1 // Copyright 2011 Google Inc. All Rights Reserved.
3 // Redistribution and use in source and binary forms, with or without
4 // modification, are permitted provided that the following conditions are
7 // * Redistributions of source code must retain the above copyright
8 // notice, this list of conditions and the following disclaimer.
9 // * Redistributions in binary form must reproduce the above
10 // copyright notice, this list of conditions and the following disclaimer
11 // in the documentation and/or other materials provided with the
13 // * Neither the name of Google Inc. nor the names of its
14 // contributors may be used to endorse or promote products derived from
15 // this software without specific prior written permission.
17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 // Various stubs for the open-source version of Snappy.
31 #ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_
32 #define UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_
45 #ifdef HAVE_SYS_MMAN_H
49 #include "snappy-stubs-public.h"
51 #if defined(__x86_64__)
53 // Enable 64-bit optimized versions of some routines.
58 // Needed by OS X, among others.
60 #define MAP_ANONYMOUS MAP_ANON
63 // Pull in std::min, std::ostream, and the likes. This is safe because this
64 // header file is never used from any public header files.
67 // The size of an array, if known at compile-time.
68 // Will give unexpected results if used on a pointer.
69 // We undefine it first, since some compilers already have a definition.
73 #define ARRAYSIZE(a) (sizeof(a) / sizeof(*(a)))
75 // Static prediction hints.
76 #ifdef HAVE_BUILTIN_EXPECT
77 #define PREDICT_FALSE(x) (__builtin_expect(x, 0))
78 #define PREDICT_TRUE(x) (__builtin_expect(!!(x), 1))
80 #define PREDICT_FALSE(x) x
81 #define PREDICT_TRUE(x) x
84 // This is only used for recomputing the tag byte table used during
85 // decompression; for simplicity we just remove it from the open-source
86 // version (anyone who wants to regenerate it can just do the call
87 // themselves within main()).
88 #define DEFINE_bool(flag_name, default_value, description) \
89 bool FLAGS_ ## flag_name = default_value
90 #define DECLARE_bool(flag_name) \
91 extern bool FLAGS_ ## flag_name
95 static const uint32 kuint32max = static_cast<uint32>(0xFFFFFFFF);
96 static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL);
100 #define LOG(level) LogMessage()
101 #define VLOG(level) true ? (void)0 : \
102 snappy::LogMessageVoidify() & snappy::LogMessage()
111 LogMessage& operator<<(const std::string& msg) {
115 LogMessage& operator<<(int x) {
121 // Asserts, both versions activated in debug mode only,
122 // and ones that are always active.
124 #define CRASH_UNLESS(condition) \
125 PREDICT_TRUE(condition) ? (void)0 : \
126 snappy::LogMessageVoidify() & snappy::LogMessageCrash()
128 class LogMessageCrash : public LogMessage {
130 LogMessageCrash() { }
137 // This class is used to explicitly ignore values in the conditional
138 // logging macros. This avoids compiler warnings like "value computed
139 // is not used" and "statement has no effect".
141 class LogMessageVoidify {
143 LogMessageVoidify() { }
144 // This has to be an operator with a precedence lower than << but
146 void operator&(const LogMessage&) { }
149 #define CHECK(cond) CRASH_UNLESS(cond)
150 #define CHECK_LE(a, b) CRASH_UNLESS((a) <= (b))
151 #define CHECK_GE(a, b) CRASH_UNLESS((a) >= (b))
152 #define CHECK_EQ(a, b) CRASH_UNLESS((a) == (b))
153 #define CHECK_NE(a, b) CRASH_UNLESS((a) != (b))
154 #define CHECK_LT(a, b) CRASH_UNLESS((a) < (b))
155 #define CHECK_GT(a, b) CRASH_UNLESS((a) > (b))
159 #define DCHECK(cond) CRASH_UNLESS(true)
160 #define DCHECK_LE(a, b) CRASH_UNLESS(true)
161 #define DCHECK_GE(a, b) CRASH_UNLESS(true)
162 #define DCHECK_EQ(a, b) CRASH_UNLESS(true)
163 #define DCHECK_NE(a, b) CRASH_UNLESS(true)
164 #define DCHECK_LT(a, b) CRASH_UNLESS(true)
165 #define DCHECK_GT(a, b) CRASH_UNLESS(true)
169 #define DCHECK(cond) CHECK(cond)
170 #define DCHECK_LE(a, b) CHECK_LE(a, b)
171 #define DCHECK_GE(a, b) CHECK_GE(a, b)
172 #define DCHECK_EQ(a, b) CHECK_EQ(a, b)
173 #define DCHECK_NE(a, b) CHECK_NE(a, b)
174 #define DCHECK_LT(a, b) CHECK_LT(a, b)
175 #define DCHECK_GT(a, b) CHECK_GT(a, b)
179 // Potentially unaligned loads and stores.
