opencv

1
/*
2
 * AVX2 optimized hash slide, based on Intel's slide_sse implementation
3
 *
4
 * Copyright (C) 2017 Intel Corporation
5
 * Authors:
6
 *   Arjan van de Ven   <arjan@linux.intel.com>
7
 *   Jim Kukunas        <james.t.kukunas@linux.intel.com>
8
 *   Mika T. Lindqvist  <postmaster@raasu.org>
9
 *
10
 * For conditions of distribution and use, see copyright notice in zlib.h
11
 */
12
#include "../../zbuild.h"
13
#include "../../deflate.h"
14

15
#include <immintrin.h>
16

17
static inline void slide_hash_chain(Pos *table, uint32_t entries, const __m256i wsize) {
18
    table += entries;
19
    table -= 16;
20

21
    do {
22
        __m256i value, result;
23

24
        value = _mm256_loadu_si256((__m256i *)table);
25
        result = _mm256_subs_epu16(value, wsize);
26
        _mm256_storeu_si256((__m256i *)table, result);
27

28
        table -= 16;
29
        entries -= 16;
30
    } while (entries > 0);
31
}
32

33
Z_INTERNAL void slide_hash_avx2(deflate_state *s) {
34
    uint16_t wsize = (uint16_t)s->w_size;
35
    const __m256i ymm_wsize = _mm256_set1_epi16((short)wsize);
36

37
    slide_hash_chain(s->head, HASH_SIZE, ymm_wsize);
38
    slide_hash_chain(s->prev, wsize, ymm_wsize);
39
}
40