@@ -194,7 +194,6 @@ set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-format")
 endif()
 
 if(DEFINED 3DS OR DEFINED PSP2 OR DEFINED WII)
-	set(USE_LZMA OFF)
 	set(USE_GDB_STUB OFF)
 endif()
 
@@ -311,7 +310,6 @@ set(FEATURE_SRC)
 set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
 
 if(DISABLE_DEPS)
-	set(USE_LZMA OFF)
 	set(USE_GDB_STUB OFF)
 endif()
 
@@ -445,16 +443,18 @@ ${CMAKE_SOURCE_DIR}/src/third-party/lzma/7zCrc.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/7zCrcOpt.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/7zDec.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/CpuArch.c
+		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Delta.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/LzmaDec.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Lzma2Dec.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Bra.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Bra86.c
+		${CMAKE_SOURCE_DIR}/src/third-party/lzma/BraIA64.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Bcj2.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Ppmd7.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/Ppmd7Dec.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/7zFile.c
 		${CMAKE_SOURCE_DIR}/src/third-party/lzma/7zStream.c)
-	list(APPEND FEATURE_SRC ${LZMA_SRC})
+	list(APPEND VFS_SRC ${LZMA_SRC})
 	list(APPEND FEATURES LZMA)
 endif()
 

@@ -1,5 +1,5 @@
 /* 7z.h -- 7z interface
-2013-01-18 : Igor Pavlov : Public domain */
+2015-11-18 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_H
 #define __7Z_H
@@ -11,7 +11,7 @@
 #define k7zStartHeaderSize 0x20
 #define k7zSignatureSize 6
 
-extern Byte k7zSignature[k7zSignatureSize];
+extern const Byte k7zSignature[k7zSignatureSize];
 
 typedef struct
 {
@@ -25,8 +25,7 @@ typedef struct
 {
   size_t PropsOffset;
   UInt32 MethodID;
-  Byte NumInStreams;
-  Byte NumOutStreams;
+  Byte NumStreams;
   Byte PropsSize;
 } CSzCoderInfo;
 
@@ -34,37 +33,25 @@ typedef struct
 {
   UInt32 InIndex;
   UInt32 OutIndex;
-} CSzBindPair;
+} CSzBond;
 
 #define SZ_NUM_CODERS_IN_FOLDER_MAX 4
-#define SZ_NUM_BINDS_IN_FOLDER_MAX 3
+#define SZ_NUM_BONDS_IN_FOLDER_MAX 3
 #define SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX 4
-#define SZ_NUM_CODERS_OUT_STREAMS_IN_FOLDER_MAX 4
 
 typedef struct
 {
   UInt32 NumCoders;
-  UInt32 NumBindPairs;
+  UInt32 NumBonds;
   UInt32 NumPackStreams;
-  UInt32 MainOutStream;
+  UInt32 UnpackStream;
   UInt32 PackStreams[SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX];
-  CSzBindPair BindPairs[SZ_NUM_BINDS_IN_FOLDER_MAX];
+  CSzBond Bonds[SZ_NUM_BONDS_IN_FOLDER_MAX];
   CSzCoderInfo Coders[SZ_NUM_CODERS_IN_FOLDER_MAX];
-  UInt64 CodersUnpackSizes[SZ_NUM_CODERS_OUT_STREAMS_IN_FOLDER_MAX];
 } CSzFolder;
 
-/*
-typedef struct
-{
-  size_t CodersDataOffset;
-  size_t UnpackSizeDataOffset;
-  // UInt32 StartCoderUnpackSizesIndex;
-  UInt32 StartPackStreamIndex;
-  // UInt32 IndexOfMainOutStream;
-} CSzFolder2;
-*/
 
-SRes SzGetNextFolderItem(CSzFolder *f, CSzData *sd, CSzData *sdSizes);
+SRes SzGetNextFolderItem(CSzFolder *f, CSzData *sd);
 
 typedef struct
 {
@@ -94,47 +81,25 @@ {
   UInt32 NumPackStreams;
   UInt32 NumFolders;
 
-  UInt64 *PackPositions; // NumPackStreams + 1
-  CSzBitUi32s FolderCRCs;
+  UInt64 *PackPositions;          // NumPackStreams + 1
+  CSzBitUi32s FolderCRCs;         // NumFolders
 
-  size_t *FoCodersOffsets;
-  size_t *FoSizesOffsets;
-  // UInt32 StartCoderUnpackSizesIndex;
-  UInt32 *FoStartPackStreamIndex;
+  size_t *FoCodersOffsets;        // NumFolders + 1
+  UInt32 *FoStartPackStreamIndex; // NumFolders + 1
+  UInt32 *FoToCoderUnpackSizes;   // NumFolders + 1
+  Byte *FoToMainUnpackSizeIndex;  // NumFolders
+  UInt64 *CoderUnpackSizes;       // for all coders in all folders
 
-  // CSzFolder2 *Folders;  // +1 item for sum values
   Byte *CodersData;
-  Byte *UnpackSizesData;
-  size_t UnpackSizesDataSize;
-  // UInt64 *CoderUnpackSizes;
 } CSzAr;
 
+UInt64 SzAr_GetFolderUnpackSize(const CSzAr *p, UInt32 folderIndex);
 
 SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
     ILookInStream *stream, UInt64 startPos,
     Byte *outBuffer, size_t outSize,
     ISzAlloc *allocMain);
 
-/*
-  SzExtract extracts file from archive
-
-  *outBuffer must be 0 before first call for each new archive.
-
-  Extracting cache:
-    If you need to decompress more than one file, you can send
-    these values from previous call:
-      *blockIndex,
-      *outBuffer,
-      *outBufferSize
-    You can consider "*outBuffer" as cache of solid block. If your archive is solid,
-    it will increase decompression speed.
-  
-    If you use external function, you can declare these 3 cache variables
-    (blockIndex, outBuffer, outBufferSize) as static in that external function.
-    
-    Free *outBuffer and set *outBuffer to 0, if you want to flush cache.
-*/
-
 typedef struct
 {
   CSzAr db;
@@ -144,7 +109,7 @@ UInt64 dataPos;
   
   UInt32 NumFiles;
 
-  UInt64 *UnpackPositions;
+  UInt64 *UnpackPositions;  // NumFiles + 1
   // Byte *IsEmptyFiles;
   Byte *IsDirs;
   CSzBitUi32s CRCs;
@@ -154,9 +119,8 @@ // CSzBitUi32s Parents;
   CSzBitUi64s MTime;
   CSzBitUi64s CTime;
 
-  // UInt32 *FolderStartPackStreamIndex;
-  UInt32 *FolderStartFileIndex; // + 1
-  UInt32 *FileIndexToFolderIndexMap;
+  UInt32 *FolderToFile;   // NumFolders + 1
+  UInt32 *FileToFolder;   // NumFiles
 
   size_t *FileNameOffsets; /* in 2-byte steps */
   Byte *FileNames;  /* UTF-16-LE */
@@ -182,6 +146,28 @@
 /*
 size_t SzArEx_GetFullNameLen(const CSzArEx *p, size_t fileIndex);
 UInt16 *SzArEx_GetFullNameUtf16_Back(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
+*/
+
+
+
+/*
+  SzArEx_Extract extracts file from archive
+
+  *outBuffer must be 0 before first call for each new archive.
+
+  Extracting cache:
+    If you need to decompress more than one file, you can send
+    these values from previous call:
+      *blockIndex,
+      *outBuffer,
+      *outBufferSize
+    You can consider "*outBuffer" as cache of solid block. If your archive is solid,
+    it will increase decompression speed.
+  
+    If you use external function, you can declare these 3 cache variables
+    (blockIndex, outBuffer, outBufferSize) as static in that external function.
+    
+    Free *outBuffer and set *outBuffer to 0, if you want to flush cache.
 */
 
 SRes SzArEx_Extract(

@@ -1,5 +1,5 @@
 /* 7zAlloc.c -- Allocation functions
-2010-10-29 : Igor Pavlov : Public domain */
+2015-11-09 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -22,11 +22,11 @@ #endif
 
 void *SzAlloc(void *p, size_t size)
 {
-  (void) p;
+  UNUSED_VAR(p);
   if (size == 0)
     return 0;
   #ifdef _SZ_ALLOC_DEBUG
-  fprintf(stderr, "\nAlloc %10d bytes; count = %10d", size, g_allocCount);
+  fprintf(stderr, "\nAlloc %10u bytes; count = %10d", (unsigned)size, g_allocCount);
   g_allocCount++;
   #endif
   return malloc(size);
@@ -34,7 +34,7 @@ }
 
 void SzFree(void *p, void *address)
 {
-  (void) p;
+  UNUSED_VAR(p);
   #ifdef _SZ_ALLOC_DEBUG
   if (address != 0)
   {
@@ -47,11 +47,11 @@ }
 
 void *SzAllocTemp(void *p, size_t size)
 {
-  (void) p;
+  UNUSED_VAR(p);
   if (size == 0)
     return 0;
   #ifdef _SZ_ALLOC_DEBUG
-  fprintf(stderr, "\nAlloc_temp %10d bytes;  count = %10d", size, g_allocCountTemp);
+  fprintf(stderr, "\nAlloc_temp %10u bytes;  count = %10d", (unsigned)size, g_allocCountTemp);
   g_allocCountTemp++;
   #ifdef _WIN32
   return HeapAlloc(GetProcessHeap(), 0, size);
@@ -62,7 +62,7 @@ }
 
 void SzFreeTemp(void *p, void *address)
 {
-  (void) p;
+  UNUSED_VAR(p);
   #ifdef _SZ_ALLOC_DEBUG
   if (address != 0)
   {

@@ -1,15 +1,23 @@
 /* 7zAlloc.h -- Allocation functions
-2010-10-29 : Igor Pavlov : Public domain */
+2013-03-25 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_ALLOC_H
 #define __7Z_ALLOC_H
 
 #include <stdlib.h>
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
 void *SzAlloc(void *p, size_t size);
 void SzFree(void *p, void *address);
 
 void *SzAllocTemp(void *p, size_t size);
 void SzFreeTemp(void *p, void *address);
+
+#ifdef __cplusplus
+}
+#endif
 
 #endif

@@ -1,5 +1,5 @@
 /* 7zArcIn.c -- 7z Input functions
-2014-06-16 : Igor Pavlov : Public domain */
+2015-11-18 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -10,8 +10,16 @@ #include "7zBuf.h"
 #include "7zCrc.h"
 #include "CpuArch.h"
 
-#define MY_ALLOC(T, p, size, alloc) { if ((size) == 0) p = 0; else \
-  if ((p = (T *)IAlloc_Alloc(alloc, (size) * sizeof(T))) == 0) return SZ_ERROR_MEM; }
+#define MY_ALLOC(T, p, size, alloc) { \
+  if ((p = (T *)IAlloc_Alloc(alloc, (size) * sizeof(T))) == NULL) return SZ_ERROR_MEM; }
+
+#define MY_ALLOC_ZE(T, p, size, alloc) { if ((size) == 0) p = NULL; else MY_ALLOC(T, p, size, alloc) }
+
+#define MY_ALLOC_AND_CPY(to, size, from, alloc) \
+  { MY_ALLOC(Byte, to, size, alloc); memcpy(to, from, size); }
+
+#define MY_ALLOC_ZE_AND_CPY(to, size, from, alloc) \
+  { if ((size) == 0) p = NULL; else { MY_ALLOC_AND_CPY(to, size, from, alloc) } }
 
 #define k7zMajorVersion 0
 
@@ -48,75 +56,69 @@ // k7zParent,
   // k7zIsReal
 };
 
-Byte k7zSignature[k7zSignatureSize] = {'7', 'z', 0xBC, 0xAF, 0x27, 0x1C};
-
-#define NUM_FOLDER_CODERS_MAX 32
-#define NUM_CODER_STREAMS_MAX 32
-
-/*
-static int SzFolder_FindBindPairForInStream(const CSzFolder *p, UInt32 inStreamIndex)
-{
-  UInt32 i;
-  for (i = 0; i < p->NumBindPairs; i++)
-    if (p->BindPairs[i].InIndex == inStreamIndex)
-      return i;
-  return -1;
-}
-*/
+const Byte k7zSignature[k7zSignatureSize] = {'7', 'z', 0xBC, 0xAF, 0x27, 0x1C};
 
-#define SzBitUi32s_Init(p) { (p)->Defs = 0; (p)->Vals = 0; }
+#define SzBitUi32s_Init(p) { (p)->Defs = NULL; (p)->Vals = NULL; }
 
 static SRes SzBitUi32s_Alloc(CSzBitUi32s *p, size_t num, ISzAlloc *alloc)
 {
-  MY_ALLOC(Byte, p->Defs, (num + 7) >> 3, alloc);
-  MY_ALLOC(UInt32, p->Vals, num, alloc);
+  if (num == 0)
+  {
+    p->Defs = NULL;
+    p->Vals = NULL;
+  }
+  else
+  {
+    MY_ALLOC(Byte, p->Defs, (num + 7) >> 3, alloc);
+    MY_ALLOC(UInt32, p->Vals, num, alloc);
+  }
   return SZ_OK;
 }
 
 void SzBitUi32s_Free(CSzBitUi32s *p, ISzAlloc *alloc)
 {
-  IAlloc_Free(alloc, p->Defs); p->Defs = 0;
-  IAlloc_Free(alloc, p->Vals); p->Vals = 0;
+  IAlloc_Free(alloc, p->Defs); p->Defs = NULL;
+  IAlloc_Free(alloc, p->Vals); p->Vals = NULL;
 }
 
-#define SzBitUi64s_Init(p) { (p)->Defs = 0; (p)->Vals = 0; }
+#define SzBitUi64s_Init(p) { (p)->Defs = NULL; (p)->Vals = NULL; }
 
 void SzBitUi64s_Free(CSzBitUi64s *p, ISzAlloc *alloc)
 {
-  IAlloc_Free(alloc, p->Defs); p->Defs = 0;
-  IAlloc_Free(alloc, p->Vals); p->Vals = 0;
+  IAlloc_Free(alloc, p->Defs); p->Defs = NULL;
+  IAlloc_Free(alloc, p->Vals); p->Vals = NULL;
 }
 
+
 static void SzAr_Init(CSzAr *p)
 {
   p->NumPackStreams = 0;
   p->NumFolders = 0;
-  p->PackPositions = 0;
+  
+  p->PackPositions = NULL;
   SzBitUi32s_Init(&p->FolderCRCs);
-  // p->Folders = 0;
-  p->FoCodersOffsets = 0;
-  p->FoSizesOffsets = 0;
-  p->FoStartPackStreamIndex = 0;
+
+  p->FoCodersOffsets = NULL;
+  p->FoStartPackStreamIndex = NULL;
+  p->FoToCoderUnpackSizes = NULL;
+  p->FoToMainUnpackSizeIndex = NULL;
+  p->CoderUnpackSizes = NULL;
 
-  p->CodersData = 0;
-  // p->CoderUnpackSizes = 0;
-  p->UnpackSizesData = 0;
+  p->CodersData = NULL;
 }
 
 static void SzAr_Free(CSzAr *p, ISzAlloc *alloc)
 {
-  IAlloc_Free(alloc, p->UnpackSizesData);
-  IAlloc_Free(alloc, p->CodersData);
-  // IAlloc_Free(alloc, p->CoderUnpackSizes);
-
   IAlloc_Free(alloc, p->PackPositions);
+  SzBitUi32s_Free(&p->FolderCRCs, alloc);
  
-  // IAlloc_Free(alloc, p->Folders);
   IAlloc_Free(alloc, p->FoCodersOffsets);
-  IAlloc_Free(alloc, p->FoSizesOffsets);
   IAlloc_Free(alloc, p->FoStartPackStreamIndex);
+  IAlloc_Free(alloc, p->FoToCoderUnpackSizes);
+  IAlloc_Free(alloc, p->FoToMainUnpackSizeIndex);
+  IAlloc_Free(alloc, p->CoderUnpackSizes);
   
-  SzBitUi32s_Free(&p->FolderCRCs, alloc);
+  IAlloc_Free(alloc, p->CodersData);
 
   SzAr_Init(p);
 }
@@ -125,18 +127,19 @@
 void SzArEx_Init(CSzArEx *p)
 {
   SzAr_Init(&p->db);
+  
   p->NumFiles = 0;
   p->dataPos = 0;
-  // p->Files = 0;
-  p->UnpackPositions = 0;
-  // p->IsEmptyFiles = 0;
-  p->IsDirs = 0;
-  // p->FolderStartPackStreamIndex = 0;
-  // p->PackStreamStartPositions = 0;
-  p->FolderStartFileIndex = 0;
-  p->FileIndexToFolderIndexMap = 0;
-  p->FileNameOffsets = 0;
-  p->FileNames = 0;
+  
+  p->UnpackPositions = NULL;
+  p->IsDirs = NULL;
+  
+  p->FolderToFile = NULL;
+  p->FileToFolder = NULL;
+  
+  p->FileNameOffsets = NULL;
+  p->FileNames = NULL;
+  
   SzBitUi32s_Init(&p->CRCs);
   SzBitUi32s_Init(&p->Attribs);
   // SzBitUi32s_Init(&p->Parents);
@@ -146,47 +149,36 @@ }
 
 void SzArEx_Free(CSzArEx *p, ISzAlloc *alloc)
 {
-  // IAlloc_Free(alloc, p->FolderStartPackStreamIndex);
-  // IAlloc_Free(alloc, p->PackStreamStartPositions);
-  IAlloc_Free(alloc, p->FolderStartFileIndex);
-  IAlloc_Free(alloc, p->FileIndexToFolderIndexMap);
+  IAlloc_Free(alloc, p->UnpackPositions);
+  IAlloc_Free(alloc, p->IsDirs);
+
+  IAlloc_Free(alloc, p->FolderToFile);
+  IAlloc_Free(alloc, p->FileToFolder);
 
   IAlloc_Free(alloc, p->FileNameOffsets);
   IAlloc_Free(alloc, p->FileNames);
 
-  SzBitUi64s_Free(&p->CTime, alloc);
-  SzBitUi64s_Free(&p->MTime, alloc);
   SzBitUi32s_Free(&p->CRCs, alloc);
+  SzBitUi32s_Free(&p->Attribs, alloc);
   // SzBitUi32s_Free(&p->Parents, alloc);
-  SzBitUi32s_Free(&p->Attribs, alloc);
-  IAlloc_Free(alloc, p->IsDirs);
-  // IAlloc_Free(alloc, p->IsEmptyFiles);
-  IAlloc_Free(alloc, p->UnpackPositions);
-  // IAlloc_Free(alloc, p->Files);
-
+  SzBitUi64s_Free(&p->MTime, alloc);
+  SzBitUi64s_Free(&p->CTime, alloc);
+  
   SzAr_Free(&p->db, alloc);
   SzArEx_Init(p);
 }
 
-static int TestSignatureCandidate(Byte *testBytes)
+
+static int TestSignatureCandidate(const Byte *testBytes)
 {
-  size_t i;
+  unsigned i;
   for (i = 0; i < k7zSignatureSize; i++)
     if (testBytes[i] != k7zSignature[i])
       return 0;
   return 1;
 }
 
-#define SzData_Clear(p) { (p)->Data = 0; (p)->Size = 0; }
-
-static SRes SzReadByte(CSzData *sd, Byte *b)
-{
-  if (sd->Size == 0)
-    return SZ_ERROR_ARCHIVE;
-  sd->Size--;
-  *b = *sd->Data++;
-  return SZ_OK;
-}
+#define SzData_Clear(p) { (p)->Data = NULL; (p)->Size = 0; }
 
 #define SZ_READ_BYTE_SD(_sd_, dest) if ((_sd_)->Size == 0) return SZ_ERROR_ARCHIVE; (_sd_)->Size--; dest = *(_sd_)->Data++;
 #define SZ_READ_BYTE(dest) SZ_READ_BYTE_SD(sd, dest)
@@ -224,7 +216,7 @@ {
     Byte b;
     if ((firstByte & mask) == 0)
     {
-      UInt64 highPart = firstByte & (mask - 1);
+      UInt64 highPart = (unsigned)firstByte & (unsigned)(mask - 1);
       *value |= (highPart << (8 * i));
       return SZ_OK;
     }
@@ -235,57 +227,6 @@ }
   return SZ_OK;
 }
 
-/*
-static MY_NO_INLINE const Byte *SzReadNumbers(const Byte *data, const Byte *dataLim, UInt64 *values, UInt32 num)
-{
-  for (; num != 0; num--)
-  {
-    Byte firstByte;
-    Byte mask;
-
-    unsigned i;
-    UInt32 v;
-    UInt64 value;
-   
-    if (data == dataLim)
-      return NULL;
-    firstByte = *data++;
-
-    if ((firstByte & 0x80) == 0)
-    {
-      *values++ = firstByte;
-      continue;
-    }
-    if (data == dataLim)
-      return NULL;
-    v = *data++;
-    if ((firstByte & 0x40) == 0)
-    {
-      *values++ = (((UInt32)firstByte & 0x3F) << 8) | v;
-      continue;
-    }
-    if (data == dataLim)
-      return NULL;
-    value = v | ((UInt32)*data++ << 8);
-    mask = 0x20;
-    for (i = 2; i < 8; i++)
-    {
-      if ((firstByte & mask) == 0)
-      {
-        UInt64 highPart = firstByte & (mask - 1);
-        value |= (highPart << (8 * i));
-        break;
-      }
-      if (data == dataLim)
-        return NULL;
-      value |= ((UInt64)*data++ << (8 * i));
-      mask >>= 1;
-    }
-    *values++ = value;
-  }
-  return data;
-}
-*/
 
 static MY_NO_INLINE SRes SzReadNumber32(CSzData *sd, UInt32 *value)
 {
@@ -322,7 +263,7 @@ SKIP_DATA(sd, size);
   return SZ_OK;
 }
 
-static SRes WaitId(CSzData *sd, UInt64 id)
+static SRes WaitId(CSzData *sd, UInt32 id)
 {
   for (;;)
   {
@@ -361,29 +302,29 @@ }
     m--;
     sum += ((b >> m) & 1);
   }
-  return sum ;
+  return sum;
 }
 
 static MY_NO_INLINE SRes ReadBitVector(CSzData *sd, UInt32 numItems, Byte **v, ISzAlloc *alloc)
 {
   Byte allAreDefined;
-  UInt32 i;
   Byte *v2;
   UInt32 numBytes = (numItems + 7) >> 3;
-  RINOK(SzReadByte(sd, &allAreDefined));
+  *v = NULL;
+  SZ_READ_BYTE(allAreDefined);
+  if (numBytes == 0)
+    return SZ_OK;
   if (allAreDefined == 0)
   {
     if (numBytes > sd->Size)
       return SZ_ERROR_ARCHIVE;
-    MY_ALLOC(Byte, *v, numBytes, alloc);
-    memcpy(*v, sd->Data, numBytes);
+    MY_ALLOC_AND_CPY(*v, numBytes, sd->Data, alloc);
     SKIP_DATA(sd, numBytes);
     return SZ_OK;
   }
   MY_ALLOC(Byte, *v, numBytes, alloc);
   v2 = *v;
-  for (i = 0; i < numBytes; i++)
-    v2[i] = 0xFF;
+  memset(v2, 0xFF, (size_t)numBytes);
   {
     unsigned numBits = (unsigned)numItems & 7;
     if (numBits != 0)
@@ -398,7 +339,7 @@ UInt32 i;
   CSzData sd;
   UInt32 *vals;
   const Byte *defs;
-  MY_ALLOC(UInt32, crcs->Vals, numItems, alloc);
+  MY_ALLOC_ZE(UInt32, crcs->Vals, numItems, alloc);
   sd = *sd2;
   defs = crcs->Defs;
   vals = crcs->Vals;
@@ -424,7 +365,7 @@ static SRes SkipBitUi32s(CSzData *sd, UInt32 numItems)
 {
   Byte allAreDefined;
   UInt32 numDefined = numItems;
-  RINOK(SzReadByte(sd, &allAreDefined));
+  SZ_READ_BYTE(allAreDefined);
   if (!allAreDefined)
   {
     size_t numBytes = (numItems + 7) >> 3;
@@ -486,19 +427,22 @@ return (external == 0) ? SZ_OK: SZ_ERROR_UNSUPPORTED;
 }
 */
 
-#define SZ_NUM_IN_STREAMS_IN_FOLDER_MAX 16
+#define k_NumCodersStreams_in_Folder_MAX (SZ_NUM_BONDS_IN_FOLDER_MAX + SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX)
 
-SRes SzGetNextFolderItem(CSzFolder *f, CSzData *sd, CSzData *sdSizes)
+SRes SzGetNextFolderItem(CSzFolder *f, CSzData *sd)
 {
-  UInt32 numCoders, numBindPairs, numPackStreams, i;
-  UInt32 numInStreams = 0, numOutStreams = 0;
+  UInt32 numCoders, i;
+  UInt32 numInStreams = 0;
   const Byte *dataStart = sd->Data;
-  Byte inStreamUsed[SZ_NUM_IN_STREAMS_IN_FOLDER_MAX];
+
+  f->NumCoders = 0;
+  f->NumBonds = 0;
+  f->NumPackStreams = 0;
+  f->UnpackStream = 0;
   
   RINOK(SzReadNumber32(sd, &numCoders));
-  if (numCoders > SZ_NUM_CODERS_IN_FOLDER_MAX)
+  if (numCoders == 0 || numCoders > SZ_NUM_CODERS_IN_FOLDER_MAX)
     return SZ_ERROR_UNSUPPORTED;
-  f->NumCoders = numCoders;
   
   for (i = 0; i < numCoders; i++)
   {
@@ -506,9 +450,11 @@ Byte mainByte;
     CSzCoderInfo *coder = f->Coders + i;
     unsigned idSize, j;
     UInt64 id;
-    RINOK(SzReadByte(sd, &mainByte));
+    
+    SZ_READ_BYTE(mainByte);
     if ((mainByte & 0xC0) != 0)
       return SZ_ERROR_UNSUPPORTED;
+    
     idSize = (unsigned)(mainByte & 0xF);
     if (idSize > sizeof(id))
       return SZ_ERROR_UNSUPPORTED;
@@ -525,108 +471,130 @@ if (id > (UInt32)0xFFFFFFFF)
       return SZ_ERROR_UNSUPPORTED;
     coder->MethodID = (UInt32)id;
     
-    coder->NumInStreams = 1;
-    coder->NumOutStreams = 1;
+    coder->NumStreams = 1;
     coder->PropsOffset = 0;
     coder->PropsSize = 0;
     
     if ((mainByte & 0x10) != 0)
     {
       UInt32 numStreams;
+      
       RINOK(SzReadNumber32(sd, &numStreams));
-      if (numStreams > NUM_CODER_STREAMS_MAX)
+      if (numStreams > k_NumCodersStreams_in_Folder_MAX)
         return SZ_ERROR_UNSUPPORTED;
-      coder->NumInStreams = (Byte)numStreams;
+      coder->NumStreams = (Byte)numStreams;
+
       RINOK(SzReadNumber32(sd, &numStreams));
-      if (numStreams > NUM_CODER_STREAMS_MAX)
+      if (numStreams != 1)
         return SZ_ERROR_UNSUPPORTED;
-      coder->NumOutStreams = (Byte)numStreams;
     }
+
+    numInStreams += coder->NumStreams;
+
+    if (numInStreams > k_NumCodersStreams_in_Folder_MAX)
+      return SZ_ERROR_UNSUPPORTED;
+
     if ((mainByte & 0x20) != 0)
     {
       UInt32 propsSize = 0;
       RINOK(SzReadNumber32(sd, &propsSize));
-      if (propsSize >= 0x40)
-        return SZ_ERROR_UNSUPPORTED;
       if (propsSize > sd->Size)
         return SZ_ERROR_ARCHIVE;
+      if (propsSize >= 0x80)
+        return SZ_ERROR_UNSUPPORTED;
       coder->PropsOffset = sd->Data - dataStart;
       coder->PropsSize = (Byte)propsSize;
       sd->Data += (size_t)propsSize;
       sd->Size -= (size_t)propsSize;
     }
-    numInStreams += coder->NumInStreams;
-    numOutStreams += coder->NumOutStreams;
   }
 
-  if (numOutStreams == 0)
-    return SZ_ERROR_UNSUPPORTED;
-
-  f->NumBindPairs = numBindPairs = numOutStreams - 1;
-  if (numInStreams < numBindPairs)
-    return SZ_ERROR_ARCHIVE;
-  if (numInStreams > SZ_NUM_IN_STREAMS_IN_FOLDER_MAX)
-    return SZ_ERROR_UNSUPPORTED;
-  f->MainOutStream = 0;
-  f->NumPackStreams = numPackStreams = numInStreams - numBindPairs;
-  if (numPackStreams > SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX)
-    return SZ_ERROR_UNSUPPORTED;
-  for (i = 0; i < numInStreams; i++)
-    inStreamUsed[i] = False;
-  if (numBindPairs != 0)
+  /*
+  if (numInStreams == 1 && numCoders == 1)
+  {
+    f->NumPackStreams = 1;
+    f->PackStreams[0] = 0;
+  }
+  else
+  */
   {
-    Byte outStreamUsed[SZ_NUM_CODERS_OUT_STREAMS_IN_FOLDER_MAX];
-
-    if (numBindPairs > SZ_NUM_BINDS_IN_FOLDER_MAX)
+    Byte streamUsed[k_NumCodersStreams_in_Folder_MAX];
+    UInt32 numBonds, numPackStreams;
+    
+    numBonds = numCoders - 1;
+    if (numInStreams < numBonds)
+      return SZ_ERROR_ARCHIVE;
+    if (numBonds > SZ_NUM_BONDS_IN_FOLDER_MAX)
       return SZ_ERROR_UNSUPPORTED;
+    f->NumBonds = numBonds;
+    
+    numPackStreams = numInStreams - numBonds;
+    if (numPackStreams > SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX)
+      return SZ_ERROR_UNSUPPORTED;
+    f->NumPackStreams = numPackStreams;
+  
+    for (i = 0; i < numInStreams; i++)
+      streamUsed[i] = False;
+    
+    if (numBonds != 0)
+    {
+      Byte coderUsed[SZ_NUM_CODERS_IN_FOLDER_MAX];
 
-    for (i = 0; i < numOutStreams; i++)
-      outStreamUsed[i] = False;
-
-    for (i = 0; i < numBindPairs; i++)
+      for (i = 0; i < numCoders; i++)
+        coderUsed[i] = False;
+      
+      for (i = 0; i < numBonds; i++)
+      {
+        CSzBond *bp = f->Bonds + i;
+        
+        RINOK(SzReadNumber32(sd, &bp->InIndex));
+        if (bp->InIndex >= numInStreams || streamUsed[bp->InIndex])
+          return SZ_ERROR_ARCHIVE;
+        streamUsed[bp->InIndex] = True;
+        
+        RINOK(SzReadNumber32(sd, &bp->OutIndex));
+        if (bp->OutIndex >= numCoders || coderUsed[bp->OutIndex])
+          return SZ_ERROR_ARCHIVE;
+        coderUsed[bp->OutIndex] = True;
+      }
+      
+      for (i = 0; i < numCoders; i++)
+        if (!coderUsed[i])
+        {
+          f->UnpackStream = i;
+          break;
+        }
+      
+      if (i == numCoders)
+        return SZ_ERROR_ARCHIVE;
+    }
+    
+    if (numPackStreams == 1)
     {
-      CSzBindPair *bp = f->BindPairs + i;
-      RINOK(SzReadNumber32(sd, &bp->InIndex));
-      if (bp->InIndex >= numInStreams)
-        return SZ_ERROR_ARCHIVE;
-      inStreamUsed[bp->InIndex] = True;
-      RINOK(SzReadNumber32(sd, &bp->OutIndex));
-      if (bp->OutIndex >= numInStreams)
+      for (i = 0; i < numInStreams; i++)
+        if (!streamUsed[i])
+          break;
+      if (i == numInStreams)
         return SZ_ERROR_ARCHIVE;
-      outStreamUsed[bp->OutIndex] = True;
+      f->PackStreams[0] = i;
     }
-    for (i = 0; i < numOutStreams; i++)
-      if (!outStreamUsed[i])
+    else
+      for (i = 0; i < numPackStreams; i++)
       {
-        f->MainOutStream = i;
-        break;
+        UInt32 index;
+        RINOK(SzReadNumber32(sd, &index));
+        if (index >= numInStreams || streamUsed[index])
+          return SZ_ERROR_ARCHIVE;
+        streamUsed[index] = True;
+        f->PackStreams[i] = index;
       }
-    if (i == numOutStreams)
-      return SZ_ERROR_ARCHIVE;
   }
 
-  if (numPackStreams == 1)
-  {
-    for (i = 0; i < numInStreams; i++)
-      if (!inStreamUsed[i])
-        break;
-    if (i == numInStreams)
-      return SZ_ERROR_ARCHIVE;
-    f->PackStreams[0] = i;
-  }
-  else
-    for (i = 0; i < numPackStreams; i++)
-    {
-      RINOK(SzReadNumber32(sd, f->PackStreams + i));
-    }
+  f->NumCoders = numCoders;
 
-  for (i = 0; i < numOutStreams; i++)
-  {
-    RINOK(ReadNumber(sdSizes, f->CodersUnpackSizes + i));
-  }
-  
   return SZ_OK;
 }
+
 
 static MY_NO_INLINE SRes SkipNumbers(CSzData *sd2, UInt32 num)
 {
@@ -658,24 +626,27 @@ *sd2 = sd;
   return SZ_OK;
 }
 
-#define k_InStreamUsed_MAX 64
-#define k_OutStreamUsed_MAX 64
+
+#define k_Scan_NumCoders_MAX 64
+#define k_Scan_NumCodersStreams_in_Folder_MAX 64
+
 
 static SRes ReadUnpackInfo(CSzAr *p,
     CSzData *sd2,
-    UInt32 numFoldersMax, const CBuf *tempBufs, UInt32 numTempBufs,
+    UInt32 numFoldersMax,
+    const CBuf *tempBufs, UInt32 numTempBufs,
     ISzAlloc *alloc)
 {
   CSzData sd;
-  Byte inStreamUsed[k_InStreamUsed_MAX];
-  Byte outStreamUsed[k_OutStreamUsed_MAX];
+  
   UInt32 fo, numFolders, numCodersOutStreams, packStreamIndex;
   const Byte *startBufPtr;
   Byte external;
   
   RINOK(WaitId(sd2, k7zIdFolder));
+  
   RINOK(SzReadNumber32(sd2, &numFolders));
-  if (p->NumFolders > numFoldersMax)
+  if (numFolders > numFoldersMax)
     return SZ_ERROR_UNSUPPORTED;
   p->NumFolders = numFolders;
 
@@ -685,7 +656,7 @@ sd = *sd2;
   else
   {
     UInt32 index;
-    SzReadNumber32(sd2, &index);
+    RINOK(SzReadNumber32(sd2, &index));
     if (index >= numTempBufs)
       return SZ_ERROR_ARCHIVE;
     sd.Data = tempBufs[index].data;
@@ -693,8 +664,9 @@ sd.Size = tempBufs[index].size;
   }
   
   MY_ALLOC(size_t, p->FoCodersOffsets, (size_t)numFolders + 1, alloc);
-  MY_ALLOC(size_t, p->FoSizesOffsets, (size_t)numFolders + 1, alloc);
   MY_ALLOC(UInt32, p->FoStartPackStreamIndex, (size_t)numFolders + 1, alloc);
+  MY_ALLOC(UInt32, p->FoToCoderUnpackSizes, (size_t)numFolders + 1, alloc);
+  MY_ALLOC(Byte, p->FoToMainUnpackSizeIndex, (size_t)numFolders, alloc);
   
   startBufPtr = sd.Data;
   
