aesni-sha1-x86_64.pl revision 392aa7cc7d2b122614c5393c3e357da07fd07af3
1#!/usr/bin/env perl 2# 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# ==================================================================== 9# 10# June 2011 11# 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" 21# subroutine: 22# 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%) 26# 27# AES-192-CBC 28# Westmere 4.51 10.11 6.97 +45% 29# Sandy Bridge 6.05 11.25(12.35) 6.34(7.27) +77%(+70%) 30# 31# AES-256-CBC 32# Westmere 5.25 10.85 7.25 +50% 33# Sandy Bridge 7.05 12.25(13.35) 7.06(7.70) +74%(+73%) 34# 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 38# don't support AVX. 39# 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: 46# 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 50 51$flavour = shift; 52$output = shift; 53if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 54 55$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); 56 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 $xlate) or 60die "can't locate x86_64-xlate.pl"; 61 62$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]+)/ && 64 $1>=2.19); 65$avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) && 66 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ && 67 $1>=2.09); 68$avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) && 69 `ml64 2>&1` =~ /Version ([0-9]+)\./ && 70 $1>=10); 71 72open STDOUT,"| $^X $xlate $flavour $output"; 73 74# void aesni_cbc_sha1_enc(const void *inp, 75# void *out, 76# size_t length, 77# const AES_KEY *key, 78# unsigned char *iv, 79# SHA_CTX *ctx, 80# const void *in0); 81 82$code.=<<___; 83.text 84.extern OPENSSL_ia32cap_P 85 86.globl aesni_cbc_sha1_enc 87.type aesni_cbc_sha1_enc,\@abi-omnipotent 88.align 16 89aesni_cbc_sha1_enc: 90 # caller should check for SSSE3 and AES-NI bits 91 mov OPENSSL_ia32cap_P+0(%rip),%r10d 92 mov OPENSSL_ia32cap_P+4(%rip),%r11d 93___ 94$code.=<<___ if ($avx); 95 and \$`1<<28`,%r11d # mask AVX bit 96 and \$`1<<30`,%r10d # mask "Intel CPU" bit 97 or %r11d,%r10d 98 cmp \$`1<<28|1<<30`,%r10d 99 je aesni_cbc_sha1_enc_avx 100___ 101$code.=<<___; 102 jmp aesni_cbc_sha1_enc_ssse3 103 ret 104.size aesni_cbc_sha1_enc,.-aesni_cbc_sha1_enc 105___ 106 107my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10"); 108 109my $Xi=4; 110my @X=map("%xmm$_",(4..7,0..3)); 111my @Tx=map("%xmm$_",(8..10)); 112my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization 113my @T=("%esi","%edi"); 114my $j=0; my $jj=0; my $r=0; my $sn=0; 115my $K_XX_XX="%r11"; 116my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13)); 117my @rndkey=("%xmm14","%xmm15"); 118 119sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm 120{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; 121 my $arg = pop; 122 $arg = "\$$arg" if ($arg*1 eq $arg); 123 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n"; 124} 125 126my $_rol=sub { &rol(@_) }; 127my $_ror=sub { &ror(@_) }; 128 129$code.=<<___; 130.type aesni_cbc_sha1_enc_ssse3,\@function,6 131.align 16 132aesni_cbc_sha1_enc_ssse3: 133 mov `($win64?56:8)`(%rsp),$inp # load 7th argument 134 #shr \$6,$len # debugging artefact 135 #jz .Lepilogue_ssse3 # debugging artefact 136 push %rbx 137 push %rbp 138 push %r12 139 push %r13 140 push %r14 141 push %r15 142 lea `-104-($win64?10*16:0)`(%rsp),%rsp 143 #mov $in0,$inp # debugging artefact 144 #lea 64(%rsp),$ctx # debugging artefact 145___ 146$code.=<<___ if ($win64); 147 movaps %xmm6,96+0(%rsp) 148 movaps %xmm7,96+16(%rsp) 149 movaps %xmm8,96+32(%rsp) 150 movaps %xmm9,96+48(%rsp) 151 movaps %xmm10,96+64(%rsp) 152 movaps %xmm11,96+80(%rsp) 153 movaps %xmm12,96+96(%rsp) 154 movaps %xmm13,96+112(%rsp) 155 movaps %xmm14,96+128(%rsp) 156 movaps %xmm15,96+144(%rsp) 157.Lprologue_ssse3: 158___ 159$code.=<<___; 160 mov $in0,%r12 # reassign arguments 161 mov $out,%r13 162 mov $len,%r14 163 mov $key,%r15 164 movdqu ($ivp),$iv # load IV 165 mov $ivp,88(%rsp) # save $ivp 166___ 167my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments 168my $rounds="${ivp}d"; 169$code.