2 # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
19 # Montgomery multiplication routine for x86_64. While it gives modest
20 # 9% improvement of rsa4096 sign on Opteron, rsa512 sign runs more
21 # than twice, >2x, as fast. Most common rsa1024 sign is improved by
22 # respectful 50%. It remains to be seen if loop unrolling and
23 # dedicated squaring routine can provide further improvement...
27 # Add dedicated squaring procedure. Performance improvement varies
28 # from platform to platform, but in average it's ~5%/15%/25%/33%
29 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
33 # Unroll and modulo-schedule inner loops in such manner that they
34 # are "fallen through" for input lengths of 8, which is critical for
35 # 1024-bit RSA *sign*. Average performance improvement in comparison
36 # to *initial* version of this module from 2005 is ~0%/30%/40%/45%
37 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
41 # Optimize reduction in squaring procedure and improve 1024+-bit RSA
42 # sign performance by 10-16% on Intel Sandy Bridge and later
43 # (virtually same on non-Intel processors).
47 # Add MULX/ADOX/ADCX code path.
51 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
53 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
55 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
56 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
57 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
58 die "can't locate x86_64-xlate.pl";
60 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
63 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
64 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
68 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
69 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
73 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
74 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
78 if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9])\.([0-9]+)/) {
79 my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
84 $rp="%rdi"; # BN_ULONG *rp,
85 $ap="%rsi"; # const BN_ULONG *ap,
86 $bp="%rdx"; # const BN_ULONG *bp,
87 $np="%rcx"; # const BN_ULONG *np,
88 $n0="%r8"; # const BN_ULONG *n0,
89 $num="%r9"; # int num);
101 .extern OPENSSL_ia32cap_P
104 .type bn_mul_mont,\@function,6
112 $code.=<<___ if ($addx);
113 mov OPENSSL_ia32cap_P+8(%rip),%r11d
135 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
136 and \$-1024,%rsp # minimize TLB usage
138 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
140 # An OS-agnostic version of __chkstk.
142 # Some OSes (Windows) insist on stack being "wired" to
143 # physical memory in strictly sequential manner, i.e. if stack
144 # allocation spans two pages, then reference to farmost one can
145 # be punishable by SEGV. But page walking can do good even on
146 # other OSes, because it guarantees that villain thread hits
147 # the guard page before it can make damage to innocent one...
153 .byte 0x66,0x2e # predict non-taken
156 mov $bp,%r12 # reassign $bp
160 mov ($n0),$n0 # pull n0[0] value
161 mov ($bp),$m0 # m0=bp[0]
168 mulq $m0 # ap[0]*bp[0]
172 imulq $lo0,$m1 # "tp[0]"*n0
176 add %rax,$lo0 # discarded
189 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
192 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
196 mulq $m0 # ap[j]*bp[0]
208 mov ($ap),%rax # ap[0]
210 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
212 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
219 mov $hi1,-8(%rsp,$num,8)
220 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
226 mov ($bp,$i,8),$m0 # m0=bp[i]
230 mulq $m0 # ap[0]*bp[i]
231 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
235 imulq $lo0,$m1 # tp[0]*n0
239 add %rax,$lo0 # discarded
242 mov 8(%rsp),$lo0 # tp[1]
253 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
256 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
260 mulq $m0 # ap[j]*bp[i]
264 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
274 mov ($ap),%rax # ap[0]
276 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
279 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
285 add $lo0,$hi1 # pull upmost overflow bit
287 mov $hi1,-8(%rsp,$num,8)
288 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
294 xor $i,$i # i=0 and clear CF!
295 mov (%rsp),%rax # tp[0]
296 lea (%rsp),$ap # borrow ap for tp
300 .Lsub: sbb ($np,$i,8),%rax
301 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
302 mov 8($ap,$i,8),%rax # tp[i+1]
304 dec $j # doesnn't affect CF!
