Add Embree-aarch64 thirdparty library

26 files changed, 4551 insertions, 0 deletions

diff --git a/thirdparty/embree-aarch64/common/sys/alloc.cpp b/thirdparty/embree-aarch64/common/sys/alloc.cpp
new file mode 100644
index 0000000000..12f143f131
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/alloc.cpp
@@ -0,0 +1,327 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "alloc.h"
+#include "intrinsics.h"
+#include "sysinfo.h"
+#include "mutex.h"
+
+////////////////////////////////////////////////////////////////////////////////
+/// All Platforms
+////////////////////////////////////////////////////////////////////////////////
+  
+namespace embree
+{
+  void* alignedMalloc(size_t size, size_t align) 
+  {
+    if (size == 0)
+      return nullptr;
+    
+    assert((align & (align-1)) == 0);
+    void* ptr = _mm_malloc(size,align);
+
+    if (size != 0 && ptr == nullptr)
+      // -- GODOT start --
+      // throw std::bad_alloc();
+      abort(); 
+      // -- GODOT end --
+    
+    return ptr;
+  }
+  
+  void alignedFree(void* ptr)
+  {
+    if (ptr)
+      _mm_free(ptr);
+  }
+
+  static bool huge_pages_enabled = false;
+  static MutexSys os_init_mutex;
+
+  __forceinline bool isHugePageCandidate(const size_t bytes) 
+  {
+    if (!huge_pages_enabled)
+      return false;
+
+    /* use huge pages only when memory overhead is low */
+    const size_t hbytes = (bytes+PAGE_SIZE_2M-1) & ~size_t(PAGE_SIZE_2M-1);
+    return 66*(hbytes-bytes) < bytes; // at most 1.5% overhead
+  }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+/// Windows Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#ifdef _WIN32
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include <malloc.h>
+
+namespace embree
+{
+  bool win_enable_selockmemoryprivilege (bool verbose)
+  {
+    HANDLE hToken;
+    if (!OpenProcessToken(GetCurrentProcess(), TOKEN_QUERY | TOKEN_ADJUST_PRIVILEGES, &hToken)) {
+      if (verbose) std::cout << "WARNING: OpenProcessToken failed while trying to enable SeLockMemoryPrivilege: " << GetLastError() << std::endl;
+      return false;
+    }
+
+    TOKEN_PRIVILEGES tp;
+    tp.PrivilegeCount = 1;
+    tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
+
+    if (!LookupPrivilegeValueW(nullptr, L"SeLockMemoryPrivilege", &tp.Privileges[0].Luid)) {
+      if (verbose) std::cout << "WARNING: LookupPrivilegeValue failed while trying to enable SeLockMemoryPrivilege: " << GetLastError() << std::endl;
+      return false;
+    }
+    
+    SetLastError(ERROR_SUCCESS);
+    if (!AdjustTokenPrivileges(hToken, FALSE, &tp, sizeof(tp), nullptr, 0)) {
+      if (verbose) std::cout << "WARNING: AdjustTokenPrivileges failed while trying to enable SeLockMemoryPrivilege" << std::endl;
+      return false;
+    }
+    
+    if (GetLastError() == ERROR_NOT_ALL_ASSIGNED) {
+      if (verbose) std::cout << "WARNING: AdjustTokenPrivileges failed to enable SeLockMemoryPrivilege: Add SeLockMemoryPrivilege for current user and run process in elevated mode (Run as administrator)." << std::endl;
+      return false;
+    } 
+
+    return true;
+  }
+
+  bool os_init(bool hugepages, bool verbose) 
+  {
+    Lock<MutexSys> lock(os_init_mutex);
+
+    if (!hugepages) {
+      huge_pages_enabled = false;
+      return true;
+    }
+
+    if (GetLargePageMinimum() != PAGE_SIZE_2M) {
+      huge_pages_enabled = false;
+      return false;
+    }
+
+    huge_pages_enabled = true;
+    return true;
+  }
+
+  void* os_malloc(size_t bytes, bool& hugepages)
+  {
+    if (bytes == 0) {
+      hugepages = false;
+      return nullptr;
+    }
+
+    /* try direct huge page allocation first */
+    if (isHugePageCandidate(bytes)) 
+    {
+      int flags = MEM_COMMIT | MEM_RESERVE | MEM_LARGE_PAGES;
+      char* ptr = (char*) VirtualAlloc(nullptr,bytes,flags,PAGE_READWRITE);
+      if (ptr != nullptr) {
+        hugepages = true;
+        return ptr;
+      }
+    } 
+
+    /* fall back to 4k pages */
+    int flags = MEM_COMMIT | MEM_RESERVE;
+    char* ptr = (char*) VirtualAlloc(nullptr,bytes,flags,PAGE_READWRITE);
+    // -- GODOT start --
+    // if (ptr == nullptr) throw std::bad_alloc();
+    if (ptr == nullptr) abort();
+    // -- GODOT end --
+    hugepages = false;
+    return ptr;
+  }
+
+  size_t os_shrink(void* ptr, size_t bytesNew, size_t bytesOld, bool hugepages) 
+  {
+    if (hugepages) // decommitting huge pages seems not to work under Windows
+      return bytesOld;
+
+    const size_t pageSize = hugepages ? PAGE_SIZE_2M : PAGE_SIZE_4K;
+    bytesNew = (bytesNew+pageSize-1) & ~(pageSize-1);
+    bytesOld = (bytesOld+pageSize-1) & ~(pageSize-1);
+    if (bytesNew >= bytesOld)
+      return bytesOld;
+
+    if (!VirtualFree((char*)ptr+bytesNew,bytesOld-bytesNew,MEM_DECOMMIT))
+      // -- GODOT start --
+      // throw std::bad_alloc();
+      abort();
+      // -- GODOT end --
+
+    return bytesNew;
+  }
+
+  void os_free(void* ptr, size_t bytes, bool hugepages) 
+  {
+    if (bytes == 0) 
+      return;
+
+    if (!VirtualFree(ptr,0,MEM_RELEASE))
+      // -- GODOT start --
+      // throw std::bad_alloc();
+      abort();
+      // -- GODOT end --
+  }
+
+  void os_advise(void *ptr, size_t bytes)
+  {
+  }
+}
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Unix Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__UNIX__)
+
+#include <sys/mman.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sstream>
+
+#if defined(__MACOSX__)
+#include <mach/vm_statistics.h>
+#endif
+
+namespace embree
+{
+  bool os_init(bool hugepages, bool verbose) 
+  {
+    Lock<MutexSys> lock(os_init_mutex);
+
+    if (!hugepages) {
+      huge_pages_enabled = false;
+      return true;
+    }
+
+#if defined(__LINUX__)
+
+    int hugepagesize = 0;
+
+    std::ifstream file; 
+    file.open("/proc/meminfo",std::ios::in);
+    if (!file.is_open()) {
+      if (verbose) std::cout << "WARNING: Could not open /proc/meminfo. Huge page support cannot get enabled!" << std::endl;
+      huge_pages_enabled = false;
+      return false;
+    }
+    
+    std::string line;
+    while (getline(file,line))
+    {
+      std::stringstream sline(line);
+      while (!sline.eof() && sline.peek() == ' ') sline.ignore();
+      std::string tag; getline(sline,tag,' ');
+      while (!sline.eof() && sline.peek() == ' ') sline.ignore();
+      std::string val; getline(sline,val,' ');
+      while (!sline.eof() && sline.peek() == ' ') sline.ignore();
+      std::string unit; getline(sline,unit,' ');
+      if (tag == "Hugepagesize:" && unit == "kB") {
+	hugepagesize = std::stoi(val)*1024;
+	break;
+      }
+    }
+    
+    if (hugepagesize != PAGE_SIZE_2M) 
+    {
+      if (verbose) std::cout << "WARNING: Only 2MB huge pages supported. Huge page support cannot get enabled!" << std::endl;
+      huge_pages_enabled = false;
+      return false;
+    }
+#endif
+
+    huge_pages_enabled = true;
+    return true;
+  }
+
+  void* os_malloc(size_t bytes, bool& hugepages)
+  { 
+    if (bytes == 0) {
+      hugepages = false;
+      return nullptr;
+    }
+
+    /* try direct huge page allocation first */
+    if (isHugePageCandidate(bytes)) 
+    {
+#if defined(__MACOSX__)
+      void* ptr = mmap(0, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, VM_FLAGS_SUPERPAGE_SIZE_2MB, 0);
+      if (ptr != MAP_FAILED) {
+        hugepages = true;
+        return ptr;
+      }
+#elif defined(MAP_HUGETLB)
+      void* ptr = mmap(0, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON | MAP_HUGETLB, -1, 0);
+      if (ptr != MAP_FAILED) {
+        hugepages = true;
+        return ptr;
+      }
+#endif
+    } 
+
+    /* fallback to 4k pages */
+    void* ptr = (char*) mmap(0, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
+    // -- GODOT start --
+    // if (ptr == MAP_FAILED) throw std::bad_alloc();
+    if (ptr == MAP_FAILED) abort();
+    // -- GODOT end --
+    hugepages = false;
+
+    /* advise huge page hint for THP */
+    os_advise(ptr,bytes);
+    return ptr;
+  }
+
+  size_t os_shrink(void* ptr, size_t bytesNew, size_t bytesOld, bool hugepages) 
+  {
+    const size_t pageSize = hugepages ? PAGE_SIZE_2M : PAGE_SIZE_4K;
+    bytesNew = (bytesNew+pageSize-1) & ~(pageSize-1);
+    bytesOld = (bytesOld+pageSize-1) & ~(pageSize-1);
+    if (bytesNew >= bytesOld)
+      return bytesOld;
+
+    if (munmap((char*)ptr+bytesNew,bytesOld-bytesNew) == -1)
+      // -- GODOT start --
+      // throw std::bad_alloc();
+      abort();
+      // -- GODOT end --
+
+    return bytesNew;
+  }
+
+  void os_free(void* ptr, size_t bytes, bool hugepages) 
+  {
+    if (bytes == 0)
+      return;
+
+    /* for hugepages we need to also align the size */
+    const size_t pageSize = hugepages ? PAGE_SIZE_2M : PAGE_SIZE_4K;
+    bytes = (bytes+pageSize-1) & ~(pageSize-1);
+    if (munmap(ptr,bytes) == -1)
+      // -- GODOT start --
+      // throw std::bad_alloc();
+      abort();
+      // -- GODOT end --
+  }
+
+  /* hint for transparent huge pages (THP) */
+  void os_advise(void* pptr, size_t bytes)
+  {
+#if defined(MADV_HUGEPAGE)
+    madvise(pptr,bytes,MADV_HUGEPAGE); 
+#endif
+  }
+}
+
+#endif
diff --git a/thirdparty/embree-aarch64/common/sys/alloc.h b/thirdparty/embree-aarch64/common/sys/alloc.h
new file mode 100644
index 0000000000..5898ecda70
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/alloc.h
@@ -0,0 +1,164 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+#include <vector>
+#include <set>
+
+namespace embree
+{
+#define ALIGNED_STRUCT_(align)                                           \
+  void* operator new(size_t size) { return alignedMalloc(size,align); } \
+  void operator delete(void* ptr) { alignedFree(ptr); }                 \
+  void* operator new[](size_t size) { return alignedMalloc(size,align); } \
+  void operator delete[](void* ptr) { alignedFree(ptr); }
+
+#define ALIGNED_CLASS_(align)                                           \
+ public:                                                               \
+    ALIGNED_STRUCT_(align)                                              \
+ private:
+  
+  /*! aligned allocation */
+  void* alignedMalloc(size_t size, size_t align);
+  void alignedFree(void* ptr);
+  
+  /*! allocator that performs aligned allocations */
+  template<typename T, size_t alignment>
+    struct aligned_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;
+
+      __forceinline pointer allocate( size_type n ) {
+        return (pointer) alignedMalloc(n*sizeof(value_type),alignment);
+      }
+
+      __forceinline void deallocate( pointer p, size_type n ) {
+        return alignedFree(p);
+      }
+
+      __forceinline void construct( pointer p, const_reference val ) {
+        new (p) T(val);
+      }
+
+      __forceinline void destroy( pointer p ) {
+        p->~T();
+      }
+    };
+
+  /*! allocates pages directly from OS */
+  bool win_enable_selockmemoryprivilege(bool verbose);
+  bool os_init(bool hugepages, bool verbose);
+  void* os_malloc (size_t bytes, bool& hugepages);
+  size_t os_shrink (void* ptr, size_t bytesNew, size_t bytesOld, bool hugepages);
+  void  os_free   (void* ptr, size_t bytes, bool hugepages);
+  void  os_advise (void* ptr, size_t bytes);
+
+  /*! allocator that performs OS allocations */
+  template<typename T>
+    struct os_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;
+
+      __forceinline os_allocator () 
+        : hugepages(false) {}
+
+      __forceinline pointer allocate( size_type n ) {
+        return (pointer) os_malloc(n*sizeof(value_type),hugepages);
+      }
+
+      __forceinline void deallocate( pointer p, size_type n ) {
+        return os_free(p,n*sizeof(value_type),hugepages);
+      }
+
+      __forceinline void construct( pointer p, const_reference val ) {
+        new (p) T(val);
+      }
+
+      __forceinline void destroy( pointer p ) {
+        p->~T();
+      }
+
+      bool hugepages;
+    };
+
+  /*! allocator for IDs */
+  template<typename T, size_t max_id>
+    struct IDPool
+    {
+      typedef T value_type;
+
+      IDPool ()
+      : nextID(0) {}
+
+      T allocate() 
+      {
+        /* return ID from list */
+        if (!IDs.empty()) 
+        {
+          T id = *IDs.begin();
+          IDs.erase(IDs.begin());
+          return id;
+        } 
+
+        /* allocate new ID */
+        else
+        {
+          if (size_t(nextID)+1 > max_id)
+            return -1;
+          
+          return nextID++;
+        }
+      }
+
+      /* adds an ID provided by the user */
+      bool add(T id)
+      {
+        if (id > max_id)
+          return false;
+        
+        /* check if ID should be in IDs set */
+        if (id < nextID) {
+          auto p = IDs.find(id);
+          if (p == IDs.end()) return false;
+          IDs.erase(p);
+          return true;
+        }
+
+        /* otherwise increase ID set */
+        else
+        {
+          for (T i=nextID; i<id; i++) {
+            IDs.insert(i);
+          }
+          nextID = id+1;
+          return true;
+        }
+      }
+
+      void deallocate( T id ) 
+      {
+        assert(id < nextID);
+        MAYBE_UNUSED auto done = IDs.insert(id).second;
+        assert(done);
+      }
+
+    private:
+      std::set<T> IDs;   //!< stores deallocated IDs to be reused
+      T nextID;          //!< next ID to use when IDs vector is empty
+    };
+}
+
diff --git a/thirdparty/embree-aarch64/common/sys/array.h b/thirdparty/embree-aarch64/common/sys/array.h
new file mode 100644
index 0000000000..77722a39f6
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/array.h
@@ -0,0 +1,222 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+#include "alloc.h"
+
+namespace embree
+{
+  /*! static array with static size */
+  template<typename T, size_t N>
+    class array_t
+    {
+    public:
+
+      /********************** Iterators  ****************************/
+
+      __forceinline T* begin() const { return items; };
+      __forceinline T* end  () const { return items+N; };
+
+
+      /********************** Capacity ****************************/
+
+      __forceinline bool   empty    () const { return N == 0; }
+      __forceinline size_t size     () const { return N; }
+      __forceinline size_t max_size () const { return N; }
+            
+
+      /******************** Element access **************************/
+
+      __forceinline       T& operator[](size_t i)       { assert(i < N); return items[i]; }
+      __forceinline const T& operator[](size_t i) const { assert(i < N); return items[i]; }
+
+      __forceinline       T& at(size_t i)       { assert(i < N); return items[i]; }
+      __forceinline const T& at(size_t i) const { assert(i < N); return items[i]; }
+
+      __forceinline T& front() const { assert(N > 0); return items[0]; };
+      __forceinline T& back () const { assert(N > 0); return items[N-1]; };
+
+      __forceinline       T* data()       { return items; };
+      __forceinline const T* data() const { return items; };
+
+    private:
+      T items[N];
+    };
+
+  /*! static array with dynamic size */
+  template<typename T, size_t N>
+    class darray_t
+    {
+    public:
+
+      __forceinline darray_t () : M(0) {}
+
+      __forceinline darray_t (const T& v) : M(0) {
+        for (size_t i=0; i<N; i++) items[i] = v;
+      }
+
+      /********************** Iterators  ****************************/
+
+      __forceinline T* begin() const { return items; };
+      __forceinline T* end  () const { return items+M; };
+
+
+      /********************** Capacity ****************************/
+
+      __forceinline bool   empty    () const { return M == 0; }
+      __forceinline size_t size     () const { return M; }
+      __forceinline size_t capacity () const { return N; }
+      __forceinline size_t max_size () const { return N; }
+      
+      void resize(size_t new_size) {
+        assert(new_size < max_size());
+        M = new_size;
+      }
+
+      /******************** Modifiers **************************/
+
+      __forceinline void push_back(const T& v) 
+      {
+        assert(M+1 < max_size());
+        items[M++] = v;
+      }
+
+      __forceinline void pop_back() 
+      {
+        assert(!empty());
+        M--;
+      }
+
+      __forceinline void clear() {
+        M = 0;
+      }
+
+      /******************** Element access **************************/
+
+      __forceinline       T& operator[](size_t i)       { assert(i < M); return items[i]; }
+      __forceinline const T& operator[](size_t i) const { assert(i < M); return items[i]; }
+
+      __forceinline       T& at(size_t i)       { assert(i < M); return items[i]; }
+      __forceinline const T& at(size_t i) const { assert(i < M); return items[i]; }
+
+      __forceinline T& front() const { assert(M > 0); return items[0]; };
+      __forceinline T& back () const { assert(M > 0); return items[M-1]; };
+
+      __forceinline       T* data()       { return items; };
+      __forceinline const T* data() const { return items; };
+
+    private:
+      size_t M;
+      T items[N];
+    };
+
+  /*! dynamic sized array that is allocated on the stack */
+#define dynamic_large_stack_array(Ty,Name,N,max_stack_bytes) StackArray<Ty,max_stack_bytes> Name(N)
+  template<typename Ty, size_t max_stack_bytes>
+    struct __aligned(64) StackArray
+  {
+    __forceinline StackArray (const size_t N)
+      : N(N)
+    {
+      if (N*sizeof(Ty) <= max_stack_bytes) 
+        data = &arr[0];
+      else
+        data = (Ty*) alignedMalloc(N*sizeof(Ty),64); 
+    }
+
+    __forceinline ~StackArray () {
