Update embree to 3.13.5

13 files changed, 300 insertions, 116 deletions

diff --git a/thirdparty/embree/common/math/bbox.h b/thirdparty/embree/common/math/bbox.h
index bc43155358..e4eb3df9a4 100644
--- a/thirdparty/embree/common/math/bbox.h
+++ b/thirdparty/embree/common/math/bbox.h
@@ -77,7 +77,7 @@ namespace embree
     return lower > upper;
   }
 
-#if defined(__SSE__)
+#if defined(__SSE__) || defined(__ARM_NEON)
   template<> __forceinline bool BBox<Vec3fa>::empty() const {
     return !all(le_mask(lower,upper));
   }
@@ -196,11 +196,11 @@ namespace embree
   }
 
   template<> __inline bool subset( const BBox<Vec3fa>& a, const BBox<Vec3fa>& b ) {
-    return all(ge_mask(a.lower,b.lower)) & all(le_mask(a.upper,b.upper));
+    return all(ge_mask(a.lower,b.lower)) && all(le_mask(a.upper,b.upper));
   }
 
   template<> __inline bool subset( const BBox<Vec3fx>& a, const BBox<Vec3fx>& b ) {
-    return all(ge_mask(a.lower,b.lower)) & all(le_mask(a.upper,b.upper));
+    return all(ge_mask(a.lower,b.lower)) && all(le_mask(a.upper,b.upper));
   }
   
   /*! blending */
@@ -228,11 +228,11 @@ namespace embree
 /// SSE / AVX / MIC specializations
 ////////////////////////////////////////////////////////////////////////////////
 
-#if defined __SSE__
+#if defined (__SSE__) || defined(__ARM_NEON)
 #include "../simd/sse.h"
 #endif
 
-#if defined __AVX__
+#if defined (__AVX__)
 #include "../simd/avx.h"
 #endif
 
diff --git a/thirdparty/embree/common/math/color.h b/thirdparty/embree/common/math/color.h
index 529584ea16..e62e4ad2a4 100644
--- a/thirdparty/embree/common/math/color.h
+++ b/thirdparty/embree/common/math/color.h
@@ -152,21 +152,38 @@ namespace embree
   }
   __forceinline const Color rcp  ( const Color& a )
   {
+#if defined(__aarch64__)
+    __m128 reciprocal = _mm_rcp_ps(a.m128);
+    reciprocal = vmulq_f32(vrecpsq_f32(a.m128, reciprocal), reciprocal);
+    reciprocal = vmulq_f32(vrecpsq_f32(a.m128, reciprocal), reciprocal);
+    return (const Color)reciprocal;
+#else
 #if defined(__AVX512VL__)
     const Color r = _mm_rcp14_ps(a.m128);
 #else
     const Color r = _mm_rcp_ps(a.m128);
 #endif
-    return _mm_sub_ps(_mm_add_ps(r, r), _mm_mul_ps(_mm_mul_ps(r, r), a));
+    return _mm_add_ps(r,_mm_mul_ps(r, _mm_sub_ps(_mm_set1_ps(1.0f), _mm_mul_ps(a, r))));   // computes r + r * (1 - a * r)
+
+#endif  //defined(__aarch64__)
   }
   __forceinline const Color rsqrt( const Color& a )
   {
+#if defined(__aarch64__)
+    __m128 r = _mm_rsqrt_ps(a.m128);
+    r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+    r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+    return r;
+#else
+
 #if defined(__AVX512VL__)
     __m128 r = _mm_rsqrt14_ps(a.m128);
 #else
     __m128 r = _mm_rsqrt_ps(a.m128);
 #endif
     return _mm_add_ps(_mm_mul_ps(_mm_set1_ps(1.5f),r), _mm_mul_ps(_mm_mul_ps(_mm_mul_ps(a, _mm_set1_ps(-0.5f)), r), _mm_mul_ps(r, r)));
+
+#endif  //defined(__aarch64__)
   }
   __forceinline const Color sqrt ( const Color& a ) { return _mm_sqrt_ps(a.m128); }
 
diff --git a/thirdparty/embree/common/math/constants.cpp b/thirdparty/embree/common/math/constants.cpp
index 03919ae20c..f51c642bfc 100644
--- a/thirdparty/embree/common/math/constants.cpp
+++ b/thirdparty/embree/common/math/constants.cpp
@@ -5,23 +5,4 @@
 
 namespace embree
 {
-  TrueTy True;
-  FalseTy False;
-  ZeroTy zero;
-  OneTy one;
-  NegInfTy neg_inf;
-  PosInfTy inf;
-  PosInfTy pos_inf;
-  NaNTy nan;
-  UlpTy ulp;
-  PiTy pi;
-  OneOverPiTy one_over_pi;
-  TwoPiTy two_pi;
-  OneOverTwoPiTy one_over_two_pi;
-  FourPiTy four_pi;
-  OneOverFourPiTy one_over_four_pi;
-  StepTy step;
-  ReverseStepTy reverse_step;
-  EmptyTy empty;
-  UndefinedTy undefined;
 }
diff --git a/thirdparty/embree/common/math/constants.h b/thirdparty/embree/common/math/constants.h
index 578473a8ab..07a1a868ba 100644
--- a/thirdparty/embree/common/math/constants.h
+++ b/thirdparty/embree/common/math/constants.h
@@ -24,13 +24,13 @@ namespace embree
     __forceinline operator bool( ) const { return true; }
   };
 
-  extern MAYBE_UNUSED TrueTy True;
+  const constexpr TrueTy True = TrueTy();
 
   struct FalseTy {
     __forceinline operator bool( ) const { return false; }
   };
 
-  extern MAYBE_UNUSED FalseTy False;
+  const constexpr FalseTy False = FalseTy();
   
   struct ZeroTy
   {
@@ -48,7 +48,7 @@ namespace embree
     __forceinline operator unsigned char     ( ) const { return 0; }
   }; 
 
-  extern MAYBE_UNUSED ZeroTy zero;
+  const constexpr ZeroTy zero = ZeroTy();
 
   struct OneTy
   {
@@ -66,7 +66,7 @@ namespace embree
     __forceinline operator unsigned char     ( ) const { return 1; }
   };
 
-  extern MAYBE_UNUSED OneTy one;
+  const constexpr OneTy one = OneTy();
 
   struct NegInfTy
   {
@@ -85,7 +85,7 @@ namespace embree
 
   };
 
-  extern MAYBE_UNUSED NegInfTy neg_inf;
+  const constexpr NegInfTy neg_inf = NegInfTy();
 
   struct PosInfTy
   {
@@ -103,8 +103,8 @@ namespace embree
     __forceinline operator unsigned char     ( ) const { return std::numeric_limits<unsigned char>::max(); }
   };
 
