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.
  13
  14 #                       sha256          |       sha512
  15 #                       -m64    -m32    |       -m64    -m32
  16 # --------------------------------------+-----------------------
  17 # PPC970,gcc-4.0.0      +50%    +38%    |       +40%    +410%(*)
  18 # Power6,xlc-7          +150%   +90%    |       +100%   +430%(*)
  19 #
  20 # (*)   64-bit code in 32-bit application context, which actually is
  21 #       on TODO list. It should be noted that for safe deployment in
  22 #       32-bit *mutli-threaded* context asyncronous signals should be
  23 #       blocked upon entry to SHA512 block routine. This is because
  24 #       32-bit signaling procedure invalidates upper halves of GPRs.
  25 #       Context switch procedure preserves them, but not signaling:-(
  26
  27 # Second version is true multi-thread safe. Trouble with the original
  28 # version was that it was using thread local storage pointer register.
  29 # Well, it scrupulously preserved it, but the problem would arise the
  30 # moment asynchronous signal was delivered and signal handler would
  31 # dereference the TLS pointer. While it's never the case in openssl
  32 # application or test suite, we have to respect this scenario and not
  33 # use TLS pointer register. Alternative would be to require caller to
  34 # block signals prior calling this routine. For the record, in 32-bit
  35 # context R2 serves as TLS pointer, while in 64-bit context - R13.
  36
  37 $flavour=shift;
  38 $output =shift;
  39
  40 if ($flavour =~ /64/) {
  41         $SIZE_T=8;
  42         $LRSAVE=2*$SIZE_T;
  43         $STU="stdu";
  44         $UCMP="cmpld";
  45         $SHL="sldi";
  46         $POP="ld";
  47         $PUSH="std";
  48 } elsif ($flavour =~ /32/) {
  49         $SIZE_T=4;
  50         $LRSAVE=$SIZE_T;
  51         $STU="stwu";
  52         $UCMP="cmplw";
  53         $SHL="slwi";
  54         $POP="lwz";
  55         $PUSH="stw";
  56 } else { die "nonsense $flavour"; }
  57
  58 $LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
  59
  60 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  61 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
  62 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
  63 die "can't locate ppc-xlate.pl";
  64
  65 open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!";
  66
  67 if ($output =~ /512/) {
  68         $func="sha512_block_ppc";
  69         $SZ=8;
  70         @Sigma0=(28,34,39);
  71         @Sigma1=(14,18,41);
  72         @sigma0=(1,  8, 7);
  73         @sigma1=(19,61, 6);
  74         $rounds=80;
  75         $LD="ld";
  76         $ST="std";
  77         $ROR="rotrdi";
  78         $SHR="srdi";
  79 } else {
  80         $func="sha256_block_ppc";
  81         $SZ=4;
  82         @Sigma0=( 2,13,22);
  83         @Sigma1=( 6,11,25);
  84         @sigma0=( 7,18, 3);
  85         @sigma1=(17,19,10);
  86         $rounds=64;
  87         $LD="lwz";
  88         $ST="stw";
  89         $ROR="rotrwi";
  90         $SHR="srwi";
  91 }
  92
  93 $FRAME=32*$SIZE_T+16*$SZ;
  94 $LOCALS=6*$SIZE_T;
  95
  96 $sp ="r1";
  97 $toc="r2";
  98 $ctx="r3";      # zapped by $a0
  99 $inp="r4";      # zapped by $a1
 100 $num="r5";      # zapped by $t0
 101
 102 $T  ="r0";
 103 $a0 ="r3";
 104 $a1 ="r4";
 105 $t0 ="r5";
 106 $t1 ="r6";
 107 $Tbl="r7";
 108
 109 $A  ="r8";
 110 $B  ="r9";
 111 $C  ="r10";
 112 $D  ="r11";
 113 $E  ="r12";
 114 $F  =$t1;       $t1 = "r0";     # stay away from "r13";
 115 $G  ="r14";
 116 $H  ="r15";
 117
 118 @V=($A,$B,$C,$D,$E,$F,$G,$H);
 119 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
 120     "r24","r25","r26","r27","r28","r29","r30","r31");
 121
 122 $inp="r31" if($SZ==4 || $SIZE_T==8);    # reassigned $inp! aliases with @X[15]
 123
 124 sub ROUND_00_15 {
 125 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 126 $code.=<<___;
 127         $ROR    $a0,$e,$Sigma1[0]
 128         $ROR    $a1,$e,$Sigma1[1]
 129         and     $t0,$f,$e
 130         xor     $a0,$a0,$a1
 131         add     $h,$h,$t1
 132         andc    $t1,$g,$e
 133         $ROR    $a1,$a1,`$Sigma1[2]-$Sigma1[1]`
 134         or      $t0,$t0,$t1             ; Ch(e,f,g)
