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