1 /* Copyright (C) 2002 Free Software Foundation, Inc.
3 This file is part of GNU CC.
5 GNU CC is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
10 GNU CC is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with GNU CC; see the file COPYING. If not, write to
17 the Free Software Foundation, 59 Temple Place - Suite 330,
18 Boston, MA 02111-1307, USA. */
20 /* As a special exception, if you include this header file into source
21 files compiled by GCC, this header file does not by itself cause
22 the resulting executable to be covered by the GNU General Public
23 License. This exception does not however invalidate any other
24 reasons why the executable file might be covered by the GNU General
27 /* Implemented from the specification included in the Intel C++ Compiler
28 User Guide and Reference, version 5.0. */
30 #ifndef _XMMINTRIN_H_INCLUDED
31 #define _XMMINTRIN_H_INCLUDED
34 # error "SSE instruction set not enabled"
37 /* We need type definitions from the MMX header file. */
40 /* The data type intended for user use. */
41 typedef int __m128 __attribute__ ((__mode__(__V4SF__)));
43 /* Internal data types for implementing the intrinsics. */
44 typedef int __v4sf __attribute__ ((__mode__(__V4SF__)));
45 typedef int __v4si __attribute__ ((__mode__(__V4SI__)));
47 /* Create a selector for use with the SHUFPS instruction. */
48 #define _MM_SHUFFLE(fp3,fp2,fp1,fp0) \
49 (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))
51 /* Constants for use with _mm_prefetch. */
60 /* Bits in the MXCSR. */
61 #define _MM_EXCEPT_MASK 0x003f
62 #define _MM_EXCEPT_INVALID 0x0001
63 #define _MM_EXCEPT_DENORM 0x0002
64 #define _MM_EXCEPT_DIV_ZERO 0x0004
65 #define _MM_EXCEPT_OVERFLOW 0x0008
66 #define _MM_EXCEPT_UNDERFLOW 0x0010
67 #define _MM_EXCEPT_INEXACT 0x0020
69 #define _MM_MASK_MASK 0x1f80
70 #define _MM_MASK_INVALID 0x0080
71 #define _MM_MASK_DENORM 0x0100
72 #define _MM_MASK_DIV_ZERO 0x0200
73 #define _MM_MASK_OVERFLOW 0x0400
74 #define _MM_MASK_UNDERFLOW 0x0800
75 #define _MM_MASK_INEXACT 0x1000
77 #define _MM_ROUND_MASK 0x6000
78 #define _MM_ROUND_NEAREST 0x0000
79 #define _MM_ROUND_DOWN 0x2000
80 #define _MM_ROUND_UP 0x4000
81 #define _MM_ROUND_TOWARD_ZERO 0x6000
83 #define _MM_FLUSH_ZERO_MASK 0x8000
84 #define _MM_FLUSH_ZERO_ON 0x8000
85 #define _MM_FLUSH_ZERO_OFF 0x0000
87 /* Perform the respective operation on the lower SPFP (single-precision
88 floating-point) values of A and B; the upper three SPFP values are
89 passed through from A. */
91 static __inline __m128
92 _mm_add_ss (__m128 __A, __m128 __B)
94 return (__m128) __builtin_ia32_addss ((__v4sf)__A, (__v4sf)__B);
97 static __inline __m128
98 _mm_sub_ss (__m128 __A, __m128 __B)
100 return (__m128) __builtin_ia32_subss ((__v4sf)__A, (__v4sf)__B);
103 static __inline __m128
104 _mm_mul_ss (__m128 __A, __m128 __B)
106 return (__m128) __builtin_ia32_mulss ((__v4sf)__A, (__v4sf)__B);
109 static __inline __m128
110 _mm_div_ss (__m128 __A, __m128 __B)
112 return (__m128) __builtin_ia32_divss ((__v4sf)__A, (__v4sf)__B);
115 static __inline __m128
116 _mm_sqrt_ss (__m128 __A)
118 return (__m128) __builtin_ia32_sqrtss ((__v4sf)__A);
121 static __inline __m128
122 _mm_rcp_ss (__m128 __A)
124 return (__m128) __builtin_ia32_rcpss ((__v4sf)__A);
127 static __inline __m128
128 _mm_rsqrt_ss (__m128 __A)
130 return (__m128) __builtin_ia32_rsqrtss ((__v4sf)__A);
133 static __inline __m128
134 _mm_min_ss (__m128 __A, __m128 __B)
136 return (__m128) __builtin_ia32_minss ((__v4sf)__A, (__v4sf)__B);
139 static __inline __m128
140 _mm_max_ss (__m128 __A, __m128 __B)
142 return (__m128) __builtin_ia32_maxss ((__v4sf)__A, (__v4sf)__B);
145 /* Perform the respective operation on the four SPFP values in A and B. */
147 static __inline __m128
148 _mm_add_ps (__m128 __A, __m128 __B)
150 return (__m128) __builtin_ia32_addps ((__v4sf)__A, (__v4sf)__B);
153 static __inline __m128
154 _mm_sub_ps (__m128 __A, __m128 __B)
156 return (__m128) __builtin_ia32_subps ((__v4sf)__A, (__v4sf)__B);
159 static __inline __m128
160 _mm_mul_ps (__m128 __A, __m128 __B)
162 return (__m128) __builtin_ia32_mulps ((__v4sf)__A, (__v4sf)__B);
165 static __inline __m128
166 _mm_div_ps (__m128 __A, __m128 __B)
168 return (__m128) __builtin_ia32_divps ((__v4sf)__A, (__v4sf)__B);
171 static __inline __m128
172 _mm_sqrt_ps (__m128 __A)
174 return (__m128) __builtin_ia32_sqrtps ((__v4sf)__A);
177 static __inline __m128
178 _mm_rcp_ps (__m128 __A)
180 return (__m128) __builtin_ia32_rcpps ((__v4sf)__A);
183 static __inline __m128
184 _mm_rsqrt_ps (__m128 __A)
186 return (__m128) __builtin_ia32_rsqrtps ((__v4sf)__A);
189 static __inline __m128
190 _mm_min_ps (__m128 __A, __m128 __B)
192 return (__m128) __builtin_ia32_minps ((__v4sf)__A, (__v4sf)__B);
195 static __inline __m128
196 _mm_max_ps (__m128 __A, __m128 __B)
198 return (__m128) __builtin_ia32_maxps ((__v4sf)__A, (__v4sf)__B);
201 /* Perform logical bit-wise operations on 128-bit values. */
203 static __inline __m128
204 _mm_and_ps (__m128 __A, __m128 __B)
206 return __builtin_ia32_andps (__A, __B);
209 static __inline __m128
210 _mm_andnot_ps (__m128 __A, __m128 __B)
212 return __builtin_ia32_andnps (__A, __B);
215 static __inline __m128
216 _mm_or_ps (__m128 __A, __m128 __B)
218 return __builtin_ia32_orps (__A, __B);
221 static __inline __m128
222 _mm_xor_ps (__m128 __A, __m128 __B)
224 return __builtin_ia32_xorps (__A, __B);
227 /* Perform a comparison on the lower SPFP values of A and B. If the
228 comparison is true, place a mask of all ones in the result, otherwise a
229 mask of zeros. The upper three SPFP values are passed through from A. */
231 static __inline __m128
232 _mm_cmpeq_ss (__m128 __A, __m128 __B)
234 return (__m128) __builtin_ia32_cmpeqss ((__v4sf)__A, (__v4sf)__B);
237 static __inline __m128
238 _mm_cmplt_ss (__m128 __A, __m128 __B)
240 return (__m128) __builtin_ia32_cmpltss ((__v4sf)__A, (__v4sf)__B);
243 static __inline __m128
244 _mm_cmple_ss (__m128 __A, __m128 __B)
246 return (__m128) __builtin_ia32_cmpless ((__v4sf)__A, (__v4sf)__B);
249 static __inline __m128
250 _mm_cmpgt_ss (__m128 __A, __m128 __B)
252 return (__m128) __builtin_ia32_movss ((__v4sf) __A,
254 __builtin_ia32_cmpltss ((__v4sf) __B,
259 static __inline __m128
260 _mm_cmpge_ss (__m128 __A, __m128 __B)
262 return (__m128) __builtin_ia32_movss ((__v4sf) __A,
264 __builtin_ia32_cmpless ((__v4sf) __B,
269 static __inline __m128
270 _mm_cmpneq_ss (__m128 __A, __m128 __B)
272 return (__m128) __builtin_ia32_cmpneqss ((__v4sf)__A, (__v4sf)__B);
275 static __inline __m128
276 _mm_cmpnlt_ss (__m128 __A, __m128 __B)
278 return (__m128) __builtin_ia32_cmpnltss ((__v4sf)__A, (__v4sf)__B);
281 static __inline __m128
282 _mm_cmpnle_ss (__m128 __A, __m128 __B)
284 return (__m128) __builtin_ia32_cmpnless ((__v4sf)__A, (__v4sf)__B);
287 static __inline __m128
288 _mm_cmpngt_ss (__m128 __A, __m128 __B)
290 return (__m128) __builtin_ia32_movss ((__v4sf) __A,
292 __builtin_ia32_cmpnltss ((__v4sf) __B,
297 static __inline __m128
298 _mm_cmpnge_ss (__m128 __A, __m128 __B)
300 return (__m128) __builtin_ia32_movss ((__v4sf) __A,
302 __builtin_ia32_cmpnless ((__v4sf) __B,
307 static __inline __m128
308 _mm_cmpord_ss (__m128 __A, __m128 __B)
310 return (__m128) __builtin_ia32_cmpordss ((__v4sf)__A, (__v4sf)__B);
313 static __inline __m128
314 _mm_cmpunord_ss (__m128 __A, __m128 __B)
316 return (__m128) __builtin_ia32_cmpunordss ((__v4sf)__A, (__v4sf)__B);
319 /* Perform a comparison on the four SPFP values of A and B. For each
