3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
12 # Montgomery multiplication routine for x86_64. While it gives modest
13 # 9% improvement of rsa4096 sign on Opteron, rsa512 sign runs more
14 # than twice, >2x, as fast. Most common rsa1024 sign is improved by
15 # respectful 50%. It remains to be seen if loop unrolling and
16 # dedicated squaring routine can provide further improvement...
20 # Add dedicated squaring procedure. Performance improvement varies
21 # from platform to platform, but in average it's ~5%/15%/25%/33%
22 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
26 # Unroll and modulo-schedule inner loops in such manner that they
27 # are "fallen through" for input lengths of 8, which is critical for
28 # 1024-bit RSA *sign*. Average performance improvement in comparison
29 # to *initial* version of this module from 2005 is ~0%/30%/40%/45%
30 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
34 # Optimize reduction in squaring procedure and improve 1024+-bit RSA
35 # sign performance by 10-16% on Intel Sandy Bridge and later
36 # (virtually same on non-Intel processors).
40 # Add MULX/ADOX/ADCX code path.
44 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
46 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
48 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
49 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
50 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
51 die "can't locate x86_64-xlate.pl";
53 open OUT,"| \"$^X\" $xlate $flavour $output";
56 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
57 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
61 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
62 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
66 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
67 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
71 if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9])\.([0-9]+)/) {
72 my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
77 $rp="%rdi"; # BN_ULONG *rp,
78 $ap="%rsi"; # const BN_ULONG *ap,
79 $bp="%rdx"; # const BN_ULONG *bp,
80 $np="%rcx"; # const BN_ULONG *np,
81 $n0="%r8"; # const BN_ULONG *n0,
82 $num="%r9"; # int num);
94 .extern OPENSSL_ia32cap_P
97 .type bn_mul_mont,\@function,6
105 $code.=<<___ if ($addx);
106 mov OPENSSL_ia32cap_P+8(%rip),%r11d
128 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
129 and \$-1024,%rsp # minimize TLB usage
131 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
133 mov $bp,%r12 # reassign $bp
137 mov ($n0),$n0 # pull n0[0] value
138 mov ($bp),$m0 # m0=bp[0]
145 mulq $m0 # ap[0]*bp[0]
149 imulq $lo0,$m1 # "tp[0]"*n0
153 add %rax,$lo0 # discarded
166 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
169 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
173 mulq $m0 # ap[j]*bp[0]
185 mov ($ap),%rax # ap[0]
187 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
189 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
196 mov $hi1,-8(%rsp,$num,8)
197 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
203 mov ($bp,$i,8),$m0 # m0=bp[i]
207 mulq $m0 # ap[0]*bp[i]
208 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
212 imulq $lo0,$m1 # tp[0]*n0
216 add %rax,$lo0 # discarded
219 mov 8(%rsp),$lo0 # tp[1]
230 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
233 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
237 mulq $m0 # ap[j]*bp[i]
241 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
251 mov ($ap),%rax # ap[0]
253 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
256 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
262 add $lo0,$hi1 # pull upmost overflow bit
264 mov $hi1,-8(%rsp,$num,8)
265 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
271 xor $i,$i # i=0 and clear CF!
272 mov (%rsp),%rax # tp[0]
273 lea (%rsp),$ap # borrow ap for tp
277 .Lsub: sbb ($np,$i,8),%rax
278 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
279 mov 8($ap,$i,8),%rax # tp[i+1]
281 dec $j # doesnn't affect CF!
