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[+5.6] 9.37 6.65 +41%
25 # Sandy Bridge 5.05[+5.2(6.3)] 10.25(11.35) 6.16(7.08) +67%(+60%)
28 # Westmere 4.51 10.11 6.97 +45%
29 # Sandy Bridge 6.05 11.25(12.35) 6.34(7.27) +77%(+70%)
32 # Westmere 5.25 10.85 7.25 +50%
33 # Sandy Bridge 7.05 12.25(13.35) 7.06(7.70) +74%(+73%)
35 # (*) There are two code paths: SSSE3 and AVX. See sha1-568.pl for
36 # background information. Above numbers in parentheses are SSSE3
37 # results collected on AVX-capable CPU, i.e. apply on OSes that
40 # Needless to mention that it makes no sense to implement "stitched"
41 # *decrypt* subroutine. Because *both* AESNI-CBC decrypt and SHA1
42 # fully utilize parallelism, so stitching would not give any gain
43 # anyway. Well, there might be some, e.g. because of better cache
44 # locality... For reference, here are performance results for
45 # standalone AESNI-CBC decrypt:
47 # AES-128-CBC AES-192-CBC AES-256-CBC
48 # Westmere 1.31 1.55 1.80
49 # Sandy Bridge 0.93 1.06 1.22
53 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
55 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
57 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
58 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
59 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
60 die "can't locate x86_64-xlate.pl";
62 $avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
63 =~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
65 $avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
66 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
68 $avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
69 `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
71 $avx=1 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/ && $2>=3.0);
73 open OUT,"| \"$^X\" $xlate $flavour $output";
76 # void aesni_cbc_sha1_enc(const void *inp,
86 .extern OPENSSL_ia32cap_P
88 .globl aesni_cbc_sha1_enc
89 .type aesni_cbc_sha1_enc,\@abi-omnipotent
92 # caller should check for SSSE3 and AES-NI bits
93 mov OPENSSL_ia32cap_P+0(%rip),%r10d
94 mov OPENSSL_ia32cap_P+4(%rip),%r11d
96 $code.=<<___ if ($avx);
97 and \$`1<<28`,%r11d # mask AVX bit
98 and \$`1<<30`,%r10d # mask "Intel CPU" bit
100 cmp \$`1<<28|1<<30`,%r10d
101 je aesni_cbc_sha1_enc_avx
104 jmp aesni_cbc_sha1_enc_ssse3
106 .size aesni_cbc_sha1_enc,.-aesni_cbc_sha1_enc
109 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
112 my @X=map("%xmm$_",(4..7,0..3));
113 my @Tx=map("%xmm$_",(8..10));
114 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
115 my @T=("%esi","%edi");
116 my $j=0; my $jj=0; my $r=0; my $sn=0;
118 my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13));
119 my @rndkey=("%xmm14","%xmm15");
121 sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm
122 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
124 $arg = "\$$arg" if ($arg*1 eq $arg);
125 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
128 my $_rol=sub { &rol(@_) };
129 my $_ror=sub { &ror(@_) };
132 .type aesni_cbc_sha1_enc_ssse3,\@function,6
134 aesni_cbc_sha1_enc_ssse3:
135 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
136 #shr \$6,$len # debugging artefact
137 #jz .Lepilogue_ssse3 # debugging artefact