320 element, if the comparison is true, place a mask of all ones in the
321 result, otherwise a mask of zeros. */
323 static __inline __m128
324 _mm_cmpeq_ps (__m128 __A, __m128 __B)
326 return (__m128) __builtin_ia32_cmpeqps ((__v4sf)__A, (__v4sf)__B);
329 static __inline __m128
330 _mm_cmplt_ps (__m128 __A, __m128 __B)
332 return (__m128) __builtin_ia32_cmpltps ((__v4sf)__A, (__v4sf)__B);
335 static __inline __m128
336 _mm_cmple_ps (__m128 __A, __m128 __B)
338 return (__m128) __builtin_ia32_cmpleps ((__v4sf)__A, (__v4sf)__B);
341 static __inline __m128
342 _mm_cmpgt_ps (__m128 __A, __m128 __B)
344 return (__m128) __builtin_ia32_cmpgtps ((__v4sf)__A, (__v4sf)__B);
347 static __inline __m128
348 _mm_cmpge_ps (__m128 __A, __m128 __B)
350 return (__m128) __builtin_ia32_cmpgeps ((__v4sf)__A, (__v4sf)__B);
353 static __inline __m128
354 _mm_cmpneq_ps (__m128 __A, __m128 __B)
356 return (__m128) __builtin_ia32_cmpneqps ((__v4sf)__A, (__v4sf)__B);
359 static __inline __m128
360 _mm_cmpnlt_ps (__m128 __A, __m128 __B)
362 return (__m128) __builtin_ia32_cmpnltps ((__v4sf)__A, (__v4sf)__B);
365 static __inline __m128
366 _mm_cmpnle_ps (__m128 __A, __m128 __B)
368 return (__m128) __builtin_ia32_cmpnleps ((__v4sf)__A, (__v4sf)__B);
371 static __inline __m128
372 _mm_cmpngt_ps (__m128 __A, __m128 __B)
374 return (__m128) __builtin_ia32_cmpngtps ((__v4sf)__A, (__v4sf)__B);
377 static __inline __m128
378 _mm_cmpnge_ps (__m128 __A, __m128 __B)
380 return (__m128) __builtin_ia32_cmpngeps ((__v4sf)__A, (__v4sf)__B);
383 static __inline __m128
384 _mm_cmpord_ps (__m128 __A, __m128 __B)
386 return (__m128) __builtin_ia32_cmpordps ((__v4sf)__A, (__v4sf)__B);
389 static __inline __m128
390 _mm_cmpunord_ps (__m128 __A, __m128 __B)
392 return (__m128) __builtin_ia32_cmpunordps ((__v4sf)__A, (__v4sf)__B);
395 /* Compare the lower SPFP values of A and B and return 1 if true
399 _mm_comieq_ss (__m128 __A, __m128 __B)
401 return __builtin_ia32_comieq ((__v4sf)__A, (__v4sf)__B);
405 _mm_comilt_ss (__m128 __A, __m128 __B)
407 return __builtin_ia32_comilt ((__v4sf)__A, (__v4sf)__B);
411 _mm_comile_ss (__m128 __A, __m128 __B)
413 return __builtin_ia32_comile ((__v4sf)__A, (__v4sf)__B);
417 _mm_comigt_ss (__m128 __A, __m128 __B)
419 return __builtin_ia32_comigt ((__v4sf)__A, (__v4sf)__B);
423 _mm_comige_ss (__m128 __A, __m128 __B)
425 return __builtin_ia32_comige ((__v4sf)__A, (__v4sf)__B);
429 _mm_comineq_ss (__m128 __A, __m128 __B)
431 return __builtin_ia32_comineq ((__v4sf)__A, (__v4sf)__B);
435 _mm_ucomieq_ss (__m128 __A, __m128 __B)
437 return __builtin_ia32_ucomieq ((__v4sf)__A, (__v4sf)__B);
441 _mm_ucomilt_ss (__m128 __A, __m128 __B)
443 return __builtin_ia32_ucomilt ((__v4sf)__A, (__v4sf)__B);
447 _mm_ucomile_ss (__m128 __A, __m128 __B)
449 return __builtin_ia32_ucomile ((__v4sf)__A, (__v4sf)__B);
453 _mm_ucomigt_ss (__m128 __A, __m128 __B)
455 return __builtin_ia32_ucomigt ((__v4sf)__A, (__v4sf)__B);
459 _mm_ucomige_ss (__m128 __A, __m128 __B)
461 return __builtin_ia32_ucomige ((__v4sf)__A, (__v4sf)__B);
465 _mm_ucomineq_ss (__m128 __A, __m128 __B)
467 return __builtin_ia32_ucomineq ((__v4sf)__A, (__v4sf)__B);
470 /* Convert the lower SPFP value to a 32-bit integer according to the current
473 _mm_cvtss_si32 (__m128 __A)
475 return __builtin_ia32_cvtss2si ((__v4sf) __A);
479 /* Convert the lower SPFP value to a 32-bit integer according to the current
481 static __inline long long
482 _mm_cvtss_si64x (__m128 __A)
484 return __builtin_ia32_cvtss2si64 ((__v4sf) __A);
488 /* Convert the two lower SPFP values to 32-bit integers according to the
489 current rounding mode. Return the integers in packed form. */
490 static __inline __m64
491 _mm_cvtps_pi32 (__m128 __A)
493 return (__m64) __builtin_ia32_cvtps2pi ((__v4sf) __A);
496 /* Truncate the lower SPFP value to a 32-bit integer. */
498 _mm_cvttss_si32 (__m128 __A)
500 return __builtin_ia32_cvttss2si ((__v4sf) __A);
504 /* Truncate the lower SPFP value to a 32-bit integer. */
505 static __inline long long
506 _mm_cvttss_si64x (__m128 __A)
508 return __builtin_ia32_cvttss2si64 ((__v4sf) __A);
512 /* Truncate the two lower SPFP values to 32-bit integers. Return the
513 integers in packed form. */
514 static __inline __m64
515 _mm_cvttps_pi32 (__m128 __A)
517 return (__m64) __builtin_ia32_cvttps2pi ((__v4sf) __A);
520 /* Convert B to a SPFP value and insert it as element zero in A. */
521 static __inline __m128
522 _mm_cvtsi32_ss (__m128 __A, int __B)
524 return (__m128) __builtin_ia32_cvtsi2ss ((__v4sf) __A, __B);
528 /* Convert B to a SPFP value and insert it as element zero in A. */
529 static __inline __m128
530 _mm_cvtsi64x_ss (__m128 __A, long long __B)
532 return (__m128) __builtin_ia32_cvtsi642ss ((__v4sf) __A, __B);
536 /* Convert the two 32-bit values in B to SPFP form and insert them
537 as the two lower elements in A. */
538 static __inline __m128
539 _mm_cvtpi32_ps (__m128 __A, __m64 __B)
541 return (__m128) __builtin_ia32_cvtpi2ps ((__v4sf) __A, (__v2si)__B);
544 /* Convert the four signed 16-bit values in A to SPFP form. */
545 static __inline __m128
546 _mm_cvtpi16_ps (__m64 __A)
549 __v2si __hisi, __losi;
552 /* This comparison against zero gives us a mask that can be used to
553 fill in the missing sign bits in the unpack operations below, so
554 that we get signed values after unpacking. */
555 __sign = (__v4hi) __builtin_ia32_mmx_zero ();
556 __sign = __builtin_ia32_pcmpgtw (__sign, (__v4hi)__A);
558 /* Convert the four words to doublewords. */
559 __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __sign);
560 __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __sign);
562 /* Convert the doublewords to floating point two at a time. */
563 __r = (__v4sf) __builtin_ia32_setzerops ();
564 __r = __builtin_ia32_cvtpi2ps (__r, __hisi);
565 __r = __builtin_ia32_movlhps (__r, __r);
566 __r = __builtin_ia32_cvtpi2ps (__r, __losi);
571 /* Convert the four unsigned 16-bit values in A to SPFP form. */
572 static __inline __m128
573 _mm_cvtpu16_ps (__m64 __A)
575 __v4hi __zero = (__v4hi) __builtin_ia32_mmx_zero ();
576 __v2si __hisi, __losi;
579 /* Convert the four words to doublewords. */
580 __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __zero);
581 __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __zero);
583 /* Convert the doublewords to floating point two at a time. */
584 __r = (__v4sf) __builtin_ia32_setzerops ();
585 __r = __builtin_ia32_cvtpi2ps (__r, __hisi);
586 __r = __builtin_ia32_movlhps (__r, __r);
587 __r = __builtin_ia32_cvtpi2ps (__r, __losi);
592 /* Convert the low four signed 8-bit values in A to SPFP form. */
593 static __inline __m128
594 _mm_cvtpi8_ps (__m64 __A)
598 /* This comparison against zero gives us a mask that can be used to
599 fill in the missing sign bits in the unpack operations below, so
600 that we get signed values after unpacking. */
601 __sign = (__v8qi) __builtin_ia32_mmx_zero ();
602 __sign = __builtin_ia32_pcmpgtb (__sign, (__v8qi)__A);
604 /* Convert the four low bytes to words. */
605 __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __sign);
607 return _mm_cvtpi16_ps(__A);
610 /* Convert the low four unsigned 8-bit values in A to SPFP form. */
611 static __inline __m128
612 _mm_cvtpu8_ps(__m64 __A)
614 __v8qi __zero = (__v8qi) __builtin_ia32_mmx_zero ();
615 __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __zero);
616 return _mm_cvtpu16_ps(__A);
619 /* Convert the four signed 32-bit values in A and B to SPFP form. */
620 static __inline __m128
621 _mm_cvtpi32x2_ps(__m64 __A, __m64 __B)