+      if (data != &arr[0]) alignedFree(data);
+    }
+
+    __forceinline operator       Ty* ()       { return data; }
+    __forceinline operator const Ty* () const { return data; }
+
+    __forceinline       Ty& operator[](const int i)       { assert(i>=0 && i<N); return data[i]; }
+    __forceinline const Ty& operator[](const int i) const { assert(i>=0 && i<N); return data[i]; }
+
+    __forceinline       Ty& operator[](const unsigned i)       { assert(i<N); return data[i]; }
+    __forceinline const Ty& operator[](const unsigned i) const { assert(i<N); return data[i]; }
+
+#if defined(__X86_64__) || defined(__aarch64__)
+    __forceinline       Ty& operator[](const size_t i)       { assert(i<N); return data[i]; }
+    __forceinline const Ty& operator[](const size_t i) const { assert(i<N); return data[i]; }
+#endif
+
+  private:
+    Ty arr[max_stack_bytes/sizeof(Ty)];
+    Ty* data;
+    size_t N;
+
+  private:
+    StackArray (const StackArray& other) DELETED; // do not implement
+    StackArray& operator= (const StackArray& other) DELETED; // do not implement
+
+  };
+
+  /*! dynamic sized array that is allocated on the stack */
+  template<typename Ty, size_t max_stack_elements, size_t max_total_elements>
+    struct __aligned(64) DynamicStackArray
+  {
+    __forceinline DynamicStackArray ()
+      : data(&arr[0]) {}
+
+    __forceinline ~DynamicStackArray ()
+    {
+      if (!isStackAllocated())
+        delete[] data;
+    }
+
+    __forceinline bool isStackAllocated() const {
+      return data == &arr[0];
+    }
+
+    __forceinline size_t size() const
+    {
+      if (isStackAllocated()) return max_stack_elements;
+      else return max_total_elements;
+    }
+
+    __forceinline void resize(size_t M)
+    {
+      assert(M <= max_total_elements);
+      if (likely(M <= max_stack_elements)) return;
+      if (likely(!isStackAllocated())) return;
+
+      data = new Ty[max_total_elements];
+      
+      for (size_t i=0; i<max_stack_elements; i++)
+        data[i] = arr[i];
+    }
+
+    __forceinline operator       Ty* ()       { return data; }
+    __forceinline operator const Ty* () const { return data; }
+
+    __forceinline       Ty& operator[](const int i)      { assert(i>=0 && i<max_total_elements); resize(i+1); return data[i]; }
+    __forceinline       Ty& operator[](const unsigned i) { assert(i<max_total_elements); resize(i+1); return data[i]; }
+
+#if defined(__X86_64__) || defined(__aarch64__)
+    __forceinline       Ty& operator[](const size_t i) { assert(i<max_total_elements); resize(i+1); return data[i]; }
+#endif
+
+    __forceinline DynamicStackArray (const DynamicStackArray& other)
+      : data(&arr[0]) 
+    {
+      for (size_t i=0; i<other.size(); i++)
+        this->operator[] (i) = other[i];
+    }
+     
+    DynamicStackArray& operator= (const DynamicStackArray& other)
+    {
+      for (size_t i=0; i<other.size(); i++)
+        this->operator[] (i) = other[i];
+
+      return *this;
+    }
+
+  private:
+    Ty arr[max_stack_elements];
+    Ty* data;
+  };
+}
diff --git a/thirdparty/embree-aarch64/common/sys/atomic.h b/thirdparty/embree-aarch64/common/sys/atomic.h
new file mode 100644
index 0000000000..ebfb8552c3
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/atomic.h
@@ -0,0 +1,59 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include <atomic>
+#include "intrinsics.h"
+
+namespace embree
+{
+/* compiler memory barriers */
+#if defined(__INTEL_COMPILER)
+//#define __memory_barrier() __memory_barrier()
+#elif defined(__GNUC__) || defined(__clang__)
+#  define __memory_barrier() asm volatile("" ::: "memory")
+#elif  defined(_MSC_VER)
+#  define __memory_barrier() _ReadWriteBarrier()
+#endif
+
+  template <typename T>
+    struct atomic : public std::atomic<T>
+  {
+    atomic () {}
+      
+    atomic (const T& a)
+      : std::atomic<T>(a) {}
+
+    atomic (const atomic<T>& a) {
+      this->store(a.load());
+    }
+
+    atomic& operator=(const atomic<T>& other) {
+      this->store(other.load());
+      return *this;
+    }
+  };
+
+  template<typename T>
+    __forceinline void atomic_min(std::atomic<T>& aref, const T& bref)
+  {
+    const T b = bref.load();
+    while (true) {
+      T a = aref.load();
+      if (a <= b) break;
+      if (aref.compare_exchange_strong(a,b)) break;
+    }
+  }
+
+  template<typename T>
+    __forceinline void atomic_max(std::atomic<T>& aref, const T& bref)
+  {
+    const T b = bref.load();
+    while (true) {
+      T a = aref.load();
+      if (a >= b) break;
+      if (aref.compare_exchange_strong(a,b)) break;
+    }
+  }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/barrier.cpp b/thirdparty/embree-aarch64/common/sys/barrier.cpp
new file mode 100644
index 0000000000..0061d18db2
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/barrier.cpp
@@ -0,0 +1,289 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "barrier.h"
+#include "condition.h"
+#include "regression.h"
+#include "thread.h"
+
+#if defined (__WIN32__)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+namespace embree
+{
+  struct BarrierSysImplementation
+  {
+    __forceinline BarrierSysImplementation (size_t N) 
+      : i(0), enterCount(0), exitCount(0), barrierSize(0) 
+    {
+      events[0] = CreateEvent(nullptr, TRUE, FALSE, nullptr);
+      events[1] = CreateEvent(nullptr, TRUE, FALSE, nullptr);
+      init(N);
+    }
+    
+    __forceinline ~BarrierSysImplementation ()
+    {
+      CloseHandle(events[0]);
+      CloseHandle(events[1]);
+    }
+    
+    __forceinline void init(size_t N) 
+    {
+      barrierSize = N;
+      enterCount.store(N);
+      exitCount.store(N);
+    }
+
+    __forceinline void wait()
+    {
+      /* every thread entering the barrier decrements this count */
+      size_t i0 = i;
+      size_t cnt0 = enterCount--;
+
+      /* all threads except the last one are wait in the barrier */
+      if (cnt0 > 1) 
+      {
+        if (WaitForSingleObject(events[i0], INFINITE) != WAIT_OBJECT_0)
+          THROW_RUNTIME_ERROR("WaitForSingleObjects failed");
+      }
+      
+      /* the last thread starts all threads waiting at the barrier */
+      else 
+      {
+        i = 1-i;
+        enterCount.store(barrierSize);
+        if (SetEvent(events[i0]) == 0)
+          THROW_RUNTIME_ERROR("SetEvent failed");
+      }
+
+      /* every thread leaving the barrier decrements this count */
+      size_t cnt1 = exitCount--;
+
+      /* the last thread that left the barrier resets the event again */
+      if (cnt1 == 1) 
+      {
+        exitCount.store(barrierSize);
+        if (ResetEvent(events[i0]) == 0)
+          THROW_RUNTIME_ERROR("ResetEvent failed");
+      }
+    }
+
+  public:
+    HANDLE events[2];
+    atomic<size_t> i;
+    atomic<size_t> enterCount;
+    atomic<size_t> exitCount;
+    size_t barrierSize;
+  };
+}
+
+#else
+
+namespace embree
+{
+  struct BarrierSysImplementation
+  {
+    __forceinline BarrierSysImplementation (size_t N) 
+      : count(0), barrierSize(0) 
+    {
+      init(N);
+    }
+    
+    __forceinline void init(size_t N) 
+    {
+      assert(count == 0);
+      count = 0;
+      barrierSize = N;
+    }
+
+    __forceinline void wait()
+    {
+      mutex.lock();
+      count++;
+      
+      if (count == barrierSize) {
+        count = 0;
+        cond.notify_all();
+        mutex.unlock();
+        return;
+      }
+      
+      cond.wait(mutex);
+      mutex.unlock();
+      return;
+    }
+
+  public:
+    MutexSys mutex;
+    ConditionSys cond;
+    volatile size_t count;
+    volatile size_t barrierSize;
+  };
+}
+
+#endif
+
+namespace embree
+{
+  BarrierSys::BarrierSys (size_t N) {
+    opaque = new BarrierSysImplementation(N);
+  }
+
+  BarrierSys::~BarrierSys () {
+    delete (BarrierSysImplementation*) opaque;
+  }
+
+  void BarrierSys::init(size_t count) {
+    ((BarrierSysImplementation*) opaque)->init(count);
+  }
+
+  void BarrierSys::wait() {
+    ((BarrierSysImplementation*) opaque)->wait();
+  }
+
+  LinearBarrierActive::LinearBarrierActive (size_t N) 
+    : count0(nullptr), count1(nullptr), mode(0), flag0(0), flag1(0), threadCount(0)
+  { 
+    if (N == 0) N = getNumberOfLogicalThreads();
+    init(N);
+  }
+
+  LinearBarrierActive::~LinearBarrierActive() 
+  {
+    delete[] count0;
+    delete[] count1;
+  }
+
+  void LinearBarrierActive::init(size_t N) 
+  {
+    if (threadCount != N) {
+      threadCount = N;
+      if (count0) delete[] count0; count0 = new unsigned char[N];
+      if (count1) delete[] count1; count1 = new unsigned char[N];
+    }
+    mode      = 0;
+    flag0     = 0;
+    flag1     = 0;
+    for (size_t i=0; i<N; i++) count0[i] = 0;
+    for (size_t i=0; i<N; i++) count1[i] = 0;
+  }
+
+  void LinearBarrierActive::wait (const size_t threadIndex)
+  {
+    if (mode == 0)
+    {			
+      if (threadIndex == 0)
+      {	
+        for (size_t i=0; i<threadCount; i++)
+          count1[i] = 0;
+        
+        for (size_t i=1; i<threadCount; i++)
+        {
+          while (likely(count0[i] == 0)) 
+            pause_cpu();
+        }
+        mode  = 1;
+        flag1 = 0;
+        __memory_barrier();
+        flag0 = 1;
+      }			
+      else
+      {					
+        count0[threadIndex] = 1;
+        {
+          while (likely(flag0 == 0))
+            pause_cpu();
+        }
+        
+      }		
+    }					
+    else						
+    {
+      if (threadIndex == 0)
+      {	
+        for (size_t i=0; i<threadCount; i++)
+          count0[i] = 0;
+        
+        for (size_t i=1; i<threadCount; i++)
+        {		
+          while (likely(count1[i] == 0))
+            pause_cpu();
+        }
+        
+        mode  = 0;
+        flag0 = 0;
+        __memory_barrier();
+        flag1 = 1;
+      }			
+      else
+      {					
+        count1[threadIndex] = 1;
+        {
+          while (likely(flag1 == 0))
+            pause_cpu();
+        }
+      }		
+    }					
+  }
+
+  struct barrier_sys_regression_test : public RegressionTest
+  {
+    BarrierSys barrier;
+    std::atomic<size_t> threadID;
+    std::atomic<size_t> numFailed;
+    std::vector<size_t> threadResults;
+
+    barrier_sys_regression_test() 
+      : RegressionTest("barrier_sys_regression_test"), threadID(0), numFailed(0)
+    {
+      registerRegressionTest(this);
+    }
+
+    static void thread_alloc(barrier_sys_regression_test* This)
+    {
+      size_t tid = This->threadID++;
+      for (size_t j=0; j<1000; j++)
+      {
+        This->barrier.wait();
+        This->threadResults[tid] = tid;
+        This->barrier.wait();
+      }
+    }
+    
+    bool run ()
+    {
+      threadID.store(0);
+      numFailed.store(0);
+
+      size_t numThreads = getNumberOfLogicalThreads();
+      threadResults.resize(numThreads);
+      barrier.init(numThreads+1);
+
+      /* create threads */
+      std::vector<thread_t> threads;
+      for (size_t i=0; i<numThreads; i++)
+        threads.push_back(createThread((thread_func)thread_alloc,this));
+
+      /* run test */ 
+      for (size_t i=0; i<1000; i++)
+      {
+        for (size_t i=0; i<numThreads; i++) threadResults[i] = 0;
+        barrier.wait();
+        barrier.wait();
+        for (size_t i=0; i<numThreads; i++) numFailed += threadResults[i] != i;
+      }
+
+      /* destroy threads */
+      for (size_t i=0; i<numThreads; i++)
+        join(threads[i]);
+
+      return numFailed == 0;
+    }
+  };
+
+  barrier_sys_regression_test barrier_sys_regression_test;
+}
+
+
diff --git a/thirdparty/embree-aarch64/common/sys/barrier.h b/thirdparty/embree-aarch64/common/sys/barrier.h
new file mode 100644
index 0000000000..89607b8685
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/barrier.h
@@ -0,0 +1,112 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "intrinsics.h"
+#include "sysinfo.h"
+#include "atomic.h"
+
+namespace embree
+{
+  /*! system barrier using operating system */
+  class BarrierSys
+  {
+  public:
+
+    /*! construction / destruction */
+    BarrierSys (size_t N = 0);
+    ~BarrierSys ();
+
+  private:
+    /*! class in non-copyable */
+    BarrierSys (const BarrierSys& other) DELETED; // do not implement
+    BarrierSys& operator= (const BarrierSys& other) DELETED; // do not implement
+
+  public:
+    /*! intializes the barrier with some number of threads */
+    void init(size_t count);
+
+    /*! lets calling thread wait in barrier */
+    void wait();
+
+  private:
+    void* opaque;
+  };
+
+  /*! fast active barrier using atomitc counter */
+  struct BarrierActive 
+  {
+  public:
+    BarrierActive () 
+      : cntr(0) {}
+    
+    void reset() {
+      cntr.store(0);
+    }
+
+    void wait (size_t numThreads) 
+    {
+      cntr++;
+      while (cntr.load() != numThreads) 
+        pause_cpu();
+    }
+
+  private:
+    std::atomic<size_t> cntr;
+  };
+
+  /*! fast active barrier that does not require initialization to some number of threads */
+  struct BarrierActiveAutoReset
+  {
+  public:
+    BarrierActiveAutoReset () 
+      : cntr0(0), cntr1(0) {}
+
+    void wait (size_t threadCount) 
+    {
+      cntr0.fetch_add(1);
+      while (cntr0 != threadCount) pause_cpu();
+      cntr1.fetch_add(1);
+      while (cntr1 != threadCount) pause_cpu();
+      cntr0.fetch_add(-1);
+      while (cntr0 != 0) pause_cpu();
+      cntr1.fetch_add(-1);
+      while (cntr1 != 0) pause_cpu();
+    }
+
+  private:
+    std::atomic<size_t> cntr0;
+    std::atomic<size_t> cntr1;
+  };
+
+  class LinearBarrierActive
+  {
+  public:
+
+    /*! construction and destruction */
+    LinearBarrierActive (size_t threadCount = 0);
+    ~LinearBarrierActive();
+    
+  private:
+    /*! class in non-copyable */
+    LinearBarrierActive (const LinearBarrierActive& other) DELETED; // do not implement
+    LinearBarrierActive& operator= (const LinearBarrierActive& other) DELETED; // do not implement
+
+  public:
+    /*! intializes the barrier with some number of threads */
+    void init(size_t threadCount);
+    
+    /*! thread with threadIndex waits in the barrier */
+    void wait (const size_t threadIndex);
+    
+  private:
+    volatile unsigned char* count0;
+    volatile unsigned char* count1; 
+    volatile unsigned int mode;
+    volatile unsigned int flag0;
+    volatile unsigned int flag1;
+    volatile size_t threadCount;
+  };
+}
+
diff --git a/thirdparty/embree-aarch64/common/sys/condition.cpp b/thirdparty/embree-aarch64/common/sys/condition.cpp
new file mode 100644
index 0000000000..0e7ca7af39
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/condition.cpp
@@ -0,0 +1,81 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "condition.h"
+
+#if defined(__WIN32__) && !defined(PTHREADS_WIN32)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+namespace embree
+{
+  struct ConditionImplementation
+  {
+    __forceinline ConditionImplementation () {
+      InitializeConditionVariable(&cond);
+    }
+
+    __forceinline ~ConditionImplementation () {
+    }
+
+    __forceinline void wait(MutexSys& mutex_in) {
+      SleepConditionVariableCS(&cond, (LPCRITICAL_SECTION)mutex_in.mutex, INFINITE);
+    }
+
+    __forceinline void notify_all() {
+      WakeAllConditionVariable(&cond);
+    }
+
+  public:
+    CONDITION_VARIABLE cond;
+  };
+}
+#endif
+
+#if defined(__UNIX__) || defined(PTHREADS_WIN32)
+#include <pthread.h>
+namespace embree
+{
+  struct ConditionImplementation
+  {
+    __forceinline ConditionImplementation () { 
+      pthread_cond_init(&cond,nullptr); 
+    }
+    
+    __forceinline ~ConditionImplementation() { 
+      pthread_cond_destroy(&cond);
+    } 
+    
+    __forceinline void wait(MutexSys& mutex) { 
+      pthread_cond_wait(&cond, (pthread_mutex_t*)mutex.mutex); 
+    }
+    
+    __forceinline void notify_all() { 
+      pthread_cond_broadcast(&cond); 
+    }
+    
+  public:
+    pthread_cond_t cond;
+  };
+}
+#endif
+
+namespace embree 
+{
+  ConditionSys::ConditionSys () { 
+    cond = new ConditionImplementation; 
+  }
+
+  ConditionSys::~ConditionSys() { 
+    delete (ConditionImplementation*) cond;
+  }
+
+  void ConditionSys::wait(MutexSys& mutex) { 
+    ((ConditionImplementation*) cond)->wait(mutex);
+  }
+
+  void ConditionSys::notify_all() { 
+    ((ConditionImplementation*) cond)->notify_all();
+  }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/condition.h b/thirdparty/embree-aarch64/common/sys/condition.h
new file mode 100644
index 0000000000..7a3a05aa81
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/condition.h
@@ -0,0 +1,31 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "mutex.h"
+
+namespace embree
+{
+  class ConditionSys
+  {
+  public:
+    ConditionSys();
+    ~ConditionSys();
+    void wait( class MutexSys& mutex );
+    void notify_all();
+
+    template<typename Predicate>
+      __forceinline void wait( class MutexSys& mutex, const Predicate& pred )
+    {
+      while (!pred()) wait(mutex);
+    }
+
+  private:
+    ConditionSys (const ConditionSys& other) DELETED; // do not implement
+    ConditionSys& operator= (const ConditionSys& other) DELETED; // do not implement
+
+  protected:
+    void* cond;
+  };
+}
diff --git a/thirdparty/embree-aarch64/common/sys/filename.cpp b/thirdparty/embree-aarch64/common/sys/filename.cpp
new file mode 100644
index 0000000000..86182c1afb
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/filename.cpp
@@ -0,0 +1,138 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "filename.h"
+#include "sysinfo.h"
+
+namespace embree
+{
+#ifdef __WIN32__
+  const char path_sep = '\\';
