aboutsummaryrefslogtreecommitdiffstats
path: root/crypto/bn/asm/ppc-mont.pl
blob: da69c6aaaf6a55b07114333a008c39571f153721 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
#!/usr/bin/env perl

# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================

# April 2006

# "Teaser" Montgomery multiplication module for PowerPC. It's possible
# to gain a bit more by modulo-scheduling outer loop, then dedicated
# squaring procedure should give further 20% and code can be adapted
# for 32-bit application running on 64-bit CPU. As for the latter.
# It won't be able to achieve "native" 64-bit performance, because in
# 32-bit application context every addc instruction will have to be
# expanded as addc, twice right shift by 32 and finally adde, etc.
# So far RSA *sign* performance improvement over pre-bn_mul_mont asm
# for 64-bit application running on PPC970/G5 is:
#
# 512-bit	+65%	
# 1024-bit	+35%
# 2048-bit	+18%
# 4096-bit	+4%

$flavour = shift;

if ($flavour =~ /32/) {
	$BITS=	32;
	$BNSZ=	$BITS/8;
	$SIZE_T=4;
	$RZONE=	224;

	$LD=	"lwz";		# load
	$LDU=	"lwzu";		# load and update
	$LDX=	"lwzx";		# load indexed
	$ST=	"stw";		# store
	$STU=	"stwu";		# store and update
	$STX=	"stwx";		# store indexed
	$STUX=	"stwux";	# store indexed and update
	$UMULL=	"mullw";	# unsigned multiply low
	$UMULH=	"mulhwu";	# unsigned multiply high
	$UCMP=	"cmplw";	# unsigned compare
	$SHRI=	"srwi";		# unsigned shift right by immediate	
	$PUSH=	$ST;
	$POP=	$LD;
} elsif ($flavour =~ /64/) {
	$BITS=	64;
	$BNSZ=	$BITS/8;
	$SIZE_T=8;
	$RZONE=	288;

	# same as above, but 64-bit mnemonics...
	$LD=	"ld";		# load
	$LDU=	"ldu";		# load and update
	$LDX=	"ldx";		# load indexed
	$ST=	"std";		# store
	$STU=	"stdu";		# store and update
	$STX=	"stdx";		# store indexed
	$STUX=	"stdux";	# store indexed and update
	$UMULL=	"mulld";	# unsigned multiply low
	$UMULH=	"mulhdu";	# unsigned multiply high
	$UCMP=	"cmpld";	# unsigned compare
	$SHRI=	"srdi";		# unsigned shift right by immediate	
	$PUSH=	$ST;
	$POP=	$LD;
} else { die "nonsense $flavour"; }

$FRAME=8*$SIZE_T+$RZONE;
$LOCALS=8*$SIZE_T;

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
die "can't locate ppc-xlate.pl";

open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";

$sp="r1";
$toc="r2";
$rp="r3";	$ovf="r3";
$ap="r4";
$bp="r5";
$np="r6";
$n0="r7";
$num="r8";
$rp="r9";	# $rp is reassigned
$aj="r10";
$nj="r11";
$tj="r12";
# non-volatile registers
$i="r20";
$j="r21";
$tp="r22";
$m0="r23";
$m1="r24";
$lo0="r25";
$hi0="r26";
$lo1="r27";
$hi1="r28";
$alo="r29";
$ahi="r30";
$nlo="r31";
#
$nhi="r0";

$code=<<___;
.machine "any"
.text

.globl	.bn_mul_mont_int
.align	4
.bn_mul_mont_int:
	cmpwi	$num,4
	mr	$rp,r3		; $rp is reassigned
	li	r3,0
	bltlr
___
$code.=<<___ if ($BNSZ==4);
	cmpwi	$num,32		; longer key performance is not better
	bgelr
___
$code.=<<___;
	slwi	$num,$num,`log($BNSZ)/log(2)`
	li	$tj,-4096
	addi	$ovf,$num,$FRAME
	subf	$ovf,$ovf,$sp	; $sp-$ovf
	and	$ovf,$ovf,$tj	; minimize TLB usage
	subf	$ovf,$sp,$ovf	; $ovf-$sp
	mr	$tj,$sp
	srwi	$num,$num,`log($BNSZ)/log(2)`
	$STUX	$sp,$sp,$ovf

