18.05

Asm/x86/7zAsm.asm

@@ -1,5 +1,5 @@
 ; 7zAsm.asm -- ASM macros
-; 2012-12-30 : Igor Pavlov : Public domain
+; 2018-02-03 : Igor Pavlov : Public domain
 
 MY_ASM_START macro
   ifdef x64
@@ -52,6 +52,15 @@ endif
   x6 equ ESI
   x7 equ EDI
 
+  x0_W equ AX
+  x1_W equ CX
+  x2_W equ DX
+  x3_W equ BX
+
+  x5_W equ BP
+  x6_W equ SI
+  x7_W equ DI
+
   x0_L equ AL
   x1_L equ CL
   x2_L equ DL
@@ -63,6 +72,10 @@ endif
   x3_H equ BH
 
 ifdef x64
+  x5_L equ BPL
+  x6_L equ SIL
+  x7_L equ DIL
+
   r0 equ RAX
   r1 equ RCX
   r2 equ RDX
@@ -103,3 +116,32 @@ MY_POP_4_REGS macro
     pop     r5
     pop     r3
 endm
+
+
+ifdef x64
+
+; for WIN64-x64 ABI:
+
+REG_PARAM_0 equ r1
+REG_PARAM_1 equ r2
+REG_PARAM_2 equ r8
+REG_PARAM_3 equ r9
+
+MY_PUSH_PRESERVED_REGS macro
+    MY_PUSH_4_REGS
+    push    r12
+    push    r13
+    push    r14
+    push    r15
+endm
+
+
+MY_POP_PRESERVED_REGS macro
+    pop     r15
+    pop     r14
+    pop     r13
+    pop     r12
+    MY_POP_4_REGS
+endm
+
+endif

C/7z.h

@@ -1,5 +1,5 @@
 /* 7z.h -- 7z interface
-2014-02-08 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_H
 #define __7Z_H
@@ -48,21 +48,10 @@ typedef struct
   UInt32 PackStreams[SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX];
   CSzBond Bonds[SZ_NUM_BONDS_IN_FOLDER_MAX];
   CSzCoderInfo Coders[SZ_NUM_CODERS_IN_FOLDER_MAX];
-  UInt64 CodersUnpackSizes[SZ_NUM_CODERS_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
 {
@@ -92,29 +81,77 @@ typedef struct
   UInt32 NumPackStreams;
   UInt32 NumFolders;
 
-  UInt64 *PackPositions; // NumPackStreams + 1
+  UInt64 *PackPositions;          // NumPackStreams + 1
-  CSzBitUi32s FolderCRCs;
+  CSzBitUi32s FolderCRCs;         // NumFolders
 
-  size_t *FoCodersOffsets;
+  size_t *FoCodersOffsets;        // NumFolders + 1
-  size_t *FoSizesOffsets;
+  UInt32 *FoStartPackStreamIndex; // NumFolders + 1
-  // UInt32 StartCoderUnpackSizesIndex;
+  UInt32 *FoToCoderUnpackSizes;   // NumFolders + 1
-  UInt32 *FoStartPackStreamIndex;
+  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);
+    ISzAllocPtr allocMain);
+
+typedef struct
+{
+  CSzAr db;
+
+  UInt64 startPosAfterHeader;
+  UInt64 dataPos;
+  
+  UInt32 NumFiles;
+
+  UInt64 *UnpackPositions;  // NumFiles + 1
+  // Byte *IsEmptyFiles;
+  Byte *IsDirs;
+  CSzBitUi32s CRCs;
+
+  CSzBitUi32s Attribs;
+  // CSzBitUi32s Parents;
+  CSzBitUi64s MTime;
+  CSzBitUi64s CTime;
+
+  UInt32 *FolderToFile;   // NumFolders + 1
+  UInt32 *FileToFolder;   // NumFiles
+
+  size_t *FileNameOffsets; /* in 2-byte steps */
+  Byte *FileNames;  /* UTF-16-LE */
+} CSzArEx;
+
+#define SzArEx_IsDir(p, i) (SzBitArray_Check((p)->IsDirs, i))
+
+#define SzArEx_GetFileSize(p, i) ((p)->UnpackPositions[(i) + 1] - (p)->UnpackPositions[i])
+
+void SzArEx_Init(CSzArEx *p);
+void SzArEx_Free(CSzArEx *p, ISzAllocPtr alloc);
+UInt64 SzArEx_GetFolderStreamPos(const CSzArEx *p, UInt32 folderIndex, UInt32 indexInFolder);
+int SzArEx_GetFolderFullPackSize(const CSzArEx *p, UInt32 folderIndex, UInt64 *resSize);
 
 /*
-  SzExtract extracts file from archive
+if dest == NULL, the return value specifies the required size of the buffer,
+  in 16-bit characters, including the null-terminating character.
+if dest != NULL, the return value specifies the number of 16-bit characters that
+  are written to the dest, including the null-terminating character. */
+
+size_t SzArEx_GetFileNameUtf16(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
+
+/*
+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.
 
@@ -133,55 +170,6 @@ SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
     Free *outBuffer and set *outBuffer to 0, if you want to flush cache.
 */
 
-typedef struct
-{
-  CSzAr db;
-
-  UInt64 startPosAfterHeader;
-  UInt64 dataPos;
-  
-  UInt32 NumFiles;
-
-  UInt64 *UnpackPositions;
-  // Byte *IsEmptyFiles;
-  Byte *IsDirs;
-  CSzBitUi32s CRCs;
-
-  CSzBitUi32s Attribs;
-  // CSzBitUi32s Parents;
-  CSzBitUi64s MTime;
-  CSzBitUi64s CTime;
-
-  // UInt32 *FolderStartPackStreamIndex;
-  UInt32 *FolderStartFileIndex; // + 1
-  UInt32 *FileIndexToFolderIndexMap;
-
-  size_t *FileNameOffsets; /* in 2-byte steps */
-  Byte *FileNames;  /* UTF-16-LE */
-} CSzArEx;
-
-#define SzArEx_IsDir(p, i) (SzBitArray_Check((p)->IsDirs, i))
-
-#define SzArEx_GetFileSize(p, i) ((p)->UnpackPositions[(i) + 1] - (p)->UnpackPositions[i])
-
-void SzArEx_Init(CSzArEx *p);
-void SzArEx_Free(CSzArEx *p, ISzAlloc *alloc);
-UInt64 SzArEx_GetFolderStreamPos(const CSzArEx *p, UInt32 folderIndex, UInt32 indexInFolder);
-int SzArEx_GetFolderFullPackSize(const CSzArEx *p, UInt32 folderIndex, UInt64 *resSize);
-
-/*
-if dest == NULL, the return value specifies the required size of the buffer,
-  in 16-bit characters, including the null-terminating character.
-if dest != NULL, the return value specifies the number of 16-bit characters that
-  are written to the dest, including the null-terminating character. */
-
-size_t SzArEx_GetFileNameUtf16(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
-
-/*
-size_t SzArEx_GetFullNameLen(const CSzArEx *p, size_t fileIndex);
-UInt16 *SzArEx_GetFullNameUtf16_Back(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
-*/
-
 SRes SzArEx_Extract(
     const CSzArEx *db,
     ILookInStream *inStream,
@@ -191,8 +179,8 @@ SRes SzArEx_Extract(
     size_t *outBufferSize,    /* buffer size for output buffer */
     size_t *offset,           /* offset of stream for required file in *outBuffer */
     size_t *outSizeProcessed, /* size of file in *outBuffer */
-    ISzAlloc *allocMain,
+    ISzAllocPtr allocMain,
-    ISzAlloc *allocTemp);
+    ISzAllocPtr allocTemp);
 
 
 /*
@@ -207,7 +195,7 @@ SZ_ERROR_FAIL
 */
 
 SRes SzArEx_Open(CSzArEx *p, ILookInStream *inStream,
-    ISzAlloc *allocMain, ISzAlloc *allocTemp);
+    ISzAllocPtr allocMain, ISzAllocPtr allocTemp);
 
 EXTERN_C_END
 

C/7zAlloc.c

@@ -1,8 +1,10 @@
 /* 7zAlloc.c -- Allocation functions
-2015-02-21 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
+#include <stdlib.h>
+
 #include "7zAlloc.h"
 
 /* #define _SZ_ALLOC_DEBUG */
@@ -20,19 +22,19 @@ int g_allocCountTemp = 0;
 
 #endif
 
-void *SzAlloc(void *p, size_t size)
+void *SzAlloc(ISzAllocPtr p, size_t size)
 {
   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);
 }
 
-void SzFree(void *p, void *address)
+void SzFree(ISzAllocPtr p, void *address)
 {
   UNUSED_VAR(p);
   #ifdef _SZ_ALLOC_DEBUG
@@ -45,13 +47,13 @@ void SzFree(void *p, void *address)
   free(address);
 }
 
-void *SzAllocTemp(void *p, size_t size)
+void *SzAllocTemp(ISzAllocPtr p, size_t size)
 {
   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);
@@ -60,7 +62,7 @@ void *SzAllocTemp(void *p, size_t size)
   return malloc(size);
 }
 
-void SzFreeTemp(void *p, void *address)
+void SzFreeTemp(ISzAllocPtr p, void *address)
 {
   UNUSED_VAR(p);
   #ifdef _SZ_ALLOC_DEBUG

C/7zAlloc.h

@@ -1,23 +1,19 @@
 /* 7zAlloc.h -- Allocation functions
-2013-03-25 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_ALLOC_H
 #define __7Z_ALLOC_H
 
-#include <stdlib.h>
+#include "7zTypes.h"
 
-#ifdef __cplusplus
+EXTERN_C_BEGIN
-extern "C" {
-#endif
 
-void *SzAlloc(void *p, size_t size);
+void *SzAlloc(ISzAllocPtr p, size_t size);
-void SzFree(void *p, void *address);
+void SzFree(ISzAllocPtr p, void *address);
 
-void *SzAllocTemp(void *p, size_t size);
+void *SzAllocTemp(ISzAllocPtr p, size_t size);
-void SzFreeTemp(void *p, void *address);
+void SzFreeTemp(ISzAllocPtr p, void *address);
 
-#ifdef __cplusplus
+EXTERN_C_END
-}
-#endif
 
 #endif

C/7zBuf.c

@@ -1,5 +1,5 @@
 /* 7zBuf.c -- Byte Buffer
-2013-01-21 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -11,7 +11,7 @@ void Buf_Init(CBuf *p)
   p->size = 0;
 }
 
-int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
+int Buf_Create(CBuf *p, size_t size, ISzAllocPtr alloc)
 {
   p->size = 0;
   if (size == 0)
@@ -19,8 +19,8 @@ int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
     p->data = 0;
     return 1;
   }
-  p->data = (Byte *)alloc->Alloc(alloc, size);
+  p->data = (Byte *)ISzAlloc_Alloc(alloc, size);
-  if (p->data != 0)
+  if (p->data)
   {
     p->size = size;
     return 1;
@@ -28,9 +28,9 @@ int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
   return 0;
 }
 
-void Buf_Free(CBuf *p, ISzAlloc *alloc)
+void Buf_Free(CBuf *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->data);
+  ISzAlloc_Free(alloc, p->data);
   p->data = 0;
   p->size = 0;
 }

C/7zBuf.h

@@ -1,5 +1,5 @@
 /* 7zBuf.h -- Byte Buffer
-2013-01-18 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_BUF_H
 #define __7Z_BUF_H
@@ -15,8 +15,8 @@ typedef struct
 } CBuf;
 
 void Buf_Init(CBuf *p);
-int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc);
+int Buf_Create(CBuf *p, size_t size, ISzAllocPtr alloc);
-void Buf_Free(CBuf *p, ISzAlloc *alloc);
+void Buf_Free(CBuf *p, ISzAllocPtr alloc);
 
 typedef struct
 {
@@ -27,8 +27,8 @@ typedef struct
 
 void DynBuf_Construct(CDynBuf *p);
 void DynBuf_SeekToBeg(CDynBuf *p);
-int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAlloc *alloc);
+int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAllocPtr alloc);
-void DynBuf_Free(CDynBuf *p, ISzAlloc *alloc);
+void DynBuf_Free(CDynBuf *p, ISzAllocPtr alloc);
 
 EXTERN_C_END
 

C/7zBuf2.c

@@ -1,5 +1,5 @@
 /* 7zBuf2.c -- Byte Buffer
-2014-08-22 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -19,19 +19,20 @@ void DynBuf_SeekToBeg(CDynBuf *p)
   p->pos = 0;
 }
 
-int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAlloc *alloc)
+int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAllocPtr alloc)
 {
   if (size > p->size - p->pos)
   {
     size_t newSize = p->pos + size;
     Byte *data;
     newSize += newSize / 4;
-    data = (Byte *)alloc->Alloc(alloc, newSize);
+    data = (Byte *)ISzAlloc_Alloc(alloc, newSize);
-    if (data == 0)
+    if (!data)
       return 0;
     p->size = newSize;
-    memcpy(data, p->data, p->pos);
+    if (p->pos != 0)
-    alloc->Free(alloc, p->data);
+      memcpy(data, p->data, p->pos);
+    ISzAlloc_Free(alloc, p->data);
     p->data = data;
   }
   if (size != 0)
@@ -42,9 +43,9 @@ int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAlloc *alloc)
   return 1;
 }
 
-void DynBuf_Free(CDynBuf *p, ISzAlloc *alloc)
+void DynBuf_Free(CDynBuf *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->data);
+  ISzAlloc_Free(alloc, p->data);
   p->data = 0;
   p->size = 0;
   p->pos = 0;

C/7zCrc.c

@@ -1,5 +1,5 @@
 /* 7zCrc.c -- CRC32 init
-2015-03-10 : Igor Pavlov : Public domain */
+2017-06-06 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -61,12 +61,12 @@ void MY_FAST_CALL CrcGenerateTable()
     UInt32 r = i;
     unsigned j;
     for (j = 0; j < 8; j++)
-      r = (r >> 1) ^ (kCrcPoly & ~((r & 1) - 1));
+      r = (r >> 1) ^ (kCrcPoly & ((UInt32)0 - (r & 1)));
     g_CrcTable[i] = r;
   }
-  for (; i < 256 * CRC_NUM_TABLES; i++)
+  for (i = 256; i < 256 * CRC_NUM_TABLES; i++)
   {
-    UInt32 r = g_CrcTable[i - 256];
+    UInt32 r = g_CrcTable[(size_t)i - 256];
     g_CrcTable[i] = g_CrcTable[r & 0xFF] ^ (r >> 8);
   }
 
@@ -86,8 +86,8 @@ void MY_FAST_CALL CrcGenerateTable()
   
       #ifdef MY_CPU_X86_OR_AMD64
       if (!CPU_Is_InOrder())
-        g_CrcUpdate = CrcUpdateT8;
       #endif
+        g_CrcUpdate = CrcUpdateT8;
     #endif
 
   #else
@@ -101,7 +101,7 @@ void MY_FAST_CALL CrcGenerateTable()
       g_CrcUpdate = CrcUpdateT4;
       #if CRC_NUM_TABLES >= 8
       g_CrcUpdateT8 = CrcUpdateT8;
-      // g_CrcUpdate = CrcUpdateT8;
+      g_CrcUpdate = CrcUpdateT8;
       #endif
     }
     else if (p[0] != 1 || p[1] != 2)
@@ -111,14 +111,14 @@ void MY_FAST_CALL CrcGenerateTable()
     {
       for (i = 256 * CRC_NUM_TABLES - 1; i >= 256; i--)
       {
-        UInt32 x = g_CrcTable[i - 256];
+        UInt32 x = g_CrcTable[(size_t)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;
+      g_CrcUpdate = CrcUpdateT1_BeT8;
       #endif
     }
   }

C/7zCrcOpt.c

@@ -1,5 +1,5 @@
 /* 7zCrcOpt.c -- CRC32 calculation
-2015-03-01 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -18,10 +18,10 @@ UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const U
   {
     v ^= *(const UInt32 *)p;
     v =
-          table[0x300 + ((v      ) & 0xFF)]
+          (table + 0x300)[((v      ) & 0xFF)]
-        ^ table[0x200 + ((v >>  8) & 0xFF)]
+        ^ (table + 0x200)[((v >>  8) & 0xFF)]
-        ^ table[0x100 + ((v >> 16) & 0xFF)]
+        ^ (table + 0x100)[((v >> 16) & 0xFF)]
-        ^ table[0x000 + ((v >> 24))];
+        ^ (table + 0x000)[((v >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2(v, *p);
@@ -38,16 +38,16 @@ UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const U
     UInt32 d;
     v ^= *(const UInt32 *)p;
     v =
-          table[0x700 + ((v      ) & 0xFF)]
+          (table + 0x700)[((v      ) & 0xFF)]
-        ^ table[0x600 + ((v >>  8) & 0xFF)]
+        ^ (table + 0x600)[((v >>  8) & 0xFF)]
-        ^ table[0x500 + ((v >> 16) & 0xFF)]
+        ^ (table + 0x500)[((v >> 16) & 0xFF)]
-        ^ table[0x400 + ((v >> 24))];
+        ^ (table + 0x400)[((v >> 24))];
     d = *((const UInt32 *)p + 1);
     v ^=
-          table[0x300 + ((d      ) & 0xFF)]
+          (table + 0x300)[((d      ) & 0xFF)]
-        ^ table[0x200 + ((d >>  8) & 0xFF)]
+        ^ (table + 0x200)[((d >>  8) & 0xFF)]
-        ^ table[0x100 + ((d >> 16) & 0xFF)]
+        ^ (table + 0x100)[((d >> 16) & 0xFF)]
-        ^ table[0x000 + ((d >> 24))];
+        ^ (table + 0x000)[((d >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2(v, *p);
@@ -74,10 +74,10 @@ UInt32 MY_FAST_CALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, co
   {
     v ^= *(const UInt32 *)p;
     v =
-          table[0x000 + ((v      ) & 0xFF)]
+          (table + 0x000)[((v      ) & 0xFF)]
-        ^ table[0x100 + ((v >>  8) & 0xFF)]
+        ^ (table + 0x100)[((v >>  8) & 0xFF)]
-        ^ table[0x200 + ((v >> 16) & 0xFF)]
+        ^ (table + 0x200)[((v >> 16) & 0xFF)]
-        ^ table[0x300 + ((v >> 24))];
+        ^ (table + 0x300)[((v >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2_BE(v, *p);
@@ -96,16 +96,16 @@ UInt32 MY_FAST_CALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, co
     UInt32 d;
     v ^= *(const UInt32 *)p;
     v =
-          table[0x400 + ((v      ) & 0xFF)]
+          (table + 0x400)[((v      ) & 0xFF)]
-        ^ table[0x500 + ((v >>  8) & 0xFF)]
+        ^ (table + 0x500)[((v >>  8) & 0xFF)]
-        ^ table[0x600 + ((v >> 16) & 0xFF)]
+        ^ (table + 0x600)[((v >> 16) & 0xFF)]
-        ^ table[0x700 + ((v >> 24))];
+        ^ (table + 0x700)[((v >> 24))];
     d = *((const UInt32 *)p + 1);
     v ^=
-          table[0x000 + ((d      ) & 0xFF)]
+          (table + 0x000)[((d      ) & 0xFF)]
-        ^ table[0x100 + ((d >>  8) & 0xFF)]
+        ^ (table + 0x100)[((d >>  8) & 0xFF)]
-        ^ table[0x200 + ((d >> 16) & 0xFF)]
+        ^ (table + 0x200)[((d >> 16) & 0xFF)]
-        ^ table[0x300 + ((d >> 24))];
+        ^ (table + 0x300)[((d >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2_BE(v, *p);

C/7zDec.c

@@ -1,5 +1,5 @@
 /* 7zDec.c -- Decoding from 7z folder
-2015-08-01 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -8,6 +8,7 @@
 /* #define _7ZIP_PPMD_SUPPPORT */
 
 #include "7z.h"
+#include "7zCrc.h"
 
 #include "Bcj2.h"
 #include "Bra.h"
@@ -38,28 +39,28 @@
 
 typedef struct
 {
-  IByteIn p;
+  IByteIn vt;
   const Byte *cur;
   const Byte *end;
   const Byte *begin;
   UInt64 processed;
   Bool extra;
   SRes res;
-  ILookInStream *inStream;
+  const ILookInStream *inStream;
 } CByteInToLook;
 
-static Byte ReadByte(void *pp)
+static Byte ReadByte(const IByteIn *pp)
 {
-  CByteInToLook *p = (CByteInToLook *)pp;
+  CByteInToLook *p = CONTAINER_FROM_VTBL(pp, CByteInToLook, vt);
   if (p->cur != p->end)
     return *p->cur++;
   if (p->res == SZ_OK)
   {
     size_t size = p->cur - p->begin;
     p->processed += size;
-    p->res = p->inStream->Skip(p->inStream, size);
+    p->res = ILookInStream_Skip(p->inStream, size);
     size = (1 << 25);
-    p->res = p->inStream->Look(p->inStream, (const void **)&p->begin, &size);
+    p->res = ILookInStream_Look(p->inStream, (const void **)&p->begin, &size);
     p->cur = p->begin;
     p->end = p->begin + size;
     if (size != 0)
@@ -69,14 +70,14 @@ static Byte ReadByte(void *pp)
   return 0;
 }
 
-static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStream *inStream,
+static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, const ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CPpmd7 ppmd;
   CByteInToLook s;
   SRes res = SZ_OK;
 
