| 1 | /*===---- bmi2intrin.h - BMI2 intrinsics -----------------------------------=== |
| 2 | * |
| 3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | * See https://llvm.org/LICENSE.txt for license information. |
| 5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | * |
| 7 | *===-----------------------------------------------------------------------=== |
| 8 | */ |
| 9 | |
| 10 | #ifndef __IMMINTRIN_H |
| 11 | #error "Never use <bmi2intrin.h> directly; include <immintrin.h> instead." |
| 12 | #endif |
| 13 | |
| 14 | #ifndef __BMI2INTRIN_H |
| 15 | #define __BMI2INTRIN_H |
| 16 | |
| 17 | /* Define the default attributes for the functions in this file. */ |
| 18 | #if defined(__cplusplus) && (__cplusplus >= 201103L) |
| 19 | #define __DEFAULT_FN_ATTRS \ |
| 20 | __attribute__((__always_inline__, __nodebug__, __target__("bmi2"))) constexpr |
| 21 | #else |
| 22 | #define __DEFAULT_FN_ATTRS \ |
| 23 | __attribute__((__always_inline__, __nodebug__, __target__("bmi2"))) |
| 24 | #endif |
| 25 | |
| 26 | /// Copies the unsigned 32-bit integer \a __X and zeroes the upper bits |
| 27 | /// starting at bit number \a __Y. |
| 28 | /// |
| 29 | /// \code{.operation} |
| 30 | /// i := __Y[7:0] |
| 31 | /// result := __X |
| 32 | /// IF i < 32 |
| 33 | /// result[31:i] := 0 |
| 34 | /// FI |
| 35 | /// \endcode |
| 36 | /// |
| 37 | /// \headerfile <immintrin.h> |
| 38 | /// |
| 39 | /// This intrinsic corresponds to the \c BZHI instruction. |
| 40 | /// |
| 41 | /// \param __X |
| 42 | /// The 32-bit source value to copy. |
| 43 | /// \param __Y |
| 44 | /// The lower 8 bits specify the bit number of the lowest bit to zero. |
| 45 | /// \returns The partially zeroed 32-bit value. |
| 46 | static __inline__ unsigned int __DEFAULT_FN_ATTRS |
| 47 | _bzhi_u32(unsigned int __X, unsigned int __Y) { |
| 48 | return __builtin_ia32_bzhi_si(__X, __Y); |
| 49 | } |
| 50 | |
| 51 | /// Deposit (scatter) low-order bits from the unsigned 32-bit integer \a __X |
| 52 | /// into the 32-bit result, according to the mask in the unsigned 32-bit |
| 53 | /// integer \a __Y. All other bits of the result are zero. |
| 54 | /// |
| 55 | /// \code{.operation} |
| 56 | /// i := 0 |
| 57 | /// result := 0 |
| 58 | /// FOR m := 0 TO 31 |
| 59 | /// IF __Y[m] == 1 |
| 60 | /// result[m] := __X[i] |
| 61 | /// i := i + 1 |
| 62 | /// ENDIF |
| 63 | /// ENDFOR |
| 64 | /// \endcode |
| 65 | /// |
| 66 | /// \headerfile <immintrin.h> |
| 67 | /// |
| 68 | /// This intrinsic corresponds to the \c PDEP instruction. |
| 69 | /// |
| 70 | /// \param __X |
| 71 | /// The 32-bit source value to copy. |
| 72 | /// \param __Y |
| 73 | /// The 32-bit mask specifying where to deposit source bits. |
| 74 | /// \returns The 32-bit result. |
| 75 | static __inline__ unsigned int __DEFAULT_FN_ATTRS |
| 76 | _pdep_u32(unsigned int __X, unsigned int __Y) { |
| 77 | return __builtin_ia32_pdep_si(__X, __Y); |
| 78 | } |
| 79 | |
| 80 | /// Extract (gather) bits from the unsigned 32-bit integer \a __X into the |
| 81 | /// low-order bits of the 32-bit result, according to the mask in the |
| 82 | /// unsigned 32-bit integer \a __Y. All other bits of the result are zero. |
| 83 | /// |
