| 1 | /* Copyright (C) 2013-2017 Povilas Kanapickas <povilas@radix.lt> |
| 2 | |
| 3 | Distributed under the Boost Software License, Version 1.0. |
| 4 | (See accompanying file LICENSE_1_0.txt or copy at |
| 5 | http://www.boost.org/LICENSE_1_0.txt) |
| 6 | */ |
| 7 | |
| 8 | #ifndef LIBSIMDPP_SIMDPP_CORE_TO_INT32_H |
| 9 | #define LIBSIMDPP_SIMDPP_CORE_TO_INT32_H |
| 10 | |
| 11 | #ifndef LIBSIMDPP_SIMD_H |
| 12 | #error "This file must be included through simd.h" |
| 13 | #endif |
| 14 | |
| 15 | #include <simdpp/types.h> |
| 16 | #include <simdpp/capabilities.h> |
| 17 | #include <simdpp/detail/insn/conv_extend_to_int32.h> |
| 18 | #include <simdpp/detail/insn/conv_shrink_to_int32.h> |
| 19 | #include <simdpp/detail/insn/conv_float_to_int32.h> |
| 20 | #include <simdpp/detail/not_implemented.h> |
| 21 | |
| 22 | namespace simdpp { |
| 23 | namespace SIMDPP_ARCH_NAMESPACE { |
| 24 | |
| 25 | /** Converts elements within a vector to 32-bit signed values. |
| 26 | |
| 27 | The conversion rules are as follows: |
| 28 | 16-bit and narrower signed integers are sign-extended to 32 bits. |
| 29 | 16-bit and narrower unsigned integers are zero-extended to 32 bits. |
| 30 | 64-bit integers are truncated. |
| 31 | floating-point numbers are converted to integer values and truncated. |
| 32 | If floating-point value can not be represented in 32-bit signed integer, |
| 33 | the behavior is different for different instruction sets. |
| 34 | |
| 35 | SSE specific: |
| 36 | If the value can not be represented by int32_t, @c 0x80000000 is returned |
| 37 | TODO: NaN handling |
| 38 | |
| 39 | NEON, ALTIVEC specific: |
| 40 | If the value can not be represented by int32_t, either @c 0x80000000 or @c |
| 41 | 0x7fffffff is returned depending on the sign of the operand (saturation |
| 42 | occurs). Conversion of NaNs results in @a 0. |
| 43 | |
| 44 | @code |
| 45 | r0 = (int32_t) a0 |
| 46 | ... |
| 47 | rN = (int32_t) aN |
| 48 | @endcode |
| 49 | */ |
| 50 | template<unsigned N, class E> SIMDPP_INL |
| 51 | int32<N,expr_empty> to_int32(const int8<N,E>& a) |
| 52 | { |
| 53 | return detail::insn::i_to_int32(a.eval()); |
| 54 | } |
| 55 | template<unsigned N, class E> SIMDPP_INL |
| 56 | int32<N,expr_empty> to_int32(const uint8<N,E>& a) |
| 57 | { |
| 58 | return (int32<N>) detail::insn::i_to_uint32(a.eval()); |
| 59 | } |
| 60 | template<unsigned N, class E> SIMDPP_INL |
| 61 | int32<N,expr_empty> to_int32(const int16<N,E>& a) |
| 62 | { |
| 63 | return detail::insn::i_to_int32(a.eval()); |
| 64 | } |
| 65 | template<unsigned N, class E> SIMDPP_INL |
| 66 | int32<N,expr_empty> to_int32(const uint16<N,E>& a) |
| 67 | { |
| 68 | return (int32<N>) detail::insn::i_to_uint32(a.eval()); |
| 69 | } |
| 70 | template<unsigned N, class E> SIMDPP_INL |
| 71 | int32<N,expr_empty> to_int32(const int32<N,E>& a) |
| 72 | { |
| 73 | return a; |
| 74 | } |
| 75 | template<unsigned N, class E> SIMDPP_INL |
| 76 | int32<N,expr_empty> to_int32(const uint32<N,E>& a) |
| 77 | { |
| 78 | return int32<N>(a.eval()); |
| 79 | } |
| 80 | template<unsigned N, class E> SIMDPP_INL |
| 81 | int32<N,expr_empty> to_int32(const int64<N,E>& a) |
| 82 | { |
| 83 | return detail::insn::i_to_uint32(uint64<N>(a.eval())); |
| 84 | } |
| 85 | template<unsigned N, class E> SIMDPP_INL |
| 86 | int32<N,expr_empty> to_int32(const uint64<N,E>& a) |
| 87 | { |
| 88 | return detail::insn::i_to_uint32(a.eval()); |
| 89 | } |
| 90 | template<unsigned N, class E> SIMDPP_INL |