 135         add     $h,$h,@X[$i%16]
 136         xor     $a0,$a0,$a1             ; Sigma1(e)
 137         add     $h,$h,$t0
 138         add     $h,$h,$a0
 139
 140         $ROR    $a0,$a,$Sigma0[0]
 141         $ROR    $a1,$a,$Sigma0[1]
 142         and     $t0,$a,$b
 143         and     $t1,$a,$c
 144         xor     $a0,$a0,$a1
 145         $ROR    $a1,$a1,`$Sigma0[2]-$Sigma0[1]`
 146         xor     $t0,$t0,$t1
 147         and     $t1,$b,$c
 148         xor     $a0,$a0,$a1             ; Sigma0(a)
 149         add     $d,$d,$h
 150         xor     $t0,$t0,$t1             ; Maj(a,b,c)
 151 ___
 152 $code.=<<___ if ($i<15);
 153         $LD     $t1,`($i+1)*$SZ`($Tbl)
 154 ___
 155 $code.=<<___;
 156         add     $h,$h,$a0
 157         add     $h,$h,$t0
 158
 159 ___
 160 }
 161
 162 sub ROUND_16_xx {
 163 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 164 $i-=16;
 165 $code.=<<___;
 166         $ROR    $a0,@X[($i+1)%16],$sigma0[0]
 167         $ROR    $a1,@X[($i+1)%16],$sigma0[1]
 168         $ROR    $t0,@X[($i+14)%16],$sigma1[0]
 169         $ROR    $t1,@X[($i+14)%16],$sigma1[1]
 170         xor     $a0,$a0,$a1
 171         $SHR    $a1,@X[($i+1)%16],$sigma0[2]
 172         xor     $t0,$t0,$t1
 173         $SHR    $t1,@X[($i+14)%16],$sigma1[2]
 174         add     @X[$i],@X[$i],@X[($i+9)%16]
 175         xor     $a0,$a0,$a1             ; sigma0(X[(i+1)&0x0f])
 176         xor     $t0,$t0,$t1             ; sigma1(X[(i+14)&0x0f])
 177         $LD     $t1,`$i*$SZ`($Tbl)
 178         add     @X[$i],@X[$i],$a0
 179         add     @X[$i],@X[$i],$t0
 180 ___
 181 &ROUND_00_15($i+16,$a,$b,$c,$d,$e,$f,$g,$h);
 182 }
 183
 184 $code=<<___;
 185 .machine        "any"
 186 .text
 187
 188 .globl  $func
 189 .align  6
 190 $func:
 191         $STU    $sp,-$FRAME($sp)
 192         mflr    r0
 193         $SHL    $num,$num,`log(16*$SZ)/log(2)`
 194
 195         $PUSH   $ctx,`$FRAME-$SIZE_T*22`($sp)
 196
 197         $PUSH   r14,`$FRAME-$SIZE_T*18`($sp)
 198         $PUSH   r15,`$FRAME-$SIZE_T*17`($sp)
 199         $PUSH   r16,`$FRAME-$SIZE_T*16`($sp)
 200         $PUSH   r17,`$FRAME-$SIZE_T*15`($sp)
 201         $PUSH   r18,`$FRAME-$SIZE_T*14`($sp)
 202         $PUSH   r19,`$FRAME-$SIZE_T*13`($sp)
 203         $PUSH   r20,`$FRAME-$SIZE_T*12`($sp)
 204         $PUSH   r21,`$FRAME-$SIZE_T*11`($sp)
 205         $PUSH   r22,`$FRAME-$SIZE_T*10`($sp)
 206         $PUSH   r23,`$FRAME-$SIZE_T*9`($sp)
 207         $PUSH   r24,`$FRAME-$SIZE_T*8`($sp)
 208         $PUSH   r25,`$FRAME-$SIZE_T*7`($sp)
 209         $PUSH   r26,`$FRAME-$SIZE_T*6`($sp)
 210         $PUSH   r27,`$FRAME-$SIZE_T*5`($sp)
 211         $PUSH   r28,`$FRAME-$SIZE_T*4`($sp)
 212         $PUSH   r29,`$FRAME-$SIZE_T*3`($sp)
 213         $PUSH   r30,`$FRAME-$SIZE_T*2`($sp)
 214         $PUSH   r31,`$FRAME-$SIZE_T*1`($sp)
 215         $PUSH   r0,`$FRAME+$LRSAVE`($sp)
 216 ___
 217
 218 if ($SZ==4 || $SIZE_T==8) {
 219 $code.=<<___;
 220         $LD     $A,`0*$SZ`($ctx)
 221         mr      $inp,r4                         ; incarnate $inp
 222         $LD     $B,`1*$SZ`($ctx)
 223         $LD     $C,`2*$SZ`($ctx)
 224         $LD     $D,`3*$SZ`($ctx)
 225         $LD     $E,`4*$SZ`($ctx)
 226         $LD     $F,`5*$SZ`($ctx)
 227         $LD     $G,`6*$SZ`($ctx)
 228         $LD     $H,`7*$SZ`($ctx)
 229 ___
 230 } else {
 231   for ($i=16;$i<32;$i++) {
 232     $code.=<<___;
 233         lwz     r$i,`$LITTLE_ENDIAN^(4*($i-16))`($ctx)
 234 ___
 235   }
 236 }
 237
 238 $code.=<<___;
 239         bl      LPICmeup
 240 LPICedup:
 241         andi.   r0,$inp,3
 242         bne     Lunaligned
 243 Laligned:
 244         add     $num,$inp,$num
 245         $PUSH   $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 246         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 247         bl      Lsha2_block_private
 248         b       Ldone
 249
 250 ; PowerPC specification allows an implementation to be ill-behaved
 251 ; upon unaligned access which crosses page boundary. "Better safe
 252 ; than sorry" principle makes me treat it specially. But I don't
 253 ; look for particular offending word, but rather for the input
 254 ; block which crosses the boundary. Once found that block is aligned
 255 ; and hashed separately...
