crypto/bn/asm/ppc64-mont.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 # December 2007
  11
  12 $output = shift;
  13
  14 if ($output =~ /32\-mont\.s/) {
  15         $SIZE_T=4;
  16         $RZONE= 224;
  17         $FRAME= $SIZE_T*16+8*12;
  18         $fname= "bn_mul_mont_ppc64";
  19
  20         $STUX=  "stwux";        # store indexed and update
  21         $PUSH=  "stw";
  22         $POP=   "lwz";
  23         die "not implemented yet";
  24 } elsif ($output =~ /64\-mont\.s/) {
  25         $SIZE_T=8;
  26         $RZONE= 288;
  27         $FRAME= $SIZE_T*16+8*12;
  28         $fname= "bn_mul_mont";
  29
  30         # same as above, but 64-bit mnemonics...
  31         $STUX=  "stdux";        # store indexed and update
  32         $PUSH=  "std";
  33         $POP=   "ld";
  34 } else { die "nonsense $output"; }
  35
  36 ( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
  37         die "can't call ../perlasm/ppc-xlate.pl: $!";
  38
  39 $TRANSFER=8*8;
  40
  41 $sp="r1";
  42 $toc="r2";
  43 $rp="r3";       $ovf="r3";
  44 $ap="r4";
  45 $bp="r5";
  46 $np="r6";
  47 $n0="r7";
  48 $num="r8";
  49 $rp="r9";       # $rp is reassigned
  50 $tp="r10";
  51 $j="r11";
  52 $i="r12";
  53 # non-volatile registers
  54 $ap_l="r14";
  55 $ap_h="r15";
  56 $np_l="r16";
  57 $np_h="r17";
  58 $carry="r18";
  59 $a0="r19";      # ap[0]
  60 $t0="r20";
  61 $t1="r21";
  62 $t2="r22";
  63 $t3="r23";
  64 $t4="r24";
  65 $t5="r25";
  66 $t6="r26";
  67 $t7="r27";
  68
  69 # PPC offers enough register bank capacity to unroll inner loops twice
  70 #
  71 #     ..A3A2A1A0
  72 #           dcba
  73 #    -----------
  74 #            A0a
  75 #           A0b
  76 #          A0c
  77 #         A0d
  78 #          A1a
  79 #         A1b
  80 #        A1c
  81 #       A1d
  82 #        A2a
  83 #       A2b
  84 #      A2c
  85 #     A2d
  86 #      A3a
  87 #     A3b
  88 #    A3c
  89 #   A3d
  90 #    ..a
  91 #   ..b
  92 #
  93 $ba="f0";
  94 $bb="f1";
  95 $bc="f2";
  96 $bd="f3";
  97 $na="f4";
  98 $nb="f5";
  99 $nc="f6";
 100 $nd="f7";
 101 $dota="f8";
 102 $dotb="f9";
 103 $A0="f10";
 104 $A1="f11";
 105 $A2="f12";
 106 $A3="f13";
 107 $N0="f14";
 108 $N1="f15";
 109 $N2="f16";
 110 $N3="f17";
 111 $T0a="f18";
 112 $T0b="f19";
 113 $T1a="f20";
 114 $T1b="f21";
 115 $T2a="f22";
 116 $T2b="f23";
 117 $T3a="f24";
 118 $T3b="f25";
 119
 120 # sp----------->+-------------------------------+
 121 #               | saved sp                      |
 122 #               +-------------------------------+
 123 #               |                               |
 124 #               +-------------------------------+
 125 #               | 14 saved gpr, r14-r27         |
 126 #               .                               .
 127 #               .                               .
 128 #   +16*size_t  +-------------------------------+
 129 #               | 12 saved fpr, f14-f25         |
 130 #               .                               .
 131 #               .                               .
 132 #   +12*8       +-------------------------------+
 133 #               | 8 gpr<->fpr transfer zone     |
 134 #               .                               .
 135 #               .                               .
 136 #   +8*8        +-------------------------------+
 137 #               | __int64 tmp[-1]               |
 138 #               +-------------------------------+
 139 #               | __int64 tmp[num]              |
 140 #               .                               .
