[test] Exercise all codes in dictionary tables

This commit is contained in:
Nick Terrell 2017-10-16 17:18:43 -07:00
parent b7977e348d
commit e600b5d0da
4 changed files with 436 additions and 15 deletions

View File

@ -130,7 +130,7 @@ zbufftest-dll : $(ZSTDDIR)/common/xxhash.c $(PRGDIR)/datagen.c zbufftest.c
$(MAKE) -C $(ZSTDDIR) libzstd
$(CC) $(CPPFLAGS) $(CFLAGS) $^ $(LDFLAGS) -o $@$(EXT)
ZSTREAMFILES := $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c zstreamtest.c
ZSTREAMFILES := $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c seqgen.c zstreamtest.c
zstreamtest : CPPFLAGS += $(MULTITHREAD_CPP)
zstreamtest : LDFLAGS += $(MULTITHREAD_LD)
zstreamtest : $(ZSTREAMFILES)

260
tests/seqgen.c Normal file
View File

@ -0,0 +1,260 @@
/*
* Copyright (c) 2017-present, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#include "seqgen.h"
#include "mem.h"
#include <string.h>
#define MIN(a, b) ((a) < (b) ? (a) : (b))
static const size_t kMatchBytes = 128;
#define SEQ_rotl32(x,r) ((x << r) | (x >> (32 - r)))
static BYTE SEQ_randByte(U32* src)
{
static const U32 prime1 = 2654435761U;
static const U32 prime2 = 2246822519U;
U32 rand32 = *src;
rand32 *= prime1;
rand32 ^= prime2;
rand32 = SEQ_rotl32(rand32, 13);
*src = rand32;
return (BYTE)(rand32 >> 5);
}
SEQ_stream SEQ_initStream(unsigned seed)
{
SEQ_stream stream;
stream.state = 0;
XXH64_reset(&stream.xxh, 0);
stream.seed = seed;
return stream;
}
/* Generates a single guard byte, then match length + 1 of a different byte,
* then another guard byte.
*/
static size_t SEQ_gen_matchLength(SEQ_stream* stream, unsigned value,
SEQ_outBuffer* out)
{
typedef enum {
ml_first_byte = 0,
ml_match_bytes,
ml_last_byte,
} ml_state;
BYTE* const ostart = (BYTE*)out->dst;
BYTE* const oend = ostart + out->size;
BYTE* op = ostart + out->pos;
switch ((ml_state)stream->state) {
case ml_first_byte:
/* Generate a single byte and pick a different byte for the match */
if (op >= oend) {
stream->bytesLeft = 1;
break;
}
*op = SEQ_randByte(&stream->seed) & 0xFF;
do {
stream->saved = SEQ_randByte(&stream->seed) & 0xFF;
} while (*op == stream->saved);
++op;
/* State transition */
stream->state = ml_match_bytes;
stream->bytesLeft = value + 1;
/* fall-through */
case ml_match_bytes: {
/* Copy matchLength + 1 bytes to the output buffer */
size_t const setLength = MIN(stream->bytesLeft, (size_t)(oend - op));
if (setLength > 0) {
memset(op, stream->saved, setLength);
op += setLength;
stream->bytesLeft -= setLength;
}
if (stream->bytesLeft > 0)
break;
/* State transition */
stream->state = ml_last_byte;
}
/* fall-through */
case ml_last_byte:
/* Generate a single byte and pick a different byte for the match */
if (op >= oend) {
stream->bytesLeft = 1;
break;
}
do {
*op = SEQ_randByte(&stream->seed) & 0xFF;
} while (*op == stream->saved);
++op;
/* State transition */
/* fall-through */
default:
stream->state = 0;
stream->bytesLeft = 0;
break;
}
XXH64_update(&stream->xxh, ostart + out->pos, (op - ostart) - out->pos);
out->pos = op - ostart;
return stream->bytesLeft;
}
/* Saves the current seed then generates kMatchBytes random bytes >= 128.
* Generates literal length - kMatchBytes random bytes < 128.
* Generates another kMatchBytes using the saved seed to generate a match.
* This way the match is easy to find for the compressors.
*/
static size_t SEQ_gen_litLength(SEQ_stream* stream, unsigned value, SEQ_outBuffer* out)
{
typedef enum {
ll_start = 0,
ll_run_bytes,
ll_literals,
ll_run_match,
} ll_state;
BYTE* const ostart = (BYTE*)out->dst;
BYTE* const oend = ostart + out->size;