@@ -703,18 +675,19 @@ numCodersOutStreams = 0;
 
   for (fo = 0; fo < numFolders; fo++)
   {
-    UInt32 numCoders, ci, numInStreams = 0, numOutStreams = 0;
+    UInt32 numCoders, ci, numInStreams = 0;
     
     p->FoCodersOffsets[fo] = sd.Data - startBufPtr;
+    
     RINOK(SzReadNumber32(&sd, &numCoders));
-    if (numCoders > NUM_FOLDER_CODERS_MAX)
+    if (numCoders == 0 || numCoders > k_Scan_NumCoders_MAX)
       return SZ_ERROR_UNSUPPORTED;
     
     for (ci = 0; ci < numCoders; ci++)
     {
       Byte mainByte;
       unsigned idSize;
-      UInt32 coderInStreams, coderOutStreams;
+      UInt32 coderInStreams;
       
       SZ_READ_BYTE_2(mainByte);
       if ((mainByte & 0xC0) != 0)
@@ -727,17 +700,18 @@ return SZ_ERROR_ARCHIVE;
       SKIP_DATA2(sd, idSize);
       
       coderInStreams = 1;
-      coderOutStreams = 1;
+      
       if ((mainByte & 0x10) != 0)
       {
+        UInt32 coderOutStreams;
         RINOK(SzReadNumber32(&sd, &coderInStreams));
         RINOK(SzReadNumber32(&sd, &coderOutStreams));
-        if (coderInStreams > NUM_CODER_STREAMS_MAX ||
-            coderOutStreams > NUM_CODER_STREAMS_MAX)
+        if (coderInStreams > k_Scan_NumCodersStreams_in_Folder_MAX || coderOutStreams != 1)
           return SZ_ERROR_UNSUPPORTED;
       }
+      
       numInStreams += coderInStreams;
-      numOutStreams += coderOutStreams;
+
       if ((mainByte & 0x20) != 0)
       {
         UInt32 propsSize;
@@ -751,75 +725,86 @@
     {
       UInt32 indexOfMainStream = 0;
       UInt32 numPackStreams = 1;
-      if (numOutStreams != 1 || numInStreams != 1)
+      
+      if (numCoders != 1 || numInStreams != 1)
       {
+        Byte streamUsed[k_Scan_NumCodersStreams_in_Folder_MAX];
+        Byte coderUsed[k_Scan_NumCoders_MAX];
+    
         UInt32 i;
-        UInt32 numBindPairs = numOutStreams - 1;
-        if (numOutStreams == 0 || numInStreams < numBindPairs)
+        UInt32 numBonds = numCoders - 1;
+        if (numInStreams < numBonds)
           return SZ_ERROR_ARCHIVE;
         
-        if (numInStreams > k_InStreamUsed_MAX ||
-            numOutStreams > k_OutStreamUsed_MAX)
+        if (numInStreams > k_Scan_NumCodersStreams_in_Folder_MAX)
           return SZ_ERROR_UNSUPPORTED;
         
         for (i = 0; i < numInStreams; i++)
-          inStreamUsed[i] = False;
-        for (i = 0; i < numOutStreams; i++)
-          outStreamUsed[i] = False;
+          streamUsed[i] = False;
+        for (i = 0; i < numCoders; i++)
+          coderUsed[i] = False;
         
-        for (i = 0; i < numBindPairs; i++)
+        for (i = 0; i < numBonds; i++)
         {
           UInt32 index;
+          
           RINOK(SzReadNumber32(&sd, &index));
-          if (index >= numInStreams || inStreamUsed[index])
+          if (index >= numInStreams || streamUsed[index])
             return SZ_ERROR_ARCHIVE;
-          inStreamUsed[index] = True;
+          streamUsed[index] = True;
+          
           RINOK(SzReadNumber32(&sd, &index));
-          if (index >= numInStreams || outStreamUsed[index])
+          if (index >= numCoders || coderUsed[index])
             return SZ_ERROR_ARCHIVE;
-          outStreamUsed[index] = True;
+          coderUsed[index] = True;
         }
         
-        numPackStreams = numInStreams - numBindPairs;
+        numPackStreams = numInStreams - numBonds;
         
         if (numPackStreams != 1)
           for (i = 0; i < numPackStreams; i++)
           {
-            UInt32 temp;
-            RINOK(SzReadNumber32(&sd, &temp));
-            if (temp >= numInStreams)
+            UInt32 index;
+            RINOK(SzReadNumber32(&sd, &index));
+            if (index >= numInStreams || streamUsed[index])
               return SZ_ERROR_ARCHIVE;
+            streamUsed[index] = True;
           }
           
-        for (i = 0; i < numOutStreams; i++)
-          if (!outStreamUsed[i])
+        for (i = 0; i < numCoders; i++)
+          if (!coderUsed[i])
           {
             indexOfMainStream = i;
             break;
           }
  
-        if (i == numOutStreams)
+        if (i == numCoders)
           return SZ_ERROR_ARCHIVE;
       }
+      
       p->FoStartPackStreamIndex[fo] = packStreamIndex;
-      p->FoSizesOffsets[fo] = (numOutStreams << 8) | indexOfMainStream;
-      numCodersOutStreams += numOutStreams;
-      if (numCodersOutStreams < numOutStreams)
+      p->FoToCoderUnpackSizes[fo] = numCodersOutStreams;
+      p->FoToMainUnpackSizeIndex[fo] = (Byte)indexOfMainStream;
+      numCodersOutStreams += numCoders;
+      if (numCodersOutStreams < numCoders)
         return SZ_ERROR_UNSUPPORTED;
+      
       packStreamIndex += numPackStreams;
       if (packStreamIndex < numPackStreams)
         return SZ_ERROR_UNSUPPORTED;
+      
       if (packStreamIndex > p->NumPackStreams)
         return SZ_ERROR_ARCHIVE;
     }
   }
+
+  p->FoToCoderUnpackSizes[fo] = numCodersOutStreams;
   
   {
     size_t dataSize = sd.Data - startBufPtr;
     p->FoStartPackStreamIndex[fo] = packStreamIndex;
     p->FoCodersOffsets[fo] = dataSize;
-    MY_ALLOC(Byte, p->CodersData, dataSize, alloc);
-    memcpy(p->CodersData, startBufPtr, dataSize);
+    MY_ALLOC_ZE_AND_CPY(p->CodersData, dataSize, startBufPtr, alloc);
   }
   
   if (external != 0)
@@ -831,28 +816,13 @@ }
   
   RINOK(WaitId(&sd, k7zIdCodersUnpackSize));
   
-  // MY_ALLOC(UInt64, p->CoderUnpackSizes, (size_t)numCodersOutStreams, alloc);
+  MY_ALLOC_ZE(UInt64, p->CoderUnpackSizes, (size_t)numCodersOutStreams, alloc);
   {
-    size_t dataSize = sd.Size;
-    /*
     UInt32 i;
     for (i = 0; i < numCodersOutStreams; i++)
     {
-    RINOK(ReadNumber(&sd, p->CoderUnpackSizes + i));
+      RINOK(ReadNumber(&sd, p->CoderUnpackSizes + i));
     }
-    */
-    RINOK(SkipNumbers(&sd, numCodersOutStreams));
-    dataSize -= sd.Size;
-    MY_ALLOC(Byte, p->UnpackSizesData, dataSize, alloc);
-    memcpy(p->UnpackSizesData, sd.Data - dataSize, dataSize);
-    p->UnpackSizesDataSize = dataSize;
-    /*
-    const Byte *data = SzReadNumbers(sd.Data, sd.Data + sd.Size, p->CoderUnpackSizes, numCodersOutStreams);
-    if (data == NULL)
-    return SZ_ERROR_ARCHIVE;
-    sd.Size = sd.Data + sd.Size - data;
-    sd.Data = data;
-    */
   }
 
   for (;;)
@@ -872,6 +842,13 @@ }
     RINOK(SkipData(&sd));
   }
 }
+
+
+UInt64 SzAr_GetFolderUnpackSize(const CSzAr *p, UInt32 folderIndex)
+{
+  return p->CoderUnpackSizes[p->FoToCoderUnpackSizes[folderIndex] + p->FoToMainUnpackSizeIndex[folderIndex]];
+}
+
 
 typedef struct
 {
@@ -882,12 +859,10 @@ CSzData sdSizes;
   CSzData sdCRCs;
 } CSubStreamInfo;
 
-#define SzUi32IndexMax (((UInt32)1 << 31) - 2)
 
 static SRes ReadSubStreamsInfo(CSzAr *p, CSzData *sd, CSubStreamInfo *ssi)
 {
   UInt64 type = 0;
-  UInt32 i;
   UInt32 numSubDigests = 0;
   UInt32 numFolders = p->NumFolders;
   UInt32 numUnpackStreams = numFolders;
@@ -898,6 +873,7 @@ {
     RINOK(ReadID(sd, &type));
     if (type == k7zIdNumUnpackStream)
     {
+      UInt32 i;
       ssi->sdNumSubStreams.Data = sd->Data;
       numUnpackStreams = 0;
       numSubDigests = 0;
@@ -1009,7 +985,6 @@ {
   UInt64 dataStartPos;
   UInt32 fo;
   CSubStreamInfo ssi;
-  CSzData sdCodersUnpSizes;
 
   RINOK(SzReadStreamsInfo(p, sd, numFoldersMax, NULL, 0, &dataStartPos, &ssi, allocTemp));
   
@@ -1017,49 +992,26 @@ dataStartPos += baseOffset;
   if (p->NumFolders == 0)
     return SZ_ERROR_ARCHIVE;
  
-  sdCodersUnpSizes.Data = p->UnpackSizesData;
-  sdCodersUnpSizes.Size = p->UnpackSizesDataSize;
   for (fo = 0; fo < p->NumFolders; fo++)
     Buf_Init(tempBufs + fo);
+  
   for (fo = 0; fo < p->NumFolders; fo++)
   {
     CBuf *tempBuf = tempBufs + fo;
-    // folder = p->Folders;
-    // unpackSize = SzAr_GetFolderUnpackSize(p, 0);
-    UInt32 mix = (UInt32)p->FoSizesOffsets[fo];
-    UInt32 mainIndex = mix & 0xFF;
-    UInt32 numOutStreams = mix >> 8;
-    UInt32 si;
-    UInt64 unpackSize = 0;
-    p->FoSizesOffsets[fo] = sdCodersUnpSizes.Data - p->UnpackSizesData;
-    for (si = 0; si < numOutStreams; si++)
-    {
-      UInt64 curSize;
-      RINOK(ReadNumber(&sdCodersUnpSizes, &curSize));
-      if (si == mainIndex)
-      {
-        unpackSize = curSize;
-        break;
-      }
-    }
-    if (si == numOutStreams)
-      return SZ_ERROR_FAIL;
+    UInt64 unpackSize = SzAr_GetFolderUnpackSize(p, fo);
     if ((size_t)unpackSize != unpackSize)
       return SZ_ERROR_MEM;
     if (!Buf_Create(tempBuf, (size_t)unpackSize, allocTemp))
       return SZ_ERROR_MEM;
   }
-  p->FoSizesOffsets[fo] = sdCodersUnpSizes.Data - p->UnpackSizesData;
-    
+  
   for (fo = 0; fo < p->NumFolders; fo++)
   {
     const CBuf *tempBuf = tempBufs + fo;
     RINOK(LookInStream_SeekTo(inStream, dataStartPos));
     RINOK(SzAr_DecodeFolder(p, fo, inStream, dataStartPos, tempBuf->data, tempBuf->size, allocTemp));
-    if (SzBitWithVals_Check(&p->FolderCRCs, fo))
-      if (CrcCalc(tempBuf->data, tempBuf->size) != p->FolderCRCs.Vals[fo])
-        return SZ_ERROR_CRC;
   }
+  
   return SZ_OK;
 }
 
@@ -1069,6 +1021,8 @@ size_t pos = 0;
   *offsets++ = 0;
   if (numFiles == 0)
     return (size == 0) ? SZ_OK : SZ_ERROR_ARCHIVE;
+  if (size < 2)
+    return SZ_ERROR_ARCHIVE;
   if (data[size - 2] != 0 || data[size - 1] != 0)
     return SZ_ERROR_ARCHIVE;
   do
@@ -1100,20 +1054,23 @@ UInt32 i;
   CNtfsFileTime *vals;
   Byte *defs;
   Byte external;
+  
   RINOK(ReadBitVector(sd2, num, &p->Defs, alloc));
-  RINOK(SzReadByte(sd2, &external));
+  
+  SZ_READ_BYTE_SD(sd2, external);
   if (external == 0)
     sd = *sd2;
   else
   {
     UInt32 index;
-    SzReadNumber32(sd2, &index);
+    RINOK(SzReadNumber32(sd2, &index));
     if (index >= numTempBufs)
       return SZ_ERROR_ARCHIVE;
     sd.Data = tempBufs[index].data;
     sd.Size = tempBufs[index].size;
   }
-  MY_ALLOC(CNtfsFileTime, p->Vals, num, alloc);
+  
+  MY_ALLOC_ZE(CNtfsFileTime, p->Vals, num, alloc);
   vals = p->Vals;
   defs = p->Defs;
   for (i = 0; i < num; i++)
@@ -1127,22 +1084,22 @@ SKIP_DATA2(sd, 8);
     }
     else
       vals[i].High = vals[i].Low = 0;
+  
   if (external == 0)
     *sd2 = sd;
+  
   return SZ_OK;
 }
 
+
 #define NUM_ADDITIONAL_STREAMS_MAX 8
+
 
 static SRes SzReadHeader2(
     CSzArEx *p,   /* allocMain */
     CSzData *sd,
-    // Byte **emptyStreamVector, /* allocTemp */
-    // Byte **emptyFileVector,   /* allocTemp */
-    // Byte **lwtVector,         /* allocTemp */
     ILookInStream *inStream,
-    CBuf *tempBufs,
-    UInt32 *numTempBufs,
+    CBuf *tempBufs, UInt32 *numTempBufs,
     ISzAlloc *allocMain,
     ISzAlloc *allocTemp
     )
@@ -1150,10 +1107,9 @@ {
   UInt64 type;
   UInt32 numFiles = 0;
   UInt32 numEmptyStreams = 0;
-  UInt32 i;
   CSubStreamInfo ssi;
-  const Byte *emptyStreams = 0;
-  const Byte *emptyFiles = 0;
+  const Byte *emptyStreams = NULL;
+  const Byte *emptyFiles = NULL;
 
   SzData_Clear(&ssi.sdSizes);
   SzData_Clear(&ssi.sdCRCs);
@@ -1177,22 +1133,19 @@ }
     RINOK(ReadID(sd, &type));
   }
 
-  // if (type == k7zIdAdditionalStreamsInfo)     return SZ_ERROR_UNSUPPORTED;
-
   if (type == k7zIdAdditionalStreamsInfo)
   {
     CSzAr tempAr;
     SRes res;
-    UInt32 numTempFolders;
     
     SzAr_Init(&tempAr);
     res = SzReadAndDecodePackedStreams(inStream, sd, tempBufs, NUM_ADDITIONAL_STREAMS_MAX,
         p->startPosAfterHeader, &tempAr, allocTemp);
-    numTempFolders = tempAr.NumFolders;
+    *numTempBufs = tempAr.NumFolders;
     SzAr_Free(&tempAr, allocTemp);
+    
     if (res != SZ_OK)
       return res;
-    *numTempBufs = numTempFolders;
     RINOK(ReadID(sd, &type));
   }
 
@@ -1206,9 +1159,9 @@ }
 
   if (type == k7zIdEnd)
   {
-    // *sd2 = sd;
     return SZ_OK;
   }
+
   if (type != k7zIdFilesInfo)
     return SZ_ERROR_ARCHIVE;
   
@@ -1225,11 +1178,12 @@ break;
     RINOK(ReadNumber(sd, &size));
     if (size > sd->Size)
       return SZ_ERROR_ARCHIVE;
-    if ((UInt64)(int)type != type)
+    
+    if (type >= ((UInt32)1 << 8))
     {
       SKIP_DATA(sd, size);
     }
-    else switch((int)type)
+    else switch ((unsigned)type)
     {
       case k7zIdName:
       {
@@ -1246,7 +1200,7 @@ }
         else
         {
           UInt32 index;
-          SzReadNumber32(sd, &index);
+          RINOK(SzReadNumber32(sd, &index));
           if (index >= *numTempBufs)
             return SZ_ERROR_ARCHIVE;
           namesData = (tempBufs)[index].data;
@@ -1255,9 +1209,8 @@ }
 
         if ((namesSize & 1) != 0)
           return SZ_ERROR_ARCHIVE;
-        MY_ALLOC(Byte, p->FileNames, namesSize, allocMain);
         MY_ALLOC(size_t, p->FileNameOffsets, numFiles + 1, allocMain);
-        memcpy(p->FileNames, namesData, namesSize);
+        MY_ALLOC_ZE_AND_CPY(p->FileNames, namesSize, namesData, allocMain);
         RINOK(SzReadFileNames(p->FileNames, namesSize, numFiles, p->FileNameOffsets))
         if (external == 0)
         {
@@ -1269,6 +1222,7 @@ case k7zIdEmptyStream:
       {
         RINOK(RememberBitVector(sd, numFiles, &emptyStreams));
         numEmptyStreams = CountDefinedBits(emptyStreams, numFiles);
+        emptyFiles = NULL;
         break;
       }
       case k7zIdEmptyFile:
@@ -1290,7 +1244,7 @@ sdPtr = sd;
         else
         {
           UInt32 index;
-          SzReadNumber32(sd, &index);
+          RINOK(SzReadNumber32(sd, &index));
           if (index >= *numTempBufs)
             return SZ_ERROR_ARCHIVE;
           sdSwitch.Data = (tempBufs)[index].data;
@@ -1332,35 +1286,31 @@ RINOK(SkipData(sd));
   }
 
   {
+    UInt32 i;
     UInt32 emptyFileIndex = 0;
-
     UInt32 folderIndex = 0;
-    UInt32 indexInFolder = 0;
+    UInt32 remSubStreams = 0;
+    UInt32 numSubStreams = 0;
     UInt64 unpackPos = 0;
-    const Byte *digestsDefs = 0;
-    const Byte *digestsVals = 0;
+    const Byte *digestsDefs = NULL;
+    const Byte *digestsVals = NULL;
     UInt32 digestsValsIndex = 0;
     UInt32 digestIndex;
     Byte allDigestsDefined = 0;
-    UInt32 curNumSubStreams = (UInt32)(Int32)-1;
     Byte isDirMask = 0;
     Byte crcMask = 0;
     Byte mask = 0x80;
-    // size_t unpSizesOffset = 0;
-    CSzData sdCodersUnpSizes;
-    sdCodersUnpSizes.Data = p->db.UnpackSizesData;
-    sdCodersUnpSizes.Size = p->db.UnpackSizesDataSize;
     
-    MY_ALLOC(UInt32, p->FolderStartFileIndex, p->db.NumFolders + 1, allocMain);
-    MY_ALLOC(UInt32, p->FileIndexToFolderIndexMap, p->NumFiles, allocMain);
+    MY_ALLOC(UInt32, p->FolderToFile, p->db.NumFolders + 1, allocMain);
+    MY_ALLOC_ZE(UInt32, p->FileToFolder, p->NumFiles, allocMain);
     MY_ALLOC(UInt64, p->UnpackPositions, p->NumFiles + 1, allocMain);
-    MY_ALLOC(Byte, p->IsDirs, (p->NumFiles + 7) >> 3, allocMain);
+    MY_ALLOC_ZE(Byte, p->IsDirs, (p->NumFiles + 7) >> 3, allocMain);
 
     RINOK(SzBitUi32s_Alloc(&p->CRCs, p->NumFiles, allocMain));
 
     if (ssi.sdCRCs.Size != 0)
     {
-      RINOK(SzReadByte(&ssi.sdCRCs, &allDigestsDefined));
+      SZ_READ_BYTE_SD(&ssi.sdCRCs, allDigestsDefined);
       if (allDigestsDefined)
         digestsVals = ssi.sdCRCs.Data;
       else
@@ -1372,6 +1322,7 @@ }
     }
 
     digestIndex = 0;
+    
     for (i = 0; i < numFiles; i++, mask >>= 1)
     {
       if (mask == 0)
@@ -1386,79 +1337,66 @@ }
 
       p->UnpackPositions[i] = unpackPos;
       p->CRCs.Vals[i] = 0;
-      // p->CRCs.Defs[i] = 0;
-      if (emptyStreams && SzBitArray_Check(emptyStreams , i))
+      
+      if (emptyStreams && SzBitArray_Check(emptyStreams, i))
       {
-        if (!emptyFiles || !SzBitArray_Check(emptyFiles, emptyFileIndex))
+        if (emptyFiles)
+        {
+          if (!SzBitArray_Check(emptyFiles, emptyFileIndex))
+            isDirMask |= mask;
+          emptyFileIndex++;
+        }
+        else
           isDirMask |= mask;
-        emptyFileIndex++;
-        if (indexInFolder == 0)
+        if (remSubStreams == 0)
         {
-          p->FileIndexToFolderIndexMap[i] = (UInt32)-1;
+          p->FileToFolder[i] = (UInt32)-1;
           continue;
         }
       }
-      if (indexInFolder == 0)
+      
+      if (remSubStreams == 0)
       {
-        /*
-        v3.13 incorrectly worked with empty folders
-        v4.07: Loop for skipping empty folders
-        */
         for (;;)
         {
           if (folderIndex >= p->db.NumFolders)
             return SZ_ERROR_ARCHIVE;
-          p->FolderStartFileIndex[folderIndex] = i;
-          if (curNumSubStreams == (UInt32)(Int32)-1)
+          p->FolderToFile[folderIndex] = i;
+          numSubStreams = 1;
+          if (ssi.sdNumSubStreams.Data)
           {
-            curNumSubStreams = 1;
-            if (ssi.sdNumSubStreams.Data != 0)
-            {
-              RINOK(SzReadNumber32(&ssi.sdNumSubStreams, &curNumSubStreams));
-            }
+            RINOK(SzReadNumber32(&ssi.sdNumSubStreams, &numSubStreams));
           }
-          if (curNumSubStreams != 0)
+          remSubStreams = numSubStreams;
+          if (numSubStreams != 0)
             break;
-          curNumSubStreams = (UInt32)(Int32)-1;
-          folderIndex++; // check it
+          {
+            UInt64 folderUnpackSize = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
+            unpackPos += folderUnpackSize;
+            if (unpackPos < folderUnpackSize)
+              return SZ_ERROR_ARCHIVE;
+          }
+
+          folderIndex++;
         }
       }
-      p->FileIndexToFolderIndexMap[i] = folderIndex;
-      if (emptyStreams && SzBitArray_Check(emptyStreams , i))
+      
+      p->FileToFolder[i] = folderIndex;
+      
+      if (emptyStreams && SzBitArray_Check(emptyStreams, i))
         continue;
       
-      indexInFolder++;
-      if (indexInFolder >= curNumSubStreams)
+      if (--remSubStreams == 0)
       {
-        UInt64 folderUnpackSize = 0;
-        UInt64 startFolderUnpackPos;
-        {
-          UInt32 mix = (UInt32)p->db.FoSizesOffsets[folderIndex];
-          UInt32 mainIndex = mix & 0xFF;
-          UInt32 numOutStreams = mix >> 8;
-          UInt32 si;
-          p->db.FoSizesOffsets[folderIndex] = sdCodersUnpSizes.Data - p->db.UnpackSizesData;
-          for (si = 0; si < numOutStreams; si++)
-          {
-            UInt64 curSize;
-            RINOK(ReadNumber(&sdCodersUnpSizes, &curSize));
-            if (si == mainIndex)
-            {
-              folderUnpackSize = curSize;
-              break;
-            }
-          }
-          if (si == numOutStreams)
-            return SZ_ERROR_FAIL;
-        }
-
-        // UInt64 folderUnpackSize = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
-        startFolderUnpackPos = p->UnpackPositions[p->FolderStartFileIndex[folderIndex]];
+        UInt64 folderUnpackSize = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
+        UInt64 startFolderUnpackPos = p->UnpackPositions[p->FolderToFile[folderIndex]];
         if (folderUnpackSize < unpackPos - startFolderUnpackPos)
           return SZ_ERROR_ARCHIVE;
         unpackPos = startFolderUnpackPos + folderUnpackSize;
+        if (unpackPos < folderUnpackSize)
+          return SZ_ERROR_ARCHIVE;
 
-        if (curNumSubStreams == 1 && SzBitWithVals_Check(&p->db.FolderCRCs, i))
+        if (numSubStreams == 1 && SzBitWithVals_Check(&p->db.FolderCRCs, i))
         {
           p->CRCs.Vals[i] = p->db.FolderCRCs.Vals[folderIndex];
           crcMask |= mask;
@@ -1469,14 +1407,16 @@ p->CRCs.Vals[i] = GetUi32(digestsVals + (size_t)digestsValsIndex * 4);
           digestsValsIndex++;
           crcMask |= mask;
         }
+        
         folderIndex++;
-        indexInFolder = 0;
       }
       else
       {
         UInt64 v;
         RINOK(ReadNumber(&ssi.sdSizes, &v));
         unpackPos += v;
+        if (unpackPos < v)
+          return SZ_ERROR_ARCHIVE;
         if (allDigestsDefined || (digestsDefs && SzBitArray_Check(digestsDefs, digestIndex)))
         {
           p->CRCs.Vals[i] = GetUi32(digestsVals + (size_t)digestsValsIndex * 4);
@@ -1485,30 +1425,55 @@ crcMask |= mask;
         }
       }
     }
+
     if (mask != 0x80)
     {
       UInt32 byteIndex = (i - 1) >> 3;
       p->IsDirs[byteIndex] = isDirMask;
       p->CRCs.Defs[byteIndex] = crcMask;
     }
+    
     p->UnpackPositions[i] = unpackPos;
-    p->FolderStartFileIndex[folderIndex] = i;
-    p->db.FoSizesOffsets[folderIndex] = sdCodersUnpSizes.Data - p->db.UnpackSizesData;
+
+    if (remSubStreams != 0)
+      return SZ_ERROR_ARCHIVE;
+
+    for (;;)
+    {
+      p->FolderToFile[folderIndex] = i;
+      if (folderIndex >= p->db.NumFolders)
+        break;
+      if (!ssi.sdNumSubStreams.Data)
+        return SZ_ERROR_ARCHIVE;
+      RINOK(SzReadNumber32(&ssi.sdNumSubStreams, &numSubStreams));
+      if (numSubStreams != 0)
+        return SZ_ERROR_ARCHIVE;
+      /*
+      {
+        UInt64 folderUnpackSize = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
+        unpackPos += folderUnpackSize;
+        if (unpackPos < folderUnpackSize)
+          return SZ_ERROR_ARCHIVE;
+      }
+      */
+      folderIndex++;
+    }
+
+    if (ssi.sdNumSubStreams.Data && ssi.sdNumSubStreams.Size != 0)
+      return SZ_ERROR_ARCHIVE;
   }
+  
   return SZ_OK;
 }
+
 
 static SRes SzReadHeader(
     CSzArEx *p,
     CSzData *sd,
     ILookInStream *inStream,
-    ISzAlloc *allocMain
-    ,ISzAlloc *allocTemp
-    )
+    ISzAlloc *allocMain,
+    ISzAlloc *allocTemp)
 {
-  // Byte *emptyStreamVector = 0;
-  // Byte *emptyFileVector = 0;
-  // Byte *lwtVector = 0;
   UInt32 i;
   UInt32 numTempBufs = 0;
   SRes res;
@@ -1516,54 +1481,21 @@ CBuf tempBufs[NUM_ADDITIONAL_STREAMS_MAX];
 
   for (i = 0; i < NUM_ADDITIONAL_STREAMS_MAX; i++)
     Buf_Init(tempBufs + i);
-  // SzBitUi32s_Init(&digests);
   
-  res = SzReadHeader2(p, sd,
-      // &emptyStreamVector,
-      // &emptyFileVector,
-      // &lwtVector,
-      inStream,
+  res = SzReadHeader2(p, sd, inStream,
       tempBufs, &numTempBufs,
-      allocMain, allocTemp
-      );
+      allocMain, allocTemp);
   
-  for (i = 0; i < numTempBufs; i++)
+  for (i = 0; i < NUM_ADDITIONAL_STREAMS_MAX; i++)
     Buf_Free(tempBufs + i, allocTemp);
 
-  // IAlloc_Free(allocTemp, emptyStreamVector);
-  // IAlloc_Free(allocTemp, emptyFileVector);
-  // IAlloc_Free(allocTemp, lwtVector);
-
   RINOK(res);
-  {
-    if (sd->Size != 0)
-      return SZ_ERROR_FAIL;
-  }
 
-  return res;
-}
-
-/*
-static UInt64 SzAr_GetFolderUnpackSize(const CSzAr *p, UInt32 folderIndex)
-{
-  const CSzFolder2 *f = p->Folders + folderIndex;
-
-  // return p->CoderUnpackSizes[f->StartCoderUnpackSizesIndex + f->IndexOfMainOutStream];
+  if (sd->Size != 0)
+    return SZ_ERROR_FAIL;
 
-  UInt32 si;
-  CSzData sdCodersUnpSizes;
-  sdCodersUnpSizes.Data = p->UnpackSizesData + f->UnpackSizeDataOffset;
-  sdCodersUnpSizes.Size = p->UnpackSizesDataSize - f->UnpackSizeDataOffset;
-  for (si = 0; si < numOutStreams; si++)
-  {
-    UInt64 curSize;
-    ReadNumber(&sdCodersUnpSizes, &curSize);
-    if (si == mainIndex)
-      return curSize;
-  }
-  return 0;
+  return res;
 }
-*/
 
 static SRes SzArEx_Open2(
     CSzArEx *p,
@@ -1622,6 +1554,7 @@ if (!Buf_Create(&buf, nextHeaderSizeT, allocTemp))
     return SZ_ERROR_MEM;
 
   res = LookInStream_Read(inStream, buf.data, nextHeaderSizeT);
+  
   if (res == SZ_OK)
   {
     res = SZ_ERROR_ARCHIVE;
@@ -1631,7 +1564,9 @@ CSzData sd;
       UInt64 type;
       sd.Data = buf.data;
       sd.Size = buf.size;
+      
       res = ReadID(&sd, &type);
+      
       if (res == SZ_OK && type == k7zIdEncodedHeader)
       {
         CSzAr tempAr;
@@ -1656,35 +1591,35 @@ sd.Size = buf.size;
           res = ReadID(&sd, &type);
         }
       }
+  
       if (res == SZ_OK)
       {
         if (type == k7zIdHeader)
         {
+          /*
           CSzData sd2;
-          int ttt;
-          for (ttt = 0; ttt < 1; ttt++)
-          // for (ttt = 0; ttt < 40000; ttt++)
+          unsigned ttt;
+          for (ttt = 0; ttt < 40000; ttt++)
           {
             SzArEx_Free(p, allocMain);
             sd2 = sd;
-            res = SzReadHeader(p, &sd2, inStream, allocMain, allocTemp
-              );
+            res = SzReadHeader(p, &sd2, inStream, allocMain, allocTemp);
             if (res != SZ_OK)
               break;
           }
-
-          // res = SzReadHeader(p, &sd, allocMain, allocTemp);
+          */
+          res = SzReadHeader(p, &sd, inStream, allocMain, allocTemp);
         }
         else
           res = SZ_ERROR_UNSUPPORTED;
       }
     }
   }
+ 
   Buf_Free(&buf, allocTemp);
   return res;
 }
 
-// #include <stdio.h>
 
 SRes SzArEx_Open(CSzArEx *p, ILookInStream *inStream,
     ISzAlloc *allocMain, ISzAlloc *allocTemp)
@@ -1692,9 +1627,9 @@ {
   SRes res = SzArEx_Open2(p, inStream, allocMain, allocTemp);
   if (res != SZ_OK)
     SzArEx_Free(p, allocMain);
-  // printf ("\nrrr=%d\n", rrr);
   return res;
 }
+
 
 SRes SzArEx_Extract(
     const CSzArEx *p,
@@ -1708,34 +1643,36 @@ size_t *outSizeProcessed,
     ISzAlloc *allocMain,
     ISzAlloc *allocTemp)
 {
-  UInt32 folderIndex = p->FileIndexToFolderIndexMap[fileIndex];
+  UInt32 folderIndex = p->FileToFolder[fileIndex];
   SRes res = SZ_OK;
+  
   *offset = 0;
   *outSizeProcessed = 0;
+  
   if (folderIndex == (UInt32)-1)
   {
     IAlloc_Free(allocMain, *tempBuf);
     *blockIndex = folderIndex;
-    *tempBuf = 0;
+    *tempBuf = NULL;
     *outBufferSize = 0;
     return SZ_OK;
   }
 
-  if (*tempBuf == 0 || *blockIndex != folderIndex)
+  if (*tempBuf == NULL || *blockIndex != folderIndex)
   {
-    // UInt64 unpackSizeSpec = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
+    UInt64 unpackSizeSpec = SzAr_GetFolderUnpackSize(&p->db, folderIndex);
+    /*
     UInt64 unpackSizeSpec =
-        p->UnpackPositions[p->FolderStartFileIndex[folderIndex + 1]] -
-        p->UnpackPositions[p->FolderStartFileIndex[folderIndex]];
+        p->UnpackPositions[p->FolderToFile[folderIndex + 1]] -
+        p->UnpackPositions[p->FolderToFile[folderIndex]];
+    */
     size_t unpackSize = (size_t)unpackSizeSpec;
 
     if (unpackSize != unpackSizeSpec)
       return SZ_ERROR_MEM;
     *blockIndex = folderIndex;
     IAlloc_Free(allocMain, *tempBuf);
-    *tempBuf = 0;
-    
-    // RINOK(LookInStream_SeekTo(inStream, startOffset));
+    *tempBuf = NULL;
     
     if (res == SZ_OK)
     {
@@ -1743,36 +1680,30 @@ *outBufferSize = unpackSize;
       if (unpackSize != 0)
       {
         *tempBuf = (Byte *)IAlloc_Alloc(allocMain, unpackSize);
-        if (*tempBuf == 0)
+        if (*tempBuf == NULL)
           res = SZ_ERROR_MEM;
       }
+  
       if (res == SZ_OK)
       {
         res = SzAr_DecodeFolder(&p->db, folderIndex,
-          inStream,
-          p->dataPos,
-          *tempBuf, unpackSize, allocTemp);
-        if (res == SZ_OK)
-        {
-          if (SzBitWithVals_Check(&p->db.FolderCRCs, folderIndex))
-          {
-            if (CrcCalc(*tempBuf, unpackSize) != p->db.FolderCRCs.Vals[folderIndex])
-              res = SZ_ERROR_CRC;
-          }
-        }
+            inStream, p->dataPos, *tempBuf, unpackSize, allocTemp);
       }
     }
   }
+
   if (res == SZ_OK)
   {
     UInt64 unpackPos = p->UnpackPositions[fileIndex];
-    *offset = (size_t)(unpackPos - p->UnpackPositions[p->FolderStartFileIndex[folderIndex]]);
+    *offset = (size_t)(unpackPos - p->UnpackPositions[p->FolderToFile[folderIndex]]);
     *outSizeProcessed = (size_t)(p->UnpackPositions[fileIndex + 1] - unpackPos);
     if (*offset + *outSizeProcessed > *outBufferSize)
       return SZ_ERROR_FAIL;
-    if (SzBitWithVals_Check(&p->CRCs, fileIndex) && CrcCalc(*tempBuf + *offset, *outSizeProcessed) != p->CRCs.Vals[fileIndex])
-      res = SZ_ERROR_CRC;
+    if (SzBitWithVals_Check(&p->CRCs, fileIndex))
+      if (CrcCalc(*tempBuf + *offset, *outSizeProcessed) != p->CRCs.Vals[fileIndex])
+        res = SZ_ERROR_CRC;
   }
+
   return res;
 }
 

@@ -1,5 +1,5 @@
 /* 7zBuf2.c -- Byte Buffer
-2013-11-12 : Igor Pavlov : Public domain */
+2014-08-22 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -34,8 +34,11 @@ memcpy(data, p->data, p->pos);
     alloc->Free(alloc, p->data);
     p->data = data;
   }
-  memcpy(p->data + p->pos, buf, size);
-  p->pos += size;
+  if (size != 0)
+  {
+    memcpy(p->data + p->pos, buf, size);
+    p->pos += size;
+  }
   return 1;
 }
 

