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]+)\./) {
72 $rp="%rdi"; # BN_ULONG *rp,
73 $ap="%rsi"; # const BN_ULONG *ap,
74 $bp="%rdx"; # const BN_ULONG *bp,
75 $np="%rcx"; # const BN_ULONG *np,
76 $n0="%r8"; # const BN_ULONG *n0,
77 $num="%r9"; # int num);
89 .extern OPENSSL_ia32cap_P
92 .type bn_mul_mont,\@function,6
100 $code.=<<___ if ($addx);
101 mov OPENSSL_ia32cap_P+8(%rip),%r11d
123 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
124 and \$-1024,%rsp # minimize TLB usage
126 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
128 mov $bp,%r12 # reassign $bp
132 mov ($n0),$n0 # pull n0[0] value
133 mov ($bp),$m0 # m0=bp[0]
140 mulq $m0 # ap[0]*bp[0]
144 imulq $lo0,$m1 # "tp[0]"*n0
148 add %rax,$lo0 # discarded
161 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
164 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
168 mulq $m0 # ap[j]*bp[0]
180 mov ($ap),%rax # ap[0]
182 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
184 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
191 mov $hi1,-8(%rsp,$num,8)
192 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
198 mov ($bp,$i,8),$m0 # m0=bp[i]
202 mulq $m0 # ap[0]*bp[i]
203 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
207 imulq $lo0,$m1 # tp[0]*n0
211 add %rax,$lo0 # discarded
214 mov 8(%rsp),$lo0 # tp[1]
225 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
228 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
232 mulq $m0 # ap[j]*bp[i]
236 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
246 mov ($ap),%rax # ap[0]
248 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
251 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
257 add $lo0,$hi1 # pull upmost overflow bit
259 mov $hi1,-8(%rsp,$num,8)
260 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
266 xor $i,$i # i=0 and clear CF!
267 mov (%rsp),%rax # tp[0]
268 lea (%rsp),$ap # borrow ap for tp
272 .Lsub: sbb ($np,$i,8),%rax
273 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
274 mov 8($ap,$i,8),%rax # tp[i+1]
276 dec $j # doesnn't affect CF!
279 sbb \$0,%rax # handle upmost overflow bit
286 or $np,$ap # ap=borrow?tp:rp
288 .Lcopy: # copy or in-place refresh
290 mov $i,(%rsp,$i,8) # zap temporary vector
291 mov %rax,($rp,$i,8) # rp[i]=tp[i]
296 mov 8(%rsp,$num,8),%rsi # restore %rsp
307 .size bn_mul_mont,.-bn_mul_mont
310 my @A=("%r10","%r11");
311 my @N=("%r13","%rdi");
313 .type bn_mul4x_mont,\@function,6
318 $code.=<<___ if ($addx);
335 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
336 and \$-1024,%rsp # minimize TLB usage
338 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
340 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
341 mov %rdx,%r12 # reassign $bp
345 mov ($n0),$n0 # pull n0[0] value
346 mov ($bp),$m0 # m0=bp[0]
353 mulq $m0 # ap[0]*bp[0]
357 imulq $A[0],$m1 # "tp[0]"*n0
361 add %rax,$A[0] # discarded
384 mulq $m0 # ap[j]*bp[0]
386 mov -16($np,$j,8),%rax
392 mov -8($ap,$j,8),%rax
394 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
396 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
399 mulq $m0 # ap[j]*bp[0]
401 mov -8($np,$j,8),%rax
409 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
411 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
414 mulq $m0 # ap[j]*bp[0]
424 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
426 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
429 mulq $m0 # ap[j]*bp[0]
438 mov -16($ap,$j,8),%rax
440 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
442 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
447 mulq $m0 # ap[j]*bp[0]
449 mov -16($np,$j,8),%rax
455 mov -8($ap,$j,8),%rax
457 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
459 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
462 mulq $m0 # ap[j]*bp[0]
464 mov -8($np,$j,8),%rax
470 mov ($ap),%rax # ap[0]
472 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
474 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
