crypto/sha/asm/sha1-ia64.pl

   1 #!/usr/bin/env perl
   2 #
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@fy.chalmers.se> 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 # Eternal question is what's wrong with compiler generated code? The
  11 # trick is that it's possible to reduce the number of shifts required
  12 # to perform rotations by maintaining copy of 32-bit value in upper
  13 # bits of 64-bit register. Just follow mux2 and shrp instructions...
  14 # Performance under big-endian OS such as HP-UX is 179MBps*1GHz, which
  15 # is >50% better than HP C and >2x better than gcc.
  16
  17 $output = pop;
  18
  19 $code=<<___;
  20 .ident  \"sha1-ia64.s, version 1.3\"
  21 .ident  \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\"
  22 .explicit
  23
  24 ___
  25
  26
  27 if ($^O eq "hpux") {
  28     $ADDP="addp4";
  29     for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); }
  30 } else { $ADDP="add"; }
  31
  32 #$human=1;
  33 if ($human) {   # useful for visual code auditing...
  34         ($A,$B,$C,$D,$E)   = ("A","B","C","D","E");
  35         ($h0,$h1,$h2,$h3,$h4) = ("h0","h1","h2","h3","h4");
  36         ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
  37             (   "K_00_19","K_20_39","K_40_59","K_60_79" );
  38         @X= (   "X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7",
  39                 "X8", "X9","X10","X11","X12","X13","X14","X15"  );
  40 }
  41 else {
  42         ($A,$B,$C,$D,$E)   =    ("loc0","loc1","loc2","loc3","loc4");
  43         ($h0,$h1,$h2,$h3,$h4) = ("loc5","loc6","loc7","loc8","loc9");
  44         ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
  45             (   "r14", "r15", "loc10", "loc11"  );
  46         @X= (   "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
  47                 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31"  );
  48 }
  49
  50 sub BODY_00_15 {
  51 local   *code=shift;
  52 my      ($i,$a,$b,$c,$d,$e)=@_;
  53 my      $j=$i+1;
  54 my      $Xn=@X[$j%16];
  55
  56 $code.=<<___ if ($i==0);
  57 { .mmi; ld1     $X[$i]=[inp],2              // MSB
  58         ld1     tmp2=[tmp3],2           };;
  59 { .mmi; ld1     tmp0=[inp],2
  60         ld1     tmp4=[tmp3],2               // LSB
  61         dep     $X[$i]=$X[$i],tmp2,8,8  };;
  62 ___
  63 if ($i<15) {
  64         $code.=<<___;
  65 { .mmi; ld1     $Xn=[inp],2                 // forward Xload
  66         nop.m   0x0
  67         dep     tmp1=tmp0,tmp4,8,8      };;
  68 { .mmi; ld1     tmp2=[tmp3],2               // forward Xload
  69         and     tmp4=$c,$b
  70         dep     $X[$i]=$X[$i],tmp1,16,16} //;;
  71 { .mmi; add     $e=$e,$K_00_19              // e+=K_00_19
  72         andcm   tmp1=$d,$b
  73         dep.z   tmp5=$a,5,27            };; // a<<5
  74 { .mmi; add     $e=$e,$X[$i]                // e+=Xload
  75         or      tmp4=tmp4,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
  76         extr.u  tmp1=$a,27,5            };; // a>>27
  77 { .mmi; ld1     tmp0=[inp],2                // forward Xload
  78         add     $e=$e,tmp4                  // e+=F_00_19(b,c,d)
