crypto/bn/asm/sparcv9-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 2005
  11 #
  12 # Pure SPARCv9/8+ and IALU-only bn_mul_mont implementation. The reasons
  13 # for undertaken effort are multiple. First of all, UltraSPARC is not
  14 # the whole SPARCv9 universe and other VIS-free implementations deserve
  15 # optimized code as much. Secondly, newly introduced UltraSPARC T1,
  16 # a.k.a. Niagara, has shared FPU and concurrent FPU-intensive pathes,
  17 # such as sparcv9a-mont, will simply sink it. Yes, T1 is equipped with
  18 # several integrated RSA/DSA accelerator circuits accessible through
  19 # kernel driver [only(*)], but having decent user-land software
  20 # implementation is important too. Finally, reasons like desire to
  21 # experiment with dedicated squaring procedure. Yes, this module
  22 # implements one, because it was easiest to draft it in SPARCv9
  23 # instructions...
  24
  25 # (*)   Engine accessing the driver in question is on my TODO list.
  26 #       For reference, acceleator is estimated to give 6 to 10 times
  27 #       improvement on single-threaded RSA sign. It should be noted
  28 #       that 6-10x improvement coefficient does not actually mean
  29 #       something extraordinary in terms of absolute [single-threaded]
  30 #       performance, as SPARCv9 instruction set is by all means least
  31 #       suitable for high performance crypto among other 64 bit
  32 #       platforms. 6-10x factor simply places T1 in same performance
  33 #       domain as say AMD64 and IA-64. Improvement of RSA verify don't
  34 #       appear impressive at all, but it's the sign operation which is
  35 #       far more critical/interesting.
  36
  37 # You might notice that inner loops are modulo-scheduled:-) This has
  38 # essentially negligible impact on UltraSPARC performance, it's
  39 # Fujitsu SPARC64 V users who should notice and hopefully appreciate
  40 # the advantage... Currently this module surpasses sparcv9a-mont.pl
  41 # by ~20% on UltraSPARC-III and later cores, but recall that sparcv9a
  42 # module still have hidden potential [see TODO list there], which is
  43 # estimated to be larger than 20%...
  44
  45 # int bn_mul_mont(
  46 $rp="%i0";      # BN_ULONG *rp,
  47 $ap="%i1";      # const BN_ULONG *ap,
  48 $bp="%i2";      # const BN_ULONG *bp,
  49 $np="%i3";      # const BN_ULONG *np,
  50 $n0="%i4";      # const BN_ULONG *n0,
  51 $num="%i5";     # int num);
  52
  53 $bits=32;
  54 for (@ARGV)     { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
  55 if ($bits==64)  { $bias=2047; $frame=192; }
  56 else            { $bias=0;    $frame=128; }
  57
  58 $car0="%o0";
  59 $car1="%o1";
  60 $car2="%o2";    # 1 bit
  61 $acc0="%o3";
  62 $acc1="%o4";
  63 $mask="%g1";    # 32 bits, what a waste...
  64 $tmp0="%g4";
  65 $tmp1="%g5";
  66
  67 $i="%l0";
  68 $j="%l1";
  69 $mul0="%l2";
  70 $mul1="%l3";
  71 $tp="%l4";
  72 $apj="%l5";
  73 $npj="%l6";
  74 $tpj="%l7";
  75
  76 $fname="bn_mul_mont_int";
  77
  78 $code=<<___;
  79 .section        ".text",#alloc,#execinstr
  80
  81 .global $fname
  82 .align  32
  83 $fname:
  84         cmp     %o5,4                   ! 128 bits minimum
  85         bge,pt  %icc,.Lenter
  86         sethi   %hi(0xffffffff),$mask
  87         retl
  88         clr     %o0
  89 .align  32
  90 .Lenter:
  91         save    %sp,-$frame,%sp
  92         sll     $num,2,$num             ! num*=4
  93         or      $mask,%lo(0xffffffff),$mask
  94         ld      [$n0],$n0
  95         cmp     $ap,$bp
  96         and     $num,$mask,$num
  97         ld      [$bp],$mul0             ! bp[0]
  98         nop
  99
 100         add     %sp,$bias,%o7           ! real top of stack
 101         ld      [$ap],$car0             ! ap[0] ! redundant in squaring context
 102         sub     %o7,$num,%o7
 103         ld      [$ap+4],$apj            ! ap[1]
 104         and     %o7,-1024,%o7
 105         ld      [$np],$car1             ! np[0]
 106         sub     %o7,$bias,%sp           ! alloca
 107         ld      [$np+4],$npj            ! np[1]
 108         be,pt   `$bits==32?"%icc":"%xcc"`,.Lbn_sqr_mont
 109         mov     12,$j
 110
 111         mulx    $car0,$mul0,$car0       ! ap[0]*bp[0]
 112         mulx    $apj,$mul0,$tmp0        !prologue! ap[1]*bp[0]
 113         and     $car0,$mask,$acc0
 114         add     %sp,$bias+$frame,$tp
 115         ld      [$ap+8],$apj            !prologue!
