Merge pull request #88783 from Chubercik/embree-4.3.1

embree: Update to 4.3.1

39 files changed, 3253 insertions, 1414 deletions

diff --git a/thirdparty/embree/kernels/common/accel.h b/thirdparty/embree/kernels/common/accel.h
index d24326ce92..7d959377ae 100644
--- a/thirdparty/embree/kernels/common/accel.h
+++ b/thirdparty/embree/kernels/common/accel.h
@@ -17,7 +17,7 @@ namespace embree
   {
     ALIGNED_CLASS_(16);
   public:
-    enum Type { TY_UNKNOWN = 0, TY_ACCELN = 1, TY_ACCEL_INSTANCE = 2, TY_BVH4 = 3, TY_BVH8 = 4 };
+    enum Type { TY_UNKNOWN = 0, TY_ACCELN = 1, TY_ACCEL_INSTANCE = 2, TY_BVH4 = 3, TY_BVH8 = 4, TY_GPU = 5 };
 
   public:
     AccelData (const Type type) 
@@ -73,61 +73,49 @@ namespace embree
     /*! Type of intersect function pointer for single rays. */
     typedef void (*IntersectFunc)(Intersectors* This,  /*!< this pointer to accel */
                                   RTCRayHit& ray,      /*!< ray to intersect */
-                                  IntersectContext* context);
+                                  RayQueryContext* context);
     
     /*! Type of intersect function pointer for ray packets of size 4. */
     typedef void (*IntersectFunc4)(const void* valid,  /*!< pointer to valid mask */
                                    Intersectors* This, /*!< this pointer to accel */
                                    RTCRayHit4& ray,    /*!< ray packet to intersect */
-                                   IntersectContext* context);
+                                   RayQueryContext* context);
     
     /*! Type of intersect function pointer for ray packets of size 8. */
     typedef void (*IntersectFunc8)(const void* valid,  /*!< pointer to valid mask */
                                    Intersectors* This, /*!< this pointer to accel */
                                    RTCRayHit8& ray,    /*!< ray packet to intersect */
-                                   IntersectContext* context);
+                                   RayQueryContext* context);
     
     /*! Type of intersect function pointer for ray packets of size 16. */
     typedef void (*IntersectFunc16)(const void* valid,  /*!< pointer to valid mask */
                                     Intersectors* This, /*!< this pointer to accel */
                                     RTCRayHit16& ray,   /*!< ray packet to intersect */
-                                    IntersectContext* context);
+                                    RayQueryContext* context);
 
-    /*! Type of intersect function pointer for ray packets of size N. */
-    typedef void (*IntersectFuncN)(Intersectors* This, /*!< this pointer to accel */
-                                   RTCRayHitN** ray,   /*!< ray stream to intersect */
-                                   const size_t N,     /*!< number of rays in stream */
-                                   IntersectContext* context /*!< layout flags */);
-    
-    
     /*! Type of occlusion function pointer for single rays. */
     typedef void (*OccludedFunc) (Intersectors* This, /*!< this pointer to accel */
                                   RTCRay& ray,        /*!< ray to test occlusion */
-                                  IntersectContext* context);
+                                  RayQueryContext* context);
     
     /*! Type of occlusion function pointer for ray packets of size 4. */
     typedef void (*OccludedFunc4) (const void* valid,  /*!< pointer to valid mask */
                                    Intersectors* This, /*!< this pointer to accel */
                                    RTCRay4& ray,       /*!< ray packet to test occlusion. */
-                                   IntersectContext* context);
+                                   RayQueryContext* context);
     
     /*! Type of occlusion function pointer for ray packets of size 8. */
     typedef void (*OccludedFunc8) (const void* valid,  /*!< pointer to valid mask */
                                    Intersectors* This, /*!< this pointer to accel */
                                    RTCRay8& ray,       /*!< ray packet to test occlusion. */
-                                   IntersectContext* context);
+                                   RayQueryContext* context);
     
     /*! Type of occlusion function pointer for ray packets of size 16. */
     typedef void (*OccludedFunc16) (const void* valid,  /*!< pointer to valid mask */
                                     Intersectors* This, /*!< this pointer to accel */
                                     RTCRay16& ray,      /*!< ray packet to test occlusion. */
-                                    IntersectContext* context);
+                                    RayQueryContext* context);
 
-    /*! Type of intersect function pointer for ray packets of size N. */
-    typedef void (*OccludedFuncN)(Intersectors* This, /*!< this pointer to accel */
-                                  RTCRayN** ray,      /*!< ray stream to test occlusion */
-                                  const size_t N,     /*!< number of rays in stream */
-                                  IntersectContext* context /*!< layout flags */);
     typedef void (*ErrorFunc) ();
 
     struct Collider
@@ -217,30 +205,13 @@ namespace embree
       const char* name;
     };
 
-    struct IntersectorN 
-    {
-      IntersectorN (ErrorFunc error = nullptr)
-      : intersect((IntersectFuncN)error), occluded((OccludedFuncN)error), name(nullptr) {}
-
-      IntersectorN (IntersectFuncN intersect, OccludedFuncN occluded, const char* name)
-      : intersect(intersect), occluded(occluded), name(name) {}
-
-      operator bool() const { return name; }
-      
-    public:
-      static const char* type;
-      IntersectFuncN intersect;
-      OccludedFuncN occluded;
-      const char* name;
-    };
-   
     struct Intersectors 
     {
       Intersectors() 
-      : ptr(nullptr), leafIntersector(nullptr), collider(nullptr), intersector1(nullptr), intersector4(nullptr), intersector8(nullptr), intersector16(nullptr), intersectorN(nullptr) {}
+      : ptr(nullptr), leafIntersector(nullptr), collider(nullptr), intersector1(nullptr), intersector4(nullptr), intersector8(nullptr), intersector16(nullptr) {}
 
       Intersectors (ErrorFunc error) 
-      : ptr(nullptr), leafIntersector(nullptr), collider(error), intersector1(error), intersector4(error), intersector8(error), intersector16(error), intersectorN(error) {}
+      : ptr(nullptr), leafIntersector(nullptr), collider(error), intersector1(error), intersector4(error), intersector8(error), intersector16(error) {}
 
       void print(size_t ident) 
       {
@@ -264,10 +235,6 @@ namespace embree
           for (size_t i=0; i<ident; i++) std::cout << " ";
           std::cout << "intersector16 = " << intersector16.name << std::endl;
         }
-        if (intersectorN.name) {
-          for (size_t i=0; i<ident; i++) std::cout << " ";
-          std::cout << "intersectorN = " << intersectorN.name << std::endl;
-        }        
       }
 
       void select(bool filter)
@@ -284,10 +251,6 @@ namespace embree
           if (filter) intersector16 = intersector16_filter;
           else         intersector16 = intersector16_nofilter;
         }
-        if (intersectorN_filter) {
-          if (filter) intersectorN = intersectorN_filter;
-          else        intersectorN = intersectorN_nofilter;
-        }        
       }
 
       __forceinline bool pointQuery (PointQuery* query, PointQueryContext* context) {
@@ -302,133 +265,138 @@ namespace embree
       }
 
       /*! Intersects a single ray with the scene. */
-      __forceinline void intersect (RTCRayHit& ray, IntersectContext* context) {
+      __forceinline void intersect (RTCRayHit& ray, RayQueryContext* context) {
         assert(intersector1.intersect);
         intersector1.intersect(this,ray,context);
       }
 
       /*! Intersects a packet of 4 rays with the scene. */
-      __forceinline void intersect4 (const void* valid, RTCRayHit4& ray, IntersectContext* context) {
+      __forceinline void intersect4 (const void* valid, RTCRayHit4& ray, RayQueryContext* context) {
         assert(intersector4.intersect);
         intersector4.intersect(valid,this,ray,context);
       }
       
       /*! Intersects a packet of 8 rays with the scene. */
-      __forceinline void intersect8 (const void* valid, RTCRayHit8& ray, IntersectContext* context) {
+      __forceinline void intersect8 (const void* valid, RTCRayHit8& ray, RayQueryContext* context) {
         assert(intersector8.intersect);
         intersector8.intersect(valid,this,ray,context);
       }
       
       /*! Intersects a packet of 16 rays with the scene. */
-      __forceinline void intersect16 (const void* valid, RTCRayHit16& ray, IntersectContext* context) {
+      __forceinline void intersect16 (const void* valid, RTCRayHit16& ray, RayQueryContext* context) {
         assert(intersector16.intersect);
         intersector16.intersect(valid,this,ray,context);
       }
+
+      /*! Intersects a packet of 4 rays with the scene. */
+      __forceinline void intersect (const void* valid, RTCRayHit4& ray, RayQueryContext* context) {
+        assert(intersector4.intersect);
+        intersector4.intersect(valid,this,ray,context);
+      }
       
-      /*! Intersects a stream of N rays in SOA layout with the scene. */
-      __forceinline void intersectN (RTCRayHitN** rayN, const size_t N, IntersectContext* context)
-      {
-        assert(intersectorN.intersect);
-        intersectorN.intersect(this,rayN,N,context);
+      /*! Intersects a packet of 8 rays with the scene. */
+      __forceinline void intersect (const void* valid, RTCRayHit8& ray, RayQueryContext* context) {
+        assert(intersector8.intersect);
+        intersector8.intersect(valid,this,ray,context);
+      }
+      
+      /*! Intersects a packet of 16 rays with the scene. */
+      __forceinline void intersect (const void* valid, RTCRayHit16& ray, RayQueryContext* context) {
+        assert(intersector16.intersect);
+        intersector16.intersect(valid,this,ray,context);
       }
       
 #if defined(__SSE__) || defined(__ARM_NEON)
-      __forceinline void intersect(const vbool4& valid, RayHitK<4>& ray, IntersectContext* context) {
+      __forceinline void intersect(const vbool4& valid, RayHitK<4>& ray, RayQueryContext* context) {
         const vint<4> mask = valid.mask32();
         intersect4(&mask,(RTCRayHit4&)ray,context);
       }
 #endif
 #if defined(__AVX__)
-      __forceinline void intersect(const vbool8& valid, RayHitK<8>& ray, IntersectContext* context) {
+      __forceinline void intersect(const vbool8& valid, RayHitK<8>& ray, RayQueryContext* context) {
         const vint<8> mask = valid.mask32();
         intersect8(&mask,(RTCRayHit8&)ray,context);
       }
 #endif
 #if defined(__AVX512F__)
-      __forceinline void intersect(const vbool16& valid, RayHitK<16>& ray, IntersectContext* context) {
+      __forceinline void intersect(const vbool16& valid, RayHitK<16>& ray, RayQueryContext* context) {
         const vint<16> mask = valid.mask32();
         intersect16(&mask,(RTCRayHit16&)ray,context);
       }
 #endif
       
-      template<int K>
-      __forceinline void intersectN (RayHitK<K>** rayN, const size_t N, IntersectContext* context)
-      {
-        intersectN((RTCRayHitN**)rayN,N,context);
-      }
-
       /*! Tests if single ray is occluded by the scene. */
-      __forceinline void occluded (RTCRay& ray, IntersectContext* context) {
+      __forceinline void occluded (RTCRay& ray, RayQueryContext* context) {
         assert(intersector1.occluded);
         intersector1.occluded(this,ray,context);
       }
       
       /*! Tests if a packet of 4 rays is occluded by the scene. */
-      __forceinline void occluded4 (const void* valid, RTCRay4& ray, IntersectContext* context) {
+      __forceinline void occluded4 (const void* valid, RTCRay4& ray, RayQueryContext* context) {
         assert(intersector4.occluded);
         intersector4.occluded(valid,this,ray,context);
       }
       
       /*! Tests if a packet of 8 rays is occluded by the scene. */
-      __forceinline void occluded8 (const void* valid, RTCRay8& ray, IntersectContext* context) {
+      __forceinline void occluded8 (const void* valid, RTCRay8& ray, RayQueryContext* context) {
         assert(intersector8.occluded);
         intersector8.occluded(valid,this,ray,context);
       }
       
       /*! Tests if a packet of 16 rays is occluded by the scene. */
-      __forceinline void occluded16 (const void* valid, RTCRay16& ray, IntersectContext* context) {
+      __forceinline void occluded16 (const void* valid, RTCRay16& ray, RayQueryContext* context) {
         assert(intersector16.occluded);
         intersector16.occluded(valid,this,ray,context);
       }
+
+      /*! Tests if a packet of 4 rays is occluded by the scene. */
+      __forceinline void occluded (const void* valid, RTCRay4& ray, RayQueryContext* context) {
+        assert(intersector4.occluded);
+        intersector4.occluded(valid,this,ray,context);
+      }
       
-      /*! Tests if a stream of N rays in SOA layout is occluded by the scene. */
-      __forceinline void occludedN (RTCRayN** rayN, const size_t N, IntersectContext* context)
-      {
-        assert(intersectorN.occluded);
-        intersectorN.occluded(this,rayN,N,context);
+      /*! Tests if a packet of 8 rays is occluded by the scene. */
+      __forceinline void occluded (const void* valid, RTCRay8& ray, RayQueryContext* context) {
+        assert(intersector8.occluded);
+        intersector8.occluded(valid,this,ray,context);
+      }
+      
+      /*! Tests if a packet of 16 rays is occluded by the scene. */
+      __forceinline void occluded (const void* valid, RTCRay16& ray, RayQueryContext* context) {
+        assert(intersector16.occluded);
+        intersector16.occluded(valid,this,ray,context);
       }
       
 #if defined(__SSE__) || defined(__ARM_NEON)
-      __forceinline void occluded(const vbool4& valid, RayK<4>& ray, IntersectContext* context) {
+      __forceinline void occluded(const vbool4& valid, RayK<4>& ray, RayQueryContext* context) {
         const vint<4> mask = valid.mask32();
         occluded4(&mask,(RTCRay4&)ray,context);
       }
 #endif
 #if defined(__AVX__)
-      __forceinline void occluded(const vbool8& valid, RayK<8>& ray, IntersectContext* context) {
+      __forceinline void occluded(const vbool8& valid, RayK<8>& ray, RayQueryContext* context) {
         const vint<8> mask = valid.mask32();
         occluded8(&mask,(RTCRay8&)ray,context);
       }
 #endif
 #if defined(__AVX512F__)
-      __forceinline void occluded(const vbool16& valid, RayK<16>& ray, IntersectContext* context) {
+      __forceinline void occluded(const vbool16& valid, RayK<16>& ray, RayQueryContext* context) {
         const vint<16> mask = valid.mask32();
         occluded16(&mask,(RTCRay16&)ray,context);
       }
 #endif
 
-      template<int K>
-      __forceinline void occludedN (RayK<K>** rayN, const size_t N, IntersectContext* context)
-      {
-        occludedN((RTCRayN**)rayN,N,context);
-      }
-
       /*! Tests if single ray is occluded by the scene. */
-      __forceinline void intersect(RTCRay& ray, IntersectContext* context) {
+      __forceinline void intersect(RTCRay& ray, RayQueryContext* context) {
         occluded(ray, context);
       }
 
       /*! Tests if a packet of K rays is occluded by the scene. */
       template<int K>
-      __forceinline void intersect(const vbool<K>& valid, RayK<K>& ray, IntersectContext* context) {
+      __forceinline void intersect(const vbool<K>& valid, RayK<K>& ray, RayQueryContext* context) {
         occluded(valid, ray, context);
       }
 
-      /*! Tests if a packet of N rays in SOA layout is occluded by the scene. */
-      template<int K>
-      __forceinline void intersectN(RayK<K>** rayN, const size_t N, IntersectContext* context) {
-        occludedN(rayN, N, context);
-      }
       
     public:
       AccelData* ptr;
@@ -444,9 +412,6 @@ namespace embree
       Intersector16 intersector16;
       Intersector16 intersector16_filter;
       Intersector16 intersector16_nofilter;
-      IntersectorN intersectorN;
-      IntersectorN intersectorN_filter;
-      IntersectorN intersectorN_nofilter;      
     };
   
   public:
@@ -506,51 +471,4 @@ namespace embree
                                 (Accel::OccludedFunc16)intersector::occluded,   \
                                 TOSTRING(isa) "::" TOSTRING(symbol));           \
   }
-
-#define DEFINE_INTERSECTORN(symbol,intersector)                               \
-  Accel::IntersectorN symbol() {                                              \
-    return Accel::IntersectorN((Accel::IntersectFuncN)intersector::intersect, \
-                               (Accel::OccludedFuncN)intersector::occluded,   \
-                               TOSTRING(isa) "::" TOSTRING(symbol));          \
-  }
-
-  /* ray stream filter interface */
-  typedef void (*intersectStreamAOS_func)(Scene* scene, RTCRayHit*  _rayN, const size_t N, const size_t stride, IntersectContext* context);
-  typedef void (*intersectStreamAOP_func)(Scene* scene, RTCRayHit** _rayN, const size_t N, IntersectContext* context);
-  typedef void (*intersectStreamSOA_func)(Scene* scene, char* rayN, const size_t N, const size_t streams, const size_t stream_offset, IntersectContext* context);
-  typedef void (*intersectStreamSOP_func)(Scene* scene, const RTCRayHitNp* rayN, const size_t N, IntersectContext* context);
-
-  typedef void (*occludedStreamAOS_func)(Scene* scene, RTCRay*  _rayN, const size_t N, const size_t stride, IntersectContext* context);
-  typedef void (*occludedStreamAOP_func)(Scene* scene, RTCRay** _rayN, const size_t N, IntersectContext* context);
-  typedef void (*occludedStreamSOA_func)(Scene* scene, char* rayN, const size_t N, const size_t streams, const size_t stream_offset, IntersectContext* context);
-  typedef void (*occludedStreamSOP_func)(Scene* scene, const RTCRayNp* rayN, const size_t N, IntersectContext* context);
-
-  struct RayStreamFilterFuncs
-  {
-    RayStreamFilterFuncs()
-    : intersectAOS(nullptr), intersectAOP(nullptr), intersectSOA(nullptr), intersectSOP(nullptr),
-      occludedAOS(nullptr),  occludedAOP(nullptr),  occludedSOA(nullptr),  occludedSOP(nullptr) {}
-
-    RayStreamFilterFuncs(void (*ptr) ())
-    : intersectAOS((intersectStreamAOS_func) ptr), intersectAOP((intersectStreamAOP_func) ptr), intersectSOA((intersectStreamSOA_func) ptr), intersectSOP((intersectStreamSOP_func) ptr),
-      occludedAOS((occludedStreamAOS_func) ptr),   occludedAOP((occludedStreamAOP_func) ptr),   occludedSOA((occludedStreamSOA_func) ptr),   occludedSOP((occludedStreamSOP_func) ptr) {}
-
-    RayStreamFilterFuncs(intersectStreamAOS_func intersectAOS, intersectStreamAOP_func intersectAOP, intersectStreamSOA_func intersectSOA, intersectStreamSOP_func intersectSOP,
-                         occludedStreamAOS_func  occludedAOS,  occludedStreamAOP_func  occludedAOP,  occludedStreamSOA_func  occludedSOA,  occludedStreamSOP_func  occludedSOP)
-    : intersectAOS(intersectAOS), intersectAOP(intersectAOP), intersectSOA(intersectSOA), intersectSOP(intersectSOP),
-      occludedAOS(occludedAOS),   occludedAOP(occludedAOP),   occludedSOA(occludedSOA),   occludedSOP(occludedSOP) {}
-
-  public:
-    intersectStreamAOS_func intersectAOS;
-    intersectStreamAOP_func intersectAOP;
-    intersectStreamSOA_func intersectSOA;
-    intersectStreamSOP_func intersectSOP;
-
-    occludedStreamAOS_func occludedAOS;
-    occludedStreamAOP_func occludedAOP;
-    occludedStreamSOA_func occludedSOA;
-    occludedStreamSOP_func occludedSOP;
-  }; 
-
-  typedef RayStreamFilterFuncs (*RayStreamFilterFuncsType)();
 }
diff --git a/thirdparty/embree/kernels/common/acceln.cpp b/thirdparty/embree/kernels/common/acceln.cpp
index 111c62083d..9edb684db7 100644
--- a/thirdparty/embree/kernels/common/acceln.cpp
+++ b/thirdparty/embree/kernels/common/acceln.cpp
@@ -3,7 +3,7 @@
 
 #include "acceln.h"
 #include "ray.h"
-#include "../../include/embree3/rtcore_ray.h"
+#include "../../include/embree4/rtcore_ray.h"
 #include "../../common/algorithms/parallel_for.h"
 
 namespace embree
@@ -41,7 +41,7 @@ namespace embree
     return changed;
   }
 
-  void AccelN::intersect (Accel::Intersectors* This_in, RTCRayHit& ray, IntersectContext* context) 
+  void AccelN::intersect (Accel::Intersectors* This_in, RTCRayHit& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++)
@@ -49,7 +49,7 @@ namespace embree
         This->accels[i]->intersectors.intersect(ray,context);
   }
 
-  void AccelN::intersect4 (const void* valid, Accel::Intersectors* This_in, RTCRayHit4& ray, IntersectContext* context) 
+  void AccelN::intersect4 (const void* valid, Accel::Intersectors* This_in, RTCRayHit4& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++)
@@ -57,7 +57,7 @@ namespace embree
         This->accels[i]->intersectors.intersect4(valid,ray,context);
   }
 
-  void AccelN::intersect8 (const void* valid, Accel::Intersectors* This_in, RTCRayHit8& ray, IntersectContext* context) 
+  void AccelN::intersect8 (const void* valid, Accel::Intersectors* This_in, RTCRayHit8& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++)
@@ -65,7 +65,7 @@ namespace embree
         This->accels[i]->intersectors.intersect8(valid,ray,context);
   }
 
-  void AccelN::intersect16 (const void* valid, Accel::Intersectors* This_in, RTCRayHit16& ray, IntersectContext* context) 
+  void AccelN::intersect16 (const void* valid, Accel::Intersectors* This_in, RTCRayHit16& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++)
@@ -73,15 +73,7 @@ namespace embree
         This->accels[i]->intersectors.intersect16(valid,ray,context);
   }
 
-  void AccelN::intersectN (Accel::Intersectors* This_in, RTCRayHitN** ray, const size_t N, IntersectContext* context)
-  {
-    AccelN* This = (AccelN*)This_in->ptr;
-    for (size_t i=0; i<This->accels.size(); i++)
-      if (!This->accels[i]->isEmpty())
-        This->accels[i]->intersectors.intersectN(ray,N,context);
-  }
-
-  void AccelN::occluded (Accel::Intersectors* This_in, RTCRay& ray, IntersectContext* context) 
+  void AccelN::occluded (Accel::Intersectors* This_in, RTCRay& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++) {
@@ -91,7 +83,7 @@ namespace embree
     }
   }
 
-  void AccelN::occluded4 (const void* valid, Accel::Intersectors* This_in, RTCRay4& ray, IntersectContext* context) 
+  void AccelN::occluded4 (const void* valid, Accel::Intersectors* This_in, RTCRay4& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++) {
@@ -105,7 +97,7 @@ namespace embree
     }
   }
 
-  void AccelN::occluded8 (const void* valid, Accel::Intersectors* This_in, RTCRay8& ray, IntersectContext* context) 
+  void AccelN::occluded8 (const void* valid, Accel::Intersectors* This_in, RTCRay8& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++) {
@@ -121,7 +113,7 @@ namespace embree
     }
   }
 
-  void AccelN::occluded16 (const void* valid, Accel::Intersectors* This_in, RTCRay16& ray, IntersectContext* context) 
+  void AccelN::occluded16 (const void* valid, Accel::Intersectors* This_in, RTCRay16& ray, RayQueryContext* context) 
   {
     AccelN* This = (AccelN*)This_in->ptr;
     for (size_t i=0; i<This->accels.size(); i++) {
@@ -141,15 +133,6 @@ namespace embree
     }
   }
 
-  void AccelN::occludedN (Accel::Intersectors* This_in, RTCRayN** ray, const size_t N, IntersectContext* context)
-  {
-    AccelN* This = (AccelN*)This_in->ptr;
-    size_t M = N;
-    for (size_t i=0; i<This->accels.size(); i++)
-      if (!This->accels[i]->isEmpty())
-        This->accels[i]->intersectors.occludedN(ray,M,context);
-  }
-
   void AccelN::accels_print(size_t ident)
   {
     for (size_t i=0; i<accels.size(); i++)
@@ -201,7 +184,6 @@ namespace embree
       intersectors.intersector4  = Intersector4(&intersect4,&occluded4,valid4 ? "AccelN::intersector4" : nullptr);
       intersectors.intersector8  = Intersector8(&intersect8,&occluded8,valid8 ? "AccelN::intersector8" : nullptr);
       intersectors.intersector16 = Intersector16(&intersect16,&occluded16,valid16 ? "AccelN::intersector16": nullptr);
-      intersectors.intersectorN  = IntersectorN(&intersectN,&occludedN,"AccelN::intersectorN");
 
       /*! calculate bounds */
       bounds = empty;
diff --git a/thirdparty/embree/kernels/common/acceln.h b/thirdparty/embree/kernels/common/acceln.h
index 0445b2e811..cc3406826c 100644
--- a/thirdparty/embree/kernels/common/acceln.h
+++ b/thirdparty/embree/kernels/common/acceln.h
@@ -22,18 +22,16 @@ namespace embree
     static bool pointQuery (Accel::Intersectors* This, PointQuery* query, PointQueryContext* context);
 
   public:
-    static void intersect (Accel::Intersectors* This, RTCRayHit& ray, IntersectContext* context);
-    static void intersect4 (const void* valid, Accel::Intersectors* This, RTCRayHit4& ray, IntersectContext* context);
-    static void intersect8 (const void* valid, Accel::Intersectors* This, RTCRayHit8& ray, IntersectContext* context);
-    static void intersect16 (const void* valid, Accel::Intersectors* This, RTCRayHit16& ray, IntersectContext* context);
-    static void intersectN (Accel::Intersectors* This, RTCRayHitN** ray, const size_t N, IntersectContext* context);
+    static void intersect (Accel::Intersectors* This, RTCRayHit& ray, RayQueryContext* context);
+    static void intersect4 (const void* valid, Accel::Intersectors* This, RTCRayHit4& ray, RayQueryContext* context);
+    static void intersect8 (const void* valid, Accel::Intersectors* This, RTCRayHit8& ray, RayQueryContext* context);
+    static void intersect16 (const void* valid, Accel::Intersectors* This, RTCRayHit16& ray, RayQueryContext* context);
 
   public:
-    static void occluded (Accel::Intersectors* This, RTCRay& ray, IntersectContext* context);
-    static void occluded4 (const void* valid, Accel::Intersectors* This, RTCRay4& ray, IntersectContext* context);
-    static void occluded8 (const void* valid, Accel::Intersectors* This, RTCRay8& ray, IntersectContext* context);
-    static void occluded16 (const void* valid, Accel::Intersectors* This, RTCRay16& ray, IntersectContext* context);
-    static void occludedN (Accel::Intersectors* This, RTCRayN** ray, const size_t N, IntersectContext* context);
+    static void occluded (Accel::Intersectors* This, RTCRay& ray, RayQueryContext* context);
+    static void occluded4 (const void* valid, Accel::Intersectors* This, RTCRay4& ray, RayQueryContext* context);
+    static void occluded8 (const void* valid, Accel::Intersectors* This, RTCRay8& ray, RayQueryContext* context);
+    static void occluded16 (const void* valid, Accel::Intersectors* This, RTCRay16& ray, RayQueryContext* context);
 
   public:
     void accels_print(size_t ident);
diff --git a/thirdparty/embree/kernels/common/accelset.h b/thirdparty/embree/kernels/common/accelset.h
index 1b67120c97..f78830e397 100644
--- a/thirdparty/embree/kernels/common/accelset.h
+++ b/thirdparty/embree/kernels/common/accelset.h
@@ -17,11 +17,15 @@ namespace embree
   struct IntersectFunctionNArguments : public RTCIntersectFunctionNArguments
   {
     Geometry* geometry;
+    RTCScene forward_scene;
+    RTCIntersectArguments* args;
   };
 
   struct OccludedFunctionNArguments : public RTCOccludedFunctionNArguments
   {
     Geometry* geometry;
+    RTCScene forward_scene;
+    RTCIntersectArguments* args;
   };
 
   /*! Base class for set of acceleration structures. */
@@ -138,10 +142,9 @@ namespace embree
   public:
 
       /*! Intersects a single ray with the scene. */
-      __forceinline void intersect (RayHit& ray, unsigned int geomID, unsigned int primID, IntersectContext* context) 
+      __forceinline bool intersect (RayHit& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context) 
       {
         assert(primID < size());
-        assert(intersectorN.intersect);
         
         int mask = -1;
         IntersectFunctionNArguments args;
@@ -153,17 +156,94 @@ namespace embree
         args.geomID = geomID;
         args.primID = primID;
         args.geometry = this;
+        args.forward_scene = nullptr;
+        args.args = context->args;
+
+        IntersectFuncN intersectFunc = nullptr;
+        intersectFunc = intersectorN.intersect;
         
-        intersectorN.intersect(&args);
+        if (context->getIntersectFunction())
+          intersectFunc = context->getIntersectFunction();
+
+        assert(intersectFunc);
+        intersectFunc(&args);
+
+        return mask != 0;
       }
 
       /*! Tests if single ray is occluded by the scene. */
-      __forceinline void occluded (Ray& ray, unsigned int geomID, unsigned int primID, IntersectContext* context)
+      __forceinline bool occluded (Ray& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context)
       {
         assert(primID < size());
-        assert(intersectorN.occluded);
+
+        int mask = -1;
+        OccludedFunctionNArguments args;
+        args.valid = &mask;
+        args.geometryUserPtr = userPtr;
+        args.context = context->user;
+        args.ray = (RTCRayN*)&ray;
+        args.N = 1;
+        args.geomID = geomID;
+        args.primID = primID;
+        args.geometry = this;
+        args.forward_scene = nullptr;
+        args.args = context->args;
+
+        OccludedFuncN occludedFunc = nullptr;
+        occludedFunc = intersectorN.occluded;
+
+        if (context->getOccludedFunction())
+          occludedFunc = context->getOccludedFunction();
+
+        assert(occludedFunc);
+        occludedFunc(&args);
+
+        return mask != 0;
+      }
+
+      /*! Intersects a single ray with the scene. */
+    __forceinline bool intersect (RayHit& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context, RTCScene& forward_scene) 
+    {
+        assert(primID < size());
         
         int mask = -1;
+        IntersectFunctionNArguments args;
+        args.valid = &mask;
+        args.geometryUserPtr = userPtr;
+        args.context = context->user;
+        args.rayhit = (RTCRayHitN*)&ray;
+        args.N = 1;
+        args.geomID = geomID;
+        args.primID = primID;
+        args.geometry = this;
+        args.forward_scene = nullptr;
+        args.args = nullptr;
+
+        typedef void (*RTCIntersectFunctionSYCL)(const void* args);
+        RTCIntersectFunctionSYCL intersectFunc = nullptr;
+        
+#if EMBREE_SYCL_GEOMETRY_CALLBACK
+        if (context->args->feature_mask & RTC_FEATURE_FLAG_USER_GEOMETRY_CALLBACK_IN_GEOMETRY)
+          intersectFunc = (RTCIntersectFunctionSYCL) intersectorN.intersect;
+#endif
+        
+        if (context->args->feature_mask & RTC_FEATURE_FLAG_USER_GEOMETRY_CALLBACK_IN_ARGUMENTS)
+          if (context->getIntersectFunction())
+            intersectFunc = (RTCIntersectFunctionSYCL) context->getIntersectFunction();
+
+        if (intersectFunc)
+          intersectFunc(&args);
+        
+        forward_scene = args.forward_scene;
+        return mask != 0;
+      }
+
+      /*! Tests if single ray is occluded by the scene. */
+    __forceinline bool occluded (Ray& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context, RTCScene& forward_scene)
+      {
+        assert(primID < size());
+
+        int mask = -1;
         OccludedFunctionNArguments args;
         args.valid = &mask;
         args.geometryUserPtr = userPtr;
@@ -173,16 +253,33 @@ namespace embree
         args.geomID = geomID;
         args.primID = primID;
         args.geometry = this;
+        args.forward_scene = nullptr;
+        args.args = nullptr;
+
+        typedef void (*RTCOccludedFunctionSYCL)(const void* args);
+        RTCOccludedFunctionSYCL occludedFunc = nullptr;
+
+#if EMBREE_SYCL_GEOMETRY_CALLBACK
+        if (context->args->feature_mask & RTC_FEATURE_FLAG_USER_GEOMETRY_CALLBACK_IN_GEOMETRY)
+          occludedFunc = (RTCOccludedFunctionSYCL) intersectorN.occluded;
+#endif
+        
+        if (context->args->feature_mask & RTC_FEATURE_FLAG_USER_GEOMETRY_CALLBACK_IN_ARGUMENTS)
+          if (context->getOccludedFunction())
+            occludedFunc = (RTCOccludedFunctionSYCL) context->getOccludedFunction();
+
+        if (occludedFunc)
+          occludedFunc(&args);
         
-        intersectorN.occluded(&args);
+        forward_scene = args.forward_scene;
+        return mask != 0;
       }
-   
+
       /*! Intersects a packet of K rays with the scene. */
       template<int K>
-        __forceinline void intersect (const vbool<K>& valid, RayHitK<K>& ray, unsigned int geomID, unsigned int primID, IntersectContext* context) 
+        __forceinline void intersect (const vbool<K>& valid, RayHitK<K>& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context) 
       {
         assert(primID < size());
-        assert(intersectorN.intersect);
         
         vint<K> mask = valid.mask32();
         IntersectFunctionNArguments args;
@@ -194,16 +291,24 @@ namespace embree
         args.geomID = geomID;
         args.primID = primID;
         args.geometry = this;
-         
-        intersectorN.intersect(&args);
+        args.forward_scene = nullptr;
+        args.args = context->args;
+
+        IntersectFuncN intersectFunc = nullptr;
+        intersectFunc = intersectorN.intersect;
+        
+        if (context->getIntersectFunction())
+          intersectFunc = context->getIntersectFunction();
+
+        assert(intersectFunc);
+        intersectFunc(&args);
       }
 
       /*! Tests if a packet of K rays is occluded by the scene. */
       template<int K>
-        __forceinline void occluded (const vbool<K>& valid, RayK<K>& ray, unsigned int geomID, unsigned int primID, IntersectContext* context)
+        __forceinline void occluded (const vbool<K>& valid, RayK<K>& ray, unsigned int geomID, unsigned int primID, RayQueryContext* context)
       {
         assert(primID < size());
-        assert(intersectorN.occluded);
         
         vint<K> mask = valid.mask32();
         OccludedFunctionNArguments args;
@@ -215,8 +320,17 @@ namespace embree
         args.geomID = geomID;
         args.primID = primID;
         args.geometry = this;
+        args.forward_scene = nullptr;
+        args.args = context->args;
+
+        OccludedFuncN occludedFunc = nullptr;
+        occludedFunc = intersectorN.occluded;
         
