diff options
Diffstat (limited to 'app/snappy')
| -rw-r--r-- | app/snappy/Android.mk | 35 | ||||
| -rw-r--r-- | app/snappy/COPYING | 28 | ||||
| -rw-r--r-- | app/snappy/conf/config.h | 129 | ||||
| -rw-r--r-- | app/snappy/snappy-c.cc | 90 | ||||
| -rw-r--r-- | app/snappy/snappy-c.h | 138 | ||||
| -rw-r--r-- | app/snappy/snappy-internal.h | 150 | ||||
| -rw-r--r-- | app/snappy/snappy-sinksource.cc | 71 | ||||
| -rw-r--r-- | app/snappy/snappy-sinksource.h | 137 | ||||
| -rw-r--r-- | app/snappy/snappy-stubs-internal.cc | 42 | ||||
| -rw-r--r-- | app/snappy/snappy-stubs-internal.h | 491 | ||||
| -rw-r--r-- | app/snappy/snappy-stubs-public.h | 85 | ||||
| -rw-r--r-- | app/snappy/snappy-stubs-public.h.in | 85 | ||||
| -rw-r--r-- | app/snappy/snappy-test.cc | 606 | ||||
| -rw-r--r-- | app/snappy/snappy-test.h | 580 | ||||
| -rw-r--r-- | app/snappy/snappy.cc | 1119 | ||||
| -rw-r--r-- | app/snappy/snappy.h | 163 | ||||
| -rw-r--r-- | app/snappy/snappy_unittest.cc | 1164 |
17 files changed, 5113 insertions, 0 deletions
diff --git a/app/snappy/Android.mk b/app/snappy/Android.mk new file mode 100644 index 00000000..cdd26003 --- /dev/null +++ b/app/snappy/Android.mk @@ -0,0 +1,35 @@ + +LOCAL_PATH:= $(call my-dir) + + +common_SRC_FILES:= \ + snappy-c.cc \ + snappy-sinksource.cc \ + snappy-stubs-internal.cc \ + snappy.cc + +common_C_INCLUDES += $(LOCAL_PATH)/ $(LOCAL_PATH)/conf + +# static library +# ===================================================== + +include $(CLEAR_VARS) +LOCAL_CPP_EXTENSION := .cc +LOCAL_SRC_FILES:= $(common_SRC_FILES) +LOCAL_C_INCLUDES:= $(common_C_INCLUDES) +LOCAL_MODULE := snappy-static +LOCAL_PRELINK_MODULE:= false +LOCAL_MODULE_TAGS := optional +include $(BUILD_STATIC_LIBRARY) + +# dynamic library +# ===================================================== + +# include $(CLEAR_VARS) +# LOCAL_SRC_FILES:= $(common_SRC_FILES) +# LOCAL_C_INCLUDES:= $(common_C_INCLUDES) +# LOCAL_MODULE := liblzo +# LOCAL_PRELINK_MODULE:= false +# LOCAL_MODULE_TAGS := optional +# include $(BUILD_SHARED_LIBRARY) + diff --git a/app/snappy/COPYING b/app/snappy/COPYING new file mode 100644 index 00000000..8d6bd9fe --- /dev/null +++ b/app/snappy/COPYING @@ -0,0 +1,28 @@ +Copyright 2011, Google Inc. +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/app/snappy/conf/config.h b/app/snappy/conf/config.h new file mode 100644 index 00000000..e0bf3214 --- /dev/null +++ b/app/snappy/conf/config.h @@ -0,0 +1,129 @@ +/* config.h. Generated from config.h.in by configure. */ +/* config.h.in. Generated from configure.ac by autoheader. */ + +/* Define if building universal (internal helper macro) */ +/* #undef AC_APPLE_UNIVERSAL_BUILD */ + +/* Define to 1 if the compiler supports __builtin_ctz and friends. */ +#define HAVE_BUILTIN_CTZ 1 + +/* Define to 1 if the compiler supports __builtin_expect. */ +#define HAVE_BUILTIN_EXPECT 1 + +/* Define to 1 if you have the <byteswap.h> header file. */ +/* #undef HAVE_BYTESWAP_H */ + +/* Define to 1 if you have the <dlfcn.h> header file. */ +#define HAVE_DLFCN_H 1 + +/* Use the gflags package for command-line parsing. */ +/* #undef HAVE_GFLAGS */ + +/* Defined when Google Test is available. */ +/* #undef HAVE_GTEST */ + +/* Define to 1 if you have the <inttypes.h> header file. */ +#define HAVE_INTTYPES_H 1 + +/* Define to 1 if you have the `fastlz' library (-lfastlz). */ +/* #undef HAVE_LIBFASTLZ */ + +/* Define to 1 if you have the `lzf' library (-llzf). */ +/* #undef HAVE_LIBLZF */ + +/* Define to 1 if you have the `lzo2' library (-llzo2). */ +/* #undef HAVE_LIBLZO2 */ + +/* Define to 1 if you have the `quicklz' library (-lquicklz). */ +/* #undef HAVE_LIBQUICKLZ */ + +/* Define to 1 if you have the `z' library (-lz). */ +#define HAVE_LIBZ 1 + +/* Define to 1 if you have the <memory.h> header file. */ +#define HAVE_MEMORY_H 1 + +/* Define to 1 if you have the <stddef.h> header file. */ +#define HAVE_STDDEF_H 1 + +/* Define to 1 if you have the <stdint.h> header file. */ +#define HAVE_STDINT_H 1 + +/* Define to 1 if you have the <stdlib.h> header file. */ +#define HAVE_STDLIB_H 1 + +/* Define to 1 if you have the <strings.h> header file. */ +#define HAVE_STRINGS_H 1 + +/* Define to 1 if you have the <string.h> header file. */ +#define HAVE_STRING_H 1 + +/* Define to 1 if you have the <sys/byteswap.h> header file. */ +/* #undef HAVE_SYS_BYTESWAP_H */ + +/* Define to 1 if you have the <sys/endian.h> header file. */ +/* #undef HAVE_SYS_ENDIAN_H */ + +/* Define to 1 if you have the <sys/mman.h> header file. */ +#define HAVE_SYS_MMAN_H 1 + +/* Define to 1 if you have the <sys/resource.h> header file. */ +#define HAVE_SYS_RESOURCE_H 1 + +/* Define to 1 if you have the <sys/stat.h> header file. */ +#define HAVE_SYS_STAT_H 1 + +/* Define to 1 if you have the <sys/time.h> header file. */ +#define HAVE_SYS_TIME_H 1 + +/* Define to 1 if you have the <sys/types.h> header file. */ +#define HAVE_SYS_TYPES_H 1 + +/* Define to 1 if you have the <unistd.h> header file. */ +#define HAVE_UNISTD_H 1 + +/* Define to 1 if you have the <windows.h> header file. */ +/* #undef HAVE_WINDOWS_H */ + +/* Define to the sub-directory in which libtool stores uninstalled libraries. + */ +#define LT_OBJDIR ".libs/" + +/* Name of package */ +#define PACKAGE "snappy" + +/* Define to the address where bug reports for this package should be sent. */ +#define PACKAGE_BUGREPORT "" + +/* Define to the full name of this package. */ +#define PACKAGE_NAME "snappy" + +/* Define to the full name and version of this package. */ +#define PACKAGE_STRING "snappy 1.1.0" + +/* Define to the one symbol short name of this package. */ +#define PACKAGE_TARNAME "snappy" + +/* Define to the home page for this package. */ +#define PACKAGE_URL "" + +/* Define to the version of this package. */ +#define PACKAGE_VERSION "1.1.0" + +/* Define to 1 if you have the ANSI C header files. */ +#define STDC_HEADERS 1 + +/* Version number of package */ +#define VERSION "1.1.0" + +/* Define WORDS_BIGENDIAN to 1 if your processor stores words with the most + significant byte first (like Motorola and SPARC, unlike Intel). */ +#if defined AC_APPLE_UNIVERSAL_BUILD +# if defined __BIG_ENDIAN__ +# define WORDS_BIGENDIAN 1 +# endif +#else +# ifndef WORDS_BIGENDIAN +/* # undef WORDS_BIGENDIAN */ +# endif +#endif diff --git a/app/snappy/snappy-c.cc b/app/snappy/snappy-c.cc new file mode 100644 index 00000000..473a0b09 --- /dev/null +++ b/app/snappy/snappy-c.cc @@ -0,0 +1,90 @@ +// Copyright 2011 Martin Gieseking <martin.gieseking@uos.de>. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "snappy.h" +#include "snappy-c.h" + +extern "C" { + +snappy_status snappy_compress(const char* input, + size_t input_length, + char* compressed, + size_t *compressed_length) { + if (*compressed_length < snappy_max_compressed_length(input_length)) { + return SNAPPY_BUFFER_TOO_SMALL; + } + snappy::RawCompress(input, input_length, compressed, compressed_length); + return SNAPPY_OK; +} + +snappy_status snappy_uncompress(const char* compressed, + size_t compressed_length, + char* uncompressed, + size_t* uncompressed_length) { + size_t real_uncompressed_length; + if (!snappy::GetUncompressedLength(compressed, + compressed_length, + &real_uncompressed_length)) { + return SNAPPY_INVALID_INPUT; + } + if (*uncompressed_length < real_uncompressed_length) { + return SNAPPY_BUFFER_TOO_SMALL; + } + if (!snappy::RawUncompress(compressed, compressed_length, uncompressed)) { + return SNAPPY_INVALID_INPUT; + } + *uncompressed_length = real_uncompressed_length; + return SNAPPY_OK; +} + +size_t snappy_max_compressed_length(size_t source_length) { + return snappy::MaxCompressedLength(source_length); +} + +snappy_status snappy_uncompressed_length(const char *compressed, + size_t compressed_length, + size_t *result) { + if (snappy::GetUncompressedLength(compressed, + compressed_length, + result)) { + return SNAPPY_OK; + } else { + return SNAPPY_INVALID_INPUT; + } +} + +snappy_status snappy_validate_compressed_buffer(const char *compressed, + size_t compressed_length) { + if (snappy::IsValidCompressedBuffer(compressed, compressed_length)) { + return SNAPPY_OK; + } else { + return SNAPPY_INVALID_INPUT; + } +} + +} // extern "C" diff --git a/app/snappy/snappy-c.h b/app/snappy/snappy-c.h new file mode 100644 index 00000000..c6c2a860 --- /dev/null +++ b/app/snappy/snappy-c.h @@ -0,0 +1,138 @@ +/* + * Copyright 2011 Martin Gieseking <martin.gieseking@uos.de>. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Google Inc. nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * Plain C interface (a wrapper around the C++ implementation). + */ + +#ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_C_H_ +#define UTIL_SNAPPY_OPENSOURCE_SNAPPY_C_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <stddef.h> + +/* + * Return values; see the documentation for each function to know + * what each can return. + */ +typedef enum { + SNAPPY_OK = 0, + SNAPPY_INVALID_INPUT = 1, + SNAPPY_BUFFER_TOO_SMALL = 2 +} snappy_status; + +/* + * Takes the data stored in "input[0..input_length-1]" and stores + * it in the array pointed to by "compressed". + * + * <compressed_length> signals the space available in "compressed". + * If it is not at least equal to "snappy_max_compressed_length(input_length)", + * SNAPPY_BUFFER_TOO_SMALL is returned. After successful compression, + * <compressed_length> contains the true length of the compressed output, + * and SNAPPY_OK is returned. + * + * Example: + * size_t output_length = snappy_max_compressed_length(input_length); + * char* output = (char*)malloc(output_length); + * if (snappy_compress(input, input_length, output, &output_length) + * == SNAPPY_OK) { + * ... Process(output, output_length) ... + * } + * free(output); + */ +snappy_status snappy_compress(const char* input, + size_t input_length, + char* compressed, + size_t* compressed_length); + +/* + * Given data in "compressed[0..compressed_length-1]" generated by + * calling the snappy_compress routine, this routine stores + * the uncompressed data to + * uncompressed[0..uncompressed_length-1]. + * Returns failure (a value not equal to SNAPPY_OK) if the message + * is corrupted and could not be decrypted. + * + * <uncompressed_length> signals the space available in "uncompressed". + * If it is not at least equal to the value returned by + * snappy_uncompressed_length for this stream, SNAPPY_BUFFER_TOO_SMALL + * is returned. After successful decompression, <uncompressed_length> + * contains the true length of the decompressed output. + * + * Example: + * size_t output_length; + * if (snappy_uncompressed_length(input, input_length, &output_length) + * != SNAPPY_OK) { + * ... fail ... + * } + * char* output = (char*)malloc(output_length); + * if (snappy_uncompress(input, input_length, output, &output_length) + * == SNAPPY_OK) { + * ... Process(output, output_length) ... + * } + * free(output); + */ +snappy_status snappy_uncompress(const char* compressed, + size_t compressed_length, + char* uncompressed, + size_t* uncompressed_length); + +/* + * Returns the maximal size of the compressed representation of + * input data that is "source_length" bytes in length. + */ +size_t snappy_max_compressed_length(size_t source_length); + +/* + * REQUIRES: "compressed[]" was produced by snappy_compress() + * Returns SNAPPY_OK and stores the length of the uncompressed data in + * *result normally. Returns SNAPPY_INVALID_INPUT on parsing error. + * This operation takes O(1) time. + */ +snappy_status snappy_uncompressed_length(const char* compressed, + size_t compressed_length, + size_t* result); + +/* + * Check if the contents of "compressed[]" can be uncompressed successfully. + * Does not return the uncompressed data; if so, returns SNAPPY_OK, + * or if not, returns SNAPPY_INVALID_INPUT. + * Takes time proportional to compressed_length, but is usually at least a + * factor of four faster than actual decompression. + */ +snappy_status snappy_validate_compressed_buffer(const char* compressed, + size_t compressed_length); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif /* UTIL_SNAPPY_OPENSOURCE_SNAPPY_C_H_ */ diff --git a/app/snappy/snappy-internal.h b/app/snappy/snappy-internal.h new file mode 100644 index 00000000..c99d3313 --- /dev/null +++ b/app/snappy/snappy-internal.h @@ -0,0 +1,150 @@ +// Copyright 2008 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Internals shared between the Snappy implementation and its unittest. + +#ifndef UTIL_SNAPPY_SNAPPY_INTERNAL_H_ +#define UTIL_SNAPPY_SNAPPY_INTERNAL_H_ + +#include "snappy-stubs-internal.h" + +namespace snappy { +namespace internal { + +class WorkingMemory { + public: + WorkingMemory() : large_table_(NULL) { } + ~WorkingMemory() { delete[] large_table_; } + + // Allocates and clears a hash table using memory in "*this", + // stores the number of buckets in "*table_size" and returns a pointer to + // the base of the hash table. + uint16* GetHashTable(size_t input_size, int* table_size); + + private: + uint16 small_table_[1<<10]; // 2KB + uint16* large_table_; // Allocated only when needed + + DISALLOW_COPY_AND_ASSIGN(WorkingMemory); +}; + +// Flat array compression that does not emit the "uncompressed length" +// prefix. Compresses "input" string to the "*op" buffer. +// +// REQUIRES: "input_length <= kBlockSize" +// REQUIRES: "op" points to an array of memory that is at least +// "MaxCompressedLength(input_length)" in size. +// REQUIRES: All elements in "table[0..table_size-1]" are initialized to zero. +// REQUIRES: "table_size" is a power of two +// +// Returns an "end" pointer into "op" buffer. +// "end - op" is the compressed size of "input". +char* CompressFragment(const char* input, + size_t input_length, + char* op, + uint16* table, + const int table_size); + +// Return the largest n such that +// +// s1[0,n-1] == s2[0,n-1] +// and n <= (s2_limit - s2). +// +// Does not read *s2_limit or beyond. +// Does not read *(s1 + (s2_limit - s2)) or beyond. +// Requires that s2_limit >= s2. +// +// Separate implementation for x86_64, for speed. Uses the fact that +// x86_64 is little endian. +#if defined(ARCH_K8) +static inline int FindMatchLength(const char* s1, + const char* s2, + const char* s2_limit) { + assert(s2_limit >= s2); + int matched = 0; + + // Find out how long the match is. We loop over the data 64 bits at a + // time until we find a 64-bit block that doesn't match; then we find + // the first non-matching bit and use that to calculate the total + // length of the match. + while (PREDICT_TRUE(s2 <= s2_limit - 8)) { + if (PREDICT_FALSE(UNALIGNED_LOAD64(s2) == UNALIGNED_LOAD64(s1 + matched))) { + s2 += 8; + matched += 8; + } else { + // On current (mid-2008) Opteron models there is a 3% more + // efficient code sequence to find the first non-matching byte. + // However, what follows is ~10% better on Intel Core 2 and newer, + // and we expect AMD's bsf instruction to improve. + uint64 x = UNALIGNED_LOAD64(s2) ^ UNALIGNED_LOAD64(s1 + matched); + int matching_bits = Bits::FindLSBSetNonZero64(x); + matched += matching_bits >> 3; + return matched; + } + } + while (PREDICT_TRUE(s2 < s2_limit)) { + if (PREDICT_TRUE(s1[matched] == *s2)) { + ++s2; + ++matched; + } else { + return matched; + } + } + return matched; +} +#else +static inline int FindMatchLength(const char* s1, + const char* s2, + const char* s2_limit) { + // Implementation based on the x86-64 version, above. + assert(s2_limit >= s2); + int matched = 0; + + while (s2 <= s2_limit - 4 && + UNALIGNED_LOAD32(s2) == UNALIGNED_LOAD32(s1 + matched)) { + s2 += 4; + matched += 4; + } + if (LittleEndian::IsLittleEndian() && s2 <= s2_limit - 4) { + uint32 x = UNALIGNED_LOAD32(s2) ^ UNALIGNED_LOAD32(s1 + matched); + int matching_bits = Bits::FindLSBSetNonZero(x); + matched += matching_bits >> 3; + } else { + while ((s2 < s2_limit) && (s1[matched] == *s2)) { + ++s2; + ++matched; + } + } + return matched; +} +#endif + +} // end namespace internal +} // end namespace snappy + +#endif // UTIL_SNAPPY_SNAPPY_INTERNAL_H_ diff --git a/app/snappy/snappy-sinksource.cc b/app/snappy/snappy-sinksource.cc new file mode 100644 index 00000000..5844552c --- /dev/null +++ b/app/snappy/snappy-sinksource.cc @@ -0,0 +1,71 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include <string.h> + +#include "snappy-sinksource.h" + +namespace snappy { + +Source::~Source() { } + +Sink::~Sink() { } + +char* Sink::GetAppendBuffer(size_t length, char* scratch) { + return scratch; +} + +ByteArraySource::~ByteArraySource() { } + +size_t ByteArraySource::Available() const { return left_; } + +const char* ByteArraySource::Peek(size_t* len) { + *len = left_; + return ptr_; +} + +void ByteArraySource::Skip(size_t n) { + left_ -= n; + ptr_ += n; +} + +UncheckedByteArraySink::~UncheckedByteArraySink() { } + +void UncheckedByteArraySink::Append(const char* data, size_t n) { + // Do no copying if the caller filled in the result of GetAppendBuffer() + if (data != dest_) { + memcpy(dest_, data, n); + } + dest_ += n; +} + +char* UncheckedByteArraySink::GetAppendBuffer(size_t len, char* scratch) { + return dest_; +} + +} diff --git a/app/snappy/snappy-sinksource.h b/app/snappy/snappy-sinksource.h new file mode 100644 index 00000000..faabfa1e --- /dev/null +++ b/app/snappy/snappy-sinksource.h @@ -0,0 +1,137 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef UTIL_SNAPPY_SNAPPY_SINKSOURCE_H_ +#define UTIL_SNAPPY_SNAPPY_SINKSOURCE_H_ + +#include <stddef.h> + + +namespace snappy { + +// A Sink is an interface that consumes a sequence of bytes. +class Sink { + public: + Sink() { } + virtual ~Sink(); + + // Append "bytes[0,n-1]" to this. + virtual void Append(const char* bytes, size_t n) = 0; + + // Returns a writable buffer of the specified length for appending. + // May return a pointer to the caller-owned scratch buffer which + // must have at least the indicated length. The returned buffer is + // only valid until the next operation on this Sink. + // + // After writing at most "length" bytes, call Append() with the + // pointer returned from this function and the number of bytes + // written. Many Append() implementations will avoid copying + // bytes if this function returned an internal buffer. + // + // If a non-scratch buffer is returned, the caller may only pass a + // prefix of it to Append(). That is, it is not correct to pass an + // interior pointer of the returned array to Append(). + // + // The default implementation always returns the scratch buffer. + virtual char* GetAppendBuffer(size_t