3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5 # project. Rights for redistribution and usage in source and binary
6 # forms are granted according to the OpenSSL license.
7 # ====================================================================
11 # "Teaser" Montgomery multiplication module for PowerPC. It's possible
12 # to gain a bit more by modulo-scheduling outer loop, then dedicated
13 # squaring procedure should give further 20% and code can be adapted
14 # for 32-bit application running on 64-bit CPU. As for the latter.
15 # It won't be able to achieve "native" 64-bit performance, because in
16 # 32-bit application context every addc instruction will have to be
17 # expanded as addc, twice right shift by 32 and finally adde, etc.
18 # So far RSA *sign* performance improvement over pre-bn_mul_mont asm
19 # for 64-bit application running on PPC970/G5 is:
28 if ($output =~ /32\-mont\.s/) {
36 $LDU= "lwzu"; # load and update
37 $LDX= "lwzx"; # load indexed
39 $STU= "stwu"; # store and update
40 $STX= "stwx"; # store indexed
41 $STUX= "stwux"; # store indexed and update
42 $UMULL= "mullw"; # unsigned multiply low
43 $UMULH= "mulhwu"; # unsigned multiply high
44 $UCMP= "cmplw"; # unsigned compare
47 } elsif ($output =~ /64\-mont\.s/) {
54 # same as above, but 64-bit mnemonics...
56 $LDU= "ldu"; # load and update
57 $LDX= "ldx"; # load indexed
59 $STU= "stdu"; # store and update
60 $STX= "stdx"; # store indexed
61 $STUX= "stdux"; # store indexed and update
62 $UMULL= "mulld"; # unsigned multiply low
63 $UMULH= "mulhdu"; # unsigned multiply high
64 $UCMP= "cmpld"; # unsigned compare
67 } else { die "nonsense $output"; }
69 ( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
70 die "can't call ../perlasm/ppc-xlate.pl: $!";
80 $rp="r9"; # $rp is reassigned
84 # non-volatile registers
108 mr $rp,r3 ; $rp is reassigned
112 slwi $num,$num,`log($BNSZ)/log(2)`
114 addi $ovf,$num,`$FRAME+$RZONE`
115 subf $ovf,$ovf,$sp ; $sp-$ovf
116 and $ovf,$ovf,$tj ; minimize TLB usage
117 subf $ovf,$sp,$ovf ; $ovf-$sp
118 srwi $num,$num,`log($BNSZ)/log(2)`
121 $PUSH r14,`4*$SIZE_T`($sp)
122 $PUSH r15,`5*$SIZE_T`($sp)
123 $PUSH r16,`6*$SIZE_T`($sp)
124 $PUSH r17,`7*$SIZE_T`($sp)
125 $PUSH r18,`8*$SIZE_T`($sp)
126 $PUSH r19,`9*$SIZE_T`($sp)
127 $PUSH r20,`10*$SIZE_T`($sp)
128 $PUSH r21,`11*$SIZE_T`($sp)
129 $PUSH r22,`12*$SIZE_T`($sp)
130 $PUSH r23,`13*$SIZE_T`($sp)
131 $PUSH r24,`14*$SIZE_T`($sp)
132 $PUSH r25,`15*$SIZE_T`($sp)
134 $LD $n0,0($n0) ; pull n0[0] value
135 addi $num,$num,-2 ; adjust $num for counter register
137 $LD $m0,0($bp) ; m0=bp[0]
138 $LD $aj,0($ap) ; ap[0]
140 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[0]
143 $LD $aj,$BNSZ($ap) ; ap[1]
144 $LD $nj,0($np) ; np[0]
146 $UMULL $m1,$lo0,$n0 ; "tp[0]"*n0
148 $UMULL $alo,$aj,$m0 ; ap[1]*bp[0]
151 $UMULL $lo1,$nj,$m1 ; np[0]*m1
153 $LD $nj,$BNSZ($np) ; np[1]
157 $UMULL $nlo,$nj,$m1 ; np[1]*m1
164 $LDX $aj,$ap,$j ; ap[j]
166 $LDX $nj,$np,$j ; np[j]
168 $UMULL $alo,$aj,$m0 ; ap[j]*bp[0]
172 $UMULL $nlo,$nj,$m1 ; np[j]*m1
173 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
176 $ST $lo1,0($tp) ; tp[j-1]
178 addi $j,$j,$BNSZ ; j++
179 addi $tp,$tp,$BNSZ ; tp++
187 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
189 $ST $lo1,0($tp) ; tp[j-1]
193 addze $ovf,$ovf ; upmost overflow bit
199 $LDX $m0,$bp,$i ; m0=bp[i]
200 $LD $aj,0($ap) ; ap[0]
202 $LD $tj,$FRAME($sp) ; tp[0]
203 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[i]
205 $LD $aj,$BNSZ($ap) ; ap[1]
206 $LD $nj,0($np) ; np[0]
207 addc $lo0,$lo0,$tj ; ap[0]*bp[i]+tp[0]
208 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
210 $UMULL $m1,$lo0,$n0 ; tp[0]*n0
212 $UMULL $lo1,$nj,$m1 ; np[0]*m1
214 $LD $nj,$BNSZ($np) ; np[1]
216 $UMULL $nlo,$nj,$m1 ; np[1]*m1
224 $LDX $aj,$ap,$j ; ap[j]
226 $LD $tj,$BNSZ($tp) ; tp[j]
228 $LDX $nj,$np,$j ; np[j]
230 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
233 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
234 $UMULL $nlo,$nj,$m1 ; np[j]*m1
237 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
238 addi $j,$j,$BNSZ ; j++
240 $ST $lo1,0($tp) ; tp[j-1]
241 addi $tp,$tp,$BNSZ ; tp++
244 $LD $tj,$BNSZ($tp) ; tp[j]
247 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
252 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
254 $ST $lo1,0($tp) ; tp[j-1]
256 addic $ovf,$ovf,-1 ; move upmost overflow to XER[CA]
262 slwi $tj,$num,`log($BNSZ)/log(2)`
267 addi $num,$num,2 ; restore $num
272 subfc. $ovf,$j,$ovf ; sets XER[CA]
280 $STX $j,$tp,$j ; zap at once
285 $POP r14,`4*$SIZE_T`($sp)
286 $POP r15,`5*$SIZE_T`($sp)
287 $POP r16,`6*$SIZE_T`($sp)
288 $POP r17,`7*$SIZE_T`($sp)
289 $POP r18,`8*$SIZE_T`($sp)
290 $POP r19,`9*$SIZE_T`($sp)
291 $POP r20,`10*$SIZE_T`($sp)
292 $POP r21,`11*$SIZE_T`($sp)
293 $POP r22,`12*$SIZE_T`($sp)
294 $POP r23,`13*$SIZE_T`($sp)
295 $POP r24,`14*$SIZE_T`($sp)
296 $POP r25,`15*$SIZE_T`($sp)
302 Lsub: $LDX $tj,$tp,$j
304 subfe $tj,$nj,$tj ; tp[j]-np[j]
319 $code =~ s/\`([^\`]*)\`/eval $1/gem;