40 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("sse")))
43 _mm_add_ss(__m128 __a, __m128
__b)
50 _mm_add_ps(__m128 __a, __m128
__b)
56 _mm_sub_ss(__m128 __a, __m128
__b)
63 _mm_sub_ps(__m128 __a, __m128
__b)
69 _mm_mul_ss(__m128 __a, __m128
__b)
76 _mm_mul_ps(__m128 __a, __m128
__b)
82 _mm_div_ss(__m128 __a, __m128
__b)
89 _mm_div_ps(__m128 __a, __m128
__b)
95 _mm_sqrt_ss(__m128 __a)
97 __m128
__c = __builtin_ia32_sqrtss(__a);
98 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
102 _mm_sqrt_ps(__m128 __a)
104 return __builtin_ia32_sqrtps(__a);
108 _mm_rcp_ss(__m128 __a)
110 __m128 __c = __builtin_ia32_rcpss(__a);
111 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
115 _mm_rcp_ps(__m128 __a)
117 return __builtin_ia32_rcpps(__a);
121 _mm_rsqrt_ss(__m128 __a)
123 __m128 __c = __builtin_ia32_rsqrtss(__a);
124 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
128 _mm_rsqrt_ps(__m128 __a)
130 return __builtin_ia32_rsqrtps(__a);
134 _mm_min_ss(__m128 __a, __m128
__b)
136 return __builtin_ia32_minss(__a, __b);
140 _mm_min_ps(__m128 __a, __m128
__b)
142 return __builtin_ia32_minps(__a, __b);
146 _mm_max_ss(__m128 __a, __m128
__b)
148 return __builtin_ia32_maxss(__a, __b);
152 _mm_max_ps(__m128 __a, __m128
__b)
154 return __builtin_ia32_maxps(__a, __b);
158 _mm_and_ps(__m128 __a, __m128
__b)
160 return (__m128)((__v4si)__a & (__v4si)
__b);
164 _mm_andnot_ps(__m128 __a, __m128 __b)
166 return (__m128)(~(__v4si)__a & (__v4si)
__b);
170 _mm_or_ps(__m128 __a, __m128 __b)
172 return (__m128)((__v4si)__a | (__v4si)
__b);
176 _mm_xor_ps(__m128 __a, __m128 __b)
178 return (__m128)((__v4si)__a ^ (__v4si)
__b);
182 _mm_cmpeq_ss(__m128 __a, __m128 __b)
184 return (__m128)__builtin_ia32_cmpeqss(__a, __b);
188 _mm_cmpeq_ps(__m128 __a, __m128 __b)
190 return (__m128)__builtin_ia32_cmpeqps(__a, __b);
194 _mm_cmplt_ss(__m128 __a, __m128 __b)
196 return (__m128)__builtin_ia32_cmpltss(__a, __b);
200 _mm_cmplt_ps(__m128 __a, __m128 __b)
202 return (__m128)__builtin_ia32_cmpltps(__a, __b);
206 _mm_cmple_ss(__m128 __a, __m128 __b)
208 return (__m128)__builtin_ia32_cmpless(__a, __b);
212 _mm_cmple_ps(__m128 __a, __m128 __b)
214 return (__m128)__builtin_ia32_cmpleps(__a, __b);
218 _mm_cmpgt_ss(__m128 __a, __m128 __b)
220 return (__m128)__builtin_shufflevector(__a,
221 __builtin_ia32_cmpltss(__b, __a),
226 _mm_cmpgt_ps(__m128 __a, __m128 __b)
228 return (__m128)__builtin_ia32_cmpltps(__b, __a);
232 _mm_cmpge_ss(__m128 __a, __m128 __b)
234 return (__m128)__builtin_shufflevector(__a,
235 __builtin_ia32_cmpless(__b, __a),
240 _mm_cmpge_ps(__m128 __a, __m128 __b)
242 return (__m128)__builtin_ia32_cmpleps(__b, __a);
