• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2013 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Eric Anholt <eric@anholt.net>
25  *    Matt Turner <mattst88@gmail.com>
26  *
27  */
28 
29 #include "main/macros.h"
30 #include "util/streaming-load-memcpy.h"
31 #include "x86/common_x86_asm.h"
32 #ifdef USE_SSE41
33 #include <smmintrin.h>
34 #endif
35 
36 /* Copies memory from src to dst, using SSE 4.1's MOVNTDQA to get streaming
37  * read performance from uncached memory.
38  */
39 void
util_streaming_load_memcpy(void * restrict dst,void * restrict src,size_t len)40 util_streaming_load_memcpy(void *restrict dst, void *restrict src, size_t len)
41 {
42    char *restrict d = dst;
43    char *restrict s = src;
44 
45 #ifdef USE_SSE41
46    /* If dst and src are not co-aligned, or if SSE4.1 is not present, fallback to memcpy(). */
47    if (((uintptr_t)d & 15) != ((uintptr_t)s & 15) || !cpu_has_sse4_1) {
48       memcpy(d, s, len);
49       return;
50    }
51 
52    /* memcpy() the misaligned header. At the end of this if block, <d> and <s>
53     * are aligned to a 16-byte boundary or <len> == 0.
54     */
55    if ((uintptr_t)d & 15) {
56       uintptr_t bytes_before_alignment_boundary = 16 - ((uintptr_t)d & 15);
57       assert(bytes_before_alignment_boundary < 16);
58 
59       memcpy(d, s, MIN2(bytes_before_alignment_boundary, len));
60 
61       d = (char *)ALIGN((uintptr_t)d, 16);
62       s = (char *)ALIGN((uintptr_t)s, 16);
63       len -= MIN2(bytes_before_alignment_boundary, len);
64    }
65 
66    if (len >= 64)
67       _mm_mfence();
68 
69    while (len >= 64) {
70       __m128i *dst_cacheline = (__m128i *)d;
71       __m128i *src_cacheline = (__m128i *)s;
72 
73       __m128i temp1 = _mm_stream_load_si128(src_cacheline + 0);
74       __m128i temp2 = _mm_stream_load_si128(src_cacheline + 1);
75       __m128i temp3 = _mm_stream_load_si128(src_cacheline + 2);
76       __m128i temp4 = _mm_stream_load_si128(src_cacheline + 3);
77 
78       _mm_store_si128(dst_cacheline + 0, temp1);
79       _mm_store_si128(dst_cacheline + 1, temp2);
80       _mm_store_si128(dst_cacheline + 2, temp3);
81       _mm_store_si128(dst_cacheline + 3, temp4);
82 
83       d += 64;
84       s += 64;
85       len -= 64;
86    }
87 #endif
88    /* memcpy() the tail. */
89    if (len) {
90       memcpy(d, s, len);
91    }
92 }
93