diff options
Diffstat (limited to 'thirdparty/embree/common/math/vec3ia.h')
| -rw-r--r-- | thirdparty/embree/common/math/vec3ia.h | 46 |
1 files changed, 28 insertions, 18 deletions
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 |