crypto/sha/asm/sha512-ppc.pl

   1 #! /usr/bin/env perl
   2 # Copyright 2006-2016 The OpenSSL Project Authors. All Rights Reserved.
   3 #
   4 # Licensed under the OpenSSL license (the "License").  You may not use
   5 # this file except in compliance with the License.  You can obtain a copy
   6 # in the file LICENSE in the source distribution or at
   7 # https://www.openssl.org/source/license.html
   8
   9
  10 # ====================================================================
  11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
  12 # project. The module is, however, dual licensed under OpenSSL and
  13 # CRYPTOGAMS licenses depending on where you obtain it. For further
  14 # details see http://www.openssl.org/~appro/cryptogams/.
  15 # ====================================================================
  16
  17 # I let hardware handle unaligned input, except on page boundaries
  18 # (see below for details). Otherwise straightforward implementation
  19 # with X vector in register bank.
  20
  21 #                       sha256          |       sha512
  22 #                       -m64    -m32    |       -m64    -m32
  23 # --------------------------------------+-----------------------
  24 # PPC970,gcc-4.0.0      +50%    +38%    |       +40%    +410%(*)
  25 # Power6,xlc-7          +150%   +90%    |       +100%   +430%(*)
  26 #
  27 # (*)   64-bit code in 32-bit application context, which actually is
  28 #       on TODO list. It should be noted that for safe deployment in
  29 #       32-bit *mutli-threaded* context asyncronous signals should be
  30 #       blocked upon entry to SHA512 block routine. This is because
  31 #       32-bit signaling procedure invalidates upper halves of GPRs.
  32 #       Context switch procedure preserves them, but not signaling:-(
  33
  34 # Second version is true multi-thread safe. Trouble with the original
  35 # version was that it was using thread local storage pointer register.
  36 # Well, it scrupulously preserved it, but the problem would arise the
  37 # moment asynchronous signal was delivered and signal handler would
  38 # dereference the TLS pointer. While it's never the case in openssl
  39 # application or test suite, we have to respect this scenario and not
  40 # use TLS pointer register. Alternative would be to require caller to
  41 # block signals prior calling this routine. For the record, in 32-bit
  42 # context R2 serves as TLS pointer, while in 64-bit context - R13.
  43
  44 $flavour=shift;
  45 $output =shift;
  46
  47 if ($flavour =~ /64/) {
  48         $SIZE_T=8;
  49         $LRSAVE=2*$SIZE_T;
  50         $STU="stdu";
  51         $UCMP="cmpld";
  52         $SHL="sldi";
  53         $POP="ld";
  54         $PUSH="std";
  55 } elsif ($flavour =~ /32/) {
  56         $SIZE_T=4;
  57         $LRSAVE=$SIZE_T;
  58         $STU="stwu";
  59         $UCMP="cmplw";
  60         $SHL="slwi";
  61         $POP="lwz";
  62         $PUSH="stw";
  63 } else { die "nonsense $flavour"; }
  64
  65 $LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
  66
  67 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  68 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
  69 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
  70 die "can't locate ppc-xlate.pl";
  71
  72 open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!";
  73
  74 if ($output =~ /512/) {
  75         $func="sha512_block_ppc";
  76         $SZ=8;
  77         @Sigma0=(28,34,39);
  78         @Sigma1=(14,18,41);
  79         @sigma0=(1,  8, 7);
  80         @sigma1=(19,61, 6);
  81         $rounds=80;
  82         $LD="ld";
  83         $ST="std";
  84         $ROR="rotrdi";
  85         $SHR="srdi";
  86 } else {
  87         $func="sha256_block_ppc";
  88         $SZ=4;
  89         @Sigma0=( 2,13,22);
  90         @Sigma1=( 6,11,25);
  91         @sigma0=( 7,18, 3);
  92         @sigma1=(17,19,10);
  93         $rounds=64;
  94         $LD="lwz";
  95         $ST="stw";
  96         $ROR="rotrwi";
  97         $SHR="srwi";
  98 }
  99
 100 $FRAME=32*$SIZE_T+16*$SZ;