-  s.p.Read = ReadByte;
+  s.vt.Read = ReadByte;
   s.inStream = inStream;
   s.begin = s.end = s.cur = NULL;
   s.extra = False;
@@ -102,7 +103,7 @@ static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, I
   {
     CPpmd7z_RangeDec rc;
     Ppmd7z_RangeDec_CreateVTable(&rc);
-    rc.Stream = &s.p;
+    rc.Stream = &s.vt;
     if (!Ppmd7z_RangeDec_Init(&rc))
       res = SZ_ERROR_DATA;
     else if (s.extra)
@@ -112,7 +113,7 @@ static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, I
       SizeT i;
       for (i = 0; i < outSize; i++)
       {
-        int sym = Ppmd7_DecodeSymbol(&ppmd, &rc.p);
+        int sym = Ppmd7_DecodeSymbol(&ppmd, &rc.vt);
         if (s.extra || sym < 0)
           break;
         outBuffer[i] = (Byte)sym;
@@ -131,7 +132,7 @@ static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, I
 
 
 static SRes SzDecodeLzma(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CLzmaDec state;
   SRes res = SZ_OK;
@@ -148,7 +149,7 @@ static SRes SzDecodeLzma(const Byte *props, unsigned propsSize, UInt64 inSize, I
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look(inStream, &inBuf, &lookahead);
+    res = ILookInStream_Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
@@ -160,15 +161,24 @@ static SRes SzDecodeLzma(const Byte *props, unsigned propsSize, UInt64 inSize, I
       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 ||
+        if (outSize != state.dicPos || inSize != 0)
-            (status != LZMA_STATUS_FINISHED_WITH_MARK &&
-             status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK))
           res = SZ_ERROR_DATA;
         break;
       }
-      res = inStream->Skip((void *)inStream, inProcessed);
+
+      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 = ILookInStream_Skip(inStream, inProcessed);
       if (res != SZ_OK)
         break;
     }
@@ -182,7 +192,7 @@ static SRes SzDecodeLzma(const Byte *props, unsigned propsSize, UInt64 inSize, I
 #ifndef _7Z_NO_METHOD_LZMA2
 
 static SRes SzDecodeLzma2(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CLzma2Dec state;
   SRes res = SZ_OK;
@@ -201,7 +211,7 @@ static SRes SzDecodeLzma2(const Byte *props, unsigned propsSize, UInt64 inSize,
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look(inStream, &inBuf, &lookahead);
+    res = ILookInStream_Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
@@ -213,14 +223,21 @@ static SRes SzDecodeLzma2(const Byte *props, unsigned propsSize, UInt64 inSize,
       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 ||
+        if (outSize != state.decoder.dicPos || inSize != 0)
-            (status != LZMA_STATUS_FINISHED_WITH_MARK))
           res = SZ_ERROR_DATA;
         break;
       }
-      res = inStream->Skip((void *)inStream, inProcessed);
+
+      if (inProcessed == 0 && dicPos == state.decoder.dicPos)
+      {
+        res = SZ_ERROR_DATA;
+        break;
+      }
+
+      res = ILookInStream_Skip(inStream, inProcessed);
       if (res != SZ_OK)
         break;
     }
@@ -241,13 +258,13 @@ static SRes SzDecodeCopy(UInt64 inSize, ILookInStream *inStream, Byte *outBuffer
     size_t curSize = (1 << 18);
     if (curSize > inSize)
       curSize = (size_t)inSize;
-    RINOK(inStream->Look(inStream, &inBuf, &curSize));
+    RINOK(ILookInStream_Look(inStream, &inBuf, &curSize));
     if (curSize == 0)
       return SZ_ERROR_INPUT_EOF;
     memcpy(outBuffer, inBuf, curSize);
     outBuffer += curSize;
     inSize -= curSize;
-    RINOK(inStream->Skip((void *)inStream, curSize));
+    RINOK(ILookInStream_Skip(inStream, curSize));
   }
   return SZ_OK;
 }
@@ -355,7 +372,7 @@ static SRes SzFolder_Decode2(const CSzFolder *folder,
     const UInt64 *unpackSizes,
     const UInt64 *packPositions,
     ILookInStream *inStream, UInt64 startPos,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain,
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain,
     Byte *tempBuf[])
 {
   UInt32 ci;
@@ -387,7 +404,7 @@ static SRes SzFolder_Decode2(const CSzFolder *folder,
           outSizeCur = (SizeT)unpackSize;
           if (outSizeCur != unpackSize)
             return SZ_ERROR_MEM;
-          temp = (Byte *)IAlloc_Alloc(allocMain, outSizeCur);
+          temp = (Byte *)ISzAlloc_Alloc(allocMain, outSizeCur);
           if (!temp && outSizeCur != 0)
             return SZ_ERROR_MEM;
           outBufCur = tempBuf[1 - ci] = temp;
@@ -404,7 +421,7 @@ static SRes SzFolder_Decode2(const CSzFolder *folder,
           return SZ_ERROR_UNSUPPORTED;
       }
       offset = packPositions[si];
-      inSize = packPositions[si + 1] - offset;
+      inSize = packPositions[(size_t)si + 1] - offset;
       RINOK(LookInStream_SeekTo(inStream, startPos + offset));
 
       if (coder->MethodID == k_Copy)
@@ -443,7 +460,7 @@ static SRes SzFolder_Decode2(const CSzFolder *folder,
       tempSizes[2] = (SizeT)s3Size;
       if (tempSizes[2] != s3Size)
         return SZ_ERROR_MEM;
-      tempBuf[2] = (Byte *)IAlloc_Alloc(allocMain, tempSizes[2]);
+      tempBuf[2] = (Byte *)ISzAlloc_Alloc(allocMain, tempSizes[2]);
       if (!tempBuf[2] && tempSizes[2] != 0)
         return SZ_ERROR_MEM;
       
@@ -532,38 +549,43 @@ static SRes SzFolder_Decode2(const CSzFolder *folder,
 SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
     ILookInStream *inStream, UInt64 startPos,
     Byte *outBuffer, size_t outSize,
-    ISzAlloc *allocMain)
+    ISzAllocPtr allocMain)
 {
   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];
+  sd.Size = p->FoCodersOffsets[(size_t)folderIndex + 1] - p->FoCodersOffsets[folderIndex];
   
-  sdSizes.Data = p->UnpackSizesData + p->FoSizesOffsets[folderIndex];
+  res = SzGetNextFolderItem(&folder, &sd);
-  sdSizes.Size =
-      p->FoSizesOffsets[folderIndex + 1] -
-      p->FoSizesOffsets[folderIndex];
-
-  res = SzGetNextFolderItem(&folder, &sd, &sdSizes);
   
   if (res != SZ_OK)
     return res;
 
-  if (sd.Size != 0 || outSize != folder.CodersUnpackSizes[folder.UnpackStream])
+  if (sd.Size != 0
+      || folder.UnpackStream != p->FoToMainUnpackSizeIndex[folderIndex]
+      || outSize != SzAr_GetFolderUnpackSize(p, folderIndex))
     return SZ_ERROR_FAIL;
   {
     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]);
+      ISzAlloc_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;
   }
 }

C/7zFile.c

@@ -1,5 +1,5 @@
 /* 7zFile.c -- File IO
-2009-11-24 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -238,49 +238,49 @@ WRes File_GetLength(CSzFile *p, UInt64 *length)
 
 /* ---------- FileSeqInStream ---------- */
 
-static SRes FileSeqInStream_Read(void *pp, void *buf, size_t *size)
+static SRes FileSeqInStream_Read(const ISeqInStream *pp, void *buf, size_t *size)
 {
-  CFileSeqInStream *p = (CFileSeqInStream *)pp;
+  CFileSeqInStream *p = CONTAINER_FROM_VTBL(pp, CFileSeqInStream, vt);
   return File_Read(&p->file, buf, size) == 0 ? SZ_OK : SZ_ERROR_READ;
 }
 
 void FileSeqInStream_CreateVTable(CFileSeqInStream *p)
 {
-  p->s.Read = FileSeqInStream_Read;
+  p->vt.Read = FileSeqInStream_Read;
 }
 
 
 /* ---------- FileInStream ---------- */
 
-static SRes FileInStream_Read(void *pp, void *buf, size_t *size)
+static SRes FileInStream_Read(const ISeekInStream *pp, void *buf, size_t *size)
 {
-  CFileInStream *p = (CFileInStream *)pp;
+  CFileInStream *p = CONTAINER_FROM_VTBL(pp, CFileInStream, vt);
   return (File_Read(&p->file, buf, size) == 0) ? SZ_OK : SZ_ERROR_READ;
 }
 
-static SRes FileInStream_Seek(void *pp, Int64 *pos, ESzSeek origin)
+static SRes FileInStream_Seek(const ISeekInStream *pp, Int64 *pos, ESzSeek origin)
 {
-  CFileInStream *p = (CFileInStream *)pp;
+  CFileInStream *p = CONTAINER_FROM_VTBL(pp, CFileInStream, vt);
   return File_Seek(&p->file, pos, origin);
 }
 
 void FileInStream_CreateVTable(CFileInStream *p)
 {
-  p->s.Read = FileInStream_Read;
+  p->vt.Read = FileInStream_Read;
-  p->s.Seek = FileInStream_Seek;
+  p->vt.Seek = FileInStream_Seek;
 }
 
 
 /* ---------- FileOutStream ---------- */
 
-static size_t FileOutStream_Write(void *pp, const void *data, size_t size)
+static size_t FileOutStream_Write(const ISeqOutStream *pp, const void *data, size_t size)
 {
-  CFileOutStream *p = (CFileOutStream *)pp;
+  CFileOutStream *p = CONTAINER_FROM_VTBL(pp, CFileOutStream, vt);
   File_Write(&p->file, data, &size);
   return size;
 }
 
 void FileOutStream_CreateVTable(CFileOutStream *p)
 {
-  p->s.Write = FileOutStream_Write;
+  p->vt.Write = FileOutStream_Write;
 }

C/7zFile.h

@@ -1,5 +1,5 @@
 /* 7zFile.h -- File IO
-2013-01-18 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_FILE_H
 #define __7Z_FILE_H
@@ -54,7 +54,7 @@ WRes File_GetLength(CSzFile *p, UInt64 *length);
 
 typedef struct
 {
-  ISeqInStream s;
+  ISeqInStream vt;
   CSzFile file;
 } CFileSeqInStream;
 
@@ -63,7 +63,7 @@ void FileSeqInStream_CreateVTable(CFileSeqInStream *p);
 
 typedef struct
 {
-  ISeekInStream s;
+  ISeekInStream vt;
   CSzFile file;
 } CFileInStream;
 
@@ -72,7 +72,7 @@ void FileInStream_CreateVTable(CFileInStream *p);
 
 typedef struct
 {
-  ISeqOutStream s;
+  ISeqOutStream vt;
   CSzFile file;
 } CFileOutStream;
 

C/7zStream.c

@@ -1,5 +1,5 @@
 /* 7zStream.c -- 7z Stream functions
-2013-11-12 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -7,12 +7,12 @@
 
 #include "7zTypes.h"
 
-SRes SeqInStream_Read2(ISeqInStream *stream, void *buf, size_t size, SRes errorType)
+SRes SeqInStream_Read2(const ISeqInStream *stream, void *buf, size_t size, SRes errorType)
 {
   while (size != 0)
   {
     size_t processed = size;
-    RINOK(stream->Read(stream, buf, &processed));
+    RINOK(ISeqInStream_Read(stream, buf, &processed));
     if (processed == 0)
       return errorType;
     buf = (void *)((Byte *)buf + processed);
@@ -21,40 +21,42 @@ SRes SeqInStream_Read2(ISeqInStream *stream, void *buf, size_t size, SRes errorT
   return SZ_OK;
 }
 
-SRes SeqInStream_Read(ISeqInStream *stream, void *buf, size_t size)
+SRes SeqInStream_Read(const ISeqInStream *stream, void *buf, size_t size)
 {
   return SeqInStream_Read2(stream, buf, size, SZ_ERROR_INPUT_EOF);
 }
 
-SRes SeqInStream_ReadByte(ISeqInStream *stream, Byte *buf)
+SRes SeqInStream_ReadByte(const ISeqInStream *stream, Byte *buf)
 {
   size_t processed = 1;
-  RINOK(stream->Read(stream, buf, &processed));
+  RINOK(ISeqInStream_Read(stream, buf, &processed));
   return (processed == 1) ? SZ_OK : SZ_ERROR_INPUT_EOF;
 }
 
-SRes LookInStream_SeekTo(ILookInStream *stream, UInt64 offset)
+
+
+SRes LookInStream_SeekTo(const ILookInStream *stream, UInt64 offset)
 {
   Int64 t = offset;
-  return stream->Seek(stream, &t, SZ_SEEK_SET);
+  return ILookInStream_Seek(stream, &t, SZ_SEEK_SET);
 }
 
-SRes LookInStream_LookRead(ILookInStream *stream, void *buf, size_t *size)
+SRes LookInStream_LookRead(const ILookInStream *stream, void *buf, size_t *size)
 {
   const void *lookBuf;
   if (*size == 0)
     return SZ_OK;
-  RINOK(stream->Look(stream, &lookBuf, size));
+  RINOK(ILookInStream_Look(stream, &lookBuf, size));
   memcpy(buf, lookBuf, *size);
-  return stream->Skip(stream, *size);
+  return ILookInStream_Skip(stream, *size);
 }
 
-SRes LookInStream_Read2(ILookInStream *stream, void *buf, size_t size, SRes errorType)
+SRes LookInStream_Read2(const ILookInStream *stream, void *buf, size_t size, SRes errorType)
 {
   while (size != 0)
   {
     size_t processed = size;
-    RINOK(stream->Read(stream, buf, &processed));
+    RINOK(ILookInStream_Read(stream, buf, &processed));
     if (processed == 0)
       return errorType;
     buf = (void *)((Byte *)buf + processed);
@@ -63,61 +65,67 @@ SRes LookInStream_Read2(ILookInStream *stream, void *buf, size_t size, SRes erro
   return SZ_OK;
 }
 
-SRes LookInStream_Read(ILookInStream *stream, void *buf, size_t size)
+SRes LookInStream_Read(const ILookInStream *stream, void *buf, size_t size)
 {
   return LookInStream_Read2(stream, buf, size, SZ_ERROR_INPUT_EOF);
 }
 
-static SRes LookToRead_Look_Lookahead(void *pp, const void **buf, size_t *size)
+
+
+#define GET_LookToRead2 CLookToRead2 *p = CONTAINER_FROM_VTBL(pp, CLookToRead2, vt);
+
+static SRes LookToRead2_Look_Lookahead(const ILookInStream *pp, const void **buf, size_t *size)
 {
   SRes res = SZ_OK;
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t size2 = p->size - p->pos;
-  if (size2 == 0 && *size > 0)
+  if (size2 == 0 && *size != 0)
   {
     p->pos = 0;
-    size2 = LookToRead_BUF_SIZE;
+    p->size = 0;
-    res = p->realStream->Read(p->realStream, p->buf, &size2);
+    size2 = p->bufSize;
+    res = ISeekInStream_Read(p->realStream, p->buf, &size2);
     p->size = size2;
   }
-  if (size2 < *size)
+  if (*size > size2)
     *size = size2;
   *buf = p->buf + p->pos;
   return res;
 }
 
-static SRes LookToRead_Look_Exact(void *pp, const void **buf, size_t *size)
+static SRes LookToRead2_Look_Exact(const ILookInStream *pp, const void **buf, size_t *size)
 {
   SRes res = SZ_OK;
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t size2 = p->size - p->pos;
-  if (size2 == 0 && *size > 0)
+  if (size2 == 0 && *size != 0)
   {
     p->pos = 0;
-    if (*size > LookToRead_BUF_SIZE)
+    p->size = 0;
-      *size = LookToRead_BUF_SIZE;
+    if (*size > p->bufSize)
-    res = p->realStream->Read(p->realStream, p->buf, size);
+      *size = p->bufSize;
+    res = ISeekInStream_Read(p->realStream, p->buf, size);
     size2 = p->size = *size;
   }
-  if (size2 < *size)
+  if (*size > size2)
     *size = size2;
   *buf = p->buf + p->pos;
   return res;
 }
 
-static SRes LookToRead_Skip(void *pp, size_t offset)
+static SRes LookToRead2_Skip(const ILookInStream *pp, size_t offset)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   p->pos += offset;
   return SZ_OK;
 }
 
-static SRes LookToRead_Read(void *pp, void *buf, size_t *size)
+static SRes LookToRead2_Read(const ILookInStream *pp, void *buf, size_t *size)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t rem = p->size - p->pos;
   if (rem == 0)
-    return p->realStream->Read(p->realStream, buf, size);
+    return ISeekInStream_Read(p->realStream, buf, size);
   if (rem > *size)
     rem = *size;
   memcpy(buf, p->buf + p->pos, rem);
@@ -126,46 +134,43 @@ static SRes LookToRead_Read(void *pp, void *buf, size_t *size)
   return SZ_OK;
 }
 
-static SRes LookToRead_Seek(void *pp, Int64 *pos, ESzSeek origin)
+static SRes LookToRead2_Seek(const ILookInStream *pp, Int64 *pos, ESzSeek origin)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   p->pos = p->size = 0;
-  return p->realStream->Seek(p->realStream, pos, origin);
+  return ISeekInStream_Seek(p->realStream, pos, origin);
 }
 
-void LookToRead_CreateVTable(CLookToRead *p, int lookahead)
+void LookToRead2_CreateVTable(CLookToRead2 *p, int lookahead)
 {
-  p->s.Look = lookahead ?
+  p->vt.Look = lookahead ?
-      LookToRead_Look_Lookahead :
+      LookToRead2_Look_Lookahead :
-      LookToRead_Look_Exact;
+      LookToRead2_Look_Exact;
-  p->s.Skip = LookToRead_Skip;
+  p->vt.Skip = LookToRead2_Skip;
-  p->s.Read = LookToRead_Read;
+  p->vt.Read = LookToRead2_Read;
-  p->s.Seek = LookToRead_Seek;
+  p->vt.Seek = LookToRead2_Seek;
 }
 
-void LookToRead_Init(CLookToRead *p)
-{
-  p->pos = p->size = 0;
-}
 
-static SRes SecToLook_Read(void *pp, void *buf, size_t *size)
+
+static SRes SecToLook_Read(const ISeqInStream *pp, void *buf, size_t *size)
 {
-  CSecToLook *p = (CSecToLook *)pp;
+  CSecToLook *p = CONTAINER_FROM_VTBL(pp, CSecToLook, vt);
   return LookInStream_LookRead(p->realStream, buf, size);
 }
 
 void SecToLook_CreateVTable(CSecToLook *p)
 {
-  p->s.Read = SecToLook_Read;
+  p->vt.Read = SecToLook_Read;
 }
 
-static SRes SecToRead_Read(void *pp, void *buf, size_t *size)
+static SRes SecToRead_Read(const ISeqInStream *pp, void *buf, size_t *size)
 {
-  CSecToRead *p = (CSecToRead *)pp;
+  CSecToRead *p = CONTAINER_FROM_VTBL(pp, CSecToRead, vt);
-  return p->realStream->Read(p->realStream, buf, size);
+  return ILookInStream_Read(p->realStream, buf, size);
 }
 
 void SecToRead_CreateVTable(CSecToRead *p)
 {
-  p->s.Read = SecToRead_Read;
+  p->vt.Read = SecToRead_Read;
 }

C/7zTypes.h

@@ -1,5 +1,5 @@
 /* 7zTypes.h -- Basic types
-2013-11-12 : Igor Pavlov : Public domain */
+2017-07-17 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_TYPES_H
 #define __7Z_TYPES_H
@@ -42,13 +42,23 @@ EXTERN_C_BEGIN
 
 typedef int SRes;
 
+
 #ifdef _WIN32
+
 /* typedef DWORD WRes; */
 typedef unsigned WRes;
+#define MY_SRes_HRESULT_FROM_WRes(x) HRESULT_FROM_WIN32(x)
+
 #else
+
 typedef int WRes;
+#define MY__FACILITY_WIN32 7
+#define MY__FACILITY__WRes MY__FACILITY_WIN32
+#define MY_SRes_HRESULT_FROM_WRes(x) ((HRESULT)(x) <= 0 ? ((HRESULT)(x)) : ((HRESULT) (((x) & 0x0000FFFF) | (MY__FACILITY__WRes << 16) | 0x80000000)))
+
 #endif
 
+
 #ifndef RINOK
 #define RINOK(x) { int __result__ = (x); if (__result__ != 0) return __result__; }
 #endif
@@ -112,48 +122,72 @@ typedef int Bool;
 #define MY_NO_INLINE
 #endif
 
+#define MY_FORCE_INLINE __forceinline
+
 #define MY_CDECL __cdecl
 #define MY_FAST_CALL __fastcall
 
 #else
 
 #define MY_NO_INLINE
+#define MY_FORCE_INLINE
 #define MY_CDECL
 #define MY_FAST_CALL
 
+/* inline keyword : for C++ / C99 */
+
+/* GCC, clang: */
+/*
+#if defined (__GNUC__) && (__GNUC__ >= 4)
+#define MY_FORCE_INLINE __attribute__((always_inline))
+#define MY_NO_INLINE __attribute__((noinline))
+#endif
+*/
+
 #endif
 
 
 /* The following interfaces use first parameter as pointer to structure */
 
-typedef struct
+typedef struct IByteIn IByteIn;
+struct IByteIn
 {
-  Byte (*Read)(void *p); /* reads one byte, returns 0 in case of EOF or error */
+  Byte (*Read)(const IByteIn *p); /* reads one byte, returns 0 in case of EOF or error */
-} IByteIn;
+};
+#define IByteIn_Read(p) (p)->Read(p)
 
