extzstd 0.3.1 → 0.3.2

data/contrib/zstd/lib/README.md

@@ -143,6 +143,14 @@ The file structure is designed to make this selection manually achievable for an
   Setting this macro will either force to generate the BMI2 dispatcher (1)
   or prevent it (0). It overrides automatic detection.
 
+- The build macro `ZSTD_NO_UNUSED_FUNCTIONS` can be defined to hide the definitions of functions
+  that zstd does not use. Not all unused functions are hidden, but they can be if needed.
+  Currently, this macro will hide function definitions in FSE and HUF that use an excessive
+  amount of stack space.
+
+- The build macro `ZSTD_NO_INTRINSICS` can be defined to disable all explicit intrinsics.
+  Compiler builtins are still used.
+
 
 #### Windows : using MinGW+MSYS to create DLL
 
@@ -160,6 +168,26 @@ file it should be linked with `dll\libzstd.dll`. For example:
 The compiled executable will require ZSTD DLL which is available at `dll\libzstd.dll`.
 
 
+#### Advanced Build options
+
+The build system requires a hash function in order to
+separate object files created with different compilation flags.
+By default, it tries to use `md5sum` or equivalent.
+The hash function can be manually switched by setting the `HASH` variable.
+For example : `make HASH=xxhsum`
+The hash function needs to generate at least 64-bit using hexadecimal format.
+When no hash function is found,
+the Makefile just generates all object files into the same default directory,
+irrespective of compilation flags.
+This functionality only matters if `libzstd` is compiled multiple times
+with different build flags.
+
+The build directory, where object files are stored
+can also be manually controlled using variable `BUILD_DIR`,
+for example `make BUILD_DIR=objectDir/v1`.
+In which case, the hash function doesn't matter.
+
+
 #### Deprecated API
 
 Obsolete API on their way out are stored in directory `lib/deprecated`.

data/contrib/zstd/lib/common/bitstream.h

@@ -17,7 +17,6 @@
 #if defined (__cplusplus)
 extern "C" {
 #endif
-
 /*
 *  This API consists of small unitary functions, which must be inlined for best performance.
 *  Since link-time-optimization is not available for all compilers,
@@ -36,10 +35,12 @@ extern "C" {
 /*=========================================
 *  Target specific
 =========================================*/
-#if defined(__BMI__) && defined(__GNUC__)
-#  include <immintrin.h>   /* support for bextr (experimental) */
-#elif defined(__ICCARM__)
-#  include <intrinsics.h>
+#ifndef ZSTD_NO_INTRINSICS
+#  if defined(__BMI__) && defined(__GNUC__)
+#    include <immintrin.h>   /* support for bextr (experimental) */
+#  elif defined(__ICCARM__)
+#    include <intrinsics.h>
+#  endif
 #endif
 
 #define STREAM_ACCUMULATOR_MIN_32  25
@@ -141,8 +142,12 @@ MEM_STATIC unsigned BIT_highbit32 (U32 val)
     assert(val != 0);
     {
 #   if defined(_MSC_VER)   /* Visual */
-        unsigned long r=0;
-        return _BitScanReverse ( &r, val ) ? (unsigned)r : 0;
+#       if STATIC_BMI2 == 1
+		return _lzcnt_u32(val) ^ 31;
+#       else
+		unsigned long r = 0;
+		return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#       endif
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
         return __builtin_clz (val) ^ 31;
 #   elif defined(__ICCARM__)    /* IAR Intrinsic */
@@ -198,7 +203,7 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
 MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
                             size_t value, unsigned nbBits)
 {
-    MEM_STATIC_ASSERT(BIT_MASK_SIZE == 32);
+    DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
     assert(nbBits < BIT_MASK_SIZE);
     assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
     bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
@@ -271,7 +276,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
  */
 MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
 {
-    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+    if (srcSize < 1) { ZSTD_memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
 
     bitD->start = (const char*)srcBuffer;
     bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
@@ -317,12 +322,12 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
     return srcSize;
 }
 
-MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
 {
     return bitContainer >> start;
 }
 
-MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
 {
     U32 const regMask = sizeof(bitContainer)*8 - 1;
     /* if start > regMask, bitstream is corrupted, and result is undefined */
@@ -330,10 +335,14 @@ MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 co
     return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
 }
 
-MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
 {
+#if defined(STATIC_BMI2) && STATIC_BMI2 == 1
+	return  _bzhi_u64(bitContainer, nbBits);
+#else
     assert(nbBits < BIT_MASK_SIZE);
     return bitContainer & BIT_mask[nbBits];
+#endif
 }
 