284 sbb \$0,%rax # handle upmost overflow bit
291 or $np,$ap # ap=borrow?tp:rp
293 .Lcopy: # copy or in-place refresh
295 mov $i,(%rsp,$i,8) # zap temporary vector
296 mov %rax,($rp,$i,8) # rp[i]=tp[i]
301 mov 8(%rsp,$num,8),%rsi # restore %rsp
312 .size bn_mul_mont,.-bn_mul_mont
315 my @A=("%r10","%r11");
316 my @N=("%r13","%rdi");
318 .type bn_mul4x_mont,\@function,6
323 $code.=<<___ if ($addx);
340 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
341 and \$-1024,%rsp # minimize TLB usage
343 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
345 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
346 mov %rdx,%r12 # reassign $bp
350 mov ($n0),$n0 # pull n0[0] value
351 mov ($bp),$m0 # m0=bp[0]
358 mulq $m0 # ap[0]*bp[0]
362 imulq $A[0],$m1 # "tp[0]"*n0
366 add %rax,$A[0] # discarded
389 mulq $m0 # ap[j]*bp[0]
391 mov -16($np,$j,8),%rax
397 mov -8($ap,$j,8),%rax
399 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
401 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
404 mulq $m0 # ap[j]*bp[0]
406 mov -8($np,$j,8),%rax
414 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
416 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
419 mulq $m0 # ap[j]*bp[0]
429 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
431 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
434 mulq $m0 # ap[j]*bp[0]
443 mov -16($ap,$j,8),%rax
445 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
447 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
452 mulq $m0 # ap[j]*bp[0]
454 mov -16($np,$j,8),%rax
460 mov -8($ap,$j,8),%rax
462 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
464 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
467 mulq $m0 # ap[j]*bp[0]
469 mov -8($np,$j,8),%rax
475 mov ($ap),%rax # ap[0]
477 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
479 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
485 mov $N[0],-8(%rsp,$j,8)
486 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
491 mov ($bp,$i,8),$m0 # m0=bp[i]
495 mulq $m0 # ap[0]*bp[i]
496 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
500 imulq $A[0],$m1 # tp[0]*n0
504 add %rax,$A[0] # "$N[0]", discarded
509 mulq $m0 # ap[j]*bp[i]
513 add 8(%rsp),$A[1] # +tp[1]
521 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
524 mov $N[1],(%rsp) # tp[j-1]
529 mulq $m0 # ap[j]*bp[i]
531 mov -16($np,$j,8),%rax
533 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
539 mov -8($ap,$j,8),%rax
543 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
546 mulq $m0 # ap[j]*bp[i]
548 mov -8($np,$j,8),%rax
550 add -8(%rsp,$j,8),$A[1]
560 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
563 mulq $m0 # ap[j]*bp[i]
567 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
577 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
580 mulq $m0 # ap[j]*bp[i]
584 add 8(%rsp,$j,8),$A[1]
591 mov -16($ap,$j,8),%rax
595 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
600 mulq $m0 # ap[j]*bp[i]
602 mov -16($np,$j,8),%rax
604 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
610 mov -8($ap,$j,8),%rax
614 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
617 mulq $m0 # ap[j]*bp[i]
619 mov -8($np,$j,8),%rax
621 add -8(%rsp,$j,8),$A[1]
628 mov ($ap),%rax # ap[0]
632 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
638 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
640 mov $N[0],-8(%rsp,$j,8)
641 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
647 my @ri=("%rax","%rdx",$m0,$m1);
649 mov 16(%rsp,$num,8),$rp # restore $rp
650 mov 0(%rsp),@ri[0] # tp[0]
652 mov 8(%rsp),@ri[1] # tp[1]
653 shr \$2,$num # num/=4
654 lea (%rsp),$ap # borrow ap for tp
655 xor $i,$i # i=0 and clear CF!
658 mov 16($ap),@ri[2] # tp[2]
659 mov 24($ap),@ri[3] # tp[3]
661 lea -1($num),$j # j=num/4-1
665 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
666 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
667 sbb 16($np,$i,8),@ri[2]
668 mov 32($ap,$i,8),@ri[0] # tp[i+1]
669 mov 40($ap,$i,8),@ri[1]
670 sbb 24($np,$i,8),@ri[3]
671 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
672 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
673 sbb 32($np,$i,8),@ri[0]
674 mov 48($ap,$i,8),@ri[2]
675 mov 56($ap,$i,8),@ri[3]
676 sbb 40($np,$i,8),@ri[1]
678 dec $j # doesnn't affect CF!