-typedef struct
-{
-  void (*Write)(void *p, Byte b);
-} IByteOut;
 
-typedef struct
+typedef struct IByteOut IByteOut;
+struct IByteOut
 {
-  SRes (*Read)(void *p, void *buf, size_t *size);
+  void (*Write)(const IByteOut *p, Byte b);
+};
+#define IByteOut_Write(p, b) (p)->Write(p, b)
+
+
+typedef struct ISeqInStream ISeqInStream;
+struct ISeqInStream
+{
+  SRes (*Read)(const ISeqInStream *p, void *buf, size_t *size);
     /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream.
        (output(*size) < input(*size)) is allowed */
-} ISeqInStream;
+};
+#define ISeqInStream_Read(p, buf, size) (p)->Read(p, buf, size)
 
 /* it can return SZ_ERROR_INPUT_EOF */
-SRes SeqInStream_Read(ISeqInStream *stream, void *buf, size_t size);
+SRes SeqInStream_Read(const ISeqInStream *stream, void *buf, size_t size);
-SRes SeqInStream_Read2(ISeqInStream *stream, void *buf, size_t size, SRes errorType);
+SRes SeqInStream_Read2(const ISeqInStream *stream, void *buf, size_t size, SRes errorType);
-SRes SeqInStream_ReadByte(ISeqInStream *stream, Byte *buf);
+SRes SeqInStream_ReadByte(const ISeqInStream *stream, Byte *buf);
 
-typedef struct
+
+typedef struct ISeqOutStream ISeqOutStream;
+struct ISeqOutStream
 {
-  size_t (*Write)(void *p, const void *buf, size_t size);
+  size_t (*Write)(const ISeqOutStream *p, const void *buf, size_t size);
     /* Returns: result - the number of actually written bytes.
        (result < size) means error */
-} ISeqOutStream;
+};
+#define ISeqOutStream_Write(p, buf, size) (p)->Write(p, buf, size)
 
 typedef enum
 {
@@ -162,78 +196,162 @@ typedef enum
   SZ_SEEK_END = 2
 } ESzSeek;
 
-typedef struct
-{
-  SRes (*Read)(void *p, void *buf, size_t *size);  /* same as ISeqInStream::Read */
-  SRes (*Seek)(void *p, Int64 *pos, ESzSeek origin);
-} ISeekInStream;
 
-typedef struct
+typedef struct ISeekInStream ISeekInStream;
+struct ISeekInStream
 {
-  SRes (*Look)(void *p, const void **buf, size_t *size);
+  SRes (*Read)(const ISeekInStream *p, void *buf, size_t *size);  /* same as ISeqInStream::Read */
+  SRes (*Seek)(const ISeekInStream *p, Int64 *pos, ESzSeek origin);
+};
+#define ISeekInStream_Read(p, buf, size)   (p)->Read(p, buf, size)
+#define ISeekInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin)
+
+
+typedef struct ILookInStream ILookInStream;
+struct ILookInStream
+{
+  SRes (*Look)(const ILookInStream *p, const void **buf, size_t *size);
     /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream.
        (output(*size) > input(*size)) is not allowed
        (output(*size) < input(*size)) is allowed */
-  SRes (*Skip)(void *p, size_t offset);
+  SRes (*Skip)(const ILookInStream *p, size_t offset);
     /* offset must be <= output(*size) of Look */
 
-  SRes (*Read)(void *p, void *buf, size_t *size);
+  SRes (*Read)(const ILookInStream *p, void *buf, size_t *size);
     /* reads directly (without buffer). It's same as ISeqInStream::Read */
-  SRes (*Seek)(void *p, Int64 *pos, ESzSeek origin);
+  SRes (*Seek)(const ILookInStream *p, Int64 *pos, ESzSeek origin);
-} ILookInStream;
+};
 
-SRes LookInStream_LookRead(ILookInStream *stream, void *buf, size_t *size);
+#define ILookInStream_Look(p, buf, size)   (p)->Look(p, buf, size)
-SRes LookInStream_SeekTo(ILookInStream *stream, UInt64 offset);
+#define ILookInStream_Skip(p, offset)      (p)->Skip(p, offset)
+#define ILookInStream_Read(p, buf, size)   (p)->Read(p, buf, size)
+#define ILookInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin)
+
+
+SRes LookInStream_LookRead(const ILookInStream *stream, void *buf, size_t *size);
+SRes LookInStream_SeekTo(const ILookInStream *stream, UInt64 offset);
 
 /* reads via ILookInStream::Read */
-SRes LookInStream_Read2(ILookInStream *stream, void *buf, size_t size, SRes errorType);
+SRes LookInStream_Read2(const ILookInStream *stream, void *buf, size_t size, SRes errorType);
-SRes LookInStream_Read(ILookInStream *stream, void *buf, size_t size);
+SRes LookInStream_Read(const ILookInStream *stream, void *buf, size_t size);
+
 
-#define LookToRead_BUF_SIZE (1 << 14)
 
 typedef struct
 {
-  ILookInStream s;
+  ILookInStream vt;
-  ISeekInStream *realStream;
+  const ISeekInStream *realStream;
+ 
   size_t pos;
-  size_t size;
+  size_t size; /* it's data size */
-  Byte buf[LookToRead_BUF_SIZE];
+  
-} CLookToRead;
+  /* the following variables must be set outside */
+  Byte *buf;
+  size_t bufSize;
+} CLookToRead2;
+
+void LookToRead2_CreateVTable(CLookToRead2 *p, int lookahead);
+
+#define LookToRead2_Init(p) { (p)->pos = (p)->size = 0; }
 
-void LookToRead_CreateVTable(CLookToRead *p, int lookahead);
-void LookToRead_Init(CLookToRead *p);
 
 typedef struct
 {
-  ISeqInStream s;
+  ISeqInStream vt;
-  ILookInStream *realStream;
+  const ILookInStream *realStream;
 } CSecToLook;
 
 void SecToLook_CreateVTable(CSecToLook *p);
 
+
+
 typedef struct
 {
-  ISeqInStream s;
+  ISeqInStream vt;
-  ILookInStream *realStream;
+  const ILookInStream *realStream;
 } CSecToRead;
 
 void SecToRead_CreateVTable(CSecToRead *p);
 
-typedef struct
+
+typedef struct ICompressProgress ICompressProgress;
+
+struct ICompressProgress
 {
-  SRes (*Progress)(void *p, UInt64 inSize, UInt64 outSize);
+  SRes (*Progress)(const ICompressProgress *p, UInt64 inSize, UInt64 outSize);
     /* Returns: result. (result != SZ_OK) means break.
        Value (UInt64)(Int64)-1 for size means unknown value. */
-} ICompressProgress;
+};
+#define ICompressProgress_Progress(p, inSize, outSize) (p)->Progress(p, inSize, outSize)
 
-typedef struct
+
+
+typedef struct ISzAlloc ISzAlloc;
+typedef const ISzAlloc * ISzAllocPtr;
+
+struct ISzAlloc
 {
-  void *(*Alloc)(void *p, size_t size);
+  void *(*Alloc)(ISzAllocPtr p, size_t size);
-  void (*Free)(void *p, void *address); /* address can be 0 */
+  void (*Free)(ISzAllocPtr p, void *address); /* address can be 0 */
-} ISzAlloc;
+};
+
+#define ISzAlloc_Alloc(p, size) (p)->Alloc(p, size)
+#define ISzAlloc_Free(p, a) (p)->Free(p, a)
+
+/* deprecated */
+#define IAlloc_Alloc(p, size) ISzAlloc_Alloc(p, size)
+#define IAlloc_Free(p, a) ISzAlloc_Free(p, a)
+
+
+
+
+
+#ifndef MY_offsetof
+  #ifdef offsetof
+    #define MY_offsetof(type, m) offsetof(type, m)
+    /*
+    #define MY_offsetof(type, m) FIELD_OFFSET(type, m)
+    */
+  #else
+    #define MY_offsetof(type, m) ((size_t)&(((type *)0)->m))
+  #endif
+#endif
+
+
+
+#ifndef MY_container_of
+
+/*
+#define MY_container_of(ptr, type, m) container_of(ptr, type, m)
+#define MY_container_of(ptr, type, m) CONTAINING_RECORD(ptr, type, m)
+#define MY_container_of(ptr, type, m) ((type *)((char *)(ptr) - offsetof(type, m)))
+#define MY_container_of(ptr, type, m) (&((type *)0)->m == (ptr), ((type *)(((char *)(ptr)) - MY_offsetof(type, m))))
+*/
+
+/*
+  GCC shows warning: "perhaps the 'offsetof' macro was used incorrectly"
+    GCC 3.4.4 : classes with constructor
+    GCC 4.8.1 : classes with non-public variable members"
+*/
+
+#define MY_container_of(ptr, type, m) ((type *)((char *)(1 ? (ptr) : &((type *)0)->m) - MY_offsetof(type, m)))
+
+
+#endif
+
+#define CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m) ((type *)(ptr))
+
+/*
+#define CONTAINER_FROM_VTBL(ptr, type, m) CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m)
+*/
+#define CONTAINER_FROM_VTBL(ptr, type, m) MY_container_of(ptr, type, m)
+
+#define CONTAINER_FROM_VTBL_CLS(ptr, type, m) CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m)
+/*
+#define CONTAINER_FROM_VTBL_CLS(ptr, type, m) CONTAINER_FROM_VTBL(ptr, type, m)
+*/
+
 
-#define IAlloc_Alloc(p, size) (p)->Alloc((p), size)
-#define IAlloc_Free(p, a) (p)->Free((p), a)
 
 #ifdef _WIN32
 

C/7zVersion.h

@@ -1,14 +1,21 @@
-#define MY_VER_MAJOR 15
+#define MY_VER_MAJOR 18
-#define MY_VER_MINOR 06
+#define MY_VER_MINOR 05
-#define MY_VER_BUILD 00
+#define MY_VER_BUILD 0
-#define MY_VERSION_NUMBERS "15.06"
+#define MY_VERSION_NUMBERS "18.05"
-#define MY_VERSION "15.06 beta"
+#define MY_VERSION MY_VERSION_NUMBERS
-#define MY_DATE "2015-08-09"
+
+#ifdef MY_CPU_NAME
+  #define MY_VERSION_CPU MY_VERSION " (" MY_CPU_NAME ")"
+#else
+  #define MY_VERSION_CPU MY_VERSION
+#endif
+
+#define MY_DATE "2018-04-30"
 #undef MY_COPYRIGHT
 #undef MY_VERSION_COPYRIGHT_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"
+#define MY_COPYRIGHT_CR "Copyright (c) 1999-2018 Igor Pavlov"
 
 #ifdef USE_COPYRIGHT_CR
   #define MY_COPYRIGHT MY_COPYRIGHT_CR
@@ -16,4 +23,5 @@
   #define MY_COPYRIGHT MY_COPYRIGHT_PD
 #endif
 
-#define MY_VERSION_COPYRIGHT_DATE MY_VERSION " : " MY_COPYRIGHT " : " MY_DATE
+#define MY_COPYRIGHT_DATE MY_COPYRIGHT " : " MY_DATE
+#define MY_VERSION_COPYRIGHT_DATE MY_VERSION_CPU " : " MY_COPYRIGHT " : " MY_DATE

C/Aes.c

@@ -1,5 +1,5 @@
 /* Aes.c -- AES encryption / decryption
-2015-02-23 : Igor Pavlov : Public domain */
+2017-01-24 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -49,7 +49,13 @@ static const Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0
 #define gb0(x) ( (x)          & 0xFF)
 #define gb1(x) (((x) >> ( 8)) & 0xFF)
 #define gb2(x) (((x) >> (16)) & 0xFF)
-#define gb3(x) (((x) >> (24)) & 0xFF)
+#define gb3(x) (((x) >> (24)))
+
+#define gb(n, x) gb ## n(x)
+
+#define TT(x) (T + (x << 8))
+#define DD(x) (D + (x << 8))
+
 
 void AesGenTables(void)
 {
@@ -63,10 +69,10 @@ void AesGenTables(void)
       UInt32 a1 = Sbox[i];
       UInt32 a2 = xtime(a1);
       UInt32 a3 = a2 ^ a1;
-      T[        i] = Ui32(a2, a1, a1, a3);
+      TT(0)[i] = Ui32(a2, a1, a1, a3);
-      T[0x100 + i] = Ui32(a3, a2, a1, a1);
+      TT(1)[i] = Ui32(a3, a2, a1, a1);
-      T[0x200 + i] = Ui32(a1, a3, a2, a1);
+      TT(2)[i] = Ui32(a1, a3, a2, a1);
-      T[0x300 + i] = Ui32(a1, a1, a3, a2);
+      TT(3)[i] = Ui32(a1, a1, a3, a2);
     }
     {
       UInt32 a1 = InvS[i];
@@ -77,10 +83,10 @@ void AesGenTables(void)
       UInt32 aB = a8 ^ a2 ^ a1;
       UInt32 aD = a8 ^ a4 ^ a1;
       UInt32 aE = a8 ^ a4 ^ a2;
-      D[        i] = Ui32(aE, a9, aD, aB);
+      DD(0)[i] = Ui32(aE, a9, aD, aB);
-      D[0x100 + i] = Ui32(aB, aE, a9, aD);
+      DD(1)[i] = Ui32(aB, aE, a9, aD);
-      D[0x200 + i] = Ui32(aD, aB, aE, a9);
+      DD(2)[i] = Ui32(aD, aB, aE, a9);
-      D[0x300 + i] = Ui32(a9, aD, aB, aE);
+      DD(3)[i] = Ui32(a9, aD, aB, aE);
     }
   }
   
@@ -99,7 +105,7 @@ void AesGenTables(void)
 }
 
 
-#define HT(i, x, s) (T + (x << 8))[gb ## x(s[(i + x) & 3])]
+#define HT(i, x, s) TT(x)[gb(x, s[(i + x) & 3])]
 
 #define HT4(m, i, s, p) m[i] = \
     HT(i, 0, s) ^ \
@@ -113,11 +119,11 @@ void AesGenTables(void)
     HT4(m, 2, s, p); \
     HT4(m, 3, s, p); \
 
-#define FT(i, x) Sbox[gb ## x(m[(i + x) & 3])]
+#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 HD(i, x, s) DD(x)[gb(x, s[(i - x) & 3])]
 
 #define HD4(m, i, s, p) m[i] = \
     HD(i, 0, s) ^ \
@@ -131,7 +137,7 @@ void AesGenTables(void)
     HD4(m, 2, s, p); \
     HD4(m, 3, s, p); \
 
-#define FD(i, x) InvS[gb ## x(m[(i - x) & 3])]
+#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];
 
 void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
@@ -147,7 +153,7 @@ void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
 
   for (; i < wSize; i++)
   {
-    UInt32 t = w[i - 1];
+    UInt32 t = w[(size_t)i - 1];
     unsigned rem = i % keySize;
     if (rem == 0)
       t = Ui32(Sbox[gb1(t)] ^ Rcon[i / keySize], Sbox[gb2(t)], Sbox[gb3(t)], Sbox[gb0(t)]);
@@ -167,10 +173,10 @@ void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
   {
     UInt32 r = w[i];
     w[i] =
-      D[        Sbox[gb0(r)]] ^
+      DD(0)[Sbox[gb0(r)]] ^
-      D[0x100 + Sbox[gb1(r)]] ^
+      DD(1)[Sbox[gb1(r)]] ^
-      D[0x200 + Sbox[gb2(r)]] ^
+      DD(2)[Sbox[gb2(r)]] ^
-      D[0x300 + Sbox[gb3(r)]];
+      DD(3)[Sbox[gb3(r)]];
   }
 }
 
@@ -276,20 +282,25 @@ void MY_FAST_CALL AesCtr_Code(UInt32 *p, Byte *data, size_t numBlocks)
   for (; numBlocks != 0; numBlocks--)
   {
     UInt32 temp[4];
-    Byte buf[16];
+    unsigned i;
-    int i;
 
     if (++p[0] == 0)
       p[1]++;
     
     Aes_Encode(p + 4, temp, p);
     
-    SetUi32(buf,      temp[0]);
+    for (i = 0; i < 4; i++, data += 4)
-    SetUi32(buf + 4,  temp[1]);
+    {
-    SetUi32(buf + 8,  temp[2]);
+      UInt32 t = temp[i];
-    SetUi32(buf + 12, temp[3]);
+
-    
+      #ifdef MY_CPU_LE_UNALIGN
-    for (i = 0; i < 16; i++)
+        *((UInt32 *)data) ^= t;
-      *data++ ^= buf[i];
+      #else
+        data[0] ^= (t & 0xFF);
+        data[1] ^= ((t >> 8) & 0xFF);
+        data[2] ^= ((t >> 16) & 0xFF);
+        data[3] ^= ((t >> 24));
+      #endif
+    }
   }
 }

C/AesOpt.c

@@ -1,12 +1,12 @@
 /* AesOpt.c -- Intel's AES
-2013-11-12 : Igor Pavlov : Public domain */
+2017-06-08 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
 #include "CpuArch.h"
 
 #ifdef MY_CPU_X86_OR_AMD64
-#if _MSC_VER >= 1500
+#if (_MSC_VER > 1500) || (_MSC_FULL_VER >= 150030729)
 #define USE_INTEL_AES
 #endif
 #endif

C/Alloc.c

@@ -1,8 +1,10 @@
 /* Alloc.c -- Memory allocation functions
-2015-02-21 : Igor Pavlov : Public domain */
+2018-04-27 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
+#include <stdio.h>
+
 #ifdef _WIN32
 #include <windows.h>
 #endif
@@ -14,20 +16,127 @@
 
 /* use _SZ_ALLOC_DEBUG to debug alloc/free operations */
 #ifdef _SZ_ALLOC_DEBUG
+
 #include <stdio.h>
 int g_allocCount = 0;
 int g_allocCountMid = 0;
 int g_allocCountBig = 0;
+
+
+#define CONVERT_INT_TO_STR(charType, tempSize) \
+  unsigned char temp[tempSize]; unsigned i = 0; \
+  while (val >= 10) { temp[i++] = (unsigned char)('0' + (unsigned)(val % 10)); val /= 10; } \
+  *s++ = (charType)('0' + (unsigned)val); \
+  while (i != 0) { i--; *s++ = temp[i]; } \
+  *s = 0;
+
+static void ConvertUInt64ToString(UInt64 val, char *s)
+{
+  CONVERT_INT_TO_STR(char, 24);
+}
+
+#define GET_HEX_CHAR(t) ((char)(((t < 10) ? ('0' + t) : ('A' + (t - 10)))))
+
+static void ConvertUInt64ToHex(UInt64 val, char *s)
+{
+  UInt64 v = val;
+  unsigned i;
+  for (i = 1;; i++)
+  {
+    v >>= 4;
+    if (v == 0)
+      break;
+  }
+  s[i] = 0;
+  do
+  {
+    unsigned t = (unsigned)(val & 0xF);
+    val >>= 4;
+    s[--i] = GET_HEX_CHAR(t);
+  }
+  while (i);
+}
+
+#define DEBUG_OUT_STREAM stderr
+
+static void Print(const char *s)
+{
+  fputs(s, DEBUG_OUT_STREAM);
+}
+
+static void PrintAligned(const char *s, size_t align)
+{
+  size_t len = strlen(s);
+  for(;;)
+  {
+    fputc(' ', DEBUG_OUT_STREAM);
+    if (len >= align)
+      break;
+    ++len;
+  }
+  Print(s);
+}
+
+static void PrintLn()
+{
+  Print("\n");
+}
+
+static void PrintHex(UInt64 v, size_t align)
+{
+  char s[32];
+  ConvertUInt64ToHex(v, s);
+  PrintAligned(s, align);
+}
+
+static void PrintDec(UInt64 v, size_t align)
+{
+  char s[32];
+  ConvertUInt64ToString(v, s);
+  PrintAligned(s, align);
+}
+
+static void PrintAddr(void *p)
+{
+  PrintHex((UInt64)(size_t)(ptrdiff_t)p, 12);
+}
+
+
+#define PRINT_ALLOC(name, cnt, size, ptr) \
+    Print(name " "); \
+    PrintDec(cnt++, 10); \
+    PrintHex(size, 10); \
+    PrintAddr(ptr); \
+    PrintLn();
+ 
+#define PRINT_FREE(name, cnt, ptr) if (ptr) { \
+    Print(name " "); \
+    PrintDec(--cnt, 10); \
+    PrintAddr(ptr); \
+    PrintLn(); }
+ 
+#else
+
+#define PRINT_ALLOC(name, cnt, size, ptr)
+#define PRINT_FREE(name, cnt, ptr)
+#define Print(s)
+#define PrintLn()
+#define PrintHex(v, align)
+#define PrintDec(v, align)
+#define PrintAddr(p)
+
 #endif
 
+
+
 void *MyAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
+    return NULL;
   #ifdef _SZ_ALLOC_DEBUG
   {
     void *p = malloc(size);
-    fprintf(stderr, "\nAlloc %10d bytes, count = %10d,  addr = %8X", size, g_allocCount++, (unsigned)p);
+    PRINT_ALLOC("Alloc    ", g_allocCount, size, p);
     return p;
   }
   #else
@@ -37,10 +146,8 @@ void *MyAlloc(size_t size)
 
 void MyFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
+  PRINT_FREE("Free    ", g_allocCount, address);
-  if (address != 0)
+  
-    fprintf(stderr, "\nFree; count = %10d,  addr = %8X", --g_allocCount, (unsigned)address);
-  #endif
   free(address);
 }
 
