1#! /usr/bin/env perl 2# Author: Marc Bevand <bevand_m (at) epita.fr> 3# Copyright 2005-2020 The OpenSSL Project Authors. All Rights Reserved. 4# 5# Licensed under the OpenSSL license (the "License"). You may not use 6# this file except in compliance with the License. You can obtain a copy 7# in the file LICENSE in the source distribution or at 8# https://www.openssl.org/source/license.html 9 10# MD5 optimized for AMD64. 11 12use strict; 13 14my $code; 15 16# round1_step() does: 17# dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s) 18# %r10d = X[k_next] 19# %r11d = z' (copy of z for the next step) 20# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC) 21sub round1_step 22{ 23 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 24 $code .= " mov 0*4(%rsi), %r10d /* (NEXT STEP) X[0] */\n" if ($pos == -1); 25 $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1); 26 $code .= <<EOF; 27 xor $y, %r11d /* y ^ ... */ 28 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 29 and $x, %r11d /* x & ... */ 30 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 31 xor $z, %r11d /* z ^ ... */ 32 add %r11d, $dst /* dst += ... */ 33 rol \$$s, $dst /* dst <<< s */ 34 mov $y, %r11d /* (NEXT STEP) z' = $y */ 35 add $x, $dst /* dst += x */ 36EOF 37} 38 39# round2_step() does: 40# dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s) 41# %r10d = X[k_next] 42# %r11d = z' (copy of z for the next step) 43# %r12d = z' (copy of z for the next step) 44# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC) 45sub round2_step 46{ 47 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 48 $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1); 49 $code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1); 50 $code .= <<EOF; 51 not %r11d /* not z */ 52 and $x, %r12d /* x & z */ 53 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 54 and $y, %r11d /* y & (not z) */ 55 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 56 or %r11d, %r12d /* (y & (not z)) | (x & z) */ 57 mov $y, %r11d /* (NEXT STEP) z' = $y */ 58 add %r12d, $dst /* dst += ... */ 59 mov $y, %r12d /* (NEXT STEP) z' = $y */ 60 rol \$$s, $dst /* dst <<< s */ 61 add $x, $dst /* dst += x */ 62EOF 63} 64 65# round3_step() does: 66# dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s) 67# %r10d = X[k_next] 68# %r11d = y' (copy of y for the next step) 69# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC) 70{ my $round3_alter=0; 71sub round3_step 72{ 73 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 74 $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1); 75 $code .= <<EOF; 76 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 77 xor $z, %r11d /* z ^ ... */ 78 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 79 xor $x, %r11d /* x ^ ... */ 80 add %r11d, $dst /* dst += ... */ 81EOF 82 $code .= <<EOF if ($round3_alter); 83 rol \$$s, $dst /* dst <<< s */ 84 mov $x, %r11d /* (NEXT STEP) y' = $x */ 85EOF 86 $code .= <<EOF if (!$round3_alter); 87 mov $x, %r11d /* (NEXT STEP) y' = $x */ 88 rol \$$s, $dst /* dst <<< s */ 89EOF 90 $code .= <<EOF; 91 add $x, $dst /* dst += x */ 92EOF 93 $round3_alter^=1; 94} 95} 96 97# round4_step() does: 98# dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s) 99# %r10d = X[k_next] 100# %r11d = not z' (copy of not z for the next step) 101# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC) 102sub round4_step 103{ 104 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 105 $code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1); 106 $code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n" 107 if ($pos == -1); 108 $code .= <<EOF; 109 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 110 or $x, %r11d /* x | ... */ 111 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 112 xor $y, %r11d /* y ^ ... */ 113 add %r11d, $dst /* dst += ... */ 114 mov \$0xffffffff, %r11d 115 rol \$$s, $dst /* dst <<< s */ 116 xor $y, %r11d /* (NEXT STEP) not z' = not $y */ 117 add $x, $dst /* dst += x */ 118EOF 119} 120 121no warnings qw(uninitialized); 122my $flavour = shift; 123my $output = shift; 124if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 125 126my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); 127 128$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate; 129( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or 130( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or 131die "can't locate x86_64-xlate.pl"; 132 133open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; 134*STDOUT=*OUT; 135 136$code .