You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@couchdb.apache.org by da...@apache.org on 2014/01/17 23:59:51 UTC
[18/50] [abbrv] Removed all google snappy files
http://git-wip-us.apache.org/repos/asf/couchdb-snappy/blob/f370a848/c_src/google-snappy/snappy.cc
----------------------------------------------------------------------
diff --git a/c_src/google-snappy/snappy.cc b/c_src/google-snappy/snappy.cc
deleted file mode 100644
index df377f1..0000000
--- a/c_src/google-snappy/snappy.cc
+++ /dev/null
@@ -1,1001 +0,0 @@
-// Copyright 2005 Google Inc. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#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) {
- DCHECK_GT(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) {
- UNALIGNED_STORE64(op, UNALIGNED_LOAD64(src));
- len -= op - src;
- op += op - src;
- }
- while (len > 0) {
- UNALIGNED_STORE64(op, UNALIGNED_LOAD64(src));
- 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) {
- UNALIGNED_STORE64(op, UNALIGNED_LOAD64(literal));
- UNALIGNED_STORE64(op + 8, UNALIGNED_LOAD64(literal + 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, int offset, int len) {
- DCHECK_LE(len, 64);
- DCHECK_GE(len, 4);
- DCHECK_LT(offset, 65536);
-
- if ((len < 12) && (offset < 2048)) {
- int 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, int 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);
- int htsize = 256;
- while (htsize < kMaxHashTableSize && htsize < input_size) {
- htsize <<= 1;
- }
- CHECK_EQ(0, htsize & (htsize - 1)) << ": must be power of two";
- CHECK_LE(htsize, kMaxHashTableSize) << ": hash table too large";
-
- 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(UNALIGNED_LOAD64(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.
-static inline uint32 GetUint32AtOffset(uint64 v, int offset) {
- DCHECK(0 <= offset && offset <= 4) << offset;
- return v >> (LittleEndian::IsLittleEndian() ? 8 * offset : 32 - 8 * offset);
-}
-
-// 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* const input,
- const 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;
- CHECK_LE(input_size, kBlockSize);
- CHECK_EQ(table_size & (table_size - 1), 0) << ": table must be power of two";
- const int shift = 32 - Bits::Log2Floor(table_size);
- DCHECK_EQ(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 int kInputMarginBytes = 15;
- if (PREDICT_TRUE(input_size >= kInputMarginBytes)) {
- const char* ip_limit = input + input_size - kInputMarginBytes;
-
- for (uint32 next_hash = Hash(++ip, shift); ; ) {
- DCHECK_LT(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;
- DCHECK_EQ(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];
- DCHECK_GE(candidate, base_ip);
- DCHECK_LT(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."
- DCHECK_LE(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.
- uint64 input_bytes = 0;
- 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;
- int offset = base - candidate;
- DCHECK_EQ(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 = UNALIGNED_LOAD64(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, uint32 length, bool allow_fast_path);
-// bool AppendFromSelf(uint32 offset, uint32 length);
-// };
-//
-// "allow_fast_path" is a parameter that says if there is at least 16
-// readable bytes in "ip". It is currently only used by SnappyArrayWriter.
-
-// -----------------------------------------------------------------------
-// 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
- DCHECK_EQ(extra, extra & 0x7); // At most 3 bits
- DCHECK_EQ(copy_offset, copy_offset & 0x7); // At most 3 bits
- DCHECK_EQ(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.
