diff --git a/Makefile b/Makefile
index d87da54..1970efb 100644
--- a/Makefile
+++ b/Makefile
@@ -31,9 +31,9 @@
 LIBVER := $(LIBVER_MAJOR).$(LIBVER_MINOR).$(LIBVER_PATCH)
 
 CFLAGS ?= -O3
-CFLAGS += -std=c99 -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
+CFLAGS += -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
           -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
-		  -Wstrict-prototypes -Wundef -pedantic
+		  -Wstrict-prototypes -Wundef
 FLAGS  := $(CFLAGS) $(CPPFLAGS) $(LDFLAGS) $(MOREFLAGS)
 XXHSUM_VERSION=$(LIBVER)
 MD2ROFF = ronn
@@ -107,7 +107,7 @@
 
 armtest: clean
 	@echo ---- test ARM compilation ----
-	$(MAKE) xxhsum CC=arm-linux-gnueabi-gcc MOREFLAGS="-Werror"
+	$(MAKE) xxhsum CC=arm-linux-gnueabi-gcc MOREFLAGS="-Werror -static"
 
 clangtest: clean
 	@echo ---- test clang compilation ----
@@ -117,7 +117,12 @@
 	@echo ---- test g++ compilation ----
 	$(MAKE) all CC=g++ CFLAGS="-O3 -Wall -Wextra -Wundef -Wshadow -Wcast-align -Werror"
 
-sanitize: clean
+c90test: clean
+	@echo ---- test strict C90 compilation [xxh32 only] ----
+	$(CC) -std=c90 -Werror -pedantic -DXXH_NO_LONG_LONG -c xxhash.c
+	$(RM) xxhash.o
+
+usan: clean
 	@echo ---- check undefined behavior - sanitize ----
 	$(MAKE) test CC=clang MOREFLAGS="-g -fsanitize=undefined"
 
@@ -142,7 +147,8 @@
 preview-man: clean-man man
 	man ./xxhsum.1
 
-test-all: clean all namespaceTest test test32 test-xxhsum-c clean-xxhsum-c armtest clangtest gpptest sanitize staticAnalyze
+test-all: clean all namespaceTest test test32 test-xxhsum-c clean-xxhsum-c \
+	armtest clangtest gpptest c90test usan staticAnalyze
 
 clean: clean-xxhsum-c
 	@$(RM) -f core *.o xxhsum$(EXT) xxhsum32$(EXT) xxhsum_inlinedXXH$(EXT) xxh32sum xxh64sum
diff --git a/README.md b/README.md
index 9a3a15f..48a9c17 100644
--- a/README.md
+++ b/README.md
@@ -2,8 +2,9 @@
 ======================================
 
 xxHash is an Extremely fast Hash algorithm, running at RAM speed limits.
-It successfully completes the [SMHasher](http://code.google.com/p/smhasher/wiki/SMHasher) test suite 
+It successfully completes the [SMHasher](http://code.google.com/p/smhasher/wiki/SMHasher) test suite
 which evaluates collision, dispersion and randomness qualities of hash functions.
+Code is highly portable, and hashes are identical on all platforms (little / big endian).
 
 |Branch      |Status   |
 |------------|---------|
@@ -38,8 +39,8 @@
 10 is a perfect score.
 Algorithms with a score < 5 are not listed on this table.
 
-A new version, XXH64, has been created thanks to [Mathias Westerdahl]'s contribution, 
-which offers superior speed and dispersion for 64-bits systems. 
+A new version, XXH64, has been created thanks to [Mathias Westerdahl]'s contribution,
+which offers superior speed and dispersion for 64-bits systems.
 Note however that 32-bits applications will still run faster using the 32-bits version.
 [Mathias Westerdahl]: https://github.com/JCash
 
@@ -60,6 +61,28 @@
 The utility `xxhsum` is GPL licensed.
 
