Updated : xxHash to r37

4 files changed, 360 insertions, 240 deletions

diff --git a/NEWS b/NEWS
index 41d5762..7edee54 100644
--- a/NEWS
+++ b/NEWS
@@ -1,3 +1,7 @@
+r124:
+Fix : LZ4F_compressBound() using NULL preferencesPtr
+Updated : xxHash, to r37
+
 r123:
 Added : experimental lz4frame API, thanks to Takayuki Matsuoka and Christopher Jackson for testings
 Fix : s390x support, thanks to Nobuhiro Iwamatsu
diff --git a/programs/lz4io.c b/programs/lz4io.c
index 20520d6..6de5206 100644
--- a/programs/lz4io.c
+++ b/programs/lz4io.c
@@ -409,7 +409,7 @@ static int compress_file_blockDependency(char* input_filename, char* output_file
     clock_t start, end;
     unsigned int blockSize, inputBufferSize;
     size_t sizeCheck, header_size;
-    void* streamChecksumState=NULL;
+    XXH32_state_t streamCRC;
 
     // Init
     start = clock();
@@ -440,7 +440,7 @@ static int compress_file_blockDependency(char* input_filename, char* output_file
     if (!in_buff || !out_buff) EXM_THROW(31, "Allocation error : not enough memory");
     in_blockStart = in_buff + 64 KB;
     if (compressionlevel>=3) in_blockStart = in_buff;
-    if (streamChecksum) streamChecksumState = XXH32_init(LZ4S_CHECKSUM_SEED);
+    if (streamChecksum) XXH32_reset(&streamCRC, LZ4S_CHECKSUM_SEED);
     ctx = initFunction(in_buff);
 
     // Write Archive Header
@@ -469,7 +469,7 @@ static int compress_file_blockDependency(char* input_filename, char* output_file
         if( inSize==0 ) break;   // No more input : end of compression
         filesize += inSize;
         DISPLAYLEVEL(3, "\rRead : %i MB   ", (int)(filesize>>20));
-        if (streamChecksum) XXH32_update(streamChecksumState, in_blockStart, inSize);
+        if (streamChecksum) XXH32_update(&streamCRC, in_blockStart, inSize);
 
         // Compress Block
         outSize = compressionFunction(ctx, in_blockStart, out_buff+4, inSize, inSize-1, compressionlevel);
@@ -521,7 +521,7 @@ static int compress_file_blockDependency(char* input_filename, char* output_file
     compressedfilesize += 4;
     if (streamChecksum)
     {
-        unsigned int checksum = XXH32_digest(streamChecksumState);
+        unsigned int checksum = XXH32_digest(&streamCRC);
         * (unsigned int*) out_buff = LITTLE_ENDIAN_32(checksum);
         sizeCheck = fwrite(out_buff, 1, 4, foutput);
         if (sizeCheck!=(size_t)(4)) EXM_THROW(37, "Write error : cannot write stream checksum");
@@ -566,7 +566,7 @@ int LZ4IO_compressFilename(char* input_filename, char* output_filename, int comp
     clock_t start, end;
     int blockSize;
     size_t sizeCheck, header_size, readSize;
-    void* streamChecksumState=NULL;
+    XXH32_state_t streamCRC;
 
     // Branch out
     if (blockIndependence==0) return compress_file_blockDependency(input_filename, output_filename, compressionLevel);
@@ -584,7 +584,7 @@ int LZ4IO_compressFilename(char* input_filename, char* output_filename, int comp
     out_buff = (char*)malloc(blockSize+CACHELINE);
     headerBuffer = (char*)malloc(LZ4S_MAXHEADERSIZE);
     if (!in_buff || !out_buff || !(headerBuffer)) EXM_THROW(31, "Allocation error : not enough memory");
-    if (streamChecksum) streamChecksumState = XXH32_init(LZ4S_CHECKSUM_SEED);
+    if (streamChecksum) XXH32_reset(&streamCRC, LZ4S_CHECKSUM_SEED);
 
     // Write Archive Header
     *(unsigned int*)headerBuffer = LITTLE_ENDIAN_32(LZ4S_MAGICNUMBER);   // Magic Number, in Little Endian convention
@@ -613,7 +613,7 @@ int LZ4IO_compressFilename(char* input_filename, char* output_filename, int comp
 
         filesize += readSize;
         DISPLAYLEVEL(3, "\rRead : %i MB   ", (int)(filesize>>20));
-        if (streamChecksum) XXH32_update(streamChecksumState, in_buff, (int)readSize);
+        if (streamChecksum) XXH32_update(&streamCRC, in_buff, (int)readSize);
 
         // Compress Block
         outSize = compressionFunction(in_buff, out_buff+4, (int)readSize, (int)readSize-1, compressionLevel);
@@ -662,8 +662,8 @@ int LZ4IO_compressFilename(char* input_filename, char* output_filename, int comp
     compressedfilesize += 4;
     if (streamChecksum)
     {
-        unsigned int checksum = XXH32_digest(streamChecksumState);
-        * (unsigned int*) out_buff = LITTLE_ENDIAN_32(checksum);
+        unsigned int checksum = XXH32_digest(&streamCRC);
+        *(unsigned int*) out_buff = LITTLE_ENDIAN_32(checksum);
         sizeCheck = fwrite(out_buff, 1, 4, foutput);
         if (sizeCheck!=(size_t)(4)) EXM_THROW(37, "Write error : cannot write stream checksum");
         compressedfilesize += 4;
@@ -755,7 +755,7 @@ static unsigned long long decodeLZ4S(FILE* finput, FILE* foutput)
     unsigned int maxBlockSize;
     size_t sizeCheck;
     int blockChecksumFlag, streamChecksumFlag, blockIndependenceFlag;
-    void* streamChecksumState=NULL;
+    XXH32_state_t streamCRC;
     int (*decompressionFunction)(LZ4_streamDecode_t* ctx, const char* src, char* dst, int cSize, int maxOSize) = LZ4_decompress_safe_continue;
     LZ4_streamDecode_t ctx;
 