- CHECK_EQ(assigned, 256);
- for (int i = 0; i < 256; i++) {
- CHECK_NE(dst[i], 0xffff);
- }
-
- 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++) {
- CHECK_EQ(dst[i], char_table[i]);
- }
-}
-REGISTER_MODULE_INITIALIZER(snappy, ComputeTable());
-#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) {
- DCHECK(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>
- bool Step(Writer* writer) {
- const char* ip = ip_;
- if (ip_limit_ - ip < 5) {
- if (!RefillTag()) return false;
- ip = ip_;
- }
-
- const unsigned char c = *(reinterpret_cast<const unsigned char*>(ip++));
- const uint32 entry = char_table[c];
- const uint32 trailer = LittleEndian::Load32(ip) & wordmask[entry >> 11];
- ip += entry >> 11;
- const uint32 length = entry & 0xff;
-
- if ((c & 0x3) == LITERAL) {
- uint32 literal_length = length + trailer;
- uint32 avail = ip_limit_ - ip;
- while (avail < literal_length) {
- bool allow_fast_path = (avail >= 16);
- if (!writer->Append(ip, avail, allow_fast_path)) return false;
- literal_length -= avail;
- reader_->Skip(peeked_);
- size_t n;
- ip = reader_->Peek(&n);
- avail = n;
- peeked_ = avail;
- if (avail == 0) return false; // Premature end of input
- ip_limit_ = ip + avail;
- }
- ip_ = ip + literal_length;
- bool allow_fast_path = (avail >= 16);
- return writer->Append(ip, literal_length, allow_fast_path);
- } else {
- ip_ = ip;
- // 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;
- return writer->AppendFromSelf(copy_offset + trailer, length);
- }
- }
-};
-
-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
- DCHECK_LT(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'
- DCHECK_LE(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);
- }
- DCHECK_EQ(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;
- // Protect against possible DoS attack
- if (static_cast<uint64>(uncompressed_len) > max_len) {
- return false;
- }
-
- writer->SetExpectedLength(uncompressed_len);
-
- // Process the entire input
- while (decompressor.Step(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;
- int 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);
- DCHECK_NE(fragment_size, 0) << ": premature end of input";
- const int num_to_read = min(N, kBlockSize);
- size_t bytes_read = fragment_size;
-
- int 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);
- }
- DCHECK_EQ(bytes_read, num_to_read);
- fragment = scratch;
- fragment_size = num_to_read;
- }
- DCHECK_EQ(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::Step().
-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, uint32 len, bool allow_fast_path) {
- char* op = op_;
- const int space_left = op_limit_ - op;
- if (allow_fast_path && len <= 16 && space_left >= 16) {
- // Fast path, used for the majority (about 90%) of dynamic invocations.
- UNALIGNED_STORE64(op, UNALIGNED_LOAD64(ip));
- UNALIGNED_STORE64(op + 8, UNALIGNED_LOAD64(ip + 8));
- } else {
- if (space_left < len) {
- return false;
- }
- memcpy(op, ip, len);
- }
- op_ = op + len;
- return true;
- }
-
- inline bool AppendFromSelf(uint32 offset, uint32 len) {
- char* op = op_;
- const int 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.
- UNALIGNED_STORE64(op, UNALIGNED_LOAD64(op - offset));
- UNALIGNED_STORE64(op + 8, UNALIGNED_LOAD64(op - offset + 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, uint32 len, bool allow_fast_path) {
- produced_ += len;
- return produced_ <= expected_;
- }
- inline bool AppendFromSelf(uint32 offset, uint32 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
-
http://git-wip-us.apache.org/repos/asf/couchdb-snappy/blob/f370a848/c_src/google-snappy/snappy.h
----------------------------------------------------------------------
diff --git a/c_src/google-snappy/snappy.h b/c_src/google-snappy/snappy.h
deleted file mode 100644
index 0c05f0b..0000000
--- a/c_src/google-snappy/snappy.h
+++ /dev/null
@@ -1,141 +0,0 @@
-// Copyright 2005 and onwards Google Inc.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-// 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);
-
- 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);
-
- // *** DO NOT CHANGE THE VALUE OF kBlockSize ***
- //
- // New Compression code chops up the input into blocks of at most
- // the following size. This ensures that back-references in the
- // output never cross kBlockSize block boundaries. This can be
- // helpful in implementing blocked decompression. However the
- // decompression code should not rely on this guarantee since older
- // compression code may not obey it.