 116
 117         mulx    $n0,$acc0,$mul1         ! "t[0]"*n0
 118         and     $mul1,$mask,$mul1
 119
 120         mulx    $car1,$mul1,$car1       ! np[0]*"t[0]"*n0
 121         mulx    $npj,$mul1,$acc1        !prologue! np[1]*"t[0]"*n0
 122         srlx    $car0,32,$car0
 123         add     $acc0,$car1,$car1
 124         ld      [$np+8],$npj            !prologue!
 125         srlx    $car1,32,$car1
 126         mov     $tmp0,$acc0             !prologue!
 127
 128 .L1st:
 129         mulx    $apj,$mul0,$tmp0
 130         mulx    $npj,$mul1,$tmp1
 131         add     $acc0,$car0,$car0
 132         ld      [$ap+$j],$apj           ! ap[j]
 133         and     $car0,$mask,$acc0
 134         add     $acc1,$car1,$car1
 135         ld      [$np+$j],$npj           ! np[j]
 136         srlx    $car0,32,$car0
 137         add     $acc0,$car1,$car1
 138         add     $j,4,$j                 ! j++
 139         mov     $tmp0,$acc0
 140         st      $car1,[$tp]
 141         cmp     $j,$num
 142         mov     $tmp1,$acc1
 143         srlx    $car1,32,$car1
 144         bl      %icc,.L1st
 145         add     $tp,4,$tp               ! tp++
 146 !.L1st
 147
 148         mulx    $apj,$mul0,$tmp0        !epilogue!
 149         mulx    $npj,$mul1,$tmp1
 150         add     $acc0,$car0,$car0
 151         and     $car0,$mask,$acc0
 152         add     $acc1,$car1,$car1
 153         srlx    $car0,32,$car0
 154         add     $acc0,$car1,$car1
 155         st      $car1,[$tp]
 156         srlx    $car1,32,$car1
 157
 158         add     $tmp0,$car0,$car0
 159         and     $car0,$mask,$acc0
 160         add     $tmp1,$car1,$car1
 161         srlx    $car0,32,$car0
 162         add     $acc0,$car1,$car1
 163         st      $car1,[$tp+4]
 164         srlx    $car1,32,$car1
 165
 166         add     $car0,$car1,$car1
 167         st      $car1,[$tp+8]
 168         srlx    $car1,32,$car2
 169 \f
 170         mov     4,$i                    ! i++
 171         ld      [$bp+4],$mul0           ! bp[1]
 172 .Louter:
 173         add     %sp,$bias+$frame,$tp
 174         ld      [$ap],$car0             ! ap[0]
 175         ld      [$ap+4],$apj            ! ap[1]
 176         ld      [$np],$car1             ! np[0]
 177         ld      [$np+4],$npj            ! np[1]
 178         ld      [$tp],$tmp1             ! tp[0]
 179         ld      [$tp+4],$tpj            ! tp[1]
 180         mov     12,$j
 181
 182         mulx    $car0,$mul0,$car0
 183         mulx    $apj,$mul0,$tmp0        !prologue!
 184         add     $tmp1,$car0,$car0
 185         ld      [$ap+8],$apj            !prologue!
 186         and     $car0,$mask,$acc0
 187
 188         mulx    $n0,$acc0,$mul1
 189         and     $mul1,$mask,$mul1
 190
 191         mulx    $car1,$mul1,$car1
 192         mulx    $npj,$mul1,$acc1        !prologue!
 193         srlx    $car0,32,$car0
 194         add     $acc0,$car1,$car1
 195         ld      [$np+8],$npj            !prologue!
 196         srlx    $car1,32,$car1
 197         mov     $tmp0,$acc0             !prologue!
