Lines Matching refs:B
43 # define eq(x, y) (B(cmpph, _mask, x, y, 0, -1) == ALL_TRUE)
45 # define eq(x, y) (B(pcmpeqb, _mask, (vqi_t)(x), (vqi_t)(y), -1) == ALL_TRUE)
47 # define eq(x, y) (B(pcmpeqw, _mask, (vhi_t)(x), (vhi_t)(y), -1) == ALL_TRUE)
49 # define eq(x, y) (B(pcmpeqd, _mask, (vsi_t)(x), (vsi_t)(y), -1) == ALL_TRUE)
51 # define eq(x, y) (B(pcmpeqq, _mask, (vdi_t)(x), (vdi_t)(y), -1) == ALL_TRUE)
168 # define to_int(x) B(cvtdq2pd, _mask, BR(cvtpd2dq, _mask, x, (vsi_half_t){}, ~0), undef(), ~0)
169 # define to_uint(x) B(cvtudq2pd, _mask, BR(cvtpd2udq, _mask, x, (vsi_half_t){}, ~0), undef(), ~0)
183 # define _half(x, lh) ((vhf_half_t)B(extracti32x4_, _mask, (vsi_t)(x), lh, (vsi_half_t){}, ~0))
187 ((vec_t)B(inserti32x4_, _mask, (vsi_t)(x), (vsi_half_t)(y), p, (vsi_t)undef(), ~0))
348 # define broadcast_octet(x) B(broadcastf32x8_, _mask, x, undef(), ~0)
349 # define insert_octet(x, y, p) B(insertf32x8_, _mask, x, y, p, undef(), ~0)
352 # define frac(x) B(reduceps, _mask, x, 0b00001011, undef(), ~0)
363 # define mix(x, y) B(blendmps_, _mask, x, y, (0b1010101010101010 & ALL_TRUE))
369 # define recip(x) B(rcp14ps, _mask, x, undef(), ~0)
370 # define rsqrt(x) B(rsqrt14ps, _mask, x, undef(), ~0)
377 # define interleave_hi(x, y) B(unpckhps, _mask, x, y, undef(), ~0)
378 # define interleave_lo(x, y) B(unpcklps, _mask, x, y, undef(), ~0)
379 # define swap(x) B(shufps, _mask, x, x, 0b00011011, undef(), ~0)
382 # define broadcast_quartet(x) B(broadcastf32x4_, _mask, x, undef(), ~0)
384 B(insertf32x4_, _mask, x, \
388 # define insert_quartet(x, y, p) B(insertf32x4_, _mask, x, y, p, undef(), ~0)
389 # define interleave_hi(x, y) B(vpermi2varps, _mask, x, interleave_hi, y, ~0)
390 # define interleave_lo(x, y) B(vpermt2varps, _mask, interleave_lo, x, y, ~0)
392 vec_t t_ = B(shuf_f32x4_, _mask, x, x, VEC_SIZE == 32 ? 0b01 : 0b00011011, undef(), ~0); \
393 B(shufps, _mask, t_, t_, 0b00011011, undef(), ~0); \
395 # define swap2(x) B(vpermilps, _mask, \
396 B(shuf_f32x4_, _mask, x, x, \
417 # define broadcast_pair(x) B(broadcastf64x2_, _mask, x, undef(), ~0)
418 # define insert_pair(x, y, p) B(insertf64x2_, _mask, x, y, p, undef(), ~0)
421 # define broadcast_quartet(x) B(broadcastf64x4_, , x, undef(), ~0)
422 # define insert_quartet(x, y, p) B(insertf64x4_, _mask, x, y, p, undef(), ~0)
425 # define frac(x) B(reducepd, _mask, x, 0b00001011, undef(), ~0)
436 # define mix(x, y) B(blendmpd_, _mask, x, y, 0b10101010)
442 # define recip(x) B(rcp14pd, _mask, x, undef(), ~0)
443 # define rsqrt(x) B(rsqrt14pd, _mask, x, undef(), ~0)
448 # define interleave_hi(x, y) B(unpckhpd, _mask, x, y, undef(), ~0)
449 # define interleave_lo(x, y) B(unpcklpd, _mask, x, y, undef(), ~0)
450 # define swap(x) B(shufpd, _mask, x, x, 0b01, undef(), ~0)
453 # define interleave_hi(x, y) B(vpermi2varpd, _mask, x, interleave_hi, y, ~0)
