2 # Copyright 2006-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (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@fy.chalmers.se> 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 # "Teaser" Montgomery multiplication module for PowerPC. It's possible
20 # to gain a bit more by modulo-scheduling outer loop, then dedicated
21 # squaring procedure should give further 20% and code can be adapted
22 # for 32-bit application running on 64-bit CPU. As for the latter.
23 # It won't be able to achieve "native" 64-bit performance, because in
24 # 32-bit application context every addc instruction will have to be
25 # expanded as addc, twice right shift by 32 and finally adde, etc.
26 # So far RSA *sign* performance improvement over pre-bn_mul_mont asm
27 # for 64-bit application running on PPC970/G5 is:
36 if ($flavour =~ /32/) {
43 $LDU= "lwzu"; # load and update
44 $LDX= "lwzx"; # load indexed
46 $STU= "stwu"; # store and update
47 $STX= "stwx"; # store indexed
48 $STUX= "stwux"; # store indexed and update
49 $UMULL= "mullw"; # unsigned multiply low
50 $UMULH= "mulhwu"; # unsigned multiply high
51 $UCMP= "cmplw"; # unsigned compare
52 $SHRI= "srwi"; # unsigned shift right by immediate
55 } elsif ($flavour =~ /64/) {
61 # same as above, but 64-bit mnemonics...
63 $LDU= "ldu"; # load and update
64 $LDX= "ldx"; # load indexed
66 $STU= "stdu"; # store and update
67 $STX= "stdx"; # store indexed
68 $STUX= "stdux"; # store indexed and update
69 $UMULL= "mulld"; # unsigned multiply low
70 $UMULH= "mulhdu"; # unsigned multiply high
71 $UCMP= "cmpld"; # unsigned compare
72 $SHRI= "srdi"; # unsigned shift right by immediate
75 } else { die "nonsense $flavour"; }
77 $FRAME=8*$SIZE_T+$RZONE;
80 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
81 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
82 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
83 die "can't locate ppc-xlate.pl";
85 open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
95 $rp="r9"; # $rp is reassigned
99 # non-volatile registers
119 .globl .bn_mul_mont_int
123 mr $rp,r3 ; $rp is reassigned
127 $code.=<<___ if ($BNSZ==4);
128 cmpwi $num,32 ; longer key performance is not better
132 slwi $num,$num,`log($BNSZ)/log(2)`
134 addi $ovf,$num,$FRAME
135 subf $ovf,$ovf,$sp ; $sp-$ovf
136 and $ovf,$ovf,$tj ; minimize TLB usage
137 subf $ovf,$sp,$ovf ; $ovf-$sp
139 srwi $num,$num,`log($BNSZ)/log(2)`
142 $PUSH r20,`-12*$SIZE_T`($tj)
143 $PUSH r21,`-11*$SIZE_T`($tj)
144 $PUSH r22,`-10*$SIZE_T`($tj)
145 $PUSH r23,`-9*$SIZE_T`($tj)
146 $PUSH r24,`-8*$SIZE_T`($tj)
147 $PUSH r25,`-7*$SIZE_T`($tj)
148 $PUSH r26,`-6*$SIZE_T`($tj)
149 $PUSH r27,`-5*$SIZE_T`($tj)
150 $PUSH r28,`-4*$SIZE_T`($tj)
151 $PUSH r29,`-3*$SIZE_T`($tj)
152 $PUSH r30,`-2*$SIZE_T`($tj)
153 $PUSH r31,`-1*$SIZE_T`($tj)