 
+### Build modifiers
+
+The following macros influence xxhash behavior. They are all disabled by default.
+
+- `XXH_FORCE_NATIVE_FORMAT` : on big-endian systems : use native number representation,
+                              resulting in system-specific results.
+                              Breaks consistency with little-endian results.
+- `XXH_ACCEPT_NULL_INPUT_POINTER` : if presented with a null-pointer,
+                              xxhash result is the same as a null-length key,
+                              instead of a dereference segfault.
+- `XXH_NO_LONG_LONG` : removes support for XXH64,
+                       useful for targets without 64-bits support.
+- `XXH_STATIC_LINKING_ONLY` : gives access to state definition for static allocation.
+                      Incompatible with dynamic linking, due to risks of ABI changes.
+- `XXH_PRIVATE_API` : Make all functions `static` and accessible through `xxhash.h` for inlining.
+                      Do not compile `xxhash.c` as a separate module in this case.
+- `XXH_NAMESPACE` : prefix all symbols with the value of `XXH_NAMESPACE`,
+                    in order to evade symbol naming collisions,
+                    in case of multiple inclusions of xxHash library
+                    (typically via intermediate libraries).
+
+
 ### Other languages
 
 Beyond the C reference version,
diff --git a/xxhash.c b/xxhash.c
index 17a54f1..4546c54 100644
--- a/xxhash.c
+++ b/xxhash.c
@@ -131,45 +131,37 @@
 *  Basic Types
 ***************************************/
 #ifndef MEM_MODULE
-# define MEM_MODULE
 # if !defined (__VMS) && (defined (__cplusplus) || (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;
-#  else
+# else
     typedef unsigned char      BYTE;
     typedef unsigned short     U16;
     typedef unsigned int       U32;
     typedef   signed int       S32;
-    typedef unsigned long long U64;   /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
-#  endif
+# endif
 #endif
 
-
 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 
 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
 static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
-static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
 
 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 
 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 /* currently only defined for gcc and icc */
-typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
+typedef union { U32 u32; } __attribute__((packed)) unalign;
 static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
 
 #else
 
 /* portable and safe solution. Generally efficient.
  * see : http://stackoverflow.com/a/32095106/646947
  */
-
 static U32 XXH_read32(const void* memPtr)
 {
     U32 val;
@@ -177,13 +169,6 @@
     return val;
 }
 
-static U64 XXH_read64(const void* memPtr)
-{
-    U64 val;
-    memcpy(&val, memPtr, sizeof(val));
-    return val;
-}
-
 #endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
 
 
@@ -203,10 +188,8 @@
 
 #if defined(_MSC_VER)     /* Visual Studio */
 #  define XXH_swap32 _byteswap_ulong
-#  define XXH_swap64 _byteswap_uint64
 #elif GCC_VERSION >= 403
 #  define XXH_swap32 __builtin_bswap32
-#  define XXH_swap64 __builtin_bswap64
 #else
 static U32 XXH_swap32 (U32 x)
 {
@@ -215,17 +198,6 @@
             ((x >>  8) & 0x0000ff00 ) |
             ((x >> 24) & 0x000000ff );
 }
-static U64 XXH_swap64 (U64 x)
-{
-    return  ((x << 56) & 0xff00000000000000ULL) |
-            ((x << 40) & 0x00ff000000000000ULL) |
-            ((x << 24) & 0x0000ff0000000000ULL) |
-            ((x << 8)  & 0x000000ff00000000ULL) |
-            ((x >> 8)  & 0x00000000ff000000ULL) |
-            ((x >> 24) & 0x0000000000ff0000ULL) |
-            ((x >> 40) & 0x000000000000ff00ULL) |
-            ((x >> 56) & 0x00000000000000ffULL);
-}
 #endif
 
 
@@ -264,67 +236,23 @@
     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
 }
 
-FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
-{
-    if (align==XXH_unaligned)
-        return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
-    else
-        return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
-}
-
-FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
-{
-    return XXH_readLE64_align(ptr, endian, XXH_unaligned);
-}
-
-static U64 XXH_readBE64(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
-}
-
 
 /* *************************************
 *  Macros
 ***************************************/
 #define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }    /* use only *after* variable declarations */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
 
