28 #error "SSE instruction set not enabled"
44 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
47 _mm_add_ss(__m128 __a, __m128
__b)
54 _mm_add_ps(__m128 __a, __m128
__b)
60 _mm_sub_ss(__m128 __a, __m128
__b)
67 _mm_sub_ps(__m128 __a, __m128
__b)
73 _mm_mul_ss(__m128 __a, __m128
__b)
80 _mm_mul_ps(__m128 __a, __m128
__b)
86 _mm_div_ss(__m128 __a, __m128
__b)
93 _mm_div_ps(__m128 __a, __m128
__b)
99 _mm_sqrt_ss(__m128 __a)
101 __m128
__c = __builtin_ia32_sqrtss(__a);
102 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
106 _mm_sqrt_ps(__m128 __a)
108 return __builtin_ia32_sqrtps(__a);
112 _mm_rcp_ss(__m128 __a)
114 __m128 __c = __builtin_ia32_rcpss(__a);
115 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
119 _mm_rcp_ps(__m128 __a)
121 return __builtin_ia32_rcpps(__a);
125 _mm_rsqrt_ss(__m128 __a)
127 __m128 __c = __builtin_ia32_rsqrtss(__a);
128 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
132 _mm_rsqrt_ps(__m128 __a)
134 return __builtin_ia32_rsqrtps(__a);
138 _mm_min_ss(__m128 __a, __m128
__b)
140 return __builtin_ia32_minss(__a, __b);
144 _mm_min_ps(__m128 __a, __m128
__b)
146 return __builtin_ia32_minps(__a, __b);
150 _mm_max_ss(__m128 __a, __m128
__b)
152 return __builtin_ia32_maxss(__a, __b);
156 _mm_max_ps(__m128 __a, __m128
__b)
158 return __builtin_ia32_maxps(__a, __b);
162 _mm_and_ps(__m128 __a, __m128
__b)
164 return (__m128)((__v4si)__a & (__v4si)
__b);
168 _mm_andnot_ps(__m128 __a, __m128 __b)
170 return (__m128)(~(__v4si)__a & (__v4si)
__b);
174 _mm_or_ps(__m128 __a, __m128 __b)
176 return (__m128)((__v4si)__a | (__v4si)
__b);
180 _mm_xor_ps(__m128 __a, __m128 __b)
182 return (__m128)((__v4si)__a ^ (__v4si)
__b);
186 _mm_cmpeq_ss(__m128 __a, __m128 __b)
188 return (__m128)__builtin_ia32_cmpeqss(__a, __b);
192 _mm_cmpeq_ps(__m128 __a, __m128 __b)
194 return (__m128)__builtin_ia32_cmpeqps(__a, __b);
198 _mm_cmplt_ss(__m128 __a, __m128 __b)
200 return (__m128)__builtin_ia32_cmpltss(__a, __b);
204 _mm_cmplt_ps(__m128 __a, __m128 __b)
206 return (__m128)__builtin_ia32_cmpltps(__a, __b);
210 _mm_cmple_ss(__m128 __a, __m128 __b)
212 return (__m128)__builtin_ia32_cmpless(__a, __b);
216 _mm_cmple_ps(__m128 __a, __m128 __b)
218 return (__m128)__builtin_ia32_cmpleps(__a, __b);
222 _mm_cmpgt_ss(__m128 __a, __m128 __b)
224 return (__m128)__builtin_shufflevector(__a,
225 __builtin_ia32_cmpltss(__b, __a),
230 _mm_cmpgt_ps(__m128 __a, __m128 __b)
232 return (__m128)__builtin_ia32_cmpltps(__b, __a);
236 _mm_cmpge_ss(__m128 __a, __m128 __b)
238 return (__m128)__builtin_shufflevector(__a,
239 __builtin_ia32_cmpless(__b, __a),
244 _mm_cmpge_ps(__m128 __a, __m128 __b)
246 return (__m128)__builtin_ia32_cmpleps(__b, __a);
250 _mm_cmpneq_ss(__m128 __a, __m128 __b)
252 return (__m128)__builtin_ia32_cmpneqss(__a, __b);
256 _mm_cmpneq_ps(__m128 __a, __m128 __b)
258 return (__m128)__builtin_ia32_cmpneqps(__a, __b);
262 _mm_cmpnlt_ss(__m128 __a, __m128 __b)
264 return (__m128)__builtin_ia32_cmpnltss(__a, __b);
268 _mm_cmpnlt_ps(__m128 __a, __m128 __b)
270 return (__m128)__builtin_ia32_cmpnltps(__a, __b);