144 lea `-104-($win64?10*16:0)`(%rsp),%rsp
145 #mov $in0,$inp # debugging artefact
146 #lea 64(%rsp),$ctx # debugging artefact
148 $code.=<<___ if ($win64);
149 movaps %xmm6,96+0(%rsp)
150 movaps %xmm7,96+16(%rsp)
151 movaps %xmm8,96+32(%rsp)
152 movaps %xmm9,96+48(%rsp)
153 movaps %xmm10,96+64(%rsp)
154 movaps %xmm11,96+80(%rsp)
155 movaps %xmm12,96+96(%rsp)
156 movaps %xmm13,96+112(%rsp)
157 movaps %xmm14,96+128(%rsp)
158 movaps %xmm15,96+144(%rsp)
162 mov $in0,%r12 # reassign arguments
166 movdqu ($ivp),$iv # load IV
167 mov $ivp,88(%rsp) # save $ivp
169 my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
170 my $rounds="${ivp}d";
174 mov 240($key),$rounds
175 add $inp,$len # end of input
177 lea K_XX_XX(%rip),$K_XX_XX
178 mov 0($ctx),$A # load context
182 mov $B,@T[0] # magic seed
185 movdqa 64($K_XX_XX),@X[2] # pbswap mask
186 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
187 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
188 movdqu 16($inp),@X[-3&7]
189 movdqu 32($inp),@X[-2&7]
190 movdqu 48($inp),@X[-1&7]
191 pshufb @X[2],@X[-4&7] # byte swap
193 pshufb @X[2],@X[-3&7]
194 pshufb @X[2],@X[-2&7]
195 pshufb @X[2],@X[-1&7]
196 paddd @Tx[1],@X[-4&7] # add K_00_19
197 paddd @Tx[1],@X[-3&7]
198 paddd @Tx[1],@X[-2&7]
199 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
200 psubd @Tx[1],@X[-4&7] # restore X[]
201 movdqa @X[-3&7],16(%rsp)
202 psubd @Tx[1],@X[-3&7]
203 movdqa @X[-2&7],32(%rsp)
204 psubd @Tx[1],@X[-2&7]
205 movups ($key),$rndkey0 # $key[0]
206 movups 16($key),$rndkey[0] # forward reference
212 my ($n,$k)=($r/10,$r%10);
215 movups `16*$n`($in0),$in # load input
218 $code.=<<___ if ($n);
219 movups $iv,`16*($n-1)`($out,$in0) # write output
223 aesenc $rndkey[0],$iv
224 movups `32+16*$k`($key),$rndkey[1]
231 movups `32+16*($k+0)`($key),$rndkey[1]
232 aesenc $rndkey[0],$iv
233 movups `32+16*($k+1)`($key),$rndkey[0]
234 aesenc $rndkey[1],$iv
236 movups `32+16*($k+2)`($key),$rndkey[1]
237 aesenc $rndkey[0],$iv
238 movups `32+16*($k+3)`($key),$rndkey[0]
239 aesenc $rndkey[1],$iv
241 aesenclast $rndkey[0],$iv
242 movups 16($key),$rndkey[1] # forward reference
246 aesenc $rndkey[0],$iv
247 movups `32+16*$k`($key),$rndkey[1]
250 $r++; unshift(@rndkey,pop(@rndkey));
253 sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4
256 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
259 &movdqa (@X[0],@X[-3&7]);
262 &movdqa (@Tx[0],@X[-1&7]);
263 &palignr(@X[0],@X[-4&7],8); # compose "X[-14]" in "X[0]"
267 &paddd (@Tx[1],@X[-1&7]);
270 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords
273 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
277 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
283 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
286 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
290 &movdqa (@Tx[2],@X[0]);
291 &movdqa (@Tx[0],@X[0]);
297 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword
298 &paddd (@X[0],@X[0]);
307 &movdqa (@Tx[1],@Tx[2]);
312 &por (@X[0],@Tx[0]); # "X[0]"<<<=1
319 &pxor (@X[0],@Tx[2]);
322 &movdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
326 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
328 foreach (@insns) { eval; } # remaining instructions [if any]
