|
|
|
| 1 |
|
/** |
| 2 |
|
*** Copyright (C) 1985-1999 Intel Corporation. All rights reserved. |
| 3 |
|
*** |
| 4 |
|
*** The information and source code contained herein is the exclusive |
| 5 |
|
*** property of Intel Corporation and may not be disclosed, examined |
| 6 |
|
*** or reproduced in whole or in part without explicit written authorization |
| 7 |
|
*** from the company. |
| 8 |
|
*** |
| 9 |
|
**/ |
| 10 |
|
|
| 11 |
|
/* |
| 12 |
|
* xmmintrin.h |
| 13 |
|
* |
| 14 |
|
* Principal header file for Streaming SIMD Extensions intrinsics |
| 15 |
|
* |
| 16 |
|
* The intrinsics package can be used in 2 ways, based whether or not |
| 17 |
|
* _MM_FUNCTIONALITY is defined; if it is, the C/x87 implementation |
| 18 |
|
* will be used (the "faux intrinsics"). |
| 19 |
|
* |
| 20 |
|
* |
| 21 |
|
* Note that the m128 datatype provided using _MM2_FUNCTIONALITY mode is |
| 22 |
|
* implemented as struct, will not be 128b aligned, will be passed |
| 23 |
|
* via the stack, etc. MM_FUNCTIONALITY mode is not intended for |
| 24 |
|
* performance, just semantics. |
| 25 |
|
* |
| 26 |
|
*/ |
| 27 |
|
|
| 28 |
|
#pragma once |
| 29 |
|
#ifndef __midl |
| 30 |
|
#ifndef _INCLUDED_MM2 |
| 31 |
|
#define _INCLUDED_MM2 |
| 32 |
|
|
| 33 |
|
#include <crtdefs.h> |
| 34 |
|
|
| 35 |
|
#if defined(_M_CEE_PURE) |
| 36 |
|
#error ERROR: XMM intrinsics not supported in the pure mode! |
| 37 |
|
#else |
| 38 |
|
|
| 39 |
|
/* |
| 40 |
|
* the m64 type is required for the integer Streaming SIMD Extensions intrinsics |
| 41 |
|
*/ |
| 42 |
|
#ifndef _MMINTRIN_H_INCLUDED |
| 43 |
|
#include <mmintrin.h> |
| 44 |
|
#endif |
| 45 |
|
|
| 46 |
|
#ifdef _MM2_FUNCTIONALITY |
| 47 |
|
/* support old notation */ |
| 48 |
|
#ifndef _MM_FUNCTIONALITY |
| 49 |
|
#define _MM_FUNCTIONALITY |
| 50 |
|
#endif |
| 51 |
|
#endif |
| 52 |
|
|
| 53 |
|
#ifdef __ICL |
| 54 |
|
#ifdef _MM_FUNCTIONALITY |
| 55 |
|
#include "xmm_func.h" |
| 56 |
|
#else |
| 57 |
|
/* using real intrinsics */ |
| 58 |
|
typedef long long __m128; |
| 59 |
|
#endif |
| 60 |
|
#else |
| 61 |
|
|
| 62 |
|
typedef union __declspec(intrin_type) _CRT_ALIGN(16) __m128 { |
| 63 |
|
float m128_f32[4]; |
| 64 |
|
unsigned __int64 m128_u64[2]; |
| 65 |
|
__int8 m128_i8[16]; |
| 66 |
|
__int16 m128_i16[8]; |
| 67 |
|
__int32 m128_i32[4]; |
| 68 |
|
__int64 m128_i64[2]; |
| 69 |
|
unsigned __int8 m128_u8[16]; |
| 70 |
|
unsigned __int16 m128_u16[8]; |
| 71 |
|
unsigned __int32 m128_u32[4]; |
| 72 |
|
} __m128; |
| 73 |
|
|
| 74 |
|
#ifndef _INC_MALLOC |
| 75 |
|