=<<___; 170 shl \$6,$len 171 sub $in0,$out 172 mov 240($key),$rounds 173 add $inp,$len # end of input 174 175 lea K_XX_XX(%rip),$K_XX_XX 176 mov 0($ctx),$A # load context 177 mov 4($ctx),$B 178 mov 8($ctx),$C 179 mov 12($ctx),$D 180 mov $B,@T[0] # magic seed 181 mov 16($ctx),$E 182 183 movdqa 64($K_XX_XX),@X[2] # pbswap mask 184 movdqa 0($K_XX_XX),@Tx[1] # K_00_19 185 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3] 186 movdqu 16($inp),@X[-3&7] 187 movdqu 32($inp),@X[-2&7] 188 movdqu 48($inp),@X[-1&7] 189 pshufb @X[2],@X[-4&7] # byte swap 190 add \$64,$inp 191 pshufb @X[2],@X[-3&7] 192 pshufb @X[2],@X[-2&7] 193 pshufb @X[2],@X[-1&7] 194 paddd @Tx[1],@X[-4&7] # add K_00_19 195 paddd @Tx[1],@X[-3&7] 196 paddd @Tx[1],@X[-2&7] 197 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU 198 psubd @Tx[1],@X[-4&7] # restore X[] 199 movdqa @X[-3&7],16(%rsp) 200 psubd @Tx[1],@X[-3&7] 201 movdqa @X[-2&7],32(%rsp) 202 psubd @Tx[1],@X[-2&7] 203 movups ($key),$rndkey0 # $key[0] 204 movups 16($key),$rndkey[0] # forward reference 205 jmp .Loop_ssse3 206___ 207 208my $aesenc=sub { 209 use integer; 210 my ($n,$k)=($r/10,$r%10); 211 if ($k==0) { 212 $code.=<<___; 213 movups `16*$n`($in0),$in # load input 214 xorps $rndkey0,$in 215___ 216 $code.=<<___ if ($n); 217 movups $iv,`16*($n-1)`($out,$in0) # write output 218___ 219 $code.=<<___; 220 xorps $in,$iv 221 aesenc $rndkey[0],$iv 222 movups `32+16*$k`($key),$rndkey[1] 223___ 224 } elsif ($k==9) { 225 $sn++; 226 $code.=<<___; 227 cmp \$11,$rounds 228 jb .Laesenclast$sn 229 movups `32+16*($k+0)`($key),$rndkey[1] 230 aesenc $rndkey[0],$iv 231 movups `32+16*($k+1)`($key),$rndkey[0] 232 aesenc $rndkey[1],$iv 233 je .Laesenclast$sn 234 movups `32+16*($k+2)`($key),$rndkey[1] 235 aesenc $rndkey[0],$iv 236 movups `32+16*($k+3)`($key),$rndkey[0] 237 aesenc $rndkey[1],$iv 238.Laesenclast$sn: 239 aesenclast $rndkey[0],$iv 240 movups 16($key),$rndkey[1] # forward reference 241___ 242 } else { 243 $code.=<<___; 244 aesenc $rndkey[0],$iv 245 movups `32+16*$k`($key),$rndkey[1] 246___ 247 } 248 $r++; unshift(@rndkey,pop(@rndkey)); 249}; 250 251sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4 252{ use integer; 253 my $body = shift; 254 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions 255 my ($a,$b,$c,$d,$e); 256 257 &movdqa (@X[0],@X[-3&7]); 258 eval(shift(@insns)); 259 eval(shift(@insns)); 260 &movdqa (@Tx[0],@X[-1&7]); 261 &palignr(@X[0],@X[-4&7],8); # compose "X[-14]" in "X[0]" 262 eval(shift(@insns)); 263 eval(shift(@insns)); 264 265 &paddd (@Tx[1],@X[-1&7]); 266 eval(shift(@insns)); 267 eval(shift(@insns)); 268 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords 269 eval(shift(@insns)); 270 eval(shift(@insns)); 271 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]" 272 eval(shift(@insns)); 273 eval(shift(@insns)); 274 275 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]" 276 eval(shift(@insns)); 277 eval(shift(@insns)); 278 eval(shift(@insns)); 279 eval(shift(@insns)); 280 281 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]" 282 eval(shift(@insns)); 283 eval(shift(@insns)); 284 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU 285 eval(shift(@insns)); 286 eval(shift(@insns)); 287 288 &movdqa (@Tx[2],@X[0]); 289 &movdqa (@Tx[0],@X[0]); 290 eval(shift(@insns)); 291 eval(shift(@insns)); 292 eval(shift(@insns)); 293 eval(shift(@insns)); 294 295 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword 296 &paddd (@X[0],@X[0]); 297 eval(shift(@insns)); 298 eval(shift(@insns)); 299 eval(shift(@insns)); 300 eval(shift(@insns)); 301 302 &psrld (@Tx[0],31); 303 eval(shift(@insns)); 304 eval(shift(@insns)); 305 &movdqa (@Tx[1],@Tx[2]); 306 eval(shift(@insns)); 307 eval(shift(@insns)); 308 309 &psrld (@Tx[2],30); 310 &por (@X[0],@Tx[0]); # "X[0]"<<<=1 311 eval(shift(@insns)); 312 eval(shift(@insns)); 313 eval(shift(@insns)); 314 eval(shift(@insns)); 315 316 &pslld (@Tx[1],2); 317 &pxor (@X[0],@Tx[2]); 318 eval(shift(@insns)); 319 eval(shift(@insns)); 320 &movdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX 321 eval(shift(@insns)); 322 eval(shift(@insns)); 323 324 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2 325 326 foreach (@insns) { eval; } # remaining instructions [if any] 327 328 $Xi++; push(@X,shift(@X)); # "rotate" X[] 329 push(@Tx,shift(@Tx)); 330} 331 332sub Xupdate_ssse3_32_79() 333{ use integer; 334 my $body = shift; 335 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions 336 my ($a,$b,$c,$d,$e); 337 338 &movdqa (@Tx[0],@X[-1&7]) if ($Xi==8); 339 eval(shift(@insns)); # body_20_39 340 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]" 341 &palignr(@Tx[0],@X[-2&7],8); # compose "X[-6]" 342 eval(shift(@insns)); 343 eval(shift(@insns)); 344 eval(shift(@insns)); # rol 345 346 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]" 347 eval(shift(@insns)); 348 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/); 349 if ($Xi%5) { 350 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX... 351 } else { # ... or load next one 352 &movdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)"); 353 } 354 &paddd (@Tx[1],@X[-1&7]); 355 eval(shift(@insns)); # ror 356 eval(shift(@insns)); 357 358 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]" 359 eval(shift(@insns)); # body_20_39 360 eval(shift(@insns)); 361 eval(shift(@insns)); 362 eval(shift(@insns)); # rol 363 364 &movdqa (@Tx[0],@X[0]); 365 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU 366 eval(shift(@insns)); 367 eval(shift(@insns)); 368 eval(shift(@insns)); # ror 369 eval(shift(@insns)); 370 371 &pslld (@X[0],2); 372 eval(shift(@insns)); # body_20_39 373 eval(shift(@insns)); 374 &psrld (@Tx[0],30); 375 eval(shift(@insns)); 376 eval(shift(@insns)); # rol 377 eval(shift(@insns)); 378 eval(shift(@insns)); 379 eval(shift(@insns)); # ror 380 eval(shift(@insns)); 381 382 &por (@X[0],@Tx[0]); # "X[0]"<<<=2 383 eval(shift(@insns)); # body_20_39 384 eval(shift(@insns)); 385 &movdqa (@Tx[1],@X[0]) if ($Xi<19); 386 eval(shift(@insns)); 387 eval(shift(@insns)); # rol 388 eval(shift(@insns)); 389 eval(shift(@insns)); 390 eval(shift(@insns)); # rol 391 eval(shift(@insns)); 392 393 foreach (@insns) { eval; } # remaining instructions 394 395 $Xi++; push(@X,shift(@X)); # "rotate" X[] 396 push(@Tx,shift(@Tx)); 397} 398 399sub Xuplast_ssse3_80() 400{ use integer; 401 my $body = shift; 402 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 403 my ($a,$b,$c,$d,$e); 404 405 eval(shift(@insns)); 406 &paddd (@Tx[1],@X[-1&7]); 407 eval(shift(@insns)); 408 eval(shift(@insns)); 409 eval(shift(@insns)); 410 eval(shift(@insns)); 411 412 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU 413 414 foreach (@insns) { eval; } # remaining instructions 415 416 &cmp ($inp,$len); 417 &je (".Ldone_ssse3"); 418 419 unshift(@Tx,pop(@Tx)); 420 421 &movdqa (@X[2],"64($K_XX_XX)"); # pbswap mask 422 &movdqa (@Tx[1],"0($K_XX_XX)"); # K_00_19 423 &movdqu (@X[-4&7],"0($inp)"); # load input 424 &movdqu (@X[-3&7],"16($inp)"); 425 &movdqu (@X[-2&7],"32($inp)"); 426 &movdqu (@X[-1&7],"48($inp)"); 427 &pshufb (@X[-4&7],@X[2]); # byte swap 428 &add ($inp,64); 429 430 $Xi=0; 431} 432 433sub Xloop_ssse3() 434{ use integer; 435 my $body = shift; 436 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 437 my ($a,$b,$c,$d,$e); 438 439 eval(shift(@insns)); 440 eval(shift(@insns)); 441 &pshufb (@X[($Xi-3)&7],@X[2]); 442 eval(shift(@insns)); 443 eval(shift(@insns)); 444 &paddd (@X[($Xi-4)&7],@Tx[1]); 445 eval(shift(@insns)); 446 eval(shift(@insns)); 447 eval(shift(@insns)); 448 eval(shift(@insns)); 449 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU 450 eval(shift(@insns)); 451 eval(shift(@insns)); 452 &psubd (@X[($Xi-4)&7],@Tx[1]); 453 454 foreach (@insns) { eval; } 455 $Xi++; 456} 457 458sub Xtail_ssse3() 459{ use integer; 460 my $body = shift; 461 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 462 my ($a,$b,$c,$d,$e); 463 464 foreach (@insns) { eval; } 465} 466 467sub body_00_19 () { 468 use integer; 469 my ($k,$n); 470 my @r=( 471 '($a,$b,$c,$d,$e)=@V;'. 472 '&add ($e,eval(4*($j&15))."(%rsp)");', # X[]+K xfer 473 '&xor ($c,$d);', 474 '&mov (@T[1],$a);', # $b in next round 475 '&$_rol ($a,5);', 476 '&and (@T[0],$c);', # ($b&($c^$d)) 477 '&xor ($c,$d);', # restore $c 478 '&xor (@T[0],$d);', 479 '&add ($e,$a);', 480 '&$_ror ($b,$j?7:2);', # $b>>>2 481 '&add ($e,@T[0]);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));' 482 ); 483 $n = scalar(@r); 484 $k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds 485 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n); 486 $jj++; 487 return @r; 488} 489 490sub body_20_39 () { 491 use integer; 492 my ($k,$n); 493 my @r=( 494 '($a,$b,$c,$d,$e)=@V;'. 