length, char* scratch); + + + private: + // No copying + Sink(const Sink&); + void operator=(const Sink&); +}; + +// A Source is an interface that yields a sequence of bytes +class Source { + public: + Source() { } + virtual ~Source(); + + // Return the number of bytes left to read from the source + virtual size_t Available() const = 0; + + // Peek at the next flat region of the source. Does not reposition + // the source. The returned region is empty iff Available()==0. + // + // Returns a pointer to the beginning of the region and store its + // length in *len. + // + // The returned region is valid until the next call to Skip() or + // until this object is destroyed, whichever occurs first. + // + // The returned region may be larger than Available() (for example + // if this ByteSource is a view on a substring of a larger source). + // The caller is responsible for ensuring that it only reads the + // Available() bytes. + virtual const char* Peek(size_t* len) = 0; + + // Skip the next n bytes. Invalidates any buffer returned by + // a previous call to Peek(). + // REQUIRES: Available() >= n + virtual void Skip(size_t n) = 0; + + private: + // No copying + Source(const Source&); + void operator=(const Source&); +}; + +// A Source implementation that yields the contents of a flat array +class ByteArraySource : public Source { + public: + ByteArraySource(const char* p, size_t n) : ptr_(p), left_(n) { } + virtual ~ByteArraySource(); + virtual size_t Available() const; + virtual const char* Peek(size_t* len); + virtual void Skip(size_t n); + private: + const char* ptr_; + size_t left_; +}; + +// A Sink implementation that writes to a flat array without any bound checks. +class UncheckedByteArraySink : public Sink { + public: + explicit UncheckedByteArraySink(char* dest) : dest_(dest) { } + virtual ~UncheckedByteArraySink(); + virtual void Append(const char* data, size_t n); + virtual char* GetAppendBuffer(size_t len, char* scratch); + + // Return the current output pointer so that a caller can see how + // many bytes were produced. + // Note: this is not a Sink method. + char* CurrentDestination() const { return dest_; } + private: + char* dest_; +}; + + +} + +#endif // UTIL_SNAPPY_SNAPPY_SINKSOURCE_H_ diff --git a/app/snappy/snappy-stubs-internal.cc b/app/snappy/snappy-stubs-internal.cc new file mode 100644 index 00000000..6ed33437 --- /dev/null +++ b/app/snappy/snappy-stubs-internal.cc @@ -0,0 +1,42 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include <algorithm> +#include <string> + +#include "snappy-stubs-internal.h" + +namespace snappy { + +void Varint::Append32(string* s, uint32 value) { + char buf[Varint::kMax32]; + const char* p = Varint::Encode32(buf, value); + s->append(buf, p - buf); +} + +} // namespace snappy diff --git a/app/snappy/snappy-stubs-internal.h b/app/snappy/snappy-stubs-internal.h new file mode 100644 index 00000000..12393b62 --- /dev/null +++ b/app/snappy/snappy-stubs-internal.h @@ -0,0 +1,491 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Various stubs for the open-source version of Snappy. + +#ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ +#define UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <string> + +#include <assert.h> +#include <stdlib.h> +#include <string.h> + +#ifdef HAVE_SYS_MMAN_H +#include <sys/mman.h> +#endif + +#include "snappy-stubs-public.h" + +#if defined(__x86_64__) + +// Enable 64-bit optimized versions of some routines. +#define ARCH_K8 1 + +#endif + +// Needed by OS X, among others. +#ifndef MAP_ANONYMOUS +#define MAP_ANONYMOUS MAP_ANON +#endif + +// Pull in std::min, std::ostream, and the likes. This is safe because this +// header file is never used from any public header files. +using namespace std; + +// The size of an array, if known at compile-time. +// Will give unexpected results if used on a pointer. +// We undefine it first, since some compilers already have a definition. +#ifdef ARRAYSIZE +#undef ARRAYSIZE +#endif +#define ARRAYSIZE(a) (sizeof(a) / sizeof(*(a))) + +// Static prediction hints. +#ifdef HAVE_BUILTIN_EXPECT +#define PREDICT_FALSE(x) (__builtin_expect(x, 0)) +#define PREDICT_TRUE(x) (__builtin_expect(!!(x), 1)) +#else +#define PREDICT_FALSE(x) x +#define PREDICT_TRUE(x) x +#endif + +// This is only used for recomputing the tag byte table used during +// decompression; for simplicity we just remove it from the open-source +// version (anyone who wants to regenerate it can just do the call +// themselves within main()). +#define DEFINE_bool(flag_name, default_value, description) \ + bool FLAGS_ ## flag_name = default_value +#define DECLARE_bool(flag_name) \ + extern bool FLAGS_ ## flag_name + +namespace snappy { + +static const uint32 kuint32max = static_cast<uint32>(0xFFFFFFFF); +static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL); + +// Potentially unaligned loads and stores. + +// x86 and PowerPC can simply do these loads and stores native. + +#if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) + +#define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p)) +#define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p)) +#define UNALIGNED_LOAD64(_p) (*reinterpret_cast<const uint64 *>(_p)) + +#define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val)) +#define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val)) +#define UNALIGNED_STORE64(_p, _val) (*reinterpret_cast<uint64 *>(_p) = (_val)) + +// ARMv7 and newer support native unaligned accesses, but only of 16-bit +// and 32-bit values (not 64-bit); older versions either raise a fatal signal, +// do an unaligned read and rotate the words around a bit, or do the reads very +// slowly (trip through kernel mode). There's no simple #define that says just +// “ARMv7 or higher”, so we have to filter away all ARMv5 and ARMv6 +// sub-architectures. +// +// This is a mess, but there's not much we can do about it. + +#elif defined(__arm__) && \ + !defined(__ARM_ARCH_4__) && \ + !defined(__ARM_ARCH_4T__) && \ + !defined(__ARM_ARCH_5__) && \ + !defined(__ARM_ARCH_5T__) && \ + !defined(__ARM_ARCH_5TE__) && \ + !defined(__ARM_ARCH_5TEJ__) && \ + !defined(__ARM_ARCH_6__) && \ + !defined(__ARM_ARCH_6J__) && \ + !defined(__ARM_ARCH_6K__) && \ + !defined(__ARM_ARCH_6Z__) && \ + !defined(__ARM_ARCH_6ZK__) && \ + !defined(__ARM_ARCH_6T2__) + +#define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p)) +#define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p)) + +#define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val)) +#define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val)) + +// TODO(user): NEON supports unaligned 64-bit loads and stores. +// See if that would be more efficient on platforms supporting it, +// at least for copies. + +inline uint64 UNALIGNED_LOAD64(const void *p) { + uint64 t; + memcpy(&t, p, sizeof t); + return t; +} + +inline void UNALIGNED_STORE64(void *p, uint64 v) { + memcpy(p, &v, sizeof v); +} + +#else + +// These functions are provided for architectures that don't support +// unaligned loads and stores. + +inline uint16 UNALIGNED_LOAD16(const void *p) { + uint16 t; + memcpy(&t, p, sizeof t); + return t; +} + +inline uint32 UNALIGNED_LOAD32(const void *p) { + uint32 t; + memcpy(&t, p, sizeof t); + return t; +} + +inline uint64 UNALIGNED_LOAD64(const void *p) { + uint64 t; + memcpy(&t, p, sizeof t); + return t; +} + +inline void UNALIGNED_STORE16(void *p, uint16 v) { + memcpy(p, &v, sizeof v); +} + +inline void UNALIGNED_STORE32(void *p, uint32 v) { + memcpy(p, &v, sizeof v); +} + +inline void UNALIGNED_STORE64(void *p, uint64 v) { + memcpy(p, &v, sizeof v); +} + +#endif + +// This can be more efficient than UNALIGNED_LOAD64 + UNALIGNED_STORE64 +// on some platforms, in particular ARM. +inline void UnalignedCopy64(const void *src, void *dst) { + if (sizeof(void *) == 8) { + UNALIGNED_STORE64(dst, UNALIGNED_LOAD64(src)); + } else { + const char *src_char = reinterpret_cast<const char *>(src); + char *dst_char = reinterpret_cast<char *>(dst); + + UNALIGNED_STORE32(dst_char, UNALIGNED_LOAD32(src_char)); + UNALIGNED_STORE32(dst_char + 4, UNALIGNED_LOAD32(src_char + 4)); + } +} + +// The following guarantees declaration of the byte swap functions. +#ifdef WORDS_BIGENDIAN + +#ifdef HAVE_SYS_BYTEORDER_H +#include <sys/byteorder.h> +#endif + +#ifdef HAVE_SYS_ENDIAN_H +#include <sys/endian.h> +#endif + +#ifdef _MSC_VER +#include <stdlib.h> +#define bswap_16(x) _byteswap_ushort(x) +#define bswap_32(x) _byteswap_ulong(x) +#define bswap_64(x) _byteswap_uint64(x) + +#elif defined(__APPLE__) +// Mac OS X / Darwin features +#include <libkern/OSByteOrder.h> +#define bswap_16(x) OSSwapInt16(x) +#define bswap_32(x) OSSwapInt32(x) +#define bswap_64(x) OSSwapInt64(x) + +#elif defined(HAVE_BYTESWAP_H) +#include <byteswap.h> + +#elif defined(bswap32) +// FreeBSD defines bswap{16,32,64} in <sys/endian.h> (already #included). +#define bswap_16(x) bswap16(x) +#define bswap_32(x) bswap32(x) +#define bswap_64(x) bswap64(x) + +#elif defined(BSWAP_64) +// Solaris 10 defines BSWAP_{16,32,64} in <sys/byteorder.h> (already #included). +#define bswap_16(x) BSWAP_16(x) +#define bswap_32(x) BSWAP_32(x) +#define bswap_64(x) BSWAP_64(x) + +#else + +inline uint16 bswap_16(uint16 x) { + return (x << 8) | (x >> 8); +} + +inline uint32 bswap_32(uint32 x) { + x = ((x & 0xff00ff00UL) >> 8) | ((x & 0x00ff00ffUL) << 8); + return (x >> 16) | (x << 16); +} + +inline uint64 bswap_64(uint64 x) { + x = ((x & 0xff00ff00ff00ff00ULL) >> 8) | ((x & 0x00ff00ff00ff00ffULL) << 8); + x = ((x & 0xffff0000ffff0000ULL) >> 16) | ((x & 0x0000ffff0000ffffULL) << 16); + return (x >> 32) | (x << 32); +} + +#endif + +#endif // WORDS_BIGENDIAN + +// Convert to little-endian storage, opposite of network format. +// Convert x from host to little endian: x = LittleEndian.FromHost(x); +// convert x from little endian to host: x = LittleEndian.ToHost(x); +// +// Store values into unaligned memory converting to little endian order: +// LittleEndian.Store16(p, x); +// +// Load unaligned values stored in little endian converting to host order: +// x = LittleEndian.Load16(p); +class LittleEndian { + public: + // Conversion functions. +#ifdef WORDS_BIGENDIAN + + static uint16 FromHost16(uint16 x) { return bswap_16(x); } + static uint16 ToHost16(uint16 x) { return bswap_16(x); } + + static uint32 FromHost32(uint32 x) { return bswap_32(x); } + static uint32 ToHost32(uint32 x) { return bswap_32(x); } + + static bool IsLittleEndian() { return false; } + +#else // !defined(WORDS_BIGENDIAN) + + static uint16 FromHost16(uint16 x) { return x; } + static uint16 ToHost16(uint16 x) { return x; } + + static uint32 FromHost32(uint32 x) { return x; } + static uint32 ToHost32(uint32 x) { return x; } + + static bool IsLittleEndian() { return true; } + +#endif // !defined(WORDS_BIGENDIAN) + + // Functions to do unaligned loads and stores in little-endian order. + static uint16 Load16(const void *p) { + return ToHost16(UNALIGNED_LOAD16(p)); + } + + static void Store16(void *p, uint16 v) { + UNALIGNED_STORE16(p, FromHost16(v)); + } + + static uint32 Load32(const void *p) { + return ToHost32(UNALIGNED_LOAD32(p)); + } + + static void Store32(void *p, uint32 v) { + UNALIGNED_STORE32(p, FromHost32(v)); + } +}; + +// Some bit-manipulation functions. +class Bits { + public: + // Return floor(log2(n)) for positive integer n. Returns -1 iff n == 0. + static int Log2Floor(uint32 n); + + // Return the first set least / most significant bit, 0-indexed. Returns an + // undefined value if n == 0. FindLSBSetNonZero() is similar to ffs() except + // that it's 0-indexed. + static int FindLSBSetNonZero(uint32 n); + static int FindLSBSetNonZero64(uint64 n); + + private: + DISALLOW_COPY_AND_ASSIGN(Bits); +}; + +#ifdef HAVE_BUILTIN_CTZ + +inline int Bits::Log2Floor(uint32 n) { + return n == 0 ? -1 : 31 ^ __builtin_clz(n); +} + +inline int Bits::FindLSBSetNonZero(uint32 n) { + return __builtin_ctz(n); +} + +inline int Bits::FindLSBSetNonZero64(uint64 n) { + return __builtin_ctzll(n); +} + +#else // Portable versions. + +inline int Bits::Log2Floor(uint32 n) { + if (n == 0) + return -1; + int log = 0; + uint32 value = n; + for (int i = 4; i >= 0; --i) { + int shift = (1 << i); + uint32 x = value >> shift; + if (x != 0) { + value = x; + log += shift; + } + } + assert(value == 1); + return log; +} + +inline int Bits::FindLSBSetNonZero(uint32 n) { + int rc = 31; + for (int i = 4, shift = 1 << 4; i >= 0; --i) { + const uint32 x = n << shift; + if (x != 0) { + n = x; + rc -= shift; + } + shift >>= 1; + } + return rc; +} + +// FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero(). +inline int Bits::FindLSBSetNonZero64(uint64 n) { + const uint32 bottombits = static_cast<uint32>(n); + if (bottombits == 0) { + // Bottom bits are zero, so scan in top bits + return 32 + FindLSBSetNonZero(static_cast<uint32>(n >> 32)); + } else { + return FindLSBSetNonZero(bottombits); + } +} + +#endif // End portable versions. + +// Variable-length integer encoding. +class Varint { + public: + // Maximum lengths of varint encoding of uint32. + static const int kMax32 = 5; + + // Attempts to parse a varint32 from a prefix of the bytes in [ptr,limit-1]. + // Never reads a character at or beyond limit. If a valid/terminated varint32 + // was found in the range, stores it in *OUTPUT and returns a pointer just + // past the last byte of the varint32. Else returns NULL. On success, + // "result <= limit". + static const char* Parse32WithLimit(const char* ptr, const char* limit, + uint32* OUTPUT); + + // REQUIRES "ptr" points to a buffer of length sufficient to hold "v". + // EFFECTS Encodes "v" into "ptr" and returns a pointer to the + // byte just past the last encoded byte. + static char* Encode32(char* ptr, uint32 v); + + // EFFECTS Appends the varint representation of "value" to "*s". + static void Append32(string* s, uint32 value); +}; + +inline const char* Varint::Parse32WithLimit(const char* p, + const char* l, + uint32* OUTPUT) { + const unsigned char* ptr = reinterpret_cast<const unsigned char*>(p); + const unsigned char* limit = reinterpret_cast<const unsigned char*>(l); + uint32 b, result; + if (ptr >= limit) return NULL; + b = *(ptr++); result = b & 127; if (b < 128) goto done; + if (ptr >= limit) return NULL; + b = *(ptr++); result |= (b & 127) << 7; if (b < 128) goto done; + if (ptr >= limit) return NULL; + b = *(ptr++); result |= (b & 127) << 14; if (b < 128) goto done; + if (ptr >= limit) return NULL; + b = *(ptr++); result |= (b & 127) << 21; if (b < 128) goto done; + if (ptr >= limit) return NULL; + b = *(ptr++); result |= (b & 127) << 28; if (b < 16) goto done; + return NULL; // Value is too long to be a varint32 + done: + *OUTPUT = result; + return reinterpret_cast<const char*>(ptr); +} + +inline char* Varint::Encode32(char* sptr, uint32 v) { + // Operate on characters as unsigneds + unsigned char* ptr = reinterpret_cast<unsigned char*>(sptr); + static const int B = 128; + if (v < (1<<7)) { + *(ptr++) = v; + } else if (v < (1<<14)) { + *(ptr++) = v | B; + *(ptr++) = v>>7; + } else if (v < (1<<21)) { + *(ptr++) = v | B; + *(ptr++) = (v>>7) | B; + *(ptr++) = v>>14; + } else if (v < (1<<28)) { + *(ptr++) = v | B; + *(ptr++) = (v>>7) | B; + *(ptr++) = (v>>14) | B; + *(ptr++) = v>>21; + } else { + *(ptr++) = v | B; + *(ptr++) = (v>>7) | B; + *(ptr++) = (v>>14) | B; + *(ptr++) = (v>>21) | B; + *(ptr++) = v>>28; + } + return reinterpret_cast<char*>(ptr); +} + +// If you know the internal layout of the std::string in use, you can +// replace this function with one that resizes the string without +// filling the new space with zeros (if applicable) -- +// it will be non-portable but faster. +inline void STLStringResizeUninitialized(string* s, size_t new_size) { + s->resize(new_size); +} + +// Return a mutable char* pointing to a string's internal buffer, +// which may not be null-terminated. Writing through this pointer will +// modify the string. +// +// string_as_array(&str)[i] is valid for 0 <= i < str.size() until the +// next call to a string method that invalidates iterators. +// +// As of 2006-04, there is no standard-blessed way of getting a +// mutable reference to a string's internal buffer. However, issue 530 +// (http://www.open-std.org/JTC1/SC22/WG21/docs/lwg-defects.html#530) +// proposes this as the method. It will officially be part of the standard +// for C++0x. This should already work on all current implementations. +inline char* string_as_array(string* str) { + return str->empty() ? NULL : &*str->begin(); +} + +} // namespace snappy + +#endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ diff --git a/app/snappy/snappy-stubs-public.h b/app/snappy/snappy-stubs-public.h new file mode 100644 index 00000000..96d88650 --- /dev/null +++ b/app/snappy/snappy-stubs-public.h @@ -0,0 +1,85 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// Author: sesse@google.com (Steinar H. Gunderson) +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Various type stubs for the open-source version of Snappy. +// +// This file cannot include config.h, as it is included from snappy.h, +// which is a public header. Instead, snappy-stubs-public.h is generated by +// from snappy-stubs-public.h.in at configure time. + +#ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ +#define UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ + +#if 1 +#include <stdint.h> +#endif + +#if 1 +#include <stddef.h> +#endif + +#define SNAPPY_MAJOR 1 +#define SNAPPY_MINOR 1 +#define SNAPPY_PATCHLEVEL 0 +#define SNAPPY_VERSION \ + ((SNAPPY_MAJOR << 16) | (SNAPPY_MINOR << 8) | SNAPPY_PATCHLEVEL) + +#include <string> + +namespace snappy { + +#if 1 +typedef int8_t int8; +typedef uint8_t uint8; +typedef int16_t int16; +typedef uint16_t uint16; +typedef int32_t int32; +typedef uint32_t uint32; +typedef int64_t int64; +typedef uint64_t uint64; +#else +typedef signed char int8; +typedef unsigned char uint8; +typedef short int16; +typedef unsigned short uint16; +typedef int int32; +typedef unsigned int uint32; +typedef long long int64; +typedef unsigned long long uint64; +#endif + +typedef std::string string; + +#define DISALLOW_COPY_AND_ASSIGN(TypeName) \ + TypeName(const TypeName&); \ + void operator=(const TypeName&) + +} // namespace snappy + +#endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ diff --git a/app/snappy/snappy-stubs-public.h.in b/app/snappy/snappy-stubs-public.h.in new file mode 100644 index 00000000..f0babcbe --- /dev/null +++ b/app/snappy/snappy-stubs-public.h.in @@ -0,0 +1,85 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// Author: sesse@google.com (Steinar H. Gunderson) +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Various type stubs for the open-source version of Snappy. +// +// This file cannot include config.h, as it is included from snappy.h, +// which is a public header. Instead, snappy-stubs-public.h is generated by +// from snappy-stubs-public.h.in at configure time. + +#ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ +#define UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ + +#if @ac_cv_have_stdint_h@ +#include <stdint.h> +#endif + +#if @ac_cv_have_stddef_h@ +#include <stddef.h> +#endif + +#define SNAPPY_MAJOR @SNAPPY_MAJOR@ +#define SNAPPY_MINOR @SNAPPY_MINOR@ +#define SNAPPY_PATCHLEVEL @SNAPPY_PATCHLEVEL@ +#define SNAPPY_VERSION \ + ((SNAPPY_MAJOR << 16) | (SNAPPY_MINOR << 8) | SNAPPY_PATCHLEVEL) + +#include <string> + +namespace snappy { + +#if @ac_cv_have_stdint_h@ +typedef int8_t int8; +typedef uint8_t uint8; +typedef int16_t int16; +typedef uint16_t uint16; +typedef int32_t int32; +typedef uint32_t uint32; +typedef int64_t int64; +typedef uint64_t uint64; +#else +typedef signed char int8; +typedef unsigned char uint8; +typedef short int16; +typedef unsigned short uint16; +typedef int int32; +typedef unsigned int uint32; +typedef long long int64; +typedef unsigned long long uint64; +#endif + +typedef std::string string; + +#define DISALLOW_COPY_AND_ASSIGN(TypeName) \ + TypeName(const TypeName&); \ + void operator=(const TypeName&) + +} // namespace snappy + +#endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ diff --git a/app/snappy/snappy-test.cc b/app/snappy/snappy-test.cc new file mode 100644 index 00000000..46194109 --- /dev/null +++ b/app/snappy/snappy-test.cc @@ -0,0 +1,606 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Various stubs for the unit tests for the open-source version of Snappy. + +#include "snappy-test.h" + +#ifdef HAVE_WINDOWS_H +#define WIN32_LEAN_AND_MEAN +#include <windows.h> +#endif + +#include <algorithm> + +DEFINE_bool(run_microbenchmarks, true, + "Run microbenchmarks before doing anything else."); + +namespace snappy { + +string ReadTestDataFile(const string& base, size_t size_limit) { + string contents; + const char* srcdir = getenv("srcdir"); // This is set by Automake. + string prefix; + if (srcdir) { + prefix = string(srcdir) + "/"; + } + file::GetContents(prefix + "testdata/" + base, &contents, file::Defaults() + ).CheckSuccess(); + if (size_limit > 0) { + contents = contents.substr(0, size_limit); + } + return contents; +} + +string ReadTestDataFile(const string& base) { + return ReadTestDataFile(base, 0); +} + +string StringPrintf(const char* format, ...) { + char buf[4096]; + va_list ap; + va_start(ap, format); + vsnprintf(buf, sizeof(buf), format, ap); + va_end(ap); + return buf; +} + +bool benchmark_running = false; +int64 benchmark_real_time_us = 0; +int64 benchmark_cpu_time_us = 0; +string *benchmark_label = NULL; +int64 benchmark_bytes_processed = 0; + +void ResetBenchmarkTiming() { + benchmark_real_time_us = 0; + benchmark_cpu_time_us = 0; +} + +#ifdef WIN32 +LARGE_INTEGER benchmark_start_real; +FILETIME benchmark_start_cpu; +#else // WIN32 +struct timeval benchmark_start_real; +struct rusage benchmark_start_cpu; +#endif // WIN32 + +void StartBenchmarkTiming() { +#ifdef WIN32 + QueryPerformanceCounter(&benchmark_start_real); + FILETIME dummy; + CHECK(GetProcessTimes( + GetCurrentProcess(), &dummy, &dummy, &dummy, &benchmark_start_cpu)); +#else + gettimeofday(&benchmark_start_real, NULL); + if (getrusage(RUSAGE_SELF, &benchmark_start_cpu) == -1) { + perror("getrusage(RUSAGE_SELF)"); + exit(1); + } +#endif + benchmark_running = true; +} + +void StopBenchmarkTiming() { + if (!benchmark_running) { + return; + } + +#ifdef WIN32 + LARGE_INTEGER benchmark_stop_real; + LARGE_INTEGER benchmark_frequency; + QueryPerformanceCounter(&benchmark_stop_real); + QueryPerformanceFrequency(&benchmark_frequency); + + double elapsed_real = static_cast<double>( + benchmark_stop_real.QuadPart - benchmark_start_real.QuadPart) / + benchmark_frequency.QuadPart; + benchmark_real_time_us += elapsed_real * 1e6 + 0.5; + + FILETIME benchmark_stop_cpu, dummy; + CHECK(GetProcessTimes( + GetCurrentProcess(), &dummy, &dummy, &dummy, &benchmark_stop_cpu)); + + ULARGE_INTEGER start_ulargeint; + start_ulargeint.LowPart = benchmark_start_cpu.dwLowDateTime; + start_ulargeint.HighPart = benchmark_start_cpu.dwHighDateTime; + + ULARGE_INTEGER stop_ulargeint; + stop_ulargeint.LowPart = benchmark_stop_cpu.dwLowDateTime; + stop_ulargeint.HighPart = benchmark_stop_cpu.dwHighDateTime; + + benchmark_cpu_time_us += + (stop_ulargeint.QuadPart - start_ulargeint.QuadPart + 5) / 10; +#else // WIN32 + struct timeval benchmark_stop_real; + gettimeofday(&benchmark_stop_real, NULL); + benchmark_real_time_us += + 1000000 * (benchmark_stop_real.tv_sec - benchmark_start_real.tv_sec); + benchmark_real_time_us += + (benchmark_stop_real.tv_usec - benchmark_start_real.tv_usec); + + struct rusage benchmark_stop_cpu; + if (getrusage(RUSAGE_SELF, &benchmark_stop_cpu) == -1) { + perror("getrusage(RUSAGE_SELF)"); + exit(1); + } + benchmark_cpu_time_us += 1000000 * (benchmark_stop_cpu.ru_utime.tv_sec - + benchmark_start_cpu.ru_utime.tv_sec); + benchmark_cpu_time_us += (benchmark_stop_cpu.ru_utime.tv_usec - + benchmark_start_cpu.ru_utime.tv_usec); +#endif // WIN32 + + benchmark_running = false; +} + +void SetBenchmarkLabel(const string& str) { + if (benchmark_label) { + delete benchmark_label; + } + benchmark_label = new string(str); +} + +void SetBenchmarkBytesProcessed(int64 bytes) { + benchmark_bytes_processed = bytes; +} + +struct BenchmarkRun { + int64 real_time_us; + int64 cpu_time_us; +}; + +struct BenchmarkCompareCPUTime { + bool operator() (const BenchmarkRun& a, const BenchmarkRun& b) const { + return a.cpu_time_us < b.cpu_time_us; + } +}; + +void Benchmark::Run() { + for (int test_case_num = start_; test_case_num <= stop_; ++test_case_num) { + // Run a few iterations first to find out approximately how fast + // the benchmark is. + const int kCalibrateIterations = 100; + ResetBenchmarkTiming(); + StartBenchmarkTiming(); + (*function_)(kCalibrateIterations, test_case_num); + StopBenchmarkTiming(); + + // Let each test case run for about 200ms, but at least as many + // as we used to calibrate. + // Run five times and pick the median. + const int kNumRuns = 5; + const int kMedianPos = kNumRuns / 2; + int num_iterations = 0; + if (benchmark_real_time_us > 0) { + num_iterations = 200000 * kCalibrateIterations / benchmark_real_time_us; + } + num_iterations = max(num_iterations, kCalibrateIterations); + BenchmarkRun benchmark_runs[kNumRuns]; + + for (int run = 0; run < kNumRuns; ++run) { + ResetBenchmarkTiming(); + StartBenchmarkTiming(); + (*function_)(num_iterations, test_case_num); + StopBenchmarkTiming(); + + benchmark_runs[run].real_time_us = benchmark_real_time_us; + benchmark_runs[run].cpu_time_us = benchmark_cpu_time_us; + } + + string heading = StringPrintf("%s/%d", name_.c_str(), test_case_num); + string human_readable_speed; + + nth_element(benchmark_runs, + benchmark_runs + kMedianPos, + benchmark_runs + kNumRuns, + BenchmarkCompareCPUTime()); + int64 real_time_us = benchmark_runs[kMedianPos].real_time_us; + int64 cpu_time_us = benchmark_runs[kMedianPos].cpu_time_us; + if (cpu_time_us <= 0) { + human_readable_speed = "?"; + } else { + int64 bytes_per_second = + benchmark_bytes_processed * 1000000 / cpu_time_us; + if (bytes_per_second < 1024) { + human_readable_speed = StringPrintf("%dB/s", bytes_per_second); + } else if (bytes_per_second < 1024 * 1024) { + human_readable_speed = StringPrintf( + "%.1fkB/s", bytes_per_second / 1024.0f); + } else if (bytes_per_second < 1024 * 1024 * 1024) { + human_readable_speed = StringPrintf( + "%.1fMB/s", bytes_per_second / (1024.0f * 1024.0f)); + } else { + human_readable_speed = StringPrintf( + "%.1fGB/s", bytes_per_second / (1024.0f * 1024.0f * 1024.0f)); + } + } + + fprintf(stderr, +#ifdef WIN32 + "%-18s %10I64d %10I64d %10d %s %s\n", +#else + "%-18s %10lld %10lld %10d %s %s\n", +#endif + heading.c_str(), + static_cast<long long>(real_time_us * 1000 / num_iterations), + static_cast<long long>(cpu_time_us * 1000 / num_iterations), + num_iterations, + human_readable_speed.c_str(), + benchmark_label->c_str()); + } +} + +#ifdef HAVE_LIBZ + +ZLib::ZLib() + : comp_init_(false), + uncomp_init_(false) { + Reinit(); +} + +ZLib::~ZLib() { + if (comp_init_) { deflateEnd(&comp_stream_); } + if (uncomp_init_) { inflateEnd(&uncomp_stream_); } +} + +void ZLib::Reinit() { + compression_level_ = Z_DEFAULT_COMPRESSION; + window_bits_ = MAX_WBITS; + mem_level_ = 8; // DEF_MEM_LEVEL + if (comp_init_) { + deflateEnd(&comp_stream_); + comp_init_ = false; + } + if (uncomp_init_) { + inflateEnd(&uncomp_stream_); + uncomp_init_ = false; + } + first_chunk_ = true; +} + +void ZLib::Reset() { + first_chunk_ = true; +} + +// --------- COMPRESS MODE + +// Initialization method to be called if we hit an error while +// compressing. On hitting an error, call this method before returning +// the error. +void ZLib::CompressErrorInit() { + deflateEnd(&comp_stream_); + comp_init_ = false; + Reset(); +} + +int ZLib::DeflateInit() { + return deflateInit2(&comp_stream_, + compression_level_, + Z_DEFLATED, + window_bits_, + mem_level_, + Z_DEFAULT_STRATEGY); +} + +int ZLib::CompressInit(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen) { + int err; + + comp_stream_.next_in = (Bytef*)source; + comp_stream_.avail_in = (uInt)*sourceLen; + if ((uLong)comp_stream_.avail_in != *sourceLen) return Z_BUF_ERROR; + comp_stream_.next_out = dest; + comp_stream_.avail_out = (uInt)*destLen; + if ((uLong)comp_stream_.avail_out != *destLen) return Z_BUF_ERROR; + + if ( !first_chunk_ ) // only need to set up stream the first time through + return Z_OK; + + if (comp_init_) { // we've already initted it + err = deflateReset(&comp_stream_); + if (err != Z_OK) { + LOG(WARNING) << "ERROR: Can't reset compress object; creating a new one"; + deflateEnd(&comp_stream_); + comp_init_ = false; + } + } + if (!comp_init_) { // first use + comp_stream_.zalloc = (alloc_func)0; + comp_stream_.zfree = (free_func)0; + comp_stream_.opaque = (voidpf)0; + err = DeflateInit(); + if (err != Z_OK) return err; + comp_init_ = true; + } + return Z_OK; +} + +// In a perfect world we'd always have the full buffer to compress +// when the time came, and we could just call Compress(). Alas, we +// want to do chunked compression on our webserver. In this +// application, we compress the header, send it off, then compress the +// results, send them off, then compress the footer. Thus we need to +// use the chunked compression features of zlib. +int ZLib::CompressAtMostOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen, + int flush_mode) { // Z_FULL_FLUSH or Z_FINISH + int err; + + if ( (err=CompressInit(dest, destLen, source, sourceLen)) != Z_OK ) + return err; + + // This is used to figure out how many bytes we wrote *this chunk* + int compressed_size = comp_stream_.total_out; + + // Some setup happens only for the first chunk we compress in a run + if ( first_chunk_ ) { + first_chunk_ = false; + } + + // flush_mode is Z_FINISH for all mode, Z_SYNC_FLUSH for incremental + // compression. + err = deflate(&comp_stream_, flush_mode); + + *sourceLen = comp_stream_.avail_in; + + if ((err == Z_STREAM_END || err == Z_OK) + && comp_stream_.avail_in == 0 + && comp_stream_.avail_out != 0 ) { + // we processed everything ok and the output buffer was large enough. + ; + } else if (err == Z_STREAM_END && comp_stream_.avail_in > 0) { + return Z_BUF_ERROR; // should never happen + } else if (err != Z_OK && err != Z_STREAM_END && err != Z_BUF_ERROR) { + // an error happened + CompressErrorInit(); + return err; + } else if (comp_stream_.avail_out == 0) { // not enough space + err = Z_BUF_ERROR; + } + + assert(err == Z_OK || err == Z_STREAM_END || err == Z_BUF_ERROR); + if (err == Z_STREAM_END) + err = Z_OK; + + // update the crc and other metadata + compressed_size = comp_stream_.total_out - compressed_size; // delta + *destLen = compressed_size; + + return err; +} + +int ZLib::CompressChunkOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen, + int flush_mode) { // Z_FULL_FLUSH or Z_FINISH + const int ret = + CompressAtMostOrAll(dest, destLen, source, &sourceLen, flush_mode); + if (ret == Z_BUF_ERROR) + CompressErrorInit(); + return ret; +} + +// This routine only initializes the compression stream once. Thereafter, it +// just does a deflateReset on the stream, which should be faster. +int ZLib::Compress(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen) { + int err; + if ( (err=CompressChunkOrAll(dest, destLen, source, sourceLen, + Z_FINISH)) != Z_OK ) + return err; + Reset(); // reset for next call to Compress + + return Z_OK; +} + + +// --------- UNCOMPRESS MODE + +int ZLib::InflateInit() { + return inflateInit2(&uncomp_stream_, MAX_WBITS); +} + +// Initialization method to be called if we hit an error while +// uncompressing. On hitting an error, call this method before +// returning the error. +void ZLib::UncompressErrorInit() { + inflateEnd(&uncomp_stream_); + uncomp_init_ = false; + Reset(); +} + +int ZLib::UncompressInit(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen) { + int err; + + uncomp_stream_.next_in = (Bytef*)source; + uncomp_stream_.avail_in = (uInt)*sourceLen; + // Check for source > 64K on 16-bit machine: + if ((uLong)uncomp_stream_.avail_in != *sourceLen) return Z_BUF_ERROR; + + uncomp_stream_.next_out = dest; + uncomp_stream_.avail_out = (uInt)*destLen; + if ((uLong)uncomp_stream_.avail_out != *destLen) return Z_BUF_ERROR; + + if ( !first_chunk_ ) // only need to set up stream the first time through + return Z_OK; + + if (uncomp_init_) { // we've already initted it + err = inflateReset(&uncomp_stream_); + if (err != Z_OK) { + LOG(WARNING) + << "ERROR: Can't reset uncompress object; creating a new one"; + UncompressErrorInit(); + } + } + if (!uncomp_init_) { + uncomp_stream_.zalloc = (alloc_func)0; + uncomp_stream_.zfree = (free_func)0; + uncomp_stream_.opaque = (voidpf)0; + err = InflateInit(); + if (err != Z_OK) return err; + uncomp_init_ = true; + } + return Z_OK; +} + +// If you compressed your data a chunk at a time, with CompressChunk, +// you can uncompress it a chunk at a time with UncompressChunk. +// Only difference bewteen chunked and unchunked uncompression +// is the flush mode we use: Z_SYNC_FLUSH (chunked) or Z_FINISH (unchunked). +int ZLib::UncompressAtMostOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen, + int flush_mode) { // Z_SYNC_FLUSH or Z_FINISH + int err = Z_OK; + + if ( (err=UncompressInit(dest, destLen, source, sourceLen)) != Z_OK ) { + LOG(WARNING) << "UncompressInit: Error: " << err << " SourceLen: " + << *sourceLen; + return err; + } + + // This is used to figure out how many output bytes we wrote *this chunk*: + const uLong old_total_out = uncomp_stream_.total_out; + + // This is used to figure out how many input bytes we read *this chunk*: + const uLong old_total_in = uncomp_stream_.total_in; + + // Some setup happens only for the first chunk we compress in a run + if ( first_chunk_ ) { + first_chunk_ = false; // so we don't do this again + + // For the first chunk *only* (to avoid infinite troubles), we let + // there be no actual data to uncompress. This sometimes triggers + // when the input is only the gzip header, say. + if ( *sourceLen == 0 ) { + *destLen = 0; + return Z_OK; + } + } + + // We'll uncompress as much as we can. If we end OK great, otherwise + // if we get an error that seems to be the gzip footer, we store the + // gzip footer and return OK, otherwise we return the error. + + // flush_mode is Z_SYNC_FLUSH for chunked mode, Z_FINISH for all mode. + err = inflate(&uncomp_stream_, flush_mode); + + // Figure out how many bytes of the input zlib slurped up: + const uLong bytes_read = uncomp_stream_.total_in - old_total_in; + CHECK_LE(source + bytes_read, source + *sourceLen); + *sourceLen = uncomp_stream_.avail_in; + + if ((err == Z_STREAM_END || err == Z_OK) // everything went ok + && uncomp_stream_.avail_in == 0) { // and we read it all + ; + } else if (err == Z_STREAM_END && uncomp_stream_.avail_in > 0) { + LOG(WARNING) + << "UncompressChunkOrAll: Received some extra data, bytes total: " + << uncomp_stream_.avail_in << " bytes: " + << string(reinterpret_cast<const char *>(uncomp_stream_.next_in), + min(int(uncomp_stream_.avail_in), 20)); + UncompressErrorInit(); + return Z_DATA_ERROR; // what's the extra data for? + } else if (err != Z_OK && err != Z_STREAM_END && err != Z_BUF_ERROR) { + // an error happened + LOG(WARNING) << "UncompressChunkOrAll: Error: " << err + << " avail_out: " << uncomp_stream_.avail_out; + UncompressErrorInit(); + return err; + } else if (uncomp_stream_.avail_out == 0) { + err = Z_BUF_ERROR; + } + + assert(err == Z_OK || err == Z_BUF_ERROR || err == Z_STREAM_END); + if (err == Z_STREAM_END) + err = Z_OK; + + *destLen = uncomp_stream_.total_out - old_total_out; // size for this call + + return err; +} + +int ZLib::UncompressChunkOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen, + int flush_mode) { // Z_SYNC_FLUSH or Z_FINISH + const int ret = + UncompressAtMostOrAll(dest, destLen, source, &sourceLen, flush_mode); + if (ret == Z_BUF_ERROR) + UncompressErrorInit(); + return ret; +} + +int ZLib::UncompressAtMost(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen) { + return UncompressAtMostOrAll(dest, destLen, source, sourceLen, Z_SYNC_FLUSH); +} + +// We make sure we've uncompressed everything, that is, the current +// uncompress stream is at a compressed-buffer-EOF boundary. In gzip +// mode, we also check the gzip footer to make sure we pass the gzip +// consistency checks. We RETURN true iff both types of checks pass. +bool ZLib::UncompressChunkDone() { + assert(!first_chunk_ && uncomp_init_); + // Make sure we're at the end-of-compressed-data point. This means + // if we call inflate with Z_FINISH we won't consume any input or + // write any output + Bytef dummyin, dummyout; + uLongf dummylen = 0; + if ( UncompressChunkOrAll(&dummyout, &dummylen, &dummyin, 0, Z_FINISH) + != Z_OK ) { + return false; + } + + // Make sure that when we exit, we can start a new round of chunks later + Reset(); + + return true; +} + +// Uncompresses the source buffer into the destination buffer. +// The destination buffer must be long enough to hold the entire +// decompressed contents. +// +// We only initialize the uncomp_stream once. Thereafter, we use +// inflateReset, which should be faster. +// +// Returns Z_OK on success, otherwise, it returns a zlib error code. +int ZLib::Uncompress(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen) { + int err; + if ( (err=UncompressChunkOrAll(dest, destLen, source, sourceLen, + Z_FINISH)) != Z_OK ) { + Reset(); // let us try to compress again + return err; + } + if ( !UncompressChunkDone() ) // calls Reset() + return Z_DATA_ERROR; + return Z_OK; // stream_end is ok +} + +#endif // HAVE_LIBZ + +} // namespace snappy diff --git a/app/snappy/snappy-test.h b/app/snappy/snappy-test.h new file mode 100644 index 00000000..f7ba79ed --- /dev/null +++ b/app/snappy/snappy-test.h @@ -0,0 +1,580 @@ +// Copyright 2011 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// Various stubs for the unit tests for the open-source version of Snappy. + +#ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_TEST_H_ +#define UTIL_SNAPPY_OPENSOURCE_SNAPPY_TEST_H_ + +#include <iostream> +#include <string> + +#include "snappy-stubs-internal.h" + +#include <stdio.h> +#include <stdarg.h> + +#ifdef HAVE_SYS_MMAN_H +#include <sys/mman.h> +#endif + +#ifdef HAVE_SYS_RESOURCE_H +#include <sys/resource.h> +#endif + +#ifdef HAVE_SYS_TIME_H +#include <sys/time.h> +#endif + +#ifdef HAVE_WINDOWS_H +#define WIN32_LEAN_AND_MEAN +#include <windows.h> +#endif + +#include <string> + +#ifdef HAVE_GTEST + +#include <gtest/gtest.h> +#undef TYPED_TEST +#define TYPED_TEST TEST +#define INIT_GTEST(argc, argv) ::testing::InitGoogleTest(argc, *argv) + +#else + +// Stubs for if the user doesn't have Google Test installed. + +#define TEST(test_case, test_subcase) \ + void Test_ ## test_case ## _ ## test_subcase() +#define INIT_GTEST(argc, argv) + +#define TYPED_TEST TEST +#define EXPECT_EQ CHECK_EQ +#define EXPECT_NE CHECK_NE +#define EXPECT_FALSE(cond) CHECK(!