 
-/* *************************************
-*  Constants
-***************************************/
+/* *******************************************************************
+*  32-bits hash functions
+*********************************************************************/
 static const U32 PRIME32_1 = 2654435761U;
 static const U32 PRIME32_2 = 2246822519U;
 static const U32 PRIME32_3 = 3266489917U;
 static const U32 PRIME32_4 =  668265263U;
 static const U32 PRIME32_5 =  374761393U;
 
-static const U64 PRIME64_1 = 11400714785074694791ULL;
-static const U64 PRIME64_2 = 14029467366897019727ULL;
-static const U64 PRIME64_3 =  1609587929392839161ULL;
-static const U64 PRIME64_4 =  9650029242287828579ULL;
-static const U64 PRIME64_5 =  2870177450012600261ULL;
-
-XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
-
-
-/* **************************
-*  Utils
-****************************/
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
-{
-    memcpy(dstState, srcState, sizeof(*dstState));
-}
-
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
-{
-    memcpy(dstState, srcState, sizeof(*dstState));
-}
-
-
-/* ***************************
-*  Simple Hash Functions
-*****************************/
-
 static U32 XXH32_round(U32 seed, U32 input)
 {
     seed += input * PRIME32_2;
@@ -417,6 +345,267 @@
 }
 
 
+
+/*======   Hash streaming   ======*/
+
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
+{
+    memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
+{
+    XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
+    memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
+    state.v1 = seed + PRIME32_1 + PRIME32_2;
+    state.v2 = seed + PRIME32_2;
+    state.v3 = seed + 0;
+    state.v4 = seed - PRIME32_1;
+    memcpy(statePtr, &state, sizeof(state));
+    return XXH_OK;
+}
+
+
+FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
+{
+    const BYTE* p = (const BYTE*)input;
+    const BYTE* const bEnd = p + len;
+
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
+    if (input==NULL) return XXH_ERROR;
+#endif
+
+    state->total_len_32 += (unsigned)len;
+    state->large_len |= (len>=16) | (state->total_len_32>=16);
+
+    if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
+        state->memsize += (unsigned)len;
+        return XXH_OK;
+    }
+
+    if (state->memsize) {   /* some data left from previous update */
+        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
+        {   const U32* p32 = state->mem32;
+            state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
+            state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
+            state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
+            state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
+        }
+        p += 16-state->memsize;
+        state->memsize = 0;
+    }
+
+    if (p <= bEnd-16) {
+        const BYTE* const limit = bEnd - 16;
+        U32 v1 = state->v1;
+        U32 v2 = state->v2;
+        U32 v3 = state->v3;
+        U32 v4 = state->v4;
+
+        do {
+            v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
+            v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
+            v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
+            v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
+        } while (p<=limit);
+
+        state->v1 = v1;
+        state->v2 = v2;
+        state->v3 = v3;
+        state->v4 = v4;
+    }
+
+    if (p < bEnd) {
+        XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+        state->memsize = (unsigned)(bEnd-p);
+    }
+
+    return XXH_OK;
+}
+
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
+    else
+        return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
+}
+
+
+
+FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
+{
+    const BYTE * p = (const BYTE*)state->mem32;
+    const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
+    U32 h32;
+
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
+    } else {
+        h32 = state->v3 /* == seed */ + PRIME32_5;
+    }
+
+    h32 += state->total_len_32;
+
+    while (p+4<=bEnd) {
+        h32 += XXH_readLE32(p, endian) * PRIME32_3;
+        h32  = XXH_rotl32(h32, 17) * PRIME32_4;
+        p+=4;
+    }
+
+    while (p<bEnd) {
+        h32 += (*p) * PRIME32_5;
+        h32  = XXH_rotl32(h32, 11) * PRIME32_1;
+        p++;
+    }
+
+    h32 ^= h32 >> 15;
+    h32 *= PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= PRIME32_3;
+    h32 ^= h32 >> 16;
+
+    return h32;
+}
+
+
+XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH32_digest_endian(state_in, XXH_littleEndian);
+    else
+        return XXH32_digest_endian(state_in, XXH_bigEndian);
+}
+
+
+/*======   Canonical representation   ======*/
+
+/*! Default XXH result types are basic unsigned 32 and 64 bits.
+*   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
+*   These functions allow transformation of hash result into and from its canonical format.
+*   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
+*/
+
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    memcpy(dst, &hash, sizeof(*dst));
+}
+
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
+}
+
+
+#ifndef XXH_NO_LONG_LONG
+
+/* *******************************************************************
+*  64-bits hash functions
+*********************************************************************/
+
+/*======   Memory access   ======*/
+
+#ifndef MEM_MODULE
+# define MEM_MODULE
+# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include <stdint.h>
+    typedef uint64_t U64;
+# else
+    typedef unsigned long long U64;   /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
+# endif
+#endif
+
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign64;
+
+static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; }
+
+#else
+
+/* portable and safe solution. Generally efficient.
+ * see : http://stackoverflow.com/a/32095106/646947
+ */
+
+static U64 XXH_read64(const void* memPtr)
+{
+    U64 val;
+    memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap64 _byteswap_uint64
+#elif GCC_VERSION >= 403
+#  define XXH_swap64 __builtin_bswap64
+#else
+static U64 XXH_swap64 (U64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
+}
+#endif
+
+FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+    else
+        return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
+}
+
+FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
+{
+    return XXH_readLE64_align(ptr, endian, XXH_unaligned);
+}
+
+static U64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+}
+
+
+/*======   xxh64   ======*/
+
+static const U64 PRIME64_1 = 11400714785074694791ULL;
+static const U64 PRIME64_2 = 14029467366897019727ULL;
+static const U64 PRIME64_3 =  1609587929392839161ULL;
+static const U64 PRIME64_4 =  9650029242287828579ULL;
+static const U64 PRIME64_5 =  2870177450012600261ULL;
+
 static U64 XXH64_round(U64 acc, U64 input)
 {
     acc += input * PRIME64_2;
@@ -528,20 +717,7 @@
 #endif
 }
 