 141 #               .                               .
 142 #               .                               .
 143 #   +(num+1)*8  +-------------------------------+
 144 #               | double a_lo[num]              |
 145 #               .                               .
 146 #               .                               .
 147 #               .                               .
 148 #   +num*8      +-------------------------------+
 149 #               | double a_hi[num]              |
 150 #               .                               .
 151 #               .                               .
 152 #               .                               .
 153 #   +num*8      +-------------------------------+
 154 #               | double n_lo[num]              |
 155 #               .                               .
 156 #               .                               .
 157 #               .                               .
 158 #   +num*8      +-------------------------------+
 159 #               | double n_hi[num]              |
 160 #               .                               .
 161 #               .                               .
 162 #               .                               .
 163 #               +-------------------------------+
 164
 165 $code=<<___;
 166 .machine "any"
 167 .text
 168
 169 .globl  .$fname
 170 .align  4
 171 .$fname:
 172         cmpwi   $num,4
 173         mr      $rp,r3          ; $rp is reassigned
 174         li      r3,0            ; possible "not handled" return code
 175         bltlr-
 176         andi.   r0,$num,1       ; $num has to be even
 177         bnelr-
 178
 179         slwi    $num,$num,3     ; num*=8
 180         li      $i,-4096
 181         slwi    $tp,$num,2      ; place for {an}p_{lh}[num], i.e. 4*num
 182         add     $tp,$tp,$num    ; place for tp[num+1]
 183         addi    $tp,$tp,`$FRAME+$TRANSFER+8+$RZONE`
 184         subf    $tp,$tp,$sp     ; $sp-$tp
 185         and     $tp,$tp,$i      ; minimize TLB usage
 186         subf    $tp,$sp,$tp     ; $tp-$sp
 187         $STUX   $sp,$sp,$tp     ; alloca
 188
 189         $PUSH   r14,`2*$SIZE_T`($sp)
 190         $PUSH   r15,`3*$SIZE_T`($sp)
 191         $PUSH   r16,`4*$SIZE_T`($sp)
 192         $PUSH   r17,`5*$SIZE_T`($sp)
 193         $PUSH   r18,`6*$SIZE_T`($sp)
 194         $PUSH   r19,`7*$SIZE_T`($sp)
 195         $PUSH   r20,`8*$SIZE_T`($sp)
 196         $PUSH   r21,`9*$SIZE_T`($sp)
 197         $PUSH   r22,`10*$SIZE_T`($sp)
 198         $PUSH   r23,`11*$SIZE_T`($sp)
 199         $PUSH   r24,`12*$SIZE_T`($sp)
 200         $PUSH   r25,`13*$SIZE_T`($sp)
 201         $PUSH   r26,`14*$SIZE_T`($sp)
 202         $PUSH   r27,`15*$SIZE_T`($sp)
 203         stfd    f14,`16*$SIZE_T+0`($sp)
 204         stfd    f15,`16*$SIZE_T+8`($sp)
 205         stfd    f16,`16*$SIZE_T+16`($sp)
 206         stfd    f17,`16*$SIZE_T+24`($sp)
 207         stfd    f18,`16*$SIZE_T+32`($sp)
 208         stfd    f19,`16*$SIZE_T+40`($sp)
 209         stfd    f20,`16*$SIZE_T+48`($sp)
 210         stfd    f21,`16*$SIZE_T+56`($sp)
 211         stfd    f22,`16*$SIZE_T+64`($sp)
 212         stfd    f23,`16*$SIZE_T+72`($sp)
 213         stfd    f24,`16*$SIZE_T+80`($sp)
 214         stfd    f25,`16*$SIZE_T+88`($sp)
 215         std     r0,$FRAME($sp)  ; r0 is still 0
 216         lfd     $dota,$FRAME($sp)
 217         lfd     $dotb,$FRAME($sp)
 218
 219         addi    $tp,$sp,`$FRAME+$TRANSFER`
 220         ; note that {an}p_{lh} are off by 1, this is because they
 221         ; are used with stfdu/lfdu instruction...
