tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
ARC: String library
Hand optimised asm code for ARC700 pipeline.
Originally written/optimized by Joern Rennecke
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Cc: Joern Rennecke <joern.rennecke@embecosm.com>
+661
+40
arch/arc/include/asm/string.h
+40
arch/arc/include/asm/string.h
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*
8
+
* vineetg: May 2011
9
+
* -We had half-optimised memset/memcpy, got better versions of those
10
+
* -Added memcmp, strchr, strcpy, strcmp, strlen
11
+
*
12
+
* Amit Bhor: Codito Technologies 2004
13
+
*/
14
+
15
+
#ifndef _ASM_ARC_STRING_H
16
+
#define _ASM_ARC_STRING_H
17
+
18
+
#include <linux/types.h>
19
+
20
+
#ifdef __KERNEL__
21
+
22
+
#define __HAVE_ARCH_MEMSET
23
+
#define __HAVE_ARCH_MEMCPY
24
+
#define __HAVE_ARCH_MEMCMP
25
+
#define __HAVE_ARCH_STRCHR
26
+
#define __HAVE_ARCH_STRCPY
27
+
#define __HAVE_ARCH_STRCMP
28
+
#define __HAVE_ARCH_STRLEN
29
+
30
+
extern void *memset(void *ptr, int, __kernel_size_t);
31
+
extern void *memcpy(void *, const void *, __kernel_size_t);
32
+
extern void memzero(void *ptr, __kernel_size_t n);
33
+
extern int memcmp(const void *, const void *, __kernel_size_t);
34
+
extern char *strchr(const char *s, int c);
35
+
extern char *strcpy(char *dest, const char *src);
36
+
extern int strcmp(const char *cs, const char *ct);
37
+
extern __kernel_size_t strlen(const char *);
38
+
39
+
#endif /* __KERNEL__ */
40
+
#endif /* _ASM_ARC_STRING_H */
+124
arch/arc/lib/memcmp.S
+124
arch/arc/lib/memcmp.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
#include <asm/linkage.h>
10
+
11
+
#ifdef __LITTLE_ENDIAN__
12
+
#define WORD2 r2
13
+
#define SHIFT r3
14
+
#else /* BIG ENDIAN */
15
+
#define WORD2 r3
16
+
#define SHIFT r2
17
+
#endif
18
+
19
+
ARC_ENTRY memcmp
20
+
or r12,r0,r1
21
+
asl_s r12,r12,30
22
+
sub r3,r2,1
23
+
brls r2,r12,.Lbytewise
24
+
ld r4,[r0,0]
25
+
ld r5,[r1,0]
26
+
lsr.f lp_count,r3,3
27
+
lpne .Loop_end
28
+
ld_s WORD2,[r0,4]
29
+
ld_s r12,[r1,4]
30
+
brne r4,r5,.Leven
31
+
ld.a r4,[r0,8]
32
+
ld.a r5,[r1,8]
33
+
brne WORD2,r12,.Lodd
34
+
.Loop_end:
35
+
asl_s SHIFT,SHIFT,3
36
+
bhs_s .Last_cmp
37
+
brne r4,r5,.Leven
38
+
ld r4,[r0,4]
39
+
ld r5,[r1,4]
40
+
#ifdef __LITTLE_ENDIAN__
41
+
nop_s
42
+
; one more load latency cycle
43
+
.Last_cmp:
44
+
xor r0,r4,r5
45
+
bset r0,r0,SHIFT
46
+
sub_s r1,r0,1
47
+
bic_s r1,r1,r0
48
+
norm r1,r1
49
+
b.d .Leven_cmp
50
+
and r1,r1,24
51
+
.Leven:
52
+
xor r0,r4,r5
53
+