-        intersectorN.occluded(&args);
+        if (context->getOccludedFunction())
+          occludedFunc = context->getOccludedFunction();
+
+        assert(occludedFunc);
+        occludedFunc(&args);
       }
 
     public:
diff --git a/thirdparty/embree/kernels/common/alloc.cpp b/thirdparty/embree/kernels/common/alloc.cpp
index 38a76225f4..cc2f9976f2 100644
--- a/thirdparty/embree/kernels/common/alloc.cpp
+++ b/thirdparty/embree/kernels/common/alloc.cpp
@@ -10,7 +10,7 @@
 namespace embree
 {
   __thread FastAllocator::ThreadLocal2* FastAllocator::thread_local_allocator2 = nullptr;
-  SpinLock FastAllocator::s_thread_local_allocators_lock;
+  MutexSys FastAllocator::s_thread_local_allocators_lock;
   std::vector<std::unique_ptr<FastAllocator::ThreadLocal2>> FastAllocator::s_thread_local_allocators;
    
   struct fast_allocator_regression_test : public RegressionTest
diff --git a/thirdparty/embree/kernels/common/alloc.h b/thirdparty/embree/kernels/common/alloc.h
index 12769df2c8..840d48c327 100644
--- a/thirdparty/embree/kernels/common/alloc.h
+++ b/thirdparty/embree/kernels/common/alloc.h
@@ -6,11 +6,9 @@
 #include "default.h"
 #include "device.h"
 #include "scene.h"
-#include "primref.h"
+#include "../builders/primref.h"
 
-#if defined(APPLE) && defined(__aarch64__)
-#include <mutex>
-#endif
+#include "../../common/tasking/taskscheduler.h"
 
 namespace embree
 {
@@ -18,7 +16,7 @@ namespace embree
   {
     /*! maximum supported alignment */
     static const size_t maxAlignment = 64;
-
+    
     /*! maximum allocation size */
 
     /* default settings */
@@ -39,14 +37,14 @@ namespace embree
     public:
 
       /*! Constructor for usage with ThreadLocalData */
-      __forceinline ThreadLocal (ThreadLocal2* parent) 
-	: parent(parent), ptr(nullptr), cur(0), end(0), allocBlockSize(0), bytesUsed(0), bytesWasted(0) {}
+      __forceinline ThreadLocal (ThreadLocal2* parent)
+        : parent(parent), ptr(nullptr), cur(0), end(0), allocBlockSize(0), bytesUsed(0), bytesWasted(0) {}
 
       /*! initialize allocator */
-      void init(FastAllocator* alloc) 
+      void init(FastAllocator* alloc)
       {
         ptr = nullptr;
-	cur = end = 0;
+        cur = end = 0;
         bytesUsed = 0;
         bytesWasted = 0;
         allocBlockSize = 0;
@@ -54,64 +52,62 @@ namespace embree
       }
 
       /* Allocate aligned memory from the threads memory block. */
-      __forceinline void* malloc(FastAllocator* alloc, size_t bytes, size_t align = 16) 
+      __forceinline void* malloc(FastAllocator* alloc, size_t bytes, size_t align = 16)
       {
         /* bind the thread local allocator to the proper FastAllocator*/
         parent->bind(alloc);
 
         assert(align <= maxAlignment);
-	bytesUsed += bytes;
+        bytesUsed += bytes;
 
         /* try to allocate in local block */
-	size_t ofs = (align - cur) & (align-1);
+        size_t ofs = (align - cur) & (align-1);
         cur += bytes + ofs;
         if (likely(cur <= end)) { bytesWasted += ofs; return &ptr[cur - bytes]; }
-	cur -= bytes + ofs;
-        
+        cur -= bytes + ofs;
+
         /* if allocation is too large allocate with parent allocator */
         if (4*bytes > allocBlockSize) {
           return alloc->malloc(bytes,maxAlignment,false);
-	}
+        }
 
         /* get new partial block if allocation failed */
         size_t blockSize = allocBlockSize;
         ptr = (char*) alloc->malloc(blockSize,maxAlignment,true);
- 	bytesWasted += end-cur;
-	cur = 0; end = blockSize;
+        bytesWasted += end-cur;
+        cur = 0; end = blockSize;
 
         /* retry allocation */
-	ofs = (align - cur) & (align-1);
+        ofs = (align - cur) & (align-1);
         cur += bytes + ofs;
         if (likely(cur <= end)) { bytesWasted += ofs; return &ptr[cur - bytes]; }
-	cur -= bytes + ofs;
+        cur -= bytes + ofs;
 
         /* get new full block if allocation failed */
         blockSize = allocBlockSize;
         ptr = (char*) alloc->malloc(blockSize,maxAlignment,false);
-	bytesWasted += end-cur;
-	cur = 0; end = blockSize;
+        bytesWasted += end-cur;
+        cur = 0; end = blockSize;
 
         /* retry allocation */
-	ofs = (align - cur) & (align-1);
+        ofs = (align - cur) & (align-1);
         cur += bytes + ofs;
         if (likely(cur <= end)) { bytesWasted += ofs; return &ptr[cur - bytes]; }
-	cur -= bytes + ofs;
+        cur -= bytes + ofs;
 
         /* should never happen as large allocations get handled specially above */
         assert(false);
         return nullptr;
       }
 
-      
-      /*! returns amount of used bytes */
       __forceinline size_t getUsedBytes() const { return bytesUsed; }
-  
+
       /*! returns amount of free bytes */
       __forceinline size_t getFreeBytes() const { return end-cur; }
-      
+
       /*! returns amount of wasted bytes */
       __forceinline size_t getWastedBytes() const { return bytesWasted; }
-  
+
     private:
       ThreadLocal2* parent;
       char*  ptr;            //!< pointer to memory block
@@ -136,11 +132,7 @@ namespace embree
       {
         assert(alloc_i);
         if (alloc.load() == alloc_i) return;
-#if defined(APPLE) && defined(__aarch64__)
-        std::scoped_lock lock(mutex);
-#else
-        Lock<SpinLock> lock(mutex);
-#endif
+        Lock<MutexSys> lock(mutex);
         //if (alloc.load() == alloc_i) return; // not required as only one thread calls bind
         if (alloc.load()) {
           alloc.load()->bytesUsed   += alloc0.getUsedBytes()   + alloc1.getUsedBytes();
@@ -158,11 +150,7 @@ namespace embree
       {
         assert(alloc_i);
         if (alloc.load() != alloc_i) return;
-#if defined(APPLE) && defined(__aarch64__)
-        std::scoped_lock lock(mutex);
-#else
-        Lock<SpinLock> lock(mutex);
-#endif
+        Lock<MutexSys> lock(mutex);
         if (alloc.load() != alloc_i) return; // required as a different thread calls unbind
         alloc.load()->bytesUsed   += alloc0.getUsedBytes()   + alloc1.getUsedBytes();
         alloc.load()->bytesFree   += alloc0.getFreeBytes()   + alloc1.getFreeBytes();
@@ -173,26 +161,47 @@ namespace embree
       }
 
     public:
-#if defined(APPLE) && defined(__aarch64__)
-      std::mutex mutex;
-#else
-      SpinLock mutex;        //!< required as unbind is called from other threads
-#endif
+      MutexSys mutex;
       std::atomic<FastAllocator*> alloc;  //!< parent allocator
       ThreadLocal alloc0;
       ThreadLocal alloc1;
     };
 
-    FastAllocator (Device* device, bool osAllocation) 
-      : device(device), slotMask(0), usedBlocks(nullptr), freeBlocks(nullptr), use_single_mode(false), defaultBlockSize(PAGE_SIZE), estimatedSize(0),
-        growSize(PAGE_SIZE), maxGrowSize(maxAllocationSize), log2_grow_size_scale(0), bytesUsed(0), bytesFree(0), bytesWasted(0), atype(osAllocation ? EMBREE_OS_MALLOC : ALIGNED_MALLOC),
-        primrefarray(device,0)
+    FastAllocator (Device* device,
+                   bool osAllocation,
+                   bool useUSM = false,
+                   bool blockAllocation = true)
+      : device(device)
+      , slotMask(0)
+      , defaultBlockSize(PAGE_SIZE)
+      , estimatedSize(0)
+      , growSize(PAGE_SIZE)
+      , maxGrowSize(maxAllocationSize)
+      , usedBlocks(nullptr)
+      , freeBlocks(nullptr)
+      , useUSM(useUSM)
+      , blockAllocation(blockAllocation)
+      , use_single_mode(false)
+      , log2_grow_size_scale(0)
+      , bytesUsed(0)
+      , bytesFree(0)
+      , bytesWasted(0)
+      , atype(osAllocation ? EMBREE_OS_MALLOC : ALIGNED_MALLOC)
+      , primrefarray(device,0)
     {
+      // -- GODOT start --
+      // if (osAllocation && useUSM)
+      //   throw std::runtime_error("USM allocation cannot be combined with OS allocation.");
+      if (osAllocation && useUSM) {
+        abort();
+      }
+      // -- GODOT end --
+
       for (size_t i=0; i<MAX_THREAD_USED_BLOCK_SLOTS; i++)
       {
         threadUsedBlocks[i] = nullptr;
         threadBlocks[i] = nullptr;
-        assert(!slotMutex[i].isLocked());
+        //assert(!slotMutex[i].isLocked());
       }
     }
 
@@ -233,11 +242,7 @@ namespace embree
       ThreadLocal2* alloc = thread_local_allocator2;
       if (alloc == nullptr) {
         thread_local_allocator2 = alloc = new ThreadLocal2;
-#if defined(APPLE) && defined(__aarch64__)
-        std::scoped_lock lock(s_thread_local_allocators_lock);
-#else
-        Lock<SpinLock> lock(s_thread_local_allocators_lock);
-#endif
+        Lock<MutexSys> lock(s_thread_local_allocators_lock);
         s_thread_local_allocators.push_back(make_unique(alloc));
       }
       return alloc;
@@ -247,11 +252,7 @@ namespace embree
 
     __forceinline void join(ThreadLocal2* alloc)
     {
-#if defined(APPLE) && defined(__aarch64__)
-      std::scoped_lock lock(s_thread_local_allocators_lock);
-#else
-      Lock<SpinLock> lock(thread_local_allocators_lock);
-#endif
+      Lock<MutexSys> lock(s_thread_local_allocators_lock);
       thread_local_allocators.push_back(alloc);
     }
 
@@ -412,7 +413,7 @@ namespace embree
       slotMask = MAX_THREAD_USED_BLOCK_SLOTS-1; // FIXME: remove
       if (usedBlocks.load() || freeBlocks.load()) { reset(); return; }
       if (bytesReserve == 0) bytesReserve = bytesAllocate;
-      freeBlocks = Block::create(device,bytesAllocate,bytesReserve,nullptr,atype);
+      freeBlocks = Block::create(device,useUSM,bytesAllocate,bytesReserve,nullptr,atype);
       estimatedSize = bytesEstimate;
       initGrowSizeAndNumSlots(bytesEstimate,true);
     }
@@ -478,8 +479,8 @@ namespace embree
       bytesUsed.store(0);
       bytesFree.store(0);
       bytesWasted.store(0);
-      if (usedBlocks.load() != nullptr) usedBlocks.load()->clear_list(device); usedBlocks = nullptr;
-      if (freeBlocks.load() != nullptr) freeBlocks.load()->clear_list(device); freeBlocks = nullptr;
+      if (usedBlocks.load() != nullptr) usedBlocks.load()->clear_list(device,useUSM); usedBlocks = nullptr;
+      if (freeBlocks.load() != nullptr) freeBlocks.load()->clear_list(device,useUSM); freeBlocks = nullptr;
       for (size_t i=0; i<MAX_THREAD_USED_BLOCK_SLOTS; i++) {
         threadUsedBlocks[i] = nullptr;
         threadBlocks[i] = nullptr;
@@ -503,9 +504,16 @@ namespace embree
         /* allocate using current block */
         size_t threadID = TaskScheduler::threadID();
         size_t slot = threadID & slotMask;
-	Block* myUsedBlocks = threadUsedBlocks[slot];
+        Block* myUsedBlocks = threadUsedBlocks[slot];
         if (myUsedBlocks) {
           void* ptr = myUsedBlocks->malloc(device,bytes,align,partial);
+          // -- GODOT start --
+          // if (ptr == nullptr && !blockAllocation)
+          //   throw std::bad_alloc();
+          if (ptr == nullptr && !blockAllocation) {
+            abort();
+          }
+          // -- GODOT end --
           if (ptr) return ptr;
         }
 
@@ -516,16 +524,12 @@ namespace embree
         /* parallel block creation in case of no freeBlocks, avoids single global mutex */
         if (likely(freeBlocks.load() == nullptr))
         {
-#if defined(APPLE) && defined(__aarch64__)
-          std::scoped_lock lock(slotMutex[slot]);
-#else
-          Lock<SpinLock> lock(slotMutex[slot]);
-#endif
+          Lock<MutexSys> lock(slotMutex[slot]);
           if (myUsedBlocks == threadUsedBlocks[slot]) {
             const size_t alignedBytes = (bytes+(align-1)) & ~(align-1);
             const size_t allocSize = max(min(growSize,maxGrowSize),alignedBytes);
             assert(allocSize >= bytes);
-            threadBlocks[slot] = threadUsedBlocks[slot] = Block::create(device,allocSize,allocSize,threadBlocks[slot],atype); // FIXME: a large allocation might throw away a block here!
+            threadBlocks[slot] = threadUsedBlocks[slot] = Block::create(device,useUSM,allocSize,allocSize,threadBlocks[slot],atype); // FIXME: a large allocation might throw away a block here!
             // FIXME: a direct allocation should allocate inside the block here, and not in the next loop! a different thread could do some allocation and make the large allocation fail.
           }
           continue;
@@ -533,24 +537,20 @@ namespace embree
 
         /* if this fails allocate new block */
         {
-#if defined(APPLE) && defined(__aarch64__)
-            std::scoped_lock lock(mutex);
-#else
-            Lock<SpinLock> lock(mutex);
-#endif
-	  if (myUsedBlocks == threadUsedBlocks[slot])
-	  {
+          Lock<MutexSys> lock(mutex);
+          if (myUsedBlocks == threadUsedBlocks[slot])
+          {
             if (freeBlocks.load() != nullptr) {
-	      Block* nextFreeBlock = freeBlocks.load()->next;
-	      freeBlocks.load()->next = usedBlocks;
-	      __memory_barrier();
-	      usedBlocks = freeBlocks.load();
+              Block* nextFreeBlock = freeBlocks.load()->next;
+              freeBlocks.load()->next = usedBlocks;
+              __memory_barrier();
+              usedBlocks = freeBlocks.load();
               threadUsedBlocks[slot] = freeBlocks.load();
-	      freeBlocks = nextFreeBlock;
-	    } else {
+              freeBlocks = nextFreeBlock;
+            } else {
               const size_t allocSize = min(growSize*incGrowSizeScale(),maxGrowSize);
-	      usedBlocks = threadUsedBlocks[slot] = Block::create(device,allocSize,allocSize,usedBlocks,atype); // FIXME: a large allocation should get delivered directly, like above!
-	    }
+              usedBlocks = threadUsedBlocks[slot] = Block::create(device,useUSM,allocSize,allocSize,usedBlocks,atype); // FIXME: a large allocation should get delivered directly, like above!
+            }
           }
         }
       }
@@ -559,11 +559,7 @@ namespace embree
     /*! add new block */
     void addBlock(void* ptr, ssize_t bytes)
     {
-#if defined(APPLE) && defined(__aarch64__)
-      std::scoped_lock lock(mutex);
-#else
-      Lock<SpinLock> lock(mutex);
-#endif
+      Lock<MutexSys> lock(mutex);
       const size_t sizeof_Header = offsetof(Block,data[0]);
       void* aptr = (void*) ((((size_t)ptr)+maxAlignment-1) & ~(maxAlignment-1));
       size_t ofs = (size_t) aptr - (size_t) ptr;
@@ -723,7 +719,12 @@ namespace embree
 
     void print_blocks()
     {
-      std::cout << "  estimatedSize = " << estimatedSize << ", slotMask = " << slotMask << ", use_single_mode = " << use_single_mode << ", maxGrowSize = " << maxGrowSize << ", defaultBlockSize = " << defaultBlockSize << std::endl;
+      std::cout << "  estimatedSize = " << estimatedSize
+                << ", slotMask = " << slotMask
+                << ", use_single_mode = " << use_single_mode
+                << ", maxGrowSize = " << maxGrowSize
+                << ", defaultBlockSize = " << defaultBlockSize
+                << std::endl;
 
       std::cout << "  used blocks = ";
       if (usedBlocks.load() != nullptr) usedBlocks.load()->print_list();
@@ -738,7 +739,19 @@ namespace embree
 
     struct Block
     {
-      static Block* create(MemoryMonitorInterface* device, size_t bytesAllocate, size_t bytesReserve, Block* next, AllocationType atype)
+      __forceinline static void* blockAlignedMalloc(Device* device, bool useUSM, size_t bytesAllocate, size_t bytesAlignment)
+      {
+        if (useUSM) return device->malloc(bytesAllocate, bytesAlignment);
+	else        return alignedMalloc (bytesAllocate, bytesAlignment);
+      }
+
+      __forceinline static void blockAlignedFree(Device* device, bool useUSM, void* ptr)
+      {
+        if (useUSM) return device->free(ptr);
+	else        return alignedFree(ptr);
+      }
+
+      static Block* create(Device* device, bool useUSM, size_t bytesAllocate, size_t bytesReserve, Block* next, AllocationType atype)
       {
         /* We avoid using os_malloc for small blocks as this could
          * cause a risk of fragmenting the virtual address space and
@@ -766,7 +779,7 @@ namespace embree
           {
             const size_t alignment = maxAlignment;
             if (device) device->memoryMonitor(bytesAllocate+alignment,false);
-            ptr = alignedMalloc(bytesAllocate,alignment);
+            ptr = blockAlignedMalloc(device,useUSM,bytesAllocate,alignment);
 
             /* give hint to transparently convert these pages to 2MB pages */
             const size_t ptr_aligned_begin = ((size_t)ptr) & ~size_t(PAGE_SIZE_2M-1);
@@ -780,7 +793,7 @@ namespace embree
           {
             const size_t alignment = maxAlignment;
             if (device) device->memoryMonitor(bytesAllocate+alignment,false);
-            ptr = alignedMalloc(bytesAllocate,alignment);
+            ptr = blockAlignedMalloc(device,useUSM,bytesAllocate,alignment);
             return new (ptr) Block(ALIGNED_MALLOC,bytesAllocate-sizeof_Header,bytesAllocate-sizeof_Header,next,alignment);
           }
         }
@@ -812,23 +825,23 @@ namespace embree
         return head;
       }
 
-      void clear_list(MemoryMonitorInterface* device)
+      void clear_list(Device* device, bool useUSM)
       {
         Block* block = this;
         while (block) {
           Block* next = block->next;
-          block->clear_block(device);
+          block->clear_block(device, useUSM);
           block = next;
         }
       }
 
-      void clear_block (MemoryMonitorInterface* device)
+      void clear_block (Device* device, bool useUSM)
       {
         const size_t sizeof_Header = offsetof(Block,data[0]);
         const ssize_t sizeof_Alloced = wasted+sizeof_Header+getBlockAllocatedBytes();
 
         if (atype == ALIGNED_MALLOC) {
-          alignedFree(this);
+          blockAlignedFree(device, useUSM, this);
           if (device) device->memoryMonitor(-sizeof_Alloced,true);
         }
 
@@ -847,16 +860,16 @@ namespace embree
         size_t bytes = bytes_in;
         assert(align <= maxAlignment);
         bytes = (bytes+(align-1)) & ~(align-1);
-	if (unlikely(cur+bytes > reserveEnd && !partial)) return nullptr;
-	const size_t i = cur.fetch_add(bytes);
+        if (unlikely(cur+bytes > reserveEnd && !partial)) return nullptr;
+        const size_t i = cur.fetch_add(bytes);
         if (unlikely(i+bytes > reserveEnd && !partial)) return nullptr;
         if (unlikely(i > reserveEnd)) return nullptr;
         bytes_in = bytes = min(bytes,reserveEnd-i);
-        
-	if (i+bytes > allocEnd) {
+
+        if (i+bytes > allocEnd) {
           if (device) device->memoryMonitor(i+bytes-max(i,allocEnd),true);
         }
-	return &data[i];
+        return &data[i];
       }
 
       void* ptr() {
@@ -874,7 +887,7 @@ namespace embree
       }
 
       size_t getBlockFreeBytes() const {
-	return getBlockAllocatedBytes() - getBlockUsedBytes();
+        return getBlockAllocatedBytes() - getBlockUsedBytes();
       }
 
       size_t getBlockAllocatedBytes() const {
@@ -963,40 +976,40 @@ namespace embree
       char data[1];              //!< here starts memory to use for allocations
     };
 
+  public:
+    static const size_t blockHeaderSize = offsetof(Block,data[0]);
+
   private:
     Device* device;
-    SpinLock mutex;
     size_t slotMask;
+    size_t defaultBlockSize;
+    size_t estimatedSize;
+    size_t growSize;
+    size_t maxGrowSize;
+
+    MutexSys mutex;
+    MutexSys slotMutex[MAX_THREAD_USED_BLOCK_SLOTS];
     std::atomic<Block*> threadUsedBlocks[MAX_THREAD_USED_BLOCK_SLOTS];
+    std::atomic<Block*> threadBlocks[MAX_THREAD_USED_BLOCK_SLOTS];
     std::atomic<Block*> usedBlocks;
     std::atomic<Block*> freeBlocks;
 
-    std::atomic<Block*> threadBlocks[MAX_THREAD_USED_BLOCK_SLOTS];
-#if defined(APPLE) && defined(__aarch64__)
-    std::mutex slotMutex[MAX_THREAD_USED_BLOCK_SLOTS];
-#else
-    PaddedSpinLock slotMutex[MAX_THREAD_USED_BLOCK_SLOTS];
-#endif
-
+    bool useUSM;
+    bool blockAllocation = true;
     bool use_single_mode;
-    size_t defaultBlockSize;
-    size_t estimatedSize;
-    size_t growSize;
-    size_t maxGrowSize;
+
     std::atomic<size_t> log2_grow_size_scale; //!< log2 of scaling factor for grow size // FIXME: remove
     std::atomic<size_t> bytesUsed;
     std::atomic<size_t> bytesFree;
     std::atomic<size_t> bytesWasted;
+
     static __thread ThreadLocal2* thread_local_allocator2;
-    static SpinLock s_thread_local_allocators_lock;
+    static MutexSys s_thread_local_allocators_lock;
     static std::vector<std::unique_ptr<ThreadLocal2>> s_thread_local_allocators;
-#if defined(APPLE) && defined(__aarch64__)
-    std::mutex thread_local_allocators_lock;
-#else
-    SpinLock thread_local_allocators_lock;
-#endif
+
     std::vector<ThreadLocal2*> thread_local_allocators;
     AllocationType atype;
+
     mvector<PrimRef> primrefarray;     //!< primrefarray used to allocate nodes
   };
 }
diff --git a/thirdparty/embree/kernels/common/buffer.h b/thirdparty/embree/kernels/common/buffer.h
index 793012c04d..831f5815e8 100644
--- a/thirdparty/embree/kernels/common/buffer.h
+++ b/thirdparty/embree/kernels/common/buffer.h
@@ -13,8 +13,8 @@ namespace embree
   {
   public:
     /*! Buffer construction */
-    Buffer() 
-      : device(nullptr), ptr(nullptr), numBytes(0), shared(false) {}
+    //Buffer() 
+    //: device(nullptr), ptr(nullptr), numBytes(0), shared(false) {}
 
     /*! Buffer construction */
     Buffer(Device* device, size_t numBytes_in, void* ptr_in = nullptr)
@@ -77,19 +77,17 @@ namespace embree
     /*! allocated buffer */
     void alloc()
     {
-      if (device)
-        device->memoryMonitor(this->bytes(), false);
+      device->memoryMonitor(this->bytes(), false);
       size_t b = (this->bytes()+15) & ssize_t(-16);
-      ptr = (char*)alignedMalloc(b,16);
+      ptr = (char*)device->malloc(b,16);
     }
     
     /*! frees the buffer */
     void free()
     {
       if (shared) return;
-      alignedFree(ptr); 
-      if (device)
-        device->memoryMonitor(-ssize_t(this->bytes()), true);
+      device->free(ptr); 
+      device->memoryMonitor(-ssize_t(this->bytes()), true);
       ptr = nullptr;
     }
     
@@ -246,6 +244,24 @@ namespace embree
   public:
     typedef Vec3fa value_type;
 
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+
+     /*! access to the ith element of the buffer */
+    __forceinline const Vec3fa operator [](size_t i) const
+    {
+      assert(i<num);
+      return Vec3fa::loadu(ptr_ofs + i*stride);
+    }
+    
+    /*! writes the i'th element */
+    __forceinline void store(size_t i, const Vec3fa& v)
+    {
+      assert(i<num);
+      Vec3fa::storeu(ptr_ofs + i*stride, v);
+    }
+    
+#else
+
     /*! access to the ith element of the buffer */
     __forceinline const Vec3fa operator [](size_t i) const
     {
@@ -259,5 +275,6 @@ namespace embree
       assert(i<num);
       vfloat4::storeu((float*)(ptr_ofs + i*stride), (vfloat4)v);
     }
+#endif
   };
 }
diff --git a/thirdparty/embree/kernels/common/builder.h b/thirdparty/embree/kernels/common/builder.h
index 07fe7b069b..4f6a226810 100644
--- a/thirdparty/embree/kernels/common/builder.h
+++ b/thirdparty/embree/kernels/common/builder.h
@@ -7,7 +7,7 @@
 #include "accel.h"
 
 namespace embree
-{
+{  
 #define MODE_HIGH_QUALITY (1<<8)
 
   /*! virtual interface for all hierarchy builders */
diff --git a/thirdparty/embree/kernels/common/context.h b/thirdparty/embree/kernels/common/context.h
index ccd88bdeac..936d03e54d 100644
--- a/thirdparty/embree/kernels/common/context.h
+++ b/thirdparty/embree/kernels/common/context.h
@@ -11,35 +11,62 @@ namespace embree
 {
   class Scene;
 
-  struct IntersectContext
+  struct RayQueryContext
   {
   public:
-    __forceinline IntersectContext(Scene* scene, RTCIntersectContext* user_context)
-      : scene(scene), user(user_context) {}
+
+    __forceinline RayQueryContext(Scene* scene, RTCRayQueryContext* user_context, RTCIntersectArguments* args)
+      : scene(scene), user(user_context), args(args) {}
+
+    __forceinline RayQueryContext(Scene* scene, RTCRayQueryContext* user_context, RTCOccludedArguments* args)
+      : scene(scene), user(user_context), args((RTCIntersectArguments*)args) {}
 
     __forceinline bool hasContextFilter() const {
-      return user->filter != nullptr;
+      return args->filter != nullptr;
+    }
+
+    RTCFilterFunctionN getFilter() const {
+      return args->filter;
+    }
+
+    RTCIntersectFunctionN getIntersectFunction() const {
+      return args->intersect;
+    }
+    
+    RTCOccludedFunctionN getOccludedFunction() const {
+      return (RTCOccludedFunctionN) args->intersect;
     }
 
     __forceinline bool isCoherent() const {
-      return embree::isCoherent(user->flags);
+      return embree::isCoherent(args->flags);
     }
 
     __forceinline bool isIncoherent() const {
-      return embree::isIncoherent(user->flags);
+      return embree::isIncoherent(args->flags);
     }
-    
+
+    __forceinline bool enforceArgumentFilterFunction() const {
+      return args->flags & RTC_RAY_QUERY_FLAG_INVOKE_ARGUMENT_FILTER;
+    }
+
+#if RTC_MIN_WIDTH
+    __forceinline float getMinWidthDistanceFactor() const {
+      return args->minWidthDistanceFactor;
+    }
+#endif
+
   public:
-    Scene* scene;
-    RTCIntersectContext* user;
+    Scene* scene = nullptr;
+    RTCRayQueryContext* user = nullptr;
+    RTCIntersectArguments* args = nullptr;
   };
 
   template<int M, typename Geometry>
-      __forceinline Vec4vf<M> enlargeRadiusToMinWidth(const IntersectContext* context, const Geometry* geom, const Vec3vf<M>& ray_org, const Vec4vf<M>& v)
+      __forceinline Vec4vf<M> enlargeRadiusToMinWidth(const RayQueryContext* context, const Geometry* geom, const Vec3vf<M>& ray_org, const Vec4vf<M>& v)
     {
 #if RTC_MIN_WIDTH
       const vfloat<M> d = length(Vec3vf<M>(v) - ray_org);
-      const vfloat<M> r = clamp(context->user->minWidthDistanceFactor*d, v.w, geom->maxRadiusScale*v.w);
+      const vfloat<M> r = clamp(context->getMinWidthDistanceFactor()*d, v.w, geom->maxRadiusScale*v.w);
       return Vec4vf<M>(v.x,v.y,v.z,r);
 #else
       return v;
@@ -47,16 +74,21 @@ namespace embree
     }
 
     template<typename Geometry>
-    __forceinline Vec3ff enlargeRadiusToMinWidth(const IntersectContext* context, const Geometry* geom, const Vec3fa& ray_org, const Vec3ff& v)
+    __forceinline Vec3ff enlargeRadiusToMinWidth(const RayQueryContext* context, const Geometry* geom, const Vec3fa& ray_org, const Vec3ff& v)
   {
 #if RTC_MIN_WIDTH
     const float d = length(Vec3fa(v) - ray_org);
-    const float r = clamp(context->user->minWidthDistanceFactor*d, v.w, geom->maxRadiusScale*v.w);
+    const float r = clamp(context->getMinWidthDistanceFactor()*d, v.w, geom->maxRadiusScale*v.w);
     return Vec3ff(v.x,v.y,v.z,r);
 #else
     return v;
 #endif
   }
+
+  template<typename Geometry>
+    __forceinline Vec3ff enlargeRadiusToMinWidth(const RayQueryContext* context, const Geometry* geom, const Vec3fa& ray_org, const Vec4f& v) {
+    return enlargeRadiusToMinWidth(context,geom,ray_org,Vec3ff(v.x,v.y,v.z,v.w));
+  }
   
   enum PointQueryType
   {
@@ -66,7 +98,7 @@ namespace embree
   };
 
   typedef bool (*PointQueryFunction)(struct RTCPointQueryFunctionArguments* args);
-  
+
   struct PointQueryContext
   {
   public:
@@ -78,6 +110,7 @@ namespace embree
                                     float similarityScale,
                                     void* userPtr)
       : scene(scene)
+      , tstate(nullptr)
       , query_ws(query_ws)
       , query_type(query_type)
       , func(func)
@@ -88,16 +121,24 @@ namespace embree
       , geomID(RTC_INVALID_GEOMETRY_ID)
       , query_radius(query_ws->radius)
     { 
+      update();
+    }
+
+  public:
+    __forceinline void update()
+    {
       if (query_type == POINT_QUERY_TYPE_AABB) {
         assert(similarityScale == 0.f);
         updateAABB();
       }
+      else{
+        query_radius = Vec3fa(query_ws->radius * similarityScale);
+      }
       if (userContext->instStackSize == 0) {
         assert(similarityScale == 1.f);
       }
     }
 
-  public:
     __forceinline void updateAABB() 
     {
       if (likely(query_ws->radius == (float)inf || userContext->instStackSize == 0)) {
@@ -113,12 +154,13 @@ namespace embree
 
 public:
     Scene* scene;
+    void* tstate;
 
     PointQuery* query_ws; // the original world space point query 
     PointQueryType query_type;
     PointQueryFunction func;
     RTCPointQueryContext* userContext;
-    const float similarityScale;
+    float similarityScale;
 
     void* userPtr;
 
diff --git a/thirdparty/embree/kernels/common/default.h b/thirdparty/embree/kernels/common/default.h
index f15d61b768..3b00ad3c88 100644
--- a/thirdparty/embree/kernels/common/default.h
+++ b/thirdparty/embree/kernels/common/default.h
@@ -13,11 +13,11 @@
 #include "../../common/sys/mutex.h"
 #include "../../common/sys/vector.h"
 #include "../../common/sys/array.h"
-#include "../../common/sys/string.h"
+#include "../../common/sys/estring.h"
 #include "../../common/sys/regression.h"
 #include "../../common/sys/vector.h"
 
-#include "../../common/math/math.h"
+#include "../../common/math/emath.h"
 #include "../../common/math/transcendental.h"
 #include "../../common/simd/simd.h"
 #include "../../common/math/vec2.h"
@@ -35,8 +35,6 @@
 #include "../../common/math/range.h"
 #include "../../common/lexers/tokenstream.h"
 
-#include "../../common/tasking/taskscheduler.h"
-
 #define COMMA ,
 
 #include "../config.h"
@@ -217,7 +215,7 @@ namespace embree
   __forceinline int getTimeSegment(float time, float numTimeSegments, float& ftime)
   {
     const float timeScaled = time * numTimeSegments;
-    const float itimef = clamp(floorf(timeScaled), 0.0f, numTimeSegments-1.0f);
+    const float itimef = clamp(floor(timeScaled), 0.0f, numTimeSegments-1.0f);
     ftime = timeScaled - itimef;
     return int(itimef);
   }
@@ -225,7 +223,7 @@ namespace embree
   __forceinline int getTimeSegment(float time, float start_time, float end_time, float numTimeSegments, float& ftime)
   {
     const float timeScaled = (time-start_time)/(end_time-start_time) * numTimeSegments;
-    const float itimef = clamp(floorf(timeScaled), 0.0f, numTimeSegments-1.0f);
+    const float itimef = clamp(floor(timeScaled), 0.0f, numTimeSegments-1.0f);
     ftime = timeScaled - itimef;
     return int(itimef);
   }
diff --git a/thirdparty/embree/kernels/common/device.cpp b/thirdparty/embree/kernels/common/device.cpp
index 833ec65139..07214532a1 100644
--- a/thirdparty/embree/kernels/common/device.cpp
+++ b/thirdparty/embree/kernels/common/device.cpp
@@ -2,6 +2,9 @@
 // SPDX-License-Identifier: Apache-2.0
 
 #include "device.h"
+
+#include "../../common/tasking/taskscheduler.h"
+
 #include "../hash.h"
 #include "scene_triangle_mesh.h"
 #include "scene_user_geometry.h"
@@ -19,9 +22,12 @@
 #include "../bvh/bvh4_factory.h"
 #include "../bvh/bvh8_factory.h"
 
