1#! /usr/bin/env perl 2# Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved. 3# 4# Licensed under the OpenSSL license (the "License"). You may not use 5# this file except in compliance with the License. You can obtain a copy 6# in the file LICENSE in the source distribution or at 7# https://www.openssl.org/source/license.html 8 9 10# ==================================================================== 11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL 12# project. The module is, however, dual licensed under OpenSSL and 13# CRYPTOGAMS licenses depending on where you obtain it. For further 14# details see http://www.openssl.org/~appro/cryptogams/. 15# ==================================================================== 16 17# October 2005. 18# 19# Montgomery multiplication routine for x86_64. While it gives modest 20# 9% improvement of rsa4096 sign on Opteron, rsa512 sign runs more 21# than twice, >2x, as fast. Most common rsa1024 sign is improved by 22# respectful 50%. It remains to be seen if loop unrolling and 23# dedicated squaring routine can provide further improvement... 24 25# July 2011. 26# 27# Add dedicated squaring procedure. Performance improvement varies 28# from platform to platform, but in average it's ~5%/15%/25%/33% 29# for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively. 30 31# August 2011. 32# 33# Unroll and modulo-schedule inner loops in such manner that they 34# are "fallen through" for input lengths of 8, which is critical for 35# 1024-bit RSA *sign*. Average performance improvement in comparison 36# to *initial* version of this module from 2005 is ~0%/30%/40%/45% 37# for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively. 38 39# June 2013. 40# 41# Optimize reduction in squaring procedure and improve 1024+-bit RSA 42# sign performance by 10-16% on Intel Sandy Bridge and later 43# (virtually same on non-Intel processors). 44 45# August 2013. 46# 47# Add MULX/ADOX/ADCX code path. 48 49$flavour = shift; 50$output = shift; 51if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 52 53$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); 54 55$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 56( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or 57( $xlate="${dir}../../../perlasm/x86_64-xlate.pl" and -f $xlate) or 58die "can't locate x86_64-xlate.pl"; 59 60open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; 61*STDOUT=*OUT; 62 63# In upstream, this is controlled by shelling out to the compiler to check 64# versions, but BoringSSL is intended to be used with pre-generated perlasm 65# output, so this isn't useful anyway. 66$addx = 1; 67 68# int bn_mul_mont( 69$rp="%rdi"; # BN_ULONG *rp, 70$ap="%rsi"; # const BN_ULONG *ap, 71$bp="%rdx"; # const BN_ULONG *bp, 72$np="%rcx"; # const BN_ULONG *np, 73$n0="%r8"; # const BN_ULONG *n0, 74# TODO(davidben): The code below treats $num as an int, but C passes in a 75# size_t. 76$num="%r9"; # size_t num); 77$lo0="%r10"; 78$hi0="%r11"; 79$hi1="%r13"; 80$i="%r14"; 81$j="%r15"; 82$m0="%rbx"; 83$m1="%rbp"; 84 85$code=<<___; 86.text 87 88.extern OPENSSL_ia32cap_P 89 90.globl bn_mul_mont 91.type bn_mul_mont,\@function,6 92.align 16 93bn_mul_mont: 94.cfi_startproc 95 mov ${num}d,${num}d 96 mov %rsp,%rax 97.cfi_def_cfa_register %rax 98 test \$3,${num}d 99 jnz .Lmul_enter 100 cmp \$8,${num}d 101 jb .Lmul_enter 102___ 103$code.=<<___ if ($addx); 104 leaq OPENSSL_ia32cap_P(%rip),%r11 105 mov 8(%r11),%r11d 106___ 107$code.=<<___; 108 cmp $ap,$bp 109 jne .Lmul4x_enter 110 test \$7,${num}d 111 jz .Lsqr8x_enter 112 jmp .Lmul4x_enter 113 114.align 16 115.Lmul_enter: 116 push %rbx 117.cfi_push %rbx 118 push %rbp 119.cfi_push %rbp 120 push %r12 121.cfi_push %r12 122 push %r13 123.cfi_push %r13 124 push %r14 125.cfi_push %r14 126 push %r15 127.cfi_push %r15 128 129 neg $num 130 mov %rsp,%r11 131 lea -16(%rsp,$num,8),%r10 # future alloca(8*(num+2)) 132 neg $num # restore $num 133 and \$-1024,%r10 # minimize TLB usage 134 135 # An OS-agnostic version of __chkstk. 136 # 137 # Some OSes (Windows) insist on stack being "wired" to 138 # physical memory in strictly sequential manner, i.e. if stack 139 # allocation spans two pages, then reference to farmost one can 140 # be punishable by SEGV. But page walking can do good even on 141 # other OSes, because it guarantees that villain thread hits 142 # the guard page before it can make damage to innocent one... 143 sub %r10,%r11 144 and \$-4096,%r11 145 lea (%r10,%r11),%rsp 146 mov (%rsp),%r11 147 cmp %r10,%rsp 148 ja .Lmul_page_walk 149 jmp .Lmul_page_walk_done 150 151.align 16 152.Lmul_page_walk: 153 lea -4096(%rsp),%rsp 154 mov (%rsp),%r11 155 cmp %r10,%rsp 156 ja .Lmul_page_walk 157.Lmul_page_walk_done: 158 159 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp 160.cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8 161.Lmul_body: 162 mov $bp,%r12 # reassign $bp 163___ 164 $bp="%r12"; 165$code.