495 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer 496 '&xor (@T[0],$d);', # ($b^$d) 497 '&mov (@T[1],$a);', # $b in next round 498 '&$_rol ($a,5);', 499 '&xor (@T[0],$c);', # ($b^$d^$c) 500 '&add ($e,$a);', 501 '&$_ror ($b,7);', # $b>>>2 502 '&add ($e,@T[0]);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));' 503 ); 504 $n = scalar(@r); 505 $k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds 506 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n); 507 $jj++; 508 return @r; 509} 510 511sub body_40_59 () { 512 use integer; 513 my ($k,$n); 514 my @r=( 515 '($a,$b,$c,$d,$e)=@V;'. 516 '&mov (@T[1],$c);', 517 '&xor ($c,$d);', 518 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer 519 '&and (@T[1],$d);', 520 '&and (@T[0],$c);', # ($b&($c^$d)) 521 '&$_ror ($b,7);', # $b>>>2 522 '&add ($e,@T[1]);', 523 '&mov (@T[1],$a);', # $b in next round 524 '&$_rol ($a,5);', 525 '&add ($e,@T[0]);', 526 '&xor ($c,$d);', # restore $c 527 '&add ($e,$a);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));' 528 ); 529 $n = scalar(@r); 530 $k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds 531 @r[$k%$n].='&$aesenc();' if ($jj==$k/$n); 532 $jj++; 533 return @r; 534} 535$code.=<<___; 536.align 16 537.Loop_ssse3: 538___ 539 &Xupdate_ssse3_16_31(\&body_00_19); 540 &Xupdate_ssse3_16_31(\&body_00_19); 541 &Xupdate_ssse3_16_31(\&body_00_19); 542 &Xupdate_ssse3_16_31(\&body_00_19); 543 &Xupdate_ssse3_32_79(\&body_00_19); 544 &Xupdate_ssse3_32_79(\&body_20_39); 545 &Xupdate_ssse3_32_79(\&body_20_39); 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_40_59); 550 &Xupdate_ssse3_32_79(\&body_40_59); 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_20_39); 555 &Xuplast_ssse3_80(\&body_20_39); # can jump to "done" 556 557 $saved_j=$j; @saved_V=@V; 558 $saved_r=$r; @saved_rndkey=@rndkey; 559 560 &Xloop_ssse3(\&body_20_39); 561 &Xloop_ssse3(\&body_20_39); 562 &Xloop_ssse3(\&body_20_39); 563 564$code.=<<___; 565 movups $iv,48($out,$in0) # write output 566 lea 64($in0),$in0 567 568 add 0($ctx),$A # update context 569 add 4($ctx),@T[0] 570 add 8($ctx),$C 571 add 12($ctx),$D 572 mov $A,0($ctx) 573 add 16($ctx),$E 574 mov @T[0],4($ctx) 575 mov @T[0],$B # magic seed 576 mov $C,8($ctx) 577 mov $D,12($ctx) 578 mov $E,16($ctx) 579 jmp .Loop_ssse3 580 581.align 16 582.Ldone_ssse3: 583___ 584 $jj=$j=$saved_j; @V=@saved_V; 585 $r=$saved_r; @rndkey=@saved_rndkey; 586 587 &Xtail_ssse3(\&body_20_39); 588 &Xtail_ssse3(\&body_20_39); 589 &Xtail_ssse3(\&body_20_39); 590 591$code.=<<___; 592 movups $iv,48($out,$in0) # write output 593 mov 88(%rsp),$ivp # restore $ivp 594 595 add 0($ctx),$A # update context 596 add 4($ctx),@T[0] 597 add 8($ctx),$C 598 mov $A,0($ctx) 599 add 12($ctx),$D 600 mov @T[0],4($ctx) 601 add 16($ctx),$E 602 mov $C,8($ctx) 603 mov $D,12($ctx) 604 mov $E,16($ctx) 605 movups $iv,($ivp) # write IV 606___ 607$code.=<<___ if ($win64); 608 movaps 96+0(%rsp),%xmm6 609 movaps 96+16(%rsp),%xmm7 610 movaps 96+32(%rsp),%xmm8 611 movaps 96+48(%rsp),%xmm9 612 movaps 96+64(%rsp),%xmm10 613 movaps 96+80(%rsp),%xmm11 614 movaps 96+96(%rsp),%xmm12 615 movaps 96+112(%rsp),%xmm13 616 movaps 96+128(%rsp),%xmm14 617 movaps 96+144(%rsp),%xmm15 618___ 619$code.=<<___; 620 lea `104+($win64?10*16:0)`(%rsp),%rsi 621 mov 0(%rsi),%r15 622 mov 8(%rsi),%r14 623 mov 16(%rsi),%r13 624 mov 24(%rsi),%r12 625 mov 32(%rsi),%rbp 626 mov 40(%rsi),%rbx 627 lea 48(%rsi),%rsp 628.Lepilogue_ssse3: 629 ret 630.size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3 631___ 632 633$j=$jj=$r=$sn=0; 634 635if ($avx) { 636my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10"); 637 638my $Xi=4; 639my @X=map("%xmm$_",(4..7,0..3)); 640my @Tx=map("%xmm$_",(8..10)); 641my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization 642my @T=("%esi","%edi"); 643 644my $_rol=sub { &shld(@_[0],@_) }; 645my $_ror=sub { &shrd(@_[0],@_) }; 646 647$code.