aboutsummaryrefslogtreecommitdiff
path: root/openssl/crypto/sha/asm/sha1-armv4-large.pl
blob: 88861af64116195bb074691243d6bdcac1354ad8 (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
#!/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/.
# ====================================================================

# sha1_block procedure for ARMv4.
#
# January 2007.

# Size/performance trade-off
# ====================================================================
# impl		size in bytes	comp cycles[*]	measured performance
# ====================================================================
# thumb		304		3212		4420
# armv4-small	392/+29%	1958/+64%	2250/+96%
# armv4-compact	740/+89%	1552/+26%	1840/+22%
# armv4-large	1420/+92%	1307/+19%	1370/+34%[***]
# full unroll	~5100/+260%	~1260/+4%	~1300/+5%
# ====================================================================
# thumb		= same as 'small' but in Thumb instructions[**] and
#		  with recurring code in two private functions;
# small		= detached Xload/update, loops are folded;
# compact	= detached Xload/update, 5x unroll;
# large		= interleaved Xload/update, 5x unroll;
# full unroll	= interleaved Xload/update, full unroll, estimated[!];
#
# [*]	Manually counted instructions in "grand" loop body. Measured
#	performance is affected by prologue and epilogue overhead,
#	i-cache availability, branch penalties, etc.
# [**]	While each Thumb instruction is twice smaller, they are not as
#	diverse as ARM ones: e.g., there are only two arithmetic
#	instructions with 3 arguments, no [fixed] rotate, addressing
#	modes are limited. As result it takes more instructions to do
#	the same job in Thumb, therefore the code is never twice as
#	small and always slower.
# [***]	which is also ~35% better than compiler generated code.

$output=shift;
open STDOUT,">$output";

$ctx="r0";
$inp="r1";
$len="r2";
$a="r3";
$b="r4";
$c="r5";
$d="r6";
$e="r7";
$K="r8";
$t0="r9";
$t1="r10";
$t2="r11";
$t3="r12";
$Xi="r14";
@V=($a,$b,$c,$d,$e);

# One can optimize this for aligned access on big-endian architecture,
# but code's endian neutrality makes it too pretty:-)
sub Xload {
my ($a,$b,$c,$d,$e)=@_;
$code.=<<___;
	ldrb	$t0,[$inp],#4
	ldrb	$t1,[$inp,#-3]
	ldrb	$t2,[$inp,#-2]
	ldrb	$t3,[$inp,#-1]
	add	$e,$K,$e,ror#2			@ E+=K_00_19
	orr	$t0,$t1,$t0,lsl#8
	add	$e,$e,$a,ror#27			@ E+=ROR(A,27)
	orr	$t0,$t2,$t0,lsl#8
	eor	$t1,$c,$d			@ F_xx_xx
	orr	$t0,$t3,$t0,lsl#8
	add	$e,$e,$t0			@ E+=X[i]
	str	$t0,[$Xi,#-4]!
___
}
sub Xupdate {
my ($a,$b,$c,$d,$e,$flag)=@_;
$code.=<<___;
	ldr	$t0,[$Xi,#15*4]
	ldr	$t1,[$Xi,#13*4]
	ldr	$t2,[$Xi,#7*4]
	ldr	$t3,[$Xi,#2*4]
	add	$e,$K,$e,ror#2			@ E+=K_xx_xx
	eor	$t0,$t0,$t1
	eor	$t0,$t0,$t2
	eor	$t0,$t0,$t3
	add	$e,$e,$a,ror#27			@ E+=ROR(A,27)
___
$code.=<<___ if (!defined($flag));
	eor	$t1,$c,$d			@ F_xx_xx, but not in 40_59
___
$code.=<<___;
	mov	$t0,$t0,ror#31
	add	$e,$e,$t0			@ E+=X[i]
	str	$t0,[$Xi,#-4]!
___
}

sub BODY_00_15 {
my ($a,$b,$c,$d,$e)=@_;
	&Xload(@_);
$code.=<<___;
	and	$t1,$b,$t1,ror#2
	eor	$t1,$t1,$d,ror#2		@ F_00_19(B,C,D)
	add	$e,$e,$t1			@ E+=F_00_19(B,C,D)
___
}

sub BODY_16_19 {
my ($a,$b,$c,$d,$e)=@_;
	&Xupdate(@_);
$code.=<<___;
	and	$t1,$b,$t1,ror#2
	eor	$t1,$t1,$d,ror#2		@ F_00_19(B,C,D)
	add	$e,$e,$t1			@ E+=F_00_19(B,C,D)
___
}