623 __v4sf __zero = (__v4sf) __builtin_ia32_setzerops ();
624 __v4sf __sfa = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__A);
625 __v4sf __sfb = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__B);
626 return (__m128) __builtin_ia32_movlhps (__sfa, __sfb);
629 /* Convert the four SPFP values in A to four signed 16-bit integers. */
630 static __inline __m64
631 _mm_cvtps_pi16(__m128 __A)
633 __v4sf __hisf = (__v4sf)__A;
634 __v4sf __losf = __builtin_ia32_movhlps (__hisf, __hisf);
635 __v2si __hisi = __builtin_ia32_cvtps2pi (__hisf);
636 __v2si __losi = __builtin_ia32_cvtps2pi (__losf);
637 return (__m64) __builtin_ia32_packssdw (__hisi, __losi);
640 /* Convert the four SPFP values in A to four signed 8-bit integers. */
641 static __inline __m64
642 _mm_cvtps_pi8(__m128 __A)
644 __v4hi __tmp = (__v4hi) _mm_cvtps_pi16 (__A);
645 __v4hi __zero = (__v4hi) __builtin_ia32_mmx_zero ();
646 return (__m64) __builtin_ia32_packsswb (__tmp, __zero);
649 /* Selects four specific SPFP values from A and B based on MASK. */
651 static __inline __m128
652 _mm_shuffle_ps (__m128 __A, __m128 __B, int __mask)
654 return (__m128) __builtin_ia32_shufps ((__v4sf)__A, (__v4sf)__B, __mask);
657 #define _mm_shuffle_ps(A, B, MASK) \
658 ((__m128) __builtin_ia32_shufps ((__v4sf)(A), (__v4sf)(B), (MASK)))
662 /* Selects and interleaves the upper two SPFP values from A and B. */
663 static __inline __m128
664 _mm_unpackhi_ps (__m128 __A, __m128 __B)
666 return (__m128) __builtin_ia32_unpckhps ((__v4sf)__A, (__v4sf)__B);
669 /* Selects and interleaves the lower two SPFP values from A and B. */
670 static __inline __m128
671 _mm_unpacklo_ps (__m128 __A, __m128 __B)
673 return (__m128) __builtin_ia32_unpcklps ((__v4sf)__A, (__v4sf)__B);
676 /* Sets the upper two SPFP values with 64-bits of data loaded from P;
677 the lower two values are passed through from A. */
678 static __inline __m128
679 _mm_loadh_pi (__m128 __A, __m64 const *__P)
681 return (__m128) __builtin_ia32_loadhps ((__v4sf)__A, (__v2si *)__P);
684 /* Stores the upper two SPFP values of A into P. */
686 _mm_storeh_pi (__m64 *__P, __m128 __A)
688 __builtin_ia32_storehps ((__v2si *)__P, (__v4sf)__A);
691 /* Moves the upper two values of B into the lower two values of A. */
692 static __inline __m128
693 _mm_movehl_ps (__m128 __A, __m128 __B)
695 return (__m128) __builtin_ia32_movhlps ((__v4sf)__A, (__v4sf)__B);
698 /* Moves the lower two values of B into the upper two values of A. */
699 static __inline __m128
700 _mm_movelh_ps (__m128 __A, __m128 __B)
702 return (__m128) __builtin_ia32_movlhps ((__v4sf)__A, (__v4sf)__B);
705 /* Sets the lower two SPFP values with 64-bits of data loaded from P;
706 the upper two values are passed through from A. */
707 static __inline __m128
708 _mm_loadl_pi (__m128 __A, __m64 const *__P)
710 return (__m128) __builtin_ia32_loadlps ((__v4sf)__A, (__v2si *)__P);
713 /* Stores the lower two SPFP values of A into P. */
715 _mm_storel_pi (__m64 *__P, __m128 __A)
717 __builtin_ia32_storelps ((__v2si *)__P, (__v4sf)__A);
720 /* Creates a 4-bit mask from the most significant bits of the SPFP values. */
722 _mm_movemask_ps (__m128 __A)
724 return __builtin_ia32_movmskps ((__v4sf)__A);
727 /* Return the contents of the control register. */
728 static __inline unsigned int
731 return __builtin_ia32_stmxcsr ();
734 /* Read exception bits from the control register. */
735 static __inline unsigned int
736 _MM_GET_EXCEPTION_STATE (void)
738 return _mm_getcsr() & _MM_EXCEPT_MASK;
741 static __inline unsigned int
742 _MM_GET_EXCEPTION_MASK (void)
744 return _mm_getcsr() & _MM_MASK_MASK;
747 static __inline unsigned int
748 _MM_GET_ROUNDING_MODE (void)
750 return _mm_getcsr() & _MM_ROUND_MASK;
753 static __inline unsigned int
754 _MM_GET_FLUSH_ZERO_MODE (void)
756 return _mm_getcsr() & _MM_FLUSH_ZERO_MASK;
759 /* Set the control register to I. */
761 _mm_setcsr (unsigned int __I)
763 __builtin_ia32_ldmxcsr (__I);
766 /* Set exception bits in the control register. */
768 _MM_SET_EXCEPTION_STATE(unsigned int __mask)
770 _mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | __mask);
774 _MM_SET_EXCEPTION_MASK (unsigned int __mask)
776 _mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | __mask);
780 _MM_SET_ROUNDING_MODE (unsigned int __mode)
782 _mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | __mode);
786 _MM_SET_FLUSH_ZERO_MODE (unsigned int __mode)
788 _mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | __mode);
791 /* Create a vector with element 0 as *P and the rest zero. */
792 static __inline __m128
793 _mm_load_ss (float const *__P)
795 return (__m128) __builtin_ia32_loadss (__P);
798 /* Create a vector with all four elements equal to *P. */
799 static __inline __m128
800 _mm_load1_ps (float const *__P)
802 __v4sf __tmp = __builtin_ia32_loadss (__P);
803 return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,0,0,0));
806 static __inline __m128
807 _mm_load_ps1 (float const *__P)
809 return _mm_load1_ps (__P);
812 /* Load four SPFP values from P. The address must be 16-byte aligned. */
813 static __inline __m128
814 _mm_load_ps (float const *__P)
816 return (__m128) __builtin_ia32_loadaps (__P);
819 /* Load four SPFP values from P. The address need not be 16-byte aligned. */
820 static __inline __m128
821 _mm_loadu_ps (float const *__P)
823 return (__m128) __builtin_ia32_loadups (__P);
826 /* Load four SPFP values in reverse order. The address must be aligned. */
827 static __inline __m128
828 _mm_loadr_ps (float const *__P)
830 __v4sf __tmp = __builtin_ia32_loadaps (__P);
831 return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,1,2,3));
834 /* Create a vector with element 0 as F and the rest zero. */
835 static __inline __m128
836 _mm_set_ss (float __F)
838 return (__m128) __builtin_ia32_loadss (&__F);
841 /* Create a vector with all four elements equal to F. */
842 static __inline __m128
843 _mm_set1_ps (float __F)
845 __v4sf __tmp = __builtin_ia32_loadss (&__F);
846 return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,0,0,0));
849 static __inline __m128
850 _mm_set_ps1 (float __F)
852 return _mm_set1_ps (__F);
855 /* Create the vector [Z Y X W]. */
856 static __inline __m128
857 _mm_set_ps (float __Z, float __Y, float __X, float __W)
872 /* Create the vector [W X Y Z]. */
873 static __inline __m128
874 _mm_setr_ps (float __Z, float __Y, float __X, float __W)
876 return _mm_set_ps (__W, __X, __Y, __Z);
879 /* Create a vector of zeros. */
880 static __inline __m128
881 _mm_setzero_ps (void)
883 return (__m128) __builtin_ia32_setzerops ();
886 /* Stores the lower SPFP value. */
888 _mm_store_ss (float *__P, __m128 __A)
890 __builtin_ia32_storess (__P, (__v4sf)__A);
893 /* Store the lower SPFP value across four words. */
895 _mm_store1_ps (float *__P, __m128 __A)
897 __v4sf __va = (__v4sf)__A;
898 __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,0,0,0));
899 __builtin_ia32_storeaps (__P, __tmp);
903 _mm_store_ps1 (float *__P, __m128 __A)
905 _mm_store1_ps (__P, __A);
908 /* Store four SPFP values. The address must be 16-byte aligned. */
910 _mm_store_ps (float *__P, __m128 __A)
912 __builtin_ia32_storeaps (__P, (__v4sf)__A);
915 /* Store four SPFP values. The address need not be 16-byte aligned. */
917 _mm_storeu_ps (float *__P, __m128 __A)
919 __builtin_ia32_storeups (__P, (__v4sf)__A);
922 /* Store four SPFP values in reverse order. The address must be aligned. */
924 _mm_storer_ps (float *__P, __m128 __A)
926 __v4sf __va = (__v4sf)__A;
927 __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,1,2,3));
