lib/libcrypto/bn/asm/alpha-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 # On 21264 RSA sign performance improves by 70/35/20/15 percent for
  11 # 512/1024/2048/4096 bit key lengths. This is against vendor compiler
  12 # instructed to '-tune host' code with in-line assembler. Other
  13 # benchmarks improve by 15-20%. To anchor it to something else, the
  14 # code provides approximately the same performance per GHz as AMD64.
  15 # I.e. if you compare 1GHz 21264 and 2GHz Opteron, you'll observe ~2x
  16 # difference.
  17
  18 # int bn_mul_mont(
  19 $rp="a0";       # BN_ULONG *rp,
  20 $ap="a1";       # const BN_ULONG *ap,
  21 $bp="a2";       # const BN_ULONG *bp,
  22 $np="a3";       # const BN_ULONG *np,
  23 $n0="a4";       # const BN_ULONG *n0,
  24 $num="a5";      # int num);
  25
  26 $lo0="t0";
  27 $hi0="t1";
  28 $lo1="t2";
  29 $hi1="t3";
  30 $aj="t4";
  31 $bi="t5";
  32 $nj="t6";
  33 $tp="t7";
  34 $alo="t8";
  35 $ahi="t9";
  36 $nlo="t10";
  37 $nhi="t11";
  38 $tj="t12";
  39 $i="s3";
  40 $j="s4";
  41 $m1="s5";
  42
  43 $code=<<___;
  44 #include <machine/asm.h>
  45
  46 .text
  47
  48 .set    noat
  49 .set    noreorder
  50
  51 .globl  bn_mul_mont
  52 .align  5
  53 .ent    bn_mul_mont
  54 bn_mul_mont:
  55         lda     sp,-48(sp)
  56         stq     ra,0(sp)
  57         stq     s3,8(sp)
  58         stq     s4,16(sp)
  59         stq     s5,24(sp)
  60         stq     fp,32(sp)
  61         mov     sp,fp
  62         .mask   0x0400f000,-48
  63         .frame  fp,48,ra
  64         .prologue 0
  65
  66         .align  4
  67         .set    reorder
  68         sextl   $num,$num
  69         mov     0,v0
  70         cmplt   $num,4,AT
  71         bne     AT,.Lexit
  72
  73         ldq     $hi0,0($ap)     # ap[0]
  74         s8addq  $num,16,AT
  75         ldq     $aj,8($ap)
  76         subq    sp,AT,sp
  77         ldq     $bi,0($bp)      # bp[0]
  78         lda     AT,-4096(zero)  # mov   -4096,AT
  79         ldq     $n0,0($n0)
  80         and     sp,AT,sp
  81
  82         mulq    $hi0,$bi,$lo0
  83         ldq     $hi1,0($np)     # np[0]
  84         umulh   $hi0,$bi,$hi0
  85         ldq     $nj,8($np)
  86
  87         mulq    $lo0,$n0,$m1
  88
  89         mulq    $hi1,$m1,$lo1
  90         umulh   $hi1,$m1,$hi1
  91
  92         addq    $lo1,$lo0,$lo1
  93         cmpult  $lo1,$lo0,AT
  94         addq    $hi1,AT,$hi1
  95
  96         mulq    $aj,$bi,$alo
  97         mov     2,$j
  98         umulh   $aj,$bi,$ahi
  99         mov     sp,$tp
 100
 101         mulq    $nj,$m1,$nlo
 102         s8addq  $j,$ap,$aj
 103         umulh   $nj,$m1,$nhi
 104         s8addq  $j,$np,$nj
 105 .align  4
 106 .L1st:
 107         .set    noreorder
 108         ldq     $aj,0($aj)
 109         addl    $j,1,$j
 110         ldq     $nj,0($nj)
 111         lda     $tp,8($tp)
 112
 113         addq    $alo,$hi0,$lo0
 114         mulq    $aj,$bi,$alo
 115         cmpult  $lo0,$hi0,AT
