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
110 .cfi_def_cfa_register %rax
116 $code.=<<___ if ($addx);
117 mov OPENSSL_ia32cap_P+8(%rip),%r11d
143 lea -16(%rsp,$num,8),%r10 # future alloca(8*(num+2))
144 neg $num # restore $num
145 and \$-1024,%r10 # minimize TLB usage
147 # An OS-agnostic version of __chkstk.
149 # Some OSes (Windows) insist on stack being "wired" to
150 # physical memory in strictly sequential manner, i.e. if stack
151 # allocation spans two pages, then reference to farmost one can
152 # be punishable by SEGV. But page walking can do good even on
153 # other OSes, because it guarantees that villain thread hits
154 # the guard page before it can make damage to innocent one...
161 jmp .Lmul_page_walk_done
169 .Lmul_page_walk_done:
171 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
172 .cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
174 mov $bp,%r12 # reassign $bp
178 mov ($n0),$n0 # pull n0[0] value
179 mov ($bp),$m0 # m0=bp[0]
186 mulq $m0 # ap[0]*bp[0]
190 imulq $lo0,$m1 # "tp[0]"*n0
194 add %rax,$lo0 # discarded
207 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
210 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
214 mulq $m0 # ap[j]*bp[0]
226 mov ($ap),%rax # ap[0]
228 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
230 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
237 mov $hi1,-8(%rsp,$num,8)
238 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
244 mov ($bp,$i,8),$m0 # m0=bp[i]
248 mulq $m0 # ap[0]*bp[i]
249 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
253 imulq $lo0,$m1 # tp[0]*n0
257 add %rax,$lo0 # discarded
260 mov 8(%rsp),$lo0 # tp[1]
271 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
274 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
278 mulq $m0 # ap[j]*bp[i]
282 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
292 mov ($ap),%rax # ap[0]
294 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
297 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
303 add $lo0,$hi1 # pull upmost overflow bit
305 mov $hi1,-8(%rsp,$num,8)
306 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
312 xor $i,$i # i=0 and clear CF!
313 mov (%rsp),%rax # tp[0]
314 lea (%rsp),$ap # borrow ap for tp
318 .Lsub: sbb ($np,$i,8),%rax
319 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
320 mov 8($ap,$i,8),%rax # tp[i+1]
322 dec $j # doesn't affect CF!
325 sbb \$0,%rax # handle upmost overflow bit
332 or $np,$ap # ap=borrow?tp:rp
334 .Lcopy: # copy or in-place refresh
336 mov $i,(%rsp,$i,8) # zap temporary vector
337 mov %rax,($rp,$i,8) # rp[i]=tp[i]
342 mov 8(%rsp,$num,8),%rsi # restore %rsp
358 .cfi_def_cfa_register %rsp
362 .size bn_mul_mont,.-bn_mul_mont
365 my @A=("%r10","%r11");
366 my @N=("%r13","%rdi");
368 .type bn_mul4x_mont,\@function,6
374 .cfi_def_cfa_register %rax
377 $code.=<<___ if ($addx);
398 lea -32(%rsp,$num,8),%r10 # future alloca(8*(num+4))
400 and \$-1024,%r10 # minimize TLB usage
408 jmp .Lmul4x_page_walk_done
415 .Lmul4x_page_walk_done:
417 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
418 .cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
420 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
421 mov %rdx,%r12 # reassign $bp
425 mov ($n0),$n0 # pull n0[0] value
426 mov ($bp),$m0 # m0=bp[0]
433 mulq $m0 # ap[0]*bp[0]
437 imulq $A[0],$m1 # "tp[0]"*n0
441 add %rax,$A[0] # discarded
464 mulq $m0 # ap[j]*bp[0]
