crypto/md5/asm/md5-sparcv9.pl

   1 #!/usr/bin/env perl
   2
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@opensl.org> for the OpenSSL
   5 # project. The module is, however, dual licensed under OpenSSL and
   6 # CRYPTOGAMS licenses depending on where you obtain it. For further
   7 # details see http://www.openssl.org/~appro/cryptogams/.
   8 # ====================================================================
   9
  10 # MD5 for SPARCv9, 7.5 cycles per byte on UltraSPARC, >40% faster than
  11 # code generated by Sun C 5.2.
  12
  13 $bits=32;
  14 for (@ARGV)     { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
  15 if ($bits==64)  { $bias=2047; $frame=192; }
  16 else            { $bias=0;    $frame=112; }
  17
  18 $output=shift;
  19 open STDOUT,">$output";
  20
  21 use integer;
  22
  23 ($ctx,$inp,$len)=("%i0","%i1","%i2");   # input arguments
  24
  25 # 64-bit values
  26 @X=("%o0","%o1","%o2","%o3","%o4","%o5","%o7","%g1","%g2");
  27 $tx="%g3";
  28 ($AB,$CD)=("%g4","%g5");
  29
  30 # 32-bit values
  31 @V=($A,$B,$C,$D)=map("%l$_",(0..3));
  32 ($t1,$t2,$t3,$saved_asi)=map("%l$_",(4..7));
  33 ($shr,$shl1,$shl2)=("%i3","%i4","%i5");
  34
  35 my @K=( 0xd76aa478,0xe8c7b756,0x242070db,0xc1bdceee,
  36         0xf57c0faf,0x4787c62a,0xa8304613,0xfd469501,
  37         0x698098d8,0x8b44f7af,0xffff5bb1,0x895cd7be,
  38         0x6b901122,0xfd987193,0xa679438e,0x49b40821,
  39
  40         0xf61e2562,0xc040b340,0x265e5a51,0xe9b6c7aa,
  41         0xd62f105d,0x02441453,0xd8a1e681,0xe7d3fbc8,
  42         0x21e1cde6,0xc33707d6,0xf4d50d87,0x455a14ed,
  43         0xa9e3e905,0xfcefa3f8,0x676f02d9,0x8d2a4c8a,
  44
  45         0xfffa3942,0x8771f681,0x6d9d6122,0xfde5380c,
  46         0xa4beea44,0x4bdecfa9,0xf6bb4b60,0xbebfbc70,
  47         0x289b7ec6,0xeaa127fa,0xd4ef3085,0x04881d05,
  48         0xd9d4d039,0xe6db99e5,0x1fa27cf8,0xc4ac5665,
  49
  50         0xf4292244,0x432aff97,0xab9423a7,0xfc93a039,
  51         0x655b59c3,0x8f0ccc92,0xffeff47d,0x85845dd1,
  52         0x6fa87e4f,0xfe2ce6e0,0xa3014314,0x4e0811a1,
  53         0xf7537e82,0xbd3af235,0x2ad7d2bb,0xeb86d391, 0  );
  54
  55 sub R0 {
  56   my ($i,$a,$b,$c,$d) = @_;
  57   my $rot = (7,12,17,22)[$i%4];
  58   my $j   = ($i+1)/2;
  59
  60   if ($i&1) {
  61     $code.=<<___;
  62          srlx   @X[$j],$shr,@X[$j]      ! align X[`$i+1`]
  63         and     $b,$t1,$t1              ! round $i
  64          sllx   @X[$j+1],$shl1,$tx
  65         add     $t2,$a,$a
  66          sllx   $tx,$shl2,$tx
  67         xor     $d,$t1,$t1
  68          or     $tx,@X[$j],@X[$j]
  69          sethi  %hi(@K[$i+1]),$t2
  70         add     $t1,$a,$a
  71          or     $t2,%lo(@K[$i+1]),$t2
  72         sll     $a,$rot,$t3
  73          add    @X[$j],$t2,$t2          ! X[`$i+1`]+K[`$i+1`]
  74         srl     $a,32-$rot,$a
  75         add     $b,$t3,$t3
  76          xor     $b,$c,$t1
  77         add     $t3,$a,$a
  78 ___
  79   } else {
  80     $code.=<<___;
  81          srlx   @X[$j],32,$tx           ! extract X[`2*$j+1`]
  82         and     $b,$t1,$t1              ! round $i
  83         add     $t2,$a,$a
  84         xor     $d,$t1,$t1
  85          sethi  %hi(@K[$i+1]),$t2
  86         add     $t1,$a,$a
  87          or     $t2,%lo(@K[$i+1]),$t2
  88         sll     $a,$rot,$t3