928 __builtin_ia32_storeaps (__P, __tmp);
931 /* Sets the low SPFP value of A from the low value of B. */
932 static __inline __m128
933 _mm_move_ss (__m128 __A, __m128 __B)
935 return (__m128) __builtin_ia32_movss ((__v4sf)__A, (__v4sf)__B);
938 /* Extracts one of the four words of A. The selector N must be immediate. */
941 _mm_extract_pi16 (__m64 __A, int __N)
943 return __builtin_ia32_pextrw ((__v4hi)__A, __N);
946 #define _mm_extract_pi16(A, N) \
947 __builtin_ia32_pextrw ((__v4hi)(A), (N))
950 /* Inserts word D into one of four words of A. The selector N must be
953 static __inline __m64
954 _mm_insert_pi16 (__m64 __A, int __D, int __N)
956 return (__m64)__builtin_ia32_pinsrw ((__v4hi)__A, __D, __N);
959 #define _mm_insert_pi16(A, D, N) \
960 ((__m64) __builtin_ia32_pinsrw ((__v4hi)(A), (D), (N)))
963 /* Compute the element-wise maximum of signed 16-bit values. */
964 static __inline __m64
965 _mm_max_pi16 (__m64 __A, __m64 __B)
967 return (__m64) __builtin_ia32_pmaxsw ((__v4hi)__A, (__v4hi)__B);
970 /* Compute the element-wise maximum of unsigned 8-bit values. */
971 static __inline __m64
972 _mm_max_pu8 (__m64 __A, __m64 __B)
974 return (__m64) __builtin_ia32_pmaxub ((__v8qi)__A, (__v8qi)__B);
977 /* Compute the element-wise minimum of signed 16-bit values. */
978 static __inline __m64
979 _mm_min_pi16 (__m64 __A, __m64 __B)
981 return (__m64) __builtin_ia32_pminsw ((__v4hi)__A, (__v4hi)__B);
984 /* Compute the element-wise minimum of unsigned 8-bit values. */
985 static __inline __m64
986 _mm_min_pu8 (__m64 __A, __m64 __B)
988 return (__m64) __builtin_ia32_pminub ((__v8qi)__A, (__v8qi)__B);
991 /* Create an 8-bit mask of the signs of 8-bit values. */
993 _mm_movemask_pi8 (__m64 __A)
995 return __builtin_ia32_pmovmskb ((__v8qi)__A);
998 /* Multiply four unsigned 16-bit values in A by four unsigned 16-bit values
999 in B and produce the high 16 bits of the 32-bit results. */
1000 static __inline __m64
1001 _mm_mulhi_pu16 (__m64 __A, __m64 __B)
1003 return (__m64) __builtin_ia32_pmulhuw ((__v4hi)__A, (__v4hi)__B);
1006 /* Return a combination of the four 16-bit values in A. The selector
1007 must be an immediate. */
1009 static __inline __m64
1010 _mm_shuffle_pi16 (__m64 __A, int __N)
1012 return (__m64) __builtin_ia32_pshufw ((__v4hi)__A, __N);
1015 #define _mm_shuffle_pi16(A, N) \
1016 ((__m64) __builtin_ia32_pshufw ((__v4hi)(A), (N)))
1019 /* Conditionally store byte elements of A into P. The high bit of each
1020 byte in the selector N determines whether the corresponding byte from
1022 static __inline void
1023 _mm_maskmove_si64 (__m64 __A, __m64 __N, char *__P)
1025 __builtin_ia32_maskmovq ((__v8qi)__A, (__v8qi)__N, __P);
1028 /* Compute the rounded averages of the unsigned 8-bit values in A and B. */
1029 static __inline __m64
1030 _mm_avg_pu8 (__m64 __A, __m64 __B)
1032 return (__m64) __builtin_ia32_pavgb ((__v8qi)__A, (__v8qi)__B);
1035 /* Compute the rounded averages of the unsigned 16-bit values in A and B. */
1036 static __inline __m64
1037 _mm_avg_pu16 (__m64 __A, __m64 __B)
1039 return (__m64) __builtin_ia32_pavgw ((__v4hi)__A, (__v4hi)__B);
1042 /* Compute the sum of the absolute differences of the unsigned 8-bit
1043 values in A and B. Return the value in the lower 16-bit word; the
1044 upper words are cleared. */
1045 static __inline __m64
1046 _mm_sad_pu8 (__m64 __A, __m64 __B)
1048 return (__m64) __builtin_ia32_psadbw ((__v8qi)__A, (__v8qi)__B);
1051 /* Loads one cache line from address P to a location "closer" to the
1052 processor. The selector I specifies the type of prefetch operation. */
1054 static __inline void
1055 _mm_prefetch (void *__P, enum _mm_hint __I)
1057 __builtin_prefetch (__P, 0, __I);
1060 #define _mm_prefetch(P, I) \
1061 __builtin_prefetch ((P), 0, (I))
1064 /* Stores the data in A to the address P without polluting the caches. */
1065 static __inline void
1066 _mm_stream_pi (__m64 *__P, __m64 __A)
1068 __builtin_ia32_movntq ((unsigned long long *)__P, (unsigned long long)__A);
1071 /* Likewise. The address must be 16-byte aligned. */
1072 static __inline void
1073 _mm_stream_ps (float *__P, __m128 __A)
1075 __builtin_ia32_movntps (__P, (__v4sf)__A);
1078 /* Guarantees that every preceding store is globally visible before
1079 any subsequent store. */
1080 static __inline void
1083 __builtin_ia32_sfence ();
1086 /* The execution of the next instruction is delayed by an implementation
1087 specific amount of time. The instruction does not modify the
1088 architectural state. */
1089 static __inline void
1092 __asm__ __volatile__ ("rep; nop" : : );
1095 /* Transpose the 4x4 matrix composed of row[0-3]. */
1096 #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
1098 __v4sf __r0 = (row0), __r1 = (row1), __r2 = (row2), __r3 = (row3); \
1099 __v4sf __t0 = __builtin_ia32_shufps (__r0, __r1, 0x44); \
1100 __v4sf __t2 = __builtin_ia32_shufps (__r0, __r1, 0xEE); \
1101 __v4sf __t1 = __builtin_ia32_shufps (__r2, __r3, 0x44); \
1102 __v4sf __t3 = __builtin_ia32_shufps (__r2, __r3, 0xEE); \
1103 (row0) = __builtin_ia32_shufps (__t0, __t1, 0x88); \
1104 (row1) = __builtin_ia32_shufps (__t0, __t1, 0xDD); \
1105 (row2) = __builtin_ia32_shufps (__t2, __t3, 0x88); \
1106 (row3) = __builtin_ia32_shufps (__t2, __t3, 0xDD); \
1111 typedef int __v2df __attribute__ ((mode (V2DF)));
1112 typedef int __v2di __attribute__ ((mode (V2DI)));
1113 typedef int __v4si __attribute__ ((mode (V4SI)));
1114 typedef int __v8hi __attribute__ ((mode (V8HI)));
1115 typedef int __v16qi __attribute__ ((mode (V16QI)));
1117 /* Create a selector for use with the SHUFPD instruction. */
1118 #define _MM_SHUFFLE2(fp1,fp0) \
1119 (((fp1) << 1) | (fp0))
1121 #define __m128i __v2di
1122 #define __m128d __v2df
1124 /* Create a vector with element 0 as *P and the rest zero. */
1125 static __inline __m128d
1126 _mm_load_sd (double const *__P)
1128 return (__m128d) __builtin_ia32_loadsd (__P);
1131 /* Create a vector with all two elements equal to *P. */
1132 static __inline __m128d
1133 _mm_load1_pd (double const *__P)
1135 __v2df __tmp = __builtin_ia32_loadsd (__P);
1136 return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,0));
1139 static __inline __m128d
1140 _mm_load_pd1 (double const *__P)
1142 return _mm_load1_pd (__P);
1145 /* Load two DPFP values from P. The address must be 16-byte aligned. */
1146 static __inline __m128d
1147 _mm_load_pd (double const *__P)
1149 return (__m128d) __builtin_ia32_loadapd (__P);
1152 /* Load two DPFP values from P. The address need not be 16-byte aligned. */
1153 static __inline __m128d
1154 _mm_loadu_pd (double const *__P)
1156 return (__m128d) __builtin_ia32_loadupd (__P);
1159 /* Load two DPFP values in reverse order. The address must be aligned. */
1160 static __inline __m128d
1161 _mm_loadr_pd (double const *__P)
1163 __v2df __tmp = __builtin_ia32_loadapd (__P);
1164 return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,1));
1167 /* Create a vector with element 0 as F and the rest zero. */
1168 static __inline __m128d
1169 _mm_set_sd (double __F)
1171 return (__m128d) __builtin_ia32_loadsd (&__F);
1174 /* Create a vector with all two elements equal to F. */
1175 static __inline __m128d
1176 _mm_set1_pd (double __F)
1178 __v2df __tmp = __builtin_ia32_loadsd (&__F);
1179 return (__m128d) __builtin_ia32_shufpd (__tmp, __tmp, _MM_SHUFFLE2 (0,0));
1182 static __inline __m128d
1183 _mm_set_pd1 (double __F)
1185 return _mm_set1_pd (__F);
1188 /* Create the vector [Z Y]. */
1189 static __inline __m128d