@@ -1,5 +1,5 @@
 /* 7zCrc.c -- CRC32 init
-2013-11-12 : Igor Pavlov : Public domain */
+2015-03-10 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -8,24 +8,28 @@ #include "CpuArch.h"
 
 #define kCrcPoly 0xEDB88320
 
-#ifdef MY_CPU_X86_OR_AMD64
+#ifdef MY_CPU_LE
   #define CRC_NUM_TABLES 8
-  UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
-#elif defined(MY_CPU_LE)
-  #define CRC_NUM_TABLES 4
 #else
-  #define CRC_NUM_TABLES 5
+  #define CRC_NUM_TABLES 9
+
   #define CRC_UINT32_SWAP(v) ((v >> 24) | ((v >> 8) & 0xFF00) | ((v << 8) & 0xFF0000) | (v << 24))
+
   UInt32 MY_FAST_CALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 MY_FAST_CALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
 #endif
 
 #ifndef MY_CPU_BE
   UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
 #endif
 
 typedef UInt32 (MY_FAST_CALL *CRC_FUNC)(UInt32 v, const void *data, size_t size, const UInt32 *table);
 
+CRC_FUNC g_CrcUpdateT4;
+CRC_FUNC g_CrcUpdateT8;
 CRC_FUNC g_CrcUpdate;
+
 UInt32 g_CrcTable[256 * CRC_NUM_TABLES];
 
 UInt32 MY_FAST_CALL CrcUpdate(UInt32 v, const void *data, size_t size)
@@ -38,6 +42,17 @@ {
   return g_CrcUpdate(CRC_INIT_VAL, data, size, g_CrcTable) ^ CRC_INIT_VAL;
 }
 
+#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
+
+UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  const Byte *pEnd = p + size;
+  for (; p != pEnd; p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  return v;
+}
+
 void MY_FAST_CALL CrcGenerateTable()
 {
   UInt32 i;
@@ -54,22 +69,43 @@ {
     UInt32 r = g_CrcTable[i - 256];
     g_CrcTable[i] = g_CrcTable[r & 0xFF] ^ (r >> 8);
   }
+
+  #if CRC_NUM_TABLES < 4
+  
+  g_CrcUpdate = CrcUpdateT1;
   
+  #else
+ 
   #ifdef MY_CPU_LE
 
-  g_CrcUpdate = CrcUpdateT4;
+    g_CrcUpdateT4 = CrcUpdateT4;
+    g_CrcUpdate = CrcUpdateT4;
+
+    #if CRC_NUM_TABLES >= 8
+      g_CrcUpdateT8 = CrcUpdateT8;
   
-  #if CRC_NUM_TABLES == 8
-  if (!CPU_Is_InOrder())
-    g_CrcUpdate = CrcUpdateT8;
-  #endif
+      #ifdef MY_CPU_X86_OR_AMD64
+      if (!CPU_Is_InOrder())
+        g_CrcUpdate = CrcUpdateT8;
+      #endif
+    #endif
 
   #else
   {
     #ifndef MY_CPU_BE
-    UInt32 k = 1;
-    if (*(const Byte *)&k == 1)
+    UInt32 k = 0x01020304;
+    const Byte *p = (const Byte *)&k;
+    if (p[0] == 4 && p[1] == 3)
+    {
+      g_CrcUpdateT4 = CrcUpdateT4;
       g_CrcUpdate = CrcUpdateT4;
+      #if CRC_NUM_TABLES >= 8
+      g_CrcUpdateT8 = CrcUpdateT8;
+      // g_CrcUpdate = CrcUpdateT8;
+      #endif
+    }
+    else if (p[0] != 1 || p[1] != 2)
+      g_CrcUpdate = CrcUpdateT1;
     else
     #endif
     {
@@ -78,8 +114,15 @@ {
         UInt32 x = g_CrcTable[i - 256];
         g_CrcTable[i] = CRC_UINT32_SWAP(x);
       }
+      g_CrcUpdateT4 = CrcUpdateT1_BeT4;
       g_CrcUpdate = CrcUpdateT1_BeT4;
+      #if CRC_NUM_TABLES >= 8
+      g_CrcUpdateT8 = CrcUpdateT1_BeT8;
+      // g_CrcUpdate = CrcUpdateT1_BeT8;
+      #endif
     }
   }
+  #endif
+
   #endif
 }

@@ -1,13 +1,13 @@
 /* 7zCrcOpt.c -- CRC32 calculation
-2013-11-12 : Igor Pavlov : Public domain */
+2015-03-01 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
 #include "CpuArch.h"
 
-#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
-
 #ifndef MY_CPU_BE
+
+#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
 
 UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
@@ -18,10 +18,10 @@ for (; size >= 4; size -= 4, p += 4)
   {
     v ^= *(const UInt32 *)p;
     v =
-      table[0x300 + (v & 0xFF)] ^
-      table[0x200 + ((v >> 8) & 0xFF)] ^
-      table[0x100 + ((v >> 16) & 0xFF)] ^
-      table[0x000 + ((v >> 24))];
+          table[0x300 + ((v      ) & 0xFF)]
+        ^ table[0x200 + ((v >>  8) & 0xFF)]
+        ^ table[0x100 + ((v >> 16) & 0xFF)]
+        ^ table[0x000 + ((v >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2(v, *p);
@@ -30,7 +30,28 @@ }
 
 UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
-  return CrcUpdateT4(v, data, size, table);
+  const Byte *p = (const Byte *)data;
+  for (; size > 0 && ((unsigned)(ptrdiff_t)p & 7) != 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  for (; size >= 8; size -= 8, p += 8)
+  {
+    UInt32 d;
+    v ^= *(const UInt32 *)p;
+    v =
+          table[0x700 + ((v      ) & 0xFF)]
+        ^ table[0x600 + ((v >>  8) & 0xFF)]
+        ^ table[0x500 + ((v >> 16) & 0xFF)]
+        ^ table[0x400 + ((v >> 24))];
+    d = *((const UInt32 *)p + 1);
+    v ^=
+          table[0x300 + ((d      ) & 0xFF)]
+        ^ table[0x200 + ((d >>  8) & 0xFF)]
+        ^ table[0x100 + ((d >> 16) & 0xFF)]
+        ^ table[0x000 + ((d >> 24))];
+  }
+  for (; size > 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  return v;
 }
 
 #endif
@@ -40,27 +61,55 @@ #ifndef MY_CPU_LE
 
 #define CRC_UINT32_SWAP(v) ((v >> 24) | ((v >> 8) & 0xFF00) | ((v << 8) & 0xFF0000) | (v << 24))
 
+#define CRC_UPDATE_BYTE_2_BE(crc, b) (table[(((crc) >> 24) ^ (b))] ^ ((crc) << 8))
+
 UInt32 MY_FAST_CALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
   const Byte *p = (const Byte *)data;
+  table += 0x100;
+  v = CRC_UINT32_SWAP(v);
   for (; size > 0 && ((unsigned)(ptrdiff_t)p & 3) != 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2(v, *p);
-  v = CRC_UINT32_SWAP(v);
-  table += 0x100;
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
   for (; size >= 4; size -= 4, p += 4)
   {
     v ^= *(const UInt32 *)p;
     v =
-      table[0x000 + (v & 0xFF)] ^
-      table[0x100 + ((v >> 8) & 0xFF)] ^
-      table[0x200 + ((v >> 16) & 0xFF)] ^
-      table[0x300 + ((v >> 24))];
+          table[0x000 + ((v      ) & 0xFF)]
+        ^ table[0x100 + ((v >>  8) & 0xFF)]
+        ^ table[0x200 + ((v >> 16) & 0xFF)]
+        ^ table[0x300 + ((v >> 24))];
   }
-  table -= 0x100;
+  for (; size > 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  return CRC_UINT32_SWAP(v);
+}
+
+UInt32 MY_FAST_CALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  table += 0x100;
   v = CRC_UINT32_SWAP(v);
+  for (; size > 0 && ((unsigned)(ptrdiff_t)p & 7) != 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  for (; size >= 8; size -= 8, p += 8)
+  {
+    UInt32 d;
+    v ^= *(const UInt32 *)p;
+    v =
+          table[0x400 + ((v      ) & 0xFF)]
+        ^ table[0x500 + ((v >>  8) & 0xFF)]
+        ^ table[0x600 + ((v >> 16) & 0xFF)]
+        ^ table[0x700 + ((v >> 24))];
+    d = *((const UInt32 *)p + 1);
+    v ^=
+          table[0x000 + ((d      ) & 0xFF)]
+        ^ table[0x100 + ((d >>  8) & 0xFF)]
+        ^ table[0x200 + ((d >> 16) & 0xFF)]
+        ^ table[0x300 + ((d >> 24))];
+  }
   for (; size > 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2(v, *p);
-  return v;
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  return CRC_UINT32_SWAP(v);
 }
 
 #endif

@@ -1,5 +1,5 @@
 /* 7zDec.c -- Decoding from 7z folder
-2014-06-16 : Igor Pavlov : Public domain */
+2015-11-18 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -8,10 +8,12 @@
 /* #define _7ZIP_PPMD_SUPPPORT */
 
 #include "7z.h"
+#include "7zCrc.h"
 
 #include "Bcj2.h"
 #include "Bra.h"
 #include "CpuArch.h"
+#include "Delta.h"
 #include "LzmaDec.h"
 #include "Lzma2Dec.h"
 #ifdef _7ZIP_PPMD_SUPPPORT
@@ -19,14 +21,17 @@ #include "Ppmd7.h"
 #endif
 
 #define k_Copy 0
+#define k_Delta 3
 #define k_LZMA2 0x21
 #define k_LZMA  0x30101
-#define k_BCJ   0x03030103
-#define k_PPC   0x03030205
-#define k_ARM   0x03030501
-#define k_ARMT  0x03030701
-#define k_SPARC 0x03030805
-#define k_BCJ2  0x0303011B
+#define k_BCJ   0x3030103
+#define k_BCJ2  0x303011B
+#define k_PPC   0x3030205
+#define k_IA64  0x3030401
+#define k_ARM   0x3030501
+#define k_ARMT  0x3030701
+#define k_SPARC 0x3030805
+
 
 #ifdef _7ZIP_PPMD_SUPPPORT
 
@@ -140,11 +145,11 @@ LzmaDec_Init(&state);
 
   for (;;)
   {
-    Byte *inBuf = NULL;
+    const void *inBuf = NULL;
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look((void *)inStream, (const void **)&inBuf, &lookahead);
+    res = inStream->Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
@@ -156,14 +161,23 @@ lookahead -= inProcessed;
       inSize -= inProcessed;
       if (res != SZ_OK)
         break;
-      if (state.dicPos == state.dicBufSize || (inProcessed == 0 && dicPos == state.dicPos))
+
+      if (status == LZMA_STATUS_FINISHED_WITH_MARK)
       {
-        if (state.dicBufSize != outSize || lookahead != 0 ||
-            (status != LZMA_STATUS_FINISHED_WITH_MARK &&
-             status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK))
+        if (outSize != state.dicPos || inSize != 0)
           res = SZ_ERROR_DATA;
         break;
       }
+
+      if (outSize == state.dicPos && inSize == 0 && status == LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK)
+        break;
+
+      if (inProcessed == 0 && dicPos == state.dicPos)
+      {
+        res = SZ_ERROR_DATA;
+        break;
+      }
+
       res = inStream->Skip((void *)inStream, inProcessed);
       if (res != SZ_OK)
         break;
@@ -173,6 +187,9 @@
   LzmaDec_FreeProbs(&state, allocMain);
   return res;
 }
+
+
+#ifndef _7Z_NO_METHOD_LZMA2
 
 static SRes SzDecodeLzma2(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStream *inStream,
     Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
@@ -190,11 +207,11 @@ Lzma2Dec_Init(&state);
 
   for (;;)
   {
-    Byte *inBuf = NULL;
+    const void *inBuf = NULL;
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look((void *)inStream, (const void **)&inBuf, &lookahead);
+    res = inStream->Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
@@ -206,13 +223,20 @@ lookahead -= inProcessed;
       inSize -= inProcessed;
       if (res != SZ_OK)
         break;
-      if (state.decoder.dicPos == state.decoder.dicBufSize || (inProcessed == 0 && dicPos == state.decoder.dicPos))
+
+      if (status == LZMA_STATUS_FINISHED_WITH_MARK)
       {
-        if (state.decoder.dicBufSize != outSize || lookahead != 0 ||
-            (status != LZMA_STATUS_FINISHED_WITH_MARK))
+        if (outSize != state.decoder.dicPos || inSize != 0)
           res = SZ_ERROR_DATA;
         break;
       }
+
+      if (inProcessed == 0 && dicPos == state.decoder.dicPos)
+      {
+        res = SZ_ERROR_DATA;
+        break;
+      }
+
       res = inStream->Skip((void *)inStream, inProcessed);
       if (res != SZ_OK)
         break;
@@ -222,16 +246,19 @@
   Lzma2Dec_FreeProbs(&state, allocMain);
   return res;
 }
+
+#endif
+
 
 static SRes SzDecodeCopy(UInt64 inSize, ILookInStream *inStream, Byte *outBuffer)
 {
   while (inSize > 0)
   {
-    void *inBuf;
+    const void *inBuf;
     size_t curSize = (1 << 18);
     if (curSize > inSize)
       curSize = (size_t)inSize;
-    RINOK(inStream->Look((void *)inStream, (const void **)&inBuf, &curSize));
+    RINOK(inStream->Look(inStream, &inBuf, &curSize));
     if (curSize == 0)
       return SZ_ERROR_INPUT_EOF;
     memcpy(outBuffer, inBuf, curSize);
@@ -248,7 +275,9 @@ switch (m)
   {
     case k_Copy:
     case k_LZMA:
+    #ifndef _7Z_NO_METHOD_LZMA2
     case k_LZMA2:
+    #endif
     #ifdef _7ZIP_PPMD_SUPPPORT
     case k_PPMD:
     #endif
@@ -260,13 +289,12 @@
 static Bool IS_SUPPORTED_CODER(const CSzCoderInfo *c)
 {
   return
-      c->NumInStreams == 1 &&
-      c->NumOutStreams == 1 &&
-      /* c->MethodID <= (UInt32)0xFFFFFFFF && */
-      IS_MAIN_METHOD((UInt32)c->MethodID);
+      c->NumStreams == 1
+      /* && c->MethodID <= (UInt32)0xFFFFFFFF */
+      && IS_MAIN_METHOD((UInt32)c->MethodID);
 }
 
-#define IS_BCJ2(c) ((c)->MethodID == k_BCJ2 && (c)->NumInStreams == 4 && (c)->NumOutStreams == 1)
+#define IS_BCJ2(c) ((c)->MethodID == k_BCJ2 && (c)->NumStreams == 4)
 
 static SRes CheckSupportedFolder(const CSzFolder *f)
 {
@@ -276,51 +304,64 @@ if (!IS_SUPPORTED_CODER(&f->Coders[0]))
     return SZ_ERROR_UNSUPPORTED;
   if (f->NumCoders == 1)
   {
-    if (f->NumPackStreams != 1 || f->PackStreams[0] != 0 || f->NumBindPairs != 0)
+    if (f->NumPackStreams != 1 || f->PackStreams[0] != 0 || f->NumBonds != 0)
       return SZ_ERROR_UNSUPPORTED;
     return SZ_OK;
   }
+  
+  
+  #ifndef _7Z_NO_METHODS_FILTERS
+
   if (f->NumCoders == 2)
   {
     const CSzCoderInfo *c = &f->Coders[1];
     if (
         /* c->MethodID > (UInt32)0xFFFFFFFF || */
-        c->NumInStreams != 1 ||
-        c->NumOutStreams != 1 ||
-        f->NumPackStreams != 1 ||
-        f->PackStreams[0] != 0 ||
-        f->NumBindPairs != 1 ||
-        f->BindPairs[0].InIndex != 1 ||
-        f->BindPairs[0].OutIndex != 0)
+        c->NumStreams != 1
+        || f->NumPackStreams != 1
+        || f->PackStreams[0] != 0
+        || f->NumBonds != 1
+        || f->Bonds[0].InIndex != 1
+        || f->Bonds[0].OutIndex != 0)
       return SZ_ERROR_UNSUPPORTED;
     switch ((UInt32)c->MethodID)
     {
+      case k_Delta:
       case k_BCJ:
+      case k_PPC:
+      case k_IA64:
+      case k_SPARC:
       case k_ARM:
+      case k_ARMT:
         break;
       default:
         return SZ_ERROR_UNSUPPORTED;
     }
     return SZ_OK;
   }
+
+  #endif
+
+  
   if (f->NumCoders == 4)
   {
-    if (!IS_SUPPORTED_CODER(&f->Coders[1]) ||
-        !IS_SUPPORTED_CODER(&f->Coders[2]) ||
-        !IS_BCJ2(&f->Coders[3]))
+    if (!IS_SUPPORTED_CODER(&f->Coders[1])
+        || !IS_SUPPORTED_CODER(&f->Coders[2])
+        || !IS_BCJ2(&f->Coders[3]))
       return SZ_ERROR_UNSUPPORTED;
-    if (f->NumPackStreams != 4 ||
-        f->PackStreams[0] != 2 ||
-        f->PackStreams[1] != 6 ||
-        f->PackStreams[2] != 1 ||
-        f->PackStreams[3] != 0 ||
-        f->NumBindPairs != 3 ||
-        f->BindPairs[0].InIndex != 5 || f->BindPairs[0].OutIndex != 0 ||
-        f->BindPairs[1].InIndex != 4 || f->BindPairs[1].OutIndex != 1 ||
-        f->BindPairs[2].InIndex != 3 || f->BindPairs[2].OutIndex != 2)
+    if (f->NumPackStreams != 4
+        || f->PackStreams[0] != 2
+        || f->PackStreams[1] != 6
+        || f->PackStreams[2] != 1
+        || f->PackStreams[3] != 0
+        || f->NumBonds != 3
+        || f->Bonds[0].InIndex != 5 || f->Bonds[0].OutIndex != 0
+        || f->Bonds[1].InIndex != 4 || f->Bonds[1].OutIndex != 1
+        || f->Bonds[2].InIndex != 3 || f->Bonds[2].OutIndex != 2)
       return SZ_ERROR_UNSUPPORTED;
     return SZ_OK;
   }
+  
   return SZ_ERROR_UNSUPPORTED;
 }
 
@@ -364,7 +405,7 @@ outSizeCur = (SizeT)unpackSize;
           if (outSizeCur != unpackSize)
             return SZ_ERROR_MEM;
           temp = (Byte *)IAlloc_Alloc(allocMain, outSizeCur);
-          if (temp == 0 && outSizeCur != 0)
+          if (!temp && outSizeCur != 0)
             return SZ_ERROR_MEM;
           outBufCur = tempBuf[1 - ci] = temp;
           tempSizes[1 - ci] = outSizeCur;
@@ -393,65 +434,117 @@ else if (coder->MethodID == k_LZMA)
       {
         RINOK(SzDecodeLzma(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain));
       }
+      #ifndef _7Z_NO_METHOD_LZMA2
       else if (coder->MethodID == k_LZMA2)
       {
         RINOK(SzDecodeLzma2(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain));
       }
-      else
+      #endif
+      #ifdef _7ZIP_PPMD_SUPPPORT
+      else if (coder->MethodID == k_PPMD)
       {
-        #ifdef _7ZIP_PPMD_SUPPPORT
         RINOK(SzDecodePpmd(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain));
-        #else
+      }
+      #endif
+      else
         return SZ_ERROR_UNSUPPORTED;
-        #endif
-      }
     }
     else if (coder->MethodID == k_BCJ2)
     {
       UInt64 offset = packPositions[1];
       UInt64 s3Size = packPositions[2] - offset;
-      SRes res;
+      
       if (ci != 3)
         return SZ_ERROR_UNSUPPORTED;
-      RINOK(LookInStream_SeekTo(inStream, startPos + offset));
+      
       tempSizes[2] = (SizeT)s3Size;
       if (tempSizes[2] != s3Size)
         return SZ_ERROR_MEM;
       tempBuf[2] = (Byte *)IAlloc_Alloc(allocMain, tempSizes[2]);
-      if (tempBuf[2] == 0 && tempSizes[2] != 0)
+      if (!tempBuf[2] && tempSizes[2] != 0)
         return SZ_ERROR_MEM;
-      res = SzDecodeCopy(s3Size, inStream, tempBuf[2]);
-      RINOK(res)
+      
+      RINOK(LookInStream_SeekTo(inStream, startPos + offset));
+      RINOK(SzDecodeCopy(s3Size, inStream, tempBuf[2]));
+
+      if ((tempSizes[0] & 3) != 0 ||
+          (tempSizes[1] & 3) != 0 ||
+          tempSize3 + tempSizes[0] + tempSizes[1] != outSize)
+        return SZ_ERROR_DATA;
+
+      {
+        CBcj2Dec p;
+        
+        p.bufs[0] = tempBuf3;   p.lims[0] = tempBuf3 + tempSize3;
+        p.bufs[1] = tempBuf[0]; p.lims[1] = tempBuf[0] + tempSizes[0];
+        p.bufs[2] = tempBuf[1]; p.lims[2] = tempBuf[1] + tempSizes[1];
+        p.bufs[3] = tempBuf[2]; p.lims[3] = tempBuf[2] + tempSizes[2];
+        
+        p.dest = outBuffer;
+        p.destLim = outBuffer + outSize;
+        
+        Bcj2Dec_Init(&p);
+        RINOK(Bcj2Dec_Decode(&p));
+
+        {
+          unsigned i;
+          for (i = 0; i < 4; i++)
+            if (p.bufs[i] != p.lims[i])
+              return SZ_ERROR_DATA;
+          
+          if (!Bcj2Dec_IsFinished(&p))
+            return SZ_ERROR_DATA;
 
-      res = Bcj2_Decode(
-          tempBuf3, tempSize3,
-          tempBuf[0], tempSizes[0],
-          tempBuf[1], tempSizes[1],
-          tempBuf[2], tempSizes[2],
-          outBuffer, outSize);
-      RINOK(res)
+          if (p.dest != p.destLim
+             || p.state != BCJ2_STREAM_MAIN)
+            return SZ_ERROR_DATA;
+        }
+      }
     }
-    else
+    #ifndef _7Z_NO_METHODS_FILTERS
+    else if (ci == 1)
     {
-      if (ci != 1)
-        return SZ_ERROR_UNSUPPORTED;
-      switch (coder->MethodID)
+      if (coder->MethodID == k_Delta)
       {
-        case k_BCJ:
+        if (coder->PropsSize != 1)
+          return SZ_ERROR_UNSUPPORTED;
         {
-          UInt32 state;
-          x86_Convert_Init(state);
-          x86_Convert(outBuffer, outSize, 0, &state, 0);
-          break;
+          Byte state[DELTA_STATE_SIZE];
+          Delta_Init(state);
+          Delta_Decode(state, (unsigned)(propsData[coder->PropsOffset]) + 1, outBuffer, outSize);
         }
-        CASE_BRA_CONV(ARM)
-        default:
+      }
+      else
+      {
+        if (coder->PropsSize != 0)
           return SZ_ERROR_UNSUPPORTED;
+        switch (coder->MethodID)
+        {
+          case k_BCJ:
+          {
+            UInt32 state;
+            x86_Convert_Init(state);
+            x86_Convert(outBuffer, outSize, 0, &state, 0);
+            break;
+          }
+          CASE_BRA_CONV(PPC)
+          CASE_BRA_CONV(IA64)
+          CASE_BRA_CONV(SPARC)
+          CASE_BRA_CONV(ARM)
+          CASE_BRA_CONV(ARMT)
+          default:
+            return SZ_ERROR_UNSUPPORTED;
+        }
       }
     }
+    #endif
+    else
+      return SZ_ERROR_UNSUPPORTED;
   }
+
   return SZ_OK;
 }
+
 
 SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
     ILookInStream *inStream, UInt64 startPos,
@@ -461,33 +554,38 @@ {
   SRes res;
   CSzFolder folder;
   CSzData sd;
-  CSzData sdSizes;
   
   const Byte *data = p->CodersData + p->FoCodersOffsets[folderIndex];
   sd.Data = data;
   sd.Size = p->FoCodersOffsets[folderIndex + 1] - p->FoCodersOffsets[folderIndex];
   
-  sdSizes.Data = p->UnpackSizesData + p->FoSizesOffsets[folderIndex];
-  sdSizes.Size =
-      p->FoSizesOffsets[folderIndex + 1] -
-      p->FoSizesOffsets[folderIndex];
-
-  res = SzGetNextFolderItem(&folder, &sd, &sdSizes);
+  res = SzGetNextFolderItem(&folder, &sd);
   
   if (res != SZ_OK)
     return res;
 
-  if (sd.Size != 0 || outSize != folder.CodersUnpackSizes[folder.MainOutStream])
+  if (sd.Size != 0
+      || folder.UnpackStream != p->FoToMainUnpackSizeIndex[folderIndex]
+      || outSize != SzAr_GetFolderUnpackSize(p, folderIndex))
     return SZ_ERROR_FAIL;
   {
-    int i;
+    unsigned i;
     Byte *tempBuf[3] = { 0, 0, 0};
-    res = SzFolder_Decode2(&folder, data, folder.CodersUnpackSizes,
+
+    res = SzFolder_Decode2(&folder, data,
+        &p->CoderUnpackSizes[p->FoToCoderUnpackSizes[folderIndex]],
         p->PackPositions + p->FoStartPackStreamIndex[folderIndex],
         inStream, startPos,
         outBuffer, (SizeT)outSize, allocMain, tempBuf);
+    
     for (i = 0; i < 3; i++)
       IAlloc_Free(allocMain, tempBuf[i]);
+
+    if (res == SZ_OK)
+      if (SzBitWithVals_Check(&p->FolderCRCs, folderIndex))
+        if (CrcCalc(outBuffer, outSize) != p->FolderCRCs.Vals[folderIndex])
+          res = SZ_ERROR_CRC;
+
     return res;
   }
 }

@@ -1,10 +1,19 @@
-#define MY_VER_MAJOR 9
-#define MY_VER_MINOR 38
-#define MY_VER_BUILD 00
-#define MY_VERSION "9.38 beta"
-// #define MY_7ZIP_VERSION "9.38"
-#define MY_DATE "2015-01-03"
+#define MY_VER_MAJOR 15
+#define MY_VER_MINOR 14
+#define MY_VER_BUILD 0
+#define MY_VERSION_NUMBERS "15.14"
+#define MY_VERSION "15.14"
+#define MY_DATE "2015-12-31"
 #undef MY_COPYRIGHT
 #undef MY_VERSION_COPYRIGHT_DATE
-#define MY_COPYRIGHT ": Igor Pavlov : Public domain"
-#define MY_VERSION_COPYRIGHT_DATE MY_VERSION " " MY_COPYRIGHT " : " MY_DATE
+#define MY_AUTHOR_NAME "Igor Pavlov"
+#define MY_COPYRIGHT_PD "Igor Pavlov : Public domain"
+#define MY_COPYRIGHT_CR "Copyright (c) 1999-2015 Igor Pavlov"
+
+#ifdef USE_COPYRIGHT_CR
+  #define MY_COPYRIGHT MY_COPYRIGHT_CR
+#else
+  #define MY_COPYRIGHT MY_COPYRIGHT_PD
+#endif
+
+#define MY_VERSION_COPYRIGHT_DATE MY_VERSION " : " MY_COPYRIGHT " : " MY_DATE

@@ -1,5 +1,5 @@
 /* Aes.c -- AES encryption / decryption
-2013-11-12 : Igor Pavlov : Public domain */
+2015-02-23 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -7,7 +7,7 @@ #include "Aes.h"
 #include "CpuArch.h"
 
 static UInt32 T[256 * 4];
-static Byte Sbox[256] = {
+static const Byte Sbox[256] = {
   0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
   0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
   0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
@@ -40,7 +40,7 @@
 static UInt32 D[256 * 4];
 static Byte InvS[256];
 
-static Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 };
+static const Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 };
 
 #define xtime(x) ((((x) << 1) ^ (((x) & 0x80) != 0 ? 0x1B : 0)) & 0xFF)
 
@@ -56,6 +56,7 @@ {
   unsigned i;
   for (i = 0; i < 256; i++)
     InvS[Sbox[i]] = (Byte)i;
+  
   for (i = 0; i < 256; i++)
   {
     {
@@ -82,9 +83,11 @@ D[0x200 + i] = Ui32(aD, aB, aE, a9);
       D[0x300 + i] = Ui32(a9, aD, aB, aE);
     }
   }
+  
   g_AesCbc_Encode = AesCbc_Encode;
   g_AesCbc_Decode = AesCbc_Decode;
   g_AesCtr_Code = AesCtr_Code;
+  
   #ifdef MY_CPU_X86_OR_AMD64
   if (CPU_Is_Aes_Supported())
   {
@@ -95,34 +98,38 @@ }
   #endif
 }
 
+
 #define HT(i, x, s) (T + (x << 8))[gb ## x(s[(i + x) & 3])]
+
 #define HT4(m, i, s, p) m[i] = \
     HT(i, 0, s) ^ \
     HT(i, 1, s) ^ \
     HT(i, 2, s) ^ \
     HT(i, 3, s) ^ w[p + i]
-/* such order (2031) in HT16 is for VC6/K8 speed optimization) */
+
 #define HT16(m, s, p) \
+    HT4(m, 0, s, p); \
+    HT4(m, 1, s, p); \
     HT4(m, 2, s, p); \
-    HT4(m, 0, s, p); \
     HT4(m, 3, s, p); \
-    HT4(m, 1, s, p); \
 
 #define FT(i, x) Sbox[gb ## x(m[(i + x) & 3])]
 #define FT4(i) dest[i] = Ui32(FT(i, 0), FT(i, 1), FT(i, 2), FT(i, 3)) ^ w[i];
 
+
 #define HD(i, x, s) (D + (x << 8))[gb ## x(s[(i - x) & 3])]
+
 #define HD4(m, i, s, p) m[i] = \
     HD(i, 0, s) ^ \
     HD(i, 1, s) ^ \
     HD(i, 2, s) ^ \
     HD(i, 3, s) ^ w[p + i];
-/* such order (0231) in HD16 is for VC6/K8 speed optimization) */
+
 #define HD16(m, s, p) \
     HD4(m, 0, s, p); \
+    HD4(m, 1, s, p); \
     HD4(m, 2, s, p); \
     HD4(m, 3, s, p); \
-    HD4(m, 1, s, p); \
 
 #define FD(i, x) InvS[gb ## x(m[(i - x) & 3])]
 #define FD4(i) dest[i] = Ui32(FD(i, 0), FD(i, 1), FD(i, 2), FD(i, 3)) ^ w[i];
@@ -169,7 +176,7 @@ }
 
 /* Aes_Encode and Aes_Decode functions work with little-endian words.
   src and dest are pointers to 4 UInt32 words.
-  arc and dest can point to same block */
+  src and dest can point to same block */
 
 static void Aes_Encode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
 {
@@ -271,13 +278,17 @@ {
     UInt32 temp[4];
     Byte buf[16];
     int i;
+
     if (++p[0] == 0)
       p[1]++;
+    
     Aes_Encode(p + 4, temp, p);
+    
     SetUi32(buf,      temp[0]);
     SetUi32(buf + 4,  temp[1]);
     SetUi32(buf + 8,  temp[2]);
     SetUi32(buf + 12, temp[3]);
+    
     for (i = 0; i < 16; i++)
       *data++ ^= buf[i];
   }

@@ -1,5 +1,5 @@
 /* Alloc.c -- Memory allocation functions
-2013-11-12 : Igor Pavlov : Public domain */
+2015-02-21 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -125,3 +125,12 @@ VirtualFree(address, 0, MEM_RELEASE);
 }
 
 #endif
+
+
+static void *SzAlloc(void *p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); }
+static void SzFree(void *p, void *address) { UNUSED_VAR(p); MyFree(address); }
+ISzAlloc g_Alloc = { SzAlloc, SzFree };
+
+static void *SzBigAlloc(void *p, size_t size) { UNUSED_VAR(p); return BigAlloc(size); }
+static void SzBigFree(void *p, void *address) { UNUSED_VAR(p); BigFree(address); }
+ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree };

@@ -1,14 +1,12 @@
 /* Alloc.h -- Memory allocation functions
-2009-02-07 : Igor Pavlov : Public domain */
+2015-02-21 : Igor Pavlov : Public domain */
 
 #ifndef __COMMON_ALLOC_H
 #define __COMMON_ALLOC_H
 
-#include <stddef.h>
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 void *MyAlloc(size_t size);
 void MyFree(void *address);
@@ -31,8 +29,9 @@ #define BigFree(address) MyFree(address)
 
 #endif
 
-#ifdef __cplusplus
-}
-#endif
+extern ISzAlloc g_Alloc;
+extern ISzAlloc g_BigAlloc;
+
+EXTERN_C_END
 
 #endif

@@ -1,134 +1,256 @@
-/* Bcj2.c -- Converter for x86 code (BCJ2)
-2008-10-04 : Igor Pavlov : Public domain */
+/* Bcj2.c -- BCJ2 Decoder (Converter for x86 code)
+2015-08-01 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
 #include "Bcj2.h"
+#include "CpuArch.h"
 
-#ifdef _LZMA_PROB32
-#define CProb UInt32
-#else
 #define CProb UInt16
-#endif
 
-#define IsJcc(b0, b1) ((b0) == 0x0F && ((b1) & 0xF0) == 0x80)
-#define IsJ(b0, b1) ((b1 & 0xFE) == 0xE8 || IsJcc(b0, b1))
-
-#define kNumTopBits 24
-#define kTopValue ((UInt32)1 << kNumTopBits)
-
-#define kNumBitModelTotalBits 11
-#define kBitModelTotal (1 << kNumBitModelTotalBits)
+#define kTopValue ((UInt32)1 << 24)
+#define kNumModelBits 11
+#define kBitModelTotal (1 << kNumModelBits)
 #define kNumMoveBits 5
 
-#define RC_READ_BYTE (*buffer++)
-#define RC_TEST { if (buffer == bufferLim) return SZ_ERROR_DATA; }
-#define RC_INIT2 code = 0; range = 0xFFFFFFFF; \
-  { int i; for (i = 0; i < 5; i++) { RC_TEST; code = (code << 8) | RC_READ_BYTE; }}
-
-#define NORMALIZE if (range < kTopValue) { RC_TEST; range <<= 8; code = (code << 8) | RC_READ_BYTE; }
-
-#define IF_BIT_0(p) ttt = *(p); bound = (range >> kNumBitModelTotalBits) * ttt; if (code < bound)
-#define UPDATE_0(p) range = bound; *(p) = (CProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits)); NORMALIZE;
-#define UPDATE_1(p) range -= bound; code -= bound; *(p) = (CProb)(ttt - (ttt >> kNumMoveBits)); NORMALIZE;
+#define _IF_BIT_0 ttt = *prob; bound = (p->range >> kNumModelBits) * ttt; if (p->code < bound)
+#define _UPDATE_0 p->range = bound; *prob = (CProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits));
+#define _UPDATE_1 p->range -= bound; p->code -= bound; *prob = (CProb)(ttt - (ttt >> kNumMoveBits));
 
-int Bcj2_Decode(
-    const Byte *buf0, SizeT size0,
-    const Byte *buf1, SizeT size1,
-    const Byte *buf2, SizeT size2,
-    const Byte *buf3, SizeT size3,
-    Byte *outBuf, SizeT outSize)
+void Bcj2Dec_Init(CBcj2Dec *p)
 {
-  CProb p[256 + 2];
-  SizeT inPos = 0, outPos = 0;
+  unsigned i;
 
-  const Byte *buffer, *bufferLim;
-  UInt32 range, code;
-  Byte prevByte = 0;
+  p->state = BCJ2_DEC_STATE_OK;
+  p->ip = 0;
+  p->temp[3] = 0;
+  p->range = 0;
+  p->code = 0;
+  for (i = 0; i < sizeof(p->probs) / sizeof(p->probs[0]); i++)
+    p->probs[i] = kBitModelTotal >> 1;
+}
 