 198
 199 .Linner:
 200         mulx    $apj,$mul0,$tmp0
 201         mulx    $npj,$mul1,$tmp1
 202         add     $tpj,$car0,$car0
 203         ld      [$ap+$j],$apj           ! ap[j]
 204         add     $acc0,$car0,$car0
 205         add     $acc1,$car1,$car1
 206         ld      [$np+$j],$npj           ! np[j]
 207         and     $car0,$mask,$acc0
 208         ld      [$tp+8],$tpj            ! tp[j]
 209         srlx    $car0,32,$car0
 210         add     $acc0,$car1,$car1
 211         add     $j,4,$j                 ! j++
 212         mov     $tmp0,$acc0
 213         st      $car1,[$tp]             ! tp[j-1]
 214         srlx    $car1,32,$car1
 215         mov     $tmp1,$acc1
 216         cmp     $j,$num
 217         bl      %icc,.Linner
 218         add     $tp,4,$tp               ! tp++
 219 !.Linner
 220
 221         mulx    $apj,$mul0,$tmp0        !epilogue!
 222         mulx    $npj,$mul1,$tmp1
 223         add     $tpj,$car0,$car0
 224         add     $acc0,$car0,$car0
 225         ld      [$tp+8],$tpj            ! tp[j]
 226         and     $car0,$mask,$acc0
 227         add     $acc1,$car1,$car1
 228         srlx    $car0,32,$car0
 229         add     $acc0,$car1,$car1
 230         st      $car1,[$tp]             ! tp[j-1]
 231         srlx    $car1,32,$car1
 232
 233         add     $tpj,$car0,$car0
 234         add     $tmp0,$car0,$car0
 235         and     $car0,$mask,$acc0
 236         add     $tmp1,$car1,$car1
 237         add     $acc0,$car1,$car1
 238         st      $car1,[$tp+4]           ! tp[j-1]
 239         srlx    $car0,32,$car0
 240         add     $i,4,$i                 ! i++
 241         srlx    $car1,32,$car1
 242
 243         add     $car0,$car1,$car1
 244         cmp     $i,$num
 245         add     $car2,$car1,$car1
 246         st      $car1,[$tp+8]
 247
 248         srlx    $car1,32,$car2
 249         bl,a    %icc,.Louter
 250         ld      [$bp+$i],$mul0          ! bp[i]
 251 !.Louter
 252
 253         add     $tp,12,$tp
 254 \f
 255 .Ltail:
 256         add     $np,$num,$np
 257         add     $rp,$num,$rp
 258         mov     $tp,$ap
 259         sub     %g0,$num,%o7            ! k=-num
 260
 261         srl     $npj,30,%o0             ! boundary condition...
 262         brz,pn  %o0,.Lcopy              ! ... is met
 263         nop
 264
 265         ba      .Lsub
 266         subcc   %g0,%g0,%g0             ! clear %icc.c
 267 .align  16
 268 .Lsub:
 269         ld      [$tp+%o7],%o0
 270         ld      [$np+%o7],%o1
 271         subccc  %o0,%o1,%o1             ! tp[j]-np[j]
 272         add     $rp,%o7,$i
 273         add     %o7,4,%o7
 274         brnz    %o7,.Lsub
 275         st      %o1,[$i]
 276         subc    $car2,0,$car2           ! handle upmost overflow bit
 277         and     $tp,$car2,$ap
 278         andn    $rp,$car2,$np
 279         or      $ap,$np,$ap
 280         sub     %g0,$num,%o7
 281
 282 .Lcopy:
 283         ld      [$ap+%o7],%o0           ! copy or in-place refresh
 284         st      %g0,[$tp+%o7]           ! zap tp
 285         st      %o0,[$rp+%o7]
 286         add     %o7,4,%o7
 287         brnz    %o7,.Lcopy
 288         nop
 289         mov     1,%i0
 290         ret
 291         restore
 292 ___
 293 \f
 294 ########
 295 ######## .Lbn_sqr_mont gives up to 20% *overall* improvement over
 296 ######## code without following dedicated squaring procedure.
 297 ########
 298 $sbit="%i2";            # re-use $bp!
 299
 300 $code.=<<___;
 301 .align  32
 302 .Lbn_sqr_mont:
 303         mulx    $mul0,$mul0,$car0               ! ap[0]*ap[0]
 304         mulx    $apj,$mul0,$tmp0                !prologue!
 305         and     $car0,$mask,$acc0
 306         add     %sp,$bias+$frame,$tp
 307         ld      [$ap+8],$apj                    !prologue!
 308
 309         mulx    $n0,$acc0,$mul1                 ! "t[0]"*n0
 310         srlx    $car0,32,$car0
 311         and     $mul1,$mask,$mul1
 312
 313         mulx    $car1,$mul1,$car1               ! np[0]*"t[0]"*n0
 314         mulx    $npj,$mul1,$acc1                !prologue!
