1/* Optimized strchrnul implementation for PowerPC32/POWER7 using cmpb insn.
2   Copyright (C) 2010-2022 Free Software Foundation, Inc.
3   This file is part of the GNU C Library.
4
5   The GNU C Library is free software; you can redistribute it and/or
6   modify it under the terms of the GNU Lesser General Public
7   License as published by the Free Software Foundation; either
8   version 2.1 of the License, or (at your option) any later version.
9
10   The GNU C Library is distributed in the hope that it will be useful,
11   but WITHOUT ANY WARRANTY; without even the implied warranty of
12   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13   Lesser General Public License for more details.
14
15   You should have received a copy of the GNU Lesser General Public
16   License along with the GNU C Library; if not, see
17   <https://www.gnu.org/licenses/>.  */
18
19#include <sysdep.h>
20
21/* int [r3] strchrnul (char *s [r3], int c [r4])  */
22	.machine  power7
23ENTRY (__strchrnul)
24	CALL_MCOUNT
25	dcbt	0,r3
26	clrrwi	r8,r3,2	      /* Align the address to word boundary.  */
27
28	/* Replicate byte to word.  */
29	insrwi  r4,r4,8,16
30	insrwi  r4,r4,16,0
31
32	rlwinm	r6,r3,3,27,28 /* Calculate padding.  */
33	lwz	r12,0(r8)     /* Load word from memory.  */
34	li	r0,0	      /* Word with null chars to use
35				 with cmpb.  */
36
37	/* Now r4 has a word of c bytes and r0 has
38	   a word of null bytes.  */
39
40	cmpb	r10,r12,r0    /* Compare each byte against c byte.  */
41	cmpb	r9,r12,r4     /* Compare each byte against null byte.  */
42
43	/* Move the words left and right to discard the bits that are
44	   not part of the string and bring them back as zeros.  */
45#ifdef __LITTLE_ENDIAN__
46	srw	r10,r10,r6
47	srw	r9,r9,r6
48	slw	r10,r10,r6
49	slw	r9,r9,r6
50#else
51	slw	r10,r10,r6
52	slw	r9,r9,r6
53	srw	r10,r10,r6
54	srw	r9,r9,r6
55#endif
56	or	r5,r9,r10     /* OR the results to speed things up.  */
57	cmpwi	cr7,r5,0      /* If r5 == 0, no c or null bytes
58				 have been found.  */
59	bne	cr7,L(done)
60
61	mtcrf   0x01,r8
62
63	/* Are we now aligned to a doubleword boundary?  If so, skip to
64	   the main loop.  Otherwise, go through the alignment code.  */
65
66	bt	29,L(loop)
67
68	/* Handle WORD2 of pair.  */
69	lwzu	r12,4(r8)
70	cmpb	r10,r12,r0
71	cmpb	r9,r12,r4
72	or	r5,r9,r10
73	cmpwi	cr7,r5,0
74	bne	cr7,L(done)
75	b	L(loop)	      /* We branch here (rather than falling through)
76				 to skip the nops due to heavy alignment
77				 of the loop below.  */
78
79	.p2align  5
80L(loop):
81	/* Load two words, compare and merge in a
82	   single register for speed.  This is an attempt
83	   to speed up the null-checking process for bigger strings.  */
84	lwz	r12,4(r8)
85	lwzu	r11,8(r8)
86	cmpb	r10,r12,r0
87	cmpb	r9,r12,r4
88	cmpb	r6,r11,r0
89	cmpb	r7,r11,r4
90	or	r5,r9,r10
91	or	r10,r6,r7
92	or	r11,r5,r10
93	cmpwi	cr7,r11,0
94	beq	cr7,L(loop)
95
96	/* OK, one (or both) of the words contains a c/null byte.  Check
97	   the first word and decrement the address in case the first
98	   word really contains a c/null byte.  */
99
100	cmpwi	cr6,r5,0
101	addi	r8,r8,-4
102	bne	cr6,L(done)
103
104	/* The c/null byte must be in the second word.  Adjust the address
105	   again and move the result of cmpb to r5 so we can calculate the
106	   pointer.  */
107	mr	r5,r10
108	addi	r8,r8,4
109
110	/* r5 has the output of the cmpb instruction, that is, it contains
111	   0xff in the same position as the c/null byte in the original
112	   word from the string.  Use that to calculate the pointer.  */
113L(done):
114#ifdef __LITTLE_ENDIAN__
115	addi    r0,r5,-1
116	andc    r0,r0,r5
117	popcntw	r0,r0
118#else
119	cntlzw	r0,r5	      /* Count leading zeros before the match.  */
120#endif
121	srwi	r0,r0,3	      /* Convert leading zeros to bytes.  */
122	add	r3,r8,r0      /* Return address of matching c/null byte.  */
123	blr
124END (__strchrnul)
125weak_alias (__strchrnul,strchrnul)
126libc_hidden_builtin_def (__strchrnul)
127