- static const int kBlockLog = 15;
- static const int kBlockSize = 1 << kBlockLog;
-
- static const int kMaxHashTableBits = 14;
- static const int kMaxHashTableSize = 1 << kMaxHashTableBits;
-
-} // end namespace snappy
-
-
-#endif // UTIL_SNAPPY_SNAPPY_H__
http://git-wip-us.apache.org/repos/asf/couchdb-snappy/blob/f370a848/c_src/google-snappy/snappy_unittest.cc
----------------------------------------------------------------------
diff --git a/c_src/google-snappy/snappy_unittest.cc b/c_src/google-snappy/snappy_unittest.cc
deleted file mode 100644
index 23c6eec..0000000
--- a/c_src/google-snappy/snappy_unittest.cc
+++ /dev/null
@@ -1,1078 +0,0 @@
-// Copyright 2005 and onwards Google Inc.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#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",
-};
-
-// Returns true if we successfully compressed, false otherwise
-static bool Compress(const char* input, size_t input_size, CompressorType comp,
- string* compressed) {
- switch (comp) {
-#ifdef ZLIB_VERSION
- case ZLIB: {
- compressed->resize(ZLib::MinCompressbufSize(input_size));
- 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);
- compressed->resize(destlen);
- return true;
- }
-#endif // ZLIB_VERSION
-
-#ifdef LZO_VERSION
- case LZO: {
- compressed->resize(input_size + input_size/64 + 16 + 3);
- 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;
- compressed->resize(destlen);
- break;
- }
-#endif // LZO_VERSION
-
-#ifdef LZF_VERSION
- case LIBLZF: {
- compressed->resize(input_size - 1);
- int destlen = lzf_compress(input,
- input_size,
- string_as_array(compressed),
- input_size - 1);
- 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;
- }
- compressed->resize(destlen);
- break;
- }
-#endif // LZF_VERSION
-
-#ifdef QLZ_VERSION_MAJOR
- case QUICKLZ: {
- compressed->resize(input_size + 36000); // 36000 is used for scratch
- 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);
- compressed->resize(destlen);
- break;
- }
-#endif // QLZ_VERSION_MAJOR
-
-#ifdef FASTLZ_VERSION
- case FASTLZ: {
- int compressed_size = max(static_cast<int>(ceil(input_size * 1.05)),
- 66);
- compressed->resize(compressed_size);
- // Use level 1 compression since we mostly care about speed.
- int destlen = fastlz_compress_level(
- 1,
- input,
- input_size,
- string_as_array(compressed));
- 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));
- 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(destlen, size);
- 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(destlen, size);
- 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 stupid string
- // append time
- compressed[b].resize(block_size * 2);
- }
-
- // First, try one trial compression to make sure the code is compiled in
- if (!Compress(input[0], input_length[0], comp, &compressed[0])) {
- LOG(WARNING) << "Skipping " << names[comp] << ": "
- << "library not compiled in";
- return;
- }
-
- for (int run = 0; run < kRuns; run++) {
- CycleTimer ctimer, utimer;
-
- 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]);
- ctimer.Stop();
-
- 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 3GBytes 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., 3221225471 bytes
- dest[4] = 'k';
- CHECK(!IsValidCompressedBuffer(TypeParam(dest)));
- CHECK(!Uncompress(TypeParam(dest), &uncmp));
- dest[0] = dest[1] = dest[2] = 0xff;
- dest[3] = 0x7f;
- CHECK(!IsValidCompressedBuffer(TypeParam(dest)));
- CHECK(!Uncompress(TypeParam(dest), &uncmp));
- } else {
- LOG(WARNING) << "Crazy decompression lengths not checked on 64-bit build";
- }
-
- // 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());
- 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 is each fragment emittedn
- 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::ReadFileToStringOrDie(fname, &fullinput);
-
- string compressed;
- compressed.resize(fullinput.size() * 2);
- Compress(fullinput.data(), fullinput.size(), SNAPPY, &compressed);
-
- File::WriteStringToFileOrDie(compressed,
- string(fname).append(".comp").c_str());
-}
-
-static void UncompressFile(const char* fname) {
- string fullinput;
- File::ReadFileToStringOrDie(fname, &fullinput);
-
- size_t uncompLength;
- CHECK(CheckUncompressedLength(fullinput, &uncompLength));
-
- string uncompressed;
- uncompressed.resize(uncompLength);
- CHECK(snappy::Uncompress(fullinput.data(), fullinput.size(), &uncompressed));
-
- File::WriteStringToFileOrDie(uncompressed,
- string(fname).append(".uncomp").c_str());
-}
-
-static void MeasureFile(const char* fname) {
- string fullinput;
- File::ReadFileToStringOrDie(fname, &fullinput);
- 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;
-} files[] = {
- { "html", "html" },
- { "urls", "urls.10K" },
- { "jpg", "house.jpg" },
- { "pdf", "mapreduce-osdi-1.pdf" },
- { "html4", "html_x_4" },
- { "cp", "cp.html" },
- { "c", "fields.c" },
- { "lsp", "grammar.lsp" },
- { "xls", "kennedy.xls" },
- { "txt1", "alice29.txt" },
- { "txt2", "asyoulik.txt" },
- { "txt3", "lcet10.txt" },
- { "txt4", "plrabn12.txt" },
- { "bin", "ptt5" },
- { "sum", "sum" },
- { "man", "xargs.1" },
- { "pb", "geo.protodata" },
- { "gaviota", "kppkn.gtb" },
-};
-
-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);
-
- 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, 17);
-
-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);
-
- 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);
-
- 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, 17);
-
-
-} // 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();
-}