=<<___; 166 mov ($n0),$n0 # pull n0[0] value 167 mov ($bp),$m0 # m0=bp[0] 168 mov ($ap),%rax 169 170 xor $i,$i # i=0 171 xor $j,$j # j=0 172 173 mov $n0,$m1 174 mulq $m0 # ap[0]*bp[0] 175 mov %rax,$lo0 176 mov ($np),%rax 177 178 imulq $lo0,$m1 # "tp[0]"*n0 179 mov %rdx,$hi0 180 181 mulq $m1 # np[0]*m1 182 add %rax,$lo0 # discarded 183 mov 8($ap),%rax 184 adc \$0,%rdx 185 mov %rdx,$hi1 186 187 lea 1($j),$j # j++ 188 jmp .L1st_enter 189 190.align 16 191.L1st: 192 add %rax,$hi1 193 mov ($ap,$j,8),%rax 194 adc \$0,%rdx 195 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0] 196 mov $lo0,$hi0 197 adc \$0,%rdx 198 mov $hi1,-16(%rsp,$j,8) # tp[j-1] 199 mov %rdx,$hi1 200 201.L1st_enter: 202 mulq $m0 # ap[j]*bp[0] 203 add %rax,$hi0 204 mov ($np,$j,8),%rax 205 adc \$0,%rdx 206 lea 1($j),$j # j++ 207 mov %rdx,$lo0 208 209 mulq $m1 # np[j]*m1 210 cmp $num,$j 211 jne .L1st 212 213 add %rax,$hi1 214 mov ($ap),%rax # ap[0] 215 adc \$0,%rdx 216 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0] 217 adc \$0,%rdx 218 mov $hi1,-16(%rsp,$j,8) # tp[j-1] 219 mov %rdx,$hi1 220 mov $lo0,$hi0 221 222 xor %rdx,%rdx 223 add $hi0,$hi1 224 adc \$0,%rdx 225 mov $hi1,-8(%rsp,$num,8) 226 mov %rdx,(%rsp,$num,8) # store upmost overflow bit 227 228 lea 1($i),$i # i++ 229 jmp .Louter 230.align 16 231.Louter: 232 mov ($bp,$i,8),$m0 # m0=bp[i] 233 xor $j,$j # j=0 234 mov $n0,$m1 235 mov (%rsp),$lo0 236 mulq $m0 # ap[0]*bp[i] 237 add %rax,$lo0 # ap[0]*bp[i]+tp[0] 238 mov ($np),%rax 239 adc \$0,%rdx 240 241 imulq $lo0,$m1 # tp[0]*n0 242 mov %rdx,$hi0 243 244 mulq $m1 # np[0]*m1 245 add %rax,$lo0 # discarded 246 mov 8($ap),%rax 247 adc \$0,%rdx 248 mov 8(%rsp),$lo0 # tp[1] 249 mov %rdx,$hi1 250 251 lea 1($j),$j # j++ 252 jmp .Linner_enter 253 254.align 16 255.Linner: 256 add %rax,$hi1 257 mov ($ap,$j,8),%rax 258 adc \$0,%rdx 259 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j] 260 mov (%rsp,$j,8),$lo0 261 adc \$0,%rdx 262 mov $hi1,-16(%rsp,$j,8) # tp[j-1] 263 mov %rdx,$hi1 264 265.Linner_enter: 266 mulq $m0 # ap[j]*bp[i] 267 add %rax,$hi0 268 mov ($np,$j,8),%rax 269 adc \$0,%rdx 270 add $hi0,$lo0 # ap[j]*bp[i]+tp[j] 271 mov %rdx,$hi0 272 adc \$0,$hi0 273 lea 1($j),$j # j++ 274 275 mulq $m1 # np[j]*m1 276 cmp $num,$j 277 jne .Linner 278 279 add %rax,$hi1 280 mov ($ap),%rax # ap[0] 281 adc \$0,%rdx 282 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j] 283 mov (%rsp,$j,8),$lo0 284 adc \$0,%rdx 285 mov $hi1,-16(%rsp,$j,8) # tp[j-1] 286 mov %rdx,$hi1 287 288 xor %rdx,%rdx 289 add $hi0,$hi1 290 adc \$0,%rdx 291 add $lo0,$hi1 # pull upmost overflow bit 292 adc \$0,%rdx 293 mov $hi1,-8(%rsp,$num,8) 294 mov %rdx,(%rsp,$num,8) # store upmost overflow bit 295 296 lea 1($i),$i # i++ 297 cmp $num,$i 298 jb .Louter 299 300 xor $i,$i # i=0 and clear CF! 301 mov (%rsp),%rax # tp[0] 302 mov $num,$j # j=num 303 304.align 16 305.Lsub: sbb ($np,$i,8),%rax 306 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i] 307 mov 8(%rsp,$i,8),%rax # tp[i+1] 308 lea 1($i),$i # i++ 309 dec $j # doesn't affect CF! 310 jnz .Lsub 311 312 sbb \$0,%rax # handle upmost overflow bit 313 mov \$-1,%rbx 314 xor %rax,%rbx # not %rax 315 xor $i,$i 316 mov $num,$j # j=num 317 318.Lcopy: # conditional copy 319 mov ($rp,$i,8),%rcx 320 mov (%rsp,$i,8),%rdx 321 and %rbx,%rcx 322 and %rax,%rdx 323 mov $num,(%rsp,$i,8) # zap temporary vector 324 or %rcx,%rdx 325 mov %rdx,($rp,$i,8) # rp[i]=tp[i] 326 lea 1($i),$i 327 sub \$1,$j 328 jnz .Lcopy 329 330 mov 8(%rsp,$num,8),%rsi # restore %rsp 331.cfi_def_cfa %rsi,8 332 mov \$1,%rax 333 mov -48(%rsi),%r15 334.cfi_restore %r15 335 mov -40(%rsi),%r14 336.cfi_restore %r14 337 mov -32(%rsi),%r13 338.cfi_restore %r13 339 mov -24(%rsi),%r12 340.cfi_restore %r12 341 mov -16(%rsi),%rbp 342.cfi_restore %rbp 343 mov -8(%rsi),%rbx 344.cfi_restore %rbx 345 lea (%rsi),%rsp 346.cfi_def_cfa_register %rsp 347.Lmul_epilogue: 348 ret 349.cfi_endproc 350.size bn_mul_mont,.-bn_mul_mont 351___ 352{{{ 353my @A=("%r10","%r11"); 354my @N=("%r13","%rdi"); 355$code.=<<___; 356.type bn_mul4x_mont,\@function,6 357.align 16 358bn_mul4x_mont: 359.cfi_startproc 360 mov ${num}d,${num}d 361 mov %rsp,%rax 362.cfi_def_cfa_register %rax 363.Lmul4x_enter: 364___ 365$code.=<<___ if ($addx); 366 and \$0x80100,%r11d 367 cmp \$0x80100,%r11d 368 je .Lmulx4x_enter 369___ 370$code.