| 1 |
/*===---- pmmintrin.h - SSE3 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 __PMMINTRIN_H |
| 11 |
#define __PMMINTRIN_H |
| 12 |
|
| 13 |
#if !defined(__i386__) && !defined(__x86_64__) |
| 14 |
#error "This header is only meant to be used on x86 and x64 architecture" |
| 15 |
#endif |
| 16 |
|
| 17 |
#include <emmintrin.h> |
| 18 |
|
| 19 |
/* Define the default attributes for the functions in this file. */ |
| 20 |
#define __DEFAULT_FN_ATTRS \ |
| 21 |
__attribute__((__always_inline__, __nodebug__, __target__("sse3"), __min_vector_width__(128))) |
| 22 |
|
| 23 |
/// Loads data from an unaligned memory location to elements in a 128-bit |
| 24 |
/// vector. |
| 25 |
/// |
| 26 |
/// If the address of the data is not 16-byte aligned, the instruction may |
| 27 |
/// read two adjacent aligned blocks of memory to retrieve the requested |
| 28 |
/// data. |
| 29 |
/// |
| 30 |
/// \headerfile <x86intrin.h> |
| 31 |
/// |
| 32 |
/// This intrinsic corresponds to the <c> VLDDQU </c> instruction. |
| 33 |
/// |
| 34 |
/// \param __p |
| 35 |
/// A pointer to a 128-bit integer vector containing integer values. |
| 36 |
/// \returns A 128-bit vector containing the moved values. |
| 37 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 38 |
_mm_lddqu_si128(__m128i_u const *__p) |
| 39 |
{ |
| 40 |
return (__m128i)__builtin_ia32_lddqu((char const *)__p); |
| 41 |
} |
| 42 |
|
| 43 |
/// Adds the even-indexed values and subtracts the odd-indexed values of |
| 44 |
/// two 128-bit vectors of [4 x float]. |
| 45 |
/// |
| 46 |
/// \headerfile <x86intrin.h> |
| 47 |
/// |
| 48 |
/// This intrinsic corresponds to the <c> VADDSUBPS </c> instruction. |
| 49 |
/// |
| 50 |
/// \param __a |
| 51 |
/// A 128-bit vector of [4 x float] containing the left source operand. |
| 52 |
/// \param __b |
| 53 |
/// A 128-bit vector of [4 x float] containing the right source operand. |
| 54 |
/// \returns A 128-bit vector of [4 x float] containing the alternating sums and |
| 55 |
/// differences of both operands. |
| 56 |
static __inline__ __m128 __DEFAULT_FN_ATTRS |
| 57 |
_mm_addsub_ps(__m128 __a, __m128 __b) |
| 58 |
{ |
| 59 |
return __builtin_ia32_addsubps((__v4sf)__a, (__v4sf)__b); |
| 60 |
} |
| 61 |
|
| 62 |
/// Horizontally adds the adjacent pairs of values contained in two |
| 63 |
/// 128-bit vectors of [4 x float]. |
| 64 |
/// |
| 65 |
/// \headerfile <x86intrin.h> |
| 66 |
/// |
| 67 |
/// This intrinsic corresponds to the <c> VHADDPS </c> instruction. |
| 68 |
/// |
| 69 |
/// \param __a |
| 70 |
/// A 128-bit vector of [4 x float] containing one of the source operands. |
| 71 |
/// The horizontal sums of the values are stored in the lower bits of the |
| 72 |
/// destination. |
| 73 |
/// \param __b |
| 74 |
/// A 128-bit vector of [4 x float] containing one of the source operands. |
| 75 |
/// The horizontal sums of the values are stored in the upper bits of the |
| 76 |
/// destination. |
| 77 |
/// \returns A 128-bit vector of [4 x float] containing the horizontal sums of |
| 78 |
/// both operands. |
| 79 |
static __inline__ __m128 __DEFAULT_FN_ATTRS |
| 80 |
_mm_hadd_ps(__m128 __a, __m128 __b) |
