crypto/sha/asm/sha1-ppc.pl

   1 #!/usr/bin/env perl
   2
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
   5 # project. Rights for redistribution and usage in source and binary
   6 # forms are granted according to the OpenSSL license.
   7 # ====================================================================
   8
   9 # I let hardware handle unaligned input(*), except on page boundaries
  10 # (see below for details). Otherwise straightforward implementation
  11 # with X vector in register bank. The module is big-endian [which is
  12 # not big deal as there're no little-endian targets left around].
  13 #
  14 # (*) this means that this module is inappropriate for PPC403? Does
  15 #     anybody know if pre-POWER3 can sustain unaligned load?
  16
  17 #                       -m64    -m32
  18 # ----------------------------------
  19 # PPC970,gcc-4.0.0      +76%    +59%
  20
  21 $output = shift;
  22
  23 if ($output =~ /64\.s/) {
  24         $SIZE_T =8;
  25         $UCMP   ="cmpld";
  26         $STU    ="stdu";
  27         $POP    ="ld";
  28         $PUSH   ="std";
  29 } elsif ($output =~ /32\.s/) {
  30         $SIZE_T =4;
  31         $UCMP   ="cmplw";
  32         $STU    ="stwu";
  33         $POP    ="lwz";
  34         $PUSH   ="stw";
  35 } else { die "nonsense $output"; }
  36
  37 ( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
  38         die "can't call ../perlasm/ppc-xlate.pl: $!";
  39
  40 $FRAME=24*$SIZE_T;
  41
  42 $K  ="r0";
  43 $sp ="r1";
  44 $toc="r2";
  45 $ctx="r3";
  46 $inp="r4";
  47 $num="r5";
  48 $t0 ="r15";
  49 $t1 ="r6";
  50
  51 $A  ="r7";
  52 $B  ="r8";
  53 $C  ="r9";
  54 $D  ="r10";
  55 $E  ="r11";
  56 $T  ="r12";
  57
  58 @V=($A,$B,$C,$D,$E,$T);
  59 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
  60     "r24","r25","r26","r27","r28","r29","r30","r31");
  61
  62 sub BODY_00_19 {
  63 my ($i,$a,$b,$c,$d,$e,$f)=@_;
  64 my $j=$i+1;
  65 $code.=<<___ if ($i==0);
  66         lwz     @X[$i],`$i*4`($inp)
  67 ___
  68 $code.=<<___ if ($i<15);
  69         lwz     @X[$j],`$j*4`($inp)
  70         add     $f,$K,$e
  71         rotlwi  $e,$a,5
  72         add     $f,$f,@X[$i]
  73         and     $t0,$c,$b
  74         add     $f,$f,$e
  75         andc    $t1,$d,$b
  76         rotlwi  $b,$b,30
  77         or      $t0,$t0,$t1
  78         add     $f,$f,$t0
  79 ___
  80 $code.=<<___ if ($i>=15);
  81         add     $f,$K,$e
  82         rotlwi  $e,$a,5
  83         xor     @X[$j%16],@X[$j%16],@X[($j+2)%16]
  84         add     $f,$f,@X[$i%16]
  85         and     $t0,$c,$b
  86         xor     @X[$j%16],@X[$j%16],@X[($j+8)%16]
  87         add     $f,$f,$e
  88         andc    $t1,$d,$b
  89         rotlwi  $b,$b,30
  90         or      $t0,$t0,$t1
  91         xor     @X[$j%16],@X[$j%16],@X[($j+13)%16]
  92         add     $f,$f,$t0
  93         rotlwi  @X[$j%16],@X[$j%16],1
  94 ___
  95 }
  96
  97 sub BODY_20_39 {
  98 my ($i,$a,$b,$c,$d,$e,$f)=@_;
  99 my $j=$i+1;
 100 $code.=<<___ if ($i<79);
 101         add     $f,$K,$e
 102         rotlwi  $e,$a,5
 103         xor     @X[$j%16],@X[$j%16],@X[($j+2)%16]
 104         add     $f,$f,@X[$i%16]
 105         xor     $t0,$b,$c
 106         xor     @X[$j%16],@X[$j%16],@X[($j+8)%16]
 107         add     $f,$f,$e
 108         rotlwi  $b,$b,30
 109         xor     $t0,$t0,$d
 110         xor     @X[$j%16],@X[$j%16],@X[($j+13)%16]
 111         add     $f,$f,$t0
 112         rotlwi  @X[$j%16],@X[$j%16],1
 113 ___
 114 $code.=<<___ if ($i==79);
 115         add     $f,$K,$e
 116         rotlwi  $e,$a,5
 117         lwz     r16,0($ctx)
 118         add     $f,$f,@X[$i%16]
 119         xor     $t0,$b,$c
