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. Rights for redistribution and usage in source and binary
   6 # forms are granted according to the OpenSSL license.
   7 # ====================================================================
   8
   9 # December 2005
  10 #
  11 # Pure SPARCv9/8+ and IALU-only bn_mul_mont implementation. The reasons
  12 # for undertaken effort are multiple. First of all, UltraSPARC is not
  13 # the whole SPARCv9 universe and other VIS-free implementations deserve
  14 # optimized code as much. Secondly, newly introduced UltraSPARC T1,
  15 # a.k.a. Niagara, has shared FPU and concurrent FPU-intensive pathes,
  16 # such as sparcv9a-mont, will simply sink it. Yes, T1 is equipped with
  17 # several integrated RSA/DSA accelerator circuits accessible through
  18 # kernel driver [only(*)], but having decent user-land software
  19 # implementation is important too. Finally, reasons like desire to
  20 # experiment with dedicated squaring procedure. Yes, this module
  21 # implements one, because it was easiest to draft it in SPARCv9
  22 # instructions...
  23
  24 # (*)   Engine accessing the driver in question is on my TODO list.
  25 #       For reference, acceleator is estimated to give 6 to 10 times
  26 #       improvement on single-threaded RSA sign. It should be noted
  27 #       that 6-10x improvement coefficient does not actually mean
  28 #       something extraordinary in terms of absolute [single-threaded]
  29 #       performance, as SPARCv9 instruction set is by all means least
  30 #       suitable for high performance crypto among other 64 bit
  31 #       platforms. 6-10x factor simply places T1 in same performance
  32 #       domain as say AMD64 and IA-64. Improvement of RSA verify don't
  33 #       appear impressive at all, but it's the sign operation which is
  34 #       far more critical/interesting.
  35
  36 # You might notice that inner loops are modulo-scheduled:-) This has
  37 # essentially negligible impact on UltraSPARC performance, it's
  38 # Fujitsu SPARC64 V users who should notice and hopefully appreciate
  39 # the advantage... Currently this module surpasses sparcv9a-mont.pl
  40 # by ~20% on UltraSPARC-III and later cores, but recall that sparcv9a
  41 # module still have hidden potential [see TODO list there], which is
  42 # estimated to be larger than 20%...
  43
  44 # int bn_mul_mont(
  45 $rp="%i0";      # BN_ULONG *rp,
  46 $ap="%i1";      # const BN_ULONG *ap,
  47 $bp="%i2";      # const BN_ULONG *bp,
  48 $np="%i3";      # const BN_ULONG *np,
  49 $n0="%i4";      # const BN_ULONG *n0,
  50 $num="%i5";     # int num);
  51
  52 $bits=32;
  53 for (@ARGV)     { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
  54 if ($bits==64)  { $bias=2047; $frame=192; }
  55 else            { $bias=0;    $frame=128; }
  56
  57 $car0="%o0";
  58 $car1="%o1";
  59 $car2="%o2";    # 1 bit
  60 $acc0="%o3";
  61 $acc1="%o4";
  62 $mask="%g1";    # 32 bits, what a waste...
  63 $tmp0="%g4";
  64 $tmp1="%g5";
  65
  66 $i="%l0";
  67 $j="%l1";
  68 $mul0="%l2";
  69 $mul1="%l3";
  70 $tp="%l4";
  71 $apj="%l5";
  72 $npj="%l6";
  73 $tpj="%l7";
  74
  75 $fname="bn_mul_mont_int";
  76
  77 $code=<<___;
  78 .section        ".text",#alloc,#execinstr
  79
  80 .global $fname
  81 .align  32
  82 $fname:
  83         cmp     %o5,4                   ! 128 bits minimum
  84         bge,pt  %icc,.Lenter
  85         sethi   %hi(0xffffffff),$mask
  86         retl
  87         clr     %o0
  88 .align  32
  89 .Lenter:
  90         save    %sp,-$frame,%sp
  91         sll     $num,2,$num             ! num*=4
  92         or      $mask,%lo(0xffffffff),$mask
  93         ld      [$n0],$n0
  94         cmp     $ap,$bp
  95         and     $num,$mask,$num
  96         ld      [$bp],$mul0             ! bp[0]
  97         nop
  98
  99         add     %sp,$bias,%o7           ! real top of stack
 100         ld      [$ap],$car0             ! ap[0] ! redundant in squaring context
 101         sub     %o7,$num,%o7
 102         ld      [$ap+4],$apj            ! ap[1]
 103         and     %o7,-1024,%o7
 104         ld      [$np],$car1             ! np[0]
 105         sub     %o7,$bias,%sp           ! alloca
 106         ld      [$np+4],$npj            ! np[1]
 107         be,pt   `$bits==32?"%icc":"%xcc"`,.Lbn_sqr_mont
 108         mov     12,$j
 109
 110         mulx    $car0,$mul0,$car0       ! ap[0]*bp[0]
 111         mulx    $apj,$mul0,$tmp0        !prologue! ap[1]*bp[0]
 112         and     $car0,$mask,$acc0
 113         add     %sp,$bias+$frame,$tp
 114         ld      [$ap+8],$apj            !prologue!