466 mov -16($np,$j,8),%rax
472 mov -8($ap,$j,8),%rax
474 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
476 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
479 mulq $m0 # ap[j]*bp[0]
481 mov -8($np,$j,8),%rax
489 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
491 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
494 mulq $m0 # ap[j]*bp[0]
504 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
506 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
509 mulq $m0 # ap[j]*bp[0]
518 mov -16($ap,$j,8),%rax
520 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
522 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
527 mulq $m0 # ap[j]*bp[0]
529 mov -16($np,$j,8),%rax
535 mov -8($ap,$j,8),%rax
537 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
539 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
542 mulq $m0 # ap[j]*bp[0]
544 mov -8($np,$j,8),%rax
550 mov ($ap),%rax # ap[0]
552 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
554 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
560 mov $N[0],-8(%rsp,$j,8)
561 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
566 mov ($bp,$i,8),$m0 # m0=bp[i]
570 mulq $m0 # ap[0]*bp[i]
571 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
575 imulq $A[0],$m1 # tp[0]*n0
579 add %rax,$A[0] # "$N[0]", discarded
584 mulq $m0 # ap[j]*bp[i]
588 add 8(%rsp),$A[1] # +tp[1]
596 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
599 mov $N[1],(%rsp) # tp[j-1]
604 mulq $m0 # ap[j]*bp[i]
606 mov -16($np,$j,8),%rax
608 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
614 mov -8($ap,$j,8),%rax
618 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
621 mulq $m0 # ap[j]*bp[i]
623 mov -8($np,$j,8),%rax
625 add -8(%rsp,$j,8),$A[1]
635 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
638 mulq $m0 # ap[j]*bp[i]
642 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
652 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
655 mulq $m0 # ap[j]*bp[i]
659 add 8(%rsp,$j,8),$A[1]
666 mov -16($ap,$j,8),%rax
670 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
675 mulq $m0 # ap[j]*bp[i]
677 mov -16($np,$j,8),%rax
679 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
685 mov -8($ap,$j,8),%rax
689 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
692 mulq $m0 # ap[j]*bp[i]
694 mov -8($np,$j,8),%rax
696 add -8(%rsp,$j,8),$A[1]
703 mov ($ap),%rax # ap[0]
707 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
713 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
715 mov $N[0],-8(%rsp,$j,8)
716 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
722 my @ri=("%rax","%rdx",$m0,$m1);
724 mov 16(%rsp,$num,8),$rp # restore $rp
726 mov 0(%rsp),@ri[0] # tp[0]
728 mov 8(%rsp),@ri[1] # tp[1]
729 shr \$2,$j # j=num/4-1
730 lea (%rsp),$ap # borrow ap for tp
731 xor $i,$i # i=0 and clear CF!
734 mov 16($ap),@ri[2] # tp[2]
735 mov 24($ap),@ri[3] # tp[3]
740 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
741 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
742 sbb 16($np,$i,8),@ri[2]
743 mov 32($ap,$i,8),@ri[0] # tp[i+1]
744 mov 40($ap,$i,8),@ri[1]
745 sbb 24($np,$i,8),@ri[3]
746 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
747 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
748 sbb 32($np,$i,8),@ri[0]
749 mov 48($ap,$i,8),@ri[2]
750 mov 56($ap,$i,8),@ri[3]
751 sbb 40($np,$i,8),@ri[1]
753 dec $j # doesn't affect CF!
