crypto/bn/asm/sparcv9-gf2m.pl

   1 #! /usr/bin/env perl
   2 # Copyright 2012-2020 The OpenSSL Project Authors. All Rights Reserved.
   3 #
   4 # Licensed under the Apache License 2.0 (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 # October 2012
  18 #
  19 # The module implements bn_GF2m_mul_2x2 polynomial multiplication used
  20 # in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
  21 # the time being... Except that it has two code paths: one suitable
  22 # for all SPARCv9 processors and one for VIS3-capable ones. Former
  23 # delivers ~25-45% more, more for longer keys, heaviest DH and DSA
  24 # verify operations on venerable UltraSPARC II. On T4 VIS3 code is
  25 # ~100-230% faster than gcc-generated code and ~35-90% faster than
  26 # the pure SPARCv9 code path.
  27
  28 $output = pop and open STDOUT,">$output";
  29
  30 $locals=16*8;
  31
  32 $tab="%l0";
  33
  34 @T=("%g2","%g3");
  35 @i=("%g4","%g5");
  36
  37 ($a1,$a2,$a4,$a8,$a12,$a48)=map("%o$_",(0..5));
  38 ($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;
  39
  40 $code.=<<___;
  41 #include <sparc_arch.h>
  42
  43 #ifdef __arch64__
  44 .register       %g2,#scratch
  45 .register       %g3,#scratch
  46 #endif
  47
  48 #ifdef __PIC__
  49 SPARC_PIC_THUNK(%g1)
  50 #endif
  51
  52 .globl  bn_GF2m_mul_2x2
  53 .align  16
  54 bn_GF2m_mul_2x2:
  55         SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
  56         ld      [%g1+0],%g1                     ! OPENSSL_sparcv9cap_P[0]
  57
  58         andcc   %g1, SPARCV9_VIS3, %g0
  59         bz,pn   %icc,.Lsoftware
  60         nop
  61
  62         sllx    %o1, 32, %o1
  63         sllx    %o3, 32, %o3
  64         or      %o2, %o1, %o1
  65         or      %o4, %o3, %o3
  66         .word   0x95b262ab                      ! xmulx   %o1, %o3, %o2
  67         .word   0x99b262cb                      ! xmulxhi %o1, %o3, %o4
  68         srlx    %o2, 32, %o1                    ! 13 cycles later
  69         st      %o2, [%o0+0]
  70         st      %o1, [%o0+4]
  71         srlx    %o4, 32, %o3
  72         st      %o4, [%o0+8]
  73         retl
  74         st      %o3, [%o0+12]
  75
  76 .align  16
  77 .Lsoftware:
  78         save    %sp,-STACK_FRAME-$locals,%sp
  79
  80         sllx    %i1,32,$a
  81         mov     -1,$a12
  82         sllx    %i3,32,$b
  83         or      %i2,$a,$a
  84         srlx    $a12,1,$a48                     ! 0x7fff...
  85         or      %i4,$b,$b
  86         srlx    $a12,2,$a12                     ! 0x3fff...
  87         add     %sp,STACK_BIAS+STACK_FRAME,$tab
  88
  89         sllx    $a,2,$a4
  90         mov     $a,$a1
  91         sllx    $a,1,$a2
  92
  93         srax    $a4,63,@i[1]                    ! broadcast 61st bit
  94         and     $a48,$a4,$a4                    ! (a<<2)&0x7fff...
  95         srlx    $a48,2,$a48
  96         srax    $a2,63,@i[0]                    ! broadcast 62nd bit
  97         and     $a12,$a2,$a2                    ! (a<<1)&0x3fff...
  98         srax    $a1,63,$lo                      ! broadcast 63rd bit
  99         and     $a48,$a1,$a1                    ! (a<<0)&0x1fff...