 101 $LOCALS=6*$SIZE_T;
 102
 103 $sp ="r1";
 104 $toc="r2";
 105 $ctx="r3";      # zapped by $a0
 106 $inp="r4";      # zapped by $a1
 107 $num="r5";      # zapped by $t0
 108
 109 $T  ="r0";
 110 $a0 ="r3";
 111 $a1 ="r4";
 112 $t0 ="r5";
 113 $t1 ="r6";
 114 $Tbl="r7";
 115
 116 $A  ="r8";
 117 $B  ="r9";
 118 $C  ="r10";
 119 $D  ="r11";
 120 $E  ="r12";
 121 $F  =$t1;       $t1 = "r0";     # stay away from "r13";
 122 $G  ="r14";
 123 $H  ="r15";
 124
 125 @V=($A,$B,$C,$D,$E,$F,$G,$H);
 126 @X=("r16","r17","r18","r19","r20","r21","r22","r23",
 127     "r24","r25","r26","r27","r28","r29","r30","r31");
 128
 129 $inp="r31" if($SZ==4 || $SIZE_T==8);    # reassigned $inp! aliases with @X[15]
 130
 131 sub ROUND_00_15 {
 132 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 133 $code.=<<___;
 134         $ROR    $a0,$e,$Sigma1[0]
 135         $ROR    $a1,$e,$Sigma1[1]
 136         and     $t0,$f,$e
 137         xor     $a0,$a0,$a1
 138         add     $h,$h,$t1
 139         andc    $t1,$g,$e
 140         $ROR    $a1,$a1,`$Sigma1[2]-$Sigma1[1]`
 141         or      $t0,$t0,$t1             ; Ch(e,f,g)
 142         add     $h,$h,@X[$i%16]
 143         xor     $a0,$a0,$a1             ; Sigma1(e)
 144         add     $h,$h,$t0
 145         add     $h,$h,$a0
 146
 147         $ROR    $a0,$a,$Sigma0[0]
 148         $ROR    $a1,$a,$Sigma0[1]
 149         and     $t0,$a,$b
 150         and     $t1,$a,$c
 151         xor     $a0,$a0,$a1
 152         $ROR    $a1,$a1,`$Sigma0[2]-$Sigma0[1]`
 153         xor     $t0,$t0,$t1
 154         and     $t1,$b,$c
 155         xor     $a0,$a0,$a1             ; Sigma0(a)
 156         add     $d,$d,$h
 157         xor     $t0,$t0,$t1             ; Maj(a,b,c)
 158 ___
 159 $code.=<<___ if ($i<15);
 160         $LD     $t1,`($i+1)*$SZ`($Tbl)
 161 ___
 162 $code.=<<___;
 163         add     $h,$h,$a0
 164         add     $h,$h,$t0
 165
 166 ___
 167 }
 168
 169 sub ROUND_16_xx {
 170 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
 171 $i-=16;
 172 $code.=<<___;
 173         $ROR    $a0,@X[($i+1)%16],$sigma0[0]
 174         $ROR    $a1,@X[($i+1)%16],$sigma0[1]
 175         $ROR    $t0,@X[($i+14)%16],$sigma1[0]
 176         $ROR    $t1,@X[($i+14)%16],$sigma1[1]
 177         xor     $a0,$a0,$a1
 178         $SHR    $a1,@X[($i+1)%16],$sigma0[2]
 179         xor     $t0,$t0,$t1
 180         $SHR    $t1,@X[($i+14)%16],$sigma1[2]
 181         add     @X[$i],@X[$i],@X[($i+9)%16]
 182         xor     $a0,$a0,$a1             ; sigma0(X[(i+1)&0x0f])
 183         xor     $t0,$t0,$t1             ; sigma1(X[(i+14)&0x0f])
 184         $LD     $t1,`$i*$SZ`($Tbl)
 185         add     @X[$i],@X[$i],$a0
 186         add     @X[$i],@X[$i],$t0
 187 ___
 188 &ROUND_00_15($i+16,$a,$b,$c,$d,$e,$f,$g,$h);
 189 }
 190
 191 $code=<<___;
 192 .machine        "any"
 193 .text
 194
 195 .globl  $func
 196 .align  6
 197 $func:
 198         $STU    $sp,-$FRAME($sp)
 199         mflr    r0
 200         $SHL    $num,$num,`log(16*$SZ)/log(2)`
 201
 202         $PUSH   $ctx,`$FRAME-$SIZE_T*22`($sp)
 203
 204         $PUSH   r14,`$FRAME-$SIZE_T*18`($sp)
 205         $PUSH   r15,`$FRAME-$SIZE_T*17`($sp)
 206         $PUSH   r16,`$FRAME-$SIZE_T*16`($sp)
 207         $PUSH   r17,`$FRAME-$SIZE_T*15`($sp)
 208         $PUSH   r18,`$FRAME-$SIZE_T*14`($sp)
 209         $PUSH   r19,`$FRAME-$SIZE_T*13`($sp)
 210         $PUSH   r20,`$FRAME-$SIZE_T*12`($sp)
 211         $PUSH   r21,`$FRAME-$SIZE_T*11`($sp)
 212         $PUSH   r22,`$FRAME-$SIZE_T*10`($sp)
 213         $PUSH   r23,`$FRAME-$SIZE_T*9`($sp)
 214         $PUSH   r24,`$FRAME-$SIZE_T*8`($sp)
 215         $PUSH   r25,`$FRAME-$SIZE_T*7`($sp)
 216         $PUSH   r26,`$FRAME-$SIZE_T*6`($sp)
 217         $PUSH   r27,`$FRAME-$SIZE_T*5`($sp)
 218         $PUSH   r28,`$FRAME-$SIZE_T*4`($sp)
 219         $PUSH   r29,`$FRAME-$SIZE_T*3`($sp)
 220         $PUSH   r30,`$FRAME-$SIZE_T*2`($sp)
 221         $PUSH   r31,`$FRAME-$SIZE_T*1`($sp)
 222         $PUSH   r0,`$FRAME+$LRSAVE`($sp)
 223 ___
 224
 225 if ($SZ==4 || $SIZE_T==8) {
 226 $code.=<<___;
 227         $LD     $A,`0*$SZ`($ctx)