756 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
757 mov 32($ap,$i,8),@ri[0] # load overflow bit
758 sbb 16($np,$i,8),@ri[2]
759 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
760 sbb 24($np,$i,8),@ri[3]
761 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
763 sbb \$0,@ri[0] # handle upmost overflow bit
764 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
771 or $np,$ap # ap=borrow?tp:rp
772 shr \$2,$j # j=num/4-1
779 .Lcopy4x: # copy or in-place refresh
780 movdqu 16($ap,$i),%xmm2
781 movdqu 32($ap,$i),%xmm1
782 movdqa %xmm0,16(%rsp,$i)
783 movdqu %xmm2,16($rp,$i)
784 movdqa %xmm0,32(%rsp,$i)
785 movdqu %xmm1,32($rp,$i)
790 movdqu 16($ap,$i),%xmm2
791 movdqa %xmm0,16(%rsp,$i)
792 movdqu %xmm2,16($rp,$i)
796 mov 8(%rsp,$num,8),%rsi # restore %rsp
812 .cfi_def_cfa_register %rsp
816 .size bn_mul4x_mont,.-bn_mul4x_mont
820 ######################################################################
821 # void bn_sqr8x_mont(
822 my $rptr="%rdi"; # const BN_ULONG *rptr,
823 my $aptr="%rsi"; # const BN_ULONG *aptr,
824 my $bptr="%rdx"; # not used
825 my $nptr="%rcx"; # const BN_ULONG *nptr,
826 my $n0 ="%r8"; # const BN_ULONG *n0);
827 my $num ="%r9"; # int num, has to be divisible by 8
829 my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
830 my @A0=("%r10","%r11");
831 my @A1=("%r12","%r13");
832 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
834 $code.=<<___ if ($addx);
835 .extern bn_sqrx8x_internal # see x86_64-mont5 module
838 .extern bn_sqr8x_internal # see x86_64-mont5 module
840 .type bn_sqr8x_mont,\@function,6
845 .cfi_def_cfa_register %rax
862 shl \$3,${num}d # convert $num to bytes
863 shl \$3+2,%r10 # 4*$num
866 ##############################################################
867 # ensure that stack frame doesn't alias with $aptr modulo
868 # 4096. this is done to allow memory disambiguation logic
871 lea -64(%rsp,$num,2),%r11
878 sub %r11,%rbp # align with $aptr
879 lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
884 lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
885 lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
899 jmp .Lsqr8x_page_walk_done
907 .Lsqr8x_page_walk_done:
913 mov %rax, 40(%rsp) # save original %rsp
914 .cfi_cfa_expression %rsp+40,deref,+8
917 movq $nptr, %xmm2 # save pointer to modulus
919 movq $rptr,%xmm1 # save $rptr
920 movq %r10, %xmm3 # -$num
922 $code.=<<___ if ($addx);
923 mov OPENSSL_ia32cap_P+8(%rip),%eax
928 call bn_sqrx8x_internal # see x86_64-mont5 module
929 # %rax top-most carry
932 # %r8 end of tp[2*num]
937 sar \$3+2,%rcx # %cf=0
944 call bn_sqr8x_internal # see x86_64-mont5 module
945 # %rax top-most carry
948 # %rdi end of tp[2*num]
953 sar \$3+2,%rcx # %cf=0
973 inc %rcx # preserves %cf
976 sbb \$0,%rax # top-most carry
977 lea (%rbx,$num),%rbx # rewind
978 lea ($rptr,$num),$rptr # rewind
982 pshufd \$0,%xmm1,%xmm1
983 mov 40(%rsp),%rsi # restore %rsp
985 jmp .Lsqr8x_cond_copy
989 movdqa 16*0(%rbx),%xmm2
990 movdqa 16*1(%rbx),%xmm3
992 movdqu 16*0($rptr),%xmm4
993 movdqu 16*1($rptr),%xmm5
994 lea 16*2($rptr),$rptr
995 movdqa %xmm0,-16*2(%rbx) # zero tp
996 movdqa %xmm0,-16*1(%rbx)
997 movdqa %xmm0,-16*2(%rbx,%rdx)
998 movdqa %xmm0,-16*1(%rbx,%rdx)
1007 movdqu %xmm4,-16*2($rptr)
1008 movdqu %xmm5,-16*1($rptr)
1010 jnz .Lsqr8x_cond_copy
1026 .cfi_def_cfa_register %rsp
1030 .size bn_sqr8x_mont,.-bn_sqr8x_mont
1035 my $bp="%rdx"; # original value
1038 .type bn_mulx4x_mont,\@function,6
1043 .cfi_def_cfa_register %rax
1059 shl \$3,${num}d # convert $num to bytes
1061 sub $num,%r10 # -$num
1063 lea -72(%rsp,%r10),%rbp # future alloca(frame+$num+8)
1068 lea (%rbp,%r11),%rsp
1071 ja .Lmulx4x_page_walk
1072 jmp .Lmulx4x_page_walk_done
1076 lea -4096(%rsp),%rsp
1079 ja .Lmulx4x_page_walk
1080 .Lmulx4x_page_walk_done:
1083 ##############################################################
1086 # +8 off-loaded &b[i]
1095 mov $num,0(%rsp) # save $num
1097 mov %r10,16(%rsp) # end of b[num]
1099 mov $n0, 24(%rsp) # save *n0
1100 mov $rp, 32(%rsp) # save $rp
1101 mov %rax,40(%rsp) # save original %rsp
1102 .cfi_cfa_expression %rsp+40,deref,+8
1103 mov $num,48(%rsp) # inner counter
1109 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
1110 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
1114 mov ($bp),%rdx # b[0], $bp==%rdx actually
1115 lea 64+32(%rsp),$tptr
1118 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
1119 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
1121 mov $bptr,8(%rsp) # off-load &b[i]
1122 mulx 2*8($aptr),%r12,%r13 # ...