 256 .align  4
 257 Lunaligned:
 258         subfic  $t1,$inp,4096
 259         andi.   $t1,$t1,`4096-16*$SZ`   ; distance to closest page boundary
 260         beq     Lcross_page
 261         $UCMP   $num,$t1
 262         ble-    Laligned                ; didn't cross the page boundary
 263         subfc   $num,$t1,$num
 264         add     $t1,$inp,$t1
 265         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real remaining num
 266         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; intermediate end pointer
 267         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 268         bl      Lsha2_block_private
 269         ; $inp equals to the intermediate end pointer here
 270         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real remaining num
 271 Lcross_page:
 272         li      $t1,`16*$SZ/4`
 273         mtctr   $t1
 274 ___
 275 if ($SZ==4 || $SIZE_T==8) {
 276 $code.=<<___;
 277         addi    r20,$sp,$LOCALS                 ; aligned spot below the frame
 278 Lmemcpy:
 279         lbz     r16,0($inp)
 280         lbz     r17,1($inp)
 281         lbz     r18,2($inp)
 282         lbz     r19,3($inp)
 283         addi    $inp,$inp,4
 284         stb     r16,0(r20)
 285         stb     r17,1(r20)
 286         stb     r18,2(r20)
 287         stb     r19,3(r20)
 288         addi    r20,r20,4
 289         bdnz    Lmemcpy
 290 ___
 291 } else {
 292 $code.=<<___;
 293         addi    r12,$sp,$LOCALS                 ; aligned spot below the frame
 294 Lmemcpy:
 295         lbz     r8,0($inp)
 296         lbz     r9,1($inp)
 297         lbz     r10,2($inp)
 298         lbz     r11,3($inp)
 299         addi    $inp,$inp,4
 300         stb     r8,0(r12)
 301         stb     r9,1(r12)
 302         stb     r10,2(r12)
 303         stb     r11,3(r12)
 304         addi    r12,r12,4
 305         bdnz    Lmemcpy
 306 ___
 307 }
 308
 309 $code.=<<___;
 310         $PUSH   $inp,`$FRAME-$SIZE_T*26`($sp)   ; save real inp
 311         addi    $t1,$sp,`$LOCALS+16*$SZ`        ; fictitious end pointer
 312         addi    $inp,$sp,$LOCALS                ; fictitious inp pointer
 313         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real num
 314         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; end pointer
 315         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 316         bl      Lsha2_block_private
 317         $POP    $inp,`$FRAME-$SIZE_T*26`($sp)   ; restore real inp
 318         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real num
 319         addic.  $num,$num,`-16*$SZ`             ; num--
 320         bne-    Lunaligned
 321
 322 Ldone:
 323         $POP    r0,`$FRAME+$LRSAVE`($sp)
 324         $POP    r14,`$FRAME-$SIZE_T*18`($sp)
 325         $POP    r15,`$FRAME-$SIZE_T*17`($sp)
 326         $POP    r16,`$FRAME-$SIZE_T*16`($sp)
 327         $POP    r17,`$FRAME-$SIZE_T*15`($sp)
 328         $POP    r18,`$FRAME-$SIZE_T*14`($sp)
 329         $POP    r19,`$FRAME-$SIZE_T*13`($sp)
 330         $POP    r20,`$FRAME-$SIZE_T*12`($sp)
 331         $POP    r21,`$FRAME-$SIZE_T*11`($sp)
 332         $POP    r22,`$FRAME-$SIZE_T*10`($sp)
 333         $POP    r23,`$FRAME-$SIZE_T*9`($sp)
 334         $POP    r24,`$FRAME-$SIZE_T*8`($sp)
 335         $POP    r25,`$FRAME-$SIZE_T*7`($sp)
 336         $POP    r26,`$FRAME-$SIZE_T*6`($sp)
 337         $POP    r27,`$FRAME-$SIZE_T*5`($sp)