274 _mm_cmpnle_ss(__m128 __a, __m128 __b)
276 return (__m128)__builtin_ia32_cmpnless(__a, __b);
280 _mm_cmpnle_ps(__m128 __a, __m128 __b)
282 return (__m128)__builtin_ia32_cmpnleps(__a, __b);
286 _mm_cmpngt_ss(__m128 __a, __m128 __b)
288 return (__m128)__builtin_shufflevector(__a,
289 __builtin_ia32_cmpnltss(__b, __a),
294 _mm_cmpngt_ps(__m128 __a, __m128 __b)
296 return (__m128)__builtin_ia32_cmpnltps(__b, __a);
300 _mm_cmpnge_ss(__m128 __a, __m128 __b)
302 return (__m128)__builtin_shufflevector(__a,
303 __builtin_ia32_cmpnless(__b, __a),
308 _mm_cmpnge_ps(__m128 __a, __m128 __b)
310 return (__m128)__builtin_ia32_cmpnleps(__b, __a);
314 _mm_cmpord_ss(__m128 __a, __m128 __b)
316 return (__m128)__builtin_ia32_cmpordss(__a, __b);
320 _mm_cmpord_ps(__m128 __a, __m128 __b)
322 return (__m128)__builtin_ia32_cmpordps(__a, __b);
326 _mm_cmpunord_ss(__m128 __a, __m128 __b)
328 return (__m128)__builtin_ia32_cmpunordss(__a, __b);
332 _mm_cmpunord_ps(__m128 __a, __m128 __b)
334 return (__m128)__builtin_ia32_cmpunordps(__a, __b);
338 _mm_comieq_ss(__m128 __a, __m128 __b)
340 return __builtin_ia32_comieq(__a, __b);
344 _mm_comilt_ss(__m128 __a, __m128 __b)
346 return __builtin_ia32_comilt(__a, __b);
350 _mm_comile_ss(__m128 __a, __m128 __b)
352 return __builtin_ia32_comile(__a, __b);
356 _mm_comigt_ss(__m128 __a, __m128 __b)
358 return __builtin_ia32_comigt(__a, __b);
362 _mm_comige_ss(__m128 __a, __m128 __b)
364 return __builtin_ia32_comige(__a, __b);
368 _mm_comineq_ss(__m128 __a, __m128 __b)
370 return __builtin_ia32_comineq(__a, __b);
374 _mm_ucomieq_ss(__m128 __a, __m128 __b)
376 return __builtin_ia32_ucomieq(__a, __b);
380 _mm_ucomilt_ss(__m128 __a, __m128 __b)
382 return __builtin_ia32_ucomilt(__a, __b);
386 _mm_ucomile_ss(__m128 __a, __m128 __b)
388 return __builtin_ia32_ucomile(__a, __b);
392 _mm_ucomigt_ss(__m128 __a, __m128 __b)
394 return __builtin_ia32_ucomigt(__a, __b);
398 _mm_ucomige_ss(__m128 __a, __m128 __b)
400 return __builtin_ia32_ucomige(__a, __b);
404 _mm_ucomineq_ss(__m128 __a, __m128 __b)
406 return __builtin_ia32_ucomineq(__a, __b);
410 _mm_cvtss_si32(__m128 __a)
412 return __builtin_ia32_cvtss2si(__a);
416 _mm_cvt_ss2si(__m128 __a)
418 return _mm_cvtss_si32(__a);
424 _mm_cvtss_si64(__m128 __a)
426 return __builtin_ia32_cvtss2si64(__a);
432 _mm_cvtps_pi32(__m128 __a)
434 return (__m64)__builtin_ia32_cvtps2pi(__a);
438 _mm_cvt_ps2pi(__m128 __a)
440 return _mm_cvtps_pi32(__a);
444 _mm_cvttss_si32(__m128 __a)
450 _mm_cvtt_ss2si(__m128 __a)
452 return _mm_cvttss_si32(__a);
456 _mm_cvttss_si64(__m128 __a)
462 _mm_cvttps_pi32(__m128 __a)
464 return (__m64)__builtin_ia32_cvttps2pi(__a);
468 _mm_cvtt_ps2pi(__m128 __a)
470 return _mm_cvttps_pi32(__a);
474 _mm_cvtsi32_ss(__m128 __a,
int __b)
481 _mm_cvt_si2ss(__m128 __a,
int __b)
483 return _mm_cvtsi32_ss(__a, __b);
489 _mm_cvtsi64_ss(__m128 __a,
long long __b)
498 _mm_cvtpi32_ps(__m128 __a, __m64 __b)
500 return __builtin_ia32_cvtpi2ps(__a, (__v2si)__b);
504 _mm_cvt_pi2ps(__m128 __a, __m64 __b)
506 return _mm_cvtpi32_ps(__a, __b);
510 _mm_cvtss_f32(__m128 __a)