 315         and     $car0,1,$sbit
 316         ld      [$np+8],$npj                    !prologue!
 317         srlx    $car0,1,$car0
 318         add     $acc0,$car1,$car1
 319         srlx    $car1,32,$car1
 320         mov     $tmp0,$acc0                     !prologue!
 321
 322 .Lsqr_1st:
 323         mulx    $apj,$mul0,$tmp0
 324         mulx    $npj,$mul1,$tmp1
 325         add     $acc0,$car0,$car0               ! ap[j]*a0+c0
 326         add     $acc1,$car1,$car1
 327         ld      [$ap+$j],$apj                   ! ap[j]
 328         and     $car0,$mask,$acc0
 329         ld      [$np+$j],$npj                   ! np[j]
 330         srlx    $car0,32,$car0
 331         add     $acc0,$acc0,$acc0
 332         or      $sbit,$acc0,$acc0
 333         mov     $tmp1,$acc1
 334         srlx    $acc0,32,$sbit
 335         add     $j,4,$j                         ! j++
 336         and     $acc0,$mask,$acc0
 337         cmp     $j,$num
 338         add     $acc0,$car1,$car1
 339         st      $car1,[$tp]
 340         mov     $tmp0,$acc0
 341         srlx    $car1,32,$car1
 342         bl      %icc,.Lsqr_1st
 343         add     $tp,4,$tp                       ! tp++
 344 !.Lsqr_1st
 345
 346         mulx    $apj,$mul0,$tmp0                ! epilogue
 347         mulx    $npj,$mul1,$tmp1
 348         add     $acc0,$car0,$car0               ! ap[j]*a0+c0
 349         add     $acc1,$car1,$car1
 350         and     $car0,$mask,$acc0
 351         srlx    $car0,32,$car0
 352         add     $acc0,$acc0,$acc0
 353         or      $sbit,$acc0,$acc0
 354         srlx    $acc0,32,$sbit
 355         and     $acc0,$mask,$acc0
 356         add     $acc0,$car1,$car1
 357         st      $car1,[$tp]
 358         srlx    $car1,32,$car1
 359
 360         add     $tmp0,$car0,$car0               ! ap[j]*a0+c0
 361         add     $tmp1,$car1,$car1
 362         and     $car0,$mask,$acc0
 363         srlx    $car0,32,$car0
 364         add     $acc0,$acc0,$acc0
 365         or      $sbit,$acc0,$acc0
 366         srlx    $acc0,32,$sbit
 367         and     $acc0,$mask,$acc0
 368         add     $acc0,$car1,$car1
 369         st      $car1,[$tp+4]
 370         srlx    $car1,32,$car1
 371
 372         add     $car0,$car0,$car0
 373         or      $sbit,$car0,$car0
 374         add     $car0,$car1,$car1
 375         st      $car1,[$tp+8]
 376         srlx    $car1,32,$car2
 377 \f
 378         ld      [%sp+$bias+$frame],$tmp0        ! tp[0]
 379         ld      [%sp+$bias+$frame+4],$tmp1      ! tp[1]
 380         ld      [%sp+$bias+$frame+8],$tpj       ! tp[2]
 381         ld      [$ap+4],$mul0                   ! ap[1]
 382         ld      [$ap+8],$apj                    ! ap[2]
 383         ld      [$np],$car1                     ! np[0]
 384         ld      [$np+4],$npj                    ! np[1]
 385         mulx    $n0,$tmp0,$mul1
 386
 387         mulx    $mul0,$mul0,$car0
 388         and     $mul1,$mask,$mul1
 389
 390         mulx    $car1,$mul1,$car1
 391         mulx    $npj,$mul1,$acc1
 392         add     $tmp0,$car1,$car1
 393         and     $car0,$mask,$acc0
 394         ld      [$np+8],$npj                    ! np[2]
 395         srlx    $car1,32,$car1
 396         add     $tmp1,$car1,$car1
 397         srlx    $car0,32,$car0
 398         add     $acc0,$car1,$car1
 399         and     $car0,1,$sbit
 400         add     $acc1,$car1,$car1
 401         srlx    $car0,1,$car0
 402         mov     12,$j
 403         st      $car1,[%sp+$bias+$frame]        ! tp[0]=
 404         srlx    $car1,32,$car1
 405         add     %sp,$bias+$frame+4,$tp
 406
 407 .Lsqr_2nd:
 408         mulx    $apj,$mul0,$acc0
 409         mulx    $npj,$mul1,$acc1