	$PUSH	r20,`-12*$SIZE_T`($tj)
	$PUSH	r21,`-11*$SIZE_T`($tj)
	$PUSH	r22,`-10*$SIZE_T`($tj)
	$PUSH	r23,`-9*$SIZE_T`($tj)
	$PUSH	r24,`-8*$SIZE_T`($tj)
	$PUSH	r25,`-7*$SIZE_T`($tj)
	$PUSH	r26,`-6*$SIZE_T`($tj)
	$PUSH	r27,`-5*$SIZE_T`($tj)
	$PUSH	r28,`-4*$SIZE_T`($tj)
	$PUSH	r29,`-3*$SIZE_T`($tj)
	$PUSH	r30,`-2*$SIZE_T`($tj)
	$PUSH	r31,`-1*$SIZE_T`($tj)

	$LD	$n0,0($n0)	; pull n0[0] value
	addi	$num,$num,-2	; adjust $num for counter register

	$LD	$m0,0($bp)	; m0=bp[0]
	$LD	$aj,0($ap)	; ap[0]
	addi	$tp,$sp,$LOCALS
	$UMULL	$lo0,$aj,$m0	; ap[0]*bp[0]
	$UMULH	$hi0,$aj,$m0

	$LD	$aj,$BNSZ($ap)	; ap[1]
	$LD	$nj,0($np)	; np[0]

	$UMULL	$m1,$lo0,$n0	; "tp[0]"*n0

	$UMULL	$alo,$aj,$m0	; ap[1]*bp[0]
	$UMULH	$ahi,$aj,$m0

	$UMULL	$lo1,$nj,$m1	; np[0]*m1
	$UMULH	$hi1,$nj,$m1
	$LD	$nj,$BNSZ($np)	; np[1]
	addc	$lo1,$lo1,$lo0
	addze	$hi1,$hi1

	$UMULL	$nlo,$nj,$m1	; np[1]*m1
	$UMULH	$nhi,$nj,$m1

	mtctr	$num
	li	$j,`2*$BNSZ`
.align	4
L1st:
	$LDX	$aj,$ap,$j	; ap[j]
	addc	$lo0,$alo,$hi0
	$LDX	$nj,$np,$j	; np[j]
	addze	$hi0,$ahi
	$UMULL	$alo,$aj,$m0	; ap[j]*bp[0]
	addc	$lo1,$nlo,$hi1
	$UMULH	$ahi,$aj,$m0
	addze	$hi1,$nhi
	$UMULL	$nlo,$nj,$m1	; np[j]*m1
	addc	$lo1,$lo1,$lo0	; np[j]*m1+ap[j]*bp[0]
	$UMULH	$nhi,$nj,$m1
	addze	$hi1,$hi1
	$ST	$lo1,0($tp)	; tp[j-1]

	addi	$j,$j,$BNSZ	; j++
	addi	$tp,$tp,$BNSZ	; tp++
	bdnz-	L1st
;L1st
	addc	$lo0,$alo,$hi0
	addze	$hi0,$ahi

	addc	$lo1,$nlo,$hi1
	addze	$hi1,$nhi
	addc	$lo1,$lo1,$lo0	; np[j]*m1+ap[j]*bp[0]
	addze	$hi1,$hi1
	$ST	$lo1,0($tp)	; tp[j-1]

	li	$ovf,0
	addc	$hi1,$hi1,$hi0
	addze	$ovf,$ovf	; upmost overflow bit
	$ST	$hi1,$BNSZ($tp)

	li	$i,$BNSZ
.align	4
Louter:
	$LDX	$m0,$bp,$i	; m0=bp[i]
	$LD	$aj,0($ap)	; ap[0]
	addi	$tp,$sp,$LOCALS
	$LD	$tj,$LOCALS($sp); tp[0]
	$UMULL	$lo0,$aj,$m0	; ap[0]*bp[i]
	$UMULH	$hi0,$aj,$m0
	$LD	$aj,$BNSZ($ap)	; ap[1]
	$LD	$nj,0($np)	; np[0]
	addc	$lo0,$lo0,$tj	; ap[0]*bp[i]+tp[0]
	$UMULL	$alo,$aj,$m0	; ap[j]*bp[i]
	addze	$hi0,$hi0
	$UMULL	$m1,$lo0,$n0	; tp[0]*n0
	$UMULH	$ahi,$aj,$m0
	$UMULL	$lo1,$nj,$m1	; np[0]*m1
	$UMULH	$hi1,$nj,$m1
	$LD	$nj,$BNSZ($np)	; np[1]
	addc	$lo1,$lo1,$lo0
	$UMULL	$nlo,$nj,$m1	; np[1]*m1
	addze	$hi1,$hi1
	$UMULH	$nhi,$nj,$m1