516 _mm_loadh_pi(__m128 __a,
const __m64 *__p)
518 typedef float __mm_loadh_pi_v2f32
__attribute__((__vector_size__(8)));
519 struct __mm_loadh_pi_struct {
520 __mm_loadh_pi_v2f32 __u;
522 __mm_loadh_pi_v2f32 __b = ((
struct __mm_loadh_pi_struct*)__p)->__u;
523 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
524 return __builtin_shufflevector(__a, __bb, 0, 1, 4, 5);
528 _mm_loadl_pi(__m128 __a,
const __m64 *__p)
530 typedef float __mm_loadl_pi_v2f32
__attribute__((__vector_size__(8)));
531 struct __mm_loadl_pi_struct {
532 __mm_loadl_pi_v2f32 __u;
534 __mm_loadl_pi_v2f32 __b = ((
struct __mm_loadl_pi_struct*)__p)->__u;
535 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
536 return __builtin_shufflevector(__a, __bb, 4, 5, 2, 3);
540 _mm_load_ss(
const float *__p)
542 struct __mm_load_ss_struct {
545 float __u = ((
struct __mm_load_ss_struct*)__p)->__u;
546 return (__m128){ __u, 0, 0, 0 };
550 _mm_load1_ps(
const float *__p)
552 struct __mm_load1_ps_struct {
555 float __u = ((
struct __mm_load1_ps_struct*)__p)->__u;
556 return (__m128){ __u, __u, __u, __u };
559 #define _mm_load_ps1(p) _mm_load1_ps(p)
562 _mm_load_ps(
const float *__p)
564 return *(__m128*)__p;
568 _mm_loadu_ps(
const float *__p)
573 return ((
struct __loadu_ps*)__p)->__v;
577 _mm_loadr_ps(
const float *__p)
579 __m128 __a = _mm_load_ps(__p);
580 return __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
584 _mm_set_ss(
float __w)
586 return (__m128){ __w, 0, 0, 0 };
590 _mm_set1_ps(
float __w)
592 return (__m128){ __w, __w, __w, __w };
597 _mm_set_ps1(
float __w)
599 return _mm_set1_ps(__w);
603 _mm_set_ps(
float __z,
float __y,
float __x,
float __w)
605 return (__m128){ __w, __x, __y, __z };
609 _mm_setr_ps(
float __z,
float __y,
float __x,
float __w)
611 return (__m128){ __z, __y, __x, __w };
617 return (__m128){ 0, 0, 0, 0 };
621 _mm_storeh_pi(__m64 *__p, __m128 __a)
623 __builtin_ia32_storehps((__v2si *)__p, __a);
627 _mm_storel_pi(__m64 *__p, __m128 __a)
629 __builtin_ia32_storelps((__v2si *)__p, __a);
633 _mm_store_ss(
float *__p, __m128 __a)
635 struct __mm_store_ss_struct {
638 ((
struct __mm_store_ss_struct*)__p)->__u = __a[0];
642 _mm_storeu_ps(
float *__p, __m128 __a)
644 __builtin_ia32_storeups(__p, __a);
648 _mm_store1_ps(
float *__p, __m128 __a)
650 __a = __builtin_shufflevector(__a, __a, 0, 0, 0, 0);
651 _mm_storeu_ps(__p, __a);
655 _mm_store_ps1(
float *__p, __m128 __a)
657 return _mm_store1_ps(__p, __a);
661 _mm_store_ps(
float *__p, __m128 __a)
663 *(__m128 *)__p = __a;
667 _mm_storer_ps(
float *__p, __m128 __a)
669 __a = __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
670 _mm_store_ps(__p, __a);
673 #define _MM_HINT_T0 3
674 #define _MM_HINT_T1 2
675 #define _MM_HINT_T2 1
676 #define _MM_HINT_NTA 0
682 #define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel)))
686 _mm_stream_pi(__m64 *__p, __m64 __a)
688 __builtin_ia32_movntq(__p, __a);
692 _mm_stream_ps(
float *__p, __m128 __a)
694 __builtin_ia32_movntps(__p, __a);
700 __builtin_ia32_sfence();
704 _mm_extract_pi16(__m64 __a,
int __n)