330 $Xi++; push(@X,shift(@X)); # "rotate" X[]
331 push(@Tx,shift(@Tx));
334 sub Xupdate_ssse3_32_79()
337 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
340 &movdqa (@Tx[0],@X[-1&7]) if ($Xi==8);
341 eval(shift(@insns)); # body_20_39
342 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
343 &palignr(@Tx[0],@X[-2&7],8); # compose "X[-6]"
346 eval(shift(@insns)); # rol
348 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
350 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
352 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
353 } else { # ... or load next one
354 &movdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
356 &paddd (@Tx[1],@X[-1&7]);
357 eval(shift(@insns)); # ror
360 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]"
361 eval(shift(@insns)); # body_20_39
364 eval(shift(@insns)); # rol
366 &movdqa (@Tx[0],@X[0]);
367 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
370 eval(shift(@insns)); # ror
374 eval(shift(@insns)); # body_20_39
378 eval(shift(@insns)); # rol
381 eval(shift(@insns)); # ror
384 &por (@X[0],@Tx[0]); # "X[0]"<<<=2
385 eval(shift(@insns)); # body_20_39
387 &movdqa (@Tx[1],@X[0]) if ($Xi<19);
389 eval(shift(@insns)); # rol
392 eval(shift(@insns)); # rol
395 foreach (@insns) { eval; } # remaining instructions
397 $Xi++; push(@X,shift(@X)); # "rotate" X[]
398 push(@Tx,shift(@Tx));
401 sub Xuplast_ssse3_80()
404 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
408 &paddd (@Tx[1],@X[-1&7]);
414 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
416 foreach (@insns) { eval; } # remaining instructions
419 &je (".Ldone_ssse3");
421 unshift(@Tx,pop(@Tx));
423 &movdqa (@X[2],"64($K_XX_XX)"); # pbswap mask
424 &movdqa (@Tx[1],"0($K_XX_XX)"); # K_00_19
425 &movdqu (@X[-4&7],"0($inp)"); # load input
426 &movdqu (@X[-3&7],"16($inp)");
427 &movdqu (@X[-2&7],"32($inp)");
428 &movdqu (@X[-1&7],"48($inp)");
429 &pshufb (@X[-4&7],@X[2]); # byte swap
438 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
443 &pshufb (@X[($Xi-3)&7],@X[2]);
446 &paddd (@X[($Xi-4)&7],@Tx[1]);
451 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU
454 &psubd (@X[($Xi-4)&7],@Tx[1]);
456 foreach (@insns) { eval; }
463 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
466 foreach (@insns) { eval; }
473 '($a,$b,$c,$d,$e)=@V;'.
474 '&add ($e,eval(4*($j&15))."(%rsp)");', # X[]+K xfer
476 '&mov (@T[1],$a);', # $b in next round
478 '&and (@T[0],$c);', # ($b&($c^$d))
479 '&xor ($c,$d);', # restore $c
482 '&$_ror ($b,$j?7:2);', # $b>>>2
483 '&add ($e,@T[0]);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
486 $k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
487 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n);
496 '($a,$b,$c,$d,$e)=@V;'.
497 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer
498 '&xor (@T[0],$d);', # ($b^$d)
499 '&mov (@T[1],$a);', # $b in next round
501 '&xor (@T[0],$c);', # ($b^$d^$c)
503 '&$_ror ($b,7);', # $b>>>2
504 '&add ($e,@T[0]);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
507 $k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds
508 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n);
517 '($a,$b,$c,$d,$e)=@V;'.
