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 6.58 +41%
25 # Sandy Bridge 5.05[+5.0(6.2)] 10.06(11.21) 6.09(7.05) +65%(+59%)
26 # Ivy Bridge 5.05[+4.6] 9.65 5.54 +74%
27 # Bulldozer 5.77[+6.0] 11.72 6.37 +84%
30 # Westmere 4.51 10.00 6.87 +46%
31 # Sandy Bridge 6.05 11.06(12.21) 6.11(7.20) +81%(+70%)
32 # Ivy Bridge 6.05 10.65 6.07 +75%
33 # Bulldozer 6.89 12.84 6.96 +84%
36 # Westmere 5.25 10.74 7.19 +49%
37 # Sandy Bridge 7.05 12.06(13.21) 7.12(7.68) +69%(+72%)
38 # Ivy Bridge 7.05 11.65 7.12 +64%
39 # Bulldozer 8.00 13.95 8.25 +69%
41 # (*) There are two code paths: SSSE3 and AVX. See sha1-568.pl for
42 # background information. Above numbers in parentheses are SSSE3
43 # results collected on AVX-capable CPU, i.e. apply on OSes that
46 # Needless to mention that it makes no sense to implement "stitched"
47 # *decrypt* subroutine. Because *both* AESNI-CBC decrypt and SHA1
48 # fully utilize parallelism, so stitching would not give any gain
49 # anyway. Well, there might be some, e.g. because of better cache
50 # locality... For reference, here are performance results for
51 # standalone AESNI-CBC decrypt:
53 # AES-128-CBC AES-192-CBC AES-256-CBC
54 # Westmere 1.31 1.55 1.80
55 # Sandy Bridge 0.74 0.91 1.09
56 # Ivy Bridge 0.74 0.90 1.11
57 # Bulldozer 0.70 0.85 0.99
61 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
63 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
65 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
66 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
67 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
68 die "can't locate x86_64-xlate.pl";
70 $avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
71 =~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
73 $avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
74 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
76 $avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
77 `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
80 open OUT,"| \"$^X\" $xlate $flavour $output";
83 # void aesni_cbc_sha1_enc(const void *inp,
93 .extern OPENSSL_ia32cap_P
95 .globl aesni_cbc_sha1_enc
96 .type aesni_cbc_sha1_enc,\@abi-omnipotent
99 # caller should check for SSSE3 and AES-NI bits
100 mov OPENSSL_ia32cap_P+0(%rip),%r10d
101 mov OPENSSL_ia32cap_P+4(%rip),%r11d
103 $code.=<<___ if ($avx);
104 and \$`1<<28`,%r11d # mask AVX bit
105 and \$`1<<30`,%r10d # mask "Intel CPU" bit
107 cmp \$`1<<28|1<<30`,%r10d
108 je aesni_cbc_sha1_enc_avx
111 jmp aesni_cbc_sha1_enc_ssse3
113 .size aesni_cbc_sha1_enc,.-aesni_cbc_sha1_enc
116 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
119 my @X=map("%xmm$_",(4..7,0..3));
120 my @Tx=map("%xmm$_",(8..10));
121 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
122 my @T=("%esi","%edi");
123 my $j=0; my $jj=0; my $r=0; my $sn=0; my $rx=0;
125 my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13));
126 my @rndkey=("%xmm14","%xmm15");
129 @X=map("%xmm$_",(4..11));
130 @Tx=map("%xmm$_",(12..14));
131 ($iv,$in,$rndkey0)=map("%xmm$_",(2,3,15));
132 @rndkey=("%xmm0","%xmm1");
135 sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm
136 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
138 $arg = "\$$arg" if ($arg*1 eq $arg);
139 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
142 my $_rol=sub { &rol(@_) };
143 my $_ror=sub { &ror(@_) };
146 .type aesni_cbc_sha1_enc_ssse3,\@function,6
148 aesni_cbc_sha1_enc_ssse3:
149 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
150 #shr \$6,$len # debugging artefact