BYTE* op = ostart + out->pos;
switch ((ll_state)stream->state) {
case ll_start:
stream->state = ll_run_bytes;
stream->saved = stream->seed;
stream->bytesLeft = MIN(kMatchBytes, value);
/* fall-through */
case ll_run_bytes:
while (stream->bytesLeft > 0 && op < oend) {
*op++ = SEQ_randByte(&stream->seed) | 0x80;
--stream->bytesLeft;
}
if (stream->bytesLeft > 0)
break;
/* State transition */
stream->state = ll_literals;
stream->bytesLeft = value - MIN(kMatchBytes, value);
/* fall-through */
case ll_literals:
while (stream->bytesLeft > 0 && op < oend) {
*op++ = SEQ_randByte(&stream->seed) & 0x7F;
--stream->bytesLeft;
}
if (stream->bytesLeft > 0)
break;
/* State transition */
stream->state = ll_run_match;
stream->bytesLeft = MIN(kMatchBytes, value);
/* fall-through */
case ll_run_match: {
while (stream->bytesLeft > 0 && op < oend) {
*op++ = SEQ_randByte(&stream->saved) | 0x80;
--stream->bytesLeft;
}
if (stream->bytesLeft > 0)
break;
}
/* fall-through */
default:
stream->state = 0;
stream->bytesLeft = 0;
break;
}
XXH64_update(&stream->xxh, ostart + out->pos, (op - ostart) - out->pos);
out->pos = op - ostart;
return stream->bytesLeft;
}
/* Saves the current seed then generates kMatchBytes random bytes >= 128.
* Generates offset - kMatchBytes of zeros to get a large offset without
* polluting the hash tables.
* Generates another kMatchBytes using the saved seed to generate a with the
* required offset.
*/
static size_t SEQ_gen_offset(SEQ_stream* stream, unsigned value, SEQ_outBuffer* out)
{
typedef enum {
of_start = 0,
of_run_bytes,
of_offset,
of_run_match,
} of_state;
BYTE* const ostart = (BYTE*)out->dst;
BYTE* const oend = ostart + out->size;
BYTE* op = ostart + out->pos;
switch ((of_state)stream->state) {
case of_start:
stream->state = of_run_bytes;
stream->saved = stream->seed;
stream->bytesLeft = MIN(value, kMatchBytes);
/* fall-through */
case of_run_bytes: {
while (stream->bytesLeft > 0 && op < oend) {
*op++ = SEQ_randByte(&stream->seed) | 0x80;
--stream->bytesLeft;
}
if (stream->bytesLeft > 0)
break;
/* State transition */
stream->state = of_offset;
stream->bytesLeft = value - MIN(value, kMatchBytes);
}
/* fall-through */
case of_offset: {
/* Copy matchLength + 1 bytes to the output buffer */
size_t const setLength = MIN(stream->bytesLeft, (size_t)(oend - op));
if (setLength > 0) {
memset(op, 0, setLength);
op += setLength;
stream->bytesLeft -= setLength;
}
if (stream->bytesLeft > 0)
break;
/* State transition */
stream->state = of_run_match;
stream->bytesLeft = MIN(value, kMatchBytes);
}
/* fall-through */
case of_run_match: {
while (stream->bytesLeft > 0 && op < oend) {
*op++ = SEQ_randByte(&stream->saved) | 0x80;
--stream->bytesLeft;
}
if (stream->bytesLeft > 0)
break;
}
/* fall-through */
default:
stream->state = 0;
stream->bytesLeft = 0;
break;
}
XXH64_update(&stream->xxh, ostart + out->pos, (op - ostart) - out->pos);
out->pos = op - ostart;
return stream->bytesLeft;
}
/* Returns the number of bytes left to generate.
* Must pass the same type/value until it returns 0.
*/
size_t SEQ_gen(SEQ_stream* stream, SEQ_gen_type type, unsigned value, SEQ_outBuffer* out)
{
switch (type) {
case SEQ_gen_ml: return SEQ_gen_matchLength(stream, value, out);
case SEQ_gen_ll: return SEQ_gen_litLength(stream, value, out);
case SEQ_gen_of: return SEQ_gen_offset(stream, value, out);
case SEQ_gen_max: /* fall-through */
default: return 0;
}
}
/* Returns the xxhash of the data produced so far */
XXH64_hash_t SEQ_digest(SEQ_stream const* stream)
{
return XXH64_digest(&stream->xxh);
}