+#else
+  const char path_sep = '/';
+#endif
+
+  /*! create an empty filename */
+  FileName::FileName () {}
+
+  /*! create a valid filename from a string */
+  FileName::FileName (const char* in) {
+    filename = in;
+    for (size_t i=0; i<filename.size(); i++)
+      if (filename[i] == '\\' || filename[i] == '/')
+        filename[i] = path_sep;
+    while (!filename.empty() && filename[filename.size()-1] == path_sep)
+      filename.resize(filename.size()-1);
+  }
+
+  /*! create a valid filename from a string */
+  FileName::FileName (const std::string& in) {
+    filename = in;
+    for (size_t i=0; i<filename.size(); i++)
+      if (filename[i] == '\\' || filename[i] == '/')
+        filename[i] = path_sep;
+    while (!filename.empty() && filename[filename.size()-1] == path_sep)
+      filename.resize(filename.size()-1);
+  }
+  
+  /*! returns path to home folder */
+  FileName FileName::homeFolder() 
+  {
+#ifdef __WIN32__
+    const char* home = getenv("UserProfile");
+#else
+    const char* home = getenv("HOME");
+#endif
+    if (home) return home;
+    return "";
+  }
+
+  /*! returns path to executable */
+  FileName FileName::executableFolder() {
+    return FileName(getExecutableFileName()).path();
+  }
+
+  /*! returns the path */
+  FileName FileName::path() const {
+    size_t pos = filename.find_last_of(path_sep);
+    if (pos == std::string::npos) return FileName();
+    return filename.substr(0,pos);
+  }
+
+  /*! returns the basename */
+  std::string FileName::base() const {
+    size_t pos = filename.find_last_of(path_sep);
+    if (pos == std::string::npos) return filename;
+    return filename.substr(pos+1);
+  }
+
+  /*! returns the extension */
+  std::string FileName::ext() const {
+    size_t pos = filename.find_last_of('.');
+    if (pos == std::string::npos) return "";
+    return filename.substr(pos+1);
+  }
+
+  /*! returns the extension */
+  FileName FileName::dropExt() const {
+    size_t pos = filename.find_last_of('.');
+    if (pos == std::string::npos) return filename;
+    return filename.substr(0,pos);
+  }
+
+  /*! returns the basename without extension */
+  std::string FileName::name() const {
+    size_t start = filename.find_last_of(path_sep);
+    if (start == std::string::npos) start = 0; else start++;
+    size_t end = filename.find_last_of('.');
+    if (end == std::string::npos || end < start) end = filename.size();
+    return filename.substr(start, end - start);
+  }
+
+  /*! replaces the extension */
+  FileName FileName::setExt(const std::string& ext) const {
+    size_t start = filename.find_last_of(path_sep);
+    if (start == std::string::npos) start = 0; else start++;
+    size_t end = filename.find_last_of('.');
+    if (end == std::string::npos || end < start) return FileName(filename+ext);
+    return FileName(filename.substr(0,end)+ext);
+  }
+
+  /*! adds the extension */
+  FileName FileName::addExt(const std::string& ext) const {
+    return FileName(filename+ext);
+  }
+
+  /*! concatenates two filenames to this/other */
+  FileName FileName::operator +( const FileName& other ) const {
+    if (filename == "") return FileName(other);
+    else return FileName(filename + path_sep + other.filename);
+  }
+
+  /*! concatenates two filenames to this/other */
+  FileName FileName::operator +( const std::string& other ) const {
+    return operator+(FileName(other));
+  }
+
+  /*! removes the base from a filename (if possible) */
+  FileName FileName::operator -( const FileName& base ) const {
+    size_t pos = filename.find_first_of(base);
+    if (pos == std::string::npos) return *this;
+    return FileName(filename.substr(pos+1));
+  }
+
+  /*! == operator */
+  bool operator== (const FileName& a, const FileName& b) {
+    return a.filename == b.filename;
+  }
+  
+  /*! != operator */
+  bool operator!= (const FileName& a, const FileName& b) {
+    return a.filename != b.filename;
+  }
+
+  /*! output operator */
+  std::ostream& operator<<(std::ostream& cout, const FileName& filename) {
+    return cout << filename.filename;
+  }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/filename.h b/thirdparty/embree-aarch64/common/sys/filename.h
new file mode 100644
index 0000000000..58f881b14d
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/filename.h
@@ -0,0 +1,81 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+
+namespace embree
+{
+  /*! Convenience class for handling file names and paths. */
+  class FileName
+  {
+  public:
+
+    /*! create an empty filename */
+    FileName ();
+
+    /*! create a valid filename from a string */
+    FileName (const char* filename);
+
+    /*! create a valid filename from a string */
+    FileName (const std::string& filename);
+    
+    /*! returns path to home folder */
+    static FileName homeFolder();
+
+    /*! returns path to executable */
+    static FileName executableFolder();
+
+    /*! auto convert into a string */
+    operator std::string() const { return filename; }
+
+    /*! returns a string of the filename */
+    const std::string str() const { return filename; }
+
+    /*! returns a c-string of the filename */
+    const char* c_str() const { return filename.c_str(); }
+
+    /*! returns the path of a filename */
+    FileName path() const;
+
+    /*! returns the file of a filename  */
+    std::string base() const;
+
+    /*! returns the base of a filename without extension */
+    std::string name() const;
+
+    /*! returns the file extension */
+    std::string ext() const;
+
+    /*! drops the file extension */
+    FileName dropExt() const;
+
+    /*! replaces the file extension */
+    FileName setExt(const std::string& ext = "") const;
+
+    /*! adds file extension */
+    FileName addExt(const std::string& ext = "") const;
+
+    /*! concatenates two filenames to this/other */
+    FileName operator +( const FileName& other ) const;
+
+    /*! concatenates two filenames to this/other */
+    FileName operator +( const std::string& other ) const;
+
+    /*! removes the base from a filename (if possible) */
+    FileName operator -( const FileName& base ) const;
+
+    /*! == operator */
+    friend bool operator==(const FileName& a, const FileName& b);
+
+    /*! != operator */
+    friend bool operator!=(const FileName& a, const FileName& b);
+
+    /*! output operator */
+    friend embree_ostream operator<<(embree_ostream cout, const FileName& filename);
+   
+  private:
+    std::string filename;
+  };
+}
diff --git a/thirdparty/embree-aarch64/common/sys/intrinsics.h b/thirdparty/embree-aarch64/common/sys/intrinsics.h
new file mode 100644
index 0000000000..44cdbd8f0f
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/intrinsics.h
@@ -0,0 +1,559 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+
+#if defined(__WIN32__)
+#include <intrin.h>
+#endif
+
+#if defined(__ARM_NEON)
+#include "../math/SSE2NEON.h"
+#if defined(NEON_AVX2_EMULATION)
+#include "../math/AVX2NEON.h"
+#endif
+#else
+#include <immintrin.h>
+#endif
+
+#if defined(__BMI__) && defined(__GNUC__) && !defined(__INTEL_COMPILER)
+  #if !defined(_tzcnt_u32)
+    #define _tzcnt_u32 __tzcnt_u32
+  #endif
+  #if !defined(_tzcnt_u64)
+    #define _tzcnt_u64 __tzcnt_u64
+  #endif
+#endif
+
+#if defined(__aarch64__)
+#if !defined(_lzcnt_u32)
+  #define _lzcnt_u32 __builtin_clz
+#endif
+#if !defined(_lzcnt_u32)
+  #define _lzcnt_u32 __builtin_clzll
+#endif
+#else
+#if defined(__LZCNT__)
+  #if !defined(_lzcnt_u32)
+    #define _lzcnt_u32 __lzcnt32
+  #endif
+  #if !defined(_lzcnt_u64)
+    #define _lzcnt_u64 __lzcnt64
+  #endif
+#endif
+#endif
+
+#if defined(__WIN32__)
+#  ifndef NOMINMAX
+#  define NOMINMAX
+#  endif
+#  include <windows.h>
+#endif
+
+/* normally defined in pmmintrin.h, but we always need this */
+#if !defined(_MM_SET_DENORMALS_ZERO_MODE)
+#define _MM_DENORMALS_ZERO_ON   (0x0040)
+#define _MM_DENORMALS_ZERO_OFF  (0x0000)
+#define _MM_DENORMALS_ZERO_MASK (0x0040)
+#define _MM_SET_DENORMALS_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_DENORMALS_ZERO_MASK) | (x)))
+#endif
+
+namespace embree
+{
+
+////////////////////////////////////////////////////////////////////////////////
+/// Windows Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__WIN32__)
+
+  __forceinline size_t read_tsc()
+  {
+    LARGE_INTEGER li;
+    QueryPerformanceCounter(&li);
+    return (size_t)li.QuadPart;
+  }
+
+  __forceinline int bsf(int v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return _tzcnt_u32(v);
+#else
+    unsigned long r = 0; _BitScanForward(&r,v); return r;
+#endif
+  }
+
+  __forceinline unsigned bsf(unsigned v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return _tzcnt_u32(v);
+#else
+    unsigned long r = 0; _BitScanForward(&r,v); return r;
+#endif
+  }
+
+#if defined(__X86_64__)
+  __forceinline size_t bsf(size_t v) {
+#if defined(__AVX2__)
+    return _tzcnt_u64(v);
+#else
+    unsigned long r = 0; _BitScanForward64(&r,v); return r;
+#endif
+  }
+#endif
+
+  __forceinline int bscf(int& v)
+  {
+    int i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+
+  __forceinline unsigned bscf(unsigned& v)
+  {
+    unsigned i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+
+#if defined(__X86_64__)
+  __forceinline size_t bscf(size_t& v)
+  {
+    size_t i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+#endif
+
+  __forceinline int bsr(int v) {
+#if defined(__AVX2__)  && !defined(__aarch64__)
+    return 31 - _lzcnt_u32(v);
+#else
+    unsigned long r = 0; _BitScanReverse(&r,v); return r;
+#endif
+  }
+
+  __forceinline unsigned bsr(unsigned v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return 31 - _lzcnt_u32(v);
+#else
+    unsigned long r = 0; _BitScanReverse(&r,v); return r;
+#endif
+  }
+
+#if defined(__X86_64__)
+  __forceinline size_t bsr(size_t v) {
+#if defined(__AVX2__)
+    return 63 -_lzcnt_u64(v);
+#else
+    unsigned long r = 0; _BitScanReverse64(&r, v); return r;
+#endif
+  }
+#endif
+
+  __forceinline int lzcnt(const int x)
+  {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return _lzcnt_u32(x);
+#else
+    if (unlikely(x == 0)) return 32;
+    return 31 - bsr(x);
+#endif
+  }
+
+  __forceinline int btc(int v, int i) {
+    long r = v; _bittestandcomplement(&r,i); return r;
+  }
+
+  __forceinline int bts(int v, int i) {
+    long r = v; _bittestandset(&r,i); return r;
+  }
+
+  __forceinline int btr(int v, int i) {
+    long r = v; _bittestandreset(&r,i); return r;
+  }
+
+#if defined(__X86_64__)
+
+  __forceinline size_t btc(size_t v, size_t i) {
+    size_t r = v; _bittestandcomplement64((__int64*)&r,i); return r;
+  }
+
+  __forceinline size_t bts(size_t v, size_t i) {
+    __int64 r = v; _bittestandset64(&r,i); return r;
+  }
+
+  __forceinline size_t btr(size_t v, size_t i) {
+    __int64 r = v; _bittestandreset64(&r,i); return r;
+  }
+
+#endif
+
+  __forceinline int32_t atomic_cmpxchg(volatile int32_t* p, const int32_t c, const int32_t v) {
+    return _InterlockedCompareExchange((volatile long*)p,v,c);
+  }
+
+////////////////////////////////////////////////////////////////////////////////
+/// Unix Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#else
+
+#if defined(__i386__) && defined(__PIC__)
+
+  __forceinline void __cpuid(int out[4], int op)
+  {
+    asm volatile ("xchg{l}\t{%%}ebx, %1\n\t"
+                  "cpuid\n\t"
+                  "xchg{l}\t{%%}ebx, %1\n\t"
+                  : "=a"(out[0]), "=r"(out[1]), "=c"(out[2]), "=d"(out[3])
+                  : "0"(op));
+  }
+
+  __forceinline void __cpuid_count(int out[4], int op1, int op2)
+  {
+    asm volatile ("xchg{l}\t{%%}ebx, %1\n\t"
+                  "cpuid\n\t"
+                  "xchg{l}\t{%%}ebx, %1\n\t"
+                  : "=a" (out[0]), "=r" (out[1]), "=c" (out[2]), "=d" (out[3])
+                  : "0" (op1), "2" (op2));
+  }
+
+#else
+
+  __forceinline void __cpuid(int out[4], int op) {
+#if defined(__ARM_NEON)
+    if (op == 0) { // Get CPU name
+      out[0] = 0x41524d20;
+      out[1] = 0x41524d20;
+      out[2] = 0x41524d20;
+      out[3] = 0x41524d20;
+    }
+#else
+    asm volatile ("cpuid" : "=a"(out[0]), "=b"(out[1]), "=c"(out[2]), "=d"(out[3]) : "a"(op));
+#endif
+  }
+
+#if !defined(__ARM_NEON)
+  __forceinline void __cpuid_count(int out[4], int op1, int op2) {
+    asm volatile ("cpuid" : "=a"(out[0]), "=b"(out[1]), "=c"(out[2]), "=d"(out[3]) : "a"(op1), "c"(op2));
+  }
+#endif
+
+#endif
+
+  __forceinline uint64_t read_tsc()  {
+#if defined(__ARM_NEON)
+    return 0; // FIXME(LTE): mimic rdtsc
+#else
+    uint32_t high,low;
+    asm volatile ("rdtsc" : "=d"(high), "=a"(low));
+    return (((uint64_t)high) << 32) + (uint64_t)low;
+#endif
+  }
+
+  __forceinline int bsf(int v) {
+#if defined(__ARM_NEON)
+    return __builtin_ctz(v);
+#else
+#if defined(__AVX2__)
+    return _tzcnt_u32(v);
+#else
+    int r = 0; asm ("bsf %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+#endif
+  }
+
+#if defined(__X86_64__) || defined(__aarch64__)
+  __forceinline unsigned bsf(unsigned v)
+  {
+#if defined(__ARM_NEON)
+    return __builtin_ctz(v);
+#else
+#if defined(__AVX2__)
+    return _tzcnt_u32(v);
+#else
+    unsigned r = 0; asm ("bsf %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+#endif
+  }
+#endif
+
+  __forceinline size_t bsf(size_t v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+#if defined(__X86_64__)
+    return _tzcnt_u64(v);
+#else
+    return _tzcnt_u32(v);
+#endif
+#elif defined(__ARM_NEON)
+    return __builtin_ctzl(v);
+#else
+    size_t r = 0; asm ("bsf %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+  }
+
+  __forceinline int bscf(int& v)
+  {
+    int i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+
+#if defined(__X86_64__) || defined(__aarch64__)
+  __forceinline unsigned int bscf(unsigned int& v)
+  {
+    unsigned int i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+#endif
+
+  __forceinline size_t bscf(size_t& v)
+  {
+    size_t i = bsf(v);
+    v &= v-1;
+    return i;
+  }
+
+  __forceinline int bsr(int v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return 31 - _lzcnt_u32(v);
+#elif defined(__ARM_NEON)
+    return __builtin_clz(v)^31;
+#else
+    int r = 0; asm ("bsr %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+  }
+
+#if defined(__X86_64__) || defined(__aarch64__)
+  __forceinline unsigned bsr(unsigned v) {
+#if defined(__AVX2__)
+    return 31 - _lzcnt_u32(v);
+#elif defined(__ARM_NEON)
+    return __builtin_clz(v)^31;
+#else
+    unsigned r = 0; asm ("bsr %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+  }
+#endif
+
+  __forceinline size_t bsr(size_t v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+#if defined(__X86_64__)
+    return 63 - _lzcnt_u64(v);
+#else
+    return 31 - _lzcnt_u32(v);
+#endif
+#elif defined(__aarch64__)
+    return (sizeof(v) * 8 - 1) - __builtin_clzl(v);
+#else
+    size_t r = 0; asm ("bsr %1,%0" : "=r"(r) : "r"(v)); return r;
+#endif
+  }
+
+  __forceinline int lzcnt(const int x)
+  {
+#if defined(__AVX2__) && !defined(__aarch64__)
+    return _lzcnt_u32(x);
+#else
+    if (unlikely(x == 0)) return 32;
+    return 31 - bsr(x);
+#endif
+  }
+
+  __forceinline size_t blsr(size_t v) {
+#if defined(__AVX2__) && !defined(__aarch64__)
+#if defined(__INTEL_COMPILER)
+    return _blsr_u64(v);
+#else
+#if defined(__X86_64__)
+    return __blsr_u64(v);
+#else
+    return __blsr_u32(v);
+#endif
+#endif
+#else
+    return v & (v-1);
+#endif
+  }
+
+  __forceinline int btc(int v, int i) {
+#if defined(__aarch64__)
+    // _bittestandcomplement(long *a, long b) {
+    // unsigned char x = (*a >> b) & 1;
+    // *a = *a ^ (1 << b);
+    // return x;
+
+    // We only need `*a`
+    return (v ^ (1 << i));
+#else
+    int r = 0; asm ("btc %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags" ); return r;
+#endif
+  }
+
+  __forceinline int bts(int v, int i) {
+#if defined(__aarch64__)
+    // _bittestandset(long *a, long b) {
+    // unsigned char x = (*a >> b) & 1;
+    //  *a = *a | (1 << b);
+    //  return x;
+    return (v | (v << i));
+#else
+    int r = 0; asm ("bts %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags"); return r;
+#endif
+  }
+
+  __forceinline int btr(int v, int i) {
+#if defined(__aarch64__)
+    // _bittestandreset(long *a, long b) {
+    // unsigned char x = (*a >> b) & 1;
+    //  *a = *a & ~(1 << b);
+    //  return x;
+    return (v & ~(v << i));
+#else
+    int r = 0; asm ("btr %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags"); return r;
+#endif
+  }
+
+  __forceinline size_t btc(size_t v, size_t i) {
+#if defined(__aarch64__)
+    return (v ^ (1 << i));
+#else
+    size_t r = 0; asm ("btc %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags" ); return r;
+#endif
+  }
+
+  __forceinline size_t bts(size_t v, size_t i) {
+#if defined(__aarch64__)
+    return (v | (v << i));
+#else
+    size_t r = 0; asm ("bts %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags"); return r;
+#endif
+  }
+
+  __forceinline size_t btr(size_t v, size_t i) {
+#if defined(__ARM_NEON)
+    return (v & ~(v << i));
+#else
+    size_t r = 0; asm ("btr %1,%0" : "=r"(r) : "r"(i), "0"(v) : "flags"); return r;
+#endif
+  }
+
+  __forceinline int32_t atomic_cmpxchg(int32_t volatile* value, int32_t comparand, const int32_t input) {
+    return __sync_val_compare_and_swap(value, comparand, input);
+  }
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// All Platforms
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__clang__) || defined(__GNUC__)
+#if !defined(_mm_undefined_ps)