@@ -805,7 +805,7 @@ static unsigned long long decodeLZ4S(FILE* finput, FILE* foutput)
         out_start = out_buff;
         out_end = out_start + outBuffSize;
         if (!in_buff || !out_buff) EXM_THROW(70, "Allocation error : not enough memory");
-        if (streamChecksumFlag) streamChecksumState = XXH32_init(LZ4S_CHECKSUM_SEED);
+        if (streamChecksumFlag) XXH32_reset(&streamCRC, LZ4S_CHECKSUM_SEED);
     }
 
     // Main Loop
@@ -843,7 +843,7 @@ static unsigned long long decodeLZ4S(FILE* finput, FILE* foutput)
             sizeCheck = fwrite(in_buff, 1, blockSize, foutput);
             if (sizeCheck != (size_t)blockSize) EXM_THROW(76, "Write error : cannot write data block");
             filesize += blockSize;
-            if (streamChecksumFlag) XXH32_update(streamChecksumState, in_buff, blockSize);
+            if (streamChecksumFlag) XXH32_update(&streamCRC, in_buff, blockSize);
             if (!blockIndependenceFlag)
             {
                 // handle dictionary for streaming
@@ -859,7 +859,7 @@ static unsigned long long decodeLZ4S(FILE* finput, FILE* foutput)
             decodedBytes = decompressionFunction(&ctx, in_buff, out_start, blockSize, maxBlockSize);
             if (decodedBytes < 0) EXM_THROW(77, "Decoding Failed ! Corrupted input detected !");
             filesize += decodedBytes;
-            if (streamChecksumFlag) XXH32_update(streamChecksumState, out_start, decodedBytes);
+            if (streamChecksumFlag) XXH32_update(&streamCRC, out_start, decodedBytes);
 
             // Write Block
             sizeCheck = fwrite(out_start, 1, decodedBytes, foutput);
@@ -872,7 +872,7 @@ static unsigned long long decodeLZ4S(FILE* finput, FILE* foutput)
     // Stream Checksum
     if (streamChecksumFlag)
     {
-        unsigned int checksum = XXH32_digest(streamChecksumState);
+        unsigned int checksum = XXH32_digest(&streamCRC);
         unsigned int readChecksum;
         sizeCheck = fread(&readChecksum, 1, 4, finput);
         if (sizeCheck != 4) EXM_THROW(74, "Read error : cannot read stream checksum");
diff --git a/xxhash.c b/xxhash.c
index be48370..529d69c 100644
--- a/xxhash.c
+++ b/xxhash.c
@@ -28,6 +28,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 You can contact the author at :
 - xxHash source repository : http://code.google.com/p/xxhash/
+- public discussion board : https://groups.google.com/forum/#!forum/lz4c
 */
 
 
@@ -80,14 +81,23 @@ You can contact the author at :
 // Includes & Memory related functions
 //**************************************
 #include "xxhash.h"
-// Modify the local functions below should you wish to use some other memory related routines
+// Modify the local functions below should you wish to use some other memory routines
 // for malloc(), free()
 #include <stdlib.h>
-FORCE_INLINE void* XXH_malloc(size_t s) { return malloc(s); }
-FORCE_INLINE void  XXH_free  (void* p)  { free(p); }
+FORCE_INLINE void* XXH_malloc(size_t s)
+{
+    return malloc(s);
+}
+FORCE_INLINE void  XXH_free  (void* p)
+{
+    free(p);
+}
 // for memcpy()
 #include <string.h>
-FORCE_INLINE void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
+FORCE_INLINE void* XXH_memcpy(void* dest, const void* src, size_t size)
+{
+    return memcpy(dest,src,size);
+}
 
 
 //**************************************
@@ -95,17 +105,17 @@ FORCE_INLINE void* XXH_memcpy(void* dest, const void* src, size_t size) { return
 //**************************************
 #if 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;
+typedef uint8_t  BYTE;
+typedef uint16_t U16;
+typedef uint32_t U32;
+typedef  int32_t S32;
+typedef uint64_t U64;
 #else
-  typedef unsigned char      BYTE;
-  typedef unsigned short     U16;
-  typedef unsigned int       U32;
-  typedef   signed int       S32;
-  typedef unsigned long long U64;
+typedef unsigned char      BYTE;
+typedef unsigned short     U16;
+typedef unsigned int       U32;
+typedef   signed int       S32;
+typedef unsigned long long U64;
 #endif
 
 #if defined(__GNUC__)  && !defined(XXH_USE_UNALIGNED_ACCESS)
@@ -122,8 +132,14 @@ FORCE_INLINE void* XXH_memcpy(void* dest, const void* src, size_t size) { return
 #  endif
 #endif
 
-typedef struct _U32_S { U32 v; } _PACKED U32_S;
-typedef struct _U64_S { U64 v; } _PACKED U64_S;
+typedef struct _U32_S
+{
+    U32 v;
+} _PACKED U32_S;
+typedef struct _U64_S
+{
+    U64 v;
+} _PACKED U64_S;
 
 #if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__)
 #  pragma pack(pop)
@@ -154,12 +170,15 @@ typedef struct _U64_S { U64 v; } _PACKED U64_S;
 #  define XXH_swap32 __builtin_bswap32
 #  define XXH_swap64 __builtin_bswap64
 #else
-static inline U32 XXH_swap32 (U32 x) {
+static inline U32 XXH_swap32 (U32 x)
+{
     return  ((x << 24) & 0xff000000 ) |
-			((x <<  8) & 0x00ff0000 ) |
-			((x >>  8) & 0x0000ff00 ) |
-			((x >> 24) & 0x000000ff );}
-static inline U64 XXH_swap64 (U64 x) {
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+static inline U64 XXH_swap64 (U64 x)
+{
     return  ((x << 56) & 0xff00000000000000ULL) |
             ((x << 40) & 0x00ff000000000000ULL) |
             ((x << 24) & 0x0000ff0000000000ULL) |
@@ -167,7 +186,8 @@ static inline U64 XXH_swap64 (U64 x) {
             ((x >> 8)  & 0x00000000ff000000ULL) |
             ((x >> 24) & 0x0000000000ff0000ULL) |
             ((x >> 40) & 0x000000000000ff00ULL) |
-            ((x >> 56) & 0x00000000000000ffULL);}
+            ((x >> 56) & 0x00000000000000ffULL);
+}
 #endif
 
