zstd-native-ruby 1.0.0

data/vendor/zstd/programs/fileio_asyncio.h

@@ -0,0 +1,195 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * 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.
+ */
+
+ /*
+  * FileIO AsyncIO exposes read/write IO pools that allow doing IO asynchronously.
+  * Current implementation relies on having one thread that reads and one that
+  * writes.
+  * Each IO pool supports up to `MAX_IO_JOBS` that can be enqueued for work, but
+  * are performed serially by the appropriate worker thread.
+  * Most systems exposes better primitives to perform asynchronous IO, such as
+  * io_uring on newer linux systems. The API is built in such a way that in the
+  * future we could replace the threads with better solutions when available.
+  */
+
+#ifndef ZSTD_FILEIO_ASYNCIO_H
+#define ZSTD_FILEIO_ASYNCIO_H
+
+#include "../lib/common/mem.h"     /* U32, U64 */
+#include "fileio_types.h"
+#include "platform.h"
+#include "util.h"
+#include "../lib/common/pool.h"
+#include "../lib/common/threading.h"
+
+#define MAX_IO_JOBS          (10)
+
+typedef struct {
+    /* These struct fields should be set only on creation and not changed afterwards */
+    POOL_ctx* threadPool;
+    int threadPoolActive;
+    int totalIoJobs;
+    const FIO_prefs_t* prefs;
+    POOL_function poolFunction;
+
+    /* Controls the file we currently write to, make changes only by using provided utility functions */
+    FILE* file;
+
+    /* The jobs and availableJobsCount fields are accessed by both the main and worker threads and should
+     * only be mutated after locking the mutex */
+    ZSTD_pthread_mutex_t ioJobsMutex;
+    void* availableJobs[MAX_IO_JOBS];
+    int availableJobsCount;
+    size_t jobBufferSize;
+} IOPoolCtx_t;
+
+typedef struct {
+    IOPoolCtx_t base;
+
+    /* State regarding the currently read file */
+    int reachedEof;
+    U64 nextReadOffset;
+    U64 waitingOnOffset;
+
+    /* We may hold an IOJob object as needed if we actively expose its buffer. */
+    void *currentJobHeld;
+
+    /* Coalesce buffer is used to join two buffers in case where we need to read more bytes than left in
+     * the first of them. Shouldn't be accessed from outside ot utility functions. */
+    U8 *coalesceBuffer;
+
+    /* Read buffer can be used by consumer code, take care when copying this pointer aside as it might
+     * change when consuming / refilling buffer. */
+    U8 *srcBuffer;
+    size_t srcBufferLoaded;
+
+    /* We need to know what tasks completed so we can use their buffers when their time comes.
+     * Should only be accessed after locking base.ioJobsMutex . */
+    void* completedJobs[MAX_IO_JOBS];
+    int completedJobsCount;
+    ZSTD_pthread_cond_t jobCompletedCond;
+} ReadPoolCtx_t;
+
+typedef struct {
+    IOPoolCtx_t base;
+    unsigned storedSkips;
+} WritePoolCtx_t;
+
+typedef struct {
+    /* These fields are automatically set and shouldn't be changed by non WritePool code. */
+    void *ctx;
+    FILE* file;
+    void *buffer;
+    size_t bufferSize;
+
+    /* This field should be changed before a job is queued for execution and should contain the number
+     * of bytes to write from the buffer. */
+    size_t usedBufferSize;
+    U64 offset;
+} IOJob_t;
+
+/* AIO_supported:
+ * Returns 1 if AsyncIO is supported on the system, 0 otherwise. */
+int AIO_supported(void);
+
+
+/* AIO_WritePool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+void AIO_WritePool_releaseIoJob(IOJob_t *job);
+
+/* AIO_WritePool_acquireJob:
+ * Returns an available write job to be used for a future write. */
+IOJob_t* AIO_WritePool_acquireJob(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_enqueueAndReacquireWriteJob:
+ * Enqueues a write job for execution and acquires a new one.
+ * After execution `job`'s pointed value would change to the newly acquired job.
