1/*
2 * Copyright 2022 ByteDance Inc.
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#pragma once
18
19#include <immintrin.h>
20#include <string.h>
21#include "native.h"
22
23static always_inline bool vec_cross_page(const void * p, size_t n) {
24#define PAGE_SIZE 4096
25 return (((size_t)(p)) & (PAGE_SIZE - 1)) > (PAGE_SIZE - n);
26#undef PAGE_SIZE
27}
28
29static always_inline void memcpy4 (void *__restrict dp, const void *__restrict sp) {
30 ((uint32_t *)dp)[0] = ((const uint32_t *)sp)[0];
31}
32
33static always_inline void memcpy8 (void *__restrict dp, const void *__restrict sp) {
34 ((uint64_t *)dp)[0] = ((const uint64_t *)sp)[0];
35}
36
37static always_inline void memcpy16 (void *__restrict dp, const void *__restrict sp) {
38 _mm_storeu_si128((void *)(dp), _mm_loadu_si128((const void *)(sp)));
39}
40
41static always_inline void memcpy32(void *__restrict dp, const void *__restrict sp) {
42#if USE_AVX2
43 _mm256_storeu_si256((void *)dp, _mm256_loadu_si256((const void *)sp));
44#else
45 _mm_storeu_si128((void *)(dp), _mm_loadu_si128((const void *)(sp)));
46 _mm_storeu_si128((void *)(dp + 16), _mm_loadu_si128((const void *)(sp + 16)));
47#endif
48}
49
50static always_inline void memcpy64(void *__restrict dp, const void *__restrict sp) {
51 memcpy32(dp, sp);
52 memcpy32(dp + 32, sp + 32);
53}
54
55static always_inline void memcpy_p4(void *__restrict dp, const void *__restrict sp, size_t nb) {
56 if (nb >= 2) { *(uint16_t *)dp = *(const uint16_t *)sp; sp += 2, dp += 2, nb -= 2; }
57 if (nb >= 1) { *(uint8_t *) dp = *(const uint8_t *)sp; }
58}
59
60static always_inline void memcpy_p8(void *__restrict dp, const void *__restrict sp, ssize_t nb) {
61 if (nb >= 4) { memcpy4(dp, sp); sp += 4, dp += 4, nb -= 4; }
62 memcpy_p4(dp, sp, nb);
63}
64
65static always_inline void memcpy_p16(void *__restrict dp, const void *__restrict sp, size_t nb) {
66 if (nb >= 8) { memcpy8(dp, sp); sp += 8, dp += 8, nb -= 8; }
67 memcpy_p8(dp, sp, nb);
68}
69
70static always_inline void memcpy_p32(void *__restrict dp, const void *__restrict sp, size_t nb) {
71 if (nb >= 16) { memcpy16(dp, sp); sp += 16, dp += 16, nb -= 16; }
72 memcpy_p16(dp, sp, nb);
73}
74
75static always_inline void memcpy_p64(void *__restrict dp, const void *__restrict sp, size_t nb) {
76 if (nb >= 32) { memcpy32(dp, sp); sp += 32, dp += 32, nb -= 32; }
77 memcpy_p32(dp, sp, nb);
78}
View as plain text