 115
 116         mulx    $n0,$acc0,$mul1         ! "t[0]"*n0
 117         and     $mul1,$mask,$mul1
 118
 119         mulx    $car1,$mul1,$car1       ! np[0]*"t[0]"*n0
 120         mulx    $npj,$mul1,$acc1        !prologue! np[1]*"t[0]"*n0
 121         srlx    $car0,32,$car0
 122         add     $acc0,$car1,$car1
 123         ld      [$np+8],$npj            !prologue!
 124         srlx    $car1,32,$car1
 125         mov     $tmp0,$acc0             !prologue!
 126
 127 .L1st:
 128         mulx    $apj,$mul0,$tmp0
 129         mulx    $npj,$mul1,$tmp1
 130         add     $acc0,$car0,$car0
 131         ld      [$ap+$j],$apj           ! ap[j]
 132         and     $car0,$mask,$acc0
 133         add     $acc1,$car1,$car1
 134         ld      [$np+$j],$npj           ! np[j]
 135         srlx    $car0,32,$car0
 136         add     $acc0,$car1,$car1
 137         add     $j,4,$j                 ! j++
 138         mov     $tmp0,$acc0
 139         st      $car1,[$tp]
 140         cmp     $j,$num
 141         mov     $tmp1,$acc1
 142         srlx    $car1,32,$car1
 143         bl      %icc,.L1st
 144         add     $tp,4,$tp               ! tp++
 145 !.L1st
 146
 147         mulx    $apj,$mul0,$tmp0        !epilogue!
 148         mulx    $npj,$mul1,$tmp1
 149         add     $acc0,$car0,$car0
 150         and     $car0,$mask,$acc0
 151         add     $acc1,$car1,$car1
 152         srlx    $car0,32,$car0
 153         add     $acc0,$car1,$car1
 154         st      $car1,[$tp]
 155         srlx    $car1,32,$car1
 156
 157         add     $tmp0,$car0,$car0
 158         and     $car0,$mask,$acc0
 159         add     $tmp1,$car1,$car1
 160         srlx    $car0,32,$car0
 161         add     $acc0,$car1,$car1
 162         st      $car1,[$tp+4]
 163         srlx    $car1,32,$car1
 164
 165         add     $car0,$car1,$car1
 166         st      $car1,[$tp+8]
 167         srlx    $car1,32,$car2
 168 \f
 169         mov     4,$i                    ! i++
 170         ld      [$bp+4],$mul0           ! bp[1]
 171 .Louter:
 172         add     %sp,$bias+$frame,$tp
 173         ld      [$ap],$car0             ! ap[0]
 174         ld      [$ap+4],$apj            ! ap[1]
 175         ld      [$np],$car1             ! np[0]
 176         ld      [$np+4],$npj            ! np[1]
 177         ld      [$tp],$tmp1             ! tp[0]
 178         ld      [$tp+4],$tpj            ! tp[1]
 179         mov     12,$j
 180
 181         mulx    $car0,$mul0,$car0
 182         mulx    $apj,$mul0,$tmp0        !prologue!
 183         add     $tmp1,$car0,$car0
 184         ld      [$ap+8],$apj            !prologue!
 185         and     $car0,$mask,$acc0
 186
 187         mulx    $n0,$acc0,$mul1
 188         and     $mul1,$mask,$mul1
 189
 190         mulx    $car1,$mul1,$car1
 191         mulx    $npj,$mul1,$acc1        !prologue!
 192         srlx    $car0,32,$car0
 193         add     $acc0,$car1,$car1
 194         ld      [$np+8],$npj            !prologue!
 195         srlx    $car1,32,$car1
 196         mov     $tmp0,$acc0             !prologue!