+  __forceinline __m128 _mm_undefined_ps() { return _mm_setzero_ps(); }
+#endif
+#if !defined(_mm_undefined_si128)
+  __forceinline __m128i _mm_undefined_si128() { return _mm_setzero_si128(); }
+#endif
+#if !defined(_mm256_undefined_ps) && defined(__AVX__)
+  __forceinline __m256 _mm256_undefined_ps() { return _mm256_setzero_ps(); }
+#endif
+#if !defined(_mm256_undefined_si256) && defined(__AVX__)
+  __forceinline __m256i _mm256_undefined_si256() { return _mm256_setzero_si256(); }
+#endif
+#if !defined(_mm512_undefined_ps) && defined(__AVX512F__)
+  __forceinline __m512 _mm512_undefined_ps() { return _mm512_setzero_ps(); }
+#endif
+#if !defined(_mm512_undefined_epi32) && defined(__AVX512F__)
+  __forceinline __m512i _mm512_undefined_epi32() { return _mm512_setzero_si512(); }
+#endif
+#endif
+
+#if defined(__SSE4_2__) || defined(__ARM_NEON)
+
+  __forceinline int popcnt(int in) {
+    return _mm_popcnt_u32(in);
+  }
+
+  __forceinline unsigned popcnt(unsigned in) {
+    return _mm_popcnt_u32(in);
+  }
+
+#if defined(__X86_64__) || defined(__ARM_NEON)
+  __forceinline size_t popcnt(size_t in) {
+    return _mm_popcnt_u64(in);
+  }
+#endif
+
+#endif
+
+  __forceinline uint64_t rdtsc()
+  {
+    int dummy[4];
+    __cpuid(dummy,0);
+    uint64_t clock = read_tsc();
+    __cpuid(dummy,0);
+    return clock;
+  }
+
+  __forceinline void pause_cpu(const size_t N = 8)
+  {
+    for (size_t i=0; i<N; i++)
+      _mm_pause();
+  }
+
+  /* prefetches */
+  __forceinline void prefetchL1 (const void* ptr) { _mm_prefetch((const char*)ptr,_MM_HINT_T0); }
+  __forceinline void prefetchL2 (const void* ptr) { _mm_prefetch((const char*)ptr,_MM_HINT_T1); }
+  __forceinline void prefetchL3 (const void* ptr) { _mm_prefetch((const char*)ptr,_MM_HINT_T2); }
+  __forceinline void prefetchNTA(const void* ptr) { _mm_prefetch((const char*)ptr,_MM_HINT_NTA); }
+  __forceinline void prefetchEX (const void* ptr) {
+#if defined(__INTEL_COMPILER)
+    _mm_prefetch((const char*)ptr,_MM_HINT_ET0);
+#else
+    _mm_prefetch((const char*)ptr,_MM_HINT_T0);
+#endif
+  }
+
+  __forceinline void prefetchL1EX(const void* ptr) {
+    prefetchEX(ptr);
+  }
+
+  __forceinline void prefetchL2EX(const void* ptr) {
+    prefetchEX(ptr);
+  }
+#if defined(__AVX2__) && !defined(__aarch64__)
+   __forceinline unsigned int pext(unsigned int a, unsigned int b) { return _pext_u32(a, b); }
+   __forceinline unsigned int pdep(unsigned int a, unsigned int b) { return _pdep_u32(a, b); }
+#if defined(__X86_64__)
+   __forceinline size_t pext(size_t a, size_t b) { return _pext_u64(a, b); }
+   __forceinline size_t pdep(size_t a, size_t b) { return _pdep_u64(a, b); }
+#endif
+#endif
+
+#if defined(__AVX512F__)
+#if defined(__INTEL_COMPILER)
+   __forceinline float mm512_cvtss_f32(__m512 v) {
+     return _mm512_cvtss_f32(v);
+   }
+   __forceinline int mm512_mask2int(__mmask16 k1) {
+     return _mm512_mask2int(k1);
+   }
+   __forceinline __mmask16 mm512_int2mask(int mask) {
+     return _mm512_int2mask(mask);
+   }
+#else
+   __forceinline float mm512_cvtss_f32(__m512 v) { // FIXME: _mm512_cvtss_f32 neither supported by clang v4.0.0 nor GCC 6.3
+     return _mm_cvtss_f32(_mm512_castps512_ps128(v));
+   }
+   __forceinline int mm512_mask2int(__mmask16 k1) { // FIXME: _mm512_mask2int not yet supported by GCC 6.3
+     return (int)k1;
+   }
+   __forceinline __mmask16 mm512_int2mask(int mask) { // FIXME: _mm512_int2mask not yet supported by GCC 6.3
+     return (__mmask16)mask;
+   }
+#endif
+#endif
+}
diff --git a/thirdparty/embree-aarch64/common/sys/library.cpp b/thirdparty/embree-aarch64/common/sys/library.cpp
new file mode 100644
index 0000000000..899267a1e4
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/library.cpp
@@ -0,0 +1,83 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "library.h"
+#include "sysinfo.h"
+#include "filename.h"
+
+////////////////////////////////////////////////////////////////////////////////
+/// Windows Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__WIN32__)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+namespace embree
+{
+  /* opens a shared library */
+  lib_t openLibrary(const std::string& file)
+  {
+    std::string fullName = file+".dll";
+    FileName executable = getExecutableFileName();
+    HANDLE handle = LoadLibrary((executable.path() + fullName).c_str());
+    return lib_t(handle);
+  }
+
+  /* returns address of a symbol from the library */
+  void* getSymbol(lib_t lib, const std::string& sym) {
+    return reinterpret_cast<void *>(GetProcAddress(HMODULE(lib),sym.c_str()));
+  }
+
+  /* closes the shared library */
+  void closeLibrary(lib_t lib) {
+    FreeLibrary(HMODULE(lib));
+  }
+}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Unix Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__UNIX__)
+
+#include <dlfcn.h>
+
+namespace embree
+{
+  /* opens a shared library */
+  lib_t openLibrary(const std::string& file)
+  {
+#if defined(__MACOSX__)
+    std::string fullName = "lib"+file+".dylib";
+#else
+    std::string fullName = "lib"+file+".so";
+#endif
+    void* lib = dlopen(fullName.c_str(), RTLD_NOW);
+    if (lib) return lib_t(lib);
+    FileName executable = getExecutableFileName();
+    lib = dlopen((executable.path() + fullName).c_str(),RTLD_NOW);
+    if (lib == nullptr) {
+      const char* error = dlerror();
+      if (error) {
+        THROW_RUNTIME_ERROR(error);
+      } else {
+        THROW_RUNTIME_ERROR("could not load library "+executable.str());
+      }
+    }
+    return lib_t(lib);
+  }
+
+  /* returns address of a symbol from the library */
+  void* getSymbol(lib_t lib, const std::string& sym) {
+    return dlsym(lib,sym.c_str());
+  }
+
+  /* closes the shared library */
+  void closeLibrary(lib_t lib) {
+    dlclose(lib);
+  }
+}
+#endif
diff --git a/thirdparty/embree-aarch64/common/sys/library.h b/thirdparty/embree-aarch64/common/sys/library.h
new file mode 100644
index 0000000000..c2164e9fbe
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/library.h
@@ -0,0 +1,21 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+
+namespace embree
+{
+  /*! type for shared library */
+  typedef struct opaque_lib_t* lib_t;
+
+  /*! loads a shared library */
+  lib_t openLibrary(const std::string& file);
+
+  /*! returns address of a symbol from the library */
+  void* getSymbol(lib_t lib, const std::string& sym);
+
+  /*! unloads a shared library */
+  void closeLibrary(lib_t lib);
+}
diff --git a/thirdparty/embree-aarch64/common/sys/mutex.cpp b/thirdparty/embree-aarch64/common/sys/mutex.cpp
new file mode 100644
index 0000000000..11779bc9b9
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/mutex.cpp
@@ -0,0 +1,58 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "mutex.h"
+#include "regression.h"
+
+#if defined(__WIN32__) && !defined(PTHREADS_WIN32)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+namespace embree
+{
+  MutexSys::MutexSys() { mutex = new CRITICAL_SECTION; InitializeCriticalSection((CRITICAL_SECTION*)mutex); }
+  MutexSys::~MutexSys() { DeleteCriticalSection((CRITICAL_SECTION*)mutex); delete (CRITICAL_SECTION*)mutex; }
+  void MutexSys::lock() { EnterCriticalSection((CRITICAL_SECTION*)mutex); }
+  bool MutexSys::try_lock() { return TryEnterCriticalSection((CRITICAL_SECTION*)mutex) != 0; }
+  void MutexSys::unlock() { LeaveCriticalSection((CRITICAL_SECTION*)mutex); }
+}
+#endif
+
+#if defined(__UNIX__) || defined(PTHREADS_WIN32)
+#include <pthread.h>
+namespace embree
+{
+  /*! system mutex using pthreads */
+  MutexSys::MutexSys() 
+  { 
+    mutex = new pthread_mutex_t; 
+    if (pthread_mutex_init((pthread_mutex_t*)mutex, nullptr) != 0)
+      THROW_RUNTIME_ERROR("pthread_mutex_init failed");
+  }
+  
+  MutexSys::~MutexSys() 
+  { 
+    MAYBE_UNUSED bool ok = pthread_mutex_destroy((pthread_mutex_t*)mutex) == 0;
+    assert(ok);
+    delete (pthread_mutex_t*)mutex; 
+    mutex = nullptr;
+  }
+  
+  void MutexSys::lock() 
+  { 
+    if (pthread_mutex_lock((pthread_mutex_t*)mutex) != 0) 
+      THROW_RUNTIME_ERROR("pthread_mutex_lock failed");
+  }
+  
+  bool MutexSys::try_lock() { 
+    return pthread_mutex_trylock((pthread_mutex_t*)mutex) == 0;
+  }
+  
+  void MutexSys::unlock() 
+  { 
+    if (pthread_mutex_unlock((pthread_mutex_t*)mutex) != 0)
+      THROW_RUNTIME_ERROR("pthread_mutex_unlock failed");
+  }
+};
+#endif
diff --git a/thirdparty/embree-aarch64/common/sys/mutex.h b/thirdparty/embree-aarch64/common/sys/mutex.h
new file mode 100644
index 0000000000..1164210f23
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/mutex.h
@@ -0,0 +1,98 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+#include "intrinsics.h"
+#include "atomic.h"
+
+namespace embree
+{
+  /*! system mutex */
+  class MutexSys {
+    friend struct ConditionImplementation;
+  public:
+    MutexSys();
+    ~MutexSys();
+
+  private:
+    MutexSys (const MutexSys& other) DELETED; // do not implement
+    MutexSys& operator= (const MutexSys& other) DELETED; // do not implement
+
+  public:
+    void lock();
+    bool try_lock();
+    void unlock();
+
+  protected:
+    void* mutex;
+  };
+
+  /*! spinning mutex */
+  class SpinLock
+  {
+  public:
+ 
+    SpinLock ()
+      : flag(false) {}
+
+    __forceinline bool isLocked() {
+      return flag.load();
+    }
+
+    __forceinline void lock()
+    {
+      while (true) 
+      {
+        while (flag.load()) 
+        {
+          _mm_pause(); 
+          _mm_pause();
+        }
+        
+        bool expected = false;
+        if (flag.compare_exchange_strong(expected,true,std::memory_order_acquire))
+          break;
+      }
+    }
+    
+    __forceinline bool try_lock()
+    {
+      bool expected = false;
+      if (flag.load() != expected) {
+        return false;
+      }
+      return flag.compare_exchange_strong(expected,true,std::memory_order_acquire);
+    }
+
+    __forceinline void unlock() {
+      flag.store(false,std::memory_order_release);
+    }
+    
+    __forceinline void wait_until_unlocked() 
+    {
+      while(flag.load())
+      {
+        _mm_pause(); 
+        _mm_pause();
+      }
+    }
+
+  public:
+    atomic<bool> flag;
+  };
+
+  /*! safe mutex lock and unlock helper */
+  template<typename Mutex> class Lock {
+  public:
+    Lock (Mutex& mutex) : mutex(mutex), locked(true) { mutex.lock(); }
+    Lock (Mutex& mutex, bool locked) : mutex(mutex), locked(locked) {}
+    ~Lock() { if (locked) mutex.unlock(); }
+    __forceinline void lock() { assert(!locked); locked = true; mutex.lock(); }
+    __forceinline bool isLocked() const { return locked; }
+  protected:
+    Mutex& mutex;
+    bool locked;
+  };
+}
diff --git a/thirdparty/embree-aarch64/common/sys/platform.h b/thirdparty/embree-aarch64/common/sys/platform.h
new file mode 100644
index 0000000000..737f14aa6e
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/platform.h
@@ -0,0 +1,387 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#define _CRT_SECURE_NO_WARNINGS
+
+#include <cstddef>
+#include <cassert>
+#include <cstdlib>
+#include <cstdio>
+#include <memory>
+#include <stdexcept>
+#include <iostream>
+#include <iomanip>
+#include <fstream>
+#include <string>
+#include <cstring>
+#include <stdint.h>
+#include <functional>
+
+////////////////////////////////////////////////////////////////////////////////
+/// detect platform
+////////////////////////////////////////////////////////////////////////////////
+
+/* detect 32 or 64 platform */
+#if defined(__x86_64__) || defined(__ia64__) || defined(_M_X64)
+#define __X86_64__
+#endif
+
+/* detect Linux platform */
+#if defined(linux) || defined(__linux__) || defined(__LINUX__)
+#  if !defined(__LINUX__)
+#     define __LINUX__
+#  endif
+#  if !defined(__UNIX__)
+#     define __UNIX__
+#  endif
+#endif
+
+/* detect FreeBSD platform */
+#if defined(__FreeBSD__) || defined(__FREEBSD__)
+#  if !defined(__FREEBSD__)
+#     define __FREEBSD__
+#  endif
+#  if !defined(__UNIX__)
+#     define __UNIX__
+#  endif
+#endif
+
+/* detect Windows 95/98/NT/2000/XP/Vista/7/8/10 platform */
+#if (defined(WIN32) || defined(_WIN32) || defined(__WIN32__) || defined(__NT__)) && !defined(__CYGWIN__)
+#  if !defined(__WIN32__)
+#     define __WIN32__
+#  endif
+#endif
+
+/* detect Cygwin platform */
+#if defined(__CYGWIN__)
+#  if !defined(__UNIX__)
+#     define __UNIX__
+#  endif
+#endif
+
+/* detect MAC OS X platform */
+#if defined(__APPLE__) || defined(MACOSX) || defined(__MACOSX__)
+#  if !defined(__MACOSX__)
+#     define __MACOSX__
+#  endif
+#  if !defined(__UNIX__)
+#     define __UNIX__
+#  endif
+#endif
+
+/* try to detect other Unix systems */
+#if defined(__unix__) || defined (unix) || defined(__unix) || defined(_unix)
+#  if !defined(__UNIX__)
+#     define __UNIX__
+#  endif
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Macros
+////////////////////////////////////////////////////////////////////////////////
+
+#ifdef __WIN32__
+#define dll_export __declspec(dllexport)
+#define dll_import __declspec(dllimport)
+#else
+#define dll_export __attribute__ ((visibility ("default")))
+#define dll_import
+#endif
+
+#ifdef __WIN32__
+#if !defined(__noinline)
+#define __noinline             __declspec(noinline)
+#endif
+//#define __forceinline        __forceinline
+//#define __restrict           __restrict
+#if defined(__INTEL_COMPILER)
+#define __restrict__           __restrict
+#else
+#define __restrict__           //__restrict // causes issues with MSVC
+#endif
+#if !defined(__thread)
+// NOTE: Require `-fms-extensions` for clang
+#define __thread               __declspec(thread)
+#endif
+#if !defined(__aligned)
+#if defined(__MINGW32__)
+#define __aligned(...)           __attribute__((aligned(__VA_ARGS__)))
+#else
+#define __aligned(...)           __declspec(align(__VA_ARGS__))
+#endif
+#endif
+//#define __FUNCTION__           __FUNCTION__
+#define debugbreak()           __debugbreak()
+
+#else
+#if !defined(__noinline)
+#define __noinline             __attribute__((noinline))
+#endif
+#if !defined(__forceinline)
+#define __forceinline          inline __attribute__((always_inline))
+#endif
+//#define __restrict             __restrict
+//#define __thread               __thread
+#if !defined(__aligned)
+#define __aligned(...)           __attribute__((aligned(__VA_ARGS__)))
+#endif
+#if !defined(__FUNCTION__)
+#define __FUNCTION__           __PRETTY_FUNCTION__
+#endif
+#define debugbreak()           asm ("int $3")
+#endif
+
+#if defined(__clang__) || defined(__GNUC__)
+  #define MAYBE_UNUSED __attribute__((unused))
+#else
+  #define MAYBE_UNUSED
+#endif
+
+#if defined(_MSC_VER) && (_MSC_VER < 1900) // before VS2015 deleted functions are not supported properly
+  #define DELETED
+#else
+  #define DELETED  = delete
+#endif
+
+// -- GODOT start --
+#ifndef likely
+// -- GODOT end --
+#if defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+#define   likely(expr) (expr)
+#define unlikely(expr) (expr)
+#else
+#define   likely(expr) __builtin_expect((bool)(expr),true )
+#define unlikely(expr) __builtin_expect((bool)(expr),false)
+#endif
+// -- GODOT start --
+#endif
+// -- GODOT end --
+
+////////////////////////////////////////////////////////////////////////////////
+/// Error handling and debugging
+////////////////////////////////////////////////////////////////////////////////
+
+/* debug printing macros */
+#define STRING(x) #x
+#define TOSTRING(x) STRING(x)
+#define PING embree_cout << __FILE__ << " (" << __LINE__ << "): " << __FUNCTION__ << embree_endl
+#define PRINT(x) embree_cout << STRING(x) << " = " << (x) << embree_endl
+#define PRINT2(x,y) embree_cout << STRING(x) << " = " << (x) << ", " << STRING(y) << " = " << (y) << embree_endl
+#define PRINT3(x,y,z) embree_cout << STRING(x) << " = " << (x) << ", " << STRING(y) << " = " << (y) << ", " << STRING(z) << " = " << (z) << embree_endl
+#define PRINT4(x,y,z,w) embree_cout << STRING(x) << " = " << (x) << ", " << STRING(y) << " = " << (y) << ", " << STRING(z) << " = " << (z) << ", " << STRING(w) << " = " << (w) << embree_endl
+
+#if defined(DEBUG) // only report file and line in debug mode
+  // -- GODOT start --
+  // #define THROW_RUNTIME_ERROR(str)
+  //   throw std::runtime_error(std::string(__FILE__) + " (" + toString(__LINE__) + "): " + std::string(str));
+  #define THROW_RUNTIME_ERROR(str) \
+    printf(std::string(__FILE__) + " (" + toString(__LINE__) + "): " + std::string(str)), abort();
+  // -- GODOT end --
+#else
+  // -- GODOT start --
+  // #define THROW_RUNTIME_ERROR(str)
+  //   throw std::runtime_error(str);
+  #define THROW_RUNTIME_ERROR(str) \
+    abort();
+  // -- GODOT end --
+#endif
+
+#define FATAL(x)   THROW_RUNTIME_ERROR(x)
+#define WARNING(x) { std::cerr << "Warning: " << x << embree_endl << std::flush; }
+
+#define NOT_IMPLEMENTED FATAL(std::string(__FUNCTION__) + " not implemented")
+
+////////////////////////////////////////////////////////////////////////////////
+/// Basic types
+////////////////////////////////////////////////////////////////////////////////
+
+/* default floating-point type */
+namespace embree {
+  typedef float real;
+}
+
+/* windows does not have ssize_t */
+#if defined(__WIN32__)
+#if defined(__X86_64__) || defined(__aarch64__)
+typedef int64_t ssize_t;
+#else
+typedef int32_t ssize_t;
+#endif
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Basic utility functions
+////////////////////////////////////////////////////////////////////////////////
+
+__forceinline std::string toString(long long value) {
+  return std::to_string(value);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+/// Disable some compiler warnings