246 _mm_cmpneq_ss(__m128 __a, __m128 __b)
248 return (__m128)__builtin_ia32_cmpneqss(__a, __b);
252 _mm_cmpneq_ps(__m128 __a, __m128 __b)
254 return (__m128)__builtin_ia32_cmpneqps(__a, __b);
258 _mm_cmpnlt_ss(__m128 __a, __m128 __b)
260 return (__m128)__builtin_ia32_cmpnltss(__a, __b);
264 _mm_cmpnlt_ps(__m128 __a, __m128 __b)
266 return (__m128)__builtin_ia32_cmpnltps(__a, __b);
270 _mm_cmpnle_ss(__m128 __a, __m128 __b)
272 return (__m128)__builtin_ia32_cmpnless(__a, __b);
276 _mm_cmpnle_ps(__m128 __a, __m128 __b)
278 return (__m128)__builtin_ia32_cmpnleps(__a, __b);
282 _mm_cmpngt_ss(__m128 __a, __m128 __b)
284 return (__m128)__builtin_shufflevector(__a,
285 __builtin_ia32_cmpnltss(__b, __a),
290 _mm_cmpngt_ps(__m128 __a, __m128 __b)
292 return (__m128)__builtin_ia32_cmpnltps(__b, __a);
296 _mm_cmpnge_ss(__m128 __a, __m128 __b)
298 return (__m128)__builtin_shufflevector(__a,
299 __builtin_ia32_cmpnless(__b, __a),
304 _mm_cmpnge_ps(__m128 __a, __m128 __b)
306 return (__m128)__builtin_ia32_cmpnleps(__b, __a);
310 _mm_cmpord_ss(__m128 __a, __m128 __b)
312 return (__m128)__builtin_ia32_cmpordss(__a, __b);
316 _mm_cmpord_ps(__m128 __a, __m128 __b)
318 return (__m128)__builtin_ia32_cmpordps(__a, __b);
322 _mm_cmpunord_ss(__m128 __a, __m128 __b)
324 return (__m128)__builtin_ia32_cmpunordss(__a, __b);
328 _mm_cmpunord_ps(__m128 __a, __m128 __b)
330 return (__m128)__builtin_ia32_cmpunordps(__a, __b);
334 _mm_comieq_ss(__m128 __a, __m128 __b)
336 return __builtin_ia32_comieq(__a, __b);
340 _mm_comilt_ss(__m128 __a, __m128 __b)
342 return __builtin_ia32_comilt(__a, __b);
346 _mm_comile_ss(__m128 __a, __m128 __b)
348 return __builtin_ia32_comile(__a, __b);
352 _mm_comigt_ss(__m128 __a, __m128 __b)
354 return __builtin_ia32_comigt(__a, __b);
358 _mm_comige_ss(__m128 __a, __m128 __b)
360 return __builtin_ia32_comige(__a, __b);
364 _mm_comineq_ss(__m128 __a, __m128 __b)
366 return __builtin_ia32_comineq(__a, __b);
370 _mm_ucomieq_ss(__m128 __a, __m128 __b)
372 return __builtin_ia32_ucomieq(__a, __b);
376 _mm_ucomilt_ss(__m128 __a, __m128 __b)
378 return __builtin_ia32_ucomilt(__a, __b);
382 _mm_ucomile_ss(__m128 __a, __m128 __b)
384 return __builtin_ia32_ucomile(__a, __b);
388 _mm_ucomigt_ss(__m128 __a, __m128 __b)
390 return __builtin_ia32_ucomigt(__a, __b);
394 _mm_ucomige_ss(__m128 __a, __m128 __b)
396 return __builtin_ia32_ucomige(__a, __b);
400 _mm_ucomineq_ss(__m128 __a, __m128 __b)
402 return __builtin_ia32_ucomineq(__a, __b);
406 _mm_cvtss_si32(__m128 __a)
408 return __builtin_ia32_cvtss2si(__a);
412 _mm_cvt_ss2si(__m128 __a)
414 return _mm_cvtss_si32(__a);
420 _mm_cvtss_si64(__m128 __a)
