Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 1 | /*===---- f16cintrin.h - F16C intrinsics -----------------------------------=== |
| 2 | * |
| 3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 4 | * of this software and associated documentation files (the "Software"), to deal |
| 5 | * in the Software without restriction, including without limitation the rights |
| 6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 7 | * copies of the Software, and to permit persons to whom the Software is |
| 8 | * furnished to do so, subject to the following conditions: |
| 9 | * |
| 10 | * The above copyright notice and this permission notice shall be included in |
| 11 | * all copies or substantial portions of the Software. |
| 12 | * |
| 13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 19 | * THE SOFTWARE. |
| 20 | * |
| 21 | *===-----------------------------------------------------------------------=== |
| 22 | */ |
| 23 | |
| 24 | #if !defined __X86INTRIN_H && !defined __EMMINTRIN_H && !defined __IMMINTRIN_H |
| 25 | #error "Never use <f16cintrin.h> directly; include <emmintrin.h> instead." |
| 26 | #endif |
| 27 | |
| 28 | #ifndef __F16CINTRIN_H |
| 29 | #define __F16CINTRIN_H |
| 30 | |
| 31 | /* Define the default attributes for the functions in this file. */ |
| 32 | #define __DEFAULT_FN_ATTRS \ |
| 33 | __attribute__((__always_inline__, __nodebug__, __target__("f16c"))) |
| 34 | |
| 35 | /// \brief Converts a 16-bit half-precision float value into a 32-bit float |
| 36 | /// value. |
| 37 | /// |
| 38 | /// \headerfile <x86intrin.h> |
| 39 | /// |
| 40 | /// This intrinsic corresponds to the \c VCVTPH2PS instruction. |
| 41 | /// |
| 42 | /// \param a |
| 43 | /// A 16-bit half-precision float value. |
| 44 | /// \returns The converted 32-bit float value. |
| 45 | static __inline float __DEFAULT_FN_ATTRS |
| 46 | _cvtsh_ss(unsigned short __a) |
| 47 | { |
| 48 | __v8hi v = {(short)__a, 0, 0, 0, 0, 0, 0, 0}; |
| 49 | __v4sf r = __builtin_ia32_vcvtph2ps(v); |
| 50 | return r[0]; |
| 51 | } |
| 52 | |
| 53 | /// \brief Converts a 32-bit single-precision float value to a 16-bit |
| 54 | /// half-precision float value. |
| 55 | /// |
| 56 | /// \headerfile <x86intrin.h> |
| 57 | /// |
| 58 | /// \code |
| 59 | /// unsigned short _cvtss_sh(float a, const int imm); |
| 60 | /// \endcode |
| 61 | /// |
| 62 | /// This intrinsic corresponds to the \c VCVTPS2PH instruction. |
| 63 | /// |
| 64 | /// \param a |
| 65 | /// A 32-bit single-precision float value to be converted to a 16-bit |
| 66 | /// half-precision float value. |
| 67 | /// \param imm |
| 68 | /// An immediate value controlling rounding using bits [2:0]: |
| 69 | /// 000: Nearest |
| 70 | /// 001: Down |
| 71 | /// 010: Up |
| 72 | /// 011: Truncate |
| 73 | /// 1XX: Use MXCSR.RC for rounding |
| 74 | /// \returns The converted 16-bit half-precision float value. |
| 75 | #define _cvtss_sh(a, imm) \ |
| 76 | ((unsigned short)(((__v8hi)__builtin_ia32_vcvtps2ph((__v4sf){a, 0, 0, 0}, \ |
| 77 | (imm)))[0])) |
| 78 | |
| 79 | /// \brief Converts a 128-bit vector containing 32-bit float values into a |
| 80 | /// 128-bit vector containing 16-bit half-precision float values. |
| 81 | /// |
| 82 | /// \headerfile <x86intrin.h> |
| 83 | /// |
| 84 | /// \code |
| 85 | /// __m128i _mm_cvtps_ph(__m128 a, const int imm); |
| 86 | /// \endcode |
| 87 | /// |
| 88 | /// This intrinsic corresponds to the \c VCVTPS2PH instruction. |
| 89 | /// |
| 90 | /// \param a |
| 91 | /// A 128-bit vector containing 32-bit float values. |
| 92 | /// \param imm |
| 93 | /// An immediate value controlling rounding using bits [2:0]: |
| 94 | /// 000: Nearest |
| 95 | /// 001: Down |
| 96 | /// 010: Up |
| 97 | /// 011: Truncate |
| 98 | /// 1XX: Use MXCSR.RC for rounding |
| 99 | /// \returns A 128-bit vector containing converted 16-bit half-precision float |
| 100 | /// values. The lower 64 bits are used to store the converted 16-bit |
| 101 | /// half-precision floating-point values. |
| 102 | #define _mm_cvtps_ph(a, imm) \ |
| 103 | ((__m128i)__builtin_ia32_vcvtps2ph((__v4sf)(__m128)(a), (imm))) |
| 104 | |
| 105 | /// \brief Converts a 128-bit vector containing 16-bit half-precision float |
| 106 | /// values into a 128-bit vector containing 32-bit float values. |
| 107 | /// |
| 108 | /// \headerfile <x86intrin.h> |
| 109 | /// |
| 110 | /// This intrinsic corresponds to the \c VCVTPH2PS instruction. |
| 111 | /// |
| 112 | /// \param a |
| 113 | /// A 128-bit vector containing 16-bit half-precision float values. The lower |
| 114 | /// 64 bits are used in the conversion. |
| 115 | /// \returns A 128-bit vector of [4 x float] containing converted float values. |
| 116 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 117 | _mm_cvtph_ps(__m128i __a) |
| 118 | { |
| 119 | return (__m128)__builtin_ia32_vcvtph2ps((__v8hi)__a); |
| 120 | } |
| 121 | |
| 122 | #undef __DEFAULT_FN_ATTRS |
| 123 | |
| 124 | #endif /* __F16CINTRIN_H */ |