1 /*
2 * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
3 * All rights reserved.
4 *
5 * This source code is licensed under both the BSD-style license (found in the
6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7 * in the COPYING file in the root directory of this source tree).
8 * You may select, at your option, one of the above-listed licenses.
9 */
10
11 #ifndef MEM_H_MODULE
12 #define MEM_H_MODULE
13
14 #if defined (__cplusplus)
15 extern "C" {
16 #endif
17
18 /*-****************************************
19 * Dependencies
20 ******************************************/
21 #include <stddef.h> /* size_t, ptrdiff_t */
22 #include "compiler.h" /* __has_builtin */
23 #include "debug.h" /* DEBUG_STATIC_ASSERT */
24 #include "zstd_deps.h" /* ZSTD_memcpy */
25
26
27 /*-****************************************
28 * Compiler specifics
29 ******************************************/
30 #if defined(_MSC_VER) /* Visual Studio */
31 # include <stdlib.h> /* _byteswap_ulong */
32 # include <intrin.h> /* _byteswap_* */
33 #endif
34 #if defined(__GNUC__)
35 # define MEM_STATIC static __inline __attribute__((unused))
36 #elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
37 # define MEM_STATIC static inline
38 #elif defined(_MSC_VER)
39 # define MEM_STATIC static __inline
40 #else
41 # define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
42 #endif
43
44 /*-**************************************************************
45 * Basic Types
46 *****************************************************************/
47 #if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
48 # if defined(_AIX)
49 # include <inttypes.h>
50 # else
51 # include <stdint.h> /* intptr_t */
52 # endif
53 typedef uint8_t BYTE;
54 typedef uint16_t U16;
55 typedef int16_t S16;
56 typedef uint32_t U32;
57 typedef int32_t S32;
58 typedef uint64_t U64;
59 typedef int64_t S64;
60 #else
61 # include <limits.h>
62 #if CHAR_BIT != 8
63 # error "this implementation requires char to be exactly 8-bit type"
64 #endif
65 typedef unsigned char BYTE;
66 #if USHRT_MAX != 65535
67 # error "this implementation requires short to be exactly 16-bit type"
68 #endif
69 typedef unsigned short U16;
70 typedef signed short S16;
71 #if UINT_MAX != 4294967295
72 # error "this implementation requires int to be exactly 32-bit type"
73 #endif
74 typedef unsigned int U32;
75 typedef signed int S32;
76 /* note : there are no limits defined for long long type in C90.
77 * limits exist in C99, however, in such case, <stdint.h> is preferred */
78 typedef unsigned long long U64;
79 typedef signed long long S64;
80 #endif
81
82
83 /*-**************************************************************
84 * Memory I/O API
85 *****************************************************************/
86 /*=== Static platform detection ===*/
87 MEM_STATIC unsigned MEM_32bits(void);
88 MEM_STATIC unsigned MEM_64bits(void);
89 MEM_STATIC unsigned MEM_isLittleEndian(void);
90
91 /*=== Native unaligned read/write ===*/
92 MEM_STATIC U16 MEM_read16(const void* memPtr);
93 MEM_STATIC U32 MEM_read32(const void* memPtr);
94 MEM_STATIC U64 MEM_read64(const void* memPtr);
95 MEM_STATIC size_t MEM_readST(const void* memPtr);
96
97 MEM_STATIC void MEM_write16(void* memPtr, U16 value);
98 MEM_STATIC void MEM_write32(void* memPtr, U32 value);
99 MEM_STATIC void MEM_write64(void* memPtr, U64 value);
100
101 /*=== Little endian unaligned read/write ===*/
102 MEM_STATIC U16 MEM_readLE16(const void* memPtr);
103 MEM_STATIC U32 MEM_readLE24(const void* memPtr);
104 MEM_STATIC U32 MEM_readLE32(const void* memPtr);
105 MEM_STATIC U64 MEM_readLE64(const void* memPtr);
106 MEM_STATIC size_t MEM_readLEST(const void* memPtr);
107
108 MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val);
109 MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val);
110 MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32);
111 MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64);
112 MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val);
113
114 /*=== Big endian unaligned read/write ===*/
115 MEM_STATIC U32 MEM_readBE32(const void* memPtr);
116 MEM_STATIC U64 MEM_readBE64(const void* memPtr);
117 MEM_STATIC size_t MEM_readBEST(const void* memPtr);
118
119 MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32);
120 MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64);
121 MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val);
122
123 /*=== Byteswap ===*/
124 MEM_STATIC U32 MEM_swap32(U32 in);
125 MEM_STATIC U64 MEM_swap64(U64 in);
126 MEM_STATIC size_t MEM_swapST(size_t in);
127
128
129 /*-**************************************************************
130 * Memory I/O Implementation
131 *****************************************************************/
132 /* MEM_FORCE_MEMORY_ACCESS :
133 * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
134 * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
135 * The below switch allow to select different access method for improved performance.