1126 mov $mi,$bptr # borrow $bptr
1127 imulq 24(%rsp),$mi # "t[0]"*n0
1128 xor $zero,$zero # cf=0, of=0
1130 mulx 3*8($aptr),%rax,%r14
1132 lea 4*8($aptr),$aptr
1134 adcx $zero,%r14 # cf=0
1136 mulx 0*8($nptr),%rax,%r10
1137 adcx %rax,$bptr # discarded
1139 mulx 1*8($nptr),%rax,%r11
1142 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
1143 mov 48(%rsp),$bptr # counter value
1144 mov %r10,-4*8($tptr)
1147 mulx 3*8($nptr),%rax,%r15
1149 mov %r11,-3*8($tptr)
1151 adox $zero,%r15 # of=0
1152 lea 4*8($nptr),$nptr
1153 mov %r12,-2*8($tptr)
1159 adcx $zero,%r15 # cf=0, modulo-scheduled
1160 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
1162 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
1164 mulx 2*8($aptr),%r12,%rax # ...
1166 mulx 3*8($aptr),%r13,%r14
1170 adcx $zero,%r14 # cf=0
1171 lea 4*8($aptr),$aptr
1172 lea 4*8($tptr),$tptr
1175 mulx 0*8($nptr),%rax,%r15
1178 mulx 1*8($nptr),%rax,%r15
1181 mulx 2*8($nptr),%rax,%r15
1182 mov %r10,-5*8($tptr)
1184 mov %r11,-4*8($tptr)
1186 mulx 3*8($nptr),%rax,%r15
1188 mov %r12,-3*8($tptr)
1191 lea 4*8($nptr),$nptr
1192 mov %r13,-2*8($tptr)
1194 dec $bptr # of=0, pass cf
1197 mov 0(%rsp),$num # load num
1198 mov 8(%rsp),$bptr # re-load &b[i]
1199 adc $zero,%r15 # modulo-scheduled
1201 sbb %r15,%r15 # top-most carry
1202 mov %r14,-1*8($tptr)
1207 mov ($bptr),%rdx # b[i]
1208 lea 8($bptr),$bptr # b++
1209 sub $num,$aptr # rewind $aptr
1210 mov %r15,($tptr) # save top-most carry
1211 lea 64+4*8(%rsp),$tptr
1212 sub $num,$nptr # rewind $nptr
1214 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
1215 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1217 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
1218 adox -4*8($tptr),$mi
1220 mulx 2*8($aptr),%r15,%r13 # ...