 338         $POP    r28,`$FRAME-$SIZE_T*4`($sp)
 339         $POP    r29,`$FRAME-$SIZE_T*3`($sp)
 340         $POP    r30,`$FRAME-$SIZE_T*2`($sp)
 341         $POP    r31,`$FRAME-$SIZE_T*1`($sp)
 342         mtlr    r0
 343         addi    $sp,$sp,$FRAME
 344         blr
 345         .long   0
 346         .byte   0,12,4,1,0x80,18,3,0
 347         .long   0
 348 ___
 349
 350 if ($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 && !$LITTLE_ENDIAN);
 358         lwz     @X[$i],`$i*$SZ`($inp)
 359 ___
 360 $code.=<<___ if ($SZ==4 && $LITTLE_ENDIAN);
 361         lwz     $a0,`$i*$SZ`($inp)
 362         rotlwi  @X[$i],$a0,8
 363         rlwimi  @X[$i],$a0,24,0,7
 364         rlwimi  @X[$i],$a0,24,16,23
 365 ___
 366 # 64-bit loads are split to 2x32-bit ones, as CPU can't handle
 367 # unaligned 64-bit loads, only 32-bit ones...
 368 $code.=<<___ if ($SZ==8 && !$LITTLE_ENDIAN);
 369         lwz     $t0,`$i*$SZ`($inp)
 370         lwz     @X[$i],`$i*$SZ+4`($inp)
 371         insrdi  @X[$i],$t0,32,0
 372 ___
 373 $code.=<<___ if ($SZ==8 && $LITTLE_ENDIAN);
 374         lwz     $a0,`$i*$SZ`($inp)
 375          lwz    $a1,`$i*$SZ+4`($inp)
 376         rotlwi  $t0,$a0,8
 377          rotlwi @X[$i],$a1,8
 378         rlwimi  $t0,$a0,24,0,7
 379          rlwimi @X[$i],$a1,24,0,7
 380         rlwimi  $t0,$a0,24,16,23
 381          rlwimi @X[$i],$a1,24,16,23
 382         insrdi  @X[$i],$t0,32,0
 383 ___
 384         &ROUND_00_15($i,@V);
 385         unshift(@V,pop(@V));
 386 }
 387 $code.=<<___;
 388         li      $t0,`$rounds/16-1`
 389         mtctr   $t0
 390 .align  4
 391 Lrounds:
 392         addi    $Tbl,$Tbl,`16*$SZ`
 393 ___
 394 for(;$i<32;$i++) {
 395         &ROUND_16_xx($i,@V);
 396         unshift(@V,pop(@V));
 397 }
 398 $code.=<<___;
 399         bdnz-   Lrounds
 400
 401         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 402         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 403         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 404         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 405
 406         $LD     r16,`0*$SZ`($ctx)
 407         $LD     r17,`1*$SZ`($ctx)
 408         $LD     r18,`2*$SZ`($ctx)
 409         $LD     r19,`3*$SZ`($ctx)
 410         $LD     r20,`4*$SZ`($ctx)
 411         $LD     r21,`5*$SZ`($ctx)
 412         $LD     r22,`6*$SZ`($ctx)
 413         addi    $inp,$inp,`16*$SZ`              ; advance inp
 414         $LD     r23,`7*$SZ`($ctx)
 415         add     $A,$A,r16
 416         add     $B,$B,r17
 417         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 418         add     $C,$C,r18
 419         $ST     $A,`0*$SZ`($ctx)
 420         add     $D,$D,r19
 421         $ST     $B,`1*$SZ`($ctx)
 422         add     $E,$E,r20
 423         $ST     $C,`2*$SZ`($ctx)
 424         add     $F,$F,r21
 425         $ST     $D,`3*$SZ`($ctx)
 426         add     $G,$G,r22
 427         $ST     $E,`4*$SZ`($ctx)
 428         add     $H,$H,r23
 429         $ST     $F,`5*$SZ`($ctx)
 430         $ST     $G,`6*$SZ`($ctx)
 431         $UCMP   $inp,$num
 432         $ST     $H,`7*$SZ`($ctx)
 433         bne     Lsha2_block_private
 434         blr
 435         .long   0
 436         .byte   0,12,0x14,0,0,0,0,0
 437 .size   $func,.-$func
 438 ___
 439 } else {
 440 ########################################################################
 441 # SHA512 for PPC32, X vector is off-loaded to stack...