@@ -49,20 +156,18 @@ void MyFree(void *address)
 void *MidAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
+    return NULL;
-  #ifdef _SZ_ALLOC_DEBUG
+  
-  fprintf(stderr, "\nAlloc_Mid %10d bytes;  count = %10d", size, g_allocCountMid++);
+  PRINT_ALLOC("Alloc-Mid", g_allocCountMid, size, NULL);
-  #endif
+  
-  return VirtualAlloc(0, size, MEM_COMMIT, PAGE_READWRITE);
+  return VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE);
 }
 
 void MidFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
+  PRINT_FREE("Free-Mid", g_allocCountMid, address);
-  if (address != 0)
+
-    fprintf(stderr, "\nFree_Mid; count = %10d", --g_allocCountMid);
+  if (!address)
-  #endif
-  if (address == 0)
     return;
   VirtualFree(address, 0, MEM_RELEASE);
 }
@@ -79,10 +184,10 @@ typedef SIZE_T (WINAPI *GetLargePageMinimumP)();
 void SetLargePageSize()
 {
   #ifdef _7ZIP_LARGE_PAGES
-  SIZE_T size = 0;
+  SIZE_T size;
   GetLargePageMinimumP largePageMinimum = (GetLargePageMinimumP)
         GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")), "GetLargePageMinimum");
-  if (largePageMinimum == 0)
+  if (!largePageMinimum)
     return;
   size = largePageMinimum();
   if (size == 0 || (size & (size - 1)) != 0)
@@ -95,31 +200,36 @@ void SetLargePageSize()
 void *BigAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
+    return NULL;
-  #ifdef _SZ_ALLOC_DEBUG
+
-  fprintf(stderr, "\nAlloc_Big %10d bytes;  count = %10d", size, g_allocCountBig++);
+  PRINT_ALLOC("Alloc-Big", g_allocCountBig, size, NULL);
-  #endif
   
   #ifdef _7ZIP_LARGE_PAGES
-  if (g_LargePageSize != 0 && g_LargePageSize <= (1 << 30) && size >= (1 << 18))
   {
-    void *res = VirtualAlloc(0, (size + g_LargePageSize - 1) & (~(g_LargePageSize - 1)),
+    SIZE_T ps = g_LargePageSize;
-        MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE);
+    if (ps != 0 && ps <= (1 << 30) && size > (ps / 2))
-    if (res != 0)
+    {
-      return res;
+      size_t size2;
+      ps--;
+      size2 = (size + ps) & ~ps;
+      if (size2 >= size)
+      {
+        void *res = VirtualAlloc(NULL, size2, MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE);
+        if (res)
+          return res;
+      }
+    }
   }
   #endif
-  return VirtualAlloc(0, size, MEM_COMMIT, PAGE_READWRITE);
+
+  return VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE);
 }
 
 void BigFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
+  PRINT_FREE("Free-Big", g_allocCountBig, address);
-  if (address != 0)
-    fprintf(stderr, "\nFree_Big; count = %10d", --g_allocCountBig);
-  #endif
   
-  if (address == 0)
+  if (!address)
     return;
   VirtualFree(address, 0, MEM_RELEASE);
 }
@@ -127,10 +237,219 @@ void BigFree(void *address)
 #endif
 
 
-static void *SzAlloc(void *p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); }
+static void *SzAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); }
-static void SzFree(void *p, void *address) { UNUSED_VAR(p); MyFree(address); }
+static void SzFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); MyFree(address); }
-ISzAlloc g_Alloc = { SzAlloc, SzFree };
+const ISzAlloc g_Alloc = { SzAlloc, SzFree };
 
-static void *SzBigAlloc(void *p, size_t size) { UNUSED_VAR(p); return BigAlloc(size); }
+static void *SzMidAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return MidAlloc(size); }
-static void SzBigFree(void *p, void *address) { UNUSED_VAR(p); BigFree(address); }
+static void SzMidFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); MidFree(address); }
-ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree };
+const ISzAlloc g_MidAlloc = { SzMidAlloc, SzMidFree };
+
+static void *SzBigAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return BigAlloc(size); }
+static void SzBigFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); BigFree(address); }
+const ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree };
+
+
+/*
+  uintptr_t : <stdint.h> C99 (optional)
+            : unsupported in VS6
+*/
+
+#ifdef _WIN32
+  typedef UINT_PTR UIntPtr;
+#else
+  /*
+  typedef uintptr_t UIntPtr;
+  */
+  typedef ptrdiff_t UIntPtr;
+#endif
+
+
+#define ADJUST_ALLOC_SIZE 0
+/*
+#define ADJUST_ALLOC_SIZE (sizeof(void *) - 1)
+*/
+/*
+  Use (ADJUST_ALLOC_SIZE = (sizeof(void *) - 1)), if
+     MyAlloc() can return address that is NOT multiple of sizeof(void *).
+*/
+
+
+/*
+#define MY_ALIGN_PTR_DOWN(p, align) ((void *)((char *)(p) - ((size_t)(UIntPtr)(p) & ((align) - 1))))
+*/
+#define MY_ALIGN_PTR_DOWN(p, align) ((void *)((((UIntPtr)(p)) & ~((UIntPtr)(align) - 1))))
+
+#define MY_ALIGN_PTR_UP_PLUS(p, align) MY_ALIGN_PTR_DOWN(((char *)(p) + (align) + ADJUST_ALLOC_SIZE), align)
+
+
+#if (_POSIX_C_SOURCE >= 200112L) && !defined(_WIN32)
+  #define USE_posix_memalign
+#endif
+
+/*
+  This posix_memalign() is for test purposes only.
+  We also need special Free() function instead of free(),
+  if this posix_memalign() is used.
+*/
+
+/*
+static int posix_memalign(void **ptr, size_t align, size_t size)
+{
+  size_t newSize = size + align;
+  void *p;
+  void *pAligned;
+  *ptr = NULL;
+  if (newSize < size)
+    return 12; // ENOMEM
+  p = MyAlloc(newSize);
+  if (!p)
+    return 12; // ENOMEM
+  pAligned = MY_ALIGN_PTR_UP_PLUS(p, align);
+  ((void **)pAligned)[-1] = p;
+  *ptr = pAligned;
+  return 0;
+}
+*/
+
+/*
+  ALLOC_ALIGN_SIZE >= sizeof(void *)
+  ALLOC_ALIGN_SIZE >= cache_line_size
+*/
+
+#define ALLOC_ALIGN_SIZE ((size_t)1 << 7)
+
+static void *SzAlignedAlloc(ISzAllocPtr pp, size_t size)
+{
+  #ifndef USE_posix_memalign
+  
+  void *p;
+  void *pAligned;
+  size_t newSize;
+  UNUSED_VAR(pp);
+
+  /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned
+     block to prevent cache line sharing with another allocated blocks */
+
+  newSize = size + ALLOC_ALIGN_SIZE * 1 + ADJUST_ALLOC_SIZE;
+  if (newSize < size)
+    return NULL;
+
+  p = MyAlloc(newSize);
+  
+  if (!p)
+    return NULL;
+  pAligned = MY_ALIGN_PTR_UP_PLUS(p, ALLOC_ALIGN_SIZE);
+
+  Print(" size="); PrintHex(size, 8);
+  Print(" a_size="); PrintHex(newSize, 8);
+  Print(" ptr="); PrintAddr(p);
+  Print(" a_ptr="); PrintAddr(pAligned);
+  PrintLn();
+
+  ((void **)pAligned)[-1] = p;
+
+  return pAligned;
+
+  #else
+
+  void *p;
+  UNUSED_VAR(pp);
+  if (posix_memalign(&p, ALLOC_ALIGN_SIZE, size))
+    return NULL;
+
+  Print(" posix_memalign="); PrintAddr(p);
+  PrintLn();
+
+  return p;
+
+  #endif
+}
+
+
+static void SzAlignedFree(ISzAllocPtr pp, void *address)
+{
+  UNUSED_VAR(pp);
+  #ifndef USE_posix_memalign
+  if (address)
+    MyFree(((void **)address)[-1]);
+  #else
+  free(address);
+  #endif
+}
+
+
+const ISzAlloc g_AlignedAlloc = { SzAlignedAlloc, SzAlignedFree };
+
+
+
+#define MY_ALIGN_PTR_DOWN_1(p) MY_ALIGN_PTR_DOWN(p, sizeof(void *))
+
+/* we align ptr to support cases where CAlignOffsetAlloc::offset is not multiply of sizeof(void *) */
+#define REAL_BLOCK_PTR_VAR(p) ((void **)MY_ALIGN_PTR_DOWN_1(p))[-1]
+/*
+#define REAL_BLOCK_PTR_VAR(p) ((void **)(p))[-1]
+*/
+
+static void *AlignOffsetAlloc_Alloc(ISzAllocPtr pp, size_t size)
+{
+  CAlignOffsetAlloc *p = CONTAINER_FROM_VTBL(pp, CAlignOffsetAlloc, vt);
+  void *adr;
+  void *pAligned;
+  size_t newSize;
+  size_t extra;
+  size_t alignSize = (size_t)1 << p->numAlignBits;
+
+  if (alignSize < sizeof(void *))
+    alignSize = sizeof(void *);
+  
+  if (p->offset >= alignSize)
+    return NULL;
+
+  /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned
+     block to prevent cache line sharing with another allocated blocks */
+  extra = p->offset & (sizeof(void *) - 1);
+  newSize = size + alignSize + extra + ADJUST_ALLOC_SIZE;
+  if (newSize < size)
+    return NULL;
+
+  adr = ISzAlloc_Alloc(p->baseAlloc, newSize);
+  
+  if (!adr)
+    return NULL;
+
+  pAligned = (char *)MY_ALIGN_PTR_DOWN((char *)adr +
+      alignSize - p->offset + extra + ADJUST_ALLOC_SIZE, alignSize) + p->offset;
+
+  PrintLn();
+  Print("- Aligned: ");
+  Print(" size="); PrintHex(size, 8);
+  Print(" a_size="); PrintHex(newSize, 8);
+  Print(" ptr="); PrintAddr(adr);
+  Print(" a_ptr="); PrintAddr(pAligned);
+  PrintLn();
+
+  REAL_BLOCK_PTR_VAR(pAligned) = adr;
+
+  return pAligned;
+}
+
+
+static void AlignOffsetAlloc_Free(ISzAllocPtr pp, void *address)
+{
+  if (address)
+  {
+    CAlignOffsetAlloc *p = CONTAINER_FROM_VTBL(pp, CAlignOffsetAlloc, vt);
+    PrintLn();
+    Print("- Aligned Free: ");
+    PrintLn();
+    ISzAlloc_Free(p->baseAlloc, REAL_BLOCK_PTR_VAR(address));
+  }
+}
+
+
+void AlignOffsetAlloc_CreateVTable(CAlignOffsetAlloc *p)
+{
+  p->vt.Alloc = AlignOffsetAlloc_Alloc;
+  p->vt.Free = AlignOffsetAlloc_Free;
+}

C/Alloc.h

@@ -1,5 +1,5 @@
 /* Alloc.h -- Memory allocation functions
-2015-02-21 : Igor Pavlov : Public domain */
+2018-02-19 : Igor Pavlov : Public domain */
 
 #ifndef __COMMON_ALLOC_H
 #define __COMMON_ALLOC_H
@@ -29,8 +29,22 @@ void BigFree(void *address);
 
 #endif
 
-extern ISzAlloc g_Alloc;
+extern const ISzAlloc g_Alloc;
-extern ISzAlloc g_BigAlloc;
+extern const ISzAlloc g_BigAlloc;
+extern const ISzAlloc g_MidAlloc;
+extern const ISzAlloc g_AlignedAlloc;
+
+
+typedef struct
+{
+  ISzAlloc vt;
+  ISzAllocPtr baseAlloc;
+  unsigned numAlignBits; /* ((1 << numAlignBits) >= sizeof(void *)) */
+  size_t offset;         /* (offset == (k * sizeof(void *)) && offset < (1 << numAlignBits) */
+} CAlignOffsetAlloc;
+
+void AlignOffsetAlloc_CreateVTable(CAlignOffsetAlloc *p);
+
 
 EXTERN_C_END
 

C/Bcj2.c

@@ -1,5 +1,5 @@
 /* Bcj2.c -- BCJ2 Decoder (Converter for x86 code)
-2015-08-01 : Igor Pavlov : Public domain */
+2018-04-28 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -61,7 +61,8 @@ SRes Bcj2Dec_Decode(CBcj2Dec *p)
       Byte *dest = p->dest;
       if (dest == p->destLim)
         return SZ_OK;
-      *dest = p->temp[p->state++ - BCJ2_DEC_STATE_ORIG_0];
+      *dest = p->temp[(size_t)p->state - BCJ2_DEC_STATE_ORIG_0];
+      p->state++;
       p->dest = dest + 1;
     }
   }
@@ -231,10 +232,10 @@ SRes Bcj2Dec_Decode(CBcj2Dec *p)
       
       if (rem < 4)
       {
-        SizeT i;
+        p->temp[0] = (Byte)val; if (rem > 0) dest[0] = (Byte)val; val >>= 8;
-        SetUi32(p->temp, val);
+        p->temp[1] = (Byte)val; if (rem > 1) dest[1] = (Byte)val; val >>= 8;
-        for (i = 0; i < rem; i++)
+        p->temp[2] = (Byte)val; if (rem > 2) dest[2] = (Byte)val; val >>= 8;
-          dest[i] = p->temp[i];
+        p->temp[3] = (Byte)val;
         p->dest = dest + rem;
         p->state = BCJ2_DEC_STATE_ORIG_0 + (unsigned)rem;
         break;

C/Bcj2Enc.c

@@ -1,5 +1,5 @@
 /* Bcj2Enc.c -- BCJ2 Encoder (Converter for x86 code)
-2014-11-10 : Igor Pavlov : Public domain */
+2017-04-28 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -12,7 +12,6 @@
 #define PRF(x)
 #endif
 
-#include <windows.h>
 #include <string.h>
 
 #include "Bcj2.h"
@@ -271,7 +270,7 @@ void Bcj2Enc_Encode(CBcj2Enc *p)
         unsigned i;
         p->tempPos = tempPos;
         for (i = 0; i < tempPos; i++)
-          p->temp[i] = p->temp[i + num];
+          p->temp[i] = p->temp[(size_t)i + num];
       
         p->src = src;
         p->srcLim = srcLim;

C/Bra.c

@@ -1,135 +1,230 @@
 /* Bra.c -- Converters for RISC code
-2010-04-16 : Igor Pavlov : Public domain */
+2017-04-04 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
+#include "CpuArch.h"
 #include "Bra.h"
 
 SizeT ARM_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
 {
-  SizeT i;
+  Byte *p;
-  if (size < 4)
+  const Byte *lim;
-    return 0;
+  size &= ~(size_t)3;
-  size -= 4;
+  ip += 4;
-  ip += 8;
+  p = data;
-  for (i = 0; i <= size; i += 4)
+  lim = data + size;
+
+  if (encoding)
+
+  for (;;)
   {
-    if (data[i + 3] == 0xEB)
+    for (;;)
     {
-      UInt32 dest;
+      if (p >= lim)
-      UInt32 src = ((UInt32)data[i + 2] << 16) | ((UInt32)data[i + 1] << 8) | (data[i + 0]);
+        return p - data;
-      src <<= 2;
+      p += 4;
-      if (encoding)
+      if (p[-1] == 0xEB)
-        dest = ip + (UInt32)i + src;
+        break;
-      else
+    }
-        dest = src - (ip + (UInt32)i);
+    {
-      dest >>= 2;
+      UInt32 v = GetUi32(p - 4);
-      data[i + 2] = (Byte)(dest >> 16);
+      v <<= 2;
-      data[i + 1] = (Byte)(dest >> 8);
+        v += ip + (UInt32)(p - data);
-      data[i + 0] = (Byte)dest;
+      v >>= 2;
+      v &= 0x00FFFFFF;
+      v |= 0xEB000000;
+      SetUi32(p - 4, v);
+    }
+  }
+
+  for (;;)
+  {
+    for (;;)
+    {
+      if (p >= lim)
+        return p - data;
+      p += 4;
+      if (p[-1] == 0xEB)
+        break;
+    }
+    {
+      UInt32 v = GetUi32(p - 4);
+      v <<= 2;
+        v -= ip + (UInt32)(p - data);
+      v >>= 2;
+      v &= 0x00FFFFFF;
+      v |= 0xEB000000;
+      SetUi32(p - 4, v);
     }
   }
-  return i;
 }
 
+
 SizeT ARMT_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
 {
-  SizeT i;
+  Byte *p;
-  if (size < 4)
+  const Byte *lim;
-    return 0;
+  size &= ~(size_t)1;
-  size -= 4;
+  p = data;
-  ip += 4;
+  lim = data + size - 4;
-  for (i = 0; i <= size; i += 2)
+
+  if (encoding)
+  
+  for (;;)
   {
-    if ((data[i + 1] & 0xF8) == 0xF0 &&
+    UInt32 b1;
-        (data[i + 3] & 0xF8) == 0xF8)
+    for (;;)
     {
-      UInt32 dest;
+      UInt32 b3;
-      UInt32 src =
+      if (p > lim)
-        (((UInt32)data[i + 1] & 0x7) << 19) |
+        return p - data;
-        ((UInt32)data[i + 0] << 11) |
+      b1 = p[1];
-        (((UInt32)data[i + 3] & 0x7) << 8) |
+      b3 = p[3];
-        (data[i + 2]);
+      p += 2;
-      
+      b1 ^= 8;
-      src <<= 1;
+      if ((b3 & b1) >= 0xF8)
-      if (encoding)
+        break;
-        dest = ip + (UInt32)i + src;
+    }
-      else
+    {
-        dest = src - (ip + (UInt32)i);
+      UInt32 v =
-      dest >>= 1;
+             ((UInt32)b1 << 19)
-      
+          + (((UInt32)p[1] & 0x7) << 8)
-      data[i + 1] = (Byte)(0xF0 | ((dest >> 19) & 0x7));
+          + (((UInt32)p[-2] << 11))
-      data[i + 0] = (Byte)(dest >> 11);
+          + (p[0]);
-      data[i + 3] = (Byte)(0xF8 | ((dest >> 8) & 0x7));
+
-      data[i + 2] = (Byte)dest;
+      p += 2;
-      i += 2;
+      {
+        UInt32 cur = (ip + (UInt32)(p - data)) >> 1;
+          v += cur;
+      }
+
+      p[-4] = (Byte)(v >> 11);
+      p[-3] = (Byte)(0xF0 | ((v >> 19) & 0x7));
+      p[-2] = (Byte)v;
+      p[-1] = (Byte)(0xF8 | (v >> 8));
+    }
+  }
+  
+  for (;;)
+  {
+    UInt32 b1;
+    for (;;)
+    {
+      UInt32 b3;
+      if (p > lim)
+        return p - data;
+      b1 = p[1];
+      b3 = p[3];
+      p += 2;
+      b1 ^= 8;
+      if ((b3 & b1) >= 0xF8)
+        break;
+    }
+    {
+      UInt32 v =
+             ((UInt32)b1 << 19)
+          + (((UInt32)p[1] & 0x7) << 8)
+          + (((UInt32)p[-2] << 11))
+          + (p[0]);
+
+      p += 2;
+      {
+        UInt32 cur = (ip + (UInt32)(p - data)) >> 1;
+          v -= cur;
+      }
+
+      /*
+      SetUi16(p - 4, (UInt16)(((v >> 11) & 0x7FF) | 0xF000));
+      SetUi16(p - 2, (UInt16)(v | 0xF800));
+      */
+      
+      p[-4] = (Byte)(v >> 11);
+      p[-3] = (Byte)(0xF0 | ((v >> 19) & 0x7));
+      p[-2] = (Byte)v;
+      p[-1] = (Byte)(0xF8 | (v >> 8));
     }
   }
-  return i;
 }
 
+
 SizeT PPC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
 {
-  SizeT i;
+  Byte *p;
-  if (size < 4)
+  const Byte *lim;
-    return 0;
+  size &= ~(size_t)3;
-  size -= 4;
+  ip -= 4;
-  for (i = 0; i <= size; i += 4)
+  p = data;
+  lim = data + size;
+
+  for (;;)
   {
-    if ((data[i] >> 2) == 0x12 && (data[i + 3] & 3) == 1)
+    for (;;)
     {
-      UInt32 src = ((UInt32)(data[i + 0] & 3) << 24) |
+      if (p >= lim)
-        ((UInt32)data[i + 1] << 16) |
+        return p - data;
-        ((UInt32)data[i + 2] << 8) |
+      p += 4;
-        ((UInt32)data[i + 3] & (~3));
+      /* if ((v & 0xFC000003) == 0x48000001) */
-      
+      if ((p[-4] & 0xFC) == 0x48 && (p[-1] & 3) == 1)
-      UInt32 dest;
+        break;
+    }
+    {
+      UInt32 v = GetBe32(p - 4);
       if (encoding)
-        dest = ip + (UInt32)i + src;
+        v += ip + (UInt32)(p - data);
       else
-        dest = src - (ip + (UInt32)i);
+        v -= ip + (UInt32)(p - data);
-      data[i + 0] = (Byte)(0x48 | ((dest >> 24) &  0x3));
+      v &= 0x03FFFFFF;
-      data[i + 1] = (Byte)(dest >> 16);
+      v |= 0x48000000;
-      data[i + 2] = (Byte)(dest >> 8);
+      SetBe32(p - 4, v);
-      data[i + 3] &= 0x3;
-      data[i + 3] |= dest;
     }
   }
-  return i;
 }
 