151 #jz .Lepilogue_ssse3 # debugging artefact
158 lea `-104-($win64?10*16:0)`(%rsp),%rsp
159 #mov $in0,$inp # debugging artefact
160 #lea 64(%rsp),$ctx # debugging artefact
162 $code.=<<___ if ($win64);
163 movaps %xmm6,96+0(%rsp)
164 movaps %xmm7,96+16(%rsp)
165 movaps %xmm8,96+32(%rsp)
166 movaps %xmm9,96+48(%rsp)
167 movaps %xmm10,96+64(%rsp)
168 movaps %xmm11,96+80(%rsp)
169 movaps %xmm12,96+96(%rsp)
170 movaps %xmm13,96+112(%rsp)
171 movaps %xmm14,96+128(%rsp)
172 movaps %xmm15,96+144(%rsp)
176 mov $in0,%r12 # reassign arguments
180 movdqu ($ivp),$iv # load IV
181 mov $ivp,88(%rsp) # save $ivp
183 my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
184 my $rounds="${ivp}d";
188 mov 240($key),$rounds
189 add $inp,$len # end of input
191 lea K_XX_XX(%rip),$K_XX_XX
192 mov 0($ctx),$A # load context
196 mov $B,@T[0] # magic seed
202 movdqa 64($K_XX_XX),@X[2] # pbswap mask
203 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
204 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
205 movdqu 16($inp),@X[-3&7]
206 movdqu 32($inp),@X[-2&7]
207 movdqu 48($inp),@X[-1&7]
208 pshufb @X[2],@X[-4&7] # byte swap
210 pshufb @X[2],@X[-3&7]
211 pshufb @X[2],@X[-2&7]
212 pshufb @X[2],@X[-1&7]
213 paddd @Tx[1],@X[-4&7] # add K_00_19
214 paddd @Tx[1],@X[-3&7]
215 paddd @Tx[1],@X[-2&7]
216 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
217 psubd @Tx[1],@X[-4&7] # restore X[]
218 movdqa @X[-3&7],16(%rsp)
219 psubd @Tx[1],@X[-3&7]
220 movdqa @X[-2&7],32(%rsp)
221 psubd @Tx[1],@X[-2&7]
222 movups ($key),$rndkey0 # $key[0]
223 movups 16($key),$rndkey[0] # forward reference
229 my ($n,$k)=($r/10,$r%10);
232 movups `16*$n`($in0),$in # load input
235 $code.=<<___ if ($n);
236 movups $iv,`16*($n-1)`($out,$in0) # write output
240 aesenc $rndkey[0],$iv
241 movups `32+16*$k`($key),$rndkey[1]
248 movups `32+16*($k+0)`($key),$rndkey[1]
249 aesenc $rndkey[0],$iv
250 movups `32+16*($k+1)`($key),$rndkey[0]
251 aesenc $rndkey[1],$iv
253 movups `32+16*($k+2)`($key),$rndkey[1]
254 aesenc $rndkey[0],$iv
255 movups `32+16*($k+3)`($key),$rndkey[0]
256 aesenc $rndkey[1],$iv
258 aesenclast $rndkey[0],$iv
259 movups 16($key),$rndkey[1] # forward reference
263 aesenc $rndkey[0],$iv
264 movups `32+16*$k`($key),$rndkey[1]
267 $r++; unshift(@rndkey,pop(@rndkey));
270 sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4
273 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
276 &movdqa (@X[0],@X[-3&7]);
279 &movdqa (@Tx[0],@X[-1&7]);
280 &palignr(@X[0],@X[-4&7],8); # compose "X[-14]" in "X[0]"
284 &paddd (@Tx[1],@X[-1&7]);
287 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords
290 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
294 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
300 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
303 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
307 &movdqa (@Tx[2],@X[0]);
308 &movdqa (@Tx[0],@X[0]);
314 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword
315 &paddd (@X[0],@X[0]);
324 &movdqa (@Tx[1],@Tx[2]);
329 &por (@X[0],@Tx[0]); # "X[0]"<<<=1
336 &pxor (@X[0],@Tx[2]);
339 &movdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
343 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
345 foreach (@insns) { eval; } # remaining instructions [if any]
347 $Xi++; push(@X,shift(@X)); # "rotate" X[]
348 push(@Tx,shift(@Tx));
351 sub Xupdate_ssse3_32_79()
354 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
357 &movdqa (@Tx[0],@X[-1&7]) if ($Xi==8);
358 eval(shift(@insns)); # body_20_39
359 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
