diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
index 784bb918a..a7b23e798 100644
--- a/lib/compress/zstd_opt.c
+++ b/lib/compress/zstd_opt.c
@@ -79,21 +79,6 @@ static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
 }
 
 
-/* ZSTD_downscaleStat() :
- * reduce all elements in table by a factor 2^(ZSTD_FREQ_DIV+malus)
- * return the resulting sum of elements */
-static U32 ZSTD_downscaleStat(unsigned* table, U32 lastEltIndex, int malus)
-{
-    U32 s, sum=0;
-    DEBUGLOG(5, "ZSTD_downscaleStat (nbElts=%u)", (unsigned)lastEltIndex+1);
-    assert(ZSTD_FREQ_DIV+malus > 0 && ZSTD_FREQ_DIV+malus < 31);
-    for (s=0; s<lastEltIndex+1; s++) {
-        table[s] = 1 + (table[s] >> (ZSTD_FREQ_DIV+malus));
-        sum += table[s];
-    }
-    return sum;
-}
-
 static U32 sum_u32(const unsigned table[], size_t nbElts)
 {
     size_t n;
@@ -104,6 +89,31 @@ static U32 sum_u32(const unsigned table[], size_t nbElts)
     return total;
 }
 
+static U32 ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift)
+{
+    U32 s, sum=0;
+    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)", (unsigned)lastEltIndex+1, (unsigned)shift);
+    assert(shift < 30);
+    for (s=0; s<lastEltIndex+1; s++) {
+        table[s] = 1 + (table[s] >> shift);
+        sum += table[s];
+    }
+    return sum;
+}
+
+/* ZSTD_scaleStats() :
+ * reduce all elements in table is sum too large
+ * return the resulting sum of elements */
+static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
+{
+    U32 const prevsum = sum_u32(table, lastEltIndex+1);
+    U32 const factor = prevsum >> logTarget;
+    DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
+    assert(logTarget < 30);
+    if (factor <= 1) return prevsum;
+    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor));
+}
+
 /* ZSTD_rescaleFreqs() :
  * if first block (detected by optPtr->litLengthSum == 0) : init statistics
  *    take hints from dictionary if there is one
@@ -185,7 +195,7 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
             if (compressedLiterals) {
                 unsigned lit = MaxLit;
                 HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
-                optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8);
             }
 
             {   unsigned const baseLLfreqs[MaxLL+1] = {
@@ -219,10 +229,10 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
     } else {   /* new block : re-use previous statistics, scaled down */
 
         if (compressedLiterals)
-            optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
-        optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);
-        optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);
-        optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);
+            optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 10);
+        optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 9);
+        optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 9);
+        optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 9);
     }
 
     ZSTD_setBasePrices(optPtr, optLevel);
diff --git a/programs/benchzstd.c b/programs/benchzstd.c
index 262efb1b1..2366cf2d7 100644
--- a/programs/benchzstd.c
+++ b/programs/benchzstd.c
@@ -70,6 +70,8 @@ static const size_t maxMemory = (sizeof(size_t)==4)  ?
 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush(NULL); }
 #define DISPLAYLEVEL(l, ...) if (displayLevel>=l) { DISPLAY(__VA_ARGS__); }
 /* 0 : no display;   1: errors;   2 : + result + interaction + warnings;   3 : + progression;   4 : + information */
+#define OUTPUT(...)          { fprintf(stdout, __VA_ARGS__); fflush(NULL); }
+#define OUTPUTLEVEL(l, ...)  if (displayLevel>=l) { OUTPUT(__VA_ARGS__); }
 
 
 /* *************************************
@@ -429,9 +431,9 @@ BMK_benchMemAdvancedNoAlloc(
         dctxprep.dictBuffer = dictBuffer;
         dctxprep.dictBufferSize = dictBufferSize;
 
-        DISPLAYLEVEL(2, "\r%70s\r", "");   /* blank line */
+        OUTPUTLEVEL(2, "\r%70s\r", "");   /* blank line */
         assert(srcSize < UINT_MAX);
-        DISPLAYLEVEL(2, "%2s-%-17.17s :%10u -> \r", marks[markNb], displayName, (unsigned)srcSize);
+        OUTPUTLEVEL(2, "%2s-%-17.17s :%10u -> \r", marks[markNb], displayName, (unsigned)srcSize);
 
         while (!(compressionCompleted && decompressionCompleted)) {
             if (!compressionCompleted) {
@@ -453,7 +455,7 @@ BMK_benchMemAdvancedNoAlloc(
 
                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
                     assert(cSize < UINT_MAX);
-                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f), %6.*f MB/s\r",
+                    OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f), %6.*f MB/s\r",
                             marks[markNb], displayName,
                             (unsigned)srcSize, (unsigned)cSize,
                             ratioAccuracy, ratio,
@@ -476,7 +478,7 @@ BMK_benchMemAdvancedNoAlloc(
                 }
 
                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f), %6.*f MB/s, %6.1f MB/s \r",
+                    OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f), %6.*f MB/s, %6.1f MB/s \r",
                             marks[markNb], displayName,
                             (unsigned)srcSize, (unsigned)cSize,
                             ratioAccuracy, ratio,
@@ -537,13 +539,13 @@ BMK_benchMemAdvancedNoAlloc(
             double const cSpeed = (double)benchResult.cSpeed / MB_UNIT;
             double const dSpeed = (double)benchResult.dSpeed / MB_UNIT;
             if (adv->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, adv->additionalParam);
+                OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, adv->additionalParam);
             } else {
-                DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName);
+                OUTPUT("-%-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);
+        OUTPUTLEVEL(2, "%2i#\n", cLevel);
     }   /* Bench */
 
     benchResult.cMem = (1ULL << (comprParams->windowLog)) + ZSTD_sizeof_CCtx(cctx);
@@ -672,7 +674,7 @@ static BMK_benchOutcome_t BMK_benchCLevel(const void* srcBuffer, size_t benchedS
     }
 
     if (displayLevel == 1 && !adv->additionalParam)   /* --quiet mode */
-        DISPLAY("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n",
+        OUTPUT("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n",
                 ZSTD_VERSION_STRING, ZSTD_GIT_COMMIT_STRING,
                 (unsigned)benchedSize, adv->nbSeconds, (unsigned)(adv->blockSize>>10));
 
@@ -762,7 +764,7 @@ static int BMK_loadFiles(void* buffer, size_t bufferSize,
         }
         {   FILE* const f = fopen(fileNamesTable[n], "rb");
             if (f==NULL) RETURN_ERROR_INT(10, "impossible to open file %s", fileNamesTable[n]);
-            DISPLAYLEVEL(2, "Loading %s...       \r", fileNamesTable[n]);
+            OUTPUTLEVEL(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) RETURN_ERROR_INT(11, "could not read %s", fileNamesTable[n]);