 197
 198 .Linner:
 199         mulx    $apj,$mul0,$tmp0
 200         mulx    $npj,$mul1,$tmp1
 201         add     $tpj,$car0,$car0
 202         ld      [$ap+$j],$apj           ! ap[j]
 203         add     $acc0,$car0,$car0
 204         add     $acc1,$car1,$car1
 205         ld      [$np+$j],$npj           ! np[j]
 206         and     $car0,$mask,$acc0
 207         ld      [$tp+8],$tpj            ! tp[j]
 208         srlx    $car0,32,$car0
 209         add     $acc0,$car1,$car1
 210         add     $j,4,$j                 ! j++
 211         mov     $tmp0,$acc0
 212         st      $car1,[$tp]             ! tp[j-1]
 213         srlx    $car1,32,$car1
 214         mov     $tmp1,$acc1
 215         cmp     $j,$num
 216         bl      %icc,.Linner
 217         add     $tp,4,$tp               ! tp++
 218 !.Linner
 219
 220         mulx    $apj,$mul0,$tmp0        !epilogue!
 221         mulx    $npj,$mul1,$tmp1
 222         add     $tpj,$car0,$car0
 223         add     $acc0,$car0,$car0
 224         ld      [$tp+8],$tpj            ! tp[j]
 225         and     $car0,$mask,$acc0
 226         add     $acc1,$car1,$car1
 227         srlx    $car0,32,$car0
 228         add     $acc0,$car1,$car1
 229         st      $car1,[$tp]             ! tp[j-1]
 230         srlx    $car1,32,$car1
 231
 232         add     $tpj,$car0,$car0
 233         add     $tmp0,$car0,$car0
 234         and     $car0,$mask,$acc0
 235         add     $tmp1,$car1,$car1
 236         add     $acc0,$car1,$car1
 237         st      $car1,[$tp+4]           ! tp[j-1]
 238         srlx    $car0,32,$car0
 239         add     $i,4,$i                 ! i++
 240         srlx    $car1,32,$car1
 241
 242         add     $car0,$car1,$car1
 243         cmp     $i,$num
 244         add     $car2,$car1,$car1
 245         st      $car1,[$tp+8]
 246
 247         srlx    $car1,32,$car2
 248         bl,a    %icc,.Louter
 249         ld      [$bp+$i],$mul0          ! bp[i]
 250 !.Louter
 251
 252         add     $tp,12,$tp
 253 \f
 254 .Ltail:
 255         add     $np,$num,$np
 256         add     $rp,$num,$rp
 257
 258         cmp     $car2,0                 ! clears %icc.c
 259         bne,pn  %icc,.Lsub
 260         sub     %g0,$num,%o7            ! k=-num
 261
 262         cmp     $car1,$npj              ! compare top-most $tp and $np words
 263         bcs,pt  %icc,.Lcopy             ! %icc.c is clean if not taken
 264         nop
 265
 266 .align  16,0x1000000
 267 .Lsub:
 268         ld      [$tp+%o7],%o0
 269         ld      [$np+%o7],%o1
 270         subccc  %o0,%o1,%o1
 271         st      %o1,[$rp+%o7]
 272         add     %o7,4,%o7
 273         brnz    %o7,.Lsub
 274         nop
 275         subccc  $car2,0,$car2
 276         bcc     %icc,.Lzap
 277         sub     %g0,$num,%o7
 278
 279 .align  16,0x1000000
 280 .Lcopy:
 281         ld      [$tp+%o7],%o0
 282         st      %o0,[$rp+%o7]
 283         add     %o7,4,%o7
 284         brnz    %o7,.Lcopy
 285         nop
 286         ba      .Lzap
 287         sub     %g0,$num,%o7
 288
 289 .align  32
 290 .Lzap:
 291         st      %g0,[$tp+%o7]
 292         add     %o7,4,%o7
 293         brnz    %o7,.Lzap
 294         nop
 295         mov     1,%i0
 296         ret
 297         restore
 298 ___
 299 \f
 300 ########
 301 ######## .Lbn_sqr_mont gives up to 20% *overall* improvement over
 302 ######## code without following dedicated squaring procedure.
 303 ########
 304 $sbit="%i2";            # re-use $bp!
 305
 306 $code.=<<___;
 307 .align  32
 308 .Lbn_sqr_mont:
 309         mulx    $mul0,$mul0,$car0               ! ap[0]*ap[0]
 310         mulx    $apj,$mul0,$tmp0                !prologue!
 311         and     $car0,$mask,$acc0
 312         add     %sp,$bias+$frame,$tp
 313         ld      [$ap+8],$apj                    !prologue!