360 &palignr(@Tx[0],@X[-2&7],8); # compose "X[-6]"
363 eval(shift(@insns)); # rol
365 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
367 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
369 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
370 } else { # ... or load next one
371 &movdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
373 &paddd (@Tx[1],@X[-1&7]);
374 eval(shift(@insns)); # ror
377 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]"
378 eval(shift(@insns)); # body_20_39
381 eval(shift(@insns)); # rol
383 &movdqa (@Tx[0],@X[0]);
384 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
387 eval(shift(@insns)); # ror
391 eval(shift(@insns)); # body_20_39
395 eval(shift(@insns)); # rol
398 eval(shift(@insns)); # ror
401 &por (@X[0],@Tx[0]); # "X[0]"<<<=2
402 eval(shift(@insns)); # body_20_39
404 &movdqa (@Tx[1],@X[0]) if ($Xi<19);
406 eval(shift(@insns)); # rol
409 eval(shift(@insns)); # rol
412 foreach (@insns) { eval; } # remaining instructions
414 $Xi++; push(@X,shift(@X)); # "rotate" X[]
415 push(@Tx,shift(@Tx));
418 sub Xuplast_ssse3_80()
421 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
425 &paddd (@Tx[1],@X[-1&7]);
431 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
433 foreach (@insns) { eval; } # remaining instructions
436 &je (".Ldone_ssse3");
438 unshift(@Tx,pop(@Tx));
440 &movdqa (@X[2],"64($K_XX_XX)"); # pbswap mask
441 &movdqa (@Tx[1],"0($K_XX_XX)"); # K_00_19
442 &movdqu (@X[-4&7],"0($inp)"); # load input
443 &movdqu (@X[-3&7],"16($inp)");
444 &movdqu (@X[-2&7],"32($inp)");
445 &movdqu (@X[-1&7],"48($inp)");
446 &pshufb (@X[-4&7],@X[2]); # byte swap
455 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
460 &pshufb (@X[($Xi-3)&7],@X[2]);
463 &paddd (@X[($Xi-4)&7],@Tx[1]);
468 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU
471 &psubd (@X[($Xi-4)&7],@Tx[1]);
473 foreach (@insns) { eval; }
480 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
483 foreach (@insns) { eval; }
486 sub body_00_19 () { # ((c^d)&b)^d
487 # on start @T[0]=(c^d)&b
488 return &body_20_39() if ($rx==19); $rx++;
493 '($a,$b,$c,$d,$e)=@V;'.
494 '&$_ror ($b,$j?7:2);', # $b>>>2
496 '&mov (@T[1],$a);', # $b for next round
498 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
499 '&xor ($b,$c);', # $c^$d for next round
503 '&and (@T[1],$b);', # ($b&($c^$d)) for next round
505 '&xor ($b,$c);', # restore $b
506 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
509 $k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
510 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n);
515 sub body_20_39 () { # b^d^c
517 return &body_40_59() if ($rx==39); $rx++;
522 '($a,$b,$c,$d,$e)=@V;'.
523 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
524 '&xor (@T[0],$d) if($j==19);'.
525 '&xor (@T[0],$c) if($j> 19);', # ($b^$d^$c)
526 '&mov (@T[1],$a);', # $b for next round
530 '&xor (@T[1],$c) if ($j< 79);', # $b^$d for next round
532 '&$_ror ($b,7);', # $b>>>2
533 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
536 $k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds
537 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=20);
542 sub body_40_59 () { # ((b^c)&(c^d))^c
543 # on entry @T[0]=(b^c), (c^=d)
549 '($a,$b,$c,$d,$e)=@V;'.
550 '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
551 '&and (@T[0],$c) if ($j>=40);', # (b^c)&(c^d)
552 '&xor ($c,$d) if ($j>=40);', # restore $c
554 '&$_ror ($b,7);', # $b>>>2
555 '&mov (@T[1],$a);', # $b for next round
560 '&xor (@T[1],$c) if ($j==59);'.
