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