454 # define interleave_lo(x, y) B(vpermt2varpd, _mask, interleave_lo, x, y, ~0)
456 vec_t t_ = B(shuf_f64x2_, _mask, x, x, VEC_SIZE == 32 ? 0b01 : 0b00011011, undef(), ~0); \
457 B(shufpd, _mask, t_, t_, 0b01010101, undef(), ~0); \
459 # define swap2(x) B(vpermilpd, _mask, \
460 B(shuf_f64x2_, _mask, x, x, \
471 # define recip(x) B(rcpph, _mask, x, undef(), ~0)
472 # define rsqrt(x) B(rsqrtph, _mask, x, undef(), ~0)
621 # define broadcast_octet(x) ((vec_t)B(broadcasti32x8_, _mask, (vsi_octet_t)(x), (vsi_t)undef(), ~…
622 # define insert_octet(x, y, p) ((vec_t)B(inserti32x8_, _mask, (vsi_t)(x), (vsi_octet_t)(y), p, (v…
625 # define interleave_hi(x, y) ((vec_t)B(punpckhdq, _mask, (vsi_t)(x), (vsi_t)(y), (vsi_t)undef(), …
626 # define interleave_lo(x, y) ((vec_t)B(punpckldq, _mask, (vsi_t)(x), (vsi_t)(y), (vsi_t)undef(), …
627 # define swap(x) ((vec_t)B(pshufd, _mask, (vsi_t)(x), 0b00011011, (vsi_t)undef(), ~0))
629 # define broadcast_quartet(x) ((vec_t)B(broadcasti32x4_, _mask, (vsi_quartet_t)(x), (vsi_t)undef(…
631 (vec_t)(B(inserti32x4_, _mask, (vsi_t)(x), \
636 # define insert_quartet(x, y, p) ((vec_t)B(inserti32x4_, _mask, (vsi_t)(x), (vsi_quartet_t)(y), p…
637 # define interleave_hi(x, y) ((vec_t)B(vpermi2vard, _mask, (vsi_t)(x), interleave_hi, (vsi_t)(y),…
638 # define interleave_lo(x, y) ((vec_t)B(vpermt2vard, _mask, interleave_lo, (vsi_t)(x), (vsi_t)(y),…
639 # define swap(x) ((vec_t)B(pshufd, _mask, \
640 B(shuf_i32x4_, _mask, (vsi_t)(x), (vsi_t)(x), \
645 # define mix(x, y) ((vec_t)B(blendmd_, _mask, (vsi_t)(x), (vsi_t)(y), \
647 # define rotr(x, n) ((vec_t)B(alignd, _mask, (vsi_t)(x), (vsi_t)(x), n, (vsi_t)undef(), ~0))
648 # define shrink1(x) ((half_t)B(pmovqd, _mask, (vdi_t)(x), (vsi_half_t){}, ~0))
664 # define broadcast_pair(x) ((vec_t)B(broadcasti64x2_, _mask, (vdi_pair_t)(x), (vdi_t)undef(), ~0))
665 # define insert_pair(x, y, p) ((vec_t)B(inserti64x2_, _mask, (vdi_t)(x), (vdi_pair_t)(y), p, (vdi…
668 # define broadcast_quartet(x) ((vec_t)B(broadcasti64x4_, , (vdi_quartet_t)(x), (vdi_t)undef(), ~0…
669 # define insert_quartet(x, y, p) ((vec_t)B(inserti64x4_, _mask, (vdi_t)(x), (vdi_quartet_t)(y), p…
672 # define interleave_hi(x, y) ((vec_t)B(punpckhqdq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(),…
673 # define interleave_lo(x, y) ((vec_t)B(punpcklqdq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(),…
674 # define swap(x) ((vec_t)B(pshufd, _mask, (vsi_t)(x), 0b01001110, (vsi_t)undef(), ~0))
676 # define interleave_hi(x, y) ((vec_t)B(vpermi2varq, _mask, (vdi_t)(x), interleave_hi, (vdi_t)(y),…
677 # define interleave_lo(x, y) ((vec_t)B(vpermt2varq, _mask, interleave_lo, (vdi_t)(x), (vdi_t)(y),…
678 # define swap(x) ((vec_t)B(pshufd, _mask, \
679 (vsi_t)B(shuf_i64x2_, _mask, (vdi_t)(x), (vdi_t)(x), \
682 # define swap2(x) ((vec_t)B(permvardi, _mask, (vdi_t)(x), (vdi_t)(inv - 1), (vdi_t)undef(), ~0))