  79         shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
  80 { .mmi; ld1     tmp4=[tmp3],2               // forward Xload
  81         or      tmp5=tmp1,tmp5              // ROTATE(a,5)
  82         mux2    tmp6=$a,0x44            };; // see b in next iteration
  83 { .mii; add     $e=$e,tmp5                  // e+=ROTATE(a,5)
  84         dep     $Xn=$Xn,tmp2,8,8            // forward Xload
  85         mux2    $X[$i]=$X[$i],0x44      } //;;
  86
  87 ___
  88         }
  89 else    {
  90         $code.=<<___;
  91 { .mii; and     tmp3=$c,$b
  92         dep     tmp1=tmp0,tmp4,8,8;;
  93         dep     $X[$i]=$X[$i],tmp1,16,16} //;;
  94 { .mmi; add     $e=$e,$K_00_19              // e+=K_00_19
  95         andcm   tmp1=$d,$b
  96         dep.z   tmp5=$a,5,27            };; // a<<5
  97 { .mmi; add     $e=$e,$X[$i]                // e+=Xupdate
  98         or      tmp4=tmp3,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
  99         extr.u  tmp1=$a,27,5            }   // a>>27
 100 { .mmi; xor     $Xn=$Xn,$X[($j+2)%16]       // forward Xupdate
 101         xor     tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate
 102         nop.i   0                       };;
 103 { .mmi; add     $e=$e,tmp4                  // e+=F_00_19(b,c,d)
 104         xor     $Xn=$Xn,tmp3                // forward Xupdate
 105         shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
 106 { .mmi; or      tmp1=tmp1,tmp5              // ROTATE(a,5)
 107         mux2    tmp6=$a,0x44            };; // see b in next iteration
 108 { .mii; add     $e=$e,tmp1                  // e+=ROTATE(a,5)
 109         shrp    $Xn=$Xn,$Xn,31              // ROTATE(x[0]^x[2]^x[8]^x[13],1)
 110         mux2    $X[$i]=$X[$i],0x44      };;
 111
 112 ___
 113         }
 114 }
 115
 116 sub BODY_16_19 {
 117 local   *code=shift;
 118 my      ($i,$a,$b,$c,$d,$e)=@_;
 119 my      $j=$i+1;
 120 my      $Xn=@X[$j%16];
 121
 122 $code.=<<___;
 123 { .mib; add     $e=$e,$K_00_19              // e+=K_00_19
 124         dep.z   tmp5=$a,5,27            }   // a<<5
 125 { .mib; andcm   tmp1=$d,$b
 126         and     tmp0=$c,$b              };;
 127 { .mmi; add     $e=$e,$X[$i%16]             // e+=Xupdate
 128         or      tmp0=tmp0,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
 129         extr.u  tmp1=$a,27,5            }   // a>>27
 130 { .mmi; xor     $Xn=$Xn,$X[($j+2)%16]       // forward Xupdate
 131         xor     tmp3=$X[($j+8)%16],$X[($j+13)%16]       // forward Xupdate
 132         nop.i   0                       };;
 133 { .mmi; add     $e=$e,tmp0                  // f+=F_00_19(b,c,d)
 134         xor     $Xn=$Xn,tmp3                // forward Xupdate
 135         shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
 136 { .mmi; or      tmp1=tmp1,tmp5              // ROTATE(a,5)
 137         mux2    tmp6=$a,0x44            };; // see b in next iteration
 138 { .mii; add     $e=$e,tmp1                  // e+=ROTATE(a,5)
 139         shrp    $Xn=$Xn,$Xn,31              // ROTATE(x[0]^x[2]^x[8]^x[13],1)
 140         nop.i   0                       };;
 141
 142 ___
 143 }
 144
 145 sub BODY_20_39 {
 146 local   *code=shift;