 228         mr      $inp,r4                         ; incarnate $inp
 229         $LD     $B,`1*$SZ`($ctx)
 230         $LD     $C,`2*$SZ`($ctx)
 231         $LD     $D,`3*$SZ`($ctx)
 232         $LD     $E,`4*$SZ`($ctx)
 233         $LD     $F,`5*$SZ`($ctx)
 234         $LD     $G,`6*$SZ`($ctx)
 235         $LD     $H,`7*$SZ`($ctx)
 236 ___
 237 } else {
 238   for ($i=16;$i<32;$i++) {
 239     $code.=<<___;
 240         lwz     r$i,`$LITTLE_ENDIAN^(4*($i-16))`($ctx)
 241 ___
 242   }
 243 }
 244
 245 $code.=<<___;
 246         bl      LPICmeup
 247 LPICedup:
 248         andi.   r0,$inp,3
 249         bne     Lunaligned
 250 Laligned:
 251         add     $num,$inp,$num
 252         $PUSH   $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 253         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 254         bl      Lsha2_block_private
 255         b       Ldone
 256
 257 ; PowerPC specification allows an implementation to be ill-behaved
 258 ; upon unaligned access which crosses page boundary. "Better safe
 259 ; than sorry" principle makes me treat it specially. But I don't
 260 ; look for particular offending word, but rather for the input
 261 ; block which crosses the boundary. Once found that block is aligned
 262 ; and hashed separately...
 263 .align  4
 264 Lunaligned:
 265         subfic  $t1,$inp,4096
 266         andi.   $t1,$t1,`4096-16*$SZ`   ; distance to closest page boundary
 267         beq     Lcross_page
 268         $UCMP   $num,$t1
 269         ble     Laligned                ; didn't cross the page boundary
 270         subfc   $num,$t1,$num
 271         add     $t1,$inp,$t1
 272         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real remaining num
 273         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; intermediate end pointer
 274         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 275         bl      Lsha2_block_private
 276         ; $inp equals to the intermediate end pointer here
 277         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real remaining num
 278 Lcross_page:
 279         li      $t1,`16*$SZ/4`
 280         mtctr   $t1
 281 ___
 282 if ($SZ==4 || $SIZE_T==8) {
 283 $code.=<<___;
 284         addi    r20,$sp,$LOCALS                 ; aligned spot below the frame
 285 Lmemcpy:
 286         lbz     r16,0($inp)
 287         lbz     r17,1($inp)
 288         lbz     r18,2($inp)
 289         lbz     r19,3($inp)
 290         addi    $inp,$inp,4
 291         stb     r16,0(r20)
 292         stb     r17,1(r20)
 293         stb     r18,2(r20)
 294         stb     r19,3(r20)
 295         addi    r20,r20,4
 296         bdnz    Lmemcpy
 297 ___
 298 } else {
 299 $code.=<<___;
 300         addi    r12,$sp,$LOCALS                 ; aligned spot below the frame
 301 Lmemcpy:
 302         lbz     r8,0($inp)
 303         lbz     r9,1($inp)
 304         lbz     r10,2($inp)
 305         lbz     r11,3($inp)
 306         addi    $inp,$inp,4
 307         stb     r8,0(r12)
 308         stb     r9,1(r12)
 309         stb     r10,2(r12)
 310         stb     r11,3(r12)
 311         addi    r12,r12,4
 312         bdnz    Lmemcpy
 313 ___
 314 }
 315
 316 $code.=<<___;
 317         $PUSH   $inp,`$FRAME-$SIZE_T*26`($sp)   ; save real inp
 318         addi    $t1,$sp,`$LOCALS+16*$SZ`        ; fictitious end pointer
 319         addi    $inp,$sp,$LOCALS                ; fictitious inp pointer
 320         $PUSH   $num,`$FRAME-$SIZE_T*25`($sp)   ; save real num
 321         $PUSH   $t1,`$FRAME-$SIZE_T*24`($sp)    ; end pointer
 322         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 323         bl      Lsha2_block_private
 324         $POP    $inp,`$FRAME-$SIZE_T*26`($sp)   ; restore real inp
 325         $POP    $num,`$FRAME-$SIZE_T*25`($sp)   ; restore real num
 326         addic.  $num,$num,`-16*$SZ`             ; num--
 327         bne     Lunaligned
 328
 329 Ldone:
 330         $POP    r0,`$FRAME+$LRSAVE`($sp)
 331         $POP    r14,`$FRAME-$SIZE_T*18`($sp)
 332         $POP    r15,`$FRAME-$SIZE_T*17`($sp)
 333         $POP    r16,`$FRAME-$SIZE_T*16`($sp)