 442 #
 443 #                       |       sha512
 444 #                       |       -m32
 445 # ----------------------+-----------------------
 446 # PPC74x0,gcc-4.0.1     |       +48%
 447 # POWER6,gcc-4.4.6      |       +124%(*)
 448 # POWER7,gcc-4.4.6      |       +79%(*)
 449 # e300,gcc-4.1.0        |       +167%
 450 #
 451 # (*)   ~1/3 of -m64 result [and ~20% better than -m32 code generated
 452 #       by xlc-12.1]
 453
 454 my $XOFF=$LOCALS;
 455
 456 my @V=map("r$_",(16..31));      # A..H
 457
 458 my ($s0,$s1,$t0,$t1,$t2,$t3,$a0,$a1,$a2,$a3)=map("r$_",(0,5,6,8..12,14,15));
 459 my ($x0,$x1)=("r3","r4");       # zaps $ctx and $inp
 460
 461 sub ROUND_00_15_ppc32 {
 462 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 463         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 464
 465 $code.=<<___;
 466         lwz     $t2,`$SZ*($i%16)+($LITTLE_ENDIAN^4)`($Tbl)
 467          xor    $a0,$flo,$glo
 468         lwz     $t3,`$SZ*($i%16)+($LITTLE_ENDIAN^0)`($Tbl)
 469          xor    $a1,$fhi,$ghi
 470         addc    $hlo,$hlo,$t0                   ; h+=x[i]
 471         stw     $t0,`$XOFF+0+$SZ*($i%16)`($sp)  ; save x[i]
 472
 473         srwi    $s0,$elo,$Sigma1[0]
 474         srwi    $s1,$ehi,$Sigma1[0]
 475          and    $a0,$a0,$elo
 476         adde    $hhi,$hhi,$t1
 477          and    $a1,$a1,$ehi
 478         stw     $t1,`$XOFF+4+$SZ*($i%16)`($sp)
 479         srwi    $t0,$elo,$Sigma1[1]
 480         srwi    $t1,$ehi,$Sigma1[1]
 481          addc   $hlo,$hlo,$t2                   ; h+=K512[i]
 482         insrwi  $s0,$ehi,$Sigma1[0],0
 483         insrwi  $s1,$elo,$Sigma1[0],0
 484          xor    $a0,$a0,$glo                    ; Ch(e,f,g)
 485          adde   $hhi,$hhi,$t3
 486          xor    $a1,$a1,$ghi
 487         insrwi  $t0,$ehi,$Sigma1[1],0
 488         insrwi  $t1,$elo,$Sigma1[1],0
 489          addc   $hlo,$hlo,$a0                   ; h+=Ch(e,f,g)
 490         srwi    $t2,$ehi,$Sigma1[2]-32
 491         srwi    $t3,$elo,$Sigma1[2]-32
 492         xor     $s0,$s0,$t0
 493         xor     $s1,$s1,$t1
 494         insrwi  $t2,$elo,$Sigma1[2]-32,0
 495         insrwi  $t3,$ehi,$Sigma1[2]-32,0
 496          xor    $a0,$alo,$blo                   ; a^b, b^c in next round
 497          adde   $hhi,$hhi,$a1
 498          xor    $a1,$ahi,$bhi
 499         xor     $s0,$s0,$t2                     ; Sigma1(e)
 500         xor     $s1,$s1,$t3
 501
 502         srwi    $t0,$alo,$Sigma0[0]
 503          and    $a2,$a2,$a0
 504          addc   $hlo,$hlo,$s0                   ; h+=Sigma1(e)
 505          and    $a3,$a3,$a1
 506         srwi    $t1,$ahi,$Sigma0[0]
 507         srwi    $s0,$ahi,$Sigma0[1]-32
 508          adde   $hhi,$hhi,$s1
 509         srwi    $s1,$alo,$Sigma0[1]-32
 510         insrwi  $t0,$ahi,$Sigma0[0],0
 511         insrwi  $t1,$alo,$Sigma0[0],0
 512          xor    $a2,$a2,$blo                    ; Maj(a,b,c)
 513          addc   $dlo,$dlo,$hlo                  ; d+=h
 514          xor    $a3,$a3,$bhi
 515         insrwi  $s0,$alo,$Sigma0[1]-32,0
 516         insrwi  $s1,$ahi,$Sigma0[1]-32,0
 517          adde   $dhi,$dhi,$hhi
 518         srwi    $t2,$ahi,$Sigma0[2]-32
 519         srwi    $t3,$alo,$Sigma0[2]-32
 520         xor     $s0,$s0,$t0
 521          addc   $hlo,$hlo,$a2                   ; h+=Maj(a,b,c)
 522         xor     $s1,$s1,$t1
 523         insrwi  $t2,$alo,$Sigma0[2]-32,0
 524         insrwi  $t3,$ahi,$Sigma0[2]-32,0
 525          adde   $hhi,$hhi,$a3
 526 ___
 527 $code.=<<___ if ($i>=15);
 528         lwz     $t0,`$XOFF+0+$SZ*(($i+2)%16)`($sp)
 529         lwz     $t1,`$XOFF+4+$SZ*(($i+2)%16)`($sp)