+
 SizeT SPARC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
 {
-  UInt32 i;
+  Byte *p;
-  if (size < 4)
+  const Byte *lim;
-    return 0;
+  size &= ~(size_t)3;
-  size -= 4;
+  ip -= 4;
-  for (i = 0; i <= size; i += 4)
+  p = data;
-  {
+  lim = data + size;
-    if ((data[i] == 0x40 && (data[i + 1] & 0xC0) == 0x00) ||
-        (data[i] == 0x7F && (data[i + 1] & 0xC0) == 0xC0))
-    {
-      UInt32 src =
-        ((UInt32)data[i + 0] << 24) |
-        ((UInt32)data[i + 1] << 16) |
-        ((UInt32)data[i + 2] << 8) |
-        ((UInt32)data[i + 3]);
-      UInt32 dest;
-      
-      src <<= 2;
-      if (encoding)
-        dest = ip + i + src;
-      else
-        dest = src - (ip + i);
-      dest >>= 2;
-      
-      dest = (((0 - ((dest >> 22) & 1)) << 22) & 0x3FFFFFFF) | (dest & 0x3FFFFF) | 0x40000000;
 
-      data[i + 0] = (Byte)(dest >> 24);
+  for (;;)
-      data[i + 1] = (Byte)(dest >> 16);
+  {
-      data[i + 2] = (Byte)(dest >> 8);
+    for (;;)
-      data[i + 3] = (Byte)dest;
+    {
+      if (p >= lim)
+        return p - data;
+      /*
+      v = GetBe32(p);
+      p += 4;
+      m = v + ((UInt32)5 << 29);
+      m ^= (UInt32)7 << 29;
+      m += (UInt32)1 << 22;
+      if ((m & ((UInt32)0x1FF << 23)) == 0)
+        break;
+      */
+      p += 4;
+      if ((p[-4] == 0x40 && (p[-3] & 0xC0) == 0) ||
+          (p[-4] == 0x7F && (p[-3] >= 0xC0)))
+        break;
+    }
+    {
+      UInt32 v = GetBe32(p - 4);
+      v <<= 2;
+      if (encoding)
+        v += ip + (UInt32)(p - data);
+      else
+        v -= ip + (UInt32)(p - data);
+      
+      v &= 0x01FFFFFF;
+      v -= (UInt32)1 << 24;
+      v ^= 0xFF000000;
+      v >>= 2;
+      v |= 0x40000000;
+      SetBe32(p - 4, v);
     }
   }
-  return i;
 }

C/Bra86.c

@@ -1,5 +1,5 @@
 /* Bra86.c -- Converter for x86 code (BCJ)
-2013-11-12 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -37,7 +37,7 @@ SizeT x86_Convert(Byte *data, SizeT size, UInt32 ip, UInt32 *state, int encoding
       else
       {
         mask >>= (unsigned)d;
-        if (mask != 0 && (mask > 4 || mask == 3 || Test86MSByte(p[(mask >> 1) + 1])))
+        if (mask != 0 && (mask > 4 || mask == 3 || Test86MSByte(p[(size_t)(mask >> 1) + 1])))
         {
           mask = (mask >> 1) | 4;
           pos++;

C/BraIA64.c

@@ -1,69 +1,53 @@
 /* BraIA64.c -- Converter for IA-64 code
-2013-11-12 : Igor Pavlov : Public domain */
+2017-01-26 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
+#include "CpuArch.h"
 #include "Bra.h"
 
-static const Byte kBranchTable[32] =
-{
-  0, 0, 0, 0, 0, 0, 0, 0,
-  0, 0, 0, 0, 0, 0, 0, 0,
-  4, 4, 6, 6, 0, 0, 7, 7,
-  4, 4, 0, 0, 4, 4, 0, 0
-};
-
 SizeT IA64_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
 {
   SizeT i;
   if (size < 16)
     return 0;
   size -= 16;
-  for (i = 0; i <= size; i += 16)
+  i = 0;
+  do
   {
-    UInt32 instrTemplate = data[i] & 0x1F;
+    unsigned m = ((UInt32)0x334B0000 >> (data[i] & 0x1E)) & 3;
-    UInt32 mask = kBranchTable[instrTemplate];
+    if (m)
-    UInt32 bitPos = 5;
-    int slot;
-    for (slot = 0; slot < 3; slot++, bitPos += 41)
     {
-      UInt32 bytePos, bitRes;
+      m++;
-      UInt64 instruction, instNorm;
+      do
-      int j;
-      if (((mask >> slot) & 1) == 0)
-        continue;
-      bytePos = (bitPos >> 3);
-      bitRes = bitPos & 0x7;
-      instruction = 0;
-      for (j = 0; j < 6; j++)
-        instruction += (UInt64)data[i + j + bytePos] << (8 * j);
-
-      instNorm = instruction >> bitRes;
-      if (((instNorm >> 37) & 0xF) == 0x5 && ((instNorm >> 9) & 0x7) == 0)
       {
-        UInt32 src = (UInt32)((instNorm >> 13) & 0xFFFFF);
+        Byte *p = data + (i + (size_t)m * 5 - 8);
-        UInt32 dest;
+        if (((p[3] >> m) & 15) == 5
-        src |= ((UInt32)(instNorm >> 36) & 1) << 20;
+            && (((p[-1] | ((UInt32)p[0] << 8)) >> m) & 0x70) == 0)
-        
+        {
-        src <<= 4;
+          unsigned raw = GetUi32(p);
-        
+          unsigned v = raw >> m;
-        if (encoding)
+          v = (v & 0xFFFFF) | ((v & (1 << 23)) >> 3);
-          dest = ip + (UInt32)i + src;
+          
-        else
+          v <<= 4;
-          dest = src - (ip + (UInt32)i);
+          if (encoding)
-        
+            v += ip + (UInt32)i;
-        dest >>= 4;
+          else
-        
+            v -= ip + (UInt32)i;
-        instNorm &= ~((UInt64)(0x8FFFFF) << 13);
+          v >>= 4;
-        instNorm |= ((UInt64)(dest & 0xFFFFF) << 13);
+          
-        instNorm |= ((UInt64)(dest & 0x100000) << (36 - 20));
+          v &= 0x1FFFFF;
-        
+          v += 0x700000;
-        instruction &= (1 << bitRes) - 1;
+          v &= 0x8FFFFF;
-        instruction |= (instNorm << bitRes);
+          raw &= ~((UInt32)0x8FFFFF << m);
-        for (j = 0; j < 6; j++)
+          raw |= (v << m);
-          data[i + j + bytePos] = (Byte)(instruction >> (8 * j));
+          SetUi32(p, raw);
+        }
       }
+      while (++m <= 4);
     }
+    i += 16;
   }
+  while (i <= size);
   return i;
 }

C/BwtSort.c

@@ -1,5 +1,5 @@
 /* BwtSort.c -- BWT block sorting
-2013-11-12 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -314,7 +314,7 @@ UInt32 NO_INLINE SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 group
       #ifndef BLOCK_SORT_EXTERNAL_FLAGS
       UInt32 subGroupSize = ((ind2[j] & ~0xC0000000) >> kNumBitsMax);
       if ((ind2[j] & 0x40000000) != 0)
-        subGroupSize += ((ind2[j + 1] >> kNumBitsMax) << kNumExtra0Bits);
+        subGroupSize += ((ind2[(size_t)j + 1] >> kNumBitsMax) << kNumExtra0Bits);
       subGroupSize++;
       for (;;)
       {
@@ -362,7 +362,7 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
   for (i = 0; i < kNumHashValues; i++)
     counters[i] = 0;
   for (i = 0; i < blockSize - 1; i++)
-    counters[((UInt32)data[i] << 8) | data[i + 1]]++;
+    counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]]++;
   counters[((UInt32)data[i] << 8) | data[0]]++;
 
   Groups = counters + BS_TEMP_SIZE;
@@ -392,11 +392,11 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
     }
 
     for (i = 0; i < blockSize - 1; i++)
-      Groups[i] = counters[((UInt32)data[i] << 8) | data[i + 1]];
+      Groups[i] = counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]];
     Groups[i] = counters[((UInt32)data[i] << 8) | data[0]];
 
     for (i = 0; i < blockSize - 1; i++)
-      Indices[counters[((UInt32)data[i] << 8) | data[i + 1]]++] = i;
+      Indices[counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]]++] = i;
     Indices[counters[((UInt32)data[i] << 8) | data[0]]++] = i;
     
     #ifndef BLOCK_SORT_EXTERNAL_FLAGS
@@ -451,8 +451,8 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
       Bool finishedGroup = ((Indices[i] & 0x80000000) == 0);
       if ((Indices[i] & 0x40000000) != 0)
       {
-        groupSize += ((Indices[i + 1] >> kNumBitsMax) << kNumExtra0Bits);
+        groupSize += ((Indices[(size_t)i + 1] >> kNumBitsMax) << kNumExtra0Bits);
-        Indices[i + 1] &= kIndexMask;
+        Indices[(size_t)i + 1] &= kIndexMask;
       }
       Indices[i] &= kIndexMask;
       groupSize++;
@@ -460,7 +460,7 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
       {
         Indices[i - finishedGroupSize] &= kIndexMask;
         if (finishedGroupSize > 1)
-          Indices[i - finishedGroupSize + 1] &= kIndexMask;
+          Indices[(size_t)(i - finishedGroupSize) + 1] &= kIndexMask;
         {
         UInt32 newGroupSize = groupSize + finishedGroupSize;
         SetFinishedGroupSize(Indices + i - finishedGroupSize, newGroupSize);
@@ -503,8 +503,8 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
     UInt32 groupSize = ((Indices[i] & ~0xC0000000) >> kNumBitsMax);
     if ((Indices[i] & 0x40000000) != 0)
     {
-      groupSize += ((Indices[i + 1] >> kNumBitsMax) << kNumExtra0Bits);
+      groupSize += ((Indices[(size_t)i + 1] >> kNumBitsMax) << kNumExtra0Bits);
-      Indices[i + 1] &= kIndexMask;
+      Indices[(size_t)i + 1] &= kIndexMask;
     }
     Indices[i] &= kIndexMask;
     groupSize++;
@@ -513,4 +513,3 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
   #endif
   return Groups[0];
 }
-

C/Compiler.h

@@ -1,5 +1,5 @@
 /* Compiler.h
-2015-08-02 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_COMPILER_H
 #define __7Z_COMPILER_H
@@ -21,6 +21,7 @@
     #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 : 4714) // function marked as __forceinline not inlined
     #pragma warning(disable : 4786) // identifier was truncated to '255' characters in the debug information
   #endif
 

C/CpuArch.c

@@ -1,5 +1,5 @@
 /* CpuArch.c -- CPU specific code
-2015-03-25: Igor Pavlov : Public domain */
+2016-02-25: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -45,7 +45,8 @@ static UInt32 CheckFlag(UInt32 flag)
     "push %%EDX\n\t"
     "popf\n\t"
     "andl %%EAX, %0\n\t":
-    "=c" (flag) : "c" (flag));
+    "=c" (flag) : "c" (flag) :
+    "%eax", "%edx");
   #endif
   return flag;
 }
@@ -79,7 +80,13 @@ void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d)
   #else
 
   __asm__ __volatile__ (
-  #if defined(MY_CPU_X86) && defined(__PIC__)
+  #if defined(MY_CPU_AMD64) && defined(__PIC__)
+    "mov %%rbx, %%rdi;"
+    "cpuid;"
+    "xchg %%rbx, %%rdi;"
+    : "=a" (*a) ,
+      "=D" (*b) ,
+  #elif defined(MY_CPU_X86) && defined(__PIC__)
     "mov %%ebx, %%edi;"
     "cpuid;"
     "xchgl %%ebx, %%edi;"

C/CpuArch.h

@@ -1,5 +1,5 @@
 /* CpuArch.h -- CPU specific code
-2015-08-02: Igor Pavlov : Public domain */
+2017-09-04 : Igor Pavlov : Public domain */
 
 #ifndef __CPU_ARCH_H
 #define __CPU_ARCH_H
@@ -10,54 +10,128 @@ 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__)
+#if  defined(_M_X64) \
-#define MY_CPU_AMD64
+  || defined(_M_AMD64) \
-#endif
+  || defined(__x86_64__) \
-
+  || defined(__AMD64__) \
-#if defined(MY_CPU_AMD64) \
+  || defined(__amd64__)
-    || defined(_M_IA64) \
+  #define MY_CPU_AMD64
-    || defined(__AARCH64EL__) \
+  #ifdef __ILP32__
-    || defined(__AARCH64EB__)
+    #define MY_CPU_NAME "x32"
+  #else
+    #define MY_CPU_NAME "x64"
+  #endif
   #define MY_CPU_64BIT
 #endif
 
-#if defined(_M_IX86) || defined(__i386__)
+
-#define MY_CPU_X86
+#if  defined(_M_IX86) \
+  || defined(__i386__)
+  #define MY_CPU_X86
+  #define MY_CPU_NAME "x86"
+  #define MY_CPU_32BIT
 #endif
 
+
+#if  defined(_M_ARM64) \
+  || defined(__AARCH64EL__) \
+  || defined(__AARCH64EB__) \
+  || defined(__aarch64__)
+  #define MY_CPU_ARM64
+  #define MY_CPU_NAME "arm64"
+  #define MY_CPU_64BIT
+#endif
+
+
+#if  defined(_M_ARM) \
+  || defined(_M_ARM_NT) \
+  || defined(_M_ARMT) \
+  || defined(__arm__) \
+  || defined(__thumb__) \
+  || defined(__ARMEL__) \
+  || defined(__ARMEB__) \
+  || defined(__THUMBEL__) \
+  || defined(__THUMBEB__)
+  #define MY_CPU_ARM
+  #define MY_CPU_NAME "arm"
+  #define MY_CPU_32BIT
+#endif
+
+
+#if  defined(_M_IA64) \
+  || defined(__ia64__)
+  #define MY_CPU_IA64
+  #define MY_CPU_NAME "ia64"
+  #define MY_CPU_64BIT
+#endif
+
+
+#if  defined(__mips64) \
+  || defined(__mips64__) \
+  || (defined(__mips) && (__mips == 64 || __mips == 4 || __mips == 3))
+  #define MY_CPU_NAME "mips64"
+  #define MY_CPU_64BIT
+#elif defined(__mips__)
+  #define MY_CPU_NAME "mips"
+  /* #define MY_CPU_32BIT */
+#endif
+
+
+#if  defined(__ppc64__) \
+  || defined(__powerpc64__)
+  #ifdef __ILP32__
+    #define MY_CPU_NAME "ppc64-32"
+  #else
+    #define MY_CPU_NAME "ppc64"
+  #endif
+  #define MY_CPU_64BIT
+#elif defined(__ppc__) \
+  || defined(__powerpc__)
+  #define MY_CPU_NAME "ppc"
+  #define MY_CPU_32BIT
+#endif
+
+
+#if  defined(__sparc64__)
+  #define MY_CPU_NAME "sparc64"
+  #define MY_CPU_64BIT
+#elif defined(__sparc__)
+  #define MY_CPU_NAME "sparc"
+  /* #define MY_CPU_32BIT */
+#endif
+
+
 #if defined(MY_CPU_X86) || defined(MY_CPU_AMD64)
 #define MY_CPU_X86_OR_AMD64
 #endif
 
-#if defined(MY_CPU_X86) \
+
-    || defined(_M_ARM) \
+#ifdef _WIN32
-    || defined(__ARMEL__) \
+
-    || defined(__THUMBEL__) \
+  #ifdef MY_CPU_ARM
-    || defined(__ARMEB__) \
+  #define MY_CPU_ARM_LE
-    || defined(__THUMBEB__)
+  #endif
-  #define MY_CPU_32BIT
+
+  #ifdef MY_CPU_ARM64
+  #define MY_CPU_ARM64_LE
+  #endif
+
+  #ifdef _M_IA64
+  #define MY_CPU_IA64_LE
+  #endif
+
 #endif
 
-#if defined(_WIN32) && defined(_M_ARM)
-#define MY_CPU_ARM_LE
-#endif
-
-#if defined(_WIN32) && defined(_M_IA64)
-#define MY_CPU_IA64_LE
-#endif
-
-#if defined(MY_CPU_X86_OR_AMD64)
-#define MY_CPU_LE_UNALIGN
-#endif
 
 #if defined(MY_CPU_X86_OR_AMD64) \
     || defined(MY_CPU_ARM_LE) \
+    || defined(MY_CPU_ARM64_LE) \
     || defined(MY_CPU_IA64_LE) \
     || defined(__LITTLE_ENDIAN__) \
     || defined(__ARMEL__) \
@@ -65,7 +139,9 @@ If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of pla
     || defined(__AARCH64EL__) \
     || defined(__MIPSEL__) \
     || defined(__MIPSEL) \
-    || defined(_MIPSEL)
+    || defined(_MIPSEL) \
+    || defined(__BFIN__) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
   #define MY_CPU_LE
 #endif
 
@@ -76,12 +152,47 @@ If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of pla
     || defined(__MIPSEB__) \
     || defined(__MIPSEB) \
     || defined(_MIPSEB) \
-    || defined(__m68k__)
+    || defined(__m68k__) \
+    || defined(__s390__) \
+    || defined(__s390x__) \
+    || defined(__zarch__) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))
   #define MY_CPU_BE
 #endif
 
+
 #if defined(MY_CPU_LE) && defined(MY_CPU_BE)
-Stop_Compiling_Bad_Endian
+  #error Stop_Compiling_Bad_Endian
+#endif
+
+
+#if defined(MY_CPU_32BIT) && defined(MY_CPU_64BIT)
+  #error Stop_Compiling_Bad_32_64_BIT
+#endif
+
+
+#ifndef MY_CPU_NAME
+  #ifdef MY_CPU_LE
+    #define MY_CPU_NAME "LE"
+  #elif defined(MY_CPU_BE)
+    #define MY_CPU_NAME "BE"
+  #else
+    /*
+    #define MY_CPU_NAME ""
+    */
+  #endif
+#endif
+
+
+
+
+
+#ifdef MY_CPU_LE
+  #if defined(MY_CPU_X86_OR_AMD64) \
+      || defined(MY_CPU_ARM64) \
+      || defined(__ARM_FEATURE_UNALIGNED)
+    #define MY_CPU_LE_UNALIGN
+  #endif
 #endif
 
 
@@ -125,18 +236,38 @@ Stop_Compiling_Bad_Endian
 
 #endif
 
+#ifdef __has_builtin
+  #define MY__has_builtin(x) __has_builtin(x)
+#else
+  #define MY__has_builtin(x) 0
+#endif
 
 #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>
 
+#pragma intrinsic(_byteswap_ushort)
 #pragma intrinsic(_byteswap_ulong)
 #pragma intrinsic(_byteswap_uint64)
+
+/* #define GetBe16(p) _byteswap_ushort(*(const UInt16 *)(const Byte *)(p)) */
 #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))) \
+    || (defined(__clang__) && MY__has_builtin(__builtin_bswap16)) )
+
+/* #define GetBe16(p) __builtin_bswap16(*(const UInt16 *)(const Byte *)(p)) */
+#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) ( \
@@ -156,10 +287,14 @@ Stop_Compiling_Bad_Endian
 #endif
 
 
+#ifndef GetBe16
+
 #define GetBe16(p) ( (UInt16) ( \
     ((UInt16)((const Byte *)(p))[0] << 8) | \
              ((const Byte *)(p))[1] ))
 
+#endif
+
 
 
 #ifdef MY_CPU_X86_OR_AMD64

C/DllSecur.c

@@ -0,0 +1,87 @@
+/* DllSecur.c -- DLL loading security
+2016-10-04 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+#ifdef _WIN32
+
+#include <windows.h>
+
+#include "DllSecur.h"
+
+#ifndef UNDER_CE
+
+typedef BOOL (WINAPI *Func_SetDefaultDllDirectories)(DWORD DirectoryFlags);
+
+#define MY_LOAD_LIBRARY_SEARCH_USER_DIRS 0x400
+#define MY_LOAD_LIBRARY_SEARCH_SYSTEM32  0x800
+
+static const char * const g_Dlls =
+  #ifndef _CONSOLE
+  "UXTHEME\0"
+  #endif
+  "USERENV\0"
+  "SETUPAPI\0"
+  "APPHELP\0"
+  "PROPSYS\0"
+  "DWMAPI\0"
+  "CRYPTBASE\0"
+  "OLEACC\0"
+  "CLBCATQ\0"
+  ;
+
+#endif
+
+void LoadSecurityDlls()
+{
+  #ifndef UNDER_CE
+  
+  wchar_t buf[MAX_PATH + 100];
+
+  {
+    // at Vista (ver 6.0) : CoCreateInstance(CLSID_ShellLink, ...) doesn't work after SetDefaultDllDirectories() : Check it ???
+    OSVERSIONINFO vi;
+    vi.dwOSVersionInfoSize = sizeof(vi);
+    if (!GetVersionEx(&vi) || vi.dwMajorVersion != 6 || vi.dwMinorVersion != 0)
+    {
+      Func_SetDefaultDllDirectories setDllDirs = (Func_SetDefaultDllDirectories)
+          GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")), "SetDefaultDllDirectories");
+      if (setDllDirs)
+        if (setDllDirs(MY_LOAD_LIBRARY_SEARCH_SYSTEM32 | MY_LOAD_LIBRARY_SEARCH_USER_DIRS))
+          return;
+    }
+  }
+
+  {
+    unsigned len = GetSystemDirectoryW(buf, MAX_PATH + 2);
+    if (len == 0 || len > MAX_PATH)
+      return;
+  }
+  {
+    const char *dll;
+    unsigned pos = (unsigned)lstrlenW(buf);
+
+    if (buf[pos - 1] != '\\')
+      buf[pos++] = '\\';
+    
+    for (dll = g_Dlls; dll[0] != 0;)
+    {
+      unsigned k = 0;
+      for (;;)
+      {
+        char c = *dll++;
+        buf[pos + k] = c;
+        k++;
+        if (c == 0)
+          break;
+      }
+
+      lstrcatW(buf, L".dll");
+      LoadLibraryExW(buf, NULL, LOAD_WITH_ALTERED_SEARCH_PATH);
+    }
+  }
+  
+  #endif
+}
+
+#endif