/* pick up _mm_malloc() and _mm_free() */ |
| 76 |
|
#include <malloc.h> |
| 77 |
|
#endif |
| 78 |
|
#endif |
| 79 |
|
|
| 80 |
|
/*******************************************************/ |
| 81 |
|
/* MACRO for shuffle parameter for _mm_shuffle_ps(). */ |
| 82 |
|
/* Argument fp3 is a digit[0123] that represents the fp*/ |
| 83 |
|
/* from argument "b" of mm_shuffle_ps that will be */ |
| 84 |
|
/* placed in fp3 of result. fp2 is the same for fp2 in */ |
| 85 |
|
/* result. fp1 is a digit[0123] that represents the fp */ |
| 86 |
|
/* from argument "a" of mm_shuffle_ps that will be */ |
| 87 |
|
/* places in fp1 of result. fp0 is the same for fp0 of */ |
| 88 |
|
/* result */ |
| 89 |
|
/*******************************************************/ |
| 90 |
|
#define _MM_SHUFFLE(fp3,fp2,fp1,fp0) (((fp3) << 6) | ((fp2) << 4) | \ |
| 91 |
|
((fp1) << 2) | ((fp0))) |
| 92 |
|
|
| 93 |
|
|
| 94 |
|
/*******************************************************/ |
| 95 |
|
/* MACRO for performing the transpose of a 4x4 matrix */ |
| 96 |
|
/* of single precision floating point values. */ |
| 97 |
|
/* Arguments row0, row1, row2, and row3 are __m128 */ |
| 98 |
|
/* values whose elements form the corresponding rows */ |
| 99 |
|
/* of a 4x4 matrix. The matrix transpose is returned */ |
| 100 |
|
/* in arguments row0, row1, row2, and row3 where row0 */ |
| 101 |
|
/* now holds column 0 of the original matrix, row1 now */ |
| 102 |
|
/* holds column 1 of the original matrix, etc. */ |
| 103 |
|
/*******************************************************/ |
| 104 |
|
#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) { \ |
| 105 |
|
__m128 tmp3, tmp2, tmp1, tmp0; \ |
| 106 |
|
\ |
| 107 |
|
tmp0 = _mm_shuffle_ps((row0), (row1), 0x44); \ |
| 108 |
|
tmp2 = _mm_shuffle_ps((row0), (row1), 0xEE); \ |
| 109 |
|
tmp1 = _mm_shuffle_ps((row2), (row3), 0x44); \ |
| 110 |
|
tmp3 = _mm_shuffle_ps((row2), (row3), 0xEE); \ |
| 111 |
|
\ |
| 112 |
|
(row0) = _mm_shuffle_ps(tmp0, tmp1, 0x88); \ |
| 113 |
|
(row1) = _mm_shuffle_ps(tmp0, tmp1, 0xDD); \ |
| 114 |
|
(row2) = _mm_shuffle_ps(tmp2, tmp3, 0x88); \ |
| 115 |
|
(row3) = _mm_shuffle_ps(tmp2, tmp3, 0xDD); \ |
| 116 |
|
} |
| 117 |
|
|
| 118 |
|
|
| 119 |
|
/* constants for use with _mm_prefetch */ |
| 120 |
|
#define _MM_HINT_T0 1 |
| 121 |
|
#define _MM_HINT_T1 2 |
| 122 |
|
#define _MM_HINT_T2 3 |
| 123 |
|
#define _MM_HINT_NTA 0 |
| 124 |
|
|
| 125 |
|
/* (this declspec not supported with 0.A or 0.B) */ |
| 126 |
|
#define _MM_ALIGN16 _CRT_ALIGN(16) |
| 127 |
|
|
| 128 |
|