	mtctr	$num
	li	$j,`2*$BNSZ`
.align	4
Linner:
	$LDX	$aj,$ap,$j	; ap[j]
	addc	$lo0,$alo,$hi0
	$LD	$tj,$BNSZ($tp)	; tp[j]
	addze	$hi0,$ahi
	$LDX	$nj,$np,$j	; np[j]
	addc	$lo1,$nlo,$hi1
	$UMULL	$alo,$aj,$m0	; ap[j]*bp[i]
	addze	$hi1,$nhi
	$UMULH	$ahi,$aj,$m0
	addc	$lo0,$lo0,$tj	; ap[j]*bp[i]+tp[j]
	$UMULL	$nlo,$nj,$m1	; np[j]*m1
	addze	$hi0,$hi0
	$UMULH	$nhi,$nj,$m1
	addc	$lo1,$lo1,$lo0	; np[j]*m1+ap[j]*bp[i]+tp[j]
	addi	$j,$j,$BNSZ	; j++
	addze	$hi1,$hi1
	$ST	$lo1,0($tp)	; tp[j-1]
	addi	$tp,$tp,$BNSZ	; tp++
	bdnz-	Linner
;Linner
	$LD	$tj,$BNSZ($tp)	; tp[j]
	addc	$lo0,$alo,$hi0
	addze	$hi0,$ahi
	addc	$lo0,$lo0,$tj	; ap[j]*bp[i]+tp[j]
	addze	$hi0,$hi0

	addc	$lo1,$nlo,$hi1
	addze	$hi1,$nhi
	addc	$lo1,$lo1,$lo0	; np[j]*m1+ap[j]*bp[i]+tp[j]
	addze	$hi1,$hi1
	$ST	$lo1,0($tp)	; tp[j-1]

	addic	$ovf,$ovf,-1	; move upmost overflow to XER[CA]
	li	$ovf,0
	adde	$hi1,$hi1,$hi0
	addze	$ovf,$ovf
	$ST	$hi1,$BNSZ($tp)
;
	slwi	$tj,$num,`log($BNSZ)/log(2)`
	$UCMP	$i,$tj
	addi	$i,$i,$BNSZ
	ble-	Louter

	addi	$num,$num,2	; restore $num
	subfc	$j,$j,$j	; j=0 and "clear" XER[CA]
	addi	$tp,$sp,$LOCALS
	mtctr	$num

.align	4
Lsub:	$LDX	$tj,$tp,$j
	$LDX	$nj,$np,$j
	subfe	$aj,$nj,$tj	; tp[j]-np[j]
	$STX	$aj,$rp,$j
	addi	$j,$j,$BNSZ
	bdnz-	Lsub

	li	$j,0
	mtctr	$num
	subfe	$ovf,$j,$ovf	; handle upmost overflow bit
	and	$ap,$tp,$ovf
	andc	$np,$rp,$ovf
	or	$ap,$ap,$np	; ap=borrow?tp:rp

.align	4
Lcopy:				; copy or in-place refresh
	$LDX	$tj,$ap,$j
	$STX	$tj,$rp,$j
	$STX	$j,$tp,$j	; zap at once
	addi	$j,$j,$BNSZ
	bdnz-	Lcopy

	$POP	$tj,0($sp)
	li	r3,1
	$POP	r20,`-12*$SIZE_T`($tj)
	$POP	r21,`-11*$SIZE_T`($tj)
	$POP	r22,`-10*$SIZE_T`($tj)
	$POP	r23,`-9*$SIZE_T`($tj)
	$POP	r24,`-8*$SIZE_T`($tj)
	$POP	r25,`-7*$SIZE_T`($tj)
	$POP	r26,`-6*$SIZE_T`($tj)
	$POP	r27,`-5*$SIZE_T`($tj)
	$POP	r28,`-4*$SIZE_T`($tj)
	$POP	r29,`-3*$SIZE_T`($tj)
	$POP	r30,`-2*$SIZE_T`($tj)
	$POP	r31,`-1*$SIZE_T`($tj)
	mr	$sp,$tj
	blr
	.long	0
	.byte	0,12,4,0,0x80,12,6,0
	.long	0
.size	.bn_mul_mont_int,.-.bn_mul_mont_int

.asciz  "Montgomery Multiplication for PPC, CRYPTOGAMS by <appro\@openssl.org>"
___

$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;