crypto/sha/asm/sha512-ppc.pl

   1 #!/usr/bin/env perl
   2
   3 # ====================================================================
   4 # Written by Andy Polyakov <appro@openssl.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 # I let hardware handle unaligned input, except on page boundaries
  11 # (see below for details). Otherwise straightforward implementation
  12 # with X vector in register bank. The module is big-endian [which is
  13 # not big deal as there're no little-endian targets left around].
  14
  15 #                       sha256          |       sha512
  16 #                       -m64    -m32    |       -m64    -m32
  17 # --------------------------------------+-----------------------
  18 # PPC970,gcc-4.0.0      +50%    +38%    |       +40%    +410%(*)
  19 # Power6,xlc-7          +150%   +90%    |       +100%   +430%(*)
  20 #
  21 # (*)   64-bit code in 32-bit application context, which actually is
  22 #       on TODO list. It should be noted that for safe deployment in
  23 #       32-bit *mutli-threaded* context asyncronous signals should be
  24 #       blocked upon entry to SHA512 block routine. This is because
  25 #       32-bit signaling procedure invalidates upper halves of GPRs.
  26 #       Context switch procedure preserves them, but not signaling:-(
  27
  28 # Second version is true multi-thread safe. Trouble with the original
  29 # version was that it was using thread local storage pointer register.
  30 # Well, it scrupulously preserved it, but the problem would arise the
  31 # moment asynchronous signal was delivered and signal handler would
  32 # dereference the TLS pointer. While it's never the case in openssl
  33 # application or test suite, we have to respect this scenario and not
  34 # use TLS pointer register. Alternative would be to require caller to
  35 # block signals prior calling this routine. For the record, in 32-bit
  36 # context R2 serves as TLS pointer, while in 64-bit context - R13.
  37
  38 $flavour=shift;
  39 $output =shift;
  40
  41 if ($flavour =~ /64/) {
  42         $SIZE_T=8;
  43         $LRSAVE=2*$SIZE_T;
  44         $STU="stdu";
  45         $UCMP="cmpld";
  46         $SHL="sldi";
  47         $POP="ld";
  48         $PUSH="std";
  49 } elsif ($flavour =~ /32/) {
  50         $SIZE_T=4;
  51         $LRSAVE=$SIZE_T;
  52         $STU="stwu";
  53         $UCMP="cmplw";
  54         $SHL="slwi";
  55         $POP="lwz";
  56         $PUSH="stw";
  57 } else { die "nonsense $flavour"; }
  58
  59 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  60 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
  61 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
  62 die "can't locate ppc-xlate.pl";
  63
  64 open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!";
  65
  66 if ($output =~ /512/) {
  67         $func="sha512_block_data_order";
  68         $SZ=8;
  69         @Sigma0=(28,34,39);
  70         @Sigma1=(14,18,41);
  71         @sigma0=(1,  8, 7);
  72         @sigma1=(19,61, 6);
  73         $rounds=80;
  74         $LD="ld";
  75         $ST="std";
  76         $ROR="rotrdi";
  77         $SHR="srdi";
  78 } else {
  79         $func="sha256_block_data_order";
  80         $SZ=4;
  81         @Sigma0=( 2,13,22);
  82         @Sigma1=( 6,11,25);
  83         @sigma0=( 7,18, 3);
  84         @sigma1=(17,19,10);
  85         $rounds=64;
  86         $LD="lwz";
  87         $ST="stw";
  88         $ROR="rotrwi";
  89         $SHR="srwi";
  90 }
  91
  92 $FRAME=32*$SIZE_T+16*$SZ;
  93 $LOCALS=6*$SIZE_T;
  94
  95 $sp ="r1";
  96 $toc="r2";
  97 $ctx="r3";      # zapped by $a0
  98 $inp="r4";      # zapped by $a1
  99 $num="r5";      # zapped by $t0
 100
 101 $T  ="r0";
 102 $a0 ="r3";
 103 $a1 ="r4";
 104 $t0 ="r5";
 105 $t1 ="r6";