684 # define mix(x, y) ((vec_t)B(blendmq_, _mask, (vdi_t)(x), (vdi_t)(y), 0b10101010))
685 # define rotr(x, n) ((vec_t)B(alignq, _mask, (vdi_t)(x), (vdi_t)(x), n, (vdi_t)undef(), ~0))
691 B(shuf_i64x2_, _mask, t_, t_, 0b01001110, (vdi_t)undef(), ~0); \
696 # define abs(x) B(pabsd, _mask, x, undef(), ~0)
697 # define max(x, y) B(pmaxsd, _mask, x, y, undef(), ~0)
698 # define min(x, y) B(pminsd, _mask, x, y, undef(), ~0)
699 # define mul_full(x, y) ((vec_t)B(pmuldq, _mask, x, y, (vdi_t)undef(), ~0))
700 # define widen1(x) ((vec_t)B(pmovsxdq, _mask, x, (vdi_t)undef(), ~0))
702 # define max(x, y) ((vec_t)B(pmaxud, _mask, (vsi_t)(x), (vsi_t)(y), (vsi_t)undef(), ~0))
703 # define min(x, y) ((vec_t)B(pminud, _mask, (vsi_t)(x), (vsi_t)(y), (vsi_t)undef(), ~0))
704 # define mul_full(x, y) ((vec_t)B(pmuludq, _mask, (vsi_t)(x), (vsi_t)(y), (vdi_t)undef(), ~0))
705 # define widen1(x) ((vec_t)B(pmovzxdq, _mask, (vsi_half_t)(x), (vdi_t)undef(), ~0))
707 # define abs(x) ((vec_t)B(pabsq, _mask, (vdi_t)(x), (vdi_t)undef(), ~0))
708 # define max(x, y) ((vec_t)B(pmaxsq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(), ~0))
709 # define min(x, y) ((vec_t)B(pminsq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(), ~0))
711 # define max(x, y) ((vec_t)B(pmaxuq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(), ~0))
712 # define min(x, y) ((vec_t)B(pminuq, _mask, (vdi_t)(x), (vdi_t)(y), (vdi_t)undef(), ~0))
729 # define interleave_hi(x, y) ((vec_t)B(punpckhbw, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), …
730 # define interleave_lo(x, y) ((vec_t)B(punpcklbw, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), …
731 # define rotr(x, n) ((vec_t)B(palignr, _mask, (vdi_t)(x), (vdi_t)(x), (n) * 8, (vdi_t)undef(), ~0…
732 # define swap(x) ((vec_t)B(pshufb, _mask, (vqi_t)(x), (vqi_t)(inv - 1), (vqi_t)undef(), ~0))
734 # define interleave_hi(x, y) ((vec_t)B(vpermi2varqi, _mask, (vqi_t)(x), interleave_hi, (vqi_t)(y)…
735 # define interleave_lo(x, y) ((vec_t)B(vpermt2varqi, _mask, interleave_lo, (vqi_t)(x), (vqi_t)(y)…
737 # define mix(x, y) ((vec_t)B(blendmb_, _mask, (vqi_t)(x), (vqi_t)(y), \
739 # define shrink1(x) ((half_t)B(pmovwb, _mask, (vhi_t)(x), (vqi_half_t){}, ~0))
740 # define shrink2(x) ((quarter_t)B(pmovdb, _mask, (vsi_t)(x), (vqi_quarter_t){}, ~0))
741 # define shrink3(x) ((eighth_t)B(pmovqb, _mask, (vdi_t)(x), (vqi_eighth_t){}, ~0))
743 # define swap2(x) ((vec_t)B(permvarqi, _mask, (vqi_t)(x), (vqi_t)(inv - 1), (vqi_t)undef(), ~0))
758 # define interleave_hi(x, y) ((vec_t)B(punpckhwd, _mask, (vhi_t)(x), (vhi_t)(y), (vhi_t)undef(), …
759 # define interleave_lo(x, y) ((vec_t)B(punpcklwd, _mask, (vhi_t)(x), (vhi_t)(y), (vhi_t)undef(), …
760 # define rotr(x, n) ((vec_t)B(palignr, _mask, (vdi_t)(x), (vdi_t)(x), (n) * 16, (vdi_t)undef(), ~…
761 # define swap(x) ((vec_t)B(pshufd, _mask, \
762 (vsi_t)B(pshufhw, _mask, \