-  extern MAYBE_UNUSED PosInfTy inf;
-  extern MAYBE_UNUSED PosInfTy pos_inf;
+  const constexpr PosInfTy     inf = PosInfTy();
+  const constexpr PosInfTy pos_inf = PosInfTy();
 
   struct NaNTy
   {
@@ -112,15 +112,15 @@ namespace embree
     __forceinline operator float ( ) const { return std::numeric_limits<float>::quiet_NaN(); }
   };
 
-  extern MAYBE_UNUSED NaNTy nan;
+  const constexpr NaNTy nan = NaNTy();
 
   struct UlpTy
   {
     __forceinline operator double( ) const { return std::numeric_limits<double>::epsilon(); }
     __forceinline operator float ( ) const { return std::numeric_limits<float>::epsilon(); }
   };
-
-  extern MAYBE_UNUSED UlpTy ulp;
+  
+  const constexpr UlpTy ulp = UlpTy();
 
   struct PiTy
   {
@@ -128,7 +128,7 @@ namespace embree
     __forceinline operator float ( ) const { return float(M_PI); }
   };
 
-  extern MAYBE_UNUSED PiTy pi;
+  const constexpr PiTy pi = PiTy();
 
   struct OneOverPiTy
   {
@@ -136,7 +136,7 @@ namespace embree
     __forceinline operator float ( ) const { return float(M_1_PI); }
   };
 
-  extern MAYBE_UNUSED OneOverPiTy one_over_pi;
+  const constexpr OneOverPiTy one_over_pi = OneOverPiTy();
 
   struct TwoPiTy
   {
@@ -144,7 +144,7 @@ namespace embree
     __forceinline operator float ( ) const { return float(2.0*M_PI); }
   };
 
-  extern MAYBE_UNUSED TwoPiTy two_pi;
+  const constexpr TwoPiTy two_pi = TwoPiTy();
 
   struct OneOverTwoPiTy
   {
@@ -152,7 +152,7 @@ namespace embree
     __forceinline operator float ( ) const { return float(0.5*M_1_PI); }
   };
 
-  extern MAYBE_UNUSED OneOverTwoPiTy one_over_two_pi;
+  const constexpr OneOverTwoPiTy one_over_two_pi = OneOverTwoPiTy();
 
   struct FourPiTy
   {
@@ -160,7 +160,7 @@ namespace embree
     __forceinline operator float ( ) const { return float(4.0*M_PI); }
   };
 
-  extern MAYBE_UNUSED FourPiTy four_pi;
+  const constexpr FourPiTy four_pi = FourPiTy();
 
   struct OneOverFourPiTy
   {
@@ -168,30 +168,42 @@ namespace embree
     __forceinline operator float ( ) const { return float(0.25*M_1_PI); }
   };
 
-  extern MAYBE_UNUSED OneOverFourPiTy one_over_four_pi;
+  const constexpr OneOverFourPiTy one_over_four_pi = OneOverFourPiTy();
 
   struct StepTy {
+    __forceinline operator          double   ( ) const { return 0; }
+    __forceinline operator          float    ( ) const { return 0; }
+    __forceinline operator          long long( ) const { return 0; }
+    __forceinline operator unsigned long long( ) const { return 0; }
+    __forceinline operator          long     ( ) const { return 0; }
+    __forceinline operator unsigned long     ( ) const { return 0; }
+    __forceinline operator          int      ( ) const { return 0; }
+    __forceinline operator unsigned int      ( ) const { return 0; }
+    __forceinline operator          short    ( ) const { return 0; }
+    __forceinline operator unsigned short    ( ) const { return 0; }
+    __forceinline operator          char     ( ) const { return 0; }
+    __forceinline operator unsigned char     ( ) const { return 0; }
   };
 
-  extern MAYBE_UNUSED StepTy step;
+  const constexpr StepTy step = StepTy();
 
   struct ReverseStepTy {
   };
 
-  extern MAYBE_UNUSED ReverseStepTy reverse_step;
+  const constexpr ReverseStepTy reverse_step = ReverseStepTy();
 
   struct EmptyTy {
   };
 
-  extern MAYBE_UNUSED EmptyTy empty;
+  const constexpr EmptyTy empty = EmptyTy();
 
   struct FullTy {
   };
 
-  extern MAYBE_UNUSED FullTy full;
+  const constexpr FullTy full = FullTy();
 
   struct UndefinedTy {
   };
 
-  extern MAYBE_UNUSED UndefinedTy undefined;
+  const constexpr UndefinedTy undefined = UndefinedTy();
 }
diff --git a/thirdparty/embree/common/math/math.h b/thirdparty/embree/common/math/math.h
index 4bc54c1a6a..7930c17727 100644
--- a/thirdparty/embree/common/math/math.h
+++ b/thirdparty/embree/common/math/math.h
@@ -53,6 +53,16 @@ namespace embree
 
   __forceinline float rcp  ( const float x )
   {
+#if defined(__aarch64__)
+      // Move scalar to vector register and do rcp.
+      __m128 a;
+      a[0] = x;
+      float32x4_t reciprocal = vrecpeq_f32(a);
+      reciprocal = vmulq_f32(vrecpsq_f32(a, reciprocal), reciprocal);
+      reciprocal = vmulq_f32(vrecpsq_f32(a, reciprocal), reciprocal);
+      return reciprocal[0];
+#else
+
     const __m128 a = _mm_set_ss(x);
 
 #if defined(__AVX512VL__)
@@ -66,30 +76,71 @@ namespace embree
 #else
     return _mm_cvtss_f32(_mm_mul_ss(r,_mm_sub_ss(_mm_set_ss(2.0f), _mm_mul_ss(r, a))));
 #endif
+
+#endif  //defined(__aarch64__)
   }
 