=<<___; 648.type aesni_cbc_sha1_enc_avx,\@function,6 649.align 16 650aesni_cbc_sha1_enc_avx: 651 mov `($win64?56:8)`(%rsp),$inp # load 7th argument 652 #shr \$6,$len # debugging artefact 653 #jz .Lepilogue_avx # debugging artefact 654 push %rbx 655 push %rbp 656 push %r12 657 push %r13 658 push %r14 659 push %r15 660 lea `-104-($win64?10*16:0)`(%rsp),%rsp 661 #mov $in0,$inp # debugging artefact 662 #lea 64(%rsp),$ctx # debugging artefact 663___ 664$code.=<<___ if ($win64); 665 movaps %xmm6,96+0(%rsp) 666 movaps %xmm7,96+16(%rsp) 667 movaps %xmm8,96+32(%rsp) 668 movaps %xmm9,96+48(%rsp) 669 movaps %xmm10,96+64(%rsp) 670 movaps %xmm11,96+80(%rsp) 671 movaps %xmm12,96+96(%rsp) 672 movaps %xmm13,96+112(%rsp) 673 movaps %xmm14,96+128(%rsp) 674 movaps %xmm15,96+144(%rsp) 675.Lprologue_avx: 676___ 677$code.=<<___; 678 vzeroall 679 mov $in0,%r12 # reassign arguments 680 mov $out,%r13 681 mov $len,%r14 682 mov $key,%r15 683 vmovdqu ($ivp),$iv # load IV 684 mov $ivp,88(%rsp) # save $ivp 685___ 686my ($in0,$out,$len,$key)=map("%r$_",(12..15)); # reassign arguments 687my $rounds="${ivp}d"; 688$code.=<<___; 689 shl \$6,$len 690 sub $in0,$out 691 mov 240($key),$rounds 692 add \$112,$key # size optimization 693 add $inp,$len # end of input 694 695 lea K_XX_XX(%rip),$K_XX_XX 696 mov 0($ctx),$A # load context 697 mov 4($ctx),$B 698 mov 8($ctx),$C 699 mov 12($ctx),$D 700 mov $B,@T[0] # magic seed 701 mov 16($ctx),$E 702 703 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask 704 vmovdqa 0($K_XX_XX),@Tx[1] # K_00_19 705 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3] 706 vmovdqu 16($inp),@X[-3&7] 707 vmovdqu 32($inp),@X[-2&7] 708 vmovdqu 48($inp),@X[-1&7] 709 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap 710 add \$64,$inp 711 vpshufb @X[2],@X[-3&7],@X[-3&7] 712 vpshufb @X[2],@X[-2&7],@X[-2&7] 713 vpshufb @X[2],@X[-1&7],@X[-1&7] 714 vpaddd @Tx[1],@X[-4&7],@X[0] # add K_00_19 715 vpaddd @Tx[1],@X[-3&7],@X[1] 716 vpaddd @Tx[1],@X[-2&7],@X[2] 717 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU 718 vmovdqa @X[1],16(%rsp) 719 vmovdqa @X[2],32(%rsp) 720 vmovups -112($key),$rndkey0 # $key[0] 721 vmovups 16-112($key),$rndkey[0] # forward reference 722 jmp .Loop_avx 723___ 724 725my $aesenc=sub { 726 use integer; 727 my ($n,$k)=($r/10,$r%10); 728 if ($k==0) { 729 $code.=<<___; 730 vmovups `16*$n`($in0),$in # load input 731 vxorps $rndkey0,$in,$in 732___ 733 $code.=<<___ if ($n); 734 vmovups $iv,`16*($n-1)`($out,$in0) # write output 735___ 736 $code.=<<___; 737 vxorps $in,$iv,$iv 738 vaesenc $rndkey[0],$iv,$iv 739 vmovups `32+16*$k-112`($key),$rndkey[1] 740___ 741 } elsif ($k==9) { 742 $sn++; 743 $code.=<<___; 744 cmp \$11,$rounds 745 jb .Lvaesenclast$sn 746 vaesenc $rndkey[0],$iv,$iv 747 vmovups `32+16*($k+0)-112`($key),$rndkey[1] 748 vaesenc $rndkey[1],$iv,$iv 749 vmovups `32+16*($k+1)-112`($key),$rndkey[0] 750 je .Lvaesenclast$sn 751 vaesenc $rndkey[0],$iv,$iv 752 vmovups `32+16*($k+2)-112`($key),$rndkey[1] 753 vaesenc $rndkey[1],$iv,$iv 754 vmovups `32+16*($k+3)-112`($key),$rndkey[0] 755.Lvaesenclast$sn: 756 vaesenclast $rndkey[0],$iv,$iv 757 vmovups 16-112($key),$rndkey[1] # forward reference 758___ 759 } else { 760 $code.=<<___; 761 vaesenc $rndkey[0],$iv,$iv 762 vmovups `32+16*$k-112`($key),$rndkey[1] 763___ 764 } 765 $r++; unshift(@rndkey,pop(@rndkey)); 766}; 767 768sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4 769{ use integer; 770 my $body = shift; 771 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions 772 my ($a,$b,$c,$d,$e); 773 774 eval(shift(@insns)); 775 eval(shift(@insns)); 776 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]" 777 eval(shift(@insns)); 778 eval(shift(@insns)); 779 780 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]); 781 eval(shift(@insns)); 782 eval(shift(@insns)); 783 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords 784 eval(shift(@insns)); 785 eval(shift(@insns)); 786 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]" 787 eval(shift(@insns)); 788 eval(shift(@insns)); 789 790 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]" 791 eval(shift(@insns)); 792 eval(shift(@insns)); 