 116         addq    $nlo,$hi1,$lo1
 117
 118         mulq    $nj,$m1,$nlo
 119         addq    $ahi,AT,$hi0
 120         cmpult  $lo1,$hi1,v0
 121         cmplt   $j,$num,$tj
 122
 123         umulh   $aj,$bi,$ahi
 124         addq    $nhi,v0,$hi1
 125         addq    $lo1,$lo0,$lo1
 126         s8addq  $j,$ap,$aj
 127
 128         umulh   $nj,$m1,$nhi
 129         cmpult  $lo1,$lo0,v0
 130         addq    $hi1,v0,$hi1
 131         s8addq  $j,$np,$nj
 132
 133         stq     $lo1,-8($tp)
 134         nop
 135         unop
 136         bne     $tj,.L1st
 137         .set    reorder
 138
 139         addq    $alo,$hi0,$lo0
 140         addq    $nlo,$hi1,$lo1
 141         cmpult  $lo0,$hi0,AT
 142         cmpult  $lo1,$hi1,v0
 143         addq    $ahi,AT,$hi0
 144         addq    $nhi,v0,$hi1
 145
 146         addq    $lo1,$lo0,$lo1
 147         cmpult  $lo1,$lo0,v0
 148         addq    $hi1,v0,$hi1
 149
 150         stq     $lo1,0($tp)
 151
 152         addq    $hi1,$hi0,$hi1
 153         cmpult  $hi1,$hi0,AT
 154         stq     $hi1,8($tp)
 155         stq     AT,16($tp)
 156
 157         mov     1,$i
 158 .align  4
 159 .Louter:
 160         s8addq  $i,$bp,$bi
 161         ldq     $hi0,0($ap)
 162         ldq     $aj,8($ap)
 163         ldq     $bi,0($bi)
 164         ldq     $hi1,0($np)
 165         ldq     $nj,8($np)
 166         ldq     $tj,0(sp)
 167
 168         mulq    $hi0,$bi,$lo0
 169         umulh   $hi0,$bi,$hi0
 170
 171         addq    $lo0,$tj,$lo0
 172         cmpult  $lo0,$tj,AT
 173         addq    $hi0,AT,$hi0
 174
 175         mulq    $lo0,$n0,$m1
 176
 177         mulq    $hi1,$m1,$lo1
 178         umulh   $hi1,$m1,$hi1
 179
 180         addq    $lo1,$lo0,$lo1
 181         cmpult  $lo1,$lo0,AT
 182         mov     2,$j
 183         addq    $hi1,AT,$hi1
 184
 185         mulq    $aj,$bi,$alo
 186         mov     sp,$tp
 187         umulh   $aj,$bi,$ahi
 188
 189         mulq    $nj,$m1,$nlo
 190         s8addq  $j,$ap,$aj
 191         umulh   $nj,$m1,$nhi
 192 .align  4
 193 .Linner:
 194         .set    noreorder
 195         ldq     $tj,8($tp)      #L0
 196         nop                     #U1
 197         ldq     $aj,0($aj)      #L1
 198         s8addq  $j,$np,$nj      #U0
 199
 200         ldq     $nj,0($nj)      #L0
 201         nop                     #U1
 202         addq    $alo,$hi0,$lo0  #L1
 203         lda     $tp,8($tp)
 204
 205         mulq    $aj,$bi,$alo    #U1
 206         cmpult  $lo0,$hi0,AT    #L0
 207         addq    $nlo,$hi1,$lo1  #L1
 208         addl    $j,1,$j
 209
 210         mulq    $nj,$m1,$nlo    #U1
 211         addq    $ahi,AT,$hi0    #L0
 212         addq    $lo0,$tj,$lo0   #L1
 213         cmpult  $lo1,$hi1,v0    #U0
 214
 215         umulh   $aj,$bi,$ahi    #U1
 216         cmpult  $lo0,$tj,AT     #L0
 217         addq    $lo1,$lo0,$lo1  #L1
 218         addq    $nhi,v0,$hi1    #U0
 219
 220         umulh   $nj,$m1,$nhi    #U1
 221         s8addq  $j,$ap,$aj      #L0