   __forceinline float signmsk ( const float x ) {
+#if defined(__aarch64__)
+      // FP and Neon shares same vector register in arm64
+      __m128 a;
+      __m128i b;
+      a[0] = x;
+      b[0] = 0x80000000;
+      a = _mm_and_ps(a, vreinterpretq_f32_s32(b));
+      return a[0];
+#else
     return _mm_cvtss_f32(_mm_and_ps(_mm_set_ss(x),_mm_castsi128_ps(_mm_set1_epi32(0x80000000))));
+#endif
   }
   __forceinline float xorf( const float x, const float y ) {
+#if defined(__aarch64__)
+      // FP and Neon shares same vector register in arm64
+      __m128 a;
+      __m128 b;
+      a[0] = x;
+      b[0] = y;
+      a = _mm_xor_ps(a, b);
+      return a[0];
+#else
     return _mm_cvtss_f32(_mm_xor_ps(_mm_set_ss(x),_mm_set_ss(y)));
+#endif
   }
   __forceinline float andf( const float x, const unsigned y ) {
+#if defined(__aarch64__)
+      // FP and Neon shares same vector register in arm64
+      __m128 a;
+      __m128i b;
+      a[0] = x;
+      b[0] = y;
+      a = _mm_and_ps(a, vreinterpretq_f32_s32(b));
+      return a[0];
+#else
     return _mm_cvtss_f32(_mm_and_ps(_mm_set_ss(x),_mm_castsi128_ps(_mm_set1_epi32(y))));
+#endif
   }
   __forceinline float rsqrt( const float x )
   {
+#if defined(__aarch64__)
+      // FP and Neon shares same vector register in arm64
+      __m128 a;
+      a[0] = x;
+      __m128 value = _mm_rsqrt_ps(a);
+      value = vmulq_f32(value, vrsqrtsq_f32(vmulq_f32(a, value), value));
+      value = vmulq_f32(value, vrsqrtsq_f32(vmulq_f32(a, value), value));
+      return value[0];
+#else
+
     const __m128 a = _mm_set_ss(x);
 #if defined(__AVX512VL__)
     __m128 r = _mm_rsqrt14_ss(_mm_set_ss(0.0f),a);
 #else
     __m128 r = _mm_rsqrt_ss(a);
 #endif
-    r = _mm_add_ss(_mm_mul_ss(_mm_set_ss(1.5f), r), _mm_mul_ss(_mm_mul_ss(_mm_mul_ss(a, _mm_set_ss(-0.5f)), r), _mm_mul_ss(r, r)));
-#if defined(__ARM_NEON)
-    r = _mm_add_ss(_mm_mul_ss(_mm_set_ss(1.5f), r), _mm_mul_ss(_mm_mul_ss(_mm_mul_ss(a, _mm_set_ss(-0.5f)), r), _mm_mul_ss(r, r)));
+    const __m128 c = _mm_add_ss(_mm_mul_ss(_mm_set_ss(1.5f), r),
+                                _mm_mul_ss(_mm_mul_ss(_mm_mul_ss(a, _mm_set_ss(-0.5f)), r), _mm_mul_ss(r, r)));
+    return _mm_cvtss_f32(c);
 #endif
-    return _mm_cvtss_f32(r);
   }
 
 #if defined(__WIN32__) && defined(_MSC_VER) && (_MSC_VER <= 1700)
@@ -146,7 +197,17 @@ namespace embree
   __forceinline double floor( const double x ) { return ::floor (x); }
   __forceinline double ceil ( const double x ) { return ::ceil (x); }
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__)
+    __forceinline float mini(float a, float b) {
+        // FP and Neon shares same vector register in arm64
+        __m128 x;
+        __m128 y;
+        x[0] = a;
+        y[0] = b;
+        x = _mm_min_ps(x, y);
+        return x[0];
+    }
+#elif defined(__SSE4_1__)
   __forceinline float mini(float a, float b) {
     const __m128i ai = _mm_castps_si128(_mm_set_ss(a));
     const __m128i bi = _mm_castps_si128(_mm_set_ss(b));
@@ -155,7 +216,17 @@ namespace embree
   }
 #endif
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__)
+    __forceinline float maxi(float a, float b) {
+        // FP and Neon shares same vector register in arm64
+        __m128 x;
+        __m128 y;
+        x[0] = a;
+        y[0] = b;
+        x = _mm_max_ps(x, y);
+        return x[0];
+    }
+#elif defined(__SSE4_1__)
   __forceinline float maxi(float a, float b) {
     const __m128i ai = _mm_castps_si128(_mm_set_ss(a));
     const __m128i bi = _mm_castps_si128(_mm_set_ss(b));
@@ -172,9 +243,12 @@ namespace embree
   __forceinline  int64_t min(int64_t  a, int64_t  b) { return a<b ? a:b; }
   __forceinline    float min(float    a, float    b) { return a<b ? a:b; }
   __forceinline   double min(double   a, double   b) { return a<b ? a:b; }
-#if defined(__64BIT__)
+#if defined(__64BIT__) || defined(__EMSCRIPTEN__)
   __forceinline   size_t min(size_t   a, size_t   b) { return a<b ? a:b; }
 #endif
+#if defined(__EMSCRIPTEN__)
+  __forceinline   long   min(long     a, long     b) { return a<b ? a:b; }
+#endif
 
   template<typename T> __forceinline T min(const T& a, const T& b, const T& c) { return min(min(a,b),c); }
   template<typename T> __forceinline T min(const T& a, const T& b, const T& c, const T& d) { return min(min(a,b),min(c,d)); }
@@ -189,9 +263,12 @@ namespace embree
   __forceinline  int64_t max(int64_t  a, int64_t  b) { return a<b ? b:a; }
   __forceinline    float max(float    a, float    b) { return a<b ? b:a; }
   __forceinline   double max(double   a, double   b) { return a<b ? b:a; }
-#if defined(__64BIT__)
+#if defined(__64BIT__) || defined(__EMSCRIPTEN__)
   __forceinline   size_t max(size_t   a, size_t   b) { return a<b ? b:a; }
 #endif
+#if defined(__EMSCRIPTEN__)
+  __forceinline   long   max(long     a, long     b) { return a<b ? b:a; }
+#endif
 
   template<typename T> __forceinline T max(const T& a, const T& b, const T& c) { return max(max(a,b),c); }
   template<typename T> __forceinline T max(const T& a, const T& b, const T& c, const T& d) { return max(max(a,b),max(c,d)); }
@@ -231,6 +308,15 @@ namespace embree
   __forceinline float msub  ( const float a, const float b, const float c) { return _mm_cvtss_f32(_mm_fmsub_ss(_mm_set_ss(a),_mm_set_ss(b),_mm_set_ss(c))); }
   __forceinline float nmadd ( const float a, const float b, const float c) { return _mm_cvtss_f32(_mm_fnmadd_ss(_mm_set_ss(a),_mm_set_ss(b),_mm_set_ss(c))); }
   __forceinline float nmsub ( const float a, const float b, const float c) { return _mm_cvtss_f32(_mm_fnmsub_ss(_mm_set_ss(a),_mm_set_ss(b),_mm_set_ss(c))); }
+  
+#elif defined (__aarch64__) && defined(__clang__)
+#pragma clang fp contract(fast)
+__forceinline float madd  ( const float a, const float b, const float c) { return a*b + c; }
+__forceinline float msub  ( const float a, const float b, const float c) { return a*b - c; }
+__forceinline float nmadd ( const float a, const float b, const float c) { return c - a*b; }
+__forceinline float nmsub ( const float a, const float b, const float c) { return -(c + a*b); }
+#pragma clang fp contract(on)
+  
 #else
   __forceinline float madd  ( const float a, const float b, const float c) { return a*b+c; }
   __forceinline float msub  ( const float a, const float b, const float c) { return a*b-c; }
@@ -326,7 +412,7 @@ namespace embree
     return x | (y << 1) | (z << 2);
   }
 
