1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* strlen.S: Sparc optimized strlen code !! 2 /*
3 * Hand optimized from GNU libc's strlen !! 3 * strlen.S (c) 1995 David Mosberger (davidm@cs.arizona.edu)
4 * Copyright (C) 1991,1996 Free Software Found !! 4 *
5 * Copyright (C) 1996,2008 David S. Miller (da !! 5 * Finds length of a 0-terminated string. Optimized for the
6 * Copyright (C) 1996, 1997 Jakub Jelinek (jj@ !! 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).
7 */ 14 */
8 <<
9 #include <linux/export.h> 15 #include <linux/export.h>
10 #include <linux/linkage.h> !! 16 .set noreorder
11 #include <asm/asm.h> !! 17 .set noat
>> 18
>> 19 .align 3
>> 20
>> 21 .globl strlen
>> 22 .ent strlen
>> 23
>> 24 strlen:
>> 25 ldq_u $1, 0($16) # load first quadword ($16 may be misaligned)
>> 26 lda $2, -1($31)
>> 27 insqh $2, $16, $2
>> 28 andnot $16, 7, $0
>> 29 or $2, $1, $1
>> 30 cmpbge $31, $1, $2 # $2 <- bitmask: bit i == 1 <==> i-th byte == 0
>> 31 bne $2, found
>> 32
>> 33 loop: ldq $1, 8($0)
>> 34 addq $0, 8, $0 # addr += 8
>> 35 nop # helps dual issue last two insns
>> 36 cmpbge $31, $1, $2
>> 37 beq $2, loop
>> 38
>> 39 found: blbs $2, done # make aligned case fast
>> 40 negq $2, $3
>> 41 and $2, $3, $2
>> 42
>> 43 and $2, 0x0f, $1
>> 44 addq $0, 4, $3
>> 45 cmoveq $1, $3, $0
>> 46
>> 47 and $2, 0x33, $1
>> 48 addq $0, 2, $3
>> 49 cmoveq $1, $3, $0
>> 50
>> 51 and $2, 0x55, $1
>> 52 addq $0, 1, $3
>> 53 cmoveq $1, $3, $0
12 54
13 #define LO_MAGIC 0x01010101 !! 55 done: subq $0, $16, $0
14 #define HI_MAGIC 0x80808080 !! 56 ret $31, ($26)
15 57
16 .text !! 58 .end strlen
17 ENTRY(strlen) !! 59 EXPORT_SYMBOL(strlen)
18 mov %o0, %o1 <<
19 andcc %o0, 3, %g0 <<
20 BRANCH32(be, pt, 9f) <<
21 sethi %hi(HI_MAGIC), %o4 <<
22 ldub [%o0], %o5 <<
23 BRANCH_REG_ZERO(pn, %o5, 11f) <<
24 add %o0, 1, %o0 <<
25 andcc %o0, 3, %g0 <<
26 BRANCH32(be, pn, 4f) <<
27 or %o4, %lo(HI_MAGIC), %o3 <<
28 ldub [%o0], %o5 <<
29 BRANCH_REG_ZERO(pn, %o5, 12f) <<
30 add %o0, 1, %o0 <<
31 andcc %o0, 3, %g0 <<
32 BRANCH32(be, pt, 5f) <<
33 sethi %hi(LO_MAGIC), %o4 <<
34 ldub [%o0], %o5 <<
35 BRANCH_REG_ZERO(pn, %o5, 13f) <<
36 add %o0, 1, %o0 <<
37 BRANCH32(ba, pt, 8f) <<
38 or %o4, %lo(LO_MAGIC), %o2 <<
39 9: <<
40 or %o4, %lo(HI_MAGIC), %o3 <<
41 4: <<
42 sethi %hi(LO_MAGIC), %o4 <<
43 5: <<
44 or %o4, %lo(LO_MAGIC), %o2 <<
45 8: <<
46 ld [%o0], %o5 <<
47 2: <<
48 sub %o5, %o2, %o4 <<
49 andcc %o4, %o3, %g0 <<
50 BRANCH32(be, pt, 8b) <<
51 add %o0, 4, %o0 <<
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.