58
tests/seqgen.h Normal file
View File

@ -0,0 +1,58 @@
/*
* Copyright (c) 2017-present, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef SEQGEN_H
#define SEQGEN_H
#define XXH_STATIC_LINKING_ONLY
#include "xxhash.h"
#include <stddef.h> /* size_t */
typedef enum {
SEQ_gen_ml = 0,
SEQ_gen_ll,
SEQ_gen_of,
SEQ_gen_max /* Must be the last value */
} SEQ_gen_type;
/* Internal state, do not use */
typedef struct {
XXH64_state_t xxh; /* xxh state for all the data produced so far (seed=0) */
unsigned seed;
int state; /* enum to control state machine (clean=0) */
unsigned saved;
size_t bytesLeft;
} SEQ_stream;
SEQ_stream SEQ_initStream(unsigned seed);
typedef struct {
void* dst;
size_t size;
size_t pos;
} SEQ_outBuffer;
/* Returns non-zero until the current type/value has been generated.
* Must pass the same type/value until it returns 0.
*
* Recommended to pick a value in the middle of the range you want, since there
* may be some noise that causes actual results to be slightly different.
* We try to be more accurate for smaller values.
*
* NOTE: Very small values don't work well (< 6).
*/
size_t SEQ_gen(SEQ_stream* stream, SEQ_gen_type type, unsigned value,
SEQ_outBuffer* out);
/* Returns the xxhash of the data produced so far */
XXH64_hash_t SEQ_digest(SEQ_stream const* stream);
#endif /* SEQGEN_H */