 147 my      ($i,$a,$b,$c,$d,$e,$Konst)=@_;
 148         $Konst = $K_20_39 if (!defined($Konst));
 149 my      $j=$i+1;
 150 my      $Xn=@X[$j%16];
 151
 152 if ($i<79) {
 153 $code.=<<___;
 154 { .mib; add     $e=$e,$Konst                // e+=K_XX_XX
 155         dep.z   tmp5=$a,5,27            }   // a<<5
 156 { .mib; xor     tmp0=$c,$b
 157         xor     $Xn=$Xn,$X[($j+2)%16]   };; // forward Xupdate
 158 { .mib; add     $e=$e,$X[$i%16]             // e+=Xupdate
 159         extr.u  tmp1=$a,27,5            }   // a>>27
 160 { .mib; xor     tmp0=tmp0,$d                // F_20_39(b,c,d)=b^c^d
 161         xor     $Xn=$Xn,$X[($j+8)%16]   };; // forward Xupdate
 162 { .mmi; add     $e=$e,tmp0                  // e+=F_20_39(b,c,d)
 163         xor     $Xn=$Xn,$X[($j+13)%16]      // forward Xupdate
 164         shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
 165 { .mmi; or      tmp1=tmp1,tmp5              // ROTATE(a,5)
 166         mux2    tmp6=$a,0x44            };; // see b in next iteration
 167 { .mii; add     $e=$e,tmp1                  // e+=ROTATE(a,5)
 168         shrp    $Xn=$Xn,$Xn,31              // ROTATE(x[0]^x[2]^x[8]^x[13],1)
 169         nop.i   0                       };;
 170
 171 ___
 172 }
 173 else {
 174 $code.=<<___;
 175 { .mib; add     $e=$e,$Konst                // e+=K_60_79
 176         dep.z   tmp5=$a,5,27            }   // a<<5
 177 { .mib; xor     tmp0=$c,$b
 178         add     $h1=$h1,$a              };; // wrap up
 179 { .mib; add     $e=$e,$X[$i%16]             // e+=Xupdate
 180         extr.u  tmp1=$a,27,5            }   // a>>27
 181 { .mib; xor     tmp0=tmp0,$d                // F_20_39(b,c,d)=b^c^d
 182         add     $h3=$h3,$c              };; // wrap up
 183 { .mmi; add     $e=$e,tmp0                  // e+=F_20_39(b,c,d)
 184         or      tmp1=tmp1,tmp5              // ROTATE(a,5)
 185         shrp    $b=tmp6,tmp6,2          };; // b=ROTATE(b,30) ;;?
 186 { .mmi; add     $e=$e,tmp1                  // e+=ROTATE(a,5)
 187         add     tmp3=1,inp                  // used in unaligned codepath
 188         add     $h4=$h4,$d              };; // wrap up
 189
 190 ___
 191 }
 192 }
 193
 194 sub BODY_40_59 {
 195 local   *code=shift;
 196 my      ($i,$a,$b,$c,$d,$e)=@_;
 197 my      $j=$i+1;
 198 my      $Xn=@X[$j%16];
 199
 200 $code.=<<___;
 201 { .mib; add     $e=$e,$K_40_59              // e+=K_40_59
 202         dep.z   tmp5=$a,5,27            }   // a<<5
 203 { .mib; and     tmp1=$c,$d
 204         xor     tmp0=$c,$d              };;
 205 { .mmi; add     $e=$e,$X[$i%16]             // e+=Xupdate
 206         add     tmp5=tmp5,tmp1              // a<<5+(c&d)
 207         extr.u  tmp1=$a,27,5            }   // a>>27
 208 { .mmi; and     tmp0=tmp0,$b
 209         xor     $Xn=$Xn,$X[($j+2)%16]       // forward Xupdate
 210         xor     tmp3=$X[($j+8)%16],$X[($j+13)%16] };;   // forward Xupdate
 211 { .mmi; add     $e=$e,tmp0                  // e+=b&(c^d)
 212         add     tmp5=tmp5,tmp1              // ROTATE(a,5)+(c&d)
 213         shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
 214 { .mmi; xor     $Xn=$Xn,tmp3