422 return __builtin_ia32_cvtss2si64(__a);
428 _mm_cvtps_pi32(__m128 __a)
430 return (__m64)__builtin_ia32_cvtps2pi(__a);
434 _mm_cvt_ps2pi(__m128 __a)
436 return _mm_cvtps_pi32(__a);
440 _mm_cvttss_si32(__m128 __a)
446 _mm_cvtt_ss2si(__m128 __a)
448 return _mm_cvttss_si32(__a);
452 _mm_cvttss_si64(__m128 __a)
458 _mm_cvttps_pi32(__m128 __a)
460 return (__m64)__builtin_ia32_cvttps2pi(__a);
464 _mm_cvtt_ps2pi(__m128 __a)
466 return _mm_cvttps_pi32(__a);
470 _mm_cvtsi32_ss(__m128 __a,
int __b)
477 _mm_cvt_si2ss(__m128 __a,
int __b)
479 return _mm_cvtsi32_ss(__a, __b);
485 _mm_cvtsi64_ss(__m128 __a,
long long __b)
494 _mm_cvtpi32_ps(__m128 __a, __m64 __b)
496 return __builtin_ia32_cvtpi2ps(__a, (__v2si)__b);
500 _mm_cvt_pi2ps(__m128 __a, __m64 __b)
502 return _mm_cvtpi32_ps(__a, __b);
506 _mm_cvtss_f32(__m128 __a)
512 _mm_loadh_pi(__m128 __a,
const __m64 *
__p)
514 typedef float __mm_loadh_pi_v2f32
__attribute__((__vector_size__(8)));
515 struct __mm_loadh_pi_struct {
516 __mm_loadh_pi_v2f32 __u;
518 __mm_loadh_pi_v2f32 __b = ((
struct __mm_loadh_pi_struct*)__p)->__u;
519 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
520 return __builtin_shufflevector(__a, __bb, 0, 1, 4, 5);
524 _mm_loadl_pi(__m128 __a,
const __m64 *__p)
526 typedef float __mm_loadl_pi_v2f32
__attribute__((__vector_size__(8)));
527 struct __mm_loadl_pi_struct {
528 __mm_loadl_pi_v2f32 __u;
530 __mm_loadl_pi_v2f32 __b = ((
struct __mm_loadl_pi_struct*)__p)->__u;
531 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
532 return __builtin_shufflevector(__a, __bb, 4, 5, 2, 3);
536 _mm_load_ss(
const float *__p)
538 struct __mm_load_ss_struct {
541 float __u = ((
struct __mm_load_ss_struct*)__p)->__u;
542 return (__m128){ __u, 0, 0, 0 };
546 _mm_load1_ps(
const float *__p)
548 struct __mm_load1_ps_struct {
551 float __u = ((
struct __mm_load1_ps_struct*)__p)->__u;
552 return (__m128){ __u, __u, __u, __u };
555 #define _mm_load_ps1(p) _mm_load1_ps(p)
558 _mm_load_ps(
const float *__p)
560 return *(__m128*)__p;
564 _mm_loadu_ps(
const float *__p)
569 return ((
struct __loadu_ps*)__p)->__v;
573 _mm_loadr_ps(
const float *__p)
575 __m128 __a = _mm_load_ps(__p);
576 return __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
582 return (__m128)__builtin_ia32_undef128();
586 _mm_set_ss(
float __w)
588 return (__m128){ __w, 0, 0, 0 };
592 _mm_set1_ps(
float __w)
594 return (__m128){ __w, __w, __w, __w };
599 _mm_set_ps1(
float __w)
601 return _mm_set1_ps(__w);
605 _mm_set_ps(
float __z,
float __y,
float __x,
float __w)
607 return (__m128){ __w,
__x,
__y, __z };
611 _mm_setr_ps(
float __z,
float __y,
float __x,