View File

@ -36,6 +36,7 @@
#include "datagen.h" /* RDG_genBuffer */
#define XXH_STATIC_LINKING_ONLY /* XXH64_state_t */
#include "xxhash.h" /* XXH64_* */
#include "seqgen.h"
/*-************************************
@ -96,15 +97,21 @@ unsigned int FUZ_rand(unsigned int* seedPtr)
return rand32 >> 5;
}
#define CHECK_Z(f) { \
size_t const err = f; \
if (ZSTD_isError(err)) { \
DISPLAY("Error => %s : %s ", \
#f, ZSTD_getErrorName(err)); \
DISPLAY(" (seed %u, test nb %u) \n", seed, testNb); \
#define CHECK(cond, ...) { \
if (cond) { \
DISPLAY("Error => "); \
DISPLAY(__VA_ARGS__); \
DISPLAY(" (seed %u, test nb %u, line %u) \n", \
seed, testNb, __LINE__); \
goto _output_error; \
} }
#define CHECK_Z(f) { \
size_t const err = f; \
CHECK(ZSTD_isError(err), "%s : %s ", \
#f, ZSTD_getErrorName(err)); \
}
/*======================================================
* Basic Unit tests
@ -144,6 +151,63 @@ static void FUZ_freeDictionary(buffer_t dict)
free(dict.start);
}
/* Round trips data and updates xxh with the decompressed data produced */
static size_t SEQ_roundTrip(ZSTD_CCtx* cctx, ZSTD_DCtx* dctx,
XXH64_state_t* xxh, void* data, size_t size,
ZSTD_EndDirective endOp)
{
static BYTE compressed[1024];
static BYTE uncompressed[1024];
ZSTD_inBuffer cin = {data, size, 0};
size_t cret;
do {
ZSTD_outBuffer cout = {compressed, sizeof(compressed), 0};
ZSTD_inBuffer din = {compressed, 0, 0};
ZSTD_outBuffer dout = {uncompressed, 0, 0};
cret = ZSTD_compress_generic(cctx, &cout, &cin, endOp);
if (ZSTD_isError(cret))
return cret;
din.size = cout.pos;
while (din.pos < din.size || (endOp == ZSTD_e_end && cret == 0)) {
size_t dret;
dout.pos = 0;
dout.size = sizeof(uncompressed);
dret = ZSTD_decompressStream(dctx, &dout, &din);
if (ZSTD_isError(dret))
return dret;
XXH64_update(xxh, dout.dst, dout.pos);
if (dret == 0)
break;
}
} while (cin.pos < cin.size || (endOp != ZSTD_e_continue && cret != 0));
return 0;
}
/* Generates some data and round trips it */
static size_t SEQ_generateRoundTrip(ZSTD_CCtx* cctx, ZSTD_DCtx* dctx,
XXH64_state_t* xxh, SEQ_stream* seq,
SEQ_gen_type type, unsigned value)
{
static BYTE data[1024];
size_t gen;
do {
SEQ_outBuffer sout = {data, sizeof(data), 0};
size_t ret;
gen = SEQ_gen(seq, type, value, &sout);
ret = SEQ_roundTrip(cctx, dctx, xxh, sout.dst, sout.pos, ZSTD_e_continue);
if (ZSTD_isError(ret))
return ret;
} while (gen != 0);
return 0;
}
static int basicUnitTests(U32 seed, double compressibility, ZSTD_customMem customMem)
{
@ -618,6 +682,53 @@ static int basicUnitTests(U32 seed, double compressibility, ZSTD_customMem custo
if (r != 0) goto _output_error; } /* error, or some data not flushed */
DISPLAYLEVEL(3, "OK \n");
DISPLAYLEVEL(3, "test%3i : check dictionary FSE tables can represent every code : ", testNb++);
{ unsigned const kMaxWindowLog = 24;
unsigned value;
ZSTD_compressionParameters cParams = ZSTD_getCParams(3, 1U << kMaxWindowLog, 1024);
ZSTD_CDict* cdict;
ZSTD_DDict* ddict;
SEQ_stream seq = SEQ_initStream(0x87654321);
SEQ_gen_type type;
XXH64_state_t xxh;
XXH64_reset(&xxh, 0);
cParams.windowLog = kMaxWindowLog;
cdict = ZSTD_createCDict_advanced(dictionary.start, dictionary.filled, ZSTD_dlm_byRef, ZSTD_dm_fullDict, cParams, ZSTD_defaultCMem);
ddict = ZSTD_createDDict(dictionary.start, dictionary.filled);
if (!cdict || !ddict) goto _output_error;
ZSTD_CCtx_reset(zc);
ZSTD_resetDStream(zd);
CHECK_Z(ZSTD_CCtx_refCDict(zc, cdict));
CHECK_Z(ZSTD_initDStream_usingDDict(zd, ddict));
CHECK_Z(ZSTD_setDStreamParameter(zd, DStream_p_maxWindowSize, 1U << kMaxWindowLog));
/* Test all values < 300 */
for (value = 0; value < 300; ++value) {
for (type = (SEQ_gen_type)0; type < SEQ_gen_max; ++type) {
CHECK_Z(SEQ_generateRoundTrip(zc, zd, &xxh, &seq, type, value));
}
}
/* Test values 2^8 to 2^17 */
for (value = (1 << 8); value < (1 << 17); value <<= 1) {
for (type = (SEQ_gen_type)0; type < SEQ_gen_max; ++type) {
CHECK_Z(SEQ_generateRoundTrip(zc, zd, &xxh, &seq, type, value));
CHECK_Z(SEQ_generateRoundTrip(zc, zd, &xxh, &seq, type, value + (value >> 2)));
}
}
/* Test offset values up to the max window log */
for (value = 8; value <= kMaxWindowLog; ++value) {
CHECK_Z(SEQ_generateRoundTrip(zc, zd, &xxh, &seq, SEQ_gen_of, (1U << value) - 1));
}
CHECK_Z(SEQ_roundTrip(zc, zd, &xxh, NULL, 0, ZSTD_e_end));
CHECK(SEQ_digest(&seq) != XXH64_digest(&xxh), "SEQ XXH64 does not match");
ZSTD_freeCDict(cdict);
ZSTD_freeDDict(ddict);
}
DISPLAYLEVEL(3, "OK \n");
/* Overlen overwriting window data bug */
DISPLAYLEVEL(3, "test%3i : wildcopy doesn't overwrite potential match data : ", testNb++);
@ -708,14 +819,6 @@ static U32 FUZ_randomClampedLength(U32* seed, U32 minVal, U32 maxVal)
return (U32)((FUZ_rand(seed) % mod) + minVal);
}
#define CHECK(cond, ...) { \
if (cond) { \
DISPLAY("Error => "); \
DISPLAY(__VA_ARGS__); \
DISPLAY(" (seed %u, test nb %u) \n", seed, testNb); \
goto _output_error; \
} }
static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, double compressibility, int bigTests)
{
U32 const maxSrcLog = bigTests ? 24 : 22;