diff --git a/lib/common/entropy_common.c b/lib/common/entropy_common.c
index 0a13d9d9..3969d2b3 100644
--- a/lib/common/entropy_common.c
+++ b/lib/common/entropy_common.c
@@ -96,7 +96,6 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             bitStream >>= 2 * repeats;
             bitCount += 2 * repeats;
 
-            assert(bitCount < 30 && (bitStream & 3) != 3);
             charnum += bitStream & 3;
             bitCount += 2;
 
@@ -186,6 +185,15 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
 size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize)
+{
+    U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp));
+}
+
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
 {
     U32 weightTotal;
     const BYTE* ip = (const BYTE*) src;
@@ -208,9 +216,8 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                 huffWeight[n+1] = ip[n/2] & 15;
     }   }   }
     else  {   /* header compressed with FSE (normal case) */
-        FSE_DTable fseWorkspace[FSE_DTABLE_SIZE_U32(6)];  /* 6 is max possible tableLog for HUF header (maybe even 5, to be tested) */
         if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
-        oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, fseWorkspace, 6);   /* max (hwSize-1) values decoded, as last one is implied */
+        oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize);   /* max (hwSize-1) values decoded, as last one is implied */
         if (FSE_isError(oSize)) return oSize;
     }
 
diff --git a/lib/common/fse.h b/lib/common/fse.h
index 55dd8f3a..8d15b340 100644
--- a/lib/common/fse.h
+++ b/lib/common/fse.h
@@ -311,7 +311,7 @@ unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsi
  * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
  * FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
  */
-#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
+#define FSE_COMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
 size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
 size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
@@ -326,14 +326,21 @@ size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
  */
 size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
+#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1 << maxTableLog) + 8)
+#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
+FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
+
 size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
 /**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
 
 size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
 /**< build a fake FSE_DTable, designed to always generate the same symbolValue */
 
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
-/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue))
+#define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)` */
 
 typedef enum {
    FSE_repeat_none,  /**< Cannot use the previous table */
diff --git a/lib/common/fse_decompress.c b/lib/common/fse_decompress.c
index 54dab255..119ee7b6 100644
--- a/lib/common/fse_decompress.c
+++ b/lib/common/fse_decompress.c
@@ -68,17 +68,24 @@ void FSE_freeDTable (FSE_DTable* dt)
     free(dt);
 }
 
-size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog) {
+    U32 wksp[FSE_BUILD_DTABLE_WKSP_SIZE_U32(FSE_TABLELOG_ABSOLUTE_MAX, FSE_MAX_SYMBOL_VALUE)];
+    return FSE_buildDTable_wksp(dt, normalizedCounter, maxSymbolValue, tableLog, wksp, sizeof(wksp));
+}
+
+size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
 {
     void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
     FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
-    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U16* symbolNext = (U16*)workSpace;
+    BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
 
     U32 const maxSV1 = maxSymbolValue + 1;
     U32 const tableSize = 1 << tableLog;
     U32 highThreshold = tableSize-1;
 
     /* Sanity Checks */
+    if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
     if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
@@ -100,7 +107,53 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
     }
 
     /* Spread symbols */
-    {   U32 const tableMask = tableSize-1;
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                pos += n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what emperically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].symbol = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        U32 const tableMask = tableSize-1;
         U32 const step = FSE_TABLESTEP(tableSize);
         U32 s, position = 0;
         for (s=0; s<maxSV1; s++) {
@@ -252,13 +305,14 @@ size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
 }
 
 
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog)
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
 {
     const BYTE* const istart = (const BYTE*)cSrc;
     const BYTE* ip = istart;
     short counting[FSE_MAX_SYMBOL_VALUE+1];
     unsigned tableLog;
     unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    FSE_DTable* const dtable = (FSE_DTable*)workSpace;
 
     /* normal FSE decoding mode */
     size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
@@ -268,9 +322,13 @@ size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size
     ip += NCountLength;
     cSrcSize -= NCountLength;
 