561 '&xor (@T[1],$b) if ($j< 59);', # b^c for next round
563 '&xor ($b,$c) if ($j< 59);', # c^d for next round
564 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
567 $k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
568 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=40);
576 &Xupdate_ssse3_16_31(\&body_00_19);
577 &Xupdate_ssse3_16_31(\&body_00_19);
578 &Xupdate_ssse3_16_31(\&body_00_19);
579 &Xupdate_ssse3_16_31(\&body_00_19);
580 &Xupdate_ssse3_32_79(\&body_00_19);
581 &Xupdate_ssse3_32_79(\&body_20_39);
582 &Xupdate_ssse3_32_79(\&body_20_39);
583 &Xupdate_ssse3_32_79(\&body_20_39);
584 &Xupdate_ssse3_32_79(\&body_20_39);
585 &Xupdate_ssse3_32_79(\&body_20_39);
586 &Xupdate_ssse3_32_79(\&body_40_59);
587 &Xupdate_ssse3_32_79(\&body_40_59);
588 &Xupdate_ssse3_32_79(\&body_40_59);
589 &Xupdate_ssse3_32_79(\&body_40_59);
590 &Xupdate_ssse3_32_79(\&body_40_59);
591 &Xupdate_ssse3_32_79(\&body_20_39);
592 &Xuplast_ssse3_80(\&body_20_39); # can jump to "done"
594 $saved_j=$j; @saved_V=@V;
595 $saved_r=$r; @saved_rndkey=@rndkey;
597 &Xloop_ssse3(\&body_20_39);
598 &Xloop_ssse3(\&body_20_39);
599 &Xloop_ssse3(\&body_20_39);
602 movups $iv,48($out,$in0) # write output
605 add 0($ctx),$A # update context
612 mov @T[0],$B # magic seed
624 $jj=$j=$saved_j; @V=@saved_V;
625 $r=$saved_r; @rndkey=@saved_rndkey;
627 &Xtail_ssse3(\&body_20_39);
628 &Xtail_ssse3(\&body_20_39);
629 &Xtail_ssse3(\&body_20_39);
632 movups $iv,48($out,$in0) # write output
633 mov 88(%rsp),$ivp # restore $ivp
635 add 0($ctx),$A # update context
645 movups $iv,($ivp) # write IV
647 $code.=<<___ if ($win64);
648 movaps 96+0(%rsp),%xmm6
649 movaps 96+16(%rsp),%xmm7
650 movaps 96+32(%rsp),%xmm8
651 movaps 96+48(%rsp),%xmm9
652 movaps 96+64(%rsp),%xmm10
653 movaps 96+80(%rsp),%xmm11
654 movaps 96+96(%rsp),%xmm12
655 movaps 96+112(%rsp),%xmm13
656 movaps 96+128(%rsp),%xmm14
657 movaps 96+144(%rsp),%xmm15
660 lea `104+($win64?10*16:0)`(%rsp),%rsi
670 .size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
676 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
679 my @X=map("%xmm$_",(4..7,0..3));
680 my @Tx=map("%xmm$_",(8..10));
681 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
682 my @T=("%esi","%edi");
683 my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13));
684 my @rndkey=("%xmm14","%xmm15");
687 my $_rol=sub { &shld(@_[0],@_) };
688 my $_ror=sub { &shrd(@_[0],@_) };
691 .type aesni_cbc_sha1_enc_avx,\@function,6
693 aesni_cbc_sha1_enc_avx:
694 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
695 #shr \$6,$len # debugging artefact
696 #jz .Lepilogue_avx # debugging artefact
703 lea `-104-($win64?10*16:0)`(%rsp),%rsp
704 #mov $in0,$inp # debugging artefact
705 #lea 64(%rsp),$ctx # debugging artefact
707 $code.=<<___ if ($win64);
708 movaps %xmm6,96+0(%rsp)
709 movaps %xmm7,96+16(%rsp)
710 movaps %xmm8,96+32(%rsp)
711 movaps %xmm9,96+48(%rsp)
712 movaps %xmm10,96+64(%rsp)
713 movaps %xmm11,96+80(%rsp)
714 movaps %xmm12,96+96(%rsp)
715 movaps %xmm13,96+112(%rsp)
716 movaps %xmm14,96+128(%rsp)
717 movaps %xmm15,96+144(%rsp)
722 mov $in0,%r12 # reassign arguments
726 vmovdqu ($ivp),$iv # load IV
727 mov $ivp,88(%rsp) # save $ivp
729 my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
730 my $rounds="${ivp}d";
734 mov 240($key),$rounds
735 add \$112,$key # size optimization
736 add $inp,$len # end of input
738 lea K_XX_XX(%rip),$K_XX_XX
