Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 1 | /* |
Damjan Marion | f1213b8 | 2016-03-13 02:22:06 +0100 | [diff] [blame] | 2 | * Copyright (c) 2016 Cisco and/or its affiliates. |
| 3 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | * you may not use this file except in compliance with the License. |
| 5 | * You may obtain a copy of the License at: |
| 6 | * |
| 7 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | * |
| 9 | * Unless required by applicable law or agreed to in writing, software |
| 10 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | * See the License for the specific language governing permissions and |
| 13 | * limitations under the License. |
| 14 | */ |
| 15 | /* |
Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 16 | Copyright (c) 2001, 2002, 2003 Eliot Dresselhaus |
| 17 | |
| 18 | Permission is hereby granted, free of charge, to any person obtaining |
| 19 | a copy of this software and associated documentation files (the |
| 20 | "Software"), to deal in the Software without restriction, including |
| 21 | without limitation the rights to use, copy, modify, merge, publish, |
| 22 | distribute, sublicense, and/or sell copies of the Software, and to |
| 23 | permit persons to whom the Software is furnished to do so, subject to |
| 24 | the following conditions: |
| 25 | |
| 26 | The above copyright notice and this permission notice shall be |
| 27 | included in all copies or substantial portions of the Software. |
| 28 | |
| 29 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 30 | EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 31 | MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 32 | NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE |
| 33 | LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION |
| 34 | OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION |
| 35 | WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
| 36 | */ |
| 37 | |
| 38 | #ifndef included_clib_string_h |
| 39 | #define included_clib_string_h |
| 40 | |
Dave Barach | c379999 | 2016-08-15 11:12:27 -0400 | [diff] [blame] | 41 | #include <vppinfra/clib.h> /* for CLIB_LINUX_KERNEL */ |
Damjan Marion | e319de0 | 2016-10-21 19:30:42 +0200 | [diff] [blame] | 42 | #include <vppinfra/vector.h> |
Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 43 | |
| 44 | #ifdef CLIB_LINUX_KERNEL |
| 45 | #include <linux/string.h> |
| 46 | #endif |
| 47 | |
| 48 | #ifdef CLIB_UNIX |
| 49 | #include <string.h> |
| 50 | #endif |
| 51 | |
| 52 | #ifdef CLIB_STANDALONE |
| 53 | #include <vppinfra/standalone_string.h> |
| 54 | #endif |
| 55 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 56 | #if _x86_64_ |
| 57 | #include <x86intrin.h> |
| 58 | #endif |
| 59 | |
Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 60 | /* Exchanges source and destination. */ |
Dave Barach | c379999 | 2016-08-15 11:12:27 -0400 | [diff] [blame] | 61 | void clib_memswap (void *_a, void *_b, uword bytes); |
Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 62 | |
Dave Barach | d4048a4 | 2016-11-07 09:55:55 -0500 | [diff] [blame] | 63 | /* |
| 64 | * the vector unit memcpy variants confuse coverity |
| 65 | * so don't let it anywhere near them. |
| 66 | */ |
| 67 | #ifndef __COVERITY__ |
Damjan Marion | f1213b8 | 2016-03-13 02:22:06 +0100 | [diff] [blame] | 68 | #if __AVX__ |
| 69 | #include <vppinfra/memcpy_avx.h> |
Damjan Marion | 793b18d | 2016-05-16 16:52:55 +0200 | [diff] [blame] | 70 | #elif __SSSE3__ |
Damjan Marion | f1213b8 | 2016-03-13 02:22:06 +0100 | [diff] [blame] | 71 | #include <vppinfra/memcpy_sse3.h> |
| 72 | #else |
| 73 | #define clib_memcpy(a,b,c) memcpy(a,b,c) |
| 74 | #endif |
Dave Barach | d4048a4 | 2016-11-07 09:55:55 -0500 | [diff] [blame] | 75 | #else /* __COVERITY__ */ |
| 76 | #define clib_memcpy(a,b,c) memcpy(a,b,c) |
| 77 | #endif |
Damjan Marion | f1213b8 | 2016-03-13 02:22:06 +0100 | [diff] [blame] | 78 | |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 79 | /* |
| 80 | * Copy 64 bytes of data to 4 destinations |
| 81 | * this function is typically used in quad-loop case when whole cacheline |