-
-/* **************************************************
-*  Advanced Hash Functions
-****************************************************/
-
-XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
-{
-    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
-}
-XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
-{
-    XXH_free(statePtr);
-    return XXH_OK;
-}
+/*======   Hash Streaming   ======*/
 
 XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
 {
@@ -553,22 +729,11 @@
     return XXH_OK;
 }
 
-
-/*** Hash feed ***/
-
-XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
 {
-    XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
-    memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
-    state.v1 = seed + PRIME32_1 + PRIME32_2;
-    state.v2 = seed + PRIME32_2;
-    state.v3 = seed + 0;
-    state.v4 = seed - PRIME32_1;
-    memcpy(statePtr, &state, sizeof(state));
-    return XXH_OK;
+    memcpy(dstState, srcState, sizeof(*dstState));
 }
 
-
 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
 {
     XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
@@ -581,127 +746,6 @@
     return XXH_OK;
 }
 
-
-FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
-{
-    const BYTE* p = (const BYTE*)input;
-    const BYTE* const bEnd = p + len;
-
-#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (input==NULL) return XXH_ERROR;
-#endif
-
-    state->total_len_32 += len;
-    state->large_len |= (len>=16) | (state->total_len_32>=16);
-
-    if (state->memsize + len < 16)  {   /* fill in tmp buffer */
-        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
-        state->memsize += (U32)len;
-        return XXH_OK;
-    }
-
-    if (state->memsize) {   /* some data left from previous update */
-        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
-        {   const U32* p32 = state->mem32;
-            state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
-            state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
-            state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
-            state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
-        }
-        p += 16-state->memsize;
-        state->memsize = 0;
-    }
-
-    if (p <= bEnd-16) {
-        const BYTE* const limit = bEnd - 16;
-        U32 v1 = state->v1;
-        U32 v2 = state->v2;
-        U32 v3 = state->v3;
-        U32 v4 = state->v4;
-
-        do {
-            v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
-            v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
-            v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
-            v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
-        } while (p<=limit);
-
-        state->v1 = v1;
-        state->v2 = v2;
-        state->v3 = v3;
-        state->v4 = v4;
-    }
-
-    if (p < bEnd) {
-        XXH_memcpy(state->mem32, p, bEnd-p);
-        state->memsize = (int)(bEnd-p);
-    }
-
-    return XXH_OK;
-}
-
-XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
-    else
-        return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
-}
-
-
-
-FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
-{
-    const BYTE * p = (const BYTE*)state->mem32;
-    const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
-    U32 h32;
-
-    if (state->large_len) {
-        h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
-    } else {
-        h32 = state->v3 /* == seed */ + PRIME32_5;
-    }
-
-    h32 += state->total_len_32;
-
-    while (p+4<=bEnd) {
-        h32 += XXH_readLE32(p, endian) * PRIME32_3;
-        h32  = XXH_rotl32(h32, 17) * PRIME32_4;
-        p+=4;
-    }
-
-    while (p<bEnd) {
