1 /* SPDX-License-Identifier: GPL-2.0 */ !! 1 /*
2 /* strlen.S: Sparc optimized strlen code !! 2 * strlen.S (c) 1995 David Mosberger (davidm@cs.arizona.edu)
3 * Hand optimized from GNU libc's strlen !! 3 *
4 * Copyright (C) 1991,1996 Free Software Found !! 4 * Finds length of a 0-terminated string. Optimized for the
5 * Copyright (C) 1996,2008 David S. Miller (da !! 5 * Alpha architecture:
6 * Copyright (C) 1996, 1997 Jakub Jelinek (jj@ !! 6 *
>> 7 * - memory accessed as aligned quadwords only
>> 8 * - uses bcmpge to compare 8 bytes in parallel
>> 9 * - does binary search to find 0 byte in last
>> 10 * quadword (HAKMEM needed 12 instructions to
>> 11 * do this instead of the 9 instructions that
>> 12 * binary search needs).
7 */ 13 */
8 14
9 #include <linux/export.h> !! 15 .set noreorder
10 #include <linux/linkage.h> !! 16 .set noat
11 #include <asm/asm.h> !! 17
12 !! 18 .align 3
13 #define LO_MAGIC 0x01010101 !! 19
14 #define HI_MAGIC 0x80808080 !! 20 .globl strlen
15 !! 21 .ent strlen
16 .text !! 22
17 ENTRY(strlen) !! 23 strlen:
18 mov %o0, %o1 !! 24 ldq_u $1, 0($16) # load first quadword ($16 may be misaligned)
19 andcc %o0, 3, %g0 !! 25 lda $2, -1($31)
20 BRANCH32(be, pt, 9f) !! 26 insqh $2, $16, $2
21 sethi %hi(HI_MAGIC), %o4 !! 27 andnot $16, 7, $0
22 ldub [%o0], %o5 !! 28 or $2, $1, $1
23 BRANCH_REG_ZERO(pn, %o5, 11f) !! 29 cmpbge $31, $1, $2 # $2 <- bitmask: bit i == 1 <==> i-th byte == 0
24 add %o0, 1, %o0 !! 30 bne $2, found
25 andcc %o0, 3, %g0 !! 31
26 BRANCH32(be, pn, 4f) !! 32 loop: ldq $1, 8($0)
27 or %o4, %lo(HI_MAGIC), %o3 !! 33 addq $0, 8, $0 # addr += 8
28 ldub [%o0], %o5 !! 34 nop # helps dual issue last two insns
29 BRANCH_REG_ZERO(pn, %o5, 12f) !! 35 cmpbge $31, $1, $2
30 add %o0, 1, %o0 !! 36 beq $2, loop
31 andcc %o0, 3, %g0 !! 37
32 BRANCH32(be, pt, 5f) !! 38 found: blbs $2, done # make aligned case fast
33 sethi %hi(LO_MAGIC), %o4 !! 39 negq $2, $3
34 ldub [%o0], %o5 !! 40 and $2, $3, $2
35 BRANCH_REG_ZERO(pn, %o5, 13f) !! 41
36 add %o0, 1, %o0 !! 42 and $2, 0x0f, $1
37 BRANCH32(ba, pt, 8f) !! 43 addq $0, 4, $3
38 or %o4, %lo(LO_MAGIC), %o2 !! 44 cmoveq $1, $3, $0
39 9: !! 45
40 or %o4, %lo(HI_MAGIC), %o3 !! 46 and $2, 0x33, $1
41 4: !! 47 addq $0, 2, $3
42 sethi %hi(LO_MAGIC), %o4 !! 48 cmoveq $1, $3, $0
43 5: !! 49
44 or %o4, %lo(LO_MAGIC), %o2 !! 50 and $2, 0x55, $1
45 8: !! 51 addq $0, 1, $3
46 ld [%o0], %o5 !! 52 cmoveq $1, $3, $0
47 2: !! 53
48 sub %o5, %o2, %o4 !! 54 done: subq $0, $16, $0
49 andcc %o4, %o3, %g0 !! 55 ret $31, ($26)
50 BRANCH32(be, pt, 8b) !! 56
51 add %o0, 4, %o0 !! 57 .end strlen
52 <<
53 /* Check every byte. */ <<
54 srl %o5, 24, %g7 <<
55 andcc %g7, 0xff, %g0 <<
56 BRANCH32(be, pn, 1f) <<
57 add %o0, -4, %o4 <<
58 srl %o5, 16, %g7 <<
59 andcc %g7, 0xff, %g0 <<
60 BRANCH32(be, pn, 1f) <<
61 add %o4, 1, %o4 <<
62 srl %o5, 8, %g7 <<
63 andcc %g7, 0xff, %g0 <<
64 BRANCH32(be, pn, 1f) <<
65 add %o4, 1, %o4 <<
66 andcc %o5, 0xff, %g0 <<
67 BRANCH32_ANNUL(bne, pt, 2b) <<
68 ld [%o0], %o5 <<
69 add %o4, 1, %o4 <<
70 1: <<
71 retl <<
72 sub %o4, %o1, %o0 <<
73 11: <<
74 retl <<
75 mov 0, %o0 <<
76 12: <<
77 retl <<
78 mov 1, %o0 <<
79 13: <<
80 retl <<
81 mov 2, %o0 <<
82 ENDPROC(strlen) <<
83 EXPORT_SYMBOL(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.