307 sbb \$0,%rax # handle upmost overflow bit
314 or $np,$ap # ap=borrow?tp:rp
316 .Lcopy: # copy or in-place refresh
318 mov $i,(%rsp,$i,8) # zap temporary vector
319 mov %rax,($rp,$i,8) # rp[i]=tp[i]
324 mov 8(%rsp,$num,8),%rsi # restore %rsp
335 .size bn_mul_mont,.-bn_mul_mont
338 my @A=("%r10","%r11");
339 my @N=("%r13","%rdi");
341 .type bn_mul4x_mont,\@function,6
346 $code.=<<___ if ($addx);
363 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
364 and \$-1024,%rsp # minimize TLB usage
366 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
373 .byte 0x2e # predict non-taken
374 jnc .Lmul4x_page_walk
376 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
377 mov %rdx,%r12 # reassign $bp
381 mov ($n0),$n0 # pull n0[0] value
382 mov ($bp),$m0 # m0=bp[0]
389 mulq $m0 # ap[0]*bp[0]
393 imulq $A[0],$m1 # "tp[0]"*n0
397 add %rax,$A[0] # discarded
420 mulq $m0 # ap[j]*bp[0]
422 mov -16($np,$j,8),%rax
428 mov -8($ap,$j,8),%rax
430 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
432 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
435 mulq $m0 # ap[j]*bp[0]
437 mov -8($np,$j,8),%rax
445 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
447 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
450 mulq $m0 # ap[j]*bp[0]
460 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
462 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
465 mulq $m0 # ap[j]*bp[0]
474 mov -16($ap,$j,8),%rax
476 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
478 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
483 mulq $m0 # ap[j]*bp[0]
485 mov -16($np,$j,8),%rax
491 mov -8($ap,$j,8),%rax
493 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
495 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
498 mulq $m0 # ap[j]*bp[0]
500 mov -8($np,$j,8),%rax
506 mov ($ap),%rax # ap[0]
508 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
510 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
516 mov $N[0],-8(%rsp,$j,8)
517 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
522 mov ($bp,$i,8),$m0 # m0=bp[i]
526 mulq $m0 # ap[0]*bp[i]
527 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
531 imulq $A[0],$m1 # tp[0]*n0
535 add %rax,$A[0] # "$N[0]", discarded
540 mulq $m0 # ap[j]*bp[i]
544 add 8(%rsp),$A[1] # +tp[1]
552 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
555 mov $N[1],(%rsp) # tp[j-1]
560 mulq $m0 # ap[j]*bp[i]
562 mov -16($np,$j,8),%rax
564 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
570 mov -8($ap,$j,8),%rax
574 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
577 mulq $m0 # ap[j]*bp[i]
579 mov -8($np,$j,8),%rax
581 add -8(%rsp,$j,8),$A[1]
591 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
594 mulq $m0 # ap[j]*bp[i]
598 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
608 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
611 mulq $m0 # ap[j]*bp[i]
615 add 8(%rsp,$j,8),$A[1]
622 mov -16($ap,$j,8),%rax
626 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
631 mulq $m0 # ap[j]*bp[i]
633 mov -16($np,$j,8),%rax
635 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
641 mov -8($ap,$j,8),%rax
645 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
648 mulq $m0 # ap[j]*bp[i]
650 mov -8($np,$j,8),%rax
652 add -8(%rsp,$j,8),$A[1]
659 mov ($ap),%rax # ap[0]
663 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
669 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
671 mov $N[0],-8(%rsp,$j,8)
672 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
678 my @ri=("%rax","%rdx",$m0,$m1);
680 mov 16(%rsp,$num,8),$rp # restore $rp
681 mov 0(%rsp),@ri[0] # tp[0]
683 mov 8(%rsp),@ri[1] # tp[1]
684 shr \$2,$num # num/=4
685 lea (%rsp),$ap # borrow ap for tp
686 xor $i,$i # i=0 and clear CF!
689 mov 16($ap),@ri[2] # tp[2]
690 mov 24($ap),@ri[3] # tp[3]
692 lea -1($num),$j # j=num/4-1
696 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
697 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
698 sbb 16($np,$i,8),@ri[2]
699 mov 32($ap,$i,8),@ri[0] # tp[i+1]
700 mov 40($ap,$i,8),@ri[1]
701 sbb 24($np,$i,8),@ri[3]
702 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
703 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
704 sbb 32($np,$i,8),@ri[0]
705 mov 48($ap,$i,8),@ri[2]
706 mov 56($ap,$i,8),@ri[3]
707 sbb 40($np,$i,8),@ri[1]
709 dec $j # doesnn't affect CF!