 106 $Tbl="r7";
 107
 108 $A  ="r8";
 109 $B  ="r9";
 110 $C  ="r10";
 111 $D  ="r11";
 112 $E  ="r12";
 113 $F  =$t1;       $t1 = "r0";     # stay away from "r13";
 114 $G  ="r14";
 115 $H  ="r15";
 116
 117 @V=($A,$B,$C,$D,$E,$F,$G,$H);
 118 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
 119     "r24","r25","r26","r27","r28","r29","r30","r31");
 120
 121 $inp="r31" if($SZ==4 || $SIZE_T==8);    # reassigned $inp! aliases with @X[15]
 122
 123 sub ROUND_00_15 {
 124 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 125 $code.=<<___;
 126         $ROR    $a0,$e,$Sigma1[0]
 127         $ROR    $a1,$e,$Sigma1[1]
 128         and     $t0,$f,$e
 129         xor     $a0,$a0,$a1
 130         add     $h,$h,$t1
 131         andc    $t1,$g,$e
 132         $ROR    $a1,$a1,`$Sigma1[2]-$Sigma1[1]`
 133         or      $t0,$t0,$t1             ; Ch(e,f,g)
 134         add     $h,$h,@X[$i%16]
 135         xor     $a0,$a0,$a1             ; Sigma1(e)
 136         add     $h,$h,$t0
 137         add     $h,$h,$a0
 138
 139         $ROR    $a0,$a,$Sigma0[0]
 140         $ROR    $a1,$a,$Sigma0[1]
 141         and     $t0,$a,$b
 142         and     $t1,$a,$c
 143         xor     $a0,$a0,$a1
 144         $ROR    $a1,$a1,`$Sigma0[2]-$Sigma0[1]`
 145         xor     $t0,$t0,$t1
 146         and     $t1,$b,$c
 147         xor     $a0,$a0,$a1             ; Sigma0(a)
 148         add     $d,$d,$h
 149         xor     $t0,$t0,$t1             ; Maj(a,b,c)
 150 ___
 151 $code.=<<___ if ($i<15);
 152         $LD     $t1,`($i+1)*$SZ`($Tbl)
 153 ___
 154 $code.=<<___;
 155         add     $h,$h,$a0
 156         add     $h,$h,$t0
 157
 158 ___
 159 }
 160
 161 sub ROUND_16_xx {
 162 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 163 $i-=16;
 164 $code.=<<___;
 165         $ROR    $a0,@X[($i+1)%16],$sigma0[0]
 166         $ROR    $a1,@X[($i+1)%16],$sigma0[1]
 167         $ROR    $t0,@X[($i+14)%16],$sigma1[0]
 168         $ROR    $t1,@X[($i+14)%16],$sigma1[1]
 169         xor     $a0,$a0,$a1
 170         $SHR    $a1,@X[($i+1)%16],$sigma0[2]
 171         xor     $t0,$t0,$t1
 172         $SHR    $t1,@X[($i+14)%16],$sigma1[2]
 173         add     @X[$i],@X[$i],@X[($i+9)%16]
 174         xor     $a0,$a0,$a1             ; sigma0(X[(i+1)&0x0f])
 175         xor     $t0,$t0,$t1             ; sigma1(X[(i+14)&0x0f])
 176         $LD     $t1,`$i*$SZ`($Tbl)
 177         add     @X[$i],@X[$i],$a0
 178         add     @X[$i],@X[$i],$t0
 179 ___
 180 &ROUND_00_15($i+16,$a,$b,$c,$d,$e,$f,$g,$h);
 181 }
 182
 183 $code=<<___;
 184 .machine        "any"
 185 .text
 186
 187 .globl  $func
 188 .align  6
 189 $func:
 190         $STU    $sp,-$FRAME($sp)
 191         mflr    r0
 192         $SHL    $num,$num,`log(16*$SZ)/log(2)`
 193
 194         $PUSH   $ctx,`$FRAME-$SIZE_T*22`($sp)
 195
 196         $PUSH   r14,`$FRAME-$SIZE_T*18`($sp)
 197         $PUSH   r15,`$FRAME-$SIZE_T*17`($sp)
 198         $PUSH   r16,`$FRAME-$SIZE_T*16`($sp)
 199         $PUSH   r17,`$FRAME-$SIZE_T*15`($sp)
 200         $PUSH   r18,`$FRAME-$SIZE_T*14`($sp)
 201         $PUSH   r19,`$FRAME-$SIZE_T*13`($sp)
 202         $PUSH   r20,`$FRAME-$SIZE_T*12`($sp)
 203         $PUSH   r21,`$FRAME-$SIZE_T*11`($sp)
 204         $PUSH   r22,`$FRAME-$SIZE_T*10`($sp)
 205         $PUSH   r23,`$FRAME-$SIZE_T*9`($sp)
 206         $PUSH   r24,`$FRAME-$SIZE_T*8`($sp)
 207         $PUSH   r25,`$FRAME-$SIZE_T*7`($sp)
 208         $PUSH   r26,`$FRAME-$SIZE_T*6`($sp)
 209         $PUSH   r27,`$FRAME-$SIZE_T*5`($sp)
 210         $PUSH   r28,`$FRAME-$SIZE_T*4`($sp)
 211         $PUSH   r29,`$FRAME-$SIZE_T*3`($sp)
 212         $PUSH   r30,`$FRAME-$SIZE_T*2`($sp)
 213         $PUSH   r31,`$FRAME-$SIZE_T*1`($sp)
 214         $PUSH   r0,`$FRAME+$LRSAVE`($sp)
 215 ___
 216
 217 if ($SZ==4 || $SIZE_T==8) {
 218 $code.=<<___;