-#include "../../common/tasking/taskscheduler.h"
 #include "../../common/sys/alloc.h"
 
+#if defined(EMBREE_SYCL_SUPPORT)
+#  include "../level_zero/ze_wrapper.h"
+#endif
+
 namespace embree
 {
   /*! some global variables that can be set via rtcSetParameter1i for debugging purposes */
@@ -30,13 +36,18 @@ namespace embree
   ssize_t Device::debug_int2 = 0;
   ssize_t Device::debug_int3 = 0;
 
-  DECLARE_SYMBOL2(RayStreamFilterFuncs,rayStreamFilterFuncs);
-
   static MutexSys g_mutex;
   static std::map<Device*,size_t> g_cache_size_map;
   static std::map<Device*,size_t> g_num_threads_map;
+  
+  struct TaskArena
+  {
+#if USE_TASK_ARENA
+    std::unique_ptr<tbb::task_arena> arena;
+#endif
+  };
 
-  Device::Device (const char* cfg)
+  Device::Device (const char* cfg) : arena(new TaskArena())
   {
     /* check that CPU supports lowest ISA */
     if (!hasISA(ISA)) {
@@ -48,12 +59,12 @@ namespace embree
     case CPU::UNKNOWN:         frequency_level = FREQUENCY_SIMD256; break;
     case CPU::XEON_ICE_LAKE:   frequency_level = FREQUENCY_SIMD256; break;
     case CPU::CORE_ICE_LAKE:   frequency_level = FREQUENCY_SIMD256; break;
-    case CPU::CORE_TIGER_LAKE: frequency_level = FREQUENCY_SIMD128; break;
-    case CPU::CORE_COMET_LAKE: frequency_level = FREQUENCY_SIMD128; break;
-    case CPU::CORE_CANNON_LAKE:frequency_level = FREQUENCY_SIMD128; break;
-    case CPU::CORE_KABY_LAKE:  frequency_level = FREQUENCY_SIMD128; break;
+    case CPU::CORE_TIGER_LAKE: frequency_level = FREQUENCY_SIMD256; break;
+    case CPU::CORE_COMET_LAKE: frequency_level = FREQUENCY_SIMD256; break;
+    case CPU::CORE_CANNON_LAKE:frequency_level = FREQUENCY_SIMD256; break;
+    case CPU::CORE_KABY_LAKE:  frequency_level = FREQUENCY_SIMD256; break;
     case CPU::XEON_SKY_LAKE:   frequency_level = FREQUENCY_SIMD128; break;
-    case CPU::CORE_SKY_LAKE:   frequency_level = FREQUENCY_SIMD128; break;
+    case CPU::CORE_SKY_LAKE:   frequency_level = FREQUENCY_SIMD256; break;
     case CPU::XEON_BROADWELL:  frequency_level = FREQUENCY_SIMD256; break;
     case CPU::CORE_BROADWELL:  frequency_level = FREQUENCY_SIMD256; break;
     case CPU::XEON_HASWELL:    frequency_level = FREQUENCY_SIMD256; break;
@@ -66,11 +77,7 @@ namespace embree
     case CPU::CORE1:           frequency_level = FREQUENCY_SIMD128; break;
     case CPU::XEON_PHI_KNIGHTS_MILL   : frequency_level = FREQUENCY_SIMD512; break;
     case CPU::XEON_PHI_KNIGHTS_LANDING: frequency_level = FREQUENCY_SIMD512; break;
-#if defined(__APPLE__)
-    case CPU::ARM:             frequency_level = FREQUENCY_SIMD256; break; // Apple M1 supports high throughput for SIMD4
-#else
-    case CPU::ARM:             frequency_level = FREQUENCY_SIMD128; break;
-#endif
+    case CPU::ARM:             frequency_level = FREQUENCY_SIMD256; break;
     }
 
     /* initialize global state */
@@ -126,13 +133,6 @@ namespace embree
 
     /* setup tasking system */
     initTaskingSystem(numThreads);
-
-    /* ray stream SOA to AOS conversion */
-#if defined(EMBREE_RAY_PACKETS)
-    RayStreamFilterFuncsType rayStreamFilterFuncs;
-    SELECT_SYMBOL_DEFAULT_SSE42_AVX_AVX2_AVX512(enabled_cpu_features,rayStreamFilterFuncs);
-    rayStreamFilters = rayStreamFilterFuncs();
-#endif
   }
 
   Device::~Device ()
@@ -174,6 +174,9 @@ namespace embree
 #if defined (EMBREE_BACKFACE_CULLING_CURVES)
     v += "backfacecullingcurves ";
 #endif
+#if defined (EMBREE_BACKFACE_CULLING_SPHERES)
+    v += "backfacecullingspheres ";
+#endif
 #if defined(EMBREE_FILTER_FUNCTION)
     v += "intersection_filter ";
 #endif
@@ -367,7 +370,7 @@ namespace embree
 #if USE_TASK_ARENA
     const size_t nThreads = min(maxNumThreads,TaskScheduler::threadCount());
     const size_t uThreads = min(max(numUserThreads,(size_t)1),nThreads);
-    arena = make_unique(new tbb::task_arena((int)nThreads,(unsigned int)uThreads));
+    arena->arena = make_unique(new tbb::task_arena((int)nThreads,(unsigned int)uThreads));
 #endif
   }
 
@@ -386,8 +389,21 @@ namespace embree
       TaskScheduler::create(maxNumThreads,State::set_affinity,State::start_threads);
     }
 #if USE_TASK_ARENA
-    arena.reset();
+    arena->arena.reset();
+#endif
+  }
+
+  void Device::execute(bool join, const std::function<void()>& func)
+  {
+#if USE_TASK_ARENA
+    if (join) {
+      arena->arena->execute(func);
+    }
+    else
 #endif
+    {
+      func();
+    }
   }
 
   void Device::setProperty(const RTCDeviceProperty prop, ssize_t val)
@@ -450,12 +466,6 @@ namespace embree
     case RTC_DEVICE_PROPERTY_NATIVE_RAY16_SUPPORTED: return 0;
 #endif
 
-#if defined(EMBREE_RAY_PACKETS)
-    case RTC_DEVICE_PROPERTY_RAY_STREAM_SUPPORTED:  return 1;
-#else
-    case RTC_DEVICE_PROPERTY_RAY_STREAM_SUPPORTED:  return 0;
-#endif
-    
 #if defined(EMBREE_RAY_MASK)
     case RTC_DEVICE_PROPERTY_RAY_MASK_SUPPORTED: return 1;
 #else
@@ -474,6 +484,12 @@ namespace embree
     case RTC_DEVICE_PROPERTY_BACKFACE_CULLING_CURVES_ENABLED: return 0;
 #endif
 
+#if defined(EMBREE_BACKFACE_CULLING_SPHERES)
+    case RTC_DEVICE_PROPERTY_BACKFACE_CULLING_SPHERES_ENABLED: return 1;
+#else
+    case RTC_DEVICE_PROPERTY_BACKFACE_CULLING_SPHERES_ENABLED: return 0;
+#endif
+
 #if defined(EMBREE_COMPACT_POLYS)
     case RTC_DEVICE_PROPERTY_COMPACT_POLYS_ENABLED: return 1;
 #else
@@ -557,4 +573,158 @@ namespace embree
     default: throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "unknown readable property"); break;
     };
   }
+
+  void* Device::malloc(size_t size, size_t align) {
+    return alignedMalloc(size,align);
+  }
+
+  void Device::free(void* ptr) {
+    alignedFree(ptr);
+  }
+
+
+#if defined(EMBREE_SYCL_SUPPORT)
+
+  DeviceGPU::DeviceGPU(sycl::context sycl_context, const char* cfg)
+    : Device(cfg), gpu_context(sycl_context)
+  {
+    /* initialize ZeWrapper */
+    if (ZeWrapper::init() != ZE_RESULT_SUCCESS)
+       throw_RTCError(RTC_ERROR_UNKNOWN, "cannot initialize ZeWrapper");
+     
+    /* take first device as default device */
+    auto devices = gpu_context.get_devices();
+    if (devices.size() == 0)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "SYCL context contains no device");
+    gpu_device = devices[0];
+
+    /* check if RTAS build extension is available */
+    sycl::platform platform = gpu_device.get_platform();
+    ze_driver_handle_t hDriver = sycl::get_native<sycl::backend::ext_oneapi_level_zero>(platform);
+    
+    uint32_t count = 0;
+    std::vector<ze_driver_extension_properties_t> extensions;
+    ze_result_t result = ZeWrapper::zeDriverGetExtensionProperties(hDriver,&count,extensions.data());
+    if (result != ZE_RESULT_SUCCESS)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "zeDriverGetExtensionProperties failed");
+    
+    extensions.resize(count);
+    result = ZeWrapper::zeDriverGetExtensionProperties(hDriver,&count,extensions.data());
+    if (result != ZE_RESULT_SUCCESS)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "zeDriverGetExtensionProperties failed");
+
+#if defined(EMBREE_SYCL_L0_RTAS_BUILDER)
+    bool ze_rtas_builder = false;
+    for (uint32_t i=0; i<extensions.size(); i++)
+    {
+      if (strncmp("ZE_experimental_rtas_builder",extensions[i].name,sizeof(extensions[i].name)) == 0)
+        ze_rtas_builder = true;
+    }
+    if (!ze_rtas_builder)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "ZE_experimental_rtas_builder extension not found");
+
+    result = ZeWrapper::initRTASBuilder(hDriver,ZeWrapper::LEVEL_ZERO);
+    if (result == ZE_RESULT_ERROR_DEPENDENCY_UNAVAILABLE)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "cannot load ZE_experimental_rtas_builder extension");
+    if (result != ZE_RESULT_SUCCESS)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "cannot initialize ZE_experimental_rtas_builder extension");
+#else
+    ZeWrapper::initRTASBuilder(hDriver,ZeWrapper::INTERNAL);
+#endif
+
+    if (State::verbosity(1))
+    {
+      if (ZeWrapper::rtas_builder == ZeWrapper::INTERNAL)
+        std::cout << "  Internal RTAS Builder" << std::endl;
+      else
+        std::cout << "  Level Zero RTAS Builder" << std::endl;
+    }
+
+    /* check if extension library can get loaded */
+    ze_rtas_parallel_operation_exp_handle_t hParallelOperation;
+    result = ZeWrapper::zeRTASParallelOperationCreateExp(hDriver, &hParallelOperation);
+    if (result == ZE_RESULT_ERROR_DEPENDENCY_UNAVAILABLE)
+      throw_RTCError(RTC_ERROR_UNKNOWN, "Level Zero RTAS Build Extension cannot get loaded");
+    if (result == ZE_RESULT_SUCCESS)
+      ZeWrapper::zeRTASParallelOperationDestroyExp(hParallelOperation);
+
+    gpu_maxWorkGroupSize = getGPUDevice().get_info<sycl::info::device::max_work_group_size>();
+    gpu_maxComputeUnits  = getGPUDevice().get_info<sycl::info::device::max_compute_units>();    
+
+    if (State::verbosity(1))
+    {
+      sycl::platform platform = gpu_context.get_platform();
+      std::cout << "  Platform              : " << platform.get_info<sycl::info::platform::name>() << std::endl;
+      std::cout << "    Device              : " << getGPUDevice().get_info<sycl::info::device::name>() << std::endl;
+      std::cout << "    Max Work Group Size : " << gpu_maxWorkGroupSize << std::endl;
+      std::cout << "    Max Compute Units   : " << gpu_maxComputeUnits  << std::endl;
+      std::cout << std::endl;
+    }
+    
+    dispatchGlobalsPtr = zeRTASInitExp(gpu_device, gpu_context);
+  }
+
+  DeviceGPU::~DeviceGPU()
+  {
+    rthwifCleanup(this,dispatchGlobalsPtr,gpu_context);
+  }
+
+  void DeviceGPU::enter() {
+    enableUSMAllocEmbree(&gpu_context,&gpu_device);
+  }
+
+  void DeviceGPU::leave() {
+    disableUSMAllocEmbree();
+  }
+
+  void* DeviceGPU::malloc(size_t size, size_t align) {
+    return alignedSYCLMalloc(&gpu_context,&gpu_device,size,align,EMBREE_USM_SHARED_DEVICE_READ_ONLY);
+  }
+
+  void DeviceGPU::free(void* ptr) {
+    alignedSYCLFree(&gpu_context,ptr);
+  }
+
+  void DeviceGPU::setSYCLDevice(const sycl::device sycl_device_in) {
+    gpu_device = sycl_device_in;
+  }
+  
+#endif
+
+  DeviceEnterLeave::DeviceEnterLeave (RTCDevice hdevice)
+    : device((Device*)hdevice)
+  {
+    assert(device);
+    device->refInc();
+    device->enter();
+  }
+  
+  DeviceEnterLeave::DeviceEnterLeave (RTCScene hscene)
+    : device(((Scene*)hscene)->device)
+  {
+    assert(device);
+    device->refInc();
+    device->enter();
+  }
+  
+  DeviceEnterLeave::DeviceEnterLeave (RTCGeometry hgeometry)
+    : device(((Geometry*)hgeometry)->device)
+  {
+    assert(device);
+    device->refInc();
+    device->enter();
+  }
+  
+  DeviceEnterLeave::DeviceEnterLeave (RTCBuffer hbuffer)
+    : device(((Buffer*)hbuffer)->device)
+  {
+    assert(device);
+    device->refInc();
+    device->enter();
+  }
+  
+  DeviceEnterLeave::~DeviceEnterLeave() {
+    device->leave();
+    device->refDec();
+  }
 }
diff --git a/thirdparty/embree/kernels/common/device.h b/thirdparty/embree/kernels/common/device.h
index 21c42c654d..c9e8888a5a 100644
--- a/thirdparty/embree/kernels/common/device.h
+++ b/thirdparty/embree/kernels/common/device.h
@@ -11,10 +11,57 @@ namespace embree
 {
   class BVH4Factory;
   class BVH8Factory;
+  struct TaskArena;
 
   class Device : public State, public MemoryMonitorInterface
   {
     ALIGNED_CLASS_(16);
+    
+  public:
+    
+    /*! allocator that performs unified shared memory allocations */
+    template<typename T, size_t alignment>
+    struct allocator
+    {
+      typedef T value_type;
+      typedef T* pointer;
+      typedef const T* const_pointer;
+      typedef T& reference;
+      typedef const T& const_reference;
+      typedef std::size_t size_type;
+      typedef std::ptrdiff_t difference_type;
+      
+      allocator() {}
+      
+      allocator(Device* device)
+        : device(device) {}
+      
+      __forceinline pointer allocate( size_type n ) {
+        assert(device);
+        return (pointer) device->malloc(n*sizeof(T),alignment);
+      }
+      
+      __forceinline void deallocate( pointer p, size_type n ) {
+        if (device) device->free(p);
+      }
+      
+      __forceinline void construct( pointer p, const_reference val ) {
+        new (p) T(val);
+      }
+      
+      __forceinline void destroy( pointer p ) {
+        p->~T();
+      }
+      
+      Device* device = nullptr;
+    };
+
+    /*! vector class that performs aligned allocations from Device object */
+    template<typename T>
+    using vector = vector_t<T,allocator<T,std::alignment_of<T>::value>>;
+
+    template<typename T, size_t alignment>
+    using avector = vector_t<T,allocator<T,alignment>>;
 
   public:
 
@@ -54,6 +101,18 @@ namespace embree
     /*! gets a property */
     ssize_t getProperty(const RTCDeviceProperty prop);
 
+    /*! enter device by setting up some global state */
+    virtual void enter() {}
+
+    /*! leave device by setting up some global state */
+    virtual void leave() {}
+
+    /*! buffer allocation */
+    virtual void* malloc(size_t size, size_t align);
+
+    /*! buffer deallocation */
+    virtual void free(void* ptr);
+
   private:
 
     /*! initializes the tasking system */
@@ -62,6 +121,13 @@ namespace embree
     /*! shuts down the tasking system */
     void exitTaskingSystem();
 
+    std::unique_ptr<TaskArena> arena;
+
+  public:
+
+    // use tasking system arena to execute func
+    void execute(bool join, const std::function<void()>& func);
+
     /*! some variables that can be set via rtcSetParameter1i for debugging purposes */
   public:
     static ssize_t debug_int0;
@@ -74,12 +140,55 @@ namespace embree
 #if defined(EMBREE_TARGET_SIMD8)
     std::unique_ptr<BVH8Factory> bvh8_factory;
 #endif
-    
-#if USE_TASK_ARENA
-    std::unique_ptr<tbb::task_arena> arena;
+  };
+
+#if defined(EMBREE_SYCL_SUPPORT)
+     
+  class DeviceGPU : public Device
+  {
+  public:
+
+    DeviceGPU(sycl::context sycl_context, const char* cfg);
+    ~DeviceGPU();
+
+    virtual void enter() override;
+    virtual void leave() override;
+    virtual void* malloc(size_t size, size_t align) override;
+    virtual void free(void* ptr) override;
+
+    /* set SYCL device */
+    void setSYCLDevice(const sycl::device sycl_device);
+
+  private:
+    sycl::context gpu_context;
+    sycl::device  gpu_device;
+        
+    unsigned int gpu_maxWorkGroupSize;
+    unsigned int gpu_maxComputeUnits;
+
+  public:
+    void* dispatchGlobalsPtr = nullptr;
+
+  public:
+    inline sycl::device  &getGPUDevice()  { return gpu_device; }        
+    inline sycl::context &getGPUContext() { return gpu_context; }    
+
+    inline unsigned int getGPUMaxWorkGroupSize() { return gpu_maxWorkGroupSize; }
+
+    void init_rthw_level_zero();
+    void init_rthw_opencl();
+  };
+
 #endif
-    
-    /* ray streams filter */
-    RayStreamFilterFuncs rayStreamFilters;
+
+  struct DeviceEnterLeave
+  {
+    DeviceEnterLeave (RTCDevice hdevice);
+    DeviceEnterLeave (RTCScene hscene);
+    DeviceEnterLeave (RTCGeometry hgeometry);
+    DeviceEnterLeave (RTCBuffer hbuffer);
+    ~DeviceEnterLeave();
+  private:
+    Device* device;
   };
 }
diff --git a/thirdparty/embree/kernels/common/geometry.cpp b/thirdparty/embree/kernels/common/geometry.cpp
index d8d3f65a5c..79a6eb00d7 100644
--- a/thirdparty/embree/kernels/common/geometry.cpp
+++ b/thirdparty/embree/kernels/common/geometry.cpp
@@ -45,12 +45,13 @@ namespace embree
   Geometry::Geometry (Device* device, GType gtype, unsigned int numPrimitives, unsigned int numTimeSteps) 
     : device(device), userPtr(nullptr),
       numPrimitives(numPrimitives), numTimeSteps(unsigned(numTimeSteps)), fnumTimeSegments(float(numTimeSteps-1)), time_range(0.0f,1.0f),
-      mask(-1),
+      mask(1),
       gtype(gtype),
       gsubtype(GTY_SUBTYPE_DEFAULT),
       quality(RTC_BUILD_QUALITY_MEDIUM),
       state((unsigned)State::MODIFIED),
       enabled(true),
+      argumentFilterEnabled(false),
       intersectionFilterN(nullptr), occlusionFilterN(nullptr), pointQueryFunc(nullptr)
   {
     device->refInc();
@@ -88,6 +89,11 @@ namespace embree
     Geometry::update();
   }
   
+  BBox1f Geometry::getTimeRange () const
+  {
+    return time_range;
+  }
+
   void Geometry::update()
   {
     ++modCounter_; // FIXME: required?
@@ -227,11 +233,11 @@ namespace embree
       }
     }
   }
-    
+
   bool Geometry::pointQuery(PointQuery* query, PointQueryContext* context)
   {
     assert(context->primID < size());
-   
+
     RTCPointQueryFunctionArguments args;
     args.query           = (RTCPointQuery*)context->query_ws;
     args.userPtr         = context->userPtr;
@@ -239,7 +245,7 @@ namespace embree
     args.geomID          = context->geomID;
     args.context         = context->userContext;
     args.similarityScale = context->similarityScale;
-    
+
     bool update = false;
     if(context->func)  update |= context->func(&args);
     if(pointQueryFunc) update |= pointQueryFunc(&args);
diff --git a/thirdparty/embree/kernels/common/geometry.h b/thirdparty/embree/kernels/common/geometry.h
index 593990f5b1..00e3c5ede3 100644
--- a/thirdparty/embree/kernels/common/geometry.h
+++ b/thirdparty/embree/kernels/common/geometry.h
@@ -8,6 +8,7 @@
 #include "buffer.h"
 #include "../common/point_query.h"
 #include "../builders/priminfo.h"
+#include "../builders/priminfo_mb.h"
 
 namespace embree
 {
@@ -26,12 +27,14 @@ namespace embree
         numUserGeometries(0), numMBUserGeometries(0), 
         numInstancesCheap(0), numMBInstancesCheap(0), 
         numInstancesExpensive(0), numMBInstancesExpensive(0), 
-        numGrids(0), numMBGrids(0), 
+        numInstanceArrays(0), numMBInstanceArrays(0),
+        numGrids(0), numMBGrids(0),
+        numSubGrids(0), numMBSubGrids(0), 
         numPoints(0), numMBPoints(0) {}
 
     __forceinline size_t size() const {
-      return    numTriangles + numQuads + numBezierCurves + numLineSegments + numSubdivPatches + numUserGeometries + numInstancesCheap + numInstancesExpensive + numGrids + numPoints
-              + numMBTriangles + numMBQuads + numMBBezierCurves + numMBLineSegments + numMBSubdivPatches + numMBUserGeometries + numMBInstancesCheap + numMBInstancesExpensive + numMBGrids + numMBPoints;
+      return    numTriangles + numQuads + numBezierCurves + numLineSegments + numSubdivPatches + numUserGeometries + numInstancesCheap + numInstancesExpensive + numInstanceArrays + numGrids + numPoints
+              + numMBTriangles + numMBQuads + numMBBezierCurves + numMBLineSegments + numMBSubdivPatches + numMBUserGeometries + numMBInstancesCheap + numMBInstancesExpensive + numMBInstanceArrays + numMBGrids + numMBPoints;
     }
 
     __forceinline unsigned int enabledGeometryTypesMask() const
@@ -44,8 +47,9 @@ namespace embree
       if (numUserGeometries) mask |= 1 << 4;
       if (numInstancesCheap) mask |= 1 << 5;
       if (numInstancesExpensive) mask |= 1 << 6;
-      if (numGrids) mask |= 1 << 7;
-      if (numPoints) mask |= 1 << 8;
+      if (numInstanceArrays) mask |= 1 << 7;
+      if (numGrids) mask |= 1 << 8;
+      if (numPoints) mask |= 1 << 9;
 
       unsigned int maskMB = 0;
       if (numMBTriangles) maskMB |= 1 << 0;
@@ -55,8 +59,9 @@ namespace embree
       if (numMBUserGeometries) maskMB |= 1 << 4;
       if (numMBInstancesCheap) maskMB |= 1 << 5;
       if (numMBInstancesExpensive) maskMB |= 1 << 6;
-      if (numMBGrids) maskMB |= 1 << 7;
-      if (numMBPoints) maskMB |= 1 << 8;
+      if (numMBInstanceArrays) maskMB |= 1 << 7;
+      if (numMBGrids) maskMB |= 1 << 8;
+      if (numMBPoints) maskMB |= 1 << 9;
       
       return (mask<<8) + maskMB;
     }
@@ -81,8 +86,12 @@ namespace embree
       ret.numMBInstancesCheap = numMBInstancesCheap + rhs.numMBInstancesCheap;
       ret.numInstancesExpensive = numInstancesExpensive + rhs.numInstancesExpensive;
       ret.numMBInstancesExpensive = numMBInstancesExpensive + rhs.numMBInstancesExpensive;
+      ret.numInstanceArrays = numInstanceArrays + rhs.numInstanceArrays;
+      ret.numMBInstanceArrays = numMBInstanceArrays + rhs.numMBInstanceArrays;
       ret.numGrids = numGrids + rhs.numGrids;
       ret.numMBGrids = numMBGrids + rhs.numMBGrids;
+      ret.numSubGrids = numSubGrids + rhs.numSubGrids;
+      ret.numMBSubGrids = numMBSubGrids + rhs.numMBSubGrids;
       ret.numPoints = numPoints + rhs.numPoints;
       ret.numMBPoints = numMBPoints + rhs.numMBPoints;
 
@@ -106,8 +115,12 @@ namespace embree
     size_t numMBInstancesCheap;      //!< number of enabled motion blurred cheap instances
     size_t numInstancesExpensive;    //!< number of enabled expensive instances
     size_t numMBInstancesExpensive;  //!< number of enabled motion blurred expensive instances
+    size_t numInstanceArrays;        //!< number of enabled instance arrays
+    size_t numMBInstanceArrays;      //!< number of enabled motion blurred instance arrays
     size_t numGrids;                 //!< number of enabled grid geometries
     size_t numMBGrids;               //!< number of enabled motion blurred grid geometries
+    size_t numSubGrids;              //!< number of enabled grid geometries
+    size_t numMBSubGrids;            //!< number of enabled motion blurred grid geometries
     size_t numPoints;                //!< number of enabled points
     size_t numMBPoints;              //!< number of enabled motion blurred points
   };
@@ -115,6 +128,8 @@ namespace embree
   /*! Base class all geometries are derived from */
   class Geometry : public RefCount
   {
+    ALIGNED_CLASS_USM_(16);
+    
     friend class Scene;
   public:
 
@@ -154,6 +169,7 @@ namespace embree
       GTY_USER_GEOMETRY = 29,
       GTY_INSTANCE_CHEAP = 30,
       GTY_INSTANCE_EXPENSIVE = 31,
+      GTY_INSTANCE_ARRAY = 24,
       GTY_END = 32,
 
       GTY_BASIS_LINEAR = 0,
@@ -222,7 +238,10 @@ namespace embree
 
       MTY_INSTANCE_CHEAP = 1ul << GTY_INSTANCE_CHEAP,
       MTY_INSTANCE_EXPENSIVE = 1ul << GTY_INSTANCE_EXPENSIVE,
-      MTY_INSTANCE = MTY_INSTANCE_CHEAP | MTY_INSTANCE_EXPENSIVE
+      MTY_INSTANCE = MTY_INSTANCE_CHEAP | MTY_INSTANCE_EXPENSIVE,
+      MTY_INSTANCE_ARRAY = 1ul << GTY_INSTANCE_ARRAY,
+
+      MTY_ALL = -1
     };
 
     static const char* gtype_names[GTY_END];
@@ -248,8 +267,13 @@ namespace embree
     /*! tests if geometry is disabled */
     __forceinline bool isDisabled() const { return !isEnabled(); }
 
+    /* checks if argument version of filter functions are enabled */
+    __forceinline bool hasArgumentFilterFunctions() const {
+      return argumentFilterEnabled;
+    }
+    
     /*! tests if that geometry has some filter function set */
-    __forceinline bool hasFilterFunctions () const {
+    __forceinline bool hasGeometryFilterFunctions () const {
       return (intersectionFilterN  != nullptr) || (occlusionFilterN  != nullptr);
     }
 
@@ -265,6 +289,11 @@ namespace embree
     /*! returns geometry type mask */
     __forceinline GTypeMask getTypeMask() const { return (GTypeMask)(1 << gtype); }
 
+    /*! returns true of geometry contains motion blur */
+    __forceinline bool hasMotionBlur () const {
+      return numTimeSteps > 1;
+    }
+
     /*! returns number of primitives */
     __forceinline size_t size() const { return numPrimitives; }
 
@@ -277,6 +306,9 @@ namespace embree
     /*! sets motion blur time range */
     void setTimeRange (const BBox1f range);
 
+    /*! gets motion blur time range */
+    BBox1f getTimeRange () const;
+
     /*! sets number of vertex attributes */
     virtual void setVertexAttributeCount (unsigned int N) {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry"); 
@@ -447,6 +479,11 @@ namespace embree
     /*! Set occlusion filter function for ray packets of size N. */
     virtual void setOcclusionFilterFunctionN (RTCFilterFunctionN filterN);
 
+    /* Enables argument version of intersection or occlusion filter function. */
+    virtual void enableFilterFunctionFromArguments (bool enable) {
+      argumentFilterEnabled = enable;
+    }
+
     /*! for instances only */
   public:
 
@@ -455,6 +492,11 @@ namespace embree
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry");
     }
 
+    /*! Sets the instanced scenes */
+    virtual void setInstancedScenes(const RTCScene* scenes, size_t numScenes) {
+      throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry");
+    }
+
     /*! Sets transformation of the instance */
     virtual void setTransform(const AffineSpace3fa& transform, unsigned int timeStep) {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry"); 
@@ -467,7 +509,12 @@ namespace embree
 
     /*! Returns the transformation of the instance */
     virtual AffineSpace3fa getTransform(float time) {
-      throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry"); 
+      throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry");
+    }
+
+    /*! Returns the transformation of the instance */
+    virtual AffineSpace3fa getTransform(size_t instance, float time) {
+      throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation not supported for this geometry");
     }
 
     /*! for user geometries only */
@@ -498,18 +545,47 @@ namespace embree
 
   public:
 
-    virtual PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const {
+    virtual PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"createPrimRefArray not implemented for this geometry"); 
     }
 
+    PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefArray(prims.data(),r,k,geomID);
+    }
+
+    PrimInfo createPrimRefArray(avector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefArray(prims.data(),r,k,geomID);
+    }
+
+    virtual PrimInfo createPrimRefArray(mvector<PrimRef>& prims, mvector<SubGridBuildData>& sgrids, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefArray(prims,r,k,geomID);
+    }
+
     virtual PrimInfo createPrimRefArrayMB(mvector<PrimRef>& prims, size_t itime, const range<size_t>& r, size_t k, unsigned int geomID) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"createPrimRefMBArray not implemented for this geometry"); 
     }
 
+    /*! Calculates the PrimRef over the complete time interval */
+    virtual PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      throw_RTCError(RTC_ERROR_INVALID_OPERATION,"createPrimRefMBArray not implemented for this geometry");
+    }
+
+    PrimInfo createPrimRefArrayMB(mvector<PrimRef>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefArrayMB(prims.data(),t0t1,r,k,geomID);
+    }
+
+    PrimInfo createPrimRefArrayMB(avector<PrimRef>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefArrayMB(prims.data(),t0t1,r,k,geomID);
+    }
+    
     virtual PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"createPrimRefMBArray not implemented for this geometry"); 
     }
 
+    virtual PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, mvector<SubGridBuildData>& sgrids, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const {
+      return createPrimRefMBArray(prims,t0t1,r,k,geomID);
+    }
+
     virtual LinearSpace3fa computeAlignedSpace(const size_t primID) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"computeAlignedSpace not implemented for this geometry"); 
     }
@@ -541,6 +617,10 @@ namespace embree
     virtual LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"vlinearBounds not implemented for this geometry"); 
     }
+
+    virtual LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range, const SubGridBuildData * const sgrids) const {
+      return vlinearBounds(primID,time_range);
+    }
     
     virtual LBBox3fa vlinearBounds(const LinearSpace3fa& space, size_t primID, const BBox1f& time_range) const {
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"vlinearBounds not implemented for this geometry"); 
@@ -566,13 +646,14 @@ namespace embree
     
     unsigned int mask;             //!< for masking out geometry
     unsigned int modCounter_ = 1; //!< counter for every modification - used to rebuild scenes when geo is modified
-    
+
     struct {
       GType gtype : 8;                //!< geometry type
       GSubType gsubtype : 8;          //!< geometry subtype
       RTCBuildQuality quality : 3;    //!< build quality for geometry
       unsigned state : 2;
-      bool enabled : 1;              //!< true if geometry is enabled
+      bool enabled : 1;               //!< true if geometry is enabled
+      bool argumentFilterEnabled : 1; //!< true if argument filter functions are enabled for this geometry
     };
        
     RTCFilterFunctionN intersectionFilterN;
diff --git a/thirdparty/embree/kernels/common/hit.h b/thirdparty/embree/kernels/common/hit.h
index fd1a9d6391..cbaeb9b73a 100644
--- a/thirdparty/embree/kernels/common/hit.h
+++ b/thirdparty/embree/kernels/common/hit.h
@@ -17,14 +17,26 @@ namespace embree
     __forceinline HitK() {}
 
     /* Constructs a hit */
-    __forceinline HitK(const RTCIntersectContext* context, const vuint<K>& geomID, const vuint<K>& primID, const vfloat<K>& u, const vfloat<K>& v, const Vec3vf<K>& Ng)
+    __forceinline HitK(const RTCRayQueryContext* context, const vuint<K>& geomID, const vuint<K>& primID, const vfloat<K>& u, const vfloat<K>& v, const Vec3vf<K>& Ng)
       : Ng(Ng), u(u), v(v), primID(primID), geomID(geomID) 
     {
-      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l) {
         instID[l] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        instPrimID[l] = RTC_INVALID_GEOMETRY_ID;
+#endif
+      }
+      
       instance_id_stack::copy_UV<K>(context->instID, instID);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+      instance_id_stack::copy_UV<K>(context->instPrimID, instPrimID);
+#endif
     }
 
+    /* Constructs a hit */
+    __forceinline HitK(const RTCRayQueryContext* context, const vuint<K>& geomID, const vuint<K>& primID, const Vec2vf<K>& uv, const Vec3vf<K>& Ng)
+      : HitK(context,geomID,primID,uv.x,uv.y,Ng) {}
+
     /* Returns the size of the hit */
     static __forceinline size_t size() { return K; }
 
@@ -35,6 +47,9 @@ namespace embree
     vuint<K> primID;      // primitive ID
     vuint<K> geomID;      // geometry ID
     vuint<K> instID[RTC_MAX_INSTANCE_LEVEL_COUNT];      // instance ID
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    vuint<K> instPrimID[RTC_MAX_INSTANCE_LEVEL_COUNT];      // instance primitive ID
+#endif
   };
 
   /* Specialization for a single hit */
@@ -45,12 +60,19 @@ namespace embree
     __forceinline HitK() {}
 
     /* Constructs a hit */
-    __forceinline HitK(const RTCIntersectContext* context, unsigned int geomID, unsigned int primID, float u, float v, const Vec3fa& Ng)
+    __forceinline HitK(const RTCRayQueryContext* context, unsigned int geomID, unsigned int primID, float u, float v, const Vec3fa& Ng)
       : Ng(Ng.x,Ng.y,Ng.z), u(u), v(v), primID(primID), geomID(geomID)
     {
-      instance_id_stack::copy_UU(context->instID, instID);
+      instance_id_stack::copy_UU(context, context->instID, instID);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+      instance_id_stack::copy_UU(context, context->instPrimID, instPrimID);
+#endif
     }
 