739 mov 0($ctx),$A # load context
743 mov $B,@T[0] # magic seed
749 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
750 vmovdqa 0($K_XX_XX),$Kx # K_00_19
751 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
752 vmovdqu 16($inp),@X[-3&7]
753 vmovdqu 32($inp),@X[-2&7]
754 vmovdqu 48($inp),@X[-1&7]
755 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
757 vpshufb @X[2],@X[-3&7],@X[-3&7]
758 vpshufb @X[2],@X[-2&7],@X[-2&7]
759 vpshufb @X[2],@X[-1&7],@X[-1&7]
760 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
761 vpaddd $Kx,@X[-3&7],@X[1]
762 vpaddd $Kx,@X[-2&7],@X[2]
763 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
764 vmovdqa @X[1],16(%rsp)
765 vmovdqa @X[2],32(%rsp)
766 vmovups -112($key),$rndkey[1] # $key[0]
767 vmovups 16-112($key),$rndkey[0] # forward reference
773 my ($n,$k)=($r/10,$r%10);
776 vmovups `16*$n`($in0),$in # load input
777 vxorps $rndkey[1],$in,$in
779 $code.=<<___ if ($n);
780 vmovups $iv,`16*($n-1)`($out,$in0) # write output
784 vaesenc $rndkey[0],$iv,$iv
785 vmovups `32+16*$k-112`($key),$rndkey[1]
792 vaesenc $rndkey[0],$iv,$iv
793 vmovups `32+16*($k+0)-112`($key),$rndkey[1]
794 vaesenc $rndkey[1],$iv,$iv
795 vmovups `32+16*($k+1)-112`($key),$rndkey[0]
797 vaesenc $rndkey[0],$iv,$iv
798 vmovups `32+16*($k+2)-112`($key),$rndkey[1]
799 vaesenc $rndkey[1],$iv,$iv
800 vmovups `32+16*($k+3)-112`($key),$rndkey[0]
802 vaesenclast $rndkey[0],$iv,$iv
803 vmovups -112($key),$rndkey[0]
804 vmovups 16-112($key),$rndkey[1] # forward reference
808 vaesenc $rndkey[0],$iv,$iv
809 vmovups `32+16*$k-112`($key),$rndkey[1]
812 $r++; unshift(@rndkey,pop(@rndkey));
815 sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
818 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
823 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
827 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
830 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
833 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
837 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
843 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
846 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
850 &vpsrld (@Tx[0],@X[0],31);
856 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
857 &vpaddd (@X[0],@X[0],@X[0]);
863 &vpsrld (@Tx[1],@Tx[2],30);
864 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
870 &vpslld (@Tx[2],@Tx[2],2);
871 &vpxor (@X[0],@X[0],@Tx[1]);
877 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
880 &vmovdqa ($Kx,eval(16*(($Xi)/5))."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
885 foreach (@insns) { eval; } # remaining instructions [if any]
887 $Xi++; push(@X,shift(@X)); # "rotate" X[]
890 sub Xupdate_avx_32_79()
893 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
896 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
897 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
898 eval(shift(@insns)); # body_20_39
901 eval(shift(@insns)); # rol
903 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
905 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
906 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
907 &vmovdqa ($Kx,eval(16*($Xi/5))."($K_XX_XX)") if ($Xi%5==0);
908 eval(shift(@insns)); # ror
911 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
912 eval(shift(@insns)); # body_20_39
915 eval(shift(@insns)); # rol
917 &vpsrld (@Tx[0],@X[0],30);
918 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
