bench.c based on zstd

1 files changed, 400 insertions, 312 deletions

diff --git a/programs/bench.c b/programs/bench.c
index bec93f9..1840bb0 100644
--- a/programs/bench.c
+++ b/programs/bench.c
@@ -23,36 +23,18 @@
     - LZ4 source repository : https://github.com/lz4/lz4
 */
 
-/*-************************************
-*  Compiler Options
-***************************************/
-#if defined(_MSC_VER) || defined(_WIN32)
-#  define _CRT_SECURE_NO_WARNINGS
-#  define _CRT_SECURE_NO_DEPRECATE     /* VS2005 */
-#endif
-
-/* Unix Large Files support (>4GB) */
-#define _FILE_OFFSET_BITS 64
-#if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
-#  define _LARGEFILE_SOURCE
-#elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
-#  define _LARGEFILE64_SOURCE
-#endif
-
-#if defined(__MINGW32__) && !defined(_POSIX_SOURCE)
-#  define _POSIX_SOURCE 1          /* disable %llu warnings with MinGW on Windows */
-#endif
 
-
-/*-************************************
+/* *************************************
 *  Includes
 ***************************************/
-#include <stdlib.h>     /* malloc */
-#include <stdio.h>      /* fprintf, fopen */
-#include <sys/types.h>  /* stat64 */
-#include <sys/stat.h>   /* stat64 */
-#include <time.h>       /* clock_t, clock, CLOCKS_PER_SEC */
-#include <string.h>     /* strlen */
+#include "util.h"        /* Compiler options, UTIL_GetFileSize, UTIL_sleep */
+#include <stdlib.h>      /* malloc, free */
+#include <string.h>      /* memset */
+#include <stdio.h>       /* fprintf, fopen, ftello64 */
+#include <time.h>        /* clock_t, clock, CLOCKS_PER_SEC */
+
+#include "datagen.h"     /* RDG_genBuffer */
+#include "xxhash.h"
 
 
 #include "lz4.h"
@@ -61,348 +43,454 @@ static int LZ4_compress_local(const char* src, char* dst, int srcSize, int dstSi
 #include "lz4hc.h"
 #define COMPRESSOR1 LZ4_compress_HC
 #define DEFAULTCOMPRESSOR COMPRESSOR0
-
-#include "xxhash.h"
+#define LZ4_isError(errcode) (errcode==0) 
 
 
-/*-************************************
-*  Compiler specifics
-***************************************/
-#if !defined(S_ISREG)
-#  define S_ISREG(x) (((x) & S_IFMT) == S_IFREG)
-#endif
-
-
-/*-************************************
-*  Basic Types
+/* *************************************
+*  Constants
 ***************************************/
-#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-# include <stdint.h>
-  typedef uint8_t  BYTE;
-  typedef uint16_t U16;
-  typedef uint32_t U32;
-  typedef  int32_t S32;
-  typedef uint64_t U64;
+#ifndef LZ4_GIT_COMMIT_STRING
+#  define LZ4_GIT_COMMIT_STRING ""
 #else
-  typedef unsigned char       BYTE;
-  typedef unsigned short      U16;
-  typedef unsigned int        U32;
-  typedef   signed int        S32;
-  typedef unsigned long long  U64;
+#  define LZ4_GIT_COMMIT_STRING LZ4_EXPAND_AND_QUOTE(LZ4_GIT_COMMIT)
 #endif
 
-
-/*-************************************
-*  Constants
-***************************************/
-#define NBLOOPS    3
-#define TIMELOOP_S 1
-#define TIMELOOP_CLOCK (TIMELOOP_S * CLOCKS_PER_SEC)
+#define NBSECONDS             3
+#define TIMELOOP_MICROSEC     1*1000000ULL /* 1 second */
+#define ACTIVEPERIOD_MICROSEC 70*1000000ULL /* 70 seconds */
+#define COOLPERIOD_SEC        10
 
 #define KB *(1 <<10)
 #define MB *(1 <<20)
 #define GB *(1U<<30)
 
-#define MAX_MEM             (2 GB - 64 MB)
-#define DEFAULT_CHUNKSIZE   (4 MB)
+static const size_t maxMemory = (sizeof(size_t)==4)  ?  (2 GB - 64 MB) : (size_t)(1ULL << ((sizeof(size_t)*8)-31));
+
+static U32 g_compressibilityDefault = 50;
 