 222         cmpult  $lo1,$lo0,v0    #L1
 223         cmplt   $j,$num,$tj     #U0     # borrow $tj
 224
 225         addq    $hi0,AT,$hi0    #L0
 226         addq    $hi1,v0,$hi1    #U1
 227         stq     $lo1,-8($tp)    #L1
 228         bne     $tj,.Linner     #U0
 229         .set    reorder
 230
 231         ldq     $tj,8($tp)
 232         addq    $alo,$hi0,$lo0
 233         addq    $nlo,$hi1,$lo1
 234         cmpult  $lo0,$hi0,AT
 235         cmpult  $lo1,$hi1,v0
 236         addq    $ahi,AT,$hi0
 237         addq    $nhi,v0,$hi1
 238
 239         addq    $lo0,$tj,$lo0
 240         cmpult  $lo0,$tj,AT
 241         addq    $hi0,AT,$hi0
 242
 243         ldq     $tj,16($tp)
 244         addq    $lo1,$lo0,$j
 245         cmpult  $j,$lo0,v0
 246         addq    $hi1,v0,$hi1
 247
 248         addq    $hi1,$hi0,$lo1
 249         stq     $j,0($tp)
 250         cmpult  $lo1,$hi0,$hi1
 251         addq    $lo1,$tj,$lo1
 252         cmpult  $lo1,$tj,AT
 253         addl    $i,1,$i
 254         addq    $hi1,AT,$hi1
 255         stq     $lo1,8($tp)
 256         cmplt   $i,$num,$tj     # borrow $tj
 257         stq     $hi1,16($tp)
 258         bne     $tj,.Louter
 259 \f
 260         s8addq  $num,sp,$tj     # &tp[num]
 261         mov     $rp,$bp         # put rp aside
 262         mov     sp,$tp
 263         mov     sp,$ap
 264         mov     0,$hi0          # clear borrow bit
 265
 266 .align  4
 267 .Lsub:  ldq     $lo0,0($tp)
 268         ldq     $lo1,0($np)
 269         lda     $tp,8($tp)
 270         lda     $np,8($np)
 271         subq    $lo0,$lo1,$lo1  # tp[i]-np[i]
 272         cmpult  $lo0,$lo1,AT
 273         subq    $lo1,$hi0,$lo0
 274         cmpult  $lo1,$lo0,$hi0
 275         or      $hi0,AT,$hi0
 276         stq     $lo0,0($rp)
 277         cmpult  $tp,$tj,v0
 278         lda     $rp,8($rp)
 279         bne     v0,.Lsub
 280
 281         subq    $hi1,$hi0,$hi0  # handle upmost overflow bit
 282         mov     sp,$tp
 283         mov     $bp,$rp         # restore rp
 284
 285         and     sp,$hi0,$ap
 286         bic     $bp,$hi0,$bp
 287         bis     $bp,$ap,$ap     # ap=borrow?tp:rp
 288
 289 .align  4
 290 .Lcopy: ldq     $aj,0($ap)      # copy or in-place refresh
 291         lda     $tp,8($tp)
 292         lda     $rp,8($rp)
 293         lda     $ap,8($ap)
 294         stq     zero,-8($tp)    # zap tp
 295         cmpult  $tp,$tj,AT
 296         stq     $aj,-8($rp)
 297         bne     AT,.Lcopy
 298         mov     1,v0
 299
 300 .Lexit:
 301         .set    noreorder
 302         mov     fp,sp
 303         /*ldq   ra,0(sp)*/
 304         ldq     s3,8(sp)
 305         ldq     s4,16(sp)
 306         ldq     s5,24(sp)
 307         ldq     fp,32(sp)
 308         lda     sp,48(sp)
 309         ret     (ra)
 310 .end    bn_mul_mont
 311 .ascii  "Montgomery Multiplication for Alpha, CRYPTOGAMS by <appro\@openssl.org>"
 312 .align  2
 313 ___
 314
 315 print $code;
 316 close STDOUT;