summaryrefslogtreecommitdiffstats
path: root/libmariadb/external/zlib/chunkcopy.h
blob: 8fd1a0d39f819ef253c97871c12b372e2571a72a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
/* chunkcopy.h -- fast chunk copy and set operations
 *
 * (C) 1995-2013 Jean-loup Gailly and Mark Adler
 *
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 *
 * Jean-loup Gailly        Mark Adler
 * jloup@gzip.org          madler@alumni.caltech.edu
 *
 * Copyright (C) 2017 ARM, Inc.
 * Copyright 2017 The Chromium Authors. All rights reserved.
 * Use of this source code is governed by a BSD-style license that can be
 * found in the Chromium source repository LICENSE file.
 */

#ifndef CHUNKCOPY_H
#define CHUNKCOPY_H

#include <stdint.h>
#include "zutil.h"

#define Z_STATIC_ASSERT(name, assert) typedef char name[(assert) ? 1 : -1]

#if __STDC_VERSION__ >= 199901L
#define Z_RESTRICT restrict
#else
#define Z_RESTRICT
#endif

#if defined(__clang__) || defined(__GNUC__) || defined(__llvm__)
#define Z_BUILTIN_MEMCPY __builtin_memcpy
#else
#define Z_BUILTIN_MEMCPY zmemcpy
#endif

#if defined(INFLATE_CHUNK_SIMD_NEON)
#include <arm_neon.h>
typedef uint8x16_t z_vec128i_t;
#elif defined(INFLATE_CHUNK_SIMD_SSE2)
#include <emmintrin.h>
typedef __m128i z_vec128i_t;
#else
#error chunkcopy.h inflate chunk SIMD is not defined for your build target
#endif

/*
 * chunk copy type: the z_vec128i_t type size should be exactly 128-bits
 * and equal to CHUNKCOPY_CHUNK_SIZE.
 */
#define CHUNKCOPY_CHUNK_SIZE sizeof(z_vec128i_t)

Z_STATIC_ASSERT(vector_128_bits_wide,
                CHUNKCOPY_CHUNK_SIZE == sizeof(int8_t) * 16);

/*
 * Ask the compiler to perform a wide, unaligned load with a machine
 * instruction appropriate for the z_vec128i_t type.
 */
static inline z_vec128i_t loadchunk(
    const unsigned char FAR* s) {
  z_vec128i_t v;
  Z_BUILTIN_MEMCPY(&v, s, sizeof(v));
  return v;
}

/*
 * Ask the compiler to perform a wide, unaligned store with a machine
 * instruction appropriate for the z_vec128i_t type.
 */
static inline void storechunk(
    unsigned char FAR* d,
    const z_vec128i_t v) {
  Z_BUILTIN_MEMCPY(d, &v, sizeof(v));
}

/*
 * Perform a memcpy-like operation, assuming that length is non-zero and that
 * it's OK to overwrite at least CHUNKCOPY_CHUNK_SIZE bytes of output even if
 * the length is shorter than this.
 *
 * It also guarantees that it will properly unroll the data if the distance
 * between `out` and `from` is at least CHUNKCOPY_CHUNK_SIZE, which we rely on
 * in chunkcopy_relaxed().
 *
 * Aside from better memory bus utilisation, this means that short copies
 * (CHUNKCOPY_CHUNK_SIZE bytes or fewer) will fall straight through the loop
 * without iteration, which will hopefully make the branch prediction more
 * reliable.
 */
static inline unsigned char FAR* chunkcopy_core(
    unsigned char FAR* out,
    const unsigned char FAR* from,
    unsigned len) {
  const int bump = (--len % CHUNKCOPY_CHUNK_SIZE) + 1;
  storechunk(out, loadchunk(from));
  out += bump;
  from += bump;
  len /= CHUNKCOPY_CHUNK_SIZE;
  while (len-- > 0) {
    storechunk(out, loadchunk(from));
    out += CHUNKCOPY_CHUNK_SIZE;
    from += CHUNKCOPY_CHUNK_SIZE;
  }
  return out;
}