681 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
682 mov 32($ap,$i,8),@ri[0] # load overflow bit
683 sbb 16($np,$i,8),@ri[2]
684 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
685 sbb 24($np,$i,8),@ri[3]
686 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
688 sbb \$0,@ri[0] # handle upmost overflow bit
689 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
696 or $np,$ap # ap=borrow?tp:rp
703 .Lcopy4x: # copy or in-place refresh
704 movdqu 16($ap,$i),%xmm2
705 movdqu 32($ap,$i),%xmm1
706 movdqa %xmm0,16(%rsp,$i)
707 movdqu %xmm2,16($rp,$i)
708 movdqa %xmm0,32(%rsp,$i)
709 movdqu %xmm1,32($rp,$i)
715 movdqu 16($ap,$i),%xmm2
716 movdqa %xmm0,16(%rsp,$i)
717 movdqu %xmm2,16($rp,$i)
721 mov 8(%rsp,$num,8),%rsi # restore %rsp
732 .size bn_mul4x_mont,.-bn_mul4x_mont
736 ######################################################################
737 # void bn_sqr8x_mont(
738 my $rptr="%rdi"; # const BN_ULONG *rptr,
739 my $aptr="%rsi"; # const BN_ULONG *aptr,
740 my $bptr="%rdx"; # not used
741 my $nptr="%rcx"; # const BN_ULONG *nptr,
742 my $n0 ="%r8"; # const BN_ULONG *n0);
743 my $num ="%r9"; # int num, has to be divisible by 8
745 my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
746 my @A0=("%r10","%r11");
747 my @A1=("%r12","%r13");
748 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
750 $code.=<<___ if ($addx);
751 .extern bn_sqrx8x_internal # see x86_64-mont5 module
754 .extern bn_sqr8x_internal # see x86_64-mont5 module
756 .type bn_sqr8x_mont,\@function,6
769 shl \$3,${num}d # convert $num to bytes
770 shl \$3+2,%r10 # 4*$num
773 ##############################################################
774 # ensure that stack frame doesn't alias with $aptr modulo
775 # 4096. this is done to allow memory disambiguation logic
778 lea -64(%rsp,$num,2),%r11
784 sub %r11,%rsp # align with $aptr
785 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
790 lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
791 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
802 mov %rax, 40(%rsp) # save original %rsp
805 movq $nptr, %xmm2 # save pointer to modulus
807 movq $rptr,%xmm1 # save $rptr
808 movq %r10, %xmm3 # -$num
810 $code.=<<___ if ($addx);
811 mov OPENSSL_ia32cap_P+8(%rip),%eax
816 call bn_sqrx8x_internal # see x86_64-mont5 module
817 # %rax top-most carry
820 # %r8 end of tp[2*num]
825 sar \$3+2,%rcx # %cf=0
832 call bn_sqr8x_internal # see x86_64-mont5 module
833 # %rax top-most carry
836 # %rdi end of tp[2*num]
841 sar \$3+2,%rcx # %cf=0
861 inc %rcx # preserves %cf
864 sbb \$0,%rax # top-most carry
865 lea (%rbx,$num),%rbx # rewind
866 lea ($rptr,$num),$rptr # rewind
870 pshufd \$0,%xmm1,%xmm1
871 mov 40(%rsp),%rsi # restore %rsp
872 jmp .Lsqr8x_cond_copy
876 movdqa 16*0(%rbx),%xmm2
877 movdqa 16*1(%rbx),%xmm3
879 movdqu 16*0($rptr),%xmm4
880 movdqu 16*1($rptr),%xmm5
881 lea 16*2($rptr),$rptr
882 movdqa %xmm0,-16*2(%rbx) # zero tp
883 movdqa %xmm0,-16*1(%rbx)
884 movdqa %xmm0,-16*2(%rbx,%rdx)
885 movdqa %xmm0,-16*1(%rbx,%rdx)
894 movdqu %xmm4,-16*2($rptr)
895 movdqu %xmm5,-16*1($rptr)
897 jnz .Lsqr8x_cond_copy
909 .size bn_sqr8x_mont,.-bn_sqr8x_mont
914 my $bp="%rdx"; # original value
917 .type bn_mulx4x_mont,\@function,6
929 shl \$3,${num}d # convert $num to bytes
932 sub $num,%r10 # -$num
934 lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
937 ##############################################################
940 # +8 off-loaded &b[i]
949 mov $num,0(%rsp) # save $num
951 mov %r10,16(%rsp) # end of b[num]
953 mov $n0, 24(%rsp) # save *n0
954 mov $rp, 32(%rsp) # save $rp
955 mov %rax,40(%rsp) # save original %rsp
956 mov $num,48(%rsp) # inner counter
962 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
963 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
967 mov ($bp),%rdx # b[0], $bp==%rdx actually
968 lea 64+32(%rsp),$tptr
971 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
972 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
974 mov $bptr,8(%rsp) # off-load &b[i]
975 mulx 2*8($aptr),%r12,%r13 # ...