763 … B(pshuflw, _mask, (vhi_t)(x), 0b00011011, (vhi_t)undef(), ~0), \
767 # define interleave_hi(x, y) ((vec_t)B(vpermi2varhi, _mask, (vhi_t)(x), interleave_hi, (vhi_t)(y)…
768 # define interleave_lo(x, y) ((vec_t)B(vpermt2varhi, _mask, interleave_lo, (vhi_t)(x), (vhi_t)(y)…
770 # define mix(x, y) ((vec_t)B(blendmw_, _mask, (vhi_t)(x), (vhi_t)(y), \
772 # define shrink1(x) ((half_t)B(pmovdw, _mask, (vsi_t)(x), (vhi_half_t){}, ~0))
773 # define shrink2(x) ((quarter_t)B(pmovqw, _mask, (vdi_t)(x), (vhi_quarter_t){}, ~0))
774 # define swap2(x) ((vec_t)B(permvarhi, _mask, (vhi_t)(x), (vhi_t)(inv - 1), (vhi_t)undef(), ~0))
777 # define abs(x) ((vec_t)B(pabsb, _mask, (vqi_t)(x), (vqi_t)undef(), ~0))
778 # define max(x, y) ((vec_t)B(pmaxsb, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), ~0))
779 # define min(x, y) ((vec_t)B(pminsb, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), ~0))
780 # define widen1(x) ((vec_t)B(pmovsxbw, _mask, (vqi_half_t)(x), (vhi_t)undef(), ~0))
781 # define widen2(x) ((vec_t)B(pmovsxbd, _mask, (vqi_quarter_t)(x), (vsi_t)undef(), ~0))
782 # define widen3(x) ((vec_t)B(pmovsxbq, _mask, (vqi_eighth_t)(x), (vdi_t)undef(), ~0))
784 # define max(x, y) ((vec_t)B(pmaxub, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), ~0))
785 # define min(x, y) ((vec_t)B(pminub, _mask, (vqi_t)(x), (vqi_t)(y), (vqi_t)undef(), ~0))
786 # define widen1(x) ((vec_t)B(pmovzxbw, _mask, (vqi_half_t)(x), (vhi_t)undef(), ~0))
787 # define widen2(x) ((vec_t)B(pmovzxbd, _mask, (vqi_quarter_t)(x), (vsi_t)undef(), ~0))
788 # define widen3(x) ((vec_t)B(pmovzxbq, _mask, (vqi_eighth_t)(x), (vdi_t)undef(), ~0))
790 # define abs(x) B(pabsw, _mask, x, undef(), ~0)
791 # define max(x, y) B(pmaxsw, _mask, x, y, undef(), ~0)
792 # define min(x, y) B(pminsw, _mask, x, y, undef(), ~0)
793 # define mul_hi(x, y) B(pmulhw, _mask, x, y, undef(), ~0)
794 # define widen1(x) ((vec_t)B(pmovsxwd, _mask, x, (vsi_t)undef(), ~0))
795 # define widen2(x) ((vec_t)B(pmovsxwq, _mask, x, (vdi_t)undef(), ~0))
797 # define max(x, y) ((vec_t)B(pmaxuw, _mask, (vhi_t)(x), (vhi_t)(y), (vhi_t)undef(), ~0))
798 # define min(x, y) ((vec_t)B(pminuw, _mask, (vhi_t)(x), (vhi_t)(y), (vhi_t)undef(), ~0))
799 # define mul_hi(x, y) ((vec_t)B(pmulhuw, _mask, (vhi_t)(x), (vhi_t)(y), (vhi_t)undef(), ~0))
800 # define widen1(x) ((vec_t)B(pmovzxwd, _mask, (vhi_half_t)(x), (vsi_t)undef(), ~0))
801 # define widen2(x) ((vec_t)B(pmovzxwq, _mask, (vhi_quarter_t)(x), (vdi_t)undef(), ~0))
995 # define dup_hi(x) B(movshdup, _mask, x, undef(), ~0)
996 # define dup_lo(x) B(movsldup, _mask, x, undef(), ~0)
998 # define dup_lo(x) B(movddup, _mask, x, undef(), ~0)
1002 # define dup_hi(x) ((vec_t)B(pshufhw, _mask, \
1003 B(pshuflw, _mask, (vhi_t)(x), 0b11110101, \
1006 # define dup_lo(x) ((vec_t)B(pshufhw, _mask, \
1007 B(pshuflw, _mask, (vhi_t)(x), 0b10100000, \