 
-/*-************************************
-*  Local structures
+/* *************************************
+*  console display
 ***************************************/
-struct chunkParameters
-{
-    U32   id;
-    char* origBuffer;
-    char* compressedBuffer;
-    int   origSize;
-    int   compressedSize;
-};
+#define DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
+#define DISPLAYLEVEL(l, ...) if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); }
+static U32 g_displayLevel = 2;   /* 0 : no display;   1: errors;   2 : + result + interaction + warnings;   3 : + progression;   4 : + information */
 
-struct compressionParameters
-{
-    int (*compressionFunction)(const char* src, char* dst, int srcSize, int dstSize, int cLevel);
-    int (*decompressionFunction)(const char* src, char* dst, int dstSize);
-};
+#define DISPLAYUPDATE(l, ...) if (g_displayLevel>=l) { \
+            if ((clock() - g_time > refreshRate) || (g_displayLevel>=4)) \
+            { g_time = clock(); DISPLAY(__VA_ARGS__); \
+            if (g_displayLevel>=4) fflush(stdout); } }
+static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
 
 
-/*-************************************
-*  Macro
+/* *************************************
+*  Exceptions
 ***************************************/
-#define DISPLAY(...) fprintf(stderr, __VA_ARGS__)
+#ifndef DEBUG
+#  define DEBUG 0
+#endif
+#define DEBUGOUTPUT(...) if (DEBUG) DISPLAY(__VA_ARGS__);
+#define EXM_THROW(error, ...)                                             \
+{                                                                         \
+    DEBUGOUTPUT("Error defined at %s, line %i : \n", __FILE__, __LINE__); \
+    DISPLAYLEVEL(1, "Error %i : ", error);                                \
+    DISPLAYLEVEL(1, __VA_ARGS__);                                         \
+    DISPLAYLEVEL(1, "\n");                                                \
+    exit(error);                                                          \
+}
 
 
-/*-************************************
+/* *************************************
 *  Benchmark Parameters
 ***************************************/
-static int g_chunkSize = DEFAULT_CHUNKSIZE;
-static int g_nbIterations = NBLOOPS;
-static int BMK_pause = 0;
+static U32 g_nbSeconds = NBSECONDS;
+static size_t g_blockSize = 0;
+int g_additionalParam = 0;
 
-void BMK_setBlocksize(int bsize) { g_chunkSize = bsize; }
+void BMK_setNotificationLevel(unsigned level) { g_displayLevel=level; }
 
-void BMK_setNbIterations(int nbLoops)
+void BMK_setAdditionalParam(int additionalParam) { g_additionalParam=additionalParam; }
+
+void BMK_SetNbSeconds(unsigned nbSeconds)
 {
-    g_nbIterations = nbLoops;
-    DISPLAY("- %i iterations -\n", g_nbIterations);
+    g_nbSeconds = nbSeconds;
+    DISPLAYLEVEL(3, "- test >= %u seconds per compression / decompression -\n", g_nbSeconds);
 }
 
-void BMK_setPause(void) { BMK_pause = 1; }
+void BMK_SetBlockSize(size_t blockSize)
+{
+    g_blockSize = blockSize;
+    DISPLAYLEVEL(2, "using blocks of size %u KB \n", (U32)(blockSize>>10));
+}
 
 
-/*-*******************************************************
-*  Private functions
+/* ********************************************************
+*  Bench functions
 **********************************************************/
+typedef struct
+{
+    const char* srcPtr;
+    size_t srcSize;
+    char*  cPtr;
+    size_t cRoom;
+    size_t cSize;
+    char*  resPtr;
+    size_t resSize;
+} blockParam_t;
+
+struct compressionParameters
+{
+    int (*compressionFunction)(const char* src, char* dst, int srcSize, int dstSize, int cLevel);
+};
+
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
 
