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 # Companion to x86_64-mont.pl that optimizes cache-timing attack
13 # countermeasures. The subroutines are produced by replacing bp[i]
14 # references in their x86_64-mont.pl counterparts with cache-neutral
15 # references to powers table computed in BN_mod_exp_mont_consttime.
16 # In addition subroutine that scatters elements of the powers table
17 # is implemented, so that scatter-/gathering can be tuned without
18 # bn_exp.c modifications.
22 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
24 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
26 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
27 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
28 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
29 die "can't locate x86_64-xlate.pl";
31 open OUT,"| \"$^X\" $xlate $flavour $output";
34 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
35 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
39 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
40 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
44 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
45 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
49 # int bn_mul_mont_gather5(
50 $rp="%rdi"; # BN_ULONG *rp,
51 $ap="%rsi"; # const BN_ULONG *ap,
52 $bp="%rdx"; # const BN_ULONG *bp,
53 $np="%rcx"; # const BN_ULONG *np,
54 $n0="%r8"; # const BN_ULONG *n0,
55 $num="%r9"; # int num,
56 # int idx); # 0 to 2^5-1, "index" in $bp holding
57 # pre-computed powers of a', interlaced
58 # in such manner that b[0] is $bp[idx],
59 # b[1] is [2^5+idx], etc.
71 .extern OPENSSL_ia32cap_P
73 .globl bn_mul_mont_gather5
74 .type bn_mul_mont_gather5,\@function,6
82 $code.=<<___ if ($addx);
83 mov OPENSSL_ia32cap_P+8(%rip),%r11d
91 mov `($win64?56:8)`(%rsp),%r10d # load 7th argument
99 $code.=<<___ if ($win64);
102 movaps %xmm7,0x10(%rsp)
109 lea (%rsp,%r11,8),%rsp # tp=alloca(8*(num+2))
110 and \$-1024,%rsp # minimize TLB usage
112 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
114 mov $bp,%r12 # reassign $bp
117 $STRIDE=2**5*8; # 5 is "window size"
118 $N=$STRIDE/4; # should match cache line size
121 shr \$`log($N/8)/log(2)`,%r10
124 lea .Lmagic_masks(%rip),%rax
125 and \$`2**5/($N/8)-1`,%r10 # 5 is "window size"
126 lea 96($bp,%r11,8),$bp # pointer within 1st cache line
127 movq 0(%rax,%r10,8),%xmm4 # set of masks denoting which
128 movq 8(%rax,%r10,8),%xmm5 # cache line contains element
129 movq 16(%rax,%r10,8),%xmm6 # denoted by 7th argument
130 movq 24(%rax,%r10,8),%xmm7
132 movq `0*$STRIDE/4-96`($bp),%xmm0
133 movq `1*$STRIDE/4-96`($bp),%xmm1
135 movq `2*$STRIDE/4-96`($bp),%xmm2
137 movq `3*$STRIDE/4-96`($bp),%xmm3
145 movq %xmm0,$m0 # m0=bp[0]
147 mov ($n0),$n0 # pull n0[0] value
153 movq `0*$STRIDE/4-96`($bp),%xmm0
154 movq `1*$STRIDE/4-96`($bp),%xmm1
156 movq `2*$STRIDE/4-96`($bp),%xmm2
160 mulq $m0 # ap[0]*bp[0]
164 movq `3*$STRIDE/4-96`($bp),%xmm3
169 imulq $lo0,$m1 # "tp[0]"*n0
177 add %rax,$lo0 # discarded
190 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
193 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
197 mulq $m0 # ap[j]*bp[0]
208 movq %xmm0,$m0 # bp[1]
211 mov ($ap),%rax # ap[0]
213 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
215 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
222 mov $hi1,-8(%rsp,$num,8)
223 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
233 movq `0*$STRIDE/4-96`($bp),%xmm0
234 movq `1*$STRIDE/4-96`($bp),%xmm1
236 movq `2*$STRIDE/4-96`($bp),%xmm2
239 mulq $m0 # ap[0]*bp[i]
240 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
244 movq `3*$STRIDE/4-96`($bp),%xmm3
249 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]
291 movq %xmm0,$m0 # bp[i+1]
294 mov ($ap),%rax # ap[0]
296 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
299 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
305 add $lo0,$hi1 # pull upmost overflow bit
307 mov $hi1,-8(%rsp,$num,8)
308 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
314 xor $i,$i # i=0 and clear CF!