 410         add     $acc0,$car0,$car0
 411         add     $tpj,$car1,$car1
 412         ld      [$ap+$j],$apj                   ! ap[j]
 413         and     $car0,$mask,$acc0
 414         ld      [$np+$j],$npj                   ! np[j]
 415         srlx    $car0,32,$car0
 416         add     $acc1,$car1,$car1
 417         ld      [$tp+8],$tpj                    ! tp[j]
 418         add     $acc0,$acc0,$acc0
 419         add     $j,4,$j                         ! j++
 420         or      $sbit,$acc0,$acc0
 421         srlx    $acc0,32,$sbit
 422         and     $acc0,$mask,$acc0
 423         cmp     $j,$num
 424         add     $acc0,$car1,$car1
 425         st      $car1,[$tp]                     ! tp[j-1]
 426         srlx    $car1,32,$car1
 427         bl      %icc,.Lsqr_2nd
 428         add     $tp,4,$tp                       ! tp++
 429 !.Lsqr_2nd
 430
 431         mulx    $apj,$mul0,$acc0
 432         mulx    $npj,$mul1,$acc1
 433         add     $acc0,$car0,$car0
 434         add     $tpj,$car1,$car1
 435         and     $car0,$mask,$acc0
 436         srlx    $car0,32,$car0
 437         add     $acc1,$car1,$car1
 438         add     $acc0,$acc0,$acc0
 439         or      $sbit,$acc0,$acc0
 440         srlx    $acc0,32,$sbit
 441         and     $acc0,$mask,$acc0
 442         add     $acc0,$car1,$car1
 443         st      $car1,[$tp]                     ! tp[j-1]
 444         srlx    $car1,32,$car1
 445
 446         add     $car0,$car0,$car0
 447         or      $sbit,$car0,$car0
 448         add     $car0,$car1,$car1
 449         add     $car2,$car1,$car1
 450         st      $car1,[$tp+4]
 451         srlx    $car1,32,$car2
 452 \f
 453         ld      [%sp+$bias+$frame],$tmp1        ! tp[0]
 454         ld      [%sp+$bias+$frame+4],$tpj       ! tp[1]
 455         ld      [$ap+8],$mul0                   ! ap[2]
 456         ld      [$np],$car1                     ! np[0]
 457         ld      [$np+4],$npj                    ! np[1]
 458         mulx    $n0,$tmp1,$mul1
 459         and     $mul1,$mask,$mul1
 460         mov     8,$i
 461
 462         mulx    $mul0,$mul0,$car0
 463         mulx    $car1,$mul1,$car1
 464         and     $car0,$mask,$acc0
 465         add     $tmp1,$car1,$car1
 466         srlx    $car0,32,$car0
 467         add     %sp,$bias+$frame,$tp
 468         srlx    $car1,32,$car1
 469         and     $car0,1,$sbit
 470         srlx    $car0,1,$car0
 471         mov     4,$j
 472
 473 .Lsqr_outer:
 474 .Lsqr_inner1:
 475         mulx    $npj,$mul1,$acc1
 476         add     $tpj,$car1,$car1
 477         add     $j,4,$j
 478         ld      [$tp+8],$tpj
 479         cmp     $j,$i
 480         add     $acc1,$car1,$car1
 481         ld      [$np+$j],$npj
 482         st      $car1,[$tp]
 483         srlx    $car1,32,$car1
 484         bl      %icc,.Lsqr_inner1
 485         add     $tp,4,$tp
 486 !.Lsqr_inner1
 487
 488         add     $j,4,$j
 489         ld      [$ap+$j],$apj                   ! ap[j]
 490         mulx    $npj,$mul1,$acc1
 491         add     $tpj,$car1,$car1
 492         ld      [$np+$j],$npj                   ! np[j]
 493         add     $acc0,$car1,$car1
 494         ld      [$tp+8],$tpj                    ! tp[j]
 495         add     $acc1,$car1,$car1
 496         st      $car1,[$tp]
 497         srlx    $car1,32,$car1
 498
 499         add     $j,4,$j
 500         cmp     $j,$num
 501         be,pn   %icc,.Lsqr_no_inner2
 502         add     $tp,4,$tp
 503
 504 .Lsqr_inner2:
 505         mulx    $apj,$mul0,$acc0
 506         mulx    $npj,$mul1,$acc1
 507         add     $tpj,$car1,$car1
 508         add     $acc0,$car0,$car0
 509         ld      [$ap+$j],$apj                   ! ap[j]
 510         and     $car0,$mask,$acc0