/* MACRO functions for setting and reading the MXCSR */ |
| 129 |
|
#define _MM_EXCEPT_MASK 0x003f |
| 130 |
|
#define _MM_EXCEPT_INVALID 0x0001 |
| 131 |
|
#define _MM_EXCEPT_DENORM 0x0002 |
| 132 |
|
#define _MM_EXCEPT_DIV_ZERO 0x0004 |
| 133 |
|
#define _MM_EXCEPT_OVERFLOW 0x0008 |
| 134 |
|
#define _MM_EXCEPT_UNDERFLOW 0x0010 |
| 135 |
|
#define _MM_EXCEPT_INEXACT 0x0020 |
| 136 |
|
|
| 137 |
|
#define _MM_MASK_MASK 0x1f80 |
| 138 |
|
#define _MM_MASK_INVALID 0x0080 |
| 139 |
|
#define _MM_MASK_DENORM 0x0100 |
| 140 |
|
#define _MM_MASK_DIV_ZERO 0x0200 |
| 141 |
|
#define _MM_MASK_OVERFLOW 0x0400 |
| 142 |
|
#define _MM_MASK_UNDERFLOW 0x0800 |
| 143 |
|
#define _MM_MASK_INEXACT 0x1000 |
| 144 |
|
|
| 145 |
|
#define _MM_ROUND_MASK 0x6000 |
| 146 |
|
#define _MM_ROUND_NEAREST 0x0000 |
| 147 |
|
#define _MM_ROUND_DOWN 0x2000 |
| 148 |
|
#define _MM_ROUND_UP 0x4000 |
| 149 |
|
#define _MM_ROUND_TOWARD_ZERO 0x6000 |
| 150 |
|
|
| 151 |
|
#define _MM_FLUSH_ZERO_MASK 0x8000 |
| 152 |
|
#define _MM_FLUSH_ZERO_ON 0x8000 |
| 153 |
|
#define _MM_FLUSH_ZERO_OFF 0x0000 |
| 154 |
|
|
| 155 |
|
#define _MM_SET_EXCEPTION_STATE(mask) \ |
| 156 |
|
_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (mask)) |
| 157 |
|
#define _MM_GET_EXCEPTION_STATE() \ |
| 158 |
|
(_mm_getcsr() & _MM_EXCEPT_MASK) |
| 159 |
|
|
| 160 |
|
#define _MM_SET_EXCEPTION_MASK(mask) \ |
| 161 |
|
_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (mask)) |
| 162 |
|
#define _MM_GET_EXCEPTION_MASK() \ |
| 163 |
|
(_mm_getcsr() & _MM_MASK_MASK) |
| 164 |
|
|
| 165 |
|
#define _MM_SET_ROUNDING_MODE(mode) \ |
| 166 |
|
_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (mode)) |
| 167 |
|
#define _MM_GET_ROUNDING_MODE() \ |
| 168 |
|
(_mm_getcsr() & _MM_ROUND_MASK) |
| 169 |
|
|
| 170 |
|
#define _MM_SET_FLUSH_ZERO_MODE(mode) \ |
| 171 |
|
_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (mode)) |
| 172 |
|
#define _MM_GET_FLUSH_ZERO_MODE(mode) \ |
| 173 |
|
(_mm_getcsr() & _MM_FLUSH_ZERO_MASK) |
| 174 |
|
|
| 175 |
|
/*****************************************************/ |
| 176 |
|
/* INTRINSICS FUNCTION PROTOTYPES START HERE */ |
| 177 |
|
/*****************************************************/ |
| 178 |
|
|
| 179 |
|
#if defined __cplusplus |
| 180 |
|
extern "C" { /* Begin "C" */ |
| 181 |
|
/* Intrinsics use C name-mangling. */ |
| 182 |
|
#endif /* __cplusplus */ |
| 183 |
|
|
| 184 |
|
/* |
| 185 |
|
* FP, arithmetic |
| 186 |
|
*/ |
| 187 |
|
|
| 188 |
|
extern __m128 _mm_add_ss(__m128 _A, __m128 _B); |
| 189 |
|
