu/common/Instance.h b/modules/cpu/common/Instance.h
@@ -24,6 +24,10 @@ struct OSPRAY_SDK_INTERFACE Instance
   box3f getBounds() const override;
 
   void setEmbreeGeom(RTCScene scene, unsigned int geomID);
+  void setEmbreeGeom(RTCTraversable traversable)
+  {
+    getSh()->traversable = traversable;
+  }
 
   Ref<Group> group;
   const Ref<Group> groupAPI;
diff --git a/modules/cpu/common/Instance.ih b/modules/cpu/common/Instance.ih
@@ -23,8 +23,11 @@ inline uniform AffineSpace3f Instance_getTransform(
     const Instance *uniform self, const uniform float time)
 {
   uniform AffineSpace3f xfm;
-  rtcGetGeometryTransformFromScene(
-      self->scene, self->geomID, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
+  rtcGetGeometryTransformFromTraversable(self->traversable,
+      self->geomID,
+      time,
+      RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
+      &xfm);
 
   return xfm;
 }
@@ -34,8 +37,11 @@ inline AffineSpace3f Instance_getTransform(
     const Instance *uniform self, const float time)
 {
   AffineSpace3f xfm;
-  rtcGetGeometryTransformFromScene(
-      self->scene, self->geomID, time, RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR, &xfm);
+  rtcGetGeometryTransformFromTraversable(self->traversable,
+      self->geomID,
+      time,
+      RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR,
+      &xfm);
 
   return xfm;
 }
diff --git a/modules/cpu/common/InstanceShared.h b/modules/cpu/common/InstanceShared.h
@@ -13,8 +13,8 @@ struct Instance
 {
   Group *group;
 
-  // Scene and geomID are used to call rtcGetGeometryTransformFromScene only
-  RTCScene scene;
+  // traversable and geomID are only for rtcGetGeometryTransformFromTraversable
+  RTCTraversable traversable;
   unsigned int geomID;
 
   AffineSpace3f xfm;
@@ -26,7 +26,7 @@ struct Instance
 #ifdef __cplusplus
   Instance()
       : group(nullptr),
-        scene(nullptr),
+        traversable(nullptr),
         geomID(0),
         xfm(one),
         rcp_xfm(one),
diff --git a/modules/cpu/common/World.cpp b/modules/cpu/common/World.cpp
@@ -47,12 +47,12 @@ static void freeAndNullifyEmbreeScene(RTCScene &scene)
 World::~World()
 {
   // Release Embree scenes
-  freeAndNullifyEmbreeScene(getSh()->embreeSceneHandleGeometries);
+  freeAndNullifyEmbreeScene(embreeSceneHandleGeometries);
 #ifdef OSPRAY_ENABLE_VOLUMES
-  freeAndNullifyEmbreeScene(getSh()->embreeSceneHandleVolumes);
+  freeAndNullifyEmbreeScene(embreeSceneHandleVolumes);
 #endif
 #ifndef OSPRAY_TARGET_SYCL
-  freeAndNullifyEmbreeScene(getSh()->embreeSceneHandleClippers);
+  freeAndNullifyEmbreeScene(embreeSceneHandleClippers);
 #endif
 }
 
@@ -69,19 +69,14 @@ std::string World::toString() const
 
 void World::commit()
 {
-  RTCScene &esGeom = getSh()->embreeSceneHandleGeometries;
-#ifdef OSPRAY_ENABLE_VOLUMES
-  RTCScene &esVol = getSh()->embreeSceneHandleVolumes;
-#endif
-#ifndef OSPRAY_TARGET_SYCL
-  RTCScene &esClip = getSh()->embreeSceneHandleClippers;
-#endif
-
+  RTCScene &esGeom = embreeSceneHandleGeometries;
   freeAndNullifyEmbreeScene(esGeom);
 #ifdef OSPRAY_ENABLE_VOLUMES
+  RTCScene &esVol = embreeSceneHandleVolumes;
   freeAndNullifyEmbreeScene(esVol);
 #endif
 #ifndef OSPRAY_TARGET_SYCL
+  RTCScene &esClip = embreeSceneHandleClippers;
   freeAndNullifyEmbreeScene(esClip);
 #endif
 
@@ -114,18 +109,13 @@ void World::commit()
 
   RTCDevice embreeDevice = getISPCDevice().getEmbreeDevice();
   if (instances) {
-    for (auto &&inst : *instances)
-#ifndef OSPRAY_TARGET_SYCL
-      if (inst->group->sceneClippers)
-        getSh()->numInvertedClippers += inst->group->numInvertedClippers;
-#endif
-
     // Create shared buffers for instance pointers
     instanceArray = devicert::make_buffer_shared_unique<ispc::Instance *>(
         getISPCDevice().getDRTDevice(),
         sizeof(ispc::Instance *) * numInstances);
     getSh()->instances = instanceArray->sharedPtr();
 