/*
 * Like chunkcopy_core(), but avoid writing beyond of legal output.
 *
 * Accepts an additional pointer to the end of safe output.  A generic safe
 * copy would use (out + len), but it's normally the case that the end of the
 * output buffer is beyond the end of the current copy, and this can still be
 * exploited.
 */
static inline unsigned char FAR* chunkcopy_core_safe(
    unsigned char FAR* out,
    const unsigned char FAR* from,
    unsigned len,
    unsigned char FAR* limit) {
  Assert(out + len <= limit, "chunk copy exceeds safety limit");
  if ((limit - out) < (ptrdiff_t)CHUNKCOPY_CHUNK_SIZE) {
    const unsigned char FAR* Z_RESTRICT rfrom = from;
    if (len & 8) {
      Z_BUILTIN_MEMCPY(out, rfrom, 8);
      out += 8;
      rfrom += 8;
    }
    if (len & 4) {
      Z_BUILTIN_MEMCPY(out, rfrom, 4);
      out += 4;
      rfrom += 4;
    }
    if (len & 2) {
      Z_BUILTIN_MEMCPY(out, rfrom, 2);
      out += 2;
      rfrom += 2;
    }
    if (len & 1) {
      *out++ = *rfrom++;
    }
    return out;
  }
  return chunkcopy_core(out, from, len);
}

/*
 * Perform short copies until distance can be rewritten as being at least
 * CHUNKCOPY_CHUNK_SIZE.
 *
 * Assumes it's OK to overwrite at least the first 2*CHUNKCOPY_CHUNK_SIZE
 * bytes of output even if the copy is shorter than this.  This assumption
 * holds within zlib inflate_fast(), which starts every iteration with at
 * least 258 bytes of output space available (258 being the maximum length
 * output from a single token; see inffast.c).
 */
static inline unsigned char FAR* chunkunroll_relaxed(
    unsigned char FAR* out,
    unsigned FAR* dist,
    unsigned FAR* len) {
  const unsigned char FAR* from = out - *dist;
  while (*dist < *len && *dist < CHUNKCOPY_CHUNK_SIZE) {
    storechunk(out, loadchunk(from));
    out += *dist;
    *len -= *dist;
    *dist += *dist;
  }
  return out;
}

#if defined(INFLATE_CHUNK_SIMD_NEON)
/*
 * v_load64_dup(): load *src as an unaligned 64-bit int and duplicate it in
 * every 64-bit component of the 128-bit result (64-bit int splat).
 */
static inline z_vec128i_t v_load64_dup(const void* src) {
  return vcombine_u8(vld1_u8(src), vld1_u8(src));
}

/*
 * v_load32_dup(): load *src as an unaligned 32-bit int and duplicate it in
 * every 32-bit component of the 128-bit result (32-bit int splat).
 */
static inline z_vec128i_t v_load32_dup(const void* src) {
  int32_t i32;
  Z_BUILTIN_MEMCPY(&i32, src, sizeof(i32));
  return vreinterpretq_u8_s32(vdupq_n_s32(i32));
}

/*
 * v_load16_dup(): load *src as an unaligned 16-bit int and duplicate it in
 * every 16-bit component of the 128-bit result (16-bit int splat).
 */
static inline z_vec128i_t v_load16_dup(const void* src) {
  int16_t i16;
  Z_BUILTIN_MEMCPY(&i16, src, sizeof(i16));
  return vreinterpretq_u8_s16(vdupq_n_s16(i16));
}

/*
 * v_load8_dup(): load the 8-bit int *src and duplicate it in every 8-bit
 * component of the 128-bit result (8-bit int splat).
 */
static inline z_vec128i_t v_load8_dup(const void* src) {
  return vld1q_dup_u8((const uint8_t*)src);
}

/*
 * v_store_128(): store the 128-bit vec in a memory destination (that might
 * not be 16-byte aligned) void* out.
 */
static inline void v_store_128(void* out, const z_vec128i_t vec) {
  vst1q_u8(out, vec);
}
#elif defined (INFLATE_CHUNK_SIMD_SSE2)
/*
 * v_load64_dup(): load *src as an unaligned 64-bit int and duplicate it in
 * every 64-bit component of the 128-bit result (64-bit int splat).
 */
static inline z_vec128i_t v_load64_dup(const void* src) {
  int64_t i64;
  Z_BUILTIN_MEMCPY(&i64, src, sizeof(i64));
  return _mm_set1_epi64x(i64);
}