315 mov (%rsp),%rax # tp[0]
316 lea (%rsp),$ap # borrow ap for tp
320 .Lsub: sbb ($np,$i,8),%rax
321 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
322 mov 8($ap,$i,8),%rax # tp[i+1]
324 dec $j # doesnn't affect CF!
327 sbb \$0,%rax # handle upmost overflow bit
334 or $np,$ap # ap=borrow?tp:rp
336 .Lcopy: # copy or in-place refresh
338 mov $i,(%rsp,$i,8) # zap temporary vector
339 mov %rax,($rp,$i,8) # rp[i]=tp[i]
344 mov 8(%rsp,$num,8),%rsi # restore %rsp
347 $code.=<<___ if ($win64);
349 movaps 0x10(%rsi),%xmm7
362 .size bn_mul_mont_gather5,.-bn_mul_mont_gather5
365 my @A=("%r10","%r11");
366 my @N=("%r13","%rdi");
368 .type bn_mul4x_mont_gather5,\@function,6
370 bn_mul4x_mont_gather5:
373 $code.=<<___ if ($addx);
380 mov `($win64?56:8)`(%rsp),%r10d # load 7th argument
388 $code.=<<___ if ($win64);
391 movaps %xmm7,0x10(%rsp)
398 lea (%rsp,%r11,8),%rsp # tp=alloca(8*(num+4))
399 and \$-1024,%rsp # minimize TLB usage
401 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
403 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
404 mov %rdx,%r12 # reassign $bp
407 $STRIDE=2**5*8; # 5 is "window size"
408 $N=$STRIDE/4; # should match cache line size
411 shr \$`log($N/8)/log(2)`,%r10
414 lea .Lmagic_masks(%rip),%rax
415 and \$`2**5/($N/8)-1`,%r10 # 5 is "window size"
416 lea 96($bp,%r11,8),$bp # pointer within 1st cache line
417 movq 0(%rax,%r10,8),%xmm4 # set of masks denoting which
418 movq 8(%rax,%r10,8),%xmm5 # cache line contains element
419 movq 16(%rax,%r10,8),%xmm6 # denoted by 7th argument
420 movq 24(%rax,%r10,8),%xmm7
422 movq `0*$STRIDE/4-96`($bp),%xmm0
423 movq `1*$STRIDE/4-96`($bp),%xmm1
425 movq `2*$STRIDE/4-96`($bp),%xmm2
427 movq `3*$STRIDE/4-96`($bp),%xmm3
435 movq %xmm0,$m0 # m0=bp[0]
436 mov ($n0),$n0 # pull n0[0] value
442 movq `0*$STRIDE/4-96`($bp),%xmm0
443 movq `1*$STRIDE/4-96`($bp),%xmm1
445 movq `2*$STRIDE/4-96`($bp),%xmm2
449 mulq $m0 # ap[0]*bp[0]
453 movq `3*$STRIDE/4-96`($bp),%xmm3
458 imulq $A[0],$m1 # "tp[0]"*n0
466 add %rax,$A[0] # discarded
489 mulq $m0 # ap[j]*bp[0]
491 mov -16($np,$j,8),%rax
497 mov -8($ap,$j,8),%rax
499 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
501 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
504 mulq $m0 # ap[j]*bp[0]
506 mov -8($np,$j,8),%rax
514 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
516 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
519 mulq $m0 # ap[j]*bp[0]
529 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
531 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
534 mulq $m0 # ap[j]*bp[0]
543 mov -16($ap,$j,8),%rax
545 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
547 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
552 mulq $m0 # ap[j]*bp[0]
554 mov -16($np,$j,8),%rax
560 mov -8($ap,$j,8),%rax
562 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