sub_s r1,r0,1
54
+
bic_s r1,r1,r0
55
+
norm r1,r1
56
+
; slow track insn
57
+
and r1,r1,24
58
+
.Leven_cmp:
59
+
asl r2,r4,r1
60
+
asl r12,r5,r1
61
+
lsr_s r2,r2,1
62
+
lsr_s r12,r12,1
63
+
j_s.d [blink]
64
+
sub r0,r2,r12
65
+
.balign 4
66
+
.Lodd:
67
+
xor r0,WORD2,r12
68
+
sub_s r1,r0,1
69
+
bic_s r1,r1,r0
70
+
norm r1,r1
71
+
; slow track insn
72
+
and r1,r1,24
73
+
asl_s r2,r2,r1
74
+
asl_s r12,r12,r1
75
+
lsr_s r2,r2,1
76
+
lsr_s r12,r12,1
77
+
j_s.d [blink]
78
+
sub r0,r2,r12
79
+
#else /* BIG ENDIAN */
80
+
.Last_cmp:
81
+
neg_s SHIFT,SHIFT
82
+
lsr r4,r4,SHIFT
83
+
lsr r5,r5,SHIFT
84
+
; slow track insn
85
+
.Leven:
86
+
sub.f r0,r4,r5
87
+
mov.ne r0,1
88
+
j_s.d [blink]
89
+
bset.cs r0,r0,31
90
+
.Lodd:
91
+
cmp_s WORD2,r12
92
+
93
+
mov_s r0,1
94
+
j_s.d [blink]
95
+
bset.cs r0,r0,31
96
+
#endif /* ENDIAN */
97
+
.balign 4
98
+
.Lbytewise:
99
+
breq r2,0,.Lnil
100
+
ldb r4,[r0,0]
101
+
ldb r5,[r1,0]
102
+
lsr.f lp_count,r3
103
+
lpne .Lbyte_end
104
+
ldb_s r3,[r0,1]
105
+
ldb r12,[r1,1]
106
+
brne r4,r5,.Lbyte_even
107
+
ldb.a r4,[r0,2]
108
+
ldb.a r5,[r1,2]
109
+
brne r3,r12,.Lbyte_odd
110
+
.Lbyte_end:
111
+
bcc .Lbyte_even
112
+
brne r4,r5,.Lbyte_even
113
+
ldb_s r3,[r0,1]
114
+
ldb_s r12,[r1,1]
115
+
.Lbyte_odd:
116
+
j_s.d [blink]
117
+
sub r0,r3,r12
118
+
.Lbyte_even:
119
+
j_s.d [blink]
120
+
sub r0,r4,r5
121
+
.Lnil:
122
+
j_s.d [blink]
123
+
mov r0,0
124
+
ARC_EXIT memcmp
+66
arch/arc/lib/memcpy-700.S
+66
arch/arc/lib/memcpy-700.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
#include <asm/linkage.h>
10
+
11
+
ARC_ENTRY memcpy
12
+
or r3,r0,r1
13
+
asl_s r3,r3,30
14
+
mov_s r5,r0
15
+
brls.d r2,r3,.Lcopy_bytewise
16
+
sub.f r3,r2,1
17
+
ld_s r12,[r1,0]
18
+
asr.f lp_count,r3,3
19
+
bbit0.d r3,2,.Lnox4
20
+
bmsk_s r2,r2,1
21
+
st.ab r12,[r5,4]
22
+
ld.a r12,[r1,4]
23
+
.Lnox4:
24
+
lppnz .Lendloop
25
+
ld_s r3,[r1,4]
26
+
st.ab r12,[r5,4]
27
+
ld.a r12,[r1,8]
28
+
st.ab r3,[r5,4]
29
+
.Lendloop:
30
+
breq r2,0,.Last_store
31
+
ld r3,[r5,0]
32
+
#ifdef __LITTLE_ENDIAN__
33
+
add3 r2,-1,r2
34
+
; uses long immediate
35
+
xor_s r12,r12,r3
36
+
bmsk r12,r12,r2
37
+
xor_s r12,r12,r3
38
+
#else /* BIG ENDIAN */
39
+
sub3 r2,31,r2
40
+
; uses long immediate
41
+
xor_s r3,r3,r12
42
+
bmsk r3,r3,r2
43
+
xor_s r12,r12,r3
44
+
#endif /* ENDIAN */
45
+
.Last_store:
46
+
j_s.d [blink]
47
+
st r12,[r5,0]
48
+
49
+
.balign 4
50
+
.Lcopy_bytewise:
51
+
jcs [blink]
52
+
ldb_s r12,[r1,0]
53
+
lsr.f lp_count,r3
54
+
bhs_s .Lnox1
55
+
stb.ab r12,[r5,1]
56
+