 120         lwz     r17,4($ctx)
 121         add     $f,$f,$e
 122         rotlwi  $b,$b,30
 123         lwz     r18,8($ctx)
 124         xor     $t0,$t0,$d
 125         lwz     r19,12($ctx)
 126         add     $f,$f,$t0
 127         lwz     r20,16($ctx)
 128 ___
 129 }
 130
 131 sub BODY_40_59 {
 132 my ($i,$a,$b,$c,$d,$e,$f)=@_;
 133 my $j=$i+1;
 134 $code.=<<___;
 135         add     $f,$K,$e
 136         rotlwi  $e,$a,5
 137         xor     @X[$j%16],@X[$j%16],@X[($j+2)%16]
 138         add     $f,$f,@X[$i%16]
 139         and     $t0,$b,$c
 140         xor     @X[$j%16],@X[$j%16],@X[($j+8)%16]
 141         add     $f,$f,$e
 142         or      $t1,$b,$c
 143         rotlwi  $b,$b,30
 144         xor     @X[$j%16],@X[$j%16],@X[($j+13)%16]
 145         and     $t1,$t1,$d
 146         or      $t0,$t0,$t1
 147         rotlwi  @X[$j%16],@X[$j%16],1
 148         add     $f,$f,$t0
 149 ___
 150 }
 151
 152 $code=<<___;
 153 .text
 154
 155 .globl  .sha1_block_asm_data_order
 156 .align  4
 157 .sha1_block_asm_data_order:
 158         mflr    r0
 159         $STU    $sp,`-($FRAME+64)`($sp)
 160         $PUSH   r0,`$FRAME-$SIZE_T*18`($sp)
 161         $PUSH   r15,`$FRAME-$SIZE_T*17`($sp)
 162         $PUSH   r16,`$FRAME-$SIZE_T*16`($sp)
 163         $PUSH   r17,`$FRAME-$SIZE_T*15`($sp)
 164         $PUSH   r18,`$FRAME-$SIZE_T*14`($sp)
 165         $PUSH   r19,`$FRAME-$SIZE_T*13`($sp)
 166         $PUSH   r20,`$FRAME-$SIZE_T*12`($sp)
 167         $PUSH   r21,`$FRAME-$SIZE_T*11`($sp)
 168         $PUSH   r22,`$FRAME-$SIZE_T*10`($sp)
 169         $PUSH   r23,`$FRAME-$SIZE_T*9`($sp)
 170         $PUSH   r24,`$FRAME-$SIZE_T*8`($sp)
 171         $PUSH   r25,`$FRAME-$SIZE_T*7`($sp)
 172         $PUSH   r26,`$FRAME-$SIZE_T*6`($sp)
 173         $PUSH   r27,`$FRAME-$SIZE_T*5`($sp)
 174         $PUSH   r28,`$FRAME-$SIZE_T*4`($sp)
 175         $PUSH   r29,`$FRAME-$SIZE_T*3`($sp)
 176         $PUSH   r30,`$FRAME-$SIZE_T*2`($sp)
 177         $PUSH   r31,`$FRAME-$SIZE_T*1`($sp)
 178         lwz     $A,0($ctx)
 179         lwz     $B,4($ctx)
 180         lwz     $C,8($ctx)
 181         lwz     $D,12($ctx)
 182         lwz     $E,16($ctx)
 183         andi.   r0,$inp,3
 184         bne     Lunaligned
 185 Laligned:
 186         mtctr   $num
 187         bl      Lsha1_block_private
 188 Ldone:
 189         $POP    r0,`$FRAME-$SIZE_T*18`($sp)
 190         $POP    r15,`$FRAME-$SIZE_T*17`($sp)
 191         $POP    r16,`$FRAME-$SIZE_T*16`($sp)
 192         $POP    r17,`$FRAME-$SIZE_T*15`($sp)
 193         $POP    r18,`$FRAME-$SIZE_T*14`($sp)
 194         $POP    r19,`$FRAME-$SIZE_T*13`($sp)
 195         $POP    r20,`$FRAME-$SIZE_T*12`($sp)
 196         $POP    r21,`$FRAME-$SIZE_T*11`($sp)
 197         $POP    r22,`$FRAME-$SIZE_T*10`($sp)
 198         $POP    r23,`$FRAME-$SIZE_T*9`($sp)
 199         $POP    r24,`$FRAME-$SIZE_T*8`($sp)
 200         $POP    r25,`$FRAME-$SIZE_T*7`($sp)
 201         $POP    r26,`$FRAME-$SIZE_T*6`($sp)
 202         $POP    r27,`$FRAME-$SIZE_T*5`($sp)
 203         $POP    r28,`$FRAME-$SIZE_T*4`($sp)
 204         $POP    r29,`$FRAME-$SIZE_T*3`($sp)
 205         $POP    r30,`$FRAME-$SIZE_T*2`($sp)
 206         $POP    r31,`$FRAME-$SIZE_T*1`($sp)
 207         mtlr    r0
 208         addi    $sp,$sp,`$FRAME+64`
 209         blr
 210 ___
 211
 212 # PowerPC specification allows an implementation to be ill-behaved
 213 # upon unaligned access which crosses page boundary. "Better safe
 214 # than sorry" principle makes me treat it specially. But I don't
 215 # look for particular offending word, but rather for 64-byte input
 216 # block which crosses the boundary. Once found that block is aligned
 217 # and hashed separately...