-  unsigned int i;
-  for (i = 0; i < sizeof(p) / sizeof(p[0]); i++)
-    p[i] = kBitModelTotal >> 1;
-
-  buffer = buf3;
-  bufferLim = buffer + size3;
-  RC_INIT2
-
-  if (outSize == 0)
-    return SZ_OK;
-
-  for (;;)
+SRes Bcj2Dec_Decode(CBcj2Dec *p)
+{
+  if (p->range <= 5)
   {
-    Byte b;
-    CProb *prob;
-    UInt32 bound;
-    UInt32 ttt;
+    p->state = BCJ2_DEC_STATE_OK;
+    for (; p->range != 5; p->range++)
+    {
+      if (p->range == 1 && p->code != 0)
+        return SZ_ERROR_DATA;
+      
+      if (p->bufs[BCJ2_STREAM_RC] == p->lims[BCJ2_STREAM_RC])
+      {
+        p->state = BCJ2_STREAM_RC;
+        return SZ_OK;
+      }
 
-    SizeT limit = size0 - inPos;
-    if (outSize - outPos < limit)
-      limit = outSize - outPos;
-    while (limit != 0)
+      p->code = (p->code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
+    }
+    
+    if (p->code == 0xFFFFFFFF)
+      return SZ_ERROR_DATA;
+    
+    p->range = 0xFFFFFFFF;
+  }
+  else if (p->state >= BCJ2_DEC_STATE_ORIG_0)
+  {
+    while (p->state <= BCJ2_DEC_STATE_ORIG_3)
     {
-      Byte b = buf0[inPos];
-      outBuf[outPos++] = b;
-      if (IsJ(prevByte, b))
-        break;
-      inPos++;
-      prevByte = b;
-      limit--;
+      Byte *dest = p->dest;
+      if (dest == p->destLim)
+        return SZ_OK;
+      *dest = p->temp[p->state++ - BCJ2_DEC_STATE_ORIG_0];
+      p->dest = dest + 1;
     }
-
-    if (limit == 0 || outPos == outSize)
-      break;
-
-    b = buf0[inPos++];
-
-    if (b == 0xE8)
-      prob = p + prevByte;
-    else if (b == 0xE9)
-      prob = p + 256;
-    else
-      prob = p + 257;
+  }
 
-    IF_BIT_0(prob)
+  /*
+  if (BCJ2_IS_32BIT_STREAM(p->state))
+  {
+    const Byte *cur = p->bufs[p->state];
+    if (cur == p->lims[p->state])
+      return SZ_OK;
+    p->bufs[p->state] = cur + 4;
+    
     {
-      UPDATE_0(prob)
-      prevByte = b;
+      UInt32 val;
+      Byte *dest;
+      SizeT rem;
+      
+      p->ip += 4;
+      val = GetBe32(cur) - p->ip;
+      dest = p->dest;
+      rem = p->destLim - dest;
+      if (rem < 4)
+      {
+        SizeT i;
+        SetUi32(p->temp, val);
+        for (i = 0; i < rem; i++)
+          dest[i] = p->temp[i];
+        p->dest = dest + rem;
+        p->state = BCJ2_DEC_STATE_ORIG_0 + (unsigned)rem;
+        return SZ_OK;
+      }
+      SetUi32(dest, val);
+      p->temp[3] = (Byte)(val >> 24);
+      p->dest = dest + 4;
+      p->state = BCJ2_DEC_STATE_OK;
     }
+  }
+  */
+
+  for (;;)
+  {
+    if (BCJ2_IS_32BIT_STREAM(p->state))
+      p->state = BCJ2_DEC_STATE_OK;
     else
     {
-      UInt32 dest;
-      const Byte *v;
-      UPDATE_1(prob)
-      if (b == 0xE8)
+      if (p->range < kTopValue)
       {
-        v = buf1;
-        if (size1 < 4)
-          return SZ_ERROR_DATA;
-        buf1 += 4;
-        size1 -= 4;
+        if (p->bufs[BCJ2_STREAM_RC] == p->lims[BCJ2_STREAM_RC])
+        {
+          p->state = BCJ2_STREAM_RC;
+          return SZ_OK;
+        }
+        p->range <<= 8;
+        p->code = (p->code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
       }
-      else
+
       {
-        v = buf2;
-        if (size2 < 4)
-          return SZ_ERROR_DATA;
-        buf2 += 4;
-        size2 -= 4;
+        const Byte *src = p->bufs[BCJ2_STREAM_MAIN];
+        const Byte *srcLim;
+        Byte *dest;
+        SizeT num = p->lims[BCJ2_STREAM_MAIN] - src;
+        
+        if (num == 0)
+        {
+          p->state = BCJ2_STREAM_MAIN;
+          return SZ_OK;
+        }
+        
+        dest = p->dest;
+        if (num > (SizeT)(p->destLim - dest))
+        {
+          num = p->destLim - dest;
+          if (num == 0)
+          {
+            p->state = BCJ2_DEC_STATE_ORIG;
+            return SZ_OK;
+          }
+        }
+       
+        srcLim = src + num;
+
+        if (p->temp[3] == 0x0F && (src[0] & 0xF0) == 0x80)
+          *dest = src[0];
+        else for (;;)
+        {
+          Byte b = *src;
+          *dest = b;
+          if (b != 0x0F)
+          {
+            if ((b & 0xFE) == 0xE8)
+              break;
+            dest++;
+            if (++src != srcLim)
+              continue;
+            break;
+          }
+          dest++;
+          if (++src == srcLim)
+            break;
+          if ((*src & 0xF0) != 0x80)
+            continue;
+          *dest = *src;
+          break;
+        }
+        
+        num = src - p->bufs[BCJ2_STREAM_MAIN];
+        
+        if (src == srcLim)
+        {
+          p->temp[3] = src[-1];
+          p->bufs[BCJ2_STREAM_MAIN] = src;
+          p->ip += (UInt32)num;
+          p->dest += num;
+          p->state =
+            p->bufs[BCJ2_STREAM_MAIN] ==
+            p->lims[BCJ2_STREAM_MAIN] ?
+              (unsigned)BCJ2_STREAM_MAIN :
+              (unsigned)BCJ2_DEC_STATE_ORIG;
+          return SZ_OK;
+        }
+        
+        {
+          UInt32 bound, ttt;
+          CProb *prob;
+          Byte b = src[0];
+          Byte prev = (Byte)(num == 0 ? p->temp[3] : src[-1]);
+          
+          p->temp[3] = b;
+          p->bufs[BCJ2_STREAM_MAIN] = src + 1;
+          num++;
+          p->ip += (UInt32)num;
+          p->dest += num;
+          
+          prob = p->probs + (unsigned)(b == 0xE8 ? 2 + (unsigned)prev : (b == 0xE9 ? 1 : 0));
+          
+          _IF_BIT_0
+          {
+            _UPDATE_0
+            continue;
+          }
+          _UPDATE_1
+            
+        }
       }
-      dest = (((UInt32)v[0] << 24) | ((UInt32)v[1] << 16) |
-          ((UInt32)v[2] << 8) | ((UInt32)v[3])) - ((UInt32)outPos + 4);
-      outBuf[outPos++] = (Byte)dest;
-      if (outPos == outSize)
-        break;
-      outBuf[outPos++] = (Byte)(dest >> 8);
-      if (outPos == outSize)
+    }
+
+    {
+      UInt32 val;
+      unsigned cj = (p->temp[3] == 0xE8) ? BCJ2_STREAM_CALL : BCJ2_STREAM_JUMP;
+      const Byte *cur = p->bufs[cj];
+      Byte *dest;
+      SizeT rem;
+      
+      if (cur == p->lims[cj])
+      {
+        p->state = cj;
         break;
-      outBuf[outPos++] = (Byte)(dest >> 16);
-      if (outPos == outSize)
+      }
+      
+      val = GetBe32(cur);
+      p->bufs[cj] = cur + 4;
+
+      p->ip += 4;
+      val -= p->ip;
+      dest = p->dest;
+      rem = p->destLim - dest;
+      
+      if (rem < 4)
+      {
+        SizeT i;
+        SetUi32(p->temp, val);
+        for (i = 0; i < rem; i++)
+          dest[i] = p->temp[i];
+        p->dest = dest + rem;
+        p->state = BCJ2_DEC_STATE_ORIG_0 + (unsigned)rem;
         break;
-      outBuf[outPos++] = prevByte = (Byte)(dest >> 24);
+      }
+      
+      SetUi32(dest, val);
+      p->temp[3] = (Byte)(val >> 24);
+      p->dest = dest + 4;
     }
   }
-  return (outPos == outSize) ? SZ_OK : SZ_ERROR_DATA;
+
+  if (p->range < kTopValue && p->bufs[BCJ2_STREAM_RC] != p->lims[BCJ2_STREAM_RC])
+  {
+    p->range <<= 8;
+    p->code = (p->code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
+  }
+
+  return SZ_OK;
 }

@@ -1,5 +1,5 @@
-/* Bcj2.h -- Converter for x86 code (BCJ2)
-2013-01-18 : Igor Pavlov : Public domain */
+/* Bcj2.h -- BCJ2 Converter for x86 code
+2014-11-10 : Igor Pavlov : Public domain */
 
 #ifndef __BCJ2_H
 #define __BCJ2_H
@@ -8,26 +8,138 @@ #include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
-/*
-Conditions:
-  outSize <= FullOutputSize,
-  where FullOutputSize is full size of output stream of x86_2 filter.
+#define BCJ2_NUM_STREAMS 4
+
+enum
+{
+  BCJ2_STREAM_MAIN,
+  BCJ2_STREAM_CALL,
+  BCJ2_STREAM_JUMP,
+  BCJ2_STREAM_RC
+};
 
-If buf0 overlaps outBuf, there are two required conditions:
-  1) (buf0 >= outBuf)
-  2) (buf0 + size0 >= outBuf + FullOutputSize).
+enum
+{
+  BCJ2_DEC_STATE_ORIG_0 = BCJ2_NUM_STREAMS,
+  BCJ2_DEC_STATE_ORIG_1,
+  BCJ2_DEC_STATE_ORIG_2,
+  BCJ2_DEC_STATE_ORIG_3,
+  
+  BCJ2_DEC_STATE_ORIG,
+  BCJ2_DEC_STATE_OK
+};
 
-Returns:
-  SZ_OK
-  SZ_ERROR_DATA - Data error
+enum
+{
+  BCJ2_ENC_STATE_ORIG = BCJ2_NUM_STREAMS,
+  BCJ2_ENC_STATE_OK
+};
+
+
+#define BCJ2_IS_32BIT_STREAM(s) ((s) == BCJ2_STREAM_CALL || (s) == BCJ2_STREAM_JUMP)
+
+/*
+CBcj2Dec / CBcj2Enc
+bufs sizes:
+  BUF_SIZE(n) = lims[n] - bufs[n]
+bufs sizes for BCJ2_STREAM_CALL and BCJ2_STREAM_JUMP must be mutliply of 4:
+    (BUF_SIZE(BCJ2_STREAM_CALL) & 3) == 0
+    (BUF_SIZE(BCJ2_STREAM_JUMP) & 3) == 0
 */
 
-int Bcj2_Decode(
-    const Byte *buf0, SizeT size0,
-    const Byte *buf1, SizeT size1,
-    const Byte *buf2, SizeT size2,
-    const Byte *buf3, SizeT size3,
-    Byte *outBuf, SizeT outSize);
+/*
+CBcj2Dec:
+dest is allowed to overlap with bufs[BCJ2_STREAM_MAIN], with the following conditions:
+  bufs[BCJ2_STREAM_MAIN] >= dest &&
+  bufs[BCJ2_STREAM_MAIN] - dest >= tempReserv +
+        BUF_SIZE(BCJ2_STREAM_CALL) +
+        BUF_SIZE(BCJ2_STREAM_JUMP)
+     tempReserv = 0 : for first call of Bcj2Dec_Decode
+     tempReserv = 4 : for any other calls of Bcj2Dec_Decode
+  overlap with offset = 1 is not allowed
+*/
+
+typedef struct
+{
+  const Byte *bufs[BCJ2_NUM_STREAMS];
+  const Byte *lims[BCJ2_NUM_STREAMS];
+  Byte *dest;
+  const Byte *destLim;
+
+  unsigned state; /* BCJ2_STREAM_MAIN has more priority than BCJ2_STATE_ORIG */
+
+  UInt32 ip;
+  Byte temp[4];
+  UInt32 range;
+  UInt32 code;
+  UInt16 probs[2 + 256];
+} CBcj2Dec;
+
+void Bcj2Dec_Init(CBcj2Dec *p);
+
+/* Returns: SZ_OK or SZ_ERROR_DATA */
+SRes Bcj2Dec_Decode(CBcj2Dec *p);
+
+#define Bcj2Dec_IsFinished(_p_) ((_p_)->code == 0)
+
+
+
+typedef enum
+{
+  BCJ2_ENC_FINISH_MODE_CONTINUE,
+  BCJ2_ENC_FINISH_MODE_END_BLOCK,
+  BCJ2_ENC_FINISH_MODE_END_STREAM
+} EBcj2Enc_FinishMode;
+
+typedef struct
+{
+  Byte *bufs[BCJ2_NUM_STREAMS];
+  const Byte *lims[BCJ2_NUM_STREAMS];
+  const Byte *src;
+  const Byte *srcLim;
+
+  unsigned state;
+  EBcj2Enc_FinishMode finishMode;
+
+  Byte prevByte;
+
+  Byte cache;
+  UInt32 range;
+  UInt64 low;
+  UInt64 cacheSize;
+
+  UInt32 ip;
+
+  /* 32-bit ralative offset in JUMP/CALL commands is
+       - (mod 4 GB)   in 32-bit mode
+       - signed Int32 in 64-bit mode
+     We use (mod 4 GB) check for fileSize.
+     Use fileSize up to 2 GB, if you want to support 32-bit and 64-bit code conversion. */
+  UInt32 fileIp;
+  UInt32 fileSize;    /* (fileSize <= ((UInt32)1 << 31)), 0 means no_limit */
+  UInt32 relatLimit;  /* (relatLimit <= ((UInt32)1 << 31)), 0 means desable_conversion */
+
+  UInt32 tempTarget;
+  unsigned tempPos;
+  Byte temp[4 * 2];
+
+  unsigned flushPos;
+  
+  UInt16 probs[2 + 256];
+} CBcj2Enc;
+
+void Bcj2Enc_Init(CBcj2Enc *p);
+void Bcj2Enc_Encode(CBcj2Enc *p);
+
+#define Bcj2Enc_Get_InputData_Size(p) ((SizeT)((p)->srcLim - (p)->src) + (p)->tempPos)
+#define Bcj2Enc_IsFinished(p) ((p)->flushPos == 5)
+
+
+#define BCJ2_RELAT_LIMIT_NUM_BITS 26
+#define BCJ2_RELAT_LIMIT ((UInt32)1 << BCJ2_RELAT_LIMIT_NUM_BITS)
+
+/* limit for CBcj2Enc::fileSize variable */
+#define BCJ2_FileSize_MAX ((UInt32)1 << 31)
 
 EXTERN_C_END
 

@@ -0,0 +1,312 @@
+/* Bcj2Enc.c -- BCJ2 Encoder (Converter for x86 code)
+2014-11-10 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+/* #define SHOW_STAT */
+
+#ifdef SHOW_STAT
+#include <stdio.h>
+#define PRF(x) x
+#else
+#define PRF(x)
+#endif
+
+#include <windows.h>
+#include <string.h>
+
+#include "Bcj2.h"
+#include "CpuArch.h"
+
+#define CProb UInt16
+
+#define kTopValue ((UInt32)1 << 24)
+#define kNumModelBits 11
+#define kBitModelTotal (1 << kNumModelBits)
+#define kNumMoveBits 5
+
+void Bcj2Enc_Init(CBcj2Enc *p)
+{
+  unsigned i;
+
+  p->state = BCJ2_ENC_STATE_OK;
+  p->finishMode = BCJ2_ENC_FINISH_MODE_CONTINUE;
+
+  p->prevByte = 0;
+
+  p->cache = 0;
+  p->range = 0xFFFFFFFF;
+  p->low = 0;
+  p->cacheSize = 1;
+
+  p->ip = 0;
+
+  p->fileIp = 0;
+  p->fileSize = 0;
+  p->relatLimit = BCJ2_RELAT_LIMIT;
+
+  p->tempPos = 0;
+
+  p->flushPos = 0;
+
+  for (i = 0; i < sizeof(p->probs) / sizeof(p->probs[0]); i++)
+    p->probs[i] = kBitModelTotal >> 1;
+}
+
+static Bool MY_FAST_CALL RangeEnc_ShiftLow(CBcj2Enc *p)
+{
+  if ((UInt32)p->low < (UInt32)0xFF000000 || (UInt32)(p->low >> 32) != 0)
+  {
+    Byte *buf = p->bufs[BCJ2_STREAM_RC];
+    do
+    {
+      if (buf == p->lims[BCJ2_STREAM_RC])
+      {
+        p->state = BCJ2_STREAM_RC;
+        p->bufs[BCJ2_STREAM_RC] = buf;
+        return True;
+      }
+      *buf++ = (Byte)(p->cache + (Byte)(p->low >> 32));
+      p->cache = 0xFF;
+    }
+    while (--p->cacheSize);
+    p->bufs[BCJ2_STREAM_RC] = buf;
+    p->cache = (Byte)((UInt32)p->low >> 24);
+  }
+  p->cacheSize++;
+  p->low = (UInt32)p->low << 8;
+  return False;
+}
+
+static void Bcj2Enc_Encode_2(CBcj2Enc *p)
+{
+  if (BCJ2_IS_32BIT_STREAM(p->state))
+  {
+    Byte *cur = p->bufs[p->state];
+    if (cur == p->lims[p->state])
+      return;
+    SetBe32(cur, p->tempTarget);
+    p->bufs[p->state] = cur + 4;
+  }
+
+  p->state = BCJ2_ENC_STATE_ORIG;
+
+  for (;;)
+  {
+    if (p->range < kTopValue)
+    {
+      if (RangeEnc_ShiftLow(p))
+        return;
+      p->range <<= 8;
+    }
+
+    {
+      {
+        const Byte *src = p->src;
+        const Byte *srcLim;
+        Byte *dest;
+        SizeT num = p->srcLim - src;
+
+        if (p->finishMode == BCJ2_ENC_FINISH_MODE_CONTINUE)
+        {
+          if (num <= 4)
+            return;
+          num -= 4;
+        }
+        else if (num == 0)
+          break;
+
+        dest = p->bufs[BCJ2_STREAM_MAIN];
+        if (num > (SizeT)(p->lims[BCJ2_STREAM_MAIN] - dest))
+        {
+          num = p->lims[BCJ2_STREAM_MAIN] - dest;
+          if (num == 0)
+          {
+            p->state = BCJ2_STREAM_MAIN;
+            return;
+          }
+        }
+       
+        srcLim = src + num;
+
+        if (p->prevByte == 0x0F && (src[0] & 0xF0) == 0x80)
+          *dest = src[0];
+        else for (;;)
+        {
+          Byte b = *src;
+          *dest = b;
+          if (b != 0x0F)
+          {
+            if ((b & 0xFE) == 0xE8)
+              break;
+            dest++;
+            if (++src != srcLim)
+              continue;
+            break;
+          }
+          dest++;
+          if (++src == srcLim)
+            break;
+          if ((*src & 0xF0) != 0x80)
+            continue;
+          *dest = *src;
+          break;
+        }
+        
+        num = src - p->src;
+        
+        if (src == srcLim)
+        {
+          p->prevByte = src[-1];
+          p->bufs[BCJ2_STREAM_MAIN] = dest;
+          p->src = src;
+          p->ip += (UInt32)num;
+          continue;
+        }
+ 
+        {
+          Byte context = (Byte)(num == 0 ? p->prevByte : src[-1]);
+          Bool needConvert;
+
+          p->bufs[BCJ2_STREAM_MAIN] = dest + 1;
+          p->ip += (UInt32)num + 1;
+          src++;
+          
+          needConvert = False;
+
+          if ((SizeT)(p->srcLim - src) >= 4)
+          {
+            UInt32 relatVal = GetUi32(src);
+            if ((p->fileSize == 0 || (UInt32)(p->ip + 4 + relatVal - p->fileIp) < p->fileSize)
+                && ((relatVal + p->relatLimit) >> 1) < p->relatLimit)
+              needConvert = True;
+          }
+
+          {
+            UInt32 bound;
+            unsigned ttt;
+            Byte b = src[-1];
+            CProb *prob = p->probs + (unsigned)(b == 0xE8 ? 2 + (unsigned)context : (b == 0xE9 ? 1 : 0));
+
+            ttt = *prob;
+            bound = (p->range >> kNumModelBits) * ttt;
+            
+            if (!needConvert)
+            {
+              p->range = bound;
+              *prob = (CProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits));
+              p->src = src;
+              p->prevByte = b;
+              continue;
+            }
+            
+            p->low += bound;
+            p->range -= bound;
+            *prob = (CProb)(ttt - (ttt >> kNumMoveBits));
+
+            {
+              UInt32 relatVal = GetUi32(src);
+              UInt32 absVal;
+              p->ip += 4;
+              absVal = p->ip + relatVal;
+              p->prevByte = src[3];
+              src += 4;
+              p->src = src;
+              {
+                unsigned cj = (b == 0xE8) ? BCJ2_STREAM_CALL : BCJ2_STREAM_JUMP;
+                Byte *cur = p->bufs[cj];
+                if (cur == p->lims[cj])
+                {
+                  p->state = cj;
+                  p->tempTarget = absVal;
+                  return;
+                }
+                SetBe32(cur, absVal);
+                p->bufs[cj] = cur + 4;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+
+  if (p->finishMode != BCJ2_ENC_FINISH_MODE_END_STREAM)
+    return;
+
+  for (; p->flushPos < 5; p->flushPos++)
+    if (RangeEnc_ShiftLow(p))
+      return;
+  p->state = BCJ2_ENC_STATE_OK;
+}
+
+
+void Bcj2Enc_Encode(CBcj2Enc *p)
+{
+  PRF(printf("\n"));
+  PRF(printf("---- ip = %8d   tempPos = %8d   src = %8d\n", p->ip, p->tempPos, p->srcLim - p->src));
+
+  if (p->tempPos != 0)
+  {
+    unsigned extra = 0;
+   
+    for (;;)
+    {
+      const Byte *src = p->src;
+      const Byte *srcLim = p->srcLim;
+      unsigned finishMode = p->finishMode;
+      
+      p->src = p->temp;
+      p->srcLim = p->temp + p->tempPos;
+      if (src != srcLim)
+        p->finishMode = BCJ2_ENC_FINISH_MODE_CONTINUE;
+      
+      PRF(printf("     ip = %8d   tempPos = %8d   src = %8d\n", p->ip, p->tempPos, p->srcLim - p->src));
+
+      Bcj2Enc_Encode_2(p);
+      
+      {
+        unsigned num = (unsigned)(p->src - p->temp);
+        unsigned tempPos = p->tempPos - num;
+        unsigned i;
+        p->tempPos = tempPos;
+        for (i = 0; i < tempPos; i++)
+          p->temp[i] = p->temp[i + num];
+      
+        p->src = src;
+        p->srcLim = srcLim;
+        p->finishMode = finishMode;
+        
+        if (p->state != BCJ2_ENC_STATE_ORIG || src == srcLim)
+          return;
+        
+        if (extra >= tempPos)
+        {
+          p->src = src - tempPos;
+          p->tempPos = 0;
+          break;
+        }
+        
+        p->temp[tempPos] = src[0];
+        p->tempPos = tempPos + 1;
+        p->src = src + 1;
+        extra++;
+      }
+    }
+  }
+
+  PRF(printf("++++ ip = %8d   tempPos = %8d   src = %8d\n", p->ip, p->tempPos, p->srcLim - p->src));
+
+  Bcj2Enc_Encode_2(p);
+  
+  if (p->state == BCJ2_ENC_STATE_ORIG)
+  {
+    const Byte *src = p->src;
+    unsigned rem = (unsigned)(p->srcLim - src);
+    unsigned i;
+    for (i = 0; i < rem; i++)
+      p->temp[i] = src[i];
+    p->tempPos = rem;
+    p->src = src + rem;
+  }
+}

@@ -1,5 +1,5 @@
-/* Compiler.h -- Compiler ypes
-2013-11-12 : Igor Pavlov : Public domain */
+/* Compiler.h
+2015-08-02 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_COMPILER_H
 #define __7Z_COMPILER_H
@@ -18,11 +18,15 @@ #pragma warning(disable : 4996) // This function or variable may be unsafe
   #else
     #pragma warning(disable : 4511) // copy constructor could not be generated
     #pragma warning(disable : 4512) // assignment operator could not be generated
+    #pragma warning(disable : 4514) // unreferenced inline function has been removed
     #pragma warning(disable : 4702) // unreachable code
     #pragma warning(disable : 4710) // not inlined
     #pragma warning(disable : 4786) // identifier was truncated to '255' characters in the debug information
   #endif
 
 #endif
+
+#define UNUSED_VAR(x) (void)x;
+/* #define UNUSED_VAR(x) x=x; */
 
 #endif

@@ -1,5 +1,5 @@
 /* CpuArch.c -- CPU specific code
-2012-05-29: Igor Pavlov : Public domain */
+2015-03-25: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -54,7 +54,7 @@ #else
 #define CHECK_CPUID_IS_SUPPORTED
 #endif
 
-static void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d)
+void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d)
 {
   #ifdef USE_ASM
 
@@ -116,7 +116,7 @@ MyCPUID(1, &p->ver, &p->b, &p->c, &p->d);
   return True;
 }
 
-static UInt32 kVendors[][3] =
+static const UInt32 kVendors[][3] =
 {
   { 0x756E6547, 0x49656E69, 0x6C65746E},
   { 0x68747541, 0x69746E65, 0x444D4163},
@@ -144,18 +144,21 @@ int firm;
   UInt32 family, model;
   if (!x86cpuid_CheckAndRead(&p))
     return True;
-  family = x86cpuid_GetFamily(&p);
-  model = x86cpuid_GetModel(&p);
+
+  family = x86cpuid_GetFamily(p.ver);
+  model = x86cpuid_GetModel(p.ver);
+  
   firm = x86cpuid_GetFirm(&p);
+
   switch (firm)
   {
     case CPU_FIRM_INTEL: return (family < 6 || (family == 6 && (
-        /* Atom CPU */
-           model == 0x100C  /* 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330 */
-        || model == 0x2006  /* 45 nm, Z6xx */
-        || model == 0x2007  /* 32 nm, Z2460 */
-        || model == 0x3005  /* 32 nm, Z2760 */
-        || model == 0x3006  /* 32 nm, N2xxx, D2xxx */
+        /* In-Order Atom CPU */
+           model == 0x1C  /* 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330 */
+        || model == 0x26  /* 45 nm, Z6xx */
+        || model == 0x27  /* 32 nm, Z2460 */
+        || model == 0x35  /* 32 nm, Z2760 */
+        || model == 0x36  /* 32 nm, N2xxx, D2xxx */
         )));
     case CPU_FIRM_AMD: return (family < 5 || (family == 5 && (model < 6 || model == 0xA)));
     case CPU_FIRM_VIA: return (family < 6 || (family == 6 && model < 0xF));

@@ -1,5 +1,5 @@
 /* CpuArch.h -- CPU specific code
-2013-11-12: Igor Pavlov : Public domain */
+2015-12-01: Igor Pavlov : Public domain */
 
 #ifndef __CPU_ARCH_H
 #define __CPU_ARCH_H
@@ -10,18 +10,25 @@ EXTERN_C_BEGIN
 
 /*
 MY_CPU_LE means that CPU is LITTLE ENDIAN.
-If MY_CPU_LE is not defined, we don't know about that property of platform (it can be LITTLE ENDIAN).
+MY_CPU_BE means that CPU is BIG ENDIAN.
+If MY_CPU_LE and MY_CPU_BE are not defined, we don't know about ENDIANNESS of platform.
 
 MY_CPU_LE_UNALIGN means that CPU is LITTLE ENDIAN and CPU supports unaligned memory accesses.
-If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of platform.
 */
 
-#if defined(_M_X64) || defined(_M_AMD64) || defined(__x86_64__)
-#define MY_CPU_AMD64
+#if defined(_M_X64) \
+   || defined(_M_AMD64) \
+   || defined(__x86_64__) \
+   || defined(__AMD64__) \
+   || defined(__amd64__)
+  #define MY_CPU_AMD64
 #endif
 
-#if defined(MY_CPU_AMD64) || defined(_M_IA64)
-#define MY_CPU_64BIT
+#if defined(MY_CPU_AMD64) \
+    || defined(_M_IA64) \
+    || defined(__AARCH64EL__) \
+    || defined(__AARCH64EB__)
+  #define MY_CPU_64BIT
 #endif
 
 #if defined(_M_IX86) || defined(__i386__)
@@ -32,8 +39,13 @@ #if defined(MY_CPU_X86) || defined(MY_CPU_AMD64)
 #define MY_CPU_X86_OR_AMD64
 #endif
 
-#if defined(MY_CPU_X86) || defined(_M_ARM)
-#define MY_CPU_32BIT
+#if defined(MY_CPU_X86) \
+    || defined(_M_ARM) \
+    || defined(__ARMEL__) \
+    || defined(__THUMBEL__) \
+    || defined(__ARMEB__) \
+    || defined(__THUMBEB__)
+  #define MY_CPU_32BIT
 #endif
 
 #if defined(_WIN32) && defined(_M_ARM)
@@ -44,34 +56,63 @@ #if defined(_WIN32) && defined(_M_IA64)
 #define MY_CPU_IA64_LE
 #endif
 
-#if defined(MY_CPU_X86_OR_AMD64)
-#define MY_CPU_LE_UNALIGN
+#if defined(MY_CPU_X86_OR_AMD64) \
+    || defined(MY_CPU_ARM_LE) \
+    || defined(MY_CPU_IA64_LE) \
+    || defined(__LITTLE_ENDIAN__) \
+    || defined(__ARMEL__) \
+    || defined(__THUMBEL__) \
+    || defined(__AARCH64EL__) \
+    || defined(__MIPSEL__) \
+    || defined(__MIPSEL) \
+    || defined(_MIPSEL) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
+  #define MY_CPU_LE
 #endif
 
-#if defined(MY_CPU_X86_OR_AMD64) || defined(MY_CPU_ARM_LE)  || defined(MY_CPU_IA64_LE) || defined(__ARMEL__) || defined(__MIPSEL__) || defined(__LITTLE_ENDIAN__)
-#define MY_CPU_LE
+#if defined(__BIG_ENDIAN__) \
+    || defined(__ARMEB__) \
+    || defined(__THUMBEB__) \
+    || defined(__AARCH64EB__) \
+    || defined(__MIPSEB__) \
+    || defined(__MIPSEB) \
+    || defined(_MIPSEB) \
+    || defined(__m68k__) \
+    || defined(__s390__) \
+    || defined(__s390x__) \
+    || defined(__zarch__) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))
+  #define MY_CPU_BE
 #endif
 
-#if defined(__BIG_ENDIAN__) || defined(__m68k__) ||  defined(__ARMEB__) || defined(__MIPSEB__)
-#define MY_CPU_BE
+#if defined(MY_CPU_LE) && defined(MY_CPU_BE)
+Stop_Compiling_Bad_Endian
 #endif
 
-#if defined(MY_CPU_LE) && defined(MY_CPU_BE)
-Stop_Compiling_Bad_Endian
+
+#ifdef MY_CPU_LE
+  #if defined(MY_CPU_X86_OR_AMD64) \
+      /* || defined(__AARCH64EL__) */
+    #define MY_CPU_LE_UNALIGN
+  #endif
 #endif
+
 
 #ifdef MY_CPU_LE_UNALIGN
 
 #define GetUi16(p) (*(const UInt16 *)(const void *)(p))
 #define GetUi32(p) (*(const UInt32 *)(const void *)(p))
 #define GetUi64(p) (*(const UInt64 *)(const void *)(p))
-#define SetUi16(p, d) *(UInt16 *)(p) = (d);
-#define SetUi32(p, d) *(UInt32 *)(p) = (d);
-#define SetUi64(p, d) *(UInt64 *)(p) = (d);
+
+#define SetUi16(p, v) { *(UInt16 *)(p) = (v); }
+#define SetUi32(p, v) { *(UInt32 *)(p) = (v); }
+#define SetUi64(p, v) { *(UInt64 *)(p) = (v); }
 
 #else
 
-#define GetUi16(p) (((const Byte *)(p))[0] | ((UInt16)((const Byte *)(p))[1] << 8))
+#define GetUi16(p) ( (UInt16) ( \
+             ((const Byte *)(p))[0] | \
+    ((UInt16)((const Byte *)(p))[1] << 8) ))
 
 #define GetUi32(p) ( \
              ((const Byte *)(p))[0]        | \
@@ -81,23 +122,26 @@ ((UInt32)((const Byte *)(p))[3] << 24))
 
 #define GetUi64(p) (GetUi32(p) | ((UInt64)GetUi32(((const Byte *)(p)) + 4) << 32))
 
-#define SetUi16(p, d) { UInt32 _x_ = (d); \
-    ((Byte *)(p))[0] = (Byte)_x_; \
-    ((Byte *)(p))[1] = (Byte)(_x_ >> 8); }
+#define SetUi16(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)_vvv_; \
+    _ppp_[1] = (Byte)(_vvv_ >> 8); }
 
-#define SetUi32(p, d) { UInt32 _x_ = (d); \
-    ((Byte *)(p))[0] = (Byte)_x_; \
-    ((Byte *)(p))[1] = (Byte)(_x_ >> 8); \
-    ((Byte *)(p))[2] = (Byte)(_x_ >> 16); \
-    ((Byte *)(p))[3] = (Byte)(_x_ >> 24); }
+#define SetUi32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)_vvv_; \
+    _ppp_[1] = (Byte)(_vvv_ >> 8); \
+    _ppp_[2] = (Byte)(_vvv_ >> 16); \
+    _ppp_[3] = (Byte)(_vvv_ >> 24); }
 
-#define SetUi64(p, d) { UInt64 _x64_ = (d); \
-    SetUi32(p, (UInt32)_x64_); \
-    SetUi32(((Byte *)(p)) + 4, (UInt32)(_x64_ >> 32)); }
+#define SetUi64(p, v) { Byte *_ppp2_ = (Byte *)(p); UInt64 _vvv2_ = (v); \
+    SetUi32(_ppp2_    , (UInt32)_vvv2_); \
+    SetUi32(_ppp2_ + 4, (UInt32)(_vvv2_ >> 32)); }
 
 #endif
 
-#if defined(MY_CPU_LE_UNALIGN) && defined(_WIN64) && (_MSC_VER >= 1300)
+
+#if defined(MY_CPU_LE_UNALIGN) && /* defined(_WIN64) && */ (_MSC_VER >= 1300)
+
+/* Note: we use bswap instruction, that is unsupported in 386 cpu */
 
 #include <stdlib.h>
 
@@ -106,6 +150,15 @@ #pragma intrinsic(_byteswap_uint64)
 #define GetBe32(p) _byteswap_ulong(*(const UInt32 *)(const Byte *)(p))
 #define GetBe64(p) _byteswap_uint64(*(const UInt64 *)(const Byte *)(p))
 
+#define SetBe32(p, v) (*(UInt32 *)(void *)(p)) = _byteswap_ulong(v)
+
+#elif defined(MY_CPU_LE_UNALIGN) && defined (__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))
+
+#define GetBe32(p) __builtin_bswap32(*(const UInt32 *)(const Byte *)(p))
+#define GetBe64(p) __builtin_bswap64(*(const UInt64 *)(const Byte *)(p))
+
+#define SetBe32(p, v) (*(UInt32 *)(void *)(p)) = __builtin_bswap32(v)
+
 #else
 
 #define GetBe32(p) ( \
@@ -116,9 +169,19 @@ ((const Byte *)(p))[3] )
 
 #define GetBe64(p) (((UInt64)GetBe32(p) << 32) | GetBe32(((const Byte *)(p)) + 4))
 
+#define SetBe32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)(_vvv_ >> 24); \
+    _ppp_[1] = (Byte)(_vvv_ >> 16); \
+    _ppp_[2] = (Byte)(_vvv_ >> 8); \
+    _ppp_[3] = (Byte)_vvv_; }
+
 #endif
 
