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 # This is AESNI-CBC+SHA1 "stitch" implementation. The idea, as spelled
13 # in http://download.intel.com/design/intarch/papers/323686.pdf, is
14 # that since AESNI-CBC encrypt exhibit *very* low instruction-level
15 # parallelism, interleaving it with another algorithm would allow to
16 # utilize processor resources better and achieve better performance.
17 # SHA1 instruction sequences(*) are taken from sha1-x86_64.pl and
18 # AESNI code is weaved into it. Below are performance numbers in
19 # cycles per processed byte, less is better, for standalone AESNI-CBC
20 # encrypt, sum of the latter and standalone SHA1, and "stitched"
23 # AES-128-CBC +SHA1 stitch gain
24 # Westmere 3.77[+5.5] 9.26 6.66 +39%
25 # Sandy Bridge 5.05[+5.0(6.2)] 10.06(11.21) 5.98(7.01) +68%(+60%)
26 # Ivy Bridge 5.05[+4.6] 9.65 5.54 +74%
27 # Haswell 4.43[+3.6(4.1)] 8.00(8.55) 4.55(5.21) +75%(+64%)
28 # Bulldozer 5.77[+6.0] 11.72 6.37 +84%
31 # Westmere 4.51 10.00 6.87 +46%
32 # Sandy Bridge 6.05 11.06(12.21) 6.11(7.20) +81%(+70%)
33 # Ivy Bridge 6.05 10.65 6.07 +75%
34 # Haswell 5.29 8.86(9.42) 5.32(5.32) +67%(+77%)
35 # Bulldozer 6.89 12.84 6.96 +84%
38 # Westmere 5.25 10.74 7.24 +48%
39 # Sandy Bridge 7.05 12.06(13.21) 7.12(7.63) +69%(+73%)
40 # Ivy Bridge 7.05 11.65 7.12 +64%
41 # Haswell 6.19 9.76(10.3) 6.21(6.25) +57%(+65%)
42 # Bulldozer 8.00 13.95 8.25 +69%
44 # (*) There are two code paths: SSSE3 and AVX. See sha1-568.pl for
45 # background information. Above numbers in parentheses are SSSE3
46 # results collected on AVX-capable CPU, i.e. apply on OSes that
49 # Needless to mention that it makes no sense to implement "stitched"
50 # *decrypt* subroutine. Because *both* AESNI-CBC decrypt and SHA1
51 # fully utilize parallelism, so stitching would not give any gain
52 # anyway. Well, there might be some, e.g. because of better cache
53 # locality... For reference, here are performance results for
54 # standalone AESNI-CBC decrypt:
56 # AES-128-CBC AES-192-CBC AES-256-CBC
57 # Westmere 1.25 1.50 1.75
58 # Sandy Bridge 0.74 0.91 1.09
59 # Ivy Bridge 0.74 0.90 1.11
60 # Haswell 0.63 0.76 0.88
61 # Bulldozer 0.70 0.85 0.99
65 # AES-256-CBC +SHA1 stitch gain
66 # Westmere 1.75 7.20 6.68 +7.8%
67 # Sandy Bridge 1.09 6.09(7.22) 5.82(6.95) +4.6%(+3.9%)
68 # Ivy Bridge 1.11 5.70 5.45 +4.6%
69 # Haswell 0.88 4.45(5.00) 4.39(4.69) +1.4%(+6.6%)
70 # Bulldozer 0.99 6.95 5.95 +17%
74 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
76 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
78 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
79 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
80 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
81 die "can't locate x86_64-xlate.pl";
83 $avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
84 =~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
86 $avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
87 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
89 $avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
90 `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
95 open OUT,"| \"$^X\" $xlate $flavour $output";
98 # void aesni_cbc_sha1_enc(const void *inp,
101 # const AES_KEY *key,
108 .extern OPENSSL_ia32cap_P
110 .globl aesni_cbc_sha1_enc
111 .type aesni_cbc_sha1_enc,\@abi-omnipotent
114 # caller should check for SSSE3 and AES-NI bits
115 mov OPENSSL_ia32cap_P+0(%rip),%r10d
116 mov OPENSSL_ia32cap_P+4(%rip),%r11d
118 $code.=<<___ if ($avx);
119 and \$`1<<28`,%r11d # mask AVX bit
120 and \$`1<<30`,%r10d # mask "Intel CPU" bit
122 cmp \$`1<<28|1<<30`,%r10d
123 je aesni_cbc_sha1_enc_avx
126 jmp aesni_cbc_sha1_enc_ssse3
128 .size aesni_cbc_sha1_enc,.-aesni_cbc_sha1_enc
131 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
134 my @X=map("%xmm$_",(4..7,0..3));
135 my @Tx=map("%xmm$_",(8..10));
136 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
137 my @T=("%esi","%edi");
138 my $j=0; my $jj=0; my $r=0; my $sn=0; my $rx=0;
140 my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13)); # for enc
141 my @rndkey=("%xmm14","%xmm15"); # for enc
142 my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15)); # for dec
144 if (1) { # reassign for Atom Silvermont
145 @X=map("%xmm$_",(8..15));
146 @Tx=map("%xmm$_",(5..7));
147 ($iv,$in,$rndkey0)=map("%xmm$_",(2..4)); # for enc
148 @rndkey=("%xmm0","%xmm1"); # for enc
149 ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(0..3)); # for dec
152 sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm
153 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
155 $arg = "\$$arg" if ($arg*1 eq $arg);
156 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
159 my $_rol=sub { &rol(@_) };
160 my $_ror=sub { &ror(@_) };
163 .type aesni_cbc_sha1_enc_ssse3,\@function,6
165 aesni_cbc_sha1_enc_ssse3:
166 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
167 #shr \$6,$len # debugging artefact
168 #jz .Lepilogue_ssse3 # debugging artefact
175 lea `-104-($win64?10*16:0)`(%rsp),%rsp
176 #mov $in0,$inp # debugging artefact
177 #lea 64(%rsp),$ctx # debugging artefact
179 $code.=<<___ if ($win64);
180 movaps %xmm6,96+0(%rsp)
181 movaps %xmm7,96+16(%rsp)
182 movaps %xmm8,96+32(%rsp)
183 movaps %xmm9,96+48(%rsp)
184 movaps %xmm10,96+64(%rsp)
185 movaps %xmm11,96+80(%rsp)
186 movaps %xmm12,96+96(%rsp)
