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