+    // Phase 1
     // Populate shared buffer with instance pointers,
     // create Embree instances
     featureFlags.reset();
@@ -144,6 +134,7 @@ void World::commit()
 #endif
 #ifndef OSPRAY_TARGET_SYCL
       if (inst->group->sceneClippers) {
+        getSh()->numInvertedClippers += inst->group->numInvertedClippers;
         addGeometryInstance(
             esClip, inst->group->sceneClippers, inst, embreeDevice, id);
       }
@@ -168,12 +159,14 @@ void World::commit()
     rtcSetSceneFlags(esGeom, static_cast<RTCSceneFlags>(sceneFlags));
     rtcSetSceneBuildQuality(esGeom, buildQuality);
     rtcCommitScene(esGeom);
+    getSh()->embreeTraversableHandleGeometries = rtcGetSceneTraversable(esGeom);
   }
 #ifdef OSPRAY_ENABLE_VOLUMES
   if (esVol) {
     rtcSetSceneFlags(esVol, static_cast<RTCSceneFlags>(sceneFlags));
     rtcSetSceneBuildQuality(esVol, buildQuality);
     rtcCommitScene(esVol);
+    getSh()->embreeTraversableHandleVolumes = rtcGetSceneTraversable(esVol);
   }
 #endif
 #ifndef OSPRAY_TARGET_SYCL
@@ -183,24 +176,42 @@ void World::commit()
             sceneFlags | RTC_SCENE_FLAG_FILTER_FUNCTION_IN_ARGUMENTS));
     rtcSetSceneBuildQuality(esClip, buildQuality);
     rtcCommitScene(esClip);
+    getSh()->embreeTraversableHandleClippers = rtcGetSceneTraversable(esClip);
   }
 #endif
+
+  if (instances) {
+    // Phase 2: set traversable at instances
+    // Note: the same instance transform is set in each scene, the instance
+    // only needs to access one for interpolation (thus fine to overwrite)
+    for (auto &&inst : *instances) {
+      if (inst->group->sceneGeometries)
+        inst->setEmbreeGeom(getSh()->embreeTraversableHandleGeometries);
+#ifdef OSPRAY_ENABLE_VOLUMES
+      if (inst->group->sceneVolumes)
+        inst->setEmbreeGeom(getSh()->embreeTraversableHandleVolumes);
+#endif
+#ifndef OSPRAY_TARGET_SYCL
+      if (inst->group->sceneClippers)
+        inst->setEmbreeGeom(getSh()->embreeTraversableHandleClippers);
+#endif
+    }
+  }
 }
 
 box3f World::getBounds() const
 {
   box3f sceneBounds;
 
   box4f bounds; // NOTE(jda) - Embree expects box4f, NOT box3f...
-  if (getSh()->embreeSceneHandleGeometries) {
-    rtcGetSceneBounds(
-        getSh()->embreeSceneHandleGeometries, (RTCBounds *)&bounds);
+  if (embreeSceneHandleGeometries) {
+    rtcGetSceneBounds(embreeSceneHandleGeometries, (RTCBounds *)&bounds);
     sceneBounds.extend(box3f(vec3f(&bounds.lower[0]), vec3f(&bounds.upper[0])));
   }
 
 #ifdef OSPRAY_ENABLE_VOLUMES
-  if (getSh()->embreeSceneHandleVolumes) {
-    rtcGetSceneBounds(getSh()->embreeSceneHandleVolumes, (RTCBounds *)&bounds);
+  if (embreeSceneHandleVolumes) {
+    rtcGetSceneBounds(embreeSceneHandleVolumes, (RTCBounds *)&bounds);
     sceneBounds.extend(box3f(vec3f(&bounds.lower[0]), vec3f(&bounds.upper[0])));
   }
 #endif
diff --git a/modules/cpu/common/World.h b/modules/cpu/common/World.h
@@ -34,6 +34,14 @@ struct OSPRAY_SDK_INTERFACE World
 
   // Data members //
 
+  RTCScene embreeSceneHandleGeometries{nullptr};
+#ifdef OSPRAY_ENABLE_VOLUMES
+  RTCScene embreeSceneHandleVolumes{nullptr};
+#endif
+#ifndef OSPRAY_TARGET_SYCL
+  RTCScene embreeSceneHandleClippers{nullptr};
+#endif
+
   Ref<const DataT<Instance *>> instances;
   Ref<const DataT<Light *>> lights;
 
diff --git a/modules/cpu/common/World.ih b/modules/cpu/common/World.ih
diff --git a/modules/cpu/common/WorldShared.h b/modules/cpu/common/WorldShared.h
diff --git a/modules/mpi/ospray/common/DistributedWorld.cpp b/modules/mpi/ospray/common/DistributedWorld.cpp
diff --git a/modules/mpi/ospray/common/DistributedWorldShared.h b/modules/mpi/ospray/common/DistributedWorldShared.h
diff --git a/modules/mpi/ospray/render/distributed/DistributedRenderer.ispc b/modules/mpi/ospray/render/distributed/DistributedRenderer.ispc
diff --git a/scripts/superbuild/CMakeLists.txt b/scripts/superbuild/CMakeLists.txt