921 eval(shift(@insns)); # ror
924 &vpslld (@X[0],@X[0],2);
925 eval(shift(@insns)); # body_20_39
928 eval(shift(@insns)); # rol
931 eval(shift(@insns)); # ror
934 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
935 eval(shift(@insns)); # body_20_39
938 eval(shift(@insns)); # rol
941 eval(shift(@insns)); # rol
944 foreach (@insns) { eval; } # remaining instructions
946 $Xi++; push(@X,shift(@X)); # "rotate" X[]
952 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
956 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
962 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
964 foreach (@insns) { eval; } # remaining instructions
969 unshift(@Tx,pop(@Tx));
971 &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask
972 &vmovdqa($Kx,"0($K_XX_XX)"); # K_00_19
973 &vmovdqu(@X[-4&7],"0($inp)"); # load input
974 &vmovdqu(@X[-3&7],"16($inp)");
975 &vmovdqu(@X[-2&7],"32($inp)");
976 &vmovdqu(@X[-1&7],"48($inp)");
977 &vpshufb(@X[-4&7],@X[-4&7],@X[2]); # byte swap
986 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
991 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
994 &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],$Kx);
999 &vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]); # X[]+K xfer to IALU
1000 eval(shift(@insns));
1001 eval(shift(@insns));
1003 foreach (@insns) { eval; }
1010 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1011 my ($a,$b,$c,$d,$e);
1013 foreach (@insns) { eval; }
1020 &Xupdate_avx_16_31(\&body_00_19);
1021 &Xupdate_avx_16_31(\&body_00_19);
1022 &Xupdate_avx_16_31(\&body_00_19);
1023 &Xupdate_avx_16_31(\&body_00_19);
1024 &Xupdate_avx_32_79(\&body_00_19);
1025 &Xupdate_avx_32_79(\&body_20_39);
1026 &Xupdate_avx_32_79(\&body_20_39);
1027 &Xupdate_avx_32_79(\&body_20_39);
1028 &Xupdate_avx_32_79(\&body_20_39);
1029 &Xupdate_avx_32_79(\&body_20_39);
1030 &Xupdate_avx_32_79(\&body_40_59);
1031 &Xupdate_avx_32_79(\&body_40_59);
1032 &Xupdate_avx_32_79(\&body_40_59);
1033 &Xupdate_avx_32_79(\&body_40_59);
1034 &Xupdate_avx_32_79(\&body_40_59);
1035 &Xupdate_avx_32_79(\&body_20_39);
1036 &Xuplast_avx_80(\&body_20_39); # can jump to "done"
1038 $saved_j=$j; @saved_V=@V;
1039 $saved_r=$r; @saved_rndkey=@rndkey;
1041 &Xloop_avx(\&body_20_39);
1042 &Xloop_avx(\&body_20_39);
1043 &Xloop_avx(\&body_20_39);
1046 vmovups $iv,48($out,$in0) # write output
1049 add 0($ctx),$A # update context
1056 mov @T[0],$B # magic seed
1068 $jj=$j=$saved_j; @V=@saved_V;
1069 $r=$saved_r; @rndkey=@saved_rndkey;
1071 &Xtail_avx(\&body_20_39);
1072 &Xtail_avx(\&body_20_39);
1073 &Xtail_avx(\&body_20_39);
1076 vmovups $iv,48($out,$in0) # write output
1077 mov 88(%rsp),$ivp # restore $ivp
1079 add 0($ctx),$A # update context
1089 vmovups $iv,($ivp) # write IV
1092 $code.=<<___ if ($win64);
1093 movaps 96+0(%rsp),%xmm6
1094 movaps 96+16(%rsp),%xmm7
1095 movaps 96+32(%rsp),%xmm8
1096 movaps 96+48(%rsp),%xmm9
1097 movaps 96+64(%rsp),%xmm10
1098 movaps 96+80(%rsp),%xmm11
1099 movaps 96+96(%rsp),%xmm12
1100 movaps 96+112(%rsp),%xmm13
1101 movaps 96+128(%rsp),%xmm14
1102 movaps 96+144(%rsp),%xmm15
1105 lea `104+($win64?10*16:0)`(%rsp),%rsi
1115 .size aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