979 mov $mi,$bptr # borrow $bptr
980 imulq 24(%rsp),$mi # "t[0]"*n0
981 xor $zero,$zero # cf=0, of=0
983 mulx 3*8($aptr),%rax,%r14
987 adcx $zero,%r14 # cf=0
989 mulx 0*8($nptr),%rax,%r10
990 adcx %rax,$bptr # discarded
992 mulx 1*8($nptr),%rax,%r11
995 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
996 mov 48(%rsp),$bptr # counter value
1000 mulx 3*8($nptr),%rax,%r15
1002 mov %r11,-3*8($tptr)
1004 adox $zero,%r15 # of=0
1005 lea 4*8($nptr),$nptr
1006 mov %r12,-2*8($tptr)
1012 adcx $zero,%r15 # cf=0, modulo-scheduled
1013 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
1015 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
1017 mulx 2*8($aptr),%r12,%rax # ...
1019 mulx 3*8($aptr),%r13,%r14
1023 adcx $zero,%r14 # cf=0
1024 lea 4*8($aptr),$aptr
1025 lea 4*8($tptr),$tptr
1028 mulx 0*8($nptr),%rax,%r15
1031 mulx 1*8($nptr),%rax,%r15
1034 mulx 2*8($nptr),%rax,%r15
1035 mov %r10,-5*8($tptr)
1037 mov %r11,-4*8($tptr)
1039 mulx 3*8($nptr),%rax,%r15
1041 mov %r12,-3*8($tptr)
1044 lea 4*8($nptr),$nptr
1045 mov %r13,-2*8($tptr)
1047 dec $bptr # of=0, pass cf
1050 mov 0(%rsp),$num # load num
1051 mov 8(%rsp),$bptr # re-load &b[i]
1052 adc $zero,%r15 # modulo-scheduled
1054 sbb %r15,%r15 # top-most carry
1055 mov %r14,-1*8($tptr)
1060 mov ($bptr),%rdx # b[i]
1061 lea 8($bptr),$bptr # b++
1062 sub $num,$aptr # rewind $aptr
1063 mov %r15,($tptr) # save top-most carry
1064 lea 64+4*8(%rsp),$tptr
1065 sub $num,$nptr # rewind $nptr
1067 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
1068 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1070 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
1071 adox -4*8($tptr),$mi
1073 mulx 2*8($aptr),%r15,%r13 # ...