-        h32 += (*p) * PRIME32_5;
-        h32  = XXH_rotl32(h32, 11) * PRIME32_1;
-        p++;
-    }
-
-    h32 ^= h32 >> 15;
-    h32 *= PRIME32_2;
-    h32 ^= h32 >> 13;
-    h32 *= PRIME32_3;
-    h32 ^= h32 >> 16;
-
-    return h32;
-}
-
-
-XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH32_digest_endian(state_in, XXH_littleEndian);
-    else
-        return XXH32_digest_endian(state_in, XXH_bigEndian);
-}
-
-
-
-/* **** XXH64 **** */
-
 FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
 {
     const BYTE* p = (const BYTE*)input;
@@ -750,8 +794,8 @@
     }
 
     if (p < bEnd) {
-        XXH_memcpy(state->mem64, p, bEnd-p);
-        state->memsize = (int)(bEnd-p);
+        XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+        state->memsize = (unsigned)(bEnd-p);
     }
 
     return XXH_OK;
@@ -767,8 +811,6 @@
         return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
 }
 
-
-
 FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
 {
     const BYTE * p = (const BYTE*)state->mem64;
@@ -820,7 +862,6 @@
     return h64;
 }
 
-
 XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
 {
     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
@@ -832,22 +873,7 @@
 }
 
 
-/* **************************
-*  Canonical representation
-****************************/
-
-/*! Default XXH result types are basic unsigned 32 and 64 bits.
-*   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
-*   These functions allow transformation of hash result into and from its canonical format.
-*   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
-*/
-
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
-    memcpy(dst, &hash, sizeof(*dst));
-}
+/*====== Canonical representation   ======*/
 
 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
 {
@@ -856,12 +882,9 @@
     memcpy(dst, &hash, sizeof(*dst));
 }
 
-XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
-{
-    return XXH_readBE32(src);
-}
-
 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
 {
     return XXH_readBE64(src);
 }
+
+#endif  /* XXH_NO_LONG_LONG */
diff --git a/xxhash.h b/xxhash.h
index 75e2ed2..9475673 100644
--- a/xxhash.h
+++ b/xxhash.h
@@ -73,6 +73,14 @@
 
 
 /* ****************************
+*  Compiler specifics
+******************************/
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* ! C99 */
+#  define restrict   /* disable restrict */
+#endif
+
+
+/* ****************************
 *  Definitions
 ******************************/
 #include <stddef.h>   /* size_t */
@@ -83,13 +91,13 @@
 *  API modifier
 ******************************/
 /** XXH_PRIVATE_API
-*   This is useful if you want to include xxhash functions in `static` mode
+*   This is useful to include xxhash functions in `static` mode
 *   in order to inline them, and remove their symbol from the public list.
 *   Methodology :
 *     #define XXH_PRIVATE_API
 *     #include "xxhash.h"
 *   `xxhash.c` is automatically included.
-*   It's not useful to compile and link it as a separate module anymore.
+*   It's not useful to compile and link it as a separate module.
 */
 #ifdef XXH_PRIVATE_API
 #  ifndef XXH_STATIC_LINKING_ONLY
@@ -111,10 +119,10 @@
 /*!XXH_NAMESPACE, aka Namespace Emulation :
 
 If you want to include _and expose_ xxHash functions from within your own library,
-but also want to avoid symbol collisions with another library which also includes xxHash,
+but also want to avoid symbol collisions with other libraries which may also include xxHash,
 
 you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library
-with the value of XXH_NAMESPACE (so avoid to keep it NULL and avoid numeric values).
+with the value of XXH_NAMESPACE (therefore, avoid NULL and numeric values).
 