+    /* Constructs a hit */
+    __forceinline HitK(const RTCRayQueryContext* context, unsigned int geomID, unsigned int primID, const Vec2f& uv, const Vec3fa& Ng)
+      : HitK<1>(context,geomID,primID,uv.x,uv.y,Ng) {}
+
     /* Returns the size of the hit */
     static __forceinline size_t size() { return 1; }
 
@@ -61,6 +83,9 @@ namespace embree
     unsigned int primID;      // primitive ID
     unsigned int geomID;      // geometry ID
     unsigned int instID[RTC_MAX_INSTANCE_LEVEL_COUNT];      // instance ID
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    unsigned int instPrimID[RTC_MAX_INSTANCE_LEVEL_COUNT];      // instance primitive ID
+#endif
   };
 
   /* Shortcuts */
@@ -68,6 +93,7 @@ namespace embree
   typedef HitK<4>  Hit4;
   typedef HitK<8>  Hit8;
   typedef HitK<16> Hit16;
+  typedef HitK<VSIZEX> Hitx;
 
   /* Outputs hit to stream */
   template<int K>
@@ -84,6 +110,13 @@ namespace embree
     {
       cout << " " << ray.instID[l];
     }
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    cout << "  instPrimID =";
+    for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+    {
+      cout << " " << ray.instPrimID[l];
+    }
+#endif
     cout << embree_endl;
     return cout << "}";
   }
@@ -97,10 +130,13 @@ namespace embree
     ray.primID = hit.primID;
     ray.geomID = hit.geomID;
     instance_id_stack::copy_UU(hit.instID, ray.instID);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    instance_id_stack::copy_UU(hit.instPrimID, ray.instPrimID);
+#endif
   }
 
   template<int K>
-    __forceinline void copyHitToRay(const vbool<K> &mask, RayHitK<K> &ray, const HitK<K> &hit)
+    __forceinline void copyHitToRay(const vbool<K>& mask, RayHitK<K>& ray, const HitK<K>& hit)
   {
     vfloat<K>::storeu(mask,&ray.Ng.x, hit.Ng.x);
     vfloat<K>::storeu(mask,&ray.Ng.y, hit.Ng.y);
@@ -110,5 +146,8 @@ namespace embree
     vuint<K>::storeu(mask,&ray.primID, hit.primID);
     vuint<K>::storeu(mask,&ray.geomID, hit.geomID);
     instance_id_stack::copy_VV<K>(hit.instID, ray.instID, mask);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    instance_id_stack::copy_VV<K>(hit.instPrimID, ray.instPrimID, mask);
+#endif
   }
 }
diff --git a/thirdparty/embree/kernels/common/instance_stack.h b/thirdparty/embree/kernels/common/instance_stack.h
index d3c0a643f1..32b57b48a3 100644
--- a/thirdparty/embree/kernels/common/instance_stack.h
+++ b/thirdparty/embree/kernels/common/instance_stack.h
@@ -19,40 +19,108 @@ static_assert(RTC_MAX_INSTANCE_LEVEL_COUNT > 0,
 /* 
  * Push an instance to the stack. 
  */
-RTC_FORCEINLINE bool push(RTCIntersectContext* context, 
-                          unsigned instanceId)
+template<typename Context>
+RTC_FORCEINLINE bool push(Context context,
+                          unsigned instanceId,
+                          unsigned instancePrimId)
 {
 #if RTC_MAX_INSTANCE_LEVEL_COUNT > 1
   const bool spaceAvailable = context->instStackSize < RTC_MAX_INSTANCE_LEVEL_COUNT;
-  /* We assert here because instances are silently dropped when the stack is full. 
+  /* We assert here because instances are silently dropped when the stack is full.
      This might be quite hard to find in production. */
-  assert(spaceAvailable); 
-  if (likely(spaceAvailable))
-    context->instID[context->instStackSize++] = instanceId;
+  assert(spaceAvailable);
+  if (likely(spaceAvailable)) {
+    context->instID[context->instStackSize] = instanceId;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    context->instPrimID[context->instStackSize] = instancePrimId;
+#endif
+    context->instStackSize++;
+  }
   return spaceAvailable;
 #else
   const bool spaceAvailable = (context->instID[0] == RTC_INVALID_GEOMETRY_ID);
-  assert(spaceAvailable); 
-  if (likely(spaceAvailable))
+  assert(spaceAvailable);
+  if (likely(spaceAvailable)) {
     context->instID[0] = instanceId;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    context->instPrimID[0] = instancePrimId;
+#endif
+  }
   return spaceAvailable;
 #endif
 }
 
-
 /* 
  * Pop the last instance pushed to the stack. 
  * Do not call on an empty stack. 
  */
-RTC_FORCEINLINE void pop(RTCIntersectContext* context)
+template<typename Context>
+RTC_FORCEINLINE void pop(Context context)
 {
   assert(context);
 #if RTC_MAX_INSTANCE_LEVEL_COUNT > 1
   assert(context->instStackSize > 0);
-  context->instID[--context->instStackSize] = RTC_INVALID_GEOMETRY_ID;
+  --context->instStackSize;
+  context->instID[context->instStackSize] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+  context->instPrimID[context->instStackSize] = RTC_INVALID_GEOMETRY_ID;
+#endif
 #else
   assert(context->instID[0] != RTC_INVALID_GEOMETRY_ID);
   context->instID[0] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+  context->instPrimID[0] = RTC_INVALID_GEOMETRY_ID;
+#endif
+#endif
+}
+
+
+/* Push an instance to the stack. Used for point queries*/
+RTC_FORCEINLINE bool push(RTCPointQueryContext* context,
+                          unsigned int instanceId,
+                          unsigned int instancePrimId,
+                          AffineSpace3fa const& w2i,
+                          AffineSpace3fa const& i2w)
+{
+  assert(context);
+  const size_t stackSize = context->instStackSize;
+  assert(stackSize < RTC_MAX_INSTANCE_LEVEL_COUNT);
+  context->instID[stackSize] = instanceId;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+  context->instPrimID[stackSize] = instancePrimId;
+#endif
+
+  AffineSpace3fa_store_unaligned(w2i,(AffineSpace3fa*)context->world2inst[stackSize]);
+  AffineSpace3fa_store_unaligned(i2w,(AffineSpace3fa*)context->inst2world[stackSize]);
+
+#if RTC_MAX_INSTANCE_LEVEL_COUNT > 1
+  if (unlikely(stackSize > 0))
+  {
+    const AffineSpace3fa world2inst = AffineSpace3fa_load_unaligned((AffineSpace3fa*)context->world2inst[stackSize  ])
+                                    * AffineSpace3fa_load_unaligned((AffineSpace3fa*)context->world2inst[stackSize-1]);
+    const AffineSpace3fa inst2world = AffineSpace3fa_load_unaligned((AffineSpace3fa*)context->inst2world[stackSize-1])
+                                    * AffineSpace3fa_load_unaligned((AffineSpace3fa*)context->inst2world[stackSize  ]);
+    AffineSpace3fa_store_unaligned(world2inst,(AffineSpace3fa*)context->world2inst[stackSize]);
+    AffineSpace3fa_store_unaligned(inst2world,(AffineSpace3fa*)context->inst2world[stackSize]);
+  }
+#endif
+  context->instStackSize++;
+  return true;
+}
+
+template<>
+RTC_FORCEINLINE void pop(RTCPointQueryContext* context)
+{
+  assert(context);
+#if RTC_MAX_INSTANCE_LEVEL_COUNT > 1
+  assert(context->instStackSize > 0);
+#else
+  assert(context->instID[0] != RTC_INVALID_GEOMETRY_ID);
+#endif
+  --context->instStackSize;
+  context->instID[context->instStackSize] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+  context->instPrimID[context->instStackSize] = RTC_INVALID_GEOMETRY_ID;
 #endif
 }
 
@@ -77,6 +145,24 @@ RTC_FORCEINLINE void copy_UU(const unsigned* src, unsigned* tgt)
 #endif
 }
 
+RTC_FORCEINLINE void copy_UU(const RTCRayQueryContext* context, const unsigned* src, unsigned* tgt)
+{
+#if (RTC_MAX_INSTANCE_LEVEL_COUNT == 1)
+  tgt[0] = src[0];
+  
+#else
+  
+  unsigned int depth = context->instStackSize;
+  
+  for (unsigned l = 0; l < depth; ++l)
+    tgt[l] = src[l];
+  
+  for (unsigned l = depth; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+    tgt[l] = RTC_INVALID_GEOMETRY_ID;
+
+#endif
+}
+  
 template <int K>
 RTC_FORCEINLINE void copy_UV(const unsigned* src, vuint<K>* tgt)
 {
diff --git a/thirdparty/embree/kernels/common/point_query.h b/thirdparty/embree/kernels/common/point_query.h
index 7d55c91fff..e92e22ae36 100644
--- a/thirdparty/embree/kernels/common/point_query.h
+++ b/thirdparty/embree/kernels/common/point_query.h
@@ -120,6 +120,7 @@ namespace embree
   typedef PointQueryK<4>  PointQuery4;
   typedef PointQueryK<8>  PointQuery8;
   typedef PointQueryK<16> PointQuery16;
+  typedef PointQueryK<VSIZEX> PointQueryx;
   struct PointQueryN;
 
   /* Outputs point query to stream */
diff --git a/thirdparty/embree/kernels/common/primref.h b/thirdparty/embree/kernels/common/primref.h
deleted file mode 100644
index d61763487b..0000000000
--- a/thirdparty/embree/kernels/common/primref.h
+++ /dev/null
@@ -1,138 +0,0 @@
-// Copyright 2009-2021 Intel Corporation
-// SPDX-License-Identifier: Apache-2.0
-
-#pragma once
-
-#include "default.h"
-
-namespace embree
-{
-  /*! A primitive reference stores the bounds of the primitive and its ID. */
-  struct __aligned(32) PrimRef 
-  {
-    __forceinline PrimRef () {}
-
-#if defined(__AVX__)
-    __forceinline PrimRef(const PrimRef& v) { 
-      vfloat8::store((float*)this,vfloat8::load((float*)&v));
-    }
-    __forceinline PrimRef& operator=(const PrimRef& v) { 
-      vfloat8::store((float*)this,vfloat8::load((float*)&v)); return *this;
-    }
-#endif
-
-    __forceinline PrimRef (const BBox3fa& bounds, unsigned int geomID, unsigned int primID) 
-    {
-      lower = Vec3fx(bounds.lower, geomID);
-      upper = Vec3fx(bounds.upper, primID);
-    }
-
-    __forceinline PrimRef (const BBox3fa& bounds, size_t id) 
-    {
-#if defined(__64BIT__)
-      lower = Vec3fx(bounds.lower, (unsigned)(id & 0xFFFFFFFF));
-      upper = Vec3fx(bounds.upper, (unsigned)((id >> 32) & 0xFFFFFFFF));
-#else
-      lower = Vec3fx(bounds.lower, (unsigned)id);
-      upper = Vec3fx(bounds.upper, (unsigned)0);
-#endif
-    }
-
-    /*! calculates twice the center of the primitive */
-    __forceinline const Vec3fa center2() const {
-      return lower+upper;
-    }
-    
-    /*! return the bounding box of the primitive */
-    __forceinline const BBox3fa bounds() const {
-      return BBox3fa(lower,upper);
-    }
-
-    /*! size for bin heuristic is 1 */
-    __forceinline unsigned size() const { 
-      return 1;
-    }
-
-    /*! returns bounds and centroid used for binning */
-    __forceinline void binBoundsAndCenter(BBox3fa& bounds_o, Vec3fa& center_o) const 
-    {
-      bounds_o = bounds();
-      center_o = embree::center2(bounds_o);
-    }
-
-    __forceinline unsigned& geomIDref() {  // FIXME: remove !!!!!!!
-      return lower.u;
-    }
-    __forceinline unsigned& primIDref() {  // FIXME: remove !!!!!!!
-      return upper.u;
-    }
-    
-    /*! returns the geometry ID */
-    __forceinline unsigned geomID() const { 
-      return lower.a;
-    }
-
-    /*! returns the primitive ID */
-    __forceinline unsigned primID() const { 
-      return upper.a;
-    }
-
-    /*! returns an size_t sized ID */
-    __forceinline size_t ID() const { 
-#if defined(__64BIT__)
-      return size_t(lower.u) + (size_t(upper.u) << 32);
-#else
-      return size_t(lower.u);
-#endif
-    }
-
-    /*! special function for operator< */
-    __forceinline uint64_t ID64() const {
-      return (((uint64_t)primID()) << 32) + (uint64_t)geomID();
-    }
-    
-    /*! allows sorting the primrefs by ID */
-    friend __forceinline bool operator<(const PrimRef& p0, const PrimRef& p1) {
-      return p0.ID64() < p1.ID64();
-    }
-
-    /*! Outputs primitive reference to a stream. */
-    friend __forceinline embree_ostream operator<<(embree_ostream cout, const PrimRef& ref) {
-      return cout << "{ lower = " << ref.lower << ", upper = " << ref.upper << ", geomID = " << ref.geomID() << ", primID = " << ref.primID() << " }";
-    }
-
-  public:
-    Vec3fx lower;     //!< lower bounds and geomID
-    Vec3fx upper;     //!< upper bounds and primID
-  };
-
-  /*! fast exchange for PrimRefs */
-  __forceinline void xchg(PrimRef& a, PrimRef& b)
-  {
-#if defined(__AVX__)
-    const vfloat8 aa = vfloat8::load((float*)&a);
-    const vfloat8 bb = vfloat8::load((float*)&b);
-    vfloat8::store((float*)&a,bb);
-    vfloat8::store((float*)&b,aa);
-#else
-    std::swap(a,b);
-#endif
-  }
-
-  /************************************************************************************/
-  /************************************************************************************/
-  /************************************************************************************/
-  /************************************************************************************/
-  
-  struct SubGridBuildData {
-    unsigned short sx,sy;
-    unsigned int primID;
-    
-    __forceinline SubGridBuildData() {};
-    __forceinline SubGridBuildData(const unsigned int sx, const unsigned int sy, const unsigned int primID) : sx(sx), sy(sy), primID(primID) {};
-    
-    __forceinline size_t x() const { return (size_t)sx & 0x7fff; }
-    __forceinline size_t y() const { return (size_t)sy & 0x7fff; }
-    
-  };
-}
diff --git a/thirdparty/embree/kernels/common/primref_mb.h b/thirdparty/embree/kernels/common/primref_mb.h
deleted file mode 100644
index fb08a05003..0000000000
--- a/thirdparty/embree/kernels/common/primref_mb.h
+++ /dev/null
@@ -1,262 +0,0 @@
-// Copyright 2009-2021 Intel Corporation
-// SPDX-License-Identifier: Apache-2.0
-
-#pragma once
-
-#include "default.h"
-
-#define MBLUR_BIN_LBBOX 1
-
-namespace embree
-{
-#if MBLUR_BIN_LBBOX
-
-  /*! A primitive reference stores the bounds of the primitive and its ID. */
-  struct PrimRefMB
-  {
-    typedef LBBox3fa BBox;
-
-    __forceinline PrimRefMB () {}
-
-    __forceinline PrimRefMB (const LBBox3fa& lbounds_i, unsigned int activeTimeSegments, BBox1f time_range, unsigned int totalTimeSegments, unsigned int geomID, unsigned int primID)
-      : lbounds((LBBox3fx)lbounds_i), time_range(time_range)
-    {
-      assert(activeTimeSegments > 0);
-      lbounds.bounds0.lower.a = geomID;
-      lbounds.bounds0.upper.a = primID;
-      lbounds.bounds1.lower.a = activeTimeSegments;
-      lbounds.bounds1.upper.a = totalTimeSegments;
-    }
-
-    __forceinline PrimRefMB (EmptyTy empty, const LBBox3fa& lbounds_i, unsigned int activeTimeSegments, BBox1f time_range, unsigned int totalTimeSegments, size_t id)
-      : lbounds((LBBox3fx)lbounds_i), time_range(time_range)
-    {
-      assert(activeTimeSegments > 0);
-#if defined(__64BIT__)
-      lbounds.bounds0.lower.a = id & 0xFFFFFFFF;
-      lbounds.bounds0.upper.a = (id >> 32) & 0xFFFFFFFF;
-#else
-      lbounds.bounds0.lower.a = id;
-      lbounds.bounds0.upper.a = 0;
-#endif
-      lbounds.bounds1.lower.a = activeTimeSegments;
-      lbounds.bounds1.upper.a = totalTimeSegments;
-    }
-    
-    __forceinline PrimRefMB (const LBBox3fa& lbounds_i, unsigned int activeTimeSegments, BBox1f time_range, unsigned int totalTimeSegments, size_t id)
-      : lbounds((LBBox3fx)lbounds_i), time_range(time_range)
-    {
-      assert(activeTimeSegments > 0);
-#if defined(__64BIT__)
-      lbounds.bounds0.lower.u = id & 0xFFFFFFFF;
-      lbounds.bounds0.upper.u = (id >> 32) & 0xFFFFFFFF;
-#else
-      lbounds.bounds0.lower.u = id;
-      lbounds.bounds0.upper.u = 0;
-#endif
-      lbounds.bounds1.lower.a = activeTimeSegments;
-      lbounds.bounds1.upper.a = totalTimeSegments;
-    }
-
-    /*! returns bounds for binning */
-    __forceinline LBBox3fa bounds() const {
-      return lbounds;
-    }
-
-    /*! returns the number of time segments of this primref */
-    __forceinline unsigned size() const {
-      return lbounds.bounds1.lower.a;
-    }
-
-    __forceinline unsigned totalTimeSegments() const {
-      return lbounds.bounds1.upper.a;
-    }
-
-     /* calculate overlapping time segment range */
-    __forceinline range<int> timeSegmentRange(const BBox1f& range) const {
-      return getTimeSegmentRange(range,time_range,float(totalTimeSegments()));
-    }
-
-     /* returns time that corresponds to time step */
-    __forceinline float timeStep(const int i) const {
-      assert(i>=0 && i<=(int)totalTimeSegments());
-      return time_range.lower + time_range.size()*float(i)/float(totalTimeSegments());
-    }
-    
-    /*! checks if time range overlaps */
-    __forceinline bool time_range_overlap(const BBox1f& range) const
-    {
-      if (0.9999f*time_range.upper <= range.lower) return false;
-      if (1.0001f*time_range.lower >= range.upper) return false;
-      return true;
-    }
-
-    /*! returns center for binning */
-    __forceinline Vec3fa binCenter() const {
-      return center2(lbounds.interpolate(0.5f));
-    }
-
-    /*! returns bounds and centroid used for binning */
-    __forceinline void binBoundsAndCenter(LBBox3fa& bounds_o, Vec3fa& center_o) const
-    {
-      bounds_o = bounds();
-      center_o = binCenter();
-    }
-
-    /*! returns the geometry ID */
-    __forceinline unsigned geomID() const {
-      return lbounds.bounds0.lower.a;
-    }
-
-    /*! returns the primitive ID */
-    __forceinline unsigned primID() const {
-      return lbounds.bounds0.upper.a;
-    }
-
-    /*! returns an size_t sized ID */
-    __forceinline size_t ID() const {
-#if defined(__64BIT__)
-      return size_t(lbounds.bounds0.lower.u) + (size_t(lbounds.bounds0.upper.u) << 32);
-#else
-      return size_t(lbounds.bounds0.lower.u);
-#endif
-    }
-
-    /*! special function for operator< */
-    __forceinline uint64_t ID64() const {
-      return (((uint64_t)primID()) << 32) + (uint64_t)geomID();
-    }
-
-    /*! allows sorting the primrefs by ID */
-    friend __forceinline bool operator<(const PrimRefMB& p0, const PrimRefMB& p1) {
-      return p0.ID64() < p1.ID64();
-    }
-
-    /*! Outputs primitive reference to a stream. */
-    friend __forceinline embree_ostream operator<<(embree_ostream cout, const PrimRefMB& ref) {
-      return cout << "{ time_range = " << ref.time_range << ", bounds = " << ref.bounds() << ", geomID = " << ref.geomID() << ", primID = " << ref.primID() << ", active_segments = " << ref.size() << ",  total_segments = " << ref.totalTimeSegments() << " }";
-    }
-
-  public:
-    LBBox3fx lbounds;
-    BBox1f time_range; // entire geometry time range
-  };
-
-#else
-
-  /*! A primitive reference stores the bounds of the primitive and its ID. */
-  struct __aligned(16) PrimRefMB
-  {
-    typedef BBox3fa BBox;
-
-    __forceinline PrimRefMB () {}
-
-    __forceinline PrimRefMB (const LBBox3fa& bounds, unsigned int activeTimeSegments, BBox1f time_range, unsigned int totalTimeSegments, unsigned int geomID, unsigned int primID)
-      : bbox(bounds.interpolate(0.5f)), _activeTimeSegments(activeTimeSegments), _totalTimeSegments(totalTimeSegments), time_range(time_range)
-    {
-      assert(activeTimeSegments > 0);
-      bbox.lower.a = geomID;
-      bbox.upper.a = primID;
-    }
-    
-    __forceinline PrimRefMB (EmptyTy empty, const LBBox3fa& bounds, unsigned int activeTimeSegments, BBox1f time_range, unsigned int totalTimeSegments, size_t id)
-      : bbox(bounds.interpolate(0.5f)), _activeTimeSegments(activeTimeSegments), _totalTimeSegments(totalTimeSegments), time_range(time_range)
-    {
-      assert(activeTimeSegments > 0);
-#if defined(__64BIT__)
-      bbox.lower.u = id & 0xFFFFFFFF;
-      bbox.upper.u = (id >> 32) & 0xFFFFFFFF;
-#else
-      bbox.lower.u = id;
-      bbox.upper.u = 0;
-#endif
-    }
-    
-    /*! returns bounds for binning */
-    __forceinline BBox3fa bounds() const {
-      return bbox;
-    }
-
-    /*! returns the number of time segments of this primref */
-    __forceinline unsigned int size() const { 
-      return _activeTimeSegments;
-    }
-
-    __forceinline unsigned int totalTimeSegments() const { 
-      return _totalTimeSegments;
-    }
-
-     /* calculate overlapping time segment range */
-    __forceinline range<int> timeSegmentRange(const BBox1f& range) const {
-      return getTimeSegmentRange(range,time_range,float(_totalTimeSegments));
-    }
-
-     /* returns time that corresponds to time step */
-    __forceinline float timeStep(const int i) const {
-      assert(i>=0 && i<=(int)_totalTimeSegments);
-      return time_range.lower + time_range.size()*float(i)/float(_totalTimeSegments);
-    }
-    
-    /*! checks if time range overlaps */
-    __forceinline bool time_range_overlap(const BBox1f& range) const
-    {
-      if (0.9999f*time_range.upper <= range.lower) return false;
-      if (1.0001f*time_range.lower >= range.upper) return false;
-      return true;
-    }
-
-    /*! returns center for binning */
-    __forceinline Vec3fa binCenter() const {
-      return center2(bounds());
-    }
-
-    /*! returns bounds and centroid used for binning */
-    __forceinline void binBoundsAndCenter(BBox3fa& bounds_o, Vec3fa& center_o) const
-    {
-      bounds_o = bounds();
-      center_o = center2(bounds());
-    }
-
-    /*! returns the geometry ID */
-    __forceinline unsigned int geomID() const { 
-      return bbox.lower.a;
-    }
-
-    /*! returns the primitive ID */
-    __forceinline unsigned int primID() const { 
-      return bbox.upper.a;
-    }
-
-    /*! returns an size_t sized ID */
-    __forceinline size_t ID() const { 
-#if defined(__64BIT__)
-      return size_t(bbox.lower.u) + (size_t(bbox.upper.u) << 32);
-#else
-      return size_t(bbox.lower.u);
-#endif
-    }
-
-    /*! special function for operator< */
-    __forceinline uint64_t ID64() const {
-      return (((uint64_t)primID()) << 32) + (uint64_t)geomID();
-    }
-    
-    /*! allows sorting the primrefs by ID */
-    friend __forceinline bool operator<(const PrimRefMB& p0, const PrimRefMB& p1) {
-      return p0.ID64() < p1.ID64();
-    }
-
-    /*! Outputs primitive reference to a stream. */
-    friend __forceinline embree_ostream operator<<(embree_ostream cout, const PrimRefMB& ref) {
-      return cout << "{ bounds = " << ref.bounds() << ", geomID = " << ref.geomID() << ", primID = " << ref.primID() << ", active_segments = " << ref.size() << ",  total_segments = " << ref.totalTimeSegments() << " }";
-    }
-
-  public:
-    BBox3fa bbox; // bounds, geomID, primID
-    unsigned int _activeTimeSegments;
-    unsigned int _totalTimeSegments;
-    BBox1f time_range; // entire geometry time range
-  };
-
-#endif
-}
diff --git a/thirdparty/embree/kernels/common/ray.h b/thirdparty/embree/kernels/common/ray.h
index 3c8ee3989c..c886013354 100644
--- a/thirdparty/embree/kernels/common/ray.h
+++ b/thirdparty/embree/kernels/common/ray.h
@@ -10,8 +10,6 @@
 
 namespace embree
 {
-  static const size_t MAX_INTERNAL_STREAM_SIZE = 32;
-
   /* Ray structure for K rays */
   template<int K>
   struct RayK
@@ -97,16 +95,24 @@ namespace embree
       : RayK<K>(org, dir, tnear, tfar, time, mask, id, flags),
         geomID(RTC_INVALID_GEOMETRY_ID) 
     {
-      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l) {
         instID[l] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        instPrimID[l] = RTC_INVALID_GEOMETRY_ID;
+#endif
+      }
     }
 
     __forceinline RayHitK(const RayK<K>& ray)
       : RayK<K>(ray),
         geomID(RTC_INVALID_GEOMETRY_ID) 
     {
-      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l) {
         instID[l] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        instPrimID[l] = RTC_INVALID_GEOMETRY_ID;
+#endif
+      }
     }
 
     __forceinline RayHitK<K>& operator =(const RayK<K>& ray)
@@ -121,8 +127,12 @@ namespace embree
       flags  = ray.flags;
 
       geomID = RTC_INVALID_GEOMETRY_ID;
-      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l) {
         instID[l] = RTC_INVALID_GEOMETRY_ID;
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        instPrimID[l] = RTC_INVALID_GEOMETRY_ID;
+#endif
+      }
 
       return *this;
     }
@@ -159,7 +169,10 @@ namespace embree
     vuint<K> primID; // primitive ID
     vuint<K> geomID; // geometry ID
     vuint<K> instID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance ID
-  };
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    vuint<K> instPrimID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance prim ID
+#endif
+};
 
   /* Specialization for a single ray */
   template<>
@@ -178,6 +191,11 @@ namespace embree
       return all(le_mask(abs(Vec3fa(org)), Vec3fa(FLT_LARGE)) & le_mask(abs(Vec3fa(dir)), Vec3fa(FLT_LARGE))) && abs(tnear()) <= float(inf) && abs(tfar) <= float(inf);
     }
 
+    /* checks if occlusion ray is done */
+    __forceinline bool occluded() const {
+      return tfar < 0.0f;
+    }
+
     /* Ray data */
     Vec3ff org;  // 3 floats for ray origin, 1 float for tnear
     //float tnear; // start of ray segment
@@ -251,6 +269,9 @@ namespace embree
     unsigned int primID; // primitive ID
     unsigned int geomID; // geometry ID
     unsigned int instID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance ID
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    unsigned int instPrimID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance primitive ID
+#endif
   };
 
   /* Converts ray packet to single rays */
@@ -293,6 +314,9 @@ namespace embree
     ray.primID = primID[i]; ray.geomID = geomID[i]; 
 
     instance_id_stack::copy_VU<K>(instID, ray.instID, i);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    instance_id_stack::copy_VU<K>(instPrimID, ray.instPrimID, i);
+#endif
   }
 
   /* Converts single rays to ray packet */
@@ -332,6 +356,9 @@ namespace embree
     primID[i] = ray.primID; geomID[i] = ray.geomID;
 
     instance_id_stack::copy_UV<K>(ray.instID, instID, i);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    instance_id_stack::copy_UV<K>(ray.instPrimID, instPrimID, i);
+#endif
   }
 
   /* copies a ray packet element into another element*/
@@ -354,6 +381,9 @@ namespace embree
     primID[dest] = primID[source]; geomID[dest] = geomID[source];  
 
     instance_id_stack::copy_VV<K>(instID, instID, source, dest);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    instance_id_stack::copy_VV<K>(instPrimID, instPrimID, source, dest);
+#endif
   }
 
   /* Shortcuts */
@@ -361,12 +391,14 @@ namespace embree
   typedef RayK<4>  Ray4;
   typedef RayK<8>  Ray8;
   typedef RayK<16> Ray16;
+  typedef RayK<VSIZEX> Rayx;
   struct RayN;
 
   typedef RayHitK<1>  RayHit;
   typedef RayHitK<4>  RayHit4;
   typedef RayHitK<8>  RayHit8;
   typedef RayHitK<16> RayHit16;
+  typedef RayHitK<VSIZEX> RayHitx;
   struct RayHitN;
 
   template<int K, bool intersect>
@@ -428,6 +460,13 @@ namespace embree
     {
       cout << " " << ray.instID[l];
     }
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    cout << "  instPrimID =";
+    for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+    {
+      cout << " " << ray.instPrimID[l];
+    }
+#endif
     cout << embree_endl;
     return cout << "}";
   }
@@ -464,6 +503,9 @@ namespace embree
     __forceinline unsigned int* primID(size_t offset = 0) { return (unsigned int*)&ptr[17*4*N+offset]; };   // primitive ID
     __forceinline unsigned int* geomID(size_t offset = 0) { return (unsigned int*)&ptr[18*4*N+offset]; };   // geometry ID
     __forceinline unsigned int* instID(size_t level, size_t offset = 0) { return (unsigned int*)&ptr[19*4*N+level*4*N+offset]; };   // instance ID
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    __forceinline unsigned int* instPrimID(size_t level, size_t offset = 0) { return (unsigned int*)&ptr[19*4*N+RTC_MAX_INSTANCE_LEVEL_COUNT*4*N+level*4*N+offset]; };   // instance primitive ID
+#endif
 
     __forceinline Ray getRayByOffset(size_t offset)
     {
@@ -578,9 +620,16 @@ namespace embree
               geomID(offset)[k] = ray.geomID[k];
 
               instID(0, offset)[k] = ray.instID[0][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+              instPrimID(0, offset)[k] = ray.instPrimID[0][k];
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-              for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l)
+              for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l) {
                 instID(l, offset)[k] = ray.instID[l][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+                instPrimID(l, offset)[k] = ray.instPrimID[l][k];
+#endif
+              }
 #endif
             }
           }
@@ -592,9 +641,16 @@ namespace embree
           vuint<K>::storeu(valid, geomID(offset), ray.geomID);
 
           vuint<K>::storeu(valid, instID(0, offset), ray.instID[0]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          vuint<K>::storeu(valid, instPrimID(0, offset), ray.instPrimID[0]);
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l)
+          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l) {
             vuint<K>::storeu(valid, instID(l, offset), ray.instID[l]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            vuint<K>::storeu(valid, instPrimID(l, offset), ray.instPrimID[l]);
+#endif
+          }
 #endif
         }
       }
@@ -698,9 +754,16 @@ namespace embree
         vuint<K>::template scatter<1>(valid, geomID(), offset, ray.geomID);
 
         vuint<K>::template scatter<1>(valid, instID(0), offset, ray.instID[0]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        vuint<K>::template scatter<1>(valid, instPrimID(0), offset, ray.instPrimID[0]);
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-        for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l)
+        for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l) {
           vuint<K>::template scatter<1>(valid, instID(l), offset, ray.instID[l]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          vuint<K>::template scatter<1>(valid, instPrimID(l), offset, ray.instPrimID[l]);
+#endif
+        }
 #endif
 #else
         size_t valid_bits = movemask(valid);
@@ -720,9 +783,16 @@ namespace embree
           *geomID(ofs) = ray.geomID[k];
 
           *instID(0, ofs) = ray.instID[0][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          *instPrimID(0, ofs) = ray.instPrimID[0][k];
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l)
+          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l) {
             *instID(l, ofs) = ray.instID[l][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            *instPrimID(l, ofs) = ray.instPrimID[l][k];
+#endif
+          }
 #endif
         }
 #endif
@@ -792,8 +862,12 @@ namespace embree
       primID = (unsigned int*)&t.primID;
       geomID = (unsigned int*)&t.geomID;
 
-      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l)
+      for (unsigned l = 0; l < RTC_MAX_INSTANCE_LEVEL_COUNT; ++l) {
         instID[l] = (unsigned int*)&t.instID[l];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        instPrimID[l] = (unsigned int*)&t.instPrimID[l];
+#endif
+      }
     }
 
     __forceinline Ray getRayByOffset(size_t offset)
@@ -859,9 +933,16 @@ namespace embree
 
         if (likely(instID[0])) {
           *(unsigned int* __restrict__)((char*)instID[0] + offset) = ray.instID[0];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          *(unsigned int* __restrict__)((char*)instPrimID[0] + offset) = ray.instPrimID[0];
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID; ++l)
+          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID; ++l) {
             *(unsigned int* __restrict__)((char*)instID[l] + offset) = ray.instID[l];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            *(unsigned int* __restrict__)((char*)instPrimID[l] + offset) = ray.instPrimID[l];
+#endif
+          }
 #endif
         }
       }
@@ -892,9 +973,16 @@ namespace embree
 
         if (likely(instID[0])) {
           vuint<K>::storeu(valid, (unsigned int* __restrict__)((char*)instID[0] + offset), ray.instID[0]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          vuint<K>::storeu(valid, (unsigned int* __restrict__)((char*)instPrimID[0] + offset), ray.instPrimID[0]);
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l)
+          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l) {
             vuint<K>::storeu(valid, (unsigned int* __restrict__)((char*)instID[l] + offset), ray.instID[l]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            vuint<K>::storeu(valid, (unsigned int* __restrict__)((char*)instPrimID[l] + offset), ray.instPrimID[l]);
+#endif
+          }
 #endif
         }
       }
@@ -1008,9 +1096,16 @@ namespace embree
 
         if (likely(instID[0])) {
           vuint<K>::template scatter<1>(valid, (unsigned int*)instID[0], offset, ray.instID[0]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          vuint<K>::template scatter<1>(valid, (unsigned int*)instPrimID[0], offset, ray.instPrimID[0]);
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l)
+          for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l) {
             vuint<K>::template scatter<1>(valid, (unsigned int*)instID[l], offset, ray.instID[l]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            vuint<K>::template scatter<1>(valid, (unsigned int*)instPrimID[l], offset, ray.instPrimID[l]);
+#endif
+          }
 #endif
         }
 #else
@@ -1032,9 +1127,16 @@ namespace embree
 
           if (likely(instID[0])) {
             *(unsigned int* __restrict__)((char*)instID[0] + ofs) = ray.instID[0][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+            *(unsigned int* __restrict__)((char*)instPrimID[0] + ofs) = ray.instPrimID[0][k];
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-            for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l)
+            for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && ray.instID[l-1][k] != RTC_INVALID_GEOMETRY_ID; ++l) {
               *(unsigned int* __restrict__)((char*)instID[l] + ofs) = ray.instID[l][k];
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+              *(unsigned int* __restrict__)((char*)instPrimID[l] + ofs) = ray.instPrimID[l][k];
+#endif
+            }
 #endif
           }
         }
@@ -1091,7 +1193,10 @@ namespace embree
 
     unsigned int* __restrict__ primID; // primitive ID
     unsigned int* __restrict__ geomID; // geometry ID
-    unsigned int* __restrict__ instID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance ID (optional)
+    unsigned int* __restrict__ instID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance ID
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+    unsigned int* __restrict__ instPrimID[RTC_MAX_INSTANCE_LEVEL_COUNT]; // instance primitive ID (optional)
+#endif
   };
 