 100
 101         sllx    $a1,3,$a8
 102         and     $b,$lo,$lo
 103         and     $b,@i[0],@i[0]
 104         and     $b,@i[1],@i[1]
 105
 106         stx     %g0,[$tab+0*8]                  ! tab[0]=0
 107         xor     $a1,$a2,$a12
 108         stx     $a1,[$tab+1*8]                  ! tab[1]=a1
 109         stx     $a2,[$tab+2*8]                  ! tab[2]=a2
 110          xor    $a4,$a8,$a48
 111         stx     $a12,[$tab+3*8]                 ! tab[3]=a1^a2
 112          xor    $a4,$a1,$a1
 113
 114         stx     $a4,[$tab+4*8]                  ! tab[4]=a4
 115         xor     $a4,$a2,$a2
 116         stx     $a1,[$tab+5*8]                  ! tab[5]=a1^a4
 117         xor     $a4,$a12,$a12
 118         stx     $a2,[$tab+6*8]                  ! tab[6]=a2^a4
 119          xor    $a48,$a1,$a1
 120         stx     $a12,[$tab+7*8]                 ! tab[7]=a1^a2^a4
 121          xor    $a48,$a2,$a2
 122
 123         stx     $a8,[$tab+8*8]                  ! tab[8]=a8
 124         xor     $a48,$a12,$a12
 125         stx     $a1,[$tab+9*8]                  ! tab[9]=a1^a8
 126          xor    $a4,$a1,$a1
 127         stx     $a2,[$tab+10*8]                 ! tab[10]=a2^a8
 128          xor    $a4,$a2,$a2
 129         stx     $a12,[$tab+11*8]                ! tab[11]=a1^a2^a8
 130
 131         xor     $a4,$a12,$a12
 132         stx     $a48,[$tab+12*8]                ! tab[12]=a4^a8
 133          srlx   $lo,1,$hi
 134         stx     $a1,[$tab+13*8]                 ! tab[13]=a1^a4^a8
 135          sllx   $lo,63,$lo
 136         stx     $a2,[$tab+14*8]                 ! tab[14]=a2^a4^a8
 137          srlx   @i[0],2,@T[0]
 138         stx     $a12,[$tab+15*8]                ! tab[15]=a1^a2^a4^a8
 139
 140         sllx    @i[0],62,$a1
 141          sllx   $b,3,@i[0]
 142         srlx    @i[1],3,@T[1]
 143          and    @i[0],`0xf<<3`,@i[0]
 144         sllx    @i[1],61,$a2
 145          ldx    [$tab+@i[0]],@i[0]
 146          srlx   $b,4-3,@i[1]
 147         xor     @T[0],$hi,$hi
 148          and    @i[1],`0xf<<3`,@i[1]
 149         xor     $a1,$lo,$lo
 150          ldx    [$tab+@i[1]],@i[1]
 151         xor     @T[1],$hi,$hi
 152
 153         xor     @i[0],$lo,$lo
 154         srlx    $b,8-3,@i[0]
 155          xor    $a2,$lo,$lo
 156         and     @i[0],`0xf<<3`,@i[0]
 157 ___
 158 for($n=1;$n<14;$n++) {
 159 $code.=<<___;
 160         sllx    @i[1],`$n*4`,@T[0]
 161         ldx     [$tab+@i[0]],@i[0]
 162         srlx    @i[1],`64-$n*4`,@T[1]
 163         xor     @T[0],$lo,$lo
 164         srlx    $b,`($n+2)*4`-3,@i[1]
 165         xor     @T[1],$hi,$hi
 166         and     @i[1],`0xf<<3`,@i[1]
 167 ___
 168         push(@i,shift(@i)); push(@T,shift(@T));
 169 }
 170 $code.=<<___;
 171         sllx    @i[1],`$n*4`,@T[0]
 172         ldx     [$tab+@i[0]],@i[0]
 173         srlx    @i[1],`64-$n*4`,@T[1]
 174         xor     @T[0],$lo,$lo
 175
 176         sllx    @i[0],`($n+1)*4`,@T[0]
 177          xor    @T[1],$hi,$hi
 178         srlx    @i[0],`64-($n+1)*4`,@T[1]
 179         xor     @T[0],$lo,$lo
 180         xor     @T[1],$hi,$hi
 181
 182         srlx    $lo,32,%i1
 183         st      $lo,[%i0+0]
 184         st      %i1,[%i0+4]
 185         srlx    $hi,32,%i2
 186         st      $hi,[%i0+8]
 187         st      %i2,[%i0+12]
 188
 189         ret
 190         restore
 191 .type   bn_GF2m_mul_2x2,#function
 192 .size   bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
 193 .asciz  "GF(2^m) Multiplication for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
 194 .align  4
 195 ___
 196
 197 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
 198 print $code;
 199 close STDOUT or die "error closing STDOUT: $!";