181 // x86 and PowerPC can simply do these loads and stores native.
183 #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__)
185 #define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p))
186 #define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p))
187 #define UNALIGNED_LOAD64(_p) (*reinterpret_cast<const uint64 *>(_p))
189 #define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val))
190 #define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val))
191 #define UNALIGNED_STORE64(_p, _val) (*reinterpret_cast<uint64 *>(_p) = (_val))
193 // ARMv7 and newer support native unaligned accesses, but only of 16-bit
194 // and 32-bit values (not 64-bit); older versions either raise a fatal signal,
195 // do an unaligned read and rotate the words around a bit, or do the reads very
196 // slowly (trip through kernel mode). There's no simple #define that says just
197 // “ARMv7 or higher”, so we have to filter away all ARMv5 and ARMv6
198 // sub-architectures.
200 // This is a mess, but there's not much we can do about it.
202 #elif defined(__arm__) && \
203 !defined(__ARM_ARCH_5__) && \
204 !defined(__ARM_ARCH_5T__) && \
205 !defined(__ARM_ARCH_5TE__) && \
206 !defined(__ARM_ARCH_5TEJ__) && \
207 !defined(__ARM_ARCH_6__) && \
208 !defined(__ARM_ARCH_6J__) && \
209 !defined(__ARM_ARCH_6K__) && \
210 !defined(__ARM_ARCH_6Z__) && \
211 !defined(__ARM_ARCH_6ZK__) && \
212 !defined(__ARM_ARCH_6T2__)
214 #define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p))
215 #define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p))
217 #define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val))
218 #define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val))
220 // TODO(user): NEON supports unaligned 64-bit loads and stores.
221 // See if that would be more efficient on platforms supporting it,
222 // at least for copies.
224 inline uint64 UNALIGNED_LOAD64(const void *p) {
226 memcpy(&t, p, sizeof t);
230 inline void UNALIGNED_STORE64(void *p, uint64 v) {
231 memcpy(p, &v, sizeof v);
236 // These functions are provided for architectures that don't support
237 // unaligned loads and stores.
239 inline uint16 UNALIGNED_LOAD16(const void *p) {
241 memcpy(&t, p, sizeof t);
245 inline uint32 UNALIGNED_LOAD32(const void *p) {
247 memcpy(&t, p, sizeof t);
251 inline uint64 UNALIGNED_LOAD64(const void *p) {
253 memcpy(&t, p, sizeof t);
257 inline void UNALIGNED_STORE16(void *p, uint16 v) {
258 memcpy(p, &v, sizeof v);
261 inline void UNALIGNED_STORE32(void *p, uint32 v) {
262 memcpy(p, &v, sizeof v);
265 inline void UNALIGNED_STORE64(void *p, uint64 v) {
266 memcpy(p, &v, sizeof v);
271 // This can be more efficient than UNALIGNED_LOAD64 + UNALIGNED_STORE64
272 // on some platforms, in particular ARM.
273 inline void UnalignedCopy64(const void *src, void *dst) {
274 if (sizeof(void *) == 8) {
275 UNALIGNED_STORE64(dst, UNALIGNED_LOAD64(src));
277 const char *src_char = reinterpret_cast<const char *>(src);
278 char *dst_char = reinterpret_cast<char *>(dst);
280 UNALIGNED_STORE32(dst_char, UNALIGNED_LOAD32(src_char));
281 UNALIGNED_STORE32(dst_char + 4, UNALIGNED_LOAD32(src_char + 4));
285 // The following guarantees declaration of the byte swap functions.
286 #ifdef WORDS_BIGENDIAN
288 #ifdef HAVE_SYS_BYTEORDER_H
289 #include <sys/byteorder.h>
292 #ifdef HAVE_SYS_ENDIAN_H
293 #include <sys/endian.h>
298 #define bswap_16(x) _byteswap_ushort(x)
299 #define bswap_32(x) _byteswap_ulong(x)
300 #define bswap_64(x) _byteswap_uint64(x)
302 #elif defined(__APPLE__)
303 // Mac OS X / Darwin features
304 #include <libkern/OSByteOrder.h>
305 #define bswap_16(x) OSSwapInt16(x)
306 #define bswap_32(x) OSSwapInt32(x)
307 #define bswap_64(x) OSSwapInt64(x)
309 #elif defined(HAVE_BYTESWAP_H)
310 #include <byteswap.h>
312 #elif defined(bswap32)
313 // FreeBSD defines bswap{16,32,64} in <sys/endian.h> (already #included).