712 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
713 mov 32($ap,$i,8),@ri[0] # load overflow bit
714 sbb 16($np,$i,8),@ri[2]
715 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
716 sbb 24($np,$i,8),@ri[3]
717 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
719 sbb \$0,@ri[0] # handle upmost overflow bit
720 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
727 or $np,$ap # ap=borrow?tp:rp
734 .Lcopy4x: # copy or in-place refresh
735 movdqu 16($ap,$i),%xmm2
736 movdqu 32($ap,$i),%xmm1
737 movdqa %xmm0,16(%rsp,$i)
738 movdqu %xmm2,16($rp,$i)
739 movdqa %xmm0,32(%rsp,$i)
740 movdqu %xmm1,32($rp,$i)
746 movdqu 16($ap,$i),%xmm2
747 movdqa %xmm0,16(%rsp,$i)
748 movdqu %xmm2,16($rp,$i)
752 mov 8(%rsp,$num,8),%rsi # restore %rsp
763 .size bn_mul4x_mont,.-bn_mul4x_mont
767 ######################################################################
768 # void bn_sqr8x_mont(
769 my $rptr="%rdi"; # const BN_ULONG *rptr,
770 my $aptr="%rsi"; # const BN_ULONG *aptr,
771 my $bptr="%rdx"; # not used
772 my $nptr="%rcx"; # const BN_ULONG *nptr,
773 my $n0 ="%r8"; # const BN_ULONG *n0);
774 my $num ="%r9"; # int num, has to be divisible by 8
776 my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
777 my @A0=("%r10","%r11");
778 my @A1=("%r12","%r13");
779 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
781 $code.=<<___ if ($addx);
782 .extern bn_sqrx8x_internal # see x86_64-mont5 module
785 .extern bn_sqr8x_internal # see x86_64-mont5 module
787 .type bn_sqr8x_mont,\@function,6
800 shl \$3,${num}d # convert $num to bytes
801 shl \$3+2,%r10 # 4*$num
804 ##############################################################
805 # ensure that stack frame doesn't alias with $aptr modulo
806 # 4096. this is done to allow memory disambiguation logic
809 lea -64(%rsp,$num,2),%r11
815 sub %r11,%rsp # align with $aptr
816 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
821 lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
822 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
835 .byte 0x2e # predict non-taken
836 jnc .Lsqr8x_page_walk
842 mov %rax, 40(%rsp) # save original %rsp
845 movq $nptr, %xmm2 # save pointer to modulus
847 movq $rptr,%xmm1 # save $rptr
848 movq %r10, %xmm3 # -$num
850 $code.=<<___ if ($addx);
851 mov OPENSSL_ia32cap_P+8(%rip),%eax
856 call bn_sqrx8x_internal # see x86_64-mont5 module
857 # %rax top-most carry
860 # %r8 end of tp[2*num]
865 sar \$3+2,%rcx # %cf=0
872 call bn_sqr8x_internal # see x86_64-mont5 module
873 # %rax top-most carry
876 # %rdi end of tp[2*num]
881 sar \$3+2,%rcx # %cf=0
901 inc %rcx # preserves %cf
904 sbb \$0,%rax # top-most carry
905 lea (%rbx,$num),%rbx # rewind
906 lea ($rptr,$num),$rptr # rewind
910 pshufd \$0,%xmm1,%xmm1
911 mov 40(%rsp),%rsi # restore %rsp
912 jmp .Lsqr8x_cond_copy
916 movdqa 16*0(%rbx),%xmm2
917 movdqa 16*1(%rbx),%xmm3
919 movdqu 16*0($rptr),%xmm4
920 movdqu 16*1($rptr),%xmm5
921 lea 16*2($rptr),$rptr
922 movdqa %xmm0,-16*2(%rbx) # zero tp