=<<___; 371 push %rbx 372.cfi_push %rbx 373 push %rbp 374.cfi_push %rbp 375 push %r12 376.cfi_push %r12 377 push %r13 378.cfi_push %r13 379 push %r14 380.cfi_push %r14 381 push %r15 382.cfi_push %r15 383 384 neg $num 385 mov %rsp,%r11 386 lea -32(%rsp,$num,8),%r10 # future alloca(8*(num+4)) 387 neg $num # restore 388 and \$-1024,%r10 # minimize TLB usage 389 390 sub %r10,%r11 391 and \$-4096,%r11 392 lea (%r10,%r11),%rsp 393 mov (%rsp),%r11 394 cmp %r10,%rsp 395 ja .Lmul4x_page_walk 396 jmp .Lmul4x_page_walk_done 397 398.Lmul4x_page_walk: 399 lea -4096(%rsp),%rsp 400 mov (%rsp),%r11 401 cmp %r10,%rsp 402 ja .Lmul4x_page_walk 403.Lmul4x_page_walk_done: 404 405 mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp 406.cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8 407.Lmul4x_body: 408 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp 409 mov %rdx,%r12 # reassign $bp 410___ 411 $bp="%r12"; 412$code.=<<___; 413 mov ($n0),$n0 # pull n0[0] value 414 mov ($bp),$m0 # m0=bp[0] 415 mov ($ap),%rax 416 417 xor $i,$i # i=0 418 xor $j,$j # j=0 419 420 mov $n0,$m1 421 mulq $m0 # ap[0]*bp[0] 422 mov %rax,$A[0] 423 mov ($np),%rax 424 425 imulq $A[0],$m1 # "tp[0]"*n0 426 mov %rdx,$A[1] 427 428 mulq $m1 # np[0]*m1 429 add %rax,$A[0] # discarded 430 mov 8($ap),%rax 431 adc \$0,%rdx 432 mov %rdx,$N[1] 433 434 mulq $m0 435 add %rax,$A[1] 436 mov 8($np),%rax 437 adc \$0,%rdx 438 mov %rdx,$A[0] 439 440 mulq $m1 441 add %rax,$N[1] 442 mov 16($ap),%rax 443 adc \$0,%rdx 444 add $A[1],$N[1] 445 lea 4($j),$j # j++ 446 adc \$0,%rdx 447 mov $N[1],(%rsp) 448 mov %rdx,$N[0] 449 jmp .L1st4x 450.align 16 451.L1st4x: 452 mulq $m0 # ap[j]*bp[0] 453 add %rax,$A[0] 454 mov -16($np,$j,8),%rax 455 adc \$0,%rdx 456 mov %rdx,$A[1] 457 458 mulq $m1 # np[j]*m1 459 add %rax,$N[0] 460 mov -8($ap,$j,8),%rax 461 adc \$0,%rdx 462 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0] 463 adc \$0,%rdx 464 mov $N[0],-24(%rsp,$j,8) # tp[j-1] 465 mov %rdx,$N[1] 466 467 mulq $m0 # ap[j]*bp[0] 468 add %rax,$A[1] 469 mov -8($np,$j,8),%rax 470 adc \$0,%rdx 471 mov %rdx,$A[0] 472 473 mulq $m1 # np[j]*m1 474 add %rax,$N[1] 475 mov ($ap,$j,8),%rax 476 adc \$0,%rdx 477 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0] 478 adc \$0,%rdx 479 mov $N[1],-16(%rsp,$j,8) # tp[j-1] 480 mov %rdx,$N[0] 481 482 mulq $m0 # ap[j]*bp[0] 483 add %rax,$A[0] 484 mov ($np,$j,8),%rax 485 adc \$0,%rdx 486 mov %rdx,$A[1] 487 488 mulq $m1 # np[j]*m1 489 add %rax,$N[0] 490 mov 8($ap,$j,8),%rax 491 adc \$0,%rdx 492 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0] 493 adc \$0,%rdx 494 mov $N[0],-8(%rsp,$j,8) # tp[j-1] 495 mov %rdx,$N[1] 496 497 mulq $m0 # ap[j]*bp[0] 498 add %rax,$A[1] 499 mov 8($np,$j,8),%rax 500 adc \$0,%rdx 501 lea 4($j),$j # j++ 502 mov %rdx,$A[0] 503 504 mulq $m1 # np[j]*m1 505 add %rax,$N[1] 506 mov -16($ap,$j,8),%rax 507 adc \$0,%rdx 508 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0] 509 adc \$0,%rdx 510 mov $N[1],-32(%rsp,$j,8) # tp[j-1] 511 mov %rdx,$N[0] 512 cmp $num,$j 513 jb .L1st4x 514 515 mulq $m0 # ap[j]*bp[0] 516 add %rax,$A[0] 517 mov -16($np,$j,8),%rax 518 adc \$0,%rdx 519 mov %rdx,$A[1] 520 521 mulq $m1 # np[j]*m1 522 add %rax,$N[0] 523 mov -8($ap,$j,8),%rax 524 adc \$0,%rdx 525 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0] 526 adc \$0,%rdx 527 mov $N[0],-24(%rsp,$j,8) # tp[j-1] 528 mov %rdx,$N[1] 529 530 mulq $m0 # ap[j]*bp[0] 531 add %rax,$A[1] 532 mov -8($np,$j,8),%rax 533 adc \$0,%rdx 534 mov %rdx,$A[0] 535 536 mulq $m1 # np[j]*m1 537 add %rax,$N[1] 538 mov ($ap),%rax # ap[0] 539 adc \$0,%rdx 540 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0] 541 adc \$0,%rdx 542 mov $N[1],-16(%rsp,$j,8) # tp[j-1] 543 mov %rdx,$N[0] 544 545 xor $N[1],$N[1] 546 add $A[0],$N[0] 547 adc \$0,$N[1] 548 mov $N[0],-8(%rsp,$j,8) 549 mov $N[1],(%rsp,$j,8) # store upmost overflow bit 550 551 lea 1($i),$i # i++ 552.align 4 553.Louter4x: 554 mov ($bp,$i,8),$m0 # m0=bp[i] 555 xor $j,$j # j=0 556 mov (%rsp),$A[0] 557 mov $n0,$m1 558 mulq $m0 # ap[0]*bp[i] 559 add %rax,$A[0] # ap[0]*bp[i]+tp[0] 560 mov ($np),%rax 561 adc \$0,%rdx 562 563 imulq $A[0],$m1 # tp[0]*n0 564 mov %rdx,$A[1] 565 566 mulq $m1 # np[0]*m1 567 add %rax,$A[0] # "$N[0]", discarded 568 mov 8($ap),%rax 569 adc \$0,%rdx 570 mov %rdx,$N[1] 571 572 mulq $m0 # ap[j]*bp[i] 573 add %rax,$A[1] 574 mov 8($np),%rax 575 adc \$0,%rdx 576 add 8(%rsp),$A[1] # +tp[1] 577 adc \$0,%rdx 578 mov %rdx,$A[0] 579 580 mulq $m1 # np[j]*m1 581 add %rax,$N[1] 582 mov 16($ap),%rax 583 adc \$0,%rdx 584 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j] 585 lea 4($j),$j # j+=2 586 adc \$0,%rdx 587 mov $N[1],(%rsp) # tp[j-1] 588 mov %rdx,$N[0] 589 jmp .Linner4x 590.align 16 591.Linner4x: 592 mulq $m0 # ap[j]*bp[i] 593 add %rax,$A[0] 594 mov -16($np,$j,8),%rax 595 adc \$0,%rdx 596 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j] 597 adc \$0,%rdx 598 mov %rdx,$A[1] 599 600 mulq $m1 # np[j]*m1 601 add %rax,$N[0] 602 mov -8($ap,$j,8),%rax 603 adc \$0,%rdx 604 add $A[0],$N[0] 605 adc \$0,%rdx 606 mov $N[0],-24(%rsp,$j,8) # tp[j-1] 607 mov %rdx,$N[1] 608 609 mulq $m0 # ap[j]*bp[i] 610 add %rax,$A[1] 611 mov -8($np,$j,8),%rax 