= <<EOF; 137.text 138.align 16 139 140.globl md5_block_asm_data_order 141.type md5_block_asm_data_order,\@function,3 142md5_block_asm_data_order: 143.cfi_startproc 144 push %rbp 145.cfi_push %rbp 146 push %rbx 147.cfi_push %rbx 148 push %r12 149.cfi_push %r12 150 push %r14 151.cfi_push %r14 152 push %r15 153.cfi_push %r15 154.Lprologue: 155 156 # rdi = arg #1 (ctx, MD5_CTX pointer) 157 # rsi = arg #2 (ptr, data pointer) 158 # rdx = arg #3 (nbr, number of 16-word blocks to process) 159 mov %rdi, %rbp # rbp = ctx 160 shl \$6, %rdx # rdx = nbr in bytes 161 lea (%rsi,%rdx), %rdi # rdi = end 162 mov 0*4(%rbp), %eax # eax = ctx->A 163 mov 1*4(%rbp), %ebx # ebx = ctx->B 164 mov 2*4(%rbp), %ecx # ecx = ctx->C 165 mov 3*4(%rbp), %edx # edx = ctx->D 166 # end is 'rdi' 167 # ptr is 'rsi' 168 # A is 'eax' 169 # B is 'ebx' 170 # C is 'ecx' 171 # D is 'edx' 172 173 cmp %rdi, %rsi # cmp end with ptr 174 je .Lend # jmp if ptr == end 175 176 # BEGIN of loop over 16-word blocks 177.Lloop: # save old values of A, B, C, D 178 mov %eax, %r8d 179 mov %ebx, %r9d 180 mov %ecx, %r14d 181 mov %edx, %r15d 182EOF 183round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7'); 184round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12'); 185round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17'); 186round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22'); 187round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7'); 188round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12'); 189round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17'); 190round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22'); 191round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7'); 192round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12'); 193round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17'); 194round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22'); 195round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7'); 196round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12'); 197round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17'); 198round1_step( 1,'%ebx','%ecx','%edx','%eax', '1','0x49b40821','22'); 199 200round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5'); 201round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9'); 202round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14'); 203round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20'); 204round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5'); 205round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9'); 206round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14'); 207round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20'); 208round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5'); 209round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9'); 210round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14'); 211round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20'); 212round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5'); 213round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9'); 214round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14'); 215round2_step( 1,'%ebx','%ecx','%edx','%eax', '5','0x8d2a4c8a','20'); 216 217round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4'); 218round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11'); 219round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16'); 220round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23'); 221round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4'); 222round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11'); 223round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16'); 224round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23'); 225round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4'); 226round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11'); 227round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16'); 228round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23'); 229round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4'); 230round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11'); 231round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16'); 232round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23'); 233 234round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6'); 235round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10'); 