C/DllSecur.h

@@ -0,0 +1,19 @@
+/* DllSecur.h -- DLL loading for security
+2016-06-08 : Igor Pavlov : Public domain */
+
+#ifndef __DLL_SECUR_H
+#define __DLL_SECUR_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+#ifdef _WIN32
+
+void LoadSecurityDlls();
+
+#endif
+
+EXTERN_C_END
+
+#endif

C/HuffEnc.c

@@ -1,5 +1,5 @@
 /* HuffEnc.c -- functions for Huffman encoding
-2009-09-02 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -113,7 +113,7 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
         if (len >= maxLen)
           for (len = maxLen - 1; lenCounters[len] == 0; len--);
         lenCounters[len]--;
-        lenCounters[len + 1] += 2;
+        lenCounters[(size_t)len + 1] += 2;
       }
       
       {
@@ -121,8 +121,8 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
         i = 0;
         for (len = maxLen; len != 0; len--)
         {
-          UInt32 num;
+          UInt32 k;
-          for (num = lenCounters[len]; num != 0; num--)
+          for (k = lenCounters[len]; k != 0; k--)
             lens[p[i++] & MASK] = (Byte)len;
         }
       }
@@ -133,14 +133,14 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
           UInt32 code = 0;
           UInt32 len;
           for (len = 1; len <= kMaxLen; len++)
-            nextCodes[len] = code = (code + lenCounters[len - 1]) << 1;
+            nextCodes[len] = code = (code + lenCounters[(size_t)len - 1]) << 1;
         }
         /* if (code + lenCounters[kMaxLen] - 1 != (1 << kMaxLen) - 1) throw 1; */
 
         {
-          UInt32 i;
+          UInt32 k;
-          for (i = 0; i < numSymbols; i++)
+          for (k = 0; k < numSymbols; k++)
-            p[i] = nextCodes[lens[i]]++;
+            p[k] = nextCodes[lens[k]]++;
         }
       }
     }

C/LzFind.c

@@ -1,5 +1,5 @@
 /* LzFind.c -- Match finder for LZ algorithms
-2015-05-15 : Igor Pavlov : Public domain */
+2017-06-10 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -11,23 +11,23 @@
 #define kEmptyHashValue 0
 #define kMaxValForNormalize ((UInt32)0xFFFFFFFF)
 #define kNormalizeStepMin (1 << 10) /* it must be power of 2 */
-#define kNormalizeMask (~(kNormalizeStepMin - 1))
+#define kNormalizeMask (~(UInt32)(kNormalizeStepMin - 1))
 #define kMaxHistorySize ((UInt32)7 << 29)
 
 #define kStartMaxLen 3
 
-static void LzInWindow_Free(CMatchFinder *p, ISzAlloc *alloc)
+static void LzInWindow_Free(CMatchFinder *p, ISzAllocPtr alloc)
 {
   if (!p->directInput)
   {
-    alloc->Free(alloc, p->bufferBase);
+    ISzAlloc_Free(alloc, p->bufferBase);
     p->bufferBase = NULL;
   }
 }
 
 /* keepSizeBefore + keepSizeAfter + keepSizeReserv must be < 4G) */
 
-static int LzInWindow_Create(CMatchFinder *p, UInt32 keepSizeReserv, ISzAlloc *alloc)
+static int LzInWindow_Create(CMatchFinder *p, UInt32 keepSizeReserv, ISzAllocPtr alloc)
 {
   UInt32 blockSize = p->keepSizeBefore + p->keepSizeAfter + keepSizeReserv;
   if (p->directInput)
@@ -39,7 +39,7 @@ static int LzInWindow_Create(CMatchFinder *p, UInt32 keepSizeReserv, ISzAlloc *a
   {
     LzInWindow_Free(p, alloc);
     p->blockSize = blockSize;
-    p->bufferBase = (Byte *)alloc->Alloc(alloc, (size_t)blockSize);
+    p->bufferBase = (Byte *)ISzAlloc_Alloc(alloc, (size_t)blockSize);
   }
   return (p->bufferBase != NULL);
 }
@@ -60,9 +60,11 @@ 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;
@@ -79,7 +81,7 @@ static void MatchFinder_ReadBlock(CMatchFinder *p)
     if (size == 0)
       return;
 
-    p->result = p->stream->Read(p->stream, dest, &size);
+    p->result = ISeqInStream_Read(p->stream, dest, &size);
     if (p->result != SZ_OK)
       return;
     if (size == 0)
@@ -97,7 +99,7 @@ void MatchFinder_MoveBlock(CMatchFinder *p)
 {
   memmove(p->bufferBase,
       p->buffer - p->keepSizeBefore,
-      (size_t)(p->streamPos - p->pos + p->keepSizeBefore));
+      (size_t)(p->streamPos - p->pos) + p->keepSizeBefore);
   p->buffer = p->bufferBase + p->keepSizeBefore;
 }
 
@@ -140,6 +142,7 @@ void MatchFinder_Construct(CMatchFinder *p)
   p->bufferBase = NULL;
   p->directInput = 0;
   p->hash = NULL;
+  p->expectedDataSize = (UInt64)(Int64)-1;
   MatchFinder_SetDefaultSettings(p);
 
   for (i = 0; i < 256; i++)
@@ -147,34 +150,34 @@ void MatchFinder_Construct(CMatchFinder *p)
     UInt32 r = i;
     unsigned j;
     for (j = 0; j < 8; j++)
-      r = (r >> 1) ^ (kCrcPoly & ~((r & 1) - 1));
+      r = (r >> 1) ^ (kCrcPoly & ((UInt32)0 - (r & 1)));
     p->crc[i] = r;
   }
 }
 
-static void MatchFinder_FreeThisClassMemory(CMatchFinder *p, ISzAlloc *alloc)
+static void MatchFinder_FreeThisClassMemory(CMatchFinder *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->hash);
+  ISzAlloc_Free(alloc, p->hash);
   p->hash = NULL;
 }
 
-void MatchFinder_Free(CMatchFinder *p, ISzAlloc *alloc)
+void MatchFinder_Free(CMatchFinder *p, ISzAllocPtr alloc)
 {
   MatchFinder_FreeThisClassMemory(p, alloc);
   LzInWindow_Free(p, alloc);
 }
 
-static CLzRef* AllocRefs(size_t num, ISzAlloc *alloc)
+static CLzRef* AllocRefs(size_t num, ISzAllocPtr alloc)
 {
   size_t sizeInBytes = (size_t)num * sizeof(CLzRef);
   if (sizeInBytes / sizeof(CLzRef) != num)
     return NULL;
-  return (CLzRef *)alloc->Alloc(alloc, sizeInBytes);
+  return (CLzRef *)ISzAlloc_Alloc(alloc, sizeInBytes);
 }
 
 int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
     UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter,
-    ISzAlloc *alloc)
+    ISzAllocPtr alloc)
 {
   UInt32 sizeReserv;
   
@@ -206,7 +209,11 @@ int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
         hs = (1 << 16) - 1;
       else
       {
-        hs = historySize - 1;
+        hs = historySize;
+        if (hs > p->expectedDataSize)
+          hs = (UInt32)p->expectedDataSize;
+        if (hs != 0)
+          hs--;
         hs |= (hs >> 1);
         hs |= (hs >> 2);
         hs |= (hs >> 4);
@@ -290,23 +297,51 @@ static void MatchFinder_SetLimits(CMatchFinder *p)
   p->posLimit = p->pos + limit;
 }
 
-void MatchFinder_Init(CMatchFinder *p)
+
+void MatchFinder_Init_LowHash(CMatchFinder *p)
+{
+  size_t i;
+  CLzRef *items = p->hash;
+  size_t numItems = p->fixedHashSize;
+  for (i = 0; i < numItems; i++)
+    items[i] = kEmptyHashValue;
+}
+
+
+void MatchFinder_Init_HighHash(CMatchFinder *p)
+{
+  size_t i;
+  CLzRef *items = p->hash + p->fixedHashSize;
+  size_t numItems = (size_t)p->hashMask + 1;
+  for (i = 0; i < numItems; i++)
+    items[i] = kEmptyHashValue;
+}
+
+
+void MatchFinder_Init_3(CMatchFinder *p, int readData)
 {
-  UInt32 i;
-  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->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_HighHash(p);
+  MatchFinder_Init_LowHash(p);
+  MatchFinder_Init_3(p, True);
+}
+
+  
 static UInt32 MatchFinder_GetSubValue(CMatchFinder *p)
 {
   return (p->pos - p->historySize - 1) & kNormalizeMask;
@@ -548,10 +583,10 @@ static UInt32 Bt3_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
 
   d2 = pos - hash[h2];
 
-  curMatch = hash[kFix3HashSize + hv];
+  curMatch = (hash + kFix3HashSize)[hv];
   
   hash[h2] = pos;
-  hash[kFix3HashSize + hv] = pos;
+  (hash + kFix3HashSize)[hv] = pos;
 
   maxLen = 2;
   offset = 0;
@@ -584,13 +619,13 @@ static UInt32 Bt4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   pos = p->pos;
 
   d2 = pos - hash[                h2];
-  d3 = pos - hash[kFix3HashSize + h3];
+  d3 = pos - (hash + kFix3HashSize)[h3];
 
-  curMatch = hash[kFix4HashSize + hv];
+  curMatch = (hash + kFix4HashSize)[hv];
 
   hash[                h2] = pos;
-  hash[kFix3HashSize + h3] = pos;
+  (hash + kFix3HashSize)[h3] = pos;
-  hash[kFix4HashSize + hv] = pos;
+  (hash + kFix4HashSize)[hv] = pos;
 
   maxLen = 0;
   offset = 0;
@@ -605,7 +640,7 @@ static UInt32 Bt4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (d2 != d3 && d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
   {
     maxLen = 3;
-    distances[offset + 1] = d3 - 1;
+    distances[(size_t)offset + 1] = d3 - 1;
     offset += 2;
     d2 = d3;
   }
@@ -613,7 +648,7 @@ static UInt32 Bt4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (offset != 0)
   {
     UPDATE_maxLen
-    distances[offset - 2] = maxLen;
+    distances[(size_t)offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
       SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p));
@@ -640,15 +675,15 @@ static UInt32 Bt5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   pos = p->pos;
 
   d2 = pos - hash[                h2];
-  d3 = pos - hash[kFix3HashSize + h3];
+  d3 = pos - (hash + kFix3HashSize)[h3];
-  d4 = pos - hash[kFix4HashSize + h4];
+  d4 = pos - (hash + kFix4HashSize)[h4];
 
-  curMatch = hash[kFix5HashSize + hv];
+  curMatch = (hash + kFix5HashSize)[hv];
 
   hash[                h2] = pos;
-  hash[kFix3HashSize + h3] = pos;
+  (hash + kFix3HashSize)[h3] = pos;
-  hash[kFix4HashSize + h4] = pos;
+  (hash + kFix4HashSize)[h4] = pos;
-  hash[kFix5HashSize + hv] = pos;
+  (hash + kFix5HashSize)[hv] = pos;
 
   maxLen = 0;
   offset = 0;
@@ -681,7 +716,7 @@ static UInt32 Bt5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
       && *(cur - d4 + 3) == *(cur + 3))
   {
     maxLen = 4;
-    distances[offset + 1] = d4 - 1;
+    distances[(size_t)offset + 1] = d4 - 1;
     offset += 2;
     d2 = d4;
   }
@@ -689,7 +724,7 @@ static UInt32 Bt5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (offset != 0)
   {
     UPDATE_maxLen
-    distances[offset - 2] = maxLen;
+    distances[(size_t)offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
       SkipMatchesSpec(lenLimit, curMatch, MF_PARAMS(p));
@@ -716,13 +751,13 @@ static UInt32 Hc4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   pos = p->pos;
   
   d2 = pos - hash[                h2];
-  d3 = pos - hash[kFix3HashSize + h3];
+  d3 = pos - (hash + kFix3HashSize)[h3];
   
-  curMatch = hash[kFix4HashSize + hv];
+  curMatch = (hash + kFix4HashSize)[hv];
 
   hash[                h2] = pos;
-  hash[kFix3HashSize + h3] = pos;
+  (hash + kFix3HashSize)[h3] = pos;
-  hash[kFix4HashSize + hv] = pos;
+  (hash + kFix4HashSize)[hv] = pos;
 
   maxLen = 0;
   offset = 0;
@@ -737,7 +772,7 @@ static UInt32 Hc4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (d2 != d3 && d3 < p->cyclicBufferSize && *(cur - d3) == *cur)
   {
     maxLen = 3;
-    distances[offset + 1] = d3 - 1;
+    distances[(size_t)offset + 1] = d3 - 1;
     offset += 2;
     d2 = d3;
   }
@@ -745,7 +780,7 @@ static UInt32 Hc4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (offset != 0)
   {
     UPDATE_maxLen
-    distances[offset - 2] = maxLen;
+    distances[(size_t)offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
       p->son[p->cyclicBufferPos] = curMatch;
@@ -774,15 +809,15 @@ static UInt32 Hc5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   pos = p->pos;
   
   d2 = pos - hash[                h2];
-  d3 = pos - hash[kFix3HashSize + h3];
+  d3 = pos - (hash + kFix3HashSize)[h3];
-  d4 = pos - hash[kFix4HashSize + h4];
+  d4 = pos - (hash + kFix4HashSize)[h4];
 
-  curMatch = hash[kFix5HashSize + hv];
+  curMatch = (hash + kFix5HashSize)[hv];
 
   hash[                h2] = pos;
-  hash[kFix3HashSize + h3] = pos;
+  (hash + kFix3HashSize)[h3] = pos;
-  hash[kFix4HashSize + h4] = pos;
+  (hash + kFix4HashSize)[h4] = pos;
-  hash[kFix5HashSize + hv] = pos;
+  (hash + kFix5HashSize)[hv] = pos;
 
   maxLen = 0;
   offset = 0;
@@ -815,7 +850,7 @@ static UInt32 Hc5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
       && *(cur - d4 + 3) == *(cur + 3))
   {
     maxLen = 4;
-    distances[offset + 1] = d4 - 1;
+    distances[(size_t)offset + 1] = d4 - 1;
     offset += 2;
     d2 = d4;
   }
@@ -823,7 +858,7 @@ static UInt32 Hc5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances)
   if (offset != 0)
   {
     UPDATE_maxLen
-    distances[offset - 2] = maxLen;
+    distances[(size_t)offset - 2] = maxLen;
     if (maxLen == lenLimit)
     {
       p->son[p->cyclicBufferPos] = curMatch;
@@ -887,9 +922,9 @@ static void Bt3_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
     SKIP_HEADER(3)
     HASH3_CALC;
     hash = p->hash;
-    curMatch = hash[kFix3HashSize + hv];
+    curMatch = (hash + kFix3HashSize)[hv];
     hash[h2] =
-    hash[kFix3HashSize + hv] = p->pos;
+    (hash + kFix3HashSize)[hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -904,10 +939,10 @@ static void Bt4_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
     SKIP_HEADER(4)
     HASH4_CALC;
     hash = p->hash;
-    curMatch = hash[kFix4HashSize + hv];
+    curMatch = (hash + kFix4HashSize)[hv];
     hash[                h2] =
-    hash[kFix3HashSize + h3] =
+    (hash + kFix3HashSize)[h3] =
-    hash[kFix4HashSize + hv] = p->pos;
+    (hash + kFix4HashSize)[hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -923,11 +958,11 @@ static void Bt5_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
     SKIP_HEADER(5)
     HASH5_CALC;
     hash = p->hash;
-    curMatch = hash[kFix5HashSize + hv];
+    curMatch = (hash + kFix5HashSize)[hv];
     hash[                h2] =
-    hash[kFix3HashSize + h3] =
+    (hash + kFix3HashSize)[h3] =
-    hash[kFix4HashSize + h4] =
+    (hash + kFix4HashSize)[h4] =
-    hash[kFix5HashSize + hv] = p->pos;
+    (hash + kFix5HashSize)[hv] = p->pos;
     SKIP_FOOTER
   }
   while (--num != 0);
@@ -943,10 +978,10 @@ static void Hc4_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
     SKIP_HEADER(4)
     HASH4_CALC;
     hash = p->hash;
-    curMatch = hash[kFix4HashSize + hv];
+    curMatch = (hash + kFix4HashSize)[hv];
     hash[                h2] =
-    hash[kFix3HashSize + h3] =
+    (hash + kFix3HashSize)[h3] =
-    hash[kFix4HashSize + hv] = p->pos;
+    (hash + kFix4HashSize)[hv] = p->pos;
     p->son[p->cyclicBufferPos] = curMatch;
     MOVE_POS
   }
@@ -963,11 +998,11 @@ static void Hc5_MatchFinder_Skip(CMatchFinder *p, UInt32 num)
     SKIP_HEADER(5)
     HASH5_CALC;
     hash = p->hash;
-    curMatch = p->hash[kFix5HashSize + hv];
+    curMatch = hash + kFix5HashSize)[hv];
     hash[                h2] =
-    hash[kFix3HashSize + h3] =
+    (hash + kFix3HashSize)[h3] =
-    hash[kFix4HashSize + h4] =
+    (hash + kFix4HashSize)[h4] =
-    hash[kFix5HashSize + hv] = p->pos;
+    (hash + kFix5HashSize)[hv] = p->pos;
     p->son[p->cyclicBufferPos] = curMatch;
     MOVE_POS
   }

C/LzFind.h

@@ -1,5 +1,5 @@
 /* LzFind.h -- Match finder for LZ algorithms
-2015-05-01 : Igor Pavlov : Public domain */
+2017-06-10 : Igor Pavlov : Public domain */
 
 #ifndef __LZ_FIND_H
 #define __LZ_FIND_H
@@ -47,12 +47,19 @@ typedef struct _CMatchFinder
   SRes result;
   UInt32 crc[256];
   size_t numRefs;
+
+  UInt64 expectedDataSize;
 } CMatchFinder;
 
 #define Inline_MatchFinder_GetPointerToCurrentPos(p) ((p)->buffer)
 
 #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);
@@ -66,8 +73,8 @@ void MatchFinder_Construct(CMatchFinder *p);
 */
 int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
     UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter,
-    ISzAlloc *alloc);
+    ISzAllocPtr alloc);
-void MatchFinder_Free(CMatchFinder *p, ISzAlloc *alloc);
+void MatchFinder_Free(CMatchFinder *p, ISzAllocPtr alloc);
 void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems);
 void MatchFinder_ReduceOffsets(CMatchFinder *p, UInt32 subValue);
 
@@ -98,9 +105,14 @@ typedef struct _IMatchFinder
 
 void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder *vTable);
 
+void MatchFinder_Init_LowHash(CMatchFinder *p);
+void MatchFinder_Init_HighHash(CMatchFinder *p);
+void MatchFinder_Init_3(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);
 

C/LzFindMt.c

@@ -1,5 +1,5 @@
 /* LzFindMt.c -- multithreaded Match finder for LZ algorithms
-2015-05-03 : Igor Pavlov : Public domain */
+2017-06-10 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -33,6 +33,8 @@ static void MtSync_GetNextBlock(CMtSync *p)
 
     Event_Set(&p->canStart);
     Event_Wait(&p->wasStarted);
+
+    // if (mt) MatchFinder_Init_LowHash(mt->MatchFinder);
   }
   else
   {
@@ -155,6 +157,9 @@ static void HashThreadFunc(CMatchFinderMt *mt)
     UInt32 numProcessedBlocks = 0;
     Event_Wait(&p->canStart);
     Event_Set(&p->wasStarted);
+
+    MatchFinder_Init_HighHash(mt->MatchFinder);
+
     for (;;)
     {
       if (p->exit)
@@ -173,12 +178,12 @@ static void HashThreadFunc(CMatchFinderMt *mt)
           CriticalSection_Enter(&mt->btSync.cs);
           CriticalSection_Enter(&mt->hashSync.cs);
           {
-            const Byte *beforePtr = MatchFinder_GetPointerToCurrentPos(mf);
+            const Byte *beforePtr = Inline_MatchFinder_GetPointerToCurrentPos(mf);
-            const Byte *afterPtr;
+            ptrdiff_t offset;
             MatchFinder_MoveBlock(mf);
-            afterPtr = MatchFinder_GetPointerToCurrentPos(mf);
+            offset = beforePtr - Inline_MatchFinder_GetPointerToCurrentPos(mf);
-            mt->pointerToCurPos -= beforePtr - afterPtr;
+            mt->pointerToCurPos -= offset;
-            mt->buffer -= beforePtr - afterPtr;
+            mt->buffer -= offset;
           }
           CriticalSection_Leave(&mt->btSync.cs);
           CriticalSection_Leave(&mt->hashSync.cs);
@@ -205,7 +210,7 @@ static void HashThreadFunc(CMatchFinderMt *mt)
             if (num > kMtHashBlockSize - 2)
               num = kMtHashBlockSize - 2;
             mt->GetHeadsFunc(mf->buffer, mf->pos, mf->hash + mf->fixedHashSize, mf->hashMask, heads + 2, num, mf->crc);
-            heads[0] += num;
+            heads[0] = 2 + num;
           }
           mf->pos += num;
           mf->buffer += num;
@@ -443,13 +448,13 @@ void MatchFinderMt_Construct(CMatchFinderMt *p)
   MtSync_Construct(&p->btSync);
 }
 