564 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
567 mulq $m0 # ap[j]*bp[0]
569 mov -8($np,$j,8),%rax
575 mov ($ap),%rax # ap[0]
577 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
579 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
582 movq %xmm0,$m0 # bp[1]
587 mov $N[0],-8(%rsp,$j,8)
588 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
594 movq `0*$STRIDE/4-96`($bp),%xmm0
595 movq `1*$STRIDE/4-96`($bp),%xmm1
597 movq `2*$STRIDE/4-96`($bp),%xmm2
602 mulq $m0 # ap[0]*bp[i]
603 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
607 movq `3*$STRIDE/4-96`($bp),%xmm3
612 imulq $A[0],$m1 # tp[0]*n0
620 add %rax,$A[0] # "$N[0]", discarded
625 mulq $m0 # ap[j]*bp[i]
629 add 8(%rsp),$A[1] # +tp[1]
637 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
644 mulq $m0 # ap[j]*bp[i]
646 mov -16($np,$j,8),%rax
648 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
654 mov -8($ap,$j,8),%rax
658 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
661 mulq $m0 # ap[j]*bp[i]
663 mov -8($np,$j,8),%rax
665 add -8(%rsp,$j,8),$A[1]
675 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
678 mulq $m0 # ap[j]*bp[i]
682 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
692 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
695 mulq $m0 # ap[j]*bp[i]
699 add 8(%rsp,$j,8),$A[1]
706 mov -16($ap,$j,8),%rax
710 mov $N[0],-40(%rsp,$j,8) # tp[j-1]
715 mulq $m0 # ap[j]*bp[i]
717 mov -16($np,$j,8),%rax
719 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
725 mov -8($ap,$j,8),%rax
729 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
732 mulq $m0 # ap[j]*bp[i]
734 mov -8($np,$j,8),%rax
736 add -8(%rsp,$j,8),$A[1]
743 mov ($ap),%rax # ap[0]
747 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
750 movq %xmm0,$m0 # bp[i+1]
751 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
756 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
758 mov $N[0],-8(%rsp,$j,8)
759 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
765 my @ri=("%rax","%rdx",$m0,$m1);
767 mov 16(%rsp,$num,8),$rp # restore $rp
768 mov 0(%rsp),@ri[0] # tp[0]
770 mov 8(%rsp),@ri[1] # tp[1]
771 shr \$2,$num # num/=4
772 lea (%rsp),$ap # borrow ap for tp
773 xor $i,$i # i=0 and clear CF!
776 mov 16($ap),@ri[2] # tp[2]
777 mov 24($ap),@ri[3] # tp[3]
779 lea -1($num),$j # j=num/4-1
783 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
784 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
785 sbb 16($np,$i,8),@ri[2]
786 mov 32($ap,$i,8),@ri[0] # tp[i+1]
787 mov 40($ap,$i,8),@ri[1]
788 sbb 24($np,$i,8),@ri[3]
789 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
790 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
791 sbb 32($np,$i,8),@ri[0]
792 mov 48($ap,$i,8),@ri[2]
793 mov 56($ap,$i,8),@ri[3]
794 sbb 40($np,$i,8),@ri[1]
796 dec $j # doesnn't affect CF!