-    CHECK_F( FSE_buildDTable (workSpace, counting, maxSymbolValue, tableLog) );
+    if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
+    workSpace = dtable + FSE_DTABLE_SIZE_U32(tableLog);
+    wkspSize -= FSE_DTABLE_SIZE(tableLog);
 
-    return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, workSpace);   /* always return, even if it is an error code */
+    CHECK_F( FSE_buildDTable_wksp(dtable, counting, maxSymbolValue, tableLog, workSpace, wkspSize) );
+
+    return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, dtable);   /* always return, even if it is an error code */
 }
 
 
@@ -278,8 +336,9 @@ typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
 
 size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
 {
-    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
-    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, dt, FSE_MAX_TABLELOG);
+    /* Static analyzer seems unable to understand this table will be properly initialized later */
+    U32 wksp[FSE_DECOMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
+    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, FSE_MAX_TABLELOG, wksp, sizeof(wksp));
 }
 
 
diff --git a/lib/common/huf.h b/lib/common/huf.h
index ef432685..90e613a1 100644
--- a/lib/common/huf.h
+++ b/lib/common/huf.h
@@ -111,6 +111,8 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
 
 /* *** Dependencies *** */
 #include "mem.h"   /* U32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
 
 
 /* *** Constants *** */
@@ -226,6 +228,17 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
                      U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize);
 
+/*! HUF_readStats_wksp() :
+ * Same as HUF_readStats() but takes an external workspace which must be
+ * 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
+ */
+#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
+#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
+                          U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
+                          const void* src, size_t srcSize,
+                          void* workspace, size_t wkspSize);
+
 /** HUF_readCTable() :
  *  Loading a CTable saved with HUF_writeCTable() */
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
diff --git a/lib/compress/fse_compress.c b/lib/compress/fse_compress.c
index 1187e3e6..48b654d0 100644
--- a/lib/compress/fse_compress.c
+++ b/lib/compress/fse_compress.c
@@ -630,7 +630,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
 
     /* init conditions */
-    if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
+    if (wkspSize < FSE_COMPRESS_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
     if (srcSize <= 1) return 0;  /* Not compressible */
     if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
@@ -674,7 +674,7 @@ typedef struct {
 size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
 {
     fseWkspMax_t scratchBuffer;
-    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_COMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
     return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
 }
diff --git a/lib/decompress/huf_decompress.c b/lib/decompress/huf_decompress.c
index 68293a13..fbe4127a 100644
--- a/lib/decompress/huf_decompress.c
+++ b/lib/decompress/huf_decompress.c
@@ -115,6 +115,70 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
 /*-***************************/
 typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decoding */
 
+/**
+ * Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
+ * a time.
+ */
+static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        D4 = symbol + (nbBits << 8);
+    } else {
+        D4 = (symbol << 8) + nbBits;
+    }
+    D4 *= 0x0001000100010001ULL;
+    return D4;
+}
+
+#if 0
+// TODO: Remove this
+/* BMI2 version that uses _pdep_u64() for weight 1 and 2 symbols.
+ * This doesn't provide much gains, so not worth the complexity.
+ * Leaving in for now but will remove before I commit.
+ */
+#include <immintrin.h>
+
+static U64 HUF_DEltX1_pack4(BYTE const* symbols, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        U64 const nbBits4 = nbBits * 0x0100010001000100ULL;
+        U64 const symbols4 = _pdep_u64(MEM_read32(symbols), 0x00FF00FF00FF00FFULL);
+        D4 = symbols4 | nbBits4;
+    } else {
+        U64 const nbBits4 = nbBits * 0x0001000100010001ULL;
+        U64 const symbols4 = _pdep_u64(MEM_read32(symbols), 0xFF00FF00FF00FF00ULL);
+        D4 = symbols4 | nbBits4;
+    }
+    return D4;
+}
+
+static U64 HUF_DEltX1_pack2(BYTE const* symbols, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        U64 const nbBits4 = nbBits * 0x0100010001000100ULL;
+        U64 symbols4 = _pdep_u64(MEM_read16(symbols), 0x000000FF000000FFULL);
+        symbols4 = symbols4 * 0x00010001ULL;
+        D4 = symbols4 | nbBits4;
+    } else {
+        U64 const nbBits4 = nbBits * 0x0001000100010001ULL;
+        U64 symbols4 = _pdep_u64(MEM_read16(symbols), 0x0000FF000000FF00ULL);
+        symbols4 *= 0x00010001ULL;
+        D4 = symbols4 | nbBits4;
+    }
+    return D4;
+}
+#endif
+
+typedef struct {
+        U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 statsWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+        BYTE symbols[HUF_SYMBOLVALUE_MAX + 1];
+        BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
+} HUF_ReadDTableX1_Workspace;
+
+
+// TODO: Template based on BMI2 (5% boost)
 size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
 {
     U32 tableLog = 0;
@@ -122,22 +186,15 @@ size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
     size_t iSize;
     void* const dtPtr = DTable + 1;
     HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
+    HUF_ReadDTableX1_Workspace* wksp = (HUF_ReadDTableX1_Workspace*)workSpace;
 