(cond)) + +#endif + +#ifdef HAVE_GFLAGS + +#include <gflags/gflags.h> + +// This is tricky; both gflags and Google Test want to look at the command line +// arguments. Google Test seems to be the most happy with unknown arguments, +// though, so we call it first and hope for the best. +#define InitGoogle(argv0, argc, argv, remove_flags) \ + INIT_GTEST(argc, argv); \ + google::ParseCommandLineFlags(argc, argv, remove_flags); + +#else + +// If we don't have the gflags package installed, these can only be +// changed at compile time. +#define DEFINE_int32(flag_name, default_value, description) \ + static int FLAGS_ ## flag_name = default_value; + +#define InitGoogle(argv0, argc, argv, remove_flags) \ + INIT_GTEST(argc, argv) + +#endif + +#ifdef HAVE_LIBZ +#include "zlib.h" +#endif + +#ifdef HAVE_LIBLZO2 +#include "lzo/lzo1x.h" +#endif + +#ifdef HAVE_LIBLZF +extern "C" { +#include "lzf.h" +} +#endif + +#ifdef HAVE_LIBFASTLZ +#include "fastlz.h" +#endif + +#ifdef HAVE_LIBQUICKLZ +#include "quicklz.h" +#endif + +namespace { + +namespace File { + void Init() { } +} // namespace File + +namespace file { + int Defaults() { } + + class DummyStatus { + public: + void CheckSuccess() { } + }; + + DummyStatus GetContents(const string& filename, string* data, int unused) { + FILE* fp = fopen(filename.c_str(), "rb"); + if (fp == NULL) { + perror(filename.c_str()); + exit(1); + } + + data->clear(); + while (!feof(fp)) { + char buf[4096]; + size_t ret = fread(buf, 1, 4096, fp); + if (ret == 0 && ferror(fp)) { + perror("fread"); + exit(1); + } + data->append(string(buf, ret)); + } + + fclose(fp); + } + + DummyStatus SetContents(const string& filename, + const string& str, + int unused) { + FILE* fp = fopen(filename.c_str(), "wb"); + if (fp == NULL) { + perror(filename.c_str()); + exit(1); + } + + int ret = fwrite(str.data(), str.size(), 1, fp); + if (ret != 1) { + perror("fwrite"); + exit(1); + } + + fclose(fp); + } +} // namespace file + +} // namespace + +namespace snappy { + +#define FLAGS_test_random_seed 301 +typedef string TypeParam; + +void Test_CorruptedTest_VerifyCorrupted(); +void Test_Snappy_SimpleTests(); +void Test_Snappy_MaxBlowup(); +void Test_Snappy_RandomData(); +void Test_Snappy_FourByteOffset(); +void Test_SnappyCorruption_TruncatedVarint(); +void Test_SnappyCorruption_UnterminatedVarint(); +void Test_Snappy_ReadPastEndOfBuffer(); +void Test_Snappy_FindMatchLength(); +void Test_Snappy_FindMatchLengthRandom(); + +string ReadTestDataFile(const string& base, size_t size_limit); + +string ReadTestDataFile(const string& base); + +// A sprintf() variant that returns a std::string. +// Not safe for general use due to truncation issues. +string StringPrintf(const char* format, ...); + +// A simple, non-cryptographically-secure random generator. +class ACMRandom { + public: + explicit ACMRandom(uint32 seed) : seed_(seed) {} + + int32 Next(); + + int32 Uniform(int32 n) { + return Next() % n; + } + uint8 Rand8() { + return static_cast<uint8>((Next() >> 1) & 0x000000ff); + } + bool OneIn(int X) { return Uniform(X) == 0; } + + // Skewed: pick "base" uniformly from range [0,max_log] and then + // return "base" random bits. The effect is to pick a number in the + // range [0,2^max_log-1] with bias towards smaller numbers. + int32 Skewed(int max_log); + + private: + static const uint32 M = 2147483647L; // 2^31-1 + uint32 seed_; +}; + +inline int32 ACMRandom::Next() { + static const uint64 A = 16807; // bits 14, 8, 7, 5, 2, 1, 0 + // We are computing + // seed_ = (seed_ * A) % M, where M = 2^31-1 + // + // seed_ must not be zero or M, or else all subsequent computed values + // will be zero or M respectively. For all other values, seed_ will end + // up cycling through every number in [1,M-1] + uint64 product = seed_ * A; + + // Compute (product % M) using the fact that ((x << 31) % M) == x. + seed_ = (product >> 31) + (product & M); + // The first reduction may overflow by 1 bit, so we may need to repeat. + // mod == M is not possible; using > allows the faster sign-bit-based test. + if (seed_ > M) { + seed_ -= M; + } + return seed_; +} + +inline int32 ACMRandom::Skewed(int max_log) { + const int32 base = (Next() - 1) % (max_log+1); + return (Next() - 1) & ((1u << base)-1); +} + +// A wall-time clock. This stub is not super-accurate, nor resistant to the +// system time changing. +class CycleTimer { + public: + CycleTimer() : real_time_us_(0) {} + + void Start() { +#ifdef WIN32 + QueryPerformanceCounter(&start_); +#else + gettimeofday(&start_, NULL); +#endif + } + + void Stop() { +#ifdef WIN32 + LARGE_INTEGER stop; + LARGE_INTEGER frequency; + QueryPerformanceCounter(&stop); + QueryPerformanceFrequency(&frequency); + + double elapsed = static_cast<double>(stop.QuadPart - start_.QuadPart) / + frequency.QuadPart; + real_time_us_ += elapsed * 1e6 + 0.5; +#else + struct timeval stop; + gettimeofday(&stop, NULL); + + real_time_us_ += 1000000 * (stop.tv_sec - start_.tv_sec); + real_time_us_ += (stop.tv_usec - start_.tv_usec); +#endif + } + + double Get() { + return real_time_us_ * 1e-6; + } + + private: + int64 real_time_us_; +#ifdef WIN32 + LARGE_INTEGER start_; +#else + struct timeval start_; +#endif +}; + +// Minimalistic microbenchmark framework. + +typedef void (*BenchmarkFunction)(int, int); + +class Benchmark { + public: + Benchmark(const string& name, BenchmarkFunction function) : + name_(name), function_(function) {} + + Benchmark* DenseRange(int start, int stop) { + start_ = start; + stop_ = stop; + return this; + } + + void Run(); + + private: + const string name_; + const BenchmarkFunction function_; + int start_, stop_; +}; +#define BENCHMARK(benchmark_name) \ + Benchmark* Benchmark_ ## benchmark_name = \ + (new Benchmark(#benchmark_name, benchmark_name)) + +extern Benchmark* Benchmark_BM_UFlat; +extern Benchmark* Benchmark_BM_UValidate; +extern Benchmark* Benchmark_BM_ZFlat; + +void ResetBenchmarkTiming(); +void StartBenchmarkTiming(); +void StopBenchmarkTiming(); +void SetBenchmarkLabel(const string& str); +void SetBenchmarkBytesProcessed(int64 bytes); + +#ifdef HAVE_LIBZ + +// Object-oriented wrapper around zlib. +class ZLib { + public: + ZLib(); + ~ZLib(); + + // Wipe a ZLib object to a virgin state. This differs from Reset() + // in that it also breaks any state. + void Reinit(); + + // Call this to make a zlib buffer as good as new. Here's the only + // case where they differ: + // CompressChunk(a); CompressChunk(b); CompressChunkDone(); vs + // CompressChunk(a); Reset(); CompressChunk(b); CompressChunkDone(); + // You'll want to use Reset(), then, when you interrupt a compress + // (or uncompress) in the middle of a chunk and want to start over. + void Reset(); + + // According to the zlib manual, when you Compress, the destination + // buffer must have size at least src + .1%*src + 12. This function + // helps you calculate that. Augment this to account for a potential + // gzip header and footer, plus a few bytes of slack. + static int MinCompressbufSize(int uncompress_size) { + return uncompress_size + uncompress_size/1000 + 40; + } + + // Compresses the source buffer into the destination buffer. + // sourceLen is the byte length of the source buffer. + // Upon entry, destLen is the total size of the destination buffer, + // which must be of size at least MinCompressbufSize(sourceLen). + // Upon exit, destLen is the actual size of the compressed buffer. + // + // This function can be used to compress a whole file at once if the + // input file is mmap'ed. + // + // Returns Z_OK if success, Z_MEM_ERROR if there was not + // enough memory, Z_BUF_ERROR if there was not enough room in the + // output buffer. Note that if the output buffer is exactly the same + // size as the compressed result, we still return Z_BUF_ERROR. + // (check CL#1936076) + int Compress(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen); + + // Uncompresses the source buffer into the destination buffer. + // The destination buffer must be long enough to hold the entire + // decompressed contents. + // + // Returns Z_OK on success, otherwise, it returns a zlib error code. + int Uncompress(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen); + + // Uncompress data one chunk at a time -- ie you can call this + // more than once. To get this to work you need to call per-chunk + // and "done" routines. + // + // Returns Z_OK if success, Z_MEM_ERROR if there was not + // enough memory, Z_BUF_ERROR if there was not enough room in the + // output buffer. + + int UncompressAtMost(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen); + + // Checks gzip footer information, as needed. Mostly this just + // makes sure the checksums match. Whenever you call this, it + // will assume the last 8 bytes from the previous UncompressChunk + // call are the footer. Returns true iff everything looks ok. + bool UncompressChunkDone(); + + private: + int InflateInit(); // sets up the zlib inflate structure + int DeflateInit(); // sets up the zlib deflate structure + + // These init the zlib data structures for compressing/uncompressing + int CompressInit(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen); + int UncompressInit(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen); + // Initialization method to be called if we hit an error while + // uncompressing. On hitting an error, call this method before + // returning the error. + void UncompressErrorInit(); + + // Helper function for Compress + int CompressChunkOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen, + int flush_mode); + int CompressAtMostOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen, + int flush_mode); + + // Likewise for UncompressAndUncompressChunk + int UncompressChunkOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong sourceLen, + int flush_mode); + + int UncompressAtMostOrAll(Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen, + int flush_mode); + + // Initialization method to be called if we hit an error while + // compressing. On hitting an error, call this method before + // returning the error. + void CompressErrorInit(); + + int compression_level_; // compression level + int window_bits_; // log base 2 of the window size used in compression + int mem_level_; // specifies the amount of memory to be used by + // compressor (1-9) + z_stream comp_stream_; // Zlib stream data structure + bool comp_init_; // True if we have initialized comp_stream_ + z_stream uncomp_stream_; // Zlib stream data structure + bool uncomp_init_; // True if we have initialized uncomp_stream_ + + // These are used only with chunked compression. + bool first_chunk_; // true if we need to emit headers with this chunk +}; + +#endif // HAVE_LIBZ + +} // namespace snappy + +DECLARE_bool(run_microbenchmarks); + +static void RunSpecifiedBenchmarks() { + if (!FLAGS_run_microbenchmarks) { + return; + } + + fprintf(stderr, "Running microbenchmarks.\n"); +#ifndef NDEBUG + fprintf(stderr, "WARNING: Compiled with assertions enabled, will be slow.\n"); +#endif +#ifndef __OPTIMIZE__ + fprintf(stderr, "WARNING: Compiled without optimization, will be slow.\n"); +#endif + fprintf(stderr, "Benchmark Time(ns) CPU(ns) Iterations\n"); + fprintf(stderr, "---------------------------------------------------\n"); + + snappy::Benchmark_BM_UFlat->Run(); + snappy::Benchmark_BM_UValidate->Run(); + snappy::Benchmark_BM_ZFlat->Run(); + + fprintf(stderr, "\n"); +} + +#ifndef HAVE_GTEST + +static inline int RUN_ALL_TESTS() { + fprintf(stderr, "Running correctness tests.\n"); + snappy::Test_CorruptedTest_VerifyCorrupted(); + snappy::Test_Snappy_SimpleTests(); + snappy::Test_Snappy_MaxBlowup(); + snappy::Test_Snappy_RandomData(); + snappy::Test_Snappy_FourByteOffset(); + snappy::Test_SnappyCorruption_TruncatedVarint(); + snappy::Test_SnappyCorruption_UnterminatedVarint(); + snappy::Test_Snappy_ReadPastEndOfBuffer(); + snappy::Test_Snappy_FindMatchLength(); + snappy::Test_Snappy_FindMatchLengthRandom(); + fprintf(stderr, "All tests passed.\n"); + + return 0; +} + +#endif // HAVE_GTEST + +// For main(). +namespace snappy { + +static void CompressFile(const char* fname); +static void UncompressFile(const char* fname); +static void MeasureFile(const char* fname); + +// Logging. + +#define LOG(level) LogMessage() +#define VLOG(level) true ? (void)0 : \ + snappy::LogMessageVoidify() & snappy::LogMessage() + +class LogMessage { + public: + LogMessage() { } + ~LogMessage() { + cerr << endl; + } + + LogMessage& operator<<(const std::string& msg) { + cerr << msg; + return *this; + } + LogMessage& operator<<(int x) { + cerr << x; + return *this; + } +}; + +// Asserts, both versions activated in debug mode only, +// and ones that are always active. + +#define CRASH_UNLESS(condition) \ + PREDICT_TRUE(condition) ? (void)0 : \ + snappy::LogMessageVoidify() & snappy::LogMessageCrash() + +class LogMessageCrash : public LogMessage { + public: + LogMessageCrash() { } + ~LogMessageCrash() { + cerr << endl; + abort(); + } +}; + +// This class is used to explicitly ignore values in the conditional +// logging macros. This avoids compiler warnings like "value computed +// is not used" and "statement has no effect". + +class LogMessageVoidify { + public: + LogMessageVoidify() { } + // This has to be an operator with a precedence lower than << but + // higher than ?: + void operator&(const LogMessage&) { } +}; + +#define CHECK(cond) CRASH_UNLESS(cond) +#define CHECK_LE(a, b) CRASH_UNLESS((a) <= (b)) +#define CHECK_GE(a, b) CRASH_UNLESS((a) >= (b)) +#define CHECK_EQ(a, b) CRASH_UNLESS((a) == (b)) +#define CHECK_NE(a, b) CRASH_UNLESS((a) != (b)) +#define CHECK_LT(a, b) CRASH_UNLESS((a) < (b)) +#define CHECK_GT(a, b) CRASH_UNLESS((a) > (b)) + +} // namespace + +using snappy::CompressFile; +using snappy::UncompressFile; +using snappy::MeasureFile; + +#endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_TEST_H_ diff --git a/app/snappy/snappy.cc b/app/snappy/snappy.cc new file mode 100644 index 00000000..1230321f --- /dev/null +++ b/app/snappy/snappy.cc @@ -0,0 +1,1119 @@ +// Copyright 2005 Google Inc. All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "snappy.h" +#include "snappy-internal.h" +#include "snappy-sinksource.h" + +#include <stdio.h> + +#include <algorithm> +#include <string> +#include <vector> + + +namespace snappy { + +// Any hash function will produce a valid compressed bitstream, but a good +// hash function reduces the number of collisions and thus yields better +// compression for compressible input, and more speed for incompressible +// input. Of course, it doesn't hurt if the hash function is reasonably fast +// either, as it gets called a lot. +static inline uint32 HashBytes(uint32 bytes, int shift) { + uint32 kMul = 0x1e35a7bd; + return (bytes * kMul) >> shift; +} +static inline uint32 Hash(const char* p, int shift) { + return HashBytes(UNALIGNED_LOAD32(p), shift); +} + +size_t MaxCompressedLength(size_t source_len) { + // Compressed data can be defined as: + // compressed := item* literal* + // item := literal* copy + // + // The trailing literal sequence has a space blowup of at most 62/60 + // since a literal of length 60 needs one tag byte + one extra byte + // for length information. + // + // Item blowup is trickier to measure. Suppose the "copy" op copies + // 4 bytes of data. Because of a special check in the encoding code, + // we produce a 4-byte copy only if the offset is < 65536. Therefore + // the copy op takes 3 bytes to encode, and this type of item leads + // to at most the 62/60 blowup for representing literals. + // + // Suppose the "copy" op copies 5 bytes of data. If the offset is big + // enough, it will take 5 bytes to encode the copy op. Therefore the + // worst case here is a one-byte literal followed by a five-byte copy. + // I.e., 6 bytes of input turn into 7 bytes of "compressed" data. + // + // This last factor dominates the blowup, so the final estimate is: + return 32 + source_len + source_len/6; +} + +enum { + LITERAL = 0, + COPY_1_BYTE_OFFSET = 1, // 3 bit length + 3 bits of offset in opcode + COPY_2_BYTE_OFFSET = 2, + COPY_4_BYTE_OFFSET = 3 +}; + +// Copy "len" bytes from "src" to "op", one byte at a time. Used for +// handling COPY operations where the input and output regions may +// overlap. For example, suppose: +// src == "ab" +// op == src + 2 +// len == 20 +// After IncrementalCopy(src, op, len), the result will have +// eleven copies of "ab" +// ababababababababababab +// Note