 530 ___
 531 $code.=<<___ if ($i<15 && !$LITTLE_ENDIAN);
 532         lwz     $t1,`$SZ*($i+1)+0`($inp)
 533         lwz     $t0,`$SZ*($i+1)+4`($inp)
 534 ___
 535 $code.=<<___ if ($i<15 && $LITTLE_ENDIAN);
 536         lwz     $a2,`$SZ*($i+1)+0`($inp)
 537          lwz    $a3,`$SZ*($i+1)+4`($inp)
 538         rotlwi  $t1,$a2,8
 539          rotlwi $t0,$a3,8
 540         rlwimi  $t1,$a2,24,0,7
 541          rlwimi $t0,$a3,24,0,7
 542         rlwimi  $t1,$a2,24,16,23
 543          rlwimi $t0,$a3,24,16,23
 544 ___
 545 $code.=<<___;
 546         xor     $s0,$s0,$t2                     ; Sigma0(a)
 547         xor     $s1,$s1,$t3
 548         addc    $hlo,$hlo,$s0                   ; h+=Sigma0(a)
 549         adde    $hhi,$hhi,$s1
 550 ___
 551 $code.=<<___ if ($i==15);
 552         lwz     $x0,`$XOFF+0+$SZ*(($i+1)%16)`($sp)
 553         lwz     $x1,`$XOFF+4+$SZ*(($i+1)%16)`($sp)
 554 ___
 555 }
 556 sub ROUND_16_xx_ppc32 {
 557 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 558         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 559
 560 $code.=<<___;
 561         srwi    $s0,$t0,$sigma0[0]
 562         srwi    $s1,$t1,$sigma0[0]
 563         srwi    $t2,$t0,$sigma0[1]
 564         srwi    $t3,$t1,$sigma0[1]
 565         insrwi  $s0,$t1,$sigma0[0],0
 566         insrwi  $s1,$t0,$sigma0[0],0
 567         srwi    $a0,$t0,$sigma0[2]
 568         insrwi  $t2,$t1,$sigma0[1],0
 569         insrwi  $t3,$t0,$sigma0[1],0
 570         insrwi  $a0,$t1,$sigma0[2],0
 571         xor     $s0,$s0,$t2
 572          lwz    $t2,`$XOFF+0+$SZ*(($i+14)%16)`($sp)
 573         srwi    $a1,$t1,$sigma0[2]
 574         xor     $s1,$s1,$t3
 575          lwz    $t3,`$XOFF+4+$SZ*(($i+14)%16)`($sp)
 576         xor     $a0,$a0,$s0
 577          srwi   $s0,$t2,$sigma1[0]
 578         xor     $a1,$a1,$s1
 579          srwi   $s1,$t3,$sigma1[0]
 580         addc    $x0,$x0,$a0                     ; x[i]+=sigma0(x[i+1])
 581          srwi   $a0,$t3,$sigma1[1]-32
 582         insrwi  $s0,$t3,$sigma1[0],0
 583         insrwi  $s1,$t2,$sigma1[0],0
 584         adde    $x1,$x1,$a1
 585          srwi   $a1,$t2,$sigma1[1]-32
 586
 587         insrwi  $a0,$t2,$sigma1[1]-32,0
 588         srwi    $t2,$t2,$sigma1[2]
 589         insrwi  $a1,$t3,$sigma1[1]-32,0
 590         insrwi  $t2,$t3,$sigma1[2],0
 591         xor     $s0,$s0,$a0
 592          lwz    $a0,`$XOFF+0+$SZ*(($i+9)%16)`($sp)
 593         srwi    $t3,$t3,$sigma1[2]
 594         xor     $s1,$s1,$a1
 595          lwz    $a1,`$XOFF+4+$SZ*(($i+9)%16)`($sp)
 596         xor     $s0,$s0,$t2
 597          addc   $x0,$x0,$a0                     ; x[i]+=x[i+9]
 598         xor     $s1,$s1,$t3
 599          adde   $x1,$x1,$a1
 600         addc    $x0,$x0,$s0                     ; x[i]+=sigma1(x[i+14])
 601         adde    $x1,$x1,$s1
 602 ___
 603         ($t0,$t1,$x0,$x1) = ($x0,$x1,$t0,$t1);
 604         &ROUND_00_15_ppc32(@_);
 605 }
 606
 607 $code.=<<___;
 608 .align  4
 609 Lsha2_block_private:
 610 ___
 611 $code.=<<___ if (!$LITTLE_ENDIAN);
 612         lwz     $t1,0($inp)
 613         xor     $a2,@V[3],@V[5]         ; B^C, magic seed
 614         lwz     $t0,4($inp)
 615         xor     $a3,@V[2],@V[4]
 616 ___
 617 $code.=<<___ if ($LITTLE_ENDIAN);
 618         lwz     $a1,0($inp)
 619         xor     $a2,@V[3],@V[5]         ; B^C, magic seed
 620         lwz     $a0,4($inp)
 621         xor     $a3,@V[2],@V[4]
 622         rotlwi  $t1,$a1,8
 623          rotlwi $t0,$a0,8
 624         rlwimi  $t1,$a1,24,0,7
 625          rlwimi $t0,$a0,24,0,7
 626         rlwimi  $t1,$a1,24,16,23
 627          rlwimi $t0,$a0,24,16,23