 
@@ -1134,9 +1239,16 @@ namespace embree
         vuint<K>::template scatter<1>(valid, (unsigned int*)&((RayHit*)ptr)->geomID, offset, ray.geomID);
 
         vuint<K>::template scatter<1>(valid, (unsigned int*)&((RayHit*)ptr)->instID[0], offset, ray.instID[0]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+        vuint<K>::template scatter<1>(valid, (unsigned int*)&((RayHit*)ptr)->instPrimID[0], offset, ray.instPrimID[0]);
+#endif
 #if (RTC_MAX_INSTANCE_LEVEL_COUNT > 1)
-        for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l)
+        for (unsigned l = 1; l < RTC_MAX_INSTANCE_LEVEL_COUNT && any(valid & (ray.instID[l-1] != RTC_INVALID_GEOMETRY_ID)); ++l) {
           vuint<K>::template scatter<1>(valid, (unsigned int*)&((RayHit*)ptr)->instID[l], offset, ray.instID[l]);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          vuint<K>::template scatter<1>(valid, (unsigned int*)&((RayHit*)ptr)->instPrimID[l], offset, ray.instPrimID[l]);
+#endif
+        }
 #endif
 #else
         size_t valid_bits = movemask(valid);
@@ -1154,6 +1266,9 @@ namespace embree
           ray_k->geomID = ray.geomID[k];
 
           instance_id_stack::copy_VU<K>(ray.instID, ray_k->instID, k);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          instance_id_stack::copy_VU<K>(ray.instPrimID, ray_k->instPrimID, k);
+#endif
         }
 #endif
       }
@@ -1183,7 +1298,7 @@ namespace embree
 
     Ray* __restrict__ ptr;
   };
-
+  
   template<>
   __forceinline Ray4 RayStreamAOS::getRayByOffset<4>(const vint4& offset)
   {
@@ -1219,7 +1334,7 @@ namespace embree
 
     return ray;
   }
-
+  
 #if defined(__AVX__)
   template<>
   __forceinline Ray8 RayStreamAOS::getRayByOffset<8>(const vint8& offset)
@@ -1358,6 +1473,9 @@ namespace embree
           ray_k->primID = ray.primID[k];
           ray_k->geomID = ray.geomID[k];
           instance_id_stack::copy_VU<K>(ray.instID, ray_k->instID, k);
+#if defined(RTC_GEOMETRY_INSTANCE_ARRAY)
+          instance_id_stack::copy_VU<K>(ray.instPrimID, ray_k->instPrimID, k);
+#endif
         }
       }
     }
@@ -1383,7 +1501,7 @@ namespace embree
 
     Ray** __restrict__ ptr;
   };
-
+  
   template<>
   __forceinline Ray4 RayStreamAOP::getRayByIndex<4>(const vint4& index)
   {
@@ -1419,7 +1537,7 @@ namespace embree
 
     return ray;
   }
-
+  
 #if defined(__AVX__)
   template<>
   __forceinline Ray8 RayStreamAOP::getRayByIndex<8>(const vint8& index)
diff --git a/thirdparty/embree/kernels/common/rtcore.cpp b/thirdparty/embree/kernels/common/rtcore.cpp
index a6ea55bfc4..eb8d2c0a58 100644
--- a/thirdparty/embree/kernels/common/rtcore.cpp
+++ b/thirdparty/embree/kernels/common/rtcore.cpp
@@ -8,11 +8,14 @@
 #include "scene.h"
 #include "context.h"
 #include "../geometry/filter.h"
-#include "../../include/embree3/rtcore_ray.h"
+#include "../../include/embree4/rtcore_ray.h"
 using namespace embree;
 
 RTC_NAMESPACE_BEGIN;
 
+#define RTC_ENTER_DEVICE(arg) \
+  DeviceEnterLeave enterleave(arg);
+
   /* mutex to make API thread safe */
   static MutexSys g_mutex;
 
@@ -27,6 +30,57 @@ RTC_NAMESPACE_BEGIN;
     return (RTCDevice) nullptr;
   }
 
+#if defined(EMBREE_SYCL_SUPPORT)
+
+  RTC_API RTCDevice rtcNewSYCLDeviceInternal(sycl::context sycl_context, const char* config)
+  {
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcNewSYCLDevice);
+    Lock<MutexSys> lock(g_mutex);
+
+    DeviceGPU* device = new DeviceGPU(sycl_context,config);
+    return (RTCDevice) device->refInc();
+    RTC_CATCH_END(nullptr);
+    return (RTCDevice) nullptr;
+  }
+
+  RTC_API bool rtcIsSYCLDeviceSupported(const sycl::device device)
+  {
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcIsSYCLDeviceSupported);
+    return rthwifIsSYCLDeviceSupported(device) > 0;
+    RTC_CATCH_END(nullptr);
+    return false;
+  }
+
+  RTC_API int rtcSYCLDeviceSelector(const sycl::device device)
+  {
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcSYCLDeviceSelector);
+    return rthwifIsSYCLDeviceSupported(device);
+    RTC_CATCH_END(nullptr);
+    return -1;
+  }
+
+  RTC_API void rtcSetDeviceSYCLDevice(RTCDevice hdevice, const sycl::device sycl_device)
+  {
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcSetDeviceSYCLDevice);
+    RTC_VERIFY_HANDLE(hdevice);
+
+    Lock<MutexSys> lock(g_mutex);
+    
+    DeviceGPU* device = dynamic_cast<DeviceGPU*>((Device*) hdevice);
+    if (device == nullptr)
+      throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "passed device must be an Embree SYCL device")
+      
+    device->setSYCLDevice(sycl_device);
+    
+    RTC_CATCH_END(nullptr);
+  }
+
+#endif
+
   RTC_API void rtcRetainDevice(RTCDevice hdevice) 
   {
     Device* device = (Device*) hdevice;
@@ -108,6 +162,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcNewBuffer);
     RTC_VERIFY_HANDLE(hdevice);
+    RTC_ENTER_DEVICE(hdevice);
     Buffer* buffer = new Buffer((Device*)hdevice, byteSize);
     return (RTCBuffer)buffer->refInc();
     RTC_CATCH_END((Device*)hdevice);
@@ -119,6 +174,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcNewSharedBuffer);
     RTC_VERIFY_HANDLE(hdevice);
+    RTC_ENTER_DEVICE(hdevice);
     Buffer* buffer = new Buffer((Device*)hdevice, byteSize, ptr);
     return (RTCBuffer)buffer->refInc();
     RTC_CATCH_END((Device*)hdevice);
@@ -131,6 +187,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetBufferData);
     RTC_VERIFY_HANDLE(hbuffer);
+    RTC_ENTER_DEVICE(hbuffer);
     return buffer->data();
     RTC_CATCH_END2(buffer);
     return nullptr;
@@ -142,6 +199,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcRetainBuffer);
     RTC_VERIFY_HANDLE(hbuffer);
+    RTC_ENTER_DEVICE(hbuffer);
     buffer->refInc();
     RTC_CATCH_END2(buffer);
   }
@@ -152,6 +210,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcReleaseBuffer);
     RTC_VERIFY_HANDLE(hbuffer);
+    RTC_ENTER_DEVICE(hbuffer);
     buffer->refDec();
     RTC_CATCH_END2(buffer);
   }
@@ -161,6 +220,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcNewScene);
     RTC_VERIFY_HANDLE(hdevice);
+    RTC_ENTER_DEVICE(hdevice);
     Scene* scene = new Scene((Device*)hdevice);
     return (RTCScene) scene->refInc();
     RTC_CATCH_END((Device*)hdevice);
@@ -184,6 +244,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetSceneProgressMonitorFunction);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     Lock<MutexSys> lock(g_mutex);
     scene->setProgressMonitorFunction(progress,ptr);
     RTC_CATCH_END2(scene);
@@ -195,13 +256,18 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetSceneBuildQuality);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
+    // -- GODOT start --
+    // if (quality != RTC_BUILD_QUALITY_LOW &&
+    //     quality != RTC_BUILD_QUALITY_MEDIUM &&
+    //     quality != RTC_BUILD_QUALITY_HIGH)
+    //   throw std::runtime_error("invalid build quality");
     if (quality != RTC_BUILD_QUALITY_LOW &&
         quality != RTC_BUILD_QUALITY_MEDIUM &&
-        quality != RTC_BUILD_QUALITY_HIGH)
-      // -- GODOT start --
-      // throw std::runtime_error("invalid build quality");
+        quality != RTC_BUILD_QUALITY_HIGH) {
       abort();
-      // -- GODOT end --
+    }
+    // -- GODOT end --
     scene->setBuildQuality(quality);
     RTC_CATCH_END2(scene);
   }
@@ -212,6 +278,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetSceneFlags);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     scene->setSceneFlags(flags);
     RTC_CATCH_END2(scene);
   }
@@ -222,6 +289,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetSceneFlags);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     return scene->getSceneFlags();
     RTC_CATCH_END2(scene);
     return RTC_SCENE_FLAG_NONE;
@@ -233,6 +301,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcCommitScene);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     scene->commit(false);
     RTC_CATCH_END2(scene);
   }
@@ -243,6 +312,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcJoinCommitScene);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     scene->commit(true);
     RTC_CATCH_END2(scene);
   }
@@ -253,6 +323,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetSceneBounds);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
     BBox3fa bounds = scene->bounds.bounds();
     bounds_o->lower_x = bounds.lower.x;
@@ -272,6 +343,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetSceneBounds);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     if (bounds_o == nullptr)
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"invalid destination pointer");
     if (scene->isModified())
@@ -447,7 +519,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_END2_FALSE(scene);
   }
 
-  RTC_API void rtcIntersect1 (RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit* rayhit) 
+  RTC_API void rtcIntersect1 (RTCScene hscene, RTCRayHit* rayhit, RTCIntersectArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -458,7 +530,21 @@ RTC_NAMESPACE_BEGIN;
     if (((size_t)rayhit) & 0x0F) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 16 bytes");   
 #endif
     STAT3(normal.travs,1,1,1);
-    IntersectContext context(scene,user_context);
+
+    RTCIntersectArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitIntersectArguments(&defaultArgs);
+      args = &defaultArgs;
+    }
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+    
     scene->intersectors.intersect(*rayhit,&context);
 #if defined(DEBUG)
     ((RayHit*)rayhit)->verifyHit();
@@ -466,7 +552,45 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_END2(scene);
   }
 
-  RTC_API void rtcIntersect4 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit4* rayhit) 
+  RTC_API void rtcForwardIntersect1 (const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay* iray_, unsigned int instID)
+  {
+    rtcForwardIntersect1Ex(args, hscene, iray_, instID, 0);
+  }
+
+  RTC_API void rtcForwardIntersect1Ex(const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay* iray_, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcForwardIntersect1Ex);
+#if defined(DEBUG)
+    RTC_VERIFY_HANDLE(hscene);
+    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
+    if (((size_t)iray_) & 0x0F) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 16 bytes");
+#endif
+
+    Ray* iray = (Ray*) iray_;
+    RayHit* oray = (RayHit*)args->rayhit;
+    RTCRayQueryContext* user_context = args->context;
+    const Vec3ff ray_org_tnear = oray->org;
+    const Vec3ff ray_dir_time = oray->dir;
+    oray->org = iray->org;
+    oray->dir = iray->dir;
+    STAT3(normal.travs,1,1,1);
+
+    RTCIntersectArguments* iargs = ((IntersectFunctionNArguments*) args)->args;
+    RayQueryContext context(scene,user_context,iargs);
+
+    instance_id_stack::push(user_context, instID, instPrimID);
+    scene->intersectors.intersect(*(RTCRayHit*)oray,&context);
+    instance_id_stack::pop(user_context);
+
+    oray->org = ray_org_tnear;
+    oray->dir = ray_dir_time;
+
+    RTC_CATCH_END2(scene);
+  }
+
+  RTC_API void rtcIntersect4 (const int* valid, RTCScene hscene, RTCRayHit4* rayhit, RTCIntersectArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -481,23 +605,119 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<4; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(normal.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit4* ray4 = (RayHit4*) rayhit;
-    for (size_t i=0; i<4; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray4->get(i,ray1);
-      scene->intersectors.intersect((RTCRayHit&)ray1,&context);
-      ray4->set(i,ray1);
+    RTCIntersectArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitIntersectArguments(&defaultArgs);
+      args = &defaultArgs;
+    }
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+
+    if (likely(scene->intersectors.intersector4))
+      scene->intersectors.intersect4(valid,*rayhit,&context);
+
+    else {
+      RayHit4* ray4 = (RayHit4*) rayhit;
+      for (size_t i=0; i<4; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray4->get(i,ray1);
+        scene->intersectors.intersect((RTCRayHit&)ray1,&context);
+        ray4->set(i,ray1);
+      }
     }
-#else
-    scene->intersectors.intersect4(valid,*rayhit,&context);
-#endif
     
     RTC_CATCH_END2(scene);
   }
+
+  template<int N> void copy(float* dst, float* src);
+
+  template<>
+  __forceinline void copy<4>(float* dst, float* src) {
+    vfloat4::storeu(&dst[0],vfloat4::loadu(&src[0]));
+  }
+
+  template<>
+  __forceinline void copy<8>(float* dst, float* src) {
+    vfloat4::storeu(&dst[0],vfloat4::loadu(&src[0]));
+    vfloat4::storeu(&dst[4],vfloat4::loadu(&src[4]));
+  }
+
+  template<>
+  __forceinline void copy<16>(float* dst, float* src) {
+    vfloat4::storeu(&dst[0],vfloat4::loadu(&src[0]));
+    vfloat4::storeu(&dst[4],vfloat4::loadu(&src[4]));
+    vfloat4::storeu(&dst[8],vfloat4::loadu(&src[8]));
+    vfloat4::storeu(&dst[12],vfloat4::loadu(&src[12]));
+  }
+
+  template<typename RTCRay, typename RTCRayHit, int N>
+  __forceinline void rtcForwardIntersectN(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTCRayHit* oray = (RTCRayHit*)args->rayhit;
+    RTCRayQueryContext* user_context = args->context;
+
+    __aligned(16) float ray_org_x[N];
+    __aligned(16) float ray_org_y[N];
+    __aligned(16) float ray_org_z[N];
+    __aligned(16) float ray_dir_x[N];
+    __aligned(16) float ray_dir_y[N];
+    __aligned(16) float ray_dir_z[N];
+    
+    copy<N>(ray_org_x,oray->ray.org_x);
+    copy<N>(ray_org_y,oray->ray.org_y);
+    copy<N>(ray_org_z,oray->ray.org_z);
+    copy<N>(ray_dir_x,oray->ray.dir_x);
+    copy<N>(ray_dir_y,oray->ray.dir_y);
+    copy<N>(ray_dir_z,oray->ray.dir_z);
+    
+    copy<N>(oray->ray.org_x,iray->org_x);
+    copy<N>(oray->ray.org_y,iray->org_y);
+    copy<N>(oray->ray.org_z,iray->org_z);
+    copy<N>(oray->ray.dir_x,iray->dir_x);
+    copy<N>(oray->ray.dir_y,iray->dir_y);
+    copy<N>(oray->ray.dir_z,iray->dir_z);
+    
+    STAT(size_t cnt=0; for (size_t i=0; i<N; i++) cnt += ((int*)valid)[i] == -1;);
+    STAT3(normal.travs,cnt,cnt,cnt);
+
+    RTCIntersectArguments* iargs = ((IntersectFunctionNArguments*) args)->args;
+    RayQueryContext context(scene,user_context,iargs);
+
+    instance_id_stack::push(user_context, instID, instPrimID);
+    scene->intersectors.intersect(valid,*oray,&context);
+    instance_id_stack::pop(user_context);
+
+    copy<N>(oray->ray.org_x,ray_org_x);
+    copy<N>(oray->ray.org_y,ray_org_y);
+    copy<N>(oray->ray.org_z,ray_org_z);
+    copy<N>(oray->ray.dir_x,ray_dir_x);
+    copy<N>(oray->ray.dir_y,ray_dir_y);
+    copy<N>(oray->ray.dir_z,ray_dir_z);
+  }
+
+  RTC_API void rtcForwardIntersect4(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay4* iray, unsigned int instID)
+  {
+    RTC_TRACE(rtcForwardIntersect4);
+    return rtcForwardIntersect4Ex(valid, args, hscene, iray, instID, 0);
+  }
+
+  RTC_API void rtcForwardIntersect4Ex(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay4* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcForwardIntersect4);
+    rtcForwardIntersectN<RTCRay4,RTCRayHit4,4>(valid,args,hscene,iray,instID,instPrimID);
+    RTC_CATCH_END2(scene);
+  }
   
-  RTC_API void rtcIntersect8 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit8* rayhit) 
+  RTC_API void rtcIntersect8 (const int* valid, RTCScene hscene, RTCRayHit8* rayhit, RTCIntersectArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -512,25 +732,53 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<8; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(normal.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit8* ray8 = (RayHit8*) rayhit;
-    for (size_t i=0; i<8; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray8->get(i,ray1);
-      scene->intersectors.intersect((RTCRayHit&)ray1,&context);
-      ray8->set(i,ray1);
+    RTCIntersectArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitIntersectArguments(&defaultArgs);
+      args = &defaultArgs;
     }
-#else
-    if (likely(scene->intersectors.intersector8))
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+    
+    if (likely(scene->intersectors.intersector8)) 
       scene->intersectors.intersect8(valid,*rayhit,&context);
+    
     else
-      scene->device->rayStreamFilters.intersectSOA(scene,(char*)rayhit,8,1,sizeof(RTCRayHit8),&context);
-#endif
+    {
+      RayHit8* ray8 = (RayHit8*) rayhit;
+      for (size_t i=0; i<8; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray8->get(i,ray1);
+        scene->intersectors.intersect((RTCRayHit&)ray1,&context);
+        ray8->set(i,ray1);
+      }
+    }
+    
     RTC_CATCH_END2(scene);
   }
-  
-  RTC_API void rtcIntersect16 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit16* rayhit) 
+
+  RTC_API void rtcForwardIntersect8(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay8* iray, unsigned int instID)
+  {
+    RTC_TRACE(rtcForwardIntersect8);
+    return rtcForwardIntersect8Ex(valid, args, hscene, iray, instID, 0);
+  }
+
+  RTC_API void rtcForwardIntersect8Ex(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay8* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcForwardIntersect8Ex);
+    rtcForwardIntersectN<RTCRay8,RTCRayHit8,8>(valid,args,hscene,iray,instID,instPrimID);
+    RTC_CATCH_END2(scene);
+  }
+
+  RTC_API void rtcIntersect16 (const int* valid, RTCScene hscene, RTCRayHit16* rayhit, RTCIntersectArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -545,179 +793,121 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<16; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(normal.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit16* ray16 = (RayHit16*) rayhit;
-    for (size_t i=0; i<16; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray16->get(i,ray1);
-      scene->intersectors.intersect((RTCRayHit&)ray1,&context);
-      ray16->set(i,ray1);
+    RTCIntersectArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitIntersectArguments(&defaultArgs);
+      args = &defaultArgs;
     }
-#else
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+
     if (likely(scene->intersectors.intersector16))
       scene->intersectors.intersect16(valid,*rayhit,&context);
-    else
-      scene->device->rayStreamFilters.intersectSOA(scene,(char*)rayhit,16,1,sizeof(RTCRayHit16),&context);
-#endif
-    RTC_CATCH_END2(scene);
-  }
-
-  RTC_API void rtcIntersect1M (RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit* rayhit, unsigned int M, size_t byteStride) 
-  {
-    Scene* scene = (Scene*) hscene;
-    RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcIntersect1M);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)rayhit ) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
-#endif
-    STAT3(normal.travs,M,M,M);
-    IntersectContext context(scene,user_context);
 
-    /* fast codepath for single rays */
-    if (likely(M == 1)) {
-      if (likely(rayhit->ray.tnear <= rayhit->ray.tfar)) 
-        scene->intersectors.intersect(*rayhit,&context);
-    } 
-
-    /* codepath for streams */
     else {
-      scene->device->rayStreamFilters.intersectAOS(scene,rayhit,M,byteStride,&context);   
+      RayHit16* ray16 = (RayHit16*) rayhit;
+      for (size_t i=0; i<16; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray16->get(i,ray1);
+        scene->intersectors.intersect((RTCRayHit&)ray1,&context);
+        ray16->set(i,ray1);
+      }
     }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcIntersect1M not supported");
-#endif
+
     RTC_CATCH_END2(scene);
   }
 
-  RTC_API void rtcIntersect1Mp (RTCScene hscene, RTCIntersectContext* user_context, RTCRayHit** rn, unsigned int M) 
+  RTC_API void rtcForwardIntersect16(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay16* iray, unsigned int instID)
+  {
+    RTC_TRACE(rtcForwardIntersect16);
+    return rtcForwardIntersect16Ex(valid, args, hscene, iray, instID, 0);
+  }
+
+  RTC_API void rtcForwardIntersect16Ex(const int* valid, const RTCIntersectFunctionNArguments* args, RTCScene hscene, RTCRay16* iray, unsigned int instID, unsigned int instPrimID)
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcIntersect1Mp);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)rn) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
-#endif
-    STAT3(normal.travs,M,M,M);
-    IntersectContext context(scene,user_context);
-
-    /* fast codepath for single rays */
-    if (likely(M == 1)) {
-      if (likely(rn[0]->ray.tnear <= rn[0]->ray.tfar)) 
-        scene->intersectors.intersect(*rn[0],&context);
-    } 
-
-    /* codepath for streams */
-    else {
-      scene->device->rayStreamFilters.intersectAOP(scene,rn,M,&context);
-    }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcIntersect1Mp not supported");
-#endif
+    RTC_TRACE(rtcForwardIntersect16Ex);
+    rtcForwardIntersectN<RTCRay16,RTCRayHit16,16>(valid,args,hscene,iray,instID,instPrimID);
     RTC_CATCH_END2(scene);
   }
 
-  RTC_API void rtcIntersectNM (RTCScene hscene, RTCIntersectContext* user_context, struct RTCRayHitN* rayhit, unsigned int N, unsigned int M, size_t byteStride) 
+  RTC_API void rtcOccluded1 (RTCScene hscene, RTCRay* ray, RTCOccludedArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcIntersectNM);
-
-#if defined (EMBREE_RAY_PACKETS)
+    RTC_TRACE(rtcOccluded1);
+    STAT3(shadow.travs,1,1,1);
 #if defined(DEBUG)
     RTC_VERIFY_HANDLE(hscene);
     if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)rayhit) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
+    if (((size_t)ray) & 0x0F) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 16 bytes");   
 #endif
-    STAT3(normal.travs,N*M,N*M,N*M);
-    IntersectContext context(scene,user_context);
 
-    /* code path for single ray streams */
-    if (likely(N == 1))
-    {
-      /* fast code path for streams of size 1 */
-      if (likely(M == 1)) {
-        if (likely(((RTCRayHit*)rayhit)->ray.tnear <= ((RTCRayHit*)rayhit)->ray.tfar))
-          scene->intersectors.intersect(*(RTCRayHit*)rayhit,&context);
-      } 
-      /* normal codepath for single ray streams */
-      else {
-        scene->device->rayStreamFilters.intersectAOS(scene,(RTCRayHit*)rayhit,M,byteStride,&context);
-      }
+    RTCOccludedArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitOccludedArguments(&defaultArgs);
+      args = &defaultArgs;
     }
-    /* code path for ray packet streams */
-    else {
-      scene->device->rayStreamFilters.intersectSOA(scene,(char*)rayhit,N,M,byteStride,&context);
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
     }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcIntersectNM not supported");
-#endif
+    RayQueryContext context(scene,user_context,args);
+    
+    scene->intersectors.occluded(*ray,&context);
     RTC_CATCH_END2(scene);
   }
 
-  RTC_API void rtcIntersectNp (RTCScene hscene, RTCIntersectContext* user_context, const RTCRayHitNp* rayhit, unsigned int N) 
+  RTC_API void rtcForwardOccluded1 (const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay* iray_, unsigned int instID)
   {
-    Scene* scene = (Scene*) hscene;
-    RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcIntersectNp);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)rayhit->ray.org_x ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.org_x not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.org_y ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.org_y not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.org_z ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.org_z not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.dir_x ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.dir_x not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.dir_y ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.dir_y not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.dir_z ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.dir_z not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.tnear ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.dir_x not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.tfar  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.tnear not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.time  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.time not aligned to 4 bytes");   
-    if (((size_t)rayhit->ray.mask  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->ray.mask not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.Ng_x  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.Ng_x not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.Ng_y  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.Ng_y not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.Ng_z  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.Ng_z not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.u     ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.u not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.v     ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.v not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.geomID) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.geomID not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.primID) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.primID not aligned to 4 bytes");   
-    if (((size_t)rayhit->hit.instID) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "rayhit->hit.instID not aligned to 4 bytes");   
-#endif
-    STAT3(normal.travs,N,N,N);
-    IntersectContext context(scene,user_context);
-    scene->device->rayStreamFilters.intersectSOP(scene,rayhit,N,&context);
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcIntersectNp not supported");
-#endif
-    RTC_CATCH_END2(scene);
+    RTC_TRACE(rtcForwardOccluded1);
+    return rtcForwardOccluded1Ex(args, hscene, iray_, instID, 0);
   }
-  
-  RTC_API void rtcOccluded1 (RTCScene hscene, RTCIntersectContext* user_context, RTCRay* ray) 
+
+  RTC_API void rtcForwardOccluded1Ex(const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay* iray_, unsigned int instID, unsigned int instPrimID)
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcOccluded1);
+    RTC_TRACE(rtcForwardOccluded1Ex);
     STAT3(shadow.travs,1,1,1);
 #if defined(DEBUG)
     RTC_VERIFY_HANDLE(hscene);
     if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)ray) & 0x0F) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 16 bytes");   
+    if (((size_t)iray_) & 0x0F) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 16 bytes");   
 #endif
-    IntersectContext context(scene,user_context);
-    scene->intersectors.occluded(*ray,&context);
+    
+    Ray* iray = (Ray*)iray_;
+    Ray* oray = (Ray*)args->ray;
+    RTCRayQueryContext* user_context = args->context;
+    const Vec3ff ray_org_tnear = oray->org;
+    const Vec3ff ray_dir_time = oray->dir;
+    oray->org = iray->org;
+    oray->dir = iray->dir;
+
+    RTCIntersectArguments* iargs = ((OccludedFunctionNArguments*) args)->args;
+    RayQueryContext context(scene,user_context,iargs);
+
+    instance_id_stack::push(user_context, instID, instPrimID);
+    scene->intersectors.occluded(*(RTCRay*)oray,&context);
+    instance_id_stack::pop(user_context);
+    
+    oray->org = ray_org_tnear;
+    oray->dir = ray_dir_time;
+
     RTC_CATCH_END2(scene);
   }
-  
-  RTC_API void rtcOccluded4 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRay4* ray) 
+
+  RTC_API void rtcOccluded4 (const int* valid, RTCScene hscene, RTCRay4* ray, RTCOccludedArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -732,23 +922,98 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<4; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(shadow.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit4* ray4 = (RayHit4*) ray;
-    for (size_t i=0; i<4; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray4->get(i,ray1);
-      scene->intersectors.occluded((RTCRay&)ray1,&context);
-      ray4->geomID[i] = ray1.geomID; 
+    RTCOccludedArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitOccludedArguments(&defaultArgs);
+      args = &defaultArgs;
+    }
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+
+    if (likely(scene->intersectors.intersector4))
+       scene->intersectors.occluded4(valid,*ray,&context);
+
+    else {
+      RayHit4* ray4 = (RayHit4*) ray;
+      for (size_t i=0; i<4; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray4->get(i,ray1);
+        scene->intersectors.occluded((RTCRay&)ray1,&context);
+        ray4->geomID[i] = ray1.geomID; 
+      }
     }
-#else
-    scene->intersectors.occluded4(valid,*ray,&context);
-#endif
     
     RTC_CATCH_END2(scene);
   }
+
+  template<typename RTCRay, int N>
+  __forceinline void rtcForwardOccludedN (const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTCRay* oray = (RTCRay*)args->ray;
+    RTCRayQueryContext* user_context = args->context;
+
+    __aligned(16) float ray_org_x[N];
+    __aligned(16) float ray_org_y[N];
+    __aligned(16) float ray_org_z[N];
+    __aligned(16) float ray_dir_x[N];
+    __aligned(16) float ray_dir_y[N];
+    __aligned(16) float ray_dir_z[N];
+    
+    copy<N>(ray_org_x,oray->org_x);
+    copy<N>(ray_org_y,oray->org_y);
+    copy<N>(ray_org_z,oray->org_z);
+    copy<N>(ray_dir_x,oray->dir_x);
+    copy<N>(ray_dir_y,oray->dir_y);
+    copy<N>(ray_dir_z,oray->dir_z);
+    
+    copy<N>(oray->org_x,iray->org_x);
+    copy<N>(oray->org_y,iray->org_y);
+    copy<N>(oray->org_z,iray->org_z);
+    copy<N>(oray->dir_x,iray->dir_x);
+    copy<N>(oray->dir_y,iray->dir_y);
+    copy<N>(oray->dir_z,iray->dir_z);
+    
+    STAT(size_t cnt=0; for (size_t i=0; i<N; i++) cnt += ((int*)valid)[i] == -1;);
+    STAT3(normal.travs,cnt,cnt,cnt);
+
+    RTCIntersectArguments* iargs = ((IntersectFunctionNArguments*) args)->args;
+    RayQueryContext context(scene,user_context,iargs);
+
+    instance_id_stack::push(user_context, instID, instPrimID);
+    scene->intersectors.occluded(valid,*oray,&context);
+    instance_id_stack::pop(user_context);
+
+    copy<N>(oray->org_x,ray_org_x);
+    copy<N>(oray->org_y,ray_org_y);
+    copy<N>(oray->org_z,ray_org_z);
+    copy<N>(oray->dir_x,ray_dir_x);
+    copy<N>(oray->dir_y,ray_dir_y);
+    copy<N>(oray->dir_z,ray_dir_z);
+  }
+
+  RTC_API void rtcForwardOccluded4(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay4* iray, unsigned int instID)
+  {
+    RTC_TRACE(rtcForwardOccluded4);
+    return rtcForwardOccluded4Ex(valid, args, hscene, iray, instID, 0);
+  }
+
+  RTC_API void rtcForwardOccluded4Ex(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay4* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcForwardOccluded4);
+    rtcForwardOccludedN<RTCRay4,4>(valid,args,hscene,iray,instID,instPrimID);
+    RTC_CATCH_END2(scene);
+  }
  
-  RTC_API void rtcOccluded8 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRay8* ray) 
+  RTC_API void rtcOccluded8 (const int* valid, RTCScene hscene, RTCRay8* ray, RTCOccludedArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -763,26 +1028,52 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<8; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(shadow.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit8* ray8 = (RayHit8*) ray;
-    for (size_t i=0; i<8; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray8->get(i,ray1);
-      scene->intersectors.occluded((RTCRay&)ray1,&context);
-      ray8->set(i,ray1);
+    RTCOccludedArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitOccludedArguments(&defaultArgs);
+      args = &defaultArgs;
     }
-#else
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+
     if (likely(scene->intersectors.intersector8))
       scene->intersectors.occluded8(valid,*ray,&context);
-    else
-      scene->device->rayStreamFilters.occludedSOA(scene,(char*)ray,8,1,sizeof(RTCRay8),&context);
-#endif
+
+    else {
+      RayHit8* ray8 = (RayHit8*) ray;
+      for (size_t i=0; i<8; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray8->get(i,ray1);
+        scene->intersectors.occluded((RTCRay&)ray1,&context);
+        ray8->set(i,ray1);
+      }
+    }
 
     RTC_CATCH_END2(scene);
   }
-  
-  RTC_API void rtcOccluded16 (const int* valid, RTCScene hscene, RTCIntersectContext* user_context, RTCRay16* ray) 
+
+  RTC_API void rtcForwardOccluded8(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay8* iray, unsigned int instID)
+  {
+    RTC_TRACE(rtcForwardOccluded8);
+    return rtcForwardOccluded8Ex(valid, args, hscene, iray, instID, 0);
+  }
+
+  RTC_API void rtcForwardOccluded8Ex(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay8* iray, unsigned int instID, unsigned int instPrimID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcForwardOccluded8Ex);
+    rtcForwardOccludedN<RTCRay8,8>(valid, args, hscene, iray, instID, instPrimID);
+    RTC_CATCH_END2(scene);
+  }
+   
+  RTC_API void rtcOccluded16 (const int* valid, RTCScene hscene, RTCRay16* ray, RTCOccludedArguments* args) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
@@ -797,159 +1088,58 @@ RTC_NAMESPACE_BEGIN;
     STAT(size_t cnt=0; for (size_t i=0; i<16; i++) cnt += ((int*)valid)[i] == -1;);
     STAT3(shadow.travs,cnt,cnt,cnt);
 