314 #define bswap_16(x) bswap16(x)
315 #define bswap_32(x) bswap32(x)
316 #define bswap_64(x) bswap64(x)
318 #elif defined(BSWAP_64)
319 // Solaris 10 defines BSWAP_{16,32,64} in <sys/byteorder.h> (already #included).
320 #define bswap_16(x) BSWAP_16(x)
321 #define bswap_32(x) BSWAP_32(x)
322 #define bswap_64(x) BSWAP_64(x)
326 inline uint16 bswap_16(uint16 x) {
327 return (x << 8) | (x >> 8);
330 inline uint32 bswap_32(uint32 x) {
331 x = ((x & 0xff00ff00UL) >> 8) | ((x & 0x00ff00ffUL) << 8);
332 return (x >> 16) | (x << 16);
335 inline uint64 bswap_64(uint64 x) {
336 x = ((x & 0xff00ff00ff00ff00ULL) >> 8) | ((x & 0x00ff00ff00ff00ffULL) << 8);
337 x = ((x & 0xffff0000ffff0000ULL) >> 16) | ((x & 0x0000ffff0000ffffULL) << 16);
338 return (x >> 32) | (x << 32);
343 #endif // WORDS_BIGENDIAN
345 // Convert to little-endian storage, opposite of network format.
346 // Convert x from host to little endian: x = LittleEndian.FromHost(x);
347 // convert x from little endian to host: x = LittleEndian.ToHost(x);
349 // Store values into unaligned memory converting to little endian order:
350 // LittleEndian.Store16(p, x);
352 // Load unaligned values stored in little endian converting to host order:
353 // x = LittleEndian.Load16(p);
356 // Conversion functions.
357 #ifdef WORDS_BIGENDIAN
359 static uint16 FromHost16(uint16 x) { return bswap_16(x); }
360 static uint16 ToHost16(uint16 x) { return bswap_16(x); }
362 static uint32 FromHost32(uint32 x) { return bswap_32(x); }
363 static uint32 ToHost32(uint32 x) { return bswap_32(x); }
365 static bool IsLittleEndian() { return false; }
367 #else // !defined(WORDS_BIGENDIAN)
369 static uint16 FromHost16(uint16 x) { return x; }
370 static uint16 ToHost16(uint16 x) { return x; }
372 static uint32 FromHost32(uint32 x) { return x; }
373 static uint32 ToHost32(uint32 x) { return x; }
375 static bool IsLittleEndian() { return true; }
377 #endif // !defined(WORDS_BIGENDIAN)
379 // Functions to do unaligned loads and stores in little-endian order.
380 static uint16 Load16(const void *p) {
381 return ToHost16(UNALIGNED_LOAD16(p));
384 static void Store16(void *p, uint16 v) {
385 UNALIGNED_STORE16(p, FromHost16(v));
388 static uint32 Load32(const void *p) {
389 return ToHost32(UNALIGNED_LOAD32(p));
392 static void Store32(void *p, uint32 v) {
393 UNALIGNED_STORE32(p, FromHost32(v));
397 // Some bit-manipulation functions.
400 // Return floor(log2(n)) for positive integer n. Returns -1 iff n == 0.
401 static int Log2Floor(uint32 n);
403 // Return the first set least / most significant bit, 0-indexed. Returns an
404 // undefined value if n == 0. FindLSBSetNonZero() is similar to ffs() except
405 // that it's 0-indexed.
406 static int FindLSBSetNonZero(uint32 n);
407 static int FindLSBSetNonZero64(uint64 n);
410 DISALLOW_COPY_AND_ASSIGN(Bits);
413 #ifdef HAVE_BUILTIN_CTZ
415 inline int Bits::Log2Floor(uint32 n) {
416 return n == 0 ? -1 : 31 ^ __builtin_clz(n);
419 inline int Bits::FindLSBSetNonZero(uint32 n) {
420 return __builtin_ctz(n);
423 inline int Bits::FindLSBSetNonZero64(uint64 n) {
424 return __builtin_ctzll(n);
427 #else // Portable versions.