-#if defined(__AVX2__)
+#if defined(__AVX2__) && !defined(__aarch64__)
 
   template<>
     __forceinline unsigned int bitInterleave(const unsigned int &xi, const unsigned int& yi, const unsigned int& zi)
diff --git a/thirdparty/embree/common/math/quaternion.h b/thirdparty/embree/common/math/quaternion.h
index 080800efcd..78efccda72 100644
--- a/thirdparty/embree/common/math/quaternion.h
+++ b/thirdparty/embree/common/math/quaternion.h
@@ -242,13 +242,17 @@ namespace embree
     T cosTheta = dot(q0, q1_);
     QuaternionT<T> q1 = select(cosTheta < 0.f, -q1_, q1_);
     cosTheta          = select(cosTheta < 0.f, -cosTheta, cosTheta);
-    if (unlikely(all(cosTheta > 0.9995f))) {
-      return normalize(lerp(q0, q1, t));
-    }
+
+    // spherical linear interpolation
     const T phi = t * fastapprox::acos(cosTheta);
     T sinPhi, cosPhi;
     fastapprox::sincos(phi, sinPhi, cosPhi);
     QuaternionT<T> qperp = sinPhi * normalize(msub(cosTheta, q0, q1));
-    return msub(cosPhi, q0, qperp);
+    QuaternionT<T> qslerp = msub(cosPhi, q0, qperp);
+
+    // regular linear interpolation as fallback
+    QuaternionT<T> qlerp = normalize(lerp(q0, q1, t));
+
+    return select(cosTheta > 0.9995f, qlerp, qslerp);
   }
 }
diff --git a/thirdparty/embree/common/math/transcendental.h b/thirdparty/embree/common/math/transcendental.h
index fd16c26e81..daf9dd96d2 100644
--- a/thirdparty/embree/common/math/transcendental.h
+++ b/thirdparty/embree/common/math/transcendental.h
@@ -27,7 +27,7 @@ __forceinline T sin(const T &v)
   // Reduced range version of x
   auto x = v - kReal * piOverTwoVec;
   auto kMod4 = k & 3;
-  auto sinUseCos = (kMod4 == 1 | kMod4 == 3);
+  auto sinUseCos = (kMod4 == 1) | (kMod4 == 3);
   auto flipSign = (kMod4 > 1);
 
   // These coefficients are from sollya with fpminimax(sin(x)/x, [|0, 2,
@@ -76,8 +76,8 @@ __forceinline T cos(const T &v)
   auto x = v - kReal * piOverTwoVec;
 
   auto kMod4 = k & 3;
-  auto cosUseCos = (kMod4 == 0 | kMod4 == 2);
-  auto flipSign = (kMod4 == 1 | kMod4 == 2);
+  auto cosUseCos = (kMod4 == 0) | (kMod4 == 2);
+  auto flipSign = (kMod4 == 1) | (kMod4 == 2);
 
   const float sinC2  = -0.16666667163372039794921875;
   const float sinC4  = +8.333347737789154052734375e-3;
diff --git a/thirdparty/embree/common/math/vec2.h b/thirdparty/embree/common/math/vec2.h
index d62aef51f3..f6d98ffa0d 100644
--- a/thirdparty/embree/common/math/vec2.h
+++ b/thirdparty/embree/common/math/vec2.h
@@ -144,7 +144,7 @@ namespace embree
   }
   
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
   template<typename T> __forceinline T       dot      ( const Vec2<T>& a, const Vec2<T>& b ) { return madd(a.x,b.x,a.y*b.y); }
@@ -205,11 +205,11 @@ namespace embree
 
 #include "vec2fa.h"
 
-#if defined __SSE__
+#if defined(__SSE__) || defined(__ARM_NEON)
 #include "../simd/sse.h"
 #endif
 
-#if defined __AVX__
+#if defined(__AVX__)
 #include "../simd/avx.h"
 #endif
 
@@ -221,7 +221,7 @@ namespace embree
 {
   template<> __forceinline Vec2<float>::Vec2(const Vec2fa& a) : x(a.x), y(a.y) {}
 
-#if defined(__SSE__)
+#if defined(__SSE__) || defined(__ARM_NEON)
   template<> __forceinline Vec2<vfloat4>::Vec2(const Vec2fa& a) : x(a.x), y(a.y) {}
 #endif
 
diff --git a/thirdparty/embree/common/math/vec2fa.h b/thirdparty/embree/common/math/vec2fa.h
index a51fb68fd0..4f222894c2 100644
--- a/thirdparty/embree/common/math/vec2fa.h
+++ b/thirdparty/embree/common/math/vec2fa.h
@@ -97,6 +97,12 @@ namespace embree
 
   __forceinline Vec2fa rcp  ( const Vec2fa& a )
   {
+#if defined(__aarch64__)
+        __m128 reciprocal = _mm_rcp_ps(a.m128);
+        reciprocal = vmulq_f32(vrecpsq_f32(a.m128, reciprocal), reciprocal);
+        reciprocal = vmulq_f32(vrecpsq_f32(a.m128, reciprocal), reciprocal);
+        return (const Vec2fa)reciprocal;
+#else
 #if defined(__AVX512VL__)
     const Vec2fa r = _mm_rcp14_ps(a.m128);
 #else
@@ -104,13 +110,15 @@ namespace embree
 #endif
 