ldb.a r12,[r1,1]
57
+
.Lnox1:
58
+
lppnz .Lendbloop
59
+
ldb_s r3,[r1,1]
60
+
stb.ab r12,[r5,1]
61
+
ldb.a r12,[r1,2]
62
+
stb.ab r3,[r5,1]
63
+
.Lendbloop:
64
+
j_s.d [blink]
65
+
stb r12,[r5,0]
66
+
ARC_EXIT memcpy
+59
arch/arc/lib/memset.S
+59
arch/arc/lib/memset.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
#include <asm/linkage.h>
10
+
11
+
#define SMALL 7 /* Must be at least 6 to deal with alignment/loop issues. */
12
+
13
+
ARC_ENTRY memset
14
+
mov_s r4,r0
15
+
or r12,r0,r2
16
+
bmsk.f r12,r12,1
17
+
extb_s r1,r1
18
+
asl r3,r1,8
19
+
beq.d .Laligned
20
+
or_s r1,r1,r3
21
+
brls r2,SMALL,.Ltiny
22
+
add r3,r2,r0
23
+
stb r1,[r3,-1]
24
+
bclr_s r3,r3,0
25
+
stw r1,[r3,-2]
26
+
bmsk.f r12,r0,1
27
+
add_s r2,r2,r12
28
+
sub.ne r2,r2,4
29
+
stb.ab r1,[r4,1]
30
+
and r4,r4,-2
31
+
stw.ab r1,[r4,2]
32
+
and r4,r4,-4
33
+
.Laligned: ; This code address should be aligned for speed.
34
+
asl r3,r1,16
35
+
lsr.f lp_count,r2,2
36
+
or_s r1,r1,r3
37
+
lpne .Loop_end
38
+
st.ab r1,[r4,4]
39
+
.Loop_end:
40
+
j_s [blink]
41
+
42
+
.balign 4
43
+
.Ltiny:
44
+
mov.f lp_count,r2
45
+
lpne .Ltiny_end
46
+
stb.ab r1,[r4,1]
47
+
.Ltiny_end:
48
+
j_s [blink]
49
+
ARC_EXIT memset
50
+
51
+
; memzero: @r0 = mem, @r1 = size_t
52
+
; memset: @r0 = mem, @r1 = char, @r2 = size_t
53
+
54
+
ARC_ENTRY memzero
55
+
; adjust bzero args to memset args
56
+
mov r2, r1
57
+
mov r1, 0
58
+
b memset ;tail call so need to tinker with blink
59
+
ARC_EXIT memzero
+123
arch/arc/lib/strchr-700.S
+123
arch/arc/lib/strchr-700.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
/* ARC700 has a relatively long pipeline and branch prediction, so we want
10
+
to avoid branches that are hard to predict. On the other hand, the
11
+
presence of the norm instruction makes it easier to operate on whole
12
+
words branch-free. */
13
+
14
+
#include <asm/linkage.h>
15
+
16
+
ARC_ENTRY strchr
17
+
extb_s r1,r1
18
+
asl r5,r1,8
19
+
bmsk r2,r0,1
20
+
or r5,r5,r1
21
+
mov_s r3,0x01010101
22
+
breq.d r2,r0,.Laligned
23
+
asl r4,r5,16
24
+
sub_s r0,r0,r2
25
+
asl r7,r2,3
26
+
ld_s r2,[r0]
27
+
#ifdef __LITTLE_ENDIAN__
28
+
asl r7,r3,r7
29
+
#else
30
+
lsr r7,r3,r7
31
+
#endif
32
+
or r5,r5,r4
33
+
ror r4,r3
34
+
sub r12,r2,r7
35
+
bic_s r12,r12,r2
36
+
and r12,r12,r4
37
+
brne.d r12,0,.Lfound0_ua
38
+
xor r6,r2,r5
39
+
ld.a r2,[r0,4]
40
+
sub r12,r6,r7
41
+
bic r12,r12,r6
42
+
and r7,r12,r4
43
+
breq r7,0,.Loop ; For speed, we want this branch to be unaligned.
44
+
b .Lfound_char ; Likewise this one.