 219         $LD     $A,`0*$SZ`($ctx)
 220         mr      $inp,r4                         ; incarnate $inp
 221         $LD     $B,`1*$SZ`($ctx)
 222         $LD     $C,`2*$SZ`($ctx)
 223         $LD     $D,`3*$SZ`($ctx)
 224         $LD     $E,`4*$SZ`($ctx)
 225         $LD     $F,`5*$SZ`($ctx)
 226         $LD     $G,`6*$SZ`($ctx)
 227         $LD     $H,`7*$SZ`($ctx)
 228 ___
 229 } else {
 230   for ($i=16;$i<32;$i++) {
 231     $code.=<<___;
 232         lwz     r$i,`4*($i-16)`($ctx)
 233 ___
 234   }
 235 }
 236
 237 $code.=<<___;
 238         bl      LPICmeup
 239 LPICedup:
 240         andi.   r0,$inp,3
 241         bne     Lunaligned
 242 Laligned:
 243         add     $num,$inp,$num
 244         $PUSH   $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 245         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 246         bl      Lsha2_block_private
 247         b       Ldone
 248
 249 ; PowerPC specification allows an implementation to be ill-behaved
 250 ; upon unaligned access which crosses page boundary. "Better safe
 251 ; than sorry" principle makes me treat it specially. But I don't
 252 ; look for particular offending word, but rather for the input
 253 ; block which crosses the boundary. Once found that block is aligned
 254 ; and hashed separately...
 255 .align  4
 256 Lunaligned:
 257         subfic  $t1,$inp,4096
 258         andi.   $t1,$t1,`4096-16*$SZ`   ; distance to closest page boundary
 259         beq     Lcross_page
 260         $UCMP   $num,$t1
 261         ble-    Laligned                ; didn't cross the page boundary
 262         subfc   $num,$t1,$num
 263         add     $t1,$inp,$t1
 264         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real remaining num
 265         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; intermediate end pointer
 266         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 267         bl      Lsha2_block_private
 268         ; $inp equals to the intermediate end pointer here
 269         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real remaining num
 270 Lcross_page:
 271         li      $t1,`16*$SZ/4`
 272         mtctr   $t1
 273 ___
 274 if ($SZ==4 || $SIZE_T==8) {
 275 $code.=<<___;
 276         addi    r20,$sp,$LOCALS                 ; aligned spot below the frame
 277 Lmemcpy:
 278         lbz     r16,0($inp)
 279         lbz     r17,1($inp)
 280         lbz     r18,2($inp)
 281         lbz     r19,3($inp)
 282         addi    $inp,$inp,4
 283         stb     r16,0(r20)
 284         stb     r17,1(r20)
 285         stb     r18,2(r20)
 286         stb     r19,3(r20)
 287         addi    r20,r20,4
 288         bdnz    Lmemcpy
 289 ___
 290 } else {
 291 $code.=<<___;
 292         addi    r12,$sp,$LOCALS                 ; aligned spot below the frame
 293 Lmemcpy:
 294         lbz     r8,0($inp)
 295         lbz     r9,1($inp)
 296         lbz     r10,2($inp)
 297         lbz     r11,3($inp)
 298         addi    $inp,$inp,4
 299         stb     r8,0(r12)
 300         stb     r9,1(r12)
 301         stb     r10,2(r12)
 302         stb     r11,3(r12)
 303         addi    r12,r12,4
 304         bdnz    Lmemcpy
 305 ___
 306 }
 307
 308 $code.=<<___;
 309         $PUSH   $inp,`$FRAME-$SIZE_T*26`($sp)   ; save real inp
 310         addi    $t1,$sp,`$LOCALS+16*$SZ`        ; fictitious end pointer
 311         addi    $inp,$sp,$LOCALS                ; fictitious inp pointer
 312         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real num
 313         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; end pointer
 314         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 315         bl      Lsha2_block_private
 316         $POP    $inp,`$FRAME-$SIZE_T*26`($sp)   ; restore real inp