793 eval(shift(@insns)); 794 eval(shift(@insns)); 795 796 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]" 797 eval(shift(@insns)); 798 eval(shift(@insns)); 799 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU 800 eval(shift(@insns)); 801 eval(shift(@insns)); 802 803 &vpsrld (@Tx[0],@X[0],31); 804 eval(shift(@insns)); 805 eval(shift(@insns)); 806 eval(shift(@insns)); 807 eval(shift(@insns)); 808 809 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword 810 &vpaddd (@X[0],@X[0],@X[0]); 811 eval(shift(@insns)); 812 eval(shift(@insns)); 813 eval(shift(@insns)); 814 eval(shift(@insns)); 815 816 &vpsrld (@Tx[1],@Tx[2],30); 817 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1 818 eval(shift(@insns)); 819 eval(shift(@insns)); 820 eval(shift(@insns)); 821 eval(shift(@insns)); 822 823 &vpslld (@Tx[2],@Tx[2],2); 824 &vpxor (@X[0],@X[0],@Tx[1]); 825 eval(shift(@insns)); 826 eval(shift(@insns)); 827 eval(shift(@insns)); 828 eval(shift(@insns)); 829 830 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2 831 eval(shift(@insns)); 832 eval(shift(@insns)); 833 &vmovdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX 834 eval(shift(@insns)); 835 eval(shift(@insns)); 836 837 838 foreach (@insns) { eval; } # remaining instructions [if any] 839 840 $Xi++; push(@X,shift(@X)); # "rotate" X[] 841 push(@Tx,shift(@Tx)); 842} 843 844sub Xupdate_avx_32_79() 845{ use integer; 846 my $body = shift; 847 my @insns = (&$body,&$body,&$body,&$body); # 32 to 48 instructions 848 my ($a,$b,$c,$d,$e); 849 850 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]" 851 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]" 852 eval(shift(@insns)); # body_20_39 853 eval(shift(@insns)); 854 eval(shift(@insns)); 855 eval(shift(@insns)); # rol 856 857 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]" 858 eval(shift(@insns)); 859 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/); 860 if ($Xi%5) { 861 &vmovdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX... 862 } else { # ... or load next one 863 &vmovdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)"); 864 } 865 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]); 866 eval(shift(@insns)); # ror 867 eval(shift(@insns)); 868 869 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]" 870 eval(shift(@insns)); # body_20_39 871 eval(shift(@insns)); 872 eval(shift(@insns)); 873 eval(shift(@insns)); # rol 874 875 &vpsrld (@Tx[0],@X[0],30); 876 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU 877 eval(shift(@insns)); 878 eval(shift(@insns)); 879 eval(shift(@insns)); # ror 880 eval(shift(@insns)); 881 882 &vpslld (@X[0],@X[0],2); 883 eval(shift(@insns)); # body_20_39 884 eval(shift(@insns)); 885 eval(shift(@insns)); 886 eval(shift(@insns)); # rol 887 eval(shift(@insns)); 888 eval(shift(@insns)); 889 eval(shift(@insns)); # ror 890 eval(shift(@insns)); 891 892 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2 893 eval(shift(@insns)); # body_20_39 894 eval(shift(@insns)); 895 &vmovdqa (@Tx[1],@X[0]) if ($Xi<19); 896 eval(shift(@insns)); 897 eval(shift(@insns)); # rol 898 eval(shift(@insns)); 899 eval(shift(@insns)); 900 eval(shift(@insns)); # rol 901 eval(shift(@insns)); 902 903 foreach (@insns) { eval; } # remaining instructions 904 905 $Xi++; push(@X,shift(@X)); # "rotate" X[] 906 push(@Tx,shift(@Tx)); 907} 908 909sub Xuplast_avx_80() 910{ use integer; 911 my $body = shift; 912 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 913 my ($a,$b,$c,$d,$e); 914 915 eval(shift(@insns)); 916 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]); 917 eval(shift(@insns)); 918 eval(shift(@insns)); 919 eval(shift(@insns)); 920 eval(shift(@insns)); 921 922 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU 923 924 foreach (@insns) { eval; } # remaining instructions 925 926 &cmp ($inp,$len); 927 &je (".Ldone_avx"); 928 929 unshift(@Tx,pop(@Tx)); 930 931 &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask 932 &vmovdqa(@Tx[1],"0($K_XX_XX)"); # K_00_19 933 &vmovdqu(@X[-4&7],"0($inp)"); # load input 934 &vmovdqu(@X[-3&7],"16($inp)"); 935 &vmovdqu(@X[-2&7],"32($inp)"); 936 &vmovdqu(@X[-1&7],"48($inp)"); 