 /*! BIT_lookBits() :
@@ -342,7 +351,7 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
  *  On 32-bits, maxNbBits==24.
  *  On 64-bits, maxNbBits==56.
  * @return : value extracted */
-MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
+MEM_STATIC  FORCE_INLINE_ATTR size_t BIT_lookBits(const BIT_DStream_t*  bitD, U32 nbBits)
 {
     /* arbitrate between double-shift and shift+mask */
 #if 1
@@ -365,7 +374,7 @@ MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
     return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
 }
 
-MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
 {
     bitD->bitsConsumed += nbBits;
 }
@@ -374,7 +383,7 @@ MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
  *  Read (consume) next n bits from local register and update.
  *  Pay attention to not read more than nbBits contained into local register.
  * @return : extracted value. */
-MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
     BIT_skipBits(bitD, nbBits);

data/contrib/zstd/lib/common/compiler.h

@@ -38,6 +38,17 @@
 
 #endif
 
+/**
+  On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
+  This explictly marks such functions as __cdecl so that the code will still compile
+  if a CC other than __cdecl has been made the default.
+*/
+#if  defined(_MSC_VER)
+#  define WIN_CDECL __cdecl
+#else
+#  define WIN_CDECL
+#endif
+
 /**
  * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
  * parameters. They must be inlined for the compiler to eliminate the constant
@@ -114,12 +125,12 @@
 #    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
 #    define PREFETCH_L1(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
 #    define PREFETCH_L2(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
-#    elif defined(__aarch64__)
-#     define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
-#     define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
 #  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
 #    define PREFETCH_L1(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
 #    define PREFETCH_L2(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
+#  elif defined(__aarch64__)
+#    define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
+#    define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
 #  else
 #    define PREFETCH_L1(ptr) (void)(ptr)  /* disabled */
 #    define PREFETCH_L2(ptr) (void)(ptr)  /* disabled */
@@ -172,4 +183,106 @@
 #  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
 #endif
 
+/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
+#ifndef STATIC_BMI2
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))
+#    ifdef __AVX2__  //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
+#       define STATIC_BMI2 1
+#    endif
+#  endif
+#endif
+
+#ifndef STATIC_BMI2
+    #define STATIC_BMI2 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_builtin
+#  define __has_builtin(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_feature
+#  define __has_feature(x) 0
+#endif
+
+/* detects whether we are being compiled under msan */
+#ifndef ZSTD_MEMORY_SANITIZER
+#  if __has_feature(memory_sanitizer)
+#    define ZSTD_MEMORY_SANITIZER 1
+#  else
+#    define ZSTD_MEMORY_SANITIZER 0
+#  endif
+#endif
+
+#if ZSTD_MEMORY_SANITIZER
+/* Not all platforms that support msan provide sanitizers/msan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+#define ZSTD_DEPS_NEED_STDINT
+#include "zstd_deps.h"  /* intptr_t */
+
+/* Make memory region fully initialized (without changing its contents). */
+void __msan_unpoison(const volatile void *a, size_t size);
+
+/* Make memory region fully uninitialized (without changing its contents).
+   This is a legacy interface that does not update origin information. Use
+   __msan_allocated_memory() instead. */
+void __msan_poison(const volatile void *a, size_t size);
+
+/* Returns the offset of the first (at least partially) poisoned byte in the
+   memory range, or -1 if the whole range is good. */
+intptr_t __msan_test_shadow(const volatile void *x, size_t size);
+#endif
+
+/* detects whether we are being compiled under asan */
+#ifndef ZSTD_ADDRESS_SANITIZER
+#  if __has_feature(address_sanitizer)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  elif defined(__SANITIZE_ADDRESS__)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  else
+#    define ZSTD_ADDRESS_SANITIZER 0
+#  endif
+#endif
+
+#if ZSTD_ADDRESS_SANITIZER
+/* Not all platforms that support asan provide sanitizers/asan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
+ *
+ * This memory must be previously allocated by your program. Instrumented
+ * code is forbidden from accessing addresses in this region until it is
+ * unpoisoned. This function is not guaranteed to poison the entire region -
+ * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
+ * alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can poison or
+ * unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_poison_memory_region(void const volatile *addr, size_t size);
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
+ *
+ * This memory must be previously allocated by your program. Accessing
+ * addresses in this region is allowed until this region is poisoned again.
+ * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
+ * to ASan alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can
+ * poison or unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
+#endif
+
 #endif /* ZSTD_COMPILER_H */

data/contrib/zstd/lib/common/cpu.h

@@ -16,8 +16,6 @@
  * https://github.com/facebook/folly/blob/master/folly/CpuId.h
  */
 
-#include <string.h>
-
 #include "mem.h"
 