706 __v4hi __b = (__v4hi)__a;
707 return (
unsigned short)__b[__n & 3];
711 _mm_insert_pi16(__m64 __a,
int __d,
int __n)
713 __v4hi __b = (__v4hi)__a;
719 _mm_max_pi16(__m64 __a, __m64 __b)
721 return (__m64)__builtin_ia32_pmaxsw((__v4hi)__a, (__v4hi)__b);
725 _mm_max_pu8(__m64 __a, __m64 __b)
727 return (__m64)__builtin_ia32_pmaxub((__v8qi)__a, (__v8qi)__b);
731 _mm_min_pi16(__m64 __a, __m64 __b)
733 return (__m64)__builtin_ia32_pminsw((__v4hi)__a, (__v4hi)__b);
737 _mm_min_pu8(__m64 __a, __m64 __b)
739 return (__m64)__builtin_ia32_pminub((__v8qi)__a, (__v8qi)__b);
743 _mm_movemask_pi8(__m64 __a)
745 return __builtin_ia32_pmovmskb((__v8qi)__a);
749 _mm_mulhi_pu16(__m64 __a, __m64 __b)
751 return (__m64)__builtin_ia32_pmulhuw((__v4hi)__a, (__v4hi)__b);
754 #define _mm_shuffle_pi16(a, n) __extension__ ({ \
756 (__m64)__builtin_ia32_pshufw((__v4hi)__a, (n)); })
759 _mm_maskmove_si64(__m64 __d, __m64 __n,
char *__p)
761 __builtin_ia32_maskmovq((__v8qi)__d, (__v8qi)__n, __p);
765 _mm_avg_pu8(__m64 __a, __m64 __b)
767 return (__m64)__builtin_ia32_pavgb((__v8qi)__a, (__v8qi)__b);
771 _mm_avg_pu16(__m64 __a, __m64 __b)
773 return (__m64)__builtin_ia32_pavgw((__v4hi)__a, (__v4hi)__b);
777 _mm_sad_pu8(__m64 __a, __m64 __b)
779 return (__m64)__builtin_ia32_psadbw((__v8qi)__a, (__v8qi)__b);
785 return __builtin_ia32_stmxcsr();
789 _mm_setcsr(
unsigned int __i)
791 __builtin_ia32_ldmxcsr(__i);
794 #define _mm_shuffle_ps(a, b, mask) __extension__ ({ \
797 (__m128)__builtin_shufflevector((__v4sf)__a, (__v4sf)__b, \
798 (mask) & 0x3, ((mask) & 0xc) >> 2, \
799 (((mask) & 0x30) >> 4) + 4, \
800 (((mask) & 0xc0) >> 6) + 4); })
803 _mm_unpackhi_ps(__m128 __a, __m128 __b)
805 return __builtin_shufflevector(__a, __b, 2, 6, 3, 7);
809 _mm_unpacklo_ps(__m128 __a, __m128 __b)
811 return __builtin_shufflevector(__a, __b, 0, 4, 1, 5);
815 _mm_move_ss(__m128 __a, __m128 __b)
817 return __builtin_shufflevector(__a, __b, 4, 1, 2, 3);
821 _mm_movehl_ps(__m128 __a, __m128 __b)
823 return __builtin_shufflevector(__a, __b, 6, 7, 2, 3);
827 _mm_movelh_ps(__m128 __a, __m128 __b)
829 return __builtin_shufflevector(__a, __b, 0, 1, 4, 5);
833 _mm_cvtpi16_ps(__m64 __a)
838 __b = _mm_setzero_si64();
839 __b = _mm_cmpgt_pi16(__b, __a);
840 __c = _mm_unpackhi_pi16(__a, __b);
841 __r = _mm_setzero_ps();
842 __r = _mm_cvtpi32_ps(__r, __c);
843 __r = _mm_movelh_ps(__r, __r);
844 __c = _mm_unpacklo_pi16(__a, __b);
845 __r = _mm_cvtpi32_ps(__r, __c);
851 _mm_cvtpu16_ps(__m64 __a)
856 __b = _mm_setzero_si64();
857 __c = _mm_unpackhi_pi16(__a, __b);
858 __r = _mm_setzero_ps();
859 __r = _mm_cvtpi32_ps(__r, __c);
860 __r = _mm_movelh_ps(__r, __r);
861 __c = _mm_unpacklo_pi16(__a, __b);
862 __r = _mm_cvtpi32_ps(__r, __c);
868 _mm_cvtpi8_ps(__m64 __a)
872 __b = _mm_setzero_si64();
873 __b = _mm_cmpgt_pi8(__b, __a);
874 __b = _mm_unpacklo_pi8(__a, __b);
876 return _mm_cvtpi16_ps(__b);
880 _mm_cvtpu8_ps(__m64 __a)
884 __b = _mm_setzero_si64();
885 __b = _mm_unpacklo_pi8(__a, __b);
887 return _mm_cvtpi16_ps(__b);
891 _mm_cvtpi32x2_ps(__m64 __a, __m64 __b)
895 __c = _mm_setzero_ps();