923 movdqa %xmm0,-16*1(%rbx)
924 movdqa %xmm0,-16*2(%rbx,%rdx)
925 movdqa %xmm0,-16*1(%rbx,%rdx)
934 movdqu %xmm4,-16*2($rptr)
935 movdqu %xmm5,-16*1($rptr)
937 jnz .Lsqr8x_cond_copy
949 .size bn_sqr8x_mont,.-bn_sqr8x_mont
954 my $bp="%rdx"; # original value
957 .type bn_mulx4x_mont,\@function,6
969 shl \$3,${num}d # convert $num to bytes
972 sub $num,%r10 # -$num
974 lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
982 .byte 0x66,0x2e # predict non-taken
983 jnc .Lmulx4x_page_walk
986 ##############################################################
989 # +8 off-loaded &b[i]
998 mov $num,0(%rsp) # save $num
1000 mov %r10,16(%rsp) # end of b[num]
1002 mov $n0, 24(%rsp) # save *n0
1003 mov $rp, 32(%rsp) # save $rp
1004 mov %rax,40(%rsp) # save original %rsp
1005 mov $num,48(%rsp) # inner counter
1011 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
1012 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
1016 mov ($bp),%rdx # b[0], $bp==%rdx actually
1017 lea 64+32(%rsp),$tptr
1020 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
1021 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
1023 mov $bptr,8(%rsp) # off-load &b[i]
1024 mulx 2*8($aptr),%r12,%r13 # ...
1028 mov $mi,$bptr # borrow $bptr
1029 imulq 24(%rsp),$mi # "t[0]"*n0
1030 xor $zero,$zero # cf=0, of=0
1032 mulx 3*8($aptr),%rax,%r14
1034 lea 4*8($aptr),$aptr
1036 adcx $zero,%r14 # cf=0
1038 mulx 0*8($nptr),%rax,%r10
1039 adcx %rax,$bptr # discarded
1041 mulx 1*8($nptr),%rax,%r11
1044 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
1045 mov 48(%rsp),$bptr # counter value
1046 mov %r10,-4*8($tptr)
1049 mulx 3*8($nptr),%rax,%r15
1051 mov %r11,-3*8($tptr)
1053 adox $zero,%r15 # of=0
1054 lea 4*8($nptr),$nptr
1055 mov %r12,-2*8($tptr)
1061 adcx $zero,%r15 # cf=0, modulo-scheduled
1062 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
1064 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
1066 mulx 2*8($aptr),%r12,%rax # ...
1068 mulx 3*8($aptr),%r13,%r14
1072 adcx $zero,%r14 # cf=0
1073 lea 4*8($aptr),$aptr
1074 lea 4*8($tptr),$tptr
1077 mulx 0*8($nptr),%rax,%r15
1080 mulx 1*8($nptr),%rax,%r15
1083 mulx 2*8($nptr),%rax,%r15
1084 mov %r10,-5*8($tptr)
1086 mov %r11,-4*8($tptr)
1088 mulx 3*8($nptr),%rax,%r15
1090 mov %r12,-3*8($tptr)
1093 lea 4*8($nptr),$nptr
1094 mov %r13,-2*8($tptr)
1096 dec $bptr # of=0, pass cf
1099 mov 0(%rsp),$num # load num
1100 mov 8(%rsp),$bptr # re-load &b[i]
1101 adc $zero,%r15 # modulo-scheduled
1103 sbb %r15,%r15 # top-most carry
1104 mov %r14,-1*8($tptr)
1109 mov ($bptr),%rdx # b[i]
1110 lea 8($bptr),$bptr # b++
1111 sub $num,$aptr # rewind $aptr
1112 mov %r15,($tptr) # save top-most carry
1113 lea 64+4*8(%rsp),$tptr
1114 sub $num,$nptr # rewind $nptr
1116 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
1117 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1119 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
1120 adox -4*8($tptr),$mi
1122 mulx 2*8($aptr),%r15,%r13 # ...