-/** BMK_getClockSpan() :
-    works even if overflow; Typical max span ~ 30 mn */
-static clock_t BMK_getClockSpan (clock_t clockStart)
+static int BMK_benchMem(const void* srcBuffer, size_t srcSize,
+                        const char* displayName, int cLevel,
+                        const size_t* fileSizes, U32 nbFiles)
 {
-    return clock() - clockStart;
+    size_t const blockSize = (g_blockSize>=32 ? g_blockSize : srcSize) + (!srcSize) /* avoid div by 0 */ ;
+    U32 const maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles;
+    blockParam_t* const blockTable = (blockParam_t*) malloc(maxNbBlocks * sizeof(blockParam_t));
+    size_t const maxCompressedSize = LZ4_compressBound((int)srcSize) + (maxNbBlocks * 1024);   /* add some room for safety */
+    void* const compressedBuffer = malloc(maxCompressedSize);
+    void* const resultBuffer = malloc(srcSize);
+    U32 nbBlocks;
+    UTIL_time_t ticksPerSecond;
+    struct compressionParameters compP;
+    int cfunctionId;
+
+    /* checks */
+    if (!compressedBuffer || !resultBuffer || !blockTable)
+        EXM_THROW(31, "allocation error : not enough memory");
+
+    /* init */
+    if (strlen(displayName)>17) displayName += strlen(displayName)-17;   /* can only display 17 characters */
+    UTIL_initTimer(&ticksPerSecond);
+
+    /* Init */
+    if (cLevel < LZ4HC_MIN_CLEVEL) cfunctionId = 0; else cfunctionId = 1;
+    switch (cfunctionId)
+    {
+#ifdef COMPRESSOR0
+    case 0 : compP.compressionFunction = COMPRESSOR0; break;
+#endif
+#ifdef COMPRESSOR1
+    case 1 : compP.compressionFunction = COMPRESSOR1; break;
+#endif
+    default : compP.compressionFunction = DEFAULTCOMPRESSOR;
+    }
+
+    /* Init blockTable data */
+    {   const char* srcPtr = (const char*)srcBuffer;
+        char* cPtr = (char*)compressedBuffer;
+        char* resPtr = (char*)resultBuffer;
+        U32 fileNb;
+        for (nbBlocks=0, fileNb=0; fileNb<nbFiles; fileNb++) {
+            size_t remaining = fileSizes[fileNb];
+            U32 const nbBlocksforThisFile = (U32)((remaining + (blockSize-1)) / blockSize);
+            U32 const blockEnd = nbBlocks + nbBlocksforThisFile;
+            for ( ; nbBlocks<blockEnd; nbBlocks++) {
+                size_t const thisBlockSize = MIN(remaining, blockSize);
+                blockTable[nbBlocks].srcPtr = srcPtr;
+                blockTable[nbBlocks].cPtr = cPtr;
+                blockTable[nbBlocks].resPtr = resPtr;
+                blockTable[nbBlocks].srcSize = thisBlockSize;
+                blockTable[nbBlocks].cRoom = LZ4_compressBound((int)thisBlockSize);
+                srcPtr += thisBlockSize;
+                cPtr += blockTable[nbBlocks].cRoom;
+                resPtr += thisBlockSize;
+                remaining -= thisBlockSize;
+    }   }   }
+
+    /* warmimg up memory */
+    RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.50, 1);
+
+    /* Bench */
+    {   U64 fastestC = (U64)(-1LL), fastestD = (U64)(-1LL);
+        U64 const crcOrig = XXH64(srcBuffer, srcSize, 0);
+        UTIL_time_t coolTime;
+        U64 const maxTime = (g_nbSeconds * TIMELOOP_MICROSEC) + 100;
+        U64 totalCTime=0, totalDTime=0;
+        U32 cCompleted=0, dCompleted=0;
+#       define NB_MARKS 4
+        const char* const marks[NB_MARKS] = { " |", " /", " =",  "\\" };
+        U32 markNb = 0;
+        size_t cSize = 0;
+        double ratio = 0.;
+
+        UTIL_getTime(&coolTime);
+        DISPLAYLEVEL(2, "\r%79s\r", "");
+        while (!cCompleted | !dCompleted) {
+            UTIL_time_t clockStart;
+            U64 clockLoop = g_nbSeconds ? TIMELOOP_MICROSEC : 1;
+
+            /* overheat protection */
+            if (UTIL_clockSpanMicro(coolTime, ticksPerSecond) > ACTIVEPERIOD_MICROSEC) {
+                DISPLAYLEVEL(2, "\rcooling down ...    \r");
+                UTIL_sleep(COOLPERIOD_SEC);
+                UTIL_getTime(&coolTime);
+            }
+
+            /* Compression */
+            DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->\r", marks[markNb], displayName, (U32)srcSize);
+            if (!cCompleted) memset(compressedBuffer, 0xE5, maxCompressedSize);  /* warm up and erase result buffer */
+
+            UTIL_sleepMilli(1);  /* give processor time to other processes */
+            UTIL_waitForNextTick(ticksPerSecond);
+            UTIL_getTime(&clockStart);
+
+            if (!cCompleted) {   /* still some time to do compression tests */
+                U32 nbLoops = 0;
+                do {
+                    U32 blockNb;
+                    for (blockNb=0; blockNb<nbBlocks; blockNb++) {
+                        size_t const rSize = compP.compressionFunction(blockTable[blockNb].srcPtr, blockTable[blockNb].cPtr, (int)blockTable[blockNb].srcSize, (int)blockTable[blockNb].cRoom, cLevel);
+                        if (LZ4_isError(rSize)) EXM_THROW(1, "LZ4_compress() failed");