 
@@ -191,7 +211,7 @@ static inline U64 XXH_swap64 (U64 x) {
 //**************************************
 typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
 #ifndef XXH_CPU_LITTLE_ENDIAN   // It is possible to define XXH_CPU_LITTLE_ENDIAN externally, for example using a compiler switch
-    static const int one = 1;
+static const int one = 1;
 #   define XXH_CPU_LITTLE_ENDIAN   (*(char*)(&one))
 #endif
 
@@ -215,7 +235,10 @@ FORCE_INLINE U32 XXH_readLE32_align(const U32* ptr, XXH_endianess endian, XXH_al
         return endian==XXH_littleEndian ? *ptr : XXH_swap32(*ptr);
 }
 
-FORCE_INLINE U32 XXH_readLE32(const U32* ptr, XXH_endianess endian) { return XXH_readLE32_align(ptr, endian, XXH_unaligned); }
+FORCE_INLINE U32 XXH_readLE32(const U32* ptr, XXH_endianess endian)
+{
+    return XXH_readLE32_align(ptr, endian, XXH_unaligned);
+}
 
 FORCE_INLINE U64 XXH_readLE64_align(const U64* ptr, XXH_endianess endian, XXH_alignment align)
 {
@@ -225,13 +248,16 @@ FORCE_INLINE U64 XXH_readLE64_align(const U64* ptr, XXH_endianess endian, XXH_al
         return endian==XXH_littleEndian ? *ptr : XXH_swap64(*ptr);
 }
 
-FORCE_INLINE U64 XXH_readLE64(const U64* ptr, XXH_endianess endian) { return XXH_readLE64_align(ptr, endian, XXH_unaligned); }
+FORCE_INLINE U64 XXH_readLE64(const U64* ptr, XXH_endianess endian)
+{
+    return XXH_readLE64_align(ptr, endian, XXH_unaligned);
+}
 
 
 //****************************
 // Simple Hash Functions
 //****************************
-FORCE_INLINE U32 XXH32_endian_align(const void* input, unsigned int len, U32 seed, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* bEnd = p + len;
@@ -239,7 +265,11 @@ FORCE_INLINE U32 XXH32_endian_align(const void* input, unsigned int len, U32 see
 #define XXH_get32bits(p) XXH_readLE32_align((const U32*)p, endian, align)
 
 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (p==NULL) { len=0; bEnd=p=(const BYTE*)(size_t)16; }
+    if (p==NULL)
+    {
+        len=0;
+        bEnd=p=(const BYTE*)(size_t)16;
+    }
 #endif
 
     if (len>=16)
@@ -252,11 +282,24 @@ FORCE_INLINE U32 XXH32_endian_align(const void* input, unsigned int len, U32 see
 
         do
         {
-            v1 += XXH_get32bits(p) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4;
-            v2 += XXH_get32bits(p) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4;
-            v3 += XXH_get32bits(p) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4;
-            v4 += XXH_get32bits(p) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4;
-        } while (p<=limit);
+            v1 += XXH_get32bits(p) * PRIME32_2;
+            v1 = XXH_rotl32(v1, 13);
+            v1 *= PRIME32_1;
+            p+=4;
+            v2 += XXH_get32bits(p) * PRIME32_2;
+            v2 = XXH_rotl32(v2, 13);
+            v2 *= PRIME32_1;
+            p+=4;
+            v3 += XXH_get32bits(p) * PRIME32_2;
+            v3 = XXH_rotl32(v3, 13);
+            v3 *= PRIME32_1;
+            p+=4;
+            v4 += XXH_get32bits(p) * PRIME32_2;
+            v4 = XXH_rotl32(v4, 13);
+            v4 *= PRIME32_1;
+            p+=4;
+        }
+        while (p<=limit);
 
         h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
     }
@@ -291,13 +334,14 @@ FORCE_INLINE U32 XXH32_endian_align(const void* input, unsigned int len, U32 see
 }
 
 
-U32 XXH32(const void* input, unsigned int len, U32 seed)
+unsigned int XXH32 (const void* input, size_t len, unsigned seed)
 {
 #if 0
     // Simple version, good for code maintenance, but unfortunately slow for small inputs
-    void* state = XXH32_init(seed);
-    XXH32_update(state, input, len);
-    return XXH32_digest(state);
+    XXH32_state_t state;
+    XXH32_reset(&state, seed);
+    XXH32_update(&state, input, len);
+    return XXH32_digest(&state);
 #else
     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 
@@ -318,7 +362,7 @@ U32 XXH32(const void* input, unsigned int len, U32 seed)
 #endif
 }
 
-FORCE_INLINE U64 XXH64_endian_align(const void* input, unsigned int len, U64 seed, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* bEnd = p + len;
@@ -326,7 +370,11 @@ FORCE_INLINE U64 XXH64_endian_align(const void* input, unsigned int len, U64 see
 #define XXH_get64bits(p) XXH_readLE64_align((const U64*)p, endian, align)
 