799 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
800 mov 32($ap,$i,8),@ri[0] # load overflow bit
801 sbb 16($np,$i,8),@ri[2]
802 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
803 sbb 24($np,$i,8),@ri[3]
804 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
806 sbb \$0,@ri[0] # handle upmost overflow bit
807 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
814 or $np,$ap # ap=borrow?tp:rp
821 .Lcopy4x: # copy or in-place refresh
822 movdqu 16($ap,$i),%xmm2
823 movdqu 32($ap,$i),%xmm1
824 movdqa %xmm0,16(%rsp,$i)
825 movdqu %xmm2,16($rp,$i)
826 movdqa %xmm0,32(%rsp,$i)
827 movdqu %xmm1,32($rp,$i)
833 movdqu 16($ap,$i),%xmm2
834 movdqa %xmm0,16(%rsp,$i)
835 movdqu %xmm2,16($rp,$i)
839 mov 8(%rsp,$num,8),%rsi # restore %rsp
842 $code.=<<___ if ($win64);
844 movaps 0x10(%rsi),%xmm7
857 .size bn_mul4x_mont_gather5,.-bn_mul4x_mont_gather5
861 my $bp="%rdx"; # original value
864 .type bn_mulx4x_mont_gather5,\@function,6
866 bn_mulx4x_mont_gather5:
876 $code.=<<___ if ($win64);
879 movaps %xmm7,0x10(%rsp)
882 shl \$3,${num}d # convert $num to bytes
884 mov %rsp,%r11 # put aside %rsp
885 sub $num,%r10 # -$num
887 lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
889 ##############################################################
892 # +8 off-loaded &b[i]
901 mov $num,0(%rsp) # save $num
903 lea 256($bp,$num),%r10
905 mov %r10,16(%rsp) # end of b[num]
907 mov $n0, 24(%rsp) # save *n0
908 mov $rp, 32(%rsp) # save $rp
909 mov $num,48(%rsp) # inner counter
910 mov %r11,56(%rsp) # save original %rsp
916 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
917 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
919 my $STRIDE=2**5*8; # 5 is "window size"
920 my $N=$STRIDE/4; # should match cache line size
922 mov `($win64?56:8)`(%rax),%r10d # load 7th argument
924 shr \$`log($N/8)/log(2)`,%r10
927 lea .Lmagic_masks(%rip),%rax
928 and \$`2**5/($N/8)-1`,%r10 # 5 is "window size"
929 lea 96($bp,%r11,8),$bptr # pointer within 1st cache line
930 movq 0(%rax,%r10,8),%xmm4 # set of masks denoting which
931 movq 8(%rax,%r10,8),%xmm5 # cache line contains element
932 movq 16(%rax,%r10,8),%xmm6 # denoted by 7th argument
933 movq 24(%rax,%r10,8),%xmm7
935 movq `0*$STRIDE/4-96`($bptr),%xmm0
936 movq `1*$STRIDE/4-96`($bptr),%xmm1
938 movq `2*$STRIDE/4-96`($bptr),%xmm2
940 movq `3*$STRIDE/4-96`($bptr),%xmm3
945 lea $STRIDE($bptr),$bptr
948 movq %xmm0,%rdx # bp[0]
949 movq `0*$STRIDE/4-96`($bptr),%xmm0
950 movq `1*$STRIDE/4-96`($bptr),%xmm1
952 movq `2*$STRIDE/4-96`($bptr),%xmm2
955 lea 64+32(%rsp),$tptr
957 xor $zero,$zero # of=0,cf=0
959 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
960 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
962 mulx 2*8($aptr),%r12,%r13 # ...
966 movq `3*$STRIDE/4-96`($bptr),%xmm3
967 lea $STRIDE($bptr),%r10 # next &b[i]
972 mov $mi,$bptr # borrow $bptr
973 imulq 24(%rsp),$mi # "t[0]"*n0
974 xor $zero,$zero # cf=0, of=0
978 mov %r10,8(%rsp) # off-load &b[i]
980 mulx 3*8($aptr),%rax,%r14
984 adcx $zero,%r14 # cf=0
986 mulx 0*8($nptr),%rax,%r10
987 adcx %rax,$bptr # discarded
989 mulx 1*8($nptr),%rax,%r11
992 mulx 2*8($nptr),%rax,%r12
993 mov 48(%rsp),$bptr # counter value
997 mulx 3*8($nptr),%rax,%r15
1001 adox $zero,%r15 # of=0
1002 lea 4*8($nptr),$nptr
1003 mov %r12,-2*8($tptr)
1009 adcx $zero,%r15 # cf=0, modulo-scheduled
1010 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
1012 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
1014 mulx 2*8($aptr),%r12,%rax # ...