 334         $POP    r17,`$FRAME-$SIZE_T*15`($sp)
 335         $POP    r18,`$FRAME-$SIZE_T*14`($sp)
 336         $POP    r19,`$FRAME-$SIZE_T*13`($sp)
 337         $POP    r20,`$FRAME-$SIZE_T*12`($sp)
 338         $POP    r21,`$FRAME-$SIZE_T*11`($sp)
 339         $POP    r22,`$FRAME-$SIZE_T*10`($sp)
 340         $POP    r23,`$FRAME-$SIZE_T*9`($sp)
 341         $POP    r24,`$FRAME-$SIZE_T*8`($sp)
 342         $POP    r25,`$FRAME-$SIZE_T*7`($sp)
 343         $POP    r26,`$FRAME-$SIZE_T*6`($sp)
 344         $POP    r27,`$FRAME-$SIZE_T*5`($sp)
 345         $POP    r28,`$FRAME-$SIZE_T*4`($sp)
 346         $POP    r29,`$FRAME-$SIZE_T*3`($sp)
 347         $POP    r30,`$FRAME-$SIZE_T*2`($sp)
 348         $POP    r31,`$FRAME-$SIZE_T*1`($sp)
 349         mtlr    r0
 350         addi    $sp,$sp,$FRAME
 351         blr
 352         .long   0
 353         .byte   0,12,4,1,0x80,18,3,0
 354         .long   0
 355 ___
 356
 357 if ($SZ==4 || $SIZE_T==8) {
 358 $code.=<<___;
 359 .align  4
 360 Lsha2_block_private:
 361         $LD     $t1,0($Tbl)
 362 ___
 363 for($i=0;$i<16;$i++) {
 364 $code.=<<___ if ($SZ==4 && !$LITTLE_ENDIAN);
 365         lwz     @X[$i],`$i*$SZ`($inp)
 366 ___
 367 $code.=<<___ if ($SZ==4 && $LITTLE_ENDIAN);
 368         lwz     $a0,`$i*$SZ`($inp)
 369         rotlwi  @X[$i],$a0,8
 370         rlwimi  @X[$i],$a0,24,0,7
 371         rlwimi  @X[$i],$a0,24,16,23
 372 ___
 373 # 64-bit loads are split to 2x32-bit ones, as CPU can't handle
 374 # unaligned 64-bit loads, only 32-bit ones...
 375 $code.=<<___ if ($SZ==8 && !$LITTLE_ENDIAN);
 376         lwz     $t0,`$i*$SZ`($inp)
 377         lwz     @X[$i],`$i*$SZ+4`($inp)
 378         insrdi  @X[$i],$t0,32,0
 379 ___
 380 $code.=<<___ if ($SZ==8 && $LITTLE_ENDIAN);
 381         lwz     $a0,`$i*$SZ`($inp)
 382          lwz    $a1,`$i*$SZ+4`($inp)
 383         rotlwi  $t0,$a0,8
 384          rotlwi @X[$i],$a1,8
 385         rlwimi  $t0,$a0,24,0,7
 386          rlwimi @X[$i],$a1,24,0,7
 387         rlwimi  $t0,$a0,24,16,23
 388          rlwimi @X[$i],$a1,24,16,23
 389         insrdi  @X[$i],$t0,32,0
 390 ___
 391         &ROUND_00_15($i,@V);
 392         unshift(@V,pop(@V));
 393 }
 394 $code.=<<___;
 395         li      $t0,`$rounds/16-1`
 396         mtctr   $t0
 397 .align  4
 398 Lrounds:
 399         addi    $Tbl,$Tbl,`16*$SZ`
 400 ___
 401 for(;$i<32;$i++) {
 402         &ROUND_16_xx($i,@V);
 403         unshift(@V,pop(@V));
 404 }
 405 $code.=<<___;
 406         bdnz    Lrounds
 407
 408         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 409         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 410         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 411         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 412
 413         $LD     r16,`0*$SZ`($ctx)
 414         $LD     r17,`1*$SZ`($ctx)
 415         $LD     r18,`2*$SZ`($ctx)
 416         $LD     r19,`3*$SZ`($ctx)
 417         $LD     r20,`4*$SZ`($ctx)
 418         $LD     r21,`5*$SZ`($ctx)
 419         $LD     r22,`6*$SZ`($ctx)
 420         addi    $inp,$inp,`16*$SZ`              ; advance inp
 421         $LD     r23,`7*$SZ`($ctx)
 422         add     $A,$A,r16
 423         add     $B,$B,r17
 424         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 425         add     $C,$C,r18
 426         $ST     $A,`0*$SZ`($ctx)
 427         add     $D,$D,r19
 428         $ST     $B,`1*$SZ`($ctx)
 429         add     $E,$E,r20
 430         $ST     $C,`2*$SZ`($ctx)
 431         add     $F,$F,r21
 432         $ST     $D,`3*$SZ`($ctx)
 433         add     $G,$G,r22
 434         $ST     $E,`4*$SZ`($ctx)
 435         add     $H,$H,r23
 436         $ST     $F,`5*$SZ`($ctx)
 437         $ST     $G,`6*$SZ`($ctx)
 438         $UCMP   $inp,$num
 439         $ST     $H,`7*$SZ`($ctx)
 440         bne     Lsha2_block_private
 441         blr
 442         .long   0
 443         .byte   0,12,0x14,0,0,0,0,0
 444 .size   $func,.-$func
 445 ___
 446 } else {
 447 ########################################################################
 448 # SHA512 for PPC32, X vector is off-loaded to stack...