 222         add     $ap_l,$tp,$num
 223         add     $ap_h,$ap_l,$num
 224         add     $np_l,$ap_h,$num
 225         add     $np_h,$np_l,$num
 226
 227         ld      $a0,0($ap)      ; pull ap[0] value
 228         ld      $n0,0($n0)      ; pull n0[0] value
 229         srwi    $j,$num,`3+1`   ; counter register, num/2
 230 \f
 231         ld      $t3,0($bp)      ; bp[0]
 232         mulld   $t7,$a0,$t3     ; ap[0]*bp[0]
 233         mulld   $t7,$t7,$n0     ; tp[0]*n0
 234
 235         ; transfer bp[0] to FPU as 4x16-bit values
 236         extrdi  $t0,$t3,16,48
 237         extrdi  $t1,$t3,16,32
 238         extrdi  $t2,$t3,16,16
 239         extrdi  $t3,$t3,16,0
 240         std     $t0,`$FRAME+0`($sp)
 241         std     $t1,`$FRAME+8`($sp)
 242         std     $t2,`$FRAME+16`($sp)
 243         std     $t3,`$FRAME+24`($sp)
 244         lfd     $ba,`$FRAME+0`($sp)
 245         lfd     $bb,`$FRAME+8`($sp)
 246         lfd     $bc,`$FRAME+16`($sp)
 247         lfd     $bd,`$FRAME+24`($sp)
 248         fcfid   $ba,$ba
 249         fcfid   $bb,$bb
 250         fcfid   $bc,$bc
 251         fcfid   $bd,$bd
 252
 253         ; transfer (ap[0]*bp[0])*n0 to FPU as 4x16-bit values
 254         extrdi  $t4,$t7,16,48
 255         extrdi  $t5,$t7,16,32
 256         extrdi  $t6,$t7,16,16
 257         extrdi  $t7,$t7,16,0
 258         std     $t4,`$FRAME+32`($sp)
 259         std     $t5,`$FRAME+40`($sp)
 260         std     $t6,`$FRAME+48`($sp)
 261         std     $t7,`$FRAME+56`($sp)
 262         lfd     $na,`$FRAME+32`($sp)
 263         lfd     $nb,`$FRAME+40`($sp)
 264         lfd     $nc,`$FRAME+48`($sp)
 265         lfd     $nd,`$FRAME+56`($sp)
 266         fcfid   $na,$na
 267         fcfid   $nb,$nb
 268         fcfid   $nc,$nc
 269         fcfid   $nd,$nd
 270 \f
 271         addi    $tp,$sp,`$FRAME+$TRANSFER-8`
 272         li      $carry,0
 273         mtctr   $j
 274 .align  4
 275 L1st:
 276         lwz     $t0,4($ap)              ; load a[j] as 32-bit word pair
 277         lwz     $t1,0($ap)
 278         lwz     $t2,4($np)              ; load n[j] as 32-bit word pair
 279         lwz     $t3,0($np)
 280         std     $t0,`$FRAME+0`($sp)
 281         std     $t1,`$FRAME+8`($sp)
 282         std     $t2,`$FRAME+16`($sp)
 283         std     $t3,`$FRAME+24`($sp)
 284         lfd     $A0,`$FRAME+0`($sp)
 285         lfd     $A1,`$FRAME+8`($sp)
 286         lfd     $N0,`$FRAME+16`($sp)
 287         lfd     $N1,`$FRAME+24`($sp)
 288         fcfid   $A0,$A0
 289         fcfid   $A1,$A1
 290         fcfid   $N0,$N0
 291         fcfid   $N1,$N1
 292         stfdu   $A0,8($ap_l)            ; save a[j] in double format
 293         stfdu   $A1,8($ap_h)
 294         stfdu   $N0,8($np_l)            ; save n[j] in double format
 295         stfdu   $N1,8($np_h)
 296
 297         lwz     $t4,12($ap)             ; load a[j+1] as 32-bit word pair