1016 mulx 3*8($aptr),%r13,%r14
1020 adcx $zero,%r14 # cf=0
1021 lea 4*8($aptr),$aptr
1022 lea 4*8($tptr),$tptr
1025 mulx 0*8($nptr),%rax,%r15
1028 mulx 1*8($nptr),%rax,%r15
1032 mulx 2*8($nptr),%rax,%r15
1033 mov %r10,-5*8($tptr)
1034 mov %r11,-4*8($tptr)
1037 mulx 3*8($nptr),%rax,%r15
1039 mov %r12,-3*8($tptr)
1042 lea 4*8($nptr),$nptr
1043 mov %r13,-2*8($tptr)
1045 dec $bptr # of=0, pass cf
1048 mov 0(%rsp),$num # load num
1049 mov 8(%rsp),$bptr # re-load &b[i]
1050 movq %xmm0,%rdx # bp[1]
1051 adc $zero,%r15 # modulo-scheduled
1053 sbb %r15,%r15 # top-most carry
1054 mov %r14,-1*8($tptr)
1059 sub $num,$aptr # rewind $aptr
1060 mov %r15,($tptr) # save top-most carry
1063 sub $num,$nptr # rewind $nptr
1064 xor $zero,$zero # cf=0, of=0
1067 movq `0*$STRIDE/4-96`($bptr),%xmm0
1068 movq `1*$STRIDE/4-96`($bptr),%xmm1
1070 movq `2*$STRIDE/4-96`($bptr),%xmm2
1073 mulx 0*8($aptr),$mi,%rax # a[0]*b[i]
1076 mulx 1*8($aptr),%r11,%r14 # a[1]*b[i]
1078 mulx 2*8($aptr),%r12,%r13 # ...
1084 movq `3*$STRIDE/4-96`($bptr),%xmm3
1085 lea $STRIDE($bptr),%r10 # next &b[i]
1090 mov $mi,$bptr # borrow $bptr
1091 imulq 24(%rsp),$mi # "t[0]"*n0
1092 xor $zero,$zero # cf=0, of=0
1096 mov %r10,8(%rsp) # off-load &b[i]
1099 mulx 3*8($aptr),%rax,%r14
1103 adox 3*8($tptr),%r13
1105 lea 4*8($aptr),$aptr
1106 lea 4*8($tptr),$tptr
1109 mulx 0*8($nptr),%rax,%r10
1110 adcx %rax,$bptr # discarded
1112 mulx 1*8($nptr),%rax,%r11
1115 mulx 2*8($nptr),%rax,%r12
1117 mov %r10,-4*8($tptr)
1122 mulx 3*8($nptr),%rax,%r15
1124 mov %r11,-3*8($tptr)
1126 adox $zero,%r15 # of=0
1127 mov 48(%rsp),$bptr # counter value
1128 mov %r12,-2*8($tptr)
1129 lea 4*8($nptr),$nptr
1135 adcx $zero,%r15 # cf=0, modulo-scheduled
1137 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1140 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1142 mulx 2*8($aptr),%r12,%rax # ...