520 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer
522 '&and (@T[0],$c);', # ($b&($c^$d))
523 '&$_ror ($b,7);', # $b>>>2
525 '&mov (@T[1],$a);', # $b in next round
528 '&xor ($c,$d);', # restore $c
529 '&add ($e,$a);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
532 $k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
533 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n);
541 &Xupdate_ssse3_16_31(\&body_00_19);
542 &Xupdate_ssse3_16_31(\&body_00_19);
543 &Xupdate_ssse3_16_31(\&body_00_19);
544 &Xupdate_ssse3_16_31(\&body_00_19);
545 &Xupdate_ssse3_32_79(\&body_00_19);
546 &Xupdate_ssse3_32_79(\&body_20_39);
547 &Xupdate_ssse3_32_79(\&body_20_39);
548 &Xupdate_ssse3_32_79(\&body_20_39);
549 &Xupdate_ssse3_32_79(\&body_20_39);
550 &Xupdate_ssse3_32_79(\&body_20_39);
551 &Xupdate_ssse3_32_79(\&body_40_59);
552 &Xupdate_ssse3_32_79(\&body_40_59);
553 &Xupdate_ssse3_32_79(\&body_40_59);
554 &Xupdate_ssse3_32_79(\&body_40_59);
555 &Xupdate_ssse3_32_79(\&body_40_59);
556 &Xupdate_ssse3_32_79(\&body_20_39);
557 &Xuplast_ssse3_80(\&body_20_39); # can jump to "done"
559 $saved_j=$j; @saved_V=@V;
560 $saved_r=$r; @saved_rndkey=@rndkey;
562 &Xloop_ssse3(\&body_20_39);
563 &Xloop_ssse3(\&body_20_39);
564 &Xloop_ssse3(\&body_20_39);
567 movups $iv,48($out,$in0) # write output
570 add 0($ctx),$A # update context
577 mov @T[0],$B # magic seed
586 $jj=$j=$saved_j; @V=@saved_V;
587 $r=$saved_r; @rndkey=@saved_rndkey;
589 &Xtail_ssse3(\&body_20_39);
590 &Xtail_ssse3(\&body_20_39);
591 &Xtail_ssse3(\&body_20_39);
594 movups $iv,48($out,$in0) # write output
595 mov 88(%rsp),$ivp # restore $ivp
597 add 0($ctx),$A # update context
607 movups $iv,($ivp) # write IV
609 $code.=<<___ if ($win64);
610 movaps 96+0(%rsp),%xmm6
611 movaps 96+16(%rsp),%xmm7
612 movaps 96+32(%rsp),%xmm8
613 movaps 96+48(%rsp),%xmm9
614 movaps 96+64(%rsp),%xmm10
615 movaps 96+80(%rsp),%xmm11
616 movaps 96+96(%rsp),%xmm12
617 movaps 96+112(%rsp),%xmm13
618 movaps 96+128(%rsp),%xmm14
619 movaps 96+144(%rsp),%xmm15
622 lea `104+($win64?10*16:0)`(%rsp),%rsi
632 .size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
638 my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
641 my @X=map("%xmm$_",(4..7,0..3));
642 my @Tx=map("%xmm$_",(8..10));
643 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
644 my @T=("%esi","%edi");
646 my $_rol=sub { &shld(@_[0],@_) };
647 my $_ror=sub { &shrd(@_[0],@_) };
650 .type aesni_cbc_sha1_enc_avx,\@function,6
652 aesni_cbc_sha1_enc_avx:
653 mov `($win64?56:8)`(%rsp),$inp # load 7th argument
654 #shr \$6,$len # debugging artefact
655 #jz .Lepilogue_avx # debugging artefact
662 lea `-104-($win64?10*16:0)`(%rsp),%rsp
663 #mov $in0,$inp # debugging artefact
664 #lea 64(%rsp),$ctx # debugging artefact
666 $code.=<<___ if ($win64);
667 movaps %xmm6,96+0(%rsp)
668 movaps %xmm7,96+16(%rsp)
669 movaps %xmm8,96+32(%rsp)
670 movaps %xmm9,96+48(%rsp)
671 movaps %xmm10,96+64(%rsp)
672 movaps %xmm11,96+80(%rsp)
673 movaps %xmm12,96+96(%rsp)
674 movaps %xmm13,96+112(%rsp)
675 movaps %xmm14,96+128(%rsp)
676 movaps %xmm15,96+144(%rsp)
681 mov $in0,%r12 # reassign arguments
685 vmovdqu ($ivp),$iv # load IV
686 mov $ivp,88(%rsp) # save $ivp
688 my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments
689 my $rounds="${ivp}d";
693 mov 240($key),$rounds
694 add \$112,$key # size optimization
695 add $inp,$len # end of input
697 lea K_XX_XX(%rip),$K_XX_XX
698 mov 0($ctx),$A # load context