+ * Make sure to set `usedBufferSize` to the wanted length before call.
+ * The queued job shouldn't be used directly after queueing it. */
+void AIO_WritePool_enqueueAndReacquireWriteJob(IOJob_t **job);
+
+/* AIO_WritePool_sparseWriteEnd:
+ * Ends sparse writes to the current file.
+ * Blocks on completion of all current write jobs before executing. */
+void AIO_WritePool_sparseWriteEnd(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_setFile:
+ * Sets the destination file for future writes in the pool.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.
+ * Also requires ending of sparse write if a previous file was used in sparse mode. */
+void AIO_WritePool_setFile(WritePoolCtx_t *ctx, FILE* file);
+
+/* AIO_WritePool_getFile:
+ * Returns the file the writePool is currently set to write to. */
+FILE* AIO_WritePool_getFile(const WritePoolCtx_t* ctx);
+
+/* AIO_WritePool_closeFile:
+ * Ends sparse write and closes the writePool's current file and sets the file to NULL.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.  */
+int AIO_WritePool_closeFile(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_create:
+ * Allocates and sets and a new write pool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to write to at a time. */
+WritePoolCtx_t* AIO_WritePool_create(const FIO_prefs_t* prefs, size_t bufferSize);
+
+/* AIO_WritePool_free:
+ * Frees and releases a writePool and its resources. Closes destination file. */
+void AIO_WritePool_free(WritePoolCtx_t* ctx);
+
+/* AIO_WritePool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_WritePool_setAsync(WritePoolCtx_t* ctx, int async);
+
+/* AIO_ReadPool_create:
+ * Allocates and sets and a new readPool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to read at a time, will also be used
+ * as our basic read size. */
+ReadPoolCtx_t* AIO_ReadPool_create(const FIO_prefs_t* prefs, size_t bufferSize);
+
+/* AIO_ReadPool_free:
+ * Frees and releases a readPool and its resources. Closes source file. */
+void AIO_ReadPool_free(ReadPoolCtx_t* ctx);
+
+/* AIO_ReadPool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_ReadPool_setAsync(ReadPoolCtx_t* ctx, int async);
+
+/* AIO_ReadPool_consumeBytes:
+ * Consumes byes from srcBuffer's beginning and updates srcBufferLoaded accordingly. */
+void AIO_ReadPool_consumeBytes(ReadPoolCtx_t *ctx, size_t n);
+
+/* AIO_ReadPool_fillBuffer:
+ * Makes sure buffer has at least n bytes loaded (as long as n is not bigger than the initialized bufferSize).
+ * Returns if srcBuffer has at least n bytes loaded or if we've reached the end of the file.
+ * Return value is the number of bytes added to the buffer.
+ * Note that srcBuffer might have up to 2 times bufferSize bytes. */
+size_t AIO_ReadPool_fillBuffer(ReadPoolCtx_t *ctx, size_t n);
+
+/* AIO_ReadPool_consumeAndRefill:
+ * Consumes the current buffer and refills it with bufferSize bytes. */
+size_t AIO_ReadPool_consumeAndRefill(ReadPoolCtx_t *ctx);
+
+/* AIO_ReadPool_setFile:
+ * Sets the source file for future read in the pool. Initiates reading immediately if file is not NULL.
