crypto/openssl-0.9/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 $code;
  13
  14 # round1_step() does:
  15 #   dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
  16 #   %r10d = X[k_next]
  17 #   %r11d = z' (copy of z for the next step)
  18 # Each round1_step() takes about 5.71 clocks (9 instructions, 1.58 IPC)
  19 sub round1_step
  20 {
  21     my ($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         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,             $dst            /* dst += x */
  34 EOF
  35 }
  36
  37 # round2_step() does:
  38 #   dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
  39 #   %r10d = X[k_next]
  40 #   %r11d = y' (copy of y for the next step)
  41 # Each round2_step() takes about 6.22 clocks (9 instructions, 1.45 IPC)
  42 sub round2_step
  43 {
  44     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
  45     $code .= " mov      1*4(%rsi),      %r10d           /* (NEXT STEP) X[1] */\n" if ($pos == -1);
  46     $code .= " mov      %ecx,           %r11d           /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
  47     $code .= <<EOF;
  48         xor     $x,             %r11d           /* x ^ ... */
  49         lea     $T_i($dst,%r10d),$dst           /* Const + dst + ... */
  50         and     $z,             %r11d           /* z & ... */
  51         xor     $y,             %r11d           /* y ^ ... */
  52         mov     $k_next*4(%rsi),%r10d           /* (NEXT STEP) X[$k_next] */
  53         add     %r11d,          $dst            /* dst += ... */
  54         rol     \$$s,           $dst            /* dst <<< s */
  55         mov     $x,             %r11d           /* (NEXT STEP) y' = $x */
  56         add     $x,             $dst            /* dst += x */
  57 EOF
  58 }
  59
  60 # round3_step() does:
  61 #   dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
  62 #   %r10d = X[k_next]
  63 #   %r11d = y' (copy of y for the next step)
  64 # Each round3_step() takes about 4.26 clocks (8 instructions, 1.88 IPC)
  65 sub round3_step
  66 {
  67     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
  68     $code .= " mov      5*4(%rsi),      %r10d           /* (NEXT STEP) X[5] */\n" if ($pos == -1);
  69     $code .= " mov      %ecx,           %r11d           /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
  70     $code .= <<EOF;
  71         lea     $T_i($dst,%r10d),$dst           /* Const + dst + ... */
  72         mov     $k_next*4(%rsi),%r10d           /* (NEXT STEP) X[$k_next] */
  73         xor     $z,             %r11d           /* z ^ ... */
  74         xor     $x,             %r11d           /* x ^ ... */
  75         add     %r11d,          $dst            /* dst += ... */
  76         rol     \$$s,           $dst            /* dst <<< s */
  77         mov     $x,             %r11d           /* (NEXT STEP) y' = $x */
  78         add     $x,             $dst            /* dst += x */
  79 EOF
  80 }
  81
  82 # round4_step() does:
  83 #   dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
  84 #   %r10d = X[k_next]
  85 #   %r11d = not z' (copy of not z for the next step)
  86 # Each round4_step() takes about 5.27 clocks (9 instructions, 1.71 IPC)
  87 sub round4_step
  88 {
  89     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
  90     $code .= " mov      0*4(%rsi),      %r10d           /* (NEXT STEP) X[0] */\n" if ($pos == -1);
  91     $code .= " mov      \$0xffffffff,   %r11d\n" if ($pos == -1);
  92     $code .= " xor      %edx,           %r11d           /* (NEXT STEP) not z' = not %edx*/\n"
  93     if ($pos == -1);
  94     $code .= <<EOF;
  95         lea     $T_i($dst,%r10d),$dst           /* Const + dst + ... */
  96         or      $x,             %r11d           /* x | ... */
  97         xor     $y,             %r11d           /* y ^ ... */
  98         add     %r11d,          $dst            /* dst += ... */
  99         mov     $k_next*4(%rsi),%r10d           /* (NEXT STEP) X[$k_next] */
 100         mov     \$0xffffffff,   %r11d
 101         rol     \$$s,           $dst            /* dst <<< s */
 102         xor     $y,             %r11d           /* (NEXT STEP) not z' = not $y */
 103         add     $x,             $dst            /* dst += x */
 104 EOF
 105 }
 106
 107 my $output = shift;
 108 open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
 109
 110 $code .= <<EOF;
 111 .text
 112 .align 16
 113
 114 .globl md5_block_asm_host_order
 115 .type md5_block_asm_host_order,\@function,3
 116 md5_block_asm_host_order:
 117         push    %rbp
 118         push    %rbx
 119         push    %r14
 120         push    %r15
 121
 122         # rdi = arg #1 (ctx, MD5_CTX pointer)
 123         # rsi = arg #2 (ptr, data pointer)
 124         # rdx = arg #3 (nbr, number of 16-word blocks to process)
 125         mov     %rdi,           %rbp    # rbp = ctx
 126         shl     \$6,            %rdx    # rdx = nbr in bytes
 127         lea     (%rsi,%rdx),    %rdi    # rdi = end
 128         mov     0*4(%rbp),      %eax    # eax = ctx->A
 129         mov     1*4(%rbp),      %ebx    # ebx = ctx->B
 130         mov     2*4(%rbp),      %ecx    # ecx = ctx->C
 131         mov     3*4(%rbp),      %edx    # edx = ctx->D
 132         # end is 'rdi'
 133         # ptr is 'rsi'
 134         # A is 'eax'
 135         # B is 'ebx'
 136         # C is 'ecx'
 137         # D is 'edx'
 138
 139         cmp     %rdi,           %rsi            # cmp end with ptr