 317         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real num
 318         addic.  $num,$num,`-16*$SZ`             ; num--
 319         bne-    Lunaligned
 320
 321 Ldone:
 322         $POP    r0,`$FRAME+$LRSAVE`($sp)
 323         $POP    r14,`$FRAME-$SIZE_T*18`($sp)
 324         $POP    r15,`$FRAME-$SIZE_T*17`($sp)
 325         $POP    r16,`$FRAME-$SIZE_T*16`($sp)
 326         $POP    r17,`$FRAME-$SIZE_T*15`($sp)
 327         $POP    r18,`$FRAME-$SIZE_T*14`($sp)
 328         $POP    r19,`$FRAME-$SIZE_T*13`($sp)
 329         $POP    r20,`$FRAME-$SIZE_T*12`($sp)
 330         $POP    r21,`$FRAME-$SIZE_T*11`($sp)
 331         $POP    r22,`$FRAME-$SIZE_T*10`($sp)
 332         $POP    r23,`$FRAME-$SIZE_T*9`($sp)
 333         $POP    r24,`$FRAME-$SIZE_T*8`($sp)
 334         $POP    r25,`$FRAME-$SIZE_T*7`($sp)
 335         $POP    r26,`$FRAME-$SIZE_T*6`($sp)
 336         $POP    r27,`$FRAME-$SIZE_T*5`($sp)
 337         $POP    r28,`$FRAME-$SIZE_T*4`($sp)
 338         $POP    r29,`$FRAME-$SIZE_T*3`($sp)
 339         $POP    r30,`$FRAME-$SIZE_T*2`($sp)
 340         $POP    r31,`$FRAME-$SIZE_T*1`($sp)
 341         mtlr    r0
 342         addi    $sp,$sp,$FRAME
 343         blr
 344         .long   0
 345         .byte   0,12,4,1,0x80,18,3,0
 346         .long   0
 347 .size   $func,.-$func
 348 ___
 349
 350 if ($SZ==4 || $SIZE_T==8) {
 351 $code.=<<___;
 352 .align  4
 353 Lsha2_block_private:
 354         $LD     $t1,0($Tbl)
 355 ___
 356 for($i=0;$i<16;$i++) {
 357 $code.=<<___ if ($SZ==4);
 358         lwz     @X[$i],`$i*$SZ`($inp)
 359 ___
 360 # 64-bit loads are split to 2x32-bit ones, as CPU can't handle
 361 # unaligned 64-bit loads, only 32-bit ones...
 362 $code.=<<___ if ($SZ==8);
 363         lwz     $t0,`$i*$SZ`($inp)
 364         lwz     @X[$i],`$i*$SZ+4`($inp)
 365         insrdi  @X[$i],$t0,32,0
 366 ___
 367         &ROUND_00_15($i,@V);
 368         unshift(@V,pop(@V));
 369 }
 370 $code.=<<___;
 371         li      $t0,`$rounds/16-1`
 372         mtctr   $t0
 373 .align  4
 374 Lrounds:
 375         addi    $Tbl,$Tbl,`16*$SZ`
 376 ___
 377 for(;$i<32;$i++) {
 378         &ROUND_16_xx($i,@V);
 379         unshift(@V,pop(@V));
 380 }
 381 $code.=<<___;
 382         bdnz-   Lrounds
 383
 384         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 385         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 386         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 387         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 388
 389         $LD     r16,`0*$SZ`($ctx)
 390         $LD     r17,`1*$SZ`($ctx)
 391         $LD     r18,`2*$SZ`($ctx)
 392         $LD     r19,`3*$SZ`($ctx)
 393         $LD     r20,`4*$SZ`($ctx)
 394         $LD     r21,`5*$SZ`($ctx)
 395         $LD     r22,`6*$SZ`($ctx)
 396         addi    $inp,$inp,`16*$SZ`              ; advance inp
 397         $LD     r23,`7*$SZ`($ctx)
 398         add     $A,$A,r16
 399         add     $B,$B,r17
 400         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 401         add     $C,$C,r18
 402         $ST     $A,`0*$SZ`($ctx)
 403         add     $D,$D,r19
 404         $ST     $B,`1*$SZ`($ctx)
 405         add     $E,$E,r20
 406         $ST     $C,`2*$SZ`($ctx)
 407         add     $F,$F,r21
 408         $ST     $D,`3*$SZ`($ctx)
 409         add     $G,$G,r22
 410         $ST     $E,`4*$SZ`($ctx)
 411         add     $H,$H,r23
 412         $ST     $F,`5*$SZ`($ctx)
 413         $ST     $G,`6*$SZ`($ctx)
 414         $UCMP   $inp,$num
 415         $ST     $H,`7*$SZ`($ctx)
 416         bne     Lsha2_block_private
 417         blr
 418         .long   0
 419         .byte   0,12,0x14,0,0,0,0,0
 420 ___
 421 } else {
 422 ########################################################################
 423 # SHA512 for PPC32, X vector is off-loaded to stack...