896 __c = _mm_cvtpi32_ps(__c, __b);
897 __c = _mm_movelh_ps(__c, __c);
899 return _mm_cvtpi32_ps(__c, __a);
903 _mm_cvtps_pi16(__m128 __a)
907 __b = _mm_cvtps_pi32(__a);
908 __a = _mm_movehl_ps(__a, __a);
909 __c = _mm_cvtps_pi32(__a);
911 return _mm_packs_pi32(__b, __c);
915 _mm_cvtps_pi8(__m128 __a)
919 __b = _mm_cvtps_pi16(__a);
920 __c = _mm_setzero_si64();
922 return _mm_packs_pi16(__b, __c);
926 _mm_movemask_ps(__m128 __a)
928 return __builtin_ia32_movmskps(__a);
931 #define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w))
933 #define _MM_EXCEPT_INVALID (0x0001)
934 #define _MM_EXCEPT_DENORM (0x0002)
935 #define _MM_EXCEPT_DIV_ZERO (0x0004)
936 #define _MM_EXCEPT_OVERFLOW (0x0008)
937 #define _MM_EXCEPT_UNDERFLOW (0x0010)
938 #define _MM_EXCEPT_INEXACT (0x0020)
939 #define _MM_EXCEPT_MASK (0x003f)
941 #define _MM_MASK_INVALID (0x0080)
942 #define _MM_MASK_DENORM (0x0100)
943 #define _MM_MASK_DIV_ZERO (0x0200)
944 #define _MM_MASK_OVERFLOW (0x0400)
945 #define _MM_MASK_UNDERFLOW (0x0800)
946 #define _MM_MASK_INEXACT (0x1000)
947 #define _MM_MASK_MASK (0x1f80)
949 #define _MM_ROUND_NEAREST (0x0000)
950 #define _MM_ROUND_DOWN (0x2000)
951 #define _MM_ROUND_UP (0x4000)
952 #define _MM_ROUND_TOWARD_ZERO (0x6000)
953 #define _MM_ROUND_MASK (0x6000)
955 #define _MM_FLUSH_ZERO_MASK (0x8000)
956 #define _MM_FLUSH_ZERO_ON (0x8000)
957 #define _MM_FLUSH_ZERO_OFF (0x0000)
959 #define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK)
960 #define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK)
961 #define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK)
962 #define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK)
964 #define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x)))
965 #define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x)))
966 #define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x)))
967 #define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x)))
969 #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
971 __m128 tmp3, tmp2, tmp1, tmp0; \
972 tmp0 = _mm_unpacklo_ps((row0), (row1)); \
973 tmp2 = _mm_unpacklo_ps((row2), (row3)); \
974 tmp1 = _mm_unpackhi_ps((row0), (row1)); \
975 tmp3 = _mm_unpackhi_ps((row2), (row3)); \
976 (row0) = _mm_movelh_ps(tmp0, tmp2); \
977 (row1) = _mm_movehl_ps(tmp2, tmp0); \
978 (row2) = _mm_movelh_ps(tmp1, tmp3); \
979 (row3) = _mm_movehl_ps(tmp3, tmp1); \
983 #define _m_pextrw _mm_extract_pi16
984 #define _m_pinsrw _mm_insert_pi16
985 #define _m_pmaxsw _mm_max_pi16
986 #define _m_pmaxub _mm_max_pu8
987 #define _m_pminsw _mm_min_pi16
988 #define _m_pminub _mm_min_pu8
989 #define _m_pmovmskb _mm_movemask_pi8
990 #define _m_pmulhuw _mm_mulhi_pu16
991 #define _m_pshufw _mm_shuffle_pi16
992 #define _m_maskmovq _mm_maskmove_si64
993 #define _m_pavgb _mm_avg_pu8
994 #define _m_pavgw _mm_avg_pu16
995 #define _m_psadbw _mm_sad_pu8
999 #undef __DEFAULT_FN_ATTRS
1002 #if defined(__SSE2__) && !__has_feature(modules)
char __v64qi __attribute__((__vector_size__(64)))
static vector float vector float __b
#define __DEFAULT_FN_ATTRS