612 adc \$0,%rdx 613 add -8(%rsp,$j,8),$A[1] 614 adc \$0,%rdx 615 mov %rdx,$A[0] 616 617 mulq $m1 # np[j]*m1 618 add %rax,$N[1] 619 mov ($ap,$j,8),%rax 620 adc \$0,%rdx 621 add $A[1],$N[1] 622 adc \$0,%rdx 623 mov $N[1],-16(%rsp,$j,8) # tp[j-1] 624 mov %rdx,$N[0] 625 626 mulq $m0 # ap[j]*bp[i] 627 add %rax,$A[0] 628 mov ($np,$j,8),%rax 629 adc \$0,%rdx 630 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j] 631 adc \$0,%rdx 632 mov %rdx,$A[1] 633 634 mulq $m1 # np[j]*m1 635 add %rax,$N[0] 636 mov 8($ap,$j,8),%rax 637 adc \$0,%rdx 638 add $A[0],$N[0] 639 adc \$0,%rdx 640 mov $N[0],-8(%rsp,$j,8) # tp[j-1] 641 mov %rdx,$N[1] 642 643 mulq $m0 # ap[j]*bp[i] 644 add %rax,$A[1] 645 mov 8($np,$j,8),%rax 646 adc \$0,%rdx 647 add 8(%rsp,$j,8),$A[1] 648 adc \$0,%rdx 649 lea 4($j),$j # j++ 650 mov %rdx,$A[0] 651 652 mulq $m1 # np[j]*m1 653 add %rax,$N[1] 654 mov -16($ap,$j,8),%rax 655 adc \$0,%rdx 656 add $A[1],$N[1] 657 adc \$0,%rdx 658 mov $N[1],-32(%rsp,$j,8) # tp[j-1] 659 mov %rdx,$N[0] 660 cmp $num,$j 661 jb .Linner4x 662 663 mulq $m0 # ap[j]*bp[i] 664 add %rax,$A[0] 665 mov -16($np,$j,8),%rax 666 adc \$0,%rdx 667 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j] 668 adc \$0,%rdx 669 mov %rdx,$A[1] 670 671 mulq $m1 # np[j]*m1 672 add %rax,$N[0] 673 mov -8($ap,$j,8),%rax 674 adc \$0,%rdx 675 add $A[0],$N[0] 676 adc \$0,%rdx 677 mov $N[0],-24(%rsp,$j,8) # tp[j-1] 678 mov %rdx,$N[1] 679 680 mulq $m0 # ap[j]*bp[i] 681 add %rax,$A[1] 682 mov -8($np,$j,8),%rax 683 adc \$0,%rdx 684 add -8(%rsp,$j,8),$A[1] 685 adc \$0,%rdx 686 lea 1($i),$i # i++ 687 mov %rdx,$A[0] 688 689 mulq $m1 # np[j]*m1 690 add %rax,$N[1] 691 mov ($ap),%rax # ap[0] 692 adc \$0,%rdx 693 add $A[1],$N[1] 694 adc \$0,%rdx 695 mov $N[1],-16(%rsp,$j,8) # tp[j-1] 696 mov %rdx,$N[0] 697 698 xor $N[1],$N[1] 699 add $A[0],$N[0] 700 adc \$0,$N[1] 701 add (%rsp,$num,8),$N[0] # pull upmost overflow bit 702 adc \$0,$N[1] 703 mov $N[0],-8(%rsp,$j,8) 704 mov $N[1],(%rsp,$j,8) # store upmost overflow bit 705 706 cmp $num,$i 707 jb .Louter4x 708___ 709{ 710my @ri=("%rax","%rdx",$m0,$m1); 711$code.=<<___; 712 mov 16(%rsp,$num,8),$rp # restore $rp 713 lea -4($num),$j 714 mov 0(%rsp),@ri[0] # tp[0] 715 mov 8(%rsp),@ri[1] # tp[1] 716 shr \$2,$j # j=num/4-1 717 lea (%rsp),$ap # borrow ap for tp 718 xor $i,$i # i=0 and clear CF! 719 720 sub 0($np),@ri[0] 721 mov 16($ap),@ri[2] # tp[2] 722 mov 24($ap),@ri[3] # tp[3] 723 sbb 8($np),@ri[1] 724 725.Lsub4x: 726 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i] 727 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i] 728 sbb 16($np,$i,8),@ri[2] 729 mov 32($ap,$i,8),@ri[0] # tp[i+1] 730 mov 40($ap,$i,8),@ri[1] 731 sbb 24($np,$i,8),@ri[3] 732 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i] 733 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i] 734 sbb 32($np,$i,8),@ri[0] 735 mov 48($ap,$i,8),@ri[2] 736 mov 56($ap,$i,8),@ri[3] 737 sbb 40($np,$i,8),@ri[1] 738 lea 4($i),$i # i++ 739 dec $j # doesn't affect CF! 740 jnz .Lsub4x 741 742 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i] 743 mov 32($ap,$i,8),@ri[0] # load overflow bit 744 sbb 16($np,$i,8),@ri[2] 745 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i] 746 sbb 24($np,$i,8),@ri[3] 747 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i] 748 749 sbb \$0,@ri[0] # handle upmost overflow bit 750 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i] 751 pxor %xmm0,%xmm0 752 movq @ri[0],%xmm4 753 pcmpeqd %xmm5,%xmm5 754 pshufd \$0,%xmm4,%xmm4 755 mov $num,$j 756 pxor %xmm4,%xmm5 757 shr \$2,$j # j=num/4 758 xor %eax,%eax # i=0 759 760 jmp .Lcopy4x 761.align 16 762.Lcopy4x: # conditional copy 763 movdqa (%rsp,%rax),%xmm1 764 movdqu ($rp,%rax),%xmm2 765 pand %xmm4,%xmm1 766 pand %xmm5,%xmm2 767 movdqa 16(%rsp,%rax),%xmm3 768 movdqa %xmm0,(%rsp,%rax) 769 por %xmm2,%xmm1 770 movdqu 16($rp,%rax),%xmm2 771 movdqu %xmm1,($rp,%rax) 772 pand %xmm4,%xmm3 773 pand %xmm5,%xmm2 774 movdqa %xmm0,16(%rsp,%rax) 775 por %xmm2,%xmm3 776 movdqu %xmm3,16($rp,%rax) 777 lea 32(%rax),%rax 778 dec $j 779 jnz .Lcopy4x 780___ 781} 782$code.=<<___; 783 mov 8(%rsp,$num,8),%rsi # restore %rsp 784.cfi_def_cfa %rsi, 8 785 mov \$1,%rax 786 mov -48(%rsi),%r15 787.cfi_restore %r15 788 mov -40(%rsi),%r14 789.cfi_restore %r14 790 mov -32(%rsi),%r13 791.cfi_restore %r13 792 mov -24(%rsi),%r12 793.cfi_restore %r12 794 mov -16(%rsi),%rbp 795.cfi_restore %rbp 796 mov -8(%rsi),%rbx 797.cfi_restore %rbx 798 lea (%rsi),%rsp 799.cfi_def_cfa_register %rsp 800.Lmul4x_epilogue: 801 ret 802.cfi_endproc 803.size bn_mul4x_mont,.