+                        blockTable[blockNb].cSize = rSize;
+                    }
+                    nbLoops++;
+                } while (UTIL_clockSpanMicro(clockStart, ticksPerSecond) < clockLoop);
+                {   U64 const clockSpan = UTIL_clockSpanMicro(clockStart, ticksPerSecond);
+                    if (clockSpan < fastestC*nbLoops) fastestC = clockSpan / nbLoops;
+                    totalCTime += clockSpan;
+                    cCompleted = totalCTime>maxTime;
+            }   }
+
+            cSize = 0;
+            { U32 blockNb; for (blockNb=0; blockNb<nbBlocks; blockNb++) cSize += blockTable[blockNb].cSize; }
+            ratio = (double)srcSize / (double)cSize;
+            markNb = (markNb+1) % NB_MARKS;
+            DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s\r",
+                    marks[markNb], displayName, (U32)srcSize, (U32)cSize, ratio,
+                    (double)srcSize / fastestC );
+
+            (void)fastestD; (void)crcOrig;   /*  unused when decompression disabled */
+#if 1
+            /* Decompression */
+            if (!dCompleted) memset(resultBuffer, 0xD6, srcSize);  /* warm result buffer */
+
+            UTIL_sleepMilli(1); /* give processor time to other processes */
+            UTIL_waitForNextTick(ticksPerSecond);
+            UTIL_getTime(&clockStart);
+
+            if (!dCompleted) {
+                U32 nbLoops = 0;
+                do {
+                    U32 blockNb;
+                    for (blockNb=0; blockNb<nbBlocks; blockNb++) {
+                        size_t const regenSize = LZ4_decompress_safe(blockTable[blockNb].cPtr, blockTable[blockNb].resPtr, (int)blockTable[blockNb].cSize, (int)blockTable[blockNb].srcSize);
+                        if (LZ4_isError(regenSize)) {
+                            DISPLAY("LZ4_decompress_safe() failed on block %u  \n", blockNb);
+                            clockLoop = 0;   /* force immediate test end */
+                            break;
+                        }
+
+                        blockTable[blockNb].resSize = regenSize;
+                    }
+                    nbLoops++;
+                } while (UTIL_clockSpanMicro(clockStart, ticksPerSecond) < clockLoop);
+                {   U64 const clockSpan = UTIL_clockSpanMicro(clockStart, ticksPerSecond);
+                    if (clockSpan < fastestD*nbLoops) fastestD = clockSpan / nbLoops;
+                    totalDTime += clockSpan;
+                    dCompleted = totalDTime>maxTime;
+            }   }
+
+            markNb = (markNb+1) % NB_MARKS;
+            DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s ,%6.1f MB/s\r",
+                    marks[markNb], displayName, (U32)srcSize, (U32)cSize, ratio,
+                    (double)srcSize / fastestC,
+                    (double)srcSize / fastestD );
+
+            /* CRC Checking */
+            {   U64 const crcCheck = XXH64(resultBuffer, srcSize, 0);
+                if (crcOrig!=crcCheck) {
+                    size_t u;
+                    DISPLAY("!!! WARNING !!! %14s : Invalid Checksum : %x != %x   \n", displayName, (unsigned)crcOrig, (unsigned)crcCheck);
+                    for (u=0; u<srcSize; u++) {
+                        if (((const BYTE*)srcBuffer)[u] != ((const BYTE*)resultBuffer)[u]) {
+                            U32 segNb, bNb, pos;
+                            size_t bacc = 0;
+                            DISPLAY("Decoding error at pos %u ", (U32)u);
+                            for (segNb = 0; segNb < nbBlocks; segNb++) {
+                                if (bacc + blockTable[segNb].srcSize > u) break;
+                                bacc += blockTable[segNb].srcSize;
+                            }
+                            pos = (U32)(u - bacc);
+                            bNb = pos / (128 KB);
+                            DISPLAY("(block %u, sub %u, pos %u) \n", segNb, bNb, pos);
+                            break;
+                        }
+                        if (u==srcSize-1) {  /* should never happen */
+                            DISPLAY("no difference detected\n");
+                    }   }
+                    break;
+            }   }   /* CRC Checking */
+#endif
+        }   /* for (testNb = 1; testNb <= (g_nbSeconds + !g_nbSeconds); testNb++) */
+
+        if (g_displayLevel == 1) {
+            double cSpeed = (double)srcSize / fastestC;
+            double dSpeed = (double)srcSize / fastestD;
+            if (g_additionalParam)
+                DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, g_additionalParam);
+            else
+                DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName);
+        }
+        DISPLAYLEVEL(2, "%2i#\n", cLevel);
+    }   /* Bench */
+
+    /* clean up */
+    free(blockTable);
+    free(compressedBuffer);
+    free(resultBuffer);
+    return 0;
 }
 