 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (p==NULL) { len=0; bEnd=p=(const BYTE*)(size_t)32; }
+    if (p==NULL)
+    {
+        len=0;
+        bEnd=p=(const BYTE*)(size_t)32;
+    }
 #endif
 
     if (len>=32)
@@ -339,25 +387,50 @@ FORCE_INLINE U64 XXH64_endian_align(const void* input, unsigned int len, U64 see
 
         do
         {
-            v1 += XXH_get64bits(p) * PRIME64_2; p+=8; v1 = XXH_rotl64(v1, 31); v1 *= PRIME64_1;
-            v2 += XXH_get64bits(p) * PRIME64_2; p+=8; v2 = XXH_rotl64(v2, 31); v2 *= PRIME64_1;
-            v3 += XXH_get64bits(p) * PRIME64_2; p+=8; v3 = XXH_rotl64(v3, 31); v3 *= PRIME64_1;
-            v4 += XXH_get64bits(p) * PRIME64_2; p+=8; v4 = XXH_rotl64(v4, 31); v4 *= PRIME64_1;
-        } while (p<=limit);
+            v1 += XXH_get64bits(p) * PRIME64_2;
+            p+=8;
+            v1 = XXH_rotl64(v1, 31);
+            v1 *= PRIME64_1;
+            v2 += XXH_get64bits(p) * PRIME64_2;
+            p+=8;
+            v2 = XXH_rotl64(v2, 31);
+            v2 *= PRIME64_1;
+            v3 += XXH_get64bits(p) * PRIME64_2;
+            p+=8;
+            v3 = XXH_rotl64(v3, 31);
+            v3 *= PRIME64_1;
+            v4 += XXH_get64bits(p) * PRIME64_2;
+            p+=8;
+            v4 = XXH_rotl64(v4, 31);
+            v4 *= PRIME64_1;
+        }
+        while (p<=limit);
 
         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 
-        v1 *= PRIME64_2; v1 = XXH_rotl64(v1, 31); v1 *= PRIME64_1; h64 ^= v1;
-    	h64 = h64 * PRIME64_1 + PRIME64_4;
-
-        v2 *= PRIME64_2; v2 = XXH_rotl64(v2, 31); v2 *= PRIME64_1; h64 ^= v2;
-    	h64 = h64 * PRIME64_1 + PRIME64_4;
-
-        v3 *= PRIME64_2; v3 = XXH_rotl64(v3, 31); v3 *= PRIME64_1; h64 ^= v3;
-    	h64 = h64 * PRIME64_1 + PRIME64_4;
-
-        v4 *= PRIME64_2; v4 = XXH_rotl64(v4, 31); v4 *= PRIME64_1; h64 ^= v4;
-    	h64 = h64 * PRIME64_1 + PRIME64_4;
+        v1 *= PRIME64_2;
+        v1 = XXH_rotl64(v1, 31);
+        v1 *= PRIME64_1;
+        h64 ^= v1;
+        h64 = h64 * PRIME64_1 + PRIME64_4;
+
+        v2 *= PRIME64_2;
+        v2 = XXH_rotl64(v2, 31);
+        v2 *= PRIME64_1;
+        h64 ^= v2;
+        h64 = h64 * PRIME64_1 + PRIME64_4;
+
+        v3 *= PRIME64_2;
+        v3 = XXH_rotl64(v3, 31);
+        v3 *= PRIME64_1;
+        h64 ^= v3;
+        h64 = h64 * PRIME64_1 + PRIME64_4;
+
+        v4 *= PRIME64_2;
+        v4 = XXH_rotl64(v4, 31);
+        v4 *= PRIME64_1;
+        h64 ^= v4;
+        h64 = h64 * PRIME64_1 + PRIME64_4;
     }
     else
     {
@@ -368,22 +441,25 @@ FORCE_INLINE U64 XXH64_endian_align(const void* input, unsigned int len, U64 see
 
     while (p+8<=bEnd)
     {
-    	U64 k1 = XXH_get64bits(p);
-    	k1 *= PRIME64_2; k1 = XXH_rotl64(k1,31); k1 *= PRIME64_1; h64 ^= k1;
-    	h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
-    	p+=8;
+        U64 k1 = XXH_get64bits(p);
+        k1 *= PRIME64_2;
+        k1 = XXH_rotl64(k1,31);
+        k1 *= PRIME64_1;
+        h64 ^= k1;
+        h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
+        p+=8;
     }
 
     if (p+4<=bEnd)
     {
-    	h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
+        h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
         h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
         p+=4;
     }
 
     while (p<bEnd)
     {
-    	h64 ^= (*p) * PRIME64_5;
+        h64 ^= (*p) * PRIME64_5;
         h64 = XXH_rotl64(h64, 11) * PRIME64_1;
         p++;
     }
@@ -398,8 +474,15 @@ FORCE_INLINE U64 XXH64_endian_align(const void* input, unsigned int len, U64 see
 }
 
 
-unsigned long long XXH64(const void* input, unsigned int len, unsigned long long seed)
+unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
 {
+#if 0
+    // Simple version, good for code maintenance, but unfortunately slow for small inputs
+    XXH64_state_t state;
+    XXH64_reset(&state, seed);
+    XXH64_update(&state, input, len);
+    return XXH64_digest(&state);
+#else
     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 
 #  if !defined(XXH_USE_UNALIGNED_ACCESS)
@@ -416,13 +499,15 @@ unsigned long long XXH64(const void* input, unsigned int len, unsigned long long
         return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
     else
         return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
+#endif
 }
 
-//****************************
-// Advanced Hash Functions
-//****************************
+/****************************************************
+ *  Advanced Hash Functions
+****************************************************/
 
-struct XXH_state32_t
+/*** Allocation ***/
+typedef struct
 {
     U64 total_len;
     U32 seed;
@@ -430,11 +515,11 @@ struct XXH_state32_t
     U32 v2;
     U32 v3;
     U32 v4;
-    int memsize;
+    U32 memsize;
     char memory[16];
-};
+} XXH_istate32_t;
 
-struct XXH_state64_t
+typedef struct
 {
     U64 total_len;
     U64 seed;
@@ -442,27 +527,39 @@ struct XXH_state64_t
     U64 v2;
     U64 v3;
     U64 v4;
-    int memsize;
+    U32 memsize;
     char memory[32];
-};
+} XXH_istate64_t;
 
 
-int XXH32_sizeofState(void)
+XXH32_state_t* XXH32_createState(void)
 {
-    XXH_STATIC_ASSERT(XXH32_SIZEOFSTATE >= sizeof(struct XXH_state32_t));   // A compilation error here means XXH32_SIZEOFSTATE is not large enough
-    return sizeof(struct XXH_state32_t);
+    XXH_STATIC_ASSERT(sizeof(XXH32_state_t) >= sizeof(XXH_istate32_t));   // A compilation error here means XXH32_state_t is not large enough
+    return (XXH32_state_t*)malloc(sizeof(XXH32_state_t));
 }
+XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    free(statePtr);
+    return XXH_OK;
+};
 