1123 adox -3*8($tptr),%r11
1128 mov $bptr,8(%rsp) # off-load &b[i]
1131 imulq 24(%rsp),$mi # "t[0]"*n0
1132 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1134 mulx 3*8($aptr),%rax,%r14
1136 adox -2*8($tptr),%r12
1138 adox -1*8($tptr),%r13
1140 lea 4*8($aptr),$aptr
1143 mulx 0*8($nptr),%rax,%r10
1144 adcx %rax,%r15 # discarded
1146 mulx 1*8($nptr),%rax,%r11
1149 mulx 2*8($nptr),%rax,%r12
1150 mov %r10,-4*8($tptr)
1153 mulx 3*8($nptr),%rax,%r15
1155 mov %r11,-3*8($tptr)
1156 lea 4*8($nptr),$nptr
1158 adox $zero,%r15 # of=0
1159 mov 48(%rsp),$bptr # counter value
1160 mov %r12,-2*8($tptr)
1166 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1167 adcx $zero,%r15 # cf=0, modulo-scheduled
1169 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1170 adcx 0*8($tptr),%r10
1172 mulx 2*8($aptr),%r12,%rax # ...
1173 adcx 1*8($tptr),%r11
1175 mulx 3*8($aptr),%r13,%r14
1177 adcx 2*8($tptr),%r12
1179 adcx 3*8($tptr),%r13
1180 adox $zero,%r14 # of=0
1181 lea 4*8($aptr),$aptr
1182 lea 4*8($tptr),$tptr
1183 adcx $zero,%r14 # cf=0
1186 mulx 0*8($nptr),%rax,%r15
1189 mulx 1*8($nptr),%rax,%r15
1192 mulx 2*8($nptr),%rax,%r15
1193 mov %r10,-5*8($tptr)
1196 mulx 3*8($nptr),%rax,%r15
1198 mov %r11,-4*8($tptr)
1199 mov %r12,-3*8($tptr)
1202 lea 4*8($nptr),$nptr
1203 mov %r13,-2*8($tptr)
1205 dec $bptr # of=0, pass cf
1208 mov 0(%rsp),$num # load num
1209 mov 8(%rsp),$bptr # re-load &b[i]
1210 adc $zero,%r15 # modulo-scheduled
1211 sub 0*8($tptr),$zero # pull top-most carry
1213 sbb %r15,%r15 # top-most carry
1214 mov %r14,-1*8($tptr)
1220 sub $num,$nptr # rewind $nptr
1223 shr \$3+2,$num # %cf=0
1224 mov 32(%rsp),$rptr # restore rp
1233 lea 8*4($tptr),$tptr
1238 lea 8*4($nptr),$nptr
1243 lea 8*4($rptr),$rptr
1244 dec $num # preserves %cf
1247 sbb \$0,%r15 # top-most carry
1249 sub %rdx,$rptr # rewind
1253 pshufd \$0,%xmm1,%xmm1
1254 mov 40(%rsp),%rsi # restore %rsp
1255 jmp .Lmulx4x_cond_copy
1259 movdqa 16*0($tptr),%xmm2
1260 movdqa 16*1($tptr),%xmm3
1261 lea 16*2($tptr),$tptr
1262 movdqu 16*0($rptr),%xmm4
1263 movdqu 16*1($rptr),%xmm5
1264 lea 16*2($rptr),$rptr
1265 movdqa %xmm0,-16*2($tptr) # zero tp
1266 movdqa %xmm0,-16*1($tptr)
1275 movdqu %xmm4,-16*2($rptr)
1276 movdqu %xmm5,-16*1($rptr)
1278 jnz .Lmulx4x_cond_copy
1292 .size bn_mulx4x_mont,.-bn_mulx4x_mont
1296 .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1300 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1301 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1309 .extern __imp_RtlVirtualUnwind
1310 .type mul_handler,\@abi-omnipotent
1324 mov 120($context),%rax # pull context->Rax
1325 mov 248($context),%rbx # pull context->Rip
1327 mov 8($disp),%rsi # disp->ImageBase
1328 mov 56($disp),%r11 # disp->HandlerData
1330 mov 0(%r11),%r10d # HandlerData[0]
1331 lea (%rsi,%r10),%r10 # end of prologue label
1332 cmp %r10,%rbx # context->Rip<end of prologue label
1333 jb .Lcommon_seh_tail
1335 mov 152($context),%rax # pull context->Rsp
1337 mov 4(%r11),%r10d # HandlerData[1]
1338 lea (%rsi,%r10),%r10 # epilogue label
1339 cmp %r10,%rbx # context->Rip>=epilogue label
1340 jae .Lcommon_seh_tail
1342 mov 192($context),%r10 # pull $num