/*
 * v_load32_dup(): load *src as an unaligned 32-bit int and duplicate it in
 * every 32-bit component of the 128-bit result (32-bit int splat).
 */
static inline z_vec128i_t v_load32_dup(const void* src) {
  int32_t i32;
  Z_BUILTIN_MEMCPY(&i32, src, sizeof(i32));
  return _mm_set1_epi32(i32);
}

/*
 * v_load16_dup(): load *src as an unaligned 16-bit int and duplicate it in
 * every 16-bit component of the 128-bit result (16-bit int splat).
 */
static inline z_vec128i_t v_load16_dup(const void* src) {
  int16_t i16;
  Z_BUILTIN_MEMCPY(&i16, src, sizeof(i16));
  return _mm_set1_epi16(i16);
}

/*
 * v_load8_dup(): load the 8-bit int *src and duplicate it in every 8-bit
 * component of the 128-bit result (8-bit int splat).
 */
static inline z_vec128i_t v_load8_dup(const void* src) {
  return _mm_set1_epi8(*(const char*)src);
}

/*
 * v_store_128(): store the 128-bit vec in a memory destination (that might
 * not be 16-byte aligned) void* out.
 */
static inline void v_store_128(void* out, const z_vec128i_t vec) {
  _mm_storeu_si128((__m128i*)out, vec);
}
#endif

/*
 * Perform an overlapping copy which behaves as a memset() operation, but
 * supporting periods other than one, and assume that length is non-zero and
 * that it's OK to overwrite at least CHUNKCOPY_CHUNK_SIZE*3 bytes of output
 * even if the length is shorter than this.
 */
static inline unsigned char FAR* chunkset_core(
    unsigned char FAR* out,
    unsigned period,
    unsigned len) {
  z_vec128i_t v;
  const int bump = ((len - 1) % sizeof(v)) + 1;

  switch (period) {
    case 1:
      v = v_load8_dup(out - 1);
      v_store_128(out, v);
      out += bump;
      len -= bump;
      while (len > 0) {
        v_store_128(out, v);
        out += sizeof(v);
        len -= sizeof(v);
      }
      return out;
    case 2:
      v = v_load16_dup(out - 2);
      v_store_128(out, v);
      out += bump;
      len -= bump;
      if (len > 0) {
        v = v_load16_dup(out - 2);
        do {
          v_store_128(out, v);
          out += sizeof(v);
          len -= sizeof(v);
        } while (len > 0);
      }
      return out;
    case 4:
      v = v_load32_dup(out - 4);
      v_store_128(out, v);
      out += bump;
      len -= bump;
      if (len > 0) {
        v = v_load32_dup(out - 4);
        do {
          v_store_128(out, v);
          out += sizeof(v);
          len -= sizeof(v);
        } while (len > 0);
      }
      return out;
    case 8:
      v = v_load64_dup(out - 8);
      v_store_128(out, v);
      out += bump;
      len -= bump;
      if (len > 0) {
        v = v_load64_dup(out - 8);
        do {
          v_store_128(out, v);
          out += sizeof(v);
          len -= sizeof(v);
        } while (len > 0);
      }
      return out;
  }
  out = chunkunroll_relaxed(out, &period, &len);
  return chunkcopy_core(out, out - period, len);
}

/*
 * Perform a memcpy-like operation, but assume that length is non-zero and that
 * it's OK to overwrite at least CHUNKCOPY_CHUNK_SIZE bytes of output even if
 * the length is shorter than this.
 *
 * Unlike chunkcopy_core() above, no guarantee is made regarding the behaviour
 * of overlapping buffers, regardless of the distance between the pointers.
 * This is reflected in the `restrict`-qualified pointers, allowing the
 * compiler to re-order loads and stores.
 */
static inline unsigned char FAR* chunkcopy_relaxed(
    unsigned char FAR* Z_RESTRICT out,
    const unsigned char FAR* Z_RESTRICT from,
    unsigned len) {
  return chunkcopy_core(out, from, len);
}