 #if defined(__AVX2__)
-    const Vec2fa res = _mm_mul_ps(r,_mm_fnmadd_ps(r, a, vfloat4(2.0f)));
+    const Vec2fa h_n = _mm_fnmadd_ps(a, r, vfloat4(1.0));  // First, compute 1 - a * r (which will be very close to 0)
+    const Vec2fa res = _mm_fmadd_ps(r, h_n, r);            // Then compute r + r * h_n
 #else
-    const Vec2fa res = _mm_mul_ps(r,_mm_sub_ps(vfloat4(2.0f), _mm_mul_ps(r, a)));
-    //return _mm_sub_ps(_mm_add_ps(r, r), _mm_mul_ps(_mm_mul_ps(r, r), a));
+    const Vec2fa h_n = _mm_sub_ps(vfloat4(1.0f), _mm_mul_ps(a, r));  // First, compute 1 - a * r (which will be very close to 0)
+    const Vec2fa res = _mm_add_ps(r,_mm_mul_ps(r, h_n));             // Then compute r + r * h_n  
 #endif
 
     return res;
+#endif  //defined(__aarch64__)
   }
 
   __forceinline Vec2fa sqrt ( const Vec2fa& a ) { return _mm_sqrt_ps(a.m128); }
@@ -118,12 +126,21 @@ namespace embree
 
   __forceinline Vec2fa rsqrt( const Vec2fa& a )
   {
+#if defined(__aarch64__)
+        __m128 r = _mm_rsqrt_ps(a.m128);
+        r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+        r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+        return r;
+#else
+
 #if defined(__AVX512VL__)
     __m128 r = _mm_rsqrt14_ps(a.m128);
 #else
     __m128 r = _mm_rsqrt_ps(a.m128);
 #endif
     return _mm_add_ps(_mm_mul_ps(_mm_set1_ps(1.5f),r), _mm_mul_ps(_mm_mul_ps(_mm_mul_ps(a, _mm_set1_ps(-0.5f)), r), _mm_mul_ps(r, r)));
+
+#endif
   }
 
   __forceinline Vec2fa zero_fix(const Vec2fa& a) {
@@ -156,7 +173,7 @@ namespace embree
   __forceinline Vec2fa min( const Vec2fa& a, const Vec2fa& b ) { return _mm_min_ps(a.m128,b.m128); }
   __forceinline Vec2fa max( const Vec2fa& a, const Vec2fa& b ) { return _mm_max_ps(a.m128,b.m128); }
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
     __forceinline Vec2fa mini(const Vec2fa& a, const Vec2fa& b) {
       const vint4 ai = _mm_castps_si128(a);
       const vint4 bi = _mm_castps_si128(b);
@@ -165,7 +182,7 @@ namespace embree
     }
 #endif
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
     __forceinline Vec2fa maxi(const Vec2fa& a, const Vec2fa& b) {
       const vint4 ai = _mm_castps_si128(a);
       const vint4 bi = _mm_castps_si128(b);
@@ -227,7 +244,7 @@ namespace embree
   __forceinline bool operator !=( const Vec2fa& a, const Vec2fa& b ) { return (_mm_movemask_ps(_mm_cmpneq_ps(a.m128, b.m128)) & 3) != 0; }
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
 #if defined(__SSE4_1__)
diff --git a/thirdparty/embree/common/math/vec3.h b/thirdparty/embree/common/math/vec3.h
index ce94eff327..254f6c4011 100644
--- a/thirdparty/embree/common/math/vec3.h
+++ b/thirdparty/embree/common/math/vec3.h
@@ -197,7 +197,7 @@ namespace embree
   template<typename T> __forceinline Vec3<bool> ge_mask( const Vec3<T>& a, const Vec3<T>& b ) { return Vec3<bool>(a.x>=b.x,a.y>=b.y,a.z>=b.z); }
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
   template<typename T> __forceinline T       sqr      ( const Vec3<T>& a )                   { return dot(a,a); }
@@ -207,7 +207,6 @@ namespace embree
   template<typename T> __forceinline Vec3<T> normalize( const Vec3<T>& a )                   { return a*rsqrt(sqr(a)); }
   template<typename T> __forceinline T       distance ( const Vec3<T>& a, const Vec3<T>& b ) { return length(a-b); }
   template<typename T> __forceinline Vec3<T> cross    ( const Vec3<T>& a, const Vec3<T>& b ) { return Vec3<T>(msub(a.y,b.z,a.z*b.y), msub(a.z,b.x,a.x*b.z), msub(a.x,b.y,a.y*b.x)); }
-
   template<typename T> __forceinline Vec3<T> stable_triangle_normal( const Vec3<T>& a, const Vec3<T>& b, const Vec3<T>& c )
   {
     const T ab_x = a.z*b.y, ab_y = a.x*b.z, ab_z = a.y*b.x;
@@ -266,11 +265,11 @@ namespace embree
 /// SSE / AVX / MIC specializations
 ////////////////////////////////////////////////////////////////////////////////
 
-#if defined __SSE__
+#if defined(__SSE__) || defined(__ARM_NEON)
 #include "../simd/sse.h"
 #endif
 
-#if defined __AVX__
+#if defined(__AVX__)
 #include "../simd/avx.h"
 #endif
 
@@ -291,14 +290,14 @@ namespace embree
   template<> __forceinline Vec3<vfloat4>::Vec3(const Vec3fa& a) {
     x = a.x; y = a.y; z = a.z;
   }
-#elif defined(__SSE__)
+#elif defined(__SSE__) || defined(__ARM_NEON)
   template<>
   __forceinline Vec3<vfloat4>::Vec3(const Vec3fa& a) {
     const vfloat4 v = vfloat4(a.m128); x = shuffle<0,0,0,0>(v); y = shuffle<1,1,1,1>(v); z = shuffle<2,2,2,2>(v);
   }
 #endif
 
-#if defined(__SSE__)
+#if defined(__SSE__) || defined(__ARM_NEON)
   template<>
   __forceinline Vec3<vfloat4> broadcast<vfloat4,vfloat4>(const Vec3<vfloat4>& a, const size_t k) {
     return Vec3<vfloat4>(vfloat4::broadcast(&a.x[k]), vfloat4::broadcast(&a.y[k]), vfloat4::broadcast(&a.z[k]));
diff --git a/thirdparty/embree/common/math/vec3fa.h b/thirdparty/embree/common/math/vec3fa.h
index 586039741d..8564cf6d10 100644
--- a/thirdparty/embree/common/math/vec3fa.h
+++ b/thirdparty/embree/common/math/vec3fa.h
@@ -55,7 +55,13 @@ namespace embree
     ////////////////////////////////////////////////////////////////////////////////
 
     static __forceinline Vec3fa load( const void* const a ) {
+#if defined(__aarch64__)
+        __m128 t = _mm_load_ps((float*)a);
+        t[3] = 0.0f;
+        return Vec3fa(t);
+#else
       return Vec3fa(_mm_and_ps(_mm_load_ps((float*)a),_mm_castsi128_ps(_mm_set_epi32(0, -1, -1, -1))));
+#endif
     }
 
     static __forceinline Vec3fa loadu( const void* const a ) {
@@ -89,12 +95,20 @@ namespace embree
 