1190 _mm_set_pd (double __Z, double __Y)
1203 /* Create the vector [Y Z]. */
1204 static __inline __m128d
1205 _mm_setr_pd (double __Z, double __Y)
1207 return _mm_set_pd (__Y, __Z);
1210 /* Create a vector of zeros. */
1211 static __inline __m128d
1212 _mm_setzero_pd (void)
1214 return (__m128d) __builtin_ia32_setzeropd ();
1217 /* Stores the lower DPFP value. */
1218 static __inline void
1219 _mm_store_sd (double *__P, __m128d __A)
1221 __builtin_ia32_storesd (__P, (__v2df)__A);
1224 /* Store the lower DPFP value acrosd two words. */
1225 static __inline void
1226 _mm_store1_pd (double *__P, __m128d __A)
1228 __v2df __va = (__v2df)__A;
1229 __v2df __tmp = __builtin_ia32_shufpd (__va, __va, _MM_SHUFFLE2 (0,0));
1230 __builtin_ia32_storeapd (__P, __tmp);
1233 static __inline void
1234 _mm_store_pd1 (double *__P, __m128d __A)
1236 _mm_store1_pd (__P, __A);
1239 /* Store two DPFP values. The address must be 16-byte aligned. */
1240 static __inline void
1241 _mm_store_pd (double *__P, __m128d __A)
1243 __builtin_ia32_storeapd (__P, (__v2df)__A);
1246 /* Store two DPFP values. The address need not be 16-byte aligned. */
1247 static __inline void
1248 _mm_storeu_pd (double *__P, __m128d __A)
1250 __builtin_ia32_storeupd (__P, (__v2df)__A);
1253 /* Store two DPFP values in reverse order. The address must be aligned. */
1254 static __inline void
1255 _mm_storer_pd (double *__P, __m128d __A)
1257 __v2df __va = (__v2df)__A;
1258 __v2df __tmp = __builtin_ia32_shufpd (__va, __va, _MM_SHUFFLE2 (0,1));
1259 __builtin_ia32_storeapd (__P, __tmp);
1262 /* Sets the low DPFP value of A from the low value of B. */
1263 static __inline __m128d
1264 _mm_move_sd (__m128d __A, __m128d __B)
1266 return (__m128d) __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
1270 static __inline __m128d
1271 _mm_add_pd (__m128d __A, __m128d __B)
1273 return (__m128d)__builtin_ia32_addpd ((__v2df)__A, (__v2df)__B);
1276 static __inline __m128d
1277 _mm_add_sd (__m128d __A, __m128d __B)
1279 return (__m128d)__builtin_ia32_addsd ((__v2df)__A, (__v2df)__B);
1282 static __inline __m128d
1283 _mm_sub_pd (__m128d __A, __m128d __B)
1285 return (__m128d)__builtin_ia32_subpd ((__v2df)__A, (__v2df)__B);
1288 static __inline __m128d
1289 _mm_sub_sd (__m128d __A, __m128d __B)
1291 return (__m128d)__builtin_ia32_subsd ((__v2df)__A, (__v2df)__B);
1294 static __inline __m128d
1295 _mm_mul_pd (__m128d __A, __m128d __B)
1297 return (__m128d)__builtin_ia32_mulpd ((__v2df)__A, (__v2df)__B);
1300 static __inline __m128d
1301 _mm_mul_sd (__m128d __A, __m128d __B)
1303 return (__m128d)__builtin_ia32_mulsd ((__v2df)__A, (__v2df)__B);
1306 static __inline __m128d
1307 _mm_div_pd (__m128d __A, __m128d __B)
1309 return (__m128d)__builtin_ia32_divpd ((__v2df)__A, (__v2df)__B);
1312 static __inline __m128d
1313 _mm_div_sd (__m128d __A, __m128d __B)
1315 return (__m128d)__builtin_ia32_divsd ((__v2df)__A, (__v2df)__B);
1318 static __inline __m128d
1319 _mm_sqrt_pd (__m128d __A)
1321 return (__m128d)__builtin_ia32_sqrtpd ((__v2df)__A);
1324 /* Return pair {sqrt (A[0), B[1]}. */
1325 static __inline __m128d
1326 _mm_sqrt_sd (__m128d __A, __m128d __B)
1328 __v2df __tmp = __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
1329 return (__m128d)__builtin_ia32_sqrtsd ((__v2df)__tmp);
1332 static __inline __m128d
1333 _mm_min_pd (__m128d __A, __m128d __B)
1335 return (__m128d)__builtin_ia32_minpd ((__v2df)__A, (__v2df)__B);
1338 static __inline __m128d
1339 _mm_min_sd (__m128d __A, __m128d __B)
1341 return (__m128d)__builtin_ia32_minsd ((__v2df)__A, (__v2df)__B);
1344 static __inline __m128d
1345 _mm_max_pd (__m128d __A, __m128d __B)
1347 return (__m128d)__builtin_ia32_maxpd ((__v2df)__A, (__v2df)__B);
1350 static __inline __m128d
1351 _mm_max_sd (__m128d __A, __m128d __B)
1353 return (__m128d)__builtin_ia32_maxsd ((__v2df)__A, (__v2df)__B);
1356 static __inline __m128d
1357 _mm_and_pd (__m128d __A, __m128d __B)
1359 return (__m128d)__builtin_ia32_andpd ((__v2df)__A, (__v2df)__B);
1362 static __inline __m128d
1363 _mm_andnot_pd (__m128d __A, __m128d __B)
1365 return (__m128d)__builtin_ia32_andnpd ((__v2df)__A, (__v2df)__B);
1368 static __inline __m128d
1369 _mm_or_pd (__m128d __A, __m128d __B)
1371 return (__m128d)__builtin_ia32_orpd ((__v2df)__A, (__v2df)__B);
1374 static __inline __m128d
1375 _mm_xor_pd (__m128d __A, __m128d __B)
1377 return (__m128d)__builtin_ia32_xorpd ((__v2df)__A, (__v2df)__B);
1380 static __inline __m128d
1381 _mm_cmpeq_pd (__m128d __A, __m128d __B)
1383 return (__m128d)__builtin_ia32_cmpeqpd ((__v2df)__A, (__v2df)__B);
1386 static __inline __m128d
1387 _mm_cmplt_pd (__m128d __A, __m128d __B)
1389 return (__m128d)__builtin_ia32_cmpltpd ((__v2df)__A, (__v2df)__B);
1392 static __inline __m128d
1393 _mm_cmple_pd (__m128d __A, __m128d __B)
1395 return (__m128d)__builtin_ia32_cmplepd ((__v2df)__A, (__v2df)__B);
1398 static __inline __m128d
1399 _mm_cmpgt_pd (__m128d __A, __m128d __B)
1401 return (__m128d)__builtin_ia32_cmpgtpd ((__v2df)__A, (__v2df)__B);
1404 static __inline __m128d
1405 _mm_cmpge_pd (__m128d __A, __m128d __B)
1407 return (__m128d)__builtin_ia32_cmpgepd ((__v2df)__A, (__v2df)__B);
1410 static __inline __m128d
1411 _mm_cmpneq_pd (__m128d __A, __m128d __B)
1413 return (__m128d)__builtin_ia32_cmpneqpd ((__v2df)__A, (__v2df)__B);
1416 static __inline __m128d
1417 _mm_cmpnlt_pd (__m128d __A, __m128d __B)
1419 return (__m128d)__builtin_ia32_cmpnltpd ((__v2df)__A, (__v2df)__B);
1422 static __inline __m128d
1423 _mm_cmpnle_pd (__m128d __A, __m128d __B)
1425 return (__m128d)__builtin_ia32_cmpnlepd ((__v2df)__A, (__v2df)__B);
1428 static __inline __m128d
1429 _mm_cmpngt_pd (__m128d __A, __m128d __B)
1431 return (__m128d)__builtin_ia32_cmpngtpd ((__v2df)__A, (__v2df)__B);
1434 static __inline __m128d
1435 _mm_cmpnge_pd (__m128d __A, __m128d __B)
1437 return (__m128d)__builtin_ia32_cmpngepd ((__v2df)__A, (__v2df)__B);
1440 static __inline __m128d
1441 _mm_cmpord_pd (__m128d __A, __m128d __B)
1443 return (__m128d)__builtin_ia32_cmpordpd ((__v2df)__A, (__v2df)__B);
1446 static __inline __m128d
1447 _mm_cmpunord_pd (__m128d __A, __m128d __B)
1449 return (__m128d)__builtin_ia32_cmpunordpd ((__v2df)__A, (__v2df)__B);
1452 static __inline __m128d
1453 _mm_cmpeq_sd (__m128d __A, __m128d __B)
1455 return (__m128d)__builtin_ia32_cmpeqsd ((__v2df)__A, (__v2df)__B);
1458 static __inline __m128d
1459 _mm_cmplt_sd (__m128d __A, __m128d __B)
1461 return (__m128d)__builtin_ia32_cmpltsd ((__v2df)__A, (__v2df)__B);
1464 static __inline __m128d
1465 _mm_cmple_sd (__m128d __A, __m128d __B)
1467 return (__m128d)__builtin_ia32_cmplesd ((__v2df)__A, (__v2df)__B);
1470 static __inline __m128d
1471 _mm_cmpgt_sd (__m128d __A, __m128d __B)
1473 return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
1475 __builtin_ia32_cmpltsd ((__v2df) __B,
1480 static __inline __m128d
1481 _mm_cmpge_sd (__m128d __A, __m128d __B)
1483 return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
1485 __builtin_ia32_cmplesd ((__v2df) __B,
1490 static __inline __m128d
1491 _mm_cmpneq_sd (__m128d __A, __m128d __B)
1493 return (__m128d)__builtin_ia32_cmpneqsd ((__v2df)__A, (__v2df)__B);
1496 static __inline __m128d
1497 _mm_cmpnlt_sd (__m128d __A, __m128d __B)
1499 return (__m128d)__builtin_ia32_cmpnltsd ((__v2df)__A, (__v2df)__B);
1502 static __inline __m128d
1503 _mm_cmpnle_sd (__m128d __A, __m128d __B)
1505 return (__m128d)__builtin_ia32_cmpnlesd ((__v2df)__A, (__v2df)__B);
1508 static __inline __m128d
1509 _mm_cmpngt_sd (__m128d __A, __m128d __B)