-bn_mul4x_mont 804___ 805}}} 806 {{{ 807###################################################################### 808# void bn_sqr8x_mont( 809my $rptr="%rdi"; # const BN_ULONG *rptr, 810my $aptr="%rsi"; # const BN_ULONG *aptr, 811my $bptr="%rdx"; # not used 812my $nptr="%rcx"; # const BN_ULONG *nptr, 813my $n0 ="%r8"; # const BN_ULONG *n0); 814my $num ="%r9"; # int num, has to be divisible by 8 815 816my ($i,$j,$tptr)=("%rbp","%rcx",$rptr); 817my @A0=("%r10","%r11"); 818my @A1=("%r12","%r13"); 819my ($a0,$a1,$ai)=("%r14","%r15","%rbx"); 820 821$code.=<<___ if ($addx); 822.extern bn_sqrx8x_internal # see x86_64-mont5 module 823___ 824$code.=<<___; 825.extern bn_sqr8x_internal # see x86_64-mont5 module 826 827.type bn_sqr8x_mont,\@function,6 828.align 32 829bn_sqr8x_mont: 830.cfi_startproc 831 mov %rsp,%rax 832.cfi_def_cfa_register %rax 833.Lsqr8x_enter: 834 push %rbx 835.cfi_push %rbx 836 push %rbp 837.cfi_push %rbp 838 push %r12 839.cfi_push %r12 840 push %r13 841.cfi_push %r13 842 push %r14 843.cfi_push %r14 844 push %r15 845.cfi_push %r15 846.Lsqr8x_prologue: 847 848 mov ${num}d,%r10d 849 shl \$3,${num}d # convert $num to bytes 850 shl \$3+2,%r10 # 4*$num 851 neg $num 852 853 ############################################################## 854 # ensure that stack frame doesn't alias with $aptr modulo 855 # 4096. this is done to allow memory disambiguation logic 856 # do its job. 857 # 858 lea -64(%rsp,$num,2),%r11 859 mov %rsp,%rbp 860 mov ($n0),$n0 # *n0 861 sub $aptr,%r11 862 and \$4095,%r11 863 cmp %r11,%r10 864 jb .Lsqr8x_sp_alt 865 sub %r11,%rbp # align with $aptr 866 lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num) 867 jmp .Lsqr8x_sp_done 868 869.align 32 870.Lsqr8x_sp_alt: 871 lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num 872 lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num) 873 sub %r10,%r11 874 mov \$0,%r10 875 cmovc %r10,%r11 876 sub %r11,%rbp 877.Lsqr8x_sp_done: 878 and \$-64,%rbp 879 mov %rsp,%r11 880 sub %rbp,%r11 881 and \$-4096,%r11 882 lea (%rbp,%r11),%rsp 883 mov (%rsp),%r10 884 cmp %rbp,%rsp 885 ja .Lsqr8x_page_walk 886 jmp .Lsqr8x_page_walk_done 887 888.align 16 889.Lsqr8x_page_walk: 890 lea -4096(%rsp),%rsp 891 mov (%rsp),%r10 892 cmp %rbp,%rsp 893 ja .Lsqr8x_page_walk 894.Lsqr8x_page_walk_done: 895 896 mov $num,%r10 897 neg $num 898 899 mov $n0, 32(%rsp) 900 mov %rax, 40(%rsp) # save original %rsp 901.cfi_cfa_expression %rsp+40,deref,+8 902.Lsqr8x_body: 903 904 movq $nptr, %xmm2 # save pointer to modulus 905 pxor %xmm0,%xmm0 906 movq $rptr,%xmm1 # save $rptr 907 movq %r10, %xmm3 # -$num 908___ 909$code.=<<___ if ($addx); 910 leaq OPENSSL_ia32cap_P(%rip),%rax 911 mov 8(%rax),%eax 912 and \$0x80100,%eax 913 cmp \$0x80100,%eax 914 jne .Lsqr8x_nox 915 916 call bn_sqrx8x_internal # see x86_64-mont5 module 917 # %rax top-most carry 918 # %rbp nptr 919 # %rcx -8*num 920 # %r8 end of tp[2*num] 921 lea (%r8,%rcx),%rbx 922 mov %rcx,$num 923 mov %rcx,%rdx 924 movq %xmm1,$rptr 925 sar \$3+2,%rcx # %cf=0 926 jmp .Lsqr8x_sub 927 928.align 32 929.Lsqr8x_nox: 930___ 931$code.=<<___; 932 call bn_sqr8x_internal # see x86_64-mont5 module 933 # %rax top-most carry 934 # %rbp nptr 935 # %r8 -8*num 936 # %rdi end of tp[2*num] 937 lea (%rdi,$num),%rbx 938 mov $num,%rcx 939 mov $num,%rdx 940 movq %xmm1,$rptr 941 sar \$3+2,%rcx # %cf=0 942 jmp .Lsqr8x_sub 943 944.align 32 945.Lsqr8x_sub: 946 mov 8*0(%rbx),%r12 947 mov 8*1(%rbx),%r13 948 mov 8*2(%rbx),%r14 949 mov 8*3(%rbx),%r15 950 lea 8*4(%rbx),%rbx 951 sbb 8*0(%rbp),%r12 952 sbb 8*1(%rbp),%r13 953 sbb 8*2(%rbp),%r14 954 sbb 8*3(%rbp),%r15 955 lea 8*4(%rbp),%rbp 956 mov %r12,8*0($rptr) 957 mov %r13,8*1($rptr) 958 mov %r14,8*2($rptr) 959 mov %r15,8*3($rptr) 960 lea 8*4($rptr),$rptr 961 inc %rcx # preserves %cf 962 jnz .Lsqr8x_sub 963 964 sbb \$0,%rax # top-most carry 965 lea (%rbx,$num),%rbx # rewind 966 lea ($rptr,$num),$rptr # rewind 967 968 movq %rax,%xmm1 969 pxor %xmm0,%xmm0 970 pshufd \$0,%xmm1,%xmm1 971 mov 40(%rsp),%rsi # restore %rsp 972.cfi_def_cfa %rsi,8 973 jmp .Lsqr8x_cond_copy 974 975.align 32 976.Lsqr8x_cond_copy: 977 movdqa 16*0(%rbx),%xmm2 978 movdqa 16*1(%rbx),%xmm3 979 lea 16*2(%rbx),%rbx 980 movdqu 16*0($rptr),%xmm4 981 movdqu 16*1($rptr),%xmm5 982 lea 16*2($rptr),$rptr 983 movdqa %xmm0,-16*2(%rbx) # zero tp 984 movdqa %xmm0,-16*1(%rbx) 985 movdqa %xmm0,-16*2(%rbx,%rdx) 986 movdqa %xmm0,-16*1(%rbx,%rdx) 987 pcmpeqd %xmm1,%xmm0 988 pand %xmm1,%xmm2 989 pand %xmm1,%xmm3 990 pand %xmm0,%xmm4 991 pand %xmm0,%xmm5 992 pxor %xmm0,%xmm0 993 por %xmm2,%xmm4 994 por %xmm3,%xmm5 995 movdqu %xmm4,-16*2($rptr) 996 movdqu %xmm5,-16*1($rptr) 997 add \$32,$num 998 jnz .Lsqr8x_cond_copy 999 1000 mov \$1,%rax 1001 mov -48(%rsi),%r15 1002.cfi_restore %r15 1003 mov -40(%rsi),%r14 1004.cfi_restore %r14 1005 mov -32(%rsi),%r13 1006.cfi_restore %r13 1007 mov -24(%rsi),%r12 1008.cfi_restore %r12 1009 mov -16(%rsi),%rbp 1010.cfi_restore %rbp 1011 mov -8(%rsi),%rbx 1012.cfi_restore %rbx 1013 lea (%rsi),%rsp 1014.cfi_def_cfa_register %rsp 1015.Lsqr8x_epilogue: 1016 ret 1017.cfi_endproc 1018.size bn_sqr8x_mont,.