   __forceinline Vec3fa operator +( const Vec3fa& a ) { return a; }
   __forceinline Vec3fa operator -( const Vec3fa& a ) {
+#if defined(__aarch64__)
+    return vnegq_f32(a.m128);
+#else
     const __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
     return _mm_xor_ps(a.m128, mask);
+#endif
   }
   __forceinline Vec3fa abs  ( const Vec3fa& a ) {
+#if defined(__aarch64__)
+    return _mm_abs_ps(a.m128);
+#else
     const __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff));
     return _mm_and_ps(a.m128, mask);
+#endif
   }
   __forceinline Vec3fa sign ( const Vec3fa& a ) {
     return blendv_ps(Vec3fa(one).m128, (-Vec3fa(one)).m128, _mm_cmplt_ps (a.m128,Vec3fa(zero).m128));
@@ -102,6 +116,10 @@ namespace embree
 
   __forceinline Vec3fa rcp  ( const Vec3fa& a )
   {
+#if defined(__aarch64__)
+  return vdivq_f32(vdupq_n_f32(1.0f),a.m128);
+#else
+
 #if defined(__AVX512VL__)
     const Vec3fa r = _mm_rcp14_ps(a.m128);
 #else
@@ -109,13 +127,15 @@ namespace embree
 #endif
 
 #if defined(__AVX2__)
-    const Vec3fa res = _mm_mul_ps(r.m128,_mm_fnmadd_ps(r.m128, a.m128, vfloat4(2.0f)));
+    const Vec3fa h_n = _mm_fnmadd_ps(a.m128, r.m128, vfloat4(1.0));  // First, compute 1 - a * r (which will be very close to 0)
+    const Vec3fa res = _mm_fmadd_ps(r.m128, h_n.m128, r.m128);       // Then compute r + r * h_n
 #else
-    const Vec3fa res = _mm_mul_ps(r.m128,_mm_sub_ps(vfloat4(2.0f), _mm_mul_ps(r.m128, a.m128)));
-    //return _mm_sub_ps(_mm_add_ps(r, r), _mm_mul_ps(_mm_mul_ps(r, r), a));
+    const Vec3fa h_n = _mm_sub_ps(vfloat4(1.0f), _mm_mul_ps(a.m128, r.m128));  // First, compute 1 - a * r (which will be very close to 0)
+    const Vec3fa res = _mm_add_ps(r.m128,_mm_mul_ps(r.m128, h_n.m128));        // Then compute r + r * h_n  
 #endif
 
     return res;
+#endif  //defined(__aarch64__)
   }
 
   __forceinline Vec3fa sqrt ( const Vec3fa& a ) { return _mm_sqrt_ps(a.m128); }
@@ -123,12 +143,20 @@ namespace embree
 
   __forceinline Vec3fa rsqrt( const Vec3fa& a )
   {
+#if defined(__aarch64__)
+        __m128 r = _mm_rsqrt_ps(a.m128);
+        r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+        r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(a.m128, r), r));
+        return r;
+#else
+
 #if defined(__AVX512VL__)
     __m128 r = _mm_rsqrt14_ps(a.m128);
 #else
     __m128 r = _mm_rsqrt_ps(a.m128);
 #endif
     return _mm_add_ps(_mm_mul_ps(_mm_set1_ps(1.5f),r), _mm_mul_ps(_mm_mul_ps(_mm_mul_ps(a.m128, _mm_set1_ps(-0.5f)), r), _mm_mul_ps(r, r)));
+#endif
   }
 
   __forceinline Vec3fa zero_fix(const Vec3fa& a) {
@@ -161,7 +189,7 @@ namespace embree
   __forceinline Vec3fa min( const Vec3fa& a, const Vec3fa& b ) { return _mm_min_ps(a.m128,b.m128); }
   __forceinline Vec3fa max( const Vec3fa& a, const Vec3fa& b ) { return _mm_max_ps(a.m128,b.m128); }
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
     __forceinline Vec3fa mini(const Vec3fa& a, const Vec3fa& b) {
       const vint4 ai = _mm_castps_si128(a.m128);
       const vint4 bi = _mm_castps_si128(b.m128);
@@ -170,7 +198,7 @@ namespace embree
     }
 #endif
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
     __forceinline Vec3fa maxi(const Vec3fa& a, const Vec3fa& b) {
       const vint4 ai = _mm_castps_si128(a.m128);
       const vint4 bi = _mm_castps_si128(b.m128);
@@ -187,16 +215,16 @@ namespace embree
   /// Ternary Operators
   ////////////////////////////////////////////////////////////////////////////////
 
-#if defined(__AVX2__)
+#if defined(__AVX2__) || defined(__ARM_NEON)
   __forceinline Vec3fa madd  ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return _mm_fmadd_ps(a.m128,b.m128,c.m128); }
   __forceinline Vec3fa msub  ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return _mm_fmsub_ps(a.m128,b.m128,c.m128); }
   __forceinline Vec3fa nmadd ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return _mm_fnmadd_ps(a.m128,b.m128,c.m128); }
   __forceinline Vec3fa nmsub ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return _mm_fnmsub_ps(a.m128,b.m128,c.m128); }
 #else
   __forceinline Vec3fa madd  ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return a*b+c; }
-  __forceinline Vec3fa msub  ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return a*b-c; }
   __forceinline Vec3fa nmadd ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return -a*b+c;}
   __forceinline Vec3fa nmsub ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return -a*b-c; }
+  __forceinline Vec3fa msub  ( const Vec3fa& a, const Vec3fa& b, const Vec3fa& c) { return a*b-c; }
 #endif
 
   __forceinline Vec3fa madd  ( const float a, const Vec3fa& b, const Vec3fa& c) { return madd(Vec3fa(a),b,c); }
@@ -218,8 +246,26 @@ namespace embree
   ////////////////////////////////////////////////////////////////////////////////
   /// Reductions
   ////////////////////////////////////////////////////////////////////////////////
+#if defined(__aarch64__)
+  __forceinline float reduce_add(const Vec3fa& v) {
+    float32x4_t t = v.m128;
+    t[3] = 0.0f;
+    return vaddvq_f32(t);
+  }
 
-  __forceinline float reduce_add(const Vec3fa& v) { 
+  __forceinline float reduce_mul(const Vec3fa& v) { return v.x*v.y*v.z; }
+  __forceinline float reduce_min(const Vec3fa& v) {
+    float32x4_t t = v.m128;
+      t[3] = t[2];
+    return vminvq_f32(t);
+  }
+  __forceinline float reduce_max(const Vec3fa& v) {
+    float32x4_t t = v.m128;
+      t[3] = t[2];
+    return vmaxvq_f32(t);
+  }
+#else
+  __forceinline float reduce_add(const Vec3fa& v) {
     const vfloat4 a(v.m128);
     const vfloat4 b = shuffle<1>(a);
     const vfloat4 c = shuffle<2>(a);
@@ -229,6 +275,7 @@ namespace embree
   __forceinline float reduce_mul(const Vec3fa& v) { return v.x*v.y*v.z; }
   __forceinline float reduce_min(const Vec3fa& v) { return min(v.x,v.y,v.z); }
   __forceinline float reduce_max(const Vec3fa& v) { return max(v.x,v.y,v.z); }
+#endif
 