 Note that no change is required within the calling program as long as it includes `xxhash.h` :
 regular symbol name will be automatically translated by this header.
@@ -122,24 +130,24 @@
 #ifdef XXH_NAMESPACE
 #  define XXH_CAT(A,B) A##B
 #  define XXH_NAME2(A,B) XXH_CAT(A,B)
-#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
-#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
 #  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
 #  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
-#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
 #  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
-#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
 #  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
-#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
 #  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
-#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
 #  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
-#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
 #  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
-#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
 #  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
-#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
 #  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
 #  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
 #endif
 
@@ -154,53 +162,28 @@
 XXH_PUBLIC_API unsigned XXH_versionNumber (void);
 
 
-/* ****************************
-*  Simple Hash Functions
-******************************/
+/*-**********************************************************************
+*  32-bits hash
+************************************************************************/
 typedef unsigned int       XXH32_hash_t;
-typedef unsigned long long XXH64_hash_t;
 
-XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, unsigned int seed);
-XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed);
-
-/*!
-XXH32() :
+/*! XXH32() :
     Calculate the 32-bits hash of sequence "length" bytes stored at memory address "input".
     The memory between input & input+length must be valid (allocated and read-accessible).
     "seed" can be used to alter the result predictably.
-    Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
-XXH64() :
-    Calculate the 64-bits hash of sequence of length "len" stored at memory address "input".
-    "seed" can be used to alter the result predictably.
-    This function runs 2x faster on 64-bits systems, but slower on 32-bits systems (see benchmark).
-*/
+    Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, unsigned int seed);
 
-
-/* ****************************
-*  Streaming Hash Functions
-******************************/
+/*======   Streaming   ======*/
 typedef struct XXH32_state_s XXH32_state_t;   /* incomplete type */
-typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
-
-/*! State allocation, compatible with dynamic libraries */
-
 XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
 XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
-
-XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
-XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
-
-
-/* hash streaming */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dst_state, const XXH32_state_t* restrict src_state);
 
 XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, unsigned int seed);
 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
 XXH_PUBLIC_API XXH32_hash_t  XXH32_digest (const XXH32_state_t* statePtr);
 
-XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, unsigned long long seed);
-XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
-XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
-
 /*
 These functions generate the xxHash of an input provided in multiple segments.
 Note that, for small input, they are slower than single-call functions, due to state management.
@@ -223,29 +206,11 @@
 When done, free XXH state space if it was allocated dynamically.
 */
 
+/*======   Canonical representation   ======*/
 
-/* **************************
-*  Utils
-****************************/
-#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* ! C99 */
-#  define restrict   /* disable restrict */
-#endif
-
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dst_state, const XXH32_state_t* restrict src_state);
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state);
-
-
-/* **************************
-*  Canonical representation
-****************************/
 typedef struct { unsigned char digest[4]; } XXH32_canonical_t;
-typedef struct { unsigned char digest[8]; } XXH64_canonical_t;
-
 XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
-
 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
-XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
 
 /* Default result type for XXH functions are primitive unsigned 32 and 64 bits.
 *  The canonical representation uses human-readable write convention, aka big-endian (large digits first).
@@ -254,6 +219,36 @@
 */
 
 
+#ifndef XXH_NO_LONG_LONG
+/*-**********************************************************************
+*  64-bits hash
+************************************************************************/
+typedef unsigned long long XXH64_hash_t;
+
+/*! XXH64() :
+    Calculate the 64-bits hash of sequence of length "len" stored at memory address "input".
+    "seed" can be used to alter the result predictably.
+    This function runs faster on 64-bits systems, but slower on 32-bits systems (see benchmark).
+*/
+XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed);
+
+/*======   Streaming   ======*/
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state);
+
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, unsigned long long seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+
+/*======   Canonical representation   ======*/
+typedef struct { unsigned char digest[8]; } XXH64_canonical_t;
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
+#endif  /* XXH_NO_LONG_LONG */
+
+
 #ifdef XXH_STATIC_LINKING_ONLY
 