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__INTEL_COMPILER)
+//#pragma warning(disable:265 ) // floating-point operation result is out of range
+//#pragma warning(disable:383 ) // value copied to temporary, reference to temporary used
+//#pragma warning(disable:869 ) // parameter was never referenced
+//#pragma warning(disable:981 ) // operands are evaluated in unspecified order
+//#pragma warning(disable:1418) // external function definition with no prior declaration
+//#pragma warning(disable:1419) // external declaration in primary source file
+//#pragma warning(disable:1572) // floating-point equality and inequality comparisons are unreliable
+//#pragma warning(disable:94  ) // the size of an array must be greater than zero
+//#pragma warning(disable:1599) // declaration hides parameter
+//#pragma warning(disable:424 ) // extra ";" ignored
+#pragma warning(disable:2196) // routine is both "inline" and "noinline"
+//#pragma warning(disable:177 ) // label was declared but never referenced
+//#pragma warning(disable:114 ) // function was referenced but not defined
+//#pragma warning(disable:819 ) // template nesting depth does not match the previous declaration of function
+#pragma warning(disable:15335)  // was not vectorized: vectorization possible but seems inefficient
+#endif
+
+#if defined(_MSC_VER)
+//#pragma warning(disable:4200) // nonstandard extension used : zero-sized array in struct/union
+#pragma warning(disable:4800) // forcing value to bool 'true' or 'false' (performance warning)
+//#pragma warning(disable:4267) // '=' : conversion from 'size_t' to 'unsigned long', possible loss of data
+#pragma warning(disable:4244) // 'argument' : conversion from 'ssize_t' to 'unsigned int', possible loss of data
+//#pragma warning(disable:4355) // 'this' : used in base member initializer list
+//#pragma warning(disable:391 ) // '<=' : signed / unsigned mismatch
+//#pragma warning(disable:4018) // '<' : signed / unsigned mismatch
+//#pragma warning(disable:4305) // 'initializing' : truncation from 'double' to 'float'
+//#pragma warning(disable:4068) // unknown pragma
+//#pragma warning(disable:4146) // unary minus operator applied to unsigned type, result still unsigned
+//#pragma warning(disable:4838) // conversion from 'unsigned int' to 'const int' requires a narrowing conversion)
+//#pragma warning(disable:4227) // anachronism used : qualifiers on reference are ignored
+#pragma warning(disable:4503) // decorated name length exceeded, name was truncated
+#pragma warning(disable:4180) // qualifier applied to function type has no meaning; ignored
+#pragma warning(disable:4258) // definition from the for loop is ignored; the definition from the enclosing scope is used
+
+#  if _MSC_VER < 1910 // prior to Visual studio 2017 (V141)
+#    pragma warning(disable:4101) // warning C4101: 'x': unreferenced local variable // a compiler bug issues wrong warnings
+#    pragma warning(disable:4789) // buffer '' of size 8 bytes will be overrun; 32 bytes will be written starting at offset 0
+#  endif
+
+#endif
+
+#if defined(__clang__) && !defined(__INTEL_COMPILER)
+//#pragma clang diagnostic ignored "-Wunknown-pragmas"
+//#pragma clang diagnostic ignored "-Wunused-variable"
+//#pragma clang diagnostic ignored "-Wreorder"
+//#pragma clang diagnostic ignored "-Wmicrosoft"
+//#pragma clang diagnostic ignored "-Wunused-private-field"
+//#pragma clang diagnostic ignored "-Wunused-local-typedef"
+//#pragma clang diagnostic ignored "-Wunused-function"
+//#pragma clang diagnostic ignored "-Wnarrowing"
+//#pragma clang diagnostic ignored "-Wc++11-narrowing"
+//#pragma clang diagnostic ignored "-Wdeprecated-register"
+//#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
+
+#if defined(__GNUC__) && !defined(__INTEL_COMPILER) && !defined(__clang__)
+#pragma GCC diagnostic ignored "-Wpragmas"
+//#pragma GCC diagnostic ignored "-Wnarrowing"
+#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"
+//#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+//#pragma GCC diagnostic ignored "-Warray-bounds"
+#pragma GCC diagnostic ignored "-Wattributes"
+#pragma GCC diagnostic ignored "-Wmisleading-indentation"
+#pragma GCC diagnostic ignored "-Wsign-compare"
+#pragma GCC diagnostic ignored "-Wparentheses"
+#endif
+
+#if defined(__clang__) && defined(__WIN32__)
+#pragma clang diagnostic ignored "-Wunused-parameter"
+#pragma clang diagnostic ignored "-Wmicrosoft-cast"
+#pragma clang diagnostic ignored "-Wmicrosoft-enum-value"
+#pragma clang diagnostic ignored "-Wmicrosoft-include"
+#pragma clang diagnostic ignored "-Wunused-function"
+#pragma clang diagnostic ignored "-Wunknown-pragmas"
+#endif
+
+/* disabling deprecated warning, please use only where use of deprecated Embree API functions is desired */
+#if defined(__WIN32__) && defined(__INTEL_COMPILER)
+#define DISABLE_DEPRECATED_WARNING __pragma(warning (disable: 1478)) // warning: function was declared deprecated
+#define ENABLE_DEPRECATED_WARNING  __pragma(warning (enable:  1478)) // warning: function was declared deprecated
+#elif defined(__INTEL_COMPILER)
+#define DISABLE_DEPRECATED_WARNING _Pragma("warning (disable: 1478)") // warning: function was declared deprecated
+#define ENABLE_DEPRECATED_WARNING  _Pragma("warning (enable : 1478)") // warning: function was declared deprecated
+#elif defined(__clang__)
+#define DISABLE_DEPRECATED_WARNING _Pragma("clang diagnostic ignored \"-Wdeprecated-declarations\"") // warning: xxx is deprecated
+#define ENABLE_DEPRECATED_WARNING  _Pragma("clang diagnostic warning \"-Wdeprecated-declarations\"") // warning: xxx is deprecated
+#elif defined(__GNUC__)
+#define DISABLE_DEPRECATED_WARNING _Pragma("GCC diagnostic ignored \"-Wdeprecated-declarations\"") // warning: xxx is deprecated
+#define ENABLE_DEPRECATED_WARNING  _Pragma("GCC diagnostic warning \"-Wdeprecated-declarations\"") // warning: xxx is deprecated
+#elif defined(_MSC_VER)
+#define DISABLE_DEPRECATED_WARNING __pragma(warning (disable: 4996)) // warning: function was declared deprecated
+#define ENABLE_DEPRECATED_WARNING  __pragma(warning (enable : 4996)) // warning: function was declared deprecated
+#endif
+
+/* embree output stream */
+#define embree_ostream std::ostream&
+#define embree_cout std::cout
+#define embree_cout_uniform std::cout
+#define embree_endl std::endl
+  
+////////////////////////////////////////////////////////////////////////////////
+/// Some macros for static profiling
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined (__GNUC__)
+#define IACA_SSC_MARK( MARK_ID )						\
+__asm__ __volatile__ (									\
+					  "\n\t  movl $"#MARK_ID", %%ebx"	\
+					  "\n\t  .byte 0x64, 0x67, 0x90"	\
+					  : : : "memory" );
+
+#define IACA_UD_BYTES __asm__ __volatile__ ("\n\t .byte 0x0F, 0x0B");
+
+#else
+#define IACA_UD_BYTES {__asm _emit 0x0F \
+	__asm _emit 0x0B}
+
+#define IACA_SSC_MARK(x) {__asm  mov ebx, x\
+	__asm  _emit 0x64 \
+	__asm  _emit 0x67 \
+	__asm  _emit 0x90 }
+
+#define IACA_VC64_START __writegsbyte(111, 111);
+#define IACA_VC64_END   __writegsbyte(222, 222);
+
+#endif
+
+#define IACA_START {IACA_UD_BYTES \
+					IACA_SSC_MARK(111)}
+#define IACA_END {IACA_SSC_MARK(222) \
+					IACA_UD_BYTES}
+
+namespace embree
+{
+  template<typename Closure>
+    struct OnScopeExitHelper
+  {
+    OnScopeExitHelper (const Closure f) : active(true), f(f) {}
+    ~OnScopeExitHelper() { if (active) f(); }
+    void deactivate() { active = false; }
+    bool active;
+    const Closure f;
+  };
+
+  template <typename Closure>
+    OnScopeExitHelper<Closure> OnScopeExit(const Closure f) {
+    return OnScopeExitHelper<Closure>(f);
+  }
+
+#define STRING_JOIN2(arg1, arg2) DO_STRING_JOIN2(arg1, arg2)
+#define DO_STRING_JOIN2(arg1, arg2) arg1 ## arg2
+#define ON_SCOPE_EXIT(code)                                             \
+  auto STRING_JOIN2(on_scope_exit_, __LINE__) = OnScopeExit([&](){code;})
+
+  template<typename Ty>
+    std::unique_ptr<Ty> make_unique(Ty* ptr) {
+    return std::unique_ptr<Ty>(ptr);
+  }
+
+}
diff --git a/thirdparty/embree-aarch64/common/sys/ref.h b/thirdparty/embree-aarch64/common/sys/ref.h
new file mode 100644
index 0000000000..24648e6234
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/ref.h
@@ -0,0 +1,122 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "atomic.h"
+
+namespace embree
+{
+  struct NullTy {
+  };
+
+  extern MAYBE_UNUSED NullTy null;
+  
+  class RefCount
+  {
+  public:
+    RefCount(int val = 0) : refCounter(val) {}
+    virtual ~RefCount() {};
+  
+    virtual RefCount* refInc() { refCounter.fetch_add(1); return this; }
+    virtual void refDec() { if (refCounter.fetch_add(-1) == 1) delete this; }
+  private:
+    std::atomic<size_t> refCounter;
+  };
+  
+  ////////////////////////////////////////////////////////////////////////////////
+  /// Reference to single object
+  ////////////////////////////////////////////////////////////////////////////////
+
+  template<typename Type>
+  class Ref
+  {
+  public:
+    Type* ptr;
+
+    ////////////////////////////////////////////////////////////////////////////////
+    /// Constructors, Assignment & Cast Operators
+    ////////////////////////////////////////////////////////////////////////////////
+
+    __forceinline Ref() : ptr(nullptr) {}
+    __forceinline Ref(NullTy) : ptr(nullptr) {}
+    __forceinline Ref(const Ref& input) : ptr(input.ptr) { if (ptr) ptr->refInc(); }
+    __forceinline Ref(Ref&& input) : ptr(input.ptr) { input.ptr = nullptr; }
+
+    __forceinline Ref(Type* const input) : ptr(input)
+    {
+      if (ptr)
+        ptr->refInc();
+    }
+
+    __forceinline ~Ref()
+    {
+      if (ptr)
+        ptr->refDec();
+    }
+
+    __forceinline Ref& operator =(const Ref& input)
+    {
+      if (input.ptr)
+        input.ptr->refInc();
+      if (ptr)
+        ptr->refDec();
+      ptr = input.ptr;
+      return *this;
+    }
+
+    __forceinline Ref& operator =(Ref&& input)
+    {
+      if (ptr)
+        ptr->refDec();
+      ptr = input.ptr;
+      input.ptr = nullptr;
+      return *this;
+    }
+
+    __forceinline Ref& operator =(Type* const input)
+    {
+      if (input)
+        input->refInc();
+      if (ptr)
+        ptr->refDec();
+      ptr = input;
+      return *this;
+    }
+
+    __forceinline Ref& operator =(NullTy)
+    {
+      if (ptr)
+        ptr->refDec();
+      ptr = nullptr;
+      return *this;
+    }
+
+    __forceinline operator bool() const { return ptr != nullptr; }
+
+    __forceinline const Type& operator  *() const { return *ptr; }
+    __forceinline       Type& operator  *()       { return *ptr; }
+    __forceinline const Type* operator ->() const { return  ptr; }
+    __forceinline       Type* operator ->()       { return  ptr; }
+
+    template<typename TypeOut>
+    __forceinline       Ref<TypeOut> cast()       { return Ref<TypeOut>(static_cast<TypeOut*>(ptr)); }
+    template<typename TypeOut>
+    __forceinline const Ref<TypeOut> cast() const { return Ref<TypeOut>(static_cast<TypeOut*>(ptr)); }
+
+    template<typename TypeOut>
+    __forceinline       Ref<TypeOut> dynamicCast()       { return Ref<TypeOut>(dynamic_cast<TypeOut*>(ptr)); }
+    template<typename TypeOut>
+    __forceinline const Ref<TypeOut> dynamicCast() const { return Ref<TypeOut>(dynamic_cast<TypeOut*>(ptr)); }
+  };
+
+  template<typename Type> __forceinline bool operator < (const Ref<Type>& a, const Ref<Type>& b) { return a.ptr   <  b.ptr;   }
+
+  template<typename Type> __forceinline bool operator ==(const Ref<Type>& a, NullTy            ) { return a.ptr   == nullptr; }
+  template<typename Type> __forceinline bool operator ==(NullTy            , const Ref<Type>& b) { return nullptr == b.ptr;   }
+  template<typename Type> __forceinline bool operator ==(const Ref<Type>& a, const Ref<Type>& b) { return a.ptr   == b.ptr;   }
+
+  template<typename Type> __forceinline bool operator !=(const Ref<Type>& a, NullTy            ) { return a.ptr   != nullptr; }
+  template<typename Type> __forceinline bool operator !=(NullTy            , const Ref<Type>& b) { return nullptr != b.ptr;   }
+  template<typename Type> __forceinline bool operator !=(const Ref<Type>& a, const Ref<Type>& b) { return a.ptr   != b.ptr;   }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/regression.cpp b/thirdparty/embree-aarch64/common/sys/regression.cpp
new file mode 100644
index 0000000000..d95ff8dfe0
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/regression.cpp
@@ -0,0 +1,30 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "regression.h"
+
+namespace embree
+{
+  /* registerRegressionTest is invoked from static initializers, thus
+   * we cannot have the regression_tests variable as global static
+   * variable due to issues with static variable initialization
+   * order. */
+  std::vector<RegressionTest*>& get_regression_tests()
+  {
+    static std::vector<RegressionTest*> regression_tests;
+    return regression_tests;
+  } 
+
+  void registerRegressionTest(RegressionTest* test) 
+  {
+    get_regression_tests().push_back(test);
+  }
+
+  RegressionTest* getRegressionTest(size_t index)
+  {
+    if (index >= get_regression_tests().size())
+      return nullptr;
+    
+    return get_regression_tests()[index];
+  }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/regression.h b/thirdparty/embree-aarch64/common/sys/regression.h
new file mode 100644
index 0000000000..632f8d92cf
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/regression.h
@@ -0,0 +1,25 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+
+#include <vector>
+
+namespace embree
+{
+  /*! virtual interface for all regression tests */
+  struct RegressionTest 
+  { 
+    RegressionTest (std::string name) : name(name) {}
+    virtual bool run() = 0;
+    std::string name;
+  };
+ 
+  /*! registers a regression test */
+  void registerRegressionTest(RegressionTest* test);
+
+  /*! run all regression tests */
+  RegressionTest* getRegressionTest(size_t index);
+}
diff --git a/thirdparty/embree-aarch64/common/sys/string.cpp b/thirdparty/embree-aarch64/common/sys/string.cpp
new file mode 100644
index 0000000000..931244383e
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/string.cpp
@@ -0,0 +1,42 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "string.h"
+
+#include <algorithm>
+#include <ctype.h>
+
+namespace embree
+{
+  char to_lower(char c) { return char(tolower(int(c))); }
+  char to_upper(char c) { return char(toupper(int(c))); }
+  std::string toLowerCase(const std::string& s) { std::string dst(s); std::transform(dst.begin(), dst.end(), dst.begin(), to_lower); return dst; }
+  std::string toUpperCase(const std::string& s) { std::string dst(s); std::transform(dst.begin(), dst.end(), dst.begin(), to_upper); return dst; }
+
+  Vec2f string_to_Vec2f ( std::string str )
+  {
+    size_t next = 0;
+    const float x = std::stof(str,&next); str = str.substr(next+1);
+    const float y = std::stof(str,&next);
+    return Vec2f(x,y);
+  }
+  
+  Vec3f string_to_Vec3f ( std::string str )
+  {
+    size_t next = 0;
+    const float x = std::stof(str,&next); str = str.substr(next+1);
+    const float y = std::stof(str,&next); str = str.substr(next+1);
+    const float z = std::stof(str,&next); 
+    return Vec3f(x,y,z);
+  }
+  
+  Vec4f string_to_Vec4f ( std::string str )
+  {
+    size_t next = 0;
+    const float x = std::stof(str,&next); str = str.substr(next+1);
+    const float y = std::stof(str,&next); str = str.substr(next+1);
+    const float z = std::stof(str,&next); str = str.substr(next+1);
+    const float w = std::stof(str,&next);
+    return Vec4f(x,y,z,w);
+  }
+}
diff --git a/thirdparty/embree-aarch64/common/sys/string.h b/thirdparty/embree-aarch64/common/sys/string.h
new file mode 100644
index 0000000000..2e9b0f88c3
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/string.h
@@ -0,0 +1,37 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+#include "../math/vec2.h"
+#include "../math/vec3.h"
+#include "../math/vec4.h"
+
+namespace embree
+{
+  class IOStreamStateRestorer 
+  {
+  public:
+    IOStreamStateRestorer(std::ostream& iostream)
+      : iostream(iostream), flags(iostream.flags()), precision(iostream.precision()) {
+    }
+
+    ~IOStreamStateRestorer() {
+      iostream.flags(flags);
+      iostream.precision(precision);
+    }
+    
+  private:
+    std::ostream& iostream;
+    std::ios::fmtflags flags;
+    std::streamsize precision;
+  };
+
+  std::string toLowerCase(const std::string& s);
+  std::string toUpperCase(const std::string& s);
+
+  Vec2f string_to_Vec2f ( std::string str );
+  Vec3f string_to_Vec3f ( std::string str );
+  Vec4f string_to_Vec4f ( std::string str );
+}
diff --git a/thirdparty/embree-aarch64/common/sys/sysinfo.cpp b/thirdparty/embree-aarch64/common/sys/sysinfo.cpp
new file mode 100644
index 0000000000..1d11436770
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/sysinfo.cpp
@@ -0,0 +1,676 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "sysinfo.h"
+#include "intrinsics.h"
+#include "string.h"
+#include "ref.h"
+#if defined(__FREEBSD__)
+#include <sys/cpuset.h>
+#include <pthread_np.h>
+typedef cpuset_t cpu_set_t;
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// All Platforms
+////////////////////////////////////////////////////////////////////////////////
+
+namespace embree
+{
+  NullTy null;
+
+  std::string getPlatformName()
+  {
+#if defined(__LINUX__) && defined(__ANDROID__) && defined(__aarch64__) && defined(__ARM_NEON)