| 81 |
{ |
| 82 |
return __builtin_ia32_haddps((__v4sf)__a, (__v4sf)__b); |
| 83 |
} |
| 84 |
|
| 85 |
/// Horizontally subtracts the adjacent pairs of values contained in two |
| 86 |
/// 128-bit vectors of [4 x float]. |
| 87 |
/// |
| 88 |
/// \headerfile <x86intrin.h> |
| 89 |
/// |
| 90 |
/// This intrinsic corresponds to the <c> VHSUBPS </c> instruction. |
| 91 |
/// |
| 92 |
/// \param __a |
| 93 |
/// A 128-bit vector of [4 x float] containing one of the source operands. |
| 94 |
/// The horizontal differences between the values are stored in the lower |
| 95 |
/// bits of the destination. |
| 96 |
/// \param __b |
| 97 |
/// A 128-bit vector of [4 x float] containing one of the source operands. |
| 98 |
/// The horizontal differences between the values are stored in the upper |
| 99 |
/// bits of the destination. |
| 100 |
/// \returns A 128-bit vector of [4 x float] containing the horizontal |
| 101 |
/// differences of both operands. |
| 102 |
static __inline__ __m128 __DEFAULT_FN_ATTRS |
| 103 |
_mm_hsub_ps(__m128 __a, __m128 __b) |
| 104 |
{ |
| 105 |
return __builtin_ia32_hsubps((__v4sf)__a, (__v4sf)__b); |
| 106 |
} |
| 107 |
|
| 108 |
/// Moves and duplicates odd-indexed values from a 128-bit vector |
| 109 |
/// of [4 x float] to float values stored in a 128-bit vector of |
| 110 |
/// [4 x float]. |
| 111 |
/// |
| 112 |
/// \headerfile <x86intrin.h> |
| 113 |
/// |
| 114 |
/// This intrinsic corresponds to the <c> VMOVSHDUP </c> instruction. |
| 115 |
/// |
| 116 |
/// \param __a |
| 117 |
/// A 128-bit vector of [4 x float]. \n |
| 118 |
/// Bits [127:96] of the source are written to bits [127:96] and [95:64] of |
| 119 |
/// the destination. \n |
| 120 |
/// Bits [63:32] of the source are written to bits [63:32] and [31:0] of the |
| 121 |
/// destination. |
| 122 |
/// \returns A 128-bit vector of [4 x float] containing the moved and duplicated |
| 123 |
/// values. |
| 124 |
static __inline__ __m128 __DEFAULT_FN_ATTRS |
| 125 |
_mm_movehdup_ps(__m128 __a) |
| 126 |
{ |
| 127 |
return __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 1, 1, 3, 3); |
| 128 |
} |
| 129 |
|
| 130 |
/// Duplicates even-indexed values from a 128-bit vector of |
| 131 |
/// [4 x float] to float values stored in a 128-bit vector of [4 x float]. |
| 132 |
/// |
| 133 |
/// \headerfile <x86intrin.h> |
| 134 |
/// |
| 135 |
/// This intrinsic corresponds to the <c> VMOVSLDUP </c> instruction. |
| 136 |
/// |
| 137 |
/// \param __a |
| 138 |
/// A 128-bit vector of [4 x float] \n |
| 139 |
/// Bits [95:64] of the source are written to bits [127:96] and [95:64] of |
| 140 |
/// the destination. \n |
| 141 |
/// Bits [31:0] of the source are written to bits [63:32] and [31:0] of the |
| 142 |
/// destination. |
| 143 |
/// \returns A 128-bit vector of [4 x float] containing the moved and duplicated |
| 144 |
/// values. |
| 145 |
static __inline__ __m128 __DEFAULT_FN_ATTRS |
| 146 |
_mm_moveldup_ps(__m128 __a) |
| 147 |
{ |
| 148 |
return __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 0, 0, 2, 2); |
| 149 |
} |
| 150 |
|
| 151 |
/// Adds the even-indexed values and subtracts the odd-indexed values of |