extern __m128 _mm_add_ps(__m128 _A, __m128 _B); |
| 190 |
|
extern __m128 _mm_sub_ss(__m128 _A, __m128 _B); |
| 191 |
|
extern __m128 _mm_sub_ps(__m128 _A, __m128 _B); |
| 192 |
|
extern __m128 _mm_mul_ss(__m128 _A, __m128 _B); |
| 193 |
|
extern __m128 _mm_mul_ps(__m128 _A, __m128 _B); |
| 194 |
|
extern __m128 _mm_div_ss(__m128 _A, __m128 _B); |
| 195 |
|
extern __m128 _mm_div_ps(__m128 _A, __m128 _B); |
| 196 |
|
extern __m128 _mm_sqrt_ss(__m128 _A); |
| 197 |
|
extern __m128 _mm_sqrt_ps(__m128 _A); |
| 198 |
|
extern __m128 _mm_rcp_ss(__m128 _A); |
| 199 |
|
extern __m128 _mm_rcp_ps(__m128 _A); |
| 200 |
|
extern __m128 _mm_rsqrt_ss(__m128 _A); |
| 201 |
|
extern __m128 _mm_rsqrt_ps(__m128 _A); |
| 202 |
|
extern __m128 _mm_min_ss(__m128 _A, __m128 _B); |
| 203 |
|
extern __m128 _mm_min_ps(__m128 _A, __m128 _B); |
| 204 |
|
extern __m128 _mm_max_ss(__m128 _A, __m128 _B); |
| 205 |
|
extern __m128 _mm_max_ps(__m128 _A, __m128 _B); |
| 206 |
|
|
| 207 |
|
/* |
| 208 |
|
* FP, logical |
| 209 |
|
*/ |
| 210 |
|
|
| 211 |
|
extern __m128 _mm_and_ps(__m128 _A, __m128 _B); |
| 212 |
|
extern __m128 _mm_andnot_ps(__m128 _A, __m128 _B); |
| 213 |
|
extern __m128 _mm_or_ps(__m128 _A, __m128 _B); |
| 214 |
|
extern __m128 _mm_xor_ps(__m128 _A, __m128 _B); |
| 215 |
|
|
| 216 |
|
/* |
| 217 |
|
* FP, comparison |
| 218 |
|
*/ |
| 219 |
|
|
| 220 |
|
extern __m128 _mm_cmpeq_ss(__m128 _A, __m128 _B); |
| 221 |
|
extern __m128 _mm_cmpeq_ps(__m128 _A, __m128 _B); |
| 222 |
|
extern __m128 _mm_cmplt_ss(__m128 _A, __m128 _B); |
| 223 |
|
extern __m128 _mm_cmplt_ps(__m128 _A, __m128 _B); |
| 224 |
|
extern __m128 _mm_cmple_ss(__m128 _A, __m128 _B); |
| 225 |
|
extern __m128 _mm_cmple_ps(__m128 _A, __m128 _B); |
| 226 |
|
extern __m128 _mm_cmpgt_ss(__m128 _A, __m128 _B); |
| 227 |
|
extern __m128 _mm_cmpgt_ps(__m128 _A, __m128 _B); |
| 228 |
|
extern __m128 _mm_cmpge_ss(__m128 _A, __m128 _B); |
| 229 |
|
extern __m128 _mm_cmpge_ps(__m128 _A, __m128 _B); |
| 230 |
|
extern __m128 _mm_cmpneq_ss(__m128 _A, __m128 _B); |
| 231 |
|
extern __m128 _mm_cmpneq_ps(__m128 _A, __m128 _B); |
| 232 |
|
extern __m128 _mm_cmpnlt_ss(__m128 _A, __m128 _B); |
| 233 |
|
extern __m128 _mm_cmpnlt_ps(__m128 _A, __m128 _B); |
| 234 |
|
extern __m128 _mm_cmpnle_ss(__m128 _A, __m128 _B); |
| 235 |
|
extern __m128 _mm_cmpnle_ps(__m128 _A, __m128 _B); |
| 236 |
|
extern __m128 _mm_cmpngt_ss(__m128 _A, __m128 _B); |
| 237 |
|
extern __m128 _mm_cmpngt_ps(__m128 _A, __m128 _B); |
| 238 |
|
extern __m128 _mm_cmpnge_ss(__m128 _A, __m128 _B); |
| 239 |
|
