10503 match(Set dst (XorV src1 src2));
10504 format %{ "vpxor $dst,$src1,$src2\t! xor vectors (64 bytes)" %}
10505 ins_encode %{
10506 int vector_len = 2;
10507 __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
10508 %}
10509 ins_pipe( pipe_slow );
10510 %}
10511
10512 instruct vxor64B_mem(vecZ dst, vecZ src, memory mem) %{
10513 predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
10514 match(Set dst (XorV src (LoadVector mem)));
10515 format %{ "vpxor $dst,$src,$mem\t! xor vectors (64 bytes)" %}
10516 ins_encode %{
10517 int vector_len = 2;
10518 __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
10519 %}
10520 ins_pipe( pipe_slow );
10521 %}
10522
|
10503 match(Set dst (XorV src1 src2));
10504 format %{ "vpxor $dst,$src1,$src2\t! xor vectors (64 bytes)" %}
10505 ins_encode %{
10506 int vector_len = 2;
10507 __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
10508 %}
10509 ins_pipe( pipe_slow );
10510 %}
10511
10512 instruct vxor64B_mem(vecZ dst, vecZ src, memory mem) %{
10513 predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
10514 match(Set dst (XorV src (LoadVector mem)));
10515 format %{ "vpxor $dst,$src,$mem\t! xor vectors (64 bytes)" %}
10516 ins_encode %{
10517 int vector_len = 2;
10518 __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
10519 %}
10520 ins_pipe( pipe_slow );
10521 %}
10522
10523 // --------------------------------- FMA --------------------------------------
10524
10525 // a * b + c
10526 instruct vfma2D_reg(vecX a, vecX b, vecX c) %{
10527 predicate(UseFMA && n->as_Vector()->length() == 2);
10528 match(Set c (FmaVD c (Binary a b)));
10529 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed2D" %}
10530 ins_cost(150);
10531 ins_encode %{
10532 int vector_len = 0;
10533 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10534 %}
10535 ins_pipe( pipe_slow );
10536 %}
10537
10538 // a * b + c
10539 instruct vfma2D_mem(vecX a, memory b, vecX c) %{
10540 predicate(UseFMA && n->as_Vector()->length() == 2);
10541 match(Set c (FmaVD c (Binary a (LoadVector b))));
10542 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed2D" %}
10543 ins_cost(150);
10544 ins_encode %{
10545 int vector_len = 0;
10546 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10547 %}
10548 ins_pipe( pipe_slow );
10549 %}
10550
10551
10552 // a * b + c
10553 instruct vfma4D_reg(vecY a, vecY b, vecY c) %{
10554 predicate(UseFMA && n->as_Vector()->length() == 4);
10555 match(Set c (FmaVD c (Binary a b)));
10556 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed4D" %}
10557 ins_cost(150);
10558 ins_encode %{
10559 int vector_len = 1;
10560 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10561 %}
10562 ins_pipe( pipe_slow );
10563 %}
10564
10565 // a * b + c
10566 instruct vfma4D_mem(vecY a, memory b, vecY c) %{
10567 predicate(UseFMA && n->as_Vector()->length() == 4);
10568 match(Set c (FmaVD c (Binary a (LoadVector b))));
10569 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed4D" %}
10570 ins_cost(150);
10571 ins_encode %{
10572 int vector_len = 1;
10573 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10574 %}
10575 ins_pipe( pipe_slow );
10576 %}
10577
10578 // a * b + c
10579 instruct vfma8D_reg(vecZ a, vecZ b, vecZ c) %{
10580 predicate(UseFMA && n->as_Vector()->length() == 8);
10581 match(Set c (FmaVD c (Binary a b)));
10582 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed8D" %}
10583 ins_cost(150);
10584 ins_encode %{
10585 int vector_len = 2;
10586 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10587 %}
10588 ins_pipe( pipe_slow );
10589 %}
10590
10591 // a * b + c
10592 instruct vfma8D_mem(vecZ a, memory b, vecZ c) %{
10593 predicate(UseFMA && n->as_Vector()->length() == 8);
10594 match(Set c (FmaVD c (Binary a (LoadVector b))));
10595 format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed8D" %}
10596 ins_cost(150);
10597 ins_encode %{
10598 int vector_len = 2;
10599 __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10600 %}
10601 ins_pipe( pipe_slow );
10602 %}
10603
10604 // a * b + c
10605 instruct vfma4F_reg(vecX a, vecX b, vecX c) %{
10606 predicate(UseFMA && n->as_Vector()->length() == 4);
10607 match(Set c (FmaVF c (Binary a b)));
10608 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed4F" %}
10609 ins_cost(150);
10610 ins_encode %{
10611 int vector_len = 0;
10612 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10613 %}
10614 ins_pipe( pipe_slow );
10615 %}
10616
10617 // a * b + c
10618 instruct vfma4F_mem(vecX a, memory b, vecX c) %{
10619 predicate(UseFMA && n->as_Vector()->length() == 4);
10620 match(Set c (FmaVF c (Binary a (LoadVector b))));
10621 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed4F" %}
10622 ins_cost(150);
10623 ins_encode %{
10624 int vector_len = 0;
10625 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10626 %}
10627 ins_pipe( pipe_slow );
10628 %}
10629
10630 // a * b + c
10631 instruct vfma8F_reg(vecY a, vecY b, vecY c) %{
10632 predicate(UseFMA && n->as_Vector()->length() == 8);
10633 match(Set c (FmaVF c (Binary a b)));
10634 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed8F" %}
10635 ins_cost(150);
10636 ins_encode %{
10637 int vector_len = 1;
10638 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10639 %}
10640 ins_pipe( pipe_slow );
10641 %}
10642
10643 // a * b + c
10644 instruct vfma8F_mem(vecY a, memory b, vecY c) %{
10645 predicate(UseFMA && n->as_Vector()->length() == 8);
10646 match(Set c (FmaVF c (Binary a (LoadVector b))));
10647 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed8F" %}
10648 ins_cost(150);
10649 ins_encode %{
10650 int vector_len = 1;
10651 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10652 %}
10653 ins_pipe( pipe_slow );
10654 %}
10655
10656 // a * b + c
10657 instruct vfma16F_reg(vecZ a, vecZ b, vecZ c) %{
10658 predicate(UseFMA && n->as_Vector()->length() == 16);
10659 match(Set c (FmaVF c (Binary a b)));
10660 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed16F" %}
10661 ins_cost(150);
10662 ins_encode %{
10663 int vector_len = 2;
10664 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
10665 %}
10666 ins_pipe( pipe_slow );
10667 %}
10668
10669 // a * b + c
10670 instruct vfma16F_mem(vecZ a, memory b, vecZ c) %{
10671 predicate(UseFMA && n->as_Vector()->length() == 16);
10672 match(Set c (FmaVF c (Binary a (LoadVector b))));
10673 format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed16F" %}
10674 ins_cost(150);
10675 ins_encode %{
10676 int vector_len = 2;
10677 __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
10678 %}
10679 ins_pipe( pipe_slow );
10680 %}
|