1145 mulx 3*8($aptr),%r13,%r14
1147 adcx 2*8($tptr),%r12
1149 adcx 3*8($tptr),%r13
1150 adox $zero,%r14 # of=0
1151 lea 4*8($aptr),$aptr
1152 .byte 0x48,0x8d,0x9b,0x20,0x00,0x00,0x00 # lea 4*8($tptr),$tptr
1153 adcx $zero,%r14 # cf=0
1156 .byte 0x3e,0xc4,0x62,0xfb,0xf6,0x79,0x00 # mulx 0*8($nptr),%rax,%r15
1159 mulx 1*8($nptr),%rax,%r15
1162 mulx 2*8($nptr),%rax,%r15
1163 mov %r10,-5*8($tptr)
1167 mulx 3*8($nptr),%rax,%r15
1169 mov %r11,-4*8($tptr)
1170 mov %r12,-3*8($tptr)
1173 lea 4*8($nptr),$nptr
1174 mov %r13,-2*8($tptr)
1176 dec $bptr # of=0, pass cf
1179 mov 0(%rsp),$num # load num
1180 mov 8(%rsp),$bptr # re-load &b[i]
1181 movq %xmm0,%rdx # bp[i+1]
1182 adc $zero,%r15 # modulo-scheduled
1183 sub %r10,$zero # pull top-most carry
1185 sbb %r15,%r15 # top-most carry
1186 mov %r14,-1*8($tptr)
1192 mov 32(%rsp),$rptr # restore rp
1197 mov 0*8($nptr,$num),%r8
1198 mov 1*8($nptr,$num),%r9
1200 jmp .Lmulx4x_sub_entry
1204 mov 0*8($nptr,$num),%r8
1205 mov 1*8($nptr,$num),%r9
1208 mov 2*8($nptr,$num),%r10
1211 mov 3*8($nptr,$num),%r11
1218 neg %rdx # mov %rdx,%cf
1221 movdqa %xmm0,($tptr)
1224 movdqa %xmm0,16($tptr)
1225 lea 4*8($tptr),$tptr
1226 sbb %rdx,%rdx # mov %cf,%rdx
1232 lea 4*8($rptr),$rptr
1237 mov 56(%rsp),%rsi # restore %rsp
1240 $code.=<<___ if ($win64);
1242 movaps 0x10(%rsi),%xmm7
1255 .size bn_mulx4x_mont_gather5,.-bn_mulx4x_mont_gather5
1259 my ($inp,$num,$tbl,$idx)=$win64?("%rcx","%rdx","%r8", "%r9") : # Win64 order
1260 ("%rdi","%rsi","%rdx","%rcx"); # Unix order
1267 .type bn_scatter5,\@abi-omnipotent
1271 jz .Lscatter_epilogue
1272 lea ($tbl,$idx,8),$tbl
1282 .size bn_scatter5,.-bn_scatter5
1285 .type bn_gather5,\@abi-omnipotent
1289 $code.=<<___ if ($win64);
1290 .LSEH_begin_bn_gather5:
1291 # I can't trust assembler to use specific encoding:-(
1292 .byte 0x48,0x83,0xec,0x28 #sub \$0x28,%rsp
1293 .byte 0x0f,0x29,0x34,0x24 #movaps %xmm6,(%rsp)
1294 .byte 0x0f,0x29,0x7c,0x24,0x10 #movdqa %xmm7,0x10(%rsp)
1298 shr \$`log($N/8)/log(2)`,$idx
1301 lea .Lmagic_masks(%rip),%rax
1302 and \$`2**5/($N/8)-1`,$idx # 5 is "window size"
1303 lea 96($tbl,%r11,8),$tbl # pointer within 1st cache line
1304 movq 0(%rax,$idx,8),%xmm4 # set of masks denoting which
1305 movq 8(%rax,$idx,8),%xmm5 # cache line contains element
1306 movq 16(%rax,$idx,8),%xmm6 # denoted by 7th argument
1307 movq 24(%rax,$idx,8),%xmm7
1311 movq `0*$STRIDE/4-96`($tbl),%xmm0
1312 movq `1*$STRIDE/4-96`($tbl),%xmm1
1314 movq `2*$STRIDE/4-96`($tbl),%xmm2
1316 movq `3*$STRIDE/4-96`($tbl),%xmm3
1321 lea $STRIDE($tbl),$tbl
1324 movq %xmm0,($out) # m0=bp[0]
1329 $code.=<<___ if ($win64);
1331 movaps 0x10(%rsp),%xmm7
1336 .LSEH_end_bn_gather5:
1337 .size bn_gather5,.-bn_gather5
1343 .long 0,0, 0,0, 0,0, -1,-1
1344 .long 0,0, 0,0, 0,0, 0,0
1345 .asciz "Montgomery Multiplication with scatter/gather for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1348 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1349 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1357 .extern __imp_RtlVirtualUnwind
1358 .type mul_handler,\@abi-omnipotent