extern __m128 _mm_cmpnge_ps(__m128 _A, __m128 _B); |
| 240 |
|
extern __m128 _mm_cmpord_ss(__m128 _A, __m128 _B); |
| 241 |
|
extern __m128 _mm_cmpord_ps(__m128 _A, __m128 _B); |
| 242 |
|
extern __m128 _mm_cmpunord_ss(__m128 _A, __m128 _B); |
| 243 |
|
extern __m128 _mm_cmpunord_ps(__m128 _A, __m128 _B); |
| 244 |
|
extern int _mm_comieq_ss(__m128 _A, __m128 _B); |
| 245 |
|
extern int _mm_comilt_ss(__m128 _A, __m128 _B); |
| 246 |
|
extern int _mm_comile_ss(__m128 _A, __m128 _B); |
| 247 |
|
extern int _mm_comigt_ss(__m128 _A, __m128 _B); |
| 248 |
|
extern int _mm_comige_ss(__m128 _A, __m128 _B); |
| 249 |
|
extern int _mm_comineq_ss(__m128 _A, __m128 _B); |
| 250 |
|
extern int _mm_ucomieq_ss(__m128 _A, __m128 _B); |
| 251 |
|
extern int _mm_ucomilt_ss(__m128 _A, __m128 _B); |
| 252 |
|
extern int _mm_ucomile_ss(__m128 _A, __m128 _B); |
| 253 |
|
extern int _mm_ucomigt_ss(__m128 _A, __m128 _B); |
| 254 |
|
extern int _mm_ucomige_ss(__m128 _A, __m128 _B); |
| 255 |
|
extern int _mm_ucomineq_ss(__m128 _A, __m128 _B); |
| 256 |
|
|
| 257 |
|
/* |
| 258 |
|
* FP, conversions |
| 259 |
|
*/ |
| 260 |
|
|
| 261 |
|
extern int _mm_cvt_ss2si(__m128 _A); |
| 262 |
|
extern __m64 _mm_cvt_ps2pi(__m128 _A); |
| 263 |
|
extern int _mm_cvtt_ss2si(__m128 _A); |
| 264 |
|
extern __m64 _mm_cvtt_ps2pi(__m128 _A); |
| 265 |
|
extern __m128 _mm_cvt_si2ss(__m128, int); |
| 266 |
|
extern __m128 _mm_cvt_pi2ps(__m128, __m64); |
| 267 |
|
extern float _mm_cvtss_f32(__m128 _A); |
| 268 |
|
|
| 269 |
|
/* |
| 270 |
|
* Support for 64-bit extension intrinsics |
| 271 |
|
*/ |
| 272 |
|
#if defined(_M_AMD64) |
| 273 |
|
extern __int64 _mm_cvtss_si64(__m128 _A); |
| 274 |
|
extern __int64 _mm_cvttss_si64(__m128 _A); |
| 275 |
|
extern __m128 _mm_cvtsi64_ss(__m128 _A, __int64 _B); |
| 276 |
|
#endif |
| 277 |
|
|
| 278 |
|
/* |
| 279 |
|
* FP, misc |
| 280 |
|
*/ |
| 281 |
|
|
| 282 |
|
extern __m128 _mm_shuffle_ps(__m128 _A, __m128 _B, unsigned int _Imm8); |
| 283 |
|
extern __m128 _mm_unpackhi_ps(__m128 _A, __m128 _B); |
| 284 |
|
extern __m128 _mm_unpacklo_ps(__m128 _A, __m128 _B); |
| 285 |
|
extern __m128 _mm_loadh_pi(__m128, __m64 const*); |
| 286 |
|
extern __m128 _mm_movehl_ps(__m128, __m128); |
| 287 |
|
extern __m128 _mm_movelh_ps(__m128, __m128); |
| 288 |
|
extern void _mm_storeh_pi(__m64 *, __m128); |
| 289 |
|
extern __m128 _mm_loadl_pi(__m128, __m64 const*); |
| 290 |
|
extern void _mm_storel_pi(__m64 *, __m128); |
| 291 |
|
extern int _mm_movemask_ps(__m128 _A); |
| 292 |
|
|
| 293 |
|
|
| 294 |
|
/* |
| 295 |
|
* Integer extensions |