+ * Waits for all current enqueued tasks to complete if a previous file was set. */
+void AIO_ReadPool_setFile(ReadPoolCtx_t *ctx, FILE* file);
+
+/* AIO_ReadPool_getFile:
+ * Returns the current file set for the read pool. */
+FILE* AIO_ReadPool_getFile(const ReadPoolCtx_t *ctx);
+
+/* AIO_ReadPool_closeFile:
+ * Closes the current set file. Waits for all current enqueued tasks to complete and resets state. */
+int AIO_ReadPool_closeFile(ReadPoolCtx_t *ctx);
+
+#endif /* ZSTD_FILEIO_ASYNCIO_H */

data/vendor/zstd/programs/fileio_common.h

@@ -0,0 +1,121 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * 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 ZSTD_FILEIO_COMMON_H
+#define ZSTD_FILEIO_COMMON_H
+
+#include "../lib/common/mem.h"     /* U32, U64 */
+#include "fileio_types.h"
+#include "platform.h"
+#include "timefn.h"     /* UTIL_getTime, UTIL_clockSpanMicro */
+
+/*-*************************************
+*  Macros
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+#undef MAX
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+#undef MIN  /* in case it would be already defined */
+#define MIN(a,b)    ((a) < (b) ? (a) : (b))
+
+extern FIO_display_prefs_t g_display_prefs;
+
+#define DISPLAY_F(f, ...)    fprintf((f), __VA_ARGS__)
+#define DISPLAYOUT(...)      DISPLAY_F(stdout, __VA_ARGS__)
+#define DISPLAY(...)         DISPLAY_F(stderr, __VA_ARGS__)
+#define DISPLAYLEVEL(l, ...) { if (g_display_prefs.displayLevel>=l) { DISPLAY(__VA_ARGS__); } }
+
+extern UTIL_time_t g_displayClock;
+
+#define REFRESH_RATE  ((U64)(SEC_TO_MICRO / 6))
+#define READY_FOR_UPDATE() (UTIL_clockSpanMicro(g_displayClock) > REFRESH_RATE || g_display_prefs.displayLevel >= 4)
+#define DELAY_NEXT_UPDATE() { g_displayClock = UTIL_getTime(); }
+#define DISPLAYUPDATE(l, ...) {                              \
+        if (g_display_prefs.displayLevel>=l && (g_display_prefs.progressSetting != FIO_ps_never)) { \
+            if (READY_FOR_UPDATE()) { \
+                DELAY_NEXT_UPDATE();                         \
+                DISPLAY(__VA_ARGS__);                        \
+                if (g_display_prefs.displayLevel>=4) fflush(stderr);       \
+    }   }   }
+
+#define SHOULD_DISPLAY_SUMMARY() \
+    (g_display_prefs.displayLevel >= 2 || g_display_prefs.progressSetting == FIO_ps_always)
+#define SHOULD_DISPLAY_PROGRESS()                       \
+    (g_display_prefs.progressSetting != FIO_ps_never && SHOULD_DISPLAY_SUMMARY())
+#define DISPLAY_PROGRESS(...) { if (SHOULD_DISPLAY_PROGRESS()) { DISPLAYLEVEL(1, __VA_ARGS__); }}
+#define DISPLAYUPDATE_PROGRESS(...) { if (SHOULD_DISPLAY_PROGRESS()) { DISPLAYUPDATE(1, __VA_ARGS__); }}
+#define DISPLAY_SUMMARY(...) { if (SHOULD_DISPLAY_SUMMARY()) { DISPLAYLEVEL(1, __VA_ARGS__); } }
+
+#define EXM_THROW(error, ...)                                             \
+{                                                                         \
+    DISPLAYLEVEL(1, "zstd: ");                                            \
+    DISPLAYLEVEL(5, "Error defined at %s, line %i : \n", __FILE__, __LINE__); \
+    DISPLAYLEVEL(1, "error %i : ", error);                                \
+    DISPLAYLEVEL(1, __VA_ARGS__);                                         \
+    DISPLAYLEVEL(1, " \n");                                               \
+    exit(error);                                                          \
+}
+
+#define CHECK_V(v, f)                                \
+    v = f;                                           \
+    if (ZSTD_isError(v)) {                           \
+        DISPLAYLEVEL(5, "%s \n", #f);                \
+        EXM_THROW(11, "%s", ZSTD_getErrorName(v));   \
+    }
+#define CHECK(f) { size_t err; CHECK_V(err, f); }
+
+
+/* Avoid fseek()'s 2GiB barrier with MSVC, macOS, *BSD, MinGW */
+#if defined(LIBC_NO_FSEEKO)
+/* Some older libc implementations don't include these functions (e.g. Bionic < 24) */