480 mov $N[0],-8(%rsp,$j,8)
481 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
486 mov ($bp,$i,8),$m0 # m0=bp[i]
490 mulq $m0 # ap[0]*bp[i]
491 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
495 imulq $A[0],$m1 # tp[0]*n0
499 add %rax,$A[0] # "$N[0]", discarded
504 mulq $m0 # ap[j]*bp[i]
508 add 8(%rsp),$A[1] # +tp[1]
516 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
519 mov $N[1],(%rsp) # tp[j-1]
524 mulq $m0 # ap[j]*bp[i]
526 mov -16($np,$j,8),%rax
528 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
534 mov -8($ap,$j,8),%rax
538 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
541 mulq $m0 # ap[j]*bp[i]
543 mov -8($np,$j,8),%rax
545 add -8(%rsp,$j,8),$A[1]
555 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
558 mulq $m0 # ap[j]*bp[i]
562 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
572 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
575 mulq $m0 # ap[j]*bp[i]
579 add 8(%rsp,$j,8),$A[1]
586 mov -16($ap,$j,8),%rax
590 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
595 mulq $m0 # ap[j]*bp[i]
597 mov -16($np,$j,8),%rax
599 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
605 mov -8($ap,$j,8),%rax
609 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
612 mulq $m0 # ap[j]*bp[i]
614 mov -8($np,$j,8),%rax
616 add -8(%rsp,$j,8),$A[1]
623 mov ($ap),%rax # ap[0]
627 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
633 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
635 mov $N[0],-8(%rsp,$j,8)
636 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
642 my @ri=("%rax","%rdx",$m0,$m1);
644 mov 16(%rsp,$num,8),$rp # restore $rp
645 mov 0(%rsp),@ri[0] # tp[0]
647 mov 8(%rsp),@ri[1] # tp[1]
648 shr \$2,$num # num/=4
649 lea (%rsp),$ap # borrow ap for tp
650 xor $i,$i # i=0 and clear CF!
653 mov 16($ap),@ri[2] # tp[2]
654 mov 24($ap),@ri[3] # tp[3]
656 lea -1($num),$j # j=num/4-1
660 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
661 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
662 sbb 16($np,$i,8),@ri[2]
663 mov 32($ap,$i,8),@ri[0] # tp[i+1]
664 mov 40($ap,$i,8),@ri[1]
665 sbb 24($np,$i,8),@ri[3]
666 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
667 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
668 sbb 32($np,$i,8),@ri[0]
669 mov 48($ap,$i,8),@ri[2]
670 mov 56($ap,$i,8),@ri[3]
671 sbb 40($np,$i,8),@ri[1]
673 dec $j # doesnn't affect CF!
676 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
677 mov 32($ap,$i,8),@ri[0] # load overflow bit
678 sbb 16($np,$i,8),@ri[2]
679 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
680 sbb 24($np,$i,8),@ri[3]
681 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
683 sbb \$0,@ri[0] # handle upmost overflow bit
684 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
691 or $np,$ap # ap=borrow?tp:rp
698 .Lcopy4x: # copy or in-place refresh
699 movdqu 16($ap,$i),%xmm2
700 movdqu 32($ap,$i),%xmm1
701 movdqa %xmm0,16(%rsp,$i)
702 movdqu %xmm2,16($rp,$i)
703 movdqa %xmm0,32(%rsp,$i)
704 movdqu %xmm1,32($rp,$i)
710 movdqu 16($ap,$i),%xmm2
711 movdqa %xmm0,16(%rsp,$i)
712 movdqu %xmm2,16($rp,$i)
716 mov 8(%rsp,$num,8),%rsi # restore %rsp
727 .size bn_mul4x_mont,.-bn_mul4x_mont
731 ######################################################################
732 # void bn_sqr8x_mont(
733 my $rptr="%rdi"; # const BN_ULONG *rptr,
734 my $aptr="%rsi"; # const BN_ULONG *aptr,
735 my $bptr="%rdx"; # not used
736 my $nptr="%rcx"; # const BN_ULONG *nptr,
737 my $n0 ="%r8"; # const BN_ULONG *n0);
738 my $num ="%r9"; # int num, has to be divisible by 8
740 my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
741 my @A0=("%r10","%r11");
742 my @A1=("%r12","%r13");
743 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
746 .extern bn_sqr8x_internal # see x86_64-mont5 module
747 .extern bn_sqrx8x_internal # see x86_64-mont5 module
749 .type bn_sqr8x_mont,\@function,6
762 shl \$3,${num}d # convert $num to bytes