937 &vpshufb(@X[-4&7],@X[-4&7],@X[2]); # byte swap 938 &add ($inp,64); 939 940 $Xi=0; 941} 942 943sub Xloop_avx() 944{ use integer; 945 my $body = shift; 946 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 947 my ($a,$b,$c,$d,$e); 948 949 eval(shift(@insns)); 950 eval(shift(@insns)); 951 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]); 952 eval(shift(@insns)); 953 eval(shift(@insns)); 954 &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],@Tx[1]); 955 eval(shift(@insns)); 956 eval(shift(@insns)); 957 eval(shift(@insns)); 958 eval(shift(@insns)); 959 &vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]); # X[]+K xfer to IALU 960 eval(shift(@insns)); 961 eval(shift(@insns)); 962 963 foreach (@insns) { eval; } 964 $Xi++; 965} 966 967sub Xtail_avx() 968{ use integer; 969 my $body = shift; 970 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions 971 my ($a,$b,$c,$d,$e); 972 973 foreach (@insns) { eval; } 974} 975 976$code.=<<___; 977.align 16 978.Loop_avx: 979___ 980 &Xupdate_avx_16_31(\&body_00_19); 981 &Xupdate_avx_16_31(\&body_00_19); 982 &Xupdate_avx_16_31(\&body_00_19); 983 &Xupdate_avx_16_31(\&body_00_19); 984 &Xupdate_avx_32_79(\&body_00_19); 985 &Xupdate_avx_32_79(\&body_20_39); 986 &Xupdate_avx_32_79(\&body_20_39); 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_40_59); 991 &Xupdate_avx_32_79(\&body_40_59); 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_20_39); 996 &Xuplast_avx_80(\&body_20_39); # can jump to "done" 997 998 $saved_j=$j; @saved_V=@V; 999 $saved_r=$r; @saved_rndkey=@rndkey; 1000 1001 &Xloop_avx(\&body_20_39); 1002 &Xloop_avx(\&body_20_39); 1003 &Xloop_avx(\&body_20_39); 1004 1005$code.=<<___; 1006 vmovups $iv,48($out,$in0) # write output 1007 lea 64($in0),$in0 1008 1009 add 0($ctx),$A # update context 1010 add 4($ctx),@T[0] 1011 add 8($ctx),$C 1012 add 12($ctx),$D 1013 mov $A,0($ctx) 1014 add 16($ctx),$E 1015 mov @T[0],4($ctx) 1016 mov @T[0],$B # magic seed 1017 mov $C,8($ctx) 1018 mov $D,12($ctx) 1019 mov $E,16($ctx) 1020 jmp .Loop_avx 1021 1022.align 16 1023.Ldone_avx: 1024___ 1025 $jj=$j=$saved_j; @V=@saved_V; 1026 $r=$saved_r; @rndkey=@saved_rndkey; 1027 1028 &Xtail_avx(\&body_20_39); 1029 &Xtail_avx(\&body_20_39); 1030 &Xtail_avx(\&body_20_39); 1031 1032$code.=<<___; 1033 vmovups $iv,48($out,$in0) # write output 1034 mov 88(%rsp),$ivp # restore $ivp 1035 1036 add 0($ctx),$A # update context 1037 add 4($ctx),@T[0] 1038 add 8($ctx),$C 1039 mov $A,0($ctx) 1040 add 12($ctx),$D 1041 mov @T[0],4($ctx) 1042 add 16($ctx),$E 1043 mov $C,8($ctx) 1044 mov $D,12($ctx) 1045 mov $E,16($ctx) 1046 vmovups $iv,($ivp) # write IV 1047 vzeroall 1048___ 1049$code.=<<___ if ($win64); 1050 movaps 96+0(%rsp),%xmm6 1051 movaps 96+16(%rsp),%xmm7 1052 movaps 96+32(%rsp),%xmm8 1053 movaps 96+48(%rsp),%xmm9 1054 movaps 96+64(%rsp),%xmm10 1055 movaps 96+80(%rsp),%xmm11 1056 movaps 96+96(%rsp),%xmm12 1057 movaps 96+112(%rsp),%xmm13 1058 movaps 96+128(%rsp),%xmm14 1059 movaps 96+144(%rsp),%xmm15 1060___ 1061$code.=<<___; 1062 lea `104+($win64?10*16:0)`(%rsp),%rsi 1063 mov 0(%rsi),%r15 1064 mov 8(%rsi),%r14 1065 mov 16(%rsi),%r13 1066 mov 24(%rsi),%r12 1067 mov 32(%rsi),%rbp 1068 mov 40(%rsi),%rbx 1069 lea 48(%rsi),%rsp 1070.Lepilogue_avx: 1071 ret 1072.size aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx 1073___ 1074} 1075$code.=<<___; 1076.align 64 1077K_XX_XX: 1078.long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19 1079.long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39 1080.long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59 1081.long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79 1082.long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask 1083 1084.asciz "AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>" 1085.align 64 1086___ 1087 1088# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1089# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1090if ($win64) { 1091$rec="%rcx"; 1092$frame="%rdx"; 1093$context="%r8"; 1094$disp="%r9"; 1095 1096$code.