136 * Method 0 (default) : use `memcpy()`. Safe and portable.
137 * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
138 * This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
139 * Method 2 : direct access. This method is portable but violate C standard.
140 * It can generate buggy code on targets depending on alignment.
141 * In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
142 * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
143 * Prefer these methods in priority order (0 > 1 > 2)
144 */
145 #ifndef MEM_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
146 # if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
147 # define MEM_FORCE_MEMORY_ACCESS 2
148 # elif defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
149 # define MEM_FORCE_MEMORY_ACCESS 1
150 # endif
151 #endif
152
MEM_32bits(void)153 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
MEM_64bits(void)154 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
155
MEM_isLittleEndian(void)156 MEM_STATIC unsigned MEM_isLittleEndian(void)
157 {
158 const union { U32 u; BYTE c[4]; } one = { 1 }; /* don't use static : performance detrimental */
159 return one.c[0];
160 }
161
162 #if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
163
164 /* violates C standard, by lying on structure alignment.
165 Only use if no other choice to achieve best performance on target platform */
MEM_read16(const void * memPtr)166 MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
MEM_read32(const void * memPtr)167 MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
MEM_read64(const void * memPtr)168 MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
MEM_readST(const void * memPtr)169 MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
170
MEM_write16(void * memPtr,U16 value)171 MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
MEM_write32(void * memPtr,U32 value)172 MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
MEM_write64(void * memPtr,U64 value)173 MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
174
175 #elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
176
177 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
178 /* currently only defined for gcc and icc */
179 #if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
180 __pragma( pack(push, 1) )
181 typedef struct { U16 v; } unalign16;
182 typedef struct { U32 v; } unalign32;
183 typedef struct { U64 v; } unalign64;
184 typedef struct { size_t v; } unalignArch;
__pragma(pack (pop))185 __pragma( pack(pop) )
186 #else
187 typedef struct { U16 v; } __attribute__((packed)) unalign16;
188 typedef struct { U32 v; } __attribute__((packed)) unalign32;
189 typedef struct { U64 v; } __attribute__((packed)) unalign64;
190 typedef struct { size_t v; } __attribute__((packed)) unalignArch;
191 #endif
192
193 MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
MEM_read32(const void * ptr)194 MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
MEM_read64(const void * ptr)195 MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
MEM_readST(const void * ptr)196 MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
197
MEM_write16(void * memPtr,U16 value)198 MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
MEM_write32(void * memPtr,U32 value)199 MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
MEM_write64(void * memPtr,U64 value)200 MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
201
202 #else
203
204 /* default method, safe and standard.