 424 #
 425 #                       |       sha512
 426 #                       |       -m32
 427 # ----------------------+-----------------------
 428 # PPC74x0,gcc-4.0.1     |       +48%
 429 # POWER6,gcc-4.4.6      |       +124%(*)
 430 # POWER7,gcc-4.4.6      |       +79%(*)
 431 # e300,gcc-4.1.0        |       +167%
 432 #
 433 # (*)   ~1/3 of -m64 result [and ~20% better than -m32 code generated
 434 #       by xlc-12.1]
 435
 436 my $XOFF=$LOCALS;
 437
 438 my @V=map("r$_",(16..31));      # A..H
 439
 440 my ($s0,$s1,$t0,$t1,$t2,$t3,$a0,$a1,$a2,$a3)=map("r$_",(0,5,6,8..12,14,15));
 441 my ($x0,$x1)=("r3","r4");       # zaps $ctx and $inp
 442
 443 sub ROUND_00_15_ppc32 {
 444 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 445         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 446
 447 $code.=<<___;
 448         lwz     $t2,`$SZ*($i%16)+4`($Tbl)
 449          xor    $a0,$flo,$glo
 450         lwz     $t3,`$SZ*($i%16)+0`($Tbl)
 451          xor    $a1,$fhi,$ghi
 452         addc    $hlo,$hlo,$t0                   ; h+=x[i]
 453         stw     $t0,`$XOFF+0+$SZ*($i%16)`($sp)  ; save x[i]
 454
 455         srwi    $s0,$elo,$Sigma1[0]
 456         srwi    $s1,$ehi,$Sigma1[0]
 457          and    $a0,$a0,$elo
 458         adde    $hhi,$hhi,$t1
 459          and    $a1,$a1,$ehi
 460         stw     $t1,`$XOFF+4+$SZ*($i%16)`($sp)
 461         srwi    $t0,$elo,$Sigma1[1]
 462         srwi    $t1,$ehi,$Sigma1[1]
 463          addc   $hlo,$hlo,$t2                   ; h+=K512[i]
 464         insrwi  $s0,$ehi,$Sigma1[0],0
 465         insrwi  $s1,$elo,$Sigma1[0],0
 466          xor    $a0,$a0,$glo                    ; Ch(e,f,g)
 467          adde   $hhi,$hhi,$t3
 468          xor    $a1,$a1,$ghi
 469         insrwi  $t0,$ehi,$Sigma1[1],0
 470         insrwi  $t1,$elo,$Sigma1[1],0
 471          addc   $hlo,$hlo,$a0                   ; h+=Ch(e,f,g)
 472         srwi    $t2,$ehi,$Sigma1[2]-32
 473         srwi    $t3,$elo,$Sigma1[2]-32
 474         xor     $s0,$s0,$t0
 475         xor     $s1,$s1,$t1
 476         insrwi  $t2,$elo,$Sigma1[2]-32,0
 477         insrwi  $t3,$ehi,$Sigma1[2]-32,0
 478          xor    $a0,$alo,$blo                   ; a^b, b^c in next round
 479          adde   $hhi,$hhi,$a1
 480          xor    $a1,$ahi,$bhi
 481         xor     $s0,$s0,$t2                     ; Sigma1(e)
 482         xor     $s1,$s1,$t3
 483
 484         srwi    $t0,$alo,$Sigma0[0]
 485          and    $a2,$a2,$a0
 486          addc   $hlo,$hlo,$s0                   ; h+=Sigma1(e)
 487          and    $a3,$a3,$a1
 488         srwi    $t1,$ahi,$Sigma0[0]
 489         srwi    $s0,$ahi,$Sigma0[1]-32
 490          adde   $hhi,$hhi,$s1
 491         srwi    $s1,$alo,$Sigma0[1]-32
 492         insrwi  $t0,$ahi,$Sigma0[0],0
 493         insrwi  $t1,$alo,$Sigma0[0],0
 494          xor    $a2,$a2,$blo                    ; Maj(a,b,c)
 495          addc   $dlo,$dlo,$hlo                  ; d+=h
 496          xor    $a3,$a3,$bhi
 497         insrwi  $s0,$alo,$Sigma0[1]-32,0
 498         insrwi  $s1,$ahi,$Sigma0[1]-32,0
 499          adde   $dhi,$dhi,$hhi
 500         srwi    $t2,$ahi,$Sigma0[2]-32