-    IntersectContext context(scene,user_context);
-#if !defined(EMBREE_RAY_PACKETS)
-    RayHit16* ray16 = (RayHit16*) ray;
-    for (size_t i=0; i<16; i++) {
-      if (!valid[i]) continue;
-      RayHit ray1; ray16->get(i,ray1);
-      scene->intersectors.occluded((RTCRay&)ray1,&context);
-      ray16->set(i,ray1);
+    RTCOccludedArguments defaultArgs;
+    if (unlikely(args == nullptr)) {
+      rtcInitOccludedArguments(&defaultArgs);
+      args = &defaultArgs;
     }
-#else
+    RTCRayQueryContext* user_context = args->context;
+    
+    RTCRayQueryContext defaultContext;
+    if (unlikely(user_context == nullptr)) {
+      rtcInitRayQueryContext(&defaultContext);
+      user_context = &defaultContext;
+    }
+    RayQueryContext context(scene,user_context,args);
+
     if (likely(scene->intersectors.intersector16))
       scene->intersectors.occluded16(valid,*ray,&context);
-    else
-      scene->device->rayStreamFilters.occludedSOA(scene,(char*)ray,16,1,sizeof(RTCRay16),&context);
-#endif
-
-    RTC_CATCH_END2(scene);
-  }
-  
-  RTC_API void rtcOccluded1M(RTCScene hscene, RTCIntersectContext* user_context, RTCRay* ray, unsigned int M, size_t byteStride) 
-  {
-    Scene* scene = (Scene*) hscene;
-    RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcOccluded1M);
 
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)ray) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
-#endif
-    STAT3(shadow.travs,M,M,M);
-    IntersectContext context(scene,user_context);
-    /* fast codepath for streams of size 1 */
-    if (likely(M == 1)) {
-      if (likely(ray->tnear <= ray->tfar)) 
-        scene->intersectors.occluded (*ray,&context);
-    } 
-    /* codepath for normal streams */
     else {
-      scene->device->rayStreamFilters.occludedAOS(scene,ray,M,byteStride,&context);
+      RayHit16* ray16 = (RayHit16*) ray;
+      for (size_t i=0; i<16; i++) {
+        if (!valid[i]) continue;
+        RayHit ray1; ray16->get(i,ray1);
+        scene->intersectors.occluded((RTCRay&)ray1,&context);
+        ray16->set(i,ray1);
+      }
     }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcOccluded1M not supported");
-#endif
-    RTC_CATCH_END2(scene);
-  }
 
-  RTC_API void rtcOccluded1Mp(RTCScene hscene, RTCIntersectContext* user_context, RTCRay** ray, unsigned int M) 
-  {
-    Scene* scene = (Scene*) hscene;
-    RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcOccluded1Mp);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)ray) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
-#endif
-    STAT3(shadow.travs,M,M,M);
-    IntersectContext context(scene,user_context);
-
-    /* fast codepath for streams of size 1 */
-    if (likely(M == 1)) {
-      if (likely(ray[0]->tnear <= ray[0]->tfar)) 
-        scene->intersectors.occluded (*ray[0],&context);
-    } 
-    /* codepath for normal streams */
-    else {
-      scene->device->rayStreamFilters.occludedAOP(scene,ray,M,&context);
-    }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcOccluded1Mp not supported");
-#endif
     RTC_CATCH_END2(scene);
   }
 
-  RTC_API void rtcOccludedNM(RTCScene hscene, RTCIntersectContext* user_context, RTCRayN* ray, unsigned int N, unsigned int M, size_t byteStride)
+  RTC_API void rtcForwardOccluded16(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay16* iray, unsigned int instID)
   {
-    Scene* scene = (Scene*) hscene;
-    RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcOccludedNM);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (byteStride < sizeof(RTCRayHit)) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"byteStride too small");
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)ray) & 0x03) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "ray not aligned to 4 bytes");   
-#endif
-    STAT3(shadow.travs,N*M,N*N,N*N);
-    IntersectContext context(scene,user_context);
-
-    /* codepath for single rays */
-    if (likely(N == 1))
-    {
-      /* fast path for streams of size 1 */
-      if (likely(M == 1)) {
-        if (likely(((RTCRay*)ray)->tnear <= ((RTCRay*)ray)->tfar))
-          scene->intersectors.occluded (*(RTCRay*)ray,&context);
-      } 
-      /* codepath for normal ray streams */
-      else {
-        scene->device->rayStreamFilters.occludedAOS(scene,(RTCRay*)ray,M,byteStride,&context);
-      }
-    }
-    /* code path for ray packet streams */
-    else {
-      scene->device->rayStreamFilters.occludedSOA(scene,(char*)ray,N,M,byteStride,&context);
-    }
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcOccludedNM not supported");
-#endif
-    RTC_CATCH_END2(scene);
+    RTC_TRACE(rtcForwardOccluded16);
+    return rtcForwardOccluded16Ex(valid, args, hscene, iray, instID, 0);
   }
 
-  RTC_API void rtcOccludedNp(RTCScene hscene, RTCIntersectContext* user_context, const RTCRayNp* ray, unsigned int N)
+  RTC_API void rtcForwardOccluded16Ex(const int* valid, const RTCOccludedFunctionNArguments* args, RTCScene hscene, RTCRay16* iray, unsigned int instID, unsigned int instPrimID)
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
-    RTC_TRACE(rtcOccludedNp);
-
-#if defined (EMBREE_RAY_PACKETS)
-#if defined(DEBUG)
-    RTC_VERIFY_HANDLE(hscene);
-    if (scene->isModified()) throw_RTCError(RTC_ERROR_INVALID_OPERATION,"scene not committed");
-    if (((size_t)ray->org_x ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "org_x not aligned to 4 bytes");   
-    if (((size_t)ray->org_y ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "org_y not aligned to 4 bytes");   
-    if (((size_t)ray->org_z ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "org_z not aligned to 4 bytes");   
-    if (((size_t)ray->dir_x ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "dir_x not aligned to 4 bytes");   
-    if (((size_t)ray->dir_y ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "dir_y not aligned to 4 bytes");   
-    if (((size_t)ray->dir_z ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "dir_z not aligned to 4 bytes");   
-    if (((size_t)ray->tnear ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "dir_x not aligned to 4 bytes");   
-    if (((size_t)ray->tfar  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "tnear not aligned to 4 bytes");   
-    if (((size_t)ray->time  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "time not aligned to 4 bytes");   
-    if (((size_t)ray->mask  ) & 0x03 ) throw_RTCError(RTC_ERROR_INVALID_ARGUMENT, "mask not aligned to 4 bytes");   
-#endif
-    STAT3(shadow.travs,N,N,N);
-    IntersectContext context(scene,user_context);
-    scene->device->rayStreamFilters.occludedSOP(scene,ray,N,&context);
-#else
-    throw_RTCError(RTC_ERROR_INVALID_OPERATION,"rtcOccludedNp not supported");
-#endif
+    RTC_TRACE(rtcForwardOccluded16Ex);
+    rtcForwardOccludedN<RTCRay16,16>(valid, args, hscene, iray, instID, instPrimID);
     RTC_CATCH_END2(scene);
   }
-
+  
   RTC_API void rtcRetainScene (RTCScene hscene) 
   {
     Scene* scene = (Scene*) hscene;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcRetainScene);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     scene->refInc();
     RTC_CATCH_END2(scene);
   }
@@ -960,6 +1150,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcReleaseScene);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hscene);
     scene->refDec();
     RTC_CATCH_END2(scene);
   }
@@ -972,10 +1163,23 @@ RTC_NAMESPACE_BEGIN;
     RTC_TRACE(rtcSetGeometryInstancedScene);
     RTC_VERIFY_HANDLE(hgeometry);
     RTC_VERIFY_HANDLE(hscene);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setInstancedScene(scene);
     RTC_CATCH_END2(geometry);
   }
 
+  RTC_API void rtcSetGeometryInstancedScenes(RTCGeometry hgeometry, RTCScene* scenes, size_t numScenes)
+  {
+    Geometry* geometry = (Geometry*) hgeometry;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcSetGeometryInstancedScene);
+    RTC_VERIFY_HANDLE(hgeometry);
+    RTC_VERIFY_HANDLE(scenes);
+    RTC_ENTER_DEVICE(hgeometry);
+    geometry->setInstancedScenes(scenes, numScenes);
+    RTC_CATCH_END2(geometry);
+  }
+
   AffineSpace3fa loadTransform(RTCFormat format, const float* xfm)
   {
     AffineSpace3fa space = one;
@@ -1009,43 +1213,14 @@ RTC_NAMESPACE_BEGIN;
     return space;
   }
 
-  void storeTransform(const AffineSpace3fa& space, RTCFormat format, float* xfm)
-  {
-    switch (format)
-    {
-    case RTC_FORMAT_FLOAT3X4_ROW_MAJOR:
-      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vy.x;  xfm[ 2] = space.l.vz.x;  xfm[ 3] = space.p.x;
-      xfm[ 4] = space.l.vx.y;  xfm[ 5] = space.l.vy.y;  xfm[ 6] = space.l.vz.y;  xfm[ 7] = space.p.y;
-      xfm[ 8] = space.l.vx.z;  xfm[ 9] = space.l.vy.z;  xfm[10] = space.l.vz.z;  xfm[11] = space.p.z;
-      break;
-
-    case RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR:
-      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vx.y;  xfm[ 2] = space.l.vx.z;
-      xfm[ 3] = space.l.vy.x;  xfm[ 4] = space.l.vy.y;  xfm[ 5] = space.l.vy.z;
-      xfm[ 6] = space.l.vz.x;  xfm[ 7] = space.l.vz.y;  xfm[ 8] = space.l.vz.z;
-      xfm[ 9] = space.p.x;     xfm[10] = space.p.y;     xfm[11] = space.p.z;
-      break;
-
-    case RTC_FORMAT_FLOAT4X4_COLUMN_MAJOR:
-      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vx.y;  xfm[ 2] = space.l.vx.z;  xfm[ 3] = 0.f;
-      xfm[ 4] = space.l.vy.x;  xfm[ 5] = space.l.vy.y;  xfm[ 6] = space.l.vy.z;  xfm[ 7] = 0.f;
-      xfm[ 8] = space.l.vz.x;  xfm[ 9] = space.l.vz.y;  xfm[10] = space.l.vz.z;  xfm[11] = 0.f;
-      xfm[12] = space.p.x;     xfm[13] = space.p.y;     xfm[14] = space.p.z;     xfm[15] = 1.f;
-      break;
-
-    default:
-      throw_RTCError(RTC_ERROR_INVALID_OPERATION, "invalid matrix format");
-      break;
-    }
-  }
-
-  RTC_API void rtcSetGeometryTransform(RTCGeometry hgeometry, unsigned int timeStep, RTCFormat format, const void* xfm)
+RTC_API void rtcSetGeometryTransform(RTCGeometry hgeometry, unsigned int timeStep, RTCFormat format, const void* xfm)
   {
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryTransform);
     RTC_VERIFY_HANDLE(hgeometry);
     RTC_VERIFY_HANDLE(xfm);
+    RTC_ENTER_DEVICE(hgeometry);
     const AffineSpace3fa transform = loadTransform(format, (const float*)xfm);
     geometry->setTransform(transform, timeStep);
     RTC_CATCH_END2(geometry);
@@ -1058,6 +1233,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_TRACE(rtcSetGeometryTransformQuaternion);
     RTC_VERIFY_HANDLE(hgeometry);
     RTC_VERIFY_HANDLE(qd);
+    RTC_ENTER_DEVICE(hgeometry);
     
     AffineSpace3fx transform;
     transform.l.vx.x = qd->scale_x;
@@ -1090,21 +1266,46 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryTransform);
+    //RTC_ENTER_DEVICE(hgeometry); // no allocation required
     const AffineSpace3fa transform = geometry->getTransform(time);
     storeTransform(transform, format, (float*)xfm);
     RTC_CATCH_END2(geometry);
   }
 
-  RTC_API void rtcFilterIntersection(const struct RTCIntersectFunctionNArguments* const args_i, const struct RTCFilterFunctionNArguments* filter_args)
+  RTC_API void rtcGetGeometryTransformEx(RTCGeometry hgeometry, unsigned int instPrimID, float time, RTCFormat format, void* xfm)
+  {
+    Geometry* geometry = (Geometry*) hgeometry;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcGetGeometryTransformEx);
+    //RTC_ENTER_DEVICE(hgeometry); // no allocation required
+    const AffineSpace3fa transform = geometry->getTransform(instPrimID, time);
+    storeTransform(transform, format, (float*)xfm);
+    RTC_CATCH_END2(geometry);
+  }
+
+  RTC_API void rtcGetGeometryTransformFromScene(RTCScene hscene, unsigned int geomID, float time, RTCFormat format, void* xfm)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcGetGeometryTransformFromScene);
+    //RTC_ENTER_DEVICE(hscene); // no allocation required
+    const AffineSpace3fa transform = scene->get(geomID)->getTransform(time);
+    storeTransform(transform, format, (float*)xfm);
+    RTC_CATCH_END2(scene);
+  }
+
+  RTC_API void rtcInvokeIntersectFilterFromGeometry(const struct RTCIntersectFunctionNArguments* const args_i, const struct RTCFilterFunctionNArguments* filter_args)
   {
     IntersectFunctionNArguments* args = (IntersectFunctionNArguments*) args_i;
-    isa::reportIntersection1(args, filter_args);
+    if (args->geometry->intersectionFilterN)
+        args->geometry->intersectionFilterN(filter_args);
   }
 
-  RTC_API void rtcFilterOcclusion(const struct RTCOccludedFunctionNArguments* const args_i, const struct RTCFilterFunctionNArguments* filter_args)
+  RTC_API void rtcInvokeOccludedFilterFromGeometry(const struct RTCOccludedFunctionNArguments* const args_i, const struct RTCFilterFunctionNArguments* filter_args)
   {
     OccludedFunctionNArguments* args = (OccludedFunctionNArguments*) args_i;
-    isa::reportOcclusion1(args,filter_args);
+    if (args->geometry->occlusionFilterN)
+      args->geometry->occlusionFilterN(filter_args);
   }
   
   RTC_API RTCGeometry rtcNewGeometry (RTCDevice hdevice, RTCGeometryType type)
@@ -1112,6 +1313,7 @@ RTC_NAMESPACE_BEGIN;
     Device* device = (Device*) hdevice;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcNewGeometry);
+    RTC_ENTER_DEVICE(hdevice);
     RTC_VERIFY_HANDLE(hdevice);
 
     switch (type)
@@ -1262,6 +1464,18 @@ RTC_NAMESPACE_BEGIN;
 #endif
     }
 
+    case RTC_GEOMETRY_TYPE_INSTANCE_ARRAY:
+    {
+#if defined(EMBREE_GEOMETRY_INSTANCE_ARRAY)
+      createInstanceArrayTy createInstanceArray = nullptr;
+      SELECT_SYMBOL_DEFAULT_AVX_AVX2_AVX512(device->enabled_cpu_features,createInstanceArray);
+      Geometry* geom = createInstanceArray(device);
+      return (RTCGeometry) geom->refInc();
+#else
+      throw_RTCError(RTC_ERROR_UNKNOWN,"RTC_GEOMETRY_TYPE_INSTANCE_ARRAY is not supported");
+#endif
+    }
+
     case RTC_GEOMETRY_TYPE_GRID:
     {
 #if defined(EMBREE_GEOMETRY_GRID)
@@ -1288,6 +1502,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryUserPrimitiveCount);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     
     if (unlikely(geometry->getType() != Geometry::GTY_USER_GEOMETRY))
       throw_RTCError(RTC_ERROR_INVALID_OPERATION,"operation only allowed for user geometries"); 
@@ -1302,6 +1517,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryTimeStepCount);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
 
     if (timeStepCount > RTC_MAX_TIME_STEP_COUNT)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"number of time steps is out of range");
@@ -1316,6 +1532,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryTimeRange);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
 
     if (startTime > endTime)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"startTime has to be smaller or equal to the endTime");
@@ -1330,6 +1547,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryVertexAttributeCount);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setVertexAttributeCount(N);
     RTC_CATCH_END2(geometry);
   }
@@ -1340,6 +1558,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryTopologyCount);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setTopologyCount(N);
     RTC_CATCH_END2(geometry);
   }
@@ -1350,14 +1569,20 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryBuildQuality);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
+    // -- GODOT start --
+    // if (quality != RTC_BUILD_QUALITY_LOW &&
+    //     quality != RTC_BUILD_QUALITY_MEDIUM &&
+    //     quality != RTC_BUILD_QUALITY_HIGH &&
+    //     quality != RTC_BUILD_QUALITY_REFIT)
+    //   throw std::runtime_error("invalid build quality");
     if (quality != RTC_BUILD_QUALITY_LOW &&
         quality != RTC_BUILD_QUALITY_MEDIUM &&
         quality != RTC_BUILD_QUALITY_HIGH &&
-        quality != RTC_BUILD_QUALITY_REFIT)
-      // -- GODOT start --
-      // throw std::runtime_error("invalid build quality");
+        quality != RTC_BUILD_QUALITY_REFIT) {
       abort();
-      // -- GODOT end --
+    }
+    // -- GODOT end --
     geometry->setBuildQuality(quality);
     RTC_CATCH_END2(geometry);
   }
@@ -1383,6 +1608,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryMask);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setMask(mask);
     RTC_CATCH_END2(geometry);
   }
@@ -1393,6 +1619,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometrySubdivisionMode);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setSubdivisionMode(topologyID,mode);
     RTC_CATCH_END2(geometry);
   }
@@ -1403,6 +1630,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryVertexAttributeTopology);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setVertexAttributeTopology(vertexAttributeID, topologyID);
     RTC_CATCH_END2(geometry);
   }
@@ -1415,6 +1643,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_TRACE(rtcSetGeometryBuffer);
     RTC_VERIFY_HANDLE(hgeometry);
     RTC_VERIFY_HANDLE(hbuffer);
+    RTC_ENTER_DEVICE(hgeometry);
     
     if (geometry->device != buffer->device)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"inputs are from different devices");
@@ -1432,10 +1661,11 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetSharedGeometryBuffer);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     
     if (itemCount > 0xFFFFFFFFu)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"buffer too large");
-    
+
     Ref<Buffer> buffer = new Buffer(geometry->device, itemCount*byteStride, (char*)ptr + byteOffset);
     geometry->setBuffer(type, slot, format, buffer, 0, byteStride, (unsigned int)itemCount);
     RTC_CATCH_END2(geometry);
@@ -1447,6 +1677,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetNewGeometryBuffer);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
 
     if (itemCount > 0xFFFFFFFFu)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"buffer too large");
@@ -1469,6 +1700,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryBufferData);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     return geometry->getBuffer(type, slot);
     RTC_CATCH_END2(geometry);
     return nullptr;
@@ -1480,6 +1712,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcEnableGeometry);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->enable();
     RTC_CATCH_END2(geometry);
   }
@@ -1490,6 +1723,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcUpdateGeometryBuffer);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->updateBuffer(type, slot);
     RTC_CATCH_END2(geometry);
   }
@@ -1500,6 +1734,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcDisableGeometry);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->disable();
     RTC_CATCH_END2(geometry);
   }
@@ -1510,6 +1745,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryTessellationRate);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setTessellationRate(tessellationRate);
     RTC_CATCH_END2(geometry);
   }
@@ -1520,6 +1756,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryUserData);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setUserData(ptr);
     RTC_CATCH_END2(geometry);
   }
@@ -1530,17 +1767,34 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryUserData);
     RTC_VERIFY_HANDLE(hgeometry);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons !
     return geometry->getUserData();
     RTC_CATCH_END2(geometry);
     return nullptr;
   }
 
+  RTC_API void* rtcGetGeometryUserDataFromScene (RTCScene hscene, unsigned int geomID)
+  {
+    Scene* scene = (Scene*) hscene;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcGetGeometryUserDataFromScene);
+#if defined(DEBUG)
+    RTC_VERIFY_HANDLE(hscene);
+    RTC_VERIFY_GEOMID(geomID);
+#endif
+    //RTC_ENTER_DEVICE(hscene); // do not enable for performance reasons
+    return scene->get(geomID)->getUserData();
+    RTC_CATCH_END2(scene);
+    return nullptr;
+  }
+
   RTC_API void rtcSetGeometryBoundsFunction (RTCGeometry hgeometry, RTCBoundsFunction bounds, void* userPtr)
   {
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryBoundsFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setBoundsFunction(bounds,userPtr);
     RTC_CATCH_END2(geometry);
   }
@@ -1551,6 +1805,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryDisplacementFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setDisplacementFunction(displacement);
     RTC_CATCH_END2(geometry);
   }
@@ -1561,6 +1816,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryIntersectFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setIntersectFunctionN(intersect);
     RTC_CATCH_END2(geometry);
   }
@@ -1571,6 +1827,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryPointQueryFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setPointQueryFunction(pointQuery);
     RTC_CATCH_END2(geometry);
   }
@@ -1580,6 +1837,7 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryFirstHalfEdge);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     return geometry->getFirstHalfEdge(faceID);
     RTC_CATCH_END2(geometry);
     return -1;
@@ -1590,6 +1848,7 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryFace);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     return geometry->getFace(edgeID);
     RTC_CATCH_END2(geometry);
     return -1;
@@ -1600,6 +1859,7 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryNextHalfEdge);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     return geometry->getNextHalfEdge(edgeID);
     RTC_CATCH_END2(geometry);
     return -1;
@@ -1610,6 +1870,7 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryPreviousHalfEdge);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     return geometry->getPreviousHalfEdge(edgeID);
     RTC_CATCH_END2(geometry);
     return -1;
@@ -1620,6 +1881,7 @@ RTC_NAMESPACE_BEGIN;
     Geometry* geometry = (Geometry*) hgeometry;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcGetGeometryOppositeHalfEdge);
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     return geometry->getOppositeHalfEdge(topologyID,edgeID);
     RTC_CATCH_END2(geometry);
     return -1;
@@ -1631,6 +1893,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetOccludedFunctionN);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setOccludedFunctionN(occluded);
     RTC_CATCH_END2(geometry);
   }
@@ -1641,6 +1904,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryIntersectFilterFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setIntersectionFilterFunctionN(filter);
     RTC_CATCH_END2(geometry);
   }
@@ -1651,10 +1915,22 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcSetGeometryOccludedFilterFunction);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->setOcclusionFilterFunctionN(filter);
     RTC_CATCH_END2(geometry);
   }
 
+  RTC_API void rtcSetGeometryEnableFilterFunctionFromArguments (RTCGeometry hgeometry, bool enable) 
+  {
+    Geometry* geometry = (Geometry*) hgeometry;
+    RTC_CATCH_BEGIN;
+    RTC_TRACE(rtcSetGeometryEnableFilterFunctionFromArguments);
+    RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
+    geometry->enableFilterFunctionFromArguments(enable);
+    RTC_CATCH_END2(geometry);
+  }
+
   RTC_API void rtcInterpolate(const RTCInterpolateArguments* const args)
   {
     Geometry* geometry = (Geometry*) args->geometry;
@@ -1663,6 +1939,7 @@ RTC_NAMESPACE_BEGIN;
 #if defined(DEBUG)
     RTC_VERIFY_HANDLE(args->geometry);
 #endif
+    //RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     geometry->interpolate(args);
     RTC_CATCH_END2(geometry);
   }
@@ -1675,6 +1952,7 @@ RTC_NAMESPACE_BEGIN;
 #if defined(DEBUG)
     RTC_VERIFY_HANDLE(args->geometry);
 #endif
+    // RTC_ENTER_DEVICE(hgeometry); // do not enable for performance reasons
     geometry->interpolateN(args);
     RTC_CATCH_END2(geometry);
   }
@@ -1685,6 +1963,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcCommitGeometry);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     return geometry->commit();
     RTC_CATCH_END2(geometry);
   }
@@ -1697,6 +1976,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_TRACE(rtcAttachGeometry);
     RTC_VERIFY_HANDLE(hscene);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     if (scene->device != geometry->device)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"inputs are from different devices");
     return scene->bind(RTC_INVALID_GEOMETRY_ID,geometry);
@@ -1713,6 +1993,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_VERIFY_HANDLE(hscene);
     RTC_VERIFY_HANDLE(hgeometry);
     RTC_VERIFY_GEOMID(geomID);
+    RTC_ENTER_DEVICE(hscene);
     if (scene->device != geometry->device)
       throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"inputs are from different devices");
     scene->bind(geomID,geometry);
@@ -1726,6 +2007,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_TRACE(rtcDetachGeometry);
     RTC_VERIFY_HANDLE(hscene);
     RTC_VERIFY_GEOMID(geomID);
+    RTC_ENTER_DEVICE(hscene);
     scene->detachGeometry(geomID);
     RTC_CATCH_END2(scene);
   }
@@ -1736,6 +2018,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcRetainGeometry);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->refInc();
     RTC_CATCH_END2(geometry);
   }
@@ -1746,6 +2029,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_CATCH_BEGIN;
     RTC_TRACE(rtcReleaseGeometry);
     RTC_VERIFY_HANDLE(hgeometry);
+    RTC_ENTER_DEVICE(hgeometry);
     geometry->refDec();
     RTC_CATCH_END2(geometry);
   }
@@ -1759,6 +2043,7 @@ RTC_NAMESPACE_BEGIN;
     RTC_VERIFY_HANDLE(hscene);
     RTC_VERIFY_GEOMID(geomID);
 #endif
+    //RTC_ENTER_DEVICE(hscene); // do not enable for performance reasons
     return (RTCGeometry) scene->get(geomID);
     RTC_CATCH_END2(scene);
     return nullptr;
diff --git a/thirdparty/embree/kernels/common/rtcore.h b/thirdparty/embree/kernels/common/rtcore.h
index ac58a84d6f..47526482c1 100644
--- a/thirdparty/embree/kernels/common/rtcore.h
+++ b/thirdparty/embree/kernels/common/rtcore.h
@@ -3,26 +3,14 @@
 
 #pragma once
 
-#include "../../include/embree3/rtcore.h"
+#include "../../include/embree4/rtcore.h"
 RTC_NAMESPACE_USE
 
 namespace embree
 {  
   /*! decoding of intersection flags */
-  __forceinline bool isCoherent  (RTCIntersectContextFlags flags) { return (flags & RTC_INTERSECT_CONTEXT_FLAG_COHERENT) == RTC_INTERSECT_CONTEXT_FLAG_COHERENT; }
-  __forceinline bool isIncoherent(RTCIntersectContextFlags flags) { return (flags & RTC_INTERSECT_CONTEXT_FLAG_COHERENT) == RTC_INTERSECT_CONTEXT_FLAG_INCOHERENT; }
-
-#if defined(TASKING_TBB) && (TBB_INTERFACE_VERSION_MAJOR >= 8)
-#  define USE_TASK_ARENA 1
-#else
-#  define USE_TASK_ARENA 0
-#endif
-
-#if defined(TASKING_TBB) && (TBB_INTERFACE_VERSION >= 11009) // TBB 2019 Update 9
-#  define TASKING_TBB_USE_TASK_ISOLATION 1
-#else
-#  define TASKING_TBB_USE_TASK_ISOLATION 0
-#endif
+  __forceinline bool isCoherent  (RTCRayQueryFlags flags) { return (flags & RTC_RAY_QUERY_FLAG_COHERENT) == RTC_RAY_QUERY_FLAG_COHERENT; }
+  __forceinline bool isIncoherent(RTCRayQueryFlags flags) { return (flags & RTC_RAY_QUERY_FLAG_COHERENT) == RTC_RAY_QUERY_FLAG_INCOHERENT; }
 
 /*! Macros used in the rtcore API implementation */
 // -- GODOT start --
@@ -30,8 +18,8 @@ namespace embree
 #define RTC_CATCH_END(device)
 #define RTC_CATCH_END2(scene)
 #define RTC_CATCH_END2_FALSE(scene) return false;
-
 #if 0
+// -- GODOT end --
 #define RTC_CATCH_BEGIN try {
   
 #define RTC_CATCH_END(device)                                                \
@@ -47,7 +35,7 @@ namespace embree
   
 #define RTC_CATCH_END2(scene)                                                \
   } catch (std::bad_alloc&) {                                                   \
-    Device* device = scene ? scene->device : nullptr;                           \
+    Device* device = scene ? scene->device : nullptr;		\
     Device::process_error(device,RTC_ERROR_OUT_OF_MEMORY,"out of memory");      \
   } catch (rtcore_error& e) {                                                   \
     Device* device = scene ? scene->device : nullptr;                           \
@@ -78,9 +66,9 @@ namespace embree
     Device::process_error(device,RTC_ERROR_UNKNOWN,"unknown exception caught"); \
     return false;                                                               \
   }
-#endif
-// -- GODOT end --
 
+#endif
+  
 #define RTC_VERIFY_HANDLE(handle)                               \
   if (handle == nullptr) {                                         \
     throw_RTCError(RTC_ERROR_INVALID_ARGUMENT,"invalid argument"); \
@@ -137,5 +125,38 @@ namespace embree
 // -- GODOT end --
 
 #define RTC_BUILD_ARGUMENTS_HAS(settings,member) \
-  (settings.byteSize > (offsetof(RTCBuildArguments,member)+sizeof(settings.member))) 
+  (settings.byteSize > (offsetof(RTCBuildArguments,member)+sizeof(settings.member)))
+
+  
+  inline void storeTransform(const AffineSpace3fa& space, RTCFormat format, float* xfm)
+  {
+    switch (format)
+    {
+    case RTC_FORMAT_FLOAT3X4_ROW_MAJOR:
+      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vy.x;  xfm[ 2] = space.l.vz.x;  xfm[ 3] = space.p.x;
+      xfm[ 4] = space.l.vx.y;  xfm[ 5] = space.l.vy.y;  xfm[ 6] = space.l.vz.y;  xfm[ 7] = space.p.y;
+      xfm[ 8] = space.l.vx.z;  xfm[ 9] = space.l.vy.z;  xfm[10] = space.l.vz.z;  xfm[11] = space.p.z;
+      break;
+
+    case RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR:
+      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vx.y;  xfm[ 2] = space.l.vx.z;
+      xfm[ 3] = space.l.vy.x;  xfm[ 4] = space.l.vy.y;  xfm[ 5] = space.l.vy.z;
+      xfm[ 6] = space.l.vz.x;  xfm[ 7] = space.l.vz.y;  xfm[ 8] = space.l.vz.z;
+      xfm[ 9] = space.p.x;     xfm[10] = space.p.y;     xfm[11] = space.p.z;
+      break;
+
+    case RTC_FORMAT_FLOAT4X4_COLUMN_MAJOR:
+      xfm[ 0] = space.l.vx.x;  xfm[ 1] = space.l.vx.y;  xfm[ 2] = space.l.vx.z;  xfm[ 3] = 0.f;
+      xfm[ 4] = space.l.vy.x;  xfm[ 5] = space.l.vy.y;  xfm[ 6] = space.l.vy.z;  xfm[ 7] = 0.f;
+      xfm[ 8] = space.l.vz.x;  xfm[ 9] = space.l.vz.y;  xfm[10] = space.l.vz.z;  xfm[11] = 0.f;
+      xfm[12] = space.p.x;     xfm[13] = space.p.y;     xfm[14] = space.p.z;     xfm[15] = 1.f;
+      break;
+
+    default:
+#if !defined(__SYCL_DEVICE_ONLY__)
+      throw_RTCError(RTC_ERROR_INVALID_OPERATION, "invalid matrix format");
+#endif
+      break;
+    }
+  }
 }
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;
     }
   }
diff --git a/thirdparty/embree/kernels/common/scene.h b/thirdparty/embree/kernels/common/scene.h
index 5ed80a63f6..d9acca1065 100644
--- a/thirdparty/embree/kernels/common/scene.h
+++ b/thirdparty/embree/kernels/common/scene.h
@@ -6,11 +6,11 @@
 #include "default.h"
 #include "device.h"
 #include "builder.h"
-#include "../../common/algorithms/parallel_any_of.h"
 #include "scene_triangle_mesh.h"
 #include "scene_quad_mesh.h"
 #include "scene_user_geometry.h"
 #include "scene_instance.h"
+#include "scene_instance_array.h"
 #include "scene_curves.h"
 #include "scene_line_segments.h"
 #include "scene_subdiv_mesh.h"
@@ -21,12 +21,18 @@
 #include "acceln.h"
 #include "geometry.h"
 
+#if defined(EMBREE_SYCL_SUPPORT)
+#include "../sycl/rthwif_embree_builder.h"
+#endif
+
 namespace embree
 {
+  struct TaskGroup;
+
   /*! Base class all scenes are derived from */
   class Scene : public AccelN
   {
-    ALIGNED_CLASS_(std::alignment_of<Scene>::value);
+    ALIGNED_CLASS_USM_(std::alignment_of<Scene>::value);
 
   public:
     template<typename Ty, bool mblur = false>
@@ -140,6 +146,7 @@ namespace embree
     ~Scene () noexcept;
 
   private:
+    
     /*! class is non-copyable */
     Scene (const Scene& other) DELETED; // do not implement
     Scene& operator= (const Scene& other) DELETED; // do not implement
@@ -159,6 +166,8 @@ namespace embree
     void createInstanceMBAccel();
     void createInstanceExpensiveAccel();
     void createInstanceExpensiveMBAccel();
+    void createInstanceArrayAccel();
+    void createInstanceArrayMBAccel();
     void createGridAccel();
     void createGridMBAccel();
 
@@ -176,13 +185,13 @@ namespace embree
     
     void setSceneFlags(RTCSceneFlags scene_flags);
     RTCSceneFlags getSceneFlags() const;
-    
+
+    void build_cpu_accels();
+    void build_gpu_accels();
     void commit (bool join);
     void commit_task ();
     void build () {}
 
-    void updateInterface();
-
     /* return number of geometries */
     __forceinline size_t size() const { return geometries.size(); }
     
@@ -205,20 +214,9 @@ namespace embree
     }
 
   protected:
-    
-    __forceinline void checkIfModifiedAndSet () 
-    {
-      if (isModified ()) return;
-      
-      auto geometryIsModified = [this](size_t geomID)->bool {
-        return isGeometryModified(geomID);
-      };
 
-      if (parallel_any_of (size_t(0), geometries.size (), geometryIsModified)) {
-        setModified ();
-      }
-    }
-    
+    void checkIfModifiedAndSet ();
+
   public:
 
     /* get mesh by ID */
@@ -247,7 +245,7 @@ namespace embree
     }
 
     __forceinline Ref<Geometry> get_locked(size_t i)  {
-      Lock<SpinLock> lock(geometriesMutex);
+      Lock<MutexSys> lock(geometriesMutex);
       assert(i < geometries.size()); 
       return geometries[i]; 
     }
@@ -259,8 +257,8 @@ namespace embree
     __forceinline bool isStaticAccel()  const { return !(scene_flags & RTC_SCENE_FLAG_DYNAMIC); }
     __forceinline bool isDynamicAccel() const { return scene_flags & RTC_SCENE_FLAG_DYNAMIC; }
     