 298         lwz     $t5,8($ap)
 299         lwz     $t6,12($np)             ; load n[j+1] as 32-bit word pair
 300         lwz     $t7,8($np)
 301         std     $t4,`$FRAME+32`($sp)
 302         std     $t5,`$FRAME+40`($sp)
 303         std     $t6,`$FRAME+48`($sp)
 304         std     $t7,`$FRAME+56`($sp)
 305         lfd     $A2,`$FRAME+32`($sp)
 306         lfd     $A3,`$FRAME+40`($sp)
 307         lfd     $N2,`$FRAME+48`($sp)
 308         lfd     $N3,`$FRAME+56`($sp)
 309         fcfid   $A2,$A2
 310         fcfid   $A3,$A3
 311         fcfid   $N2,$N2
 312         fcfid   $N3,$N3
 313         stfdu   $A2,8($ap_l)            ; save a[j+1] in double format
 314         stfdu   $A3,8($ap_h)
 315         stfdu   $N2,8($np_l)            ; save n[j+1] in double format
 316         stfdu   $N3,8($np_h)
 317         addi    $ap,$ap,16
 318         addi    $np,$np,16
 319
 320         fmadd   $T0a,$A0,$ba,$dota
 321         fmadd   $T0b,$A0,$bb,$dotb
 322         fmul    $T1a,$A1,$ba
 323         fmul    $T1b,$A1,$bb
 324         fmul    $T2a,$A2,$ba
 325         fmul    $T2b,$A2,$bb
 326         fmul    $T3a,$A3,$ba
 327         fmul    $T3b,$A3,$bb
 328
 329         fmadd   $T1a,$A0,$bc,$T1a
 330         fmadd   $T1b,$A0,$bd,$T1b
 331         fmadd   $T2a,$A1,$bc,$T2a
 332         fmadd   $T2b,$A1,$bd,$T2b
 333         fmadd   $T3a,$A2,$bc,$T3a
 334         fmadd   $T3b,$A2,$bd,$T3b
 335         fmul    $dota,$A3,$bc
 336         fmul    $dotb,$A3,$bd
 337
 338         fmadd   $T0a,$N0,$na,$T0a
 339         fmadd   $T0b,$N0,$nb,$T0b
 340         fmadd   $T1a,$N1,$na,$T1a
 341         fmadd   $T1b,$N1,$nb,$T1b
 342         fmadd   $T2a,$N2,$na,$T2a
 343         fmadd   $T2b,$N2,$nb,$T2b
 344         fmadd   $T3a,$N3,$na,$T3a
 345         fmadd   $T3b,$N3,$nb,$T3b
 346
 347         fmadd   $T1a,$N0,$nc,$T1a
 348         fmadd   $T1b,$N0,$nd,$T1b
 349         fmadd   $T2a,$N1,$nc,$T2a
 350         fmadd   $T2b,$N1,$nd,$T2b
 351         fmadd   $T3a,$N2,$nc,$T3a
 352         fmadd   $T3b,$N2,$nd,$T3b
 353         fmadd   $dota,$N3,$nc,$dota
 354         fmadd   $dotb,$N3,$nd,$dotb
 355
 356         fctid   $T0a,$T0a
 357         fctid   $T0b,$T0b
 358         fctid   $T1a,$T1a
 359         fctid   $T1b,$T1b
 360         fctid   $T2a,$T2a
 361         fctid   $T2b,$T2b
 362         fctid   $T3a,$T3a
 363         fctid   $T3b,$T3b
 364
 365         stfd    $T0a,`$FRAME+0`($sp)
 366         stfd    $T0b,`$FRAME+8`($sp)
 367         stfd    $T1a,`$FRAME+16`($sp)
 368         stfd    $T1b,`$FRAME+24`($sp)
 369         stfd    $T2a,`$FRAME+32`($sp)
 370         stfd    $T2b,`$FRAME+40`($sp)
 371         stfd    $T3a,`$FRAME+48`($sp)
 372         stfd    $T3b,`$FRAME+56`($sp)
 373         ld      $t0,`$FRAME+0`($sp)
 374         ld      $t1,`$FRAME+8`($sp)
 375         ld      $t2,`$FRAME+16`($sp)
 376         ld      $t3,`$FRAME+24`($sp)
