68b2e7f763
Some checks failed
sm-rpc / build (Debug, arm-linux-gnueabihf) (push) Successful in 1m34s
sm-rpc / build (Debug, aarch64-linux-gnu) (push) Successful in 2m46s
sm-rpc / build (Debug, host.gcc) (push) Failing after 1m28s
sm-rpc / build (Release, aarch64-linux-gnu) (push) Successful in 2m14s
sm-rpc / build (Release, arm-linux-gnueabihf) (push) Successful in 2m8s
sm-rpc / build (Debug, mipsel-linux-gnu) (push) Successful in 5m35s
sm-rpc / build (Release, host.gcc) (push) Failing after 1m55s
sm-rpc / build (Release, mipsel-linux-gnu) (push) Successful in 7m21s
41 lines
1.1 KiB
C
41 lines
1.1 KiB
C
/*
|
|
* AVX2 optimized hash slide, based on Intel's slide_sse implementation
|
|
*
|
|
* Copyright (C) 2017 Intel Corporation
|
|
* Authors:
|
|
* Arjan van de Ven <arjan@linux.intel.com>
|
|
* Jim Kukunas <james.t.kukunas@linux.intel.com>
|
|
* Mika T. Lindqvist <postmaster@raasu.org>
|
|
*
|
|
* For conditions of distribution and use, see copyright notice in zlib.h
|
|
*/
|
|
#include "zbuild.h"
|
|
#include "deflate.h"
|
|
|
|
#include <immintrin.h>
|
|
|
|
static inline void slide_hash_chain(Pos *table, uint32_t entries, const __m256i wsize) {
|
|
table += entries;
|
|
table -= 16;
|
|
|
|
do {
|
|
__m256i value, result;
|
|
|
|
value = _mm256_loadu_si256((__m256i *)table);
|
|
result = _mm256_subs_epu16(value, wsize);
|
|
_mm256_storeu_si256((__m256i *)table, result);
|
|
|
|
table -= 16;
|
|
entries -= 16;
|
|
} while (entries > 0);
|
|
}
|
|
|
|
Z_INTERNAL void slide_hash_avx2(deflate_state *s) {
|
|
Assert(s->w_size <= UINT16_MAX, "w_size should fit in uint16_t");
|
|
uint16_t wsize = (uint16_t)s->w_size;
|
|
const __m256i ymm_wsize = _mm256_set1_epi16((short)wsize);
|
|
|
|
slide_hash_chain(s->head, HASH_SIZE, ymm_wsize);
|
|
slide_hash_chain(s->prev, wsize, ymm_wsize);
|
|
}
|