-int XXH64_sizeofState(void)
+XXH64_state_t* XXH64_createState(void)
 {
-    XXH_STATIC_ASSERT(XXH64_SIZEOFSTATE >= sizeof(struct XXH_state64_t));   // A compilation error here means XXH64_SIZEOFSTATE is not large enough
-    return sizeof(struct XXH_state64_t);
+    XXH_STATIC_ASSERT(sizeof(XXH64_state_t) >= sizeof(XXH_istate64_t));   // A compilation error here means XXH64_state_t is not large enough
+    return (XXH64_state_t*)malloc(sizeof(XXH64_state_t));
 }
+XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
+{
+    free(statePtr);
+    return XXH_OK;
+};
+
 
+/*** Hash feed ***/
 
-XXH_errorcode XXH32_resetState(void* state_in, U32 seed)
+XXH_errorcode XXH32_reset(XXH32_state_t* state_in, U32 seed)
 {
-    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;
+    XXH_istate32_t* state = (XXH_istate32_t*) state_in;
     state->seed = seed;
     state->v1 = seed + PRIME32_1 + PRIME32_2;
     state->v2 = seed + PRIME32_2;
@@ -473,9 +570,9 @@ XXH_errorcode XXH32_resetState(void* state_in, U32 seed)
     return XXH_OK;
 }
 
-XXH_errorcode XXH64_resetState(void* state_in, unsigned long long seed)
+XXH_errorcode XXH64_reset(XXH64_state_t* state_in, unsigned long long seed)
 {
-    struct XXH_state64_t * state = (struct XXH_state64_t *) state_in;
+    XXH_istate64_t* state = (XXH_istate64_t*) state_in;
     state->seed = seed;
     state->v1 = seed + PRIME64_1 + PRIME64_2;
     state->v2 = seed + PRIME64_2;
@@ -487,24 +584,9 @@ XXH_errorcode XXH64_resetState(void* state_in, unsigned long long seed)
 }
 
 
-void* XXH32_init (U32 seed)
+FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state_in, const void* input, size_t len, XXH_endianess endian)
 {
-    void* state = XXH_malloc (sizeof(struct XXH_state32_t));
-    if (state != NULL) XXH32_resetState(state, seed);
-    return state;
-}
-
-void* XXH64_init (unsigned long long seed)
-{
-    void* state = XXH_malloc (sizeof(struct XXH_state64_t));
-    if (state != NULL) XXH64_resetState(state, seed);
-    return state;
-}
-
-
-FORCE_INLINE XXH_errorcode XXH32_update_endian (void* state_in, const void* input, int len, XXH_endianess endian)
-{
-    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;
+    XXH_istate32_t* state = (XXH_istate32_t *) state_in;
     const BYTE* p = (const BYTE*)input;
     const BYTE* const bEnd = p + len;
 
@@ -517,7 +599,7 @@ FORCE_INLINE XXH_errorcode XXH32_update_endian (void* state_in, const void* inpu
     if (state->memsize + len < 16)   // fill in tmp buffer
     {
         XXH_memcpy(state->memory + state->memsize, input, len);
-        state->memsize +=  len;
+        state->memsize += (U32)len;
         return XXH_OK;
     }
 
@@ -526,10 +608,22 @@ FORCE_INLINE XXH_errorcode XXH32_update_endian (void* state_in, const void* inpu
         XXH_memcpy(state->memory + state->memsize, input, 16-state->memsize);
         {
             const U32* p32 = (const U32*)state->memory;
-            state->v1 += XXH_readLE32(p32, endian) * PRIME32_2; state->v1 = XXH_rotl32(state->v1, 13); state->v1 *= PRIME32_1; p32++;
-            state->v2 += XXH_readLE32(p32, endian) * PRIME32_2; state->v2 = XXH_rotl32(state->v2, 13); state->v2 *= PRIME32_1; p32++;
-            state->v3 += XXH_readLE32(p32, endian) * PRIME32_2; state->v3 = XXH_rotl32(state->v3, 13); state->v3 *= PRIME32_1; p32++;
-            state->v4 += XXH_readLE32(p32, endian) * PRIME32_2; state->v4 = XXH_rotl32(state->v4, 13); state->v4 *= PRIME32_1; p32++;
+            state->v1 += XXH_readLE32(p32, endian) * PRIME32_2;
+            state->v1 = XXH_rotl32(state->v1, 13);
+            state->v1 *= PRIME32_1;
+            p32++;
+            state->v2 += XXH_readLE32(p32, endian) * PRIME32_2;
+            state->v2 = XXH_rotl32(state->v2, 13);
+            state->v2 *= PRIME32_1;
+            p32++;
+            state->v3 += XXH_readLE32(p32, endian) * PRIME32_2;
+            state->v3 = XXH_rotl32(state->v3, 13);
+            state->v3 *= PRIME32_1;
+            p32++;
+            state->v4 += XXH_readLE32(p32, endian) * PRIME32_2;
+            state->v4 = XXH_rotl32(state->v4, 13);
+            state->v4 *= PRIME32_1;
+            p32++;
         }
         p += 16-state->memsize;
         state->memsize = 0;
@@ -545,11 +639,24 @@ FORCE_INLINE XXH_errorcode XXH32_update_endian (void* state_in, const void* inpu
 
         do
         {
-            v1 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4;
-            v2 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4;
-            v3 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4;
-            v4 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4;
-        } while (p<=limit);
+            v1 += XXH_readLE32((const U32*)p, endian) * PRIME32_2;
+            v1 = XXH_rotl32(v1, 13);
+            v1 *= PRIME32_1;
+            p+=4;
+            v2 += XXH_readLE32((const U32*)p, endian) * PRIME32_2;
+            v2 = XXH_rotl32(v2, 13);
+            v2 *= PRIME32_1;
+            p+=4;
+            v3 += XXH_readLE32((const U32*)p, endian) * PRIME32_2;
+            v3 = XXH_rotl32(v3, 13);
+            v3 *= PRIME32_1;
+            p+=4;
+            v4 += XXH_readLE32((const U32*)p, endian) * PRIME32_2;
+            v4 = XXH_rotl32(v4, 13);
+            v4 *= PRIME32_1;
+            p+=4;
+        }
+        while (p<=limit);
 
         state->v1 = v1;
         state->v2 = v2;
@@ -566,7 +673,7 @@ FORCE_INLINE XXH_errorcode XXH32_update_endian (void* state_in, const void* inpu
     return XXH_OK;
 }
 