float __w)
613 return (__m128){ __z,
__y,
__x, __w };
619 return (__m128){ 0, 0, 0, 0 };
623 _mm_storeh_pi(__m64 *__p, __m128 __a)
625 __builtin_ia32_storehps((__v2si *)__p, __a);
629 _mm_storel_pi(__m64 *__p, __m128 __a)
631 __builtin_ia32_storelps((__v2si *)__p, __a);
635 _mm_store_ss(
float *__p, __m128 __a)
637 struct __mm_store_ss_struct {
640 ((
struct __mm_store_ss_struct*)__p)->__u = __a[0];
644 _mm_storeu_ps(
float *__p, __m128 __a)
646 __builtin_ia32_storeups(__p, __a);
650 _mm_store1_ps(
float *__p, __m128 __a)
652 __a = __builtin_shufflevector(__a, __a, 0, 0, 0, 0);
653 _mm_storeu_ps(__p, __a);
657 _mm_store_ps1(
float *__p, __m128 __a)
659 return _mm_store1_ps(__p, __a);
663 _mm_store_ps(
float *__p, __m128 __a)
665 *(__m128 *)__p = __a;
669 _mm_storer_ps(
float *__p, __m128 __a)
671 __a = __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
672 _mm_store_ps(__p, __a);
675 #define _MM_HINT_T0 3
676 #define _MM_HINT_T1 2
677 #define _MM_HINT_T2 1
678 #define _MM_HINT_NTA 0
684 #define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel)))
688 _mm_stream_pi(__m64 *__p, __m64 __a)
690 __builtin_ia32_movntq(__p, __a);
694 _mm_stream_ps(
float *__p, __m128 __a)
696 __builtin_ia32_movntps(__p, __a);
702 __builtin_ia32_sfence();
706 _mm_extract_pi16(__m64 __a,
int __n)
708 __v4hi __b = (__v4hi)__a;
709 return (
unsigned short)__b[__n & 3];
713 _mm_insert_pi16(__m64 __a,
int __d,
int __n)
715 __v4hi __b = (__v4hi)__a;
721 _mm_max_pi16(__m64 __a, __m64 __b)
723 return (__m64)__builtin_ia32_pmaxsw((__v4hi)__a, (__v4hi)__b);
727 _mm_max_pu8(__m64 __a, __m64 __b)
729 return (__m64)__builtin_ia32_pmaxub((__v8qi)__a, (__v8qi)__b);
733 _mm_min_pi16(__m64 __a, __m64 __b)
735 return (__m64)__builtin_ia32_pminsw((__v4hi)__a, (__v4hi)__b);
739 _mm_min_pu8(__m64 __a, __m64 __b)
741 return (__m64)__builtin_ia32_pminub((__v8qi)__a, (__v8qi)__b);
745 _mm_movemask_pi8(__m64 __a)
747 return __builtin_ia32_pmovmskb((__v8qi)__a);
751 _mm_mulhi_pu16(__m64 __a, __m64 __b)
753 return (__m64)__builtin_ia32_pmulhuw((__v4hi)__a, (__v4hi)__b);
756 #define _mm_shuffle_pi16(a, n) __extension__ ({ \
757 (__m64)__builtin_ia32_pshufw((__v4hi)(__m64)(a), (n)); })
760 _mm_maskmove_si64(__m64 __d, __m64 __n,
char *__p)
762 __builtin_ia32_maskmovq((__v8qi)__d, (__v8qi)__n, __p);
766 _mm_avg_pu8(__m64 __a, __m64 __b)
768 return (__m64)__builtin_ia32_pavgb((__v8qi)__a, (__v8qi)__b);
772 _mm_avg_pu16(__m64 __a, __m64 __b)
774 return (__m64)__builtin_ia32_pavgw((__v4hi)__a, (__v4hi)__b);
778 _mm_sad_pu8(__m64 __a, __m64 __b)
780 return (__m64)__builtin_ia32_psadbw((__v8qi)__a, (__v8qi)__b);
786 return __builtin_ia32_stmxcsr();