-static void MatchFinderMt_FreeMem(CMatchFinderMt *p, ISzAlloc *alloc)
+static void MatchFinderMt_FreeMem(CMatchFinderMt *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->hashBuf);
+  ISzAlloc_Free(alloc, p->hashBuf);
   p->hashBuf = NULL;
 }
 
-void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAlloc *alloc)
+void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAllocPtr alloc)
 {
   MtSync_Destruct(&p->hashSync);
   MtSync_Destruct(&p->btSync);
@@ -472,7 +477,7 @@ static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE BtThreadFunc2(void *p)
 }
 
 SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddBufferBefore,
-    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAlloc *alloc)
+    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAllocPtr alloc)
 {
   CMatchFinder *mf = p->MatchFinder;
   p->historySize = historySize;
@@ -480,7 +485,7 @@ SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddB
     return SZ_ERROR_PARAM;
   if (!p->hashBuf)
   {
-    p->hashBuf = (UInt32 *)alloc->Alloc(alloc, (kHashBufferSize + kBtBufferSize) * sizeof(UInt32));
+    p->hashBuf = (UInt32 *)ISzAlloc_Alloc(alloc, (kHashBufferSize + kBtBufferSize) * sizeof(UInt32));
     if (!p->hashBuf)
       return SZ_ERROR_MEM;
     p->btBuf = p->hashBuf + kHashBufferSize;
@@ -496,13 +501,20 @@ SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddB
 }
 
 /* Call it after ReleaseStream / SetStream */
-void MatchFinderMt_Init(CMatchFinderMt *p)
+static void MatchFinderMt_Init(CMatchFinderMt *p)
 {
   CMatchFinder *mf = p->MatchFinder;
-  p->btBufPos = p->btBufPosLimit = 0;
+  
-  p->hashBufPos = p->hashBufPosLimit = 0;
+  p->btBufPos =
-  MatchFinder_Init(mf);
+  p->btBufPosLimit = 0;
-  p->pointerToCurPos = MatchFinder_GetPointerToCurrentPos(mf);
+  p->hashBufPos =
+  p->hashBufPosLimit = 0;
+
+  /* Init without data reading. We don't want to read data in this thread */
+  MatchFinder_Init_3(mf, False);
+  MatchFinder_Init_LowHash(mf);
+  
+  p->pointerToCurPos = Inline_MatchFinder_GetPointerToCurrentPos(mf);
   p->btNumAvailBytes = 0;
   p->lzPos = p->historySize + 1;
 
@@ -588,10 +600,10 @@ static UInt32 * MixMatches3(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *dista
   MT_HASH3_CALC
 
   curMatch2 = hash[                h2];
-  curMatch3 = hash[kFix3HashSize + h3];
+  curMatch3 = (hash + kFix3HashSize)[h3];
   
   hash[                h2] = lzPos;
-  hash[kFix3HashSize + h3] = lzPos;
+  (hash + kFix3HashSize)[h3] = lzPos;
 
   if (curMatch2 >= matchMinPos && cur[(ptrdiff_t)curMatch2 - lzPos] == cur[0])
   {
@@ -624,12 +636,12 @@ static UInt32 *MixMatches4(CMatchFinderMt *p, UInt32 matchMinPos, UInt32 *distan
   MT_HASH4_CALC
       
   curMatch2 = hash[                h2];
-  curMatch3 = hash[kFix3HashSize + h3];
+  curMatch3 = (hash + kFix3HashSize)[h3];
-  curMatch4 = hash[kFix4HashSize + h4];
+  curMatch4 = (hash + kFix4HashSize)[h4];
   
   hash[                h2] = lzPos;
-  hash[kFix3HashSize + h3] = lzPos;
+  (hash + kFix3HashSize)[h3] = lzPos;
-  hash[kFix4HashSize + h4] = lzPos;
+  (hash + kFix4HashSize)[h4] = lzPos;
 
   if (curMatch2 >= matchMinPos && cur[(ptrdiff_t)curMatch2 - lzPos] == cur[0])
   {
@@ -681,8 +693,12 @@ static UInt32 MatchFinderMt2_GetMatches(CMatchFinderMt *p, UInt32 *distances)
     UInt32 i;
     for (i = 0; i < len; i += 2)
     {
-      *distances++ = *btBuf++;
+      UInt32 v0 = btBuf[0];
-      *distances++ = *btBuf++;
+      UInt32 v1 = btBuf[1];
+      btBuf += 2;
+      distances[0] = v0;
+      distances[1] = v1;
+      distances += 2;
     }
   }
   INCREASE_LZ_POS
@@ -709,8 +725,12 @@ static UInt32 MatchFinderMt_GetMatches(CMatchFinderMt *p, UInt32 *distances)
     distances2 = p->MixMatchesFunc(p, p->lzPos - btBuf[1], distances);
     do
     {
-      *distances2++ = *btBuf++;
+      UInt32 v0 = btBuf[0];
-      *distances2++ = *btBuf++;
+      UInt32 v1 = btBuf[1];
+      btBuf += 2;
+      distances2[0] = v0;
+      distances2[1] = v1;
+      distances2 += 2;
     }
     while ((len -= 2) != 0);
     len = (UInt32)(distances2 - (distances));
@@ -743,7 +763,7 @@ static void MatchFinderMt3_Skip(CMatchFinderMt *p, UInt32 num)
   SKIP_HEADER_MT(3)
       UInt32 h2, h3;
       MT_HASH3_CALC
-      hash[kFix3HashSize + h3] =
+      (hash + kFix3HashSize)[h3] =
       hash[                h2] =
         p->lzPos;
   SKIP_FOOTER_MT
@@ -755,8 +775,8 @@ static void MatchFinderMt4_Skip(CMatchFinderMt *p, UInt32 num)
   SKIP_HEADER_MT(4)
       UInt32 h2, h3, h4;
       MT_HASH4_CALC
-      hash[kFix4HashSize + h4] =
+      (hash + kFix4HashSize)[h4] =
-      hash[kFix3HashSize + h3] =
+      (hash + kFix3HashSize)[h3] =
       hash[                h2] =
         p->lzPos;
   SKIP_FOOTER_MT
@@ -774,7 +794,7 @@ void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder *vTable)
   {
     case 2:
       p->GetHeadsFunc = GetHeads2;
-      p->MixMatchesFunc = (Mf_Mix_Matches)0;
+      p->MixMatchesFunc = (Mf_Mix_Matches)NULL;
       vTable->Skip = (Mf_Skip_Func)MatchFinderMt0_Skip;
       vTable->GetMatches = (Mf_GetMatches_Func)MatchFinderMt2_GetMatches;
       break;

C/LzFindMt.h

@@ -1,5 +1,5 @@
 /* LzFindMt.h -- multithreaded Match finder for LZ algorithms
-2015-05-03 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
 
 #ifndef __LZ_FIND_MT_H
 #define __LZ_FIND_MT_H
@@ -90,9 +90,9 @@ typedef struct _CMatchFinderMt
 } CMatchFinderMt;
 
 void MatchFinderMt_Construct(CMatchFinderMt *p);
-void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAlloc *alloc);
+void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAllocPtr alloc);
 SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddBufferBefore,
-    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAlloc *alloc);
+    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAllocPtr alloc);
 void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder *vTable);
 void MatchFinderMt_ReleaseStream(CMatchFinderMt *p);
 

C/Lzma2Dec.c

@@ -1,5 +1,5 @@
 /* Lzma2Dec.c -- LZMA2 Decoder
-2014-10-29 : Igor Pavlov : Public domain */
+2018-02-19 : Igor Pavlov : Public domain */
 
 /* #define SHOW_DEBUG_INFO */
 
@@ -14,28 +14,22 @@
 #include "Lzma2Dec.h"
 
 /*
-00000000  -  EOS
+00000000  -  End of data
-00000001 U U  -  Uncompressed Reset Dic
+00000001 U U  -  Uncompressed, reset dic, need reset state and set new prop
-00000010 U U  -  Uncompressed No Reset
+00000010 U U  -  Uncompressed, no reset
-100uuuuu U U P P  -  LZMA no reset
+100uuuuu U U P P  -  LZMA, no reset
-101uuuuu U U P P  -  LZMA reset state
+101uuuuu U U P P  -  LZMA, reset state
-110uuuuu U U P P S  -  LZMA reset state + new prop
+110uuuuu U U P P S  -  LZMA, reset state + set new prop
-111uuuuu U U P P S  -  LZMA reset state + new prop + reset dic
+111uuuuu U U P P S  -  LZMA, reset state + set new prop, reset dic
 
   u, U - Unpack Size
   P - Pack Size
   S - Props
 */
 
-#define LZMA2_CONTROL_LZMA (1 << 7)
-#define LZMA2_CONTROL_COPY_NO_RESET 2
 #define LZMA2_CONTROL_COPY_RESET_DIC 1
-#define LZMA2_CONTROL_EOF 0
 
-#define LZMA2_IS_UNCOMPRESSED_STATE(p) (((p)->control & LZMA2_CONTROL_LZMA) == 0)
+#define LZMA2_IS_UNCOMPRESSED_STATE(p) (((p)->control & (1 << 7)) == 0)
-
-#define LZMA2_GET_LZMA_MODE(p) (((p)->control >> 5) & 3)
-#define LZMA2_IS_THERE_PROP(mode) ((mode) >= 2)
 
 #define LZMA2_LCLP_MAX 4
 #define LZMA2_DIC_SIZE_FROM_PROP(p) (((UInt32)2 | ((p) & 1)) << ((p) / 2 + 11))
@@ -74,14 +68,14 @@ static SRes Lzma2Dec_GetOldProps(Byte prop, Byte *props)
   return SZ_OK;
 }
 
-SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAlloc *alloc)
+SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc)
 {
   Byte props[LZMA_PROPS_SIZE];
   RINOK(Lzma2Dec_GetOldProps(prop, props));
   return LzmaDec_AllocateProbs(&p->decoder, props, LZMA_PROPS_SIZE, alloc);
 }
 
-SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAlloc *alloc)
+SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc)
 {
   Byte props[LZMA_PROPS_SIZE];
   RINOK(Lzma2Dec_GetOldProps(prop, props));
@@ -91,9 +85,11 @@ SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAlloc *alloc)
 void Lzma2Dec_Init(CLzma2Dec *p)
 {
   p->state = LZMA2_STATE_CONTROL;
-  p->needInitDic = True;
+  p->needInitLevel = 0xE0;
-  p->needInitState = True;
+  p->isExtraMode = False;
-  p->needInitProp = True;
+  p->unpackSize = 0;
+  
+  // p->decoder.dicPos = 0; // we can use it instead of full init
   LzmaDec_Init(&p->decoder);
 }
 
@@ -102,19 +98,26 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
   switch (p->state)
   {
     case LZMA2_STATE_CONTROL:
+      p->isExtraMode = False;
       p->control = b;
-      PRF(printf("\n %4X ", p->decoder.dicPos));
+      PRF(printf("\n %8X", (unsigned)p->decoder.dicPos));
-      PRF(printf(" %2X", b));
+      PRF(printf(" %02X", (unsigned)b));
-      if (p->control == 0)
+      if (b == 0)
         return LZMA2_STATE_FINISHED;
       if (LZMA2_IS_UNCOMPRESSED_STATE(p))
       {
-        if ((p->control & 0x7F) > 2)
+        if (b == LZMA2_CONTROL_COPY_RESET_DIC)
+          p->needInitLevel = 0xC0;
+        else if (b > 2 || p->needInitLevel == 0xE0)
           return LZMA2_STATE_ERROR;
-        p->unpackSize = 0;
       }
       else
-        p->unpackSize = (UInt32)(p->control & 0x1F) << 16;
+      {
+        if (b < p->needInitLevel)
+          return LZMA2_STATE_ERROR;
+        p->needInitLevel = 0;
+        p->unpackSize = (UInt32)(b & 0x1F) << 16;
+      }
       return LZMA2_STATE_UNPACK0;
     
     case LZMA2_STATE_UNPACK0:
@@ -124,8 +127,8 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
     case LZMA2_STATE_UNPACK1:
       p->unpackSize |= (UInt32)b;
       p->unpackSize++;
-      PRF(printf(" %8d", p->unpackSize));
+      PRF(printf(" %7u", (unsigned)p->unpackSize));
-      return (LZMA2_IS_UNCOMPRESSED_STATE(p)) ? LZMA2_STATE_DATA : LZMA2_STATE_PACK0;
+      return LZMA2_IS_UNCOMPRESSED_STATE(p) ? LZMA2_STATE_DATA : LZMA2_STATE_PACK0;
     
     case LZMA2_STATE_PACK0:
       p->packSize = (UInt32)b << 8;
@@ -134,9 +137,9 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
     case LZMA2_STATE_PACK1:
       p->packSize |= (UInt32)b;
       p->packSize++;
-      PRF(printf(" %8d", p->packSize));
+      // if (p->packSize < 5) return LZMA2_STATE_ERROR;
-      return LZMA2_IS_THERE_PROP(LZMA2_GET_LZMA_MODE(p)) ? LZMA2_STATE_PROP:
+      PRF(printf(" %5u", (unsigned)p->packSize));
-        (p->needInitProp ? LZMA2_STATE_ERROR : LZMA2_STATE_DATA);
+      return (p->control & 0x40) ? LZMA2_STATE_PROP : LZMA2_STATE_DATA;
 
     case LZMA2_STATE_PROP:
     {
@@ -145,13 +148,12 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
         return LZMA2_STATE_ERROR;
       lc = b % 9;
       b /= 9;
-      p->decoder.prop.pb = b / 5;
+      p->decoder.prop.pb = (Byte)(b / 5);
       lp = b % 5;
       if (lc + lp > LZMA2_LCLP_MAX)
         return LZMA2_STATE_ERROR;
-      p->decoder.prop.lc = lc;
+      p->decoder.prop.lc = (Byte)lc;
-      p->decoder.prop.lp = lp;
+      p->decoder.prop.lp = (Byte)lp;
-      p->needInitProp = False;
       return LZMA2_STATE_DATA;
     }
   }
@@ -169,6 +171,7 @@ static void LzmaDec_UpdateWithUncompressed(CLzmaDec *p, const Byte *src, SizeT s
 
 void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState);
 
+
 SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
     const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
 {
@@ -176,12 +179,17 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
   *srcLen = 0;
   *status = LZMA_STATUS_NOT_SPECIFIED;
 
-  while (p->state != LZMA2_STATE_FINISHED)
+  while (p->state != LZMA2_STATE_ERROR)
   {
-    SizeT dicPos = p->decoder.dicPos;
+    SizeT dicPos;
+
+    if (p->state == LZMA2_STATE_FINISHED)
+    {
+      *status = LZMA_STATUS_FINISHED_WITH_MARK;
+      return SZ_OK;
+    }
     
-    if (p->state == LZMA2_STATE_ERROR)
+    dicPos = p->decoder.dicPos;
-      return SZ_ERROR_DATA;
     
     if (dicPos == dicLimit && finishMode == LZMA_FINISH_ANY)
     {
@@ -198,29 +206,25 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
       }
       (*srcLen)++;
       p->state = Lzma2Dec_UpdateState(p, *src++);
-
       if (dicPos == dicLimit && p->state != LZMA2_STATE_FINISHED)
-      {
+        break;
-        p->state = LZMA2_STATE_ERROR;
-        return SZ_ERROR_DATA;
-      }
       continue;
     }
     
     {
-      SizeT destSizeCur = dicLimit - dicPos;
+      SizeT inCur = inSize - *srcLen;
-      SizeT srcSizeCur = inSize - *srcLen;
+      SizeT outCur = dicLimit - dicPos;
       ELzmaFinishMode curFinishMode = LZMA_FINISH_ANY;
       
-      if (p->unpackSize <= destSizeCur)
+      if (outCur >= p->unpackSize)
       {
-        destSizeCur = (SizeT)p->unpackSize;
+        outCur = (SizeT)p->unpackSize;
         curFinishMode = LZMA_FINISH_END;
       }
 
       if (LZMA2_IS_UNCOMPRESSED_STATE(p))
       {
-        if (*srcLen == inSize)
+        if (inCur == 0)
         {
           *status = LZMA_STATUS_NEEDS_MORE_INPUT;
           return SZ_OK;
@@ -229,134 +233,240 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
         if (p->state == LZMA2_STATE_DATA)
         {
           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);
         }
 
-        if (srcSizeCur > destSizeCur)
+        if (inCur > outCur)
-          srcSizeCur = destSizeCur;
+          inCur = outCur;
+        if (inCur == 0)
+          break;
 
-        if (srcSizeCur == 0)
+        LzmaDec_UpdateWithUncompressed(&p->decoder, src, inCur);
-        {
-          p->state = LZMA2_STATE_ERROR;
-          return SZ_ERROR_DATA;
-        }
 
-        LzmaDec_UpdateWithUncompressed(&p->decoder, src, srcSizeCur);
+        src += inCur;
-
+        *srcLen += inCur;
-        src += srcSizeCur;
+        p->unpackSize -= (UInt32)inCur;
-        *srcLen += srcSizeCur;
-        p->unpackSize -= (UInt32)srcSizeCur;
         p->state = (p->unpackSize == 0) ? LZMA2_STATE_CONTROL : LZMA2_STATE_DATA_CONT;
       }
       else
       {
-        SizeT outSizeProcessed;
         SRes res;
 
         if (p->state == LZMA2_STATE_DATA)
         {
-          unsigned mode = LZMA2_GET_LZMA_MODE(p);
+          Bool initDic = (p->control >= 0xE0);
-          Bool initDic = (mode == 3);
+          Bool initState = (p->control >= 0xA0);
-          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)
+        if (inCur > p->packSize)
-          srcSizeCur = (SizeT)p->packSize;
+          inCur = (SizeT)p->packSize;
-          
-        res = LzmaDec_DecodeToDic(&p->decoder, dicPos + destSizeCur, src, &srcSizeCur, curFinishMode, status);
         
-        src += srcSizeCur;
+        res = LzmaDec_DecodeToDic(&p->decoder, dicPos + outCur, src, &inCur, curFinishMode, status);
-        *srcLen += srcSizeCur;
-        p->packSize -= (UInt32)srcSizeCur;
 
-        outSizeProcessed = p->decoder.dicPos - dicPos;
+        src += inCur;
-        p->unpackSize -= (UInt32)outSizeProcessed;
+        *srcLen += inCur;
+        p->packSize -= (UInt32)inCur;
+        outCur = p->decoder.dicPos - dicPos;
+        p->unpackSize -= (UInt32)outCur;
 
-        RINOK(res);
+        if (res != 0)
+          break;
+        
         if (*status == LZMA_STATUS_NEEDS_MORE_INPUT)
-          return res;
+        {
+          if (p->packSize == 0)
+            break;
+          return SZ_OK;
+        }
 
-        if (srcSizeCur == 0 && outSizeProcessed == 0)
+        if (inCur == 0 && outCur == 0)
         {
           if (*status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK
               || p->unpackSize != 0
               || p->packSize != 0)
-          {
+            break;
-            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_SPECIFIED;
-          *status = LZMA_STATUS_NOT_FINISHED;
       }
     }
   }
   
-  *status = LZMA_STATUS_FINISHED_WITH_MARK;
+  *status = LZMA_STATUS_NOT_SPECIFIED;
-  return SZ_OK;
+  p->state = LZMA2_STATE_ERROR;
+  return SZ_ERROR_DATA;
 }
 
