3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> 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 # ====================================================================
12 # SPARCv9 VIS3 Montgomery multiplicaion procedure suitable for T3 and
13 # onward. There are three new instructions used here: umulxhi,
14 # addxc[cc] and initializing store. On T3 RSA private key operations
15 # are 1.54/1.87/2.11/2.26 times faster for 512/1024/2048/4096-bit key
16 # lengths. This is without dedicated squaring procedure. On T4
17 # corresponding coefficients are 1.47/2.10/2.80/2.90x, which is mostly
18 # for reference purposes, because T4 has dedicated Montgomery
19 # multiplication and squaring *instructions* that deliver even more.
22 open STDOUT,">$output";
25 for (@ARGV) { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
26 if ($bits==64) { $bias=2047; $frame=192; }
27 else { $bias=0; $frame=112; }
29 $code.=<<___ if ($bits==64);
30 .register %g2,#scratch
31 .register %g3,#scratch
34 .section ".text",#alloc,#execinstr
37 ($n0,$m0,$m1,$lo0,$hi0, $lo1,$hi1,$aj,$alo,$nj,$nlo,$tj)=
38 (map("%g$_",(1..5)),map("%o$_",(0..5,7)));
41 $rp="%o0"; # BN_ULONG *rp,
42 $ap="%o1"; # const BN_ULONG *ap,
43 $bp="%o2"; # const BN_ULONG *bp,
44 $np="%o3"; # const BN_ULONG *np,
45 $n0p="%o4"; # const BN_ULONG *n0,
46 $num="%o5"; # int num); # caller ensures that num is even
49 .globl bn_mul_mont_vis3
52 add %sp, $bias, %g4 ! real top of stack
53 sll $num, 2, $num ! size in bytes
55 andn %g5, 63, %g5 ! buffer size rounded up to 64 bytes
57 add %g5, %g1, %g1 ! 3*buffer size
59 andn %g1, 63, %g1 ! align at 64 byte
60 sub %g1, $frame, %g1 ! new top of stack
66 # +-------------------------------+<----- %sp
68 # +-------------------------------+<----- aligned at 64 bytes
70 # +-------------------------------+
73 # +-------------------------------+<----- aligned at 64 bytes
74 # | __int64 ap[1..0] | converted ap[]
75 # +-------------------------------+
76 # | __int64 np[1..0] | converted np[]
77 # +-------------------------------+
78 # | __int64 ap[3..2] |
81 # +-------------------------------+
82 ($rp,$ap,$bp,$np,$n0p,$num)=map("%i$_",(0..5));
83 ($t0,$t1,$t2,$t3,$cnt,$tp,$bufsz,$anp)=map("%l$_",(0..7));
86 ld [$n0p+0], $t0 ! pull n0[0..1] value
87 add %sp, $bias+$frame, $tp
90 ld [$bp+0], $t2 ! m0=bp[0]
96 ld [$ap+0], $t0 ! ap[0]
101 ld [$ap+8], $t2 ! ap[1]
106 stx $aj, [$anp] ! converted ap[0]
108 mulx $aj, $m0, $lo0 ! ap[0]*bp[0]
109 umulxhi $aj, $m0, $hi0
111 ld [$np+0], $t0 ! np[0]
116 ld [$np+8], $t2 ! np[1]
121 stx $nj, [$anp+8] ! converted np[0]
123 mulx $lo0, $n0, $m1 ! "tp[0]"*n0
124 stx $aj, [$anp+16] ! converted ap[1]
126 mulx $aj, $m0, $alo ! ap[1]*bp[0]
127 umulxhi $aj, $m0, $aj ! ahi=aj
129 mulx $nj, $m1, $lo1 ! np[0]*m1
130 umulxhi $nj, $m1, $hi1
134 stx $nj, [$anp+24] ! converted np[1]
137 addcc $lo0, $lo1, $lo1
138 addxc %g0, $hi1, $hi1
140 mulx $nj, $m1, $nlo ! np[1]*m1
141 umulxhi $nj, $m1, $nj ! nhi=nj
144 sub $num, 24, $cnt ! cnt=num-3
148 ld [$ap+0], $t0 ! ap[j]
149 addcc $alo, $hi0, $lo0
156 stx $aj, [$anp] ! converted ap[j]
158 ld [$np+0], $t2 ! np[j]
159 addcc $nlo, $hi1, $lo1
161 addxc $nj, %g0, $hi1 ! nhi=nj
165 mulx $aj, $m0, $alo ! ap[j]*bp[0]
167 umulxhi $aj, $m0, $aj ! ahi=aj
168 stx $nj, [$anp+8] ! converted np[j]
169 add $anp, 16, $anp ! anp++
171 mulx $nj, $m1, $nlo ! np[j]*m1
172 addcc $lo0, $lo1, $lo1 ! np[j]*m1+ap[j]*bp[0]
173 umulxhi $nj, $m1, $nj ! nhi=nj
174 addxc %g0, $hi1, $hi1
175 stx $lo1, [$tp] ! tp[j-1]