205 can sometimes prove slower */
206
MEM_read16(const void * memPtr)207 MEM_STATIC U16 MEM_read16(const void* memPtr)
208 {
209 U16 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
210 }
211
MEM_read32(const void * memPtr)212 MEM_STATIC U32 MEM_read32(const void* memPtr)
213 {
214 U32 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
215 }
216
MEM_read64(const void * memPtr)217 MEM_STATIC U64 MEM_read64(const void* memPtr)
218 {
219 U64 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
220 }
221
MEM_readST(const void * memPtr)222 MEM_STATIC size_t MEM_readST(const void* memPtr)
223 {
224 size_t val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
225 }
226
MEM_write16(void * memPtr,U16 value)227 MEM_STATIC void MEM_write16(void* memPtr, U16 value)
228 {
229 ZSTD_memcpy(memPtr, &value, sizeof(value));
230 }
231
MEM_write32(void * memPtr,U32 value)232 MEM_STATIC void MEM_write32(void* memPtr, U32 value)
233 {
234 ZSTD_memcpy(memPtr, &value, sizeof(value));
235 }
236
MEM_write64(void * memPtr,U64 value)237 MEM_STATIC void MEM_write64(void* memPtr, U64 value)
238 {
239 ZSTD_memcpy(memPtr, &value, sizeof(value));
240 }
241
242 #endif /* MEM_FORCE_MEMORY_ACCESS */
243
MEM_swap32(U32 in)244 MEM_STATIC U32 MEM_swap32(U32 in)
245 {
246 #if defined(_MSC_VER) /* Visual Studio */
247 return _byteswap_ulong(in);
248 #elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
249 || (defined(__clang__) && __has_builtin(__builtin_bswap32))
250 return __builtin_bswap32(in);
251 #else
252 return ((in << 24) & 0xff000000 ) |
253 ((in << 8) & 0x00ff0000 ) |
254 ((in >> 8) & 0x0000ff00 ) |
255 ((in >> 24) & 0x000000ff );
256 #endif
257 }
258
MEM_swap64(U64 in)259 MEM_STATIC U64 MEM_swap64(U64 in)
260 {
261 #if defined(_MSC_VER) /* Visual Studio */
262 return _byteswap_uint64(in);
263 #elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
264 || (defined(__clang__) && __has_builtin(__builtin_bswap64))
265 return __builtin_bswap64(in);
266 #else
267 return ((in << 56) & 0xff00000000000000ULL) |
268 ((in << 40) & 0x00ff000000000000ULL) |
269 ((in << 24) & 0x0000ff0000000000ULL) |
270 ((in << 8) & 0x000000ff00000000ULL) |
271 ((in >> 8) & 0x00000000ff000000ULL) |
272 ((in >> 24) & 0x0000000000ff0000ULL) |
273 ((in >> 40) & 0x000000000000ff00ULL) |
274 ((in >> 56) & 0x00000000000000ffULL);
275 #endif
276 }
277
MEM_swapST(size_t in)278 MEM_STATIC size_t MEM_swapST(size_t in)
279 {
280 if (MEM_32bits())
281 return (size_t)MEM_swap32((U32)in);
282 else
283 return (size_t)MEM_swap64((U64)in);
284 }
285
286 /*=== Little endian r/w ===*/
287
MEM_readLE16(const void * memPtr)288 MEM_STATIC U16 MEM_readLE16(const void* memPtr)
289 {
290 if (MEM_isLittleEndian())
291 return MEM_read16(memPtr);
292 else {
293 const BYTE* p = (const BYTE*)memPtr;
294 return (U16)(p[0] + (p[1]<<8));
295 }
296 }
297
MEM_writeLE16(void * memPtr,U16 val)298 MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
299 {
300 if (MEM_isLittleEndian()) {
301 MEM_write16(memPtr, val);
302 } else {
303 BYTE* p = (BYTE*)memPtr;
304 p[0] = (BYTE)val;
305 p[1] = (BYTE)(val>>8);
306 }
307 }
308
MEM_readLE24(const void * memPtr)309 MEM_STATIC U32 MEM_readLE24(const void* memPtr)
310 {