-    __forceinline bool hasContextFilterFunction() const {
-      return scene_flags & RTC_SCENE_FLAG_CONTEXT_FILTER_FUNCTION;
+    __forceinline bool hasArgumentFilterFunction() const {
+      return scene_flags & RTC_SCENE_FLAG_FILTER_FUNCTION_IN_ARGUMENTS;
     }
     
     __forceinline bool hasGeometryFilterFunction() {
@@ -268,21 +266,21 @@ namespace embree
     }
       
     __forceinline bool hasFilterFunction() {
-      return hasContextFilterFunction() || hasGeometryFilterFunction();
+      return hasArgumentFilterFunction() || hasGeometryFilterFunction();
     }
     
-    /* test if scene got already build */
-    __forceinline bool isBuild() const { return is_build; }
+    void* createQBVH6Accel();
+
+  public:
+    Device* device;
 
   public:
     IDPool<unsigned,0xFFFFFFFE> id_pool;
-    vector<Ref<Geometry>> geometries; //!< list of all user geometries
-    vector<unsigned int> geometryModCounters_;
-    vector<float*> vertices;
+    Device::vector<Ref<Geometry>> geometries = device; //!< list of all user geometries
+    avector<unsigned int> geometryModCounters_;
+    Device::vector<float*> vertices = device;
     
   public:
-    Device* device;
-
     /* these are to detect if we need to recreate the acceleration structures */
     bool flags_modified;
     unsigned int enabled_geometry_types;
@@ -290,24 +288,20 @@ namespace embree
     RTCSceneFlags scene_flags;
     RTCBuildQuality quality_flags;
     MutexSys buildMutex;
-    SpinLock geometriesMutex;
-    bool is_build;
+    MutexSys geometriesMutex;
+
+#if defined(EMBREE_SYCL_SUPPORT)
+  public:
+    BBox3f hwaccel_bounds = empty;
+    AccelBuffer hwaccel;
+#endif
+    
   private:
     bool modified;                   //!< true if scene got modified
 
   public:
-    
-    /*! 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
+
+    std::unique_ptr<TaskGroup> taskGroup;
     
   public:
     struct BuildProgressMonitorInterface : public BuildProgressMonitor {
@@ -363,12 +357,28 @@ namespace embree
       
       if (mask & Geometry::MTY_INSTANCE_EXPENSIVE)
         count += mblur  ? world.numMBInstancesExpensive : world.numInstancesExpensive;
-      
+
+      if (mask & Geometry::MTY_INSTANCE_ARRAY)
+        count += mblur  ? world.numMBInstanceArrays : world.numInstanceArrays;
+
       if (mask & Geometry::MTY_GRID_MESH)
         count += mblur  ? world.numMBGrids : world.numGrids;
       
       return count;
     }
+
+    __forceinline size_t getNumSubPrimitives(Geometry::GTypeMask mask, bool mblur) const
+    {
+      size_t count = 0;
+      
+      if (mask & Geometry::MTY_GRID_MESH)
+        count += mblur  ? world.numMBSubGrids : world.numSubGrids;
+
+      Geometry::GTypeMask new_mask = (Geometry::GTypeMask)(mask & ~Geometry::MTY_GRID_MESH);
+      count += getNumPrimitives(new_mask, mblur);
+      
+      return count;
+    }
     
     template<typename Mesh, bool mblur>
     __forceinline unsigned getNumTimeSteps()
diff --git a/thirdparty/embree/kernels/common/scene_curves.h b/thirdparty/embree/kernels/common/scene_curves.h
index a1ea45d3c7..fd6ed81d7d 100644
--- a/thirdparty/embree/kernels/common/scene_curves.h
+++ b/thirdparty/embree/kernels/common/scene_curves.h
@@ -119,6 +119,15 @@ namespace embree
       p3 = vertex(i+3,itime);
     }
 
+    /*! gathers the curve normals starting with i'th vertex */
+    __forceinline void gather_normals(Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i) const
+    {
+      n0 = normal(i+0);
+      n1 = normal(i+1);
+      n2 = normal(i+2);
+      n3 = normal(i+3);
+    }
+
     /*! gathers the curve starting with i'th vertex */
     __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i) const
     {
@@ -178,6 +187,13 @@ namespace embree
     }
 
     /*! loads curve vertices for specified time */
+    __forceinline void gather_safe(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, size_t i, float time) const
+    {
+      if (hasMotionBlur()) gather(p0,p1,p2,p3,i,time);
+      else                 gather(p0,p1,p2,p3,i);
+    }
+    
+    /*! loads curve vertices for specified time */
     __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i, float time) const
     {
       float ftime;
@@ -199,8 +215,15 @@ namespace embree
       n3 = madd(Vec3ff(t0),an3,t1*bn3);
     }
 
+    /*! loads curve vertices for specified time for mblur and non-mblur case */
+    __forceinline void gather_safe(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i, float time) const
+    {
+      if (hasMotionBlur()) gather(p0,p1,p2,p3,n0,n1,n2,n3,i,time);
+      else                 gather(p0,p1,p2,p3,n0,n1,n2,n3,i);
+    }
+
     template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
-    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
+    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
     {
       Vec3ff v0,v1,v2,v3; Vec3fa n0,n1,n2,n3;
       unsigned int vertexID = curve(primID);
@@ -212,7 +235,7 @@ namespace embree
     }
 
     template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
-    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
+    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
     {
       float ftime;
       const size_t itime = timeSegment(time, ftime);
@@ -221,6 +244,19 @@ namespace embree
       return clerp(curve0,curve1,ftime);
     }
 
+    template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
+    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurveSafe(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
+    {
+      float ftime = 0.0f;
+      const size_t itime = hasMotionBlur() ? timeSegment(time, ftime) : 0;
+      const TensorLinearCubicBezierSurface3fa curve0 = getNormalOrientedCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context,ray_org,primID,itime+0);
+      if (hasMotionBlur()) {
+        const TensorLinearCubicBezierSurface3fa curve1 = getNormalOrientedCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context,ray_org,primID,itime+1);
+        return clerp(curve0,curve1,ftime);
+      }
+      return curve0;
+    }
+
     /*! gathers the hermite curve starting with i'th vertex */
     __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3ff& p1, Vec3ff& t1, size_t i) const
     {
@@ -255,6 +291,13 @@ namespace embree
       t1 = madd(Vec3ff(f0),at1,f1*bt1);
     }
 
+    /*! loads curve vertices for specified time for mblur and non-mblur geometry */
+    __forceinline void gather_hermite_safe(Vec3ff& p0, Vec3ff& t0, Vec3ff& p1, Vec3ff& t1, size_t i, float time) const
+    {
+      if (hasMotionBlur()) gather_hermite(p0,t0,p1,t1,i,time);
+      else                 gather_hermite(p0,t0,p1,t1,i);
+    }
+
     /*! gathers the hermite curve starting with i'th vertex */
     __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3ff& t1, Vec3fa& n1, Vec3fa& dn1, size_t i) const
     {
@@ -282,7 +325,7 @@ namespace embree
     }
 
     /*! loads curve vertices for specified time */
-    __forceinline void gather_hermite(Vec3ff& p0, Vec3fa& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3fa& t1, Vec3fa& n1, Vec3fa& dn1, size_t i, float time) const
+    __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3ff& t1, Vec3fa& n1, Vec3fa& dn1, size_t i, float time) const
     {
       float ftime;
       const size_t itime = timeSegment(time, ftime);
@@ -301,8 +344,15 @@ namespace embree
       dn1= madd(Vec3ff(f0),adn1,f1*bdn1);
     }
 
+    /*! loads curve vertices for specified time */
+    __forceinline void gather_hermite_safe(Vec3ff& p0, Vec3ff& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3ff& t1, Vec3fa& n1, Vec3fa& dn1, size_t i, float time) const
+    {
+      if (hasMotionBlur()) gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,i,time);
+      else                 gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,i);
+    }
+
     template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
-      __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
+      __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
     {
       Vec3ff v0,t0,v1,t1; Vec3fa n0,dn0,n1,dn1;
       unsigned int vertexID = curve(primID);
@@ -315,7 +365,7 @@ namespace embree
     }
 
     template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
-    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
+    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
     {
       float ftime;
       const size_t itime = timeSegment(time, ftime);
@@ -324,6 +374,24 @@ namespace embree
       return clerp(curve0,curve1,ftime);
     }
 
+    template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
+    __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurveSafe(RayQueryContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
+    {
+      float ftime = 0.0f;
+      const size_t itime = hasMotionBlur() ? timeSegment(time, ftime) : 0;
+      const TensorLinearCubicBezierSurface3fa curve0 = getNormalOrientedHermiteCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,itime+0);
+      if (hasMotionBlur()) {
+        const TensorLinearCubicBezierSurface3fa curve1 = getNormalOrientedHermiteCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,itime+1);
+        return clerp(curve0,curve1,ftime);
+      }
+      return curve0;
+    }
+
+    /* returns the projected area */
+    __forceinline float projectedPrimitiveArea(const size_t i) const {
+      return 1.0f;
+    }
+  
   private:
     void resizeBuffers(unsigned int numSteps);
 
@@ -333,12 +401,12 @@ namespace embree
     BufferView<Vec3fa> normals0;            //!< fast access to first normal buffer
     BufferView<Vec3ff> tangents0;           //!< fast access to first tangent buffer
     BufferView<Vec3fa> dnormals0;           //!< fast access to first normal derivative buffer
-    vector<BufferView<Vec3ff>> vertices;    //!< vertex array for each timestep
-    vector<BufferView<Vec3fa>> normals;     //!< normal array for each timestep
-    vector<BufferView<Vec3ff>> tangents;    //!< tangent array for each timestep
-    vector<BufferView<Vec3fa>> dnormals;    //!< normal derivative array for each timestep
+    Device::vector<BufferView<Vec3ff>> vertices = device;    //!< vertex array for each timestep
+    Device::vector<BufferView<Vec3fa>> normals = device;     //!< normal array for each timestep
+    Device::vector<BufferView<Vec3ff>> tangents = device;    //!< tangent array for each timestep
+    Device::vector<BufferView<Vec3fa>> dnormals = device;    //!< normal derivative array for each timestep
     BufferView<char> flags;                 //!< start, end flag per segment
-    vector<BufferView<char>> vertexAttribs; //!< user buffers
+    Device::vector<BufferView<char>> vertexAttribs = device; //!< user buffers
     int tessellationRate;                   //!< tessellation rate for flat curve
     float maxRadiusScale = 1.0;             //!< maximal min-width scaling of curve radii
   };
@@ -486,7 +554,7 @@ namespace embree
         src    = vertices[bufferSlot].getPtr();
         stride = vertices[bufferSlot].getStride();
       }
-      
+
       for (unsigned int i=0; i<valueCount; i+=N)
       {
         size_t ofs = i*sizeof(float);
diff --git a/thirdparty/embree/kernels/common/scene_grid_mesh.h b/thirdparty/embree/kernels/common/scene_grid_mesh.h
index fb6fed445b..eb2048b286 100644
--- a/thirdparty/embree/kernels/common/scene_grid_mesh.h
+++ b/thirdparty/embree/kernels/common/scene_grid_mesh.h
@@ -133,12 +133,26 @@ namespace embree
         }
       }
     }
-    
+
     void addElementsToCount (GeometryCounts & counts) const;
     
-    __forceinline unsigned int getNumSubGrids(const size_t gridID)
+    __forceinline unsigned int getNumTotalQuads() const
     {
-      const Grid &g = grid(gridID);
+      size_t quads = 0;
+      for (size_t primID=0; primID<numPrimitives; primID++)
+        quads += getNumQuads(primID);
+      return quads;
+    }
+
+    __forceinline unsigned int getNumQuads(const size_t gridID) const
+    {
+      const Grid& g = grid(gridID);
+      return (unsigned int) max((int)1,((int)g.resX-1) * ((int)g.resY-1));
+    }
+    
+    __forceinline unsigned int getNumSubGrids(const size_t gridID) const
+    {
+      const Grid& g = grid(gridID);
       return max((unsigned int)1,((unsigned int)g.resX >> 1) * ((unsigned int)g.resY >> 1));
     }
 
@@ -174,6 +188,18 @@ namespace embree
       return vertices[itime][i];
     }
 
+    /*! returns i'th vertex of for specified time */
+    __forceinline const Vec3fa vertex(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3fa v0 = vertex(i, itime+0);
+      Vec3fa v1 = vertex(i, itime+1);
+      return madd(Vec3fa(t0),v0,t1*v1);
+    }
+
     /*! returns i'th vertex of itime'th timestep */
     __forceinline const char* vertexPtr(size_t i, size_t itime) const {
       return vertices[itime].getPtr(i);
@@ -198,6 +224,56 @@ namespace embree
       return vertex(index,itime);
     }
 
+    /*! returns i'th vertex of the itime'th timestep */
+    __forceinline const Vec3fa grid_vertex(const Grid& g, size_t x, size_t y, float time) const {
+      const size_t index = grid_vertex_index(g,x,y);
+      return vertex(index,time);
+    }
+    
+    /*! gathers quad vertices */
+    __forceinline void gather_quad_vertices(Vec3fa& v0, Vec3fa& v1, Vec3fa& v2, Vec3fa& v3, const Grid& g, size_t x, size_t y) const
+    {
+      v0 = grid_vertex(g,x+0,y+0);
+      v1 = grid_vertex(g,x+1,y+0);
+      v2 = grid_vertex(g,x+1,y+1);
+      v3 = grid_vertex(g,x+0,y+1);
+    }
+    
+    /*! gathers quad vertices for specified time */
+    __forceinline void gather_quad_vertices(Vec3fa& v0, Vec3fa& v1, Vec3fa& v2, Vec3fa& v3, const Grid& g, size_t x, size_t y, float time) const
+    {
+      v0 = grid_vertex(g,x+0,y+0,time);
+      v1 = grid_vertex(g,x+1,y+0,time);
+      v2 = grid_vertex(g,x+1,y+1,time);
+      v3 = grid_vertex(g,x+0,y+1,time);
+    }
+
+    /*! gathers quad vertices for mblur and non-mblur meshes */
+    __forceinline void gather_quad_vertices_safe(Vec3fa& v0, Vec3fa& v1, Vec3fa& v2, Vec3fa& v3, const Grid& g, size_t x, size_t y, float time) const
+    {
+      if (hasMotionBlur()) gather_quad_vertices(v0,v1,v2,v3,g,x,y,time);
+      else                 gather_quad_vertices(v0,v1,v2,v3,g,x,y);
+    }
+
+    /*! calculates the build bounds of the i'th quad, if it's valid */
+    __forceinline bool buildBoundsQuad(const Grid& g, size_t sx, size_t sy, BBox3fa& bbox) const
+    {
+      BBox3fa b(empty);
+      for (size_t t=0; t<numTimeSteps; t++)
+      {
+        for (size_t y=sy;y<sy+2;y++)
+          for (size_t x=sx;x<sx+2;x++)
+          {
+            const Vec3fa v = grid_vertex(g,x,y,t);
+            if (unlikely(!isvalid(v))) return false;
+            b.extend(v);
+          }
+      }
+
+      bbox = b;
+      return true;
+    }
+    
     /*! calculates the build bounds of the i'th primitive, if it's valid */
     __forceinline bool buildBounds(const Grid& g, size_t sx, size_t sy, BBox3fa& bbox) const
     {
@@ -254,7 +330,6 @@ namespace embree
       return true;
     }
 
-
     __forceinline BBox3fa bounds(const Grid& g, size_t sx, size_t sy, size_t itime) const
     {
       BBox3fa box(empty);
@@ -274,11 +349,22 @@ namespace embree
       return LBBox3fa([&] (size_t itime) { return bounds(g,sx,sy,itime); }, dt, time_range, fnumTimeSegments);
     }
 
+    __forceinline float projectedPrimitiveArea(const size_t i) const {
+      return pos_inf;
+    }
+
   public:
     BufferView<Grid> grids;      //!< array of triangles
     BufferView<Vec3fa> vertices0;        //!< fast access to first vertex buffer
-    vector<BufferView<Vec3fa>> vertices; //!< vertex array for each timestep
-    vector<RawBufferView> vertexAttribs; //!< vertex attributes
+    Device::vector<BufferView<Vec3fa>> vertices = device; //!< vertex array for each timestep
+    Device::vector<RawBufferView> vertexAttribs = device; //!< vertex attributes
+
+#if defined(EMBREE_SYCL_SUPPORT)
+    
+  public:
+    struct PrimID_XY { uint32_t primID; uint16_t x,y; };
+    Device::vector<PrimID_XY> quadID_to_primID_xy = device;  //!< maps a quad to the primitive ID and grid coordinates
+#endif
   };
 
   namespace isa
@@ -287,6 +373,94 @@ namespace embree
     {
       GridMeshISA (Device* device)
         : GridMesh(device) {}
+
+      LBBox3fa vlinearBounds(size_t buildID, const BBox1f& time_range, const SubGridBuildData * const sgrids) const override {
+        const SubGridBuildData &subgrid = sgrids[buildID];                      
+        const unsigned int primID = subgrid.primID;
+        const size_t x = subgrid.x();
+        const size_t y = subgrid.y();
+        return linearBounds(grid(primID),x,y,time_range);
+      }
+
+#if defined(EMBREE_SYCL_SUPPORT)
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const override
+      {
+        PrimInfo pinfo(empty);
+        for (size_t j=r.begin(); j<r.end(); j++)
+        {
+          BBox3fa bounds = empty;
+          const PrimID_XY& quad = quadID_to_primID_xy[j];
+          if (!buildBoundsQuad(grids[quad.primID],quad.x,quad.y,bounds)) continue;
+          const PrimRef prim(bounds,geomID,unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+#endif
+      
+      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, mvector<SubGridBuildData>& sgrids, const range<size_t>& r, size_t k, unsigned int geomID) const override 
+      {
+        PrimInfo pinfo(empty);
+        for (size_t j=r.begin(); j<r.end(); j++)
+        {
+          if (!valid(j)) continue;
+          const GridMesh::Grid &g = grid(j);
+          
+          for (unsigned int y=0; y<g.resY-1u; y+=2)
+          {
+            for (unsigned int x=0; x<g.resX-1u; x+=2)
+            {
+              BBox3fa bounds = empty;
+              if (!buildBounds(g,x,y,bounds)) continue; // get bounds of subgrid
+              const PrimRef prim(bounds,(unsigned)geomID,(unsigned)k);
+              pinfo.add_center2(prim);
+              sgrids[k] = SubGridBuildData(x | g.get3x3FlagsX(x), y | g.get3x3FlagsY(y), unsigned(j));
+              prims[k++] = prim;                
+            }
+          }
+        }
+        return pinfo;
+      }
+
+#if defined(EMBREE_SYCL_SUPPORT)
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const override
+      {
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        PrimInfo pinfo(empty);
+        for (size_t j=r.begin(); j<r.end(); j++)
+        {
+          const PrimID_XY& quad = quadID_to_primID_xy[j];
+          const LBBox3fa lbounds = linearBounds(grids[quad.primID],quad.x,quad.y,t0t1);
+          const PrimRef prim(lbounds.bounds(), unsigned(geomID), unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+#endif
+
+      PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, mvector<SubGridBuildData>& sgrids, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const override
+      {
+        PrimInfoMB pinfoMB(empty);
+        for (size_t j=r.begin(); j<r.end(); j++)
+        {
+          if (!valid(j, timeSegmentRange(t0t1))) continue;
+          const GridMesh::Grid &g = grid(j);
+          
+          for (unsigned int y=0; y<g.resY-1u; y+=2)
+          {
+            for (unsigned int x=0; x<g.resX-1u; x+=2)
+            {
+              const PrimRefMB prim(linearBounds(g,x,y,t0t1),numTimeSegments(),time_range,numTimeSegments(),unsigned(geomID),unsigned(k));
+              pinfoMB.add_primref(prim);
+              sgrids[k] = SubGridBuildData(x | g.get3x3FlagsX(x), y | g.get3x3FlagsY(y), unsigned(j));
+              prims[k++] = prim;
+            }
+          }
+        }
+        return pinfoMB;
+      }
     };
   }
 
diff --git a/thirdparty/embree/kernels/common/scene_instance.h b/thirdparty/embree/kernels/common/scene_instance.h
index 773f2b6fec..1176018777 100644
--- a/thirdparty/embree/kernels/common/scene_instance.h
+++ b/thirdparty/embree/kernels/common/scene_instance.h
@@ -13,7 +13,7 @@ namespace embree
   /*! Instanced acceleration structure */
   struct Instance : public Geometry
   {
-    ALIGNED_STRUCT_(16);
+    //ALIGNED_STRUCT_(16);
     static const Geometry::GTypeMask geom_type = Geometry::MTY_INSTANCE;
 
   public:
@@ -50,6 +50,7 @@ namespace embree
     virtual void setTransform(const AffineSpace3fa& local2world, unsigned int timeStep) override;
     virtual void setQuaternionDecomposition(const AffineSpace3ff& qd, unsigned int timeStep) override;
     virtual AffineSpace3fa getTransform(float time) override;
+    virtual AffineSpace3fa getTransform(size_t, float time) override;
     virtual void setMask (unsigned mask) override;
     virtual void build() {}
     virtual void addElementsToCount (GeometryCounts & counts) const override;
@@ -132,10 +133,13 @@ namespace embree
 
     __forceinline AffineSpace3fa getLocal2World(float t) const
     {
-      float ftime; const unsigned int itime = timeSegment(t, ftime);
-      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
-        return slerp(local2world[itime+0],local2world[itime+1],ftime);
-      return lerp(local2world[itime+0],local2world[itime+1],ftime);
+      if (numTimeSegments() > 0) {
+        float ftime; const unsigned int itime = timeSegment(t, ftime);
+        if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+          return slerp(local2world[itime+0],local2world[itime+1],ftime);
+        return lerp(local2world[itime+0],local2world[itime+1],ftime);
+      }
+      return getLocal2World();
     }
 
     __forceinline AffineSpace3fa getWorld2Local() const {
@@ -143,7 +147,9 @@ namespace embree
     }
 
     __forceinline AffineSpace3fa getWorld2Local(float t) const {
-      return rcp(getLocal2World(t));
+      if (numTimeSegments() > 0)
+        return rcp(getLocal2World(t));
+      return getWorld2Local();
     }
 
     template<int K>
@@ -154,6 +160,10 @@ namespace embree
       return getWorld2LocalLerp<K>(valid, t);
     }
 
+    __forceinline float projectedPrimitiveArea(const size_t i) const {
+      return area(bounds(i));
+    }
+
     private:
 
     template<int K>
@@ -220,7 +230,11 @@ namespace embree
       InstanceISA (Device* device)
         : Instance(device) {}
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
+        return linearBounds(primID,time_range);
+      }
+
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         assert(r.begin() == 0);
         assert(r.end()   == 1);
@@ -252,7 +266,23 @@ namespace embree
         prims[k++] = prim;
         return pinfo;
       }
-      
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        assert(r.begin() == 0);
+        assert(r.end()   == 1);
+
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+        
+        const BBox3fa bounds = linearBounds(0, t0t1).bounds();
+        const PrimRef prim(bounds, geomID, unsigned(0));
+        pinfo.add_center2(prim);
+        prims[k++] = prim;
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         assert(r.begin() == 0);
diff --git a/thirdparty/embree/kernels/common/scene_instance_array.h b/thirdparty/embree/kernels/common/scene_instance_array.h
new file mode 100644
index 0000000000..3cf4d68feb
--- /dev/null
+++ b/thirdparty/embree/kernels/common/scene_instance_array.h
@@ -0,0 +1,385 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "geometry.h"
+#include "accel.h"
+
+namespace embree
+{
+  struct MotionDerivativeCoefficients;
+
+  /*! Instanced acceleration structure */
+  struct InstanceArray : public Geometry
+  {
+    //ALIGNED_STRUCT_(16);
+    static const Geometry::GTypeMask geom_type = Geometry::MTY_INSTANCE_ARRAY;
+
+  public:
+    InstanceArray (Device* device, unsigned int numTimeSteps = 1);
+    ~InstanceArray();
+
+  private:
+    InstanceArray (const InstanceArray& other) DELETED; // do not implement
+    InstanceArray& operator= (const InstanceArray& other) DELETED; // do not implement
+
+  private:
+    LBBox3fa nonlinearBounds(size_t i,
+                             const BBox1f& time_range_in,
+                             const BBox1f& geom_time_range,
+                             float geom_time_segments) const;
+
+    BBox3fa boundSegment(size_t i, size_t itime,
+      BBox3fa const& obbox0, BBox3fa const& obbox1,
+      BBox3fa const& bbox0, BBox3fa const& bbox1,
+      float t_min, float t_max) const;
+
+    /* calculates the (correct) interpolated bounds */
+    __forceinline BBox3fa bounds(size_t i, size_t itime0, size_t itime1, float f) const
+    {
+      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+        return xfmBounds(slerp(l2w(i, itime0), l2w(i, itime1), f),
+                         lerp(getObjectBounds(i, itime0), getObjectBounds(i, itime1), f));
+      return xfmBounds(lerp(l2w(i, itime0), l2w(i, itime1), f),
+                        lerp(getObjectBounds(i, itime0), getObjectBounds(i, itime1), f));
+    }
+
+  public:
+
+    virtual void setBuffer(RTCBufferType type, unsigned int slot, RTCFormat format, const Ref<Buffer>& buffer, size_t offset, size_t stride, unsigned int num) override;
+    virtual void* getBuffer(RTCBufferType type, unsigned int slot) override;
+    virtual void updateBuffer(RTCBufferType type, unsigned int slot) override;
+
+    virtual void setNumTimeSteps (unsigned int numTimeSteps) override;
+    virtual void setInstancedScene(const Ref<Scene>& scene) override;
+    virtual void setInstancedScenes(const RTCScene* scenes, size_t numScenes) override;
+    virtual AffineSpace3fa getTransform(size_t, float time) override;
+    virtual void setMask (unsigned mask) override;
+    virtual void build() {}
+    virtual void addElementsToCount (GeometryCounts & counts) const override;
+    virtual void commit() override;
+
+  public:
+
+     /*! calculates the bounds of instance */
+    __forceinline BBox3fa bounds(size_t i) const {
+      if (!valid(i))
+        return BBox3fa();
+
+      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+        return xfmBounds(quaternionDecompositionToAffineSpace(l2w(i, 0)),getObject(i)->bounds.bounds());
+      return xfmBounds(l2w(i, 0),getObject(i)->bounds.bounds());
+    }
+
+    /*! gets the bounds of the instanced scene */
+    __forceinline BBox3fa getObjectBounds(size_t i, size_t itime) const {
+      if (!valid(i))
+        return BBox3fa();
+
+      return getObject(i)->getBounds(timeStep(itime));
+    }
+
+     /*! calculates the bounds of instance */
+    __forceinline BBox3fa bounds(size_t i, size_t itime) const {
+      if (!valid(i))
+        return BBox3fa();
+
+      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+        return xfmBounds(quaternionDecompositionToAffineSpace(l2w(i, itime)),getObjectBounds(i, itime));
+      return xfmBounds(l2w(i, itime),getObjectBounds(i, itime));
+    }
+
+    /*! calculates the linear bounds of the i'th primitive for the specified time range */
+    __forceinline LBBox3fa linearBounds(size_t i, const BBox1f& dt) const {
+      if (!valid(i))
+        return LBBox3fa();
+
+      LBBox3fa lbbox = nonlinearBounds(i, dt, time_range, fnumTimeSegments);
+      return lbbox;
+    }
+
+    /*! calculates the build bounds of the i'th item, if it's valid */
+    __forceinline bool buildBounds(size_t i, BBox3fa* bbox = nullptr) const
+    {
+      if (!valid(i))
+        return false;
+
+      const BBox3fa b = bounds(i);
+      if (bbox) *bbox = b;
+      return isvalid(b);
+    }
+
+     /*! calculates the build bounds of the i'th item at the itime'th time segment, if it's valid */
+    __forceinline bool buildBounds(size_t i, size_t itime, BBox3fa& bbox) const
+    {
+      if (!valid(i))
+        return false;
+
+      const LBBox3fa bounds = linearBounds(i,itime);
+      bbox = bounds.bounds ();
+      return isvalid(bounds);
+    }
+
+    /* gets version info of topology */
+    unsigned int getTopologyVersion() const {
+      return numPrimitives;
+    }
+  
+    /* returns true if topology changed */
+    bool topologyChanged(unsigned int otherVersion) const {
+      return numPrimitives != otherVersion;
+    }
+
+    /*! check if the i'th primitive is valid between the specified time range */
+    __forceinline bool valid(size_t i) const
+    {
+      if (object) return true;
+      return (object_ids[i] != (unsigned int)(-1));
+    }
+
+    /*! check if the i'th primitive is valid between the specified time range */
+    __forceinline bool valid(size_t i, const range<size_t>& itime_range) const
+    {
+      for (size_t itime = itime_range.begin(); itime <= itime_range.end(); itime++)
+        if (!isvalid(bounds(i,itime))) return false;
+
+      return true;
+    }
+
+    __forceinline AffineSpace3fa getLocal2World(size_t i) const
+    {
+      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+        return quaternionDecompositionToAffineSpace(l2w(i,0));
+      return l2w(i, 0);
+    }
+
+    __forceinline AffineSpace3fa getLocal2World(size_t i, float t) const
+    {
+      if (numTimeSegments() > 0) {
+        float ftime; const unsigned int itime = timeSegment(t, ftime);
+        if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+          return slerp(l2w(i, itime+0),l2w(i, itime+1),ftime);
+        return lerp(l2w(i, itime+0),l2w(i, itime+1),ftime);
+      }
+      return getLocal2World(i);
+    }
+
+    __forceinline AffineSpace3fa getWorld2Local(size_t i) const {
+      return rcp(getLocal2World(i));
+    }
+
+    __forceinline AffineSpace3fa getWorld2Local(size_t i, float t) const {
+      return rcp(getLocal2World(i, t));
+    }
+
+    template<int K>
+    __forceinline AffineSpace3vf<K> getWorld2Local(size_t i, const vbool<K>& valid, const vfloat<K>& t) const
+    {
+      if (unlikely(gsubtype == GTY_SUBTYPE_INSTANCE_QUATERNION))
+        return getWorld2LocalSlerp<K>(i, valid, t);
+      return getWorld2LocalLerp<K>(i, valid, t);
+    }
+
+    __forceinline float projectedPrimitiveArea(const size_t i) const {
+      return area(bounds(i));
+    }
+
+    inline Accel* getObject(size_t i) const {
+      if (object) {
+        return object;
+      }
+
+      assert(objects);
+      assert(i < numPrimitives);
+      if (object_ids[i] == (unsigned int)(-1))
+        return nullptr;
+
+      assert(object_ids[i] < numObjects);
+      return objects[object_ids[i]];
+    }
+
+    private:
+
+    template<int K>
+    __forceinline AffineSpace3vf<K> getWorld2LocalSlerp(size_t i, const vbool<K>& valid, const vfloat<K>& t) const
+    {
+      vfloat<K> ftime;
+      const vint<K> itime_k = timeSegment<K>(t, ftime);
+      assert(any(valid));
+      const size_t index = bsf(movemask(valid));
+      const int itime = itime_k[index];
+      if (likely(all(valid, itime_k == vint<K>(itime)))) {
+        return rcp(slerp(AffineSpace3vff<K>(l2w(i, itime+0)),
+                         AffineSpace3vff<K>(l2w(i, itime+1)),
+                         ftime));
+      }
+      else {
+        AffineSpace3vff<K> space0,space1;
+        vbool<K> valid1 = valid;
+        while (any(valid1)) {
+          vbool<K> valid2;
+          const int itime = next_unique(valid1, itime_k, valid2);
+          space0 = select(valid2, AffineSpace3vff<K>(l2w(i, itime+0)), space0);
+          space1 = select(valid2, AffineSpace3vff<K>(l2w(i, itime+1)), space1);
+        }
+        return rcp(slerp(space0, space1, ftime));
+      }
+    }
+
+    template<int K>
+    __forceinline AffineSpace3vf<K> getWorld2LocalLerp(size_t i, const vbool<K>& valid, const vfloat<K>& t) const
+    {
+      vfloat<K> ftime;
+      const vint<K> itime_k = timeSegment<K>(t, ftime);
+      assert(any(valid));
+      const size_t index = bsf(movemask(valid));
+      const int itime = itime_k[index];
+      if (likely(all(valid, itime_k == vint<K>(itime)))) {
+        return rcp(lerp(AffineSpace3vf<K>((AffineSpace3fa)l2w(i, itime+0)),
+                        AffineSpace3vf<K>((AffineSpace3fa)l2w(i, itime+1)),
+                        ftime));
+      } else {
+        AffineSpace3vf<K> space0,space1;
+        vbool<K> valid1 = valid;
+        while (any(valid1)) {
+          vbool<K> valid2;
+          const int itime = next_unique(valid1, itime_k, valid2);
+          space0 = select(valid2, AffineSpace3vf<K>((AffineSpace3fa)l2w(i, itime+0)), space0);
+          space1 = select(valid2, AffineSpace3vf<K>((AffineSpace3fa)l2w(i, itime+1)), space1);
+        }
+        return rcp(lerp(space0, space1, ftime));
+      }
+    }
+
+  private:
+
+    __forceinline AffineSpace3ff l2w(size_t i, size_t itime) const {
+      if (l2w_buf[itime].getFormat() == RTC_FORMAT_FLOAT4X4_COLUMN_MAJOR) {
+        return *(AffineSpace3ff*)(l2w_buf[itime].getPtr(i));
+      }
+      else if(l2w_buf[itime].getFormat() == RTC_FORMAT_QUATERNION_DECOMPOSITION) {
+        AffineSpace3ff transform;
+        QuaternionDecomposition* qd = (QuaternionDecomposition*)l2w_buf[itime].getPtr(i);
+        transform.l.vx.x = qd->scale_x;
+        transform.l.vy.y = qd->scale_y;
+        transform.l.vz.z = qd->scale_z;
+        transform.l.vy.x = qd->skew_xy;
+        transform.l.vz.x = qd->skew_xz;
+        transform.l.vz.y = qd->skew_yz;
+        transform.l.vx.y = qd->translation_x;
+        transform.l.vx.z = qd->translation_y;
+        transform.l.vy.z = qd->translation_z;
+        transform.p.x    = qd->shift_x;
+        transform.p.y    = qd->shift_y;
+        transform.p.z    = qd->shift_z;
+        // normalize quaternion
+        Quaternion3f q(qd->quaternion_r, qd->quaternion_i, qd->quaternion_j, qd->quaternion_k);
+        q = normalize(q);
+        transform.l.vx.w = q.i;
+        transform.l.vy.w = q.j;
+        transform.l.vz.w = q.k;
+        transform.p.w    = q.r;
+        return transform;
+      }
+      else if (l2w_buf[itime].getFormat() == RTC_FORMAT_FLOAT3X4_COLUMN_MAJOR) {
+        AffineSpace3f* l2w = reinterpret_cast<AffineSpace3f*>(l2w_buf[itime].getPtr(i));
+        return AffineSpace3ff(*l2w);
+      }
+      else if (l2w_buf[itime].getFormat() == RTC_FORMAT_FLOAT3X4_ROW_MAJOR) {
+        float* data = reinterpret_cast<float*>(l2w_buf[itime].getPtr(i));
+        AffineSpace3f l2w;
+        l2w.l.vx.x = data[0]; l2w.l.vy.x = data[1]; l2w.l.vz.x = data[2]; l2w.p.x = data[3];
+        l2w.l.vx.y = data[4]; l2w.l.vy.y = data[5]; l2w.l.vz.y = data[6]; l2w.p.y = data[7];
+        l2w.l.vx.z = data[8]; l2w.l.vy.z = data[9]; l2w.l.vz.z = data[10]; l2w.p.z = data[11];
+        return l2w;
+      }
+      assert(false);
+      return AffineSpace3ff();
+    }
+
+    inline AffineSpace3ff l2w(size_t i) const {
+      return l2w(i, 0);
+    }
+
+  private:
+    Accel* object;                   //!< fast path if only one scene is instanced
+    Accel** objects;
+    uint32_t numObjects;
+    Device::vector<RawBufferView> l2w_buf = device; //!< transformation from local space to world space for each timestep (either normal matrix or quaternion decomposition)
+    BufferView<uint32_t> object_ids; //!< array of scene ids per instance array primitive
+  };
+
+  namespace isa
+  {
+    struct InstanceArrayISA : public InstanceArray
+    {
+      InstanceArrayISA (Device* device)
+        : InstanceArray(device) {}
+
+      LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
+        return linearBounds(primID,time_range);
+      }
+
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          BBox3fa bounds = empty;
+          if (!buildBounds(j, &bounds) || !valid(j))
+            continue;
+          const PrimRef prim(bounds, geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
+      PrimInfo createPrimRefArrayMB(mvector<PrimRef>& prims, size_t itime, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          BBox3fa bounds = empty;
+          if (!buildBounds(j, itime, bounds))
+            continue;
+          const PrimRef prim(bounds, geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = linearBounds(j, t0t1);
+          if (!isvalid(lbounds.bounds()))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
+      PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfoMB pinfo(empty);
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          if (!valid(j, timeSegmentRange(t0t1)))
+            continue;
+          const PrimRefMB prim(linearBounds(j, t0t1), this->numTimeSegments(), this->time_range, this->numTimeSegments(), geomID, unsigned(j));
+          pinfo.add_primref(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+    };
+  }
+
+  DECLARE_ISA_FUNCTION(InstanceArray*, createInstanceArray, Device*);
+}
diff --git a/thirdparty/embree/kernels/common/scene_line_segments.h b/thirdparty/embree/kernels/common/scene_line_segments.h
index 3c9fdb39db..e58fd1b7eb 100644
--- a/thirdparty/embree/kernels/common/scene_line_segments.h
+++ b/thirdparty/embree/kernels/common/scene_line_segments.h
@@ -84,6 +84,14 @@ namespace embree
       return segments[i];
     }
 