| 296 |
|
*/ |
| 297 |
|
extern int _m_pextrw(__m64, int); |
| 298 |
|
extern __m64 _m_pinsrw(__m64, int, int); |
| 299 |
|
extern __m64 _m_pmaxsw(__m64, __m64); |
| 300 |
|
extern __m64 _m_pmaxub(__m64, __m64); |
| 301 |
|
extern __m64 _m_pminsw(__m64, __m64); |
| 302 |
|
extern __m64 _m_pminub(__m64, __m64); |
| 303 |
|
extern int _m_pmovmskb(__m64); |
| 304 |
|
extern __m64 _m_pmulhuw(__m64, __m64); |
| 305 |
|
extern __m64 _m_pshufw(__m64, int); |
| 306 |
|
extern void _m_maskmovq(__m64, __m64, char *); |
| 307 |
|
extern __m64 _m_pavgb(__m64, __m64); |
| 308 |
|
extern __m64 _m_pavgw(__m64, __m64); |
| 309 |
|
extern __m64 _m_psadbw(__m64, __m64); |
| 310 |
|
|
| 311 |
|
/* |
| 312 |
|
* memory & initialization |
| 313 |
|
*/ |
| 314 |
|
|
| 315 |
|
extern __m128 _mm_set_ss(float _A); |
| 316 |
|
extern __m128 _mm_set_ps1(float _A); |
| 317 |
|
extern __m128 _mm_set_ps(float _A, float _B, float _C, float _D); |
| 318 |
|
extern __m128 _mm_setr_ps(float _A, float _B, float _C, float _D); |
| 319 |
|
extern __m128 _mm_setzero_ps(void); |
| 320 |
|
extern __m128 _mm_load_ss(float const*_A); |
| 321 |
|
extern __m128 _mm_load_ps1(float const*_A); |
| 322 |
|
extern __m128 _mm_load_ps(float const*_A); |
| 323 |
|
extern __m128 _mm_loadr_ps(float const*_A); |
| 324 |
|
extern __m128 _mm_loadu_ps(float const*_A); |
| 325 |
|
extern void _mm_store_ss(float *_V, __m128 _A); |
| 326 |
|
extern void _mm_store_ps1(float *_V, __m128 _A); |
| 327 |
|
extern void _mm_store_ps(float *_V, __m128 _A); |
| 328 |
|
extern void _mm_storer_ps(float *_V, __m128 _A); |
| 329 |
|
extern void _mm_storeu_ps(float *_V, __m128 _A); |
| 330 |
|
extern void _mm_prefetch(char const*_A, int _Sel); |
| 331 |
|
extern void _mm_stream_pi(__m64 *, __m64); |
| 332 |
|
extern void _mm_stream_ps(float *, __m128); |
| 333 |
|
extern __m128 _mm_move_ss(__m128 _A, __m128 _B); |
| 334 |
|
|
| 335 |
|
extern void _mm_sfence(void); |
| 336 |
|
extern unsigned int _mm_getcsr(void); |
| 337 |
|
extern void _mm_setcsr(unsigned int); |
| 338 |
|
|
| 339 |
|
#ifdef __ICL |
| 340 |
|
extern void* __cdecl _mm_malloc(size_t _Siz, size_t _Al); |
| 341 |
|
extern void __cdecl _mm_free(void *_P); |
| 342 |
|
#endif |
| 343 |
|
|
| 344 |
|
/* Alternate intrinsic names definition */ |
| 345 |
|
#define _mm_cvtss_si32 _mm_cvt_ss2si |
| 346 |
|
#define _mm_cvtps_pi32 _mm_cvt_ps2pi |
| 347 |
|
#define _mm_cvttss_si32 _mm_cvtt_ss2si |
| 348 |
|
#define _mm_cvttps_pi32 _mm_cvtt_ps2pi |
| 349 |
|
#define _mm_cvtsi32_ss _mm_cvt_si2ss |
| 350 |
|