311 return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
312 }
313
MEM_writeLE24(void * memPtr,U32 val)314 MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
315 {
316 MEM_writeLE16(memPtr, (U16)val);
317 ((BYTE*)memPtr)[2] = (BYTE)(val>>16);
318 }
319
MEM_readLE32(const void * memPtr)320 MEM_STATIC U32 MEM_readLE32(const void* memPtr)
321 {
322 if (MEM_isLittleEndian())
323 return MEM_read32(memPtr);
324 else
325 return MEM_swap32(MEM_read32(memPtr));
326 }
327
MEM_writeLE32(void * memPtr,U32 val32)328 MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32)
329 {
330 if (MEM_isLittleEndian())
331 MEM_write32(memPtr, val32);
332 else
333 MEM_write32(memPtr, MEM_swap32(val32));
334 }
335
MEM_readLE64(const void * memPtr)336 MEM_STATIC U64 MEM_readLE64(const void* memPtr)
337 {
338 if (MEM_isLittleEndian())
339 return MEM_read64(memPtr);
340 else
341 return MEM_swap64(MEM_read64(memPtr));
342 }
343
MEM_writeLE64(void * memPtr,U64 val64)344 MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64)
345 {
346 if (MEM_isLittleEndian())
347 MEM_write64(memPtr, val64);
348 else
349 MEM_write64(memPtr, MEM_swap64(val64));
350 }
351
MEM_readLEST(const void * memPtr)352 MEM_STATIC size_t MEM_readLEST(const void* memPtr)
353 {
354 if (MEM_32bits())
355 return (size_t)MEM_readLE32(memPtr);
356 else
357 return (size_t)MEM_readLE64(memPtr);
358 }
359
MEM_writeLEST(void * memPtr,size_t val)360 MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val)
361 {
362 if (MEM_32bits())
363 MEM_writeLE32(memPtr, (U32)val);
364 else
365 MEM_writeLE64(memPtr, (U64)val);
366 }
367
368 /*=== Big endian r/w ===*/
369
MEM_readBE32(const void * memPtr)370 MEM_STATIC U32 MEM_readBE32(const void* memPtr)
371 {
372 if (MEM_isLittleEndian())
373 return MEM_swap32(MEM_read32(memPtr));
374 else
375 return MEM_read32(memPtr);
376 }
377
MEM_writeBE32(void * memPtr,U32 val32)378 MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32)
379 {
380 if (MEM_isLittleEndian())
381 MEM_write32(memPtr, MEM_swap32(val32));
382 else
383 MEM_write32(memPtr, val32);
384 }
385
MEM_readBE64(const void * memPtr)386 MEM_STATIC U64 MEM_readBE64(const void* memPtr)
387 {
388 if (MEM_isLittleEndian())
389 return MEM_swap64(MEM_read64(memPtr));
390 else
391 return MEM_read64(memPtr);
392 }
393
MEM_writeBE64(void * memPtr,U64 val64)394 MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64)
395 {
396 if (MEM_isLittleEndian())
397 MEM_write64(memPtr, MEM_swap64(val64));
398 else
399 MEM_write64(memPtr, val64);
400 }
401
MEM_readBEST(const void * memPtr)402 MEM_STATIC size_t MEM_readBEST(const void* memPtr)
403 {
404 if (MEM_32bits())
405 return (size_t)MEM_readBE32(memPtr);
406 else
407 return (size_t)MEM_readBE64(memPtr);
408 }
409
MEM_writeBEST(void * memPtr,size_t val)410 MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
411 {
412 if (MEM_32bits())
413 MEM_writeBE32(memPtr, (U32)val);
414 else
415 MEM_writeBE64(memPtr, (U64)val);
416 }
417
418 /* code only tested on 32 and 64 bits systems */
MEM_check(void)419 MEM_STATIC void MEM_check(void) { DEBUG_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
420
421
422 #if defined (__cplusplus)
423 }
424 #endif
425
426 #endif /* MEM_H_MODULE */
427