-#define GetBe16(p) ((UInt16)(((UInt16)((const Byte *)(p))[0] << 8) | ((const Byte *)(p))[1]))
+
+#define GetBe16(p) ( (UInt16) ( \
+    ((UInt16)((const Byte *)(p))[0] << 8) | \
+             ((const Byte *)(p))[1] ))
+
 
 
 #ifdef MY_CPU_X86_OR_AMD64
@@ -140,12 +203,14 @@ CPU_FIRM_AMD,
   CPU_FIRM_VIA
 };
 
+void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d);
+
 Bool x86cpuid_CheckAndRead(Cx86cpuid *p);
 int x86cpuid_GetFirm(const Cx86cpuid *p);
 
-#define x86cpuid_GetFamily(p) (((p)->ver >> 8) & 0xFF00F)
-#define x86cpuid_GetModel(p) (((p)->ver >> 4) & 0xF00F)
-#define x86cpuid_GetStepping(p) ((p)->ver & 0xF)
+#define x86cpuid_GetFamily(ver) (((ver >> 16) & 0xFF0) | ((ver >> 8) & 0xF))
+#define x86cpuid_GetModel(ver)  (((ver >> 12) &  0xF0) | ((ver >> 4) & 0xF))
+#define x86cpuid_GetStepping(ver) (ver & 0xF)
 
 Bool CPU_Is_InOrder();
 Bool CPU_Is_Aes_Supported();

@@ -1,5 +1,5 @@
 /* LzFind.c -- Match finder for LZ algorithms
-2009-04-22 : Igor Pavlov : Public domain */
+2015-10-15 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -11,8 +11,8 @@
 #define kEmptyHashValue 0
 #define kMaxValForNormalize ((UInt32)0xFFFFFFFF)
 #define kNormalizeStepMin (1 << 10) /* it must be power of 2 */
-#define kNormalizeMask (~(kNormalizeStepMin - 1))
-#define kMaxHistorySize ((UInt32)3 << 30)
+#define kNormalizeMask (~(UInt32)(kNormalizeStepMin - 1))
+#define kMaxHistorySize ((UInt32)7 << 29)
 
 #define kStartMaxLen 3
 
@@ -21,7 +21,7 @@ {
   if (!p->directInput)
   {
     alloc->Free(alloc, p->bufferBase);
-    p->bufferBase = 0;
+    p->bufferBase = NULL;
   }
 }
 
@@ -35,17 +35,16 @@ {
     p->blockSize = blockSize;
     return 1;
   }
-  if (p->bufferBase == 0 || p->blockSize != blockSize)
+  if (!p->bufferBase || p->blockSize != blockSize)
   {
     LzInWindow_Free(p, alloc);
     p->blockSize = blockSize;
     p->bufferBase = (Byte *)alloc->Alloc(alloc, (size_t)blockSize);
   }
-  return (p->bufferBase != 0);
+  return (p->bufferBase != NULL);
 }
 
 Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p) { return p->buffer; }
-Byte MatchFinder_GetIndexByte(CMatchFinder *p, Int32 index) { return p->buffer[index]; }
 
 UInt32 MatchFinder_GetNumAvailableBytes(CMatchFinder *p) { return p->streamPos - p->pos; }
 
@@ -60,9 +59,12 @@ static void MatchFinder_ReadBlock(CMatchFinder *p)
 {
   if (p->streamEndWasReached || p->result != SZ_OK)
     return;
+
+  /* We use (p->streamPos - p->pos) value. (p->streamPos < p->pos) is allowed. */
+
   if (p->directInput)
   {
-    UInt32 curSize = 0xFFFFFFFF - p->streamPos;
+    UInt32 curSize = 0xFFFFFFFF - (p->streamPos - p->pos);
     if (curSize > p->directInputRem)
       curSize = (UInt32)p->directInputRem;
     p->directInputRem -= curSize;
@@ -71,12 +73,14 @@ if (p->directInputRem == 0)
       p->streamEndWasReached = 1;
     return;
   }
+  
   for (;;)
   {
     Byte *dest = p->buffer + (p->streamPos - p->pos);
     size_t size = (p->bufferBase + p->blockSize - dest);
     if (size == 0)
       return;
+
     p->result = p->stream->Read(p->stream, dest, &size);
     if (p->result != SZ_OK)
       return;
@@ -94,8 +98,8 @@
 void MatchFinder_MoveBlock(CMatchFinder *p)
 {
   memmove(p->bufferBase,
-    p->buffer - p->keepSizeBefore,
-    (size_t)(p->streamPos - p->pos + p->keepSizeBefore));
+      p->buffer - p->keepSizeBefore,
+      (size_t)(p->streamPos - p->pos) + p->keepSizeBefore);
   p->buffer = p->bufferBase + p->keepSizeBefore;
 }
 
@@ -135,15 +139,15 @@
 void MatchFinder_Construct(CMatchFinder *p)
 {
   UInt32 i;
-  p->bufferBase = 0;
+  p->bufferBase = NULL;
   p->directInput = 0;
-  p->hash = 0;
+  p->hash = NULL;
   MatchFinder_SetDefaultSettings(p);
 
   for (i = 0; i < 256; i++)
   {
     UInt32 r = i;
-    int j;
+    unsigned j;
     for (j = 0; j < 8; j++)
       r = (r >> 1) ^ (kCrcPoly & ~((r & 1) - 1));
     p->crc[i] = r;
@@ -153,7 +157,7 @@
 static void MatchFinder_FreeThisClassMemory(CMatchFinder *p, ISzAlloc *alloc)
 {
   alloc->Free(alloc, p->hash);
-  p->hash = 0;
+  p->hash = NULL;
 }
 
 void MatchFinder_Free(CMatchFinder *p, ISzAlloc *alloc)
@@ -162,11 +166,11 @@ MatchFinder_FreeThisClassMemory(p, alloc);
   LzInWindow_Free(p, alloc);
 }
 
-static CLzRef* AllocRefs(UInt32 num, ISzAlloc *alloc)
+static CLzRef* AllocRefs(size_t num, ISzAlloc *alloc)
 {
   size_t sizeInBytes = (size_t)num * sizeof(CLzRef);
   if (sizeInBytes / sizeof(CLzRef) != num)
-    return 0;
+    return NULL;
   return (CLzRef *)alloc->Alloc(alloc, sizeInBytes);
 }
 
@@ -175,19 +179,24 @@ UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter,
     ISzAlloc *alloc)
 {
   UInt32 sizeReserv;
+  
   if (historySize > kMaxHistorySize)
   {
     MatchFinder_Free(p, alloc);
     return 0;
   }
+  
   sizeReserv = historySize >> 1;
-  if (historySize > ((UInt32)2 << 30))
-    sizeReserv = historySize >> 2;
+       if (historySize >= ((UInt32)3 << 30)) sizeReserv = historySize >> 3;
+  else if (historySize >= ((UInt32)2 << 30)) sizeReserv = historySize >> 2;
+  
   sizeReserv += (keepAddBufferBefore + matchMaxLen + keepAddBufferAfter) / 2 + (1 << 19);
 
   p->keepSizeBefore = historySize + keepAddBufferBefore + 1;
   p->keepSizeAfter = matchMaxLen + keepAddBufferAfter;
+  
   /* we need one additional byte, since we use MoveBlock after pos++ and before dictionary using */
+  
   if (LzInWindow_Create(p, sizeReserv, alloc))
   {
     UInt32 newCyclicBufferSize = historySize + 1;
@@ -212,6 +221,7 @@ if (p->numHashBytes == 3)
             hs = (1 << 24) - 1;
           else
             hs >>= 1;
+          /* if (bigHash) mode, GetHeads4b() in LzFindMt.c needs (hs >= ((1 << 24) - 1))) */
         }
       }
       p->hashMask = hs;
@@ -223,24 +233,32 @@ hs += p->fixedHashSize;
     }
 
     {
-      UInt32 prevSize = p->hashSizeSum + p->numSons;
-      UInt32 newSize;
+      size_t newSize;
+      size_t numSons;
       p->historySize = historySize;
       p->hashSizeSum = hs;
       p->cyclicBufferSize = newCyclicBufferSize;
-      p->numSons = (p->btMode ? newCyclicBufferSize * 2 : newCyclicBufferSize);
-      newSize = p->hashSizeSum + p->numSons;
-      if (p->hash != 0 && prevSize == newSize)
+      
+      numSons = newCyclicBufferSize;
+      if (p->btMode)
+        numSons <<= 1;
+      newSize = hs + numSons;
+
+      if (p->hash && p->numRefs == newSize)
         return 1;
+      
       MatchFinder_FreeThisClassMemory(p, alloc);
+      p->numRefs = newSize;
       p->hash = AllocRefs(newSize, alloc);
-      if (p->hash != 0)
+      
+      if (p->hash)
       {
         p->son = p->hash + p->hashSizeSum;
         return 1;
       }
     }
   }
+
   MatchFinder_Free(p, alloc);
   return 0;
 }
@@ -249,9 +267,11 @@ static void MatchFinder_SetLimits(CMatchFinder *p)
 {
   UInt32 limit = kMaxValForNormalize - p->pos;
   UInt32 limit2 = p->cyclicBufferSize - p->cyclicBufferPos;
+  
   if (limit2 < limit)
     limit = limit2;
   limit2 = p->streamPos - p->pos;
+  
   if (limit2 <= p->keepSizeAfter)
   {
     if (limit2 > 0)
@@ -259,8 +279,10 @@ limit2 = 1;
   }
   else
     limit2 -= p->keepSizeAfter;
+  
   if (limit2 < limit)
     limit = limit2;
+  
   {
     UInt32 lenLimit = p->streamPos - p->pos;
     if (lenLimit > p->matchMaxLen)
@@ -270,28 +292,39 @@ }
   p->posLimit = p->pos + limit;
 }
 
-void MatchFinder_Init(CMatchFinder *p)
+void MatchFinder_Init_2(CMatchFinder *p, int readData)
 {
   UInt32 i;
-  for (i = 0; i < p->hashSizeSum; i++)
-    p->hash[i] = kEmptyHashValue;
+  UInt32 *hash = p->hash;
+  UInt32 num = p->hashSizeSum;
+  for (i = 0; i < num; i++)
+    hash[i] = kEmptyHashValue;
+  
   p->cyclicBufferPos = 0;
   p->buffer = p->bufferBase;
   p->pos = p->streamPos = p->cyclicBufferSize;
   p->result = SZ_OK;
   p->streamEndWasReached = 0;
-  MatchFinder_ReadBlock(p);
+  
+  if (readData)
+    MatchFinder_ReadBlock(p);
+  
   MatchFinder_SetLimits(p);
 }
 
+void MatchFinder_Init(CMatchFinder *p)
+{
+  MatchFinder_Init_2(p, True);
+}
+  
 static UInt32 MatchFinder_GetSubValue(CMatchFinder *p)
 {
   return (p->pos - p->historySize - 1) & kNormalizeMask;
 }
 
-void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, UInt32 numItems)
+void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems)
 {
-  UInt32 i;
+  size_t i;
   for (i = 0; i < numItems; i++)
   {
     UInt32 value = items[i];
@@ -306,7 +339,7 @@
 static void MatchFinder_Normalize(CMatchFinder *p)
 {
   UInt32 subValue = MatchFinder_GetSubValue(p);
-  MatchFinder_Normalize3(subValue, p->hash, p->hashSizeSum + p->numSons);
+  MatchFinder_Normalize3(subValue, p->hash, p->numRefs);
   MatchFinder_ReduceOffsets(p, subValue);
 }
 
@@ -467,7 +500,7 @@
 static void MatchFinder_MovePos(CMatchFinder *p) { MOVE_POS; }
 
 #define GET_MATCHES_HEADER2(minLen, ret_op) \
-  UInt32 lenLimit; UInt32 hashValue; const Byte *cur; UInt32 curMatch; \
+  UInt32 lenLimit; UInt32 hv; const Byte *cur; UInt32 curMatch; \
   lenLimit = p->lenLimit; { if (lenLimit < minLen) { MatchFinder_MovePos(p); ret_op; }} \
   cur = p->buffer;
 
@@ -483,13 +516,20 @@
 #define SKIP_FOOTER \
   SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p)); MOVE_POS;
 
+#define UPDATE_maxLen { \
+    ptrdiff_t diff = (ptrdiff_t)0 - d2; \
+    const Byte *c = cur + maxLen; \
+    const Byte *lim = cur + lenLimit; \
+    for (; c != lim; c++) if (*(c + diff) != *c) break; \
+    maxLen = (UInt32)(c - cur); }
+
 static UInt32 Bt2_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 {
   UInt32 offset;
   GET_MATCHES_HEADER(2)
   HASH2_CALC;
-  curMatch = p->hash[hashValue];
-  p->hash[hashValue] = p->pos;
+  curMatch = p->hash[hv];
+  p->hash[hv] = p->pos;
   offset = 0;
   GET_MATCHES_FOOTER(offset, 1)
 }
@@ -499,35 +539,38 @@ {
   UInt32 offset;
   GET_MATCHES_HEADER(3)
   HASH_ZIP_CALC;
-  curMatch = p->hash[hashValue];
-  p->hash[hashValue] = p->pos;
+  curMatch = p->hash[hv];
+  p->hash[hv] = p->pos;
   offset = 0;
   GET_MATCHES_FOOTER(offset, 2)
 }
 
 static UInt32 Bt3_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 {
-  UInt32 hash2Value, delta2, maxLen, offset;
+  UInt32 h2, d2, maxLen, offset, pos;
+  UInt32 *hash;
   GET_MATCHES_HEADER(3)
 
   HASH3_CALC;
 
-  delta2 = p->pos - p->hash[hash2Value];
-  curMatch = p->hash[kFix3HashSize + hashValue];
+  hash = p->hash;
+  pos = p->pos;
+
+  d2 = pos - hash[h2];
+
+  curMatch = hash[kFix3HashSize + hv];
   
-  p->hash[hash2Value] =
-  p->hash[kFix3HashSize + hashValue] = p->pos;
-
+  hash[h2] = pos;
+  hash[kFix3HashSize + hv] = pos;
 
   maxLen = 2;
   offset = 0;
-  if (delta2 < p->cyclicBufferSize && *(cur - delta2) == *cur)
+
+  if (d2 < p->cyclicBufferSize && *(cur - d2) == *cur)
   {
-    for (; maxLen != lenLimit; maxLen++)
-      if (cur[(ptrdiff_t)maxLen - delta2] != cur[maxLen])
-        break;
+    UPDATE_maxLen
     distances[0] = maxLen;
-    distances[1] = delta2 - 1;
+    distances[1] = d2 - 1;
     offset = 2;
     if (maxLen == lenLimit)
     {
@@ -535,44 +578,51 @@ SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p));
       MOVE_POS_RET;
     }
   }
+  
   GET_MATCHES_FOOTER(offset, maxLen)
 }
 
 static UInt32 Bt4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 {
-  UInt32 hash2Value, hash3Value, delta2, delta3, maxLen, offset;
+  UInt32 h2, h3, d2, d3, maxLen, offset, pos;
+  UInt32 *hash;
   GET_MATCHES_HEADER(4)
 
   HASH4_CALC;
 
-  delta2 = p->pos - p->hash[                hash2Value];
-  delta3 = p->pos - p->hash[kFix3HashSize + hash3Value];
-  curMatch = p->hash[kFix4HashSize + hashValue];
-  
-  p->hash[                hash2Value] =
-  p->hash[kFix3HashSize + hash3Value] =
-  p->hash[kFix4HashSize + hashValue] = p->pos;
+  hash = p->hash;
+  pos = p->pos;
+
+  d2 = pos - hash[                h2];
+  d3 = pos - hash[kFix3HashSize + h3];
+
+  curMatch = hash[kFix4HashSize + hv];
 
-  maxLen = 1;
+  hash[                h2] = pos;
+  hash[kFix3HashSize + h3] = pos;
+  hash[kFix4HashSize + hv] = pos;
+
+  maxLen = 0;
   offset = 0;
-  if (delta2 < p->cyclicBufferSize && *(cur - delta2) == *cur)
+  
+  if (d2 < p->cyclicBufferSize && *(cur - d2) == *cur)
   {
     distances[0] = maxLen = 2;
-    distances[1] = delta2 - 1;
+    distances[1] = d2 - 1;
     offset = 2;
   }
-  if (delta2 != delta3 && delta3 < p->cyclicBufferSize && *(cur - delta3) == *cur)
+  
+  if (d2 != d3 && d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
   {
     maxLen = 3;
-    distances[offset + 1] = delta3 - 1;
+    distances[offset + 1] = d3 - 1;
     offset += 2;
-    delta2 = delta3;
+    d2 = d3;
   }
+  
   if (offset != 0)
   {
-    for (; maxLen != lenLimit; maxLen++)
-      if (cur[(ptrdiff_t)maxLen - delta2] != cur[maxLen])
-        break;
+    UPDATE_maxLen
     distances[offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
@@ -580,46 +630,131 @@ SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p));
       MOVE_POS_RET;
     }
   }
+  
   if (maxLen < 3)
     maxLen = 3;
+  
   GET_MATCHES_FOOTER(offset, maxLen)
 }
 
+/*
+static UInt32 Bt5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
+{
+  UInt32 h2, h3, h4, d2, d3, d4, maxLen, offset, pos;
+  UInt32 *hash;
+  GET_MATCHES_HEADER(5)
+
+  HASH5_CALC;
+
+  hash = p->hash;
+  pos = p->pos;
+
+  d2 = pos - hash[                h2];
+  d3 = pos - hash[kFix3HashSize + h3];
+  d4 = pos - hash[kFix4HashSize + h4];
+
+  curMatch = hash[kFix5HashSize + hv];
+
+  hash[                h2] = pos;
+  hash[kFix3HashSize + h3] = pos;
+  hash[kFix4HashSize + h4] = pos;
+  hash[kFix5HashSize + hv] = pos;
+
+  maxLen = 0;
+  offset = 0;
+
+  if (d2 < p->cyclicBufferSize && *(cur - d2) == *cur)
+  {
+    distances[0] = maxLen = 2;
+    distances[1] = d2 - 1;
+    offset = 2;
+    if (*(cur - d2 + 2) == cur[2])
+      distances[0] = maxLen = 3;
+    else if (d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
+    {
+      distances[2] = maxLen = 3;
+      distances[3] = d3 - 1;
+      offset = 4;
+      d2 = d3;
+    }
+  }
+  else if (d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
+  {
+    distances[0] = maxLen = 3;
+    distances[1] = d3 - 1;
+    offset = 2;
+    d2 = d3;
+  }
+  
+  if (d2 != d4 && d4 < p->cyclicBufferSize
+      && *(cur - d4) == *cur
+      && *(cur - d4 + 3) == *(cur + 3))
+  {
+    maxLen = 4;
+    distances[offset + 1] = d4 - 1;
+    offset += 2;
+    d2 = d4;
+  }
+  
+  if (offset != 0)
+  {
+    UPDATE_maxLen
+    distances[offset - 2] = maxLen;
+    if (maxLen == lenLimit)
+    {
+      SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p));
+      MOVE_POS_RET;
+    }
+  }
+
+  if (maxLen < 4)
+    maxLen = 4;
+  
+  GET_MATCHES_FOOTER(offset, maxLen)
+}
+*/
+
 static UInt32 Hc4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 {
-  UInt32 hash2Value, hash3Value, delta2, delta3, maxLen, offset;
+  UInt32 h2, h3, d2, d3, maxLen, offset, pos;
+  UInt32 *hash;
   GET_MATCHES_HEADER(4)
 
   HASH4_CALC;
 
-  delta2 = p->pos - p->hash[                hash2Value];
-  delta3 = p->pos - p->hash[kFix3HashSize + hash3Value];
-  curMatch = p->hash[kFix4HashSize + hashValue];
+  hash = p->hash;
+  pos = p->pos;
+  
+  d2 = pos - hash[                h2];
+  d3 = pos - hash[kFix3HashSize + h3];
+  
+  curMatch = hash[kFix4HashSize + hv];
 
-  p->hash[                hash2Value] =
-  p->hash[kFix3HashSize + hash3Value] =
-  p->hash[kFix4HashSize + hashValue] = p->pos;
+  hash[                h2] = pos;
+  hash[kFix3HashSize + h3] = pos;
+  hash[kFix4HashSize + hv] = pos;
 
-  maxLen = 1;
+  maxLen = 0;
   offset = 0;
-  if (delta2 < p->cyclicBufferSize && *(cur - delta2) == *cur)
+
+  if (d2 < p->cyclicBufferSize && *(cur - d2) == *cur)
   {
     distances[0] = maxLen = 2;
-    distances[1] = delta2 - 1;
+    distances[1] = d2 - 1;
     offset = 2;
   }
-  if (delta2 != delta3 && delta3 < p->cyclicBufferSize && *(cur - delta3) == *cur)
+  
+  if (d2 != d3 && d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
   {
     maxLen = 3;
-    distances[offset + 1] = delta3 - 1;
+    distances[offset + 1] = d3 - 1;
     offset += 2;
-    delta2 = delta3;
+    d2 = d3;
   }
+  
   if (offset != 0)
   {
-    for (; maxLen != lenLimit; maxLen++)
-      if (cur[(ptrdiff_t)maxLen - delta2] != cur[maxLen])
-        break;
+    UPDATE_maxLen
     distances[offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
@@ -627,22 +762,103 @@ p->son[p->cyclicBufferPos] = curMatch;
       MOVE_POS_RET;
     }
   }
+  
   if (maxLen < 3)
     maxLen = 3;
+
   offset = (UInt32)(Hc_GetMatchesSpec(lenLimit, curMatch, MF_PARAMS(p),
-    distances + offset, maxLen) - (distances));
+      distances + offset, maxLen) - (distances));
   MOVE_POS_RET
 }
 
+/*
+static UInt32 Hc5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
+{
+  UInt32 h2, h3, h4, d2, d3, d4, maxLen, offset, pos
+  UInt32 *hash;
+  GET_MATCHES_HEADER(5)
+
+  HASH5_CALC;
+
+  hash = p->hash;
+  pos = p->pos;
+  
+  d2 = pos - hash[                h2];
+  d3 = pos - hash[kFix3HashSize + h3];
+  d4 = pos - hash[kFix4HashSize + h4];
+
+  curMatch = hash[kFix5HashSize + hv];
+
+  hash[                h2] = pos;
+  hash[kFix3HashSize + h3] = pos;
+  hash[kFix4HashSize + h4] = pos;
+  hash[kFix5HashSize + hv] = pos;
+
+  maxLen = 0;
+  offset = 0;
+
+  if (d2 < p->cyclicBufferSize && *(cur - d2) == *cur)
+  {
+    distances[0] = maxLen = 2;
+    distances[1] = d2 - 1;
+    offset = 2;
+    if (*(cur - d2 + 2) == cur[2])
+      distances[0] = maxLen = 3;
+    else if (d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
+    {
+      distances[2] = maxLen = 3;
+      distances[3] = d3 - 1;
+      offset = 4;
+      d2 = d3;
+    }
+  }
+  else if (d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
+  {
+    distances[0] = maxLen = 3;
+    distances[1] = d3 - 1;
+    offset = 2;
+    d2 = d3;
+  }
+  
+  if (d2 != d4 && d4 < p->cyclicBufferSize
+      && *(cur - d4) == *cur
+      && *(cur - d4 + 3) == *(cur + 3))
+  {
+    maxLen = 4;
+    distances[offset + 1] = d4 - 1;
+    offset += 2;
+    d2 = d4;
+  }
+  
+  if (offset != 0)
+  {
+    UPDATE_maxLen
+    distances[offset - 2] = maxLen;
+    if (maxLen == lenLimit)
+    {
+      p->son[p->cyclicBufferPos] = curMatch;
+      MOVE_POS_RET;
+    }
+  }
+  
+  if (maxLen < 4)
+    maxLen = 4;
+
+  offset = (UInt32)(Hc_GetMatchesSpec(lenLimit, curMatch, MF_PARAMS(p),
+      distances + offset, maxLen) - (distances));
+  MOVE_POS_RET
+}
+*/
+
 UInt32 Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 {
   UInt32 offset;
   GET_MATCHES_HEADER(3)
   HASH_ZIP_CALC;
-  curMatch = p->hash[hashValue];
-  p->hash[hashValue] = p->pos;
+  curMatch = p->hash[hv];
+  p->hash[hv] = p->pos;
   offset = (UInt32)(Hc_GetMatchesSpec(lenLimit, curMatch, MF_PARAMS(p),
-    distances, 2) - (distances));
+      distances, 2) - (distances));
   MOVE_POS_RET
 }
 
@@ -652,8 +868,8 @@ do
   {
     SKIP_HEADER(2)
     HASH2_CALC;
-    curMatch = p->hash[hashValue];
-    p->hash[hashValue] = p->pos;
+    curMatch = p->hash[hv];
+    p->hash[hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -665,8 +881,8 @@ do
   {
     SKIP_HEADER(3)
     HASH_ZIP_CALC;
-    curMatch = p->hash[hashValue];
-    p->hash[hashValue] = p->pos;
+    curMatch = p->hash[hv];
+    p->hash[hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -676,12 +892,14 @@ static void Bt3_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
 {
   do
   {
-    UInt32 hash2Value;
+    UInt32 h2;
+    UInt32 *hash;
     SKIP_HEADER(3)
     HASH3_CALC;
-    curMatch = p->hash[kFix3HashSize + hashValue];
-    p->hash[hash2Value] =
-    p->hash[kFix3HashSize + hashValue] = p->pos;
+    hash = p->hash;
+    curMatch = hash[kFix3HashSize + hv];
+    hash[h2] =
+    hash[kFix3HashSize + hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -691,43 +909,90 @@ static void Bt4_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
 {
   do
   {
-    UInt32 hash2Value, hash3Value;
+    UInt32 h2, h3;
+    UInt32 *hash;
     SKIP_HEADER(4)
     HASH4_CALC;
-    curMatch = p->hash[kFix4HashSize + hashValue];
-    p->hash[                hash2Value] =
-    p->hash[kFix3HashSize + hash3Value] = p->pos;
-    p->hash[kFix4HashSize + hashValue] = p->pos;
+    hash = p->hash;
+    curMatch = hash[kFix4HashSize + hv];
+    hash[                h2] =
+    hash[kFix3HashSize + h3] =
+    hash[kFix4HashSize + hv] = p->pos;
+    SKIP_FOOTER
+  }
+  while (--num != 0);
+}
+
+/*
+static void Bt5_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
+{
+  do
+  {
+    UInt32 h2, h3, h4;
+    UInt32 *hash;
+    SKIP_HEADER(5)
+    HASH5_CALC;
+    hash = p->hash;
+    curMatch = hash[kFix5HashSize + hv];
+    hash[                h2] =
+    hash[kFix3HashSize + h3] =
+    hash[kFix4HashSize + h4] =
+    hash[kFix5HashSize + hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
 }
+*/
 
 static void Hc4_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
 {
   do
   {
-    UInt32 hash2Value, hash3Value;
+    UInt32 h2, h3;
+    UInt32 *hash;
     SKIP_HEADER(4)
     HASH4_CALC;
-    curMatch = p->hash[kFix4HashSize + hashValue];
-    p->hash[                hash2Value] =
-    p->hash[kFix3HashSize + hash3Value] =
-    p->hash[kFix4HashSize + hashValue] = p->pos;
+    hash = p->hash;
+    curMatch = hash[kFix4HashSize + hv];
+    hash[                h2] =
+    hash[kFix3HashSize + h3] =
+    hash[kFix4HashSize + hv] = p->pos;
     p->son[p->cyclicBufferPos] = curMatch;
     MOVE_POS
   }
   while (--num != 0);
 }
 
+/*
+static void Hc5_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
+{
+  do
+  {
+    UInt32 h2, h3, h4;
+    UInt32 *hash;
+    SKIP_HEADER(5)
+    HASH5_CALC;
+    hash = p->hash;
+    curMatch = p->hash[kFix5HashSize + hv];
+    hash[                h2] =
+    hash[kFix3HashSize + h3] =
+    hash[kFix4HashSize + h4] =
+    hash[kFix5HashSize + hv] = p->pos;
+    p->son[p->cyclicBufferPos] = curMatch;
+    MOVE_POS
+  }
+  while (--num != 0);
+}
+*/
+
 void Hc3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
 {
   do
   {
     SKIP_HEADER(3)
     HASH_ZIP_CALC;
-    curMatch = p->hash[hashValue];
-    p->hash[hashValue] = p->pos;
+    curMatch = p->hash[hv];
+    p->hash[hv] = p->pos;
     p->son[p->cyclicBufferPos] = curMatch;
     MOVE_POS
   }
@@ -737,13 +1002,22 @@
 void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder *vTable)
 {
   vTable->Init = (Mf_Init_Func)MatchFinder_Init;
-  vTable->GetIndexByte = (Mf_GetIndexByte_Func)MatchFinder_GetIndexByte;
   vTable->GetNumAvailableBytes = (Mf_GetNumAvailableBytes_Func)MatchFinder_GetNumAvailableBytes;
   vTable->GetPointerToCurrentPos = (Mf_GetPointerToCurrentPos_Func)MatchFinder_GetPointerToCurrentPos;
   if (!p->btMode)
   {
-    vTable->GetMatches = (Mf_GetMatches_Func)Hc4_MatchFinder_GetMatches;
-    vTable->Skip = (Mf_Skip_Func)Hc4_MatchFinder_Skip;
+    /* if (p->numHashBytes <= 4) */
+    {
+      vTable->GetMatches = (Mf_GetMatches_Func)Hc4_MatchFinder_GetMatches;
+      vTable->Skip = (Mf_Skip_Func)Hc4_MatchFinder_Skip;
+    }
+    /*
+    else
+    {
+      vTable->GetMatches = (Mf_GetMatches_Func)Hc5_MatchFinder_GetMatches;
+      vTable->Skip = (Mf_Skip_Func)Hc5_MatchFinder_Skip;
+    }
+    */
   }
   else if (p->numHashBytes == 2)
   {
@@ -755,9 +1029,16 @@ {
     vTable->GetMatches = (Mf_GetMatches_Func)Bt3_MatchFinder_GetMatches;
     vTable->Skip = (Mf_Skip_Func)Bt3_MatchFinder_Skip;
   }
-  else
+  else /* if (p->numHashBytes == 4) */
   {
     vTable->GetMatches = (Mf_GetMatches_Func)Bt4_MatchFinder_GetMatches;
     vTable->Skip = (Mf_Skip_Func)Bt4_MatchFinder_Skip;
   }
+  /*
+  else
+  {
+    vTable->GetMatches = (Mf_GetMatches_Func)Bt5_MatchFinder_GetMatches;
+    vTable->Skip = (Mf_Skip_Func)Bt5_MatchFinder_Skip;
+  }
+  */
 }

@@ -1,5 +1,5 @@
 /* LzFind.h -- Match finder for LZ algorithms
-2013-01-18 : Igor Pavlov : Public domain */
+2015-10-15 : Igor Pavlov : Public domain */
 
 #ifndef __LZ_FIND_H
 #define __LZ_FIND_H
@@ -20,6 +20,11 @@ UInt32 lenLimit;
 
   UInt32 cyclicBufferPos;
   UInt32 cyclicBufferSize; /* it must be = (historySize + 1) */
+
+  Byte streamEndWasReached;
+  Byte btMode;
+  Byte bigHash;
+  Byte directInput;
 
   UInt32 matchMaxLen;
   CLzRef *hash;
@@ -29,30 +34,30 @@ UInt32 cutValue;
 
   Byte *bufferBase;
   ISeqInStream *stream;
-  int streamEndWasReached;
-
+  
   UInt32 blockSize;
   UInt32 keepSizeBefore;
   UInt32 keepSizeAfter;
 
   UInt32 numHashBytes;
-  int directInput;
   size_t directInputRem;
-  int btMode;
-  int bigHash;
   UInt32 historySize;
   UInt32 fixedHashSize;
   UInt32 hashSizeSum;
-  UInt32 numSons;
   SRes result;
   UInt32 crc[256];
+  size_t numRefs;
 } CMatchFinder;
 
 #define Inline_MatchFinder_GetPointerToCurrentPos(p) ((p)->buffer)
-#define Inline_MatchFinder_GetIndexByte(p, index) ((p)->buffer[(Int32)(index)])
 
 #define Inline_MatchFinder_GetNumAvailableBytes(p) ((p)->streamPos - (p)->pos)
 
+#define Inline_MatchFinder_IsFinishedOK(p) \
+    ((p)->streamEndWasReached \
+        && (p)->streamPos == (p)->pos \
+        && (!(p)->directInput || (p)->directInputRem == 0))
+      
 int MatchFinder_NeedMove(CMatchFinder *p);
 Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p);
 void MatchFinder_MoveBlock(CMatchFinder *p);
@@ -68,7 +73,7 @@ int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
     UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter,
     ISzAlloc *alloc);
 void MatchFinder_Free(CMatchFinder *p, ISzAlloc *alloc);
-void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, UInt32 numItems);
+void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems);
 void MatchFinder_ReduceOffsets(CMatchFinder *p, UInt32 subValue);
 
 UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byte *buffer, CLzRef *son,
@@ -82,7 +87,6 @@ Mf_GetPointerToCurrentPos_Func's result must be used only before any other function
 */
 
 typedef void (*Mf_Init_Func)(void *object);
-typedef Byte (*Mf_GetIndexByte_Func)(void *object, Int32 index);
 typedef UInt32 (*Mf_GetNumAvailableBytes_Func)(void *object);
 typedef const Byte * (*Mf_GetPointerToCurrentPos_Func)(void *object);
 typedef UInt32 (*Mf_GetMatches_Func)(void *object, UInt32 *distances);
@@ -91,7 +95,6 @@
 typedef struct _IMatchFinder
 {
   Mf_Init_Func Init;
-  Mf_GetIndexByte_Func GetIndexByte;
   Mf_GetNumAvailableBytes_Func GetNumAvailableBytes;
   Mf_GetPointerToCurrentPos_Func GetPointerToCurrentPos;
   Mf_GetMatches_Func GetMatches;
@@ -100,9 +103,12 @@ } IMatchFinder;
 
 void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder *vTable);
 
+void MatchFinder_Init_2(CMatchFinder *p, int readData);
 void MatchFinder_Init(CMatchFinder *p);
+
 UInt32 Bt3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
 UInt32 Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
+
 void Bt3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num);
 void Hc3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num);
 

@@ -1,5 +1,5 @@
 /* LzFindMt.c -- multithreaded Match finder for LZ algorithms
-2014-12-29 : Igor Pavlov : Public domain */
+2015-10-15 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -7,7 +7,7 @@ #include "LzHash.h"
 
 #include "LzFindMt.h"
 
-void MtSync_Construct(CMtSync *p)
+static void MtSync_Construct(CMtSync *p)
 {
   p->wasCreated = False;
   p->csWasInitialized = False;
@@ -20,7 +20,7 @@ Semaphore_Construct(&p->freeSemaphore);
   Semaphore_Construct(&p->filledSemaphore);
 }
 
-void MtSync_GetNextBlock(CMtSync *p)
+static void MtSync_GetNextBlock(CMtSync *p)
 {
   if (p->needStart)
   {
@@ -48,7 +48,7 @@ }
 
 /* MtSync_StopWriting must be called if Writing was started */
 
-void MtSync_StopWriting(CMtSync *p)
+static void MtSync_StopWriting(CMtSync *p)
 {
   UInt32 myNumBlocks = p->numProcessedBlocks;
   if (!Thread_WasCreated(&p->thread) || p->needStart)
@@ -71,7 +71,7 @@ }
   p->needStart = True;
 }
 
-void MtSync_Destruct(CMtSync *p)
+static void MtSync_Destruct(CMtSync *p)
 {
   if (Thread_WasCreated(&p->thread))
   {
@@ -134,20 +134,20 @@
 #define kMtMaxValForNormalize 0xFFFFFFFF
 