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    /*! returns the i'th segment */
+    template<int M>
+    __forceinline const vuint<M> vsegment(const vuint<M>& i) const {
+      return segments[i.v];
+    }
+#endif
+
     /*! returns the segment to the left of the i'th segment */
     __forceinline bool segmentLeftExists(size_t i) const {
       assert (flags);
@@ -136,6 +144,219 @@ namespace embree
       return vertices[itime][i].w;
     }
 
+    /*! gathers the curve starting with i'th vertex */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, unsigned int vid) const
+    {
+      p0 = vertex(vid+0);
+      p1 = vertex(vid+1);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, const vuint<M>& vid) const
+    {
+      p0 = vertex(vid.v+0);
+      p1 = vertex(vid.v+1);
+    }
+#endif
+
+    /*! gathers the curve starting with i'th vertex of itime'th timestep */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, unsigned int vid, size_t itime) const
+    {
+      p0 = vertex(vid+0,itime);
+      p1 = vertex(vid+1,itime);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, const vuint<M>& vid, const vint<M>& itime) const
+    {
+      p0 = vertex(vid.v+0,itime.v);
+      p1 = vertex(vid.v+1,itime.v);
+    }
+#endif
+
+     /*! loads curve vertices for specified time */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, unsigned int vid, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3ff a0,a1; gather(a0,a1,vid,itime);
+      Vec3ff b0,b1; gather(b0,b1,vid,itime+1);
+      p0 = madd(Vec3ff(t0),a0,t1*b0);
+      p1 = madd(Vec3ff(t0),a1,t1*b1);
+    }
+
+    /*! loads curve vertices for specified time for mblur and non-mblur case */
+    __forceinline void gather_safe(Vec3ff& p0, Vec3ff& p1, unsigned int vid, float time) const
+    {
+      if (hasMotionBlur()) gather(p0,p1,vid,time);
+      else                 gather(p0,p1,vid);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, const vuint<M>& vid, const vfloat<M>& time) const
+    {
+      vfloat<M> ftime;
+      const vint<M> itime = timeSegment<M>(time, ftime);
+
+      const vfloat<M> t0 = 1.0f - ftime;
+      const vfloat<M> t1 = ftime;
+      Vec4vf<M> a0,a1; vgather<M>(a0,a1,vid,itime);
+      Vec4vf<M> b0,b1; vgather<M>(b0,b1,vid,itime+1);
+      p0 = madd(Vec4vf<M>(t0),a0,t1*b0);
+      p1 = madd(Vec4vf<M>(t0),a1,t1*b1);
+    }
+#endif
+    
+    /*! gathers the cone curve starting with i'th vertex */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, bool& cL, bool& cR, unsigned int primID, unsigned int vid) const
+    {
+      gather(p0,p1,vid);
+      cL = !segmentLeftExists (primID);
+      cR = !segmentRightExists(primID);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, vbool<M>& cL, vbool<M>& cR, const vuint<M>& primID, const vuint<M>& vid) const
+    {
+      vgather<M>(p0,p1,vid);
+      cL = !segmentLeftExists (primID.v);
+      cR = !segmentRightExists(primID.v);
+    }
+#endif
+
+    /*! gathers the cone curve starting with i'th vertex of itime'th timestep */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, bool& cL, bool& cR, unsigned int primID, size_t vid, size_t itime) const
+    {
+      gather(p0,p1,vid,itime);
+      cL = !segmentLeftExists (primID);
+      cR = !segmentRightExists(primID);
+    }
+
+     /*! loads cone curve vertices for specified time */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, bool& cL, bool& cR, unsigned int primID, size_t vid, float time) const
+    {
+      gather(p0,p1,vid,time);
+      cL = !segmentLeftExists (primID);
+      cR = !segmentRightExists(primID);
+    }
+
+    /*! loads cone curve vertices for specified time for mblur and non-mblur geometry */
+    __forceinline void gather_safe(Vec3ff& p0, Vec3ff& p1, bool& cL, bool& cR, unsigned int primID, size_t vid, float time) const
+    {
+      if (hasMotionBlur()) gather(p0,p1,cL,cR,primID,vid,time);
+      else                 gather(p0,p1,cL,cR,primID,vid);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, vbool<M>& cL, vbool<M>& cR, const vuint<M>& primID, const vuint<M>& vid, const vfloat<M>& time) const
+    {
+      vgather<M>(p0,p1,vid,time);
+      cL = !segmentLeftExists (primID.v);
+      cR = !segmentRightExists(primID.v);
+    }
+#endif
+
+    /*! gathers the curve starting with i'th vertex */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, unsigned int primID, size_t vid) const
+    {
+      p0 = vertex(vid+0);
+      p1 = vertex(vid+1);
+      p2 = segmentLeftExists (primID) ? vertex(vid-1) : Vec3ff(inf);
+      p3 = segmentRightExists(primID) ? vertex(vid+2) : Vec3ff(inf);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, Vec4vf<M>& p2, Vec4vf<M>& p3, const vuint<M>& primID, const vuint<M>& vid) const
+    {
+      p0 = vertex(vid.v+0);
+      p1 = vertex(vid.v+1);
+      vbool<M> left  = segmentLeftExists (primID.v);
+      vbool<M> right = segmentRightExists(primID.v);
+      vuint<M> i2 = select(left, vid-1,vid+0);
+      vuint<M> i3 = select(right,vid+2,vid+1);
+      p2 = vertex(i2.v);
+      p3 = vertex(i3.v);
+      p2 =  select(left, p2,Vec4vf<M>(inf));
+      p3 =  select(right,p3,Vec4vf<M>(inf));
+    }
+#endif
+
+     /*! gathers the curve starting with i'th vertex of itime'th timestep */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, unsigned int primID, size_t vid, size_t itime) const
+    {
+      p0 = vertex(vid+0,itime);
+      p1 = vertex(vid+1,itime);
+      p2 = segmentLeftExists (primID) ? vertex(vid-1,itime) : Vec3ff(inf);
+      p3 = segmentRightExists(primID) ? vertex(vid+2,itime) : Vec3ff(inf);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, Vec4vf<M>& p2, Vec4vf<M>& p3, const vuint<M>& primID, const vuint<M>& vid, const vint<M>& itime) const
+    {
+      p0 = vertex(vid.v+0, itime.v);
+      p1 = vertex(vid.v+1, itime.v);
+      vbool<M> left  = segmentLeftExists (primID.v);
+      vbool<M> right = segmentRightExists(primID.v);
+      vuint<M> i2 = select(left, vid-1,vid+0);
+      vuint<M> i3 = select(right,vid+2,vid+1);
+      p2 = vertex(i2.v, itime.v);
+      p3 = vertex(i3.v, itime.v);
+      p2 =  select(left, p2,Vec4vf<M>(inf));
+      p3 =  select(right,p3,Vec4vf<M>(inf));
+    }
+#endif
+    
+     /*! loads curve vertices for specified time */
+    __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, unsigned int primID, size_t vid, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3ff a0,a1,a2,a3; gather(a0,a1,a2,a3,primID,vid,itime);
+      Vec3ff b0,b1,b2,b3; gather(b0,b1,b2,b3,primID,vid,itime+1);
+      p0 = madd(Vec3ff(t0),a0,t1*b0);
+      p1 = madd(Vec3ff(t0),a1,t1*b1);
+      p2 = madd(Vec3ff(t0),a2,t1*b2);
+      p3 = madd(Vec3ff(t0),a3,t1*b3);
+    }
+
+    /*! loads curve vertices for specified time for mblur and non-mblur geometry */
+    __forceinline void gather_safe(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, unsigned int primID, size_t vid, float time) const
+    {
+      if (hasMotionBlur()) gather(p0,p1,p2,p3,primID,vid,time);
+      else                 gather(p0,p1,p2,p3,primID,vid);
+    }
+
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    template<int M>
+    __forceinline void vgather(Vec4vf<M>& p0, Vec4vf<M>& p1, Vec4vf<M>& p2, Vec4vf<M>& p3, const vuint<M>& primID, const vuint<M>& vid, const vfloat<M>& time) const
+    {
+      vfloat<M> ftime;
+      const vint<M> itime = timeSegment<M>(time, ftime);
+
+      const vfloat<M> t0 = 1.0f - ftime;
+      const vfloat<M> t1 = ftime;
+      Vec4vf<M> a0,a1,a2,a3; vgather<M>(a0,a1,a2,a3,primID,vid,itime);
+      Vec4vf<M> b0,b1,b2,b3; vgather<M>(b0,b1,b2,b3,primID,vid,itime+1);
+      p0 = madd(Vec4vf<M>(t0),a0,t1*b0);
+      p1 = madd(Vec4vf<M>(t0),a1,t1*b1);
+      p2 = madd(Vec4vf<M>(t0),a2,t1*b2);
+      p3 = madd(Vec4vf<M>(t0),a3,t1*b3);
+    }
+#endif
+    
     /*! calculates bounding box of i'th line segment */
     __forceinline BBox3fa bounds(const Vec3ff& v0, const Vec3ff& v1) const
     {
@@ -183,6 +404,18 @@ namespace embree
       return bounds(w0,w1);
     }
 
+    /*! calculates bounding box of i'th segment */
+    __forceinline BBox3fa bounds(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t i, size_t itime = 0) const
+    {
+      const float r_scale = r_scale0*scale;
+      const unsigned int index = segment(i);
+      const Vec3ff v0 = vertex(index+0,itime);
+      const Vec3ff v1 = vertex(index+1,itime);
+      const Vec3ff w0(xfmVector(space,(v0-ofs)*Vec3fa(scale)),maxRadiusScale*v0.w*r_scale);
+      const Vec3ff w1(xfmVector(space,(v1-ofs)*Vec3fa(scale)),maxRadiusScale*v1.w*r_scale);
+      return bounds(w0,w1);
+    }     
+
     /*! check if the i'th primitive is valid at the itime'th timestep */
     __forceinline bool valid(size_t i, size_t itime) const {
       return valid(i, make_range(itime, itime));
@@ -193,13 +426,17 @@ namespace embree
     {
       const unsigned int index = segment(i);
       if (index+1 >= numVertices()) return false;
-      
+
+#if !defined(__SYCL_DEVICE_ONLY__)
+
       for (size_t itime = itime_range.begin(); itime <= itime_range.end(); itime++)
       {
         const Vec3ff v0 = vertex(index+0,itime); if (unlikely(!isvalid4(v0))) return false;
         const Vec3ff v1 = vertex(index+1,itime); if (unlikely(!isvalid4(v1))) return false;
         if (min(v0.w,v1.w) < 0.0f) return false;
       }
+#endif
+      
       return true;
     }
 
@@ -235,6 +472,11 @@ namespace embree
     }
 
     /*! calculates the linear bounds of the i'th primitive for the specified time range */
+    __forceinline LBBox3fa linearBounds(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t primID, const BBox1f& dt) const {
+      return LBBox3fa([&] (size_t itime) { return bounds(ofs, scale, r_scale0, space, primID, itime); }, dt, this->time_range, fnumTimeSegments);
+    }
+    
+    /*! calculates the linear bounds of the i'th primitive for the specified time range */
     __forceinline bool linearBounds(size_t i, const BBox1f& time_range, LBBox3fa& bbox) const
     {
       if (!valid(i, timeSegmentRange(time_range))) return false;
@@ -252,9 +494,9 @@ namespace embree
     BufferView<Vec3ff> vertices0;           //!< fast access to first vertex buffer
     BufferView<Vec3fa> normals0;            //!< fast access to first normal buffer
     BufferView<char> flags;                 //!< start, end flag per segment
-    vector<BufferView<Vec3ff>> vertices;    //!< vertex array for each timestep
-    vector<BufferView<Vec3fa>> normals;     //!< normal array for each timestep
-    vector<BufferView<char>> vertexAttribs; //!< user buffers
+    Device::vector<BufferView<Vec3ff>> vertices = device;    //!< vertex array for each timestep
+    Device::vector<BufferView<Vec3fa>> normals = device;     //!< normal array for each timestep
+    Device::vector<BufferView<char>> vertexAttribs = device; //!< user buffers
     int tessellationRate;                   //!< tessellation rate for bezier curve
     float maxRadiusScale = 1.0;             //!< maximal min-width scaling of curve radii
   };
@@ -266,6 +508,28 @@ namespace embree
       LineSegmentsISA (Device* device, Geometry::GType gtype)
         : LineSegments(device,gtype) {}
 
+      LinearSpace3fa computeAlignedSpace(const size_t primID) const
+      {
+        const Vec3fa dir = normalize(computeDirection(primID));
+        if (is_finite(dir)) return frame(dir);
+        else return LinearSpace3fa(one);
+      }
+
+      LinearSpace3fa computeAlignedSpaceMB(const size_t primID, const BBox1f time_range) const
+      {
+        Vec3fa axisz(0,0,1);
+        Vec3fa axisy(0,1,0);
+
+        const range<int> tbounds = this->timeSegmentRange(time_range);
+        if (tbounds.size() == 0) return frame(axisz);
+        
+        const size_t itime = (tbounds.begin()+tbounds.end())/2;
+
+        const Vec3fa dir = normalize(computeDirection(primID,itime));
+        if (is_finite(dir)) return frame(dir);
+        else return LinearSpace3fa(one);
+      }     
+
       Vec3fa computeDirection(unsigned int primID) const
       {
         const unsigned vtxID = segment(primID);
@@ -282,7 +546,7 @@ namespace embree
         return v1-v0;
       }
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfo pinfo(empty);
         for (size_t j=r.begin(); j<r.end(); j++)
@@ -309,7 +573,24 @@ namespace embree
         }
         return pinfo;
       }
-      
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+        
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = empty;
+          if (!linearBounds(j, t0t1, lbounds))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfoMB pinfo(empty);
@@ -331,6 +612,10 @@ namespace embree
         return bounds(space,i);
       }
 
+       BBox3fa vbounds(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t i, size_t itime = 0) const {
+        return bounds(ofs,scale,r_scale0,space,i,itime);
+      }
+
       LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
         return linearBounds(primID,time_range);
       }
@@ -338,6 +623,10 @@ namespace embree
       LBBox3fa vlinearBounds(const LinearSpace3fa& space, size_t primID, const BBox1f& time_range) const {
         return linearBounds(space,primID,time_range);
       }
+
+       LBBox3fa vlinearBounds(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t primID, const BBox1f& time_range) const {
+        return linearBounds(ofs,scale,r_scale0,space,primID,time_range);
+      }
     };
   }
 
diff --git a/thirdparty/embree/kernels/common/scene_points.h b/thirdparty/embree/kernels/common/scene_points.h
index 017e098a51..937a8f1806 100644
--- a/thirdparty/embree/kernels/common/scene_points.h
+++ b/thirdparty/embree/kernels/common/scene_points.h
@@ -68,6 +68,25 @@ namespace embree
       return vertices[itime][i];
     }
 
+    /*! returns i'th vertex of for specified time */
+    __forceinline Vec3ff vertex(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3ff v0 = vertex(i, itime+0);
+      Vec3ff v1 = vertex(i, itime+1);
+      return madd(Vec3ff(t0),v0,t1*v1);
+    }
+
+    /*! returns i'th vertex of for specified time */
+    __forceinline Vec3ff vertex_safe(size_t i, float time) const
+    {
+      if (hasMotionBlur()) return vertex(i,time);
+      else                 return vertex(i);
+    }
+
     /*! returns i'th vertex of itime'th timestep */
     __forceinline const char* vertexPtr(size_t i, size_t itime) const {
       return vertices[itime].getPtr(i);
@@ -78,11 +97,49 @@ namespace embree
       return normals[itime][i];
     }
 
+    /*! returns i'th normal of for specified time */
+    __forceinline Vec3fa normal(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3fa n0 = normal(i, itime+0);
+      Vec3fa n1 = normal(i, itime+1);
+      return madd(Vec3fa(t0),n0,t1*n1);
+    }
+
+    /*! returns i'th normal of for specified time */
+    __forceinline Vec3fa normal_safe(size_t i, float time) const
+    {
+      if (hasMotionBlur()) return normal(i,time);
+      else                 return normal(i);
+    }
+
     /*! returns i'th radius of itime'th timestep */
     __forceinline float radius(size_t i, size_t itime) const {
       return vertices[itime][i].w;
     }
 
+    /*! returns i'th radius of for specified time */
+    __forceinline float radius(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      float r0 = radius(i, itime+0);
+      float r1 = radius(i, itime+1);
+      return madd(t0,r0,t1*r1);
+    }
+
+    /*! returns i'th radius of for specified time */
+    __forceinline float radius_safe(size_t i, float time) const
+    {
+      if (hasMotionBlur()) return radius(i,time);
+      else                 return radius(i);
+    }
+
     /*! calculates bounding box of i'th line segment */
     __forceinline BBox3fa bounds(const Vec3ff& v0) const {
       return enlarge(BBox3fa(v0), maxRadiusScale*Vec3fa(v0.w));
@@ -185,13 +242,18 @@ namespace embree
     __forceinline float * getCompactVertexArray () const {
       return (float*) vertices0.getPtr();
     }
+    
+    __forceinline float projectedPrimitiveArea(const size_t i) const {
+      const float R = radius(i);
+      return 1 + 2*M_PI*R*R;
+    }
 
    public:
     BufferView<Vec3ff> vertices0;            //!< fast access to first vertex buffer
     BufferView<Vec3fa> normals0;             //!< fast access to first normal buffer
-    vector<BufferView<Vec3ff>> vertices;     //!< vertex array for each timestep
-    vector<BufferView<Vec3fa>> normals;      //!< normal array for each timestep
-    vector<BufferView<char>> vertexAttribs;  //!< user buffers
+    Device::vector<BufferView<Vec3ff>> vertices = device;     //!< vertex array for each timestep
+    Device::vector<BufferView<Vec3fa>> normals = device;      //!< normal array for each timestep
+    Device::vector<BufferView<char>> vertexAttribs = device;  //!< user buffers
     float maxRadiusScale = 1.0;              //!< maximal min-width scaling of curve radii
   };
 
@@ -211,7 +273,7 @@ namespace embree
         return Vec3fa(1, 0, 0);
       }
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfo pinfo(empty);
         for (size_t j = r.begin(); j < r.end(); j++) {
@@ -239,6 +301,23 @@ namespace embree
         return pinfo;
       }
 
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+        
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = empty;
+          if (!linearBounds(j, t0t1, lbounds))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims,
                                       const BBox1f& t0t1,
                                       const range<size_t>& r,
diff --git a/thirdparty/embree/kernels/common/scene_quad_mesh.h b/thirdparty/embree/kernels/common/scene_quad_mesh.h
index bd8eeaaeb7..09a8b8ddd9 100644
--- a/thirdparty/embree/kernels/common/scene_quad_mesh.h
+++ b/thirdparty/embree/kernels/common/scene_quad_mesh.h
@@ -17,12 +17,18 @@ namespace embree
     /*! triangle indices */
     struct Quad
     {
-      uint32_t v[4];
+      Quad() {}
+
+      Quad (uint32_t v0, uint32_t v1, uint32_t v2, uint32_t v3) {
+        v[0] = v0; v[1] = v1; v[2] = v2; v[3] = v3;
+      }
 
       /*! outputs triangle indices */
       __forceinline friend embree_ostream operator<<(embree_ostream cout, const Quad& q) {
         return cout << "Quad {" << q.v[0] << ", " << q.v[1] << ", " << q.v[2] << ", " << q.v[3] << " }";
       }
+
+      uint32_t v[4];
     };
 
   public:
@@ -135,6 +141,18 @@ namespace embree
       return vertices[itime].getPtr(i);
     }
 
+    /*! returns i'th vertex of for specified time */
+    __forceinline Vec3fa vertex(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3fa v0 = vertex(i, itime+0);
+      Vec3fa v1 = vertex(i, itime+1);
+      return madd(Vec3fa(t0),v0,t1*v1);
+    }
+
     /*! calculates the bounds of the i'th quad */
     __forceinline BBox3fa bounds(size_t i) const 
     {
@@ -196,7 +214,7 @@ namespace embree
       if (q.v[2] >= numVertices()) return false;
       if (q.v[3] >= numVertices()) return false;
 
-      for (unsigned int t=0; t<numTimeSteps; t++)
+      for (size_t t=0; t<numTimeSteps; t++)
       {
         const Vec3fa v0 = vertex(q.v[0],t);
         const Vec3fa v1 = vertex(q.v[1],t);
@@ -279,8 +297,8 @@ namespace embree
   public:
     BufferView<Quad> quads;                 //!< array of quads
     BufferView<Vec3fa> vertices0;           //!< fast access to first vertex buffer
-    vector<BufferView<Vec3fa>> vertices;    //!< vertex array for each timestep
-    vector<BufferView<char>> vertexAttribs; //!< vertex attribute buffers
+    Device::vector<BufferView<Vec3fa>> vertices = device; //!< vertex array for each timestep
+    Device::vector<RawBufferView> vertexAttribs = device; //!< vertex attribute buffers
   };
 
   namespace isa
@@ -290,7 +308,11 @@ namespace embree
       QuadMeshISA (Device* device)
         : QuadMesh(device) {}
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
+        return linearBounds(primID,time_range);
+      }
+
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfo pinfo(empty);
         for (size_t j=r.begin(); j<r.end(); j++)
@@ -317,7 +339,24 @@ namespace embree
         }
         return pinfo;
       }
-      
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+                
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = empty;
+          if (!linearBounds(j, t0t1, lbounds))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfoMB pinfo(empty);
diff --git a/thirdparty/embree/kernels/common/scene_subdiv_mesh.h b/thirdparty/embree/kernels/common/scene_subdiv_mesh.h
index 1db170196d..b213a9b7ba 100644
--- a/thirdparty/embree/kernels/common/scene_subdiv_mesh.h
+++ b/thirdparty/embree/kernels/common/scene_subdiv_mesh.h
@@ -9,11 +9,13 @@
 #include "../subdiv/tessellation_cache.h"
 #include "../subdiv/catmullclark_coefficients.h"
 #include "../subdiv/patch.h"
-#include "../../common/algorithms/parallel_map.h"
-#include "../../common/algorithms/parallel_set.h"
 
 namespace embree
 {
+  struct HoleSet;
+  struct VertexCreaseMap;
+  struct EdgeCreaseMap;
+
   class SubdivMesh : public Geometry
   {
     ALIGNED_CLASS_(16);
@@ -49,6 +51,7 @@ namespace embree
 
     /*! subdiv mesh construction */
     SubdivMesh(Device* device);
+    ~SubdivMesh();
 
   public:
     void setMask (unsigned mask);
@@ -272,7 +275,7 @@ namespace embree
     mvector<uint32_t> halfEdgeFace;
 
     /*! set with all holes */
-    parallel_set<uint32_t> holeSet;
+    std::unique_ptr<HoleSet> holeSet;
 
     /*! fast lookup table to detect invalid faces */
     mvector<char> invalid_face;
@@ -299,10 +302,10 @@ namespace embree
   private:
 
     /*! map with all vertex creases */
-    parallel_map<uint32_t,float> vertexCreaseMap;
+    std::unique_ptr<VertexCreaseMap> vertexCreaseMap;
     
     /*! map with all edge creases */
-    parallel_map<uint64_t,float> edgeCreaseMap;
+    std::unique_ptr<EdgeCreaseMap> edgeCreaseMap;
 
   protected:
     
diff --git a/thirdparty/embree/kernels/common/scene_triangle_mesh.cpp b/thirdparty/embree/kernels/common/scene_triangle_mesh.cpp
index 3bbd7e51ae..6cdd542a65 100644
--- a/thirdparty/embree/kernels/common/scene_triangle_mesh.cpp
+++ b/thirdparty/embree/kernels/common/scene_triangle_mesh.cpp
@@ -134,7 +134,7 @@ namespace embree
     Geometry::update();
   }
 
-  void TriangleMesh::commit() 
+  void TriangleMesh::commit()
   {
     /* verify that stride of all time steps are identical */
     for (unsigned int t=0; t<numTimeSteps; t++)
diff --git a/thirdparty/embree/kernels/common/scene_triangle_mesh.h b/thirdparty/embree/kernels/common/scene_triangle_mesh.h
index ad3f602fde..0d28219b96 100644
--- a/thirdparty/embree/kernels/common/scene_triangle_mesh.h
+++ b/thirdparty/embree/kernels/common/scene_triangle_mesh.h
@@ -129,6 +129,18 @@ namespace embree
       return vertices[itime].getPtr(i);
     }
 
+    /*! returns i'th vertex of for specified time */
+    __forceinline Vec3fa vertex(size_t i, float time) const
+    {
+      float ftime;
+      const size_t itime = timeSegment(time, ftime);
+      const float t0 = 1.0f - ftime;
+      const float t1 = ftime;
+      Vec3fa v0 = vertex(i, itime+0);
+      Vec3fa v1 = vertex(i, itime+1);
+      return madd(Vec3fa(t0),v0,t1*v1);
+    }
+
     /*! calculates the bounds of the i'th triangle */
     __forceinline BBox3fa bounds(size_t i) const 
     {
@@ -260,8 +272,8 @@ namespace embree
   public:
     BufferView<Triangle> triangles;      //!< array of triangles
     BufferView<Vec3fa> vertices0;        //!< fast access to first vertex buffer
-    vector<BufferView<Vec3fa>> vertices; //!< vertex array for each timestep
-    vector<RawBufferView> vertexAttribs; //!< vertex attributes
+    Device::vector<BufferView<Vec3fa>> vertices = device; //!< vertex array for each timestep
+    Device::vector<RawBufferView> vertexAttribs = device; //!< vertex attributes
   };
 
   namespace isa
@@ -271,7 +283,11 @@ namespace embree
       TriangleMeshISA (Device* device)
         : TriangleMesh(device) {}
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      LBBox3fa vlinearBounds(size_t primID, const BBox1f& time_range) const {
+        return linearBounds(primID,time_range);
+      }
+
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfo pinfo(empty);
         for (size_t j=r.begin(); j<r.end(); j++)
@@ -298,7 +314,24 @@ namespace embree
         }
         return pinfo;
       }
-      
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+        
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = empty;
+          if (!linearBounds(j, t0t1, lbounds))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfoMB pinfo(empty);
diff --git a/thirdparty/embree/kernels/common/scene_user_geometry.h b/thirdparty/embree/kernels/common/scene_user_geometry.h
index 2867b18b79..033476f658 100644
--- a/thirdparty/embree/kernels/common/scene_user_geometry.h
+++ b/thirdparty/embree/kernels/common/scene_user_geometry.h
@@ -21,6 +21,8 @@ namespace embree
     virtual void setOccludedFunctionN (RTCOccludedFunctionN occluded);
     virtual void build() {}
     virtual void addElementsToCount (GeometryCounts & counts) const;
+
+    __forceinline float projectedPrimitiveArea(const size_t i) const { return 0.0f; }
   };
 
   namespace isa
@@ -30,7 +32,7 @@ namespace embree
       UserGeometryISA (Device* device)
         : UserGeometry(device) {}
 
-      PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const
+      PrimInfo createPrimRefArray(PrimRef* prims, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfo pinfo(empty);
         for (size_t j=r.begin(); j<r.end(); j++)
@@ -57,7 +59,24 @@ namespace embree
         }
         return pinfo;
       }
-      
+
+      PrimInfo createPrimRefArrayMB(PrimRef* prims, const BBox1f& time_range, const range<size_t>& r, size_t k, unsigned int geomID) const
+      {
+        PrimInfo pinfo(empty);
+        const BBox1f t0t1 = BBox1f::intersect(getTimeRange(), time_range);
+        if (t0t1.empty()) return pinfo;
+        
+        for (size_t j = r.begin(); j < r.end(); j++) {
+          LBBox3fa lbounds = empty;
+          if (!linearBounds(j, t0t1, lbounds))
+            continue;
+          const PrimRef prim(lbounds.bounds(), geomID, unsigned(j));
+          pinfo.add_center2(prim);
+          prims[k++] = prim;
+        }
+        return pinfo;
+      }
+
       PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const
       {
         PrimInfoMB pinfo(empty);
diff --git a/thirdparty/embree/kernels/common/scene_verify.cpp b/thirdparty/embree/kernels/common/scene_verify.cpp
new file mode 100644
index 0000000000..1db7844f4f
--- /dev/null
+++ b/thirdparty/embree/kernels/common/scene_verify.cpp
@@ -0,0 +1,24 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "scene.h"
+
+#include "../../common/algorithms/parallel_any_of.h"
+
+namespace embree
+{
+
+void Scene::checkIfModifiedAndSet ()
+{
+  if (isModified ()) return;
+
+  auto geometryIsModified = [this](size_t geomID)->bool {
+    return isGeometryModified(geomID);
+  };
+
+  if (parallel_any_of (size_t(0), geometries.size (), geometryIsModified)) {
+    setModified ();
+  }
+}
+
+}
\ No newline at end of file
diff --git a/thirdparty/embree/kernels/common/stat.cpp b/thirdparty/embree/kernels/common/stat.cpp
index ebb77cd534..9a8c8fac4e 100644
--- a/thirdparty/embree/kernels/common/stat.cpp
+++ b/thirdparty/embree/kernels/common/stat.cpp
@@ -17,7 +17,7 @@ namespace embree
 #endif
   }
 
-  void Stat::print(std::ostream& cout)
+  void Stat::print(embree_ostream cout)
   {
     Counters& cntrs = instance.cntrs;
     Counters::Data& data = instance.cntrs.code;
diff --git a/thirdparty/embree/kernels/common/state.cpp b/thirdparty/embree/kernels/common/state.cpp
index db6b803041..1d73ae9629 100644
--- a/thirdparty/embree/kernels/common/state.cpp
+++ b/thirdparty/embree/kernels/common/state.cpp
@@ -192,10 +192,17 @@ namespace embree
   const char* symbols[3] = { "=", ",", "|" };
 
   bool State::parseFile(const FileName& fileName)
-  {
-    FILE* f = fopen(fileName.c_str(),"r");
-    if (!f) return false;
-    Ref<Stream<int> > file = new FileStream(f,fileName);
+  { 
+    Ref<Stream<int> > file;
+    // -- GODOT start --
+    // try {
+      file = new FileStream(fileName);
+    // }
+    // catch (std::runtime_error& e) {
+    //   (void) e;
+    //   return false;
+    // }
+    // -- GODOT end --
     
     std::vector<std::string> syms;
     for (size_t i=0; i<sizeof(symbols)/sizeof(void*); i++) 
@@ -393,7 +400,7 @@ namespace embree
         grid_accel = cin->get().Identifier();
       else if (tok == Token::Id("grid_accel_mb") && cin->trySymbol("="))
         grid_accel_mb = cin->get().Identifier();
-      
+
       else if (tok == Token::Id("verbose") && cin->trySymbol("="))
         verbose = cin->get().Int();
       else if (tok == Token::Id("benchmark") && cin->trySymbol("="))
@@ -419,7 +426,7 @@ namespace embree
           } while (cin->trySymbol("|"));
         }
       }
-      
+
       else if (tok == Token::Id("max_spatial_split_replications") && cin->trySymbol("="))
         max_spatial_split_replications = cin->get().Float();
 
diff --git a/thirdparty/embree/kernels/common/state.h b/thirdparty/embree/kernels/common/state.h
index 33bcc843b2..8c34614185 100644
--- a/thirdparty/embree/kernels/common/state.h
+++ b/thirdparty/embree/kernels/common/state.h
@@ -189,7 +189,7 @@ namespace embree
       memory_monitor_function = fptr;
       memory_monitor_userptr = uptr;
     }
-      
+
     RTCMemoryMonitorFunction memory_monitor_function;
     void* memory_monitor_userptr;
   };