187 movaps %xmm13,96+112(%rsp)
188 movaps %xmm14,96+128(%rsp)
189 movaps %xmm15,96+144(%rsp)
193 mov $in0,%r12 # reassign arguments
197 movdqu ($ivp),$iv # load IV
198 mov $ivp,88(%rsp) # save $ivp
200 ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
201 my $rounds="${ivp}d";
205 mov 240($key),$rounds
206 add $inp,$len # end of input
208 lea K_XX_XX(%rip),$K_XX_XX
209 mov 0($ctx),$A # load context
213 mov $B,@T[0] # magic seed
219 movdqa 64($K_XX_XX),@X[2] # pbswap mask
220 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
221 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
222 movdqu 16($inp),@X[-3&7]
223 movdqu 32($inp),@X[-2&7]
224 movdqu 48($inp),@X[-1&7]
225 pshufb @X[2],@X[-4&7] # byte swap
227 pshufb @X[2],@X[-3&7]
228 pshufb @X[2],@X[-2&7]
229 pshufb @X[2],@X[-1&7]
230 paddd @Tx[1],@X[-4&7] # add K_00_19
231 paddd @Tx[1],@X[-3&7]
232 paddd @Tx[1],@X[-2&7]
233 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
234 psubd @Tx[1],@X[-4&7] # restore X[]
235 movdqa @X[-3&7],16(%rsp)
236 psubd @Tx[1],@X[-3&7]
237 movdqa @X[-2&7],32(%rsp)
238 psubd @Tx[1],@X[-2&7]
239 movups ($key),$rndkey0 # $key[0]
240 movups 16($key),$rndkey[0] # forward reference
246 my ($n,$k)=($r/10,$r%10);
249 movups `16*$n`($in0),$in # load input
252 $code.=<<___ if ($n);
253 movups $iv,`16*($n-1)`($out,$in0) # write output
257 aesenc $rndkey[0],$iv
258 movups `32+16*$k`($key),$rndkey[1]
265 movups `32+16*($k+0)`($key),$rndkey[1]
266 aesenc $rndkey[0],$iv
267 movups `32+16*($k+1)`($key),$rndkey[0]
268 aesenc $rndkey[1],$iv
270 movups `32+16*($k+2)`($key),$rndkey[1]
271 aesenc $rndkey[0],$iv
272 movups `32+16*($k+3)`($key),$rndkey[0]
273 aesenc $rndkey[1],$iv
275 aesenclast $rndkey[0],$iv
276 movups 16($key),$rndkey[1] # forward reference
280 aesenc $rndkey[0],$iv
281 movups `32+16*$k`($key),$rndkey[1]
284 $r++; unshift(@rndkey,pop(@rndkey));
287 sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4
290 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
293 &pshufd (@X[0],@X[-4&7],0xee); # was &movdqa(@X[0],@X[-3&7]);
296 &movdqa (@Tx[0],@X[-1&7]);
297 &punpcklqdq(@X[0],@X[-3&7]); # compose "X[-14]" in "X[0]", was &palignr(@X[0],@X[-4&7],8);
301 &paddd (@Tx[1],@X[-1&7]);
304 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords
307 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
311 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
317 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
320 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
324 &movdqa (@Tx[2],@X[0]);
325 &movdqa (@Tx[0],@X[0]);
331 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword
332 &paddd (@X[0],@X[0]);
341 &movdqa (@Tx[1],@Tx[2]);
346 &por (@X[0],@Tx[0]); # "X[0]"<<<=1
353 &pxor (@X[0],@Tx[2]);
356 &movdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
360 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
362 foreach (@insns) { eval; } # remaining instructions [if any]
364 $Xi++; push(@X,shift(@X)); # "rotate" X[]
365 push(@Tx,shift(@Tx));
368 sub Xupdate_ssse3_32_79()
371 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
374 &pshufd (@Tx[0],@X[-2&7],0xee) if ($Xi==8); # was &movdqa (@Tx[0],@X[-1&7])
375 eval(shift(@insns)); # body_20_39
376 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
377 &punpcklqdq(@Tx[0],@X[-1&7]); # compose "X[-6]", was &palignr(@Tx[0],@X[-2&7],8);
380 eval(shift(@insns)); # rol
382 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
384 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
386 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
387 } else { # ... or load next one
388 &movdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
390 &paddd (@Tx[1],@X[-1&7]);
391 eval(shift(@insns)); # ror
394 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]"
395 eval(shift(@insns)); # body_20_39
398 eval(shift(@insns)); # rol
400 &movdqa (@Tx[0],@X[0]);
401 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
404 eval(shift(@insns)); # ror
408 eval(shift(@insns)); # body_20_39
412 eval(shift(@insns)); # rol
415 eval(shift(@insns)); # ror
418 &por (@X[0],@Tx[0]); # "X[0]"<<<=2
419 eval(shift(@insns)); # body_20_39
421 &pshufd(@Tx[1],@X[-1&7],0xee) if ($Xi<19); # was &movdqa (@Tx[1],@X[0])
423 eval(shift(@insns)); # rol
426 eval(shift(@insns)); # rol
429 foreach (@insns) { eval; } # remaining instructions
431 $Xi++; push(@X,shift(@X)); # "rotate" X[]
432 push(@Tx,shift(@Tx));
435 sub Xuplast_ssse3_80()
438 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
442 &paddd (@Tx[1],@X[-1&7]);
448 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
450 foreach (@insns) { eval; } # remaining instructions
455 unshift(@Tx,pop(@Tx));
457 &movdqa (@Tx[2],"64($K_XX_XX)"); # pbswap mask
458 &movdqa (@Tx[1],"0($K_XX_XX)"); # K_00_19
459 &movdqu (@X[-4&7],"0($inp)"); # load input
460 &movdqu (@X[-3&7],"16($inp)");
461 &movdqu (@X[-2&7],"32($inp)");
462 &movdqu (@X[-1&7],"48($inp)");
463 &pshufb (@X[-4&7],@Tx[2]); # byte swap
472 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
477 &pshufb (@X[($Xi-3)&7],@Tx[2]);
480 &paddd (@X[($Xi-4)&7],@Tx[1]);
485 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU
488 &psubd (@X[($Xi-4)&7],@Tx[1]);
490 foreach (@insns) { eval; }
497 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
500 foreach (@insns) { eval; }
504 '($a,$b,$c,$d,$e)=@V;'.