1121 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1122 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1123 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1124 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1125 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1127 .asciz "AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1131 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1132 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1140 .extern __imp_RtlVirtualUnwind
1141 .type ssse3_handler,\@abi-omnipotent
1155 mov 120($context),%rax # pull context->Rax
1156 mov 248($context),%rbx # pull context->Rip
1158 mov 8($disp),%rsi # disp->ImageBase
1159 mov 56($disp),%r11 # disp->HandlerData
1161 mov 0(%r11),%r10d # HandlerData[0]
1162 lea (%rsi,%r10),%r10 # prologue label
1163 cmp %r10,%rbx # context->Rip<prologue label
1164 jb .Lcommon_seh_tail
1166 mov 152($context),%rax # pull context->Rsp
1168 mov 4(%r11),%r10d # HandlerData[1]
1169 lea (%rsi,%r10),%r10 # epilogue label
1170 cmp %r10,%rbx # context->Rip>=epilogue label
1171 jae .Lcommon_seh_tail
1174 lea 512($context),%rdi # &context.Xmm6
1176 .long 0xa548f3fc # cld; rep movsq
1177 lea `104+10*16`(%rax),%rax # adjust stack pointer
1186 mov %rbx,144($context) # restore context->Rbx
1187 mov %rbp,160($context) # restore context->Rbp
1188 mov %r12,216($context) # restore context->R12
1189 mov %r13,224($context) # restore context->R13
1190 mov %r14,232($context) # restore context->R14
1191 mov %r15,240($context) # restore context->R15
1196 mov %rax,152($context) # restore context->Rsp
1197 mov %rsi,168($context) # restore context->Rsi
1198 mov %rdi,176($context) # restore context->Rdi
1200 mov 40($disp),%rdi # disp->ContextRecord
1201 mov $context,%rsi # context
1202 mov \$154,%ecx # sizeof(CONTEXT)
1203 .long 0xa548f3fc # cld; rep movsq
1206 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1207 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1208 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1209 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1210 mov 40(%rsi),%r10 # disp->ContextRecord
1211 lea 56(%rsi),%r11 # &disp->HandlerData
1212 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1213 mov %r10,32(%rsp) # arg5
1214 mov %r11,40(%rsp) # arg6
1215 mov %r12,48(%rsp) # arg7
1216 mov %rcx,56(%rsp) # arg8, (NULL)
1217 call *__imp_RtlVirtualUnwind(%rip)
1219 mov \$1,%eax # ExceptionContinueSearch
1231 .size ssse3_handler,.-ssse3_handler
1235 .rva .LSEH_begin_aesni_cbc_sha1_enc_ssse3
1236 .rva .LSEH_end_aesni_cbc_sha1_enc_ssse3
1237 .rva .LSEH_info_aesni_cbc_sha1_enc_ssse3
1239 $code.=<<___ if ($avx);
1240 .rva .LSEH_begin_aesni_cbc_sha1_enc_avx
1241 .rva .LSEH_end_aesni_cbc_sha1_enc_avx
1242 .rva .LSEH_info_aesni_cbc_sha1_enc_avx
1247 .LSEH_info_aesni_cbc_sha1_enc_ssse3:
1250 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[]
1252 $code.=<<___ if ($avx);
1253 .LSEH_info_aesni_cbc_sha1_enc_avx:
1256 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
1260 ####################################################################
1262 local *opcode=shift;
1266 $rex|=0x04 if($dst>=8);
1267 $rex|=0x01 if($src>=8);
1268 push @opcode,$rex|0x40 if($rex);
1275 if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1277 "aesenc" => 0xdc, "aesenclast" => 0xdd
1279 return undef if (!defined($opcodelet{$1}));
1280 rex(\@opcode,$3,$2);
1281 push @opcode,0x0f,0x38,$opcodelet{$1};
1282 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1283 return ".byte\t".join(',',@opcode);
1288 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1289 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;