1372 mov 120($context),%rax # pull context->Rax
1373 mov 248($context),%rbx # pull context->Rip
1375 mov 8($disp),%rsi # disp->ImageBase
1376 mov 56($disp),%r11 # disp->HandlerData
1378 mov 0(%r11),%r10d # HandlerData[0]
1379 lea (%rsi,%r10),%r10 # end of prologue label
1380 cmp %r10,%rbx # context->Rip<end of prologue label
1381 jb .Lcommon_seh_tail
1383 lea `40+48`(%rax),%rax
1385 mov 4(%r11),%r10d # HandlerData[1]
1386 lea (%rsi,%r10),%r10 # end of alloca label
1387 cmp %r10,%rbx # context->Rip<end of alloca label
1388 jb .Lcommon_seh_tail
1390 mov 152($context),%rax # pull context->Rsp
1392 mov 8(%r11),%r10d # HandlerData[2]
1393 lea (%rsi,%r10),%r10 # epilogue label
1394 cmp %r10,%rbx # context->Rip>=epilogue label
1395 jae .Lcommon_seh_tail
1397 mov 192($context),%r10 # pull $num
1398 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1401 movaps 16(%rax),%xmm1
1402 lea `40+48`(%rax),%rax
1410 mov %rbx,144($context) # restore context->Rbx
1411 mov %rbp,160($context) # restore context->Rbp
1412 mov %r12,216($context) # restore context->R12
1413 mov %r13,224($context) # restore context->R13
1414 mov %r14,232($context) # restore context->R14
1415 mov %r15,240($context) # restore context->R15
1416 movups %xmm0,512($context) # restore context->Xmm6
1417 movups %xmm1,528($context) # restore context->Xmm7
1422 mov %rax,152($context) # restore context->Rsp
1423 mov %rsi,168($context) # restore context->Rsi
1424 mov %rdi,176($context) # restore context->Rdi
1426 mov 40($disp),%rdi # disp->ContextRecord
1427 mov $context,%rsi # context
1428 mov \$154,%ecx # sizeof(CONTEXT)
1429 .long 0xa548f3fc # cld; rep movsq
1432 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1433 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1434 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1435 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1436 mov 40(%rsi),%r10 # disp->ContextRecord
1437 lea 56(%rsi),%r11 # &disp->HandlerData
1438 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1439 mov %r10,32(%rsp) # arg5
1440 mov %r11,40(%rsp) # arg6
1441 mov %r12,48(%rsp) # arg7
1442 mov %rcx,56(%rsp) # arg8, (NULL)
1443 call *__imp_RtlVirtualUnwind(%rip)
1445 mov \$1,%eax # ExceptionContinueSearch
1457 .size mul_handler,.-mul_handler
1461 .rva .LSEH_begin_bn_mul_mont_gather5
1462 .rva .LSEH_end_bn_mul_mont_gather5
1463 .rva .LSEH_info_bn_mul_mont_gather5
1465 .rva .LSEH_begin_bn_mul4x_mont_gather5
1466 .rva .LSEH_end_bn_mul4x_mont_gather5
1467 .rva .LSEH_info_bn_mul4x_mont_gather5
1469 .rva .LSEH_begin_bn_gather5
1470 .rva .LSEH_end_bn_gather5
1471 .rva .LSEH_info_bn_gather5
1475 .LSEH_info_bn_mul_mont_gather5:
1478 .rva .Lmul_alloca,.Lmul_body,.Lmul_epilogue # HandlerData[]
1480 .LSEH_info_bn_mul4x_mont_gather5:
1483 .rva .Lmul4x_alloca,.Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1485 .LSEH_info_bn_gather5:
1486 .byte 0x01,0x0d,0x05,0x00
1487 .byte 0x0d,0x78,0x01,0x00 #movaps 0x10(rsp),xmm7
1488 .byte 0x08,0x68,0x00,0x00 #movaps (rsp),xmm6
1489 .byte 0x04,0x42,0x00,0x00 #sub rsp,0x28
1494 $code =~ s/\`([^\`]*)\`/eval($1)/gem;