 314
 315         mulx    $n0,$acc0,$mul1                 ! "t[0]"*n0
 316         srlx    $car0,32,$car0
 317         and     $mul1,$mask,$mul1
 318
 319         mulx    $car1,$mul1,$car1               ! np[0]*"t[0]"*n0
 320         mulx    $npj,$mul1,$acc1                !prologue!
 321         and     $car0,1,$sbit
 322         ld      [$np+8],$npj                    !prologue!
 323         srlx    $car0,1,$car0
 324         add     $acc0,$car1,$car1
 325         srlx    $car1,32,$car1
 326         mov     $tmp0,$acc0                     !prologue!
 327
 328 .Lsqr_1st:
 329         mulx    $apj,$mul0,$tmp0
 330         mulx    $npj,$mul1,$tmp1
 331         add     $acc0,$car0,$car0               ! ap[j]*a0+c0
 332         add     $acc1,$car1,$car1
 333         ld      [$ap+$j],$apj                   ! ap[j]
 334         and     $car0,$mask,$acc0
 335         ld      [$np+$j],$npj                   ! np[j]
 336         srlx    $car0,32,$car0
 337         add     $acc0,$acc0,$acc0
 338         or      $sbit,$acc0,$acc0
 339         mov     $tmp1,$acc1
 340         srlx    $acc0,32,$sbit
 341         add     $j,4,$j                         ! j++
 342         and     $acc0,$mask,$acc0
 343         cmp     $j,$num
 344         add     $acc0,$car1,$car1
 345         st      $car1,[$tp]
 346         mov     $tmp0,$acc0
 347         srlx    $car1,32,$car1
 348         bl      %icc,.Lsqr_1st
 349         add     $tp,4,$tp                       ! tp++
 350 !.Lsqr_1st
 351
 352         mulx    $apj,$mul0,$tmp0                ! epilogue
 353         mulx    $npj,$mul1,$tmp1
 354         add     $acc0,$car0,$car0               ! ap[j]*a0+c0
 355         add     $acc1,$car1,$car1
 356         and     $car0,$mask,$acc0
 357         srlx    $car0,32,$car0
 358         add     $acc0,$acc0,$acc0
 359         or      $sbit,$acc0,$acc0
 360         srlx    $acc0,32,$sbit
 361         and     $acc0,$mask,$acc0
 362         add     $acc0,$car1,$car1
 363         st      $car1,[$tp]
 364         srlx    $car1,32,$car1
 365
 366         add     $tmp0,$car0,$car0               ! ap[j]*a0+c0
 367         add     $tmp1,$car1,$car1
 368         and     $car0,$mask,$acc0
 369         srlx    $car0,32,$car0
 370         add     $acc0,$acc0,$acc0
 371         or      $sbit,$acc0,$acc0
 372         srlx    $acc0,32,$sbit
 373         and     $acc0,$mask,$acc0
 374         add     $acc0,$car1,$car1
 375         st      $car1,[$tp+4]
 376         srlx    $car1,32,$car1
 377
 378         add     $car0,$car0,$car0
 379         or      $sbit,$car0,$car0
 380         add     $car0,$car1,$car1
 381         st      $car1,[$tp+8]
 382         srlx    $car1,32,$car2
 383 \f
 384         ld      [%sp+$bias+$frame],$tmp0        ! tp[0]
 385         ld      [%sp+$bias+$frame+4],$tmp1      ! tp[1]
 386         ld      [%sp+$bias+$frame+8],$tpj       ! tp[2]
 387         ld      [$ap+4],$mul0                   ! ap[1]
 388         ld      [$ap+8],$apj                    ! ap[2]
 389         ld      [$np],$car1                     ! np[0]
 390         ld      [$np+4],$npj                    ! np[1]
 391         mulx    $n0,$tmp0,$mul1
 392
 393         mulx    $mul0,$mul0,$car0
 394         and     $mul1,$mask,$mul1
 395
 396         mulx    $car1,$mul1,$car1
 397         mulx    $npj,$mul1,$acc1
 398         add     $tmp0,$car1,$car1
 399         and     $car0,$mask,$acc0
 400         ld      [$np+8],$npj                    ! np[2]
 401         srlx    $car1,32,$car1
 402         add     $tmp1,$car1,$car1
 403         srlx    $car0,32,$car0
 404         add     $acc0,$car1,$car1
 405         and     $car0,1,$sbit
 406         add     $acc1,$car1,$car1
 407         srlx    $car0,1,$car0
 408         mov     12,$j