-XXH_errorcode XXH32_update (void* state_in, const void* input, unsigned int len)
+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;
 
@@ -578,9 +685,9 @@ XXH_errorcode XXH32_update (void* state_in, const void* input, unsigned int len)
 
 
 
-FORCE_INLINE U32 XXH32_intermediateDigest_endian (void* state_in, XXH_endianess endian)
+FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state_in, XXH_endianess endian)
 {
-    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;
+    XXH_istate32_t* state = (XXH_istate32_t*) state_in;
     const BYTE * p = (const BYTE*)state->memory;
     BYTE* bEnd = (BYTE*)state->memory + state->memsize;
     U32 h32;
@@ -620,30 +727,20 @@ FORCE_INLINE U32 XXH32_intermediateDigest_endian (void* state_in, XXH_endianess
 }
 
 
-U32 XXH32_intermediateDigest (void* state_in)
+U32 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_intermediateDigest_endian(state_in, XXH_littleEndian);
+        return XXH32_digest_endian(state_in, XXH_littleEndian);
     else
-        return XXH32_intermediateDigest_endian(state_in, XXH_bigEndian);
-}
-
-
-U32 XXH32_digest (void* state_in)
-{
-    U32 h32 = XXH32_intermediateDigest(state_in);
-
-    XXH_free(state_in);
-
-    return h32;
+        return XXH32_digest_endian(state_in, XXH_bigEndian);
 }
 
 
-FORCE_INLINE XXH_errorcode XXH64_update_endian (void* state_in, const void* input, int len, XXH_endianess endian)
+FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state_in, const void* input, size_t len, XXH_endianess endian)
 {
-    struct XXH_state64_t * state = (struct XXH_state64_t *) state_in;
+    XXH_istate64_t * state = (XXH_istate64_t *) state_in;
     const BYTE* p = (const BYTE*)input;
     const BYTE* const bEnd = p + len;
 
@@ -656,7 +753,7 @@ FORCE_INLINE XXH_errorcode XXH64_update_endian (void* state_in, const void* inpu
     if (state->memsize + len < 32)   // fill in tmp buffer
     {
         XXH_memcpy(state->memory + state->memsize, input, len);
-        state->memsize +=  len;
+        state->memsize += (U32)len;
         return XXH_OK;
     }
 
@@ -665,10 +762,22 @@ FORCE_INLINE XXH_errorcode XXH64_update_endian (void* state_in, const void* inpu
         XXH_memcpy(state->memory + state->memsize, input, 32-state->memsize);
         {
             const U64* p64 = (const U64*)state->memory;
-            state->v1 += XXH_readLE64(p64, endian) * PRIME64_2; state->v1 = XXH_rotl64(state->v1, 31); state->v1 *= PRIME64_1; p64++;
-            state->v2 += XXH_readLE64(p64, endian) * PRIME64_2; state->v2 = XXH_rotl64(state->v2, 31); state->v2 *= PRIME64_1; p64++;
-            state->v3 += XXH_readLE64(p64, endian) * PRIME64_2; state->v3 = XXH_rotl64(state->v3, 31); state->v3 *= PRIME64_1; p64++;
-            state->v4 += XXH_readLE64(p64, endian) * PRIME64_2; state->v4 = XXH_rotl64(state->v4, 31); state->v4 *= PRIME64_1; p64++;
+            state->v1 += XXH_readLE64(p64, endian) * PRIME64_2;
+            state->v1 = XXH_rotl64(state->v1, 31);
+            state->v1 *= PRIME64_1;
+            p64++;
+            state->v2 += XXH_readLE64(p64, endian) * PRIME64_2;
+            state->v2 = XXH_rotl64(state->v2, 31);
+            state->v2 *= PRIME64_1;
+            p64++;
+            state->v3 += XXH_readLE64(p64, endian) * PRIME64_2;
+            state->v3 = XXH_rotl64(state->v3, 31);
+            state->v3 *= PRIME64_1;
+            p64++;
+            state->v4 += XXH_readLE64(p64, endian) * PRIME64_2;
+            state->v4 = XXH_rotl64(state->v4, 31);
+            state->v4 *= PRIME64_1;
+            p64++;
         }
         p += 32-state->memsize;
         state->memsize = 0;
@@ -684,11 +793,24 @@ FORCE_INLINE XXH_errorcode XXH64_update_endian (void* state_in, const void* inpu
 
         do
         {
-            v1 += XXH_readLE64((const U64*)p, endian) * PRIME64_2; v1 = XXH_rotl64(v1, 31); v1 *= PRIME64_1; p+=8;
-            v2 += XXH_readLE64((const U64*)p, endian) * PRIME64_2; v2 = XXH_rotl64(v2, 31); v2 *= PRIME64_1; p+=8;
-            v3 += XXH_readLE64((const U64*)p, endian) * PRIME64_2; v3 = XXH_rotl64(v3, 31); v3 *= PRIME64_1; p+=8;
-            v4 += XXH_readLE64((const U64*)p, endian) * PRIME64_2; v4 = XXH_rotl64(v4, 31); v4 *= PRIME64_1; p+=8;
-        } while (p<=limit);
+            v1 += XXH_readLE64((const U64*)p, endian) * PRIME64_2;
+            v1 = XXH_rotl64(v1, 31);
+            v1 *= PRIME64_1;
+            p+=8;
+            v2 += XXH_readLE64((const U64*)p, endian) * PRIME64_2;
+            v2 = XXH_rotl64(v2, 31);
+            v2 *= PRIME64_1;
+            p+=8;
+            v3 += XXH_readLE64((const U64*)p, endian) * PRIME64_2;
+            v3 = XXH_rotl64(v3, 31);
+            v3 *= PRIME64_1;
+            p+=8;
+            v4 += XXH_readLE64((const U64*)p, endian) * PRIME64_2;
+            v4 = XXH_rotl64(v4, 31);
+            v4 *= PRIME64_1;
+            p+=8;
+        }
+        while (p<=limit);
 
         state->v1 = v1;
         state->v2 = v2;
@@ -705,7 +827,7 @@ FORCE_INLINE XXH_errorcode XXH64_update_endian (void* state_in, const void* inpu
     return XXH_OK;
 }
 