| 84 | /// \code{.operation} |
| 85 | /// i := 0 |
| 86 | /// result := 0 |
| 87 | /// FOR m := 0 TO 31 |
| 88 | /// IF __Y[m] == 1 |
| 89 | /// result[i] := __X[m] |
| 90 | /// i := i + 1 |
| 91 | /// ENDIF |
| 92 | /// ENDFOR |
| 93 | /// \endcode |
| 94 | /// |
| 95 | /// \headerfile <immintrin.h> |
| 96 | /// |
| 97 | /// This intrinsic corresponds to the \c PEXT instruction. |
| 98 | /// |
| 99 | /// \param __X |
| 100 | /// The 32-bit source value to copy. |
| 101 | /// \param __Y |
| 102 | /// The 32-bit mask specifying which source bits to extract. |
| 103 | /// \returns The 32-bit result. |
| 104 | static __inline__ unsigned int __DEFAULT_FN_ATTRS |
| 105 | _pext_u32(unsigned int __X, unsigned int __Y) { |
| 106 | return __builtin_ia32_pext_si(__X, __Y); |
| 107 | } |
| 108 | |
| 109 | /// Multiplies the unsigned 32-bit integers \a __X and \a __Y to form a |
| 110 | /// 64-bit product. Stores the upper 32 bits of the product in the |
| 111 | /// memory at \a __P and returns the lower 32 bits. |
| 112 | /// |
| 113 | /// \code{.operation} |
| 114 | /// Store32(__P, (__X * __Y)[63:32]) |
| 115 | /// result := (__X * __Y)[31:0] |
| 116 | /// \endcode |
| 117 | /// |
| 118 | /// \headerfile <immintrin.h> |
| 119 | /// |
| 120 | /// This intrinsic corresponds to the \c MULX instruction. |
| 121 | /// |
| 122 | /// \param __X |
| 123 | /// An unsigned 32-bit multiplicand. |
| 124 | /// \param __Y |
| 125 | /// An unsigned 32-bit multiplicand. |
| 126 | /// \param __P |
| 127 | /// A pointer to memory for storing the upper half of the product. |
| 128 | /// \returns The lower half of the product. |
| 129 | static __inline__ unsigned int __DEFAULT_FN_ATTRS |
| 130 | _mulx_u32(unsigned int __X, unsigned int __Y, unsigned int *__P) { |
| 131 | unsigned long long __res = (unsigned long long) __X * __Y; |
| 132 | *__P = (unsigned int)(__res >> 32); |
| 133 | return (unsigned int)__res; |
| 134 | } |
| 135 | |
| 136 | #ifdef __x86_64__ |
| 137 | |
| 138 | /// Copies the unsigned 64-bit integer \a __X and zeroes the upper bits |
| 139 | /// starting at bit number \a __Y. |
| 140 | /// |
| 141 | /// \code{.operation} |
| 142 | /// i := __Y[7:0] |
| 143 | /// result := __X |
| 144 | /// IF i < 64 |
| 145 | /// result[63:i] := 0 |
| 146 | /// FI |
| 147 | /// \endcode |
| 148 | /// |
| 149 | /// \headerfile <immintrin.h> |
| 150 | /// |
| 151 | /// This intrinsic corresponds to the \c BZHI instruction. |
| 152 | /// |
| 153 | /// \param __X |
| 154 | /// The 64-bit source value to copy. |
| 155 | /// \param __Y |
| 156 | /// The lower 8 bits specify the bit number of the lowest bit to zero. |
| 157 | /// \returns The partially zeroed 64-bit value. |
| 158 | static __inline__ unsigned long long __DEFAULT_FN_ATTRS |
| 159 | _bzhi_u64(unsigned long long __X, unsigned long long __Y) { |
| 160 | return __builtin_ia32_bzhi_di(__X, __Y); |
| 161 | } |
| 162 | |
| 163 | /// Deposit (scatter) low-order bits from the unsigned 64-bit integer \a __X |
| 164 | /// into the 64-bit result, according to the mask in the unsigned 64-bit |
| 165 | /// integer \a __Y. All other bits of the result are zero. |
| 166 | /// |
| 167 | /// \code{.operation} |
| 168 | /// i := 0 |