 /* ================================================================================================
@@ -279,6 +274,7 @@
        unsigned reserved;   /* never read nor write, will be removed in a future version */
    };   /* typedef'd to XXH32_state_t */
 
+#ifndef XXH_NO_LONG_LONG
    struct XXH64_state_s {
        unsigned long long total_len;
        unsigned long long v1;
@@ -289,10 +285,10 @@
        unsigned memsize;
        unsigned reserved[2];          /* never read nor write, will be removed in a future version */
    };   /* typedef'd to XXH64_state_t */
-
+#endif
 
 #  ifdef XXH_PRIVATE_API
-#    include "xxhash.c"   /* include xxhash functions as `static`, for inlining */
+#    include "xxhash.c"   /* include xxhash function bodies as `static`, for inlining */
 #  endif
 
 #endif /* XXH_STATIC_LINKING_ONLY */
diff --git a/xxhsum.c b/xxhsum.c
index bbe9400..049bc72 100644
--- a/xxhsum.c
+++ b/xxhsum.c
@@ -364,11 +364,11 @@
 }
 
 
-static void BMK_testSequence64(void* sentence, int len, U64 seed, U64 Nresult)
+static void BMK_testSequence64(void* sentence, size_t len, U64 seed, U64 Nresult)
 {
     XXH64_state_t state;
     U64 Dresult;
-    int index;
+    size_t pos;
 
     Dresult = XXH64(sentence, len, seed);
     BMK_checkResult64(Dresult, Nresult);
@@ -379,7 +379,7 @@
     BMK_checkResult64(Dresult, Nresult);
 
     XXH64_reset(&state, seed);
-    for (index=0; index<len; index++) XXH64_update(&state, ((char*)sentence)+index, 1);
+    for (pos=0; pos<len; pos++) XXH64_update(&state, ((char*)sentence)+pos, 1);
     Dresult = XXH64_digest(&state);
     BMK_checkResult64(Dresult, Nresult);
 }
@@ -389,7 +389,7 @@
 {
     XXH32_state_t state;
     U32 Dresult;
-    size_t index;
+    size_t pos;
 
     Dresult = XXH32(sequence, len, seed);
     BMK_checkResult(Dresult, Nresult);
@@ -400,7 +400,7 @@
     BMK_checkResult(Dresult, Nresult);
 
     XXH32_reset(&state, seed);
-    for (index=0; index<len; index++) XXH32_update(&state, ((const char*)sequence)+index, 1);
+    for (pos=0; pos<len; pos++) XXH32_update(&state, ((const char*)sequence)+pos, 1);
     Dresult = XXH32_digest(&state);
     BMK_checkResult(Dresult, Nresult);
 }
@@ -449,17 +449,17 @@
 static void BMK_display_LittleEndian(const void* ptr, size_t length)
 {
     const BYTE* p = (const BYTE*)ptr;
-    size_t index;
-    for (index=length-1; index<length; index--)    /* intentional underflow to negative to detect end */
-        DISPLAYRESULT("%02x", p[index]);
+    size_t idx;
+    for (idx=length-1; idx<length; idx--)    /* intentional underflow to negative to detect end */
+        DISPLAYRESULT("%02x", p[idx]);
 }
 
 static void BMK_display_BigEndian(const void* ptr, size_t length)
 {
     const BYTE* p = (const BYTE*)ptr;
-    size_t index;
-    for (index=0; index<length; index++)
-        DISPLAYRESULT("%02x", p[index]);
+    size_t idx;
+    for (idx=0; idx<length; idx++)
+        DISPLAYRESULT("%02x", p[idx]);
 }
 
 static void BMK_hashStream(void* xxhHashValue, const algoType hashType, FILE* inFile, void* buffer, size_t blockSize)