 449 #
 450 #                       |       sha512
 451 #                       |       -m32
 452 # ----------------------+-----------------------
 453 # PPC74x0,gcc-4.0.1     |       +48%
 454 # POWER6,gcc-4.4.6      |       +124%(*)
 455 # POWER7,gcc-4.4.6      |       +79%(*)
 456 # e300,gcc-4.1.0        |       +167%
 457 #
 458 # (*)   ~1/3 of -m64 result [and ~20% better than -m32 code generated
 459 #       by xlc-12.1]
 460
 461 my $XOFF=$LOCALS;
 462
 463 my @V=map("r$_",(16..31));      # A..H
 464
 465 my ($s0,$s1,$t0,$t1,$t2,$t3,$a0,$a1,$a2,$a3)=map("r$_",(0,5,6,8..12,14,15));
 466 my ($x0,$x1)=("r3","r4");       # zaps $ctx and $inp
 467
 468 sub ROUND_00_15_ppc32 {
 469 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 470         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 471
 472 $code.=<<___;
 473         lwz     $t2,`$SZ*($i%16)+($LITTLE_ENDIAN^4)`($Tbl)
 474          xor    $a0,$flo,$glo
 475         lwz     $t3,`$SZ*($i%16)+($LITTLE_ENDIAN^0)`($Tbl)
 476          xor    $a1,$fhi,$ghi
 477         addc    $hlo,$hlo,$t0                   ; h+=x[i]
 478         stw     $t0,`$XOFF+0+$SZ*($i%16)`($sp)  ; save x[i]
 479
 480         srwi    $s0,$elo,$Sigma1[0]
 481         srwi    $s1,$ehi,$Sigma1[0]
 482          and    $a0,$a0,$elo
 483         adde    $hhi,$hhi,$t1
 484          and    $a1,$a1,$ehi
 485         stw     $t1,`$XOFF+4+$SZ*($i%16)`($sp)
 486         srwi    $t0,$elo,$Sigma1[1]
 487         srwi    $t1,$ehi,$Sigma1[1]
 488          addc   $hlo,$hlo,$t2                   ; h+=K512[i]
 489         insrwi  $s0,$ehi,$Sigma1[0],0
 490         insrwi  $s1,$elo,$Sigma1[0],0
 491          xor    $a0,$a0,$glo                    ; Ch(e,f,g)
 492          adde   $hhi,$hhi,$t3
 493          xor    $a1,$a1,$ghi
 494         insrwi  $t0,$ehi,$Sigma1[1],0
 495         insrwi  $t1,$elo,$Sigma1[1],0
 496          addc   $hlo,$hlo,$a0                   ; h+=Ch(e,f,g)
 497         srwi    $t2,$ehi,$Sigma1[2]-32
 498         srwi    $t3,$elo,$Sigma1[2]-32
 499         xor     $s0,$s0,$t0
 500         xor     $s1,$s1,$t1
 501         insrwi  $t2,$elo,$Sigma1[2]-32,0
 502         insrwi  $t3,$ehi,$Sigma1[2]-32,0
 503          xor    $a0,$alo,$blo                   ; a^b, b^c in next round
 504          adde   $hhi,$hhi,$a1
 505          xor    $a1,$ahi,$bhi
 506         xor     $s0,$s0,$t2                     ; Sigma1(e)
 507         xor     $s1,$s1,$t3
 508
 509         srwi    $t0,$alo,$Sigma0[0]
 510          and    $a2,$a2,$a0
 511          addc   $hlo,$hlo,$s0                   ; h+=Sigma1(e)
 512          and    $a3,$a3,$a1
 513         srwi    $t1,$ahi,$Sigma0[0]
 514         srwi    $s0,$ahi,$Sigma0[1]-32
 515          adde   $hhi,$hhi,$s1
 516         srwi    $s1,$alo,$Sigma0[1]-32
 517         insrwi  $t0,$ahi,$Sigma0[0],0
 518         insrwi  $t1,$alo,$Sigma0[0],0
 519          xor    $a2,$a2,$blo                    ; Maj(a,b,c)
 520          addc   $dlo,$dlo,$hlo                  ; d+=h
 521          xor    $a3,$a3,$bhi
 522         insrwi  $s0,$alo,$Sigma0[1]-32,0
 523         insrwi  $s1,$ahi,$Sigma0[1]-32,0
 524          adde   $dhi,$dhi,$hhi
 525         srwi    $t2,$ahi,$Sigma0[2]-32
 526         srwi    $t3,$alo,$Sigma0[2]-32
 527         xor     $s0,$s0,$t0
 528          addc   $hlo,$hlo,$a2                   ; h+=Maj(a,b,c)
 529         xor     $s1,$s1,$t1
 530         insrwi  $t2,$alo,$Sigma0[2]-32,0
 531         insrwi  $t3,$ahi,$Sigma0[2]-32,0
 532          adde   $hhi,$hhi,$a3
 533 ___
 534 $code.=<<___ if ($i>=15);
 535         lwz     $t0,`$XOFF+0+$SZ*(($i+2)%16)`($sp)
 536         lwz     $t1,`$XOFF+4+$SZ*(($i+2)%16)`($sp)