 #ifdef _MSC_VER

data/contrib/zstd/lib/common/debug.h

@@ -51,15 +51,6 @@ extern "C" {
 #endif
 
 
-/* DEBUGFILE can be defined externally,
- * typically through compiler command line.
- * note : currently useless.
- * Value must be stderr or stdout */
-#ifndef DEBUGFILE
-#  define DEBUGFILE stderr
-#endif
-
-
 /* recommended values for DEBUGLEVEL :
  * 0 : release mode, no debug, all run-time checks disabled
  * 1 : enables assert() only, no display
@@ -76,7 +67,8 @@ extern "C" {
  */
 
 #if (DEBUGLEVEL>=1)
-#  include <assert.h>
+#  define ZSTD_DEPS_NEED_ASSERT
+#  include "zstd_deps.h"
 #else
 #  ifndef assert   /* assert may be already defined, due to prior #include <assert.h> */
 #    define assert(condition) ((void)0)   /* disable assert (default) */
@@ -84,7 +76,8 @@ extern "C" {
 #endif
 
 #if (DEBUGLEVEL>=2)
-#  include <stdio.h>
+#  define ZSTD_DEPS_NEED_IO
+#  include "zstd_deps.h"
 extern int g_debuglevel; /* the variable is only declared,
                             it actually lives in debug.c,
                             and is shared by the whole process.
@@ -92,14 +85,14 @@ extern int g_debuglevel; /* the variable is only declared,
                             It's useful when enabling very verbose levels
                             on selective conditions (such as position in src) */
 
-#  define RAWLOG(l, ...) {                                      \
-                if (l<=g_debuglevel) {                          \
-                    fprintf(stderr, __VA_ARGS__);               \
+#  define RAWLOG(l, ...) {                                       \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__VA_ARGS__);               \
             }   }
-#  define DEBUGLOG(l, ...) {                                    \
-                if (l<=g_debuglevel) {                          \
-                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
-                    fprintf(stderr, " \n");                     \
+#  define DEBUGLOG(l, ...) {                                     \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
+                    ZSTD_DEBUG_PRINT(" \n");                     \
             }   }
 #else
 #  define RAWLOG(l, ...)      {}    /* disabled */

data/contrib/zstd/lib/common/entropy_common.c

@@ -38,8 +38,31 @@ const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
-                 const void* headerBuffer, size_t hbSize)
+static U32 FSE_ctz(U32 val)
+{
+    assert(val != 0);
+    {
+#   if defined(_MSC_VER)   /* Visual */
+        unsigned long r=0;
+        return _BitScanForward(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
+        return __builtin_ctz(val);
+#   elif defined(__ICCARM__)    /* IAR Intrinsic */
+        return __CTZ(val);
+#   else   /* Software version */
+        U32 count = 0;
+        while ((val & 1) == 0) {
+            val >>= 1;
+            ++count;
+        }
+        return count;
+#   endif
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                           const void* headerBuffer, size_t hbSize)
 {
     const BYTE* const istart = (const BYTE*) headerBuffer;
     const BYTE* const iend = istart + hbSize;
@@ -50,23 +73,23 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     U32 bitStream;
     int bitCount;
     unsigned charnum = 0;
+    unsigned const maxSV1 = *maxSVPtr + 1;
     int previous0 = 0;
 
-    if (hbSize < 4) {
-        /* This function only works when hbSize >= 4 */
-        char buffer[4];
-        memset(buffer, 0, sizeof(buffer));
-        memcpy(buffer, headerBuffer, hbSize);
+    if (hbSize < 8) {
+        /* This function only works when hbSize >= 8 */
+        char buffer[8] = {0};
+        ZSTD_memcpy(buffer, headerBuffer, hbSize);
         {   size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
                                                     buffer, sizeof(buffer));
             if (FSE_isError(countSize)) return countSize;
             if (countSize > hbSize) return ERROR(corruption_detected);
             return countSize;
     }   }
-    assert(hbSize >= 4);
+    assert(hbSize >= 8);
 
     /* init */
-    memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
+    ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
     bitStream = MEM_readLE32(ip);
     nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
     if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
@@ -77,36 +100,58 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     threshold = 1<<nbBits;
     nbBits++;
 