 377         ld      $t4,`$FRAME+32`($sp)
 378         ld      $t5,`$FRAME+40`($sp)
 379         ld      $t6,`$FRAME+48`($sp)
 380         ld      $t7,`$FRAME+56`($sp)
 381
 382         add     $t0,$t0,$carry          ; can not overflow
 383         srdi    $carry,$t0,16
 384         add     $t1,$t1,$carry
 385         srdi    $carry,$t1,16
 386         add     $t2,$t2,$carry
 387         srdi    $carry,$t2,16
 388         add     $t3,$t3,$carry
 389         srdi    $carry,$t3,16
 390         add     $t4,$t4,$carry
 391         srdi    $carry,$t4,16
 392         add     $t5,$t5,$carry
 393         srdi    $carry,$t5,16
 394         add     $t6,$t6,$carry
 395         srdi    $carry,$t6,16
 396         add     $t7,$t7,$carry
 397
 398         insrdi  $t0,$t1,16,32
 399         insrdi  $t0,$t2,16,16
 400         insrdi  $t0,$t3,16,0            ; 0..63 bits
 401         insrdi  $t4,$t5,16,32
 402         insrdi  $t4,$t6,16,16
 403         insrdi  $t4,$t7,16,0            ; 64..127 bits
 404         srdi    $carry,$t7,16           ; upper 33 bits
 405
 406         std     $t0,8($tp)              ; tp[j-1]
 407         stdu    $t4,16($tp)             ; tp[j]
 408         bdnz-   L1st
 409 \f
 410         fctid   $dota,$dota
 411         fctid   $dotb,$dotb
 412         stfd    $dota,`$FRAME+0`($sp)
 413         stfd    $dotb,`$FRAME+8`($sp)
 414         ld      $t0,`$FRAME+0`($sp)
 415         ld      $t1,`$FRAME+8`($sp)
 416         add     $t0,$t0,$carry          ; can not overflow
 417         srdi    $carry,$t0,16
 418         add     $t1,$t1,$carry
 419         insrdi  $t0,$t1,48,0
 420         srdi    $ovf,$t1,48
 421         std     $t0,8($tp)              ; tp[num-1]
 422
 423         subf    $ap_l,$num,$ap_l        ; rewind pointers
 424         subf    $ap_h,$num,$ap_h
 425         subf    $np_l,$num,$np_l
 426         subf    $np_h,$num,$np_h
 427 \f
 428         li      $i,8                    ; i=1
 429 .align  4
 430 Louter:
 431         ldx     $t3,$bp,$i      ; bp[i]
 432         ld      $t0,`$FRAME+$TRANSFER+8`($sp)   ; tp[0]
 433         mulld   $t7,$a0,$t3     ; ap[0]*bp[i]
 434         add     $t7,$t7,$t0     ; ap[0]*bp[i]+tp[0]
 435         mulld   $t7,$t7,$n0     ; tp[0]*n0
 436
 437         ; transfer b[i] to FPU as 4x16-bit values
 438         extrdi  $t0,$t3,16,48
 439         extrdi  $t1,$t3,16,32
 440         extrdi  $t2,$t3,16,16
 441         extrdi  $t3,$t3,16,0
 442         std     $t0,`$FRAME+0`($sp)
 443         std     $t1,`$FRAME+8`($sp)
 444         std     $t2,`$FRAME+16`($sp)
 445         std     $t3,`$FRAME+24`($sp)
 446         lfd     $ba,`$FRAME+0`($sp)
 447         lfd     $bb,`$FRAME+8`($sp)
 448         lfd     $bc,`$FRAME+16`($sp)
 449         lfd     $bd,`$FRAME+24`($sp)
 450         fcfid   $ba,$ba
 451         fcfid   $bb,$bb
 452         fcfid   $bc,$bc
