diff --git a/Makefile b/Makefile
index 2003da90..cea6cb60 100644
--- a/Makefile
+++ b/Makefile
@@ -32,7 +32,7 @@
 # ################################################################
 
 # Version number
-export VERSION := 0.3.5
+export VERSION := 0.3.6
 
 PRGDIR  = programs
 ZSTDDIR = lib
diff --git a/NEWS b/NEWS
index a5e18c12..0b635d80 100644
--- a/NEWS
+++ b/NEWS
@@ -1,3 +1,6 @@
+v0.3.6
+small blocks params
+
 v0.3.5
 minor generic compression improvements
 
diff --git a/images/CSpeed.png b/images/CSpeed.png
index 338c27b2..97497101 100755
Binary files a/images/CSpeed.png and b/images/CSpeed.png differ
diff --git a/lib/zstd.h b/lib/zstd.h
index b13400ec..81437885 100644
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -48,7 +48,7 @@ extern "C" {
 ***************************************/
 #define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
 #define ZSTD_VERSION_MINOR    3    /* for new (non-breaking) interface capabilities */
-#define ZSTD_VERSION_RELEASE  5    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_RELEASE  6    /* for tweaks, bug-fixes, or development */
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
 unsigned ZSTD_versionNumber (void);
 
diff --git a/lib/zstd_internal.h b/lib/zstd_internal.h
index dffa1fe6..d606e23d 100644
--- a/lib/zstd_internal.h
+++ b/lib/zstd_internal.h
@@ -226,6 +226,14 @@ static const U32 g_searchStrength = 8;
 */
 MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals, size_t offsetCode, size_t matchCode)
 {
+#if 0
+    static const BYTE* g_start = NULL;
+    if (g_start==NULL) g_start = literals;
+    if (literals - g_start == 8695)
+    printf("pos %6u : %3u literals & match %3u bytes at distance %6u \n",
+           (U32)(literals - g_start), (U32)litLength, (U32)matchCode+4, (U32)offsetCode);
+#endif
+
     /* copy Literals */
     ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
     seqStorePtr->lit += litLength;
diff --git a/lib/zstdhc.c b/lib/zstdhc.c
index cfc098ac..fdf6dc95 100644
--- a/lib/zstdhc.c
+++ b/lib/zstdhc.c
@@ -111,7 +111,7 @@ size_t ZSTD_HC_freeCCtx(ZSTD_HC_CCtx* cctx)
 /** ZSTD_HC_validateParams
     correct params value to remain within authorized range
     optimize for srcSize if srcSize > 0 */
-void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, size_t srcSize)
+void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, U64 srcSizeHint)
 {
     const U32 btPlus = (params->strategy == ZSTD_HC_btlazy2);
 
@@ -120,9 +120,9 @@ void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, size_t srcSize)
     if (params->windowLog   < ZSTD_HC_WINDOWLOG_MIN) params->windowLog = ZSTD_HC_WINDOWLOG_MIN;
 
     /* correct params, to use less memory */
-    if (srcSize > 0)
+    if ((srcSizeHint > 0) && (srcSizeHint < (1<<ZSTD_HC_WINDOWLOG_MAX)))
     {
-        U32 srcLog = ZSTD_highbit((U32)srcSize-1) + 1;
+        U32 srcLog = ZSTD_highbit((U32)srcSizeHint-1) + 1;
         if (params->windowLog > srcLog) params->windowLog = srcLog;
     }
 
@@ -139,9 +139,10 @@ void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, size_t srcSize)
 
 
 static size_t ZSTD_HC_resetCCtx_advanced (ZSTD_HC_CCtx* zc,
-                                          ZSTD_HC_parameters params)
+                                          ZSTD_HC_parameters params,
+                                          U64 srcSizeHint)
 {
-    ZSTD_HC_validateParams(&params, 0);
+    ZSTD_HC_validateParams(&params, srcSizeHint);
 
     /* reserve table memory */
     {
@@ -930,7 +931,7 @@ size_t ZSTD_HC_compressContinue (ZSTD_HC_CCtx* ctxPtr,
     if (ip != ctxPtr->end)
     {
         if (ctxPtr->end != NULL)
-            ZSTD_HC_resetCCtx_advanced(ctxPtr, ctxPtr->params);
+            ZSTD_HC_resetCCtx_advanced(ctxPtr, ctxPtr->params, srcSize);
         ctxPtr->base = ip;
     }
 