 409         st      $car1,[%sp+$bias+$frame]        ! tp[0]=
 410         srlx    $car1,32,$car1
 411         add     %sp,$bias+$frame+4,$tp
 412
 413 .Lsqr_2nd:
 414         mulx    $apj,$mul0,$acc0
 415         mulx    $npj,$mul1,$acc1
 416         add     $acc0,$car0,$car0
 417         add     $tpj,$car1,$car1
 418         ld      [$ap+$j],$apj                   ! ap[j]
 419         and     $car0,$mask,$acc0
 420         ld      [$np+$j],$npj                   ! np[j]
 421         srlx    $car0,32,$car0
 422         add     $acc1,$car1,$car1
 423         ld      [$tp+8],$tpj                    ! tp[j]
 424         add     $acc0,$acc0,$acc0
 425         add     $j,4,$j                         ! j++
 426         or      $sbit,$acc0,$acc0
 427         srlx    $acc0,32,$sbit
 428         and     $acc0,$mask,$acc0
 429         cmp     $j,$num
 430         add     $acc0,$car1,$car1
 431         st      $car1,[$tp]                     ! tp[j-1]
 432         srlx    $car1,32,$car1
 433         bl      %icc,.Lsqr_2nd
 434         add     $tp,4,$tp                       ! tp++
 435 !.Lsqr_2nd
 436
 437         mulx    $apj,$mul0,$acc0
 438         mulx    $npj,$mul1,$acc1
 439         add     $acc0,$car0,$car0
 440         add     $tpj,$car1,$car1
 441         and     $car0,$mask,$acc0
 442         srlx    $car0,32,$car0
 443         add     $acc1,$car1,$car1
 444         add     $acc0,$acc0,$acc0
 445         or      $sbit,$acc0,$acc0
 446         srlx    $acc0,32,$sbit
 447         and     $acc0,$mask,$acc0
 448         add     $acc0,$car1,$car1
 449         st      $car1,[$tp]                     ! tp[j-1]
 450         srlx    $car1,32,$car1
 451
 452         add     $car0,$car0,$car0
 453         or      $sbit,$car0,$car0
 454         add     $car0,$car1,$car1
 455         add     $car2,$car1,$car1
 456         st      $car1,[$tp+4]
 457         srlx    $car1,32,$car2
 458 \f
 459         ld      [%sp+$bias+$frame],$tmp1        ! tp[0]
 460         ld      [%sp+$bias+$frame+4],$tpj       ! tp[1]
 461         ld      [$ap+8],$mul0                   ! ap[2]
 462         ld      [$np],$car1                     ! np[0]
 463         ld      [$np+4],$npj                    ! np[1]
 464         mulx    $n0,$tmp1,$mul1
 465         and     $mul1,$mask,$mul1
 466         mov     8,$i
 467
 468         mulx    $mul0,$mul0,$car0
 469         mulx    $car1,$mul1,$car1
 470         and     $car0,$mask,$acc0
 471         add     $tmp1,$car1,$car1
 472         srlx    $car0,32,$car0
 473         add     %sp,$bias+$frame,$tp
 474         srlx    $car1,32,$car1
 475         and     $car0,1,$sbit
 476         srlx    $car0,1,$car0
 477         mov     4,$j
 478
 479 .Lsqr_outer:
 480 .Lsqr_inner1:
 481         mulx    $npj,$mul1,$acc1
 482         add     $tpj,$car1,$car1
 483         add     $j,4,$j
 484         ld      [$tp+8],$tpj
 485         cmp     $j,$i
 486         add     $acc1,$car1,$car1
 487         ld      [$np+$j],$npj
 488         st      $car1,[$tp]
 489         srlx    $car1,32,$car1
 490         bl      %icc,.Lsqr_inner1
 491         add     $tp,4,$tp
 492 !.Lsqr_inner1
 493
 494         add     $j,4,$j
 495         ld      [$ap+$j],$apj                   ! ap[j]
 496         mulx    $npj,$mul1,$acc1
 497         add     $tpj,$car1,$car1
 498         ld      [$np+$j],$npj                   ! np[j]
 499         add     $acc0,$car1,$car1
 500         ld      [$tp+8],$tpj                    ! tp[j]
 501         add     $acc1,$car1,$car1
 502         st      $car1,[$tp]
 503         srlx    $car1,32,$car1
 504
 505         add     $j,4,$j
 506         cmp     $j,$num
 507         be,pn   %icc,.Lsqr_no_inner2
 508         add     $tp,4,$tp
 509
 510 .Lsqr_inner2:
 511         mulx    $apj,$mul0,$acc0
 512         mulx    $npj,$mul1,$acc1