429 inline int Bits::Log2Floor(uint32 n) {
434 for (int i = 4; i >= 0; --i) {
435 int shift = (1 << i);
436 uint32 x = value >> shift;
446 inline int Bits::FindLSBSetNonZero(uint32 n) {
448 for (int i = 4, shift = 1 << 4; i >= 0; --i) {
449 const uint32 x = n << shift;
459 // FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero().
460 inline int Bits::FindLSBSetNonZero64(uint64 n) {
461 const uint32 bottombits = static_cast<uint32>(n);
462 if (bottombits == 0) {
463 // Bottom bits are zero, so scan in top bits
464 return 32 + FindLSBSetNonZero(static_cast<uint32>(n >> 32));
466 return FindLSBSetNonZero(bottombits);
470 #endif // End portable versions.
472 // Variable-length integer encoding.
475 // Maximum lengths of varint encoding of uint32.
476 static const int kMax32 = 5;
478 // Attempts to parse a varint32 from a prefix of the bytes in [ptr,limit-1].
479 // Never reads a character at or beyond limit. If a valid/terminated varint32
480 // was found in the range, stores it in *OUTPUT and returns a pointer just
481 // past the last byte of the varint32. Else returns NULL. On success,
482 // "result <= limit".
483 static const char* Parse32WithLimit(const char* ptr, const char* limit,
486 // REQUIRES "ptr" points to a buffer of length sufficient to hold "v".
487 // EFFECTS Encodes "v" into "ptr" and returns a pointer to the
488 // byte just past the last encoded byte.
489 static char* Encode32(char* ptr, uint32 v);
491 // EFFECTS Appends the varint representation of "value" to "*s".
492 static void Append32(string* s, uint32 value);
495 inline const char* Varint::Parse32WithLimit(const char* p,
498 const unsigned char* ptr = reinterpret_cast<const unsigned char*>(p);
499 const unsigned char* limit = reinterpret_cast<const unsigned char*>(l);
501 if (ptr >= limit) return NULL;
502 b = *(ptr++); result = b & 127; if (b < 128) goto done;
503 if (ptr >= limit) return NULL;
504 b = *(ptr++); result |= (b & 127) << 7; if (b < 128) goto done;
505 if (ptr >= limit) return NULL;
506 b = *(ptr++); result |= (b & 127) << 14; if (b < 128) goto done;
507 if (ptr >= limit) return NULL;
508 b = *(ptr++); result |= (b & 127) << 21; if (b < 128) goto done;
509 if (ptr >= limit) return NULL;
510 b = *(ptr++); result |= (b & 127) << 28; if (b < 16) goto done;
511 return NULL; // Value is too long to be a varint32
514 return reinterpret_cast<const char*>(ptr);
517 inline char* Varint::Encode32(char* sptr, uint32 v) {
518 // Operate on characters as unsigneds
519 unsigned char* ptr = reinterpret_cast<unsigned char*>(sptr);
520 static const int B = 128;
523 } else if (v < (1<<14)) {
526 } else if (v < (1<<21)) {
528 *(ptr++) = (v>>7) | B;
530 } else if (v < (1<<28)) {
532 *(ptr++) = (v>>7) | B;
533 *(ptr++) = (v>>14) | B;
537 *(ptr++) = (v>>7) | B;
538 *(ptr++) = (v>>14) | B;
539 *(ptr++) = (v>>21) | B;
542 return reinterpret_cast<char*>(ptr);
545 // If you know the internal layout of the std::string in use, you can
546 // replace this function with one that resizes the string without
547 // filling the new space with zeros (if applicable) --
548 // it will be non-portable but faster.
549 inline void STLStringResizeUninitialized(string* s, size_t new_size) {
553 // Return a mutable char* pointing to a string's internal buffer,
554 // which may not be null-terminated. Writing through this pointer will
555 // modify the string.
557 // string_as_array(&str)[i] is valid for 0 <= i < str.size() until the
558 // next call to a string method that invalidates iterators.
560 // As of 2006-04, there is no standard-blessed way of getting a
561 // mutable reference to a string's internal buffer. However, issue 530
562 // (http://www.open-std.org/JTC1/SC22/WG21/docs/lwg-defects.html#530)
563 // proposes this as the method. It will officially be part of the standard
564 // for C++0x. This should already work on all current implementations.
565 inline char* string_as_array(string* str) {
566 return str->empty() ? NULL : &*str->begin();
569 } // namespace snappy
571 #endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_