-XXH_errorcode XXH64_update (void* state_in, const void* input, unsigned int len)
+XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
 {
     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 
@@ -717,33 +839,45 @@ XXH_errorcode XXH64_update (void* state_in, const void* input, unsigned int len)
 
 
 
-FORCE_INLINE U64 XXH64_intermediateDigest_endian (void* state_in, XXH_endianess endian)
+FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state_in, XXH_endianess endian)
 {
-    struct XXH_state64_t * state = (struct XXH_state64_t *) state_in;
+    XXH_istate64_t * state = (XXH_istate64_t *) state_in;
     const BYTE * p = (const BYTE*)state->memory;
     BYTE* bEnd = (BYTE*)state->memory + state->memsize;
     U64 h64;
 
     if (state->total_len >= 32)
     {
-    	U64 v1 = state->v1;
-    	U64 v2 = state->v2;
-    	U64 v3 = state->v3;
-    	U64 v4 = state->v4;
+        U64 v1 = state->v1;
+        U64 v2 = state->v2;
+        U64 v3 = state->v3;
+        U64 v4 = state->v4;
 
         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 
-        v1 *= PRIME64_2; v1 = XXH_rotl64(v1, 31); v1 *= PRIME64_1; h64 ^= v1;
-    	h64 = h64*PRIME64_1 + PRIME64_4;
-
-        v2 *= PRIME64_2; v2 = XXH_rotl64(v2, 31); v2 *= PRIME64_1; h64 ^= v2;
-    	h64 = h64*PRIME64_1 + PRIME64_4;
-
-        v3 *= PRIME64_2; v3 = XXH_rotl64(v3, 31); v3 *= PRIME64_1; h64 ^= v3;
-    	h64 = h64*PRIME64_1 + PRIME64_4;
-
-        v4 *= PRIME64_2; v4 = XXH_rotl64(v4, 31); v4 *= PRIME64_1; h64 ^= v4;
-    	h64 = h64*PRIME64_1 + PRIME64_4;
+        v1 *= PRIME64_2;
+        v1 = XXH_rotl64(v1, 31);
+        v1 *= PRIME64_1;
+        h64 ^= v1;
+        h64 = h64*PRIME64_1 + PRIME64_4;
+
+        v2 *= PRIME64_2;
+        v2 = XXH_rotl64(v2, 31);
+        v2 *= PRIME64_1;
+        h64 ^= v2;
+        h64 = h64*PRIME64_1 + PRIME64_4;
+
+        v3 *= PRIME64_2;
+        v3 = XXH_rotl64(v3, 31);
+        v3 *= PRIME64_1;
+        h64 ^= v3;
+        h64 = h64*PRIME64_1 + PRIME64_4;
+
+        v4 *= PRIME64_2;
+        v4 = XXH_rotl64(v4, 31);
+        v4 *= PRIME64_1;
+        h64 ^= v4;
+        h64 = h64*PRIME64_1 + PRIME64_4;
     }
     else
     {
@@ -754,22 +888,25 @@ FORCE_INLINE U64 XXH64_intermediateDigest_endian (void* state_in, XXH_endianess
 
     while (p+8<=bEnd)
     {
-    	U64 k1 = XXH_readLE64((const U64*)p, endian);
-    	k1 *= PRIME64_2; k1 = XXH_rotl64(k1,31); k1 *= PRIME64_1; h64 ^= k1;
-    	h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
-    	p+=8;
+        U64 k1 = XXH_readLE64((const U64*)p, endian);
+        k1 *= PRIME64_2;
+        k1 = XXH_rotl64(k1,31);
+        k1 *= PRIME64_1;
+        h64 ^= k1;
+        h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
+        p+=8;
     }
 
     if (p+4<=bEnd)
     {
-    	h64 ^= (U64)(XXH_readLE32((const U32*)p, endian)) * PRIME64_1;
+        h64 ^= (U64)(XXH_readLE32((const U32*)p, endian)) * PRIME64_1;
         h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
         p+=4;
     }
 
     while (p<bEnd)
     {
-    	h64 ^= (*p) * PRIME64_5;
+        h64 ^= (*p) * PRIME64_5;
         h64 = XXH_rotl64(h64, 11) * PRIME64_1;
         p++;
     }
@@ -784,23 +921,14 @@ FORCE_INLINE U64 XXH64_intermediateDigest_endian (void* state_in, XXH_endianess
 }
 
 
-unsigned long long XXH64_intermediateDigest (void* state_in)
+unsigned long long XXH64_digest (const XXH64_state_t* state_in)
 {
     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 
     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH64_intermediateDigest_endian(state_in, XXH_littleEndian);
+        return XXH64_digest_endian(state_in, XXH_littleEndian);
     else
-        return XXH64_intermediateDigest_endian(state_in, XXH_bigEndian);
+        return XXH64_digest_endian(state_in, XXH_bigEndian);
 }
 
 
-unsigned long long XXH64_digest (void* state_in)
-{
-    U64 h64 = XXH64_intermediateDigest(state_in);
-
-    XXH_free(state_in);
-
-    return h64;
-}
-
diff --git a/xxhash.h b/xxhash.h
index 4311485..55b4501 100644
--- a/xxhash.h
+++ b/xxhash.h
@@ -65,6 +65,12 @@ extern "C" {
 