790 _mm_setcsr(
unsigned int __i)
792 __builtin_ia32_ldmxcsr(__i);
795 #define _mm_shuffle_ps(a, b, mask) __extension__ ({ \
796 (__m128)__builtin_shufflevector((__v4sf)(__m128)(a), (__v4sf)(__m128)(b), \
797 (mask) & 0x3, ((mask) & 0xc) >> 2, \
798 (((mask) & 0x30) >> 4) + 4, \
799 (((mask) & 0xc0) >> 6) + 4); })
802 _mm_unpackhi_ps(__m128 __a, __m128 __b)
804 return __builtin_shufflevector(__a, __b, 2, 6, 3, 7);
808 _mm_unpacklo_ps(__m128 __a, __m128 __b)
810 return __builtin_shufflevector(__a, __b, 0, 4, 1, 5);
814 _mm_move_ss(__m128 __a, __m128 __b)
816 return __builtin_shufflevector(__a, __b, 4, 1, 2, 3);
820 _mm_movehl_ps(__m128 __a, __m128 __b)
822 return __builtin_shufflevector(__a, __b, 6, 7, 2, 3);
826 _mm_movelh_ps(__m128 __a, __m128 __b)
828 return __builtin_shufflevector(__a, __b, 0, 1, 4, 5);
832 _mm_cvtpi16_ps(__m64 __a)
837 __b = _mm_setzero_si64();
838 __b = _mm_cmpgt_pi16(__b, __a);
839 __c = _mm_unpackhi_pi16(__a, __b);
840 __r = _mm_setzero_ps();
841 __r = _mm_cvtpi32_ps(__r, __c);
842 __r = _mm_movelh_ps(__r, __r);
843 __c = _mm_unpacklo_pi16(__a, __b);
844 __r = _mm_cvtpi32_ps(__r, __c);
850 _mm_cvtpu16_ps(__m64 __a)
855 __b = _mm_setzero_si64();
856 __c = _mm_unpackhi_pi16(__a, __b);
857 __r = _mm_setzero_ps();
858 __r = _mm_cvtpi32_ps(__r, __c);
859 __r = _mm_movelh_ps(__r, __r);
860 __c = _mm_unpacklo_pi16(__a, __b);
861 __r = _mm_cvtpi32_ps(__r, __c);
867 _mm_cvtpi8_ps(__m64 __a)
871 __b = _mm_setzero_si64();
872 __b = _mm_cmpgt_pi8(__b, __a);
873 __b = _mm_unpacklo_pi8(__a, __b);
875 return _mm_cvtpi16_ps(__b);
879 _mm_cvtpu8_ps(__m64 __a)
883 __b = _mm_setzero_si64();
884 __b = _mm_unpacklo_pi8(__a, __b);
886 return _mm_cvtpi16_ps(__b);
890 _mm_cvtpi32x2_ps(__m64 __a, __m64 __b)
894 __c = _mm_setzero_ps();
895 __c = _mm_cvtpi32_ps(__c, __b);
896 __c = _mm_movelh_ps(__c, __c);
898 return _mm_cvtpi32_ps(__c, __a);
902 _mm_cvtps_pi16(__m128 __a)
906 __b = _mm_cvtps_pi32(__a);
907 __a = _mm_movehl_ps(__a, __a);
908 __c = _mm_cvtps_pi32(__a);
910 return _mm_packs_pi32(__b, __c);
914 _mm_cvtps_pi8(__m128 __a)
918 __b = _mm_cvtps_pi16(__a);
919 __c = _mm_setzero_si64();
921 return _mm_packs_pi16(__b, __c);
925 _mm_movemask_ps(__m128 __a)
927 return __builtin_ia32_movmskps(__a);
932 #define _MM_ALIGN16 __declspec(align(16))
935 #define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w))
937 #define _MM_EXCEPT_INVALID (0x0001)
938 #define _MM_EXCEPT_DENORM (0x0002)
939 #define _MM_EXCEPT_DIV_ZERO (0x0004)
940 #define _MM_EXCEPT_OVERFLOW (0x0008)
941 #define _MM_EXCEPT_UNDERFLOW (0x0010)
942 #define _MM_EXCEPT_INEXACT (0x0020)
943 #define _MM_EXCEPT_MASK (0x003f)