+#   define LONG_SEEK fseek
+#   define LONG_TELL ftell
+#elif defined(_MSC_VER) && _MSC_VER >= 1400
+#   define LONG_SEEK _fseeki64
+#   define LONG_TELL _ftelli64
+#elif !defined(__64BIT__) && (PLATFORM_POSIX_VERSION >= 200112L) /* No point defining Large file for 64 bit */
+#   define LONG_SEEK fseeko
+#   define LONG_TELL ftello
+#elif defined(__MINGW32__) && !defined(__STRICT_ANSI__) && !defined(__NO_MINGW_LFS) && defined(__MSVCRT__)
+#   define LONG_SEEK fseeko64
+#   define LONG_TELL ftello64
+#elif defined(_WIN32) && !defined(__DJGPP__)
+#   include <windows.h>
+    static int LONG_SEEK(FILE* file, __int64 offset, int origin) {
+        LARGE_INTEGER off;
+        DWORD method;
+        off.QuadPart = offset;
+        if (origin == SEEK_END)
+            method = FILE_END;
+        else if (origin == SEEK_CUR)
+            method = FILE_CURRENT;
+        else
+            method = FILE_BEGIN;
+
+        if (SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, NULL, method))
+            return 0;
+        else
+            return -1;
+    }
+    static __int64 LONG_TELL(FILE* file) {
+        LARGE_INTEGER off, newOff;
+        off.QuadPart = 0;
+        newOff.QuadPart = 0;
+        SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, &newOff, FILE_CURRENT);
+        return newOff.QuadPart;
+    }
+#else
+#   define LONG_SEEK fseek
+#   define LONG_TELL ftell
+#endif
+
+#endif /* ZSTD_FILEIO_COMMON_H */

data/vendor/zstd/programs/fileio_types.h

@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * 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 FILEIO_TYPES_HEADER
+#define FILEIO_TYPES_HEADER
+
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_compressionParameters */
+#include "../lib/zstd.h"           /* ZSTD_* */
+
+/*-*************************************
+*  Parameters: FIO_prefs_t
+***************************************/
+
+typedef struct FIO_display_prefs_s FIO_display_prefs_t;
+
+typedef enum { FIO_ps_auto, FIO_ps_never, FIO_ps_always } FIO_progressSetting_e;
+
+struct FIO_display_prefs_s {
+    int displayLevel;   /* 0 : no display;  1: errors;  2: + result + interaction + warnings;  3: + progression;  4: + information */
+    FIO_progressSetting_e progressSetting;
+};
+
+
+typedef enum { FIO_zstdCompression, FIO_gzipCompression, FIO_xzCompression, FIO_lzmaCompression, FIO_lz4Compression } FIO_compressionType_t;
+
+typedef struct FIO_prefs_s {
+
+    /* Algorithm preferences */
+    FIO_compressionType_t compressionType;
+    int sparseFileSupport;   /* 0: no sparse allowed; 1: auto (file yes, stdout no); 2: force sparse */
+    int dictIDFlag;
+    int checksumFlag;
+    int jobSize;
+    int overlapLog;
+    int adaptiveMode;
+    int useRowMatchFinder;
+    int rsyncable;
+    int minAdaptLevel;
+    int maxAdaptLevel;
+    int ldmFlag;
+    int ldmHashLog;
+    int ldmMinMatch;
+    int ldmBucketSizeLog;
+    int ldmHashRateLog;
+    size_t streamSrcSize;
+    size_t targetCBlockSize;
+    int srcSizeHint;
+    int testMode;
+    ZSTD_ParamSwitch_e literalCompressionMode;
+
+    /* IO preferences */
+    int removeSrcFile;
+    int overwrite;
+    int asyncIO;
+
+    /* Computation resources preferences */
+    unsigned memLimit;
+    int nbWorkers;
+
+    int excludeCompressedFiles;
+    int patchFromMode;
+    int contentSize;
+    int allowBlockDevices;
+    int passThrough;
+    ZSTD_ParamSwitch_e mmapDict;
+} FIO_prefs_t;
+
+typedef enum {FIO_mallocDict, FIO_mmapDict} FIO_dictBufferType_t;
+
+typedef struct {
+    void* dictBuffer;
+    size_t dictBufferSize;
+    FIO_dictBufferType_t dictBufferType;
+#if defined(_MSC_VER) || defined(_WIN32)
+    HANDLE dictHandle;
+#endif
+} FIO_Dict_t;
+
+#endif /* FILEIO_TYPES_HEADER */

data/vendor/zstd/programs/lorem.c

@@ -0,0 +1,285 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * 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.