702 mov $B,@T[0] # magic seed
705 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
706 vmovdqa 0($K_XX_XX),@Tx[1] # K_00_19
707 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
708 vmovdqu 16($inp),@X[-3&7]
709 vmovdqu 32($inp),@X[-2&7]
710 vmovdqu 48($inp),@X[-1&7]
711 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
713 vpshufb @X[2],@X[-3&7],@X[-3&7]
714 vpshufb @X[2],@X[-2&7],@X[-2&7]
715 vpshufb @X[2],@X[-1&7],@X[-1&7]
716 vpaddd @Tx[1],@X[-4&7],@X[0] # add K_00_19
717 vpaddd @Tx[1],@X[-3&7],@X[1]
718 vpaddd @Tx[1],@X[-2&7],@X[2]
719 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
720 vmovdqa @X[1],16(%rsp)
721 vmovdqa @X[2],32(%rsp)
722 vmovups -112($key),$rndkey0 # $key[0]
723 vmovups 16-112($key),$rndkey[0] # forward reference
729 my ($n,$k)=($r/10,$r%10);
732 vmovups `16*$n`($in0),$in # load input
733 vxorps $rndkey0,$in,$in
735 $code.=<<___ if ($n);
736 vmovups $iv,`16*($n-1)`($out,$in0) # write output
740 vaesenc $rndkey[0],$iv,$iv
741 vmovups `32+16*$k-112`($key),$rndkey[1]
748 vaesenc $rndkey[0],$iv,$iv
749 vmovups `32+16*($k+0)-112`($key),$rndkey[1]
750 vaesenc $rndkey[1],$iv,$iv
751 vmovups `32+16*($k+1)-112`($key),$rndkey[0]
753 vaesenc $rndkey[0],$iv,$iv
754 vmovups `32+16*($k+2)-112`($key),$rndkey[1]
755 vaesenc $rndkey[1],$iv,$iv
756 vmovups `32+16*($k+3)-112`($key),$rndkey[0]
758 vaesenclast $rndkey[0],$iv,$iv
759 vmovups 16-112($key),$rndkey[1] # forward reference
763 vaesenc $rndkey[0],$iv,$iv
764 vmovups `32+16*$k-112`($key),$rndkey[1]
767 $r++; unshift(@rndkey,pop(@rndkey));
770 sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
773 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
778 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
782 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
785 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
788 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
792 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
798 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
801 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
805 &vpsrld (@Tx[0],@X[0],31);
811 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
812 &vpaddd (@X[0],@X[0],@X[0]);
818 &vpsrld (@Tx[1],@Tx[2],30);
819 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
825 &vpslld (@Tx[2],@Tx[2],2);
826 &vpxor (@X[0],@X[0],@Tx[1]);
832 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
835 &vmovdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
840 foreach (@insns) { eval; } # remaining instructions [if any]
842 $Xi++; push(@X,shift(@X)); # "rotate" X[]
843 push(@Tx,shift(@Tx));
846 sub Xupdate_avx_32_79()
849 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions
852 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
853 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
854 eval(shift(@insns)); # body_20_39
857 eval(shift(@insns)); # rol
859 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
861 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
863 &vmovdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
864 } else { # ... or load next one
865 &vmovdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
867 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
868 eval(shift(@insns)); # ror
871 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
872 eval(shift(@insns)); # body_20_39
875 eval(shift(@insns)); # rol
877 &vpsrld (@Tx[0],@X[0],30);