1343 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1352 mov %rbx,144($context) # restore context->Rbx
1353 mov %rbp,160($context) # restore context->Rbp
1354 mov %r12,216($context) # restore context->R12
1355 mov %r13,224($context) # restore context->R13
1356 mov %r14,232($context) # restore context->R14
1357 mov %r15,240($context) # restore context->R15
1359 jmp .Lcommon_seh_tail
1360 .size mul_handler,.-mul_handler
1362 .type sqr_handler,\@abi-omnipotent
1376 mov 120($context),%rax # pull context->Rax
1377 mov 248($context),%rbx # pull context->Rip
1379 mov 8($disp),%rsi # disp->ImageBase
1380 mov 56($disp),%r11 # disp->HandlerData
1382 mov 0(%r11),%r10d # HandlerData[0]
1383 lea (%rsi,%r10),%r10 # end of prologue label
1384 cmp %r10,%rbx # context->Rip<.Lsqr_body
1385 jb .Lcommon_seh_tail
1387 mov 152($context),%rax # pull context->Rsp
1389 mov 4(%r11),%r10d # HandlerData[1]
1390 lea (%rsi,%r10),%r10 # epilogue label
1391 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
1392 jae .Lcommon_seh_tail
1394 mov 40(%rax),%rax # pull saved stack pointer
1402 mov %rbx,144($context) # restore context->Rbx
1403 mov %rbp,160($context) # restore context->Rbp
1404 mov %r12,216($context) # restore context->R12
1405 mov %r13,224($context) # restore context->R13
1406 mov %r14,232($context) # restore context->R14
1407 mov %r15,240($context) # restore context->R15
1412 mov %rax,152($context) # restore context->Rsp
1413 mov %rsi,168($context) # restore context->Rsi
1414 mov %rdi,176($context) # restore context->Rdi
1416 mov 40($disp),%rdi # disp->ContextRecord
1417 mov $context,%rsi # context
1418 mov \$154,%ecx # sizeof(CONTEXT)
1419 .long 0xa548f3fc # cld; rep movsq
1422 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1423 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1424 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1425 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1426 mov 40(%rsi),%r10 # disp->ContextRecord
1427 lea 56(%rsi),%r11 # &disp->HandlerData
1428 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1429 mov %r10,32(%rsp) # arg5
1430 mov %r11,40(%rsp) # arg6
1431 mov %r12,48(%rsp) # arg7
1432 mov %rcx,56(%rsp) # arg8, (NULL)
1433 call *__imp_RtlVirtualUnwind(%rip)
1435 mov \$1,%eax # ExceptionContinueSearch
1447 .size sqr_handler,.-sqr_handler
1451 .rva .LSEH_begin_bn_mul_mont
1452 .rva .LSEH_end_bn_mul_mont
1453 .rva .LSEH_info_bn_mul_mont
1455 .rva .LSEH_begin_bn_mul4x_mont
1456 .rva .LSEH_end_bn_mul4x_mont
1457 .rva .LSEH_info_bn_mul4x_mont
1459 .rva .LSEH_begin_bn_sqr8x_mont
1460 .rva .LSEH_end_bn_sqr8x_mont
1461 .rva .LSEH_info_bn_sqr8x_mont
1463 $code.=<<___ if ($addx);
1464 .rva .LSEH_begin_bn_mulx4x_mont
1465 .rva .LSEH_end_bn_mulx4x_mont
1466 .rva .LSEH_info_bn_mulx4x_mont
1471 .LSEH_info_bn_mul_mont:
1474 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
1475 .LSEH_info_bn_mul4x_mont:
1478 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1479 .LSEH_info_bn_sqr8x_mont:
1482 .rva .Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
1484 $code.=<<___ if ($addx);
1485 .LSEH_info_bn_mulx4x_mont:
1488 .rva .Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]