that this does not match the semantics of either memcpy() +// or memmove(). +static inline void IncrementalCopy(const char* src, char* op, int len) { + assert(len > 0); + do { + *op++ = *src++; + } while (--len > 0); +} + +// Equivalent to IncrementalCopy except that it can write up to ten extra +// bytes after the end of the copy, and that it is faster. +// +// The main part of this loop is a simple copy of eight bytes at a time until +// we've copied (at least) the requested amount of bytes. However, if op and +// src are less than eight bytes apart (indicating a repeating pattern of +// length < 8), we first need to expand the pattern in order to get the correct +// results. For instance, if the buffer looks like this, with the eight-byte +// <src> and <op> patterns marked as intervals: +// +// abxxxxxxxxxxxx +// [------] src +// [------] op +// +// a single eight-byte copy from <src> to <op> will repeat the pattern once, +// after which we can move <op> two bytes without moving <src>: +// +// ababxxxxxxxxxx +// [------] src +// [------] op +// +// and repeat the exercise until the two no longer overlap. +// +// This allows us to do very well in the special case of one single byte +// repeated many times, without taking a big hit for more general cases. +// +// The worst case of extra writing past the end of the match occurs when +// op - src == 1 and len == 1; the last copy will read from byte positions +// [0..7] and write to [4..11], whereas it was only supposed to write to +// position 1. Thus, ten excess bytes. + +namespace { + +const int kMaxIncrementCopyOverflow = 10; + +} // namespace + +static inline void IncrementalCopyFastPath(const char* src, char* op, int len) { + while (op - src < 8) { + UnalignedCopy64(src, op); + len -= op - src; + op += op - src; + } + while (len > 0) { + UnalignedCopy64(src, op); + src += 8; + op += 8; + len -= 8; + } +} + +static inline char* EmitLiteral(char* op, + const char* literal, + int len, + bool allow_fast_path) { + int n = len - 1; // Zero-length literals are disallowed + if (n < 60) { + // Fits in tag byte + *op++ = LITERAL | (n << 2); + + // The vast majority of copies are below 16 bytes, for which a + // call to memcpy is overkill. This fast path can sometimes + // copy up to 15 bytes too much, but that is okay in the + // main loop, since we have a bit to go on for both sides: + // + // - The input will always have kInputMarginBytes = 15 extra + // available bytes, as long as we're in the main loop, and + // if not, allow_fast_path = false. + // - The output will always have 32 spare bytes (see + // MaxCompressedLength). + if (allow_fast_path && len <= 16) { + UnalignedCopy64(literal, op); + UnalignedCopy64(literal + 8, op + 8); + return op + len; + } + } else { + // Encode in upcoming bytes + char* base = op; + int count = 0; + op++; + while (n > 0) { + *op++ = n & 0xff; + n >>= 8; + count++; + } + assert(count >= 1); + assert(count <= 4); + *base = LITERAL | ((59+count) << 2); + } + memcpy(op, literal, len); + return op + len; +} + +static inline char* EmitCopyLessThan64(char* op, size_t offset, int len) { + assert(len <= 64); + assert(len >= 4); + assert(offset < 65536); + + if ((len < 12) && (offset < 2048)) { + size_t len_minus_4 = len - 4; + assert(len_minus_4 < 8); // Must fit in 3 bits + *op++ = COPY_1_BYTE_OFFSET + ((len_minus_4) << 2) + ((offset >> 8) << 5); + *op++ = offset & 0xff; + } else { + *op++ = COPY_2_BYTE_OFFSET + ((len-1) << 2); + LittleEndian::Store16(op, offset); + op += 2; + } + return op; +} + +static inline char* EmitCopy(char* op, size_t offset, int len) { + // Emit 64 byte copies but make sure to keep at least four bytes reserved + while (len >= 68) { + op = EmitCopyLessThan64(op, offset, 64); + len -= 64; + } + + // Emit an extra 60 byte copy if have too much data to fit in one copy + if (len > 64) { + op = EmitCopyLessThan64(op, offset, 60); + len -= 60; + } + + // Emit remainder + op = EmitCopyLessThan64(op, offset, len); + return op; +} + + +bool GetUncompressedLength(const char* start, size_t n, size_t* result) { + uint32 v = 0; + const char* limit = start + n; + if (Varint::Parse32WithLimit(start, limit, &v) != NULL) { + *result = v; + return true; + } else { + return false; + } +} + +namespace internal { +uint16* WorkingMemory::GetHashTable(size_t input_size, int* table_size) { + // Use smaller hash table when input.size() is smaller, since we + // fill the table, incurring O(hash table size) overhead for + // compression, and if the input is short, we won't need that + // many hash table entries anyway. + assert(kMaxHashTableSize >= 256); + size_t htsize = 256; + while (htsize < kMaxHashTableSize && htsize < input_size) { + htsize <<= 1; + } + + uint16* table; + if (htsize <= ARRAYSIZE(small_table_)) { + table = small_table_; + } else { + if (large_table_ == NULL) { + large_table_ = new uint16[kMaxHashTableSize]; + } + table = large_table_; + } + + *table_size = htsize; + memset(table, 0, htsize * sizeof(*table)); + return table; +} +} // end namespace internal + +// For 0 <= offset <= 4, GetUint32AtOffset(GetEightBytesAt(p), offset) will +// equal UNALIGNED_LOAD32(p + offset). Motivation: On x86-64 hardware we have +// empirically found that overlapping loads such as +// UNALIGNED_LOAD32(p) ... UNALIGNED_LOAD32(p+1) ... UNALIGNED_LOAD32(p+2) +// are slower than UNALIGNED_LOAD64(p) followed by shifts and casts to uint32. +// +// We have different versions for 64- and 32-bit; ideally we would avoid the +// two functions and just inline the UNALIGNED_LOAD64 call into +// GetUint32AtOffset, but GCC (at least not as of 4.6) is seemingly not clever +// enough to avoid loading the value multiple times then. For 64-bit, the load +// is done when GetEightBytesAt() is called, whereas for 32-bit, the load is +// done at GetUint32AtOffset() time. + +#ifdef ARCH_K8 + +typedef uint64 EightBytesReference; + +static inline EightBytesReference GetEightBytesAt(const char* ptr) { + return UNALIGNED_LOAD64(ptr); +} + +static inline uint32 GetUint32AtOffset(uint64 v, int offset) { + assert(offset >= 0); + assert(offset <= 4); + return v >> (LittleEndian::IsLittleEndian() ? 8 * offset : 32 - 8 * offset); +} + +#else + +typedef const char* EightBytesReference; + +static inline EightBytesReference GetEightBytesAt(const char* ptr) { + return ptr; +} + +static inline uint32 GetUint32AtOffset(const char* v, int offset) { + assert(offset >= 0); + assert(offset <= 4); + return UNALIGNED_LOAD32(v + offset); +} + +#endif + +// Flat array compression that does not emit the "uncompressed length" +// prefix. Compresses "input" string to the "*op" buffer. +// +// REQUIRES: "input" is at most "kBlockSize" bytes long. +// REQUIRES: "op" points to an array of memory that is at least +// "MaxCompressedLength(input.size())" in size. +// REQUIRES: All elements in "table[0..table_size-1]" are initialized to zero. +// REQUIRES: "table_size" is a power of two +// +// Returns an "end" pointer into "op" buffer. +// "end - op" is the compressed size of "input". +namespace internal { +char* CompressFragment(const char* input, + size_t input_size, + char* op, + uint16* table, + const int table_size) { + // "ip" is the input pointer, and "op" is the output pointer. + const char* ip = input; + assert(input_size <= kBlockSize); + assert((table_size & (table_size - 1)) == 0); // table must be power of two + const int shift = 32 - Bits::Log2Floor(table_size); + assert(static_cast<int>(kuint32max >> shift) == table_size - 1); + const char* ip_end = input + input_size; + const char* base_ip = ip; + // Bytes in [next_emit, ip) will be emitted as literal bytes. Or + // [next_emit, ip_end) after the main loop. + const char* next_emit = ip; + + const size_t kInputMarginBytes = 15; + if (PREDICT_TRUE(input_size >= kInputMarginBytes)) { + const char* ip_limit = input + input_size - kInputMarginBytes; + + for (uint32 next_hash = Hash(++ip, shift); ; ) { + assert(next_emit < ip); + // The body of this loop calls EmitLiteral once and then EmitCopy one or + // more times. (The exception is that when we're close to exhausting + // the input we goto emit_remainder.) + // + // In the first iteration of this loop we're just starting, so + // there's nothing to copy, so calling EmitLiteral once is + // necessary. And we only start a new iteration when the + // current iteration has determined that a call to EmitLiteral will + // precede the next call to EmitCopy (if any). + // + // Step 1: Scan forward in the input looking for a 4-byte-long match. + // If we get close to exhausting the input then goto emit_remainder. + // + // Heuristic match skipping: If 32 bytes are scanned with no matches + // found, start looking only at every other byte. If 32 more bytes are + // scanned, look at every third byte, etc.. When a match is found, + // immediately go back to looking at every byte. This is a small loss + // (~5% performance, ~0.1% density) for compressible data due to more + // bookkeeping, but for non-compressible data (such as JPEG) it's a huge + // win since the compressor quickly "realizes" the data is incompressible + // and doesn't bother looking for matches everywhere. + // + // The "skip" variable keeps track of how many bytes there are since the + // last match; dividing it by 32 (ie. right-shifting by five) gives the + // number of bytes to move ahead for each iteration. + uint32 skip = 32; + + const char* next_ip = ip; + const char* candidate; + do { + ip = next_ip; + uint32 hash = next_hash; + assert(hash == Hash(ip, shift)); + uint32 bytes_between_hash_lookups = skip++ >> 5; + next_ip = ip + bytes_between_hash_lookups; + if (PREDICT_FALSE(next_ip > ip_limit)) { + goto emit_remainder; + } + next_hash = Hash(next_ip, shift); + candidate = base_ip + table[hash]; + assert(candidate >= base_ip); + assert(candidate < ip); + + table[hash] = ip - base_ip; + } while (PREDICT_TRUE(UNALIGNED_LOAD32(ip) != + UNALIGNED_LOAD32(candidate))); + + // Step 2: A 4-byte match has been found. We'll later see if more + // than 4 bytes match. But, prior to the match, input + // bytes [next_emit, ip) are unmatched. Emit them as "literal bytes." + assert(next_emit + 16 <= ip_end); + op = EmitLiteral(op, next_emit, ip - next_emit, true); + + // Step 3: Call EmitCopy, and then see if another EmitCopy could + // be our next move. Repeat until we find no match for the + // input immediately after what was consumed by the last EmitCopy call. + // + // If we exit this loop normally then we need to call EmitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can exit + // this loop via goto if we get close to exhausting the input. + EightBytesReference input_bytes; + uint32 candidate_bytes = 0; + + do { + // We have a 4-byte match at ip, and no need to emit any + // "literal bytes" prior to ip. + const char* base = ip; + int matched = 4 + FindMatchLength(candidate + 4, ip + 4, ip_end); + ip += matched; + size_t offset = base - candidate; + assert(0 == memcmp(base, candidate, matched)); + op = EmitCopy(op, offset, matched); + // We could immediately start working at ip now, but to improve + // compression we first update table[Hash(ip - 1, ...)]. + const char* insert_tail = ip - 1; + next_emit = ip; + if (PREDICT_FALSE(ip >= ip_limit)) { + goto emit_remainder; + } + input_bytes = GetEightBytesAt(insert_tail); + uint32 prev_hash = HashBytes(GetUint32AtOffset(input_bytes, 0), shift); + table[prev_hash] = ip - base_ip - 1; + uint32 cur_hash = HashBytes(GetUint32AtOffset(input_bytes, 1), shift); + candidate = base_ip + table[cur_hash]; + candidate_bytes = UNALIGNED_LOAD32(candidate); + table[cur_hash] = ip - base_ip; + } while (GetUint32AtOffset(input_bytes, 1) == candidate_bytes); + + next_hash = HashBytes(GetUint32AtOffset(input_bytes, 2), shift); + ++ip; + } + } + + emit_remainder: + // Emit the remaining bytes as a literal + if (next_emit < ip_end) { + op = EmitLiteral(op, next_emit, ip_end - next_emit, false); + } + + return op; +} +} // end namespace internal + +// Signature of output types needed by decompression code. +// The decompression code is templatized on a type that obeys this +// signature so that we do not pay virtual function call overhead in +// the middle of a tight decompression loop. +// +// class DecompressionWriter { +// public: +// // Called before decompression +// void SetExpectedLength(size_t length); +// +// // Called after decompression +// bool CheckLength() const; +// +// // Called repeatedly during decompression +// bool Append(const char* ip, size_t length); +// bool AppendFromSelf(uint32 offset, size_t length); +// +// // The difference between TryFastAppend and Append is that TryFastAppend +// // is allowed to read up to <available> bytes from the input buffer, +// // whereas Append is allowed to read <length>. +// // +// // Also, TryFastAppend is allowed to return false, declining the append, +// // without it being a fatal error -- just "return false" would be +// // a perfectly legal implementation of TryFastAppend. The intention +// // is for TryFastAppend to allow a fast path in the common case of +// // a small append. +// // +// // NOTE(user): TryFastAppend must always return decline (return false) +// // if <length> is 61 or more, as in this case the literal length is not +// // decoded fully. In practice, this should not be a big problem, +// // as it is unlikely that one would implement a fast path accepting +// // this much data. +// bool TryFastAppend(const char* ip, size_t available, size_t length); +// }; + +// ----------------------------------------------------------------------- +// Lookup table for decompression code. Generated by ComputeTable() below. +// ----------------------------------------------------------------------- + +// Mapping from i in range [0,4] to a mask to extract the bottom 8*i bits +static const uint32 wordmask[] = { + 0u, 0xffu, 0xffffu, 0xffffffu, 0xffffffffu +}; + +// Data stored per entry in lookup table: +// Range Bits-used Description +// ------------------------------------ +// 1..64 0..7 Literal/copy length encoded in opcode byte +// 0..7 8..10 Copy offset encoded in opcode byte / 256 +// 0..4 11..13 Extra bytes after opcode +// +// We use eight bits for the length even though 7 would have sufficed +// because of efficiency reasons: +// (1) Extracting a byte is faster than a bit-field +// (2) It properly aligns copy offset so we do not need a <<8 +static const uint16 char_table[256] = { + 0x0001, 0x0804, 0x1001, 0x2001, 0x0002, 0x0805, 0x1002, 0x2002, + 0x0003, 0x0806, 0x1003, 0x2003, 0x0004, 0x0807, 0x1004, 0x2004, + 0x0005, 0x0808, 0x1005, 0x2005, 0x0006, 0x0809, 0x1006, 0x2006, + 0x0007, 0x080a, 0x1007, 0x2007, 0x0008, 0x080b, 0x1008, 0x2008, + 0x0009, 0x0904, 0x1009, 0x2009, 0x000a, 0x0905, 0x100a, 0x200a, + 0x000b, 0x0906, 0x100b, 0x200b, 0x000c, 0x0907, 0x100c, 0x200c, + 0x000d, 0x0908, 0x100d, 0x200d, 0x000e, 0x0909, 0x100e, 0x200e, + 0x000f, 0x090a, 0x100f, 0x200f, 0x0010, 0x090b, 0x1010, 0x2010, + 0x0011, 0x0a04, 0x1011, 0x2011, 0x0012, 0x0a05, 0x1012, 0x2012, + 0x0013, 0x0a06, 0x1013, 0x2013, 0x0014, 0x0a07, 0x1014, 0x2014, + 0x0015, 0x0a08, 0x1015, 0x2015, 0x0016, 0x0a09, 0x1016, 0x2016, + 0x0017, 0x0a0a, 0x1017, 0x2017, 0x0018, 0x0a0b, 0x1018, 0x2018, + 0x0019, 0x0b04, 0x1019, 0x2019, 0x001a, 0x0b05, 0x101a, 0x201a, + 0x001b, 0x0b06, 0x101b, 0x201b, 0x001c, 0x0b07, 0x101c, 0x201c, + 0x001d, 0x0b08, 0x101d, 0x201d, 0x001e, 0x0b09, 0x101e, 0x201e, + 0x001f, 0x0b0a, 0x101f, 0x201f, 0x0020, 0x0b0b, 0x1020, 0x2020, + 0x0021, 0x0c04, 0x1021, 0x2021, 0x0022, 0x0c05, 0x1022, 0x2022, + 0x0023, 0x0c06, 0x1023, 0x2023, 0x0024, 0x0c07, 0x1024, 0x2024, + 0x0025, 0x0c08, 0x1025, 0x2025, 0x0026, 0x0c09, 0x1026, 0x2026, + 0x0027, 0x0c0a, 0x1027, 0x2027, 0x0028, 0x0c0b, 0x1028, 0x2028, + 0x0029, 0x0d04, 0x1029, 0x2029, 0x002a, 0x0d05, 0x102a, 0x202a, + 0x002b, 0x0d06, 0x102b, 0x202b, 0x002c, 0x0d07, 0x102c, 0x202c, + 0x002d, 0x0d08, 0x102d, 0x202d, 0x002e, 0x0d09, 0x102e, 0x202e, + 0x002f, 0x0d0a, 0x102f, 0x202f, 0x0030, 0x0d0b, 0x1030, 0x2030, + 0x0031, 0x0e04, 0x1031, 0x2031, 0x0032, 0x0e05, 0x1032, 0x2032, + 0x0033, 0x0e06, 0x1033, 0x2033, 0x0034, 0x0e07, 0x1034, 0x2034, + 0x0035, 0x0e08, 0x1035, 0x2035, 0x0036, 0x0e09, 0x1036, 0x2036, + 0x0037, 0x0e0a, 0x1037, 0x2037, 0x0038, 0x0e0b, 0x1038, 0x2038, + 0x0039, 0x0f04, 0x1039, 0x2039, 0x003a, 0x0f05, 0x103a, 0x203a, + 0x003b, 0x0f06, 0x103b, 0x203b, 0x003c, 0x0f07, 0x103c, 0x203c, + 0x0801, 0x0f08, 0x103d, 0x203d, 0x1001, 0x0f09, 0x103e, 0x203e, + 0x1801, 0x0f0a, 0x103f, 0x203f, 0x2001, 0x0f0b, 0x1040, 0x2040 +}; + +// In debug mode, allow optional computation of the table at startup. +// Also, check that the decompression table is correct. +#ifndef NDEBUG +DEFINE_bool(snappy_dump_decompression_table, false, + "If true, we print the decompression table at startup."); + +static uint16 MakeEntry(unsigned int extra, + unsigned int len, + unsigned int copy_offset) { + // Check that all of the fields fit within the allocated space + assert(extra == (extra & 0x7)); // At most 3 bits + assert(copy_offset == (copy_offset & 0x7)); // At most 3 bits + assert(len == (len & 0x7f)); // At most 7 bits + return len | (copy_offset << 8) | (extra << 11); +} + +static void ComputeTable() { + uint16 dst[256]; + + // Place invalid entries in all places to detect missing initialization + int assigned = 0; + for (int i = 0; i < 256; i++) { + dst[i] = 0xffff; + } + + // Small LITERAL entries. We store (len-1) in the top 6 bits. + for (unsigned int len = 1; len <= 60; len++) { + dst[LITERAL | ((len-1) << 2)] = MakeEntry(0, len, 0); + assigned++; + } + + // Large LITERAL entries. We use 60..63 in the high 6 bits to + // encode the number of bytes of length info that follow the opcode. + for (unsigned int extra_bytes = 1; extra_bytes <= 4; extra_bytes++) { + // We set the length field in the lookup table to 1 because extra + // bytes encode len-1. + dst[LITERAL | ((extra_bytes+59) << 2)] = MakeEntry(extra_bytes, 1, 0); + assigned++; + } + + // COPY_1_BYTE_OFFSET. + // + // The tag byte in the compressed data stores len-4 in 3 bits, and + // offset/256 in 5 bits. offset%256 is stored in the next byte. + // + // This format is used for length in range [4..11] and offset in + // range [0..2047] + for (unsigned int len = 4; len < 12; len++) { + for (unsigned int offset = 0; offset < 2048; offset += 256) { + dst[COPY_1_BYTE_OFFSET | ((len-4)<<2) | ((offset>>8)<<5)] = + MakeEntry(1, len, offset>>8); + assigned++; + } + } + + // COPY_2_BYTE_OFFSET. + // Tag