| 1 | // Copyright 2009-2021 Intel Corporation |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| 4 | #pragma once |
| 5 | |
| 6 | #include "../sys/platform.h" |
| 7 | |
| 8 | namespace embree |
| 9 | { |
| 10 | /* Varying numeric types */ |
| 11 | template<int N> |
| 12 | struct vfloat_impl |
| 13 | { |
| 14 | union { float f[N]; int i[N]; }; |
| 15 | __forceinline const float& operator [](size_t index) const { assert(index < N); return f[index]; } |
| 16 | __forceinline float& operator [](size_t index) { assert(index < N); return f[index]; } |
| 17 | }; |
| 18 | |
| 19 | template<int N> |
| 20 | struct vdouble_impl |
| 21 | { |
| 22 | union { double f[N]; long long i[N]; }; |
| 23 | __forceinline const double& operator [](size_t index) const { assert(index < N); return f[index]; } |
| 24 | __forceinline double& operator [](size_t index) { assert(index < N); return f[index]; } |
| 25 | }; |
| 26 | |
| 27 | template<int N> |
| 28 | struct vint_impl |
| 29 | { |
| 30 | int i[N]; |
| 31 | __forceinline const int& operator [](size_t index) const { assert(index < N); return i[index]; } |
| 32 | __forceinline int& operator [](size_t index) { assert(index < N); return i[index]; } |
| 33 | }; |
| 34 | |
| 35 | template<int N> |
| 36 | struct vuint_impl |
| 37 | { |
| 38 | unsigned int i[N]; |
| 39 | __forceinline const unsigned int& operator [](size_t index) const { assert(index < N); return i[index]; } |
| 40 | __forceinline unsigned int& operator [](size_t index) { assert(index < N); return i[index]; } |
| 41 | }; |
| 42 | |
| 43 | template<int N> |
| 44 | struct vllong_impl |
| 45 | { |
| 46 | long long i[N]; |
| 47 | __forceinline const long long& operator [](size_t index) const { assert(index < N); return i[index]; } |
| 48 | __forceinline long long& operator [](size_t index) { assert(index < N); return i[index]; } |
| 49 | }; |
| 50 | |
| 51 | /* Varying bool types */ |
| 52 | template<int N> struct vboolf_impl { int i[N]; }; // for float/int |
| 53 | template<int N> struct vboold_impl { long long i[N]; }; // for double/long long |
| 54 | |
| 55 | /* Varying size constants */ |
| 56 | #if defined(__AVX512VL__) // SKX |
| 57 | const int VSIZEX = 8; // default size |
| 58 | const int VSIZEL = 16; // large size |
| 59 | #elif defined(__AVX__) |
| 60 | const int VSIZEX = 8; |
| 61 | const int VSIZEL = 8; |
| 62 | #else |
| 63 | const int VSIZEX = 4; |
| 64 | const int VSIZEL = 4; |
| 65 | #endif |
| 66 | |
| 67 | template<int N> |
| 68 | struct vtypes { |
| 69 | using vbool = vboolf_impl<N>; |
| 70 | using vboolf = vboolf_impl<N>; |
| 71 | using vboold = vboold_impl<N>; |
| 72 | using vint = vint_impl<N>; |
| 73 | using vuint = vuint_impl<N>; |
| 74 | using vllong = vllong_impl<N>; |
| 75 | using vfloat = vfloat_impl<N>; |
| 76 | using vdouble = vdouble_impl<N>; |
| 77 | }; |
| 78 | |
| 79 | template<> |
| 80 | struct vtypes<1> { |
| 81 | using vbool = bool; |
| 82 | using vboolf = bool; |
| 83 | using vboold = bool; |
| 84 | using vint = int; |
| 85 | using vuint = unsigned int; |
| 86 | using vllong = long long; |
| 87 | using vfloat = float; |
| 88 | using vdouble = double; |
| 89 | }; |
| 90 | |
| 91 | /* Aliases to default types */ |
| 92 | template<int N> using vbool = typename vtypes<N>::vbool; |
| 93 | template<int N> using vboolf = typename vtypes<N>::vboolf; |
| 94 | template<int N> using vboold = typename vtypes<N>::vboold; |
| 95 | template<int N> using vint = typename vtypes<N>::vint; |
| 96 | template<int N> using vuint = typename vtypes<N>::vuint; |
| 97 | template<int N> using vllong = typename vtypes<N>::vllong; |
| 98 | template<int N> using vreal = typename vtypes<N>::vfloat; |
| 99 | template<int N> using vfloat = typename vtypes<N>::vfloat; |
| 100 | template<int N> using vdouble = typename vtypes<N>::vdouble; |
| 101 | |
| 102 | /* 4-wide shortcuts */ |
| 103 | typedef vfloat<4> vfloat4; |
| 104 | typedef vdouble<4> vdouble4; |
| 105 | typedef vreal<4> vreal4; |
| 106 | typedef vint<4> vint4; |
| 107 | typedef vuint<4> vuint4; |
| 108 | typedef vllong<4> vllong4; |
| 109 | typedef vbool<4> vbool4; |
| 110 | typedef vboolf<4> vboolf4; |
| 111 | typedef vboold<4> vboold4; |
| 112 | |
| 113 | /* 8-wide shortcuts */ |
| 114 | typedef vfloat<8> vfloat8; |
| 115 | typedef vdouble<8> vdouble8; |
| 116 | typedef vreal<8> vreal8; |
| 117 | typedef vint<8> vint8; |
| 118 | typedef vuint<8> vuint8; |
| 119 | typedef vllong<8> vllong8; |
| 120 | typedef vbool<8> vbool8; |
| 121 | typedef vboolf<8> vboolf8; |
| 122 | typedef vboold<8> vboold8; |
| 123 | |
| 124 | /* 16-wide shortcuts */ |
| 125 | typedef vfloat<16> vfloat16; |
| 126 | typedef vdouble<16> vdouble16; |
| 127 | typedef vreal<16> vreal16; |
| 128 | typedef vint<16> vint16; |
| 129 | typedef vuint<16> vuint16; |
| 130 | typedef vllong<16> vllong16; |
| 131 | typedef vbool<16> vbool16; |
| 132 | typedef vboolf<16> vboolf16; |
| 133 | typedef vboold<16> vboold16; |
| 134 | |
| 135 | /* Default shortcuts */ |
| 136 | typedef vfloat<VSIZEX> vfloatx; |
| 137 | typedef vdouble<VSIZEX> vdoublex; |
| 138 | typedef vreal<VSIZEX> vrealx; |
| 139 | typedef vint<VSIZEX> vintx; |
| 140 | typedef vuint<VSIZEX> vuintx; |
| 141 | typedef vllong<VSIZEX> vllongx; |
| 142 | typedef vbool<VSIZEX> vboolx; |
| 143 | typedef vboolf<VSIZEX> vboolfx; |
| 144 | typedef vboold<VSIZEX> vbooldx; |
| 145 | } |
| 146 | |