 501         srwi    $t3,$alo,$Sigma0[2]-32
 502         xor     $s0,$s0,$t0
 503          addc   $hlo,$hlo,$a2                   ; h+=Maj(a,b,c)
 504         xor     $s1,$s1,$t1
 505         insrwi  $t2,$alo,$Sigma0[2]-32,0
 506         insrwi  $t3,$ahi,$Sigma0[2]-32,0
 507          adde   $hhi,$hhi,$a3
 508 ___
 509 $code.=<<___ if ($i>=15);
 510         lwz     $t0,`$XOFF+0+$SZ*(($i+2)%16)`($sp)
 511         lwz     $t1,`$XOFF+4+$SZ*(($i+2)%16)`($sp)
 512 ___
 513 $code.=<<___ if ($i<15);
 514         lwz     $t1,`$SZ*($i+1)+0`($inp)
 515         lwz     $t0,`$SZ*($i+1)+4`($inp)
 516 ___
 517 $code.=<<___;
 518         xor     $s0,$s0,$t2                     ; Sigma0(a)
 519         xor     $s1,$s1,$t3
 520         addc    $hlo,$hlo,$s0                   ; h+=Sigma0(a)
 521         adde    $hhi,$hhi,$s1
 522 ___
 523 $code.=<<___ if ($i==15);
 524         lwz     $x0,`$XOFF+0+$SZ*(($i+1)%16)`($sp)
 525         lwz     $x1,`$XOFF+4+$SZ*(($i+1)%16)`($sp)
 526 ___
 527 }
 528 sub ROUND_16_xx_ppc32 {
 529 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 530         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 531
 532 $code.=<<___;
 533         srwi    $s0,$t0,$sigma0[0]
 534         srwi    $s1,$t1,$sigma0[0]
 535         srwi    $t2,$t0,$sigma0[1]
 536         srwi    $t3,$t1,$sigma0[1]
 537         insrwi  $s0,$t1,$sigma0[0],0
 538         insrwi  $s1,$t0,$sigma0[0],0
 539         srwi    $a0,$t0,$sigma0[2]
 540         insrwi  $t2,$t1,$sigma0[1],0
 541         insrwi  $t3,$t0,$sigma0[1],0
 542         insrwi  $a0,$t1,$sigma0[2],0
 543         xor     $s0,$s0,$t2
 544          lwz    $t2,`$XOFF+0+$SZ*(($i+14)%16)`($sp)
 545         srwi    $a1,$t1,$sigma0[2]
 546         xor     $s1,$s1,$t3
 547          lwz    $t3,`$XOFF+4+$SZ*(($i+14)%16)`($sp)
 548         xor     $a0,$a0,$s0
 549          srwi   $s0,$t2,$sigma1[0]
 550         xor     $a1,$a1,$s1
 551          srwi   $s1,$t3,$sigma1[0]
 552         addc    $x0,$x0,$a0                     ; x[i]+=sigma0(x[i+1])
 553          srwi   $a0,$t3,$sigma1[1]-32
 554         insrwi  $s0,$t3,$sigma1[0],0
 555         insrwi  $s1,$t2,$sigma1[0],0
 556         adde    $x1,$x1,$a1
 557          srwi   $a1,$t2,$sigma1[1]-32
 558
 559         insrwi  $a0,$t2,$sigma1[1]-32,0
 560         srwi    $t2,$t2,$sigma1[2]
 561         insrwi  $a1,$t3,$sigma1[1]-32,0
 562         insrwi  $t2,$t3,$sigma1[2],0
 563         xor     $s0,$s0,$a0
 564          lwz    $a0,`$XOFF+0+$SZ*(($i+9)%16)`($sp)
 565         srwi    $t3,$t3,$sigma1[2]
 566         xor     $s1,$s1,$a1
 567          lwz    $a1,`$XOFF+4+$SZ*(($i+9)%16)`($sp)
 568         xor     $s0,$s0,$t2
 569          addc   $x0,$x0,$a0                     ; x[i]+=x[i+9]
 570         xor     $s1,$s1,$t3
 571          adde   $x1,$x1,$a1
 572         addc    $x0,$x0,$s0                     ; x[i]+=sigma1(x[i+14])
 573         adde    $x1,$x1,$s1
 574 ___
 575         ($t0,$t1,$x0,$x1) = ($x0,$x1,$t0,$t1);
 576         &ROUND_00_15_ppc32(@_);
 577 }
 578
 579 $code.=<<___;
 580 .align  4
 581 Lsha2_block_private:
 582         lwz     $t1,0($inp)
 583         xor     $a2,@V[3],@V[5]         ; B^C, magic seed