  89          add    $tx,$t2,$t2             ! X[`2*$j+1`]+K[`$i+1`]
  90         srl     $a,32-$rot,$a
  91         add     $b,$t3,$t3
  92          xor     $b,$c,$t1
  93         add     $t3,$a,$a
  94 ___
  95   }
  96 }
  97
  98 sub R0_1 {
  99   my ($i,$a,$b,$c,$d) = @_;
 100   my $rot = (7,12,17,22)[$i%4];
 101
 102 $code.=<<___;
 103          srlx   @X[0],32,$tx            ! extract X[1]
 104         and     $b,$t1,$t1              ! round $i
 105         add     $t2,$a,$a
 106         xor     $d,$t1,$t1
 107          sethi  %hi(@K[$i+1]),$t2
 108         add     $t1,$a,$a
 109          or     $t2,%lo(@K[$i+1]),$t2
 110         sll     $a,$rot,$t3
 111          add    $tx,$t2,$t2             ! X[1]+K[`$i+1`]
 112         srl     $a,32-$rot,$a
 113         add     $b,$t3,$t3
 114          andn    $b,$c,$t1
 115         add     $t3,$a,$a
 116 ___
 117 }
 118
 119 sub R1 {
 120   my ($i,$a,$b,$c,$d) = @_;
 121   my $rot = (5,9,14,20)[$i%4];
 122   my $j   = $i<31 ? (1+5*($i+1))%16 : (5+3*($i+1))%16;
 123   my $xi  = @X[$j/2];
 124
 125 $code.=<<___ if ($j&1 && ($xi=$tx));
 126          srlx   @X[$j/2],32,$xi         ! extract X[$j]
 127 ___
 128 $code.=<<___;
 129         and     $b,$d,$t3               ! round $i
 130         add     $t2,$a,$a
 131         or      $t3,$t1,$t1
 132          sethi  %hi(@K[$i+1]),$t2
 133         add     $t1,$a,$a
 134          or     $t2,%lo(@K[$i+1]),$t2
 135         sll     $a,$rot,$t3
 136          add    $xi,$t2,$t2             ! X[$j]+K[`$i+1`]
 137         srl     $a,32-$rot,$a
 138         add     $b,$t3,$t3
 139          `$i<31?"andn":"xor"`    $b,$c,$t1
 140         add     $t3,$a,$a
 141 ___
 142 }
 143
 144 sub R2 {
 145   my ($i,$a,$b,$c,$d) = @_;
 146   my $rot = (4,11,16,23)[$i%4];
 147   my $j   = $i<47 ? (5+3*($i+1))%16 : (0+7*($i+1))%16;
 148   my $xi  = @X[$j/2];
 149
 150 $code.=<<___ if ($j&1 && ($xi=$tx));
 151          srlx   @X[$j/2],32,$xi         ! extract X[$j]
 152 ___
 153 $code.=<<___;
 154         add     $t2,$a,$a               ! round $i
 155         xor     $b,$t1,$t1
 156          sethi  %hi(@K[$i+1]),$t2
 157         add     $t1,$a,$a
 158          or     $t2,%lo(@K[$i+1]),$t2
 159         sll     $a,$rot,$t3
 160          add    $xi,$t2,$t2             ! X[$j]+K[`$i+1`]
 161         srl     $a,32-$rot,$a
 162         add     $b,$t3,$t3
 163          xor     $b,$c,$t1
 164         add     $t3,$a,$a
 165 ___
 166 }
 167
 168 sub R3 {
 169   my ($i,$a,$b,$c,$d) = @_;
 170   my $rot = (6,10,15,21)[$i%4];
 171   my $j   = (0+7*($i+1))%16;
 172   my $xi  = @X[$j/2];
 173
 174 $code.=<<___;
 175         add     $t2,$a,$a               ! round $i
 176 ___
 177 $code.=<<___ if ($j&1 && ($xi=$tx));
 178          srlx   @X[$j/2],32,$xi         ! extract X[$j]
 179 ___
 180 $code.=<<___;
 181         orn     $b,$d,$t1
 182          sethi  %hi(@K[$i+1]),$t2
 183         xor     $c,$t1,$t1
 184          or     $t2,%lo(@K[$i+1]),$t2
 185         add     $t1,$a,$a
 186         sll     $a,$rot,$t3
 187          add    $xi,$t2,$t2             ! X[$j]+K[`$i+1`]
 188         srl     $a,32-$rot,$a
 189         add     $b,$t3,$t3
 190         add     $t3,$a,$a
 191 ___
 192 }
 193
 194 $code.=<<___ if ($bits==64);
 195 .register       %g2,#scratch
 196 .register       %g3,#scratch
 197 ___
