summaryrefslogtreecommitdiff
path: root/main/openssl/crypto/bn/asm/armv4-gf2m.pl
blob: 3f1f4f67f59b67d272c11f54f0018ef1820520fa (plain)
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
#!/usr/bin/env perl
#
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> 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/.
# ====================================================================
#
# May 2011
#
# The module implements bn_GF2m_mul_2x2 polynomial multiplication
# used in bn_gf2m.c. It's kind of low-hanging mechanical port from
# C for the time being... Except that it has two code paths: pure
# integer code suitable for any ARMv4 and later CPU and NEON code
# suitable for ARMv7. Pure integer 1x1 multiplication subroutine runs
# in ~45 cycles on dual-issue core such as Cortex A8, which is ~50%
# faster than compiler-generated code. For ECDH and ECDSA verify (but
# not for ECDSA sign) it means 25%-45% improvement depending on key
# length, more for longer keys. Even though NEON 1x1 multiplication
# runs in even less cycles, ~30, improvement is measurable only on
# longer keys. One has to optimize code elsewhere to get NEON glow...
#
# April 2014
#
# Double bn_GF2m_mul_2x2 performance by using algorithm from paper
# referred below, which improves ECDH and ECDSA verify benchmarks
# by 18-40%.
#
# Câmara, D.; Gouvêa, C. P. L.; López, J. & Dahab, R.: Fast Software
# Polynomial Multiplication on ARM Processors using the NEON Engine.
# 
# http://conradoplg.cryptoland.net/files/2010/12/mocrysen13.pdf

while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";

$code=<<___;
#include "arm_arch.h"

.text
.code	32

#if __ARM_ARCH__>=7
.fpu	neon
#endif
___
################
# private interface to mul_1x1_ialu
#
$a="r1";
$b="r0";

($a0,$a1,$a2,$a12,$a4,$a14)=
($hi,$lo,$t0,$t1, $i0,$i1 )=map("r$_",(4..9),12);

$mask="r12";

$code.=<<___;
.type	mul_1x1_ialu,%function
.align	5
mul_1x1_ialu:
	mov	$a0,#0
	bic	$a1,$a,#3<<30		@ a1=a&0x3fffffff
	str	$a0,[sp,#0]		@ tab[0]=0
	add	$a2,$a1,$a1		@ a2=a1<<1
	str	$a1,[sp,#4]		@ tab[1]=a1
	eor	$a12,$a1,$a2		@ a1^a2
	str	$a2,[sp,#8]		@ tab[2]=a2
	mov	$a4,$a1,lsl#2		@ a4=a1<<2
	str	$a12,[sp,#12]		@ tab[3]=a1^a2
	eor	$a14,$a1,$a4		@ a1^a4
	str	$a4,[sp,#16]		@ tab[4]=a4
	eor	$a0,$a2,$a4		@ a2^a4
	str	$a14,[sp,#20]		@ tab[5]=a1^a4
	eor	$a12,$a12,$a4		@ a1^a2^a4
	str	$a0,[sp,#24]		@ tab[6]=a2^a4
	and	$i0,$mask,$b,lsl#2
	str	$a12,[sp,#28]		@ tab[7]=a1^a2^a4

	and	$i1,$mask,$b,lsr#1
	ldr	$lo,[sp,$i0]		@ tab[b       & 0x7]
	and	$i0,$mask,$b,lsr#4
	ldr	$t1,[sp,$i1]		@ tab[b >>  3 & 0x7]
	and	$i1,$mask,$b,lsr#7
	ldr	$t0,[sp,$i0]		@ tab[b >>  6 & 0x7]
	eor	$lo,$lo,$t1,lsl#3	@ stall
	mov	$hi,$t1,lsr#29
	ldr	$t1,[sp,$i1]		@ tab[b >>  9 & 0x7]

	and	$i0,$mask,$b,lsr#10
	eor	$lo,$lo,$t0,lsl#6
	eor	$hi,$hi,$t0,lsr#26
	ldr	$t0,[sp,$i0]		@ tab[b >> 12 & 0x7]