 #define DEF_GetHeads2(name, v, action) \
-static void GetHeads ## name(const Byte *p, UInt32 pos, \
-UInt32 *hash, UInt32 hashMask, UInt32 *heads, UInt32 numHeads, const UInt32 *crc) \
-{ action; for (; numHeads != 0; numHeads--) { \
-const UInt32 value = (v); p++; *heads++ = pos - hash[value]; hash[value] = pos++;  } }
+  static void GetHeads ## name(const Byte *p, UInt32 pos, \
+      UInt32 *hash, UInt32 hashMask, UInt32 *heads, UInt32 numHeads, const UInt32 *crc) \
+    { action; for (; numHeads != 0; numHeads--) { \
+      const UInt32 value = (v); p++; *heads++ = pos - hash[value]; hash[value] = pos++;  } }
 
 #define DEF_GetHeads(name, v) DEF_GetHeads2(name, v, ;)
 
-DEF_GetHeads2(2,  (p[0] | ((UInt32)p[1] << 8)), hashMask = hashMask; crc = crc; )
+DEF_GetHeads2(2,  (p[0] | ((UInt32)p[1] << 8)), UNUSED_VAR(hashMask); UNUSED_VAR(crc); )
 DEF_GetHeads(3,  (crc[p[0]] ^ p[1] ^ ((UInt32)p[2] << 8)) & hashMask)
 DEF_GetHeads(4,  (crc[p[0]] ^ p[1] ^ ((UInt32)p[2] << 8) ^ (crc[p[3]] << 5)) & hashMask)
 DEF_GetHeads(4b, (crc[p[0]] ^ p[1] ^ ((UInt32)p[2] << 8) ^ ((UInt32)p[3] << 16)) & hashMask)
 /* DEF_GetHeads(5,  (crc[p[0]] ^ p[1] ^ ((UInt32)p[2] << 8) ^ (crc[p[3]] << 5) ^ (crc[p[4]] << 3)) & hashMask) */
 
-void HashThreadFunc(CMatchFinderMt *mt)
+static void HashThreadFunc(CMatchFinderMt *mt)
 {
   CMtSync *p = &mt->hashSync;
   for (;;)
@@ -173,12 +173,12 @@ {
           CriticalSection_Enter(&mt->btSync.cs);
           CriticalSection_Enter(&mt->hashSync.cs);
           {
-            const Byte *beforePtr = MatchFinder_GetPointerToCurrentPos(mf);
-            const Byte *afterPtr;
+            const Byte *beforePtr = Inline_MatchFinder_GetPointerToCurrentPos(mf);
+            ptrdiff_t offset;
             MatchFinder_MoveBlock(mf);
-            afterPtr = MatchFinder_GetPointerToCurrentPos(mf);
-            mt->pointerToCurPos -= beforePtr - afterPtr;
-            mt->buffer -= beforePtr - afterPtr;
+            offset = beforePtr - Inline_MatchFinder_GetPointerToCurrentPos(mf);
+            mt->pointerToCurPos -= offset;
+            mt->buffer -= offset;
           }
           CriticalSection_Leave(&mt->btSync.cs);
           CriticalSection_Leave(&mt->hashSync.cs);
@@ -192,7 +192,7 @@ if (mf->pos > (kMtMaxValForNormalize - kMtHashBlockSize))
         {
           UInt32 subValue = (mf->pos - mf->historySize - 1);
           MatchFinder_ReduceOffsets(mf, subValue);
-          MatchFinder_Normalize3(subValue, mf->hash + mf->fixedHashSize, mf->hashMask + 1);
+          MatchFinder_Normalize3(subValue, mf->hash + mf->fixedHashSize, (size_t)mf->hashMask + 1);
         }
         {
           UInt32 *heads = mt->hashBuf + ((numProcessedBlocks++) & kMtHashNumBlocksMask) * kMtHashBlockSize;
@@ -217,7 +217,7 @@ }
   }
 }
 
-void MatchFinderMt_GetNextBlock_Hash(CMatchFinderMt *p)
+static void MatchFinderMt_GetNextBlock_Hash(CMatchFinderMt *p)
 {
   MtSync_GetNextBlock(&p->hashSync);
   p->hashBufPosLimit = p->hashBufPos = ((p->hashSync.numProcessedBlocks - 1) & kMtHashNumBlocksMask) * kMtHashBlockSize;
@@ -233,7 +233,7 @@ #ifdef MFMT_GM_INLINE
 
 #define NO_INLINE MY_FAST_CALL
 
-Int32 NO_INLINE GetMatchesSpecN(UInt32 lenLimit, UInt32 pos, const Byte *cur, CLzRef *son,
+static Int32 NO_INLINE GetMatchesSpecN(UInt32 lenLimit, UInt32 pos, const Byte *cur, CLzRef *son,
     UInt32 _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 _cutValue,
     UInt32 *_distances, UInt32 _maxLen, const UInt32 *hash, Int32 limit, UInt32 size, UInt32 *posRes)
 {
@@ -310,12 +310,14 @@ }
 
 #endif
 
-void BtGetMatches(CMatchFinderMt *p, UInt32 *distances)
+static void BtGetMatches(CMatchFinderMt *p, UInt32 *distances)
 {
   UInt32 numProcessed = 0;
   UInt32 curPos = 2;
   UInt32 limit = kMtBtBlockSize - (p->matchMaxLen * 2);
+  
   distances[1] = p->hashNumAvail;
+  
   while (curPos < limit)
   {
     if (p->hashBufPos == p->hashBufPosLimit)
@@ -324,9 +326,11 @@ MatchFinderMt_GetNextBlock_Hash(p);
       distances[1] = numProcessed + p->hashNumAvail;
       if (p->hashNumAvail >= p->numHashBytes)
         continue;
+      distances[0] = curPos + p->hashNumAvail;
+      distances += curPos;
       for (; p->hashNumAvail != 0; p->hashNumAvail--)
-        distances[curPos++] = 0;
-      break;
+        *distances++ = 0;
+      return;
     }
     {
       UInt32 size = p->hashBufPosLimit - p->hashBufPos;
@@ -343,13 +347,14 @@ size2 = p->cyclicBufferSize - cyclicBufferPos;
         if (size2 < size)
           size = size2;
       }
+      
       #ifndef MFMT_GM_INLINE
       while (curPos < limit && size-- != 0)
       {
         UInt32 *startDistances = distances + curPos;
         UInt32 num = (UInt32)(GetMatchesSpec1(lenLimit, pos - p->hashBuf[p->hashBufPos++],
-          pos, p->buffer, p->son, cyclicBufferPos, p->cyclicBufferSize, p->cutValue,
-          startDistances + 1, p->numHashBytes - 1) - startDistances);
+            pos, p->buffer, p->son, cyclicBufferPos, p->cyclicBufferSize, p->cutValue,
+            startDistances + 1, p->numHashBytes - 1) - startDistances);
         *startDistances = num - 1;
         curPos += num;
         cyclicBufferPos++;
@@ -360,7 +365,7 @@ #else
       {
         UInt32 posRes;
         curPos = limit - GetMatchesSpecN(lenLimit, pos, p->buffer, p->son, cyclicBufferPos, p->cyclicBufferSize, p->cutValue,
-          distances + curPos, p->numHashBytes - 1, p->hashBuf + p->hashBufPos, (Int32)(limit - curPos) , size, &posRes);
+            distances + curPos, p->numHashBytes - 1, p->hashBuf + p->hashBufPos, (Int32)(limit - curPos), size, &posRes);
         p->hashBufPos += posRes - pos;
         cyclicBufferPos += posRes - pos;
         p->buffer += posRes - pos;
@@ -376,10 +381,11 @@ cyclicBufferPos = 0;
       p->cyclicBufferPos = cyclicBufferPos;
     }
   }
+  
   distances[0] = curPos;
 }
 
-void BtFillBlock(CMatchFinderMt *p, UInt32 globalBlockIndex)
+static void BtFillBlock(CMatchFinderMt *p, UInt32 globalBlockIndex)
 {
   CMtSync *sync = &p->hashSync;
   if (!sync->needStart)
@@ -393,7 +399,7 @@
   if (p->pos > kMtMaxValForNormalize - kMtBtBlockSize)
   {
     UInt32 subValue = p->pos - p->cyclicBufferSize;
-    MatchFinder_Normalize3(subValue, p->son, p->cyclicBufferSize * 2);
+    MatchFinder_Normalize3(subValue, p->son, (size_t)p->cyclicBufferSize * 2);
     p->pos -= subValue;
   }
 
@@ -432,15 +438,15 @@ }
 
 void MatchFinderMt_Construct(CMatchFinderMt *p)
 {
-  p->hashBuf = 0;
+  p->hashBuf = NULL;
   MtSync_Construct(&p->hashSync);
   MtSync_Construct(&p->btSync);
 }
 
-void MatchFinderMt_FreeMem(CMatchFinderMt *p, ISzAlloc *alloc)
+static void MatchFinderMt_FreeMem(CMatchFinderMt *p, ISzAlloc *alloc)
 {
   alloc->Free(alloc, p->hashBuf);
-  p->hashBuf = 0;
+  p->hashBuf = NULL;
 }
 
 void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAlloc *alloc)
@@ -457,9 +463,11 @@ static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE HashThreadFunc2(void *p) { HashThreadFunc((CMatchFinderMt *)p);  return 0; }
 static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE BtThreadFunc2(void *p)
 {
   Byte allocaDummy[0x180];
-  allocaDummy[0] = 0;
-  allocaDummy[1] = allocaDummy[0];
-  BtThreadFunc((CMatchFinderMt *)p);
+  unsigned i = 0;
+  for (i = 0; i < 16; i++)
+    allocaDummy[i] = (Byte)0;
+  if (allocaDummy[0] == 0)
+    BtThreadFunc((CMatchFinderMt *)p);
   return 0;
 }
 
@@ -470,10 +478,10 @@ CMatchFinder *mf = p->MatchFinder;
   p->historySize = historySize;
   if (kMtBtBlockSize <= matchMaxLen * 4)
     return SZ_ERROR_PARAM;
-  if (p->hashBuf == 0)
+  if (!p->hashBuf)
   {
     p->hashBuf = (UInt32 *)alloc->Alloc(alloc, (kHashBufferSize + kBtBufferSize) * sizeof(UInt32));
-    if (p->hashBuf == 0)
+    if (!p->hashBuf)
       return SZ_ERROR_MEM;
     p->btBuf = p->hashBuf + kHashBufferSize;
   }
@@ -493,8 +501,11 @@ {
   CMatchFinder *mf = p->MatchFinder;
   p->btBufPos = p->btBufPosLimit = 0;
   p->hashBufPos = p->hashBufPosLimit = 0;
-  MatchFinder_Init(mf);
-  p->pointerToCurPos = MatchFinder_GetPointerToCurrentPos(mf);
+
+  /* Init without data reading. We don't want to read data in this thread */
+  MatchFinder_Init_2(mf, False);
+  
+  p->pointerToCurPos = Inline_MatchFinder_GetPointerToCurrentPos(mf);
   p->btNumAvailBytes = 0;
   p->lzPos = p->historySize + 1;
 
@@ -519,13 +530,13 @@ MtSync_StopWriting(&p->btSync);
   /* p->MatchFinder->ReleaseStream(); */
 }
 
-void MatchFinderMt_Normalize(CMatchFinderMt *p)
+static void MatchFinderMt_Normalize(CMatchFinderMt *p)
 {
   MatchFinder_Normalize3(p->lzPos - p->historySize - 1, p->hash, p->fixedHashSize);
   p->lzPos = p->historySize + 1;
 }
 
-void MatchFinderMt_GetNextBlock_Bt(CMatchFinderMt *p)
+static void MatchFinderMt_GetNextBlock_Bt(CMatchFinderMt *p)
 {
   UInt32 blockIndex;
   MtSync_GetNextBlock(&p->btSync);
@@ -537,34 +548,29 @@ if (p->lzPos >= kMtMaxValForNormalize - kMtBtBlockSize)
     MatchFinderMt_Normalize(p);
 }
 
-const Byte * MatchFinderMt_GetPointerToCurrentPos(CMatchFinderMt *p)
+static const Byte * MatchFinderMt_GetPointerToCurrentPos(CMatchFinderMt *p)
 {
   return p->pointerToCurPos;
 }
 
 #define GET_NEXT_BLOCK_IF_REQUIRED if (p->btBufPos == p->btBufPosLimit) MatchFinderMt_GetNextBlock_Bt(p);
 
-UInt32 MatchFinderMt_GetNumAvailableBytes(CMatchFinderMt *p)
+static UInt32 MatchFinderMt_GetNumAvailableBytes(CMatchFinderMt *p)
 {
   GET_NEXT_BLOCK_IF_REQUIRED;
   return p->btNumAvailBytes;
 }
 
-Byte MatchFinderMt_GetIndexByte(CMatchFinderMt *p, Int32 index)
-{
-  return p->pointerToCurPos[index];
-}
-
-UInt32 * MixMatches2(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
+static UInt32 * MixMatches2(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
 {
-  UInt32 hash2Value, curMatch2;
+  UInt32 h2, curMatch2;
   UInt32 *hash = p->hash;
   const Byte *cur = p->pointerToCurPos;
   UInt32 lzPos = p->lzPos;
   MT_HASH2_CALC
       
-  curMatch2 = hash[hash2Value];
-  hash[hash2Value] = lzPos;
+  curMatch2 = hash[h2];
+  hash[h2] = lzPos;
 
   if (curMatch2 >= matchMinPos)
     if (cur[(ptrdiff_t)curMatch2 - lzPos] == cur[0])
@@ -572,23 +578,23 @@ {
       *distances++ = 2;
       *distances++ = lzPos - curMatch2 - 1;
     }
+  
   return distances;
 }
 
-UInt32 * MixMatches3(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
+static UInt32 * MixMatches3(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
 {
-  UInt32 hash2Value, hash3Value, curMatch2, curMatch3;
+  UInt32 h2, h3, curMatch2, curMatch3;
   UInt32 *hash = p->hash;
   const Byte *cur = p->pointerToCurPos;
   UInt32 lzPos = p->lzPos;
   MT_HASH3_CALC
 
-  curMatch2 = hash[                hash2Value];
-  curMatch3 = hash[kFix3HashSize + hash3Value];
+  curMatch2 = hash[                h2];
+  curMatch3 = hash[kFix3HashSize + h3];
   
-  hash[                hash2Value] =
-  hash[kFix3HashSize + hash3Value] =
-    lzPos;
+  hash[                h2] = lzPos;
+  hash[kFix3HashSize + h3] = lzPos;
 
   if (curMatch2 >= matchMinPos && cur[(ptrdiff_t)curMatch2 - lzPos] == cur[0])
   {
@@ -601,43 +607,45 @@ }
     distances[0] = 2;
     distances += 2;
   }
+  
   if (curMatch3 >= matchMinPos && cur[(ptrdiff_t)curMatch3 - lzPos] == cur[0])
   {
     *distances++ = 3;
     *distances++ = lzPos - curMatch3 - 1;
   }
+  
   return distances;
 }
 
 /*
-UInt32 *MixMatches4(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
+static UInt32 *MixMatches4(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distances)
 {
-  UInt32 hash2Value, hash3Value, hash4Value, curMatch2, curMatch3, curMatch4;
+  UInt32 h2, h3, h4, curMatch2, curMatch3, curMatch4;
   UInt32 *hash = p->hash;
   const Byte *cur = p->pointerToCurPos;
   UInt32 lzPos = p->lzPos;
   MT_HASH4_CALC
       
-  curMatch2 = hash[                hash2Value];
-  curMatch3 = hash[kFix3HashSize + hash3Value];
-  curMatch4 = hash[kFix4HashSize + hash4Value];
+  curMatch2 = hash[                h2];
+  curMatch3 = hash[kFix3HashSize + h3];
+  curMatch4 = hash[kFix4HashSize + h4];
   
-  hash[                hash2Value] =
-  hash[kFix3HashSize + hash3Value] =
-  hash[kFix4HashSize + hash4Value] =
-    lzPos;
+  hash[                h2] = lzPos;
+  hash[kFix3HashSize + h3] = lzPos;
+  hash[kFix4HashSize + h4] = lzPos;
 
   if (curMatch2 >= matchMinPos && cur[(ptrdiff_t)curMatch2 - lzPos] == cur[0])
   {
     distances[1] = lzPos - curMatch2 - 1;
     if (cur[(ptrdiff_t)curMatch2 - lzPos + 2] == cur[2])
     {
-      distances[0] =  (cur[(ptrdiff_t)curMatch2 - lzPos + 3] == cur[3]) ? 4 : 3;
+      distances[0] = (cur[(ptrdiff_t)curMatch2 - lzPos + 3] == cur[3]) ? 4 : 3;
       return distances + 2;
     }
     distances[0] = 2;
     distances += 2;
   }
+  
   if (curMatch3 >= matchMinPos && cur[(ptrdiff_t)curMatch3 - lzPos] == cur[0])
   {
     distances[1] = lzPos - curMatch3 - 1;
@@ -659,13 +667,14 @@ {
       *distances++ = 4;
       *distances++ = lzPos - curMatch4 - 1;
     }
+  
   return distances;
 }
 */
 
 #define INCREASE_LZ_POS p->lzPos++; p->pointerToCurPos++;
 
-UInt32 MatchFinderMt2_GetMatches(CMatchFinderMt *p, UInt32 *distances)
+static UInt32 MatchFinderMt2_GetMatches(CMatchFinderMt *p, UInt32 *distances)
 {
   const UInt32 *btBuf = p->btBuf + p->btBufPos;
   UInt32 len = *btBuf++;
@@ -683,7 +692,7 @@ INCREASE_LZ_POS
   return len;
 }
 
-UInt32 MatchFinderMt_GetMatches(CMatchFinderMt *p, UInt32 *distances)
+static UInt32 MatchFinderMt_GetMatches(CMatchFinderMt *p, UInt32 *distances)
 {
   const UInt32 *btBuf = p->btBuf + p->btBufPos;
   UInt32 len = *btBuf++;
@@ -691,6 +700,7 @@ p->btBufPos += 1 + len;
 
   if (len == 0)
   {
+    /* change for bt5 ! */
     if (p->btNumAvailBytes-- >= 4)
       len = (UInt32)(p->MixMatchesFunc(p, p->lzPos - p->historySize, distances) - (distances));
   }
@@ -706,7 +716,7 @@ *distances2++ = *btBuf++;
       *distances2++ = *btBuf++;
     }
     while ((len -= 2) != 0);
-    len  = (UInt32)(distances2 - (distances));
+    len = (UInt32)(distances2 - (distances));
   }
   INCREASE_LZ_POS
   return len;
@@ -716,41 +726,41 @@ #define SKIP_HEADER2_MT  do { GET_NEXT_BLOCK_IF_REQUIRED
 #define SKIP_HEADER_MT(n) SKIP_HEADER2_MT if (p->btNumAvailBytes-- >= (n)) { const Byte *cur = p->pointerToCurPos; UInt32 *hash = p->hash;
 #define SKIP_FOOTER_MT } INCREASE_LZ_POS p->btBufPos += p->btBuf[p->btBufPos] + 1; } while (--num != 0);
 
-void MatchFinderMt0_Skip(CMatchFinderMt *p, UInt32 num)
+static void MatchFinderMt0_Skip(CMatchFinderMt *p, UInt32 num)
 {
   SKIP_HEADER2_MT { p->btNumAvailBytes--;
   SKIP_FOOTER_MT
 }
 
-void MatchFinderMt2_Skip(CMatchFinderMt *p, UInt32 num)
+static void MatchFinderMt2_Skip(CMatchFinderMt *p, UInt32 num)
 {
   SKIP_HEADER_MT(2)
-      UInt32 hash2Value;
+      UInt32 h2;
       MT_HASH2_CALC
-      hash[hash2Value] = p->lzPos;
+      hash[h2] = p->lzPos;
   SKIP_FOOTER_MT
 }
 
-void MatchFinderMt3_Skip(CMatchFinderMt *p, UInt32 num)
+static void MatchFinderMt3_Skip(CMatchFinderMt *p, UInt32 num)
 {
   SKIP_HEADER_MT(3)
-      UInt32 hash2Value, hash3Value;
+      UInt32 h2, h3;
       MT_HASH3_CALC
-      hash[kFix3HashSize + hash3Value] =
-      hash[                hash2Value] =
+      hash[kFix3HashSize + h3] =
+      hash[                h2] =
         p->lzPos;
   SKIP_FOOTER_MT
 }
 
 /*
-void MatchFinderMt4_Skip(CMatchFinderMt *p, UInt32 num)
+static void MatchFinderMt4_Skip(CMatchFinderMt *p, UInt32 num)
 {
   SKIP_HEADER_MT(4)
-      UInt32 hash2Value, hash3Value, hash4Value;
+      UInt32 h2, h3, h4;
       MT_HASH4_CALC
-      hash[kFix4HashSize + hash4Value] =
-      hash[kFix3HashSize + hash3Value] =
-      hash[                hash2Value] =
+      hash[kFix4HashSize + h4] =
+      hash[kFix3HashSize + h3] =
+      hash[                h2] =
         p->lzPos;
   SKIP_FOOTER_MT
 }
@@ -759,11 +769,11 @@
 void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder *vTable)
 {
   vTable->Init = (Mf_Init_Func)MatchFinderMt_Init;
-  vTable->GetIndexByte = (Mf_GetIndexByte_Func)MatchFinderMt_GetIndexByte;
   vTable->GetNumAvailableBytes = (Mf_GetNumAvailableBytes_Func)MatchFinderMt_GetNumAvailableBytes;
   vTable->GetPointerToCurrentPos = (Mf_GetPointerToCurrentPos_Func)MatchFinderMt_GetPointerToCurrentPos;
   vTable->GetMatches = (Mf_GetMatches_Func)MatchFinderMt_GetMatches;
-  switch(p->MatchFinder->numHashBytes)
+  
+  switch (p->MatchFinder->numHashBytes)
   {
     case 2:
       p->GetHeadsFunc = GetHeads2;
@@ -779,7 +789,6 @@ break;
     default:
     /* case 4: */
       p->GetHeadsFunc = p->MatchFinder->bigHash ? GetHeads4b : GetHeads4;
-      /* p->GetHeadsFunc = GetHeads4; */
       p->MixMatchesFunc = (Mf_Mix_Matches)MixMatches3;
       vTable->Skip = (Mf_Skip_Func)MatchFinderMt3_Skip;
       break;

@@ -1,5 +1,5 @@
 /* LzFindMt.h -- multithreaded Match finder for LZ algorithms
-2013-01-18 : Igor Pavlov : Public domain */
+2015-05-03 : Igor Pavlov : Public domain */
 
 #ifndef __LZ_FIND_MT_H
 #define __LZ_FIND_MT_H
@@ -75,7 +75,7 @@ CLzRef *son;
   UInt32 matchMaxLen;
   UInt32 numHashBytes;
   UInt32 pos;
-  Byte *buffer;
+  const Byte *buffer;
   UInt32 cyclicBufferPos;
   UInt32 cyclicBufferSize; /* it must be historySize + 1 */
   UInt32 cutValue;

@@ -1,5 +1,5 @@
 /* LzHash.h -- HASH functions for LZ algorithms
-2009-02-07 : Igor Pavlov : Public domain */
+2015-04-12 : Igor Pavlov : Public domain */
 
 #ifndef __LZ_HASH_H
 #define __LZ_HASH_H
@@ -12,43 +12,46 @@ #define kFix3HashSize (kHash2Size)
 #define kFix4HashSize (kHash2Size + kHash3Size)
 #define kFix5HashSize (kHash2Size + kHash3Size + kHash4Size)
 
-#define HASH2_CALC hashValue = cur[0] | ((UInt32)cur[1] << 8);
+#define HASH2_CALC hv = cur[0] | ((UInt32)cur[1] << 8);
 
 #define HASH3_CALC { \
   UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hashValue = (temp ^ ((UInt32)cur[2] << 8)) & p->hashMask; }
+  h2 = temp & (kHash2Size - 1); \
+  hv = (temp ^ ((UInt32)cur[2] << 8)) & p->hashMask; }
 
 #define HASH4_CALC { \
   UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hashValue = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)) & p->hashMask; }
+  h2 = temp & (kHash2Size - 1); \
+  temp ^= ((UInt32)cur[2] << 8); \
+  h3 = temp & (kHash3Size - 1); \
+  hv = (temp ^ (p->crc[cur[3]] << 5)) & p->hashMask; }
 
 #define HASH5_CALC { \
   UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hash4Value = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)); \
-  hashValue = (hash4Value ^ (p->crc[cur[4]] << 3)) & p->hashMask; \
-  hash4Value &= (kHash4Size - 1); }
+  h2 = temp & (kHash2Size - 1); \
+  temp ^= ((UInt32)cur[2] << 8); \
+  h3 = temp & (kHash3Size - 1); \
+  temp ^= (p->crc[cur[3]] << 5); \
+  h4 = temp & (kHash4Size - 1); \
+  hv = (temp ^ (p->crc[cur[4]] << 3)) & p->hashMask; }
 
-/* #define HASH_ZIP_CALC hashValue = ((cur[0] | ((UInt32)cur[1] << 8)) ^ p->crc[cur[2]]) & 0xFFFF; */
-#define HASH_ZIP_CALC hashValue = ((cur[2] | ((UInt32)cur[0] << 8)) ^ p->crc[cur[1]]) & 0xFFFF;
+/* #define HASH_ZIP_CALC hv = ((cur[0] | ((UInt32)cur[1] << 8)) ^ p->crc[cur[2]]) & 0xFFFF; */
+#define HASH_ZIP_CALC hv = ((cur[2] | ((UInt32)cur[0] << 8)) ^ p->crc[cur[1]]) & 0xFFFF;
 
 
 #define MT_HASH2_CALC \
-  hash2Value = (p->crc[cur[0]] ^ cur[1]) & (kHash2Size - 1);
+  h2 = (p->crc[cur[0]] ^ cur[1]) & (kHash2Size - 1);
 
 #define MT_HASH3_CALC { \
   UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); }
+  h2 = temp & (kHash2Size - 1); \
+  h3 = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); }
 
 #define MT_HASH4_CALC { \
   UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hash4Value = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)) & (kHash4Size - 1); }
+  h2 = temp & (kHash2Size - 1); \
+  temp ^= ((UInt32)cur[2] << 8); \
+  h3 = temp & (kHash3Size - 1); \
+  h4 = (temp ^ (p->crc[cur[3]] << 5)) & (kHash4Size - 1); }
 
 #endif

@@ -1,5 +1,5 @@
 /* Lzma2Dec.c -- LZMA2 Decoder
-2010-12-15 : Igor Pavlov : Public domain */
+2015-11-09 : Igor Pavlov : Public domain */
 
 /* #define SHOW_DEBUG_INFO */
 
@@ -99,12 +99,12 @@ }
 
 static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
 {
-  switch(p->state)
+  switch (p->state)
   {
     case LZMA2_STATE_CONTROL:
       p->control = b;
-      PRF(printf("\n %4X ", p->decoder.dicPos));
-      PRF(printf(" %2X", b));
+      PRF(printf("\n %4X ", (unsigned)p->decoder.dicPos));
+      PRF(printf(" %2X", (unsigned)b));
       if (p->control == 0)
         return LZMA2_STATE_FINISHED;
       if (LZMA2_IS_UNCOMPRESSED_STATE(p))
@@ -124,7 +124,7 @@
     case LZMA2_STATE_UNPACK1:
       p->unpackSize |= (UInt32)b;
       p->unpackSize++;
-      PRF(printf(" %8d", p->unpackSize));
+      PRF(printf(" %8u", (unsigned)p->unpackSize));
       return (LZMA2_IS_UNCOMPRESSED_STATE(p)) ? LZMA2_STATE_DATA : LZMA2_STATE_PACK0;
     
     case LZMA2_STATE_PACK0:
@@ -134,13 +134,13 @@
     case LZMA2_STATE_PACK1:
       p->packSize |= (UInt32)b;
       p->packSize++;
-      PRF(printf(" %8d", p->packSize));
+      PRF(printf(" %8u", (unsigned)p->packSize));
       return LZMA2_IS_THERE_PROP(LZMA2_GET_LZMA_MODE(p)) ? LZMA2_STATE_PROP:
         (p->needInitProp ? LZMA2_STATE_ERROR : LZMA2_STATE_DATA);
 
     case LZMA2_STATE_PROP:
     {
-      int lc, lp;
+      unsigned lc, lp;
       if (b >= (9 * 5 * 5))
         return LZMA2_STATE_ERROR;
       lc = b % 9;
@@ -179,13 +179,16 @@
   while (p->state != LZMA2_STATE_FINISHED)
   {
     SizeT dicPos = p->decoder.dicPos;
+    
     if (p->state == LZMA2_STATE_ERROR)
       return SZ_ERROR_DATA;
+    
     if (dicPos == dicLimit && finishMode == LZMA_FINISH_ANY)
     {
       *status = LZMA_STATUS_NOT_FINISHED;
       return SZ_OK;
     }
+
     if (p->state != LZMA2_STATE_DATA && p->state != LZMA2_STATE_DATA_CONT)
     {
       if (*srcLen == inSize)
@@ -195,8 +198,15 @@ return SZ_OK;
       }
       (*srcLen)++;
       p->state = Lzma2Dec_UpdateState(p, *src++);
+
+      if (dicPos == dicLimit && p->state != LZMA2_STATE_FINISHED)
+      {
+        p->state = LZMA2_STATE_ERROR;
+        return SZ_ERROR_DATA;
+      }
       continue;
     }
+    
     {
       SizeT destSizeCur = dicLimit - dicPos;
       SizeT srcSizeCur = inSize - *srcLen;
@@ -222,7 +232,10 @@ Bool initDic = (p->control == LZMA2_CONTROL_COPY_RESET_DIC);
           if (initDic)
             p->needInitProp = p->needInitState = True;
           else if (p->needInitDic)
+          {
+            p->state = LZMA2_STATE_ERROR;
             return SZ_ERROR_DATA;
+          }
           p->needInitDic = False;
           LzmaDec_InitDicAndState(&p->decoder, initDic, False);
         }
@@ -231,7 +244,10 @@ if (srcSizeCur > destSizeCur)
           srcSizeCur = destSizeCur;
 
         if (srcSizeCur == 0)
+        {
+          p->state = LZMA2_STATE_ERROR;
           return SZ_ERROR_DATA;
+        }
 
         LzmaDec_UpdateWithUncompressed(&p->decoder, src, srcSizeCur);
 
@@ -247,17 +263,21 @@ SRes res;
 
         if (p->state == LZMA2_STATE_DATA)
         {
-          int mode = LZMA2_GET_LZMA_MODE(p);
+          unsigned mode = LZMA2_GET_LZMA_MODE(p);
           Bool initDic = (mode == 3);
-          Bool initState = (mode > 0);
+          Bool initState = (mode != 0);
           if ((!initDic && p->needInitDic) || (!initState && p->needInitState))
+          {
+            p->state = LZMA2_STATE_ERROR;
             return SZ_ERROR_DATA;
+          }
           
           LzmaDec_InitDicAndState(&p->decoder, initDic, initState);
           p->needInitDic = False;
           p->needInitState = False;
           p->state = LZMA2_STATE_DATA_CONT;
         }
+  
         if (srcSizeCur > p->packSize)
           srcSizeCur = (SizeT)p->packSize;
           
@@ -276,16 +296,22 @@ return res;
 
         if (srcSizeCur == 0 && outSizeProcessed == 0)
         {
-          if (*status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK ||
-              p->unpackSize != 0 || p->packSize != 0)
+          if (*status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK
+              || p->unpackSize != 0
+              || p->packSize != 0)
+          {
+            p->state = LZMA2_STATE_ERROR;
             return SZ_ERROR_DATA;
+          }
           p->state = LZMA2_STATE_CONTROL;
         }
+        
         if (*status == LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK)
           *status = LZMA_STATUS_NOT_FINISHED;
       }
     }
   }
+  
   *status = LZMA_STATUS_FINISHED_WITH_MARK;
   return SZ_OK;
 }

@@ -1,5 +1,5 @@
 /* Lzma2Dec.h -- LZMA2 Decoder
-2013-01-18 : Igor Pavlov : Public domain */
+2015-05-13 : Igor Pavlov : Public domain */
 
 #ifndef __LZMA2_DEC_H
 #define __LZMA2_DEC_H
@@ -15,7 +15,7 @@ {
   CLzmaDec decoder;
   UInt32 packSize;
   UInt32 unpackSize;
-  int state;
+  unsigned state;
   Byte control;
   Bool needInitDic;
   Bool needInitState;

@@ -1,5 +1,5 @@
 /* Lzma2Enc.c -- LZMA2 Encoder
-2012-06-19 : Igor Pavlov : Public domain */
+2015-10-04 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -109,6 +109,7 @@ if (useCopyBlock)
   {
     size_t destPos = 0;
     PRF(printf("################# COPY           "));
+
     while (unpackSize > 0)
     {
       UInt32 u = (unpackSize < LZMA2_COPY_CHUNK_SIZE) ? unpackSize : LZMA2_COPY_CHUNK_SIZE;
@@ -121,6 +122,7 @@ memcpy(outBuf + destPos, LzmaEnc_GetCurBuf(p->enc) - unpackSize, u);
       unpackSize -= u;
       destPos += u;
       p->srcPos += u;
+      
       if (outStream)
       {
         *packSizeRes += destPos;
@@ -132,9 +134,11 @@ else
         *packSizeRes = destPos;
       /* needInitState = True; */
     }
+    
     LzmaEnc_RestoreState(p->enc);
     return SZ_OK;
   }
+
   {
     size_t destPos = 0;
     UInt32 u = unpackSize - 1;
@@ -160,10 +164,12 @@
     if (outStream)
       if (outStream->Write(outStream, outBuf, destPos) != destPos)
         return SZ_ERROR_WRITE;
+    
     *packSizeRes = destPos;
     return SZ_OK;
   }
 }
+
 
 /* ---------- Lzma2 Props ---------- */
 
@@ -221,6 +227,8 @@ p->lzmaProps.numThreads = t1;
 
   LzmaEncProps_Normalize(&p->lzmaProps);
 
+  t1 = p->lzmaProps.numThreads;
+
   if (p->blockSize == 0)
   {
     UInt32 dictSize = p->lzmaProps.dictSize;
@@ -232,27 +240,33 @@ if (blockSize > kMaxSize) blockSize = kMaxSize;
     if (blockSize < dictSize) blockSize = dictSize;
     p->blockSize = (size_t)blockSize;
   }
-  if (t2 > 1)
+  
+  if (t2 > 1 && p->lzmaProps.reduceSize != (UInt64)(Int64)-1)
   {
     UInt64 temp = p->lzmaProps.reduceSize + p->blockSize - 1;
     if (temp > p->lzmaProps.reduceSize)
     {
       UInt64 numBlocks = temp / p->blockSize;
-      if (numBlocks < t2)
+      if (numBlocks < (unsigned)t2)
       {
-        t2 = (UInt32)numBlocks;
+        t2 = (unsigned)numBlocks;
+        if (t2 == 0)
+          t2 = 1;
         t3 = t1 * t2;
       }
     }
   }
+  
   p->numBlockThreads = t2;
   p->numTotalThreads = t3;
 }
+
 
 static SRes Progress(ICompressProgress *p, UInt64 inSize, UInt64 outSize)
 {
   return (p && p->Progress(p, inSize, outSize) != SZ_OK) ? SZ_ERROR_PROGRESS : SZ_OK;
 }
+
 
 /* ---------- Lzma2 ---------- */
 
@@ -283,15 +297,17 @@ {
   UInt64 packTotal = 0;
   SRes res = SZ_OK;
 
-  if (mainEncoder->outBuf == 0)
+  if (!mainEncoder->outBuf)
   {
     mainEncoder->outBuf = (Byte *)IAlloc_Alloc(mainEncoder->alloc, LZMA2_CHUNK_SIZE_COMPRESSED_MAX);
-    if (mainEncoder->outBuf == 0)
+    if (!mainEncoder->outBuf)
       return SZ_ERROR_MEM;
   }
+  
   RINOK(Lzma2EncInt_Init(p, &mainEncoder->props));
   RINOK(LzmaEnc_PrepareForLzma2(p->enc, inStream, LZMA2_KEEP_WINDOW_SIZE,
       mainEncoder->alloc, mainEncoder->allocBig));
+  
   for (;;)
   {
     size_t packSize = LZMA2_CHUNK_SIZE_COMPRESSED_MAX;
@@ -305,15 +321,19 @@ break;
     if (packSize == 0)
       break;
   }
+  
   LzmaEnc_Finish(p->enc);
+
   if (res == SZ_OK)
   {
     Byte b = 0;
     if (outStream->Write(outStream, &b, 1) != 1)
       return SZ_ERROR_WRITE;
   }
+  
   return res;
 }
+
 
