1fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil/*
2fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * Copyright (c) 2017 Imagination Technologies.
3fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *
4fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * All rights reserved.
5fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *
6fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * Redistribution and use in source and binary forms, with or without
7fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * modification, are permitted provided that the following conditions
8fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * are met:
9fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *
10fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *      * Redistributions of source code must retain the above copyright
11fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        notice, this list of conditions and the following disclaimer.
12fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *      * Redistributions in binary form must reproduce the above copyright
13fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        notice, this list of conditions and the following disclaimer
14fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        in the documentation and/or other materials provided with
15fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        the distribution.
16fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *      * Neither the name of Imagination Technologies nor the names of its
17fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        contributors may be used to endorse or promote products derived
18fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *        from this software without specific prior written permission.
19fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil *
20fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil */
32fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#include <string.h>
33fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
34fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define ENABLE_PREFETCH     1
35fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define op_t                unsigned long int
36fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define op_size             sizeof (op_t)
37fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
38fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#if ENABLE_PREFETCH
39fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define PREFETCH(addr)  __builtin_prefetch (addr, 0, 1);
40fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#else
41fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define PREFETCH(addr)
42fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#endif
43fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
44fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#if __mips64 || __mips_isa_rev >= 2
45fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patilstatic inline void * __attribute__ ((always_inline))
46fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patildo_bytes (const op_t* w, op_t inval)
47fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil{
48fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  const unsigned char *p = (const unsigned char *) w;
49fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t outval = 0;
50fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#if __mips64
51fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  __asm__ volatile (
52fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "dsbh %1, %0 \n\t"
53fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "dshd %0, %1 \n\t"
54fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "dclz %1, %0 \n\t"
55fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    : "+r" (inval), "+r" (outval)
56fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  );
57fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#else
58fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  __asm__ volatile (
59fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "wsbh %1, %0 \n\t"
60fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "rotr %0, %1, 16 \n\t"
61fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    "clz %1, %0 \n\t"
62fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    : "+r" (inval), "+r" (outval)
63fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  );
64fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#endif
65fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  p += (outval >> 3);
66fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  return (void *) p;
67fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
68fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
69fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define DO_WORD(in, val) {                        \
70fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t tmp = ((val - mask_1) & ~val) & mask_128;  \
71fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  if (tmp != 0)                                   \
72fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    return do_bytes(in, tmp);                     \
73fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
74fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#else
75fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patilstatic inline void * __attribute__ ((always_inline))
76fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patildo_bytes (const op_t* w, unsigned char ch)
77fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil{
78fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  const unsigned char *p = (const unsigned char *) w;
79fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  for (; *p != ch; ++p);
80fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  return (void *) p;
81fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
82fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
83fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define DO_WORD(in, val) {                        \
84fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t tmp = ((val - mask_1) & ~val) & mask_128;  \
85fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  if (tmp != 0)                                   \
86fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    return do_bytes(in, ch);                      \
87fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
88fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#endif
89fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
90fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#define DO_WORDS(w) {          \
91fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t* w1 = (op_t*) w;        \
92fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t val0 = w1[0] ^ mask_c;  \
93fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t val1 = w1[1] ^ mask_c;  \
94fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t val2 = w1[2] ^ mask_c;  \
95fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  op_t val3 = w1[3] ^ mask_c;  \
96fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  DO_WORD(w1, val0)            \
97fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  DO_WORD(w1 + 1, val1)        \
98fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  DO_WORD(w1 + 2, val2)        \
99fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  DO_WORD(w1 + 3, val3)        \
100fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
101fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
102fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patilvoid *
103fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patilmemchr (void const *s, int c_in, size_t n) __overloadable
104fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil{
105fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  if (n != 0) {
106fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    const unsigned char *p = (const unsigned char *) s;
107fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    const op_t *w;
108fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    op_t mask_1, mask_128, mask_c;
109fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    unsigned char ch = (unsigned char) c_in;
110fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
111fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    /*
112fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     * Check bytewize till initial alignment
113fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     */
114fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    for (; n > 0 && ((size_t) p % op_size) != 0; --n, ++p) {
115fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      if (*p == ch)
116fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil        return (void *) p;
117fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
118fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
119fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    w = (const op_t *) p;
120fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
121fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    mask_c = ch | (ch << 8);
122fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    mask_c |= mask_c << 16;
123fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    __asm__ volatile (
124fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      "li %0, 0x01010101 \n\t"
125fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      : "=r" (mask_1)
126fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    );
127fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#if __mips64
128fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    mask_1 |= mask_1 << 32;
129fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    mask_c |= mask_c << 32;
130fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#endif
131fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    mask_128 = mask_1 << 7;
132fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
133fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    /*
134fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     * Check op_size byteswize after initial alignment
135fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     */
136fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#if ((_MIPS_SIM == _ABIO32) || _MIPS_TUNE_I6400)
137fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    PREFETCH (w);
138fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    PREFETCH (w + 8);
139fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    while (n >= 24 * op_size) {
140fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      PREFETCH(w + 16);
141fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w);
142fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w + 4);
143fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      w += 8;
144fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      n -= 8 * op_size;
145fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
146fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    while (n >= 8 * op_size) {
147fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w);
148fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w + 4);
149fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      w += 8;
150fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      n -= 8 * op_size;
151fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
152fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#else
153fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    PREFETCH (w);
154fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    PREFETCH (w + 4);
155fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    while (n >= 12 * op_size) {
156fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      PREFETCH(w + 8);
157fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w);
158fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      w += 4;
159fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      n -= 4 * op_size;
160fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
161fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    while (n >= 4 * op_size) {
162fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORDS(w);
163fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      w += 4;
164fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      n -= 4 * op_size;
165fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
166fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil#endif
167fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
168fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    while (n >= op_size) {
169fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      op_t val = *w ^ mask_c;
170fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      DO_WORD(w, val);
171fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      w++;
172fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      n -= op_size;
173fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
174fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil
175fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    /*
176fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     * Check bytewize for remaining bytes
177fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil     */
178fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    p = (const unsigned char *) w;
179fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    for (; n > 0; --n, ++p) {
180fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil      if (*p == ch)
181fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil        return (void *) p;
182fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil    }
183fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  }
184fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil  return NULL;
185fcb877ac4b05e65f64b9b3199d71dfeaae299177Prashant Patil}
186