1074 adox -3*8($tptr),%r11
1079 mov $bptr,8(%rsp) # off-load &b[i]
1082 imulq 24(%rsp),$mi # "t[0]"*n0
1083 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1085 mulx 3*8($aptr),%rax,%r14
1087 adox -2*8($tptr),%r12
1089 adox -1*8($tptr),%r13
1091 lea 4*8($aptr),$aptr
1094 mulx 0*8($nptr),%rax,%r10
1095 adcx %rax,%r15 # discarded
1097 mulx 1*8($nptr),%rax,%r11
1100 mulx 2*8($nptr),%rax,%r12
1101 mov %r10,-4*8($tptr)
1104 mulx 3*8($nptr),%rax,%r15
1106 mov %r11,-3*8($tptr)
1107 lea 4*8($nptr),$nptr
1109 adox $zero,%r15 # of=0
1110 mov 48(%rsp),$bptr # counter value
1111 mov %r12,-2*8($tptr)
1117 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1118 adcx $zero,%r15 # cf=0, modulo-scheduled
1120 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1121 adcx 0*8($tptr),%r10
1123 mulx 2*8($aptr),%r12,%rax # ...
1124 adcx 1*8($tptr),%r11
1126 mulx 3*8($aptr),%r13,%r14
1128 adcx 2*8($tptr),%r12
1130 adcx 3*8($tptr),%r13
1131 adox $zero,%r14 # of=0
1132 lea 4*8($aptr),$aptr
1133 lea 4*8($tptr),$tptr
1134 adcx $zero,%r14 # cf=0
1137 mulx 0*8($nptr),%rax,%r15
1140 mulx 1*8($nptr),%rax,%r15
1143 mulx 2*8($nptr),%rax,%r15
1144 mov %r10,-5*8($tptr)
1147 mulx 3*8($nptr),%rax,%r15
1149 mov %r11,-4*8($tptr)
1150 mov %r12,-3*8($tptr)
1153 lea 4*8($nptr),$nptr
1154 mov %r13,-2*8($tptr)
1156 dec $bptr # of=0, pass cf
1159 mov 0(%rsp),$num # load num
1160 mov 8(%rsp),$bptr # re-load &b[i]
1161 adc $zero,%r15 # modulo-scheduled
1162 sub 0*8($tptr),$zero # pull top-most carry
1164 sbb %r15,%r15 # top-most carry
1165 mov %r14,-1*8($tptr)
1171 sub $num,$nptr # rewind $nptr
1174 shr \$3+2,$num # %cf=0
1175 mov 32(%rsp),$rptr # restore rp
1184 lea 8*4($tptr),$tptr
1189 lea 8*4($nptr),$nptr
1194 lea 8*4($rptr),$rptr
1195 dec $num # preserves %cf
1198 sbb \$0,%r15 # top-most carry
1200 sub %rdx,$rptr # rewind
1204 pshufd \$0,%xmm1,%xmm1
1205 mov 40(%rsp),%rsi # restore %rsp
1206 jmp .Lmulx4x_cond_copy
1210 movdqa 16*0($tptr),%xmm2
1211 movdqa 16*1($tptr),%xmm3
1212 lea 16*2($tptr),$tptr
1213 movdqu 16*0($rptr),%xmm4
1214 movdqu 16*1($rptr),%xmm5
1215 lea 16*2($rptr),$rptr
1216 movdqa %xmm0,-16*2($tptr) # zero tp
1217 movdqa %xmm0,-16*1($tptr)
1226 movdqu %xmm4,-16*2($rptr)
1227 movdqu %xmm5,-16*1($rptr)
1229 jnz .Lmulx4x_cond_copy
1243 .size bn_mulx4x_mont,.-bn_mulx4x_mont
1247 .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1251 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1252 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1260 .extern __imp_RtlVirtualUnwind
1261 .type mul_handler,\@abi-omnipotent
1275 mov 120($context),%rax # pull context->Rax
1276 mov 248($context),%rbx # pull context->Rip
1278 mov 8($disp),%rsi # disp->ImageBase
1279 mov 56($disp),%r11 # disp->HandlerData
1281 mov 0(%r11),%r10d # HandlerData[0]
1282 lea (%rsi,%r10),%r10 # end of prologue label
1283 cmp %r10,%rbx # context->Rip<end of prologue label
1284 jb .Lcommon_seh_tail
1286 mov 152($context),%rax # pull context->Rsp
1288 mov 4(%r11),%r10d # HandlerData[1]
1289 lea (%rsi,%r10),%r10 # epilogue label
1290 cmp %r10,%rbx # context->Rip>=epilogue label
1291 jae .Lcommon_seh_tail
1293 mov 192($context),%r10 # pull $num