@@ -941,22 +942,24 @@ size_t ZSTD_HC_compressContinue (ZSTD_HC_CCtx* ctxPtr,
 
 size_t ZSTD_HC_compressBegin_advanced(ZSTD_HC_CCtx* ctx,
                                       void* dst, size_t maxDstSize,
-                                      const ZSTD_HC_parameters params)
+                                      const ZSTD_HC_parameters params,
+                                      U64 srcSizeHint)
 {
     size_t errorCode;
     if (maxDstSize < 4) return ERROR(dstSize_tooSmall);
-    errorCode = ZSTD_HC_resetCCtx_advanced(ctx, params);
+    errorCode = ZSTD_HC_resetCCtx_advanced(ctx, params, srcSizeHint);
     if (ZSTD_isError(errorCode)) return errorCode;
     MEM_writeLE32(dst, ZSTD_magicNumber); /* Write Header */
     return 4;
 }
 
 
-size_t ZSTD_HC_compressBegin(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, int compressionLevel)
+size_t ZSTD_HC_compressBegin(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, int compressionLevel, U64 srcSizeHint)
 {
+    int tableID = ((srcSizeHint-1) > 128 KB);   /* intentional underflow for 0 */
     if (compressionLevel<=0) compressionLevel = 1;
     if (compressionLevel > ZSTD_HC_MAX_CLEVEL) compressionLevel = ZSTD_HC_MAX_CLEVEL;
-    return ZSTD_HC_compressBegin_advanced(ctx, dst, maxDstSize, ZSTD_HC_defaultParameters[compressionLevel]);
+    return ZSTD_HC_compressBegin_advanced(ctx, dst, maxDstSize, ZSTD_HC_defaultParameters[tableID][compressionLevel], srcSizeHint);
 }
 
 
@@ -994,7 +997,7 @@ size_t ZSTD_HC_compress_advanced (ZSTD_HC_CCtx* ctx,
     }
 
     /* Header */
-    oSize = ZSTD_HC_compressBegin_advanced(ctx, dst, maxDstSize, params);
+    oSize = ZSTD_HC_compressBegin_advanced(ctx, dst, maxDstSize, params, srcSize);
     if(ZSTD_isError(oSize)) return oSize;
     op += oSize;
     maxDstSize -= oSize;
@@ -1016,9 +1019,10 @@ size_t ZSTD_HC_compress_advanced (ZSTD_HC_CCtx* ctx,
 
 size_t ZSTD_HC_compressCCtx (ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize, int compressionLevel)
 {
+    const int tableID = (srcSize > 128 KB);
     if (compressionLevel<=1) return ZSTD_compress(dst, maxDstSize, src, srcSize);   /* fast mode */
     if (compressionLevel > ZSTD_HC_MAX_CLEVEL) compressionLevel = ZSTD_HC_MAX_CLEVEL;
-    return ZSTD_HC_compress_advanced(ctx, dst, maxDstSize, src, srcSize, ZSTD_HC_defaultParameters[compressionLevel]);
+    return ZSTD_HC_compress_advanced(ctx, dst, maxDstSize, src, srcSize, ZSTD_HC_defaultParameters[tableID][compressionLevel]);
 }
 
 size_t ZSTD_HC_compress(void* dst, size_t maxDstSize, const void* src, size_t srcSize, int compressionLevel)
diff --git a/lib/zstdhc_static.h b/lib/zstdhc_static.h
index f96d4055..a08b00bf 100644
--- a/lib/zstdhc_static.h
+++ b/lib/zstdhc_static.h
@@ -83,14 +83,14 @@ size_t ZSTD_HC_compress_advanced (ZSTD_HC_CCtx* ctx,
 
 /** ZSTD_HC_validateParams
     correct params value to remain within authorized range
-    optimize for srcSize if srcSize > 0 */
-void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, size_t srcSize);
+    srcSizeHint value is optional, select 0 if not known */
+void ZSTD_HC_validateParams(ZSTD_HC_parameters* params, U64 srcSizeHint);
 
 
 /* *************************************
 *  Streaming functions
 ***************************************/
-size_t ZSTD_HC_compressBegin(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, int compressionLevel);
+size_t ZSTD_HC_compressBegin(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, int compressionLevel, U64 srcSizeHint);
 size_t ZSTD_HC_compressContinue(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
 size_t ZSTD_HC_compressEnd(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize);
 