+
 static size_t BMK_findMaxMem(U64 requiredMem)
 {
     size_t const step = 64 MB;
-    void* testmem = NULL;
+    BYTE* testmem = NULL;
 
     requiredMem = (((requiredMem >> 26) + 1) << 26);
-    requiredMem += 2*step;
-    if (requiredMem > MAX_MEM) requiredMem = MAX_MEM;
+    requiredMem += step;
+    if (requiredMem > maxMemory) requiredMem = maxMemory;
 
-    while (!testmem) {
-        if (requiredMem > step) requiredMem -= step;
-        else requiredMem >>= 1;
-        testmem = malloc ((size_t)requiredMem);
-    }
-    free (testmem);
+    do {
+        testmem = (BYTE*)malloc((size_t)requiredMem);
+        requiredMem -= step;
+    } while (!testmem);
 
-    /* keep some space available */
-    if (requiredMem > step) requiredMem -= step;
-    else requiredMem >>= 1;
-
-    return (size_t)requiredMem;
+    free(testmem);
+    return (size_t)(requiredMem);
 }
 
-static U64 BMK_GetFileSize(const char* infilename)
+static void BMK_benchCLevel(void* srcBuffer, size_t benchedSize,
+                            const char* displayName, int cLevel, int cLevelLast,
+                            const size_t* fileSizes, unsigned nbFiles)
 {
-    int r;
-#if defined(_MSC_VER)
-    struct _stat64 statbuf;
-    r = _stat64(infilename, &statbuf);
-#else
-    struct stat statbuf;
-    r = stat(infilename, &statbuf);
-#endif
-    if (r || !S_ISREG(statbuf.st_mode)) return 0;   /* No good... */
-    return (U64)statbuf.st_size;
-}
+    int l;
 
+    const char* pch = strrchr(displayName, '\\'); /* Windows */
+    if (!pch) pch = strrchr(displayName, '/'); /* Linux */
+    if (pch) displayName = pch+1;
 
-/*-*******************************************************
-*  Public function
-**********************************************************/
+    SET_HIGH_PRIORITY;
 