155 $LD $n0,0($n0) ; pull n0[0] value
156 addi $num,$num,-2 ; adjust $num for counter register
158 $LD $m0,0($bp) ; m0=bp[0]
159 $LD $aj,0($ap) ; ap[0]
161 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[0]
164 $LD $aj,$BNSZ($ap) ; ap[1]
165 $LD $nj,0($np) ; np[0]
167 $UMULL $m1,$lo0,$n0 ; "tp[0]"*n0
169 $UMULL $alo,$aj,$m0 ; ap[1]*bp[0]
172 $UMULL $lo1,$nj,$m1 ; np[0]*m1
174 $LD $nj,$BNSZ($np) ; np[1]
178 $UMULL $nlo,$nj,$m1 ; np[1]*m1
185 $LDX $aj,$ap,$j ; ap[j]
187 $LDX $nj,$np,$j ; np[j]
189 $UMULL $alo,$aj,$m0 ; ap[j]*bp[0]
193 $UMULL $nlo,$nj,$m1 ; np[j]*m1
194 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
197 $ST $lo1,0($tp) ; tp[j-1]
199 addi $j,$j,$BNSZ ; j++
200 addi $tp,$tp,$BNSZ ; tp++
208 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[0]
210 $ST $lo1,0($tp) ; tp[j-1]
214 addze $ovf,$ovf ; upmost overflow bit
220 $LDX $m0,$bp,$i ; m0=bp[i]
221 $LD $aj,0($ap) ; ap[0]
223 $LD $tj,$LOCALS($sp); tp[0]
224 $UMULL $lo0,$aj,$m0 ; ap[0]*bp[i]
226 $LD $aj,$BNSZ($ap) ; ap[1]
227 $LD $nj,0($np) ; np[0]
228 addc $lo0,$lo0,$tj ; ap[0]*bp[i]+tp[0]
229 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
231 $UMULL $m1,$lo0,$n0 ; tp[0]*n0
233 $UMULL $lo1,$nj,$m1 ; np[0]*m1
235 $LD $nj,$BNSZ($np) ; np[1]
237 $UMULL $nlo,$nj,$m1 ; np[1]*m1
245 $LDX $aj,$ap,$j ; ap[j]
247 $LD $tj,$BNSZ($tp) ; tp[j]
249 $LDX $nj,$np,$j ; np[j]
251 $UMULL $alo,$aj,$m0 ; ap[j]*bp[i]
254 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
255 $UMULL $nlo,$nj,$m1 ; np[j]*m1
258 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
259 addi $j,$j,$BNSZ ; j++
261 $ST $lo1,0($tp) ; tp[j-1]
262 addi $tp,$tp,$BNSZ ; tp++
265 $LD $tj,$BNSZ($tp) ; tp[j]
268 addc $lo0,$lo0,$tj ; ap[j]*bp[i]+tp[j]
273 addc $lo1,$lo1,$lo0 ; np[j]*m1+ap[j]*bp[i]+tp[j]
275 $ST $lo1,0($tp) ; tp[j-1]
277 addic $ovf,$ovf,-1 ; move upmost overflow to XER[CA]
283 slwi $tj,$num,`log($BNSZ)/log(2)`
288 addi $num,$num,2 ; restore $num
289 subfc $j,$j,$j ; j=0 and "clear" XER[CA]
294 Lsub: $LDX $tj,$tp,$j
296 subfe $aj,$nj,$tj ; tp[j]-np[j]
303 subfe $ovf,$j,$ovf ; handle upmost overflow bit
306 or $ap,$ap,$np ; ap=borrow?tp:rp
309 Lcopy: ; copy or in-place refresh
312 $STX $j,$tp,$j ; zap at once
318 $POP r20,`-12*$SIZE_T`($tj)
319 $POP r21,`-11*$SIZE_T`($tj)
320 $POP r22,`-10*$SIZE_T`($tj)
321 $POP r23,`-9*$SIZE_T`($tj)
322 $POP r24,`-8*$SIZE_T`($tj)
323 $POP r25,`-7*$SIZE_T`($tj)
324 $POP r26,`-6*$SIZE_T`($tj)
325 $POP r27,`-5*$SIZE_T`($tj)
326 $POP r28,`-4*$SIZE_T`($tj)
327 $POP r29,`-3*$SIZE_T`($tj)
328 $POP r30,`-2*$SIZE_T`($tj)
329 $POP r31,`-1*$SIZE_T`($tj)
333 .byte 0,12,4,0,0x80,12,6,0
335 .size .bn_mul_mont_int,.-.bn_mul_mont_int
337 .asciz "Montgomery Multiplication for PPC, CRYPTOGAMS by <appro\@openssl.org>"
340 $code =~ s/\`([^\`]*)\`/eval $1/gem;