1 /* SPDX-License-Identifier: GPL-2.0-only */ !! 1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* 2 /*
3 * Copyright (C) 2004, 2007-2010, 2011-2012 Sy !! 3 * strlen.S (c) 1995 David Mosberger (davidm@cs.arizona.edu)
>> 4 *
>> 5 * Finds length of a 0-terminated string. Optimized for the
>> 6 * Alpha architecture:
>> 7 *
>> 8 * - memory accessed as aligned quadwords only
>> 9 * - uses bcmpge to compare 8 bytes in parallel
>> 10 * - does binary search to find 0 byte in last
>> 11 * quadword (HAKMEM needed 12 instructions to
>> 12 * do this instead of the 9 instructions that
>> 13 * binary search needs).
4 */ 14 */
>> 15 #include <linux/export.h>
>> 16 .set noreorder
>> 17 .set noat
5 18
6 #include <linux/linkage.h> !! 19 .align 3
7 20
8 ENTRY_CFI(strlen) !! 21 .globl strlen
9 or r3,r0,7 !! 22 .ent strlen
10 ld r2,[r3,-7] !! 23
11 ld.a r6,[r3,-3] !! 24 strlen:
12 mov r4,0x01010101 !! 25 ldq_u $1, 0($16) # load first quadword ($16 may be misaligned)
13 ; uses long immediate !! 26 lda $2, -1($31)
14 #ifdef __LITTLE_ENDIAN__ !! 27 insqh $2, $16, $2
15 asl_s r1,r0,3 !! 28 andnot $16, 7, $0
16 btst_s r0,2 !! 29 or $2, $1, $1
17 asl r7,r4,r1 !! 30 cmpbge $31, $1, $2 # $2 <- bitmask: bit i == 1 <==> i-th byte == 0
18 ror r5,r4 !! 31 bne $2, found
19 sub r1,r2,r7 !! 32
20 bic_s r1,r1,r2 !! 33 loop: ldq $1, 8($0)
21 mov.eq r7,r4 !! 34 addq $0, 8, $0 # addr += 8
22 sub r12,r6,r7 !! 35 nop # helps dual issue last two insns
23 bic r12,r12,r6 !! 36 cmpbge $31, $1, $2
24 or.eq r12,r12,r1 !! 37 beq $2, loop
25 and r12,r12,r5 !! 38
26 brne r12,0,.Learly_end !! 39 found: blbs $2, done # make aligned case fast
27 #else /* BIG ENDIAN */ !! 40 negq $2, $3
28 ror r5,r4 !! 41 and $2, $3, $2
29 btst_s r0,2 !! 42
30 mov_s r1,31 !! 43 and $2, 0x0f, $1
31 sub3 r7,r1,r0 !! 44 addq $0, 4, $3
32 sub r1,r2,r4 !! 45 cmoveq $1, $3, $0
33 bic_s r1,r1,r2 !! 46
34 bmsk r1,r1,r7 !! 47 and $2, 0x33, $1
35 sub r12,r6,r4 !! 48 addq $0, 2, $3
36 bic r12,r12,r6 !! 49 cmoveq $1, $3, $0
37 bmsk.ne r12,r12,r7 !! 50
38 or.eq r12,r12,r1 !! 51 and $2, 0x55, $1
39 and r12,r12,r5 !! 52 addq $0, 1, $3
40 brne r12,0,.Learly_end !! 53 cmoveq $1, $3, $0
41 #endif /* ENDIAN */ !! 54
42 !! 55 done: subq $0, $16, $0
43 .Loop: !! 56 ret $31, ($26)
44 ld_s r2,[r3,4] !! 57
45 ld.a r6,[r3,8] !! 58 .end strlen
46 ; stall for load result !! 59 EXPORT_SYMBOL(strlen)
47 sub r1,r2,r4 <<
48 bic_s r1,r1,r2 <<
49 sub r12,r6,r4 <<
50 bic r12,r12,r6 <<
51 or r12,r12,r1 <<
52 and r12,r12,r5 <<
53 breq r12,0,.Loop <<
54 .Lend: <<
55 and.f r1,r1,r5 <<
56 sub.ne r3,r3,4 <<
57 mov.eq r1,r12 <<
58 #ifdef __LITTLE_ENDIAN__ <<
59 sub_s r2,r1,1 <<
60 bic_s r2,r2,r1 <<
61 norm r1,r2 <<
62 sub_s r0,r0,3 <<
63 lsr_s r1,r1,3 <<
64 sub r0,r3,r0 <<
65 j_s.d [blink] <<
66 sub r0,r0,r1 <<
67 #else /* BIG ENDIAN */ <<
68 lsr_s r1,r1,7 <<
69 mov.eq r2,r6 <<
70 bic_s r1,r1,r2 <<
71 norm r1,r1 <<
72 sub r0,r3,r0 <<
73 lsr_s r1,r1,3 <<
74 j_s.d [blink] <<
75 add r0,r0,r1 <<
76 #endif /* ENDIAN */ <<
77 .Learly_end: <<
78 b.d .Lend <<
79 sub_s.ne r1,r1,r1 <<
80 END_CFI(strlen) <<
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.