505 '&$_ror ($b,$j?7:2);', # $b>>>2
507 '&mov (@T[1],$a);', # $b for next round
509 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
510 '&xor ($b,$c);', # $c^$d for next round
514 '&and (@T[1],$b);', # ($b&($c^$d)) for next round
516 '&xor ($b,$c);', # restore $b
517 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
520 sub body_00_19 () { # ((c^d)&b)^d
521 # on start @T[0]=(c^d)&b
522 return &body_20_39() if ($rx==19); $rx++;
529 $k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
530 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n);
537 '($a,$b,$c,$d,$e)=@V;'.
538 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
539 '&xor (@T[0],$d) if($j==19);'.
540 '&xor (@T[0],$c) if($j> 19);', # ($b^$d^$c)
541 '&mov (@T[1],$a);', # $b for next round
545 '&xor (@T[1],$c) if ($j< 79);', # $b^$d for next round
547 '&$_ror ($b,7);', # $b>>>2
548 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
551 sub body_20_39 () { # b^d^c
553 return &body_40_59() if ($rx==39); $rx++;
560 $k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds
561 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=20);
568 '($a,$b,$c,$d,$e)=@V;'.
569 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
570 '&and (@T[0],$c) if ($j>=40);', # (b^c)&(c^d)
571 '&xor ($c,$d) if ($j>=40);', # restore $c
573 '&$_ror ($b,7);', # $b>>>2
574 '&mov (@T[1],$a);', # $b for next round
579 '&xor (@T[1],$c) if ($j==59);'.
580 '&xor (@T[1],$b) if ($j< 59);', # b^c for next round
582 '&xor ($b,$c) if ($j< 59);', # c^d for next round
583 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
586 sub body_40_59 () { # ((b^c)&(c^d))^c
587 # on entry @T[0]=(b^c), (c^=d)
595 $k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
596 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=40);
605 &Xupdate_ssse3_16_31(\&body_00_19);
606 &Xupdate_ssse3_16_31(\&body_00_19);
607 &Xupdate_ssse3_16_31(\&body_00_19);
608 &Xupdate_ssse3_16_31(\&body_00_19);
609 &Xupdate_ssse3_32_79(\&body_00_19);
610 &Xupdate_ssse3_32_79(\&body_20_39);
611 &Xupdate_ssse3_32_79(\&body_20_39);
612 &Xupdate_ssse3_32_79(\&body_20_39);
613 &Xupdate_ssse3_32_79(\&body_20_39);
614 &Xupdate_ssse3_32_79(\&body_20_39);
615 &Xupdate_ssse3_32_79(\&body_40_59);
616 &Xupdate_ssse3_32_79(\&body_40_59);
617 &Xupdate_ssse3_32_79(\&body_40_59);
618 &Xupdate_ssse3_32_79(\&body_40_59);
619 &Xupdate_ssse3_32_79(\&body_40_59);
620 &Xupdate_ssse3_32_79(\&body_20_39);
621 &Xuplast_ssse3_80(\&body_20_39,".Ldone_ssse3"); # can jump to "done"
623 $saved_j=$j; @saved_V=@V;
624 $saved_r=$r; @saved_rndkey=@rndkey;
626 &Xloop_ssse3(\&body_20_39);
627 &Xloop_ssse3(\&body_20_39);
628 &Xloop_ssse3(\&body_20_39);
631 movups $iv,48($out,$in0) # write output
634 add 0($ctx),$A # update context
641 mov @T[0],$B # magic seed
652 $jj=$j=$saved_j; @V=@saved_V;
653 $r=$saved_r; @rndkey=@saved_rndkey;
655 &Xtail_ssse3(\&body_20_39);
656 &Xtail_ssse3(\&body_20_39);
657 &Xtail_ssse3(\&body_20_39);
660 movups $iv,48($out,$in0) # write output
661 mov 88(%rsp),$ivp # restore $ivp
663 add 0($ctx),$A # update context
673 movups $iv,($ivp) # write IV
675 $code.=<<___ if ($win64);
676 movaps 96+0(%rsp),%xmm6
677 movaps 96+16(%rsp),%xmm7
678 movaps 96+32(%rsp),%xmm8
679 movaps 96+48(%rsp),%xmm9
680 movaps 96+64(%rsp),%xmm10
681 movaps 96+80(%rsp),%xmm11
682 movaps 96+96(%rsp),%xmm12
683 movaps 96+112(%rsp),%xmm13
684 movaps 96+128(%rsp),%xmm14
685 movaps 96+144(%rsp),%xmm15
688 lea `104+($win64?10*16:0)`(%rsp),%rsi
698 .size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
701 if ($stitched_decrypt) {{{
703 ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
708 '&movdqu($inout0,"0x00($in0)");',
709 '&movdqu($inout1,"0x10($in0)"); &pxor ($inout0,$rndkey0);',
710 '&movdqu($inout2,"0x20($in0)"); &pxor ($inout1,$rndkey0);',
711 '&movdqu($inout3,"0x30($in0)"); &pxor ($inout2,$rndkey0);',
713 '&pxor ($inout3,$rndkey0); &movups ($rndkey0,"16-112($key)");',
714 '&movaps("64(%rsp)",@X[2]);', # save IV, originally @X[3]
717 for ($i=0;$i<13;$i++) {
719 '&aesdec ($inout0,$rndkey0);',
720 '&aesdec ($inout1,$rndkey0);',
721 '&aesdec ($inout2,$rndkey0);',
722 '&aesdec ($inout3,$rndkey0); &movups($rndkey0,"'.(16*($i+2)-112).'($key)");'
724 push (@aes256_dec,(undef,undef)) if (($i>=3 && $i<=5) || $i>=11);
725 push (@aes256_dec,(undef,undef)) if ($i==5);
728 '&aesdeclast ($inout0,$rndkey0); &movups (@X[0],"0x00($in0)");',
729 '&aesdeclast ($inout1,$rndkey0); &movups (@X[1],"0x10($in0)");',
730 '&aesdeclast ($inout2,$rndkey0); &movups (@X[2],"0x20($in0)");',
731 '&aesdeclast ($inout3,$rndkey0); &movups (@X[3],"0x30($in0)");',