   ////////////////////////////////////////////////////////////////////////////////
   /// Comparison Operators
@@ -241,8 +288,13 @@ namespace embree
   __forceinline Vec3ba neq_mask(const Vec3fa& a, const Vec3fa& b ) { return _mm_cmpneq_ps(a.m128, b.m128); }
   __forceinline Vec3ba lt_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmplt_ps (a.m128, b.m128); }
   __forceinline Vec3ba le_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmple_ps (a.m128, b.m128); }
-  __forceinline Vec3ba gt_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmpnle_ps(a.m128, b.m128); }
-  __forceinline Vec3ba ge_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmpnlt_ps(a.m128, b.m128); }
+ #if defined(__aarch64__)
+  __forceinline Vec3ba gt_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmpgt_ps (a.m128, b.m128); }
+  __forceinline Vec3ba ge_mask( const Vec3fa& a, const Vec3fa& b ) { return _mm_cmpge_ps (a.m128, b.m128); }
+#else
+  __forceinline Vec3ba gt_mask(const Vec3fa& a, const Vec3fa& b) { return _mm_cmpnle_ps(a.m128, b.m128); }
+  __forceinline Vec3ba ge_mask(const Vec3fa& a, const Vec3fa& b) { return _mm_cmpnlt_ps(a.m128, b.m128); }
+#endif
 
   __forceinline bool isvalid ( const Vec3fa& v ) {
     return all(gt_mask(v,Vec3fa(-FLT_LARGE)) & lt_mask(v,Vec3fa(+FLT_LARGE)));
@@ -261,7 +313,7 @@ namespace embree
   }
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
 #if defined(__SSE4_1__)
@@ -335,7 +387,11 @@ namespace embree
   /// Rounding Functions
   ////////////////////////////////////////////////////////////////////////////////
 
-#if defined (__SSE4_1__)
+#if defined(__aarch64__)
+  __forceinline Vec3fa floor(const Vec3fa& a) { return vrndmq_f32(a.m128); }
+  __forceinline Vec3fa ceil (const Vec3fa& a) { return vrndpq_f32(a.m128); }
+  __forceinline Vec3fa trunc(const Vec3fa& a) { return vrndq_f32(a.m128); }
+#elif defined (__SSE4_1__)
   __forceinline Vec3fa trunc( const Vec3fa& a ) { return _mm_round_ps(a.m128, _MM_FROUND_TO_NEAREST_INT); }
   __forceinline Vec3fa floor( const Vec3fa& a ) { return _mm_round_ps(a.m128, _MM_FROUND_TO_NEG_INF    ); }
   __forceinline Vec3fa ceil ( const Vec3fa& a ) { return _mm_round_ps(a.m128, _MM_FROUND_TO_POS_INF    ); }
@@ -393,8 +449,10 @@ namespace embree
 
     __forceinline Vec3fx( const Vec3fa& other, const int      a1) { m128 = other.m128; a = a1; }
     __forceinline Vec3fx( const Vec3fa& other, const unsigned a1) { m128 = other.m128; u = a1; }
-    __forceinline Vec3fx( const Vec3fa& other, const float    w1) {      
-#if defined (__SSE4_1__)
+    __forceinline Vec3fx( const Vec3fa& other, const float    w1) {
+#if defined (__aarch64__)
+      m128 = other.m128; m128[3] = w1;
+#elif defined (__SSE4_1__)
       m128 = _mm_insert_ps(other.m128, _mm_set_ss(w1),3 << 4);
 #else
       const vint4 mask(-1,-1,-1,0);
@@ -526,7 +584,7 @@ namespace embree
   __forceinline Vec3fx min( const Vec3fx& a, const Vec3fx& b ) { return _mm_min_ps(a.m128,b.m128); }
   __forceinline Vec3fx max( const Vec3fx& a, const Vec3fx& b ) { return _mm_max_ps(a.m128,b.m128); }
 
-#if defined(__SSE4_1__)
+#if defined(__SSE4_1__) || defined(__aarch64__)
     __forceinline Vec3fx mini(const Vec3fx& a, const Vec3fx& b) {
       const vint4 ai = _mm_castps_si128(a.m128);
       const vint4 bi = _mm_castps_si128(b.m128);
@@ -535,7 +593,7 @@ namespace embree
     }
 #endif
 
-#if defined(__SSE4_1__)
+#if defined(__SSE4_1__) || defined(__aarch64__)
     __forceinline Vec3fx maxi(const Vec3fx& a, const Vec3fx& b) {
       const vint4 ai = _mm_castps_si128(a.m128);
       const vint4 bi = _mm_castps_si128(b.m128);
@@ -626,7 +684,7 @@ namespace embree
   }
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
 #if defined(__SSE4_1__)
diff --git a/thirdparty/embree/common/math/vec3ia.h b/thirdparty/embree/common/math/vec3ia.h
index 694804c40d..d4cc3125cd 100644
--- a/thirdparty/embree/common/math/vec3ia.h
+++ b/thirdparty/embree/common/math/vec3ia.h
@@ -65,7 +65,9 @@ namespace embree
 
   __forceinline Vec3ia operator +( const Vec3ia& a ) { return a; }
   __forceinline Vec3ia operator -( const Vec3ia& a ) { return _mm_sub_epi32(_mm_setzero_si128(), a.m128); }
-#if defined(__SSSE3__)
+#if (defined(__aarch64__))
+  __forceinline Vec3ia abs       ( const Vec3ia& a ) { return vabsq_s32(a.m128); }
+#elif defined(__SSSE3__)
   __forceinline Vec3ia abs       ( const Vec3ia& a ) { return _mm_abs_epi32(a.m128); }
 #endif
 
@@ -81,7 +83,7 @@ namespace embree
   __forceinline Vec3ia operator -( const Vec3ia& a, const int     b ) { return a-Vec3ia(b); }
   __forceinline Vec3ia operator -( const int     a, const Vec3ia& b ) { return Vec3ia(a)-b; }
 