| 152 |
/// two 128-bit vectors of [2 x double]. |
| 153 |
/// |
| 154 |
/// \headerfile <x86intrin.h> |
| 155 |
/// |
| 156 |
/// This intrinsic corresponds to the <c> VADDSUBPD </c> instruction. |
| 157 |
/// |
| 158 |
/// \param __a |
| 159 |
/// A 128-bit vector of [2 x double] containing the left source operand. |
| 160 |
/// \param __b |
| 161 |
/// A 128-bit vector of [2 x double] containing the right source operand. |
| 162 |
/// \returns A 128-bit vector of [2 x double] containing the alternating sums |
| 163 |
/// and differences of both operands. |
| 164 |
static __inline__ __m128d __DEFAULT_FN_ATTRS |
| 165 |
_mm_addsub_pd(__m128d __a, __m128d __b) |
| 166 |
{ |
| 167 |
return __builtin_ia32_addsubpd((__v2df)__a, (__v2df)__b); |
| 168 |
} |
| 169 |
|
| 170 |
/// Horizontally adds the pairs of values contained in two 128-bit |
| 171 |
/// vectors of [2 x double]. |
| 172 |
/// |
| 173 |
/// \headerfile <x86intrin.h> |
| 174 |
/// |
| 175 |
/// This intrinsic corresponds to the <c> VHADDPD </c> instruction. |
| 176 |
/// |
| 177 |
/// \param __a |
| 178 |
/// A 128-bit vector of [2 x double] containing one of the source operands. |
| 179 |
/// The horizontal sum of the values is stored in the lower bits of the |
| 180 |
/// destination. |
| 181 |
/// \param __b |
| 182 |
/// A 128-bit vector of [2 x double] containing one of the source operands. |
| 183 |
/// The horizontal sum of the values is stored in the upper bits of the |
| 184 |
/// destination. |
| 185 |
/// \returns A 128-bit vector of [2 x double] containing the horizontal sums of |
| 186 |
/// both operands. |
| 187 |
static __inline__ __m128d __DEFAULT_FN_ATTRS |
| 188 |
_mm_hadd_pd(__m128d __a, __m128d __b) |
| 189 |
{ |
| 190 |
return __builtin_ia32_haddpd((__v2df)__a, (__v2df)__b); |
| 191 |
} |
| 192 |
|
| 193 |
/// Horizontally subtracts the pairs of values contained in two 128-bit |
| 194 |
/// vectors of [2 x double]. |
| 195 |
/// |
| 196 |
/// \headerfile <x86intrin.h> |
| 197 |
/// |
| 198 |
/// This intrinsic corresponds to the <c> VHSUBPD </c> instruction. |
| 199 |
/// |
| 200 |
/// \param __a |
| 201 |
/// A 128-bit vector of [2 x double] containing one of the source operands. |
| 202 |
/// The horizontal difference of the values is stored in the lower bits of |
| 203 |
/// the destination. |
| 204 |
/// \param __b |
| 205 |
/// A 128-bit vector of [2 x double] containing one of the source operands. |
| 206 |
/// The horizontal difference of the values is stored in the upper bits of |
| 207 |
/// the destination. |
| 208 |
/// \returns A 128-bit vector of [2 x double] containing the horizontal |
| 209 |
/// differences of both operands. |
| 210 |
static __inline__ __m128d __DEFAULT_FN_ATTRS |
| 211 |
_mm_hsub_pd(__m128d __a, __m128d __b) |
| 212 |
{ |
| 213 |
return __builtin_ia32_hsubpd((__v2df)__a, (__v2df)__b); |
| 214 |
} |
| 215 |
|
| 216 |
/// Moves and duplicates one double-precision value to double-precision |
| 217 |
/// values stored in a 128-bit vector of [2 x double]. |
| 218 |
/// |
| 219 |
/// \headerfile <x86intrin.h> |
| 220 |
/// |
| 221 |
/// \code |
| 222 |
/// __m128d _mm_loaddup_pd(double const *dp); |