-bn_sqr8x_mont 1019___ 1020}}} 1021 1022if ($addx) {{{ 1023my $bp="%rdx"; # original value 1024 1025$code.=<<___; 1026.type bn_mulx4x_mont,\@function,6 1027.align 32 1028bn_mulx4x_mont: 1029.cfi_startproc 1030 mov %rsp,%rax 1031.cfi_def_cfa_register %rax 1032.Lmulx4x_enter: 1033 push %rbx 1034.cfi_push %rbx 1035 push %rbp 1036.cfi_push %rbp 1037 push %r12 1038.cfi_push %r12 1039 push %r13 1040.cfi_push %r13 1041 push %r14 1042.cfi_push %r14 1043 push %r15 1044.cfi_push %r15 1045.Lmulx4x_prologue: 1046 1047 shl \$3,${num}d # convert $num to bytes 1048 xor %r10,%r10 1049 sub $num,%r10 # -$num 1050 mov ($n0),$n0 # *n0 1051 lea -72(%rsp,%r10),%rbp # future alloca(frame+$num+8) 1052 and \$-128,%rbp 1053 mov %rsp,%r11 1054 sub %rbp,%r11 1055 and \$-4096,%r11 1056 lea (%rbp,%r11),%rsp 1057 mov (%rsp),%r10 1058 cmp %rbp,%rsp 1059 ja .Lmulx4x_page_walk 1060 jmp .Lmulx4x_page_walk_done 1061 1062.align 16 1063.Lmulx4x_page_walk: 1064 lea -4096(%rsp),%rsp 1065 mov (%rsp),%r10 1066 cmp %rbp,%rsp 1067 ja .Lmulx4x_page_walk 1068.Lmulx4x_page_walk_done: 1069 1070 lea ($bp,$num),%r10 1071 ############################################################## 1072 # Stack layout 1073 # +0 num 1074 # +8 off-loaded &b[i] 1075 # +16 end of b[num] 1076 # +24 saved n0 1077 # +32 saved rp 1078 # +40 saved %rsp 1079 # +48 inner counter 1080 # +56 1081 # +64 tmp[num+1] 1082 # 1083 mov $num,0(%rsp) # save $num 1084 shr \$5,$num 1085 mov %r10,16(%rsp) # end of b[num] 1086 sub \$1,$num 1087 mov $n0, 24(%rsp) # save *n0 1088 mov $rp, 32(%rsp) # save $rp 1089 mov %rax,40(%rsp) # save original %rsp 1090.cfi_cfa_expression %rsp+40,deref,+8 1091 mov $num,48(%rsp) # inner counter 1092 jmp .Lmulx4x_body 1093 1094.align 32 1095.Lmulx4x_body: 1096___ 1097my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)= 1098 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax"); 1099my $rptr=$bptr; 1100$code.=<<___; 1101 lea 8($bp),$bptr 1102 mov ($bp),%rdx # b[0], $bp==%rdx actually 1103 lea 64+32(%rsp),$tptr 1104 mov %rdx,$bi 1105 1106 mulx 0*8($aptr),$mi,%rax # a[0]*b[0] 1107 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0] 1108 add %rax,%r11 1109 mov $bptr,8(%rsp) # off-load &b[i] 1110 mulx 2*8($aptr),%r12,%r13 # ... 1111 adc %r14,%r12 1112 adc \$0,%r13 1113 1114 mov $mi,$bptr # borrow $bptr 1115 imulq 24(%rsp),$mi # "t[0]"*n0 1116 xor $zero,$zero # cf=0, of=0 1117 1118 mulx 3*8($aptr),%rax,%r14 1119 mov $mi,%rdx 1120 lea 4*8($aptr),$aptr 1121 adcx %rax,%r13 1122 adcx $zero,%r14 # cf=0 1123 1124 mulx 0*8($nptr),%rax,%r10 1125 adcx %rax,$bptr # discarded 1126 adox %r11,%r10 1127 mulx 1*8($nptr),%rax,%r11 1128 adcx %rax,%r10 1129 adox %r12,%r11 1130 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12 1131 mov 48(%rsp),$bptr # counter value 1132 mov %r10,-4*8($tptr) 1133 adcx %rax,%r11 1134 adox %r13,%r12 1135 mulx 3*8($nptr),%rax,%r15 1136 mov $bi,%rdx 1137 mov %r11,-3*8($tptr) 1138 adcx %rax,%r12 1139 adox $zero,%r15 # of=0 1140 lea 4*8($nptr),$nptr 1141 mov %r12,-2*8($tptr) 1142 1143 jmp .Lmulx4x_1st 1144 1145.align 32 1146.Lmulx4x_1st: 1147 adcx $zero,%r15 # cf=0, modulo-scheduled 1148 mulx 0*8($aptr),%r10,%rax # a[4]*b[0] 1149 adcx %r14,%r10 1150 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0] 1151 adcx %rax,%r11 1152 mulx 2*8($aptr),%r12,%rax # ... 1153 adcx %r14,%r12 1154 mulx 3*8($aptr),%r13,%r14 1155 .byte 0x67,0x67 1156 mov $mi,%rdx 1157 adcx %rax,%r13 1158 adcx $zero,%r14 # cf=0 1159 lea 4*8($aptr),$aptr 1160 lea 4*8($tptr),$tptr 1161 1162 adox %r15,%r10 1163 mulx 0*8($nptr),%rax,%r15 1164 adcx %rax,%r10 1165 adox %r15,%r11 1166 mulx 1*8($nptr),%rax,%r15 1167 adcx %rax,%r11 1168 adox %r15,%r12 1169 mulx 2*8($nptr),%rax,%r15 1170 mov %r10,-5*8($tptr) 1171 adcx %rax,%r12 1172 mov %r11,-4*8($tptr) 1173 adox %r15,%r13 1174 mulx 3*8($nptr),%rax,%r15 1175 mov $bi,%rdx 1176 mov %r12,-3*8($tptr) 1177 adcx %rax,%r13 1178 adox $zero,%r15 1179 lea 4*8($nptr),$nptr 1180 mov %r13,-2*8($tptr) 1181 1182 dec $bptr # of=0, pass cf 1183 jnz .Lmulx4x_1st 1184 1185 mov 0(%rsp),$num # load num 1186 mov 8(%rsp),$bptr # re-load &b[i] 1187 adc $zero,%r15 # modulo-scheduled 1188 add %r15,%r14 1189 sbb %r15,%r15 # top-most carry 1190 mov %r14,-1*8($tptr) 1191 jmp .Lmulx4x_outer 1192 1193.align 32 1194.Lmulx4x_outer: 1195 mov ($bptr),%rdx # b[i] 1196 lea 8($bptr),$bptr # b++ 1197 sub $num,$aptr # rewind $aptr 1198 mov %r15,($tptr) # save top-most carry 1199 lea 64+4*8(%rsp),$tptr 1200 sub $num,$nptr # rewind $nptr 1201 1202 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i] 1203 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0 1204 mov %rdx,$bi 1205 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i] 1206 adox -4*8($tptr),$mi 1207 adcx %r14,%r11 1208 mulx 2*8($aptr),%r15,%r13 # ... 