| 82 | * needs to be copied to 4 different places. First it reads whole cacheline |
| 83 | * to 1/2/4 SIMD registers and then it writes data to 4 destinations. |
| 84 | */ |
| 85 | |
| 86 | static_always_inline void |
| 87 | clib_memcpy64_x4 (void *d0, void *d1, void *d2, void *d3, void *s) |
| 88 | { |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 89 | #if defined (__AVX512F__) |
| 90 | __m512i r0 = _mm512_loadu_si512 (s); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 91 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 92 | _mm512_storeu_si512 (d0, r0); |
| 93 | _mm512_storeu_si512 (d1, r0); |
| 94 | _mm512_storeu_si512 (d2, r0); |
| 95 | _mm512_storeu_si512 (d3, r0); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 96 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 97 | #elif defined (__AVX2__) |
| 98 | __m256i r0 = _mm256_loadu_si256 ((__m256i *) s + 0 * 32); |
| 99 | __m256i r1 = _mm256_loadu_si256 ((__m256i *) s + 1 * 32); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 100 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 101 | _mm256_storeu_si256 ((__m256i *) d0 + 0 * 32, r0); |
| 102 | _mm256_storeu_si256 ((__m256i *) d0 + 1 * 32, r1); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 103 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 104 | _mm256_storeu_si256 ((__m256i *) d1 + 0 * 32, r0); |
| 105 | _mm256_storeu_si256 ((__m256i *) d1 + 1 * 32, r1); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 106 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 107 | _mm256_storeu_si256 ((__m256i *) d2 + 0 * 32, r0); |
| 108 | _mm256_storeu_si256 ((__m256i *) d2 + 1 * 32, r1); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 109 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 110 | _mm256_storeu_si256 ((__m256i *) d3 + 0 * 32, r0); |
| 111 | _mm256_storeu_si256 ((__m256i *) d3 + 1 * 32, r1); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 112 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 113 | #elif defined (__SSSE3__) |
| 114 | __m128i r0 = _mm_loadu_si128 ((__m128i *) s + 0 * 16); |
| 115 | __m128i r1 = _mm_loadu_si128 ((__m128i *) s + 1 * 16); |
| 116 | __m128i r2 = _mm_loadu_si128 ((__m128i *) s + 2 * 16); |
| 117 | __m128i r3 = _mm_loadu_si128 ((__m128i *) s + 3 * 16); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 118 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 119 | _mm_storeu_si128 ((__m128i *) d0 + 0 * 16, r0); |
| 120 | _mm_storeu_si128 ((__m128i *) d0 + 1 * 16, r1); |
| 121 | _mm_storeu_si128 ((__m128i *) d0 + 2 * 16, r2); |
| 122 | _mm_storeu_si128 ((__m128i *) d0 + 3 * 16, r3); |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 123 | |
Damjan Marion | b2e1fe9 | 2017-11-22 12:41:32 +0100 | [diff] [blame^] | 124 | _mm_storeu_si128 ((__m128i *) d1 + 0 * 16, r0); |
| 125 | _mm_storeu_si128 ((__m128i *) d1 + 1 * 16, r1); |
| 126 | _mm_storeu_si128 ((__m128i *) d1 + 2 * 16, r2); |
| 127 | _mm_storeu_si128 ((__m128i *) d1 + 3 * 16, r3); |
| 128 | |
| 129 | _mm_storeu_si128 ((__m128i *) d2 + 0 * 16, r0); |
| 130 | _mm_storeu_si128 ((__m128i *) d2 + 1 * 16, r1); |
| 131 | _mm_storeu_si128 ((__m128i *) d2 + 2 * 16, r2); |
| 132 | _mm_storeu_si128 ((__m128i *) d2 + 3 * 16, r3); |
| 133 | |
| 134 | _mm_storeu_si128 ((__m128i *) d3 + 0 * 16, r0); |
| 135 | _mm_storeu_si128 ((__m128i *) d3 + 1 * 16, r1); |
| 136 | _mm_storeu_si128 ((__m128i *) d3 + 2 * 16, r2); |
| 137 | _mm_storeu_si128 ((__m128i *) d3 + 3 * 16, r3); |
| 138 | |
Damjan Marion | 04f3db3 | 2017-11-10 21:55:45 +0100 | [diff] [blame] | 139 | #else |
| 140 | clib_memcpy (d0, s, 64); |
| 141 | clib_memcpy (d1, s, 64); |
| 142 | clib_memcpy (d2, s, 64); |
| 143 | clib_memcpy (d3, s, 64); |
| 144 | #endif |
| 145 | } |
| 146 | |
Ed Warnicke | cb9cada | 2015-12-08 15:45:58 -0700 | [diff] [blame] | 147 | #endif /* included_clib_string_h */ |
Dave Barach | c379999 | 2016-08-15 11:12:27 -0400 | [diff] [blame] | 148 | |
| 149 | /* |
| 150 | * fd.io coding-style-patch-verification: ON |
| 151 | * |
| 152 | * Local Variables: |
| 153 | * eval: (c-set-style "gnu") |
| 154 | * End: |
| 155 | */ |