 #ifndef _7ZIP_ST
 
@@ -362,10 +382,12 @@ res = SZ_ERROR_PROGRESS;
           break;
         }
       }
+      
       LzmaEnc_Finish(p->enc);
       if (res != SZ_OK)
         return res;
     }
+    
     if (finished)
     {
       if (*destSize == destLim)
@@ -377,13 +399,14 @@ return res;
 }
 
 #endif
+
 
 /* ---------- Lzma2Enc ---------- */
 
 CLzma2EncHandle Lzma2Enc_Create(ISzAlloc *alloc, ISzAlloc *allocBig)
 {
   CLzma2Enc *p = (CLzma2Enc *)alloc->Alloc(alloc, sizeof(CLzma2Enc));
-  if (p == 0)
+  if (!p)
     return NULL;
   Lzma2EncProps_Init(&p->props);
   Lzma2EncProps_Normalize(&p->props);
@@ -395,6 +418,7 @@ unsigned i;
     for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
       p->coders[i].enc = 0;
   }
+  
   #ifndef _7ZIP_ST
   MtCoder_Construct(&p->mtCoder);
   #endif
@@ -455,22 +479,17 @@ int i;
 
   for (i = 0; i < p->props.numBlockThreads; i++)
   {
-    CLzma2EncInt *t = &p->coders[i];
-    if (t->enc == NULL)
+    CLzma2EncInt *t = &p->coders[(unsigned)i];
+    if (!t->enc)
     {
       t->enc = LzmaEnc_Create(p->alloc);
-      if (t->enc == NULL)
+      if (!t->enc)
         return SZ_ERROR_MEM;
     }
   }
 
   #ifndef _7ZIP_ST
-  if (p->props.numBlockThreads <= 1)
-  #endif
-    return Lzma2Enc_EncodeMt1(&p->coders[0], p, outStream, inStream, progress);
-
-  #ifndef _7ZIP_ST
-
+  if (p->props.numBlockThreads > 1)
   {
     CMtCallbackImp mtCallback;
 
@@ -485,9 +504,17 @@ p->mtCoder.mtCallback = &mtCallback.funcTable;
 
     p->mtCoder.blockSize = p->props.blockSize;
     p->mtCoder.destBlockSize = p->props.blockSize + (p->props.blockSize >> 10) + 16;
+    if (p->mtCoder.destBlockSize < p->props.blockSize)
+    {
+      p->mtCoder.destBlockSize = (size_t)0 - 1;
+      if (p->mtCoder.destBlockSize < p->props.blockSize)
+        return SZ_ERROR_FAIL;
+    }
     p->mtCoder.numThreads = p->props.numBlockThreads;
     
     return MtCoder_Code(&p->mtCoder);
   }
   #endif
+
+  return Lzma2Enc_EncodeMt1(&p->coders[0], p, outStream, inStream, progress);
 }

@@ -1,5 +1,7 @@
 /* Lzma86Dec.c -- LZMA + x86 (BCJ) Filter Decoder
-2009-08-14 : Igor Pavlov : Public domain */
+2015-11-08 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Lzma86.h"
 
@@ -7,8 +9,8 @@ #include "Alloc.h"
 #include "Bra.h"
 #include "LzmaDec.h"
 
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
+static void *SzAlloc(void *p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); }
+static void SzFree(void *p, void *address) { UNUSED_VAR(p); MyFree(address); }
 
 SRes Lzma86_GetUnpackSize(const Byte *src, SizeT srcLen, UInt64 *unpackSize)
 {

@@ -1,5 +1,7 @@
 /* Lzma86Enc.c -- LZMA + x86 (BCJ) Filter Encoder
-2009-08-14 : Igor Pavlov : Public domain */
+2015-11-08 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
 
@@ -11,8 +13,8 @@ #include "LzmaEnc.h"
 
 #define SZE_OUT_OVERFLOW SZE_DATA_ERROR
 
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
+static void *SzAlloc(void *p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); }
+static void SzFree(void *p, void *address) { UNUSED_VAR(p); MyFree(address); }
 
 int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
     int level, UInt32 dictSize, int filterMode)

@@ -1,5 +1,5 @@
 /* LzmaDec.c -- LZMA Decoder
-2015-01-01 : Igor Pavlov : Public domain */
+2015-06-23 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -114,13 +114,13 @@ #define RepLenCoder (LenCoder + kNumLenProbs)
 #define Literal (RepLenCoder + kNumLenProbs)
 
 #define LZMA_BASE_SIZE 1846
-#define LZMA_LIT_SIZE 768
-
-#define LzmaProps_GetNumProbs(p) ((UInt32)LZMA_BASE_SIZE + (LZMA_LIT_SIZE << ((p)->lc + (p)->lp)))
+#define LZMA_LIT_SIZE 0x300
 
 #if Literal != LZMA_BASE_SIZE
 StopCompilingDueBUG
 #endif
+
+#define LzmaProps_GetNumProbs(p) (Literal + ((UInt32)LZMA_LIT_SIZE << ((p)->lc + (p)->lp)))
 
 #define LZMA_DIC_MIN (1 << 12)
 
@@ -133,8 +133,8 @@ SZ_ERROR_DATA - Error
   p->remainLen:
     < kMatchSpecLenStart : normal remain
     = kMatchSpecLenStart : finished
-    = kMatchSpecLenStart + 1 : Flush marker
-    = kMatchSpecLenStart + 2 : State Init Marker
+    = kMatchSpecLenStart + 1 : Flush marker (unused now)
+    = kMatchSpecLenStart + 2 : State Init Marker (unused now)
 */
 
 static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte *bufLimit)
@@ -172,9 +172,10 @@ {
       unsigned symbol;
       UPDATE_0(prob);
       prob = probs + Literal;
-      if (checkDicSize != 0 || processedPos != 0)
-        prob += (LZMA_LIT_SIZE * (((processedPos & lpMask) << lc) +
-        (dic[(dicPos == 0 ? dicBufSize : dicPos) - 1] >> (8 - lc))));
+      if (processedPos != 0 || checkDicSize != 0)
+        prob += ((UInt32)LZMA_LIT_SIZE * (((processedPos & lpMask) << lc) +
+            (dic[(dicPos == 0 ? dicBufSize : dicPos) - 1] >> (8 - lc))));
+      processedPos++;
 
       if (state < kNumLitStates)
       {
@@ -195,7 +196,7 @@ #endif
       }
       else
       {
-        unsigned matchByte = dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
+        unsigned matchByte = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)];
         unsigned offs = 0x100;
         state -= (state < 10) ? 3 : 6;
         symbol = 1;
@@ -222,11 +223,11 @@ MATCHED_LITER_DEC
         }
         #endif
       }
+
       dic[dicPos++] = (Byte)symbol;
-      processedPos++;
       continue;
     }
-    else
+    
     {
       UPDATE_1(prob);
       prob = probs + IsRep + state;
@@ -249,7 +250,7 @@ prob = probs + IsRep0Long + (state << kNumPosBitsMax) + posState;
           IF_BIT_0(prob)
           {
             UPDATE_0(prob);
-            dic[dicPos] = dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
+            dic[dicPos] = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)];
             dicPos++;
             processedPos++;
             state = state < kNumLitStates ? 9 : 11;
@@ -290,6 +291,8 @@ }
         state = state < kNumLitStates ? 8 : 11;
         prob = probs + RepLenCoder;
       }
+      
+      #ifdef _LZMA_SIZE_OPT
       {
         unsigned limit, offset;
         CLzmaProb *probLen = prob + LenChoice;
@@ -322,6 +325,42 @@ }
         TREE_DECODE(probLen, limit, len);
         len += offset;
       }
+      #else
+      {
+        CLzmaProb *probLen = prob + LenChoice;
+        IF_BIT_0(probLen)
+        {
+          UPDATE_0(probLen);
+          probLen = prob + LenLow + (posState << kLenNumLowBits);
+          len = 1;
+          TREE_GET_BIT(probLen, len);
+          TREE_GET_BIT(probLen, len);
+          TREE_GET_BIT(probLen, len);
+          len -= 8;
+        }
+        else
+        {
+          UPDATE_1(probLen);
+          probLen = prob + LenChoice2;
+          IF_BIT_0(probLen)
+          {
+            UPDATE_0(probLen);
+            probLen = prob + LenMid + (posState << kLenNumMidBits);
+            len = 1;
+            TREE_GET_BIT(probLen, len);
+            TREE_GET_BIT(probLen, len);
+            TREE_GET_BIT(probLen, len);
+          }
+          else
+          {
+            UPDATE_1(probLen);
+            probLen = prob + LenHigh;
+            TREE_DECODE(probLen, (1 << kLenNumHighBits), len);
+            len += kLenNumLowSymbols + kLenNumMidSymbols;
+          }
+        }
+      }
+      #endif
 
       if (state >= kNumStates)
       {
@@ -332,7 +371,7 @@ TREE_6_DECODE(prob, distance);
         if (distance >= kStartPosModelIndex)
         {
           unsigned posSlot = (unsigned)distance;
-          int numDirectBits = (int)(((distance >> 1) - 1));
+          unsigned numDirectBits = (unsigned)(((distance >> 1) - 1));
           distance = (2 | (distance & 1));
           if (posSlot < kEndPosModelIndex)
           {
@@ -391,6 +430,7 @@ break;
             }
           }
         }
+        
         rep3 = rep2;
         rep2 = rep1;
         rep1 = rep0;
@@ -398,26 +438,39 @@ rep0 = distance + 1;
         if (checkDicSize == 0)
         {
           if (distance >= processedPos)
+          {
+            p->dicPos = dicPos;
             return SZ_ERROR_DATA;
+          }
         }
         else if (distance >= checkDicSize)
+        {
+          p->dicPos = dicPos;
           return SZ_ERROR_DATA;
+        }
         state = (state < kNumStates + kNumLitStates) ? kNumLitStates : kNumLitStates + 3;
       }
 
       len += kMatchMinLen;
 
-      if (limit == dicPos)
-        return SZ_ERROR_DATA;
       {
-        SizeT rem = limit - dicPos;
-        unsigned curLen = ((rem < len) ? (unsigned)rem : len);
-        SizeT pos = (dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0);
+        SizeT rem;
+        unsigned curLen;
+        SizeT pos;
+        
+        if ((rem = limit - dicPos) == 0)
+        {
+          p->dicPos = dicPos;
+          return SZ_ERROR_DATA;
+        }
+        
+        curLen = ((rem < len) ? (unsigned)rem : len);
+        pos = dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0);
 
         processedPos += curLen;
 
         len -= curLen;
-        if (pos + curLen <= dicBufSize)
+        if (curLen <= dicBufSize - pos)
         {
           Byte *dest = dic + dicPos;
           ptrdiff_t src = (ptrdiff_t)pos - (ptrdiff_t)dicPos;
@@ -441,7 +494,9 @@ }
     }
   }
   while (dicPos < limit && buf < bufLimit);
+
   NORMALIZE;
+  
   p->buf = buf;
   p->range = range;
   p->code = code;
@@ -465,9 +520,10 @@ Byte *dic = p->dic;
     SizeT dicPos = p->dicPos;
     SizeT dicBufSize = p->dicBufSize;
     unsigned len = p->remainLen;
-    UInt32 rep0 = p->reps[0];
-    if (limit - dicPos < len)
-      len = (unsigned)(limit - dicPos);
+    SizeT rep0 = p->reps[0]; /* we use SizeT to avoid the BUG of VC14 for AMD64 */
+    SizeT rem = limit - dicPos;
+    if (rem < len)
+      len = (unsigned)(rem);
 
     if (p->checkDicSize == 0 && p->prop.dicSize - p->processedPos <= len)
       p->checkDicSize = p->prop.dicSize;
@@ -477,7 +533,7 @@ p->remainLen -= len;
     while (len != 0)
     {
       len--;
-      dic[dicPos] = dic[(dicPos - rep0) + ((dicPos < rep0) ? dicBufSize : 0)];
+      dic[dicPos] = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)];
       dicPos++;
     }
     p->dicPos = dicPos;
@@ -495,17 +551,19 @@ UInt32 rem = p->prop.dicSize - p->processedPos;
       if (limit - p->dicPos > rem)
         limit2 = p->dicPos + rem;
     }
+    
     RINOK(LzmaDec_DecodeReal(p, limit2, bufLimit));
-    if (p->processedPos >= p->prop.dicSize)
+    
+    if (p->checkDicSize == 0 && p->processedPos >= p->prop.dicSize)
       p->checkDicSize = p->prop.dicSize;
+    
     LzmaDec_WriteRem(p, limit);
   }
   while (p->dicPos < limit && p->buf < bufLimit && p->remainLen < kMatchSpecLenStart);
 
   if (p->remainLen > kMatchSpecLenStart)
-  {
     p->remainLen = kMatchSpecLenStart;
-  }
+
   return 0;
 }
 
@@ -522,12 +580,12 @@ {
   UInt32 range = p->range;
   UInt32 code = p->code;
   const Byte *bufLimit = buf + inSize;
-  CLzmaProb *probs = p->probs;
+  const CLzmaProb *probs = p->probs;
   unsigned state = p->state;
   ELzmaDummy res;
 
   {
-    CLzmaProb *prob;
+    const CLzmaProb *prob;
     UInt32 bound;
     unsigned ttt;
     unsigned posState = (p->processedPos) & ((1 << p->prop.pb) - 1);
@@ -541,9 +599,9 @@ /* if (bufLimit - buf >= 7) return DUMMY_LIT; */
 
       prob = probs + Literal;
       if (p->checkDicSize != 0 || p->processedPos != 0)
-        prob += (LZMA_LIT_SIZE *
-          ((((p->processedPos) & ((1 << (p->prop.lp)) - 1)) << p->prop.lc) +
-          (p->dic[(p->dicPos == 0 ? p->dicBufSize : p->dicPos) - 1] >> (8 - p->prop.lc))));
+        prob += ((UInt32)LZMA_LIT_SIZE *
+            ((((p->processedPos) & ((1 << (p->prop.lp)) - 1)) << p->prop.lc) +
+            (p->dic[(p->dicPos == 0 ? p->dicBufSize : p->dicPos) - 1] >> (8 - p->prop.lc))));
 
       if (state < kNumLitStates)
       {
@@ -553,13 +611,13 @@ }
       else
       {
         unsigned matchByte = p->dic[p->dicPos - p->reps[0] +
-            ((p->dicPos < p->reps[0]) ? p->dicBufSize : 0)];
+            (p->dicPos < p->reps[0] ? p->dicBufSize : 0)];
         unsigned offs = 0x100;
         unsigned symbol = 1;
         do
         {
           unsigned bit;
-          CLzmaProb *probLit;
+          const CLzmaProb *probLit;
           matchByte <<= 1;
           bit = (matchByte & offs);
           probLit = prob + offs + bit + symbol;
@@ -629,7 +687,7 @@ prob = probs + RepLenCoder;
       }
       {
         unsigned limit, offset;
-        CLzmaProb *probLen = prob + LenChoice;
+        const CLzmaProb *probLen = prob + LenChoice;
         IF_BIT_0_CHECK(probLen)
         {
           UPDATE_0_CHECK;
@@ -669,7 +727,7 @@ kNumPosSlotBits);
         TREE_DECODE_CHECK(prob, 1 << kNumPosSlotBits, posSlot);
         if (posSlot >= kStartPosModelIndex)
         {
-          int numDirectBits = ((posSlot >> 1) - 1);
+          unsigned numDirectBits = ((posSlot >> 1) - 1);
 
           /* if (bufLimit - buf >= 8) return DUMMY_MATCH; */
 
@@ -708,13 +766,6 @@ return res;
 }
 
 
-static void LzmaDec_InitRc(CLzmaDec *p, const Byte *data)
-{
-  p->code = ((UInt32)data[1] << 24) | ((UInt32)data[2] << 16) | ((UInt32)data[3] << 8) | ((UInt32)data[4]);
-  p->range = 0xFFFFFFFF;
-  p->needFlush = 0;
-}
-
 void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState)
 {
   p->needFlush = 1;
@@ -739,8 +790,8 @@ }
 
 static void LzmaDec_InitStateReal(CLzmaDec *p)
 {
-  UInt32 numProbs = Literal + ((UInt32)LZMA_LIT_SIZE << (p->prop.lc + p->prop.lp));
-  UInt32 i;
+  SizeT numProbs = LzmaProps_GetNumProbs(&p->prop);
+  SizeT i;
   CLzmaProb *probs = p->probs;
   for (i = 0; i < numProbs; i++)
     probs[i] = kBitModelTotal >> 1;
@@ -762,7 +813,7 @@ while (p->remainLen != kMatchSpecLenStart)
   {
       int checkEndMarkNow;
 
-      if (p->needFlush != 0)
+      if (p->needFlush)
       {
         for (; inSize > 0 && p->tempBufSize < RC_INIT_SIZE; (*srcLen)++, inSize--)
           p->tempBuf[p->tempBufSize++] = *src++;
@@ -773,8 +824,13 @@ return SZ_OK;
         }
         if (p->tempBuf[0] != 0)
           return SZ_ERROR_DATA;
-
-        LzmaDec_InitRc(p, p->tempBuf);
+        p->code =
+              ((UInt32)p->tempBuf[1] << 24)
+            | ((UInt32)p->tempBuf[2] << 16)
+            | ((UInt32)p->tempBuf[3] << 8)
+            | ((UInt32)p->tempBuf[4]);
+        p->range = 0xFFFFFFFF;
+        p->needFlush = 0;
         p->tempBufSize = 0;
       }
 
@@ -858,7 +914,16 @@ }
         p->buf = p->tempBuf;
         if (LzmaDec_DecodeReal2(p, dicLimit, p->buf) != 0)
           return SZ_ERROR_DATA;
-        lookAhead -= (rem - (unsigned)(p->buf - p->tempBuf));
+        
+        {
+          unsigned kkk = (unsigned)(p->buf - p->tempBuf);
+          if (rem < kkk)
+            return SZ_ERROR_FAIL; /* some internal error */
+          rem -= kkk;
+          if (lookAhead < rem)
+            return SZ_ERROR_FAIL; /* some internal error */
+          lookAhead -= rem;
+        }
         (*srcLen) += lookAhead;
         src += lookAhead;
         inSize -= lookAhead;
@@ -913,13 +978,13 @@
 void LzmaDec_FreeProbs(CLzmaDec *p, ISzAlloc *alloc)
 {
   alloc->Free(alloc, p->probs);
-  p->probs = 0;
+  p->probs = NULL;
 }
 
 static void LzmaDec_FreeDict(CLzmaDec *p, ISzAlloc *alloc)
 {
   alloc->Free(alloc, p->dic);
-  p->dic = 0;
+  p->dic = NULL;
 }
 
 void LzmaDec_Free(CLzmaDec *p, ISzAlloc *alloc)
@@ -957,12 +1022,12 @@
 static SRes LzmaDec_AllocateProbs2(CLzmaDec *p, const CLzmaProps *propNew, ISzAlloc *alloc)
 {
   UInt32 numProbs = LzmaProps_GetNumProbs(propNew);
-  if (p->probs == 0 || numProbs != p->numProbs)
+  if (!p->probs || numProbs != p->numProbs)
   {
     LzmaDec_FreeProbs(p, alloc);
     p->probs = (CLzmaProb *)alloc->Alloc(alloc, numProbs * sizeof(CLzmaProb));
     p->numProbs = numProbs;
-    if (p->probs == 0)
+    if (!p->probs)
       return SZ_ERROR_MEM;
   }
   return SZ_OK;
@@ -983,12 +1048,22 @@ CLzmaProps propNew;
   SizeT dicBufSize;
   RINOK(LzmaProps_Decode(&propNew, props, propsSize));
   RINOK(LzmaDec_AllocateProbs2(p, &propNew, alloc));
-  dicBufSize = propNew.dicSize;
-  if (p->dic == 0 || dicBufSize != p->dicBufSize)
+
+  {
+    UInt32 dictSize = propNew.dicSize;
+    SizeT mask = ((UInt32)1 << 12) - 1;
+         if (dictSize >= ((UInt32)1 << 30)) mask = ((UInt32)1 << 22) - 1;
+    else if (dictSize >= ((UInt32)1 << 22)) mask = ((UInt32)1 << 20) - 1;;
+    dicBufSize = ((SizeT)dictSize + mask) & ~mask;
+    if (dicBufSize < dictSize)
+      dicBufSize = dictSize;
+  }
+
+  if (!p->dic || dicBufSize != p->dicBufSize)
   {
     LzmaDec_FreeDict(p, alloc);
     p->dic = (Byte *)alloc->Alloc(alloc, dicBufSize);
-    if (p->dic == 0)
+    if (!p->dic)
     {
       LzmaDec_FreeProbs(p, alloc);
       return SZ_ERROR_MEM;

@@ -1,5 +1,5 @@
 /* LzmaEnc.c -- LZMA Encoder
-2014-12-29 : Igor Pavlov : Public domain */
+2015-11-08 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -22,6 +22,9 @@
 #ifdef SHOW_STAT
 static unsigned g_STAT_OFFSET = 0;
 #endif
+
+#define kMaxHistorySize ((UInt32)3 << 29)
+/* #define kMaxHistorySize ((UInt32)7 << 29) */
 
 #define kBlockSizeMax ((1 << LZMA_NUM_BLOCK_SIZE_BITS) - 1)
 
@@ -58,6 +61,7 @@ {
   int level = p->level;
   if (level < 0) level = 5;
   p->level = level;
+  
   if (p->dictSize == 0) p->dictSize = (level <= 5 ? (1 << (level * 2 + 14)) : (level == 6 ? (1 << 25) : (1 << 26)));
   if (p->dictSize > p->reduceSize)
   {
@@ -68,14 +72,17 @@ if ((UInt32)p->reduceSize <= ((UInt32)2 << i)) { p->dictSize = ((UInt32)2 << i); break; }
       if ((UInt32)p->reduceSize <= ((UInt32)3 << i)) { p->dictSize = ((UInt32)3 << i); break; }
     }
   }
+
   if (p->lc < 0) p->lc = 3;
   if (p->lp < 0) p->lp = 0;
   if (p->pb < 0) p->pb = 2;
+
   if (p->algo < 0) p->algo = (level < 5 ? 0 : 1);
   if (p->fb < 0) p->fb = (level < 7 ? 32 : 64);
   if (p->btMode < 0) p->btMode = (p->algo == 0 ? 0 : 1);
   if (p->numHashBytes < 0) p->numHashBytes = 4;
-  if (p->mc == 0)  p->mc = (16 + (p->fb >> 1)) >> (p->btMode ? 0 : 1);
+  if (p->mc == 0) p->mc = (16 + (p->fb >> 1)) >> (p->btMode ? 0 : 1);
+  
   if (p->numThreads < 0)
     p->numThreads =
       #ifndef _7ZIP_ST
@@ -92,17 +99,18 @@ LzmaEncProps_Normalize(&props);
   return props.dictSize;
 }
 
+#if (_MSC_VER >= 1400)
+/* BSR code is fast for some new CPUs */
 /* #define LZMA_LOG_BSR */
-/* Define it for Intel's CPU */
-
+#endif
 
 #ifdef LZMA_LOG_BSR
 
-#define kDicLogSizeMaxCompress 30
+#define kDicLogSizeMaxCompress 32
 
 #define BSR2_RET(pos, res) { unsigned long i; _BitScanReverse(&i, (pos)); res = (i + i) + ((pos >> (i - 1)) & 1); }
 
-UInt32 GetPosSlot1(UInt32 pos)
+static UInt32 GetPosSlot1(UInt32 pos)
 {
   UInt32 res;
   BSR2_RET(pos, res);
@@ -113,27 +121,44 @@ #define GetPosSlot(pos, res) { if (pos < 2) res = pos; else BSR2_RET(pos, res); }
 
 #else
 
-#define kNumLogBits (9 + (int)sizeof(size_t) / 2)
+#define kNumLogBits (9 + sizeof(size_t) / 2)
+/* #define kNumLogBits (11 + sizeof(size_t) / 8 * 3) */
+
 #define kDicLogSizeMaxCompress ((kNumLogBits - 1) * 2 + 7)
 
-void LzmaEnc_FastPosInit(Byte *g_FastPos)
+static void LzmaEnc_FastPosInit(Byte *g_FastPos)
 {
-  int c = 2, slotFast;
+  unsigned slot;
   g_FastPos[0] = 0;
   g_FastPos[1] = 1;
+  g_FastPos += 2;
   
-  for (slotFast = 2; slotFast < kNumLogBits * 2; slotFast++)
+  for (slot = 2; slot < kNumLogBits * 2; slot++)
   {
-    UInt32 k = (1 << ((slotFast >> 1) - 1));
-    UInt32 j;
-    for (j = 0; j < k; j++, c++)
-      g_FastPos[c] = (Byte)slotFast;
+    size_t k = ((size_t)1 << ((slot >> 1) - 1));
+    size_t j;
+    for (j = 0; j < k; j++)
+      g_FastPos[j] = (Byte)slot;
+    g_FastPos += k;
   }
 }
 
+/* we can use ((limit - pos) >> 31) only if (pos < ((UInt32)1 << 31)) */
+/*
 #define BSR2_RET(pos, res) { UInt32 i = 6 + ((kNumLogBits - 1) & \
   (0 - (((((UInt32)1 << (kNumLogBits + 6)) - 1) - pos) >> 31))); \
   res = p->g_FastPos[pos >> i] + (i * 2); }
+*/
+
+/*
+#define BSR2_RET(pos, res) { UInt32 i = 6 + ((kNumLogBits - 1) & \
+  (0 - (((((UInt32)1 << (kNumLogBits)) - 1) - (pos >> 6)) >> 31))); \
+  res = p->g_FastPos[pos >> i] + (i * 2); }
+*/
+
+#define BSR2_RET(pos, res) { UInt32 i = (pos < (1 << (kNumLogBits + 6))) ? 6 : 6 + kNumLogBits - 1; \
+  res = p->g_FastPos[pos >> i] + (i * 2); }
+
 /*
 #define BSR2_RET(pos, res) { res = (pos < (1 << (kNumLogBits + 6))) ? \
   p->g_FastPos[pos >> 6] + 12 : \
@@ -213,6 +238,7 @@ #define LZMA_MATCH_LEN_MAX (LZMA_MATCH_LEN_MIN + kLenNumSymbolsTotal - 1)
 
 #define kNumStates 12
 
+
 typedef struct
 {
   CLzmaProb choice;
@@ -222,13 +248,15 @@ CLzmaProb mid[LZMA_NUM_PB_STATES_MAX << kLenNumMidBits];
   CLzmaProb high[kLenNumHighSymbols];
 } CLenEnc;
 
+
 typedef struct
 {
   CLenEnc p;
-  UInt32 prices[LZMA_NUM_PB_STATES_MAX][kLenNumSymbolsTotal];
   UInt32 tableSize;
+  UInt32 prices[LZMA_NUM_PB_STATES_MAX][kLenNumSymbolsTotal];
   UInt32 counters[LZMA_NUM_PB_STATES_MAX];
 } CLenPriceEnc;
+
 
 typedef struct
 {
@@ -243,11 +271,15 @@ ISeqOutStream *outStream;
   UInt64 processed;
   SRes res;
 } CRangeEnc;
+
 
 typedef struct
 {
   CLzmaProb *litProbs;
 
+  UInt32 state;
+  UInt32 reps[LZMA_NUM_REPS];
+
   CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX];
   CLzmaProb isRep[kNumStates];
   CLzmaProb isRepG0[kNumStates];
@@ -261,15 +293,49 @@ CLzmaProb posAlignEncoder[1 << kNumAlignBits];
   
   CLenPriceEnc lenEnc;
   CLenPriceEnc repLenEnc;
-
-  UInt32 reps[LZMA_NUM_REPS];
-  UInt32 state;
 } CSaveState;
+
 
 typedef struct
 {
+  void *matchFinderObj;
   IMatchFinder matchFinder;
-  void *matchFinderObj;
+
+  UInt32 optimumEndIndex;
+  UInt32 optimumCurrentIndex;
+
+  UInt32 longestMatchLength;
+  UInt32 numPairs;
+  UInt32 numAvail;
+
+  UInt32 numFastBytes;
+  UInt32 additionalOffset;
+  UInt32 reps[LZMA_NUM_REPS];
+  UInt32 state;
+
+  unsigned lc, lp, pb;
+  unsigned lpMask, pbMask;
+  unsigned lclp;
+
+  CLzmaProb *litProbs;
+
+  Bool fastMode;
+  Bool writeEndMark;
+  Bool finished;
+  Bool multiThread;
+  Bool needInit;
+
+  UInt64 nowPos64;
+  
+  UInt32 matchPriceCount;
+  UInt32 alignPriceCount;
+
+  UInt32 distTableSize;
+
+  UInt32 dictSize;
+  SRes result;
+
+  CRangeEnc rc;
 
   #ifndef _7ZIP_ST
   Bool mtMode;
@@ -282,12 +348,6 @@ #ifndef _7ZIP_ST
   Byte pad[128];
   #endif
   
-  UInt32 optimumEndIndex;
-  UInt32 optimumCurrentIndex;
-
-  UInt32 longestMatchLength;
-  UInt32 numPairs;
-  UInt32 numAvail;
   COptimal opt[kNumOpts];
   
   #ifndef LZMA_LOG_BSR
@@ -296,22 +356,10 @@ #endif
 
   UInt32 ProbPrices[kBitModelTotal >> kNumMoveReducingBits];
   UInt32 matches[LZMA_MATCH_LEN_MAX * 2 + 2 + 1];
-  UInt32 numFastBytes;
-  UInt32 additionalOffset;
-  UInt32 reps[LZMA_NUM_REPS];
-  UInt32 state;
 
   UInt32 posSlotPrices[kNumLenToPosStates][kDistTableSizeMax];
   UInt32 distancesPrices[kNumLenToPosStates][kNumFullDistances];
   UInt32 alignPrices[kAlignTableSize];
-  UInt32 alignPriceCount;
-
-  UInt32 distTableSize;
-
-  unsigned lc, lp, pb;
-  unsigned lpMask, pbMask;
-
-  CLzmaProb *litProbs;
 
   CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX];
   CLzmaProb isRep[kNumStates];
@@ -327,25 +375,13 @@
   CLenPriceEnc lenEnc;
   CLenPriceEnc repLenEnc;
 
-  unsigned lclp;
+  CSaveState saveState;
 
-  Bool fastMode;
-  
-  CRangeEnc rc;
+  #ifndef _7ZIP_ST
+  Byte pad2[128];
+  #endif
+} CLzmaEnc;
 
-  Bool writeEndMark;
-  UInt64 nowPos64;
-  UInt32 matchPriceCount;
-  Bool finished;
-  Bool multiThread;
-
-  SRes result;
-  UInt32 dictSize;
-
-  int needInit;
-
-  CSaveState saveState;
-} CLzmaEnc;
 
 void LzmaEnc_SaveState(CLzmaEncHandle pp)
 {
@@ -370,7 +406,7 @@ memcpy(dest->isRepG2, p->isRepG2, sizeof(p->isRepG2));
   memcpy(dest->posEncoders, p->posEncoders, sizeof(p->posEncoders));
   memcpy(dest->posAlignEncoder, p->posAlignEncoder, sizeof(p->posAlignEncoder));
   memcpy(dest->reps, p->reps, sizeof(p->reps));
-  memcpy(dest->litProbs, p->litProbs, (0x300 << p->lclp) * sizeof(CLzmaProb));
+  memcpy(dest->litProbs, p->litProbs, ((UInt32)0x300 << p->lclp) * sizeof(CLzmaProb));
 }
 
 void LzmaEnc_RestoreState(CLzmaEncHandle pp)
@@ -396,7 +432,7 @@ memcpy(dest->isRepG2, p->isRepG2, sizeof(p->isRepG2));
   memcpy(dest->posEncoders, p->posEncoders, sizeof(p->posEncoders));
   memcpy(dest->posAlignEncoder, p->posAlignEncoder, sizeof(p->posAlignEncoder));
   memcpy(dest->reps, p->reps, sizeof(p->reps));
-  memcpy(dest->litProbs, p->litProbs, (0x300 << dest->lclp) * sizeof(CLzmaProb));
+  memcpy(dest->litProbs, p->litProbs, ((UInt32)0x300 << dest->lclp) * sizeof(CLzmaProb));
 }
 
 SRes LzmaEnc_SetProps(CLzmaEncHandle pp, const CLzmaEncProps *props2)
@@ -405,9 +441,13 @@ CLzmaEnc *p = (CLzmaEnc *)pp;
   CLzmaEncProps props = *props2;
   LzmaEncProps_Normalize(&props);
 
-  if (props.lc > LZMA_LC_MAX || props.lp > LZMA_LP_MAX || props.pb > LZMA_PB_MAX ||
-      props.dictSize > ((UInt32)1 << kDicLogSizeMaxCompress) || props.dictSize > ((UInt32)1 << 30))
+  if (props.lc > LZMA_LC_MAX
+      || props.lp > LZMA_LP_MAX
+      || props.pb > LZMA_PB_MAX
+      || props.dictSize > ((UInt64)1 << kDicLogSizeMaxCompress)
+      || props.dictSize > kMaxHistorySize)
     return SZ_ERROR_PARAM;
+
   p->dictSize = props.dictSize;
   {
     unsigned fb = props.fb;
@@ -421,7 +461,7 @@ p->lc = props.lc;
   p->lp = props.lp;
   p->pb = props.pb;
   p->fastMode = (props.algo == 0);
-  p->matchFinderBase.btMode = props.btMode;
+  p->matchFinderBase.btMode = (Byte)(props.btMode ? 1 : 0);
   {
     UInt32 numHashBytes = 4;
     if (props.btMode)
@@ -465,8 +505,8 @@ #define kInfinityPrice (1 << 30)
 
 static void RangeEnc_Construct(CRangeEnc *p)
 {
-  p->outStream = 0;
-  p->bufBase = 0;
+  p->outStream = NULL;
+  p->bufBase = NULL;
 }
 
 #define RangeEnc_GetProcessed(p) ((p)->processed + ((p)->buf - (p)->bufBase) + (p)->cacheSize)
@@ -474,10 +514,10 @@
 #define RC_BUF_SIZE (1 << 16)
 static int RangeEnc_Alloc(CRangeEnc *p, ISzAlloc *alloc)
 {
-  if (p->bufBase == 0)
+  if (!p->bufBase)
   {
     p->bufBase = (Byte *)alloc->Alloc(alloc, RC_BUF_SIZE);
-    if (p->bufBase == 0)
+    if (!p->bufBase)
       return 0;
     p->bufLim = p->bufBase + RC_BUF_SIZE;
   }
@@ -607,7 +647,7 @@ }
   while (symbol < 0x10000);
 }
 
-void LzmaEnc_InitPriceTables(UInt32 *ProbPrices)
+static void LzmaEnc_InitPriceTables(UInt32 *ProbPrices)
 {
   UInt32 i;
   for (i = (1 << kNumMoveReducingBits) / 2; i < kBitModelTotal; i += (1 << kNumMoveReducingBits))
@@ -643,7 +683,7 @@
 #define GET_PRICE_0a(prob) ProbPrices[(prob) >> kNumMoveReducingBits]
 #define GET_PRICE_1a(prob) ProbPrices[((prob) ^ (kBitModelTotal - 1)) >> kNumMoveReducingBits]
 
-static UInt32 LitEnc_GetPrice(const CLzmaProb *probs, UInt32 symbol, UInt32 *ProbPrices)
+static UInt32 LitEnc_GetPrice(const CLzmaProb *probs, UInt32 symbol, const UInt32 *ProbPrices)
 {
   UInt32 price = 0;
   symbol |= 0x100;
@@ -656,7 +696,7 @@ while (symbol < 0x10000);
   return price;
 }
 