 628 ___
 629 for($i=0;$i<16;$i++) {
 630         &ROUND_00_15_ppc32($i,@V);
 631         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 632         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 633 }
 634 $code.=<<___;
 635         li      $a0,`$rounds/16-1`
 636         mtctr   $a0
 637 .align  4
 638 Lrounds:
 639         addi    $Tbl,$Tbl,`16*$SZ`
 640 ___
 641 for(;$i<32;$i++) {
 642         &ROUND_16_xx_ppc32($i,@V);
 643         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 644         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 645 }
 646 $code.=<<___;
 647         bdnz-   Lrounds
 648
 649         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 650         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 651         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 652         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 653
 654         lwz     $t0,`$LITTLE_ENDIAN^0`($ctx)
 655         lwz     $t1,`$LITTLE_ENDIAN^4`($ctx)
 656         lwz     $t2,`$LITTLE_ENDIAN^8`($ctx)
 657         lwz     $t3,`$LITTLE_ENDIAN^12`($ctx)
 658         lwz     $a0,`$LITTLE_ENDIAN^16`($ctx)
 659         lwz     $a1,`$LITTLE_ENDIAN^20`($ctx)
 660         lwz     $a2,`$LITTLE_ENDIAN^24`($ctx)
 661         addc    @V[1],@V[1],$t1
 662         lwz     $a3,`$LITTLE_ENDIAN^28`($ctx)
 663         adde    @V[0],@V[0],$t0
 664         lwz     $t0,`$LITTLE_ENDIAN^32`($ctx)
 665         addc    @V[3],@V[3],$t3
 666         lwz     $t1,`$LITTLE_ENDIAN^36`($ctx)
 667         adde    @V[2],@V[2],$t2
 668         lwz     $t2,`$LITTLE_ENDIAN^40`($ctx)
 669         addc    @V[5],@V[5],$a1
 670         lwz     $t3,`$LITTLE_ENDIAN^44`($ctx)
 671         adde    @V[4],@V[4],$a0
 672         lwz     $a0,`$LITTLE_ENDIAN^48`($ctx)
 673         addc    @V[7],@V[7],$a3
 674         lwz     $a1,`$LITTLE_ENDIAN^52`($ctx)
 675         adde    @V[6],@V[6],$a2
 676         lwz     $a2,`$LITTLE_ENDIAN^56`($ctx)
 677         addc    @V[9],@V[9],$t1
 678         lwz     $a3,`$LITTLE_ENDIAN^60`($ctx)
 679         adde    @V[8],@V[8],$t0
 680         stw     @V[0],`$LITTLE_ENDIAN^0`($ctx)
 681         stw     @V[1],`$LITTLE_ENDIAN^4`($ctx)
 682         addc    @V[11],@V[11],$t3
 683         stw     @V[2],`$LITTLE_ENDIAN^8`($ctx)
 684         stw     @V[3],`$LITTLE_ENDIAN^12`($ctx)
 685         adde    @V[10],@V[10],$t2
 686         stw     @V[4],`$LITTLE_ENDIAN^16`($ctx)
 687         stw     @V[5],`$LITTLE_ENDIAN^20`($ctx)
 688         addc    @V[13],@V[13],$a1
 689         stw     @V[6],`$LITTLE_ENDIAN^24`($ctx)
 690         stw     @V[7],`$LITTLE_ENDIAN^28`($ctx)
 691         adde    @V[12],@V[12],$a0
 692         stw     @V[8],`$LITTLE_ENDIAN^32`($ctx)
 693         stw     @V[9],`$LITTLE_ENDIAN^36`($ctx)
 694         addc    @V[15],@V[15],$a3
 695         stw     @V[10],`$LITTLE_ENDIAN^40`($ctx)
 696         stw     @V[11],`$LITTLE_ENDIAN^44`($ctx)
 697         adde    @V[14],@V[14],$a2
 698         stw     @V[12],`$LITTLE_ENDIAN^48`($ctx)
 699         stw     @V[13],`$LITTLE_ENDIAN^52`($ctx)
 700         stw     @V[14],`$LITTLE_ENDIAN^56`($ctx)
 701         stw     @V[15],`$LITTLE_ENDIAN^60`($ctx)
 702
 703         addi    $inp,$inp,`16*$SZ`              ; advance inp
 704         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 705         $UCMP   $inp,$num
 706         bne     Lsha2_block_private
 707         blr
 708         .long   0
 709         .byte   0,12,0x14,0,0,0,0,0
 710 .size   $func,.-$func
 711 ___
 712 }
 713
 714 # Ugly hack here, because PPC assembler syntax seem to vary too
 715 # much from platforms to platform...