 511         ld      [$np+$j],$npj                   ! np[j]
 512         srlx    $car0,32,$car0
 513         add     $acc0,$acc0,$acc0
 514         ld      [$tp+8],$tpj                    ! tp[j]
 515         or      $sbit,$acc0,$acc0
 516         add     $j,4,$j                         ! j++
 517         srlx    $acc0,32,$sbit
 518         and     $acc0,$mask,$acc0
 519         cmp     $j,$num
 520         add     $acc0,$car1,$car1
 521         add     $acc1,$car1,$car1
 522         st      $car1,[$tp]                     ! tp[j-1]
 523         srlx    $car1,32,$car1
 524         bl      %icc,.Lsqr_inner2
 525         add     $tp,4,$tp                       ! tp++
 526
 527 .Lsqr_no_inner2:
 528         mulx    $apj,$mul0,$acc0
 529         mulx    $npj,$mul1,$acc1
 530         add     $tpj,$car1,$car1
 531         add     $acc0,$car0,$car0
 532         and     $car0,$mask,$acc0
 533         srlx    $car0,32,$car0
 534         add     $acc0,$acc0,$acc0
 535         or      $sbit,$acc0,$acc0
 536         srlx    $acc0,32,$sbit
 537         and     $acc0,$mask,$acc0
 538         add     $acc0,$car1,$car1
 539         add     $acc1,$car1,$car1
 540         st      $car1,[$tp]                     ! tp[j-1]
 541         srlx    $car1,32,$car1
 542
 543         add     $car0,$car0,$car0
 544         or      $sbit,$car0,$car0
 545         add     $car0,$car1,$car1
 546         add     $car2,$car1,$car1
 547         st      $car1,[$tp+4]
 548         srlx    $car1,32,$car2
 549 \f
 550         add     $i,4,$i                         ! i++
 551         ld      [%sp+$bias+$frame],$tmp1        ! tp[0]
 552         ld      [%sp+$bias+$frame+4],$tpj       ! tp[1]
 553         ld      [$ap+$i],$mul0                  ! ap[j]
 554         ld      [$np],$car1                     ! np[0]
 555         ld      [$np+4],$npj                    ! np[1]
 556         mulx    $n0,$tmp1,$mul1
 557         and     $mul1,$mask,$mul1
 558         add     $i,4,$tmp0
 559
 560         mulx    $mul0,$mul0,$car0
 561         mulx    $car1,$mul1,$car1
 562         and     $car0,$mask,$acc0
 563         add     $tmp1,$car1,$car1
 564         srlx    $car0,32,$car0
 565         add     %sp,$bias+$frame,$tp
 566         srlx    $car1,32,$car1
 567         and     $car0,1,$sbit
 568         srlx    $car0,1,$car0
 569
 570         cmp     $tmp0,$num                      ! i<num-1
 571         bl      %icc,.Lsqr_outer
 572         mov     4,$j
 573 \f
 574 .Lsqr_last:
 575         mulx    $npj,$mul1,$acc1
 576         add     $tpj,$car1,$car1
 577         add     $j,4,$j
 578         ld      [$tp+8],$tpj
 579         cmp     $j,$i
 580         add     $acc1,$car1,$car1
 581         ld      [$np+$j],$npj
 582         st      $car1,[$tp]
 583         srlx    $car1,32,$car1
 584         bl      %icc,.Lsqr_last
 585         add     $tp,4,$tp
 586 !.Lsqr_last
 587
 588         mulx    $npj,$mul1,$acc1
 589         add     $tpj,$car1,$car1
 590         add     $acc0,$car1,$car1
 591         add     $acc1,$car1,$car1
 592         st      $car1,[$tp]
 593         srlx    $car1,32,$car1
 594
 595         add     $car0,$car0,$car0               ! recover $car0
 596         or      $sbit,$car0,$car0
 597         add     $car0,$car1,$car1
 598         add     $car2,$car1,$car1
 599         st      $car1,[$tp+4]
 600         srlx    $car1,32,$car2
 601
 602         ba      .Ltail
 603         add     $tp,8,$tp
 604 .type   $fname,#function
 605 .size   $fname,(.-$fname)
 606 .asciz  "Montgomery Multipltication for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
 607 .align  32
 608 ___
 609 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 610 print $code;
 611 close STDOUT;