 513         add     $tpj,$car1,$car1
 514         add     $acc0,$car0,$car0
 515         ld      [$ap+$j],$apj                   ! ap[j]
 516         and     $car0,$mask,$acc0
 517         ld      [$np+$j],$npj                   ! np[j]
 518         srlx    $car0,32,$car0
 519         add     $acc0,$acc0,$acc0
 520         ld      [$tp+8],$tpj                    ! tp[j]
 521         or      $sbit,$acc0,$acc0
 522         add     $j,4,$j                         ! j++
 523         srlx    $acc0,32,$sbit
 524         and     $acc0,$mask,$acc0
 525         cmp     $j,$num
 526         add     $acc0,$car1,$car1
 527         add     $acc1,$car1,$car1
 528         st      $car1,[$tp]                     ! tp[j-1]
 529         srlx    $car1,32,$car1
 530         bl      %icc,.Lsqr_inner2
 531         add     $tp,4,$tp                       ! tp++
 532
 533 .Lsqr_no_inner2:
 534         mulx    $apj,$mul0,$acc0
 535         mulx    $npj,$mul1,$acc1
 536         add     $tpj,$car1,$car1
 537         add     $acc0,$car0,$car0
 538         and     $car0,$mask,$acc0
 539         srlx    $car0,32,$car0
 540         add     $acc0,$acc0,$acc0
 541         or      $sbit,$acc0,$acc0
 542         srlx    $acc0,32,$sbit
 543         and     $acc0,$mask,$acc0
 544         add     $acc0,$car1,$car1
 545         add     $acc1,$car1,$car1
 546         st      $car1,[$tp]                     ! tp[j-1]
 547         srlx    $car1,32,$car1
 548
 549         add     $car0,$car0,$car0
 550         or      $sbit,$car0,$car0
 551         add     $car0,$car1,$car1
 552         add     $car2,$car1,$car1
 553         st      $car1,[$tp+4]
 554         srlx    $car1,32,$car2
 555 \f
 556         add     $i,4,$i                         ! i++
 557         ld      [%sp+$bias+$frame],$tmp1        ! tp[0]
 558         ld      [%sp+$bias+$frame+4],$tpj       ! tp[1]
 559         ld      [$ap+$i],$mul0                  ! ap[j]
 560         ld      [$np],$car1                     ! np[0]
 561         ld      [$np+4],$npj                    ! np[1]
 562         mulx    $n0,$tmp1,$mul1
 563         and     $mul1,$mask,$mul1
 564         add     $i,4,$tmp0
 565
 566         mulx    $mul0,$mul0,$car0
 567         mulx    $car1,$mul1,$car1
 568         and     $car0,$mask,$acc0
 569         add     $tmp1,$car1,$car1
 570         srlx    $car0,32,$car0
 571         add     %sp,$bias+$frame,$tp
 572         srlx    $car1,32,$car1
 573         and     $car0,1,$sbit
 574         srlx    $car0,1,$car0
 575
 576         cmp     $tmp0,$num                      ! i<num-1
 577         bl      %icc,.Lsqr_outer
 578         mov     4,$j
 579 \f
 580 .Lsqr_last:
 581         mulx    $npj,$mul1,$acc1
 582         add     $tpj,$car1,$car1
 583         add     $j,4,$j
 584         ld      [$tp+8],$tpj
 585         cmp     $j,$i
 586         add     $acc1,$car1,$car1
 587         ld      [$np+$j],$npj
 588         st      $car1,[$tp]
 589         srlx    $car1,32,$car1
 590         bl      %icc,.Lsqr_last
 591         add     $tp,4,$tp
 592 !.Lsqr_last
 593
 594         mulx    $npj,$mul1,$acc1
 595         add     $tpj,$car1,$car1
 596         add     $acc0,$car1,$car1
 597         add     $acc1,$car1,$car1
 598         st      $car1,[$tp]
 599         srlx    $car1,32,$car1
 600
 601         add     $car0,$car0,$car0               ! recover $car0
 602         or      $sbit,$car0,$car0
 603         add     $car0,$car1,$car1
 604         add     $car2,$car1,$car1
 605         st      $car1,[$tp+4]
 606         srlx    $car1,32,$car2
 607
 608         ba      .Ltail
 609         add     $tp,8,$tp
 610 .type   $fname,#function
 611 .size   $fname,(.-$fname)
 612 ___
 613 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 614 print $code;
 615 close STDOUT;