2 # Copyright 2012-2016 The OpenSSL Project Authors. All Rights Reserved.
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
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 # ====================================================================
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.
28 $output = pop and open STDOUT,">$output";
37 ($a1,$a2,$a4,$a8,$a12,$a48)=map("%o$_",(0..5));
38 ($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;
41 #include <sparc_arch.h>
44 .register %g2,#scratch
45 .register %g3,#scratch
52 .globl bn_GF2m_mul_2x2
55 SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
56 ld [%g1+0],%g1 ! OPENSSL_sparcv9cap_P[0]
58 andcc %g1, SPARCV9_VIS3, %g0
66 .word 0x95b262ab ! xmulx %o1, %o3, %o2
67 .word 0x99b262cb ! xmulxhi %o1, %o3, %o4
68 srlx %o2, 32, %o1 ! 13 cycles later
78 save %sp,-STACK_FRAME-$locals,%sp
84 srlx $a12,1,$a48 ! 0x7fff...
86 srlx $a12,2,$a12 ! 0x3fff...
87 add %sp,STACK_BIAS+STACK_FRAME,$tab
93 srax $a4,63,@i[1] ! broadcast 61st bit
94 and $a48,$a4,$a4 ! (a<<2)&0x7fff...
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,$a1 ! (a<<0)&0x1fff...
106 stx %g0,[$tab+0*8] ! tab[0]=0
108 stx $a1,[$tab+1*8] ! tab[1]=a1
109 stx $a2,[$tab+2*8] ! tab[2]=a2
111 stx $a12,[$tab+3*8] ! tab[3]=a1^a2
114 stx $a4,[$tab+4*8] ! tab[4]=a4
116 stx $a1,[$tab+5*8] ! tab[5]=a1^a4
118 stx $a2,[$tab+6*8] ! tab[6]=a2^a4
120 stx $a12,[$tab+7*8] ! tab[7]=a1^a2^a4
123 stx $a8,[$tab+8*8] ! tab[8]=a8
125 stx $a1,[$tab+9*8] ! tab[9]=a1^a8
127 stx $a2,[$tab+10*8] ! tab[10]=a2^a8
129 stx $a12,[$tab+11*8] ! tab[11]=a1^a2^a8
132 stx $a48,[$tab+12*8] ! tab[12]=a4^a8
134 stx $a1,[$tab+13*8] ! tab[13]=a1^a4^a8
136 stx $a2,[$tab+14*8] ! tab[14]=a2^a4^a8
138 stx $a12,[$tab+15*8] ! tab[15]=a1^a2^a4^a8
143 and @i[0],`0xf<<3`,@i[0]
145 ldx [$tab+@i[0]],@i[0]
148 and @i[1],`0xf<<3`,@i[1]
150 ldx [$tab+@i[1]],@i[1]
156 and @i[0],`0xf<<3`,@i[0]
158 for($n=1;$n<14;$n++) {
160 sllx @i[1],`$n*4`,@T[0]
161 ldx [$tab+@i[0]],@i[0]
162 srlx @i[1],`64-$n*4`,@T[1]
164 srlx $b,`($n+2)*4`-3,@i[1]
166 and @i[1],`0xf<<3`,@i[1]
168 push(@i,shift(@i)); push(@T,shift(@T));
171 sllx @i[1],`$n*4`,@T[0]
172 ldx [$tab+@i[0]],@i[0]
173 srlx @i[1],`64-$n*4`,@T[1]
176 sllx @i[0],`($n+1)*4`,@T[0]
178 srlx @i[0],`64-($n+1)*4`,@T[1]
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>"
197 $code =~ s/\`([^\`]*)\`/eval($1)/gem;