1209 adox -3*8($tptr),%r11 1210 adcx %r15,%r12 1211 adox -2*8($tptr),%r12 1212 adcx $zero,%r13 1213 adox $zero,%r13 1214 1215 mov $bptr,8(%rsp) # off-load &b[i] 1216 mov $mi,%r15 1217 imulq 24(%rsp),$mi # "t[0]"*n0 1218 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0 1219 1220 mulx 3*8($aptr),%rax,%r14 1221 mov $mi,%rdx 1222 adcx %rax,%r13 1223 adox -1*8($tptr),%r13 1224 adcx $zero,%r14 1225 lea 4*8($aptr),$aptr 1226 adox $zero,%r14 1227 1228 mulx 0*8($nptr),%rax,%r10 1229 adcx %rax,%r15 # discarded 1230 adox %r11,%r10 1231 mulx 1*8($nptr),%rax,%r11 1232 adcx %rax,%r10 1233 adox %r12,%r11 1234 mulx 2*8($nptr),%rax,%r12 1235 mov %r10,-4*8($tptr) 1236 adcx %rax,%r11 1237 adox %r13,%r12 1238 mulx 3*8($nptr),%rax,%r15 1239 mov $bi,%rdx 1240 mov %r11,-3*8($tptr) 1241 lea 4*8($nptr),$nptr 1242 adcx %rax,%r12 1243 adox $zero,%r15 # of=0 1244 mov 48(%rsp),$bptr # counter value 1245 mov %r12,-2*8($tptr) 1246 1247 jmp .Lmulx4x_inner 1248 1249.align 32 1250.Lmulx4x_inner: 1251 mulx 0*8($aptr),%r10,%rax # a[4]*b[i] 1252 adcx $zero,%r15 # cf=0, modulo-scheduled 1253 adox %r14,%r10 1254 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i] 1255 adcx 0*8($tptr),%r10 1256 adox %rax,%r11 1257 mulx 2*8($aptr),%r12,%rax # ... 1258 adcx 1*8($tptr),%r11 1259 adox %r14,%r12 1260 mulx 3*8($aptr),%r13,%r14 1261 mov $mi,%rdx 1262 adcx 2*8($tptr),%r12 1263 adox %rax,%r13 1264 adcx 3*8($tptr),%r13 1265 adox $zero,%r14 # of=0 1266 lea 4*8($aptr),$aptr 1267 lea 4*8($tptr),$tptr 1268 adcx $zero,%r14 # cf=0 1269 1270 adox %r15,%r10 1271 mulx 0*8($nptr),%rax,%r15 1272 adcx %rax,%r10 1273 adox %r15,%r11 1274 mulx 1*8($nptr),%rax,%r15 1275 adcx %rax,%r11 1276 adox %r15,%r12 1277 mulx 2*8($nptr),%rax,%r15 1278 mov %r10,-5*8($tptr) 1279 adcx %rax,%r12 1280 adox %r15,%r13 1281 mulx 3*8($nptr),%rax,%r15 1282 mov $bi,%rdx 1283 mov %r11,-4*8($tptr) 1284 mov %r12,-3*8($tptr) 1285 adcx %rax,%r13 1286 adox $zero,%r15 1287 lea 4*8($nptr),$nptr 1288 mov %r13,-2*8($tptr) 1289 1290 dec $bptr # of=0, pass cf 1291 jnz .Lmulx4x_inner 1292 1293 mov 0(%rsp),$num # load num 1294 mov 8(%rsp),$bptr # re-load &b[i] 1295 adc $zero,%r15 # modulo-scheduled 1296 sub 0*8($tptr),$zero # pull top-most carry 1297 adc %r15,%r14 1298 sbb %r15,%r15 # top-most carry 1299 mov %r14,-1*8($tptr) 1300 1301 cmp 16(%rsp),$bptr 1302 jne .Lmulx4x_outer 1303 1304 lea 64(%rsp),$tptr 1305 sub $num,$nptr # rewind $nptr 1306 neg %r15 1307 mov $num,%rdx 1308 shr \$3+2,$num # %cf=0 1309 mov 32(%rsp),$rptr # restore rp 1310 jmp .Lmulx4x_sub 1311 1312.align 32 1313.Lmulx4x_sub: 1314 mov 8*0($tptr),%r11 1315 mov 8*1($tptr),%r12 1316 mov 8*2($tptr),%r13 1317 mov 8*3($tptr),%r14 1318 lea 8*4($tptr),$tptr 1319 sbb 8*0($nptr),%r11 1320 sbb 8*1($nptr),%r12 1321 sbb 8*2($nptr),%r13 1322 sbb 8*3($nptr),%r14 1323 lea 8*4($nptr),$nptr 1324 mov %r11,8*0($rptr) 1325 mov %r12,8*1($rptr) 1326 mov %r13,8*2($rptr) 1327 mov %r14,8*3($rptr) 1328 lea 8*4($rptr),$rptr 1329 dec $num # preserves %cf 1330 jnz .Lmulx4x_sub 1331 1332 sbb \$0,%r15 # top-most carry 1333 lea 64(%rsp),$tptr 1334 sub %rdx,$rptr # rewind 1335 1336 movq %r15,%xmm1 1337 pxor %xmm0,%xmm0 1338 pshufd \$0,%xmm1,%xmm1 1339 mov 40(%rsp),%rsi # restore %rsp 1340.cfi_def_cfa %rsi,8 1341 jmp .Lmulx4x_cond_copy 1342 1343.align 32 1344.Lmulx4x_cond_copy: 1345 movdqa 16*0($tptr),%xmm2 1346 movdqa 16*1($tptr),%xmm3 1347 lea 16*2($tptr),$tptr 1348 movdqu 16*0($rptr),%xmm4 1349 movdqu 16*1($rptr),%xmm5 1350 lea 16*2($rptr),$rptr 1351 movdqa %xmm0,-16*2($tptr) # zero tp 1352 movdqa %xmm0,-16*1($tptr) 1353 pcmpeqd %xmm1,%xmm0 1354 pand %xmm1,%xmm2 1355 pand %xmm1,%xmm3 1356 pand %xmm0,%xmm4 1357 pand %xmm0,%xmm5 1358 pxor %xmm0,%xmm0 1359 por %xmm2,%xmm4 1360 por %xmm3,%xmm5 1361 movdqu %xmm4,-16*2($rptr) 1362 movdqu %xmm5,-16*1($rptr) 1363 sub \$32,%rdx 1364 jnz .Lmulx4x_cond_copy 1365 1366 mov %rdx,($tptr) 1367 1368 mov \$1,%rax 1369 mov -48(%rsi),%r15 1370.cfi_restore %r15 1371 mov -40(%rsi),%r14 1372.cfi_restore %r14 1373 mov -32(%rsi),%r13 1374.cfi_restore %r13 1375 mov -24(%rsi),%r12 1376.cfi_restore %r12 1377 mov -16(%rsi),%rbp 1378.cfi_restore %rbp 1379 mov -8(%rsi),%rbx 1380.cfi_restore %rbx 1381 lea (%rsi),%rsp 1382.cfi_def_cfa_register %rsp 1383.Lmulx4x_epilogue: 1384 ret 1385.cfi_endproc 1386.size bn_mulx4x_mont,.-bn_mulx4x_mont 1387___ 1388}}} 1389$code.=<<___; 1390.asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>" 1391.align 16 1392___ 1393 1394# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 1395# CONTEXT *context,DISPATCHER_CONTEXT *disp) 1396if ($win64) { 1397$rec="%rcx"; 1398$frame="%rdx"; 1399$context="%r8"; 1400$disp="%r9"; 1401 1402$code.