45
+
; /* We require this code address to be unaligned for speed... */
46
+
.Laligned:
47
+
ld_s r2,[r0]
48
+
or r5,r5,r4
49
+
ror r4,r3
50
+
; /* ... so that this code address is aligned, for itself and ... */
51
+
.Loop:
52
+
sub r12,r2,r3
53
+
bic_s r12,r12,r2
54
+
and r12,r12,r4
55
+
brne.d r12,0,.Lfound0
56
+
xor r6,r2,r5
57
+
ld.a r2,[r0,4]
58
+
sub r12,r6,r3
59
+
bic r12,r12,r6
60
+
and r7,r12,r4
61
+
breq r7,0,.Loop /* ... so that this branch is unaligned. */
62
+
; Found searched-for character. r0 has already advanced to next word.
63
+
#ifdef __LITTLE_ENDIAN__
64
+
/* We only need the information about the first matching byte
65
+
(i.e. the least significant matching byte) to be exact,
66
+
hence there is no problem with carry effects. */
67
+
.Lfound_char:
68
+
sub r3,r7,1
69
+
bic r3,r3,r7
70
+
norm r2,r3
71
+
sub_s r0,r0,1
72
+
asr_s r2,r2,3
73
+
j.d [blink]
74
+
sub_s r0,r0,r2
75
+
76
+
.balign 4
77
+
.Lfound0_ua:
78
+
mov r3,r7
79
+
.Lfound0:
80
+
sub r3,r6,r3
81
+
bic r3,r3,r6
82
+
and r2,r3,r4
83
+
or_s r12,r12,r2
84
+
sub_s r3,r12,1
85
+
bic_s r3,r3,r12
86
+
norm r3,r3
87
+
add_s r0,r0,3
88
+
asr_s r12,r3,3
89
+
asl.f 0,r2,r3
90
+
sub_s r0,r0,r12
91
+
j_s.d [blink]
92
+
mov.pl r0,0
93
+
#else /* BIG ENDIAN */
94
+
.Lfound_char:
95
+
lsr r7,r7,7
96
+
97
+
bic r2,r7,r6
98
+
norm r2,r2
99
+
sub_s r0,r0,4
100
+
asr_s r2,r2,3
101
+
j.d [blink]
102
+
add_s r0,r0,r2
103
+
104
+
.Lfound0_ua:
105
+
mov_s r3,r7
106
+
.Lfound0:
107
+
asl_s r2,r2,7
108
+
or r7,r6,r4
109
+
bic_s r12,r12,r2
110
+
sub r2,r7,r3
111
+
or r2,r2,r6
112
+
bic r12,r2,r12
113
+
bic.f r3,r4,r12
114
+
norm r3,r3
115
+
116
+
add.pl r3,r3,1
117
+
asr_s r12,r3,3
118
+
asl.f 0,r2,r3
119
+
add_s r0,r0,r12
120
+
j_s.d [blink]
121
+
mov.mi r0,0
122
+
#endif /* ENDIAN */
123
+
ARC_EXIT strchr
+96
arch/arc/lib/strcmp.S
+96
arch/arc/lib/strcmp.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
/* This is optimized primarily for the ARC700.
10
+
It would be possible to speed up the loops by one cycle / word
11
+
respective one cycle / byte by forcing double source 1 alignment, unrolling
12
+
by a factor of two, and speculatively loading the second word / byte of
13
+
source 1; however, that would increase the overhead for loop setup / finish,
14
+
and strcmp might often terminate early. */
15
+
16
+
#include <asm/linkage.h>
17
+
18
+
ARC_ENTRY strcmp
19
+
or r2,r0,r1
20
+
bmsk_s r2,r2,1
21
+
brne r2,0,.Lcharloop
22
+
mov_s r12,0x01010101
23
+
ror r5,r12
24
+
.Lwordloop:
25
+
ld.ab r2,[r0,4]
26
+
ld.ab r3,[r1,4]
27
+
nop_s
28
+
sub r4,r2,r12
29
+
bic r4,r4,r2
30
+
and r4,r4,r5
31
+
brne r4,0,.Lfound0
32
+
breq r2,r3,.Lwordloop
33
+
#ifdef __LITTLE_ENDIAN__
34
+
xor r0,r2,r3 ; mask for difference
35
+
sub_s r1,r0,1
36
+
bic_s r0,r0,r1 ; mask for least significant difference bit
37
+
sub r1,r5,r0
38
+
xor r0,r5,r1 ; mask for least significant difference byte
39
+
and_s r2,r2,r0
40
+
and_s r3,r3,r0
41
+
#endif /* LITTLE ENDIAN */
42
+
cmp_s r2,r3
43
+
mov_s r0,1
44
+
j_s.d [blink]
45
+
bset.lo r0,r0,31
46
+
47
+
.balign 4
48
+
#ifdef __LITTLE_ENDIAN__
49
+
.Lfound0:
50
+
xor r0,r2,r3 ; mask for difference
51
+
or r0,r0,r4 ; or in zero indicator
52
+
sub_s r1,r0,1
53
+
bic_s r0,r0,r1 ; mask for least significant difference bit
54
+
sub r1,r5,r0
55
+
xor r0,r5,r1 ; mask for least significant difference byte
56
+
and_s r2,r2,r0
57
+
and_s r3,r3,r0
58
+
sub.f r0,r2,r3
59
+
mov.hi r0,1
60
+
j_s.d [blink]
61
+
bset.lo r0,r0,31
62
+
#else /* BIG ENDIAN */
63
+
/* The zero-detection above can mis-detect 0x01 bytes as zeroes
64
+
because of carry-propagateion from a lower significant zero byte.