733 '&xorps ($inout0,"64(%rsp)"); &movdqu ($rndkey0,"-112($key)");',
734 '&xorps ($inout1,@X[0]); &movups ("0x00($out,$in0)",$inout0);',
735 '&xorps ($inout2,@X[1]); &movups ("0x10($out,$in0)",$inout1);',
736 '&xorps ($inout3,@X[2]); &movups ("0x20($out,$in0)",$inout2);',
738 '&movups ("0x30($out,$in0)",$inout3);'
741 sub body_00_19_dec () { # ((c^d)&b)^d
742 # on start @T[0]=(c^d)&b
743 return &body_20_39_dec() if ($rx==19);
747 unshift (@r,@aes256_dec[$rx]) if (@aes256_dec[$rx]);
753 sub body_20_39_dec () { # b^d^c
755 return &body_40_59_dec() if ($rx==39);
759 unshift (@r,@aes256_dec[$rx]) if (@aes256_dec[$rx]);
765 sub body_40_59_dec () { # ((b^c)&(c^d))^c
766 # on entry @T[0]=(b^c), (c^=d)
770 unshift (@r,@aes256_dec[$rx]) if (@aes256_dec[$rx]);
777 .globl aesni256_cbc_sha1_dec
778 .type aesni256_cbc_sha1_dec,\@abi-omnipotent
780 aesni256_cbc_sha1_dec:
781 # caller should check for SSSE3 and AES-NI bits
782 mov OPENSSL_ia32cap_P+0(%rip),%r10d
783 mov OPENSSL_ia32cap_P+4(%rip),%r11d
785 $code.=<<___ if ($avx);
786 and \$`1<<28`,%r11d # mask AVX bit
787 and \$`1<<30`,%r10d # mask "Intel CPU" bit
789 cmp \$`1<<28|1<<30`,%r10d
790 je aesni256_cbc_sha1_dec_avx
793 jmp aesni256_cbc_sha1_dec_ssse3
795 .size aesni256_cbc_sha1_dec,.-aesni256_cbc_sha1_dec
797 .type aesni256_cbc_sha1_dec_ssse3,\@function,6
799 aesni256_cbc_sha1_dec_ssse3:
800 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
807 lea `-104-($win64?10*16:0)`(%rsp),%rsp
809 $code.=<<___ if ($win64);
810 movaps %xmm6,96+0(%rsp)
811 movaps %xmm7,96+16(%rsp)
812 movaps %xmm8,96+32(%rsp)
813 movaps %xmm9,96+48(%rsp)
814 movaps %xmm10,96+64(%rsp)
815 movaps %xmm11,96+80(%rsp)
816 movaps %xmm12,96+96(%rsp)
817 movaps %xmm13,96+112(%rsp)
818 movaps %xmm14,96+128(%rsp)
819 movaps %xmm15,96+144(%rsp)
820 .Lprologue_dec_ssse3:
823 mov $in0,%r12 # reassign arguments
826 lea 112($key),%r15 # size optimization
827 movdqu ($ivp),@X[3] # load IV
828 #mov $ivp,88(%rsp) # save $ivp
830 ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
834 add $inp,$len # end of input
836 lea K_XX_XX(%rip),$K_XX_XX
837 mov 0($ctx),$A # load context
841 mov $B,@T[0] # magic seed
847 movdqa 64($K_XX_XX),@X[2] # pbswap mask
848 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
849 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
850 movdqu 16($inp),@X[-3&7]
851 movdqu 32($inp),@X[-2&7]
852 movdqu 48($inp),@X[-1&7]
853 pshufb @X[2],@X[-4&7] # byte swap
855 pshufb @X[2],@X[-3&7]
856 pshufb @X[2],@X[-2&7]
857 pshufb @X[2],@X[-1&7]
858 paddd @Tx[1],@X[-4&7] # add K_00_19
859 paddd @Tx[1],@X[-3&7]
860 paddd @Tx[1],@X[-2&7]
861 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
862 psubd @Tx[1],@X[-4&7] # restore X[]
863 movdqa @X[-3&7],16(%rsp)
864 psubd @Tx[1],@X[-3&7]
865 movdqa @X[-2&7],32(%rsp)
866 psubd @Tx[1],@X[-2&7]
867 movdqu -112($key),$rndkey0 # $key[0]
873 &Xupdate_ssse3_16_31(\&body_00_19_dec);
874 &Xupdate_ssse3_16_31(\&body_00_19_dec);
875 &Xupdate_ssse3_16_31(\&body_00_19_dec);
876 &Xupdate_ssse3_16_31(\&body_00_19_dec);
877 &Xupdate_ssse3_32_79(\&body_00_19_dec);
878 &Xupdate_ssse3_32_79(\&body_20_39_dec);
879 &Xupdate_ssse3_32_79(\&body_20_39_dec);
880 &Xupdate_ssse3_32_79(\&body_20_39_dec);
881 &Xupdate_ssse3_32_79(\&body_20_39_dec);
882 &Xupdate_ssse3_32_79(\&body_20_39_dec);
883 &Xupdate_ssse3_32_79(\&body_40_59_dec);
884 &Xupdate_ssse3_32_79(\&body_40_59_dec);
885 &Xupdate_ssse3_32_79(\&body_40_59_dec);
886 &Xupdate_ssse3_32_79(\&body_40_59_dec);
887 &Xupdate_ssse3_32_79(\&body_40_59_dec);
888 &Xupdate_ssse3_32_79(\&body_20_39_dec);
889 &Xuplast_ssse3_80(\&body_20_39_dec,".Ldone_dec_ssse3"); # can jump to "done"
891 $saved_j=$j; @saved_V=@V;
894 &Xloop_ssse3(\&body_20_39_dec);
895 &Xloop_ssse3(\&body_20_39_dec);
896 &Xloop_ssse3(\&body_20_39_dec);
898 eval(@aes256_dec[-1]); # last store
902 add 0($ctx),$A # update context
909 mov @T[0],$B # magic seed
920 $jj=$j=$saved_j; @V=@saved_V;
923 &Xtail_ssse3(\&body_20_39_dec);
924 &Xtail_ssse3(\&body_20_39_dec);
925 &Xtail_ssse3(\&body_20_39_dec);
927 eval(@aes256_dec[-1]); # last store
929 add 0($ctx),$A # update context
939 movups @X[3],($ivp) # write IV
941 $code.=<<___ if ($win64);
942 movaps 96+0(%rsp),%xmm6
943 movaps 96+16(%rsp),%xmm7
944 movaps 96+32(%rsp),%xmm8
945 movaps 96+48(%rsp),%xmm9
946 movaps 96+64(%rsp),%xmm10
947 movaps 96+80(%rsp),%xmm11
948 movaps 96+96(%rsp),%xmm12
949 movaps 96+112(%rsp),%xmm13
950 movaps 96+128(%rsp),%xmm14
951 movaps 96+144(%rsp),%xmm15
954 lea `104+($win64?10*16:0)`(%rsp),%rsi
962 .Lepilogue_dec_ssse3:
964 .size aesni256_cbc_sha1_dec_ssse3,.-aesni256_cbc_sha1_dec_ssse3
970 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