=<<___; 1403.extern __imp_RtlVirtualUnwind 1404.type mul_handler,\@abi-omnipotent 1405.align 16 1406mul_handler: 1407 push %rsi 1408 push %rdi 1409 push %rbx 1410 push %rbp 1411 push %r12 1412 push %r13 1413 push %r14 1414 push %r15 1415 pushfq 1416 sub \$64,%rsp 1417 1418 mov 120($context),%rax # pull context->Rax 1419 mov 248($context),%rbx # pull context->Rip 1420 1421 mov 8($disp),%rsi # disp->ImageBase 1422 mov 56($disp),%r11 # disp->HandlerData 1423 1424 mov 0(%r11),%r10d # HandlerData[0] 1425 lea (%rsi,%r10),%r10 # end of prologue label 1426 cmp %r10,%rbx # context->Rip<end of prologue label 1427 jb .Lcommon_seh_tail 1428 1429 mov 152($context),%rax # pull context->Rsp 1430 1431 mov 4(%r11),%r10d # HandlerData[1] 1432 lea (%rsi,%r10),%r10 # epilogue label 1433 cmp %r10,%rbx # context->Rip>=epilogue label 1434 jae .Lcommon_seh_tail 1435 1436 mov 192($context),%r10 # pull $num 1437 mov 8(%rax,%r10,8),%rax # pull saved stack pointer 1438 1439 jmp .Lcommon_pop_regs 1440.size mul_handler,.-mul_handler 1441 1442.type sqr_handler,\@abi-omnipotent 1443.align 16 1444sqr_handler: 1445 push %rsi 1446 push %rdi 1447 push %rbx 1448 push %rbp 1449 push %r12 1450 push %r13 1451 push %r14 1452 push %r15 1453 pushfq 1454 sub \$64,%rsp 1455 1456 mov 120($context),%rax # pull context->Rax 1457 mov 248($context),%rbx # pull context->Rip 1458 1459 mov 8($disp),%rsi # disp->ImageBase 1460 mov 56($disp),%r11 # disp->HandlerData 1461 1462 mov 0(%r11),%r10d # HandlerData[0] 1463 lea (%rsi,%r10),%r10 # end of prologue label 1464 cmp %r10,%rbx # context->Rip<.Lsqr_prologue 1465 jb .Lcommon_seh_tail 1466 1467 mov 4(%r11),%r10d # HandlerData[1] 1468 lea (%rsi,%r10),%r10 # body label 1469 cmp %r10,%rbx # context->Rip<.Lsqr_body 1470 jb .Lcommon_pop_regs 1471 1472 mov 152($context),%rax # pull context->Rsp 1473 1474 mov 8(%r11),%r10d # HandlerData[2] 1475 lea (%rsi,%r10),%r10 # epilogue label 1476 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue 1477 jae .Lcommon_seh_tail 1478 1479 mov 40(%rax),%rax # pull saved stack pointer 1480 1481.Lcommon_pop_regs: 1482 mov -8(%rax),%rbx 1483 mov -16(%rax),%rbp 1484 mov -24(%rax),%r12 1485 mov -32(%rax),%r13 1486 mov -40(%rax),%r14 1487 mov -48(%rax),%r15 1488 mov %rbx,144($context) # restore context->Rbx 1489 mov %rbp,160($context) # restore context->Rbp 1490 mov %r12,216($context) # restore context->R12 1491 mov %r13,224($context) # restore context->R13 1492 mov %r14,232($context) # restore context->R14 1493 mov %r15,240($context) # restore context->R15 1494 1495.Lcommon_seh_tail: 1496 mov 8(%rax),%rdi 1497 mov 16(%rax),%rsi 1498 mov %rax,152($context) # restore context->Rsp 1499 mov %rsi,168($context) # restore context->Rsi 1500 mov %rdi,176($context) # restore context->Rdi 1501 1502 mov 40($disp),%rdi # disp->ContextRecord 1503 mov $context,%rsi # context 1504 mov \$154,%ecx # sizeof(CONTEXT) 1505 .long 0xa548f3fc # cld; rep movsq 1506 1507 mov $disp,%rsi 1508 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER 1509 mov 8(%rsi),%rdx # arg2, disp->ImageBase 1510 mov 0(%rsi),%r8 # arg3, disp->ControlPc 1511 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry 1512 mov 40(%rsi),%r10 # disp->ContextRecord 1513 lea 56(%rsi),%r11 # &disp->HandlerData 1514 lea 24(%rsi),%r12 # &disp->EstablisherFrame 1515 mov %r10,32(%rsp) # arg5 1516 mov %r11,40(%rsp) # arg6 1517 mov %r12,48(%rsp) # arg7 1518 mov %rcx,56(%rsp) # arg8, (NULL) 1519 call *__imp_RtlVirtualUnwind(%rip) 1520 1521 mov \$1,%eax # ExceptionContinueSearch 1522 add \$64,%rsp 1523 popfq 1524 pop %r15 1525 pop %r14 1526 pop %r13 1527 pop %r12 1528 pop %rbp 1529 pop %rbx 1530 pop %rdi 1531 pop %rsi 1532 ret 1533.size sqr_handler,.-sqr_handler 1534 1535.section .pdata 1536.align 4 1537 .rva .LSEH_begin_bn_mul_mont 1538 .rva .LSEH_end_bn_mul_mont 1539 .rva .LSEH_info_bn_mul_mont 1540 1541 .rva .LSEH_begin_bn_mul4x_mont 1542 .rva .LSEH_end_bn_mul4x_mont 1543 .rva .LSEH_info_bn_mul4x_mont 1544 1545 .rva .LSEH_begin_bn_sqr8x_mont 1546 .rva .LSEH_end_bn_sqr8x_mont 1547 .rva .LSEH_info_bn_sqr8x_mont 1548___ 1549$code.=<<___ if ($addx); 1550 .rva .LSEH_begin_bn_mulx4x_mont 1551 .rva .LSEH_end_bn_mulx4x_mont 1552 .rva .LSEH_info_bn_mulx4x_mont 1553___ 1554$code.=<<___; 1555.section .xdata 1556.align 8 1557.LSEH_info_bn_mul_mont: 1558 .byte 9,0,0,0 1559 .rva mul_handler 1560 .rva .Lmul_body,.Lmul_epilogue # HandlerData[] 1561.LSEH_info_bn_mul4x_mont: 1562 .byte 9,0,0,0 1563 .rva mul_handler 1564 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[] 1565.LSEH_info_bn_sqr8x_mont: 1566 .byte 9,0,0,0 1567 .rva sqr_handler 1568 .rva .Lsqr8x_prologue,.Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[] 1569.align 8 1570___ 1571$code.=<<___ if ($addx); 1572.LSEH_info_bn_mulx4x_mont: 1573 .byte 9,0,0,0 1574 .rva sqr_handler 1575 .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[] 1576.align 8 1577___ 1578} 1579 1580print $code; 1581close STDOUT or die "error closing STDOUT"; 1582