-int BMK_benchLevel(const char** fileNamesTable, int nbFiles, int cLevel)
-{
-  int fileIdx=0;
-  char* orig_buff;
-  struct compressionParameters compP;
-  int cfunctionId;
-
-  U64 totals = 0;
-  U64 totalz = 0;
-  double totalc = 0.;
-  double totald = 0.;
-
-  /* Init */
-  if (cLevel < LZ4HC_MIN_CLEVEL) cfunctionId = 0; else cfunctionId = 1;
-  switch (cfunctionId)
-  {
-#ifdef COMPRESSOR0
-  case 0 : compP.compressionFunction = COMPRESSOR0; break;
-#endif
-#ifdef COMPRESSOR1
-  case 1 : compP.compressionFunction = COMPRESSOR1; break;
-#endif
-  default : compP.compressionFunction = DEFAULTCOMPRESSOR;
-  }
-
-  /* Loop for each file */
-  while (fileIdx<nbFiles) {
-      const char* inFileName = fileNamesTable[fileIdx++];
-      FILE* const inFile = fopen( inFileName, "rb" );
-      U64 const inFileSize = BMK_GetFileSize(inFileName);
-      size_t benchedSize = BMK_findMaxMem(inFileSize * 2) / 2;
-      unsigned nbChunks;
-      int maxCompressedChunkSize;
-      size_t readSize;
-      char* compressedBuffer;
-      struct chunkParameters* chunkP;
-      U32 crcOrig;
-
-      /* Check file and memory conditions */
-      if (inFile==NULL) { DISPLAY( "Pb opening %s\n", inFileName); return 11; }
-      if (inFileSize==0) { DISPLAY( "file is empty\n"); fclose(inFile); return 11; }
-      if (benchedSize==0) { DISPLAY( "not enough memory\n"); fclose(inFile); return 11; }
-      if ((U64)benchedSize > inFileSize) benchedSize = (size_t)inFileSize;
-      if (benchedSize < inFileSize) {
-          DISPLAY("Not enough memory for '%s' full size; testing %u MB only...\n", inFileName, (unsigned)(benchedSize>>20));
-      }
-
-      /* Allocation */
-      nbChunks = (unsigned)(benchedSize / g_chunkSize) + 1;
-      chunkP = (struct chunkParameters*) malloc(nbChunks * sizeof(struct chunkParameters));
-      orig_buff = (char*)malloc(benchedSize);
-      maxCompressedChunkSize = LZ4_compressBound(g_chunkSize);
-      { size_t const compressedBuffSize = (size_t)(nbChunks * maxCompressedChunkSize);
-        compressedBuffer = (char*)malloc(compressedBuffSize); }
-
-      if (!orig_buff || !compressedBuffer){
-        DISPLAY("\nError: not enough memory!\n");
-        free(orig_buff);
-        free(compressedBuffer);
-        free(chunkP);
-        fclose(inFile);
-        return 12;
-      }
-
-      /* Init chunks data */
-      {   unsigned i;
-          size_t remaining = benchedSize;
-          char* in = orig_buff;
-          char* out = compressedBuffer;
-          for (i=0; i<nbChunks; i++) {
-              chunkP[i].id = i;
-              chunkP[i].origBuffer = in; in += g_chunkSize;
-              if (remaining > (size_t)g_chunkSize) { chunkP[i].origSize = g_chunkSize; remaining -= g_chunkSize; } else { chunkP[i].origSize = (int)remaining; remaining = 0; }
-              chunkP[i].compressedBuffer = out; out += maxCompressedChunkSize;
-              chunkP[i].compressedSize = 0;
-      }   }
-
-      /* Fill input buffer */
-      DISPLAY("Loading %s...       \r", inFileName);
-      if (strlen(inFileName)>16) inFileName += strlen(inFileName)-16;   /* can only display 16 characters */
-      readSize = fread(orig_buff, 1, benchedSize, inFile);
-      fclose(inFile);
-
-      if (readSize != benchedSize) {
-        DISPLAY("\nError: problem reading file '%s' !! \n", inFileName);
-        free(orig_buff);
-        free(compressedBuffer);
-        free(chunkP);
-        return 13;
-      }
-
-      /* Calculating input Checksum */
-      crcOrig = XXH32(orig_buff, benchedSize,0);
-
-      /* Bench */
-      { int loopNb;
-        size_t cSize = 0;
-        double fastestC = 100000000., fastestD = 100000000.;
-        double ratio = 0.;
-        U32 crcCheck = 0;
-
-        DISPLAY("\r%79s\r", "");
-        for (loopNb = 1; loopNb <= g_nbIterations; loopNb++) {
-          int nbLoops = 0;
-          clock_t clockStart, clockEnd;
-          unsigned chunkNb;
-
-          /* Compression */
-          DISPLAY("%2i#%1i-%-14.14s : %9i ->\r", cLevel, loopNb, inFileName, (int)benchedSize);
-          { size_t i; for (i=0; i<benchedSize; i++) compressedBuffer[i]=(char)i; }     /* warmimg up memory */
-
-          clockStart = clock();
-          while (clock() == clockStart);
-          clockStart = clock();
-          while (BMK_getClockSpan(clockStart) < TIMELOOP_CLOCK) {
-             for (chunkNb=0; chunkNb<nbChunks; chunkNb++)
-                chunkP[chunkNb].compressedSize = compP.compressionFunction(chunkP[chunkNb].origBuffer, chunkP[chunkNb].compressedBuffer, chunkP[chunkNb].origSize, maxCompressedChunkSize, cLevel);
-             nbLoops++;
-          }
-          clockEnd = BMK_getClockSpan(clockStart);
-
-          nbLoops += !nbLoops;   /* avoid division by zero */
-          if ((double)clockEnd < fastestC*nbLoops) fastestC = (double)clockEnd/nbLoops;
-          cSize=0; for (chunkNb=0; chunkNb<nbChunks; chunkNb++) cSize += chunkP[chunkNb].compressedSize;
-          ratio = (double)cSize/(double)benchedSize*100.;
-
-          DISPLAY("%2i#%1i-%-14.14s : %9i -> %9i (%5.2f%%),%7.1f MB/s\r",
-                  cLevel, loopNb, inFileName, (int)benchedSize, (int)cSize, ratio, (double)benchedSize / (fastestC / CLOCKS_PER_SEC) / 1000000);
-
-          /* Decompression */
-          { size_t i; for (i=0; i<benchedSize; i++) orig_buff[i]=0; }     /* zeroing area, for CRC checking */
-
-          nbLoops = 0;
-          clockStart = clock();
-          while (clock() == clockStart);
-          clockStart = clock();
-          while (BMK_getClockSpan(clockStart) < TIMELOOP_CLOCK) {
-             for (chunkNb=0; chunkNb<nbChunks; chunkNb++)
-                //chunkP[chunkNb].compressedSize = LZ4_decompress_fast(chunkP[chunkNb].compressedBuffer, chunkP[chunkNb].origBuffer, chunkP[chunkNb].origSize);
-                LZ4_decompress_safe(chunkP[chunkNb].compressedBuffer, chunkP[chunkNb].origBuffer, chunkP[chunkNb].compressedSize, chunkP[chunkNb].origSize);
-             nbLoops++;
-          }
-          clockEnd = BMK_getClockSpan(clockStart);
-
-          nbLoops += !nbLoops;   /* avoid division by zero */
-          if ((double)clockEnd < fastestD*nbLoops) fastestD = (double)clockEnd/nbLoops;
-          DISPLAY("%2i#%1i-%-14.14s : %9i -> %9i (%5.2f%%),%7.1f MB/s ,%7.1f MB/s \r",
-                  cLevel, loopNb, inFileName, (int)benchedSize, (int)cSize, ratio,
-                  (double)benchedSize / (fastestC / CLOCKS_PER_SEC) / 1000000, (double)benchedSize / (fastestD / CLOCKS_PER_SEC) / 1000000 );
-
-          /* CRC Checking */
-          crcCheck = XXH32(orig_buff, benchedSize,0);
-          if (crcOrig!=crcCheck) { DISPLAY("\n!!! WARNING !!! %14s : Invalid Checksum : %x != %x\n", inFileName, (unsigned)crcOrig, (unsigned)crcCheck); break; }
-        }
+    if (g_displayLevel == 1 && !g_additionalParam)
+        DISPLAY("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n", LZ4_VERSION_STRING, LZ4_GIT_COMMIT_STRING, (U32)benchedSize, g_nbSeconds, (U32)(g_blockSize>>10));
 