 198 $code.=<<___;
 199 .section        ".text",#alloc,#execinstr
 200
 201 .globl  md5_block_asm_data_order
 202 .align  32
 203 md5_block_asm_data_order:
 204         save    %sp,-$frame,%sp
 205
 206         rd      %asi,$saved_asi
 207         wr      %g0,0x88,%asi           ! ASI_PRIMARY_LITTLE
 208         and     $inp,7,$shr
 209         andn    $inp,7,$inp
 210
 211         sll     $shr,3,$shr             ! *=8
 212         mov     56,$shl2
 213         ld      [$ctx+0],$A
 214         sub     $shl2,$shr,$shl2
 215         ld      [$ctx+4],$B
 216         and     $shl2,32,$shl1
 217         add     $shl2,8,$shl2
 218         ld      [$ctx+8],$C
 219         sub     $shl2,$shl1,$shl2       ! shr+shl1+shl2==64
 220         ld      [$ctx+12],$D
 221         nop
 222
 223 .Loop:
 224          cmp    $shr,0                  ! was inp aligned?
 225         ldxa    [$inp+0]%asi,@X[0]      ! load little-endian input
 226         ldxa    [$inp+8]%asi,@X[1]
 227         ldxa    [$inp+16]%asi,@X[2]
 228         ldxa    [$inp+24]%asi,@X[3]
 229         ldxa    [$inp+32]%asi,@X[4]
 230          sllx   $A,32,$AB               ! pack A,B
 231         ldxa    [$inp+40]%asi,@X[5]
 232          sllx   $C,32,$CD               ! pack C,D
 233         ldxa    [$inp+48]%asi,@X[6]
 234          or     $B,$AB,$AB
 235         ldxa    [$inp+56]%asi,@X[7]
 236          or     $D,$CD,$CD
 237         bnz,a,pn        %icc,.+8
 238         ldxa    [$inp+64]%asi,@X[8]
 239
 240         srlx    @X[0],$shr,@X[0]        ! align X[0]
 241         sllx    @X[1],$shl1,$tx
 242          sethi  %hi(@K[0]),$t2
 243         sllx    $tx,$shl2,$tx
 244          or     $t2,%lo(@K[0]),$t2
 245         or      $tx,@X[0],@X[0]
 246          xor    $C,$D,$t1
 247          add    @X[0],$t2,$t2           ! X[0]+K[0]
 248 ___
 249         for ($i=0;$i<15;$i++)   { &R0($i,@V);   unshift(@V,pop(@V)); }
 250         for (;$i<16;$i++)       { &R0_1($i,@V); unshift(@V,pop(@V)); }
 251         for (;$i<32;$i++)       { &R1($i,@V);   unshift(@V,pop(@V)); }
 252         for (;$i<48;$i++)       { &R2($i,@V);   unshift(@V,pop(@V)); }
 253         for (;$i<64;$i++)       { &R3($i,@V);   unshift(@V,pop(@V)); }
 254 $code.=<<___;
 255         srlx    $AB,32,$t1              ! unpack A,B,C,D and accumulate
 256         add     $inp,64,$inp            ! advance inp
 257         srlx    $CD,32,$t2
 258         add     $t1,$A,$A
 259         subcc   $len,1,$len             ! done yet?
 260         add     $AB,$B,$B
 261         add     $t2,$C,$C
 262         add     $CD,$D,$D
 263         srl     $B,0,$B                 ! clruw $B
 264         bne     `$bits==64?"%xcc":"%icc"`,.Loop
 265         srl     $D,0,$D                 ! clruw $D
 266
 267         st      $A,[$ctx+0]             ! write out ctx
 268         st      $B,[$ctx+4]
 269         st      $C,[$ctx+8]
 270         st      $D,[$ctx+12]
 271
 272         wr      %g0,$saved_asi,%asi
 273         ret
 274         restore
 275 .type   md5_block_asm_data_order,#function
 276 .size   md5_block_asm_data_order,(.-md5_block_asm_data_order)
 277
 278 .asciz  "MD5 block transform for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
 279 .align  4
 280 ___
 281
 282 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 283 print $code;
 284 close STDOUT;