973 my @X=map("%xmm$_",(4..7,0..3));
974 my @Tx=map("%xmm$_",(8..10));
975 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
976 my @T=("%esi","%edi");
977 my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13));
978 my @rndkey=("%xmm14","%xmm15");
979 my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15)); # for dec
982 my $_rol=sub { &shld(@_[0],@_) };
983 my $_ror=sub { &shrd(@_[0],@_) };
986 .type aesni_cbc_sha1_enc_avx,\@function,6
988 aesni_cbc_sha1_enc_avx:
989 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
990 #shr \$6,$len # debugging artefact
991 #jz .Lepilogue_avx # debugging artefact
998 lea `-104-($win64?10*16:0)`(%rsp),%rsp
999 #mov $in0,$inp # debugging artefact
1000 #lea 64(%rsp),$ctx # debugging artefact
1002 $code.=<<___ if ($win64);
1003 movaps %xmm6,96+0(%rsp)
1004 movaps %xmm7,96+16(%rsp)
1005 movaps %xmm8,96+32(%rsp)
1006 movaps %xmm9,96+48(%rsp)
1007 movaps %xmm10,96+64(%rsp)
1008 movaps %xmm11,96+80(%rsp)
1009 movaps %xmm12,96+96(%rsp)
1010 movaps %xmm13,96+112(%rsp)
1011 movaps %xmm14,96+128(%rsp)
1012 movaps %xmm15,96+144(%rsp)
1017 mov $in0,%r12 # reassign arguments
1021 vmovdqu ($ivp),$iv # load IV
1022 mov $ivp,88(%rsp) # save $ivp
1024 ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
1025 my $rounds="${ivp}d";
1029 mov 240($key),$rounds
1030 add \$112,$key # size optimization
1031 add $inp,$len # end of input
1033 lea K_XX_XX(%rip),$K_XX_XX
1034 mov 0($ctx),$A # load context
1038 mov $B,@T[0] # magic seed
1044 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
1045 vmovdqa 0($K_XX_XX),$Kx # K_00_19
1046 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
1047 vmovdqu 16($inp),@X[-3&7]
1048 vmovdqu 32($inp),@X[-2&7]
1049 vmovdqu 48($inp),@X[-1&7]
1050 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
1052 vpshufb @X[2],@X[-3&7],@X[-3&7]
1053 vpshufb @X[2],@X[-2&7],@X[-2&7]
1054 vpshufb @X[2],@X[-1&7],@X[-1&7]
1055 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
1056 vpaddd $Kx,@X[-3&7],@X[1]
1057 vpaddd $Kx,@X[-2&7],@X[2]
1058 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
1059 vmovdqa @X[1],16(%rsp)
1060 vmovdqa @X[2],32(%rsp)
1061 vmovups -112($key),$rndkey[1] # $key[0]
1062 vmovups 16-112($key),$rndkey[0] # forward reference
1068 my ($n,$k)=($r/10,$r%10);
1071 vmovdqu `16*$n`($in0),$in # load input
1072 vpxor $rndkey[1],$in,$in
1074 $code.=<<___ if ($n);
1075 vmovups $iv,`16*($n-1)`($out,$in0) # write output
1079 vaesenc $rndkey[0],$iv,$iv
1080 vmovups `32+16*$k-112`($key),$rndkey[1]
1087 vaesenc $rndkey[0],$iv,$iv
1088 vmovups `32+16*($k+0)-112`($key),$rndkey[1]
1089 vaesenc $rndkey[1],$iv,$iv
1090 vmovups `32+16*($k+1)-112`($key),$rndkey[0]
1092 vaesenc $rndkey[0],$iv,$iv
1093 vmovups `32+16*($k+2)-112`($key),$rndkey[1]
1094 vaesenc $rndkey[1],$iv,$iv
1095 vmovups `32+16*($k+3)-112`($key),$rndkey[0]
1097 vaesenclast $rndkey[0],$iv,$iv
1098 vmovups -112($key),$rndkey[0]
1099 vmovups 16-112($key),$rndkey[1] # forward reference
1103 vaesenc $rndkey[0],$iv,$iv
1104 vmovups `32+16*$k-112`($key),$rndkey[1]
1107 $r++; unshift(@rndkey,pop(@rndkey));
1110 sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
1113 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
1114 my ($a,$b,$c,$d,$e);
1116 eval(shift(@insns));
1117 eval(shift(@insns));
1118 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
1119 eval(shift(@insns));
1120 eval(shift(@insns));
1122 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
1123 eval(shift(@insns));
1124 eval(shift(@insns));
1125 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
1126 eval(shift(@insns));
1127 eval(shift(@insns));
1128 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
1129 eval(shift(@insns));
1130 eval(shift(@insns));
1132 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
1133 eval(shift(@insns));
1134 eval(shift(@insns));
1135 eval(shift(@insns));
1136 eval(shift(@insns));
1138 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
1139 eval(shift(@insns));
1140 eval(shift(@insns));
1141 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
1142 eval(shift(@insns));
1143 eval(shift(@insns));
1145 &vpsrld (@Tx[0],@X[0],31);
1146 eval(shift(@insns));
1147 eval(shift(@insns));
1148 eval(shift(@insns));
1149 eval(shift(@insns));
1151 &vpslldq(@Tx[1],@X[0],12); # "X[0]"<<96, extract one dword
1152 &vpaddd (@X[0],@X[0],@X[0]);
1153 eval(shift(@insns));
1154 eval(shift(@insns));
1155 eval(shift(@insns));
1156 eval(shift(@insns));
1158 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
1159 &vpsrld (@Tx[0],@Tx[1],30);
1160 eval(shift(@insns));
1161 eval(shift(@insns));
1162 eval(shift(@insns));