 537 ___
 538 $code.=<<___ if ($i<15 && !$LITTLE_ENDIAN);
 539         lwz     $t1,`$SZ*($i+1)+0`($inp)
 540         lwz     $t0,`$SZ*($i+1)+4`($inp)
 541 ___
 542 $code.=<<___ if ($i<15 && $LITTLE_ENDIAN);
 543         lwz     $a2,`$SZ*($i+1)+0`($inp)
 544          lwz    $a3,`$SZ*($i+1)+4`($inp)
 545         rotlwi  $t1,$a2,8
 546          rotlwi $t0,$a3,8
 547         rlwimi  $t1,$a2,24,0,7
 548          rlwimi $t0,$a3,24,0,7
 549         rlwimi  $t1,$a2,24,16,23
 550          rlwimi $t0,$a3,24,16,23
 551 ___
 552 $code.=<<___;
 553         xor     $s0,$s0,$t2                     ; Sigma0(a)
 554         xor     $s1,$s1,$t3
 555         addc    $hlo,$hlo,$s0                   ; h+=Sigma0(a)
 556         adde    $hhi,$hhi,$s1
 557 ___
 558 $code.=<<___ if ($i==15);
 559         lwz     $x0,`$XOFF+0+$SZ*(($i+1)%16)`($sp)
 560         lwz     $x1,`$XOFF+4+$SZ*(($i+1)%16)`($sp)
 561 ___
 562 }
 563 sub ROUND_16_xx_ppc32 {
 564 my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
 565         $ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
 566
 567 $code.=<<___;
 568         srwi    $s0,$t0,$sigma0[0]
 569         srwi    $s1,$t1,$sigma0[0]
 570         srwi    $t2,$t0,$sigma0[1]
 571         srwi    $t3,$t1,$sigma0[1]
 572         insrwi  $s0,$t1,$sigma0[0],0
 573         insrwi  $s1,$t0,$sigma0[0],0
 574         srwi    $a0,$t0,$sigma0[2]
 575         insrwi  $t2,$t1,$sigma0[1],0
 576         insrwi  $t3,$t0,$sigma0[1],0
 577         insrwi  $a0,$t1,$sigma0[2],0
 578         xor     $s0,$s0,$t2
 579          lwz    $t2,`$XOFF+0+$SZ*(($i+14)%16)`($sp)
 580         srwi    $a1,$t1,$sigma0[2]
 581         xor     $s1,$s1,$t3
 582          lwz    $t3,`$XOFF+4+$SZ*(($i+14)%16)`($sp)
 583         xor     $a0,$a0,$s0
 584          srwi   $s0,$t2,$sigma1[0]
 585         xor     $a1,$a1,$s1
 586          srwi   $s1,$t3,$sigma1[0]
 587         addc    $x0,$x0,$a0                     ; x[i]+=sigma0(x[i+1])
 588          srwi   $a0,$t3,$sigma1[1]-32
 589         insrwi  $s0,$t3,$sigma1[0],0
 590         insrwi  $s1,$t2,$sigma1[0],0
 591         adde    $x1,$x1,$a1
 592          srwi   $a1,$t2,$sigma1[1]-32
 593
 594         insrwi  $a0,$t2,$sigma1[1]-32,0
 595         srwi    $t2,$t2,$sigma1[2]
 596         insrwi  $a1,$t3,$sigma1[1]-32,0
 597         insrwi  $t2,$t3,$sigma1[2],0
 598         xor     $s0,$s0,$a0
 599          lwz    $a0,`$XOFF+0+$SZ*(($i+9)%16)`($sp)
 600         srwi    $t3,$t3,$sigma1[2]
 601         xor     $s1,$s1,$a1
 602          lwz    $a1,`$XOFF+4+$SZ*(($i+9)%16)`($sp)
 603         xor     $s0,$s0,$t2
 604          addc   $x0,$x0,$a0                     ; x[i]+=x[i+9]
 605         xor     $s1,$s1,$t3
 606          adde   $x1,$x1,$a1
 607         addc    $x0,$x0,$s0                     ; x[i]+=sigma1(x[i+14])
 608         adde    $x1,$x1,$s1
 609 ___
 610         ($t0,$t1,$x0,$x1) = ($x0,$x1,$t0,$t1);
 611         &ROUND_00_15_ppc32(@_);
 612 }
 613
 614 $code.=<<___;
 615 .align  4
 616 Lsha2_block_private:
 617 ___
 618 $code.=<<___ if (!$LITTLE_ENDIAN);
 619         lwz     $t1,0($inp)
 620         xor     $a2,@V[3],@V[5]         ; B^C, magic seed
 621         lwz     $t0,4($inp)
 622         xor     $a3,@V[2],@V[4]
 623 ___
 624 $code.=<<___ if ($LITTLE_ENDIAN);
 625         lwz     $a1,0($inp)
 626         xor     $a2,@V[3],@V[5]         ; B^C, magic seed
 627         lwz     $a0,4($inp)
 628         xor     $a3,@V[2],@V[4]
 629         rotlwi  $t1,$a1,8
 630          rotlwi $t0,$a0,8
 631         rlwimi  $t1,$a1,24,0,7
 632          rlwimi $t0,$a0,24,0,7
 633         rlwimi  $t1,$a1,24,16,23