1511 return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
1513 __builtin_ia32_cmpnltsd ((__v2df) __B,
1518 static __inline __m128d
1519 _mm_cmpnge_sd (__m128d __A, __m128d __B)
1521 return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
1523 __builtin_ia32_cmpnlesd ((__v2df) __B,
1528 static __inline __m128d
1529 _mm_cmpord_sd (__m128d __A, __m128d __B)
1531 return (__m128d)__builtin_ia32_cmpordsd ((__v2df)__A, (__v2df)__B);
1534 static __inline __m128d
1535 _mm_cmpunord_sd (__m128d __A, __m128d __B)
1537 return (__m128d)__builtin_ia32_cmpunordsd ((__v2df)__A, (__v2df)__B);
1541 _mm_comieq_sd (__m128d __A, __m128d __B)
1543 return __builtin_ia32_comisdeq ((__v2df)__A, (__v2df)__B);
1547 _mm_comilt_sd (__m128d __A, __m128d __B)
1549 return __builtin_ia32_comisdlt ((__v2df)__A, (__v2df)__B);
1553 _mm_comile_sd (__m128d __A, __m128d __B)
1555 return __builtin_ia32_comisdle ((__v2df)__A, (__v2df)__B);
1559 _mm_comigt_sd (__m128d __A, __m128d __B)
1561 return __builtin_ia32_comisdgt ((__v2df)__A, (__v2df)__B);
1565 _mm_comige_sd (__m128d __A, __m128d __B)
1567 return __builtin_ia32_comisdge ((__v2df)__A, (__v2df)__B);
1571 _mm_comineq_sd (__m128d __A, __m128d __B)
1573 return __builtin_ia32_comisdneq ((__v2df)__A, (__v2df)__B);
1577 _mm_ucomieq_sd (__m128d __A, __m128d __B)
1579 return __builtin_ia32_ucomisdeq ((__v2df)__A, (__v2df)__B);
1583 _mm_ucomilt_sd (__m128d __A, __m128d __B)
1585 return __builtin_ia32_ucomisdlt ((__v2df)__A, (__v2df)__B);
1589 _mm_ucomile_sd (__m128d __A, __m128d __B)
1591 return __builtin_ia32_ucomisdle ((__v2df)__A, (__v2df)__B);
1595 _mm_ucomigt_sd (__m128d __A, __m128d __B)
1597 return __builtin_ia32_ucomisdgt ((__v2df)__A, (__v2df)__B);
1601 _mm_ucomige_sd (__m128d __A, __m128d __B)
1603 return __builtin_ia32_ucomisdge ((__v2df)__A, (__v2df)__B);
1607 _mm_ucomineq_sd (__m128d __A, __m128d __B)
1609 return __builtin_ia32_ucomisdneq ((__v2df)__A, (__v2df)__B);
1612 /* Create a vector with element 0 as *P and the rest zero. */
1614 static __inline __m128i
1615 _mm_load_si128 (__m128i const *__P)
1617 return (__m128i) __builtin_ia32_loaddqa ((char const *)__P);
1620 static __inline __m128i
1621 _mm_loadu_si128 (__m128i const *__P)
1623 return (__m128i) __builtin_ia32_loaddqu ((char const *)__P);
1626 static __inline __m128i
1627 _mm_loadl_epi64 (__m128i const *__P)
1629 return (__m128i) __builtin_ia32_movq2dq (*(unsigned long long *)__P);
1632 static __inline void
1633 _mm_store_si128 (__m128i *__P, __m128i __B)
1635 __builtin_ia32_storedqa ((char *)__P, (__v16qi)__B);
1638 static __inline void
1639 _mm_storeu_si128 (__m128i *__P, __m128i __B)
1641 __builtin_ia32_storedqu ((char *)__P, (__v16qi)__B);
1644 static __inline void
1645 _mm_storel_epi64 (__m128i *__P, __m128i __B)
1647 *(long long *)__P = __builtin_ia32_movdq2q ((__v2di)__B);
1650 static __inline __m64
1651 _mm_movepi64_pi64 (__m128i __B)
1653 return (__m64) __builtin_ia32_movdq2q ((__v2di)__B);
1656 static __inline __m128i
1657 _mm_move_epi64 (__m128i __A)
1659 return (__m128i) __builtin_ia32_movq ((__v2di)__A);
1662 /* Create a vector of zeros. */
1663 static __inline __m128i
1664 _mm_setzero_si128 (void)
1666 return (__m128i) __builtin_ia32_setzero128 ();
1669 static __inline __m128i
1670 _mm_set_epi64 (__m64 __A, __m64 __B)
1672 __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
1673 __v2di __tmp2 = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__B);
1674 return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp2, __tmp);
1677 /* Create the vector [Z Y X W]. */
1678 static __inline __m128i
1679 _mm_set_epi32 (int __Z, int __Y, int __X, int __W)
1695 /* Create the vector [Z Y]. */
1696 static __inline __m128i
1697 _mm_set_epi64x (long long __Z, long long __Y)
1711 /* Create the vector [S T U V Z Y X W]. */
1712 static __inline __m128i
1713 _mm_set_epi16 (short __Z, short __Y, short __X, short __W,
1714 short __V, short __U, short __T, short __S)
1733 /* Create the vector [S T U V Z Y X W]. */
1734 static __inline __m128i
1735 _mm_set_epi8 (char __Z, char __Y, char __X, char __W,
1736 char __V, char __U, char __T, char __S,
1737 char __Z1, char __Y1, char __X1, char __W1,
1738 char __V1, char __U1, char __T1, char __S1)
1765 static __inline __m128i
1766 _mm_set1_epi64 (__m64 __A)
1768 __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
1769 return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp, __tmp);
1772 static __inline __m128i
1773 _mm_set1_epi32 (int __A)
1775 __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__A);
1776 return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
1780 static __inline __m128i
1781 _mm_set1_epi64x (long long __A)
1783 __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
1784 return (__m128i) __builtin_ia32_shufpd ((__v2df)__tmp, (__v2df)__tmp, _MM_SHUFFLE2 (0,0));
1788 static __inline __m128i
1789 _mm_set1_epi16 (short __A)
1791 int __Acopy = (unsigned short)__A;
1792 __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__Acopy);
1793 __tmp = (__v4si)__builtin_ia32_punpcklwd128 ((__v8hi)__tmp, (__v8hi)__tmp);
1794 return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
1797 static __inline __m128i
1798 _mm_set1_epi8 (char __A)
1800 int __Acopy = (unsigned char)__A;
1801 __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__Acopy);
1802 __tmp = (__v4si)__builtin_ia32_punpcklbw128 ((__v16qi)__tmp, (__v16qi)__tmp);
1803 __tmp = (__v4si)__builtin_ia32_punpcklbw128 ((__v16qi)__tmp, (__v16qi)__tmp);
1804 return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
1807 static __inline __m128i
1808 _mm_setr_epi64 (__m64 __A, __m64 __B)
1810 __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
1811 __v2di __tmp2 = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__B);
1812 return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp, __tmp2);
1815 /* Create the vector [Z Y X W]. */
1816 static __inline __m128i
1817 _mm_setr_epi32 (int __W, int __X, int __Y, int __Z)
1831 /* Create the vector [S T U V Z Y X W]. */
1832 static __inline __m128i
1833 _mm_setr_epi16 (short __S, short __T, short __U, short __V,
1834 short __W, short __X, short __Y, short __Z)
1853 /* Create the vector [S T U V Z Y X W]. */
1854 static __inline __m128i
1855 _mm_setr_epi8 (char __S1, char __T1, char __U1, char __V1,
1856 char __W1, char __X1, char __Y1, char __Z1,
1857 char __S, char __T, char __U, char __V,
1858 char __W, char __X, char __Y, char __Z)
1885 static __inline __m128d
1886 _mm_cvtepi32_pd (__m128i __A)
1888 return (__m128d)__builtin_ia32_cvtdq2pd ((__v4si) __A);
1891 static __inline __m128
1892 _mm_cvtepi32_ps (__m128i __A)
1894 return (__m128)__builtin_ia32_cvtdq2ps ((__v4si) __A);
1897 static __inline __m128i
1898 _mm_cvtpd_epi32 (__m128d __A)
1900 return (__m128i)__builtin_ia32_cvtpd2dq ((__v2df) __A);
1903 static __inline __m64
1904 _mm_cvtpd_pi32 (__m128d __A)
1906 return (__m64)__builtin_ia32_cvtpd2pi ((__v2df) __A);
1909 static __inline __m128
1910 _mm_cvtpd_ps (__m128d __A)
1912 return (__m128)__builtin_ia32_cvtpd2ps ((__v2df) __A);
1915 static __inline __m128i
1916 _mm_cvttpd_epi32 (__m128d __A)
1918 return (__m128i)__builtin_ia32_cvttpd2dq ((__v2df) __A);
1921 static __inline __m64
1922 _mm_cvttpd_pi32 (__m128d __A)
1924 return (__m64)__builtin_ia32_cvttpd2pi ((__v2df) __A);
1927 static __inline __m128d
1928 _mm_cvtpi32_pd (__m64 __A)
1930 return (__m128d)__builtin_ia32_cvtpi2pd ((__v2si) __A);
1933 static __inline __m128i
1934 _mm_cvtps_epi32 (__m128 __A)
1936 return (__m128i)__builtin_ia32_cvtps2dq ((__v4sf) __A);
1939 static __inline __m128i