 716 $code.=<<___;
 717 .align  6
 718 LPICmeup:
 719         mflr    r0
 720         bcl     20,31,\$+4
 721         mflr    $Tbl    ; vvvvvv "distance" between . and 1st data entry
 722         addi    $Tbl,$Tbl,`64-8`
 723         mtlr    r0
 724         blr
 725         .long   0
 726         .byte   0,12,0x14,0,0,0,0,0
 727         .space  `64-9*4`
 728 ___
 729 $code.=<<___ if ($SZ==8);
 730         .quad   0x428a2f98d728ae22,0x7137449123ef65cd
 731         .quad   0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
 732         .quad   0x3956c25bf348b538,0x59f111f1b605d019
 733         .quad   0x923f82a4af194f9b,0xab1c5ed5da6d8118
 734         .quad   0xd807aa98a3030242,0x12835b0145706fbe
 735         .quad   0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
 736         .quad   0x72be5d74f27b896f,0x80deb1fe3b1696b1
 737         .quad   0x9bdc06a725c71235,0xc19bf174cf692694
 738         .quad   0xe49b69c19ef14ad2,0xefbe4786384f25e3
 739         .quad   0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
 740         .quad   0x2de92c6f592b0275,0x4a7484aa6ea6e483
 741         .quad   0x5cb0a9dcbd41fbd4,0x76f988da831153b5
 742         .quad   0x983e5152ee66dfab,0xa831c66d2db43210
 743         .quad   0xb00327c898fb213f,0xbf597fc7beef0ee4
 744         .quad   0xc6e00bf33da88fc2,0xd5a79147930aa725
 745         .quad   0x06ca6351e003826f,0x142929670a0e6e70
 746         .quad   0x27b70a8546d22ffc,0x2e1b21385c26c926
 747         .quad   0x4d2c6dfc5ac42aed,0x53380d139d95b3df
 748         .quad   0x650a73548baf63de,0x766a0abb3c77b2a8
 749         .quad   0x81c2c92e47edaee6,0x92722c851482353b
 750         .quad   0xa2bfe8a14cf10364,0xa81a664bbc423001
 751         .quad   0xc24b8b70d0f89791,0xc76c51a30654be30
 752         .quad   0xd192e819d6ef5218,0xd69906245565a910
 753         .quad   0xf40e35855771202a,0x106aa07032bbd1b8
 754         .quad   0x19a4c116b8d2d0c8,0x1e376c085141ab53
 755         .quad   0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
 756         .quad   0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
 757         .quad   0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
 758         .quad   0x748f82ee5defb2fc,0x78a5636f43172f60
 759         .quad   0x84c87814a1f0ab72,0x8cc702081a6439ec
 760         .quad   0x90befffa23631e28,0xa4506cebde82bde9
 761         .quad   0xbef9a3f7b2c67915,0xc67178f2e372532b
 762         .quad   0xca273eceea26619c,0xd186b8c721c0c207
 763         .quad   0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
 764         .quad   0x06f067aa72176fba,0x0a637dc5a2c898a6
 765         .quad   0x113f9804bef90dae,0x1b710b35131c471b
 766         .quad   0x28db77f523047d84,0x32caab7b40c72493
 767         .quad   0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
 768         .quad   0x4cc5d4becb3e42b6,0x597f299cfc657e2a
 769         .quad   0x5fcb6fab3ad6faec,0x6c44198c4a475817
 770 ___
 771 $code.=<<___ if ($SZ==4);
 772         .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
 773         .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
 774         .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
 775         .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
 776         .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
 777         .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
 778         .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
 779         .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
 780         .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
 781         .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
 782         .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
 783         .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
 784         .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
 785         .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
 786         .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
 787         .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
 788 ___
 789
 790 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 791 print $code;
 792 close STDOUT;