1221 adox -3*8($tptr),%r11
1223 adox -2*8($tptr),%r12
1227 mov $bptr,8(%rsp) # off-load &b[i]
1229 imulq 24(%rsp),$mi # "t[0]"*n0
1230 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1232 mulx 3*8($aptr),%rax,%r14
1235 adox -1*8($tptr),%r13
1237 lea 4*8($aptr),$aptr
1240 mulx 0*8($nptr),%rax,%r10
1241 adcx %rax,%r15 # discarded
1243 mulx 1*8($nptr),%rax,%r11
1246 mulx 2*8($nptr),%rax,%r12
1247 mov %r10,-4*8($tptr)
1250 mulx 3*8($nptr),%rax,%r15
1252 mov %r11,-3*8($tptr)
1253 lea 4*8($nptr),$nptr
1255 adox $zero,%r15 # of=0
1256 mov 48(%rsp),$bptr # counter value
1257 mov %r12,-2*8($tptr)
1263 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1264 adcx $zero,%r15 # cf=0, modulo-scheduled
1266 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1267 adcx 0*8($tptr),%r10
1269 mulx 2*8($aptr),%r12,%rax # ...
1270 adcx 1*8($tptr),%r11
1272 mulx 3*8($aptr),%r13,%r14
1274 adcx 2*8($tptr),%r12
1276 adcx 3*8($tptr),%r13
1277 adox $zero,%r14 # of=0
1278 lea 4*8($aptr),$aptr
1279 lea 4*8($tptr),$tptr
1280 adcx $zero,%r14 # cf=0
1283 mulx 0*8($nptr),%rax,%r15
1286 mulx 1*8($nptr),%rax,%r15
1289 mulx 2*8($nptr),%rax,%r15
1290 mov %r10,-5*8($tptr)
1293 mulx 3*8($nptr),%rax,%r15
1295 mov %r11,-4*8($tptr)
1296 mov %r12,-3*8($tptr)
1299 lea 4*8($nptr),$nptr
1300 mov %r13,-2*8($tptr)
1302 dec $bptr # of=0, pass cf
1305 mov 0(%rsp),$num # load num
1306 mov 8(%rsp),$bptr # re-load &b[i]
1307 adc $zero,%r15 # modulo-scheduled
1308 sub 0*8($tptr),$zero # pull top-most carry
1310 sbb %r15,%r15 # top-most carry
1311 mov %r14,-1*8($tptr)
1317 sub $num,$nptr # rewind $nptr
1320 shr \$3+2,$num # %cf=0
1321 mov 32(%rsp),$rptr # restore rp
1330 lea 8*4($tptr),$tptr
1335 lea 8*4($nptr),$nptr
1340 lea 8*4($rptr),$rptr
1341 dec $num # preserves %cf
1344 sbb \$0,%r15 # top-most carry
1346 sub %rdx,$rptr # rewind
1350 pshufd \$0,%xmm1,%xmm1
1351 mov 40(%rsp),%rsi # restore %rsp
1353 jmp .Lmulx4x_cond_copy
1357 movdqa 16*0($tptr),%xmm2
1358 movdqa 16*1($tptr),%xmm3
1359 lea 16*2($tptr),$tptr
1360 movdqu 16*0($rptr),%xmm4
1361 movdqu 16*1($rptr),%xmm5
1362 lea 16*2($rptr),$rptr
1363 movdqa %xmm0,-16*2($tptr) # zero tp
1364 movdqa %xmm0,-16*1($tptr)
1373 movdqu %xmm4,-16*2($rptr)
1374 movdqu %xmm5,-16*1($rptr)
1376 jnz .Lmulx4x_cond_copy
1394 .cfi_def_cfa_register %rsp
1398 .size bn_mulx4x_mont,.-bn_mulx4x_mont
1402 .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1406 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1407 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1415 .extern __imp_RtlVirtualUnwind
1416 .type mul_handler,\@abi-omnipotent
1430 mov 120($context),%rax # pull context->Rax
1431 mov 248($context),%rbx # pull context->Rip
1433 mov 8($disp),%rsi # disp->ImageBase
1434 mov 56($disp),%r11 # disp->HandlerData
1436 mov 0(%r11),%r10d # HandlerData[0]
1437 lea (%rsi,%r10),%r10 # end of prologue label
1438 cmp %r10,%rbx # context->Rip<end of prologue label
1439 jb .Lcommon_seh_tail
1441 mov 152($context),%rax # pull context->Rsp
1443 mov 4(%r11),%r10d # HandlerData[1]
1444 lea (%rsi,%r10),%r10 # epilogue label
1445 cmp %r10,%rbx # context->Rip>=epilogue label