65
+
We can compensate for this by checking that bit0 is zero.
66
+
This compensation is not necessary in the step where we
67
+
get a low estimate for r2, because in any affected bytes
68
+
we already have 0x00 or 0x01, which will remain unchanged
69
+
when bit 7 is cleared. */
70
+
.balign 4
71
+
.Lfound0:
72
+
lsr r0,r4,8
73
+
lsr_s r1,r2
74
+
bic_s r2,r2,r0 ; get low estimate for r2 and get ...
75
+
bic_s r0,r0,r1 ; <this is the adjusted mask for zeros>
76
+
or_s r3,r3,r0 ; ... high estimate r3 so that r2 > r3 will ...
77
+
cmp_s r3,r2 ; ... be independent of trailing garbage
78
+
or_s r2,r2,r0 ; likewise for r3 > r2
79
+
bic_s r3,r3,r0
80
+
rlc r0,0 ; r0 := r2 > r3 ? 1 : 0
81
+
cmp_s r2,r3
82
+
j_s.d [blink]
83
+
bset.lo r0,r0,31
84
+
#endif /* ENDIAN */
85
+
86
+
.balign 4
87
+
.Lcharloop:
88
+
ldb.ab r2,[r0,1]
89
+
ldb.ab r3,[r1,1]
90
+
nop_s
91
+
breq r2,0,.Lcmpend
92
+
breq r2,r3,.Lcharloop
93
+
.Lcmpend:
94
+
j_s.d [blink]
95
+
sub r0,r2,r3
96
+
ARC_EXIT strcmp
+70
arch/arc/lib/strcpy-700.S
+70
arch/arc/lib/strcpy-700.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
/* If dst and src are 4 byte aligned, copy 8 bytes at a time.
10
+
If the src is 4, but not 8 byte aligned, we first read 4 bytes to get
11
+
it 8 byte aligned. Thus, we can do a little read-ahead, without
12
+
dereferencing a cache line that we should not touch.
13
+
Note that short and long instructions have been scheduled to avoid
14
+
branch stalls.