1940 _mm_cvttps_epi32 (__m128 __A)
1942 return (__m128i)__builtin_ia32_cvttps2dq ((__v4sf) __A);
1945 static __inline __m128d
1946 _mm_cvtps_pd (__m128 __A)
1948 return (__m128d)__builtin_ia32_cvtps2pd ((__v4sf) __A);
1952 _mm_cvtsd_si32 (__m128d __A)
1954 return __builtin_ia32_cvtsd2si ((__v2df) __A);
1958 static __inline long long
1959 _mm_cvtsd_si64x (__m128d __A)
1961 return __builtin_ia32_cvtsd2si64 ((__v2df) __A);
1966 _mm_cvttsd_si32 (__m128d __A)
1968 return __builtin_ia32_cvttsd2si ((__v2df) __A);
1972 static __inline long long
1973 _mm_cvttsd_si64x (__m128d __A)
1975 return __builtin_ia32_cvttsd2si64 ((__v2df) __A);
1979 static __inline __m128
1980 _mm_cvtsd_ss (__m128 __A, __m128d __B)
1982 return (__m128)__builtin_ia32_cvtsd2ss ((__v4sf) __A, (__v2df) __B);
1985 static __inline __m128d
1986 _mm_cvtsi32_sd (__m128d __A, int __B)
1988 return (__m128d)__builtin_ia32_cvtsi2sd ((__v2df) __A, __B);
1992 static __inline __m128d
1993 _mm_cvtsi64x_sd (__m128d __A, long long __B)
1995 return (__m128d)__builtin_ia32_cvtsi642sd ((__v2df) __A, __B);
1999 static __inline __m128d
2000 _mm_cvtss_sd (__m128d __A, __m128 __B)
2002 return (__m128d)__builtin_ia32_cvtss2sd ((__v2df) __A, (__v4sf)__B);
2005 #define _mm_shuffle_pd(__A, __B, __C) ((__m128d)__builtin_ia32_shufpd ((__v2df)__A, (__v2df)__B, (__C)))
2007 static __inline __m128d
2008 _mm_unpackhi_pd (__m128d __A, __m128d __B)
2010 return (__m128d)__builtin_ia32_unpckhpd ((__v2df)__A, (__v2df)__B);
2013 static __inline __m128d
2014 _mm_unpacklo_pd (__m128d __A, __m128d __B)
2016 return (__m128d)__builtin_ia32_unpcklpd ((__v2df)__A, (__v2df)__B);
2019 static __inline __m128d
2020 _mm_loadh_pd (__m128d __A, double const *__B)
2022 return (__m128d)__builtin_ia32_loadhpd ((__v2df)__A, (__v2si *)__B);
2025 static __inline void
2026 _mm_storeh_pd (double *__A, __m128d __B)
2028 __builtin_ia32_storehpd ((__v2si *)__A, (__v2df)__B);
2031 static __inline __m128d
2032 _mm_loadl_pd (__m128d __A, double const *__B)
2034 return (__m128d)__builtin_ia32_loadlpd ((__v2df)__A, (__v2si *)__B);
2037 static __inline void
2038 _mm_storel_pd (double *__A, __m128d __B)
2040 __builtin_ia32_storelpd ((__v2si *)__A, (__v2df)__B);
2044 _mm_movemask_pd (__m128d __A)
2046 return __builtin_ia32_movmskpd ((__v2df)__A);
2049 static __inline __m128i
2050 _mm_packs_epi16 (__m128i __A, __m128i __B)
2052 return (__m128i)__builtin_ia32_packsswb128 ((__v8hi)__A, (__v8hi)__B);
2055 static __inline __m128i
2056 _mm_packs_epi32 (__m128i __A, __m128i __B)
2058 return (__m128i)__builtin_ia32_packssdw128 ((__v4si)__A, (__v4si)__B);
2061 static __inline __m128i
2062 _mm_packus_epi16 (__m128i __A, __m128i __B)
2064 return (__m128i)__builtin_ia32_packuswb128 ((__v8hi)__A, (__v8hi)__B);
2067 static __inline __m128i
2068 _mm_unpackhi_epi8 (__m128i __A, __m128i __B)
2070 return (__m128i)__builtin_ia32_punpckhbw128 ((__v16qi)__A, (__v16qi)__B);
2073 static __inline __m128i
2074 _mm_unpackhi_epi16 (__m128i __A, __m128i __B)
2076 return (__m128i)__builtin_ia32_punpckhwd128 ((__v8hi)__A, (__v8hi)__B);
2079 static __inline __m128i
2080 _mm_unpackhi_epi32 (__m128i __A, __m128i __B)
2082 return (__m128i)__builtin_ia32_punpckhdq128 ((__v4si)__A, (__v4si)__B);
2085 static __inline __m128i
2086 _mm_unpackhi_epi64 (__m128i __A, __m128i __B)
2088 return (__m128i)__builtin_ia32_punpckhqdq128 ((__v2di)__A, (__v2di)__B);
2091 static __inline __m128i
2092 _mm_unpacklo_epi8 (__m128i __A, __m128i __B)
2094 return (__m128i)__builtin_ia32_punpcklbw128 ((__v16qi)__A, (__v16qi)__B);
2097 static __inline __m128i
2098 _mm_unpacklo_epi16 (__m128i __A, __m128i __B)
2100 return (__m128i)__builtin_ia32_punpcklwd128 ((__v8hi)__A, (__v8hi)__B);
2103 static __inline __m128i
2104 _mm_unpacklo_epi32 (__m128i __A, __m128i __B)
2106 return (__m128i)__builtin_ia32_punpckldq128 ((__v4si)__A, (__v4si)__B);
2109 static __inline __m128i
2110 _mm_unpacklo_epi64 (__m128i __A, __m128i __B)
2112 return (__m128i)__builtin_ia32_punpcklqdq128 ((__v2di)__A, (__v2di)__B);
2115 static __inline __m128i
2116 _mm_add_epi8 (__m128i __A, __m128i __B)
2118 return (__m128i)__builtin_ia32_paddb128 ((__v16qi)__A, (__v16qi)__B);
2121 static __inline __m128i
2122 _mm_add_epi16 (__m128i __A, __m128i __B)
2124 return (__m128i)__builtin_ia32_paddw128 ((__v8hi)__A, (__v8hi)__B);
2127 static __inline __m128i
2128 _mm_add_epi32 (__m128i __A, __m128i __B)
2130 return (__m128i)__builtin_ia32_paddd128 ((__v4si)__A, (__v4si)__B);
2133 static __inline __m128i
2134 _mm_add_epi64 (__m128i __A, __m128i __B)
2136 return (__m128i)__builtin_ia32_paddq128 ((__v2di)__A, (__v2di)__B);
2139 static __inline __m128i
2140 _mm_adds_epi8 (__m128i __A, __m128i __B)
2142 return (__m128i)__builtin_ia32_paddsb128 ((__v16qi)__A, (__v16qi)__B);
2145 static __inline __m128i
2146 _mm_adds_epi16 (__m128i __A, __m128i __B)
2148 return (__m128i)__builtin_ia32_paddsw128 ((__v8hi)__A, (__v8hi)__B);
2151 static __inline __m128i
2152 _mm_adds_epu8 (__m128i __A, __m128i __B)
2154 return (__m128i)__builtin_ia32_paddusb128 ((__v16qi)__A, (__v16qi)__B);
2157 static __inline __m128i
2158 _mm_adds_epu16 (__m128i __A, __m128i __B)
2160 return (__m128i)__builtin_ia32_paddusw128 ((__v8hi)__A, (__v8hi)__B);
2163 static __inline __m128i
2164 _mm_sub_epi8 (__m128i __A, __m128i __B)
2166 return (__m128i)__builtin_ia32_psubb128 ((__v16qi)__A, (__v16qi)__B);
2169 static __inline __m128i
2170 _mm_sub_epi16 (__m128i __A, __m128i __B)
2172 return (__m128i)__builtin_ia32_psubw128 ((__v8hi)__A, (__v8hi)__B);
2175 static __inline __m128i
2176 _mm_sub_epi32 (__m128i __A, __m128i __B)
2178 return (__m128i)__builtin_ia32_psubd128 ((__v4si)__A, (__v4si)__B);
2181 static __inline __m128i
2182 _mm_sub_epi64 (__m128i __A, __m128i __B)
2184 return (__m128i)__builtin_ia32_psubq128 ((__v2di)__A, (__v2di)__B);
2187 static __inline __m128i
2188 _mm_subs_epi8 (__m128i __A, __m128i __B)
2190 return (__m128i)__builtin_ia32_psubsb128 ((__v16qi)__A, (__v16qi)__B);
2193 static __inline __m128i
2194 _mm_subs_epi16 (__m128i __A, __m128i __B)
2196 return (__m128i)__builtin_ia32_psubsw128 ((__v8hi)__A, (__v8hi)__B);
2199 static __inline __m128i
2200 _mm_subs_epu8 (__m128i __A, __m128i __B)
2202 return (__m128i)__builtin_ia32_psubusb128 ((__v16qi)__A, (__v16qi)__B);
2205 static __inline __m128i
2206 _mm_subs_epu16 (__m128i __A, __m128i __B)
2208 return (__m128i)__builtin_ia32_psubusw128 ((__v8hi)__A, (__v8hi)__B);
2211 static __inline __m128i
2212 _mm_madd_epi16 (__m128i __A, __m128i __B)
2214 return (__m128i)__builtin_ia32_pmaddwd128 ((__v8hi)__A, (__v8hi)__B);
2217 static __inline __m128i
2218 _mm_mulhi_epi16 (__m128i __A, __m128i __B)
2220 return (__m128i)__builtin_ia32_pmulhw128 ((__v8hi)__A, (__v8hi)__B);
2223 static __inline __m128i
2224 _mm_mullo_epi16 (__m128i __A, __m128i __B)
2226 return (__m128i)__builtin_ia32_pmullw128 ((__v8hi)__A, (__v8hi)__B);
2229 static __inline __m64
2230 _mm_mul_su32 (__m64 __A, __m64 __B)
2232 return (__m64)__builtin_ia32_pmuludq ((__v2si)__A, (__v2si)__B);
2235 static __inline __m128i
2236 _mm_mul_epu32 (__m128i __A, __m128i __B)
2238 return (__m128i)__builtin_ia32_pmuludq128 ((__v4si)__A, (__v4si)__B);
2241 static __inline __m128i
2242 _mm_sll_epi16 (__m128i __A, __m128i __B)
2244 return (__m128i)__builtin_ia32_psllw128 ((__v8hi)__A, (__v2di)__B);
2247 static __inline __m128i
2248 _mm_sll_epi32 (__m128i __A, __m128i __B)
2250 return (__m128i)__builtin_ia32_pslld128 ((__v4si)__A, (__v2di)__B);
2253 static __inline __m128i
2254 _mm_sll_epi64 (__m128i __A, __m128i __B)
