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 # ====================================================================
12 # "Teaser" Montgomery multiplication module for PowerPC. It's possible
13 # to gain a bit more by modulo-scheduling outer loop, then dedicated
14 # squaring procedure should give further 20% and code can be adapted
15 # for 32-bit application running on 64-bit CPU. As for the latter.
16 # It won't be able to achieve "native" 64-bit performance, because in
17 # 32-bit application context every addc instruction will have to be
18 # expanded as addc, twice right shift by 32 and finally adde, etc.
19 # So far RSA *sign* performance improvement over pre-bn_mul_mont asm
20 # for 64-bit application running on PPC970/G5 is:
29 if ($output =~ /32\-mont\.s/) {
37 $LDU= "lwzu"; # load and update
38 $LDX= "lwzx"; # load indexed
40 $STU= "stwu"; # store and update
41 $STX= "stwx"; # store indexed
42 $STUX= "stwux"; # store indexed and update
43 $UMULL= "mullw"; # unsigned multiply low
44 $UMULH= "mulhwu"; # unsigned multiply high
45 $UCMP= "cmplw"; # unsigned compare
46 $SHRI= "srwi"; # unsigned shift right by immediate
49 } elsif ($output =~ /64\-mont\.s/) {
56 # same as above, but 64-bit mnemonics...
58 $LDU= "ldu"; # load and update
59 $LDX= "ldx"; # load indexed
61 $STU= "stdu"; # store and update
62 $STX= "stdx"; # store indexed
63 $STUX= "stdux"; # store indexed and update
64 $UMULL= "mulld"; # unsigned multiply low
65 $UMULH= "mulhdu"; # unsigned multiply high
66 $UCMP= "cmpld"; # unsigned compare
67 $SHRI= "srdi"; # unsigned shift right by immediate
70 } else { die "nonsense $output"; }
72 ( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
73 die "can't call ../perlasm/ppc-xlate.pl: $!";
83 $rp="r9"; # $rp is reassigned
87 # non-volatile registers
111 mr $rp,r3 ; $rp is reassigned
115 slwi $num,$num,`log($BNSZ)/log(2)`
117 addi $ovf,$num,`$FRAME+$RZONE`
118 subf $ovf,$ovf,$sp ; $sp-$ovf
119 and $ovf,$ovf,$tj ; minimize TLB usage
120 subf $ovf,$sp,$ovf ; $ovf-$sp
121 srwi $num,$num,`log($BNSZ)/log(2)`
124 $PUSH r14,`4*$SIZE_T`($sp)
125 $PUSH r15,`5*$SIZE_T`($sp)
126 $PUSH r16,`6*$SIZE_T`($sp)
127 $PUSH r17,`7*$SIZE_T`($sp)
128 $PUSH r18,`8*$SIZE_T`($sp)
129 $PUSH r19,`9*$SIZE_T`($sp)
130 $PUSH r20,`10*$SIZE_T`($sp)
131 $PUSH r21,`11*$SIZE_T`($sp)
132 $PUSH r22,`12*$SIZE_T`($sp)
133 $PUSH r23,`13*$SIZE_T`($sp)
134 $PUSH r24,`14*$SIZE_T`($sp)
135 $PUSH r25,`15*$SIZE_T`($sp)
137 $LD $n0,0($n0) ; pull n0[0] value
138 addi $num,$num,-2 ; adjust $num for counter register
140 $LD $m0,0($bp) ; m0=bp[0]
141 $LD $aj,0($ap) ; ap[0]
143 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[0]
146 $LD $aj,$BNSZ($ap) ; ap[1]
147 $LD $nj,0($np) ; np[0]
149 $UMULL $m1,$lo0,$n0 ; "tp[0]"*n0
151 $UMULL $alo,$aj,$m0 ; ap[1]*bp[0]
154 $UMULL $lo1,$nj,$m1 ; np[0]*m1
156 $LD $nj,$BNSZ($np) ; np[1]
160 $UMULL $nlo,$nj,$m1 ; np[1]*m1
167 $LDX $aj,$ap,$j ; ap[j]
169 $LDX $nj,$np,$j ; np[j]
171 $UMULL $alo,$aj,$m0 ; ap[j]*bp[0]
175 $UMULL $nlo,$nj,$m1 ; np[j]*m1
176 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
179 $ST $lo1,0($tp) ; tp[j-1]
181 addi $j,$j,$BNSZ ; j++
182 addi $tp,$tp,$BNSZ ; tp++
190 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
192 $ST $lo1,0($tp) ; tp[j-1]
196 addze $ovf,$ovf ; upmost overflow bit
202 $LDX $m0,$bp,$i ; m0=bp[i]
203 $LD $aj,0($ap) ; ap[0]
205 $LD $tj,$FRAME($sp) ; tp[0]
206 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[i]
208 $LD $aj,$BNSZ($ap) ; ap[1]
209 $LD $nj,0($np) ; np[0]
210 addc $lo0,$lo0,$tj ; ap[0]*bp[i]+tp[0]
211 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
213 $UMULL $m1,$lo0,$n0 ; tp[0]*n0
215 $UMULL $lo1,$nj,$m1 ; np[0]*m1
217 $LD $nj,$BNSZ($np) ; np[1]
219 $UMULL $nlo,$nj,$m1 ; np[1]*m1
227 $LDX $aj,$ap,$j ; ap[j]
229 $LD $tj,$BNSZ($tp) ; tp[j]
231 $LDX $nj,$np,$j ; np[j]
233 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
236 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
237 $UMULL $nlo,$nj,$m1 ; np[j]*m1
240 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
241 addi $j,$j,$BNSZ ; j++
243 $ST $lo1,0($tp) ; tp[j-1]
244 addi $tp,$tp,$BNSZ ; tp++
247 $LD $tj,$BNSZ($tp) ; tp[j]
250 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
255 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
257 $ST $lo1,0($tp) ; tp[j-1]
259 addic $ovf,$ovf,-1 ; move upmost overflow to XER[CA]
265 slwi $tj,$num,`log($BNSZ)/log(2)`
270 $SHRI. $nj,$nj,$BITS-2 ; check boundary condition
271 addi $num,$num,2 ; restore $num
272 subfc $j,$j,$j ; j=0 and "clear" XER[CA]
276 beq Lcopy ; boundary condition is met
279 Lsub: $LDX $tj,$tp,$j
281 subfe $aj,$nj,$tj ; tp[j]-np[j]
288 subfe $ovf,$j,$ovf ; handle upmost overflow bit
291 or $ap,$ap,$np ; ap=borrow?tp:rp
294 Lcopy: ; copy or in-place refresh
297 $STX $j,$tp,$j ; zap at once
301 $POP r14,`4*$SIZE_T`($sp)
302 $POP r15,`5*$SIZE_T`($sp)
303 $POP r16,`6*$SIZE_T`($sp)
304 $POP r17,`7*$SIZE_T`($sp)
305 $POP r18,`8*$SIZE_T`($sp)
306 $POP r19,`9*$SIZE_T`($sp)
307 $POP r20,`10*$SIZE_T`($sp)
308 $POP r21,`11*$SIZE_T`($sp)
309 $POP r22,`12*$SIZE_T`($sp)
310 $POP r23,`13*$SIZE_T`($sp)
311 $POP r24,`14*$SIZE_T`($sp)
312 $POP r25,`15*$SIZE_T`($sp)
317 .asciz "Montgomery Multiplication for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"
320 $code =~ s/\`([^\`]*)\`/eval $1/gem;