@@ -99,7 +99,32 @@ size_t ZSTD_HC_compressEnd(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstSize);
 *  Pre-defined compression levels
 ***************************************/
 #define ZSTD_HC_MAX_CLEVEL 20
-static const ZSTD_HC_parameters ZSTD_HC_defaultParameters[ZSTD_HC_MAX_CLEVEL+1] = {
+static const ZSTD_HC_parameters ZSTD_HC_defaultParameters[2][ZSTD_HC_MAX_CLEVEL+1] = {
+{   /* for <= 128 KB */
+    /* W,  C,  H,  S,  L, strat */
+    { 17, 12, 12,  1,  4, ZSTD_HC_fast    },  /* level  0 - never used */
+    { 17, 12, 13,  1,  6, ZSTD_HC_fast    },  /* level  1 */
+    { 17, 15, 16,  1,  5, ZSTD_HC_fast    },  /* level  2 */
+    { 17, 16, 17,  1,  5, ZSTD_HC_fast    },  /* level  3 */
+    { 17, 13, 15,  2,  4, ZSTD_HC_greedy  },  /* level  4 */
+    { 17, 15, 17,  3,  4, ZSTD_HC_greedy  },  /* level  5 */
+    { 17, 14, 17,  3,  4, ZSTD_HC_lazy    },  /* level  6 */
+    { 17, 16, 17,  4,  4, ZSTD_HC_lazy    },  /* level  7 */
+    { 17, 16, 17,  4,  4, ZSTD_HC_lazy2   },  /* level  8 */
+    { 17, 17, 16,  5,  4, ZSTD_HC_lazy2   },  /* level  9 */
+    { 17, 17, 16,  6,  4, ZSTD_HC_lazy2   },  /* level 10 */
+    { 17, 17, 16,  7,  4, ZSTD_HC_lazy2   },  /* level 11 */
+    { 17, 17, 16,  8,  4, ZSTD_HC_lazy2   },  /* level 12 */
+    { 17, 18, 16,  4,  4, ZSTD_HC_btlazy2 },  /* level 13 */
+    { 17, 18, 16,  5,  4, ZSTD_HC_btlazy2 },  /* level 14 */
+    { 17, 18, 16,  6,  4, ZSTD_HC_btlazy2 },  /* level 15 */
+    { 17, 18, 16,  7,  4, ZSTD_HC_btlazy2 },  /* level 16 */
+    { 17, 18, 16,  8,  4, ZSTD_HC_btlazy2 },  /* level 17 */
+    { 17, 18, 16,  9,  4, ZSTD_HC_btlazy2 },  /* level 18 */
+    { 17, 18, 16, 10,  4, ZSTD_HC_btlazy2 },  /* level 19 */
+    { 17, 18, 18, 12,  4, ZSTD_HC_btlazy2 },  /* level 20 */
+},
+{   /* for > 128 KB */
     /* W,  C,  H,  S,  L, strat */
     { 18, 12, 12,  1,  4, ZSTD_HC_fast    },  /* level  0 - never used */
     { 18, 14, 14,  1,  7, ZSTD_HC_fast    },  /* level  1 - in fact redirected towards zstd fast */
@@ -122,6 +147,7 @@ static const ZSTD_HC_parameters ZSTD_HC_defaultParameters[ZSTD_HC_MAX_CLEVEL+1]
     { 25, 24, 23,  5,  5, ZSTD_HC_btlazy2 },  /* level 18 */
     { 25, 26, 23,  5,  5, ZSTD_HC_btlazy2 },  /* level 19 */
     { 26, 27, 24,  6,  5, ZSTD_HC_btlazy2 },  /* level 20 */
+}
 };
 
 
diff --git a/programs/Makefile b/programs/Makefile
index ea51bbf7..472222c4 100644
--- a/programs/Makefile
+++ b/programs/Makefile
@@ -30,7 +30,7 @@
 # fullbench32: Same as fullbench, but forced to compile in 32-bits mode
 # ##########################################################################
 
-VERSION?= 0.3.5
+VERSION?= 0.3.6
 
 DESTDIR?=
 PREFIX ?= /usr/local
diff --git a/programs/bench.c b/programs/bench.c
index 0009e4ae..5d10f9fb 100644
--- a/programs/bench.c
+++ b/programs/bench.c
@@ -380,7 +380,7 @@ static int BMK_benchMem(void* srcBuffer, size_t srcSize, const char* fileName, i
 }
 
 
-static U64 BMK_GetFileSize(char* infilename)
+static U64 BMK_GetFileSize(const char* infilename)
 {
     int r;
 #if defined(_MSC_VER)
diff --git a/programs/fileio.c b/programs/fileio.c
index d168f8a4..4ef808fa 100644
--- a/programs/fileio.c
+++ b/programs/fileio.c
@@ -36,7 +36,7 @@
 /**LEGACY_SUPPORT :
 *  decompressor can decode older formats (starting from Zstd 0.1+) */
 #  define ZSTD_LEGACY_SUPPORT 1
-#endif // ZSTD_LEGACY_SUPPORT
+#endif
 