945 #define _MM_MASK_INVALID (0x0080)
946 #define _MM_MASK_DENORM (0x0100)
947 #define _MM_MASK_DIV_ZERO (0x0200)
948 #define _MM_MASK_OVERFLOW (0x0400)
949 #define _MM_MASK_UNDERFLOW (0x0800)
950 #define _MM_MASK_INEXACT (0x1000)
951 #define _MM_MASK_MASK (0x1f80)
953 #define _MM_ROUND_NEAREST (0x0000)
954 #define _MM_ROUND_DOWN (0x2000)
955 #define _MM_ROUND_UP (0x4000)
956 #define _MM_ROUND_TOWARD_ZERO (0x6000)
957 #define _MM_ROUND_MASK (0x6000)
959 #define _MM_FLUSH_ZERO_MASK (0x8000)
960 #define _MM_FLUSH_ZERO_ON (0x8000)
961 #define _MM_FLUSH_ZERO_OFF (0x0000)
963 #define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK)
964 #define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK)
965 #define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK)
966 #define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK)
968 #define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x)))
969 #define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x)))
970 #define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x)))
971 #define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x)))
973 #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
975 __m128 tmp3, tmp2, tmp1, tmp0; \
976 tmp0 = _mm_unpacklo_ps((row0), (row1)); \
977 tmp2 = _mm_unpacklo_ps((row2), (row3)); \
978 tmp1 = _mm_unpackhi_ps((row0), (row1)); \
979 tmp3 = _mm_unpackhi_ps((row2), (row3)); \
980 (row0) = _mm_movelh_ps(tmp0, tmp2); \
981 (row1) = _mm_movehl_ps(tmp2, tmp0); \
982 (row2) = _mm_movelh_ps(tmp1, tmp3); \
983 (row3) = _mm_movehl_ps(tmp3, tmp1); \
987 #define _m_pextrw _mm_extract_pi16
988 #define _m_pinsrw _mm_insert_pi16
989 #define _m_pmaxsw _mm_max_pi16
990 #define _m_pmaxub _mm_max_pu8
991 #define _m_pminsw _mm_min_pi16
992 #define _m_pminub _mm_min_pu8
993 #define _m_pmovmskb _mm_movemask_pi8
994 #define _m_pmulhuw _mm_mulhi_pu16
995 #define _m_pshufw _mm_shuffle_pi16
996 #define _m_maskmovq _mm_maskmove_si64
997 #define _m_pavgb _mm_avg_pu8
998 #define _m_pavgw _mm_avg_pu16
999 #define _m_psadbw _mm_sad_pu8
1003 #undef __DEFAULT_FN_ATTRS
1006 #if defined(__SSE2__) && !__has_feature(modules)
static __inline unsigned char unsigned int unsigned int unsigned int * __p
Definition: adxintrin.h:38
static __inline unsigned char unsigned int unsigned int __y
Definition: adxintrin.h:36
int __v4si __attribute__((__vector_size__(16)))
Definition: xmmintrin.h:29
static vector float vector float __b
Definition: altivec.h:419
static __inline unsigned char unsigned int __x
Definition: adxintrin.h:36
#define __DEFAULT_FN_ATTRS
Definition: xmmintrin.h:40
static void int int __c
Definition: altivec.h:2051