+    return "Android Linux (aarch64 / arm64)";
+#elif defined(__LINUX__) && defined(__ANDROID__) && defined(__X86_64__)
+    return "Android Linux (x64)";
+#elif defined(__LINUX__) && defined(__ANDROID__) && (defined(_X86_) || defined(__X86__) || defined(_M_IX86))
+    return "Android Linux (x86)";
+#elif defined(__LINUX__) && !defined(__X86_64__)
+    return "Linux (32bit)";
+#elif defined(__LINUX__) && defined(__X86_64__)
+    return "Linux (64bit)";
+#elif defined(__FREEBSD__) && !defined(__X86_64__)
+    return "FreeBSD (32bit)";
+#elif defined(__FREEBSD__) && defined(__X86_64__)
+    return "FreeBSD (64bit)";
+#elif defined(__CYGWIN__) && !defined(__X86_64__)
+    return "Cygwin (32bit)";
+#elif defined(__CYGWIN__) && defined(__X86_64__)
+    return "Cygwin (64bit)";
+#elif defined(__WIN32__) && !defined(__X86_64__)
+    return "Windows (32bit)";
+#elif defined(__WIN32__) && defined(__X86_64__)
+    return "Windows (64bit)";
+#elif defined(TARGET_IPHONE_SIMULATOR) && defined(__X86_64__)
+    return "iOS Simulator (x64)";
+#elif defined(TARGET_OS_IPHONE) && defined(__aarch64__) && defined(__ARM_NEON)
+    return "iOS (aarch64 / arm64)";
+#elif defined(__MACOSX__) && !defined(__X86_64__)
+    return "Mac OS X (32bit)";
+#elif defined(__MACOSX__) && defined(__X86_64__)
+    return "Mac OS X (64bit)";
+#elif defined(__UNIX__) && defined(__aarch64__)
+    return "Unix (aarch64)";
+#elif defined(__UNIX__) && !defined(__X86_64__)
+    return "Unix (32bit)";
+#elif defined(__UNIX__) && defined(__X86_64__)
+    return "Unix (64bit)";
+#else
+    return "Unknown";
+#endif
+  }
+
+  std::string getCompilerName()
+  {
+#if defined(__INTEL_COMPILER)
+    int icc_mayor = __INTEL_COMPILER / 100 % 100;
+    int icc_minor = __INTEL_COMPILER % 100;
+    std::string version = "Intel Compiler ";
+    version += toString(icc_mayor);
+    version += "." + toString(icc_minor);
+#if defined(__INTEL_COMPILER_UPDATE)
+    version += "." + toString(__INTEL_COMPILER_UPDATE);
+#endif
+    return version;
+#elif defined(__clang__)
+    return "CLANG " __clang_version__;
+#elif defined (__GNUC__)
+    return "GCC " __VERSION__;
+#elif defined(_MSC_VER)
+    std::string version = toString(_MSC_FULL_VER);
+    version.insert(4,".");
+    version.insert(9,".");
+    version.insert(2,".");
+    return "Visual C++ Compiler " + version;
+#else
+    return "Unknown Compiler";
+#endif
+  }
+
+  std::string getCPUVendor()
+  {
+    int cpuinfo[4];
+    __cpuid (cpuinfo, 0);
+    int name[4];
+    name[0] = cpuinfo[1];
+    name[1] = cpuinfo[3];
+    name[2] = cpuinfo[2];
+    name[3] = 0;
+    return (char*)name;
+  }
+
+  CPU getCPUModel()
+  {
+    if (getCPUVendor() != "GenuineIntel")
+      return CPU::UNKNOWN;
+
+    int out[4];
+    __cpuid(out, 0);
+    if (out[0] < 1) return CPU::UNKNOWN;
+    __cpuid(out, 1);
+
+    /* please see CPUID documentation for these formulas */
+    uint32_t family_ID          = (out[0] >>  8) & 0x0F;
+    uint32_t extended_family_ID = (out[0] >> 20) & 0xFF;
+    
+    uint32_t model_ID           = (out[0] >>  4) & 0x0F;
+    uint32_t extended_model_ID  = (out[0] >> 16) & 0x0F;
+    
+    uint32_t DisplayFamily = family_ID;
+    if (family_ID == 0x0F)
+      DisplayFamily += extended_family_ID;
+    
+    uint32_t DisplayModel = model_ID;
+    if (family_ID == 0x06 || family_ID == 0x0F)
+      DisplayModel += extended_model_ID << 4;
+
+    uint32_t DisplayFamily_DisplayModel = (DisplayFamily << 8) + (DisplayModel << 0);
+
+    // Data from Intel® 64 and IA-32 Architectures, Volume 4, Chapter 2, Table 2-1 (CPUID Signature Values of DisplayFamily_DisplayModel)
+    if (DisplayFamily_DisplayModel == 0x067D) return CPU::CORE_ICE_LAKE;
+    if (DisplayFamily_DisplayModel == 0x067E) return CPU::CORE_ICE_LAKE;
+    if (DisplayFamily_DisplayModel == 0x068C) return CPU::CORE_TIGER_LAKE;
+    if (DisplayFamily_DisplayModel == 0x06A5) return CPU::CORE_COMET_LAKE;
+    if (DisplayFamily_DisplayModel == 0x06A6) return CPU::CORE_COMET_LAKE;
+    if (DisplayFamily_DisplayModel == 0x0666) return CPU::CORE_CANNON_LAKE;
+    if (DisplayFamily_DisplayModel == 0x068E) return CPU::CORE_KABY_LAKE;
+    if (DisplayFamily_DisplayModel == 0x069E) return CPU::CORE_KABY_LAKE;
+    if (DisplayFamily_DisplayModel == 0x066A) return CPU::XEON_ICE_LAKE;
+    if (DisplayFamily_DisplayModel == 0x066C) return CPU::XEON_ICE_LAKE;
+    if (DisplayFamily_DisplayModel == 0x0655) return CPU::XEON_SKY_LAKE;
+    if (DisplayFamily_DisplayModel == 0x064E) return CPU::CORE_SKY_LAKE;
+    if (DisplayFamily_DisplayModel == 0x065E) return CPU::CORE_SKY_LAKE;
+    if (DisplayFamily_DisplayModel == 0x0656) return CPU::XEON_BROADWELL;
+    if (DisplayFamily_DisplayModel == 0x064F) return CPU::XEON_BROADWELL;
+    if (DisplayFamily_DisplayModel == 0x0647) return CPU::CORE_BROADWELL;
+    if (DisplayFamily_DisplayModel == 0x063D) return CPU::CORE_BROADWELL;
+    if (DisplayFamily_DisplayModel == 0x063F) return CPU::XEON_HASWELL;
+    if (DisplayFamily_DisplayModel == 0x063C) return CPU::CORE_HASWELL;
+    if (DisplayFamily_DisplayModel == 0x0645) return CPU::CORE_HASWELL;
+    if (DisplayFamily_DisplayModel == 0x0646) return CPU::CORE_HASWELL;
+    if (DisplayFamily_DisplayModel == 0x063E) return CPU::XEON_IVY_BRIDGE;
+    if (DisplayFamily_DisplayModel == 0x063A) return CPU::CORE_IVY_BRIDGE;
+    if (DisplayFamily_DisplayModel == 0x062D) return CPU::SANDY_BRIDGE;
+    if (DisplayFamily_DisplayModel == 0x062F) return CPU::SANDY_BRIDGE;
+    if (DisplayFamily_DisplayModel == 0x062A) return CPU::SANDY_BRIDGE;
+    if (DisplayFamily_DisplayModel == 0x062E) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x0625) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x062C) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x061E) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x061F) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x061A) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x061D) return CPU::NEHALEM;
+    if (DisplayFamily_DisplayModel == 0x0617) return CPU::CORE2;
+    if (DisplayFamily_DisplayModel == 0x060F) return CPU::CORE2;
+    if (DisplayFamily_DisplayModel == 0x060E) return CPU::CORE1;
+
+    if (DisplayFamily_DisplayModel == 0x0685) return CPU::XEON_PHI_KNIGHTS_MILL;
+    if (DisplayFamily_DisplayModel == 0x0657) return CPU::XEON_PHI_KNIGHTS_LANDING;
+    
+    return CPU::UNKNOWN;
+  }
+
+  std::string stringOfCPUModel(CPU model)
+  {
+    switch (model) {
+    case CPU::XEON_ICE_LAKE           : return "Xeon Ice Lake";
+    case CPU::CORE_ICE_LAKE           : return "Core Ice Lake";
+    case CPU::CORE_TIGER_LAKE         : return "Core Tiger Lake";
+    case CPU::CORE_COMET_LAKE         : return "Core Comet Lake";
+    case CPU::CORE_CANNON_LAKE        : return "Core Cannon Lake";
+    case CPU::CORE_KABY_LAKE          : return "Core Kaby Lake";
+    case CPU::XEON_SKY_LAKE           : return "Xeon Sky Lake";
+    case CPU::CORE_SKY_LAKE           : return "Core Sky Lake";
+    case CPU::XEON_PHI_KNIGHTS_MILL   : return "Xeon Phi Knights Mill";
+    case CPU::XEON_PHI_KNIGHTS_LANDING: return "Xeon Phi Knights Landing";
+    case CPU::XEON_BROADWELL          : return "Xeon Broadwell";
+    case CPU::CORE_BROADWELL          : return "Core Broadwell";
+    case CPU::XEON_HASWELL            : return "Xeon Haswell";
+    case CPU::CORE_HASWELL            : return "Core Haswell";
+    case CPU::XEON_IVY_BRIDGE         : return "Xeon Ivy Bridge";
+    case CPU::CORE_IVY_BRIDGE         : return "Core Ivy Bridge";
+    case CPU::SANDY_BRIDGE            : return "Sandy Bridge";
+    case CPU::NEHALEM                 : return "Nehalem";
+    case CPU::CORE2                   : return "Core2";
+    case CPU::CORE1                   : return "Core";
+    case CPU::ARM                     : return "Arm";
+    case CPU::UNKNOWN                 : return "Unknown CPU";
+    }
+    return "Unknown CPU (error)";
+  }
+
+#if !defined(__ARM_NEON)
+  /* constants to access destination registers of CPUID instruction */
+  static const int EAX = 0;
+  static const int EBX = 1;
+  static const int ECX = 2;
+  static const int EDX = 3;
+
+  /* cpuid[eax=1].ecx */
+  static const int CPU_FEATURE_BIT_SSE3   = 1 << 0;
+  static const int CPU_FEATURE_BIT_SSSE3  = 1 << 9;
+  static const int CPU_FEATURE_BIT_FMA3   = 1 << 12;
+  static const int CPU_FEATURE_BIT_SSE4_1 = 1 << 19;
+  static const int CPU_FEATURE_BIT_SSE4_2 = 1 << 20;
+  //static const int CPU_FEATURE_BIT_MOVBE  = 1 << 22;
+  static const int CPU_FEATURE_BIT_POPCNT = 1 << 23;
+  //static const int CPU_FEATURE_BIT_XSAVE  = 1 << 26;
+  static const int CPU_FEATURE_BIT_OXSAVE = 1 << 27;
+  static const int CPU_FEATURE_BIT_AVX    = 1 << 28;
+  static const int CPU_FEATURE_BIT_F16C   = 1 << 29;
+  static const int CPU_FEATURE_BIT_RDRAND = 1 << 30;
+
+  /* cpuid[eax=1].edx */
+  static const int CPU_FEATURE_BIT_SSE  = 1 << 25;
+  static const int CPU_FEATURE_BIT_SSE2 = 1 << 26;
+
+  /* cpuid[eax=0x80000001].ecx */
+  static const int CPU_FEATURE_BIT_LZCNT = 1 << 5;
+
+  /* cpuid[eax=7,ecx=0].ebx */
+  static const int CPU_FEATURE_BIT_BMI1    = 1 << 3;
+  static const int CPU_FEATURE_BIT_AVX2    = 1 << 5;
+  static const int CPU_FEATURE_BIT_BMI2    = 1 << 8;
+  static const int CPU_FEATURE_BIT_AVX512F = 1 << 16;     // AVX512F  (foundation)
+  static const int CPU_FEATURE_BIT_AVX512DQ = 1 << 17;    // AVX512DQ (doubleword and quadword instructions)
+  static const int CPU_FEATURE_BIT_AVX512PF = 1 << 26;    // AVX512PF (prefetch gather/scatter instructions)
+  static const int CPU_FEATURE_BIT_AVX512ER = 1 << 27;    // AVX512ER (exponential and reciprocal instructions)
+  static const int CPU_FEATURE_BIT_AVX512CD = 1 << 28;    // AVX512CD (conflict detection instructions)
+  static const int CPU_FEATURE_BIT_AVX512BW = 1 << 30;    // AVX512BW (byte and word instructions)
+  static const int CPU_FEATURE_BIT_AVX512VL = 1 << 31;    // AVX512VL (vector length extensions)
+  static const int CPU_FEATURE_BIT_AVX512IFMA = 1 << 21;  // AVX512IFMA (integer fused multiple-add instructions)
+
+  /* cpuid[eax=7,ecx=0].ecx */
+  static const int CPU_FEATURE_BIT_AVX512VBMI = 1 << 1;   // AVX512VBMI (vector bit manipulation instructions)
+#endif
+
+#if !defined(__ARM_NEON)
+  __noinline int64_t get_xcr0()
+  {
+    // https://github.com/opencv/opencv/blob/master/modules/core/src/system.cpp#L466
+#if defined (__WIN32__) && defined(_XCR_XFEATURE_ENABLED_MASK)
+    int64_t xcr0 = 0; // int64_t is workaround for compiler bug under VS2013, Win32
+    xcr0 = _xgetbv(0);
+    return xcr0;
+#else
+    int xcr0 = 0;
+    __asm__ ("xgetbv" : "=a" (xcr0) : "c" (0) : "%edx" );
+    return xcr0;
+#endif
+  }
+#endif
+
+  int getCPUFeatures()
+  {
+#if defined(__ARM_NEON)
+      int cpu_features = CPU_FEATURE_NEON|CPU_FEATURE_SSE|CPU_FEATURE_SSE2;
+#if defined(NEON_AVX2_EMULATION)
+      cpu_features |= CPU_FEATURE_SSE3|CPU_FEATURE_SSSE3|CPU_FEATURE_SSE42;
+      cpu_features |= CPU_FEATURE_XMM_ENABLED;
+      cpu_features |= CPU_FEATURE_YMM_ENABLED;
+      cpu_features |= CPU_FEATURE_SSE41 | CPU_FEATURE_RDRAND | CPU_FEATURE_F16C;
+      cpu_features |= CPU_FEATURE_POPCNT;
+      cpu_features |= CPU_FEATURE_AVX;
+      cpu_features |= CPU_FEATURE_AVX2;
+      cpu_features |= CPU_FEATURE_FMA3;
+      cpu_features |= CPU_FEATURE_LZCNT;
+      cpu_features |= CPU_FEATURE_BMI1;
+      cpu_features |= CPU_FEATURE_BMI2;
+      cpu_features |= CPU_FEATURE_NEON_2X;
+
+
+ 
+#endif
+     return cpu_features;
+      
+#else
+    /* cache CPU features access */
+    static int cpu_features = 0;
+    if (cpu_features)
+      return cpu_features;
+
+    /* get number of CPUID leaves */
+    int cpuid_leaf0[4];
+    __cpuid(cpuid_leaf0, 0x00000000);
+    unsigned nIds = cpuid_leaf0[EAX];
+
+    /* get number of extended CPUID leaves */
+    int cpuid_leafe[4];
+    __cpuid(cpuid_leafe, 0x80000000);
+    unsigned nExIds = cpuid_leafe[EAX];
+
+    /* get CPUID leaves for EAX = 1,7, and 0x80000001 */
+    int cpuid_leaf_1[4] = { 0,0,0,0 };
+    int cpuid_leaf_7[4] = { 0,0,0,0 };
+    int cpuid_leaf_e1[4] = { 0,0,0,0 };
+    if (nIds >= 1) __cpuid (cpuid_leaf_1,0x00000001);
+#if _WIN32
+#if _MSC_VER && (_MSC_FULL_VER < 160040219)
+#else
+    if (nIds >= 7) __cpuidex(cpuid_leaf_7,0x00000007,0);
+#endif
+#else
+    if (nIds >= 7) __cpuid_count(cpuid_leaf_7,0x00000007,0);
+#endif
+    if (nExIds >= 0x80000001) __cpuid(cpuid_leaf_e1,0x80000001);
+
+    /* detect if OS saves XMM, YMM, and ZMM states */
+    bool xmm_enabled = true;
+    bool ymm_enabled = false;
+    bool zmm_enabled = false;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_OXSAVE) {
+      int64_t xcr0 = get_xcr0();
+      xmm_enabled = ((xcr0 & 0x02) == 0x02);                /* checks if xmm are enabled in XCR0 */
+      ymm_enabled = xmm_enabled && ((xcr0 & 0x04) == 0x04); /* checks if ymm state are enabled in XCR0 */
+      zmm_enabled = ymm_enabled && ((xcr0 & 0xE0) == 0xE0); /* checks if OPMASK state, upper 256-bit of ZMM0-ZMM15 and ZMM16-ZMM31 state are enabled in XCR0 */
+    }
+    if (xmm_enabled) cpu_features |= CPU_FEATURE_XMM_ENABLED;
+    if (ymm_enabled) cpu_features |= CPU_FEATURE_YMM_ENABLED;
+    if (zmm_enabled) cpu_features |= CPU_FEATURE_ZMM_ENABLED;
+
+    if (cpuid_leaf_1[EDX] & CPU_FEATURE_BIT_SSE   ) cpu_features |= CPU_FEATURE_SSE;
+    if (cpuid_leaf_1[EDX] & CPU_FEATURE_BIT_SSE2  ) cpu_features |= CPU_FEATURE_SSE2;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_SSE3  ) cpu_features |= CPU_FEATURE_SSE3;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_SSSE3 ) cpu_features |= CPU_FEATURE_SSSE3;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_SSE4_1) cpu_features |= CPU_FEATURE_SSE41;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_SSE4_2) cpu_features |= CPU_FEATURE_SSE42;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_POPCNT) cpu_features |= CPU_FEATURE_POPCNT;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_AVX   ) cpu_features |= CPU_FEATURE_AVX;
+
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_F16C  ) cpu_features |= CPU_FEATURE_F16C;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_RDRAND) cpu_features |= CPU_FEATURE_RDRAND;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX2  ) cpu_features |= CPU_FEATURE_AVX2;
+    if (cpuid_leaf_1[ECX] & CPU_FEATURE_BIT_FMA3  ) cpu_features |= CPU_FEATURE_FMA3;
+    if (cpuid_leaf_e1[ECX] & CPU_FEATURE_BIT_LZCNT) cpu_features |= CPU_FEATURE_LZCNT;
+    if (cpuid_leaf_7 [EBX] & CPU_FEATURE_BIT_BMI1 ) cpu_features |= CPU_FEATURE_BMI1;
+    if (cpuid_leaf_7 [EBX] & CPU_FEATURE_BIT_BMI2 ) cpu_features |= CPU_FEATURE_BMI2;
+
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512F   ) cpu_features |= CPU_FEATURE_AVX512F;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512DQ  ) cpu_features |= CPU_FEATURE_AVX512DQ;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512PF  ) cpu_features |= CPU_FEATURE_AVX512PF;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512ER  ) cpu_features |= CPU_FEATURE_AVX512ER;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512CD  ) cpu_features |= CPU_FEATURE_AVX512CD;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512BW  ) cpu_features |= CPU_FEATURE_AVX512BW;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512IFMA) cpu_features |= CPU_FEATURE_AVX512IFMA;
+    if (cpuid_leaf_7[EBX] & CPU_FEATURE_BIT_AVX512VL  ) cpu_features |= CPU_FEATURE_AVX512VL;
+    if (cpuid_leaf_7[ECX] & CPU_FEATURE_BIT_AVX512VBMI) cpu_features |= CPU_FEATURE_AVX512VBMI;
+
+    return cpu_features;
+#endif
+  }
+
+  std::string stringOfCPUFeatures(int features)
+  {
+    std::string str;
+    if (features & CPU_FEATURE_XMM_ENABLED) str += "XMM ";
+    if (features & CPU_FEATURE_YMM_ENABLED) str += "YMM ";
+    if (features & CPU_FEATURE_ZMM_ENABLED) str += "ZMM ";
+    if (features & CPU_FEATURE_SSE   ) str += "SSE ";
+    if (features & CPU_FEATURE_SSE2  ) str += "SSE2 ";
+    if (features & CPU_FEATURE_SSE3  ) str += "SSE3 ";
+    if (features & CPU_FEATURE_SSSE3 ) str += "SSSE3 ";
+    if (features & CPU_FEATURE_SSE41 ) str += "SSE4.1 ";
+    if (features & CPU_FEATURE_SSE42 ) str += "SSE4.2 ";
+    if (features & CPU_FEATURE_POPCNT) str += "POPCNT ";
+    if (features & CPU_FEATURE_AVX   ) str += "AVX ";
+    if (features & CPU_FEATURE_F16C  ) str += "F16C ";
+    if (features & CPU_FEATURE_RDRAND) str += "RDRAND ";
+    if (features & CPU_FEATURE_AVX2  ) str += "AVX2 ";
+    if (features & CPU_FEATURE_FMA3  ) str += "FMA3 ";
+    if (features & CPU_FEATURE_LZCNT ) str += "LZCNT ";
+    if (features & CPU_FEATURE_BMI1  ) str += "BMI1 ";
+    if (features & CPU_FEATURE_BMI2  ) str += "BMI2 ";
+    if (features & CPU_FEATURE_AVX512F) str += "AVX512F ";
+    if (features & CPU_FEATURE_AVX512DQ) str += "AVX512DQ ";
+    if (features & CPU_FEATURE_AVX512PF) str += "AVX512PF ";
+    if (features & CPU_FEATURE_AVX512ER) str += "AVX512ER ";
+    if (features & CPU_FEATURE_AVX512CD) str += "AVX512CD ";
+    if (features & CPU_FEATURE_AVX512BW) str += "AVX512BW ";
+    if (features & CPU_FEATURE_AVX512VL) str += "AVX512VL ";
+    if (features & CPU_FEATURE_AVX512IFMA) str += "AVX512IFMA ";
+    if (features & CPU_FEATURE_AVX512VBMI) str += "AVX512VBMI ";
+    if (features & CPU_FEATURE_NEON) str += "NEON ";
+    if (features & CPU_FEATURE_NEON_2X) str += "2xNEON ";
+    return str;
+  }
+
+  std::string stringOfISA (int isa)
+  {
+    if (isa == SSE) return "SSE";
+    if (isa == SSE2) return "SSE2";
+    if (isa == SSE3) return "SSE3";