1163 eval(shift(@insns));
1165 &vpslld (@Tx[1],@Tx[1],2);
1166 &vpxor (@X[0],@X[0],@Tx[0]);
1167 eval(shift(@insns));
1168 eval(shift(@insns));
1169 eval(shift(@insns));
1170 eval(shift(@insns));
1172 &vpxor (@X[0],@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
1173 eval(shift(@insns));
1174 eval(shift(@insns));
1175 &vmovdqa ($Kx,eval(16*(($Xi)/5))."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
1176 eval(shift(@insns));
1177 eval(shift(@insns));
1180 foreach (@insns) { eval; } # remaining instructions [if any]
1182 $Xi++; push(@X,shift(@X)); # "rotate" X[]
1185 sub Xupdate_avx_32_79()
1188 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
1189 my ($a,$b,$c,$d,$e);
1191 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
1192 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
1193 eval(shift(@insns)); # body_20_39
1194 eval(shift(@insns));
1195 eval(shift(@insns));
1196 eval(shift(@insns)); # rol
1198 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
1199 eval(shift(@insns));
1200 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
1201 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
1202 &vmovdqa ($Kx,eval(16*($Xi/5))."($K_XX_XX)") if ($Xi%5==0);
1203 eval(shift(@insns)); # ror
1204 eval(shift(@insns));
1206 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
1207 eval(shift(@insns)); # body_20_39
1208 eval(shift(@insns));
1209 eval(shift(@insns));
1210 eval(shift(@insns)); # rol
1212 &vpsrld (@Tx[0],@X[0],30);
1213 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
1214 eval(shift(@insns));
1215 eval(shift(@insns));
1216 eval(shift(@insns)); # ror
1217 eval(shift(@insns));
1219 &vpslld (@X[0],@X[0],2);
1220 eval(shift(@insns)); # body_20_39
1221 eval(shift(@insns));
1222 eval(shift(@insns));
1223 eval(shift(@insns)); # rol
1224 eval(shift(@insns));
1225 eval(shift(@insns));
1226 eval(shift(@insns)); # ror
1227 eval(shift(@insns));
1229 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
1230 eval(shift(@insns)); # body_20_39
1231 eval(shift(@insns));
1232 eval(shift(@insns));
1233 eval(shift(@insns)); # rol
1234 eval(shift(@insns));
1235 eval(shift(@insns));
1236 eval(shift(@insns)); # rol
1237 eval(shift(@insns));
1239 foreach (@insns) { eval; } # remaining instructions
1241 $Xi++; push(@X,shift(@X)); # "rotate" X[]
1244 sub Xuplast_avx_80()
1247 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1248 my ($a,$b,$c,$d,$e);
1250 eval(shift(@insns));
1251 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
1252 eval(shift(@insns));
1253 eval(shift(@insns));
1254 eval(shift(@insns));
1255 eval(shift(@insns));
1257 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
1259 foreach (@insns) { eval; } # remaining instructions
1264 &vmovdqa(@Tx[1],"64($K_XX_XX)"); # pbswap mask
1265 &vmovdqa($Kx,"0($K_XX_XX)"); # K_00_19
1266 &vmovdqu(@X[-4&7],"0($inp)"); # load input
1267 &vmovdqu(@X[-3&7],"16($inp)");
1268 &vmovdqu(@X[-2&7],"32($inp)");
1269 &vmovdqu(@X[-1&7],"48($inp)");
1270 &vpshufb(@X[-4&7],@X[-4&7],@Tx[1]); # byte swap
1279 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1280 my ($a,$b,$c,$d,$e);
1282 eval(shift(@insns));
1283 eval(shift(@insns));
1284 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@Tx[1]);
1285 eval(shift(@insns));
1286 eval(shift(@insns));
1287 &vpaddd (@Tx[0],@X[($Xi-4)&7],$Kx);
1288 eval(shift(@insns));
1289 eval(shift(@insns));
1290 eval(shift(@insns));
1291 eval(shift(@insns));
1292 &vmovdqa(eval(16*$Xi)."(%rsp)",@Tx[0]); # X[]+K xfer to IALU
1293 eval(shift(@insns));
1294 eval(shift(@insns));
1296 foreach (@insns) { eval; }
1303 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1304 my ($a,$b,$c,$d,$e);
1306 foreach (@insns) { eval; }
1313 &Xupdate_avx_16_31(\&body_00_19);
1314 &Xupdate_avx_16_31(\&body_00_19);
1315 &Xupdate_avx_16_31(\&body_00_19);
1316 &Xupdate_avx_16_31(\&body_00_19);
1317 &Xupdate_avx_32_79(\&body_00_19);
1318 &Xupdate_avx_32_79(\&body_20_39);
1319 &Xupdate_avx_32_79(\&body_20_39);
1320 &Xupdate_avx_32_79(\&body_20_39);
1321 &Xupdate_avx_32_79(\&body_20_39);
1322 &Xupdate_avx_32_79(\&body_20_39);
1323 &Xupdate_avx_32_79(\&body_40_59);
1324 &Xupdate_avx_32_79(\&body_40_59);
1325 &Xupdate_avx_32_79(\&body_40_59);
1326 &Xupdate_avx_32_79(\&body_40_59);
1327 &Xupdate_avx_32_79(\&body_40_59);
1328 &Xupdate_avx_32_79(\&body_20_39);
1329 &Xuplast_avx_80(\&body_20_39,".Ldone_avx"); # can jump to "done"
1331 $saved_j=$j; @saved_V=@V;
1332 $saved_r=$r; @saved_rndkey=@rndkey;
1334 &Xloop_avx(\&body_20_39);
1335 &Xloop_avx(\&body_20_39);
1336 &Xloop_avx(\&body_20_39);
1339 vmovups $iv,48($out,$in0) # write output