| 169 | /// result := 0 |
| 170 | /// FOR m := 0 TO 63 |
| 171 | /// IF __Y[m] == 1 |
| 172 | /// result[m] := __X[i] |
| 173 | /// i := i + 1 |
| 174 | /// ENDIF |
| 175 | /// ENDFOR |
| 176 | /// \endcode |
| 177 | /// |
| 178 | /// \headerfile <immintrin.h> |
| 179 | /// |
| 180 | /// This intrinsic corresponds to the \c PDEP instruction. |
| 181 | /// |
| 182 | /// \param __X |
| 183 | /// The 64-bit source value to copy. |
| 184 | /// \param __Y |
| 185 | /// The 64-bit mask specifying where to deposit source bits. |
| 186 | /// \returns The 64-bit result. |
| 187 | static __inline__ unsigned long long __DEFAULT_FN_ATTRS |
| 188 | _pdep_u64(unsigned long long __X, unsigned long long __Y) { |
| 189 | return __builtin_ia32_pdep_di(__X, __Y); |
| 190 | } |
| 191 | |
| 192 | /// Extract (gather) bits from the unsigned 64-bit integer \a __X into the |
| 193 | /// low-order bits of the 64-bit result, according to the mask in the |
| 194 | /// unsigned 64-bit integer \a __Y. All other bits of the result are zero. |
| 195 | /// |
| 196 | /// \code{.operation} |
| 197 | /// i := 0 |
| 198 | /// result := 0 |
| 199 | /// FOR m := 0 TO 63 |
| 200 | /// IF __Y[m] == 1 |
| 201 | /// result[i] := __X[m] |
| 202 | /// i := i + 1 |
| 203 | /// ENDIF |
| 204 | /// ENDFOR |
| 205 | /// \endcode |
| 206 | /// |
| 207 | /// \headerfile <immintrin.h> |
| 208 | /// |
| 209 | /// This intrinsic corresponds to the \c PEXT instruction. |
| 210 | /// |
| 211 | /// \param __X |
| 212 | /// The 64-bit source value to copy. |
| 213 | /// \param __Y |
| 214 | /// The 64-bit mask specifying which source bits to extract. |
| 215 | /// \returns The 64-bit result. |
| 216 | static __inline__ unsigned long long __DEFAULT_FN_ATTRS |
| 217 | _pext_u64(unsigned long long __X, unsigned long long __Y) { |
| 218 | return __builtin_ia32_pext_di(__X, __Y); |
| 219 | } |
| 220 | |
| 221 | /// Multiplies the unsigned 64-bit integers \a __X and \a __Y to form a |
| 222 | /// 128-bit product. Stores the upper 64 bits of the product to the |
| 223 | /// memory addressed by \a __P and returns the lower 64 bits. |
| 224 | /// |
| 225 | /// \code{.operation} |
| 226 | /// Store64(__P, (__X * __Y)[127:64]) |
| 227 | /// result := (__X * __Y)[63:0] |
| 228 | /// \endcode |
| 229 | /// |
| 230 | /// \headerfile <immintrin.h> |
| 231 | /// |
| 232 | /// This intrinsic corresponds to the \c MULX instruction. |
| 233 | /// |
| 234 | /// \param __X |
| 235 | /// An unsigned 64-bit multiplicand. |
| 236 | /// \param __Y |
| 237 | /// An unsigned 64-bit multiplicand. |
| 238 | /// \param __P |
| 239 | /// A pointer to memory for storing the upper half of the product. |
| 240 | /// \returns The lower half of the product. |
| 241 | static __inline__ unsigned long long __DEFAULT_FN_ATTRS |
| 242 | _mulx_u64 (unsigned long long __X, unsigned long long __Y, |
| 243 | unsigned long long *__P) { |
| 244 | unsigned __int128 __res = (unsigned __int128) __X * __Y; |
| 245 | *__P = (unsigned long long) (__res >> 64); |
| 246 | return (unsigned long long) __res; |
| 247 | } |
| 248 | |
| 249 | #endif /* __x86_64__ */ |
| 250 | |
| 251 | #undef __DEFAULT_FN_ATTRS |
| 252 | |
| 253 | #endif /* __BMI2INTRIN_H */ |