 584         lwz     $t0,4($inp)
 585         xor     $a3,@V[2],@V[4]
 586 ___
 587 for($i=0;$i<16;$i++) {
 588         &ROUND_00_15_ppc32($i,@V);
 589         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 590         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 591 }
 592 $code.=<<___;
 593         li      $a0,`$rounds/16-1`
 594         mtctr   $a0
 595 .align  4
 596 Lrounds:
 597         addi    $Tbl,$Tbl,`16*$SZ`
 598 ___
 599 for(;$i<32;$i++) {
 600         &ROUND_16_xx_ppc32($i,@V);
 601         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 602         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 603 }
 604 $code.=<<___;
 605         bdnz-   Lrounds
 606
 607         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 608         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 609         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 610         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 611
 612         lwz     $t0,0($ctx)
 613         lwz     $t1,4($ctx)
 614         lwz     $t2,8($ctx)
 615         lwz     $t3,12($ctx)
 616         lwz     $a0,16($ctx)
 617         lwz     $a1,20($ctx)
 618         lwz     $a2,24($ctx)
 619         addc    @V[1],@V[1],$t1
 620         lwz     $a3,28($ctx)
 621         adde    @V[0],@V[0],$t0
 622         lwz     $t0,32($ctx)
 623         addc    @V[3],@V[3],$t3
 624         lwz     $t1,36($ctx)
 625         adde    @V[2],@V[2],$t2
 626         lwz     $t2,40($ctx)
 627         addc    @V[5],@V[5],$a1
 628         lwz     $t3,44($ctx)
 629         adde    @V[4],@V[4],$a0
 630         lwz     $a0,48($ctx)
 631         addc    @V[7],@V[7],$a3
 632         lwz     $a1,52($ctx)
 633         adde    @V[6],@V[6],$a2
 634         lwz     $a2,56($ctx)
 635         addc    @V[9],@V[9],$t1
 636         lwz     $a3,60($ctx)
 637         adde    @V[8],@V[8],$t0
 638         stw     @V[0],0($ctx)
 639         stw     @V[1],4($ctx)
 640         addc    @V[11],@V[11],$t3
 641         stw     @V[2],8($ctx)
 642         stw     @V[3],12($ctx)
 643         adde    @V[10],@V[10],$t2
 644         stw     @V[4],16($ctx)
 645         stw     @V[5],20($ctx)
 646         addc    @V[13],@V[13],$a1
 647         stw     @V[6],24($ctx)
 648         stw     @V[7],28($ctx)
 649         adde    @V[12],@V[12],$a0
 650         stw     @V[8],32($ctx)
 651         stw     @V[9],36($ctx)
 652         addc    @V[15],@V[15],$a3
 653         stw     @V[10],40($ctx)
 654         stw     @V[11],44($ctx)
 655         adde    @V[14],@V[14],$a2
 656         stw     @V[12],48($ctx)
 657         stw     @V[13],52($ctx)
 658         stw     @V[14],56($ctx)
 659         stw     @V[15],60($ctx)
 660
 661         addi    $inp,$inp,`16*$SZ`              ; advance inp
 662         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 663         $UCMP   $inp,$num
 664         bne     Lsha2_block_private
 665         blr
 666         .long   0
 667         .byte   0,12,0x14,0,0,0,0,0
 668 ___
 669 }
 670
 671 # Ugly hack here, because PPC assembler syntax seem to vary too
 672 # much from platforms to platform...