+ */
+
+/* Implementation notes:
+ *
+ * This is a very simple lorem ipsum generator
+ * which features a static list of words
+ * and print them one after another randomly
+ * with a fake sentence / paragraph structure.
+ *
+ * The goal is to generate a printable text
+ * that can be used to fake a text compression scenario.
+ * The resulting compression / ratio curve of the lorem ipsum generator
+ * is more satisfying than the previous statistical generator,
+ * which was initially designed for entropy compression,
+ * and lacks a regularity more representative of text.
+ *
+ * The compression ratio achievable on the generated lorem ipsum
+ * is still a bit too good, presumably because the dictionary is a bit too
+ * small. It would be possible to create some more complex scheme, notably by
+ * enlarging the dictionary with a word generator, and adding grammatical rules
+ * (composition) and syntax rules. But that's probably overkill for the intended
+ * goal.
+ */
+
+#include "lorem.h"
+#include <assert.h>
+#include <limits.h> /* INT_MAX */
+#include <string.h> /* memcpy */
+
+#define WORD_MAX_SIZE 20
+
+/* Define the word pool */
+static const char* kWords[] = {
+    "lorem",        "ipsum",      "dolor",       "sit",          "amet",
+    "consectetur",  "adipiscing", "elit",        "sed",          "do",
+    "eiusmod",      "tempor",     "incididunt",  "ut",           "labore",
+    "et",           "dolore",     "magna",       "aliqua",       "dis",
+    "lectus",       "vestibulum", "mattis",      "ullamcorper",  "velit",
+    "commodo",      "a",          "lacus",       "arcu",         "magnis",
+    "parturient",   "montes",     "nascetur",    "ridiculus",    "mus",
+    "mauris",       "nulla",      "malesuada",   "pellentesque", "eget",
+    "gravida",      "in",         "dictum",      "non",          "erat",
+    "nam",          "voluptat",   "maecenas",    "blandit",      "aliquam",
+    "etiam",        "enim",       "lobortis",    "scelerisque",  "fermentum",
+    "dui",          "faucibus",   "ornare",      "at",           "elementum",
+    "eu",           "facilisis",  "odio",        "morbi",        "quis",
+    "eros",         "donec",      "ac",          "orci",         "purus",
+    "turpis",       "cursus",     "leo",         "vel",          "porta",
+    "consequat",    "interdum",   "varius",      "vulputate",    "aliquet",
+    "pharetra",     "nunc",       "auctor",      "urna",         "id",
+    "metus",        "viverra",    "nibh",        "cras",         "mi",
+    "unde",         "omnis",      "iste",        "natus",        "error",
+    "perspiciatis", "voluptatem", "accusantium", "doloremque",   "laudantium",
+    "totam",        "rem",        "aperiam",     "eaque",        "ipsa",
+    "quae",         "ab",         "illo",        "inventore",    "veritatis",
+    "quasi",        "architecto", "beatae",      "vitae",        "dicta",
+    "sunt",         "explicabo",  "nemo",        "ipsam",        "quia",
+    "voluptas",     "aspernatur", "aut",         "odit",         "fugit",
+    "consequuntur", "magni",      "dolores",     "eos",          "qui",
+    "ratione",      "sequi",      "nesciunt",    "neque",        "porro",
+    "quisquam",     "est",        "dolorem",     "adipisci",     "numquam",