contains len-1 in top 6 bits, and offset in next two bytes. + for (unsigned int len = 1; len <= 64; len++) { + dst[COPY_2_BYTE_OFFSET | ((len-1)<<2)] = MakeEntry(2, len, 0); + assigned++; + } + + // COPY_4_BYTE_OFFSET. + // Tag contents len-1 in top 6 bits, and offset in next four bytes. + for (unsigned int len = 1; len <= 64; len++) { + dst[COPY_4_BYTE_OFFSET | ((len-1)<<2)] = MakeEntry(4, len, 0); + assigned++; + } + + // Check that each entry was initialized exactly once. + if (assigned != 256) { + fprintf(stderr, "ComputeTable: assigned only %d of 256\n", assigned); + abort(); + } + for (int i = 0; i < 256; i++) { + if (dst[i] == 0xffff) { + fprintf(stderr, "ComputeTable: did not assign byte %d\n", i); + abort(); + } + } + + if (FLAGS_snappy_dump_decompression_table) { + printf("static const uint16 char_table[256] = {\n "); + for (int i = 0; i < 256; i++) { + printf("0x%04x%s", + dst[i], + ((i == 255) ? "\n" : (((i%8) == 7) ? ",\n " : ", "))); + } + printf("};\n"); + } + + // Check that computed table matched recorded table + for (int i = 0; i < 256; i++) { + if (dst[i] != char_table[i]) { + fprintf(stderr, "ComputeTable: byte %d: computed (%x), expect (%x)\n", + i, static_cast<int>(dst[i]), static_cast<int>(char_table[i])); + abort(); + } + } +} +#endif /* !NDEBUG */ + +// Helper class for decompression +class SnappyDecompressor { + private: + Source* reader_; // Underlying source of bytes to decompress + const char* ip_; // Points to next buffered byte + const char* ip_limit_; // Points just past buffered bytes + uint32 peeked_; // Bytes peeked from reader (need to skip) + bool eof_; // Hit end of input without an error? + char scratch_[5]; // Temporary buffer for PeekFast() boundaries + + // Ensure that all of the tag metadata for the next tag is available + // in [ip_..ip_limit_-1]. Also ensures that [ip,ip+4] is readable even + // if (ip_limit_ - ip_ < 5). + // + // Returns true on success, false on error or end of input. + bool RefillTag(); + + public: + explicit SnappyDecompressor(Source* reader) + : reader_(reader), + ip_(NULL), + ip_limit_(NULL), + peeked_(0), + eof_(false) { + } + + ~SnappyDecompressor() { + // Advance past any bytes we peeked at from the reader + reader_->Skip(peeked_); + } + + // Returns true iff we have hit the end of the input without an error. + bool eof() const { + return eof_; + } + + // Read the uncompressed length stored at the start of the compressed data. + // On succcess, stores the length in *result and returns true. + // On failure, returns false. + bool ReadUncompressedLength(uint32* result) { + assert(ip_ == NULL); // Must not have read anything yet + // Length is encoded in 1..5 bytes + *result = 0; + uint32 shift = 0; + while (true) { + if (shift >= 32) return false; + size_t n; + const char* ip = reader_->Peek(&n); + if (n == 0) return false; + const unsigned char c = *(reinterpret_cast<const unsigned char*>(ip)); + reader_->Skip(1); + *result |= static_cast<uint32>(c & 0x7f) << shift; + if (c < 128) { + break; + } + shift += 7; + } + return true; + } + + // Process the next item found in the input. + // Returns true if successful, false on error or end of input. + template <class Writer> + void DecompressAllTags(Writer* writer) { + const char* ip = ip_; + + // We could have put this refill fragment only at the beginning of the loop. + // However, duplicating it at the end of each branch gives the compiler more + // scope to optimize the <ip_limit_ - ip> expression based on the local + // context, which overall increases speed. + #define MAYBE_REFILL() \ + if (ip_limit_ - ip < 5) { \ + ip_ = ip; \ + if (!RefillTag()) return; \ + ip = ip_; \ + } + + MAYBE_REFILL(); + for ( ;; ) { + const unsigned char c = *(reinterpret_cast<const unsigned char*>(ip++)); + + if ((c & 0x3) == LITERAL) { + size_t literal_length = (c >> 2) + 1u; + if (writer->TryFastAppend(ip, ip_limit_ - ip, literal_length)) { + assert(literal_length < 61); + ip += literal_length; + MAYBE_REFILL(); + continue; + } + if (PREDICT_FALSE(literal_length >= 61)) { + // Long literal. + const size_t literal_length_length = literal_length - 60; + literal_length = + (LittleEndian::Load32(ip) & wordmask[literal_length_length]) + 1; + ip += literal_length_length; + } + + size_t avail = ip_limit_ - ip; + while (avail < literal_length) { + if (!writer->Append(ip, avail)) return; + literal_length -= avail; + reader_->Skip(peeked_); + size_t n; + ip = reader_->Peek(&n); + avail = n; + peeked_ = avail; + if (avail == 0) return; // Premature end of input + ip_limit_ = ip + avail; + } + if (!writer->Append(ip, literal_length)) { + return; + } + ip += literal_length; + MAYBE_REFILL(); + } else { + const uint32 entry = char_table[c]; + const uint32 trailer = LittleEndian::Load32(ip) & wordmask[entry >> 11]; + const uint32 length = entry & 0xff; + ip += entry >> 11; + + // copy_offset/256 is encoded in bits 8..10. By just fetching + // those bits, we get copy_offset (since the bit-field starts at + // bit 8). + const uint32 copy_offset = entry & 0x700; + if (!writer->AppendFromSelf(copy_offset + trailer, length)) { + return; + } + MAYBE_REFILL(); + } + } + +#undef MAYBE_REFILL + } +}; + +bool SnappyDecompressor::RefillTag() { + const char* ip = ip_; + if (ip == ip_limit_) { + // Fetch a new fragment from the reader + reader_->Skip(peeked_); // All peeked bytes are used up + size_t n; + ip = reader_->Peek(&n); + peeked_ = n; + if (n == 0) { + eof_ = true; + return false; + } + ip_limit_ = ip + n; + } + + // Read the tag character + assert(ip < ip_limit_); + const unsigned char c = *(reinterpret_cast<const unsigned char*>(ip)); + const uint32 entry = char_table[c]; + const uint32 needed = (entry >> 11) + 1; // +1 byte for 'c' + assert(needed <= sizeof(scratch_)); + + // Read more bytes from reader if needed + uint32 nbuf = ip_limit_ - ip; + if (nbuf < needed) { + // Stitch together bytes from ip and reader to form the word + // contents. We store the needed bytes in "scratch_". They + // will be consumed immediately by the caller since we do not + // read more than we need. + memmove(scratch_, ip, nbuf); + reader_->Skip(peeked_); // All peeked bytes are used up + peeked_ = 0; + while (nbuf < needed) { + size_t length; + const char* src = reader_->Peek(&length); + if (length == 0) return false; + uint32 to_add = min<uint32>(needed - nbuf, length); + memcpy(scratch_ + nbuf, src, to_add); + nbuf += to_add; + reader_->Skip(to_add); + } + assert(nbuf == needed); + ip_ = scratch_; + ip_limit_ = scratch_ + needed; + } else if (nbuf < 5) { + // Have enough bytes, but move into scratch_ so that we do not + // read past end of input + memmove(scratch_, ip, nbuf); + reader_->Skip(peeked_); // All peeked bytes are used up + peeked_ = 0; + ip_ = scratch_; + ip_limit_ = scratch_ + nbuf; + } else { + // Pass pointer to buffer returned by reader_. + ip_ = ip; + } + return true; +} + +template <typename Writer> +static bool InternalUncompress(Source* r, + Writer* writer, + uint32 max_len) { + // Read the uncompressed length from the front of the compressed input + SnappyDecompressor decompressor(r); + uint32 uncompressed_len = 0; + if (!decompressor.ReadUncompressedLength(&uncompressed_len)) return false; + return InternalUncompressAllTags( + &decompressor, writer, uncompressed_len, max_len); +} + +template <typename Writer> +static bool InternalUncompressAllTags(SnappyDecompressor* decompressor, + Writer* writer, + uint32 uncompressed_len, + uint32 max_len) { + // Protect against possible DoS attack + if (static_cast<uint64>(uncompressed_len) > max_len) { + return false; + } + + writer->SetExpectedLength(uncompressed_len); + + // Process the entire input + decompressor->DecompressAllTags(writer); + return (decompressor->eof() && writer->CheckLength()); +} + +bool GetUncompressedLength(Source* source, uint32* result) { + SnappyDecompressor decompressor(source); + return decompressor.ReadUncompressedLength(result); +} + +size_t Compress(Source* reader, Sink* writer) { + size_t written = 0; + size_t N = reader->Available(); + char ulength[Varint::kMax32]; + char* p = Varint::Encode32(ulength, N); + writer->Append(ulength, p-ulength); + written += (p - ulength); + + internal::WorkingMemory wmem; + char* scratch = NULL; + char* scratch_output = NULL; + + while (N > 0) { + // Get next block to compress (without copying if possible) + size_t fragment_size; + const char* fragment = reader->Peek(&fragment_size); + assert(fragment_size != 0); // premature end of input + const size_t num_to_read = min(N, kBlockSize); + size_t bytes_read = fragment_size; + + size_t pending_advance = 0; + if (bytes_read >= num_to_read) { + // Buffer returned by reader is large enough + pending_advance = num_to_read; + fragment_size = num_to_read; + } else { + // Read into scratch buffer + if (scratch == NULL) { + // If this is the last iteration, we want to allocate N bytes + // of space, otherwise the max possible kBlockSize space. + // num_to_read contains exactly the correct value + scratch = new char[num_to_read]; + } + memcpy(scratch, fragment, bytes_read); + reader->Skip(bytes_read); + + while (bytes_read < num_to_read) { + fragment = reader->Peek(&fragment_size); + size_t n = min<size_t>(fragment_size, num_to_read - bytes_read); + memcpy(scratch + bytes_read, fragment, n); + bytes_read += n; + reader->Skip(n); + } + assert(bytes_read == num_to_read); + fragment = scratch; + fragment_size = num_to_read; + } + assert(fragment_size == num_to_read); + + // Get encoding table for compression + int table_size; + uint16* table = wmem.GetHashTable(num_to_read, &table_size); + + // Compress input_fragment and append to dest + const int max_output = MaxCompressedLength(num_to_read); + + // Need a scratch buffer for the output, in case the byte sink doesn't + // have room for us directly. + if (scratch_output == NULL) { + scratch_output = new char[max_output]; + } else { + // Since we encode kBlockSize regions followed by a region + // which is <= kBlockSize in length, a previously allocated + // scratch_output[] region is big enough for this iteration. + } + char* dest = writer->GetAppendBuffer(max_output, scratch_output); + char* end = internal::CompressFragment(fragment, fragment_size, + dest, table, table_size); + writer->Append(dest, end - dest); + written += (end - dest); + + N -= num_to_read; + reader->Skip(pending_advance); + } + + delete[] scratch; + delete[] scratch_output; + + return written; +} + +// ----------------------------------------------------------------------- +// Flat array interfaces +// ----------------------------------------------------------------------- + +// A type that writes to a flat array. +// Note that this is not a "ByteSink", but a type that matches the +// Writer template argument to SnappyDecompressor::DecompressAllTags(). +class SnappyArrayWriter { + private: + char* base_; + char* op_; + char* op_limit_; + + public: + inline explicit SnappyArrayWriter(char* dst) + : base_(dst), + op_(dst) { + } + + inline void SetExpectedLength(size_t len) { + op_limit_ = op_ + len; + } + + inline bool CheckLength() const { + return op_ == op_limit_; + } + + inline bool Append(const char* ip, size_t len) { + char* op = op_; + const size_t space_left = op_limit_ - op; + if (space_left < len) { + return false; + } + memcpy(op, ip, len); + op_ = op + len; + return true; + } + + inline bool TryFastAppend(const char* ip, size_t available, size_t len) { + char* op = op_; + const size_t space_left = op_limit_ - op; + if (len <= 16 && available >= 16 && space_left >= 16) { + // Fast path, used for the majority (about 95%) of invocations. + UnalignedCopy64(ip, op); + UnalignedCopy64(ip + 8, op + 8); + op_ = op + len; + return true; + } else { + return false; + } + } + + inline bool AppendFromSelf(size_t offset, size_t len) { + char* op = op_; + const size_t space_left = op_limit_ - op; + + if (op - base_ <= offset - 1u) { // -1u catches offset==0 + return false; + } + if (len <= 16 && offset >= 8 && space_left >= 16) { + // Fast path, used for the majority (70-80%) of dynamic invocations. + UnalignedCopy64(op - offset, op); + UnalignedCopy64(op - offset + 8, op + 8); + } else { + if (space_left >= len + kMaxIncrementCopyOverflow) { + IncrementalCopyFastPath(op - offset, op, len); + } else { + if (space_left < len) { + return false; + } + IncrementalCopy(op - offset, op, len); + } + } + + op_ = op + len; + return true; + } +}; + +bool RawUncompress(const char* compressed, size_t n, char* uncompressed) { + ByteArraySource reader(compressed, n); + return RawUncompress(&reader, uncompressed); +} + +bool RawUncompress(Source* compressed, char* uncompressed) { + SnappyArrayWriter output(uncompressed); + return InternalUncompress(compressed, &output, kuint32max); +} + +bool Uncompress(const char* compressed, size_t n, string* uncompressed) { + size_t ulength; + if (!GetUncompressedLength(compressed, n, &ulength)) { + return false; + } + // Protect against possible DoS attack + if ((static_cast<uint64>(ulength) + uncompressed->size()) > + uncompressed->max_size()) { + return false; + } + STLStringResizeUninitialized(uncompressed, ulength); + return RawUncompress(compressed, n, string_as_array(uncompressed)); +} + + +// A Writer that drops everything on the floor and just does validation +class SnappyDecompressionValidator { + private: + size_t expected_; + size_t produced_; + + public: + inline SnappyDecompressionValidator() : produced_(0) { } + inline void SetExpectedLength(size_t len) { + expected_ = len; + } + inline bool CheckLength() const { + return expected_ == produced_; + } + inline bool Append(const char* ip, size_t len) { + produced_ += len; + return produced_ <= expected_; + } + inline bool TryFastAppend(const char* ip, size_t available, size_t length) { + return false; + } + inline bool AppendFromSelf(size_t offset, size_t len) { + if (produced_ <= offset - 1u) return false; // -1u catches offset==0 + produced_ += len; + return produced_ <= expected_; + } +}; + +bool IsValidCompressedBuffer(const char* compressed, size_t n) { + ByteArraySource reader(compressed, n); + SnappyDecompressionValidator writer; + return InternalUncompress(&reader, &writer, kuint32max); +} + +void RawCompress(const char* input, + size_t input_length, + char* compressed, + size_t* compressed_length) { + ByteArraySource reader(input, input_length); + UncheckedByteArraySink writer(compressed); + Compress(&reader, &writer); + + // Compute how many bytes were added + *compressed_length = (writer.CurrentDestination() - compressed); +} + +size_t Compress(const char* input, size_t input_length, string* compressed) { + // Pre-grow the buffer to the max length of the compressed output + compressed->resize(MaxCompressedLength(input_length)); + + size_t compressed_length; + RawCompress(input, input_length, string_as_array(compressed), + &compressed_length); + compressed->resize(compressed_length); + return compressed_length; +} + + +} // end namespace snappy + diff --git a/app/snappy/snappy.h b/app/snappy/snappy.h new file mode 100644 index 00000000..03ef6ce5 --- /dev/null +++ b/app/snappy/snappy.h @@ -0,0 +1,163 @@ +// Copyright 2005 and onwards Google Inc. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +// +// A light-weight compression algorithm. It is designed for speed of +// compression and decompression, rather than for the utmost in space +// savings. +// +// For getting better compression ratios when you are compressing data +// with long repeated sequences or compressing data that is similar to +// other data, while still compressing fast, you might look at first +// using BMDiff and then compressing the output of BMDiff with +// Snappy. + +#ifndef UTIL_SNAPPY_SNAPPY_H__ +#define UTIL_SNAPPY_SNAPPY_H__ + +#include <stddef.h> +#include <string> + +#include "snappy-stubs-public.h" + +namespace snappy { + class Source; + class Sink; + + // ------------------------------------------------------------------------ + // Generic compression/decompression routines. + // ------------------------------------------------------------------------ + + // Compress the bytes read from "*source" and append to "*sink". Return the + // number of bytes written. + size_t Compress(Source* source, Sink* sink); + + // Find the uncompressed length of the given stream, as given by the header. + // Note that the true length could deviate from this; the stream could e.g. + // be truncated. + // + // Also note that this leaves "*source" in a state that is unsuitable for + // further operations, such as RawUncompress(). You will need to rewind + // or recreate the source yourself before attempting any further calls. + bool GetUncompressedLength(Source* source, uint32* result); + + // ------------------------------------------------------------------------ + // Higher-level string based routines (should be sufficient for most users) + // ------------------------------------------------------------------------ + + // Sets "*output" to the compressed version of "input[0,input_length-1]". + // Original contents of *output are lost. + // + // REQUIRES: "input[]" is not an alias of "*output". + size_t Compress(const char* input, size_t input_length, string* output); + + // Decompresses "compressed[0,compressed_length-1]" to "*uncompressed". + // Original contents of "*uncompressed" are lost. + // + // REQUIRES: "compressed[]" is not an alias of "*uncompressed". + // + // returns false if the message is corrupted and could not be decompressed + bool Uncompress(const char* compressed, size_t compressed_length, + string* uncompressed); + + + // ------------------------------------------------------------------------ + // Lower-level character array based routines. May be useful for + // efficiency reasons in certain circumstances. + // ------------------------------------------------------------------------ + + // REQUIRES: "compressed" must point to an area of memory that is at + // least "MaxCompressedLength(input_length)" bytes in length. + // + // Takes the data stored in "input[0..input_length]" and stores + // it in the array pointed to by "compressed". + // + // "*compressed_length" is set to the length of the compressed output. + // + // Example: + // char* output = new char[snappy::MaxCompressedLength(input_length)]; + // size_t output_length; + // RawCompress(input, input_length, output, &output_length); + // ... Process(output, output_length) ... + // delete [] output; + void RawCompress(const char* input, + size_t input_length, + char* compressed, + size_t* compressed_length); + + // Given data in "compressed[0..compressed_length-1]" generated by + // calling the Snappy::Compress routine, this routine + // stores the uncompressed data to + // uncompressed[0..GetUncompressedLength(compressed)-1] + // returns false if the message is corrupted and could not be decrypted + bool RawUncompress(const char* compressed, size_t compressed_length, + char* uncompressed); + + // Given data from the byte source 'compressed' generated by calling + // the Snappy::Compress routine, this routine stores the uncompressed + // data to + // uncompressed[0..GetUncompressedLength(compressed,compressed_length)-1] + // returns false if the message is corrupted and could not be decrypted + bool RawUncompress(Source* compressed, char* uncompressed); + + // Returns the maximal size of the compressed representation of + // input data that is "source_bytes" bytes in length; + size_t MaxCompressedLength(size_t source_bytes); + + // REQUIRES: "compressed[]" was produced by RawCompress() or Compress() + // Returns true and stores the length of the uncompressed data in + // *result normally. Returns false on parsing error. + // This operation takes O(1) time. + bool GetUncompressedLength(const char* compressed, size_t compressed_length, + size_t* result); + + // Returns true iff the contents of "compressed[]" can be uncompressed + // successfully. Does not return the uncompressed data. Takes + // time proportional to compressed_length, but is usually at least + // a factor of four faster than actual decompression. + bool IsValidCompressedBuffer(const char* compressed, + size_t compressed_length); + + // The size of a compression block. Note that many parts of the compression + // code assumes that kBlockSize <= 65536; in particular, the hash table + // can only store 16-bit offsets, and EmitCopy() also assumes the offset + // is 65535 bytes or less. Note also that if you change this, it will + // affect the framing format (see framing_format.txt). + // + // Note that there might be older data around that is compressed with larger + // block sizes, so the decompression code should not rely on the + // non-existence of long backreferences. + static const int kBlockLog = 16; + static const size_t kBlockSize = 1 << kBlockLog; + + static const int kMaxHashTableBits = 14; + static const size_t kMaxHashTableSize = 1 << kMaxHashTableBits; + +} // end namespace snappy + + +#endif // UTIL_SNAPPY_SNAPPY_H__ diff --git a/app/snappy/snappy_unittest.cc b/app/snappy/snappy_unittest.cc new file mode 100644 index 00000000..59c108f4 --- /dev/null +++ b/app/snappy/snappy_unittest.cc @@ -0,0 +1,1164 @@ +// Copyright 2005 and onwards Google Inc. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include <math.h> +#include <stdlib.h> + + +#include <algorithm> +#include <string> +#include <vector> + +#include "snappy.h" +#include "snappy-internal.h" +#include "snappy-test.h" +#include "snappy-sinksource.h" + +DEFINE_int32(start_len, -1, + "Starting prefix size for testing (-1: just full file contents)"); +DEFINE_int32(end_len, -1, + "Starting prefix size for testing (-1: just full file contents)"); +DEFINE_int32(bytes, 10485760, + "How many bytes to compress/uncompress per file for timing"); + +DEFINE_bool(zlib, false, + "Run zlib compression (http://www.zlib.net)"); +DEFINE_bool(lzo, false, + "Run LZO compression (http://www.oberhumer.com/opensource/lzo/)"); +DEFINE_bool(quicklz, false, + "Run quickLZ compression (http://www.quicklz.com/)"); +DEFINE_bool(liblzf, false, + "Run libLZF compression " + "(http://www.goof.com/pcg/marc/liblzf.html)"); +DEFINE_bool(fastlz, false, + "Run FastLZ compression (http://www.fastlz.org/"); +DEFINE_bool(snappy, true, "Run snappy compression"); + + +DEFINE_bool(write_compressed, false, + "Write compressed versions of each file to <file>.comp"); +DEFINE_bool(write_uncompressed, false, + "Write uncompressed versions of each file to <file>.uncomp"); + +namespace snappy { + + +#ifdef HAVE_FUNC_MMAP + +// To test against code that reads beyond its input, this class copies a +// string to a newly allocated group of pages, the last of which +// is made unreadable via mprotect. Note that we need to allocate the +// memory with mmap(), as POSIX allows mprotect() only on memory allocated +// with mmap(), and some malloc/posix_memalign implementations expect to +// be able to read previously allocated memory while doing heap allocations. +class DataEndingAtUnreadablePage { + public: + explicit DataEndingAtUnreadablePage(const string& s) { + const size_t page_size = getpagesize(); + const size_t size = s.size(); + // Round up space for string to a multiple of page_size. + size_t space_for_string = (size + page_size - 1) & ~(page_size - 1); + alloc_size_ = space_for_string + page_size; + mem_ = mmap(NULL, alloc_size_, + PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + CHECK_NE(MAP_FAILED, mem_); + protected_page_ = reinterpret_cast<char*>(mem_) + space_for_string; + char* dst = protected_page_ - size; + memcpy(dst, s.data(), size); + data_ = dst; + size_ = size; + // Make guard page unreadable. + CHECK_EQ(0, mprotect(protected_page_, page_size, PROT_NONE)); + } + + ~DataEndingAtUnreadablePage() { + // Undo the mprotect. + CHECK_EQ(0, mprotect(protected_page_, getpagesize(), PROT_READ|PROT_WRITE)); + CHECK_EQ(0, munmap(mem_, alloc_size_)); + } + + const char* data() const { return data_; } + size_t size() const { return size_; } + + private: + size_t alloc_size_; + void* mem_; + char* protected_page_; + const char* data_; + size_t size_; +}; + +#else // HAVE_FUNC_MMAP + +// Fallback for systems without mmap. +typedef string DataEndingAtUnreadablePage; + +#endif + +enum CompressorType { + ZLIB, LZO, LIBLZF, QUICKLZ, FASTLZ, SNAPPY +}; + +const char* names[] = { + "ZLIB", "LZO", "LIBLZF", "QUICKLZ", "FASTLZ", "SNAPPY" +}; + +static size_t MinimumRequiredOutputSpace(size_t input_size, + CompressorType comp) { + switch (comp) { +#ifdef ZLIB_VERSION + case ZLIB: + return ZLib::MinCompressbufSize(input_size); +#endif // ZLIB_VERSION + +#ifdef LZO_VERSION + case LZO: + return input_size + input_size/64 + 16 + 3; +#endif // LZO_VERSION + +#ifdef LZF_VERSION + case LIBLZF: + return input_size; +#endif // LZF_VERSION + +#ifdef QLZ_VERSION_MAJOR + case QUICKLZ: + return input_size + 36000; // 36000 is used for scratch. +#endif // QLZ_VERSION_MAJOR + +#ifdef FASTLZ_VERSION + case FASTLZ: + return max(static_cast<int>(ceil(input_size * 1.05)), 66); +#endif // FASTLZ_VERSION + + case SNAPPY: + return snappy::MaxCompressedLength(input_size); + + default: + LOG(FATAL) << "Unknown compression type number " << comp; + } +} + +// Returns true if we successfully compressed, false otherwise. +// +// If compressed_is_preallocated is set, do not resize the compressed buffer. +// This is typically what you want for a benchmark, in order to not spend +// time in the memory allocator. If you do set this flag, however, +// "compressed" must be preinitialized to at least MinCompressbufSize(comp) +// number of bytes, and may contain junk bytes at the end after return. +static bool Compress(const char* input, size_t input_size, CompressorType comp, + string* compressed, bool compressed_is_preallocated) { + if (!compressed_is_preallocated) { + compressed->resize(MinimumRequiredOutputSpace(input_size, comp)); + } + + switch (comp) { +#ifdef ZLIB_VERSION + case ZLIB: { + ZLib zlib; + uLongf destlen = compressed->size(); + int ret = zlib.Compress( + reinterpret_cast<Bytef*>(string_as_array(compressed)), + &destlen, + reinterpret_cast<const Bytef*>(input), + input_size); + CHECK_EQ(Z_OK, ret); + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + return true; + } +#endif // ZLIB_VERSION + +#ifdef LZO_VERSION + case LZO: { + unsigned char* mem = new unsigned char[LZO1X_1_15_MEM_COMPRESS]; + lzo_uint destlen; + int ret = lzo1x_1_15_compress( + reinterpret_cast<const uint8*>(input), + input_size, + reinterpret_cast<uint8*>(string_as_array(compressed)), + &destlen, + mem); + CHECK_EQ(LZO_E_OK, ret); + delete[] mem; + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + break; + } +#endif // LZO_VERSION + +#ifdef LZF_VERSION + case LIBLZF: { + int destlen = lzf_compress(input, + input_size, + string_as_array(compressed), + input_size); + if (destlen == 0) { + // lzf *can* cause lots of blowup when compressing, so they + // recommend to limit outsize to insize, and just not compress + // if it's bigger. Ideally, we'd just swap input and output. + compressed->assign(input, input_size); + destlen = input_size; + } + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + break; + } +#endif // LZF_VERSION + +#ifdef QLZ_VERSION_MAJOR + case QUICKLZ: { + qlz_state_compress *state_compress = new qlz_state_compress; + int destlen = qlz_compress(input, + string_as_array(compressed), + input_size, + state_compress); + delete state_compress; + CHECK_NE(0, destlen); + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + break; + } +#endif // QLZ_VERSION_MAJOR + +#ifdef FASTLZ_VERSION + case FASTLZ: { + // Use level 1 compression since we mostly care about speed. + int destlen = fastlz_compress_level( + 1, + input, + input_size, + string_as_array(compressed)); + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + CHECK_NE(destlen, 0); + break; + } +#endif // FASTLZ_VERSION + + case SNAPPY: { + size_t destlen; + snappy::RawCompress(input, input_size, + string_as_array(compressed), + &destlen); + CHECK_LE(destlen, snappy::MaxCompressedLength(input_size)); + if (!compressed_is_preallocated) { + compressed->resize(destlen); + } + break; + } + + + default: { + return false; // the asked-for library wasn't compiled in + } + } + return true; +} + +static bool Uncompress(const string& compressed, CompressorType comp, + int size, string* output) { + switch (comp) { +#ifdef ZLIB_VERSION + case ZLIB: { + output->resize(size); + ZLib zlib; + uLongf destlen = output->size(); + int ret = zlib.Uncompress( + reinterpret_cast<Bytef*>(string_as_array(output)), + &destlen, + reinterpret_cast<const Bytef*>(compressed.data()), + compressed.size()); + CHECK_EQ(Z_OK, ret); + CHECK_EQ(static_cast<uLongf>(size), destlen); + break; + } +#endif // ZLIB_VERSION + +#ifdef LZO_VERSION + case LZO: { + output->resize(size); + lzo_uint destlen; + int ret = lzo1x_decompress( + reinterpret_cast<const uint8*>(compressed.data()), + compressed.size(), + reinterpret_cast<uint8*>(string_as_array(output)), + &destlen, + NULL); + CHECK_EQ(LZO_E_OK, ret); + CHECK_EQ(static_cast<lzo_uint>(size), destlen); + break; + } +#endif // LZO_VERSION + +#ifdef LZF_VERSION + case LIBLZF: { + output->resize(size); + int destlen = lzf_decompress(compressed.data(), + compressed.size(), + string_as_array(output), + output->size()); + if (destlen == 0) { + // This error probably means we had decided not to compress, + // and thus have stored input in output directly. + output->assign(compressed.data(), compressed.size()); + destlen = compressed.size(); + } + CHECK_EQ(destlen, size); + break; + } +#endif // LZF_VERSION + +#ifdef QLZ_VERSION_MAJOR + case QUICKLZ: { + output->resize(size); + qlz_state_decompress *state_decompress = new qlz_state_decompress; + int destlen = qlz_decompress(compressed.data(), + string_as_array(output), + state_decompress); + delete state_decompress; + CHECK_EQ(destlen, size); + break; + } +#endif // QLZ_VERSION_MAJOR + +#ifdef FASTLZ_VERSION + case FASTLZ: { + output->resize(size); + int destlen = fastlz_decompress(compressed.data(), + compressed.length(), + string_as_array(output), + size); + CHECK_EQ(destlen, size); + break; + } +#endif // FASTLZ_VERSION + + case SNAPPY: { + snappy::RawUncompress(compressed.data(), compressed.size(), + string_as_array(output)); + break; + } + + + default: { + return false; // the asked-for library wasn't compiled in + } + } + return true; +} + +static void Measure(const char* data, + size_t length, + CompressorType comp, + int repeats, + int block_size) { + // Run tests a few time and pick median running times + static const int kRuns = 5; + double ctime[kRuns]; + double utime[kRuns]; + int compressed_size = 0; + + { + // Chop the input into blocks + int num_blocks = (length + block_size - 1) / block_size; + vector<const char*> input(num_blocks); + vector<size_t> input_length(num_blocks); + vector<string> compressed(num_blocks); + vector<string> output(num_blocks); + for (int b = 0; b < num_blocks; b++) { + int input_start = b * block_size; + int input_limit = min<int>((b+1)*block_size, length); + input[b] = data+input_start; + input_length[b] = input_limit-input_start; + + // Pre-grow the output buffer so we don't measure string append time. + compressed[b].resize(MinimumRequiredOutputSpace(block_size, comp)); + } + + // First, try one trial compression to make sure the code is compiled in + if (!Compress(input[0], input_length[0], comp, &compressed[0], true)) { + LOG(WARNING) << "Skipping " << names[comp] << ": " + << "library not compiled in"; + return; + } + + for (int run = 0; run < kRuns; run++) { + CycleTimer ctimer, utimer; + + for (int b = 0; b < num_blocks; b++) { + // Pre-grow the output buffer so we don't measure string append time. + compressed[b].resize(MinimumRequiredOutputSpace(block_size, comp)); + } + + ctimer.Start(); + for (int b = 0; b < num_blocks; b++) + for (int i = 0; i < repeats; i++) + Compress(input[b], input_length[b], comp, &compressed[b], true); + ctimer.Stop(); + + // Compress once more, with resizing, so we don't leave junk + // at the end that will confuse the decompressor. + for (int b = 0; b < num_blocks; b++) { + Compress(input[b], input_length[b], comp, &compressed[b], false); + } + + for (int b = 0; b < num_blocks; b++) { + output[b].resize(input_length[b]); + } + + utimer.Start(); + for (int i = 0; i < repeats; i++) + for (int b = 0; b < num_blocks; b++) + Uncompress(compressed[b], comp, input_length[b], &output[b]); + utimer.Stop(); + + ctime[run] = ctimer.Get(); + utime[run] = utimer.Get(); + } + + compressed_size = 0; + for (int i = 0; i < compressed.size(); i++) { + compressed_size += compressed[i].size(); + } + } + + sort(ctime, ctime + kRuns); + sort(utime, utime + kRuns); + const int med = kRuns/2; + + float comp_rate = (length / ctime[med]) * repeats / 1048576.0; + float uncomp_rate = (length / utime[med]) * repeats / 1048576.0; + string x = names[comp]; + x += ":"; + string urate = (uncomp_rate >= 0) + ? StringPrintf("%.1f", uncomp_rate) + : string("?"); + printf("%-7s [b %dM] bytes %6d -> %6d %4.1f%% " + "comp %5.1f MB/s uncomp %5s MB/s\n", + x.c_str(), + block_size/(1<<20), + static_cast<int>(length), static_cast<uint32>(compressed_size), + (compressed_size * 100.0) / max<int>(1, length), + comp_rate, + urate.c_str()); +} + + +static int VerifyString(const string& input) { + string compressed; + DataEndingAtUnreadablePage i(input); + const size_t written = snappy::Compress(i.data(), i.size(), &compressed); + CHECK_EQ(written, compressed.size()); + CHECK_LE(compressed.size(), + snappy::MaxCompressedLength(input.size())); + CHECK(snappy::IsValidCompressedBuffer(compressed.data(), compressed.size())); + + string uncompressed; + DataEndingAtUnreadablePage c(compressed); + CHECK(snappy::Uncompress(c.data(), c.size(), &uncompressed)); + CHECK_EQ(uncompressed, input); + return uncompressed.size(); +} + + +// Test that data compressed by a compressor that does not +// obey block sizes is uncompressed properly. +static void VerifyNonBlockedCompression(const string& input) { + if (input.length() > snappy::kBlockSize) { + // We cannot test larger blocks than the maximum block size, obviously. + return; + } + + string prefix; + Varint::Append32(&prefix, input.size()); + + // Setup compression table + snappy::internal::WorkingMemory wmem; + int table_size; + uint16* table = wmem.GetHashTable(input.size(), &table_size); + + // Compress entire input in one shot + string compressed; + compressed += prefix; + compressed.resize(prefix.size()+snappy::MaxCompressedLength(input.size())); + char* dest = string_as_array(&compressed) + prefix.size(); + char* end = snappy::internal::CompressFragment(input.data(), input.size(), + dest, table, table_size); + compressed.resize(end - compressed.data()); + + // Uncompress into string + string uncomp_str; + CHECK(snappy::Uncompress(compressed.data(), compressed.size(), &uncomp_str)); + CHECK_EQ(uncomp_str, input); + +} + +// Expand the input so that it is at least K times as big as block size +static string Expand(const string& input) { + static const int K = 3; + string data = input; + while (data.size() < K * snappy::kBlockSize) { + data += input; + } + return data; +} + +static int Verify(const string& input) { + VLOG(1) << "Verifying input of size " << input.size(); + + // Compress using string based routines + const int result = VerifyString(input); + + + VerifyNonBlockedCompression(input); + if (!input.empty()) { + VerifyNonBlockedCompression(Expand(input)); + } + + + return result; +} + +// This test checks to ensure that snappy doesn't coredump if it gets +// corrupted data. + +static bool IsValidCompressedBuffer(const string& c) { + return snappy::IsValidCompressedBuffer(c.data(), c.size()); +} +static bool Uncompress(const string& c, string* u) { + return snappy::Uncompress(c.data(), c.size(), u); +} + +TYPED_TEST(CorruptedTest, VerifyCorrupted) { + string source = "making sure we don't crash with corrupted input"; + VLOG(1) << source; + string dest; + TypeParam uncmp; + snappy::Compress(source.data(), source.size(), &dest); + + // Mess around with the data. It's hard to simulate all possible + // corruptions; this is just one example ... + CHECK_GT(dest.size(), 3); + dest[1]--; + dest[3]++; + // this really ought to fail. + CHECK(!IsValidCompressedBuffer(TypeParam(dest))); + CHECK(!Uncompress(TypeParam(dest), &uncmp)); + + // This is testing for a security bug - a buffer that decompresses to 100k + // but we lie in the snappy header and only reserve 0 bytes of memory :) + source.resize(100000); + for (int i = 0; i < source.length(); ++i) { + source[i] = 'A'; + } + snappy::Compress(source.data(), source.size(), &dest); + dest[0] = dest[1] = dest[2] = dest[3] = 0; + CHECK(!IsValidCompressedBuffer(TypeParam(dest))); + CHECK(!Uncompress(TypeParam(dest), &uncmp)); + + if (sizeof(void *) == 4) { + // Another security check; check a crazy big length can't DoS us with an + // over-allocation. + // Currently this