15
+
The beq_s to r3z could be made unaligned & long to avoid a stall
16
+
there, but the it is not likely to be taken often, and it
17
+
would also be likey to cost an unaligned mispredict at the next call. */
18
+
19
+
#include <asm/linkage.h>
20
+
21
+
ARC_ENTRY strcpy
22
+
or r2,r0,r1
23
+
bmsk_s r2,r2,1
24
+
brne.d r2,0,charloop
25
+
mov_s r10,r0
26
+
ld_s r3,[r1,0]
27
+
mov r8,0x01010101
28
+
bbit0.d r1,2,loop_start
29
+
ror r12,r8
30
+
sub r2,r3,r8
31
+
bic_s r2,r2,r3
32
+
tst_s r2,r12
33
+
bne r3z
34
+
mov_s r4,r3
35
+
.balign 4
36
+
loop:
37
+
ld.a r3,[r1,4]
38
+
st.ab r4,[r10,4]
39
+
loop_start:
40
+
ld.a r4,[r1,4]
41
+
sub r2,r3,r8
42
+
bic_s r2,r2,r3
43
+
tst_s r2,r12
44
+
bne_s r3z
45
+
st.ab r3,[r10,4]
46
+
sub r2,r4,r8
47
+
bic r2,r2,r4
48
+
tst r2,r12
49
+
beq loop
50
+
mov_s r3,r4
51
+
#ifdef __LITTLE_ENDIAN__
52
+
r3z: bmsk.f r1,r3,7
53
+
lsr_s r3,r3,8
54
+
#else
55
+
r3z: lsr.f r1,r3,24
56
+
asl_s r3,r3,8
57
+
#endif
58
+
bne.d r3z
59
+
stb.ab r1,[r10,1]
60
+
j_s [blink]
61
+
62
+
.balign 4
63
+
charloop:
64
+
ldb.ab r3,[r1,1]
65
+
66
+
67
+
brne.d r3,0,charloop
68
+
stb.ab r3,[r10,1]
69
+
j [blink]
70
+
ARC_EXIT strcpy
+83
arch/arc/lib/strlen.S
+83
arch/arc/lib/strlen.S
···
1
+
/*
2
+
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
3
+
*
4
+
* This program is free software; you can redistribute it and/or modify
5
+
* it under the terms of the GNU General Public License version 2 as
6
+
* published by the Free Software Foundation.
7
+
*/
8
+
9
+
#include <asm/linkage.h>
10
+
11
+
ARC_ENTRY strlen
12
+
or r3,r0,7
13
+
ld r2,[r3,-7]
14
+
ld.a r6,[r3,-3]
15
+
mov r4,0x01010101
16
+
; uses long immediate
17
+
#ifdef __LITTLE_ENDIAN__
18
+
asl_s r1,r0,3
19
+
btst_s r0,2
20
+
asl r7,r4,r1
21
+
ror r5,r4
22
+
sub r1,r2,r7
23
+
bic_s r1,r1,r2
24
+
mov.eq r7,r4
25
+
sub r12,r6,r7
26
+
bic r12,r12,r6
27
+
or.eq r12,r12,r1
28
+
and r12,r12,r5
29
+
brne r12,0,.Learly_end
30
+
#else /* BIG ENDIAN */
31
+
ror r5,r4
32
+
btst_s r0,2
33
+
mov_s r1,31
34
+
sub3 r7,r1,r0
35
+
sub r1,r2,r4
36
+
bic_s r1,r1,r2
37
+
bmsk r1,r1,r7
38
+
sub r12,r6,r4
39
+
bic r12,r12,r6
40
+
bmsk.ne r12,r12,r7
41
+
or.eq r12,r12,r1
42
+
and r12,r12,r5
43
+
brne r12,0,.Learly_end
44
+
#endif /* ENDIAN */
45
+
46
+
.Loop:
47
+
ld_s r2,[r3,4]
48
+
ld.a r6,[r3,8]
49
+
; stall for load result
50
+
sub r1,r2,r4
51
+
bic_s r1,r1,r2
52
+
sub r12,r6,r4
53
+
bic r12,r12,r6
54
+
or r12,r12,r1
55
+
and r12,r12,r5
56
+
breq r12,0,.Loop
57
+
.Lend:
58
+
and.f r1,r1,r5
59
+
sub.ne r3,r3,4
60
+
mov.eq r1,r12
61
+
#ifdef __LITTLE_ENDIAN__
62
+
sub_s r2,r1,1
63
+
bic_s r2,r2,r1
64
+
norm r1,r2
65
+
sub_s r0,r0,3
66
+
lsr_s r1,r1,3
67
+
sub r0,r3,r0
68
+
j_s.d [blink]
69
+
sub r0,r0,r1
70
+
#else /* BIG ENDIAN */
71
+
lsr_s r1,r1,7
72
+
mov.eq r2,r6
73
+
bic_s r1,r1,r2
74
+
norm r1,r1
75
+
sub r0,r3,r0
76
+
lsr_s r1,r1,3
77
+
j_s.d [blink]
78
+
add r0,r0,r1
79
+
#endif /* ENDIAN */
80
+
.Learly_end:
81
+
b.d .Lend
82
+
sub_s.ne r1,r1,r1
83
+
ARC_EXIT strlen