+    "eius",         "modi",       "tempora",     "incidunt",     "magnam",
+    "quaerat",      "ad",         "minima",      "veniam",       "nostrum",
+    "ullam",        "corporis",   "suscipit",    "laboriosam",   "nisi",
+    "aliquid",      "ex",         "ea",          "commodi",      "consequatur",
+    "autem",        "eum",        "iure",        "voluptate",    "esse",
+    "quam",         "nihil",      "molestiae",   "illum",        "fugiat",
+    "quo",          "pariatur",   "vero",        "accusamus",    "iusto",
+    "dignissimos",  "ducimus",    "blanditiis",  "praesentium",  "voluptatum",
+    "deleniti",     "atque",      "corrupti",    "quos",         "quas",
+    "molestias",    "excepturi",  "sint",        "occaecati",    "cupiditate",
+    "provident",    "similique",  "culpa",       "officia",      "deserunt",
+    "mollitia",     "animi",      "laborum",     "dolorum",      "fuga",
+    "harum",        "quidem",     "rerum",       "facilis",      "expedita",
+    "distinctio",   "libero",     "tempore",     "cum",          "soluta",
+    "nobis",        "eligendi",   "optio",       "cumque",       "impedit",
+    "minus",        "quod",       "maxime",      "placeat",      "facere",
+    "possimus",     "assumenda",  "repellendus", "temporibus",   "quibusdam",
+    "officiis",     "debitis",    "saepe",       "eveniet",      "voluptates",
+    "repudiandae",  "recusandae", "itaque",      "earum",        "hic",
+    "tenetur",      "sapiente",   "delectus",    "reiciendis",   "cillum",
+    "maiores",      "alias",      "perferendis", "doloribus",    "asperiores",
+    "repellat",     "minim",      "nostrud",     "exercitation", "ullamco",
+    "laboris",      "aliquip",    "duis",        "aute",         "irure",
+};
+static const unsigned kNbWords = sizeof(kWords) / sizeof(kWords[0]);
+
+/* simple 1-dimension distribution, based on word's length, favors small words
+ */
+static const int kWeights[]    = { 0, 8, 6, 4, 3, 2 };
+static const size_t kNbWeights = sizeof(kWeights) / sizeof(kWeights[0]);
+
+#define DISTRIB_SIZE_MAX 650
+static int g_distrib[DISTRIB_SIZE_MAX] = { 0 };
+static unsigned g_distribCount         = 0;
+
+static void countFreqs(
+        const char* words[],
+        size_t nbWords,
+        const int* weights,
+        size_t nbWeights)
+{
+    unsigned total = 0;
+    size_t w;
+    for (w = 0; w < nbWords; w++) {
+        size_t len = strlen(words[w]);
+        int lmax;
+        if (len >= nbWeights)
+            len = nbWeights - 1;
+        lmax = weights[len];
+        total += (unsigned)lmax;
+    }
+    g_distribCount = total;
+    assert(g_distribCount <= DISTRIB_SIZE_MAX);
+}
+
+static void init_word_distrib(
+        const char* words[],
+        size_t nbWords,
+        const int* weights,
+        size_t nbWeights)
+{
+    size_t w, d = 0;
+    countFreqs(words, nbWords, weights, nbWeights);
+    for (w = 0; w < nbWords; w++) {
+        size_t len = strlen(words[w]);
+        int l, lmax;
+        if (len >= nbWeights)
+            len = nbWeights - 1;
+        lmax = weights[len];
+        for (l = 0; l < lmax; l++) {
+            g_distrib[d++] = (int)w;
+        }
+    }
+}
+
+/* Note: this unit only works when invoked sequentially.