878 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
881 eval(shift(@insns)); # ror
884 &vpslld (@X[0],@X[0],2);
885 eval(shift(@insns)); # body_20_39
888 eval(shift(@insns)); # rol
891 eval(shift(@insns)); # ror
894 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
895 eval(shift(@insns)); # body_20_39
897 &vmovdqa (@Tx[1],@X[0]) if ($Xi<19);
899 eval(shift(@insns)); # rol
902 eval(shift(@insns)); # rol
905 foreach (@insns) { eval; } # remaining instructions
907 $Xi++; push(@X,shift(@X)); # "rotate" X[]
908 push(@Tx,shift(@Tx));
914 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
918 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
924 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
926 foreach (@insns) { eval; } # remaining instructions
931 unshift(@Tx,pop(@Tx));
933 &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask
934 &vmovdqa(@Tx[1],"0($K_XX_XX)"); # K_00_19
935 &vmovdqu(@X[-4&7],"0($inp)"); # load input
936 &vmovdqu(@X[-3&7],"16($inp)");
937 &vmovdqu(@X[-2&7],"32($inp)");
938 &vmovdqu(@X[-1&7],"48($inp)");
939 &vpshufb(@X[-4&7],@X[-4&7],@X[2]); # byte swap
948 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
953 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
956 &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],@Tx[1]);
961 &vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]); # X[]+K xfer to IALU
965 foreach (@insns) { eval; }
972 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
975 foreach (@insns) { eval; }
982 &Xupdate_avx_16_31(\&body_00_19);
983 &Xupdate_avx_16_31(\&body_00_19);
984 &Xupdate_avx_16_31(\&body_00_19);
985 &Xupdate_avx_16_31(\&body_00_19);
986 &Xupdate_avx_32_79(\&body_00_19);
987 &Xupdate_avx_32_79(\&body_20_39);
988 &Xupdate_avx_32_79(\&body_20_39);
989 &Xupdate_avx_32_79(\&body_20_39);
990 &Xupdate_avx_32_79(\&body_20_39);
991 &Xupdate_avx_32_79(\&body_20_39);
992 &Xupdate_avx_32_79(\&body_40_59);
993 &Xupdate_avx_32_79(\&body_40_59);
994 &Xupdate_avx_32_79(\&body_40_59);
995 &Xupdate_avx_32_79(\&body_40_59);
996 &Xupdate_avx_32_79(\&body_40_59);
997 &Xupdate_avx_32_79(\&body_20_39);
998 &Xuplast_avx_80(\&body_20_39); # can jump to "done"
1000 $saved_j=$j; @saved_V=@V;
1001 $saved_r=$r; @saved_rndkey=@rndkey;
1003 &Xloop_avx(\&body_20_39);
1004 &Xloop_avx(\&body_20_39);
1005 &Xloop_avx(\&body_20_39);
1008 vmovups $iv,48($out,$in0) # write output
1011 add 0($ctx),$A # update context
1018 mov @T[0],$B # magic seed
1027 $jj=$j=$saved_j; @V=@saved_V;
1028 $r=$saved_r; @rndkey=@saved_rndkey;
1030 &Xtail_avx(\&body_20_39);
1031 &Xtail_avx(\&body_20_39);
1032 &Xtail_avx(\&body_20_39);
1035 vmovups $iv,48($out,$in0) # write output
1036 mov 88(%rsp),$ivp # restore $ivp
1038 add 0($ctx),$A # update context
1048 vmovups $iv,($ivp) # write IV
1051 $code.=<<___ if ($win64);
1052 movaps 96+0(%rsp),%xmm6
1053 movaps 96+16(%rsp),%xmm7
1054 movaps 96+32(%rsp),%xmm8
1055 movaps 96+48(%rsp),%xmm9
1056 movaps 96+64(%rsp),%xmm10
1057 movaps 96+80(%rsp),%xmm11
1058 movaps 96+96(%rsp),%xmm12
1059 movaps 96+112(%rsp),%xmm13
1060 movaps 96+128(%rsp),%xmm14
1061 movaps 96+144(%rsp),%xmm15
1064 lea `104+($win64?10*16:0)`(%rsp),%rsi
1074 .size aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