-    U32* rankVal;
-    BYTE* huffWeight;
-    size_t spaceUsed32 = 0;
-
-    rankVal = (U32 *)workSpace + spaceUsed32;
-    spaceUsed32 += HUF_TABLELOG_ABSOLUTEMAX + 1;
-    huffWeight = (BYTE *)((U32 *)workSpace + spaceUsed32);
-    spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
-
-    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
+    DEBUG_STATIC_ASSERT(HUF_DECOMPRESS_WORKSPACE_SIZE >= sizeof(*wksp));
+    if (sizeof(*wksp) > wkspSize) return ERROR(tableLog_tooLarge);
 
     DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp));
     if (HUF_isError(iSize)) return iSize;
 
     /* Table header */
@@ -148,39 +205,103 @@ size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
         memcpy(DTable, &dtd, sizeof(dtd));
     }
 
-    /* Calculate starting value for each rank */
-    {   U32 n, nextRankStart = 0;
-        for (n=1; n<tableLog+1; n++) {
+    /* Compute symbols and rankStart given rankVal:
+     *
+     * rankVal already contains the number of values of each weight.
+     *
+     * symbols contains the symbols ordered by weight. First are the rankVal[0]
+     * weight 0 symbols, followed by the rankVal[1] weight 1 symbols, and so on.
+     * symbols[0] is filled (but unused) to avoid a branch.
+     *
+     * rankStart contains the offset where each rank belongs in the DTable.
+     * rankStart[0] is not filled because there are no entries in the table for
+     * weight 0.
+     */
+    {
+        int n;
+        int nextRankStart = 0;
+        for (n=0; n<(int)tableLog+1; n++) {
             U32 const current = nextRankStart;
-            nextRankStart += (rankVal[n] << (n-1));
-            rankVal[n] = current;
-    }   }
+            nextRankStart += wksp->rankVal[n];
+            wksp->rankStart[n] = current;
+        }
+        // TODO: This loop is now the bottleneck: Can this be made faster?
+        for (n=0; n < (int)nbSymbols; ++n) {
+            size_t const w = wksp->huffWeight[n];
+            wksp->symbols[wksp->rankStart[w]++] = n;
+        }
+    }
 