-        if (crcOrig==crcCheck) {
-            if (ratio < 100.)
-                DISPLAY("%2i#%-16.16s : %9i -> %9i (%5.2f%%),%7.1f MB/s ,%7.1f MB/s \n",
-                        cLevel, inFileName, (int)benchedSize, (int)cSize, ratio,
-                        (double)benchedSize / (fastestC / CLOCKS_PER_SEC) / 1000000, (double)benchedSize / (fastestD / CLOCKS_PER_SEC) / 1000000 );
-            else
-                DISPLAY("%2i#%-16.16s : %9i -> %9i (%5.1f%%),%7.1f MB/s ,%7.1f MB/s  \n",
-                        cLevel, inFileName, (int)benchedSize, (int)cSize, ratio,
-                        (double)benchedSize / (fastestC / CLOCKS_PER_SEC) / 1000000, (double)benchedSize / (fastestD / CLOCKS_PER_SEC) / 1000000 );
-        }
-        totals += benchedSize;
-        totalz += cSize;
-        totalc += fastestC;
-        totald += fastestD;
-      }
+    if (cLevelLast < cLevel) cLevelLast = cLevel;
 
-      free(orig_buff);
-      free(compressedBuffer);
-      free(chunkP);
-  }
+    for (l=cLevel; l <= cLevelLast; l++) {
+        BMK_benchMem(srcBuffer, benchedSize,
+                     displayName, l,
+                     fileSizes, nbFiles);
+    }
+}
 
-  if (nbFiles > 1)
-        DISPLAY("%2i#%-16.16s :%10llu ->%10llu (%5.2f%%), %6.1f MB/s , %6.1f MB/s\n", cLevel, "  TOTAL",
-                (long long unsigned)totals, (long long unsigned int)totalz, (double)totalz/(double)totals*100.,
-                (double)totals/(totalc/CLOCKS_PER_SEC)/1000000, (double)totals/(totald/CLOCKS_PER_SEC)/1000000);
 
-  if (BMK_pause) { DISPLAY("\npress enter...\n"); (void)getchar(); }
+/*! BMK_loadFiles() :
+    Loads `buffer` with content of files listed within `fileNamesTable`.
+    At most, fills `buffer` entirely */
+static void BMK_loadFiles(void* buffer, size_t bufferSize,
+                          size_t* fileSizes,
+                          const char** fileNamesTable, unsigned nbFiles)
+{
+    size_t pos = 0, totalSize = 0;
+    unsigned n;
+    for (n=0; n<nbFiles; n++) {
+        FILE* f;
+        U64 fileSize = UTIL_getFileSize(fileNamesTable[n]);
+        if (UTIL_isDirectory(fileNamesTable[n])) {
+            DISPLAYLEVEL(2, "Ignoring %s directory...       \n", fileNamesTable[n]);
+            fileSizes[n] = 0;
+            continue;
+        }
+        f = fopen(fileNamesTable[n], "rb");
+        if (f==NULL) EXM_THROW(10, "impossible to open file %s", fileNamesTable[n]);
+        DISPLAYUPDATE(2, "Loading %s...       \r", fileNamesTable[n]);
+        if (fileSize > bufferSize-pos) fileSize = bufferSize-pos, nbFiles=n;   /* buffer too small - stop after this file */
+        { size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f);
+          if (readSize != (size_t)fileSize) EXM_THROW(11, "could not read %s", fileNamesTable[n]);
+          pos += readSize; }
+        fileSizes[n] = (size_t)fileSize;
+        totalSize += (size_t)fileSize;
+        fclose(f);
+    }
 