1446 jae .Lcommon_seh_tail
1448 mov 192($context),%r10 # pull $num
1449 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1451 jmp .Lcommon_pop_regs
1452 .size mul_handler,.-mul_handler
1454 .type sqr_handler,\@abi-omnipotent
1468 mov 120($context),%rax # pull context->Rax
1469 mov 248($context),%rbx # pull context->Rip
1471 mov 8($disp),%rsi # disp->ImageBase
1472 mov 56($disp),%r11 # disp->HandlerData
1474 mov 0(%r11),%r10d # HandlerData[0]
1475 lea (%rsi,%r10),%r10 # end of prologue label
1476 cmp %r10,%rbx # context->Rip<.Lsqr_prologue
1477 jb .Lcommon_seh_tail
1479 mov 4(%r11),%r10d # HandlerData[1]
1480 lea (%rsi,%r10),%r10 # body label
1481 cmp %r10,%rbx # context->Rip<.Lsqr_body
1482 jb .Lcommon_pop_regs
1484 mov 152($context),%rax # pull context->Rsp
1486 mov 8(%r11),%r10d # HandlerData[2]
1487 lea (%rsi,%r10),%r10 # epilogue label
1488 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
1489 jae .Lcommon_seh_tail
1491 mov 40(%rax),%rax # pull saved stack pointer
1500 mov %rbx,144($context) # restore context->Rbx
1501 mov %rbp,160($context) # restore context->Rbp
1502 mov %r12,216($context) # restore context->R12
1503 mov %r13,224($context) # restore context->R13
1504 mov %r14,232($context) # restore context->R14
1505 mov %r15,240($context) # restore context->R15
1510 mov %rax,152($context) # restore context->Rsp
1511 mov %rsi,168($context) # restore context->Rsi
1512 mov %rdi,176($context) # restore context->Rdi
1514 mov 40($disp),%rdi # disp->ContextRecord
1515 mov $context,%rsi # context
1516 mov \$154,%ecx # sizeof(CONTEXT)
1517 .long 0xa548f3fc # cld; rep movsq
1520 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1521 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1522 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1523 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1524 mov 40(%rsi),%r10 # disp->ContextRecord
1525 lea 56(%rsi),%r11 # &disp->HandlerData
1526 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1527 mov %r10,32(%rsp) # arg5
1528 mov %r11,40(%rsp) # arg6
1529 mov %r12,48(%rsp) # arg7
1530 mov %rcx,56(%rsp) # arg8, (NULL)
1531 call *__imp_RtlVirtualUnwind(%rip)
1533 mov \$1,%eax # ExceptionContinueSearch
1545 .size sqr_handler,.-sqr_handler
1549 .rva .LSEH_begin_bn_mul_mont
1550 .rva .LSEH_end_bn_mul_mont
1551 .rva .LSEH_info_bn_mul_mont
1553 .rva .LSEH_begin_bn_mul4x_mont
1554 .rva .LSEH_end_bn_mul4x_mont
1555 .rva .LSEH_info_bn_mul4x_mont
1557 .rva .LSEH_begin_bn_sqr8x_mont
1558 .rva .LSEH_end_bn_sqr8x_mont
1559 .rva .LSEH_info_bn_sqr8x_mont
1561 $code.=<<___ if ($addx);
1562 .rva .LSEH_begin_bn_mulx4x_mont
1563 .rva .LSEH_end_bn_mulx4x_mont
1564 .rva .LSEH_info_bn_mulx4x_mont
1569 .LSEH_info_bn_mul_mont:
1572 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
1573 .LSEH_info_bn_mul4x_mont:
1576 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1577 .LSEH_info_bn_sqr8x_mont:
1580 .rva .Lsqr8x_prologue,.Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
1583 $code.=<<___ if ($addx);
1584 .LSEH_info_bn_mulx4x_mont:
1587 .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]