 
 /* *************************************
@@ -58,11 +58,13 @@
 /* *************************************
 *  Includes
 ***************************************/
-#include <stdio.h>     /* fprintf, fopen, fread, _fileno, stdin, stdout */
-#include <stdlib.h>    /* malloc, free */
-#include <string.h>    /* strcmp, strlen */
-#include <time.h>      /* clock */
-#include <errno.h>     /* errno */
+#include <stdio.h>      /* fprintf, fopen, fread, _fileno, stdin, stdout */
+#include <stdlib.h>     /* malloc, free */
+#include <string.h>     /* strcmp, strlen */
+#include <time.h>       /* clock */
+#include <errno.h>      /* errno */
+#include <sys/types.h>  /* stat64 */
+#include <sys/stat.h>   /* stat64 */
 #include "mem.h"
 #include "fileio.h"
 #include "zstd_static.h"
@@ -91,6 +93,10 @@
 #  define IS_CONSOLE(stdStream) isatty(fileno(stdStream))
 #endif
 
+#if !defined(S_ISREG)
+#  define S_ISREG(x) (((x) & S_IFMT) == S_IFREG)
+#endif
+
 
 /* *************************************
 *  Constants
@@ -215,19 +221,35 @@ static void FIO_getFileHandles(FILE** pfinput, FILE** pfoutput, const char* inpu
     if ( *pfoutput==0) EXM_THROW(13, "Pb opening dst : %s", output_filename);
 }
 
+
+static U64 FIO_getFileSize(const char* infilename)
+{
+    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;
+    return (U64)statbuf.st_size;
+}
+
+
 typedef void* (*FIO_createC) (void);
 static void* local_ZSTD_createCCtx(void) { return (void*) ZSTD_createCCtx(); }
 static void* local_ZSTD_HC_createCCtx(void) { return (void*) ZSTD_HC_createCCtx(); }
 
-typedef size_t (*FIO_initC) (void* ctx, void* dst, size_t maxDstSize, int cLevel);
-static size_t local_ZSTD_compressBegin (void* ctx, void* dst, size_t maxDstSize, int cLevel)
+typedef size_t (*FIO_initC) (void* ctx, void* dst, size_t maxDstSize, int cLevel, U64 srcSizeHint);
+static size_t local_ZSTD_compressBegin (void* ctx, void* dst, size_t maxDstSize, int cLevel, U64 srcSizeHint)
 {
-    (void)cLevel;
+    (void)cLevel; (void)srcSizeHint;
     return ZSTD_compressBegin((ZSTD_CCtx*)ctx, dst, maxDstSize);
 }
-static size_t local_ZSTD_HC_compressBegin (void* ctx, void* dst, size_t maxDstSize, int cLevel)
+static size_t local_ZSTD_HC_compressBegin (void* ctx, void* dst, size_t maxDstSize, int cLevel, U64 srcSizeHint)
 {
-    return ZSTD_HC_compressBegin((ZSTD_HC_CCtx*)ctx, dst, maxDstSize, cLevel);
+    return ZSTD_HC_compressBegin((ZSTD_HC_CCtx*)ctx, dst, maxDstSize, cLevel, srcSizeHint);
 }
 
 typedef size_t (*FIO_continueC) (void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
@@ -294,6 +316,7 @@ unsigned long long FIO_compressFilename(const char* output_filename, const char*
         freeC = local_ZSTD_HC_freeCCtx;
     }
     FIO_getFileHandles(&finput, &foutput, input_filename, output_filename);
+    filesize = FIO_getFileSize(input_filename);
 
     /* Allocate Memory */
     ctx = createC();
@@ -304,12 +327,13 @@ unsigned long long FIO_compressFilename(const char* output_filename, const char*
     inEnd = inBuff + inBuffSize;
 
     /* Write Frame Header */
-    cSize = initC(ctx, outBuff, outBuffSize, cLevel);
+    cSize = initC(ctx, outBuff, outBuffSize, cLevel, filesize);
     if (ZSTD_isError(cSize)) EXM_THROW(22, "Compression error : cannot create frame header");
 
     sizeCheck = fwrite(outBuff, 1, cSize, foutput);
     if (sizeCheck!=cSize) EXM_THROW(23, "Write error : cannot write header into %s", output_filename);
     compressedfilesize += cSize;
+    filesize = 0;
 