763 shl \$3+2,%r10 # 4*$num
766 ##############################################################
767 # ensure that stack frame doesn't alias with $aptr modulo
768 # 4096. this is done to allow memory disambiguation logic
771 lea -64(%rsp,$num,4),%r11
777 sub %r11,%rsp # align with $aptr
778 lea -64(%rsp,$num,4),%rsp # alloca(frame+4*$num)
783 lea 4096-64(,$num,4),%r10 # 4096-frame-4*$num
784 lea -64(%rsp,$num,4),%rsp # alloca(frame+4*$num)
794 lea 64(%rsp,$num,2),%r11 # copy of modulus
796 mov %rax, 40(%rsp) # save original %rsp
800 movq %r11, %xmm2 # save pointer to modulus copy
802 mov OPENSSL_ia32cap_P+8(%rip),%eax
807 movq 8*0($nptr),%xmm0
808 movq 8*1($nptr),%xmm1
809 movq 8*2($nptr),%xmm3
810 movq 8*3($nptr),%xmm4
812 movdqa %xmm0,16*0(%r11)
813 movdqa %xmm1,16*1(%r11)
814 movdqa %xmm3,16*2(%r11)
815 movdqa %xmm4,16*3(%r11)
821 movq $rptr,%xmm1 # save $rptr
822 movq %r10, %xmm3 # -$num
824 $code.=<<___ if ($addx);
829 call bn_sqrx8x_internal # see x86_64-mont5 module
833 lea 64(%rsp,$num,2),%rdx
835 mov 40(%rsp),%rsi # restore %rsp
842 call bn_sqr8x_internal # see x86_64-mont5 module
846 lea 64(%rsp,$num,2),%rdx
848 mov 40(%rsp),%rsi # restore %rsp
853 movdqa %xmm0,16*0(%rax) # wipe t
854 movdqa %xmm0,16*1(%rax)
855 movdqa %xmm0,16*2(%rax)
856 movdqa %xmm0,16*3(%rax)
858 movdqa %xmm0,16*0(%rdx) # wipe n
859 movdqa %xmm0,16*1(%rdx)
860 movdqa %xmm0,16*2(%rdx)
861 movdqa %xmm0,16*3(%rdx)
876 .size bn_sqr8x_mont,.-bn_sqr8x_mont
881 my $bp="%rdx"; # original value
884 .type bn_mulx4x_mont,\@function,6
896 shl \$3,${num}d # convert $num to bytes
899 sub $num,%r10 # -$num
901 lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
904 ##############################################################
907 # +8 off-loaded &b[i]
916 mov $num,0(%rsp) # save $num
918 mov %r10,16(%rsp) # end of b[num]
920 mov $n0, 24(%rsp) # save *n0
921 mov $rp, 32(%rsp) # save $rp
922 mov %rax,40(%rsp) # save original %rsp
923 mov $num,48(%rsp) # inner counter
929 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
930 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
934 mov ($bp),%rdx # b[0], $bp==%rdx actually
935 lea 64+32(%rsp),$tptr
938 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
939 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
941 mov $bptr,8(%rsp) # off-load &b[i]
942 mulx 2*8($aptr),%r12,%r13 # ...
946 mov $mi,$bptr # borrow $bptr
947 imulq 24(%rsp),$mi # "t[0]"*n0
948 xor $zero,$zero # cf=0, of=0
950 mulx 3*8($aptr),%rax,%r14
954 adcx $zero,%r14 # cf=0
956 mulx 0*8($nptr),%rax,%r10
957 adcx %rax,$bptr # discarded
959 mulx 1*8($nptr),%rax,%r11
962 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
963 mov 48(%rsp),$bptr # counter value
967 mulx 3*8($nptr),%rax,%r15
971 adox $zero,%r15 # of=0
979 adcx $zero,%r15 # cf=0, modulo-scheduled
980 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
982 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
984 mulx 2*8($aptr),%r12,%rax # ...
986 mulx 3*8($aptr),%r13,%r14
990 adcx $zero,%r14 # cf=0
995 mulx 0*8($nptr),%rax,%r15
998 mulx 1*8($nptr),%rax,%r15
1001 mulx 2*8($nptr),%rax,%r15
1002 mov %r10,-5*8($tptr)
1004 mov %r11,-4*8($tptr)
1006 mulx 3*8($nptr),%rax,%r15
1008 mov %r12,-3*8($tptr)
1011 lea 4*8($nptr),$nptr
1012 mov %r13,-2*8($tptr)
1014 dec $bptr # of=0, pass cf
1017 mov 0(%rsp),$num # load num
1018 mov 8(%rsp),$bptr # re-load &b[i]
1019 adc $zero,%r15 # modulo-scheduled
1021 sbb %r15,%r15 # top-most carry
1022 mov %r14,-1*8($tptr)
1027 mov ($bptr),%rdx # b[i]
1028 lea 8($bptr),$bptr # b++
1029 sub $num,$aptr # rewind $aptr
1030 mov %r15,($tptr) # save top-most carry
1031 lea 64+4*8(%rsp),$tptr
1032 sub $num,$nptr # rewind $nptr
1034 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
1035 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1037 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
1038 adox -4*8($tptr),$mi
1040 mulx 2*8($aptr),%r15,%r13 # ...