 218 $code.=<<___;
 219 .align  4
 220 Lunaligned:
 221         subfic  $t1,$inp,4096
 222         andi.   $t1,$t1,4095    ; distance to closest page boundary
 223         srwi.   $t1,$t1,6       ; t1/=64
 224         beq     Lcross_page
 225         $UCMP   $num,$t1
 226         ble-    Laligned        ; didn't cross the page boundary
 227         mtctr   $t1
 228         subfc   $num,$t1,$num
 229         bl      Lsha1_block_private
 230 Lcross_page:
 231         li      $t1,16
 232         mtctr   $t1
 233         addi    r20,$sp,$FRAME  ; spot below the frame
 234 Lmemcpy:
 235         lbz     r16,0($inp)
 236         lbz     r17,1($inp)
 237         lbz     r18,2($inp)
 238         lbz     r19,3($inp)
 239         addi    $inp,$inp,4
 240         stb     r16,0(r20)
 241         stb     r17,1(r20)
 242         stb     r18,2(r20)
 243         stb     r19,3(r20)
 244         addi    r20,r20,4
 245         bdnz    Lmemcpy
 246
 247         $PUSH   $inp,`$FRAME-$SIZE_T*19`($sp)
 248         li      $t1,1
 249         addi    $inp,$sp,$FRAME
 250         mtctr   $t1
 251         bl      Lsha1_block_private
 252         $POP    $inp,`$FRAME-$SIZE_T*19`($sp)
 253         addic.  $num,$num,-1
 254         bne-    Lunaligned
 255         b       Ldone
 256 ___
 257
 258 # This is private block function, which uses tailored calling
 259 # interface, namely upon entry SHA_CTX is pre-loaded to given
 260 # registers and counter register contains amount of chunks to
 261 # digest...
 262 $code.=<<___;
 263 .align  4
 264 Lsha1_block_private:
 265 ___
 266 $code.=<<___;   # load K_00_19
 267         lis     $K,0x5a82
 268         ori     $K,$K,0x7999
 269 ___
 270 for($i=0;$i<20;$i++)    { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
 271 $code.=<<___;   # load K_20_39
 272         lis     $K,0x6ed9
 273         ori     $K,$K,0xeba1
 274 ___
 275 for(;$i<40;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
 276 $code.=<<___;   # load K_40_59
 277         lis     $K,0x8f1b
 278         ori     $K,$K,0xbcdc
 279 ___
 280 for(;$i<60;$i++)        { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
 281 $code.=<<___;   # load K_60_79
 282         lis     $K,0xca62
 283         ori     $K,$K,0xc1d6
 284 ___
 285 for(;$i<80;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
 286 $code.=<<___;
 287         add     r16,r16,$E
 288         add     r17,r17,$T
 289         add     r18,r18,$A
 290         add     r19,r19,$B
 291         add     r20,r20,$C
 292         stw     r16,0($ctx)
 293         mr      $A,r16
 294         stw     r17,4($ctx)
 295         mr      $B,r17
 296         stw     r18,8($ctx)
 297         mr      $C,r18
 298         stw     r19,12($ctx)
 299         mr      $D,r19
 300         stw     r20,16($ctx)
 301         mr      $E,r20
 302         addi    $inp,$inp,`16*4`
 303         bdnz-   Lsha1_block_private
 304         blr
 305 ___
 306
 307 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 308 print $code;
 309 close STDOUT;