 
 /*****************************
+   Includes
+*****************************/
+#include <stddef.h>   /* size_t */
+
+
+/*****************************
    Type
 *****************************/
 typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
@@ -75,18 +81,16 @@ typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
    Simple Hash Functions
 *****************************/
 
-unsigned int       XXH32 (const void* input, unsigned int len, unsigned int seed);
-unsigned long long XXH64 (const void* input, unsigned int len, unsigned long long seed);
+unsigned int       XXH32 (const void* input, size_t length, unsigned seed);
+unsigned long long XXH64 (const void* input, size_t length, unsigned long long seed);
 
 /*
 XXH32() :
-    Calculate the 32-bits hash of sequence of length "len" stored at memory address "input".
-    The memory between input & input+len must be valid (allocated and read-accessible).
+    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.
     This function successfully passes all SMHasher tests.
     Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
-    Note that "len" is type "int", which means it is limited to 2^31-1.
-    If your data is larger, use the advanced functions below.
 XXH64() :
     Calculate the 64-bits hash of sequence of length "len" stored at memory address "input".
 */
@@ -96,70 +100,54 @@ XXH64() :
 /*****************************
    Advanced Hash Functions
 *****************************/
-
-void*         XXH32_init   (unsigned int seed);
-XXH_errorcode XXH32_update (void* state, const void* input, unsigned int len);
-unsigned int  XXH32_digest (void* state);
-
-void*         		XXH64_init   (unsigned long long seed);
-XXH_errorcode 		XXH64_update (void* state, const void* input, unsigned int len);
-unsigned long long  XXH64_digest (void* state);
+typedef struct { long long ll[ 6]; } XXH32_state_t;
+typedef struct { long long ll[11]; } XXH64_state_t;
 
 /*
-These functions calculate the xxhash of an input provided in several small packets,
-as opposed to an input provided as a single block.
-
-It must be started with :
-void* XXHnn_init()
-The function returns a pointer which holds the state of calculation.
-If the pointer is NULL, allocation has failed, so no state can be tracked.
-
-The state pointer must be provided as "void* state" parameter for XXHnn_update().
-XXHnn_update() can be called as many times as necessary.
-The user must provide a valid (allocated) input.
-The function returns an error code, with 0 meaning OK, and any other value meaning there is an error.
-Note that "len" is type "int", which means it is limited to 2^31-1.
-If your data is larger, it is recommended to chunk your data into blocks
-of size for example 2^30 (1GB) to avoid any "int" overflow issue.
+These structures allow static allocation of XXH states.
+States must then be initialized using XXHnn_reset() before first use.
 
-Finally, you can end the calculation anytime, by using XXHnn_digest().
-This function returns the final nn-bits hash.
-You must provide the same "void* state" parameter created by XXHnn_init().
-Memory will be freed by XXHnn_digest().
+If you prefer dynamic allocation, please refer to functions below.
 */
 
+XXH32_state_t* XXH32_createState(void);
+XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
 
-int           XXH32_sizeofState(void);
-XXH_errorcode XXH32_resetState(void* state, unsigned int seed);
+XXH64_state_t* XXH64_createState(void);
+XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
 
-#define       XXH32_SIZEOFSTATE 48
-typedef struct { long long ll[(XXH32_SIZEOFSTATE+(sizeof(long long)-1))/sizeof(long long)]; } XXH32_stateSpace_t;
+/*
+These functions create and release memory for XXH state.
+States must then be initialized using XXHnn_reset() before first use.
+*/
 
-int           XXH64_sizeofState(void);
-XXH_errorcode XXH64_resetState(void* state, unsigned long long seed);
 
-#define       XXH64_SIZEOFSTATE 88
-typedef struct { long long ll[(XXH64_SIZEOFSTATE+(sizeof(long long)-1))/sizeof(long long)]; } XXH64_stateSpace_t;
+XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, unsigned seed);
+XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+unsigned int  XXH32_digest (const XXH32_state_t* statePtr);
+
+XXH_errorcode      XXH64_reset  (XXH64_state_t* statePtr, unsigned long long seed);
+XXH_errorcode      XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+unsigned long long XXH64_digest (const XXH64_state_t* statePtr);
 
 /*
-These functions allow user application to make its own allocation for state.
+These functions calculate the xxHash of an input provided in multiple smaller packets,
+as opposed to an input provided as a single block.
 
-XXHnn_sizeofState() is used to know how much space must be allocated for the xxHash nn-bits state.
-Note that the state must be aligned to access 'long long' fields. Memory must be allocated and referenced by a pointer.
-This pointer must then be provided as 'state' into XXHnn_resetState(), which initializes the state.
+XXH state space must first be allocated, using either static or dynamic method provided above.
 
-For static allocation purposes (such as allocation on stack, or freestanding systems without malloc()),
-use the structure XXHnn_stateSpace_t, which will ensure that memory space is large enough and correctly aligned to access 'long long' fields.
-*/
+Start a new hash by initializing state with a seed, using XXHnn_reset().
 
+Then, feed the hash state by calling XXHnn_update() as many times as necessary.
+Obviously, input must be valid, meaning allocated and read accessible.
+The function returns an error code, with 0 meaning OK, and any other value meaning there is an error.
 
-unsigned int       XXH32_intermediateDigest (void* state);
-unsigned long long XXH64_intermediateDigest (void* state);
-/*
-These functions do the same as XXHnn_digest(), generating a nn-bit hash,
-but preserve memory context.
-This way, it becomes possible to generate intermediate hashes, and then continue feeding data with XXHnn_update().
-To free memory context, use XXHnn_digest(), or free().
+Finally, you can produce a hash anytime, by using XXHnn_digest().
+This function returns the final nn-bits hash.
+You can nonetheless continue feeding the hash state with more input,
+and therefore get some new hashes, by calling again XXHnn_digest().
+
+When you are done, don't forget to free XXH state space, using typically XXHnn_freeState().
 */