1041 adox -3*8($tptr),%r11
1046 mov $bptr,8(%rsp) # off-load &b[i]
1049 imulq 24(%rsp),$mi # "t[0]"*n0
1050 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1052 mulx 3*8($aptr),%rax,%r14
1054 adox -2*8($tptr),%r12
1056 adox -1*8($tptr),%r13
1058 lea 4*8($aptr),$aptr
1061 mulx 0*8($nptr),%rax,%r10
1062 adcx %rax,%r15 # discarded
1064 mulx 1*8($nptr),%rax,%r11
1067 mulx 2*8($nptr),%rax,%r12
1068 mov %r10,-4*8($tptr)
1071 mulx 3*8($nptr),%rax,%r15
1073 mov %r11,-3*8($tptr)
1074 lea 4*8($nptr),$nptr
1076 adox $zero,%r15 # of=0
1077 mov 48(%rsp),$bptr # counter value
1078 mov %r12,-2*8($tptr)
1084 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1085 adcx $zero,%r15 # cf=0, modulo-scheduled
1087 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1088 adcx 0*8($tptr),%r10
1090 mulx 2*8($aptr),%r12,%rax # ...
1091 adcx 1*8($tptr),%r11
1093 mulx 3*8($aptr),%r13,%r14
1095 adcx 2*8($tptr),%r12
1097 adcx 3*8($tptr),%r13
1098 adox $zero,%r14 # of=0
1099 lea 4*8($aptr),$aptr
1100 lea 4*8($tptr),$tptr
1101 adcx $zero,%r14 # cf=0
1104 mulx 0*8($nptr),%rax,%r15
1107 mulx 1*8($nptr),%rax,%r15
1110 mulx 2*8($nptr),%rax,%r15
1111 mov %r10,-5*8($tptr)
1114 mulx 3*8($nptr),%rax,%r15
1116 mov %r11,-4*8($tptr)
1117 mov %r12,-3*8($tptr)
1120 lea 4*8($nptr),$nptr
1121 mov %r13,-2*8($tptr)
1123 dec $bptr # of=0, pass cf
1126 mov 0(%rsp),$num # load num
1127 mov 8(%rsp),$bptr # re-load &b[i]
1128 adc $zero,%r15 # modulo-scheduled
1129 sub 0*8($tptr),$zero # pull top-most carry
1132 sbb %r15,%r15 # top-most carry
1133 mov %r14,-1*8($tptr)
1138 sub %r14,$mi # compare top-most words
1144 mov 32(%rsp),$rptr # restore rp
1148 mov 0*8($nptr,$num),%r8
1149 mov 1*8($nptr,$num),%r9
1151 jmp .Lmulx4x_sub_entry
1155 mov 0*8($nptr,$num),%r8
1156 mov 1*8($nptr,$num),%r9
1159 mov 2*8($nptr,$num),%r10
1162 mov 3*8($nptr,$num),%r11
1169 neg %rdx # mov %rdx,%cf
1172 movdqa %xmm0,($tptr)
1175 movdqa %xmm0,16($tptr)
1176 lea 4*8($tptr),$tptr
1177 sbb %rdx,%rdx # mov %cf,%rdx
1183 lea 4*8($rptr),$rptr
1188 mov 40(%rsp),%rsi # restore %rsp
1199 .size bn_mulx4x_mont,.-bn_mulx4x_mont
1203 .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1207 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1208 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1216 .extern __imp_RtlVirtualUnwind
1217 .type mul_handler,\@abi-omnipotent
1231 mov 120($context),%rax # pull context->Rax
1232 mov 248($context),%rbx # pull context->Rip
1234 mov 8($disp),%rsi # disp->ImageBase
1235 mov 56($disp),%r11 # disp->HandlerData
1237 mov 0(%r11),%r10d # HandlerData[0]
1238 lea (%rsi,%r10),%r10 # end of prologue label
1239 cmp %r10,%rbx # context->Rip<end of prologue label
1240 jb .Lcommon_seh_tail
1242 mov 152($context),%rax # pull context->Rsp
1244 mov 4(%r11),%r10d # HandlerData[1]
1245 lea (%rsi,%r10),%r10 # epilogue label
1246 cmp %r10,%rbx # context->Rip>=epilogue label
1247 jae .Lcommon_seh_tail
1249 mov 192($context),%r10 # pull $num
1250 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1259 mov %rbx,144($context) # restore context->Rbx