 453         fcfid   $bd,$bd
 454
 455         ; transfer (ap[0]*b[i]+t[0])*n0 to FPU as 4x16-bit values
 456         extrdi  $t4,$t7,16,48
 457         extrdi  $t5,$t7,16,32
 458         extrdi  $t6,$t7,16,16
 459         extrdi  $t7,$t7,16,0
 460         std     $t4,`$FRAME+32`($sp)
 461         std     $t5,`$FRAME+40`($sp)
 462         std     $t6,`$FRAME+48`($sp)
 463         std     $t7,`$FRAME+56`($sp)
 464         lfd     $na,`$FRAME+32`($sp)
 465         lfd     $nb,`$FRAME+40`($sp)
 466         lfd     $nc,`$FRAME+48`($sp)
 467         lfd     $nd,`$FRAME+56`($sp)
 468         fcfid   $na,$na
 469         fcfid   $nb,$nb
 470         fcfid   $nc,$nc
 471         fcfid   $nd,$nd
 472 \f
 473         addi    $tp,$sp,`$FRAME+$TRANSFER`
 474         fsub    $dota,$dota,$dota
 475         fsub    $dotb,$dotb,$dotb
 476         li      $carry,0
 477         mtctr   $j
 478 .align  4
 479 Linner:
 480         lfdu    $A0,8($ap_l)            ; load a[j] in double format
 481         lfdu    $A1,8($ap_h)
 482         lfdu    $N0,8($np_l)            ; load n[j] in double format
 483         lfdu    $N1,8($np_h)
 484         lfdu    $A2,8($ap_l)            ; load a[j+1] in double format
 485         lfdu    $A3,8($ap_h)
 486         lfdu    $N2,8($np_l)            ; load n[j+1] in double format
 487         lfdu    $N3,8($np_h)
 488
 489         fmadd   $T0a,$A0,$ba,$dota
 490         fmadd   $T0b,$A0,$bb,$dotb
 491         fmul    $T1a,$A1,$ba
 492         fmul    $T1b,$A1,$bb
 493         fmul    $T2a,$A2,$ba
 494         fmul    $T2b,$A2,$bb
 495         fmul    $T3a,$A3,$ba
 496         fmul    $T3b,$A3,$bb
 497
 498         fmadd   $T1a,$A0,$bc,$T1a
 499         fmadd   $T1b,$A0,$bd,$T1b
 500         fmadd   $T2a,$A1,$bc,$T2a
 501         fmadd   $T2b,$A1,$bd,$T2b
 502         fmadd   $T3a,$A2,$bc,$T3a
 503         fmadd   $T3b,$A2,$bd,$T3b
 504         fmul    $dota,$A3,$bc
 505         fmul    $dotb,$A3,$bd
 506
 507         fmadd   $T0a,$N0,$na,$T0a
 508         fmadd   $T0b,$N0,$nb,$T0b
 509         fmadd   $T1a,$N1,$na,$T1a
 510         fmadd   $T1b,$N1,$nb,$T1b
 511         fmadd   $T2a,$N2,$na,$T2a
 512         fmadd   $T2b,$N2,$nb,$T2b
 513         fmadd   $T3a,$N3,$na,$T3a
 514         fmadd   $T3b,$N3,$nb,$T3b
 515
 516         fmadd   $T1a,$N0,$nc,$T1a
 517         fmadd   $T1b,$N0,$nd,$T1b
 518         fmadd   $T2a,$N1,$nc,$T2a
 519         fmadd   $T2b,$N1,$nd,$T2b
 520         fmadd   $T3a,$N2,$nc,$T3a
 521         fmadd   $T3b,$N2,$nd,$T3b
 522         fmadd   $dota,$N3,$nc,$dota
 523         fmadd   $dotb,$N3,$nd,$dotb
 524
 525         fctid   $T0a,$T0a
 526         fctid   $T0b,$T0b
 527         fctid   $T1a,$T1a
 528         fctid   $T1b,$T1b
 529         fctid   $T2a,$T2a
 530         fctid   $T2b,$T2b
 531         fctid   $T3a,$T3a
 532         fctid   $T3b,$T3b
 533
 534         stfd    $T0a,`$FRAME+0`($sp)