 140         je      .Lend                           # jmp if ptr == end
 141
 142         # BEGIN of loop over 16-word blocks
 143 .Lloop: # save old values of A, B, C, D
 144         mov     %eax,           %r8d
 145         mov     %ebx,           %r9d
 146         mov     %ecx,           %r14d
 147         mov     %edx,           %r15d
 148 EOF
 149 round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
 150 round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
 151 round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
 152 round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
 153 round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
 154 round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
 155 round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
 156 round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
 157 round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
 158 round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
 159 round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
 160 round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
 161 round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
 162 round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
 163 round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
 164 round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');
 165
 166 round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
 167 round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
 168 round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
 169 round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
 170 round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
 171 round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
 172 round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
 173 round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
 174 round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
 175 round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
 176 round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
 177 round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
 178 round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
 179 round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
 180 round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
 181 round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');
 182
 183 round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
 184 round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
 185 round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
 186 round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
 187 round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
 188 round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
 189 round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
 190 round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
 191 round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
 192 round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
 193 round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
 194 round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
 195 round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
 196 round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
 197 round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
 198 round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
 199
 200 round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
 201 round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
 202 round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
 203 round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
 204 round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
 205 round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
 206 round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
 207 round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
 208 round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
 209 round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
 210 round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
 211 round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
 212 round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
 213 round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
 214 round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
 215 round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
 216 $code .= <<EOF;
 217         # add old values of A, B, C, D
 218         add     %r8d,   %eax
 219         add     %r9d,   %ebx
 220         add     %r14d,  %ecx
 221         add     %r15d,  %edx
 222
 223         # loop control
 224         add     \$64,           %rsi            # ptr += 64
 225         cmp     %rdi,           %rsi            # cmp end with ptr
 226         jb      .Lloop                          # jmp if ptr < end
 227         # END of loop over 16-word blocks
 228
 229 .Lend:
 230         mov     %eax,           0*4(%rbp)       # ctx->A = A
 231         mov     %ebx,           1*4(%rbp)       # ctx->B = B
 232         mov     %ecx,           2*4(%rbp)       # ctx->C = C
 233         mov     %edx,           3*4(%rbp)       # ctx->D = D
 234
 235         pop     %r15
 236         pop     %r14
 237         pop     %rbx
 238         pop     %rbp
 239         ret
 240 .size md5_block_asm_host_order,.-md5_block_asm_host_order
 241 EOF
 242
 243 print $code;
 244
 245 close STDOUT;