| 91 | int32<N,expr_empty> to_int32(const float32<N,E>& a) |
| 92 | { |
| 93 | return detail::insn::i_to_int32(a.eval()); |
| 94 | } |
| 95 | template<unsigned N, class E> SIMDPP_INL |
| 96 | int32<N,expr_empty> to_int32(const float64<N,E>& a) |
| 97 | { |
| 98 | return detail::insn::i_to_int32(a.eval()); |
| 99 | } |
| 100 | |
| 101 | /** Converts elements within a vector to 32-bit unsigned values. |
| 102 | |
| 103 | The conversion rules are as follows: |
| 104 | 16-bit and narrower signed integers are sign-extended to 32 bits. |
| 105 | 16-bit and narrower unsigned integers are zero-extended to 32 bits. |
| 106 | 64-bit integers are truncated. |
| 107 | If floating-point value can not be represented in 32-bit unsigned integer, |
| 108 | the behavior is different for different instruction sets. |
| 109 | |
| 110 | SSE specific: |
| 111 | If the value can not be represented by uint32_t, @c 0x80000000 is returned |
| 112 | @todo NaN handling |
| 113 | |
| 114 | NEON VFP specific: |
| 115 | If the value can not be represented by uint32_t, either @c 0x80000000 or @c |
| 116 | 0x7fffffff is returned depending on the sign of the operand. Conversion of |
| 117 | NaNs results in @a 0. |
| 118 | |
| 119 | @code |
| 120 | r0 = (uint32_t) a0 |
| 121 | ... |
| 122 | rN = (uint32_t) aN |
| 123 | @endcode |
| 124 | */ |
| 125 | template<unsigned N, class E> SIMDPP_INL |
| 126 | uint32<N,expr_empty> to_uint32(const int8<N,E>& a) |
| 127 | { |
| 128 | return (uint32<N>) detail::insn::i_to_int32(a.eval()); |
| 129 | } |
| 130 | template<unsigned N, class E> SIMDPP_INL |
| 131 | uint32<N,expr_empty> to_uint32(const uint8<N,E>& a) |
| 132 | { |
| 133 | return detail::insn::i_to_uint32(a.eval()); |
| 134 | } |
| 135 | template<unsigned N, class E> SIMDPP_INL |
| 136 | uint32<N,expr_empty> to_uint32(const int16<N,E>& a) |
| 137 | { |
| 138 | return (uint32<N>) detail::insn::i_to_int32(a.eval()); |
| 139 | } |
| 140 | template<unsigned N, class E> SIMDPP_INL |
| 141 | uint32<N,expr_empty> to_uint32(const uint16<N,E>& a) |
| 142 | { |
| 143 | return detail::insn::i_to_uint32(a.eval()); |
| 144 | } |
| 145 | template<unsigned N, class E> SIMDPP_INL |
| 146 | uint32<N,expr_empty> to_uint32(const int32<N,E>& a) |
| 147 | { |
| 148 | return a.eval(); |
| 149 | } |
| 150 | template<unsigned N, class E> SIMDPP_INL |
| 151 | uint32<N,expr_empty> to_uint32(const uint32<N,E>& a) |
| 152 | { |
| 153 | return a; |
| 154 | } |
| 155 | template<unsigned N, class E> SIMDPP_INL |
| 156 | uint32<N,expr_empty> to_uint32(const int64<N,E>& a) |
| 157 | { |
| 158 | return detail::insn::i_to_uint32(uint64<N>(a.eval())); |
| 159 | } |
| 160 | template<unsigned N, class E> SIMDPP_INL |
| 161 | uint32<N,expr_empty> to_uint32(const uint64<N,E>& a) |
| 162 | { |
| 163 | return detail::insn::i_to_uint32(a.eval()); |
| 164 | } |
| 165 | template<unsigned N, class E> SIMDPP_INL |
| 166 | uint32<N,expr_empty> to_uint32(const float32<N,E>& a) |
| 167 | { |
| 168 | return detail::insn::i_to_uint32(a.eval()); |
| 169 | } |
| 170 | template<unsigned N, class E> SIMDPP_INL |
| 171 | uint32<N,expr_empty> to_uint32(const float64<N,E>& a) |
| 172 | { |
| 173 | #if SIMDPP_HAS_FLOAT64_TO_UINT32_CONVERSION |
| 174 | return detail::insn::i_to_uint32(a.eval()); |
| 175 | #else |
| 176 | return SIMDPP_NOT_IMPLEMENTED_TEMPLATE1(E, a); |
| 177 | #endif |
| 178 | } |
| 179 | |
| 180 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 181 | } // namespace simdpp |
| 182 | |
| 183 | #endif |
| 184 | |
| 185 | |