 673 $code.=<<___;
 674 .align  6
 675 LPICmeup:
 676         mflr    r0
 677         bcl     20,31,\$+4
 678         mflr    $Tbl    ; vvvvvv "distance" between . and 1st data entry
 679         addi    $Tbl,$Tbl,`64-8`
 680         mtlr    r0
 681         blr
 682         .long   0
 683         .byte   0,12,0x14,0,0,0,0,0
 684         .space  `64-9*4`
 685 ___
 686 $code.=<<___ if ($SZ==8);
 687         .long   0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd
 688         .long   0xb5c0fbcf,0xec4d3b2f,0xe9b5dba5,0x8189dbbc
 689         .long   0x3956c25b,0xf348b538,0x59f111f1,0xb605d019
 690         .long   0x923f82a4,0xaf194f9b,0xab1c5ed5,0xda6d8118
 691         .long   0xd807aa98,0xa3030242,0x12835b01,0x45706fbe
 692         .long   0x243185be,0x4ee4b28c,0x550c7dc3,0xd5ffb4e2
 693         .long   0x72be5d74,0xf27b896f,0x80deb1fe,0x3b1696b1
 694         .long   0x9bdc06a7,0x25c71235,0xc19bf174,0xcf692694
 695         .long   0xe49b69c1,0x9ef14ad2,0xefbe4786,0x384f25e3
 696         .long   0x0fc19dc6,0x8b8cd5b5,0x240ca1cc,0x77ac9c65
 697         .long   0x2de92c6f,0x592b0275,0x4a7484aa,0x6ea6e483
 698         .long   0x5cb0a9dc,0xbd41fbd4,0x76f988da,0x831153b5
 699         .long   0x983e5152,0xee66dfab,0xa831c66d,0x2db43210
 700         .long   0xb00327c8,0x98fb213f,0xbf597fc7,0xbeef0ee4
 701         .long   0xc6e00bf3,0x3da88fc2,0xd5a79147,0x930aa725
 702         .long   0x06ca6351,0xe003826f,0x14292967,0x0a0e6e70
 703         .long   0x27b70a85,0x46d22ffc,0x2e1b2138,0x5c26c926
 704         .long   0x4d2c6dfc,0x5ac42aed,0x53380d13,0x9d95b3df
 705         .long   0x650a7354,0x8baf63de,0x766a0abb,0x3c77b2a8
 706         .long   0x81c2c92e,0x47edaee6,0x92722c85,0x1482353b
 707         .long   0xa2bfe8a1,0x4cf10364,0xa81a664b,0xbc423001
 708         .long   0xc24b8b70,0xd0f89791,0xc76c51a3,0x0654be30
 709         .long   0xd192e819,0xd6ef5218,0xd6990624,0x5565a910
 710         .long   0xf40e3585,0x5771202a,0x106aa070,0x32bbd1b8
 711         .long   0x19a4c116,0xb8d2d0c8,0x1e376c08,0x5141ab53
 712         .long   0x2748774c,0xdf8eeb99,0x34b0bcb5,0xe19b48a8
 713         .long   0x391c0cb3,0xc5c95a63,0x4ed8aa4a,0xe3418acb
 714         .long   0x5b9cca4f,0x7763e373,0x682e6ff3,0xd6b2b8a3
 715         .long   0x748f82ee,0x5defb2fc,0x78a5636f,0x43172f60
 716         .long   0x84c87814,0xa1f0ab72,0x8cc70208,0x1a6439ec
 717         .long   0x90befffa,0x23631e28,0xa4506ceb,0xde82bde9
 718         .long   0xbef9a3f7,0xb2c67915,0xc67178f2,0xe372532b
 719         .long   0xca273ece,0xea26619c,0xd186b8c7,0x21c0c207
 720         .long   0xeada7dd6,0xcde0eb1e,0xf57d4f7f,0xee6ed178
 721         .long   0x06f067aa,0x72176fba,0x0a637dc5,0xa2c898a6
 722         .long   0x113f9804,0xbef90dae,0x1b710b35,0x131c471b
 723         .long   0x28db77f5,0x23047d84,0x32caab7b,0x40c72493
 724         .long   0x3c9ebe0a,0x15c9bebc,0x431d67c4,0x9c100d4c
 725         .long   0x4cc5d4be,0xcb3e42b6,0x597f299c,0xfc657e2a
 726         .long   0x5fcb6fab,0x3ad6faec,0x6c44198c,0x4a475817
 727 ___
 728 $code.=<<___ if ($SZ==4);
 729         .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
 730         .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
 731         .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
 732         .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
 733         .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
 734         .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
 735         .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
 736         .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
 737         .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
 738         .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
 739         .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
 740         .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
 741         .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
 742         .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
 743         .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
 744         .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
 745 ___
 746
 747 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 748 print $code;
 749 close STDOUT;