sub BODY_20_39 {
my ($a,$b,$c,$d,$e)=@_;
	&Xupdate(@_);
$code.=<<___;
	eor	$t1,$b,$t1,ror#2		@ F_20_39(B,C,D)
	add	$e,$e,$t1			@ E+=F_20_39(B,C,D)
___
}

sub BODY_40_59 {
my ($a,$b,$c,$d,$e)=@_;
	&Xupdate(@_,1);
$code.=<<___;
	and	$t1,$b,$c,ror#2
	orr	$t2,$b,$c,ror#2
	and	$t2,$t2,$d,ror#2
	orr	$t1,$t1,$t2			@ F_40_59(B,C,D)
	add	$e,$e,$t1			@ E+=F_40_59(B,C,D)
___
}

$code=<<___;
.text

.global	sha1_block_data_order
.type	sha1_block_data_order,%function

.align	2
sha1_block_data_order:
	stmdb	sp!,{r4-r12,lr}
	add	$len,$inp,$len,lsl#6	@ $len to point at the end of $inp
	ldmia	$ctx,{$a,$b,$c,$d,$e}
.Lloop:
	ldr	$K,.LK_00_19
	mov	$Xi,sp
	sub	sp,sp,#15*4
	mov	$c,$c,ror#30
	mov	$d,$d,ror#30
	mov	$e,$e,ror#30		@ [6]
.L_00_15:
___
for($i=0;$i<5;$i++) {
	&BODY_00_15(@V);	unshift(@V,pop(@V));
}
$code.=<<___;
	teq	$Xi,sp
	bne	.L_00_15		@ [((11+4)*5+2)*3]
___
	&BODY_00_15(@V);	unshift(@V,pop(@V));
	&BODY_16_19(@V);	unshift(@V,pop(@V));
	&BODY_16_19(@V);	unshift(@V,pop(@V));
	&BODY_16_19(@V);	unshift(@V,pop(@V));
	&BODY_16_19(@V);	unshift(@V,pop(@V));
$code.=<<___;

	ldr	$K,.LK_20_39		@ [+15+16*4]
	sub	sp,sp,#25*4
	cmn	sp,#0			@ [+3], clear carry to denote 20_39
.L_20_39_or_60_79:
___
for($i=0;$i<5;$i++) {
	&BODY_20_39(@V);	unshift(@V,pop(@V));
}
$code.=<<___;
	teq	$Xi,sp			@ preserve carry
	bne	.L_20_39_or_60_79	@ [+((12+3)*5+2)*4]
	bcs	.L_done			@ [+((12+3)*5+2)*4], spare 300 bytes

	ldr	$K,.LK_40_59
	sub	sp,sp,#20*4		@ [+2]
.L_40_59:
___
for($i=0;$i<5;$i++) {
	&BODY_40_59(@V);	unshift(@V,pop(@V));
}
$code.=<<___;
	teq	$Xi,sp
	bne	.L_40_59		@ [+((12+5)*5+2)*4]

	ldr	$K,.LK_60_79
	sub	sp,sp,#20*4
	cmp	sp,#0			@ set carry to denote 60_79
	b	.L_20_39_or_60_79	@ [+4], spare 300 bytes
.L_done:
	add	sp,sp,#80*4		@ "deallocate" stack frame
	ldmia	$ctx,{$K,$t0,$t1,$t2,$t3}
	add	$a,$K,$a
	add	$b,$t0,$b
	add	$c,$t1,$c,ror#2
	add	$d,$t2,$d,ror#2
	add	$e,$t3,$e,ror#2
	stmia	$ctx,{$a,$b,$c,$d,$e}
	teq	$inp,$len
	bne	.Lloop			@ [+18], total 1307

	ldmia	sp!,{r4-r12,lr}
	tst	lr,#1
	moveq	pc,lr			@ be binary compatible with V4, yet
	bx	lr			@ interoperable with Thumb ISA:-)
.align	2
.LK_00_19:	.word	0x5a827999
.LK_20_39:	.word	0x6ed9eba1
.LK_40_59:	.word	0x8f1bbcdc
.LK_60_79:	.word	0xca62c1d6
.size	sha1_block_data_order,.-sha1_block_data_order
.asciz	"SHA1 block transform for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
.align	2
___

$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
print $code;
close STDOUT; # enforce flush