-    while ((remaining>1) & (charnum<=*maxSVPtr)) {
+    for (;;) {
         if (previous0) {
-            unsigned n0 = charnum;
-            while ((bitStream & 0xFFFF) == 0xFFFF) {
-                n0 += 24;
-                if (ip < iend-5) {
-                    ip += 2;
-                    bitStream = MEM_readLE32(ip) >> bitCount;
+            /* Count the number of repeats. Each time the
+             * 2-bit repeat code is 0b11 there is another
+             * repeat.
+             * Avoid UB by setting the high bit to 1.
+             */
+            int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+            while (repeats >= 12) {
+                charnum += 3 * 12;
+                if (LIKELY(ip <= iend-7)) {
+                    ip += 3;
                 } else {
-                    bitStream >>= 16;
-                    bitCount   += 16;
-            }   }
-            while ((bitStream & 3) == 3) {
-                n0 += 3;
-                bitStream >>= 2;
-                bitCount += 2;
+                    bitCount -= (int)(8 * (iend - 7 - ip));
+                    bitCount &= 31;
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> bitCount;
+                repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
             }
-            n0 += bitStream & 3;
+            charnum += 3 * repeats;
+            bitStream >>= 2 * repeats;
+            bitCount += 2 * repeats;
+
+            /* Add the final repeat which isn't 0b11. */
+            assert((bitStream & 3) < 3);
+            charnum += bitStream & 3;
             bitCount += 2;
-            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
-            while (charnum < n0) normalizedCounter[charnum++] = 0;
-            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+
+            /* This is an error, but break and return an error
+             * at the end, because returning out of a loop makes
+             * it harder for the compiler to optimize.
+             */
+            if (charnum >= maxSV1) break;
+
+            /* We don't need to set the normalized count to 0
+             * because we already memset the whole buffer to 0.
+             */
+
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
                 assert((bitCount >> 3) <= 3); /* For first condition to work */
                 ip += bitCount>>3;
                 bitCount &= 7;
-                bitStream = MEM_readLE32(ip) >> bitCount;
             } else {
-                bitStream >>= 2;
-        }   }
-        {   int const max = (2*threshold-1) - remaining;
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> bitCount;
+        }
+        {
+            int const max = (2*threshold-1) - remaining;
             int count;
 
             if ((bitStream & (threshold-1)) < (U32)max) {
@@ -119,24 +164,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             }
 
             count--;   /* extra accuracy */
-            remaining -= count < 0 ? -count : count;   /* -1 means +1 */
+            /* When it matters (small blocks), this is a
+             * predictable branch, because we don't use -1.
+             */
+            if (count >= 0) {
+                remaining -= count;
+            } else {
+                assert(count == -1);
+                remaining += count;
+            }
             normalizedCounter[charnum++] = (short)count;
             previous0 = !count;
-            while (remaining < threshold) {
-                nbBits--;
-                threshold >>= 1;
+
+            assert(threshold > 1);
+            if (remaining < threshold) {
+                /* This branch can be folded into the
+                 * threshold update condition because we
+                 * know that threshold > 1.
+                 */
+                if (remaining <= 1) break;
+                nbBits = BIT_highbit32(remaining) + 1;
+                threshold = 1 << (nbBits - 1);
             }
+            if (charnum >= maxSV1) break;
 
-            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
                 ip += bitCount>>3;
                 bitCount &= 7;
             } else {
                 bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
                 ip = iend - 4;
             }
-            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
-    }   }   /* while ((remaining>1) & (charnum<=*maxSVPtr)) */
+            bitStream = MEM_readLE32(ip) >> bitCount;
+    }   }
     if (remaining != 1) return ERROR(corruption_detected);
+    /* Only possible when there are too many zeros. */
+    if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall);
     if (bitCount > 32) return ERROR(corruption_detected);
     *maxSVPtr = charnum-1;
 
@@ -144,6 +208,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     return ip-istart;
 }
 
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t FSE_readNCount_body_default(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+#if DYNAMIC_BMI2
+TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+#endif
+
+size_t FSE_readNCount_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+    }
+#endif
+    (void)bmi2;
+    return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+size_t FSE_readNCount(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0);
+}
+
 
 /*! HUF_readStats() :
     Read compact Huffman tree, saved by HUF_writeCTable().
@@ -155,6 +256,17 @@ 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), /* bmi2 */ 0);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                   U32* nbSymbolsPtr, U32* tableLogPtr,
+                   const void* src, size_t srcSize,
+                   void* workSpace, size_t wkspSize,
+                   int bmi2)
 {
     U32 weightTotal;
     const BYTE* ip = (const BYTE*) src;
@@ -163,7 +275,7 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
 
     if (!srcSize) return ERROR(srcSize_wrong);
     iSize = ip[0];
-    /* memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
+    /* ZSTD_memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
 
     if (iSize >= 128) {  /* special header */
         oSize = iSize - 127;
@@ -177,14 +289,14 @@ 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 */
+        /* max (hwSize-1) values decoded, as last one is implied */
+        oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2);
         if (FSE_isError(oSize)) return oSize;
     }
 
     /* collect weight stats */
-    memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
+    ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
     weightTotal = 0;
     {   U32 n; for (n=0; n<oSize; n++) {
             if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
@@ -214,3 +326,37 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
     *nbSymbolsPtr = (U32)(oSize+1);
     return iSize+1;
 }
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0);
+}
+
+#if DYNAMIC_BMI2
+static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1);
+}
+#endif
+
+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,
+                     int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+    }
+#endif
+    (void)bmi2;
+    return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+}