176 add $tp, 8, $tp ! tp++
179 sub $cnt, 8, $cnt ! j--
181 addcc $alo, $hi0, $lo0
182 addxc $aj, %g0, $hi0 ! ahi=aj
184 addcc $nlo, $hi1, $lo1
186 addcc $lo0, $lo1, $lo1 ! np[j]*m1+ap[j]*bp[0]
187 addxc %g0, $hi1, $hi1
188 stx $lo1, [$tp] ! tp[j-1]
191 addcc $hi0, $hi1, $hi1
192 addxc %g0, %g0, $ovf ! upmost overflow bit
197 sub $num, 16, $i ! i=num-2
201 ld [$bp+0], $t2 ! m0=bp[i]
204 sub $anp, $num, $anp ! rewind
210 ldx [$anp+0], $aj ! ap[0]
212 ldx [$anp+8], $nj ! np[0]
214 mulx $aj, $m0, $lo0 ! ap[0]*bp[i]
215 ldx [$tp], $tj ! tp[0]
216 umulxhi $aj, $m0, $hi0
217 ldx [$anp+16], $aj ! ap[1]
218 addcc $lo0, $tj, $lo0 ! ap[0]*bp[i]+tp[0]
219 mulx $aj, $m0, $alo ! ap[1]*bp[i]
220 addxc %g0, $hi0, $hi0
221 mulx $lo0, $n0, $m1 ! tp[0]*n0
222 umulxhi $aj, $m0, $aj ! ahi=aj
223 mulx $nj, $m1, $lo1 ! np[0]*m1
224 umulxhi $nj, $m1, $hi1
225 ldx [$anp+24], $nj ! np[1]
227 addcc $lo1, $lo0, $lo1
228 mulx $nj, $m1, $nlo ! np[1]*m1
229 addxc %g0, $hi1, $hi1
230 umulxhi $nj, $m1, $nj ! nhi=nj
233 sub $num, 24, $cnt ! cnt=num-3
236 addcc $alo, $hi0, $lo0
237 ldx [$tp+8], $tj ! tp[j]
238 addxc $aj, %g0, $hi0 ! ahi=aj
239 ldx [$anp+0], $aj ! ap[j]
240 addcc $nlo, $hi1, $lo1
241 mulx $aj, $m0, $alo ! ap[j]*bp[i]
242 addxc $nj, %g0, $hi1 ! nhi=nj
243 ldx [$anp+8], $nj ! np[j]
245 umulxhi $aj, $m0, $aj ! ahi=aj
246 addcc $lo0, $tj, $lo0 ! ap[j]*bp[i]+tp[j]
247 mulx $nj, $m1, $nlo ! np[j]*m1
248 addxc %g0, $hi0, $hi0
249 umulxhi $nj, $m1, $nj ! nhi=nj
250 addcc $lo1, $lo0, $lo1 ! np[j]*m1+ap[j]*bp[i]+tp[j]
251 addxc %g0, $hi1, $hi1
252 stx $lo1, [$tp] ! tp[j-1]
254 brnz,pt $cnt, .Linner
257 ldx [$tp+8], $tj ! tp[j]
258 addcc $alo, $hi0, $lo0
259 addxc $aj, %g0, $hi0 ! ahi=aj
260 addcc $lo0, $tj, $lo0 ! ap[j]*bp[i]+tp[j]
261 addxc %g0, $hi0, $hi0
263 addcc $nlo, $hi1, $lo1
264 addxc $nj, %g0, $hi1 ! nhi=nj
265 addcc $lo1, $lo0, $lo1 ! np[j]*m1+ap[j]*bp[i]+tp[j]
266 addxc %g0, $hi1, $hi1
267 stx $lo1, [$tp] ! tp[j-1]
269 subcc %g0, $ovf, %g0 ! move upmost overflow to CCR.xcc
270 addxccc $hi1, $hi0, $hi1
278 sub $anp, $num, $anp ! rewind
282 subcc $num, 8, $cnt ! cnt=num-1 and clear CCR.xcc
290 subccc $tj, $nj, $t2 ! tp[j]-np[j]
295 st $t2, [$rp-4] ! reverse order
300 sub $anp, $num, $anp ! rewind
305 subc $ovf, %g0, $ovf ! handle upmost overflow bit
308 or $np, $ap, $ap ! ap=borrow?tp:rp
313 .Lcopy: ! copy or in-place refresh
319 stx %g0, [$anp] ! zap
322 st $t3, [$rp+0] ! flip order
331 .type bn_mul_mont_vis3, #function
332 .size bn_mul_mont_vis3, .-bn_mul_mont_vis3
333 .asciz "Montgomery Multiplication for SPARCv9 VIS3, CRYPTOGAMS by <appro\@openssl.org>"
337 # Purpose of these subroutines is to explicitly encode VIS instructions,
338 # so that one can compile the module without having to specify VIS
339 # extensions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
340 # Idea is to reserve for option to produce "universal" binary and let
341 # programmer detect if current CPU is VIS capable at run-time.
343 my ($mnemonic,$rs1,$rs2,$rd)=@_;
344 my %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
346 my %visopf = ( "addxc" => 0x011,
348 "umulxhi" => 0x016 );
350 $ref = "$mnemonic\t$rs1,$rs2,$rd";
352 if ($opf=$visopf{$mnemonic}) {
353 foreach ($rs1,$rs2,$rd) {
354 return $ref if (!/%([goli])([0-9])/);
358 return sprintf ".word\t0x%08x !%s",
359 0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
366 foreach (split("\n",$code)) {
367 s/\`([^\`]*)\`/eval $1/ge;
369 s/\b(umulxhi|addxc[c]{0,2})\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/