-  return 0;
+    if (totalSize == 0) EXM_THROW(12, "no data to bench");
 }
 
+static void BMK_benchFileTable(const char** fileNamesTable, unsigned nbFiles,
+                               int cLevel, int cLevelLast)
+{
+    void* srcBuffer;
+    size_t benchedSize;
+    size_t* fileSizes = (size_t*)malloc(nbFiles * sizeof(size_t));
+    U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles);
+    char mfName[20] = {0};
+
+    if (!fileSizes) EXM_THROW(12, "not enough memory for fileSizes");
+
+    /* Memory allocation & restrictions */
+    benchedSize = BMK_findMaxMem(totalSizeToLoad * 3) / 3;
+    if (benchedSize==0) EXM_THROW(12, "not enough memory");
+    if ((U64)benchedSize > totalSizeToLoad) benchedSize = (size_t)totalSizeToLoad;
+    if (benchedSize < totalSizeToLoad)
+        DISPLAY("Not enough memory; testing %u MB only...\n", (U32)(benchedSize >> 20));
+    srcBuffer = malloc(benchedSize);
+    if (!srcBuffer) EXM_THROW(12, "not enough memory");
+
+    /* Load input buffer */
+    BMK_loadFiles(srcBuffer, benchedSize, fileSizes, fileNamesTable, nbFiles);
+
+    /* Bench */
+    snprintf (mfName, sizeof(mfName), " %u files", nbFiles);
+    {   const char* displayName = (nbFiles > 1) ? mfName : fileNamesTable[0];
+        BMK_benchCLevel(srcBuffer, benchedSize,
+                        displayName, cLevel, cLevelLast,
+                        fileSizes, nbFiles);
+    }
 
-int BMK_benchFiles(const char** fileNamesTable, int nbFiles, int cLevel, int cLevelLast)
+    /* clean up */
+    free(srcBuffer);
+    free(fileSizes);
+}
+
+
+static void BMK_syntheticTest(int cLevel, int cLevelLast, double compressibility)
 {
-    int i, res = 0;
+    char name[20] = {0};
+    size_t benchedSize = 10000000;
+    void* const srcBuffer = malloc(benchedSize);
 
-    if (cLevel > LZ4HC_MAX_CLEVEL) cLevel = LZ4HC_MAX_CLEVEL;
-    if (cLevelLast > LZ4HC_MAX_CLEVEL) cLevelLast = LZ4HC_MAX_CLEVEL;
-    if (cLevelLast < cLevel) cLevelLast = cLevel;
+    /* Memory allocation */
+    if (!srcBuffer) EXM_THROW(21, "not enough memory");
 
-    if (cLevelLast > cLevel) DISPLAY("Benchmarking levels from %d to %d\n", cLevel, cLevelLast);
-    for (i=cLevel; i<=cLevelLast; i++) {
-        res = BMK_benchLevel(fileNamesTable, nbFiles, i);
-        if (res != 0) break;
-    }
+    /* Fill input buffer */
+    RDG_genBuffer(srcBuffer, benchedSize, compressibility, 0.0, 0);
+
+    /* Bench */
+    snprintf (name, sizeof(name), "Synthetic %2u%%", (unsigned)(compressibility*100));
+    BMK_benchCLevel(srcBuffer, benchedSize, name, cLevel, cLevelLast, &benchedSize, 1);
+
+    /* clean up */
+    free(srcBuffer);
+}
+
+
+int BMK_benchFiles(const char** fileNamesTable, unsigned nbFiles,
+                   int cLevel, int cLevelLast)
+{
+    double const compressibility = (double)g_compressibilityDefault / 100;
 
-    return res;
+    if (nbFiles == 0)
+        BMK_syntheticTest(cLevel, cLevelLast, compressibility);
+    else
+        BMK_benchFileTable(fileNamesTable, nbFiles, cLevel, cLevelLast);
+    return 0;
 }