 535         stfd    $T0b,`$FRAME+8`($sp)
 536         stfd    $T1a,`$FRAME+16`($sp)
 537         stfd    $T1b,`$FRAME+24`($sp)
 538         stfd    $T2a,`$FRAME+32`($sp)
 539         stfd    $T2b,`$FRAME+40`($sp)
 540         stfd    $T3a,`$FRAME+48`($sp)
 541         stfd    $T3b,`$FRAME+56`($sp)
 542         ld      $t0,`$FRAME+0`($sp)
 543         ld      $t1,`$FRAME+8`($sp)
 544         ld      $t2,`$FRAME+16`($sp)
 545         ld      $t3,`$FRAME+24`($sp)
 546         ld      $t4,`$FRAME+32`($sp)
 547         ld      $t5,`$FRAME+40`($sp)
 548         ld      $t6,`$FRAME+48`($sp)
 549         ld      $t7,`$FRAME+56`($sp)
 550
 551         add     $t0,$t0,$carry          ; can not overflow
 552         srdi    $carry,$t0,16
 553         add     $t1,$t1,$carry
 554         srdi    $carry,$t1,16
 555         add     $t2,$t2,$carry
 556         srdi    $carry,$t2,16
 557         add     $t3,$t3,$carry
 558         srdi    $carry,$t3,16
 559         add     $t4,$t4,$carry
 560         srdi    $carry,$t4,16
 561         add     $t5,$t5,$carry
 562         srdi    $carry,$t5,16
 563         add     $t6,$t6,$carry
 564         srdi    $carry,$t6,16
 565         add     $t7,$t7,$carry
 566
 567         insrdi  $t0,$t1,16,32
 568         insrdi  $t0,$t2,16,16
 569         insrdi  $t0,$t3,16,0            ; 0..63 bits
 570         insrdi  $t4,$t5,16,32
 571         insrdi  $t4,$t6,16,16
 572         insrdi  $t4,$t7,16,0            ; 64..127 bits
 573         srdi    $carry,$t7,16           ; upper 33 bits
 574
 575         ld      $t1,8($tp)              ; tp[j]
 576         ldu     $t2,16($tp)             ; tp[j+1]
 577
 578         addc    $t3,$t0,$t1
 579         adde    $t5,$t4,$t2
 580         addze   $carry,$carry
 581
 582         std     $t3,-16($tp)            ; tp[j-1]
 583         std     $t5,-8($tp)             ; tp[j]
 584         bdnz-   Linner
 585 \f
 586         fctid   $dota,$dota
 587         fctid   $dotb,$dotb
 588         stfd    $dota,`$FRAME+0`($sp)
 589         stfd    $dotb,`$FRAME+8`($sp)
 590         ld      $t0,`$FRAME+0`($sp)
 591         ld      $t1,`$FRAME+8`($sp)
 592         add     $carry,$carry,$ovf      ; comsume upmost overflow
 593         add     $t0,$t0,$carry          ; can not overflow
 594         srdi    $carry,$t0,16
 595         add     $t1,$t1,$carry
 596         insrdi  $t0,$t1,48,0
 597         srdi    $ovf,$t1,48
 598         std     $t0,0($tp)              ; tp[num-1]
 599
 600         subf    $ap_l,$num,$ap_l        ; rewind pointers
 601         subf    $ap_h,$num,$ap_h
 602         subf    $np_l,$num,$np_l
 603         subf    $np_h,$num,$np_h
 604         addi    $i,$i,8
 605         cmpw    $i,$num
 606         blt-    Louter
 607 \f
 608         subf    $np,$num,$np    ; rewind np
 609         subfc   $i,$i,$i        ; j=0 and "clear" XER[CA]
 610         addi    $tp,$sp,`$FRAME+$TRANSFER+8`