 634          rlwimi $t0,$a0,24,16,23
 635 ___
 636 for($i=0;$i<16;$i++) {
 637         &ROUND_00_15_ppc32($i,@V);
 638         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 639         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 640 }
 641 $code.=<<___;
 642         li      $a0,`$rounds/16-1`
 643         mtctr   $a0
 644 .align  4
 645 Lrounds:
 646         addi    $Tbl,$Tbl,`16*$SZ`
 647 ___
 648 for(;$i<32;$i++) {
 649         &ROUND_16_xx_ppc32($i,@V);
 650         unshift(@V,pop(@V));    unshift(@V,pop(@V));
 651         ($a0,$a1,$a2,$a3) = ($a2,$a3,$a0,$a1);
 652 }
 653 $code.=<<___;
 654         bdnz    Lrounds
 655
 656         $POP    $ctx,`$FRAME-$SIZE_T*22`($sp)
 657         $POP    $inp,`$FRAME-$SIZE_T*23`($sp)   ; inp pointer
 658         $POP    $num,`$FRAME-$SIZE_T*24`($sp)   ; end pointer
 659         subi    $Tbl,$Tbl,`($rounds-16)*$SZ`    ; rewind Tbl
 660
 661         lwz     $t0,`$LITTLE_ENDIAN^0`($ctx)
 662         lwz     $t1,`$LITTLE_ENDIAN^4`($ctx)
 663         lwz     $t2,`$LITTLE_ENDIAN^8`($ctx)
 664         lwz     $t3,`$LITTLE_ENDIAN^12`($ctx)
 665         lwz     $a0,`$LITTLE_ENDIAN^16`($ctx)
 666         lwz     $a1,`$LITTLE_ENDIAN^20`($ctx)
 667         lwz     $a2,`$LITTLE_ENDIAN^24`($ctx)
 668         addc    @V[1],@V[1],$t1
 669         lwz     $a3,`$LITTLE_ENDIAN^28`($ctx)
 670         adde    @V[0],@V[0],$t0
 671         lwz     $t0,`$LITTLE_ENDIAN^32`($ctx)
 672         addc    @V[3],@V[3],$t3
 673         lwz     $t1,`$LITTLE_ENDIAN^36`($ctx)
 674         adde    @V[2],@V[2],$t2
 675         lwz     $t2,`$LITTLE_ENDIAN^40`($ctx)
 676         addc    @V[5],@V[5],$a1
 677         lwz     $t3,`$LITTLE_ENDIAN^44`($ctx)
 678         adde    @V[4],@V[4],$a0
 679         lwz     $a0,`$LITTLE_ENDIAN^48`($ctx)
 680         addc    @V[7],@V[7],$a3
 681         lwz     $a1,`$LITTLE_ENDIAN^52`($ctx)
 682         adde    @V[6],@V[6],$a2
 683         lwz     $a2,`$LITTLE_ENDIAN^56`($ctx)
 684         addc    @V[9],@V[9],$t1
 685         lwz     $a3,`$LITTLE_ENDIAN^60`($ctx)
 686         adde    @V[8],@V[8],$t0
 687         stw     @V[0],`$LITTLE_ENDIAN^0`($ctx)
 688         stw     @V[1],`$LITTLE_ENDIAN^4`($ctx)
 689         addc    @V[11],@V[11],$t3
 690         stw     @V[2],`$LITTLE_ENDIAN^8`($ctx)
 691         stw     @V[3],`$LITTLE_ENDIAN^12`($ctx)
 692         adde    @V[10],@V[10],$t2
 693         stw     @V[4],`$LITTLE_ENDIAN^16`($ctx)
 694         stw     @V[5],`$LITTLE_ENDIAN^20`($ctx)
 695         addc    @V[13],@V[13],$a1
 696         stw     @V[6],`$LITTLE_ENDIAN^24`($ctx)
 697         stw     @V[7],`$LITTLE_ENDIAN^28`($ctx)
 698         adde    @V[12],@V[12],$a0
 699         stw     @V[8],`$LITTLE_ENDIAN^32`($ctx)
 700         stw     @V[9],`$LITTLE_ENDIAN^36`($ctx)
 701         addc    @V[15],@V[15],$a3
 702         stw     @V[10],`$LITTLE_ENDIAN^40`($ctx)
 703         stw     @V[11],`$LITTLE_ENDIAN^44`($ctx)
 704         adde    @V[14],@V[14],$a2
 705         stw     @V[12],`$LITTLE_ENDIAN^48`($ctx)
 706         stw     @V[13],`$LITTLE_ENDIAN^52`($ctx)
 707         stw     @V[14],`$LITTLE_ENDIAN^56`($ctx)
 708         stw     @V[15],`$LITTLE_ENDIAN^60`($ctx)
 709
 710         addi    $inp,$inp,`16*$SZ`              ; advance inp
 711         $PUSH   $inp,`$FRAME-$SIZE_T*23`($sp)
 712         $UCMP   $inp,$num
 713         bne     Lsha2_block_private
 714         blr
 715         .long   0
 716         .byte   0,12,0x14,0,0,0,0,0
 717 .size   $func,.-$func
 718 ___
 719 }
 720
 721 # Ugly hack here, because PPC assembler syntax seem to vary too
 722 # much from platforms to platform...