1080 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1081 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1082 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1083 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1084 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1086 .asciz "AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1090 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1091 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1099 .extern __imp_RtlVirtualUnwind
1100 .type ssse3_handler,\@abi-omnipotent
1114 mov 120($context),%rax # pull context->Rax
1115 mov 248($context),%rbx # pull context->Rip
1117 mov 8($disp),%rsi # disp->ImageBase
1118 mov 56($disp),%r11 # disp->HandlerData
1120 mov 0(%r11),%r10d # HandlerData[0]
1121 lea (%rsi,%r10),%r10 # prologue label
1122 cmp %r10,%rbx # context->Rip<prologue label
1123 jb .Lcommon_seh_tail
1125 mov 152($context),%rax # pull context->Rsp
1127 mov 4(%r11),%r10d # HandlerData[1]
1128 lea (%rsi,%r10),%r10 # epilogue label
1129 cmp %r10,%rbx # context->Rip>=epilogue label
1130 jae .Lcommon_seh_tail
1133 lea 512($context),%rdi # &context.Xmm6
1135 .long 0xa548f3fc # cld; rep movsq
1136 lea `104+10*16`(%rax),%rax # adjust stack pointer
1145 mov %rbx,144($context) # restore context->Rbx
1146 mov %rbp,160($context) # restore context->Rbp
1147 mov %r12,216($context) # restore context->R12
1148 mov %r13,224($context) # restore context->R13
1149 mov %r14,232($context) # restore context->R14
1150 mov %r15,240($context) # restore context->R15
1155 mov %rax,152($context) # restore context->Rsp
1156 mov %rsi,168($context) # restore context->Rsi
1157 mov %rdi,176($context) # restore context->Rdi
1159 mov 40($disp),%rdi # disp->ContextRecord
1160 mov $context,%rsi # context
1161 mov \$154,%ecx # sizeof(CONTEXT)
1162 .long 0xa548f3fc # cld; rep movsq
1165 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1166 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1167 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1168 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1169 mov 40(%rsi),%r10 # disp->ContextRecord
1170 lea 56(%rsi),%r11 # &disp->HandlerData
1171 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1172 mov %r10,32(%rsp) # arg5
1173 mov %r11,40(%rsp) # arg6
1174 mov %r12,48(%rsp) # arg7
1175 mov %rcx,56(%rsp) # arg8, (NULL)
1176 call *__imp_RtlVirtualUnwind(%rip)
1178 mov \$1,%eax # ExceptionContinueSearch
1190 .size ssse3_handler,.-ssse3_handler
1194 .rva .LSEH_begin_aesni_cbc_sha1_enc_ssse3
1195 .rva .LSEH_end_aesni_cbc_sha1_enc_ssse3
1196 .rva .LSEH_info_aesni_cbc_sha1_enc_ssse3
1198 $code.=<<___ if ($avx);
1199 .rva .LSEH_begin_aesni_cbc_sha1_enc_avx
1200 .rva .LSEH_end_aesni_cbc_sha1_enc_avx
1201 .rva .LSEH_info_aesni_cbc_sha1_enc_avx
1206 .LSEH_info_aesni_cbc_sha1_enc_ssse3:
1209 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[]
1211 $code.=<<___ if ($avx);
1212 .LSEH_info_aesni_cbc_sha1_enc_avx:
1215 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
1219 ####################################################################
1221 local *opcode=shift;
1225 $rex|=0x04 if($dst>=8);
1226 $rex|=0x01 if($src>=8);
1227 push @opcode,$rex|0x40 if($rex);
1234 if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1236 "aesenc" => 0xdc, "aesenclast" => 0xdd
1238 return undef if (!defined($opcodelet{$1}));
1239 rex(\@opcode,$3,$2);
1240 push @opcode,0x0f,0x38,$opcodelet{$1};
1241 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1242 return ".byte\t".join(',',@opcode);
1247 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1248 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;