-static UInt32 LitEnc_GetPriceMatched(const CLzmaProb *probs, UInt32 symbol, UInt32 matchByte, UInt32 *ProbPrices)
+static UInt32 LitEnc_GetPriceMatched(const CLzmaProb *probs, UInt32 symbol, UInt32 matchByte, const UInt32 *ProbPrices)
 {
   UInt32 price = 0;
   UInt32 offs = 0x100;
@@ -700,7 +740,7 @@ symbol >>= 1;
   }
 }
 
-static UInt32 RcTree_GetPrice(const CLzmaProb *probs, int numBitLevels, UInt32 symbol, UInt32 *ProbPrices)
+static UInt32 RcTree_GetPrice(const CLzmaProb *probs, int numBitLevels, UInt32 symbol, const UInt32 *ProbPrices)
 {
   UInt32 price = 0;
   symbol |= (1 << numBitLevels);
@@ -712,7 +752,7 @@ }
   return price;
 }
 
-static UInt32 RcTree_ReverseGetPrice(const CLzmaProb *probs, int numBitLevels, UInt32 symbol, UInt32 *ProbPrices)
+static UInt32 RcTree_ReverseGetPrice(const CLzmaProb *probs, int numBitLevels, UInt32 symbol, const UInt32 *ProbPrices)
 {
   UInt32 price = 0;
   UInt32 m = 1;
@@ -763,7 +803,7 @@ }
   }
 }
 
-static void LenEnc_SetPrices(CLenEnc *p, UInt32 posState, UInt32 numSymbols, UInt32 *prices, UInt32 *ProbPrices)
+static void LenEnc_SetPrices(CLenEnc *p, UInt32 posState, UInt32 numSymbols, UInt32 *prices, const UInt32 *ProbPrices)
 {
   UInt32 a0 = GET_PRICE_0a(p->choice);
   UInt32 a1 = GET_PRICE_1a(p->choice);
@@ -786,20 +826,20 @@ for (; i < numSymbols; i++)
     prices[i] = b1 + RcTree_GetPrice(p->high, kLenNumHighBits, i - kLenNumLowSymbols - kLenNumMidSymbols, ProbPrices);
 }
 
-static void MY_FAST_CALL LenPriceEnc_UpdateTable(CLenPriceEnc *p, UInt32 posState, UInt32 *ProbPrices)
+static void MY_FAST_CALL LenPriceEnc_UpdateTable(CLenPriceEnc *p, UInt32 posState, const UInt32 *ProbPrices)
 {
   LenEnc_SetPrices(&p->p, posState, p->tableSize, p->prices[posState], ProbPrices);
   p->counters[posState] = p->tableSize;
 }
 
-static void LenPriceEnc_UpdateTables(CLenPriceEnc *p, UInt32 numPosStates, UInt32 *ProbPrices)
+static void LenPriceEnc_UpdateTables(CLenPriceEnc *p, UInt32 numPosStates, const UInt32 *ProbPrices)
 {
   UInt32 posState;
   for (posState = 0; posState < numPosStates; posState++)
     LenPriceEnc_UpdateTable(p, posState, ProbPrices);
 }
 
-static void LenEnc_Encode2(CLenPriceEnc *p, CRangeEnc *rc, UInt32 symbol, UInt32 posState, Bool updatePrice, UInt32 *ProbPrices)
+static void LenEnc_Encode2(CLenPriceEnc *p, CRangeEnc *rc, UInt32 symbol, UInt32 posState, Bool updatePrice, const UInt32 *ProbPrices)
 {
   LenEnc_Encode(&p->p, rc, symbol, posState);
   if (updatePrice)
@@ -814,7 +854,7 @@ static void MovePos(CLzmaEnc *p, UInt32 num)
 {
   #ifdef SHOW_STAT
   g_STAT_OFFSET += num;
-  printf("\n MovePos %d", num);
+  printf("\n MovePos %u", num);
   #endif
   
   if (num != 0)
@@ -831,12 +871,12 @@ p->numAvail = p->matchFinder.GetNumAvailableBytes(p->matchFinderObj);
   numPairs = p->matchFinder.GetMatches(p->matchFinderObj, p->matches);
   
   #ifdef SHOW_STAT
-  printf("\n i = %d numPairs = %d    ", g_STAT_OFFSET, numPairs / 2);
+  printf("\n i = %u numPairs = %u    ", g_STAT_OFFSET, numPairs / 2);
   g_STAT_OFFSET++;
   {
     UInt32 i;
     for (i = 0; i < numPairs; i += 2)
-      printf("%2d %6d   | ", p->matches[i], p->matches[i + 1]);
+      printf("%2u %6u   | ", p->matches[i], p->matches[i + 1]);
   }
   #endif
   
@@ -845,14 +885,16 @@ {
     lenRes = p->matches[numPairs - 2];
     if (lenRes == p->numFastBytes)
     {
-      const Byte *pby = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1;
-      UInt32 distance = p->matches[numPairs - 1] + 1;
       UInt32 numAvail = p->numAvail;
       if (numAvail > LZMA_MATCH_LEN_MAX)
         numAvail = LZMA_MATCH_LEN_MAX;
       {
-        const Byte *pby2 = pby - distance;
-        for (; lenRes < numAvail && pby[lenRes] == pby2[lenRes]; lenRes++);
+        const Byte *pbyCur = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1;
+        const Byte *pby = pbyCur + lenRes;
+        ptrdiff_t dif = (ptrdiff_t)-1 - p->matches[numPairs - 1];
+        const Byte *pbyLim = pbyCur + numAvail;
+        for (; pby != pbyLim && *pby == pby[dif]; pby++);
+        lenRes = (UInt32)(pby - pbyCur);
       }
     }
   }
@@ -937,7 +979,7 @@ p->optimumCurrentIndex  = p->opt[0].posPrev;
   return p->optimumCurrentIndex;
 }
 
-#define LIT_PROBS(pos, prevByte) (p->litProbs + ((((pos) & p->lpMask) << p->lc) + ((prevByte) >> (8 - p->lc))) * 0x300)
+#define LIT_PROBS(pos, prevByte) (p->litProbs + ((((pos) & p->lpMask) << p->lc) + ((prevByte) >> (8 - p->lc))) * (UInt32)0x300)
 
 static UInt32 GetOptimum(CLzmaEnc *p, UInt32 position, UInt32 *backRes)
 {
@@ -981,7 +1023,7 @@ {
     UInt32 lenTest;
     const Byte *data2;
     reps[i] = p->reps[i];
-    data2 = data - (reps[i] + 1);
+    data2 = data - reps[i] - 1;
     if (data[0] != data2[0] || data[1] != data2[1])
     {
       repLens[i] = 0;
@@ -1125,12 +1167,12 @@
   cur = 0;
 
     #ifdef SHOW_STAT2
-    if (position >= 0)
+    /* if (position >= 0) */
     {
       unsigned i;
       printf("\n pos = %4X", position);
       for (i = cur; i <= lenEnd; i++)
-      printf("\nprice[%4X] = %d", position - cur + i, p->opt[i].price);
+      printf("\nprice[%4X] = %u", position - cur + i, p->opt[i].price);
     }
     #endif
 
@@ -1282,7 +1324,7 @@ {
       /* try Literal + rep0 */
       UInt32 temp;
       UInt32 lenTest2;
-      const Byte *data2 = data - (reps[0] + 1);
+      const Byte *data2 = data - reps[0] - 1;
       UInt32 limit = p->numFastBytes + 1;
       if (limit > numAvailFull)
         limit = numAvailFull;
@@ -1325,7 +1367,7 @@ {
       UInt32 lenTest;
       UInt32 lenTestTemp;
       UInt32 price;
-      const Byte *data2 = data - (reps[repIndex] + 1);
+      const Byte *data2 = data - reps[repIndex] - 1;
       if (data[0] != data2[0] || data[1] != data2[1])
         continue;
       for (lenTest = 2; lenTest < numAvail && data[lenTest] == data2[lenTest]; lenTest++);
@@ -1355,13 +1397,13 @@ /* if (_maxMode) */
         {
           UInt32 lenTest2 = lenTest + 1;
           UInt32 limit = lenTest2 + p->numFastBytes;
-          UInt32 nextRepMatchPrice;
           if (limit > numAvailFull)
             limit = numAvailFull;
           for (; lenTest2 < limit && data[lenTest2] == data2[lenTest2]; lenTest2++);
           lenTest2 -= lenTest + 1;
           if (lenTest2 >= 2)
           {
+            UInt32 nextRepMatchPrice;
             UInt32 state2 = kRepNextStates[state];
             UInt32 posStateNext = (position + lenTest) & p->pbMask;
             UInt32 curAndLenCharPrice =
@@ -1442,16 +1484,16 @@
         if (/*_maxMode && */lenTest == matches[offs])
         {
           /* Try Match + Literal + Rep0 */
-          const Byte *data2 = data - (curBack + 1);
+          const Byte *data2 = data - curBack - 1;
           UInt32 lenTest2 = lenTest + 1;
           UInt32 limit = lenTest2 + p->numFastBytes;
-          UInt32 nextRepMatchPrice;
           if (limit > numAvailFull)
             limit = numAvailFull;
           for (; lenTest2 < limit && data[lenTest2] == data2[lenTest2]; lenTest2++);
           lenTest2 -= lenTest + 1;
           if (lenTest2 >= 2)
           {
+            UInt32 nextRepMatchPrice;
             UInt32 state2 = kMatchNextStates[state];
             UInt32 posStateNext = (position + lenTest) & p->pbMask;
             UInt32 curAndLenCharPrice = curAndLenPrice +
@@ -1525,7 +1567,7 @@ repLen = repIndex = 0;
   for (i = 0; i < LZMA_NUM_REPS; i++)
   {
     UInt32 len;
-    const Byte *data2 = data - (p->reps[i] + 1);
+    const Byte *data2 = data - p->reps[i] - 1;
     if (data[0] != data2[0] || data[1] != data2[1])
       continue;
     for (len = 2; len < numAvail && data[len] == data2[len]; len++);
@@ -1594,7 +1636,7 @@ data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1;
   for (i = 0; i < LZMA_NUM_REPS; i++)
   {
     UInt32 len, limit;
-    const Byte *data2 = data - (p->reps[i] + 1);
+    const Byte *data2 = data - p->reps[i] - 1;
     if (data[0] != data2[0] || data[1] != data2[1])
       continue;
     limit = mainLen - 1;
@@ -1690,6 +1732,7 @@ void LzmaEnc_Construct(CLzmaEnc *p)
 {
   RangeEnc_Construct(&p->rc);
   MatchFinder_Construct(&p->matchFinderBase);
+  
   #ifndef _7ZIP_ST
   MatchFinderMt_Construct(&p->matchFinderMt);
   p->matchFinderMt.MatchFinder = &p->matchFinderBase;
@@ -1706,15 +1749,15 @@ LzmaEnc_FastPosInit(p->g_FastPos);
   #endif
 
   LzmaEnc_InitPriceTables(p->ProbPrices);
-  p->litProbs = 0;
-  p->saveState.litProbs = 0;
+  p->litProbs = NULL;
+  p->saveState.litProbs = NULL;
 }
 
 CLzmaEncHandle LzmaEnc_Create(ISzAlloc *alloc)
 {
   void *p;
   p = alloc->Alloc(alloc, sizeof(CLzmaEnc));
-  if (p != 0)
+  if (p)
     LzmaEnc_Construct((CLzmaEnc *)p);
   return p;
 }
@@ -1723,8 +1766,8 @@ void LzmaEnc_FreeLits(CLzmaEnc *p, ISzAlloc *alloc)
 {
   alloc->Free(alloc, p->litProbs);
   alloc->Free(alloc, p->saveState.litProbs);
-  p->litProbs = 0;
-  p->saveState.litProbs = 0;
+  p->litProbs = NULL;
+  p->saveState.litProbs = NULL;
 }
 
 void LzmaEnc_Destruct(CLzmaEnc *p, ISzAlloc *alloc, ISzAlloc *allocBig)
@@ -1732,6 +1775,7 @@ {
   #ifndef _7ZIP_ST
   MatchFinderMt_Destruct(&p->matchFinderMt, allocBig);
   #endif
+  
   MatchFinder_Free(&p->matchFinderBase, allocBig);
   LzmaEnc_FreeLits(p, alloc);
   RangeEnc_Free(&p->rc, alloc);
@@ -1768,7 +1812,7 @@ return Flush(p, nowPos32);
     ReadMatchDistances(p, &numPairs);
     RangeEnc_EncodeBit(&p->rc, &p->isMatch[p->state][0], 0);
     p->state = kLiteralNextStates[p->state];
-    curByte = p->matchFinder.GetIndexByte(p->matchFinderObj, 0 - p->additionalOffset);
+    curByte = *(p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - p->additionalOffset);
     LitEnc_Encode(&p->rc, p->litProbs, curByte);
     p->additionalOffset--;
     nowPos32++;
@@ -1785,7 +1829,7 @@ else
       len = GetOptimum(p, nowPos32, &pos);
 
     #ifdef SHOW_STAT2
-    printf("\n pos = %4X,   len = %d   pos = %d", nowPos32, len, pos);
+    printf("\n pos = %4X,   len = %u   pos = %u", nowPos32, len, pos);
     #endif
 
     posState = nowPos32 & p->pbMask;
@@ -1894,7 +1938,7 @@ if (processed + kNumOpts + 300 >= maxUnpackSize ||
             RangeEnc_GetProcessed(&p->rc) + kNumOpts * 2 >= maxPackSize)
           break;
       }
-      else if (processed >= (1 << 15))
+      else if (processed >= (1 << 17))
       {
         p->nowPos64 += nowPos32 - startPos32;
         return CheckErrors(p);
@@ -1912,18 +1956,19 @@ {
   UInt32 beforeSize = kNumOpts;
   if (!RangeEnc_Alloc(&p->rc, alloc))
     return SZ_ERROR_MEM;
+
   #ifndef _7ZIP_ST
   p->mtMode = (p->multiThread && !p->fastMode && (p->matchFinderBase.btMode != 0));
   #endif
 
   {
     unsigned lclp = p->lc + p->lp;
-    if (p->litProbs == 0 || p->saveState.litProbs == 0 || p->lclp != lclp)
+    if (!p->litProbs || !p->saveState.litProbs || p->lclp != lclp)
     {
       LzmaEnc_FreeLits(p, alloc);
-      p->litProbs = (CLzmaProb *)alloc->Alloc(alloc, (0x300 << lclp) * sizeof(CLzmaProb));
-      p->saveState.litProbs = (CLzmaProb *)alloc->Alloc(alloc, (0x300 << lclp) * sizeof(CLzmaProb));
-      if (p->litProbs == 0 || p->saveState.litProbs == 0)
+      p->litProbs = (CLzmaProb *)alloc->Alloc(alloc, ((UInt32)0x300 << lclp) * sizeof(CLzmaProb));
+      p->saveState.litProbs = (CLzmaProb *)alloc->Alloc(alloc, ((UInt32)0x300 << lclp) * sizeof(CLzmaProb));
+      if (!p->litProbs || !p->saveState.litProbs)
       {
         LzmaEnc_FreeLits(p, alloc);
         return SZ_ERROR_MEM;
@@ -1932,7 +1977,7 @@ p->lclp = lclp;
     }
   }
 
-  p->matchFinderBase.bigHash = (p->dictSize > kBigHashDicLimit);
+  p->matchFinderBase.bigHash = (Byte)(p->dictSize > kBigHashDicLimit ? 1 : 0);
 
   if (beforeSize + p->dictSize < keepWindowSize)
     beforeSize = keepWindowSize - p->dictSize;
@@ -1952,6 +1997,7 @@ return SZ_ERROR_MEM;
     p->matchFinderObj = &p->matchFinderBase;
     MatchFinder_CreateVTable(&p->matchFinderBase, &p->matchFinder);
   }
+  
   return SZ_OK;
 }
 
@@ -1980,9 +2026,10 @@ p->isRepG2[i] = kProbInitValue;
   }
 
   {
-    UInt32 num = 0x300 << (p->lp + p->lc);
+    UInt32 num = (UInt32)0x300 << (p->lp + p->lc);
+    CLzmaProb *probs = p->litProbs;
     for (i = 0; i < num; i++)
-      p->litProbs[i] = kProbInitValue;
+      probs[i] = kProbInitValue;
   }
 
   {
@@ -2089,9 +2136,10 @@ CLzmaEnc *p = (CLzmaEnc *)pp;
   if (p->mtMode)
     MatchFinderMt_ReleaseStream(&p->matchFinderMt);
   #else
-  pp = pp;
+  UNUSED_VAR(pp);
   #endif
 }
+
 
 typedef struct
 {
@@ -2121,6 +2169,7 @@ {
   const CLzmaEnc *p = (CLzmaEnc *)pp;
   return p->matchFinder.GetNumAvailableBytes(p->matchFinderObj);
 }
+
 
 const Byte *LzmaEnc_GetCurBuf(CLzmaEncHandle pp)
 {
@@ -2128,6 +2177,7 @@ const CLzmaEnc *p = (CLzmaEnc *)pp;
   return p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - p->additionalOffset;
 }
 
+
 SRes LzmaEnc_CodeOneMemBlock(CLzmaEncHandle pp, Bool reInit,
     Byte *dest, size_t *destLen, UInt32 desiredPackSize, UInt32 *unpackSize)
 {
@@ -2162,6 +2212,7 @@
   return res;
 }
 
+
 static SRes LzmaEnc_Encode2(CLzmaEnc *p, ICompressProgress *progress)
 {
   SRes res = SZ_OK;
@@ -2175,9 +2226,9 @@
   for (;;)
   {
     res = LzmaEnc_CodeOneBlock(p, False, 0, 0);
-    if (res != SZ_OK || p->finished != 0)
+    if (res != SZ_OK || p->finished)
       break;
-    if (progress != 0)
+    if (progress)
     {
       res = progress->Progress(progress, p->nowPos64, RangeEnc_GetProcessed(&p->rc));
       if (res != SZ_OK)
@@ -2187,9 +2238,18 @@ break;
       }
     }
   }
+  
   LzmaEnc_Finish(p);
+
+  /*
+  if (res == S_OK && !Inline_MatchFinder_IsFinishedOK(&p->matchFinderBase))
+    res = SZ_ERROR_FAIL;
+  }
+  */
+
   return res;
 }
+
 
 SRes LzmaEnc_Encode(CLzmaEncHandle pp, ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress,
     ISzAlloc *alloc, ISzAlloc *allocBig)
@@ -2198,34 +2258,34 @@ RINOK(LzmaEnc_Prepare(pp, outStream, inStream, alloc, allocBig));
   return LzmaEnc_Encode2((CLzmaEnc *)pp, progress);
 }
 
+
 SRes LzmaEnc_WriteProperties(CLzmaEncHandle pp, Byte *props, SizeT *size)
 {
   CLzmaEnc *p = (CLzmaEnc *)pp;
-  int i;
+  unsigned i;
   UInt32 dictSize = p->dictSize;
   if (*size < LZMA_PROPS_SIZE)
     return SZ_ERROR_PARAM;
   *size = LZMA_PROPS_SIZE;
   props[0] = (Byte)((p->pb * 5 + p->lp) * 9 + p->lc);
 
-  for (i = 11; i <= 30; i++)
+  if (dictSize >= ((UInt32)1 << 22))
   {
-    if (dictSize <= ((UInt32)2 << i))
-    {
-      dictSize = (2 << i);
-      break;
-    }
-    if (dictSize <= ((UInt32)3 << i))
-    {
-      dictSize = (3 << i);
-      break;
-    }
+    UInt32 kDictMask = ((UInt32)1 << 20) - 1;
+    if (dictSize < (UInt32)0xFFFFFFFF - kDictMask)
+      dictSize = (dictSize + kDictMask) & ~kDictMask;
+  }
+  else for (i = 11; i <= 30; i++)
+  {
+    if (dictSize <= ((UInt32)2 << i)) { dictSize = (2 << i); break; }
+    if (dictSize <= ((UInt32)3 << i)) { dictSize = (3 << i); break; }
   }
 
   for (i = 0; i < 4; i++)
     props[1 + i] = (Byte)(dictSize >> (8 * i));
   return SZ_OK;
 }
+
 
 SRes LzmaEnc_MemEncode(CLzmaEncHandle pp, Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
     int writeEndMark, ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig)
@@ -2235,25 +2295,29 @@ CLzmaEnc *p = (CLzmaEnc *)pp;
 
   CSeqOutStreamBuf outStream;
 
-  LzmaEnc_SetInputBuf(p, src, srcLen);
-
   outStream.funcTable.Write = MyWrite;
   outStream.data = dest;
   outStream.rem = *destLen;
   outStream.overflow = False;
 
   p->writeEndMark = writeEndMark;
+  p->rc.outStream = &outStream.funcTable;
 
-  p->rc.outStream = &outStream.funcTable;
   res = LzmaEnc_MemPrepare(pp, src, srcLen, 0, alloc, allocBig);
+  
   if (res == SZ_OK)
+  {
     res = LzmaEnc_Encode2(p, progress);
+    if (res == SZ_OK && p->nowPos64 != srcLen)
+      res = SZ_ERROR_FAIL;
+  }
 
   *destLen -= outStream.rem;
   if (outStream.overflow)
     return SZ_ERROR_OUTPUT_EOF;
   return res;
 }
+
 
 SRes LzmaEncode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
     const CLzmaEncProps *props, Byte *propsEncoded, SizeT *propsSize, int writeEndMark,
@@ -2261,7 +2325,7 @@ ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig)
 {
   CLzmaEnc *p = (CLzmaEnc *)LzmaEnc_Create(alloc);
   SRes res;
-  if (p == 0)
+  if (!p)
     return SZ_ERROR_MEM;
 
   res = LzmaEnc_SetProps(p, props);

@@ -1,18 +1,12 @@
 /* LzmaLib.c -- LZMA library wrapper
-2008-08-05
-Igor Pavlov
-Public domain */
+2015-06-13 : Igor Pavlov : Public domain */
 
-#include "LzmaEnc.h"
+#include "Alloc.h"
 #include "LzmaDec.h"
-#include "Alloc.h"
+#include "LzmaEnc.h"
 #include "LzmaLib.h"
 
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
-static ISzAlloc g_Alloc = { SzAlloc, SzFree };
-
-MY_STDAPI LzmaCompress(unsigned char *dest, size_t  *destLen, const unsigned char *src, size_t  srcLen,
+MY_STDAPI LzmaCompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t srcLen,
   unsigned char *outProps, size_t *outPropsSize,
   int level, /* 0 <= level <= 9, default = 5 */
   unsigned dictSize, /* use (1 << N) or (3 << N). 4 KB < dictSize <= 128 MB */
@@ -38,7 +32,7 @@ NULL, &g_Alloc, &g_Alloc);
 }
 
 
-MY_STDAPI LzmaUncompress(unsigned char *dest, size_t  *destLen, const unsigned char *src, size_t  *srcLen,
+MY_STDAPI LzmaUncompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t *srcLen,
   const unsigned char *props, size_t propsSize)
 {
   ELzmaStatus status;

@@ -1,9 +1,7 @@
 /* MtCoder.c -- Multi-thread Coder
-2010-09-24 : Igor Pavlov : Public domain */
+2015-10-13 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
-
-#include <stdio.h>
 
 #include "MtCoder.h"
 
@@ -120,7 +118,7 @@ Event_Construct(&p->canWrite);
   LoopThread_Construct(&p->thread);
 }
 
-#define RINOK_THREAD(x) { if((x) != 0) return SZ_ERROR_THREAD; }
+#define RINOK_THREAD(x) { if ((x) != 0) return SZ_ERROR_THREAD; }
 
 static void CMtThread_CloseEvents(CMtThread *p)
 {

@@ -1,10 +1,10 @@
 /* Ppmd7.c -- PPMdH codec
-2010-03-12 : Igor Pavlov : Public domain
+2015-09-28 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 #include "Precomp.h"
 
-#include <memory.h>
+#include <string.h>
 
 #include "Ppmd7.h"
 
@@ -66,7 +66,7 @@
   for (i = 0, k = 0; i < PPMD_NUM_INDEXES; i++)
   {
     unsigned step = (i >= 12 ? 4 : (i >> 2) + 1);
-    do { p->Units2Indx[k++] = (Byte)i; } while(--step);
+    do { p->Units2Indx[k++] = (Byte)i; } while (--step);
     p->Indx2Units[i] = (Byte)k;
   }
 
@@ -257,7 +257,7 @@ }
 
 #define MyMem12Cpy(dest, src, num) \
   { UInt32 *d = (UInt32 *)dest; const UInt32 *s = (const UInt32 *)src; UInt32 n = num; \
-    do { d[0] = s[0]; d[1] = s[1]; d[2] = s[2]; s += 3; d += 3; } while(--n); }
+    do { d[0] = s[0]; d[1] = s[1]; d[2] = s[2]; s += 3; d += 3; } while (--n); }
 
 static void *ShrinkUnits(CPpmd7 *p, void *oldPtr, unsigned oldNU, unsigned newNU)
 {

@@ -1,5 +1,5 @@
 /* Ppmd7Enc.c -- PPMdH Encoder
-2010-03-12 : Igor Pavlov : Public domain
+2015-09-28 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 #include "Precomp.h"
@@ -26,7 +26,7 @@ {
       p->Stream->Write(p->Stream, (Byte)(temp + (Byte)(p->Low >> 32)));
       temp = 0xFF;
     }
-    while(--p->CacheSize != 0);
+    while (--p->CacheSize != 0);
     p->Cache = (Byte)((UInt32)p->Low >> 24);
   }
   p->CacheSize++;

@@ -1,5 +1,5 @@
 /* RotateDefs.h -- Rotate functions
-2013-11-12 : Igor Pavlov : Public domain */
+2015-03-25 : Igor Pavlov : Public domain */
 
 #ifndef __ROTATE_DEFS_H
 #define __ROTATE_DEFS_H
@@ -8,15 +8,19 @@ #ifdef _MSC_VER
 
 #include <stdlib.h>
 
-// #if (_MSC_VER >= 1200)
+/* don't use _rotl with MINGW. It can insert slow call to function. */
+ 
+/* #if (_MSC_VER >= 1200) */
 #pragma intrinsic(_rotl)
 #pragma intrinsic(_rotr)
-// #endif
+/* #endif */
 
 #define rotlFixed(x, n) _rotl((x), (n))
 #define rotrFixed(x, n) _rotr((x), (n))
 
 #else
+
+/* new compilers can translate these macros to fast commands. */
 
 #define rotlFixed(x, n) (((x) << (n)) | ((x) >> (32 - (n))))
 #define rotrFixed(x, n) (((x) >> (n)) | ((x) << (32 - (n))))

@@ -1,14 +1,21 @@
 /* Crypto/Sha256.c -- SHA-256 Hash
-2010-06-11 : Igor Pavlov : Public domain
+2015-11-14 : Igor Pavlov : Public domain
 This code is based on public domain code from Wei Dai's Crypto++ library. */
 
 #include "Precomp.h"
 
+#include <string.h>
+
+#include "CpuArch.h"
 #include "RotateDefs.h"
 #include "Sha256.h"
 
 /* define it for speed optimization */
-/* #define _SHA256_UNROLL */
+#ifndef _SFX
+#define _SHA256_UNROLL
+#define _SHA256_UNROLL2
+#endif
+
 /* #define _SHA256_UNROLL2 */
 
 void Sha256_Init(CSha256 *p)
@@ -29,26 +36,18 @@ #define S1(x) (rotrFixed(x, 6) ^ rotrFixed(x,11) ^ rotrFixed(x, 25))
 #define s0(x) (rotrFixed(x, 7) ^ rotrFixed(x,18) ^ (x >> 3))
 #define s1(x) (rotrFixed(x,17) ^ rotrFixed(x,19) ^ (x >> 10))
 
-#define blk0(i) (W[i] = data[i])
-#define blk2(i) (W[i&15] += s1(W[(i-2)&15]) + W[(i-7)&15] + s0(W[(i-15)&15]))
+#define blk0(i) (W[i])
+#define blk2(i) (W[i] += s1(W[((i)-2)&15]) + W[((i)-7)&15] + s0(W[((i)-15)&15]))
 
 #define Ch(x,y,z) (z^(x&(y^z)))
 #define Maj(x,y,z) ((x&y)|(z&(x|y)))
 
-#define a(i) T[(0-(i))&7]
-#define b(i) T[(1-(i))&7]
-#define c(i) T[(2-(i))&7]
-#define d(i) T[(3-(i))&7]
-#define e(i) T[(4-(i))&7]
-#define f(i) T[(5-(i))&7]
-#define g(i) T[(6-(i))&7]
-#define h(i) T[(7-(i))&7]
-
-
 #ifdef _SHA256_UNROLL2
 
-#define R(a,b,c,d,e,f,g,h, i) h += S1(e) + Ch(e,f,g) + K[i+j] + (j?blk2(i):blk0(i));\
-  d += h; h += S0(a) + Maj(a, b, c)
+#define R(a,b,c,d,e,f,g,h, i) \
+    h += S1(e) + Ch(e,f,g) + K[(i)+(j)] + (j ? blk2(i) : blk0(i)); \
+    d += h; \
+    h += S0(a) + Maj(a, b, c)
 
 #define RX_8(i) \
   R(a,b,c,d,e,f,g,h, i); \
@@ -60,14 +59,32 @@ R(d,e,f,g,h,a,b,c, i+5); \
   R(c,d,e,f,g,h,a,b, i+6); \
   R(b,c,d,e,f,g,h,a, i+7)
 
+#define RX_16  RX_8(0); RX_8(8);
+
 #else
 
-#define R(i) h(i) += S1(e(i)) + Ch(e(i),f(i),g(i)) + K[i+j] + (j?blk2(i):blk0(i));\
-  d(i) += h(i); h(i) += S0(a(i)) + Maj(a(i), b(i), c(i))
+#define a(i) T[(0-(i))&7]
+#define b(i) T[(1-(i))&7]
+#define c(i) T[(2-(i))&7]
+#define d(i) T[(3-(i))&7]
+#define e(i) T[(4-(i))&7]
+#define f(i) T[(5-(i))&7]
+#define g(i) T[(6-(i))&7]
+#define h(i) T[(7-(i))&7]
+
+#define R(i) \
+    h(i) += S1(e(i)) + Ch(e(i),f(i),g(i)) + K[(i)+(j)] + (j ? blk2(i) : blk0(i)); \
+    d(i) += h(i); \
+    h(i) += S0(a(i)) + Maj(a(i), b(i), c(i)) \
 
 #ifdef _SHA256_UNROLL
 
-#define RX_8(i) R(i+0); R(i+1); R(i+2); R(i+3); R(i+4); R(i+5); R(i+6); R(i+7);
+#define RX_8(i)  R(i+0); R(i+1); R(i+2); R(i+3); R(i+4); R(i+5); R(i+6); R(i+7);
+#define RX_16  RX_8(0); RX_8(8);
+
+#else
+
+#define RX_16  unsigned i; for (i = 0; i < 16; i++) { R(i); }
 
 #endif
 
@@ -92,12 +109,30 @@ 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,
   0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
 };
 
-static void Sha256_Transform(UInt32 *state, const UInt32 *data)
+static void Sha256_WriteByteBlock(CSha256 *p)
 {
   UInt32 W[16];
   unsigned j;
+  UInt32 *state;
+
   #ifdef _SHA256_UNROLL2
   UInt32 a,b,c,d,e,f,g,h;
+  #else
+  UInt32 T[8];
+  #endif
+
+  for (j = 0; j < 16; j += 4)
+  {
+    const Byte *ccc = p->buffer + j * 4;
+    W[j    ] = GetBe32(ccc);
+    W[j + 1] = GetBe32(ccc + 4);
+    W[j + 2] = GetBe32(ccc + 8);
+    W[j + 3] = GetBe32(ccc + 12);
+  }
+
+  state = p->state;
+
+  #ifdef _SHA256_UNROLL2
   a = state[0];
   b = state[1];
   c = state[2];
@@ -107,19 +142,13 @@ f = state[5];
   g = state[6];
   h = state[7];
   #else
-  UInt32 T[8];
   for (j = 0; j < 8; j++)
     T[j] = state[j];
   #endif
 
   for (j = 0; j < 64; j += 16)
   {
-    #if defined(_SHA256_UNROLL) || defined(_SHA256_UNROLL2)
-    RX_8(0); RX_8(8);
-    #else
-    unsigned i;
-    for (i = 0; i < 16; i++) { R(i); }
-    #endif
+    RX_16
   }
 
   #ifdef _SHA256_UNROLL2
@@ -146,61 +175,74 @@ #undef S1
 #undef s0
 #undef s1
 
-static void Sha256_WriteByteBlock(CSha256 *p)
+void Sha256_Update(CSha256 *p, const Byte *data, size_t size)
 {
-  UInt32 data32[16];
-  unsigned i;
-  for (i = 0; i < 16; i++)
-    data32[i] =
-      ((UInt32)(p->buffer[i * 4    ]) << 24) +
-      ((UInt32)(p->buffer[i * 4 + 1]) << 16) +
-      ((UInt32)(p->buffer[i * 4 + 2]) <<  8) +
-      ((UInt32)(p->buffer[i * 4 + 3]));
-  Sha256_Transform(p->state, data32);
-}
+  if (size == 0)
+    return;
 
-void Sha256_Update(CSha256 *p, const Byte *data, size_t size)
-{
-  UInt32 curBufferPos = (UInt32)p->count & 0x3F;
-  while (size > 0)
   {
-    p->buffer[curBufferPos++] = *data++;
-    p->count++;
-    size--;
-    if (curBufferPos == 64)
+    unsigned pos = (unsigned)p->count & 0x3F;
+    unsigned num;
+    
+    p->count += size;
+    
+    num = 64 - pos;
+    if (num > size)
     {
-      curBufferPos = 0;
-      Sha256_WriteByteBlock(p);
+      memcpy(p->buffer + pos, data, size);
+      return;
     }
+    
+    size -= num;
+    memcpy(p->buffer + pos, data, num);
+    data += num;
   }
+
+  for (;;)
+  {
+    Sha256_WriteByteBlock(p);
+    if (size < 64)
+      break;
+    size -= 64;
+    memcpy(p->buffer, data, 64);
+    data += 64;
+  }
+
+  if (size != 0)
+    memcpy(p->buffer, data, size);
 }
 
 void Sha256_Final(CSha256 *p, Byte *digest)
 {
-  UInt64 lenInBits = (p->count << 3);
-  UInt32 curBufferPos = (UInt32)p->count & 0x3F;
+  unsigned pos = (unsigned)p->count & 0x3F;
   unsigned i;
-  p->buffer[curBufferPos++] = 0x80;
-  while (curBufferPos != (64 - 8))
+  
+  p->buffer[pos++] = 0x80;
+  
+  while (pos != (64 - 8))
   {
-    curBufferPos &= 0x3F;
-    if (curBufferPos == 0)
+    pos &= 0x3F;
+    if (pos == 0)
       Sha256_WriteByteBlock(p);
-    p->buffer[curBufferPos++] = 0;
+    p->buffer[pos++] = 0;
   }
-  for (i = 0; i < 8; i++)
+
   {
-    p->buffer[curBufferPos++] = (Byte)(lenInBits >> 56);
-    lenInBits <<= 8;
+    UInt64 numBits = (p->count << 3);
+    SetBe32(p->buffer + 64 - 8, (UInt32)(numBits >> 32));
+    SetBe32(p->buffer + 64 - 4, (UInt32)(numBits));
   }
+  
   Sha256_WriteByteBlock(p);
 
-  for (i = 0; i < 8; i++)
+  for (i = 0; i < 8; i += 2)
   {
-    *digest++ = (Byte)(p->state[i] >> 24);
-    *digest++ = (Byte)(p->state[i] >> 16);
-    *digest++ = (Byte)(p->state[i] >> 8);
-    *digest++ = (Byte)(p->state[i]);
+    UInt32 v0 = p->state[i];
+    UInt32 v1 = p->state[i + 1];
+    SetBe32(digest    , v0);
+    SetBe32(digest + 4, v1);
+    digest += 8;
   }
+  
   Sha256_Init(p);
 }