     /* Main compression loop */
     while (1)
diff --git a/programs/fuzzer.c b/programs/fuzzer.c
index ba402e71..7458bba8 100644
--- a/programs/fuzzer.c
+++ b/programs/fuzzer.c
@@ -355,7 +355,7 @@ int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, double compressibilit
         dSize = ZSTD_decompress(dstBuffer, sampleSize + dSupSize, cBuffer, cSize);
         CHECK(dSize != sampleSize, "ZSTD_decompress failed (%s) (srcSize : %u ; cSize : %u)", ZSTD_getErrorName(dSize), (U32)sampleSize, (U32)cSize);
         crcDest = XXH64(dstBuffer, sampleSize, 0);
-        CHECK(crcOrig != crcDest, "dstBuffer corrupted (pos %u / %u)", (U32)findDiff(srcBuffer+sampleStart, dstBuffer, sampleSize), (U32)sampleSize);
+        CHECK(crcOrig != crcDest, "decompression result corrupted (pos %u / %u)", (U32)findDiff(srcBuffer+sampleStart, dstBuffer, sampleSize), (U32)sampleSize);
 
         /* truncated src decompression test */
         {
diff --git a/programs/paramgrill.c b/programs/paramgrill.c
index eb265e22..90cfdd77 100644
--- a/programs/paramgrill.c
+++ b/programs/paramgrill.c
@@ -122,7 +122,7 @@ static U32 g_rand = 1;
 static U32 g_singleRun = 0;
 static U32 g_target = 0;
 static U32 g_noSeed = 0;
-static const ZSTD_HC_parameters* g_seedParams = ZSTD_HC_defaultParameters;
+static const ZSTD_HC_parameters* g_seedParams = ZSTD_HC_defaultParameters[0];
 static ZSTD_HC_parameters g_params = { 0, 0, 0, 0, 0, ZSTD_HC_greedy };
 
 void BMK_SetNbIterations(int nbLoops)
@@ -703,12 +703,13 @@ static void BMK_benchMem(void* srcBuffer, size_t srcSize)
     int i;
     const char* rfName = "grillResults.txt";
     FILE* f;
-    const U32 srcLog = BMK_highbit((U32)( (g_blockSize ? g_blockSize : srcSize) -1))+1;
+    const size_t blockSize = g_blockSize ? g_blockSize : srcSize;
+    const U32 srcLog = BMK_highbit((U32)(blockSize-1))+1;
 
     if (g_singleRun)
     {
         BMK_result_t testResult;
-        ZSTD_HC_validateParams(&g_params, g_blockSize ? g_blockSize : srcSize);
+        ZSTD_HC_validateParams(&g_params, blockSize);
         BMK_benchParam(&testResult, srcBuffer, srcSize, ctx, g_params);
         DISPLAY("\n");
         return;
@@ -725,12 +726,13 @@ static void BMK_benchMem(void* srcBuffer, size_t srcSize)
     {
         /* baseline config for level 1 */
         BMK_result_t testResult;
-        params.windowLog = MIN(srcLog, 18);
+        params.windowLog = 18;
         params.hashLog = 14;
         params.contentLog = 1;
         params.searchLog = 1;
         params.searchLength = 7;
         params.strategy = ZSTD_HC_fast;
+        ZSTD_HC_validateParams(&params, blockSize);
         BMK_benchParam(&testResult, srcBuffer, srcSize, ctx, params);
         g_cSpeedTarget[1] = (testResult.cSpeed * 15) >> 4;
     }
@@ -741,7 +743,9 @@ static void BMK_benchMem(void* srcBuffer, size_t srcSize)
 
     /* populate initial solution */
     {
+        const int tableID = (blockSize > 128 KB);
         const int maxSeeds = g_noSeed ? 1 : ZSTD_HC_MAX_CLEVEL;
+        g_seedParams = ZSTD_HC_defaultParameters[tableID];
         for (i=1; i<=maxSeeds; i++)
         {
             const U32 btPlus = (params.strategy == ZSTD_HC_btlazy2);
diff --git a/programs/zstdcli.c b/programs/zstdcli.c
index c9946c4d..65b23550 100644
--- a/programs/zstdcli.c
+++ b/programs/zstdcli.c
@@ -70,7 +70,7 @@
 **************************************/
 #define COMPRESSOR_NAME "zstd command line interface"
 #ifndef ZSTD_VERSION
-#  define ZSTD_VERSION "v0.3.5"
+#  define ZSTD_VERSION "v0.3.6"
 #endif
 #define AUTHOR "Yann Collet"
 #define WELCOME_MESSAGE "*** %s %i-bits %s, by %s (%s) ***\n", COMPRESSOR_NAME, (int)(sizeof(void*)*8), ZSTD_VERSION, AUTHOR, __DATE__