-    /* fill DTable */
-    {   U32 n;
-        size_t const nEnd = nbSymbols;
-        for (n=0; n<nEnd; n++) {
-            size_t const w = huffWeight[n];
-            size_t const length = (1 << w) >> 1;
-            size_t const uStart = rankVal[w];
-            size_t const uEnd = uStart + length;
-            size_t u;
-            HUF_DEltX1 D;
-            D.byte = (BYTE)n;
-            D.nbBits = (BYTE)(tableLog + 1 - w);
-            rankVal[w] = (U32)uEnd;
-            if (length < 4) {
-                /* Use length in the loop bound so the compiler knows it is short. */
-                for (u = 0; u < length; ++u)
-                    dt[uStart + u] = D;
-            } else {
-                /* Unroll the loop 4 times, we know it is a power of 2. */
-                for (u = uStart; u < uEnd; u += 4) {
-                    dt[u + 0] = D;
-                    dt[u + 1] = D;
-                    dt[u + 2] = D;
-                    dt[u + 3] = D;
-    }   }   }   }
+    /* fill DTable
+     * We fill all entries of each weight in order.
+     * That way length is a constant for each iteration of the outter loop.
+     * We can switch based on the length to a different inner loop which is
+     * optimized for that particular case.
+     */
+    {
+        U32 w;
+        int symbol=wksp->rankVal[0];
+        int rankStart=0;
+        for (w=1; w<tableLog+1; ++w) {
+            int const symbolCount = wksp->rankVal[w];
+            int const length = (1 << w) >> 1;
+            int uStart = rankStart;
+            BYTE const nbBits = tableLog + 1 - w;
+            int s;
+            int u;
+            switch (length) {
+            case 1:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart] = D;
+                    uStart += 1;
+                }
+                break;
+            case 2:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart+0] = D;
+                    dt[uStart+1] = D;
+                    uStart += 2;
+                }
+                break;
+            case 4:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    uStart += 4;
+                }
+                break;
+            case 8:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    MEM_write64(dt + uStart + 4, D4);
+                    uStart += 8;
+                }
+                break;
+            default:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    for (u=0; u < length; u += 16) {
+                        MEM_write64(dt + uStart + u + 0, D4);
+                        MEM_write64(dt + uStart + u + 4, D4);
+                        MEM_write64(dt + uStart + u + 8, D4);
+                        MEM_write64(dt + uStart + u + 12, D4);
+                    }
+                    assert(u == length);
+                    uStart += length;
+                }
+                break;
+            }
+            symbol += symbolCount;
+            rankStart += symbolCount * length;
+        }
+    }
     return iSize;
 }
 
diff --git a/lib/decompress/zstd_decompress_block.c b/lib/decompress/zstd_decompress_block.c
index 95afcaa3..51ef977f 100644
--- a/lib/decompress/zstd_decompress_block.c
+++ b/lib/decompress/zstd_decompress_block.c
@@ -368,14 +368,17 @@ void
 ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
             const short* normalizedCounter, unsigned maxSymbolValue,
             const U32* baseValue, const U32* nbAdditionalBits,
-            unsigned tableLog, U32* wksp, size_t wkspSize)
+            unsigned tableLog, void* wksp, size_t wkspSize)
 {
     ZSTD_seqSymbol* const tableDecode = dt+1;
-    U16 symbolNext[MaxSeq+1];
-
     U32 const maxSV1 = maxSymbolValue + 1;
     U32 const tableSize = 1 << tableLog;
 
+    U16* symbolNext = (U16*)wksp;
+    BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
+
+    assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
+
     /* Sanity Checks */
     assert(maxSymbolValue <= MaxSeq);
     assert(tableLog <= MaxFSELog);
@@ -414,9 +417,6 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
          * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
          * our buffer to handle the over-write.
          */
-        BYTE* spread = (BYTE*)wksp;
-        assert(wkspSize >= (1u << MaxFSELog) + sizeof(U64));
-        (void)wkspSize;
         {
             U64 const add = 0x0101010101010101ull;
             size_t pos = 0;
diff --git a/lib/decompress/zstd_decompress_block.h b/lib/decompress/zstd_decompress_block.h
index 201d6a9f..03afdde4 100644
--- a/lib/decompress/zstd_decompress_block.h
+++ b/lib/decompress/zstd_decompress_block.h
@@ -48,12 +48,14 @@ size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
  * this function must be called with valid parameters only
  * (dt is large enough, normalizedCounter distribution total is a power of 2, max is within range, etc.)
  * in which case it cannot fail.
+ * The workspace must be 4-byte aligned and at least ZSTD_BUILD_FSE_TABLE_WKSP_SIZE bytes.
  * Internal use only.
  */
+#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
 void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
              const short* normalizedCounter, unsigned maxSymbolValue,
              const U32* baseValue, const U32* nbAdditionalBits,
-                   unsigned tableLog, U32* wksp, size_t wkspSize);
+                   unsigned tableLog, void* wksp, size_t wkspSize);
 
 
 #endif /* ZSTD_DEC_BLOCK_H */