 611         addi    $t4,$sp,`$FRAME+$TRANSFER+16`
 612         addi    $t5,$np,8
 613         addi    $t6,$rp,8
 614         mtctr   $j
 615
 616 .align  4
 617 Lsub:   ldx     $t0,$tp,$i
 618         ldx     $t1,$np,$i
 619         ldx     $t2,$t4,$i
 620         ldx     $t3,$t5,$i
 621         subfe   $t0,$t1,$t0     ; tp[j]-np[j]
 622         subfe   $t2,$t3,$t2     ; tp[j+1]-np[j+1]
 623         stdx    $t0,$rp,$i
 624         stdx    $t2,$t6,$i
 625         addi    $i,$i,16
 626         bdnz-   Lsub
 627
 628         li      $i,0
 629         subfe   $ovf,$i,$ovf    ; handle upmost overflow bit
 630         and     $ap,$tp,$ovf
 631         andc    $np,$rp,$ovf
 632         or      $ap,$ap,$np     ; ap=borrow?tp:rp
 633         addi    $t7,$ap,8
 634         mtctr   $j
 635
 636 .align  4
 637 Lcopy:                          ; copy or in-place refresh
 638         ldx     $t0,$ap,$i
 639         ldx     $t1,$t7,$i
 640         stdu    $i,8($ap_l)     ; zap {an}p_{lh}
 641         stdu    $i,8($ap_h)
 642         stdu    $i,8($np_l)
 643         stdu    $i,8($np_h)
 644         stdu    $i,8($ap_l)
 645         stdu    $i,8($ap_h)
 646         stdu    $i,8($np_l)
 647         stdu    $i,8($np_h)
 648         stdx    $t0,$rp,$i
 649         stdx    $t1,$t6,$i
 650         stdx    $i,$tp,$i       ; zap tp at once
 651         stdx    $i,$t4,$i
 652         addi    $i,$i,16
 653         bdnz-   Lcopy
 654
 655         $POP    r14,`2*$SIZE_T`($sp)
 656         $POP    r15,`3*$SIZE_T`($sp)
 657         $POP    r16,`4*$SIZE_T`($sp)
 658         $POP    r17,`5*$SIZE_T`($sp)
 659         $POP    r18,`6*$SIZE_T`($sp)
 660         $POP    r19,`7*$SIZE_T`($sp)
 661         $POP    r20,`8*$SIZE_T`($sp)
 662         $POP    r21,`9*$SIZE_T`($sp)
 663         $POP    r22,`10*$SIZE_T`($sp)
 664         $POP    r23,`11*$SIZE_T`($sp)
 665         $POP    r24,`12*$SIZE_T`($sp)
 666         $POP    r25,`13*$SIZE_T`($sp)
 667         $POP    r26,`14*$SIZE_T`($sp)
 668         $POP    r27,`15*$SIZE_T`($sp)
 669         lfd     f14,`16*$SIZE_T+0`($sp)
 670         lfd     f15,`16*$SIZE_T+8`($sp)
 671         lfd     f16,`16*$SIZE_T+16`($sp)
 672         lfd     f17,`16*$SIZE_T+24`($sp)
 673         lfd     f18,`16*$SIZE_T+32`($sp)
 674         lfd     f19,`16*$SIZE_T+40`($sp)
 675         lfd     f20,`16*$SIZE_T+48`($sp)
 676         lfd     f21,`16*$SIZE_T+56`($sp)
 677         lfd     f22,`16*$SIZE_T+64`($sp)
 678         lfd     f23,`16*$SIZE_T+72`($sp)
 679         lfd     f24,`16*$SIZE_T+80`($sp)
 680         lfd     f25,`16*$SIZE_T+88`($sp)
 681         $POP    $sp,0($sp)
 682         li      r3,1    ; signal "handled"
 683         blr
 684         .long   0
 685 .asciz  "Montgomery Multiplication for PPC64, CRYPTOGAMS by <appro\@fy.chalmers.se>"
 686 ___
 687
 688 $code =~ s/\`([^\`]*)\`/eval $1/gem;
 689 print $code;
 690 close STDOUT;