 723 $code.=<<___;
 724 .align  6
 725 LPICmeup:
 726         mflr    r0
 727         bcl     20,31,\$+4
 728         mflr    $Tbl    ; vvvvvv "distance" between . and 1st data entry
 729         addi    $Tbl,$Tbl,`64-8`
 730         mtlr    r0
 731         blr
 732         .long   0
 733         .byte   0,12,0x14,0,0,0,0,0
 734         .space  `64-9*4`
 735 ___
 736 $code.=<<___ if ($SZ==8);
 737         .quad   0x428a2f98d728ae22,0x7137449123ef65cd
 738         .quad   0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
 739         .quad   0x3956c25bf348b538,0x59f111f1b605d019
 740         .quad   0x923f82a4af194f9b,0xab1c5ed5da6d8118
 741         .quad   0xd807aa98a3030242,0x12835b0145706fbe
 742         .quad   0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
 743         .quad   0x72be5d74f27b896f,0x80deb1fe3b1696b1
 744         .quad   0x9bdc06a725c71235,0xc19bf174cf692694
 745         .quad   0xe49b69c19ef14ad2,0xefbe4786384f25e3
 746         .quad   0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
 747         .quad   0x2de92c6f592b0275,0x4a7484aa6ea6e483
 748         .quad   0x5cb0a9dcbd41fbd4,0x76f988da831153b5
 749         .quad   0x983e5152ee66dfab,0xa831c66d2db43210
 750         .quad   0xb00327c898fb213f,0xbf597fc7beef0ee4
 751         .quad   0xc6e00bf33da88fc2,0xd5a79147930aa725
 752         .quad   0x06ca6351e003826f,0x142929670a0e6e70
 753         .quad   0x27b70a8546d22ffc,0x2e1b21385c26c926
 754         .quad   0x4d2c6dfc5ac42aed,0x53380d139d95b3df
 755         .quad   0x650a73548baf63de,0x766a0abb3c77b2a8
 756         .quad   0x81c2c92e47edaee6,0x92722c851482353b
 757         .quad   0xa2bfe8a14cf10364,0xa81a664bbc423001
 758         .quad   0xc24b8b70d0f89791,0xc76c51a30654be30
 759         .quad   0xd192e819d6ef5218,0xd69906245565a910
 760         .quad   0xf40e35855771202a,0x106aa07032bbd1b8
 761         .quad   0x19a4c116b8d2d0c8,0x1e376c085141ab53
 762         .quad   0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
 763         .quad   0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
 764         .quad   0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
 765         .quad   0x748f82ee5defb2fc,0x78a5636f43172f60
 766         .quad   0x84c87814a1f0ab72,0x8cc702081a6439ec
 767         .quad   0x90befffa23631e28,0xa4506cebde82bde9
 768         .quad   0xbef9a3f7b2c67915,0xc67178f2e372532b
 769         .quad   0xca273eceea26619c,0xd186b8c721c0c207
 770         .quad   0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
 771         .quad   0x06f067aa72176fba,0x0a637dc5a2c898a6
 772         .quad   0x113f9804bef90dae,0x1b710b35131c471b
 773         .quad   0x28db77f523047d84,0x32caab7b40c72493
 774         .quad   0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
 775         .quad   0x4cc5d4becb3e42b6,0x597f299cfc657e2a
 776         .quad   0x5fcb6fab3ad6faec,0x6c44198c4a475817
 777 ___
 778 $code.=<<___ if ($SZ==4);
 779         .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
 780         .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
 781         .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
 782         .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
 783         .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
 784         .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
 785         .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
 786         .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
 787         .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
 788         .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
 789         .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
 790         .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
 791         .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
 792         .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
 793         .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
 794         .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
 795 ___
 796
 797 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 798 print $code;
 799 close STDOUT;