1260 mov %rbp,160($context) # restore context->Rbp
1261 mov %r12,216($context) # restore context->R12
1262 mov %r13,224($context) # restore context->R13
1263 mov %r14,232($context) # restore context->R14
1264 mov %r15,240($context) # restore context->R15
1266 jmp .Lcommon_seh_tail
1267 .size mul_handler,.-mul_handler
1269 .type sqr_handler,\@abi-omnipotent
1283 mov 120($context),%rax # pull context->Rax
1284 mov 248($context),%rbx # pull context->Rip
1286 mov 8($disp),%rsi # disp->ImageBase
1287 mov 56($disp),%r11 # disp->HandlerData
1289 mov 0(%r11),%r10d # HandlerData[0]
1290 lea (%rsi,%r10),%r10 # end of prologue label
1291 cmp %r10,%rbx # context->Rip<.Lsqr_body
1292 jb .Lcommon_seh_tail
1294 mov 152($context),%rax # pull context->Rsp
1296 mov 4(%r11),%r10d # HandlerData[1]
1297 lea (%rsi,%r10),%r10 # epilogue label
1298 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
1299 jae .Lcommon_seh_tail
1301 mov 40(%rax),%rax # pull saved stack pointer
1309 mov %rbx,144($context) # restore context->Rbx
1310 mov %rbp,160($context) # restore context->Rbp
1311 mov %r12,216($context) # restore context->R12
1312 mov %r13,224($context) # restore context->R13
1313 mov %r14,232($context) # restore context->R14
1314 mov %r15,240($context) # restore context->R15
1319 mov %rax,152($context) # restore context->Rsp
1320 mov %rsi,168($context) # restore context->Rsi
1321 mov %rdi,176($context) # restore context->Rdi
1323 mov 40($disp),%rdi # disp->ContextRecord
1324 mov $context,%rsi # context
1325 mov \$154,%ecx # sizeof(CONTEXT)
1326 .long 0xa548f3fc # cld; rep movsq
1329 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1330 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1331 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1332 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1333 mov 40(%rsi),%r10 # disp->ContextRecord
1334 lea 56(%rsi),%r11 # &disp->HandlerData
1335 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1336 mov %r10,32(%rsp) # arg5
1337 mov %r11,40(%rsp) # arg6
1338 mov %r12,48(%rsp) # arg7
1339 mov %rcx,56(%rsp) # arg8, (NULL)
1340 call *__imp_RtlVirtualUnwind(%rip)
1342 mov \$1,%eax # ExceptionContinueSearch
1354 .size sqr_handler,.-sqr_handler
1358 .rva .LSEH_begin_bn_mul_mont
1359 .rva .LSEH_end_bn_mul_mont
1360 .rva .LSEH_info_bn_mul_mont
1362 .rva .LSEH_begin_bn_mul4x_mont
1363 .rva .LSEH_end_bn_mul4x_mont
1364 .rva .LSEH_info_bn_mul4x_mont
1366 .rva .LSEH_begin_bn_sqr8x_mont
1367 .rva .LSEH_end_bn_sqr8x_mont
1368 .rva .LSEH_info_bn_sqr8x_mont
1370 $code.=<<___ if ($addx);
1371 .rva .LSEH_begin_bn_mulx4x_mont
1372 .rva .LSEH_end_bn_mulx4x_mont
1373 .rva .LSEH_info_bn_mulx4x_mont
1378 .LSEH_info_bn_mul_mont:
1381 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
1382 .LSEH_info_bn_mul4x_mont:
1385 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1386 .LSEH_info_bn_sqr8x_mont:
1389 .rva .Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
1391 $code.=<<___ if ($addx);
1392 .LSEH_info_bn_mulx4x_mont:
1395 .rva .Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]