236round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15'); 237round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21'); 238round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6'); 239round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10'); 240round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15'); 241round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21'); 242round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6'); 243round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10'); 244round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15'); 245round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21'); 246round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6'); 247round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10'); 248round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15'); 249round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21'); 250$code .= <<EOF; 251 # add old values of A, B, C, D 252 add %r8d, %eax 253 add %r9d, %ebx 254 add %r14d, %ecx 255 add %r15d, %edx 256 257 # loop control 258 add \$64, %rsi # ptr += 64 259 cmp %rdi, %rsi # cmp end with ptr 260 jb .Lloop # jmp if ptr < end 261 # END of loop over 16-word blocks 262 263.Lend: 264 mov %eax, 0*4(%rbp) # ctx->A = A 265 mov %ebx, 1*4(%rbp) # ctx->B = B 266 mov %ecx, 2*4(%rbp) # ctx->C = C 267 mov %edx, 3*4(%rbp) # ctx->D = D 268 269 mov (%rsp),%r15 270.cfi_restore %r15 271 mov 8(%rsp),%r14 272.cfi_restore %r14 273 mov 16(%rsp),%r12 274.cfi_restore %r12 275 mov 24(%rsp),%rbx 276.cfi_restore %rbx 277 mov 32(%rsp),%rbp 278.cfi_restore %rbp 279 add \$40,%rsp 280.cfi_adjust_cfa_offset -40 281.Lepilogue: 282 ret 283.cfi_endproc 284.size md5_block_asm_data_order,.-md5_block_asm_data_order 285EOF 286 287# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 288# CONTEXT *context,DISPATCHER_CONTEXT *disp) 289if ($win64) { 290my $rec="%rcx"; 291my $frame="%rdx"; 292my $context="%r8"; 293my $disp="%r9"; 294 295$code.=<<___; 296.extern __imp_RtlVirtualUnwind 297.type se_handler,\@abi-omnipotent 298.align 16 299se_handler: 300 push %rsi 301 push %rdi 302 push %rbx 303 push %rbp 304 push %r12 305 push %r13 306 push %r14 307 push %r15 308 pushfq 309 sub \$64,%rsp 310 311 mov 120($context),%rax # pull context->Rax 312 mov 248($context),%rbx # pull context->Rip 313 314 lea .Lprologue(%rip),%r10 315 cmp %r10,%rbx # context->Rip<.Lprologue 316 jb .Lin_prologue 317 318 mov 152($context),%rax # pull context->Rsp 319 320 lea .Lepilogue(%rip),%r10 321 cmp %r10,%rbx # context->Rip>=.Lepilogue 322 jae .Lin_prologue 323 324 lea 40(%rax),%rax 325 326 mov -8(%rax),%rbp 327 mov -16(%rax),%rbx 328 mov -24(%rax),%r12 329 mov -32(%rax),%r14 330 mov -40(%rax),%r15 331 mov %rbx,144($context) # restore context->Rbx 332 mov %rbp,160($context) # restore context->Rbp 333 mov %r12,216($context) # restore context->R12 334 mov %r14,232($context) # restore context->R14 335 mov %r15,240($context) # restore context->R15 336 337.Lin_prologue: 338 mov 8(%rax),%rdi 339 mov 16(%rax),%rsi 340 mov %rax,152($context) # restore context->Rsp 341 mov %rsi,168($context) # restore context->Rsi 342 mov %rdi,176($context) # restore context->Rdi 343 344 mov 40($disp),%rdi # disp->ContextRecord 345 mov $context,%rsi # context 346 mov \$154,%ecx # sizeof(CONTEXT) 347 .long 0xa548f3fc # cld; rep movsq 348 349 mov $disp,%rsi 350 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER 351 mov 8(%rsi),%rdx # arg2, disp->ImageBase 352 mov 0(%rsi),%r8 # arg3, disp->ControlPc 353 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry 354 mov 40(%rsi),%r10 # disp->ContextRecord 355 lea 56(%rsi),%r11 # &disp->HandlerData 356 lea 24(%rsi),%r12 # &disp->EstablisherFrame 357 mov %r10,32(%rsp) # arg5 358 mov %r11,40(%rsp) # arg6 359 mov %r12,48(%rsp) # arg7 360 mov %rcx,56(%rsp) # arg8, (NULL) 361 call *__imp_RtlVirtualUnwind(%rip) 362 363 mov \$1,%eax # ExceptionContinueSearch 364 add \$64,%rsp 365 popfq 366 pop %r15 367 pop %r14 368 pop %r13 369 pop %r12 370 pop %rbp 371 pop %rbx 372 pop %rdi 373 pop %rsi 374 ret 375.size se_handler,.-se_handler 376 377.section .pdata 378.align 4 379 .rva .LSEH_begin_md5_block_asm_data_order 380 .rva .LSEH_end_md5_block_asm_data_order 381 .rva .LSEH_info_md5_block_asm_data_order 382 383.section .xdata 384.align 8 385.LSEH_info_md5_block_asm_data_order: 386 .byte 9,0,0,0 387 .rva se_handler 388___ 389} 390 391print $code; 392 393close STDOUT or die "error closing STDOUT: $!"; 394