/*
 * Like chunkcopy_relaxed(), but avoid writing beyond of legal output.
 *
 * Unlike chunkcopy_core_safe() above, no guarantee is made regarding the
 * behaviour of overlapping buffers, regardless of the distance between the
 * pointers.  This is reflected in the `restrict`-qualified pointers, allowing
 * the compiler to re-order loads and stores.
 *
 * Accepts an additional pointer to the end of safe output.  A generic safe
 * copy would use (out + len), but it's normally the case that the end of the
 * output buffer is beyond the end of the current copy, and this can still be
 * exploited.
 */
static inline unsigned char FAR* chunkcopy_safe(
    unsigned char FAR* out,
    const unsigned char FAR* Z_RESTRICT from,
    unsigned len,
    unsigned char FAR* limit) {
  Assert(out + len <= limit, "chunk copy exceeds safety limit");
  return chunkcopy_core_safe(out, from, len, limit);
}

/*
 * Perform chunky copy within the same buffer, where the source and destination
 * may potentially overlap.
 *
 * Assumes that len > 0 on entry, and that it's safe to write at least
 * CHUNKCOPY_CHUNK_SIZE*3 bytes to the output.
 */
static inline unsigned char FAR* chunkcopy_lapped_relaxed(
    unsigned char FAR* out,
    unsigned dist,
    unsigned len) {
  if (dist < len && dist < CHUNKCOPY_CHUNK_SIZE) {
    return chunkset_core(out, dist, len);
  }
  return chunkcopy_core(out, out - dist, len);
}

/*
 * Behave like chunkcopy_lapped_relaxed(), but avoid writing beyond of legal
 * output.
 *
 * Accepts an additional pointer to the end of safe output.  A generic safe
 * copy would use (out + len), but it's normally the case that the end of the
 * output buffer is beyond the end of the current copy, and this can still be
 * exploited.
 */
static inline unsigned char FAR* chunkcopy_lapped_safe(
    unsigned char FAR* out,
    unsigned dist,
    unsigned len,
    unsigned char FAR* limit) {
  Assert(out + len <= limit, "chunk copy exceeds safety limit");
  if ((limit - out) < (ptrdiff_t)(3 * CHUNKCOPY_CHUNK_SIZE)) {
    /* TODO(cavalcantii): try harder to optimise this */
    while (len-- > 0) {
      *out = *(out - dist);
      out++;
    }
    return out;
  }
  return chunkcopy_lapped_relaxed(out, dist, len);
}

/* TODO(cavalcanti): see crbug.com/1110083. */
static inline unsigned char FAR* chunkcopy_safe_ugly(unsigned char FAR* out,
                                                     unsigned dist,
                                                     unsigned len,
                                                     unsigned char FAR* limit) {
#if defined(__GNUC__) && !defined(__clang__)
  /* Speed is the same as using chunkcopy_safe
     w/ GCC on ARM (tested gcc 6.3 and 7.5) and avoids
     undefined behavior.
  */
  return chunkcopy_core_safe(out, out - dist, len, limit);
#elif defined(__clang__) && !defined(__aarch64__)
  /* Seems to perform better on 32bit (i.e. Android). */
  return chunkcopy_core_safe(out, out - dist, len, limit);
#else
  /* Seems to perform better on 64-bit. */
  return chunkcopy_lapped_safe(out, dist, len, limit);
#endif
}

/*
 * The chunk-copy code above deals with writing the decoded DEFLATE data to
 * the output with SIMD methods to increase decode speed. Reading the input
 * to the DEFLATE decoder with a wide, SIMD method can also increase decode
 * speed. This option is supported on little endian machines, and reads the
 * input data in 64-bit (8 byte) chunks.
 */

#ifdef INFLATE_CHUNK_READ_64LE
/*
 * Buffer the input in a uint64_t (8 bytes) in the wide input reading case.
 */
typedef uint64_t inflate_holder_t;

/*
 * Ask the compiler to perform a wide, unaligned load of a uint64_t using a
 * machine instruction appropriate for the uint64_t type.
 */
static inline inflate_holder_t read64le(const unsigned char FAR *in) {
    inflate_holder_t input;
    Z_BUILTIN_MEMCPY(&input, in, sizeof(input));
    return input;
}
#else
/*
 * Otherwise, buffer the input bits using zlib's default input buffer type.
 */
typedef unsigned long inflate_holder_t;

#endif /* INFLATE_CHUNK_READ_64LE */

#undef Z_STATIC_ASSERT
#undef Z_RESTRICT
#undef Z_BUILTIN_MEMCPY

#endif /* CHUNKCOPY_H */