=<<___; 1097.extern __imp_RtlVirtualUnwind 1098.type ssse3_handler,\@abi-omnipotent 1099.align 16 1100ssse3_handler: 1101 push %rsi 1102 push %rdi 1103 push %rbx 1104 push %rbp 1105 push %r12 1106 push %r13 1107 push %r14 1108 push %r15 1109 pushfq 1110 sub \$64,%rsp 1111 1112 mov 120($context),%rax # pull context->Rax 1113 mov 248($context),%rbx # pull context->Rip 1114 1115 mov 8($disp),%rsi # disp->ImageBase 1116 mov 56($disp),%r11 # disp->HandlerData 1117 1118 mov 0(%r11),%r10d # HandlerData[0] 1119 lea (%rsi,%r10),%r10 # prologue label 1120 cmp %r10,%rbx # context->Rip<prologue label 1121 jb .Lcommon_seh_tail 1122 1123 mov 152($context),%rax # pull context->Rsp 1124 1125 mov 4(%r11),%r10d # HandlerData[1] 1126 lea (%rsi,%r10),%r10 # epilogue label 1127 cmp %r10,%rbx # context->Rip>=epilogue label 1128 jae .Lcommon_seh_tail 1129 1130 lea 96(%rax),%rsi 1131 lea 512($context),%rdi # &context.Xmm6 1132 mov \$20,%ecx 1133 .long 0xa548f3fc # cld; rep movsq 1134 lea `104+10*16`(%rax),%rax # adjust stack pointer 1135 1136 mov 0(%rax),%r15 1137 mov 8(%rax),%r14 1138 mov 16(%rax),%r13 1139 mov 24(%rax),%r12 1140 mov 32(%rax),%rbp 1141 mov 40(%rax),%rbx 1142 lea 48(%rax),%rax 1143 mov %rbx,144($context) # restore context->Rbx 1144 mov %rbp,160($context) # restore context->Rbp 1145 mov %r12,216($context) # restore context->R12 1146 mov %r13,224($context) # restore context->R13 1147 mov %r14,232($context) # restore context->R14 1148 mov %r15,240($context) # restore context->R15 1149 1150.Lcommon_seh_tail: 1151 mov 8(%rax),%rdi 1152 mov 16(%rax),%rsi 1153 mov %rax,152($context) # restore context->Rsp 1154 mov %rsi,168($context) # restore context->Rsi 1155 mov %rdi,176($context) # restore context->Rdi 1156 1157 mov 40($disp),%rdi # disp->ContextRecord 1158 mov $context,%rsi # context 1159 mov \$154,%ecx # sizeof(CONTEXT) 1160 .long 0xa548f3fc # cld; rep movsq 1161 1162 mov $disp,%rsi 1163 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER 1164 mov 8(%rsi),%rdx # arg2, disp->ImageBase 1165 mov 0(%rsi),%r8 # arg3, disp->ControlPc 1166 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry 1167 mov 40(%rsi),%r10 # disp->ContextRecord 1168 lea 56(%rsi),%r11 # &disp->HandlerData 1169 lea 24(%rsi),%r12 # &disp->EstablisherFrame 1170 mov %r10,32(%rsp) # arg5 1171 mov %r11,40(%rsp) # arg6 1172 mov %r12,48(%rsp) # arg7 1173 mov %rcx,56(%rsp) # arg8, (NULL) 1174 call *__imp_RtlVirtualUnwind(%rip) 1175 1176 mov \$1,%eax # ExceptionContinueSearch 1177 add \$64,%rsp 1178 popfq 1179 pop %r15 1180 pop %r14 1181 pop %r13 1182 pop %r12 1183 pop %rbp 1184 pop %rbx 1185 pop %rdi 1186 pop %rsi 1187 ret 1188.size ssse3_handler,.-ssse3_handler 1189 1190.section .pdata 1191.align 4 1192 .rva .LSEH_begin_aesni_cbc_sha1_enc_ssse3 1193 .rva .LSEH_end_aesni_cbc_sha1_enc_ssse3 1194 .rva .LSEH_info_aesni_cbc_sha1_enc_ssse3 1195___ 1196$code.=<<___ if ($avx); 1197 .rva .LSEH_begin_aesni_cbc_sha1_enc_avx 1198 .rva .LSEH_end_aesni_cbc_sha1_enc_avx 1199 .rva .LSEH_info_aesni_cbc_sha1_enc_avx 1200___ 1201$code.=<<___; 1202.section .xdata 1203.align 8 1204.LSEH_info_aesni_cbc_sha1_enc_ssse3: 1205 .byte 9,0,0,0 1206 .rva ssse3_handler 1207 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[] 1208___ 1209$code.=<<___ if ($avx); 1210.LSEH_info_aesni_cbc_sha1_enc_avx: 1211 .byte 9,0,0,0 1212 .rva ssse3_handler 1213 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[] 1214___ 1215} 1216 1217#################################################################### 1218sub rex { 1219 local *opcode=shift; 1220 my ($dst,$src)=@_; 1221 my $rex=0; 1222 1223 $rex|=0x04 if($dst>=8); 1224 $rex|=0x01 if($src>=8); 1225 push @opcode,$rex|0x40 if($rex); 1226} 1227 1228sub aesni { 1229 my $line=shift; 1230 my @opcode=(0x66); 1231 1232 if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) { 1233 my %opcodelet = ( 1234 "aesenc" => 0xdc, "aesenclast" => 0xdd 1235 ); 1236 return undef if (!defined($opcodelet{$1})); 1237 rex(\@opcode,$3,$2); 1238 push @opcode,0x0f,0x38,$opcodelet{$1}; 1239 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 1240 return ".byte\t".join(',',@opcode); 1241 } 1242 return $line; 1243} 1244 1245$code =~ s/\`([^\`]*)\`/eval($1)/gem; 1246$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem; 1247 1248print $code; 1249close STDOUT; 1250