+    if (isa == SSSE3) return "SSSE3";
+    if (isa == SSE41) return "SSE4.1";
+    if (isa == SSE42) return "SSE4.2";
+    if (isa == AVX) return "AVX";
+    if (isa == AVX2) return "AVX2";
+    if (isa == AVX512KNL) return "AVX512KNL";
+    if (isa == AVX512SKX) return "AVX512SKX";
+    if (isa == NEON) return "NEON";    
+    if (isa == NEON_2X) return "2xNEON";
+    return "UNKNOWN";
+  }
+
+  bool hasISA(int features, int isa) {
+    return (features & isa) == isa;
+  }
+
+  std::string supportedTargetList (int features)
+  {
+    std::string v;
+    if (hasISA(features,SSE)) v += "SSE ";
+    if (hasISA(features,SSE2)) v += "SSE2 ";
+    if (hasISA(features,SSE3)) v += "SSE3 ";
+    if (hasISA(features,SSSE3)) v += "SSSE3 ";
+    if (hasISA(features,SSE41)) v += "SSE4.1 ";
+    if (hasISA(features,SSE42)) v += "SSE4.2 ";
+    if (hasISA(features,AVX)) v += "AVX ";
+    if (hasISA(features,AVXI)) v += "AVXI ";
+    if (hasISA(features,AVX2)) v += "AVX2 ";
+    if (hasISA(features,AVX512KNL)) v += "AVX512KNL ";
+    if (hasISA(features,AVX512SKX)) v += "AVX512SKX ";
+    if (hasISA(features,NEON)) v += "NEON ";
+    if (hasISA(features,NEON_2X)) v += "2xNEON ";
+    return v;
+  }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+/// Windows Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__WIN32__)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include <psapi.h>
+
+namespace embree
+{
+  std::string getExecutableFileName() {
+    char filename[1024];
+    if (!GetModuleFileName(nullptr, filename, sizeof(filename)))
+      return std::string();
+    return std::string(filename);
+  }
+
+  unsigned int getNumberOfLogicalThreads()
+  {
+    static int nThreads = -1;
+    if (nThreads != -1) return nThreads;
+
+    typedef WORD (WINAPI *GetActiveProcessorGroupCountFunc)();
+    typedef DWORD (WINAPI *GetActiveProcessorCountFunc)(WORD);
+    HMODULE hlib = LoadLibrary("Kernel32");
+    GetActiveProcessorGroupCountFunc pGetActiveProcessorGroupCount = (GetActiveProcessorGroupCountFunc)GetProcAddress(hlib, "GetActiveProcessorGroupCount");
+    GetActiveProcessorCountFunc      pGetActiveProcessorCount      = (GetActiveProcessorCountFunc)     GetProcAddress(hlib, "GetActiveProcessorCount");
+
+    if (pGetActiveProcessorGroupCount && pGetActiveProcessorCount)
+    {
+      int groups = pGetActiveProcessorGroupCount();
+      int totalProcessors = 0;
+      for (int i = 0; i < groups; i++)
+        totalProcessors += pGetActiveProcessorCount(i);
+      nThreads = totalProcessors;
+    }
+    else
+    {
+      SYSTEM_INFO sysinfo;
+      GetSystemInfo(&sysinfo);
+      nThreads = sysinfo.dwNumberOfProcessors;
+    }
+    assert(nThreads);
+    return nThreads;
+  }
+
+  int getTerminalWidth()
+  {
+    HANDLE handle = GetStdHandle(STD_OUTPUT_HANDLE);
+    if (handle == INVALID_HANDLE_VALUE) return 80;
+    CONSOLE_SCREEN_BUFFER_INFO info;
+    memset(&info,0,sizeof(info));
+    GetConsoleScreenBufferInfo(handle, &info);
+    return info.dwSize.X;
+  }
+
+  double getSeconds()
+  {
+    LARGE_INTEGER freq, val;
+    QueryPerformanceFrequency(&freq);
+    QueryPerformanceCounter(&val);
+    return (double)val.QuadPart / (double)freq.QuadPart;
+  }
+
+  void sleepSeconds(double t) {
+    Sleep(DWORD(1000.0*t));
+  }
+
+  size_t getVirtualMemoryBytes()
+  {
+    PROCESS_MEMORY_COUNTERS info;
+    GetProcessMemoryInfo( GetCurrentProcess( ), &info, sizeof(info) );
+    return (size_t)info.QuotaPeakPagedPoolUsage;
+  }
+
+  size_t getResidentMemoryBytes()
+  {
+    PROCESS_MEMORY_COUNTERS info;
+    GetProcessMemoryInfo( GetCurrentProcess( ), &info, sizeof(info) );
+    return (size_t)info.WorkingSetSize;
+  }
+}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Linux Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__LINUX__)
+
+#include <stdio.h>
+#include <unistd.h>
+
+namespace embree
+{
+  std::string getExecutableFileName()
+  {
+    std::string pid = "/proc/" + toString(getpid()) + "/exe";
+    char buf[4096];
+    memset(buf,0,sizeof(buf));
+    if (readlink(pid.c_str(), buf, sizeof(buf)-1) == -1)
+      return std::string();
+    return std::string(buf);
+  }
+
+  size_t getVirtualMemoryBytes()
+  {
+    size_t virt, resident, shared;
+    std::ifstream buffer("/proc/self/statm");
+    buffer >> virt >> resident >> shared;
+    return virt*sysconf(_SC_PAGE_SIZE);
+  }
+
+  size_t getResidentMemoryBytes()
+  {
+    size_t virt, resident, shared;
+    std::ifstream buffer("/proc/self/statm");
+    buffer >> virt >> resident >> shared;
+    return resident*sysconf(_SC_PAGE_SIZE);
+  }
+}
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// FreeBSD Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined (__FreeBSD__)
+
+#include <sys/sysctl.h>
+
+namespace embree
+{
+  std::string getExecutableFileName()
+  {
+    const int mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1 };
+    char buf[4096];
+    memset(buf,0,sizeof(buf));
+    size_t len = sizeof(buf)-1;
+    if (sysctl(mib, 4, buf, &len, 0x0, 0) == -1)
+      return std::string();
+    return std::string(buf);
+  }
+
+  size_t getVirtualMemoryBytes() {
+    return 0;
+  }
+
+  size_t getResidentMemoryBytes() {
+    return 0;
+  }
+}
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Mac OS X Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__MACOSX__)
+
+#include <mach-o/dyld.h>
+
+namespace embree
+{
+  std::string getExecutableFileName()
+  {
+    char buf[4096];
+    uint32_t size = sizeof(buf);
+    if (_NSGetExecutablePath(buf, &size) != 0)
+      return std::string();
+    return std::string(buf);
+  }
+
+  size_t getVirtualMemoryBytes() {
+    return 0;
+  }
+
+  size_t getResidentMemoryBytes() {
+    return 0;
+  }
+}
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Unix Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__UNIX__)
+
+#include <unistd.h>
+#include <sys/ioctl.h>
+#include <sys/time.h>
+#include <pthread.h>
+
+namespace embree
+{
+  unsigned int getNumberOfLogicalThreads()
+  {
+    static int nThreads = -1;
+    if (nThreads != -1) return nThreads;
+
+#if defined(__MACOSX__) || defined(__ANDROID__)
+    nThreads = sysconf(_SC_NPROCESSORS_ONLN); // does not work in Linux LXC container
+    assert(nThreads);
+#else
+    cpu_set_t set;
+    if (pthread_getaffinity_np(pthread_self(), sizeof(set), &set) == 0)
+      nThreads = CPU_COUNT(&set);
+#endif
+
+    assert(nThreads);
+    return nThreads;
+  }
+
+  int getTerminalWidth()
+  {
+    struct winsize info;
+    if (ioctl(STDOUT_FILENO, TIOCGWINSZ, &info) < 0) return 80;
+    return info.ws_col;
+  }
+
+  double getSeconds() {
+    struct timeval tp; gettimeofday(&tp,nullptr);
+    return double(tp.tv_sec) + double(tp.tv_usec)/1E6;
+  }
+
+  void sleepSeconds(double t) {
+    usleep(1000000.0*t);
+  }
+}
+#endif
+
diff --git a/thirdparty/embree-aarch64/common/sys/sysinfo.h b/thirdparty/embree-aarch64/common/sys/sysinfo.h
new file mode 100644
index 0000000000..8e313a59b3
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/sysinfo.h
@@ -0,0 +1,192 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#define CACHELINE_SIZE 64
+
+#if !defined(PAGE_SIZE)
+  #define PAGE_SIZE 4096
+#endif
+
+#define PAGE_SIZE_2M (2*1024*1024)
+#define PAGE_SIZE_4K (4*1024)
+
+#include "platform.h"
+
+/* define isa namespace and ISA bitvector */
+#if defined (__AVX512VL__)
+#  define isa avx512skx
+#  define ISA AVX512SKX
+#  define ISA_STR "AVX512SKX"
+#elif defined (__AVX512F__)
+#  define isa avx512knl
+#  define ISA AVX512KNL
+#  define ISA_STR "AVX512KNL"
+#elif defined (__AVX2__)
+#  define isa avx2
+#  define ISA AVX2
+#  define ISA_STR "AVX2"
+#elif defined(__AVXI__)
+#  define isa avxi
+#  define ISA AVXI
+#  define ISA_STR "AVXI"
+#elif defined(__AVX__)
+#  define isa avx
+#  define ISA AVX
+#  define ISA_STR "AVX"
+#elif defined (__SSE4_2__)
+#  define isa sse42
+#  define ISA SSE42
+#  define ISA_STR "SSE4.2"
+//#elif defined (__SSE4_1__) //  we demote this to SSE2, MacOSX code compiles with SSE41 by default with XCode 11
+//#  define isa sse41
+//#  define ISA SSE41
+//#  define ISA_STR "SSE4.1"
+//#elif defined(__SSSE3__) // we demote this to SSE2, MacOSX code compiles with SSSE3 by default with ICC
+//#  define isa ssse3
+//#  define ISA SSSE3
+//#  define ISA_STR "SSSE3"
+//#elif defined(__SSE3__) // we demote this to SSE2, MacOSX code compiles with SSE3 by default with clang
+//#  define isa sse3
+//#  define ISA SSE3
+//#  define ISA_STR "SSE3"
+#elif defined(__SSE2__) || defined(__SSE3__) || defined(__SSSE3__)
+#  define isa sse2
+#  define ISA SSE2
+#  define ISA_STR "SSE2"
+#elif defined(__SSE__)
+#  define isa sse
+#  define ISA SSE
+#  define ISA_STR "SSE"
+#elif defined(__ARM_NEON)
+// NOTE(LTE): Use sse2 for `isa` for the compatibility at the moment.
+#define isa sse2
+#define ISA NEON
+#define ISA_STR "NEON"
+#else
+#error Unknown ISA
+#endif
+
+namespace embree
+{
+  enum class CPU
+  {
+    XEON_ICE_LAKE,
+    CORE_ICE_LAKE,
+    CORE_TIGER_LAKE,
+    CORE_COMET_LAKE,
+    CORE_CANNON_LAKE,
+    CORE_KABY_LAKE,
+    XEON_SKY_LAKE,
+    CORE_SKY_LAKE,
+    XEON_PHI_KNIGHTS_MILL,
+    XEON_PHI_KNIGHTS_LANDING,
+    XEON_BROADWELL,
+    CORE_BROADWELL,
+    XEON_HASWELL,
+    CORE_HASWELL,
+    XEON_IVY_BRIDGE,
+    CORE_IVY_BRIDGE,
+    SANDY_BRIDGE,
+    NEHALEM,
+    CORE2,
+    CORE1,
+    ARM,
+    UNKNOWN,
+  };
+  
+  /*! get the full path to the running executable */
+  std::string getExecutableFileName();
+
+  /*! return platform name */
+  std::string getPlatformName();
+
+  /*! get the full name of the compiler */
+  std::string getCompilerName();
+
+  /*! return the name of the CPU */
+  std::string getCPUVendor();
+
+  /*! get microprocessor model */
+  CPU getCPUModel(); 
+
+  /*! converts CPU model into string */
+  std::string stringOfCPUModel(CPU model);
+
+  /*! CPU features */
+  static const int CPU_FEATURE_SSE    = 1 << 0;
+  static const int CPU_FEATURE_SSE2   = 1 << 1;
+  static const int CPU_FEATURE_SSE3   = 1 << 2;
+  static const int CPU_FEATURE_SSSE3  = 1 << 3;
+  static const int CPU_FEATURE_SSE41  = 1 << 4;
+  static const int CPU_FEATURE_SSE42  = 1 << 5;
+  static const int CPU_FEATURE_POPCNT = 1 << 6;
+  static const int CPU_FEATURE_AVX    = 1 << 7;
+  static const int CPU_FEATURE_F16C   = 1 << 8;
+  static const int CPU_FEATURE_RDRAND = 1 << 9;
+  static const int CPU_FEATURE_AVX2   = 1 << 10;
+  static const int CPU_FEATURE_FMA3   = 1 << 11;
+  static const int CPU_FEATURE_LZCNT  = 1 << 12;
+  static const int CPU_FEATURE_BMI1   = 1 << 13;
+  static const int CPU_FEATURE_BMI2   = 1 << 14;
+  static const int CPU_FEATURE_AVX512F = 1 << 16;
+  static const int CPU_FEATURE_AVX512DQ = 1 << 17;
+  static const int CPU_FEATURE_AVX512PF = 1 << 18;
+  static const int CPU_FEATURE_AVX512ER = 1 << 19;
+  static const int CPU_FEATURE_AVX512CD = 1 << 20;
+  static const int CPU_FEATURE_AVX512BW = 1 << 21;
+  static const int CPU_FEATURE_AVX512VL = 1 << 22;
+  static const int CPU_FEATURE_AVX512IFMA = 1 << 23;
+  static const int CPU_FEATURE_AVX512VBMI = 1 << 24;
+  static const int CPU_FEATURE_XMM_ENABLED = 1 << 25;
+  static const int CPU_FEATURE_YMM_ENABLED = 1 << 26;
+  static const int CPU_FEATURE_ZMM_ENABLED = 1 << 27;
+  static const int CPU_FEATURE_NEON = 1 << 28;
+  static const int CPU_FEATURE_NEON_2X = 1 << 29;
+
+  /*! get CPU features */
+  int getCPUFeatures();
+
+  /*! convert CPU features into a string */
+  std::string stringOfCPUFeatures(int features);
+
+  /*! creates a string of all supported targets that are supported */
+  std::string supportedTargetList (int isa);
+
+  /*! ISAs */
+  static const int SSE    = CPU_FEATURE_SSE | CPU_FEATURE_XMM_ENABLED;
+  static const int SSE2   = SSE | CPU_FEATURE_SSE2;
+  static const int SSE3   = SSE2 | CPU_FEATURE_SSE3;
+  static const int SSSE3  = SSE3 | CPU_FEATURE_SSSE3;
+  static const int SSE41  = SSSE3 | CPU_FEATURE_SSE41;
+  static const int SSE42  = SSE41 | CPU_FEATURE_SSE42 | CPU_FEATURE_POPCNT;
+  static const int AVX    = SSE42 | CPU_FEATURE_AVX | CPU_FEATURE_YMM_ENABLED;
+  static const int AVXI   = AVX | CPU_FEATURE_F16C | CPU_FEATURE_RDRAND;
+  static const int AVX2   = AVXI | CPU_FEATURE_AVX2 | CPU_FEATURE_FMA3 | CPU_FEATURE_BMI1 | CPU_FEATURE_BMI2 | CPU_FEATURE_LZCNT;
+  static const int AVX512KNL = AVX2 | CPU_FEATURE_AVX512F | CPU_FEATURE_AVX512PF | CPU_FEATURE_AVX512ER | CPU_FEATURE_AVX512CD | CPU_FEATURE_ZMM_ENABLED;
+  static const int AVX512SKX = AVX2 | CPU_FEATURE_AVX512F | CPU_FEATURE_AVX512DQ | CPU_FEATURE_AVX512CD | CPU_FEATURE_AVX512BW | CPU_FEATURE_AVX512VL | CPU_FEATURE_ZMM_ENABLED;
+  static const int NEON = CPU_FEATURE_NEON | CPU_FEATURE_SSE | CPU_FEATURE_SSE2;
+  static const int NEON_2X = CPU_FEATURE_NEON_2X | AVX2;
+
+  /*! converts ISA bitvector into a string */
+  std::string stringOfISA(int features);
+
+  /*! return the number of logical threads of the system */
+  unsigned int getNumberOfLogicalThreads();
+
+  /*! returns the size of the terminal window in characters */
+  int getTerminalWidth();
+
+  /*! returns performance counter in seconds */
+  double getSeconds();
+
+  /*! sleeps the specified number of seconds */
+  void sleepSeconds(double t);
+
+  /*! returns virtual address space occupied by process */
+  size_t getVirtualMemoryBytes();
+
+  /*! returns resident memory required by process */
+  size_t getResidentMemoryBytes();
+}
diff --git a/thirdparty/embree-aarch64/common/sys/thread.cpp b/thirdparty/embree-aarch64/common/sys/thread.cpp
new file mode 100644
index 0000000000..f9ea5b7d96
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/thread.cpp
@@ -0,0 +1,429 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "thread.h"
+#include "sysinfo.h"
+#include "string.h"
+
+#include <iostream>
+#if defined(__ARM_NEON)
+#include "../math/SSE2NEON.h"
+#else
+#include <xmmintrin.h>
+#endif
+
+#if defined(PTHREADS_WIN32)
+#pragma comment (lib, "pthreadVC.lib")
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Windows Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__WIN32__)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+namespace embree
+{
+  /*! set the affinity of a given thread */
+  void setAffinity(HANDLE thread, ssize_t affinity)
+  {
+    typedef WORD (WINAPI *GetActiveProcessorGroupCountFunc)();
+    typedef DWORD (WINAPI *GetActiveProcessorCountFunc)(WORD);
+    typedef BOOL (WINAPI *SetThreadGroupAffinityFunc)(HANDLE, const GROUP_AFFINITY *, PGROUP_AFFINITY);
+    typedef BOOL (WINAPI *SetThreadIdealProcessorExFunc)(HANDLE, PPROCESSOR_NUMBER, PPROCESSOR_NUMBER);
+    HMODULE hlib = LoadLibrary("Kernel32");
+    GetActiveProcessorGroupCountFunc pGetActiveProcessorGroupCount = (GetActiveProcessorGroupCountFunc)GetProcAddress(hlib, "GetActiveProcessorGroupCount");
+    GetActiveProcessorCountFunc pGetActiveProcessorCount = (GetActiveProcessorCountFunc)GetProcAddress(hlib, "GetActiveProcessorCount");
+    SetThreadGroupAffinityFunc pSetThreadGroupAffinity = (SetThreadGroupAffinityFunc)GetProcAddress(hlib, "SetThreadGroupAffinity");
+    SetThreadIdealProcessorExFunc pSetThreadIdealProcessorEx = (SetThreadIdealProcessorExFunc)GetProcAddress(hlib, "SetThreadIdealProcessorEx");
+    if (pGetActiveProcessorGroupCount && pGetActiveProcessorCount && pSetThreadGroupAffinity && pSetThreadIdealProcessorEx)
+    {
+      int groups = pGetActiveProcessorGroupCount();
+      int totalProcessors = 0, group = 0, number = 0;
+      for (int i = 0; i<groups; i++) {
+        int processors = pGetActiveProcessorCount(i);
+        if (totalProcessors + processors > affinity) {
+          group = i;
+          number = (int)affinity - totalProcessors;
+          break;
+        }
+        totalProcessors += processors;
+      }
+
+      GROUP_AFFINITY groupAffinity;
+      groupAffinity.Group = (WORD)group;
+      groupAffinity.Mask = (KAFFINITY)(uint64_t(1) << number);
+      groupAffinity.Reserved[0] = 0;
+      groupAffinity.Reserved[1] = 0;
+      groupAffinity.Reserved[2] = 0;
+      if (!pSetThreadGroupAffinity(thread, &groupAffinity, nullptr))
+        WARNING("SetThreadGroupAffinity failed"); // on purpose only a warning
+
+      PROCESSOR_NUMBER processorNumber;
+      processorNumber.Group = group;
+      processorNumber.Number = number;
+      processorNumber.Reserved = 0;
+      if (!pSetThreadIdealProcessorEx(thread, &processorNumber, nullptr))
+        WARNING("SetThreadIdealProcessorEx failed"); // on purpose only a warning
+    }
+    else
+    {
+      if (!SetThreadAffinityMask(thread, DWORD_PTR(uint64_t(1) << affinity)))
+        WARNING("SetThreadAffinityMask failed"); // on purpose only a warning
+      if (SetThreadIdealProcessor(thread, (DWORD)affinity) == (DWORD)-1)
+        WARNING("SetThreadIdealProcessor failed"); // on purpose only a warning
+      }
+  }
+
+  /*! set affinity of the calling thread */
+  void setAffinity(ssize_t affinity) {
+    setAffinity(GetCurrentThread(), affinity);
+  }
+
+  struct ThreadStartupData
+  {
+  public:
+    ThreadStartupData (thread_func f, void* arg)
+      : f(f), arg(arg) {}
+  public:
+    thread_func f;
+    void* arg;
+  };
+
+  DWORD WINAPI threadStartup(LPVOID ptr)
+  {
+    ThreadStartupData* parg = (ThreadStartupData*) ptr;
+    _mm_setcsr(_mm_getcsr() | /*FTZ:*/ (1<<15) | /*DAZ:*/ (1<<6));
+    parg->f(parg->arg);
+    delete parg;
+    parg = nullptr;
+    return 0;
+  }
+
+#if !defined(PTHREADS_WIN32)
+
+  /*! creates a hardware thread running on specific core */
+  thread_t createThread(thread_func f, void* arg, size_t stack_size, ssize_t threadID)
+  {