| 223 |
/// \endcode |
| 224 |
/// |
| 225 |
/// This intrinsic corresponds to the <c> VMOVDDUP </c> instruction. |
| 226 |
/// |
| 227 |
/// \param dp |
| 228 |
/// A pointer to a double-precision value to be moved and duplicated. |
| 229 |
/// \returns A 128-bit vector of [2 x double] containing the moved and |
| 230 |
/// duplicated values. |
| 231 |
#define _mm_loaddup_pd(dp) _mm_load1_pd(dp) |
| 232 |
|
| 233 |
/// Moves and duplicates the double-precision value in the lower bits of |
| 234 |
/// a 128-bit vector of [2 x double] to double-precision values stored in a |
| 235 |
/// 128-bit vector of [2 x double]. |
| 236 |
/// |
| 237 |
/// \headerfile <x86intrin.h> |
| 238 |
/// |
| 239 |
/// This intrinsic corresponds to the <c> VMOVDDUP </c> instruction. |
| 240 |
/// |
| 241 |
/// \param __a |
| 242 |
/// A 128-bit vector of [2 x double]. Bits [63:0] are written to bits |
| 243 |
/// [127:64] and [63:0] of the destination. |
| 244 |
/// \returns A 128-bit vector of [2 x double] containing the moved and |
| 245 |
/// duplicated values. |
| 246 |
static __inline__ __m128d __DEFAULT_FN_ATTRS |
| 247 |
_mm_movedup_pd(__m128d __a) |
| 248 |
{ |
| 249 |
return __builtin_shufflevector((__v2df)__a, (__v2df)__a, 0, 0); |
| 250 |
} |
| 251 |
|
| 252 |
/// Establishes a linear address memory range to be monitored and puts |
| 253 |
/// the processor in the monitor event pending state. Data stored in the |
| 254 |
/// monitored address range causes the processor to exit the pending state. |
| 255 |
/// |
| 256 |
/// \headerfile <x86intrin.h> |
| 257 |
/// |
| 258 |
/// This intrinsic corresponds to the <c> MONITOR </c> instruction. |
| 259 |
/// |
| 260 |
/// \param __p |
| 261 |
/// The memory range to be monitored. The size of the range is determined by |
| 262 |
/// CPUID function 0000_0005h. |
| 263 |
/// \param __extensions |
| 264 |
/// Optional extensions for the monitoring state. |
| 265 |
/// \param __hints |
| 266 |
/// Optional hints for the monitoring state. |
| 267 |
static __inline__ void __DEFAULT_FN_ATTRS |
| 268 |
_mm_monitor(void const *__p, unsigned __extensions, unsigned __hints) |
| 269 |
{ |
| 270 |
__builtin_ia32_monitor(__p, __extensions, __hints); |
| 271 |
} |
| 272 |
|
| 273 |
/// Used with the MONITOR instruction to wait while the processor is in |
| 274 |
/// the monitor event pending state. Data stored in the monitored address |
| 275 |
/// range causes the processor to exit the pending state. |
| 276 |
/// |
| 277 |
/// \headerfile <x86intrin.h> |
| 278 |
/// |
| 279 |
/// This intrinsic corresponds to the <c> MWAIT </c> instruction. |
| 280 |
/// |
| 281 |
/// \param __extensions |
| 282 |
/// Optional extensions for the monitoring state, which may vary by |
| 283 |
/// processor. |
| 284 |
/// \param __hints |
| 285 |
/// Optional hints for the monitoring state, which may vary by processor. |
| 286 |
static __inline__ void __DEFAULT_FN_ATTRS |
| 287 |
_mm_mwait(unsigned __extensions, unsigned __hints) |
| 288 |
{ |
| 289 |
__builtin_ia32_mwait(__extensions, __hints); |
| 290 |
} |
| 291 |
|
| 292 |
#undef __DEFAULT_FN_ATTRS |
| 293 |
|
| 294 |
#endif /* __PMMINTRIN_H */ |