lib/libcrypto/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         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,             $dst            /* dst += x */
  34 EOF
  35 }
  36
  37 # round2_step() does:
  38 #   dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
  39 #   %r10d = X[k_next]
  40 #   %r11d = z' (copy of z for the next step)
  41 #   %r12d = z' (copy of z for the next step)
  42 # Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
  43 sub round2_step
  44 {
  45     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
  46     $code .= " mov      1*4(%rsi),      %r10d           /* (NEXT STEP) X[1] */\n" if ($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,             $dst            /* dst += x */
  61 EOF
  62 }
  63
  64 # round3_step() does:
  65 #   dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
  66 #   %r10d = X[k_next]
  67 #   %r11d = y' (copy of y for the next step)
  68 # Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
  69 sub round3_step
  70 {
  71     my ($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,             $dst            /* dst += x */
  83 EOF
  84 }
  85
  86 # round4_step() does:
  87 #   dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
  88 #   %r10d = X[k_next]
  89 #   %r11d = not z' (copy of not z for the next step)
  90 # Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
  91 sub round4_step
  92 {
  93     my ($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 */
 108 EOF
 109 }
 110
 111 my $flavour = shift;
 112 my $output  = shift;
 113 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 114
 115 $0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
 116 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
 117 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
 118 die "can't locate x86_64-xlate.pl";
 119
 120 no warnings qw(uninitialized);
 121 open OUT,"| \"$^X\" $xlate $flavour $output";
 122 *STDOUT=*OUT;
 123
 124 $code .= <<EOF;
 125 .text
 126 .align 16
 127
 128 .globl md5_block_asm_data_order
 129 .type md5_block_asm_data_order,\@function,3
 130 md5_block_asm_data_order:
 131         push    %rbp
 132         push    %rbx
 133         push    %r12
 134         push    %r14
 135         push    %r15
 136 .Lprologue:
 137
 138         # rdi = arg #1 (ctx, MD5_CTX pointer)
 139         # rsi = arg #2 (ptr, data pointer)
 140         # rdx = arg #3 (nbr, number of 16-word blocks to process)
 141         mov     %rdi,           %rbp    # rbp = ctx
 142         shl     \$6,            %rdx    # rdx = nbr in bytes
 143         lea     (%rsi,%rdx),    %rdi    # rdi = end
 144         mov     0*4(%rbp),      %eax    # eax = ctx->A
 145         mov     1*4(%rbp),      %ebx    # ebx = ctx->B
 146         mov     2*4(%rbp),      %ecx    # ecx = ctx->C
 147         mov     3*4(%rbp),      %edx    # edx = ctx->D
 148         # end is 'rdi'
 149         # ptr is 'rsi'
 150         # A is 'eax'
 151         # B is 'ebx'
 152         # C is 'ecx'
 153         # D is 'edx'
 154
 155         cmp     %rdi,           %rsi            # cmp end with ptr
 156         je      .Lend                           # jmp if ptr == end
 157
 158         # BEGIN of loop over 16-word blocks
 159 .Lloop: # save old values of A, B, C, D
 160         mov     %eax,           %r8d
 161         mov     %ebx,           %r9d
 162         mov     %ecx,           %r14d
 163         mov     %edx,           %r15d
 164 EOF
 165 round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
 166 round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
 167 round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
 168 round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
 169 round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
 170 round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
 171 round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
 172 round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
 173 round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
 174 round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
 175 round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
 176 round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
 177 round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
 178 round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
 179 round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
 180 round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');
 181
 182 round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
 183 round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
 184 round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
 185 round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
 186 round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
 187 round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
 188 round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
 189 round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
 190 round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
 191 round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
 192 round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
 193 round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
 194 round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
 195 round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
 196 round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
 197 round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');
 198
 199 round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
 200 round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
 201 round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
 202 round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
 203 round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
 204 round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
 205 round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
 206 round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
 207 round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
 208 round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
 209 round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
 210 round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
 211 round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
 212 round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
 213 round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
 214 round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
 215
 216 round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
 217 round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
 218 round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
 219 round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
 220 round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
 221 round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
 222 round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
 223 round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
 224 round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
 225 round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
 226 round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
 227 round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
 228 round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
 229 round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
 230 round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
 231 round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
 232 $code .= <<EOF;
 233         # add old values of A, B, C, D
 234         add     %r8d,   %eax
 235         add     %r9d,   %ebx
 236         add     %r14d,  %ecx
 237         add     %r15d,  %edx
 238
 239         # loop control
 240         add     \$64,           %rsi            # ptr += 64
 241         cmp     %rdi,           %rsi            # cmp end with ptr
 242         jb      .Lloop                          # jmp if ptr < end
 243         # END of loop over 16-word blocks
 244
 245 .Lend:
 246         mov     %eax,           0*4(%rbp)       # ctx->A = A
 247         mov     %ebx,           1*4(%rbp)       # ctx->B = B
 248         mov     %ecx,           2*4(%rbp)       # ctx->C = C
 249         mov     %edx,           3*4(%rbp)       # ctx->D = D
 250
 251         mov     (%rsp),%r15
 252         mov     8(%rsp),%r14
 253         mov     16(%rsp),%r12
 254         mov     24(%rsp),%rbx
 255         mov     32(%rsp),%rbp
 256         add     \$40,%rsp
 257 .Lepilogue:
 258         ret
 259 .size md5_block_asm_data_order,.-md5_block_asm_data_order
 260 EOF
 261
 262 print $code;
 263
 264 close STDOUT;