+ * No concurrent access is allowed */
+static char* g_ptr         = NULL;
+static size_t g_nbChars    = 0;
+static size_t g_maxChars   = 10000000;
+static unsigned g_randRoot = 0;
+
+#define RDG_rotl32(x, r) ((x << r) | (x >> (32 - r)))
+static unsigned LOREM_rand(unsigned range)
+{
+    static const unsigned prime1 = 2654435761U;
+    static const unsigned prime2 = 2246822519U;
+    unsigned rand32              = g_randRoot;
+    rand32 *= prime1;
+    rand32 ^= prime2;
+    rand32     = RDG_rotl32(rand32, 13);
+    g_randRoot = rand32;
+    return (unsigned)(((unsigned long long)rand32 * range) >> 32);
+}
+
+static void writeLastCharacters(void)
+{
+    size_t lastChars = g_maxChars - g_nbChars;
+    assert(g_maxChars >= g_nbChars);
+    if (lastChars == 0)
+        return;
+    g_ptr[g_nbChars++] = '.';
+    if (lastChars > 2) {
+        memset(g_ptr + g_nbChars, ' ', lastChars - 2);
+    }
+    if (lastChars > 1) {
+        g_ptr[g_maxChars - 1] = '\n';
+    }
+    g_nbChars = g_maxChars;
+}
+
+static void generateWord(const char* word, const char* separator, int upCase)
+{
+    size_t const len = strlen(word) + strlen(separator);
+    if (g_nbChars + len > g_maxChars) {
+        writeLastCharacters();
+        return;
+    }
+    memcpy(g_ptr + g_nbChars, word, strlen(word));
+    if (upCase) {
+        static const char toUp = 'A' - 'a';
+        g_ptr[g_nbChars]       = (char)(g_ptr[g_nbChars] + toUp);
+    }
+    g_nbChars += strlen(word);
+    memcpy(g_ptr + g_nbChars, separator, strlen(separator));
+    g_nbChars += strlen(separator);
+}
+
+static int about(unsigned target)
+{
+    return (int)(LOREM_rand(target) + LOREM_rand(target) + 1);
+}
+
+/* Function to generate a random sentence */
+static void generateSentence(int nbWords)
+{
+    int commaPos       = about(9);
+    int comma2         = commaPos + about(7);
+    int qmark          = (LOREM_rand(11) == 7);
+    const char* endSep = qmark ? "? " : ". ";
+    int i;
+    for (i = 0; i < nbWords; i++) {
+        int const wordID       = g_distrib[LOREM_rand(g_distribCount)];
+        const char* const word = kWords[wordID];
+        const char* sep        = " ";
+        if (i == commaPos)
+            sep = ", ";
+        if (i == comma2)
+            sep = ", ";
+        if (i == nbWords - 1)
+            sep = endSep;
+        generateWord(word, sep, i == 0);
+    }
+}
+
+static void generateParagraph(int nbSentences)
+{
+    int i;
+    for (i = 0; i < nbSentences; i++) {
+        int wordsPerSentence = about(11);
+        generateSentence(wordsPerSentence);
+    }
+    if (g_nbChars < g_maxChars) {
+        g_ptr[g_nbChars++] = '\n';
+    }
+    if (g_nbChars < g_maxChars) {
+        g_ptr[g_nbChars++] = '\n';
+    }
+}
+
+/* It's "common" for lorem ipsum generators to start with the same first
+ * pre-defined sentence */
+static void generateFirstSentence(void)
+{
+    int i;
+    for (i = 0; i < 18; i++) {
+        const char* word      = kWords[i];
+        const char* separator = " ";
+        if (i == 4)
+            separator = ", ";
+        if (i == 7)
+            separator = ", ";
+        generateWord(word, separator, i == 0);
+    }
+    generateWord(kWords[18], ". ", 0);
+}
+
+size_t
+LOREM_genBlock(void* buffer, size_t size, unsigned seed, int first, int fill)
+{
+    g_ptr = (char*)buffer;
+    assert(size < INT_MAX);
+    g_maxChars = size;
+    g_nbChars  = 0;
+    g_randRoot = seed;
+    if (g_distribCount == 0) {
+        init_word_distrib(kWords, kNbWords, kWeights, kNbWeights);
+    }
+
+    if (first) {
+        generateFirstSentence();
+    }
+    while (g_nbChars < g_maxChars) {
+        int sentencePerParagraph = about(7);
+        generateParagraph(sentencePerParagraph);
+        if (!fill)
+            break; /* only generate one paragraph in not-fill mode */
+    }
+    g_ptr = NULL;
+    return g_nbChars;
+}
+
+void LOREM_genBuffer(void* buffer, size_t size, unsigned seed)
+{
+    LOREM_genBlock(buffer, size, seed, 1, 1);
+}