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
   __forceinline Vec3ia operator *( const Vec3ia& a, const Vec3ia& b ) { return _mm_mullo_epi32(a.m128, b.m128); }
   __forceinline Vec3ia operator *( const Vec3ia& a, const int     b ) { return a * Vec3ia(b); }
   __forceinline Vec3ia operator *( const int     a, const Vec3ia& b ) { return Vec3ia(a) * b; }
@@ -116,7 +118,7 @@ namespace embree
   __forceinline Vec3ia& operator -=( Vec3ia& a, const Vec3ia& b ) { return a = a - b; }
   __forceinline Vec3ia& operator -=( Vec3ia& a, const int&   b ) { return a = a - b; }
   
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
   __forceinline Vec3ia& operator *=( Vec3ia& a, const Vec3ia& b ) { return a = a * b; }
   __forceinline Vec3ia& operator *=( Vec3ia& a, const int&    b ) { return a = a * b; }
 #endif
@@ -127,18 +129,38 @@ namespace embree
   __forceinline Vec3ia& operator |=( Vec3ia& a, const Vec3ia& b ) { return a = a | b; }
   __forceinline Vec3ia& operator |=( Vec3ia& a, const int&    b ) { return a = a | b; }
   
+#if !defined(__ARM_NEON)
   __forceinline Vec3ia& operator <<=( Vec3ia& a, const int& b ) { return a = a << b; }
   __forceinline Vec3ia& operator >>=( Vec3ia& a, const int& b ) { return a = a >> b; }
+#endif
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Reductions
+  /// Select
   ////////////////////////////////////////////////////////////////////////////////
 
+  __forceinline Vec3ia select( const Vec3ba& m, const Vec3ia& t, const Vec3ia& f ) {
+#if defined(__aarch64__) || defined(__SSE4_1__)
+    return _mm_castps_si128(_mm_blendv_ps(_mm_castsi128_ps(f), _mm_castsi128_ps(t), m));
+#else
+    return _mm_or_si128(_mm_and_si128(_mm_castps_si128(m), t), _mm_andnot_si128(_mm_castps_si128(m), f)); 
+#endif
+  }
+
+  ////////////////////////////////////////////////////////////////////////////////
+  /// Reductions
+  ////////////////////////////////////////////////////////////////////////////////
+#if defined(__aarch64__)
+  __forceinline int reduce_add(const Vec3ia& v) { return vaddvq_s32(select(Vec3ba(1,1,1),v,Vec3ia(0))); }
+  __forceinline int reduce_mul(const Vec3ia& v) { return v.x*v.y*v.z; }
+  __forceinline int reduce_min(const Vec3ia& v) { return vminvq_s32(select(Vec3ba(1,1,1),v,Vec3ia(0x7FFFFFFF))); }
+  __forceinline int reduce_max(const Vec3ia& v) { return vmaxvq_s32(select(Vec3ba(1,1,1),v,Vec3ia(0x80000000))); }
+#else
   __forceinline int reduce_add(const Vec3ia& v) { return v.x+v.y+v.z; }
   __forceinline int reduce_mul(const Vec3ia& v) { return v.x*v.y*v.z; }
   __forceinline int reduce_min(const Vec3ia& v) { return min(v.x,v.y,v.z); }
   __forceinline int reduce_max(const Vec3ia& v) { return max(v.x,v.y,v.z); }
-
+#endif
+  
   ////////////////////////////////////////////////////////////////////////////////
   /// Comparison Operators
   ////////////////////////////////////////////////////////////////////////////////
@@ -156,19 +178,7 @@ namespace embree
   __forceinline Vec3ba lt_mask( const Vec3ia& a, const Vec3ia& b ) { return _mm_castsi128_ps(_mm_cmplt_epi32 (a.m128, b.m128)); }
   __forceinline Vec3ba gt_mask( const Vec3ia& a, const Vec3ia& b ) { return _mm_castsi128_ps(_mm_cmpgt_epi32 (a.m128, b.m128)); }
 
-  ////////////////////////////////////////////////////////////////////////////////
-  /// Select
-  ////////////////////////////////////////////////////////////////////////////////
-
-  __forceinline Vec3ia select( const Vec3ba& m, const Vec3ia& t, const Vec3ia& f ) {
-#if defined(__SSE4_1__)
-    return _mm_castps_si128(_mm_blendv_ps(_mm_castsi128_ps(f), _mm_castsi128_ps(t), m));
-#else
-    return _mm_or_si128(_mm_and_si128(_mm_castps_si128(m), t), _mm_andnot_si128(_mm_castps_si128(m), f)); 
-#endif
-  }
-
-#if defined(__SSE4_1__)
+#if defined(__aarch64__) || defined(__SSE4_1__)
   __forceinline Vec3ia min( const Vec3ia& a, const Vec3ia& b ) { return _mm_min_epi32(a.m128,b.m128); }
   __forceinline Vec3ia max( const Vec3ia& a, const Vec3ia& b ) { return _mm_max_epi32(a.m128,b.m128); }
 #else
diff --git a/thirdparty/embree/common/math/vec4.h b/thirdparty/embree/common/math/vec4.h
index 0ed107928a..10c53f47b4 100644
--- a/thirdparty/embree/common/math/vec4.h
+++ b/thirdparty/embree/common/math/vec4.h
@@ -149,7 +149,7 @@ namespace embree
   }
 
   ////////////////////////////////////////////////////////////////////////////////
-  /// Euclidian Space Operators
+  /// Euclidean Space Operators
   ////////////////////////////////////////////////////////////////////////////////
 
   template<typename T> __forceinline T       dot      ( const Vec4<T>& a, const Vec4<T>& b ) { return madd(a.x,b.x,madd(a.y,b.y,madd(a.z,b.z,a.w*b.w))); }
@@ -205,7 +205,7 @@ namespace embree
 /// SSE / AVX / MIC specializations
 ////////////////////////////////////////////////////////////////////////////////
 
-#if defined __SSE__
+#if defined(__SSE__) || defined(__ARM_NEON)
 #include "../simd/sse.h"
 #endif
 
@@ -225,7 +225,7 @@ namespace embree
   template<> __forceinline Vec4<vfloat4>::Vec4( const Vec3fx& a ) {
     x = a.x; y = a.y; z = a.z; w = a.w;
   }
-#elif defined(__SSE__)
+#elif defined(__SSE__) || defined(__ARM_NEON)
   template<> __forceinline Vec4<vfloat4>::Vec4( const Vec3fx& a ) {
     const vfloat4 v = vfloat4(a.m128); x = shuffle<0,0,0,0>(v); y = shuffle<1,1,1,1>(v); z = shuffle<2,2,2,2>(v); w = shuffle<3,3,3,3>(v);
   }