 215         mux2    tmp6=$a,0x44            };; // see b in next iteration
 216 { .mii; add     $e=$e,tmp5                  // e+=ROTATE(a,5)+(c&d)
 217         shrp    $Xn=$Xn,$Xn,31              // ROTATE(x[0]^x[2]^x[8]^x[13],1)
 218         nop.i   0x0                     };;
 219
 220 ___
 221 }
 222 sub BODY_60_79  { &BODY_20_39(@_,$K_60_79); }
 223
 224 $code.=<<___;
 225 .text
 226
 227 tmp0=r8;
 228 tmp1=r9;
 229 tmp2=r10;
 230 tmp3=r11;
 231 ctx=r32;        // in0
 232 inp=r33;        // in1
 233
 234 // void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num);
 235 .global sha1_block_data_order#
 236 .proc   sha1_block_data_order#
 237 .align  32
 238 sha1_block_data_order:
 239         .prologue
 240 { .mmi; alloc   tmp1=ar.pfs,3,14,0,0
 241         $ADDP   tmp0=4,ctx
 242         .save   ar.lc,r3
 243         mov     r3=ar.lc                }
 244 { .mmi; $ADDP   ctx=0,ctx
 245         $ADDP   inp=0,inp
 246         mov     r2=pr                   };;
 247 tmp4=in2;
 248 tmp5=loc12;
 249 tmp6=loc13;
 250         .body
 251 { .mlx; ld4     $h0=[ctx],8
 252         movl    $K_00_19=0x5a827999     }
 253 { .mlx; ld4     $h1=[tmp0],8
 254         movl    $K_20_39=0x6ed9eba1     };;
 255 { .mlx; ld4     $h2=[ctx],8
 256         movl    $K_40_59=0x8f1bbcdc     }
 257 { .mlx; ld4     $h3=[tmp0]
 258         movl    $K_60_79=0xca62c1d6     };;
 259 { .mmi; ld4     $h4=[ctx],-16
 260         add     in2=-1,in2                  // adjust num for ar.lc
 261         mov     ar.ec=1                 };;
 262 { .mmi; nop.m   0
 263         add     tmp3=1,inp
 264         mov     ar.lc=in2               };; // brp.loop.imp: too far
 265
 266 .Ldtop:
 267 { .mmi; mov     $A=$h0
 268         mov     $B=$h1
 269         mux2    tmp6=$h1,0x44           }
 270 { .mmi; mov     $C=$h2
 271         mov     $D=$h3
 272         mov     $E=$h4                  };;
 273
 274 ___
 275
 276 { my $i;
 277   my @V=($A,$B,$C,$D,$E);
 278
 279         for($i=0;$i<16;$i++)    { &BODY_00_15(\$code,$i,@V); unshift(@V,pop(@V)); }
 280         for(;$i<20;$i++)        { &BODY_16_19(\$code,$i,@V); unshift(@V,pop(@V)); }
 281         for(;$i<40;$i++)        { &BODY_20_39(\$code,$i,@V); unshift(@V,pop(@V)); }
 282         for(;$i<60;$i++)        { &BODY_40_59(\$code,$i,@V); unshift(@V,pop(@V)); }
 283         for(;$i<80;$i++)        { &BODY_60_79(\$code,$i,@V); unshift(@V,pop(@V)); }
 284
 285         (($V[0] eq $A) and ($V[4] eq $E)) or die;       # double-check
 286 }
 287
 288 $code.=<<___;
 289 { .mmb; add     $h0=$h0,$A
 290         add     $h2=$h2,$C
 291         br.ctop.dptk.many       .Ldtop  };;
 292 .Ldend:
 293 { .mmi; add     tmp0=4,ctx
 294         mov     ar.lc=r3                };;
 295 { .mmi; st4     [ctx]=$h0,8
 296         st4     [tmp0]=$h1,8            };;
 297 { .mmi; st4     [ctx]=$h2,8
 298         st4     [tmp0]=$h3              };;
 299 { .mib; st4     [ctx]=$h4,-16
 300         mov     pr=r2,0x1ffff
 301         br.ret.sptk.many        b0      };;
 302 .endp   sha1_block_data_order#
 303 stringz "SHA1 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
 304 ___
 305
 306 open STDOUT,">$output" if $output;
 307 print $code;