is done only for 32-bit builds. On 64-bit builds, + // where 3 GB might be an acceptable allocation size, Uncompress() + // attempts to decompress, and sometimes causes the test to run out of + // memory. + dest[0] = dest[1] = dest[2] = dest[3] = 0xff; + // This decodes to a really large size, i.e., about 3 GB. + dest[4] = 'k'; + CHECK(!IsValidCompressedBuffer(TypeParam(dest))); + CHECK(!Uncompress(TypeParam(dest), &uncmp)); + } else { + LOG(WARNING) << "Crazy decompression lengths not checked on 64-bit build"; + } + + // This decodes to about 2 MB; much smaller, but should still fail. + dest[0] = dest[1] = dest[2] = 0xff; + dest[3] = 0x00; + CHECK(!IsValidCompressedBuffer(TypeParam(dest))); + CHECK(!Uncompress(TypeParam(dest), &uncmp)); + + // try reading stuff in from a bad file. + for (int i = 1; i <= 3; ++i) { + string data = ReadTestDataFile(StringPrintf("baddata%d.snappy", i).c_str(), + 0); + string uncmp; + // check that we don't return a crazy length + size_t ulen; + CHECK(!snappy::GetUncompressedLength(data.data(), data.size(), &ulen) + || (ulen < (1<<20))); + uint32 ulen2; + snappy::ByteArraySource source(data.data(), data.size()); + CHECK(!snappy::GetUncompressedLength(&source, &ulen2) || + (ulen2 < (1<<20))); + CHECK(!IsValidCompressedBuffer(TypeParam(data))); + CHECK(!Uncompress(TypeParam(data), &uncmp)); + } +} + +// Helper routines to construct arbitrary compressed strings. +// These mirror the compression code in snappy.cc, but are copied +// here so that we can bypass some limitations in the how snappy.cc +// invokes these routines. +static void AppendLiteral(string* dst, const string& literal) { + if (literal.empty()) return; + int n = literal.size() - 1; + if (n < 60) { + // Fit length in tag byte + dst->push_back(0 | (n << 2)); + } else { + // Encode in upcoming bytes + char number[4]; + int count = 0; + while (n > 0) { + number[count++] = n & 0xff; + n >>= 8; + } + dst->push_back(0 | ((59+count) << 2)); + *dst += string(number, count); + } + *dst += literal; +} + +static void AppendCopy(string* dst, int offset, int length) { + while (length > 0) { + // Figure out how much to copy in one shot + int to_copy; + if (length >= 68) { + to_copy = 64; + } else if (length > 64) { + to_copy = 60; + } else { + to_copy = length; + } + length -= to_copy; + + if ((to_copy < 12) && (offset < 2048)) { + assert(to_copy-4 < 8); // Must fit in 3 bits + dst->push_back(1 | ((to_copy-4) << 2) | ((offset >> 8) << 5)); + dst->push_back(offset & 0xff); + } else if (offset < 65536) { + dst->push_back(2 | ((to_copy-1) << 2)); + dst->push_back(offset & 0xff); + dst->push_back(offset >> 8); + } else { + dst->push_back(3 | ((to_copy-1) << 2)); + dst->push_back(offset & 0xff); + dst->push_back((offset >> 8) & 0xff); + dst->push_back((offset >> 16) & 0xff); + dst->push_back((offset >> 24) & 0xff); + } + } +} + +TEST(Snappy, SimpleTests) { + Verify(""); + Verify("a"); + Verify("ab"); + Verify("abc"); + + Verify("aaaaaaa" + string(16, 'b') + string("aaaaa") + "abc"); + Verify("aaaaaaa" + string(256, 'b') + string("aaaaa") + "abc"); + Verify("aaaaaaa" + string(2047, 'b') + string("aaaaa") + "abc"); + Verify("aaaaaaa" + string(65536, 'b') + string("aaaaa") + "abc"); + Verify("abcaaaaaaa" + string(65536, 'b') + string("aaaaa") + "abc"); +} + +// Verify max blowup (lots of four-byte copies) +TEST(Snappy, MaxBlowup) { + string input; + for (int i = 0; i < 20000; i++) { + ACMRandom rnd(i); + uint32 bytes = static_cast<uint32>(rnd.Next()); + input.append(reinterpret_cast<char*>(&bytes), sizeof(bytes)); + } + for (int i = 19999; i >= 0; i--) { + ACMRandom rnd(i); + uint32 bytes = static_cast<uint32>(rnd.Next()); + input.append(reinterpret_cast<char*>(&bytes), sizeof(bytes)); + } + Verify(input); +} + +TEST(Snappy, RandomData) { + ACMRandom rnd(FLAGS_test_random_seed); + + const int num_ops = 20000; + for (int i = 0; i < num_ops; i++) { + if ((i % 1000) == 0) { + VLOG(0) << "Random op " << i << " of " << num_ops; + } + + string x; + int len = rnd.Uniform(4096); + if (i < 100) { + len = 65536 + rnd.Uniform(65536); + } + while (x.size() < len) { + int run_len = 1; + if (rnd.OneIn(10)) { + run_len = rnd.Skewed(8); + } + char c = (i < 100) ? rnd.Uniform(256) : rnd.Skewed(3); + while (run_len-- > 0 && x.size() < len) { + x += c; + } + } + + Verify(x); + } +} + +TEST(Snappy, FourByteOffset) { + // The new compressor cannot generate four-byte offsets since + // it chops up the input into 32KB pieces. So we hand-emit the + // copy manually. + + // The two fragments that make up the input string. + string fragment1 = "012345689abcdefghijklmnopqrstuvwxyz"; + string fragment2 = "some other string"; + + // How many times each fragment is emitted. + const int n1 = 2; + const int n2 = 100000 / fragment2.size(); + const int length = n1 * fragment1.size() + n2 * fragment2.size(); + + string compressed; + Varint::Append32(&compressed, length); + + AppendLiteral(&compressed, fragment1); + string src = fragment1; + for (int i = 0; i < n2; i++) { + AppendLiteral(&compressed, fragment2); + src += fragment2; + } + AppendCopy(&compressed, src.size(), fragment1.size()); + src += fragment1; + CHECK_EQ(length, src.size()); + + string uncompressed; + CHECK(snappy::IsValidCompressedBuffer(compressed.data(), compressed.size())); + CHECK(snappy::Uncompress(compressed.data(), compressed.size(), + &uncompressed)); + CHECK_EQ(uncompressed, src); +} + + +static bool CheckUncompressedLength(const string& compressed, + size_t* ulength) { + const bool result1 = snappy::GetUncompressedLength(compressed.data(), + compressed.size(), + ulength); + + snappy::ByteArraySource source(compressed.data(), compressed.size()); + uint32 length; + const bool result2 = snappy::GetUncompressedLength(&source, &length); + CHECK_EQ(result1, result2); + return result1; +} + +TEST(SnappyCorruption, TruncatedVarint) { + string compressed, uncompressed; + size_t ulength; + compressed.push_back('\xf0'); + CHECK(!CheckUncompressedLength(compressed, &ulength)); + CHECK(!snappy::IsValidCompressedBuffer(compressed.data(), compressed.size())); + CHECK(!snappy::Uncompress(compressed.data(), compressed.size(), + &uncompressed)); +} + +TEST(SnappyCorruption, UnterminatedVarint) { + string compressed, uncompressed; + size_t ulength; + compressed.push_back(128); + compressed.push_back(128); + compressed.push_back(128); + compressed.push_back(128); + compressed.push_back(128); + compressed.push_back(10); + CHECK(!CheckUncompressedLength(compressed, &ulength)); + CHECK(!snappy::IsValidCompressedBuffer(compressed.data(), compressed.size())); + CHECK(!snappy::Uncompress(compressed.data(), compressed.size(), + &uncompressed)); +} + +TEST(Snappy, ReadPastEndOfBuffer) { + // Check that we do not read past end of input + + // Make a compressed string that ends with a single-byte literal + string compressed; + Varint::Append32(&compressed, 1); + AppendLiteral(&compressed, "x"); + + string uncompressed; + DataEndingAtUnreadablePage c(compressed); + CHECK(snappy::Uncompress(c.data(), c.size(), &uncompressed)); + CHECK_EQ(uncompressed, string("x")); +} + +// Check for an infinite loop caused by a copy with offset==0 +TEST(Snappy, ZeroOffsetCopy) { + const char* compressed = "\x40\x12\x00\x00"; + // \x40 Length (must be > kMaxIncrementCopyOverflow) + // \x12\x00\x00 Copy with offset==0, length==5 + char uncompressed[100]; + EXPECT_FALSE(snappy::RawUncompress(compressed, 4, uncompressed)); +} + +TEST(Snappy, ZeroOffsetCopyValidation) { + const char* compressed = "\x05\x12\x00\x00"; + // \x05 Length + // \x12\x00\x00 Copy with offset==0, length==5 + EXPECT_FALSE(snappy::IsValidCompressedBuffer(compressed, 4)); +} + + +namespace { + +int TestFindMatchLength(const char* s1, const char *s2, unsigned length) { + return snappy::internal::FindMatchLength(s1, s2, s2 + length); +} + +} // namespace + +TEST(Snappy, FindMatchLength) { + // Exercise all different code paths through the function. + // 64-bit version: + + // Hit s1_limit in 64-bit loop, hit s1_limit in single-character loop. + EXPECT_EQ(6, TestFindMatchLength("012345", "012345", 6)); + EXPECT_EQ(11, TestFindMatchLength("01234567abc", "01234567abc", 11)); + + // Hit s1_limit in 64-bit loop, find a non-match in single-character loop. + EXPECT_EQ(9, TestFindMatchLength("01234567abc", "01234567axc", 9)); + + // Same, but edge cases. + EXPECT_EQ(11, TestFindMatchLength("01234567abc!", "01234567abc!", 11)); + EXPECT_EQ(11, TestFindMatchLength("01234567abc!", "01234567abc?", 11)); + + // Find non-match at once in first loop. + EXPECT_EQ(0, TestFindMatchLength("01234567xxxxxxxx", "?1234567xxxxxxxx", 16)); + EXPECT_EQ(1, TestFindMatchLength("01234567xxxxxxxx", "0?234567xxxxxxxx", 16)); + EXPECT_EQ(4, TestFindMatchLength("01234567xxxxxxxx", "01237654xxxxxxxx", 16)); + EXPECT_EQ(7, TestFindMatchLength("01234567xxxxxxxx", "0123456?xxxxxxxx", 16)); + + // Find non-match in first loop after one block. + EXPECT_EQ(8, TestFindMatchLength("abcdefgh01234567xxxxxxxx", + "abcdefgh?1234567xxxxxxxx", 24)); + EXPECT_EQ(9, TestFindMatchLength("abcdefgh01234567xxxxxxxx", + "abcdefgh0?234567xxxxxxxx", 24)); + EXPECT_EQ(12, TestFindMatchLength("abcdefgh01234567xxxxxxxx", + "abcdefgh01237654xxxxxxxx", 24)); + EXPECT_EQ(15, TestFindMatchLength("abcdefgh01234567xxxxxxxx", + "abcdefgh0123456?xxxxxxxx", 24)); + + // 32-bit version: + + // Short matches. + EXPECT_EQ(0, TestFindMatchLength("01234567", "?1234567", 8)); + EXPECT_EQ(1, TestFindMatchLength("01234567", "0?234567", 8)); + EXPECT_EQ(2, TestFindMatchLength("01234567", "01?34567", 8)); + EXPECT_EQ(3, TestFindMatchLength("01234567", "012?4567", 8)); + EXPECT_EQ(4, TestFindMatchLength("01234567", "0123?567", 8)); + EXPECT_EQ(5, TestFindMatchLength("01234567", "01234?67", 8)); + EXPECT_EQ(6, TestFindMatchLength("01234567", "012345?7", 8)); + EXPECT_EQ(7, TestFindMatchLength("01234567", "0123456?", 8)); + EXPECT_EQ(7, TestFindMatchLength("01234567", "0123456?", 7)); + EXPECT_EQ(7, TestFindMatchLength("01234567!", "0123456??", 7)); + + // Hit s1_limit in 32-bit loop, hit s1_limit in single-character loop. + EXPECT_EQ(10, TestFindMatchLength("xxxxxxabcd", "xxxxxxabcd", 10)); + EXPECT_EQ(10, TestFindMatchLength("xxxxxxabcd?", "xxxxxxabcd?", 10)); + EXPECT_EQ(13, TestFindMatchLength("xxxxxxabcdef", "xxxxxxabcdef", 13)); + + // Same, but edge cases. + EXPECT_EQ(12, TestFindMatchLength("xxxxxx0123abc!", "xxxxxx0123abc!", 12)); + EXPECT_EQ(12, TestFindMatchLength("xxxxxx0123abc!", "xxxxxx0123abc?", 12)); + + // Hit s1_limit in 32-bit loop, find a non-match in single-character loop. + EXPECT_EQ(11, TestFindMatchLength("xxxxxx0123abc", "xxxxxx0123axc", 13)); + + // Find non-match at once in first loop. + EXPECT_EQ(6, TestFindMatchLength("xxxxxx0123xxxxxxxx", + "xxxxxx?123xxxxxxxx", 18)); + EXPECT_EQ(7, TestFindMatchLength("xxxxxx0123xxxxxxxx", + "xxxxxx0?23xxxxxxxx", 18)); + EXPECT_EQ(8, TestFindMatchLength("xxxxxx0123xxxxxxxx", + "xxxxxx0132xxxxxxxx", 18)); + EXPECT_EQ(9, TestFindMatchLength("xxxxxx0123xxxxxxxx", + "xxxxxx012?xxxxxxxx", 18)); + + // Same, but edge cases. + EXPECT_EQ(6, TestFindMatchLength("xxxxxx0123", "xxxxxx?123", 10)); + EXPECT_EQ(7, TestFindMatchLength("xxxxxx0123", "xxxxxx0?23", 10)); + EXPECT_EQ(8, TestFindMatchLength("xxxxxx0123", "xxxxxx0132", 10)); + EXPECT_EQ(9, TestFindMatchLength("xxxxxx0123", "xxxxxx012?", 10)); + + // Find non-match in first loop after one block. + EXPECT_EQ(10, TestFindMatchLength("xxxxxxabcd0123xx", + "xxxxxxabcd?123xx", 16)); + EXPECT_EQ(11, TestFindMatchLength("xxxxxxabcd0123xx", + "xxxxxxabcd0?23xx", 16)); + EXPECT_EQ(12, TestFindMatchLength("xxxxxxabcd0123xx", + "xxxxxxabcd0132xx", 16)); + EXPECT_EQ(13, TestFindMatchLength("xxxxxxabcd0123xx", + "xxxxxxabcd012?xx", 16)); + + // Same, but edge cases. + EXPECT_EQ(10, TestFindMatchLength("xxxxxxabcd0123", "xxxxxxabcd?123", 14)); + EXPECT_EQ(11, TestFindMatchLength("xxxxxxabcd0123", "xxxxxxabcd0?23", 14)); + EXPECT_EQ(12, TestFindMatchLength("xxxxxxabcd0123", "xxxxxxabcd0132", 14)); + EXPECT_EQ(13, TestFindMatchLength("xxxxxxabcd0123", "xxxxxxabcd012?", 14)); +} + +TEST(Snappy, FindMatchLengthRandom) { + const int kNumTrials = 10000; + const int kTypicalLength = 10; + ACMRandom rnd(FLAGS_test_random_seed); + + for (int i = 0; i < kNumTrials; i++) { + string s, t; + char a = rnd.Rand8(); + char b = rnd.Rand8(); + while (!rnd.OneIn(kTypicalLength)) { + s.push_back(rnd.OneIn(2) ? a : b); + t.push_back(rnd.OneIn(2) ? a : b); + } + DataEndingAtUnreadablePage u(s); + DataEndingAtUnreadablePage v(t); + int matched = snappy::internal::FindMatchLength( + u.data(), v.data(), v.data() + t.size()); + if (matched == t.size()) { + EXPECT_EQ(s, t); + } else { + EXPECT_NE(s[matched], t[matched]); + for (int j = 0; j < matched; j++) { + EXPECT_EQ(s[j], t[j]); + } + } + } +} + + +static void CompressFile(const char* fname) { + string fullinput; + file::GetContents(fname, &fullinput, file::Defaults()).CheckSuccess(); + + string compressed; + Compress(fullinput.data(), fullinput.size(), SNAPPY, &compressed, false); + + file::SetContents(string(fname).append(".comp"), compressed, file::Defaults()) + .CheckSuccess(); +} + +static void UncompressFile(const char* fname) { + string fullinput; + file::GetContents(fname, &fullinput, file::Defaults()).CheckSuccess(); + + size_t uncompLength; + CHECK(CheckUncompressedLength(fullinput, &uncompLength)); + + string uncompressed; + uncompressed.resize(uncompLength); + CHECK(snappy::Uncompress(fullinput.data(), fullinput.size(), &uncompressed)); + + file::SetContents(string(fname).append(".uncomp"), uncompressed, + file::Defaults()).CheckSuccess(); +} + +static void MeasureFile(const char* fname) { + string fullinput; + file::GetContents(fname, &fullinput, file::Defaults()).CheckSuccess(); + printf("%-40s :\n", fname); + + int start_len = (FLAGS_start_len < 0) ? fullinput.size() : FLAGS_start_len; + int end_len = fullinput.size(); + if (FLAGS_end_len >= 0) { + end_len = min<int>(fullinput.size(), FLAGS_end_len); + } + for (int len = start_len; len <= end_len; len++) { + const char* const input = fullinput.data(); + int repeats = (FLAGS_bytes + len) / (len + 1); + if (FLAGS_zlib) Measure(input, len, ZLIB, repeats, 1024<<10); + if (FLAGS_lzo) Measure(input, len, LZO, repeats, 1024<<10); + if (FLAGS_liblzf) Measure(input, len, LIBLZF, repeats, 1024<<10); + if (FLAGS_quicklz) Measure(input, len, QUICKLZ, repeats, 1024<<10); + if (FLAGS_fastlz) Measure(input, len, FASTLZ, repeats, 1024<<10); + if (FLAGS_snappy) Measure(input, len, SNAPPY, repeats, 4096<<10); + + // For block-size based measurements + if (0 && FLAGS_snappy) { + Measure(input, len, SNAPPY, repeats, 8<<10); + Measure(input, len, SNAPPY, repeats, 16<<10); + Measure(input, len, SNAPPY, repeats, 32<<10); + Measure(input, len, SNAPPY, repeats, 64<<10); + Measure(input, len, SNAPPY, repeats, 256<<10); + Measure(input, len, SNAPPY, repeats, 1024<<10); + } + } +} + +static struct { + const char* label; + const char* filename; + size_t size_limit; +} files[] = { + { "html", "html", 0 }, + { "urls", "urls.10K", 0 }, + { "jpg", "house.jpg", 0 }, + { "jpg_200", "house.jpg", 200 }, + { "pdf", "mapreduce-osdi-1.pdf", 0 }, + { "html4", "html_x_4", 0 }, + { "cp", "cp.html", 0 }, + { "c", "fields.c", 0 }, + { "lsp", "grammar.lsp", 0 }, + { "xls", "kennedy.xls", 0 }, + { "xls_200", "kennedy.xls", 200 }, + { "txt1", "alice29.txt", 0 }, + { "txt2", "asyoulik.txt", 0 }, + { "txt3", "lcet10.txt", 0 }, + { "txt4", "plrabn12.txt", 0 }, + { "bin", "ptt5", 0 }, + { "bin_200", "ptt5", 200 }, + { "sum", "sum", 0 }, + { "man", "xargs.1", 0 }, + { "pb", "geo.protodata", 0 }, + { "gaviota", "kppkn.gtb", 0 }, +}; + +static void BM_UFlat(int iters, int arg) { + StopBenchmarkTiming(); + + // Pick file to process based on "arg" + CHECK_GE(arg, 0); + CHECK_LT(arg, ARRAYSIZE(files)); + string contents = ReadTestDataFile(files[arg].filename, + files[arg].size_limit); + + string zcontents; + snappy::Compress(contents.data(), contents.size(), &zcontents); + char* dst = new char[contents.size()]; + + SetBenchmarkBytesProcessed(static_cast<int64>(iters) * + static_cast<int64>(contents.size())); + SetBenchmarkLabel(files[arg].label); + StartBenchmarkTiming(); + while (iters-- > 0) { + CHECK(snappy::RawUncompress(zcontents.data(), zcontents.size(), dst)); + } + StopBenchmarkTiming(); + + delete[] dst; +} +BENCHMARK(BM_UFlat)->DenseRange(0, ARRAYSIZE(files) - 1); + +static void BM_UValidate(int iters, int arg) { + StopBenchmarkTiming(); + + // Pick file to process based on "arg" + CHECK_GE(arg, 0); + CHECK_LT(arg, ARRAYSIZE(files)); + string contents = ReadTestDataFile(files[arg].filename, + files[arg].size_limit); + + string zcontents; + snappy::Compress(contents.data(), contents.size(), &zcontents); + + SetBenchmarkBytesProcessed(static_cast<int64>(iters) * + static_cast<int64>(contents.size())); + SetBenchmarkLabel(files[arg].label); + StartBenchmarkTiming(); + while (iters-- > 0) { + CHECK(snappy::IsValidCompressedBuffer(zcontents.data(), zcontents.size())); + } + StopBenchmarkTiming(); +} +BENCHMARK(BM_UValidate)->DenseRange(0, 4); + + +static void BM_ZFlat(int iters, int arg) { + StopBenchmarkTiming(); + + // Pick file to process based on "arg" + CHECK_GE(arg, 0); + CHECK_LT(arg, ARRAYSIZE(files)); + string contents = ReadTestDataFile(files[arg].filename, + files[arg].size_limit); + + char* dst = new char[snappy::MaxCompressedLength(contents.size())]; + + SetBenchmarkBytesProcessed(static_cast<int64>(iters) * + static_cast<int64>(contents.size())); + StartBenchmarkTiming(); + + size_t zsize = 0; + while (iters-- > 0) { + snappy::RawCompress(contents.data(), contents.size(), dst, &zsize); + } + StopBenchmarkTiming(); + const double compression_ratio = + static_cast<double>(zsize) / std::max<size_t>(1, contents.size()); + SetBenchmarkLabel(StringPrintf("%s (%.2f %%)", + files[arg].label, 100.0 * compression_ratio)); + VLOG(0) << StringPrintf("compression for %s: %zd -> %zd bytes", + files[arg].label, contents.size(), zsize); + delete[] dst; +} +BENCHMARK(BM_ZFlat)->DenseRange(0, ARRAYSIZE(files) - 1); + + +} // namespace snappy + + +int main(int argc, char** argv) { + InitGoogle(argv[0], &argc, &argv, true); + File::Init(); + RunSpecifiedBenchmarks(); + + + if (argc >= 2) { + for (int arg = 1; arg < argc; arg++) { + if (FLAGS_write_compressed) { + CompressFile(argv[arg]); + } else if (FLAGS_write_uncompressed) { + UncompressFile(argv[arg]); + } else { + MeasureFile(argv[arg]); + } + } + return 0; + } + + return RUN_ALL_TESTS(); +} |