1/*
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/*-
16 *   BSD LICENSE
17 *
18 *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
19 *   All rights reserved.
20 *
21 *   Redistribution and use in source and binary forms, with or without
22 *   modification, are permitted provided that the following conditions
23 *   are met:
24 *
25 *     * Redistributions of source code must retain the above copyright
26 *       notice, this list of conditions and the following disclaimer.
27 *     * Redistributions in binary form must reproduce the above copyright
28 *       notice, this list of conditions and the following disclaimer in
29 *       the documentation and/or other materials provided with the
30 *       distribution.
31 *     * Neither the name of Intel Corporation nor the names of its
32 *       contributors may be used to endorse or promote products derived
33 *       from this software without specific prior written permission.
34 *
35 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
36 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
37 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
38 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
39 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
40 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
41 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
42 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
43 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
44 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
45 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
46 */
47
48#ifndef included_clib_memcpy_avx2_h
49#define included_clib_memcpy_avx2_h
50
51#include <stdint.h>
52#include <x86intrin.h>
53#include <vppinfra/warnings.h>
54
55/* *INDENT-OFF* */
56WARN_OFF (stringop-overflow)
57/* *INDENT-ON* */
58
59static inline void
60clib_mov16 (u8 * dst, const u8 * src)
61{
62  __m128i xmm0;
63
64  xmm0 = _mm_loadu_si128 ((const __m128i *) src);
65  _mm_storeu_si128 ((__m128i *) dst, xmm0);
66}
67
68static inline void
69clib_mov32 (u8 * dst, const u8 * src)
70{
71  __m256i ymm0;
72
73  ymm0 = _mm256_loadu_si256 ((const __m256i *) src);
74  _mm256_storeu_si256 ((__m256i *) dst, ymm0);
75}
76
77static inline void
78clib_mov64 (u8 * dst, const u8 * src)
79{
80  clib_mov32 ((u8 *) dst + 0 * 32, (const u8 *) src + 0 * 32);
81  clib_mov32 ((u8 *) dst + 1 * 32, (const u8 *) src + 1 * 32);
82}
83
84static inline void
85clib_mov128 (u8 * dst, const u8 * src)
86{
87  clib_mov64 ((u8 *) dst + 0 * 64, (const u8 *) src + 0 * 64);
88  clib_mov64 ((u8 *) dst + 1 * 64, (const u8 *) src + 1 * 64);
89}
90
91static inline void
92clib_mov128blocks (u8 * dst, const u8 * src, size_t n)
93{
94  __m256i ymm0, ymm1, ymm2, ymm3;
95
96  while (n >= 128)
97    {
98      ymm0 =
99	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 0 * 32));
100      n -= 128;
101      ymm1 =
102	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 1 * 32));
103      ymm2 =
104	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 2 * 32));
105      ymm3 =
106	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 3 * 32));
107      src = (const u8 *) src + 128;
108      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 0 * 32), ymm0);
109      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 1 * 32), ymm1);
110      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 2 * 32), ymm2);
111      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 3 * 32), ymm3);
112      dst = (u8 *) dst + 128;
113    }
114}
115
116static inline void *
117clib_memcpy_fast (void *dst, const void *src, size_t n)
118{
119  uword dstu = (uword) dst;
120  uword srcu = (uword) src;
121  void *ret = dst;
122  size_t dstofss;
123  size_t bits;
124
125  /**
126   * Copy less than 16 bytes
127   */
128  if (n < 16)
129    {
130      if (n & 0x01)
131	{
132	  *(u8 *) dstu = *(const u8 *) srcu;
133	  srcu = (uword) ((const u8 *) srcu + 1);
134	  dstu = (uword) ((u8 *) dstu + 1);
135	}
136      if (n & 0x02)
137	{
138	  *(u16 *) dstu = *(const u16 *) srcu;
139	  srcu = (uword) ((const u16 *) srcu + 1);
140	  dstu = (uword) ((u16 *) dstu + 1);
141	}
142      if (n & 0x04)
143	{
144	  *(u32 *) dstu = *(const u32 *) srcu;
145	  srcu = (uword) ((const u32 *) srcu + 1);
146	  dstu = (uword) ((u32 *) dstu + 1);
147	}
148      if (n & 0x08)
149	{
150	  *(u64 *) dstu = *(const u64 *) srcu;
151	}
152      return ret;
153    }
154
155  /**
156    * Fast way when copy size doesn't exceed 512 bytes
157    */
158  if (n <= 32)
159    {
160      clib_mov16 ((u8 *) dst, (const u8 *) src);
161      clib_mov16 ((u8 *) dst - 16 + n, (const u8 *) src - 16 + n);
162      return ret;
163    }
164  if (n <= 48)
165    {
166      clib_mov16 ((u8 *) dst, (const u8 *) src);
167      clib_mov16 ((u8 *) dst + 16, (const u8 *) src + 16);
168      clib_mov16 ((u8 *) dst - 16 + n, (const u8 *) src - 16 + n);
169      return ret;
170    }
171  if (n <= 64)
172    {
173      clib_mov32 ((u8 *) dst, (const u8 *) src);
174      clib_mov32 ((u8 *) dst - 32 + n, (const u8 *) src - 32 + n);
175      return ret;
176    }
177  if (n <= 256)
178    {
179      if (n >= 128)
180	{
181	  n -= 128;
182	  clib_mov128 ((u8 *) dst, (const u8 *) src);
183	  src = (const u8 *) src + 128;
184	  dst = (u8 *) dst + 128;
185	}
186    COPY_BLOCK_128_BACK31:
187      if (n >= 64)
188	{
189	  n -= 64;
190	  clib_mov64 ((u8 *) dst, (const u8 *) src);
191	  src = (const u8 *) src + 64;
192	  dst = (u8 *) dst + 64;
193	}
194      if (n > 32)
195	{
196	  clib_mov32 ((u8 *) dst, (const u8 *) src);
197	  clib_mov32 ((u8 *) dst - 32 + n, (const u8 *) src - 32 + n);
198	  return ret;
199	}
200      if (n > 0)
201	{
202	  clib_mov32 ((u8 *) dst - 32 + n, (const u8 *) src - 32 + n);
203	}
204      return ret;
205    }
206
207    /**
208      * Make store aligned when copy size exceeds 256 bytes
209      */
210  dstofss = (uword) dst & 0x1F;
211  if (dstofss > 0)
212    {
213      dstofss = 32 - dstofss;
214      n -= dstofss;
215      clib_mov32 ((u8 *) dst, (const u8 *) src);
216      src = (const u8 *) src + dstofss;
217      dst = (u8 *) dst + dstofss;
218    }
219
220  /**
221    * Copy 128-byte blocks.
222    */
223  clib_mov128blocks ((u8 *) dst, (const u8 *) src, n);
224  bits = n;
225  n = n & 127;
226  bits -= n;
227  src = (const u8 *) src + bits;
228  dst = (u8 *) dst + bits;
229
230  /**
231   * Copy whatever left
232   */
233  goto COPY_BLOCK_128_BACK31;
234}
235
236/* *INDENT-OFF* */
237WARN_ON (stringop-overflow)
238/* *INDENT-ON* */
239
240#endif /* included_clib_memcpy_avx2_h */
241
242
243/*
244 * fd.io coding-style-patch-verification: ON
245 *
246 * Local Variables:
247 * eval: (c-set-style "gnu")
248 * End:
249 */
250