1294 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1303 mov %rbx,144($context) # restore context->Rbx
1304 mov %rbp,160($context) # restore context->Rbp
1305 mov %r12,216($context) # restore context->R12
1306 mov %r13,224($context) # restore context->R13
1307 mov %r14,232($context) # restore context->R14
1308 mov %r15,240($context) # restore context->R15
1310 jmp .Lcommon_seh_tail
1311 .size mul_handler,.-mul_handler
1313 .type sqr_handler,\@abi-omnipotent
1327 mov 120($context),%rax # pull context->Rax
1328 mov 248($context),%rbx # pull context->Rip
1330 mov 8($disp),%rsi # disp->ImageBase
1331 mov 56($disp),%r11 # disp->HandlerData
1333 mov 0(%r11),%r10d # HandlerData[0]
1334 lea (%rsi,%r10),%r10 # end of prologue label
1335 cmp %r10,%rbx # context->Rip<.Lsqr_body
1336 jb .Lcommon_seh_tail
1338 mov 152($context),%rax # pull context->Rsp
1340 mov 4(%r11),%r10d # HandlerData[1]
1341 lea (%rsi,%r10),%r10 # epilogue label
1342 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
1343 jae .Lcommon_seh_tail
1345 mov 40(%rax),%rax # pull saved stack pointer
1353 mov %rbx,144($context) # restore context->Rbx
1354 mov %rbp,160($context) # restore context->Rbp
1355 mov %r12,216($context) # restore context->R12
1356 mov %r13,224($context) # restore context->R13
1357 mov %r14,232($context) # restore context->R14
1358 mov %r15,240($context) # restore context->R15
1363 mov %rax,152($context) # restore context->Rsp
1364 mov %rsi,168($context) # restore context->Rsi
1365 mov %rdi,176($context) # restore context->Rdi
1367 mov 40($disp),%rdi # disp->ContextRecord
1368 mov $context,%rsi # context
1369 mov \$154,%ecx # sizeof(CONTEXT)
1370 .long 0xa548f3fc # cld; rep movsq
1373 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1374 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1375 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1376 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1377 mov 40(%rsi),%r10 # disp->ContextRecord
1378 lea 56(%rsi),%r11 # &disp->HandlerData
1379 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1380 mov %r10,32(%rsp) # arg5
1381 mov %r11,40(%rsp) # arg6
1382 mov %r12,48(%rsp) # arg7
1383 mov %rcx,56(%rsp) # arg8, (NULL)
1384 call *__imp_RtlVirtualUnwind(%rip)
1386 mov \$1,%eax # ExceptionContinueSearch
1398 .size sqr_handler,.-sqr_handler
1402 .rva .LSEH_begin_bn_mul_mont
1403 .rva .LSEH_end_bn_mul_mont
1404 .rva .LSEH_info_bn_mul_mont
1406 .rva .LSEH_begin_bn_mul4x_mont
1407 .rva .LSEH_end_bn_mul4x_mont
1408 .rva .LSEH_info_bn_mul4x_mont
1410 .rva .LSEH_begin_bn_sqr8x_mont
1411 .rva .LSEH_end_bn_sqr8x_mont
1412 .rva .LSEH_info_bn_sqr8x_mont
1414 $code.=<<___ if ($addx);
1415 .rva .LSEH_begin_bn_mulx4x_mont
1416 .rva .LSEH_end_bn_mulx4x_mont
1417 .rva .LSEH_info_bn_mulx4x_mont
1422 .LSEH_info_bn_mul_mont:
1425 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
1426 .LSEH_info_bn_mul4x_mont:
1429 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1430 .LSEH_info_bn_sqr8x_mont:
1433 .rva .Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
1435 $code.=<<___ if ($addx);
1436 .LSEH_info_bn_mulx4x_mont:
1439 .rva .Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]