crypto/md5/asm/md5-x86_64.pl

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