	and	$i1,$mask,$b,lsr#13
	eor	$lo,$lo,$t1,lsl#9
	eor	$hi,$hi,$t1,lsr#23
	ldr	$t1,[sp,$i1]		@ tab[b >> 15 & 0x7]

	and	$i0,$mask,$b,lsr#16
	eor	$lo,$lo,$t0,lsl#12
	eor	$hi,$hi,$t0,lsr#20
	ldr	$t0,[sp,$i0]		@ tab[b >> 18 & 0x7]

	and	$i1,$mask,$b,lsr#19
	eor	$lo,$lo,$t1,lsl#15
	eor	$hi,$hi,$t1,lsr#17
	ldr	$t1,[sp,$i1]		@ tab[b >> 21 & 0x7]

	and	$i0,$mask,$b,lsr#22
	eor	$lo,$lo,$t0,lsl#18
	eor	$hi,$hi,$t0,lsr#14
	ldr	$t0,[sp,$i0]		@ tab[b >> 24 & 0x7]

	and	$i1,$mask,$b,lsr#25
	eor	$lo,$lo,$t1,lsl#21
	eor	$hi,$hi,$t1,lsr#11
	ldr	$t1,[sp,$i1]		@ tab[b >> 27 & 0x7]

	tst	$a,#1<<30
	and	$i0,$mask,$b,lsr#28
	eor	$lo,$lo,$t0,lsl#24
	eor	$hi,$hi,$t0,lsr#8
	ldr	$t0,[sp,$i0]		@ tab[b >> 30      ]

	eorne	$lo,$lo,$b,lsl#30
	eorne	$hi,$hi,$b,lsr#2
	tst	$a,#1<<31
	eor	$lo,$lo,$t1,lsl#27
	eor	$hi,$hi,$t1,lsr#5
	eorne	$lo,$lo,$b,lsl#31
	eorne	$hi,$hi,$b,lsr#1
	eor	$lo,$lo,$t0,lsl#30
	eor	$hi,$hi,$t0,lsr#2

	mov	pc,lr
.size	mul_1x1_ialu,.-mul_1x1_ialu
___
################
# void	bn_GF2m_mul_2x2(BN_ULONG *r,
#	BN_ULONG a1,BN_ULONG a0,
#	BN_ULONG b1,BN_ULONG b0);	# r[3..0]=a1a0·b1b0
{
my ($r,$t0,$t1,$t2,$t3)=map("q$_",(0..3,8..12));
my ($a,$b,$k48,$k32,$k16)=map("d$_",(26..31));

$code.=<<___;
.global	bn_GF2m_mul_2x2
.type	bn_GF2m_mul_2x2,%function
.align	5
bn_GF2m_mul_2x2:
#if __ARM_ARCH__>=7
	ldr	r12,.LOPENSSL_armcap
.Lpic:	ldr	r12,[pc,r12]
	tst	r12,#1
	beq	.Lialu

	ldr		r12, [sp]		@ 5th argument
	vmov.32		$a, r2, r1
	vmov.32		$b, r12, r3
	vmov.i64	$k48, #0x0000ffffffffffff
	vmov.i64	$k32, #0x00000000ffffffff
	vmov.i64	$k16, #0x000000000000ffff