1342 add 0($ctx),$A # update context
1349 mov @T[0],$B # magic seed
1360 $jj=$j=$saved_j; @V=@saved_V;
1361 $r=$saved_r; @rndkey=@saved_rndkey;
1363 &Xtail_avx(\&body_20_39);
1364 &Xtail_avx(\&body_20_39);
1365 &Xtail_avx(\&body_20_39);
1368 vmovups $iv,48($out,$in0) # write output
1369 mov 88(%rsp),$ivp # restore $ivp
1371 add 0($ctx),$A # update context
1381 vmovups $iv,($ivp) # write IV
1384 $code.=<<___ if ($win64);
1385 movaps 96+0(%rsp),%xmm6
1386 movaps 96+16(%rsp),%xmm7
1387 movaps 96+32(%rsp),%xmm8
1388 movaps 96+48(%rsp),%xmm9
1389 movaps 96+64(%rsp),%xmm10
1390 movaps 96+80(%rsp),%xmm11
1391 movaps 96+96(%rsp),%xmm12
1392 movaps 96+112(%rsp),%xmm13
1393 movaps 96+128(%rsp),%xmm14
1394 movaps 96+144(%rsp),%xmm15
1397 lea `104+($win64?10*16:0)`(%rsp),%rsi
1407 .size aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
1410 if ($stiched_decrypt) {{{
1412 ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
1414 $j=$jj=$r=$sn=$rx=0;
1418 '&vpxor ($inout0,$rndkey0,"0x00($in0)");',
1419 '&vpxor ($inout1,$rndkey0,"0x10($in0)");',
1420 '&vpxor ($inout2,$rndkey0,"0x20($in0)");',
1421 '&vpxor ($inout3,$rndkey0,"0x30($in0)");',
1423 '&vmovups($rndkey0,"16-112($key)");',
1424 '&vmovups("64(%rsp)",@X[2]);', # save IV, originally @X[3]
1427 for ($i=0;$i<13;$i++) {
1429 '&vaesdec ($inout0,$inout0,$rndkey0);',
1430 '&vaesdec ($inout1,$inout1,$rndkey0);',
1431 '&vaesdec ($inout2,$inout2,$rndkey0);',
1432 '&vaesdec ($inout3,$inout3,$rndkey0); &vmovups($rndkey0,"'.(16*($i+2)-112).'($key)");'
1434 push (@aes256_dec,(undef,undef)) if (($i>=3 && $i<=5) || $i>=11);
1435 push (@aes256_dec,(undef,undef)) if ($i==5);
1438 '&vaesdeclast ($inout0,$inout0,$rndkey0); &vmovups(@X[0],"0x00($in0)");',
1439 '&vaesdeclast ($inout1,$inout1,$rndkey0); &vmovups(@X[1],"0x10($in0)");',
1440 '&vaesdeclast ($inout2,$inout2,$rndkey0); &vmovups(@X[2],"0x20($in0)");',
1441 '&vaesdeclast ($inout3,$inout3,$rndkey0); &vmovups(@X[3],"0x30($in0)");',
1443 '&vxorps ($inout0,$inout0,"64(%rsp)"); &vmovdqu($rndkey0,"-112($key)");',
1444 '&vxorps ($inout1,$inout1,@X[0]); &vmovups("0x00($out,$in0)",$inout0);',
1445 '&vxorps ($inout2,$inout2,@X[1]); &vmovups("0x10($out,$in0)",$inout1);',
1446 '&vxorps ($inout3,$inout3,@X[2]); &vmovups("0x20($out,$in0)",$inout2);',
1448 '&vmovups ("0x30($out,$in0)",$inout3);'
1452 .type aesni256_cbc_sha1_dec_avx,\@function,6
1454 aesni256_cbc_sha1_dec_avx:
1455 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
1462 lea `-104-($win64?10*16:0)`(%rsp),%rsp
1464 $code.=<<___ if ($win64);
1465 movaps %xmm6,96+0(%rsp)
1466 movaps %xmm7,96+16(%rsp)
1467 movaps %xmm8,96+32(%rsp)
1468 movaps %xmm9,96+48(%rsp)
1469 movaps %xmm10,96+64(%rsp)
1470 movaps %xmm11,96+80(%rsp)
1471 movaps %xmm12,96+96(%rsp)
1472 movaps %xmm13,96+112(%rsp)
1473 movaps %xmm14,96+128(%rsp)
1474 movaps %xmm15,96+144(%rsp)
1479 mov $in0,%r12 # reassign arguments
1482 lea 112($key),%r15 # size optimization
1483 vmovdqu ($ivp),@X[3] # load IV
1485 ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
1489 add $inp,$len # end of input
1491 lea K_XX_XX(%rip),$K_XX_XX
1492 mov 0($ctx),$A # load context
1496 mov $B,@T[0] # magic seed
1502 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
1503 vmovdqa 0($K_XX_XX),$Kx # K_00_19
1504 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
1505 vmovdqu 16($inp),@X[-3&7]
1506 vmovdqu 32($inp),@X[-2&7]
1507 vmovdqu 48($inp),@X[-1&7]
1508 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
1510 vpshufb @X[2],@X[-3&7],@X[-3&7]
1511 vpshufb @X[2],@X[-2&7],@X[-2&7]
1512 vpshufb @X[2],@X[-1&7],@X[-1&7]
1513 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
1514 vpaddd $Kx,@X[-3&7],@X[1]
1515 vpaddd $Kx,@X[-2&7],@X[2]
1516 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
1517 vmovdqa @X[1],16(%rsp)
1518 vmovdqa @X[2],32(%rsp)
1519 vmovups -112($key),$rndkey0 # $key[0]
1525 &Xupdate_avx_16_31(\&body_00_19_dec);
1526 &Xupdate_avx_16_31(\&body_00_19_dec);
1527 &Xupdate_avx_16_31(\&body_00_19_dec);
1528 &Xupdate_avx_16_31(\&body_00_19_dec);
1529 &Xupdate_avx_32_79(\&body_00_19_dec);
1530 &Xupdate_avx_32_79(\&body_20_39_dec);
1531 &Xupdate_avx_32_79(\&body_20_39_dec);
1532 &Xupdate_avx_32_79(\&body_20_39_dec);
1533 &Xupdate_avx_32_79(\&body_20_39_dec);
1534 &Xupdate_avx_32_79(\&body_20_39_dec);
1535 &Xupdate_avx_32_79(\&body_40_59_dec);
1536 &Xupdate_avx_32_79(\&body_40_59_dec);
1537 &Xupdate_avx_32_79(\&body_40_59_dec);
1538 &Xupdate_avx_32_79(\&body_40_59_dec);
1539 &Xupdate_avx_32_79(\&body_40_59_dec);
1540 &Xupdate_avx_32_79(\&body_20_39_dec);
1541 &Xuplast_avx_80(\&body_20_39_dec,".Ldone_dec_avx"); # can jump to "done"
1543 $saved_j=$j; @saved_V=@V;