#define _mm_cvtpi32_ps _mm_cvt_pi2ps |
| 351 |
|
#define _mm_extract_pi16 _m_pextrw |
| 352 |
|
#define _mm_insert_pi16 _m_pinsrw |
| 353 |
|
#define _mm_max_pi16 _m_pmaxsw |
| 354 |
|
#define _mm_max_pu8 _m_pmaxub |
| 355 |
|
#define _mm_min_pi16 _m_pminsw |
| 356 |
|
#define _mm_min_pu8 _m_pminub |
| 357 |
|
#define _mm_movemask_pi8 _m_pmovmskb |
| 358 |
|
#define _mm_mulhi_pu16 _m_pmulhuw |
| 359 |
|
#define _mm_shuffle_pi16 _m_pshufw |
| 360 |
|
#define _mm_maskmove_si64 _m_maskmovq |
| 361 |
|
#define _mm_avg_pu8 _m_pavgb |
| 362 |
|
#define _mm_avg_pu16 _m_pavgw |
| 363 |
|
#define _mm_sad_pu8 _m_psadbw |
| 364 |
|
#define _mm_set1_ps _mm_set_ps1 |
| 365 |
|
#define _mm_load1_ps _mm_load_ps1 |
| 366 |
|
#define _mm_store1_ps _mm_store_ps1 |
| 367 |
|
|
| 368 |
|
/******************************************************/ |
| 369 |
|
/* UTILITY INTRINSICS FUNCTION DEFINITIONS START HERE */ |
| 370 |
|
/******************************************************/ |
| 371 |
|
|
| 372 |
|
/*********************************************************/ |
| 373 |
|
/* NAME : _mm_cvtpi16_ps */ |
| 374 |
|
/* DESCRIPTION : Convert 4 16-bit signed integer values */ |
| 375 |
|
/* to 4 single-precision float values */ |
| 376 |
|
/* IN : __m64 a */ |
| 377 |
|
/* OUT : none */ |
| 378 |
|
/* RETURN : __m128 : (float)a */ |
| 379 |
|
/*********************************************************/ |
| 380 |
|
__inline __m128 _mm_cvtpi16_ps(__m64 a) |
| 381 |
|
{ |
| 382 |
|
__m128 tmp; |
| 383 |
|
__m64 ext_val = _mm_cmpgt_pi16(_mm_setzero_si64(), a); |
| 384 |
|
|
| 385 |
|
tmp = _mm_cvtpi32_ps(_mm_setzero_ps(), _mm_unpackhi_pi16(a, ext_val)); |
| 386 |
|
return(_mm_cvtpi32_ps(_mm_movelh_ps(tmp, tmp), |
| 387 |
|
_mm_unpacklo_pi16(a, ext_val))); |
| 388 |
|
} |
| 389 |
|
|
| 390 |
|
|
| 391 |
|
/***********************************************************/ |
| 392 |
|
/* NAME : _mm_cvtpu16_ps */ |
| 393 |
|
/* DESCRIPTION : Convert 4 16-bit unsigned integer values */ |
| 394 |
|
/* to 4 single-precision float values */ |
| 395 |
|
/* IN : __m64 a */ |
| 396 |
|
/* OUT : none */ |
| 397 |
|
/* RETURN : __m128 : (float)a */ |
| 398 |
|
/***********************************************************/ |
| 399 |
|
__inline __m128 _mm_cvtpu16_ps(__m64 a) |
| 400 |
|
{ |
| 401 |
|
__m128 tmp; |
| 402 |
|
__m64 ext_val = _mm_setzero_si64(); |
| 403 |
|
|
| 404 |
|
tmp = _mm_cvtpi32_ps(_mm_setzero_ps(), _mm_unpackhi_pi16(a, ext_val)); |
| 405 |
|
return(_mm_cvtpi32_ps(_mm_movelh_ps(tmp, tmp), |
| 406 |
|
_mm_unpacklo_pi16(a, ext_val))); |
| 407 |
|
} |