	vext.8		$t0#lo, $a, $a, #1	@ A1
	vmull.p8	$t0, $t0#lo, $b		@ F = A1*B
	vext.8		$r#lo, $b, $b, #1	@ B1
	vmull.p8	$r, $a, $r#lo		@ E = A*B1
	vext.8		$t1#lo, $a, $a, #2	@ A2
	vmull.p8	$t1, $t1#lo, $b		@ H = A2*B
	vext.8		$t3#lo, $b, $b, #2	@ B2
	vmull.p8	$t3, $a, $t3#lo		@ G = A*B2
	vext.8		$t2#lo, $a, $a, #3	@ A3
	veor		$t0, $t0, $r		@ L = E + F
	vmull.p8	$t2, $t2#lo, $b		@ J = A3*B
	vext.8		$r#lo, $b, $b, #3	@ B3
	veor		$t1, $t1, $t3		@ M = G + H
	vmull.p8	$r, $a, $r#lo		@ I = A*B3
	veor		$t0#lo, $t0#lo, $t0#hi	@ t0 = (L) (P0 + P1) << 8
	vand		$t0#hi, $t0#hi, $k48
	vext.8		$t3#lo, $b, $b, #4	@ B4
	veor		$t1#lo, $t1#lo, $t1#hi	@ t1 = (M) (P2 + P3) << 16
	vand		$t1#hi, $t1#hi, $k32
	vmull.p8	$t3, $a, $t3#lo		@ K = A*B4
	veor		$t2, $t2, $r		@ N = I + J
	veor		$t0#lo, $t0#lo, $t0#hi
	veor		$t1#lo, $t1#lo, $t1#hi
	veor		$t2#lo, $t2#lo, $t2#hi	@ t2 = (N) (P4 + P5) << 24
	vand		$t2#hi, $t2#hi, $k16
	vext.8		$t0, $t0, $t0, #15
	veor		$t3#lo, $t3#lo, $t3#hi	@ t3 = (K) (P6 + P7) << 32
	vmov.i64	$t3#hi, #0
	vext.8		$t1, $t1, $t1, #14
	veor		$t2#lo, $t2#lo, $t2#hi
	vmull.p8	$r, $a, $b		@ D = A*B
	vext.8		$t3, $t3, $t3, #12
	vext.8		$t2, $t2, $t2, #13
	veor		$t0, $t0, $t1
	veor		$t2, $t2, $t3
	veor		$r, $r, $t0
	veor		$r, $r, $t2

	vst1.32		{$r}, [r0]
	ret		@ bx lr
.align	4
.Lialu:
#endif
___
}
$ret="r10";	# reassigned 1st argument
$code.=<<___;
	stmdb	sp!,{r4-r10,lr}
	mov	$ret,r0			@ reassign 1st argument
	mov	$b,r3			@ $b=b1
	ldr	r3,[sp,#32]		@ load b0
	mov	$mask,#7<<2
	sub	sp,sp,#32		@ allocate tab[8]

	bl	mul_1x1_ialu		@ a1·b1
	str	$lo,[$ret,#8]
	str	$hi,[$ret,#12]

	eor	$b,$b,r3		@ flip b0 and b1
	 eor	$a,$a,r2		@ flip a0 and a1
	eor	r3,r3,$b
	 eor	r2,r2,$a
	eor	$b,$b,r3
	 eor	$a,$a,r2
	bl	mul_1x1_ialu		@ a0·b0
	str	$lo,[$ret]
	str	$hi,[$ret,#4]

	eor	$a,$a,r2
	eor	$b,$b,r3
	bl	mul_1x1_ialu		@ (a1+a0)·(b1+b0)
___
@r=map("r$_",(6..9));
$code.=<<___;
	ldmia	$ret,{@r[0]-@r[3]}
	eor	$lo,$lo,$hi
	eor	$hi,$hi,@r[1]
	eor	$lo,$lo,@r[0]
	eor	$hi,$hi,@r[2]
	eor	$lo,$lo,@r[3]
	eor	$hi,$hi,@r[3]
	str	$hi,[$ret,#8]
	eor	$lo,$lo,$hi
	add	sp,sp,#32		@ destroy tab[8]
	str	$lo,[$ret,#4]

#if __ARM_ARCH__>=5
	ldmia	sp!,{r4-r10,pc}
#else
	ldmia	sp!,{r4-r10,lr}
	tst	lr,#1
	moveq	pc,lr			@ be binary compatible with V4, yet
	bx	lr			@ interoperable with Thumb ISA:-)
#endif
.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
#if __ARM_ARCH__>=7
.align	5
.LOPENSSL_armcap:
.word	OPENSSL_armcap_P-(.Lpic+8)
#endif
.asciz	"GF(2^m) Multiplication for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
.align	5

.comm	OPENSSL_armcap_P,4,4
___

foreach (split("\n",$code)) {
	s/\`([^\`]*)\`/eval $1/geo;

	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo	or
	s/\bret\b/bx	lr/go		or
	s/\bbx\s+lr\b/.word\t0xe12fff1e/go;    # make it possible to compile with -march=armv4

	print $_,"\n";
}
close STDOUT;   # enforce flush