1546 &Xloop_avx(\&body_20_39_dec);
1547 &Xloop_avx(\&body_20_39_dec);
1548 &Xloop_avx(\&body_20_39_dec);
1550 eval(@aes256_dec[-1]); # last store
1554 add 0($ctx),$A # update context
1561 mov @T[0],$B # magic seed
1572 $jj=$j=$saved_j; @V=@saved_V;
1575 &Xtail_avx(\&body_20_39_dec);
1576 &Xtail_avx(\&body_20_39_dec);
1577 &Xtail_avx(\&body_20_39_dec);
1579 eval(@aes256_dec[-1]); # last store
1582 add 0($ctx),$A # update context
1592 vmovups @X[3],($ivp) # write IV
1595 $code.=<<___ if ($win64);
1596 movaps 96+0(%rsp),%xmm6
1597 movaps 96+16(%rsp),%xmm7
1598 movaps 96+32(%rsp),%xmm8
1599 movaps 96+48(%rsp),%xmm9
1600 movaps 96+64(%rsp),%xmm10
1601 movaps 96+80(%rsp),%xmm11
1602 movaps 96+96(%rsp),%xmm12
1603 movaps 96+112(%rsp),%xmm13
1604 movaps 96+128(%rsp),%xmm14
1605 movaps 96+144(%rsp),%xmm15
1608 lea `104+($win64?10*16:0)`(%rsp),%rsi
1618 .size aesni256_cbc_sha1_dec_avx,.-aesni256_cbc_sha1_dec_avx
1625 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1626 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1627 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1628 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1629 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1631 .asciz "AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1635 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1636 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1644 .extern __imp_RtlVirtualUnwind
1645 .type ssse3_handler,\@abi-omnipotent
1659 mov 120($context),%rax # pull context->Rax
1660 mov 248($context),%rbx # pull context->Rip
1662 mov 8($disp),%rsi # disp->ImageBase
1663 mov 56($disp),%r11 # disp->HandlerData
1665 mov 0(%r11),%r10d # HandlerData[0]
1666 lea (%rsi,%r10),%r10 # prologue label
1667 cmp %r10,%rbx # context->Rip<prologue label
1668 jb .Lcommon_seh_tail
1670 mov 152($context),%rax # pull context->Rsp
1672 mov 4(%r11),%r10d # HandlerData[1]
1673 lea (%rsi,%r10),%r10 # epilogue label
1674 cmp %r10,%rbx # context->Rip>=epilogue label
1675 jae .Lcommon_seh_tail
1678 lea 512($context),%rdi # &context.Xmm6
1680 .long 0xa548f3fc # cld; rep movsq
1681 lea `104+10*16`(%rax),%rax # adjust stack pointer
1690 mov %rbx,144($context) # restore context->Rbx
1691 mov %rbp,160($context) # restore context->Rbp
1692 mov %r12,216($context) # restore context->R12
1693 mov %r13,224($context) # restore context->R13
1694 mov %r14,232($context) # restore context->R14
1695 mov %r15,240($context) # restore context->R15
1700 mov %rax,152($context) # restore context->Rsp
1701 mov %rsi,168($context) # restore context->Rsi
1702 mov %rdi,176($context) # restore context->Rdi
1704 mov 40($disp),%rdi # disp->ContextRecord
1705 mov $context,%rsi # context
1706 mov \$154,%ecx # sizeof(CONTEXT)
1707 .long 0xa548f3fc # cld; rep movsq
1710 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1711 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1712 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1713 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1714 mov 40(%rsi),%r10 # disp->ContextRecord
1715 lea 56(%rsi),%r11 # &disp->HandlerData
1716 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1717 mov %r10,32(%rsp) # arg5
1718 mov %r11,40(%rsp) # arg6
1719 mov %r12,48(%rsp) # arg7
1720 mov %rcx,56(%rsp) # arg8, (NULL)
1721 call *__imp_RtlVirtualUnwind(%rip)
1723 mov \$1,%eax # ExceptionContinueSearch
1735 .size ssse3_handler,.-ssse3_handler
1739 .rva .LSEH_begin_aesni_cbc_sha1_enc_ssse3
1740 .rva .LSEH_end_aesni_cbc_sha1_enc_ssse3
1741 .rva .LSEH_info_aesni_cbc_sha1_enc_ssse3
1743 $code.=<<___ if ($avx);
1744 .rva .LSEH_begin_aesni_cbc_sha1_enc_avx
1745 .rva .LSEH_end_aesni_cbc_sha1_enc_avx
1746 .rva .LSEH_info_aesni_cbc_sha1_enc_avx
1751 .LSEH_info_aesni_cbc_sha1_enc_ssse3:
1754 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[]
1756 $code.=<<___ if ($avx);
1757 .LSEH_info_aesni_cbc_sha1_enc_avx:
1760 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
1764 ####################################################################
1766 local *opcode=shift;
1770 $rex|=0x04 if($dst>=8);
1771 $rex|=0x01 if($src>=8);
1772 push @opcode,$rex|0x40 if($rex);
1779 if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1781 "aesenc" => 0xdc, "aesenclast" => 0xdd,
1782 "aesdec" => 0xde, "aesdeclast" => 0xdf
1784 return undef if (!defined($opcodelet{$1}));
1785 rex(\@opcode,$3,$2);
1786 push @opcode,0x0f,0x38,$opcodelet{$1};
1787 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1788 return ".byte\t".join(',',@opcode);
1793 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1794 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;