2256 return (__m128i)__builtin_ia32_psllq128 ((__v2di)__A, (__v2di)__B);
2259 static __inline __m128i
2260 _mm_sra_epi16 (__m128i __A, __m128i __B)
2262 return (__m128i)__builtin_ia32_psraw128 ((__v8hi)__A, (__v2di)__B);
2265 static __inline __m128i
2266 _mm_sra_epi32 (__m128i __A, __m128i __B)
2268 return (__m128i)__builtin_ia32_psrad128 ((__v4si)__A, (__v2di)__B);
2271 static __inline __m128i
2272 _mm_srl_epi16 (__m128i __A, __m128i __B)
2274 return (__m128i)__builtin_ia32_psrlw128 ((__v8hi)__A, (__v2di)__B);
2277 static __inline __m128i
2278 _mm_srl_epi32 (__m128i __A, __m128i __B)
2280 return (__m128i)__builtin_ia32_psrld128 ((__v4si)__A, (__v2di)__B);
2283 static __inline __m128i
2284 _mm_srl_epi64 (__m128i __A, __m128i __B)
2286 return (__m128i)__builtin_ia32_psrlq128 ((__v2di)__A, (__v2di)__B);
2289 static __inline __m128i
2290 _mm_slli_epi16 (__m128i __A, int __B)
2292 return (__m128i)__builtin_ia32_psllwi128 ((__v8hi)__A, __B);
2295 static __inline __m128i
2296 _mm_slli_epi32 (__m128i __A, int __B)
2298 return (__m128i)__builtin_ia32_pslldi128 ((__v4si)__A, __B);
2301 static __inline __m128i
2302 _mm_slli_epi64 (__m128i __A, int __B)
2304 return (__m128i)__builtin_ia32_psllqi128 ((__v2di)__A, __B);
2307 static __inline __m128i
2308 _mm_srai_epi16 (__m128i __A, int __B)
2310 return (__m128i)__builtin_ia32_psrawi128 ((__v8hi)__A, __B);
2313 static __inline __m128i
2314 _mm_srai_epi32 (__m128i __A, int __B)
2316 return (__m128i)__builtin_ia32_psradi128 ((__v4si)__A, __B);
2320 static __m128i __attribute__((__always_inline__))
2321 _mm_srli_si128 (__m128i __A, const int __B)
2323 return ((__m128i)__builtin_ia32_psrldqi128 (__A, __B))
2326 static __m128i __attribute__((__always_inline__))
2327 _mm_srli_si128 (__m128i __A, const int __B)
2329 return ((__m128i)__builtin_ia32_pslldqi128 (__A, __B))
2332 #define _mm_srli_si128(__A, __B) ((__m128i)__builtin_ia32_psrldqi128 (__A, __B))
2333 #define _mm_slli_si128(__A, __B) ((__m128i)__builtin_ia32_pslldqi128 (__A, __B))
2335 static __inline __m128i
2336 _mm_srli_epi16 (__m128i __A, int __B)
2338 return (__m128i)__builtin_ia32_psrlwi128 ((__v8hi)__A, __B);
2341 static __inline __m128i
2342 _mm_srli_epi32 (__m128i __A, int __B)
2344 return (__m128i)__builtin_ia32_psrldi128 ((__v4si)__A, __B);
2347 static __inline __m128i
2348 _mm_srli_epi64 (__m128i __A, int __B)
2350 return (__m128i)__builtin_ia32_psrlqi128 ((__v2di)__A, __B);
2353 static __inline __m128i
2354 _mm_and_si128 (__m128i __A, __m128i __B)
2356 return (__m128i)__builtin_ia32_pand128 ((__v2di)__A, (__v2di)__B);
2359 static __inline __m128i
2360 _mm_andnot_si128 (__m128i __A, __m128i __B)
2362 return (__m128i)__builtin_ia32_pandn128 ((__v2di)__A, (__v2di)__B);
2365 static __inline __m128i
2366 _mm_or_si128 (__m128i __A, __m128i __B)
2368 return (__m128i)__builtin_ia32_por128 ((__v2di)__A, (__v2di)__B);
2371 static __inline __m128i
2372 _mm_xor_si128 (__m128i __A, __m128i __B)
2374 return (__m128i)__builtin_ia32_pxor128 ((__v2di)__A, (__v2di)__B);
2377 static __inline __m128i
2378 _mm_cmpeq_epi8 (__m128i __A, __m128i __B)
2380 return (__m128i)__builtin_ia32_pcmpeqb128 ((__v16qi)__A, (__v16qi)__B);
2383 static __inline __m128i
2384 _mm_cmpeq_epi16 (__m128i __A, __m128i __B)
2386 return (__m128i)__builtin_ia32_pcmpeqw128 ((__v8hi)__A, (__v8hi)__B);
2389 static __inline __m128i
2390 _mm_cmpeq_epi32 (__m128i __A, __m128i __B)
2392 return (__m128i)__builtin_ia32_pcmpeqd128 ((__v4si)__A, (__v4si)__B);
2395 static __inline __m128i
2396 _mm_cmplt_epi8 (__m128i __A, __m128i __B)
2398 return (__m128i)__builtin_ia32_pcmpgtb128 ((__v16qi)__B, (__v16qi)__A);
2401 static __inline __m128i
2402 _mm_cmplt_epi16 (__m128i __A, __m128i __B)
2404 return (__m128i)__builtin_ia32_pcmpgtw128 ((__v8hi)__B, (__v8hi)__A);
2407 static __inline __m128i
2408 _mm_cmplt_epi32 (__m128i __A, __m128i __B)
2410 return (__m128i)__builtin_ia32_pcmpgtd128 ((__v4si)__B, (__v4si)__A);
2413 static __inline __m128i
2414 _mm_cmpgt_epi8 (__m128i __A, __m128i __B)
2416 return (__m128i)__builtin_ia32_pcmpgtb128 ((__v16qi)__A, (__v16qi)__B);
2419 static __inline __m128i
2420 _mm_cmpgt_epi16 (__m128i __A, __m128i __B)
2422 return (__m128i)__builtin_ia32_pcmpgtw128 ((__v8hi)__A, (__v8hi)__B);
2425 static __inline __m128i
2426 _mm_cmpgt_epi32 (__m128i __A, __m128i __B)
2428 return (__m128i)__builtin_ia32_pcmpgtd128 ((__v4si)__A, (__v4si)__B);
2431 #define _mm_extract_epi16(__A, __B) __builtin_ia32_pextrw128 ((__v8hi)__A, __B)
2433 #define _mm_insert_epi16(__A, __B, __C) ((__m128i)__builtin_ia32_pinsrw128 ((__v8hi)__A, __B, __C))
2435 static __inline __m128i
2436 _mm_max_epi16 (__m128i __A, __m128i __B)
2438 return (__m128i)__builtin_ia32_pmaxsw128 ((__v8hi)__A, (__v8hi)__B);
2441 static __inline __m128i
2442 _mm_max_epu8 (__m128i __A, __m128i __B)
2444 return (__m128i)__builtin_ia32_pmaxub128 ((__v16qi)__A, (__v16qi)__B);
2447 static __inline __m128i
2448 _mm_min_epi16 (__m128i __A, __m128i __B)
2450 return (__m128i)__builtin_ia32_pminsw128 ((__v8hi)__A, (__v8hi)__B);
2453 static __inline __m128i
2454 _mm_min_epu8 (__m128i __A, __m128i __B)
2456 return (__m128i)__builtin_ia32_pminub128 ((__v16qi)__A, (__v16qi)__B);
2460 _mm_movemask_epi8 (__m128i __A)
2462 return __builtin_ia32_pmovmskb128 ((__v16qi)__A);
2465 static __inline __m128i
2466 _mm_mulhi_epu16 (__m128i __A, __m128i __B)
2468 return (__m128i)__builtin_ia32_pmulhuw128 ((__v8hi)__A, (__v8hi)__B);
2471 #define _mm_shufflehi_epi16(__A, __B) ((__m128i)__builtin_ia32_pshufhw ((__v8hi)__A, __B))
2472 #define _mm_shufflelo_epi16(__A, __B) ((__m128i)__builtin_ia32_pshuflw ((__v8hi)__A, __B))
2473 #define _mm_shuffle_epi32(__A, __B) ((__m128i)__builtin_ia32_pshufd ((__v4si)__A, __B))
2475 static __inline void
2476 _mm_maskmoveu_si128 (__m128i __A, __m128i __B, char *__C)
2478 __builtin_ia32_maskmovdqu ((__v16qi)__A, (__v16qi)__B, __C);
2481 static __inline __m128i
2482 _mm_avg_epu8 (__m128i __A, __m128i __B)
2484 return (__m128i)__builtin_ia32_pavgb128 ((__v16qi)__A, (__v16qi)__B);
2487 static __inline __m128i
2488 _mm_avg_epu16 (__m128i __A, __m128i __B)
2490 return (__m128i)__builtin_ia32_pavgw128 ((__v8hi)__A, (__v8hi)__B);
2493 static __inline __m128i
2494 _mm_sad_epu8 (__m128i __A, __m128i __B)
2496 return (__m128i)__builtin_ia32_psadbw128 ((__v16qi)__A, (__v16qi)__B);
2499 static __inline void
2500 _mm_stream_si32 (int *__A, int __B)
2502 __builtin_ia32_movnti (__A, __B);
2505 static __inline void
2506 _mm_stream_si128 (__m128i *__A, __m128i __B)
2508 __builtin_ia32_movntdq ((__v2di *)__A, (__v2di)__B);
2511 static __inline void
2512 _mm_stream_pd (double *__A, __m128d __B)
2514 __builtin_ia32_movntpd (__A, (__v2df)__B);
2517 static __inline __m128i
2518 _mm_movpi64_epi64 (__m64 __A)
2520 return (__m128i)__builtin_ia32_movq2dq ((unsigned long long)__A);
2523 static __inline void
2524 _mm_clflush (void const *__A)
2526 return __builtin_ia32_clflush (__A);
2529 static __inline void
2532 __builtin_ia32_lfence ();
2535 static __inline void
2538 __builtin_ia32_mfence ();
2541 static __inline __m128i
2542 _mm_cvtsi32_si128 (int __A)
2544 return (__m128i) __builtin_ia32_loadd (&__A);
2548 static __inline __m128i
2549 _mm_cvtsi64x_si128 (long long __A)
2551 return (__m128i) __builtin_ia32_movq2dq (__A);
2556 _mm_cvtsi128_si32 (__m128i __A)
2559 __builtin_ia32_stored (&__tmp, (__v4si)__A);
2564 static __inline long long
2565 _mm_cvtsi128_si64x (__m128i __A)
2567 return __builtin_ia32_movdq2q ((__v2di)__A);
2571 #endif /* __SSE2__ */
2573 #endif /* __SSE__ */
2574 #endif /* _XMMINTRIN_H_INCLUDED */