+    HANDLE thread = CreateThread(nullptr, stack_size, threadStartup, new ThreadStartupData(f,arg), 0, nullptr);
+    if (thread == nullptr) FATAL("CreateThread failed");
+    if (threadID >= 0) setAffinity(thread, threadID);
+    return thread_t(thread);
+  }
+
+  /*! the thread calling this function gets yielded */
+  void yield() {
+    SwitchToThread();
+  }
+
+  /*! waits until the given thread has terminated */
+  void join(thread_t tid) {
+    WaitForSingleObject(HANDLE(tid), INFINITE);
+    CloseHandle(HANDLE(tid));
+  }
+
+  /*! creates thread local storage */
+  tls_t createTls() {
+    return tls_t(size_t(TlsAlloc()));
+  }
+
+  /*! set the thread local storage pointer */
+  void setTls(tls_t tls, void* const ptr) {
+    TlsSetValue(DWORD(size_t(tls)), ptr);
+  }
+
+  /*! return the thread local storage pointer */
+  void* getTls(tls_t tls) {
+    return TlsGetValue(DWORD(size_t(tls)));
+  }
+
+  /*! destroys thread local storage identifier */
+  void destroyTls(tls_t tls) {
+    TlsFree(DWORD(size_t(tls)));
+  }
+#endif
+}
+
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Linux Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__LINUX__)
+
+#include <fstream>
+#include <sstream>
+#include <algorithm>
+
+#if defined(__ANDROID__)
+#include <pthread.h>
+#endif
+
+namespace embree
+{
+  static MutexSys mutex;
+  static std::vector<size_t> threadIDs;
+
+#if !defined(__ANDROID__) // TODO(LTE): Implement for Android target
+  /* changes thread ID mapping such that we first fill up all thread on one core */
+  size_t mapThreadID(size_t threadID)
+  {
+    Lock<MutexSys> lock(mutex);
+
+    if (threadIDs.size() == 0)
+    {
+      /* parse thread/CPU topology */
+      for (size_t cpuID=0;;cpuID++)
+      {
+        std::fstream fs;
+        std::string cpu = std::string("/sys/devices/system/cpu/cpu") + std::to_string((long long)cpuID) + std::string("/topology/thread_siblings_list");
+        fs.open (cpu.c_str(), std::fstream::in);
+        if (fs.fail()) break;
+
+        int i;
+        while (fs >> i)
+        {
+          if (std::none_of(threadIDs.begin(),threadIDs.end(),[&] (int id) { return id == i; }))
+            threadIDs.push_back(i);
+          if (fs.peek() == ',')
+            fs.ignore();
+        }
+        fs.close();
+      }
+
+#if 0
+      for (size_t i=0;i<threadIDs.size();i++)
+        std::cout << i << " -> " << threadIDs[i] << std::endl;
+#endif
+
+      /* verify the mapping and do not use it if the mapping has errors */
+      for (size_t i=0;i<threadIDs.size();i++) {
+        for (size_t j=0;j<threadIDs.size();j++) {
+          if (i != j && threadIDs[i] == threadIDs[j]) {
+            threadIDs.clear();
+          }
+        }
+      }
+    }
+
+    /* re-map threadIDs if mapping is available */
+    size_t ID = threadID;
+    if (threadID < threadIDs.size())
+      ID = threadIDs[threadID];
+
+    /* find correct thread to affinitize to */
+    cpu_set_t set;
+    if (pthread_getaffinity_np(pthread_self(), sizeof(set), &set) == 0)
+    {
+      for (int i=0, j=0; i<CPU_SETSIZE; i++)
+      {
+        if (!CPU_ISSET(i,&set)) continue;
+
+        if (j == ID) {
+          ID = i;
+          break;
+        }
+        j++;
+      }
+    }
+
+    return ID;
+  }
+#endif
+
+  /*! set affinity of the calling thread */
+  void setAffinity(ssize_t affinity)
+  {
+#if defined(__ANDROID__)
+    // TODO(LTE): Implement
+#else
+    cpu_set_t cset;
+    CPU_ZERO(&cset);
+    size_t threadID = mapThreadID(affinity);
+    CPU_SET(threadID, &cset);
+
+    pthread_setaffinity_np(pthread_self(), sizeof(cset), &cset);
+#endif
+  }
+}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// FreeBSD Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__FreeBSD__)
+
+#include <pthread_np.h>
+
+namespace embree
+{
+  /*! set affinity of the calling thread */
+  void setAffinity(ssize_t affinity)
+  {
+    cpuset_t cset;
+    CPU_ZERO(&cset);
+    CPU_SET(affinity, &cset);
+
+    pthread_setaffinity_np(pthread_self(), sizeof(cset), &cset);
+  }
+}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// MacOSX Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__MACOSX__)
+
+#include <mach/thread_act.h>
+#include <mach/thread_policy.h>
+#include <mach/mach_init.h>
+
+namespace embree
+{
+  /*! set affinity of the calling thread */
+  void setAffinity(ssize_t affinity)
+  {
+    thread_affinity_policy ap;
+    ap.affinity_tag = affinity;
+    if (thread_policy_set(mach_thread_self(),THREAD_AFFINITY_POLICY,(thread_policy_t)&ap,THREAD_AFFINITY_POLICY_COUNT) != KERN_SUCCESS)
+      WARNING("setting thread affinity failed"); // on purpose only a warning
+  }
+}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////
+/// Unix Platform
+////////////////////////////////////////////////////////////////////////////////
+
+#if defined(__UNIX__) || defined(PTHREADS_WIN32)
+
+#include <pthread.h>
+#include <sched.h>
+
+#if defined(__USE_NUMA__)
+#include <numa.h>
+#endif
+
+namespace embree
+{
+  struct ThreadStartupData
+  {
+  public:
+    ThreadStartupData (thread_func f, void* arg, int affinity)
+      : f(f), arg(arg), affinity(affinity) {}
+  public:
+    thread_func f;
+    void* arg;
+    ssize_t affinity;
+  };
+
+  static void* threadStartup(ThreadStartupData* parg)
+  {
+    _mm_setcsr(_mm_getcsr() | /*FTZ:*/ (1<<15) | /*DAZ:*/ (1<<6));
+
+    /*! Mac OS X does not support setting affinity at thread creation time */
+#if defined(__MACOSX__)
+    if (parg->affinity >= 0)
+	setAffinity(parg->affinity);
+#endif
+
+    parg->f(parg->arg);
+    delete parg;
+    parg = nullptr;
+    return nullptr;
+  }
+
+  /*! creates a hardware thread running on specific core */
+  thread_t createThread(thread_func f, void* arg, size_t stack_size, ssize_t threadID)
+  {
+    /* set stack size */
+    pthread_attr_t attr;
+    pthread_attr_init(&attr);
+    if (stack_size > 0) pthread_attr_setstacksize (&attr, stack_size);
+
+    /* create thread */
+    pthread_t* tid = new pthread_t;
+    if (pthread_create(tid,&attr,(void*(*)(void*))threadStartup,new ThreadStartupData(f,arg,threadID)) != 0) {
+      pthread_attr_destroy(&attr);
+      delete tid;
+      FATAL("pthread_create failed");
+    }
+    pthread_attr_destroy(&attr);
+
+    /* set affinity */
+#if defined(__LINUX__) && !defined(__ANDROID__)
+    if (threadID >= 0) {
+      cpu_set_t cset;
+      CPU_ZERO(&cset);
+      threadID = mapThreadID(threadID);
+      CPU_SET(threadID, &cset);
+      pthread_setaffinity_np(*tid, sizeof(cset), &cset);
+    }
+#elif defined(__FreeBSD__)
+    if (threadID >= 0) {
+      cpuset_t cset;
+      CPU_ZERO(&cset);
+      CPU_SET(threadID, &cset);
+      pthread_setaffinity_np(*tid, sizeof(cset), &cset);
+    }
+#endif
+
+    return thread_t(tid);
+  }
+
+  /*! the thread calling this function gets yielded */
+  void yield() {
+    sched_yield();
+  }
+
+  /*! waits until the given thread has terminated */
+  void join(thread_t tid) {
+    if (pthread_join(*(pthread_t*)tid, nullptr) != 0)
+      FATAL("pthread_join failed");
+    delete (pthread_t*)tid;
+  }
+
+  /*! creates thread local storage */
+  tls_t createTls()
+  {
+    pthread_key_t* key = new pthread_key_t;
+    if (pthread_key_create(key,nullptr) != 0) {
+      delete key;
+      FATAL("pthread_key_create failed");
+    }
+
+    return tls_t(key);
+  }
+
+  /*! return the thread local storage pointer */
+  void* getTls(tls_t tls)
+  {
+    assert(tls);
+    return pthread_getspecific(*(pthread_key_t*)tls);
+  }
+
+  /*! set the thread local storage pointer */
+  void setTls(tls_t tls, void* const ptr)
+  {
+    assert(tls);
+    if (pthread_setspecific(*(pthread_key_t*)tls, ptr) != 0)
+      FATAL("pthread_setspecific failed");
+  }
+
+  /*! destroys thread local storage identifier */
+  void destroyTls(tls_t tls)
+  {
+    assert(tls);
+    if (pthread_key_delete(*(pthread_key_t*)tls) != 0)
+      FATAL("pthread_key_delete failed");
+    delete (pthread_key_t*)tls;
+  }
+}
+
+#endif
diff --git a/thirdparty/embree-aarch64/common/sys/thread.h b/thirdparty/embree-aarch64/common/sys/thread.h
new file mode 100644
index 0000000000..45da6e6a70
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/thread.h
@@ -0,0 +1,46 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "platform.h"
+#include "mutex.h"
+#include "alloc.h"
+#include "vector.h"
+#include <vector>
+
+namespace embree
+{
+  /*! type for thread */
+  typedef struct opaque_thread_t* thread_t;
+
+  /*! signature of thread start function */
+  typedef void (*thread_func)(void*);
+
+  /*! creates a hardware thread running on specific logical thread */
+  thread_t createThread(thread_func f, void* arg, size_t stack_size = 0, ssize_t threadID = -1);
+
+  /*! set affinity of the calling thread */
+  void setAffinity(ssize_t affinity);
+
+  /*! the thread calling this function gets yielded */
+  void yield();
+
+  /*! waits until the given thread has terminated */
+  void join(thread_t tid);
+
+  /*! type for handle to thread local storage */
+  typedef struct opaque_tls_t* tls_t;
+
+  /*! creates thread local storage */
+  tls_t createTls();
+
+  /*! set the thread local storage pointer */
+  void setTls(tls_t tls, void* const ptr);
+
+  /*! return the thread local storage pointer */
+  void* getTls(tls_t tls);
+
+  /*! destroys thread local storage identifier */
+  void destroyTls(tls_t tls);
+}
diff --git a/thirdparty/embree-aarch64/common/sys/vector.h b/thirdparty/embree-aarch64/common/sys/vector.h
new file mode 100644
index 0000000000..e41794de7c
--- /dev/null
+++ b/thirdparty/embree-aarch64/common/sys/vector.h
@@ -0,0 +1,242 @@
+// Copyright 2009-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "alloc.h"
+#include <algorithm>
+
+namespace embree
+{
+   template<typename T, typename allocator>
+    class vector_t
+    {
+    public:
+      typedef T value_type;
+      typedef T* iterator;
+      typedef const T* const_iterator;
+    
+      __forceinline vector_t () 
+        : size_active(0), size_alloced(0), items(nullptr) {}
+    
+      __forceinline explicit vector_t (size_t sz) 
+        : size_active(0), size_alloced(0), items(nullptr) { internal_resize_init(sz); }
+    
+      template<typename M>
+      __forceinline explicit vector_t (M alloc, size_t sz) 
+      : alloc(alloc), size_active(0), size_alloced(0), items(nullptr) { internal_resize_init(sz); }
+    
+      __forceinline ~vector_t() {
+        clear();
+      }
+    
+      __forceinline vector_t (const vector_t& other)
+      {
+        size_active = other.size_active;
+        size_alloced = other.size_alloced;
+        items = alloc.allocate(size_alloced);
+        for (size_t i=0; i<size_active; i++) 
+          ::new (&items[i]) value_type(other.items[i]);
+      }
+    
+      __forceinline vector_t (vector_t&& other)
+        : alloc(std::move(other.alloc))
+      {
+        size_active = other.size_active; other.size_active = 0;
+        size_alloced = other.size_alloced; other.size_alloced = 0;
+        items = other.items; other.items = nullptr;
+      }
+
+      __forceinline vector_t& operator=(const vector_t& other) 
+      {
+        resize(other.size_active);
+        for (size_t i=0; i<size_active; i++)
+          items[i] = value_type(other.items[i]);
+        return *this;
+      }
+
+      __forceinline vector_t& operator=(vector_t&& other) 
+      {
+        clear();
+        alloc = std::move(other.alloc);
+        size_active = other.size_active; other.size_active = 0;
+        size_alloced = other.size_alloced; other.size_alloced = 0;
+        items = other.items; other.items = nullptr;
+        return *this;
+      }
+
+      /********************** Iterators  ****************************/
+    
+      __forceinline       iterator begin()       { return items; };
+      __forceinline const_iterator begin() const { return items; };
+
+      __forceinline       iterator end  ()       { return items+size_active; };
+      __forceinline const_iterator end  () const { return items+size_active; };
+
+
+      /********************** Capacity ****************************/
+
+      __forceinline bool   empty    () const { return size_active == 0; }
+      __forceinline size_t size     () const { return size_active; }
+      __forceinline size_t capacity () const { return size_alloced; }
+
+
+      __forceinline void resize(size_t new_size) {
+        internal_resize(new_size,internal_grow_size(new_size));
+      }
+
+      __forceinline void reserve(size_t new_alloced) 
+      {
+        /* do nothing if container already large enough */
+        if (new_alloced <= size_alloced) 
+          return;
+
+        /* resize exact otherwise */
+        internal_resize(size_active,new_alloced);
+      }
+
+      __forceinline void shrink_to_fit() {
+        internal_resize(size_active,size_active);
+      }
+
+      /******************** Element access **************************/
+
+      __forceinline       T& operator[](size_t i)       { assert(i < size_active); return items[i]; }
+      __forceinline const T& operator[](size_t i) const { assert(i < size_active); return items[i]; }
+
+      __forceinline       T& at(size_t i)       { assert(i < size_active); return items[i]; }
+      __forceinline const T& at(size_t i) const { assert(i < size_active); return items[i]; }
+
+      __forceinline T& front() const { assert(size_active > 0); return items[0]; };
+      __forceinline T& back () const { assert(size_active > 0); return items[size_active-1]; };
+
+      __forceinline       T* data()       { return items; };
+      __forceinline const T* data() const { return items; };
+
+     
+      /******************** Modifiers **************************/
+
+      __forceinline void push_back(const T& nt) 
+      {
+        const T v = nt; // need local copy as input reference could point to this vector
+        internal_resize(size_active,internal_grow_size(size_active+1));
+        ::new (&items[size_active++]) T(v);
+      }
+
+      __forceinline void pop_back() 
+      {
+        assert(!empty());
+        size_active--;
+        alloc.destroy(&items[size_active]);
+      }
+
+      __forceinline void clear() 
+      {
+        /* destroy elements */
+        for (size_t i=0; i<size_active; i++)
+          alloc.destroy(&items[i]);
+        
+        /* free memory */
+        alloc.deallocate(items,size_alloced); 
+        items = nullptr;
+        size_active = size_alloced = 0;
+      }
+
+    /******************** Comparisons **************************/
+    
+    friend bool operator== (const vector_t& a, const vector_t& b) 
+    {
+      if (a.size() != b.size()) return false;
+      for (size_t i=0; i<a.size(); i++)
+        if (a[i] != b[i])
+          return false;
+      return true;
+    }
+
+    friend bool operator!= (const vector_t& a, const vector_t& b) {
+      return !(a==b);
+    }
+
+    private:
+
+      __forceinline void internal_resize_init(size_t new_active)
+      {
+        assert(size_active == 0); 
+        assert(size_alloced == 0);
+        assert(items == nullptr);
+        if (new_active == 0) return;
+        items = alloc.allocate(new_active);
+        for (size_t i=0; i<new_active; i++) ::new (&items[i]) T();
+        size_active = new_active;
+        size_alloced = new_active;
+      }
+
+      __forceinline void internal_resize(size_t new_active, size_t new_alloced)
+      {
+        assert(new_active <= new_alloced); 
+
+        /* destroy elements */
+        if (new_active < size_active) 
+        {
+          for (size_t i=new_active; i<size_active; i++)
+            alloc.destroy(&items[i]);
+          size_active = new_active;
+        }
+
+        /* only reallocate if necessary */
+        if (new_alloced == size_alloced) {
+          for (size_t i=size_active; i<new_active; i++) ::new (&items[i]) T;
+          size_active = new_active;
+          return;
+        }
+
+        /* reallocate and copy items */
+        T* old_items = items;
+        items = alloc.allocate(new_alloced);
+        for (size_t i=0; i<size_active; i++) {
+          ::new (&items[i]) T(std::move(old_items[i]));
+          alloc.destroy(&old_items[i]);
+        }
+
+        for (size_t i=size_active; i<new_active; i++) {
+          ::new (&items[i]) T;
+        }
+
+        alloc.deallocate(old_items,size_alloced);
+        size_active = new_active;
+        size_alloced = new_alloced;
+      }
+
+      __forceinline size_t internal_grow_size(size_t new_alloced)
+      {
+        /* do nothing if container already large enough */
+        if (new_alloced <= size_alloced) 
+          return size_alloced;
+
+        /* resize to next power of 2 otherwise */
+        size_t new_size_alloced = size_alloced;
+        while (new_size_alloced < new_alloced) {
+          new_size_alloced = std::max(size_t(1),2*new_size_alloced);
+        }
+        return new_size_alloced;
+      }
+
+    private:
+      allocator alloc;
+      size_t size_active;    // number of valid items
+      size_t size_alloced;   // number of items allocated
+      T* items;              // data array
+    };
+
+  /*! vector class that performs standard allocations */
+  template<typename T>
+    using vector = vector_t<T,std::allocator<T>>;
+
+  /*! vector class that performs aligned allocations */
+  template<typename T>
+    using avector = vector_t<T,aligned_allocator<T,std::alignment_of<T>::value> >;
+  
+  /*! vector class that performs OS allocations */
+  template<typename T>
+    using ovector = vector_t<T,os_allocator<T> >;
+}