+
+
+
+ELzma2ParseStatus Lzma2Dec_Parse(CLzma2Dec *p,
+    SizeT outSize,
+    const Byte *src, SizeT *srcLen,
+    int checkFinishBlock)
+{
+  SizeT inSize = *srcLen;
+  *srcLen = 0;
+
+  while (p->state != LZMA2_STATE_ERROR)
+  {
+    if (p->state == LZMA2_STATE_FINISHED)
+      return LZMA_STATUS_FINISHED_WITH_MARK;
+
+    if (outSize == 0 && !checkFinishBlock)
+      return LZMA_STATUS_NOT_FINISHED;
+    
+    if (p->state != LZMA2_STATE_DATA && p->state != LZMA2_STATE_DATA_CONT)
+    {
+      if (*srcLen == inSize)
+        return LZMA_STATUS_NEEDS_MORE_INPUT;
+      (*srcLen)++;
+
+      p->state = Lzma2Dec_UpdateState(p, *src++);
+
+      if (p->state == LZMA2_STATE_UNPACK0)
+      {
+        // if (p->decoder.dicPos != 0)
+        if (p->control == LZMA2_CONTROL_COPY_RESET_DIC || p->control >= 0xE0)
+          return LZMA2_PARSE_STATUS_NEW_BLOCK;
+        // if (outSize == 0) return LZMA_STATUS_NOT_FINISHED;
+      }
+
+      // The following code can be commented.
+      // It's not big problem, if we read additional input bytes.
+      // It will be stopped later in LZMA2_STATE_DATA / LZMA2_STATE_DATA_CONT state.
+
+      if (outSize == 0 && p->state != LZMA2_STATE_FINISHED)
+      {
+        // checkFinishBlock is true. So we expect that block must be finished,
+        // We can return LZMA_STATUS_NOT_SPECIFIED or LZMA_STATUS_NOT_FINISHED here
+        // break;
+        return LZMA_STATUS_NOT_FINISHED;
+      }
+
+      if (p->state == LZMA2_STATE_DATA)
+        return LZMA2_PARSE_STATUS_NEW_CHUNK;
+
+      continue;
+    }
+
+    if (outSize == 0)
+      return LZMA_STATUS_NOT_FINISHED;
+
+    {
+      SizeT inCur = inSize - *srcLen;
+
+      if (LZMA2_IS_UNCOMPRESSED_STATE(p))
+      {
+        if (inCur == 0)
+          return LZMA_STATUS_NEEDS_MORE_INPUT;
+        if (inCur > p->unpackSize)
+          inCur = p->unpackSize;
+        if (inCur > outSize)
+          inCur = outSize;
+        p->decoder.dicPos += inCur;
+        src += inCur;
+        *srcLen += inCur;
+        outSize -= inCur;
+        p->unpackSize -= (UInt32)inCur;
+        p->state = (p->unpackSize == 0) ? LZMA2_STATE_CONTROL : LZMA2_STATE_DATA_CONT;
+      }
+      else
+      {
+        p->isExtraMode = True;
+
+        if (inCur == 0)
+        {
+          if (p->packSize != 0)
+            return LZMA_STATUS_NEEDS_MORE_INPUT;
+        }
+        else if (p->state == LZMA2_STATE_DATA)
+        {
+          p->state = LZMA2_STATE_DATA_CONT;
+          if (*src != 0)
+          {
+            // first byte of lzma chunk must be Zero
+            *srcLen += 1;
+            p->packSize--;
+            break;
+          }
+        }
+  
+        if (inCur > p->packSize)
+          inCur = (SizeT)p->packSize;
+
+        src += inCur;
+        *srcLen += inCur;
+        p->packSize -= (UInt32)inCur;
+
+        if (p->packSize == 0)
+        {
+          SizeT rem = outSize;
+          if (rem > p->unpackSize)
+            rem = p->unpackSize;
+          p->decoder.dicPos += rem;
+          p->unpackSize -= (UInt32)rem;
+          outSize -= rem;
+          if (p->unpackSize == 0)
+            p->state = LZMA2_STATE_CONTROL;
+        }
+      }
+    }
+  }
+  
+  p->state = LZMA2_STATE_ERROR;
+  return LZMA_STATUS_NOT_SPECIFIED;
+}
+
+
+
+
 SRes Lzma2Dec_DecodeToBuf(CLzma2Dec *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
 {
   SizeT outSize = *destLen, inSize = *srcLen;
   *srcLen = *destLen = 0;
+  
   for (;;)
   {
-    SizeT srcSizeCur = inSize, outSizeCur, dicPos;
+    SizeT inCur = inSize, outCur, dicPos;
     ELzmaFinishMode curFinishMode;
     SRes res;
+    
     if (p->decoder.dicPos == p->decoder.dicBufSize)
       p->decoder.dicPos = 0;
     dicPos = p->decoder.dicPos;
-    if (outSize > p->decoder.dicBufSize - dicPos)
+    curFinishMode = LZMA_FINISH_ANY;
+    outCur = p->decoder.dicBufSize - dicPos;
+    
+    if (outCur >= outSize)
     {
-      outSizeCur = p->decoder.dicBufSize;
+      outCur = outSize;
-      curFinishMode = LZMA_FINISH_ANY;
-    }
-    else
-    {
-      outSizeCur = dicPos + outSize;
       curFinishMode = finishMode;
     }
 
-    res = Lzma2Dec_DecodeToDic(p, outSizeCur, src, &srcSizeCur, curFinishMode, status);
+    res = Lzma2Dec_DecodeToDic(p, dicPos + outCur, src, &inCur, curFinishMode, status);
-    src += srcSizeCur;
+    
-    inSize -= srcSizeCur;
+    src += inCur;
-    *srcLen += srcSizeCur;
+    inSize -= inCur;
-    outSizeCur = p->decoder.dicPos - dicPos;
+    *srcLen += inCur;
-    memcpy(dest, p->decoder.dic + dicPos, outSizeCur);
+    outCur = p->decoder.dicPos - dicPos;
-    dest += outSizeCur;
+    memcpy(dest, p->decoder.dic + dicPos, outCur);
-    outSize -= outSizeCur;
+    dest += outCur;
-    *destLen += outSizeCur;
+    outSize -= outCur;
+    *destLen += outCur;
     if (res != 0)
       return res;
-    if (outSizeCur == 0 || outSize == 0)
+    if (outCur == 0 || outSize == 0)
       return SZ_OK;
   }
 }
 
+
 SRes Lzma2Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
-    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAlloc *alloc)
+    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAllocPtr alloc)
 {
   CLzma2Dec p;
   SRes res;

C/Lzma2Dec.h

@@ -1,5 +1,5 @@
 /* Lzma2Dec.h -- LZMA2 Decoder
-2015-05-13 : Igor Pavlov : Public domain */
+2018-02-19 : Igor Pavlov : Public domain */
 
 #ifndef __LZMA2_DEC_H
 #define __LZMA2_DEC_H
@@ -12,25 +12,24 @@ EXTERN_C_BEGIN
 
 typedef struct
 {
-  CLzmaDec decoder;
-  UInt32 packSize;
-  UInt32 unpackSize;
   unsigned state;
   Byte control;
-  Bool needInitDic;
+  Byte needInitLevel;
-  Bool needInitState;
+  Byte isExtraMode;
-  Bool needInitProp;
+  Byte _pad_;
+  UInt32 packSize;
+  UInt32 unpackSize;
+  CLzmaDec decoder;
 } CLzma2Dec;
 
 #define Lzma2Dec_Construct(p) LzmaDec_Construct(&(p)->decoder)
-#define Lzma2Dec_FreeProbs(p, alloc) LzmaDec_FreeProbs(&(p)->decoder, alloc);
+#define Lzma2Dec_FreeProbs(p, alloc) LzmaDec_FreeProbs(&(p)->decoder, alloc)
-#define Lzma2Dec_Free(p, alloc) LzmaDec_Free(&(p)->decoder, alloc);
+#define Lzma2Dec_Free(p, alloc) LzmaDec_Free(&(p)->decoder, alloc)
 
-SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAlloc *alloc);
+SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc);
-SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAlloc *alloc);
+SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc);
 void Lzma2Dec_Init(CLzma2Dec *p);
 
-
 /*
 finishMode:
   It has meaning only if the decoding reaches output limit (*destLen or dicLimit).
@@ -53,6 +52,47 @@ SRes Lzma2Dec_DecodeToBuf(CLzma2Dec *p, Byte *dest, SizeT *destLen,
     const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status);
 
 
+/* ---------- LZMA2 block and chunk parsing ---------- */
+
+/*
+Lzma2Dec_Parse() parses compressed data stream up to next independent block or next chunk data.
+It can return LZMA_STATUS_* code or LZMA2_PARSE_STATUS_* code:
+  - LZMA2_PARSE_STATUS_NEW_BLOCK - there is new block, and 1 additional byte (control byte of next block header) was read from input.
+  - LZMA2_PARSE_STATUS_NEW_CHUNK - there is new chunk, and only lzma2 header of new chunk was read.
+                                   CLzma2Dec::unpackSize contains unpack size of that chunk
+*/
+
+typedef enum
+{
+/*
+  LZMA_STATUS_NOT_SPECIFIED                 // data error
+  LZMA_STATUS_FINISHED_WITH_MARK
+  LZMA_STATUS_NOT_FINISHED                  //
+  LZMA_STATUS_NEEDS_MORE_INPUT
+  LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK   // unused
+*/
+  LZMA2_PARSE_STATUS_NEW_BLOCK = LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK + 1,
+  LZMA2_PARSE_STATUS_NEW_CHUNK
+} ELzma2ParseStatus;
+
+ELzma2ParseStatus Lzma2Dec_Parse(CLzma2Dec *p,
+    SizeT outSize,   // output size
+    const Byte *src, SizeT *srcLen,
+    int checkFinishBlock   // set (checkFinishBlock = 1), if it must read full input data, if decoder.dicPos reaches blockMax position.
+    );
+
+/*
+LZMA2 parser doesn't decode LZMA chunks, so we must read
+  full input LZMA chunk to decode some part of LZMA chunk.
+
+Lzma2Dec_GetUnpackExtra() returns the value that shows
+    max possible number of output bytes that can be output by decoder
+    at current input positon.
+*/
+
+#define Lzma2Dec_GetUnpackExtra(p) ((p)->isExtraMode ? (p)->unpackSize : 0);
+
+
 /* ---------- One Call Interface ---------- */
 
 /*
@@ -73,7 +113,7 @@ Returns:
 */
 
 SRes Lzma2Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
-    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAlloc *alloc);
+    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAllocPtr alloc);
 
 EXTERN_C_END
 

C/Lzma2DecMt.h

@@ -0,0 +1,79 @@
+/* Lzma2DecMt.h -- LZMA2 Decoder Multi-thread
+2018-02-17 : Igor Pavlov : Public domain */
+
+#ifndef __LZMA2_DEC_MT_H
+#define __LZMA2_DEC_MT_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+typedef struct
+{
+  size_t inBufSize_ST;
+  size_t outStep_ST;
+  
+  #ifndef _7ZIP_ST
+  unsigned numThreads;
+  size_t inBufSize_MT;
+  size_t outBlockMax;
+  size_t inBlockMax;
+  #endif
+} CLzma2DecMtProps;
+
+/* init to single-thread mode */
+void Lzma2DecMtProps_Init(CLzma2DecMtProps *p);
+
+
+/* ---------- CLzma2DecMtHandle Interface ---------- */
+
+/* Lzma2DecMt_ * functions can return the following exit codes:
+SRes:
+  SZ_OK           - OK
+  SZ_ERROR_MEM    - Memory allocation error
+  SZ_ERROR_PARAM  - Incorrect paramater in props
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  // SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
+  SZ_ERROR_PROGRESS - some break from progress callback
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
+*/
+
+typedef void * CLzma2DecMtHandle;
+
+CLzma2DecMtHandle Lzma2DecMt_Create(ISzAllocPtr alloc, ISzAllocPtr allocMid);
+void Lzma2DecMt_Destroy(CLzma2DecMtHandle p);
+
+SRes Lzma2DecMt_Decode(CLzma2DecMtHandle p,
+    Byte prop,
+    const CLzma2DecMtProps *props,
+    ISeqOutStream *outStream,
+    const UInt64 *outDataSize, // NULL means undefined
+    int finishMode,            // 0 - partial unpacking is allowed, 1 - if lzma2 stream must be finished
+    // Byte *outBuf, size_t *outBufSize,
+    ISeqInStream *inStream,
+    // const Byte *inData, size_t inDataSize,
+    
+    // out variables:
+    UInt64 *inProcessed,
+    int *isMT,  /* out: (*isMT == 0), if single thread decoding was used */
+
+    // UInt64 *outProcessed,
+    ICompressProgress *progress);
+
+
+/* ---------- Read from CLzma2DecMtHandle Interface ---------- */
+
+SRes Lzma2DecMt_Init(CLzma2DecMtHandle pp,
+    Byte prop,
+    const CLzma2DecMtProps *props,
+    const UInt64 *outDataSize, int finishMode,
+    ISeqInStream *inStream);
+
+SRes Lzma2DecMt_Read(CLzma2DecMtHandle pp,
+    Byte *data, size_t *outSize,
+    UInt64 *inStreamProcessed);
+
+
+EXTERN_C_END
+
+#endif

C/Lzma2Enc.c

@@ -1,9 +1,8 @@
 /* Lzma2Enc.c -- LZMA2 Encoder
-2012-06-19 : Igor Pavlov : Public domain */
+2018-04-27 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
-/* #include <stdio.h> */
 #include <string.h>
 
 /* #define _7ZIP_ST */
@@ -13,7 +12,7 @@
 #ifndef _7ZIP_ST
 #include "MtCoder.h"
 #else
-#define NUM_MT_CODER_THREADS_MAX 1
+#define MTCODER__THREADS_MAX 1
 #endif
 
 #define LZMA2_CONTROL_LZMA (1 << 7)
@@ -35,34 +34,87 @@
 
 #define PRF(x) /* x */
 
+
+/* ---------- CLimitedSeqInStream ---------- */
+
+typedef struct
+{
+  ISeqInStream vt;
+  ISeqInStream *realStream;
+  UInt64 limit;
+  UInt64 processed;
+  int finished;
+} CLimitedSeqInStream;
+
+static void LimitedSeqInStream_Init(CLimitedSeqInStream *p)
+{
+  p->limit = (UInt64)(Int64)-1;
+  p->processed = 0;
+  p->finished = 0;
+}
+
+static SRes LimitedSeqInStream_Read(const ISeqInStream *pp, void *data, size_t *size)
+{
+  CLimitedSeqInStream *p = CONTAINER_FROM_VTBL(pp, CLimitedSeqInStream, vt);
+  size_t size2 = *size;
+  SRes res = SZ_OK;
+  
+  if (p->limit != (UInt64)(Int64)-1)
+  {
+    UInt64 rem = p->limit - p->processed;
+    if (size2 > rem)
+      size2 = (size_t)rem;
+  }
+  if (size2 != 0)
+  {
+    res = ISeqInStream_Read(p->realStream, data, &size2);
+    p->finished = (size2 == 0 ? 1 : 0);
+    p->processed += size2;
+  }
+  *size = size2;
+  return res;
+}
+
+
 /* ---------- CLzma2EncInt ---------- */
 
 typedef struct
 {
   CLzmaEncHandle enc;
+  Byte propsAreSet;
+  Byte propsByte;
+  Byte needInitState;
+  Byte needInitProp;
   UInt64 srcPos;
-  Byte props;
-  Bool needInitState;
-  Bool needInitProp;
 } CLzma2EncInt;
 
-static SRes Lzma2EncInt_Init(CLzma2EncInt *p, const CLzma2EncProps *props)
+
+static SRes Lzma2EncInt_InitStream(CLzma2EncInt *p, const CLzma2EncProps *props)
 {
-  Byte propsEncoded[LZMA_PROPS_SIZE];
+  if (!p->propsAreSet)
-  SizeT propsSize = LZMA_PROPS_SIZE;
+  {
-  RINOK(LzmaEnc_SetProps(p->enc, &props->lzmaProps));
+    SizeT propsSize = LZMA_PROPS_SIZE;
-  RINOK(LzmaEnc_WriteProperties(p->enc, propsEncoded, &propsSize));
+    Byte propsEncoded[LZMA_PROPS_SIZE];
-  p->srcPos = 0;
+    RINOK(LzmaEnc_SetProps(p->enc, &props->lzmaProps));
-  p->props = propsEncoded[0];
+    RINOK(LzmaEnc_WriteProperties(p->enc, propsEncoded, &propsSize));
-  p->needInitState = True;
+    p->propsByte = propsEncoded[0];
-  p->needInitProp = True;
+    p->propsAreSet = True;
+  }
   return SZ_OK;
 }
 
+static void Lzma2EncInt_InitBlock(CLzma2EncInt *p)
+{
+  p->srcPos = 0;
+  p->needInitState = True;
+  p->needInitProp = True;
+}
+
+
 SRes LzmaEnc_PrepareForLzma2(CLzmaEncHandle pp, ISeqInStream *inStream, UInt32 keepWindowSize,
-    ISzAlloc *alloc, ISzAlloc *allocBig);
+    ISzAllocPtr alloc, ISzAllocPtr allocBig);
 SRes LzmaEnc_MemPrepare(CLzmaEncHandle pp, const Byte *src, SizeT srcLen,
-    UInt32 keepWindowSize, ISzAlloc *alloc, ISzAlloc *allocBig);
+    UInt32 keepWindowSize, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 SRes LzmaEnc_CodeOneMemBlock(CLzmaEncHandle pp, Bool reInit,
     Byte *dest, size_t *destLen, UInt32 desiredPackSize, UInt32 *unpackSize);
 const Byte *LzmaEnc_GetCurBuf(CLzmaEncHandle pp);
@@ -70,6 +122,9 @@ void LzmaEnc_Finish(CLzmaEncHandle pp);
 void LzmaEnc_SaveState(CLzmaEncHandle pp);
 void LzmaEnc_RestoreState(CLzmaEncHandle pp);
 
+/*
+UInt32 LzmaEnc_GetNumAvailableBytes(CLzmaEncHandle pp);
+*/
 
 static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
     size_t *packSizeRes, ISeqOutStream *outStream)
@@ -109,6 +164,7 @@ static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
   {
     size_t destPos = 0;
     PRF(printf("################# COPY           "));
+
     while (unpackSize > 0)
     {
       UInt32 u = (unpackSize < LZMA2_COPY_CHUNK_SIZE) ? unpackSize : LZMA2_COPY_CHUNK_SIZE;
@@ -121,10 +177,11 @@ static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
       unpackSize -= u;
       destPos += u;
       p->srcPos += u;
+      
       if (outStream)
       {
         *packSizeRes += destPos;
-        if (outStream->Write(outStream, outBuf, destPos) != destPos)
+        if (ISeqOutStream_Write(outStream, outBuf, destPos) != destPos)
           return SZ_ERROR_WRITE;
         destPos = 0;
       }
@@ -132,9 +189,11 @@ static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
         *packSizeRes = destPos;
       /* needInitState = True; */
     }
+    
     LzmaEnc_RestoreState(p->enc);
     return SZ_OK;
   }
+
   {
     size_t destPos = 0;
     UInt32 u = unpackSize - 1;
@@ -150,7 +209,7 @@ static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
     outBuf[destPos++] = (Byte)pm;
     
     if (p->needInitProp)
-      outBuf[destPos++] = p->props;
+      outBuf[destPos++] = p->propsByte;
     
     p->needInitProp = False;
     p->needInitState = False;
@@ -158,26 +217,30 @@ static SRes Lzma2EncInt_EncodeSubblock(CLzma2EncInt *p, Byte *outBuf,
     p->srcPos += unpackSize;
 
     if (outStream)
-      if (outStream->Write(outStream, outBuf, destPos) != destPos)
+      if (ISeqOutStream_Write(outStream, outBuf, destPos) != destPos)
         return SZ_ERROR_WRITE;
+    
     *packSizeRes = destPos;
     return SZ_OK;
   }
 }
 
+
 /* ---------- Lzma2 Props ---------- */
 
 void Lzma2EncProps_Init(CLzma2EncProps *p)
 {
   LzmaEncProps_Init(&p->lzmaProps);
+  p->blockSize = LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO;
+  p->numBlockThreads_Reduced = -1;
+  p->numBlockThreads_Max = -1;
   p->numTotalThreads = -1;
-  p->numBlockThreads = -1;
-  p->blockSize = 0;
 }
 
 void Lzma2EncProps_Normalize(CLzma2EncProps *p)
 {
-  int t1, t1n, t2, t3;
+  UInt64 fileSize;
+  int t1, t1n, t2, t2r, t3;
   {
     CLzmaEncProps lzmaProps = p->lzmaProps;
     LzmaEncProps_Normalize(&lzmaProps);
@@ -185,11 +248,11 @@ void Lzma2EncProps_Normalize(CLzma2EncProps *p)
   }
 
   t1 = p->lzmaProps.numThreads;
-  t2 = p->numBlockThreads;
+  t2 = p->numBlockThreads_Max;
   t3 = p->numTotalThreads;
 
-  if (t2 > NUM_MT_CODER_THREADS_MAX)
+  if (t2 > MTCODER__THREADS_MAX)
-    t2 = NUM_MT_CODER_THREADS_MAX;
+    t2 = MTCODER__THREADS_MAX;
 
   if (t3 <= 0)
   {
@@ -205,8 +268,8 @@ void Lzma2EncProps_Normalize(CLzma2EncProps *p)
       t1 = 1;
       t2 = t3;
     }
-    if (t2 > NUM_MT_CODER_THREADS_MAX)
+    if (t2 > MTCODER__THREADS_MAX)
-      t2 = NUM_MT_CODER_THREADS_MAX;
+      t2 = MTCODER__THREADS_MAX;
   }
   else if (t1 <= 0)
   {
@@ -219,211 +282,186 @@ void Lzma2EncProps_Normalize(CLzma2EncProps *p)
 
   p->lzmaProps.numThreads = t1;
 
+  t2r = t2;
+
+  fileSize = p->lzmaProps.reduceSize;
+
+  if (   p->blockSize != LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID
+      && p->blockSize != LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO
+      && (p->blockSize < fileSize || fileSize == (UInt64)(Int64)-1))
+    p->lzmaProps.reduceSize = p->blockSize;
+
   LzmaEncProps_Normalize(&p->lzmaProps);
 
-  if (p->blockSize == 0)
+  p->lzmaProps.reduceSize = fileSize;
+
+  t1 = p->lzmaProps.numThreads;
+
+  if (p->blockSize == LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID)
   {
-    UInt32 dictSize = p->lzmaProps.dictSize;
+    t2r = t2 = 1;
-    UInt64 blockSize = (UInt64)dictSize << 2;
+    t3 = t1;
-    const UInt32 kMinSize = (UInt32)1 << 20;
-    const UInt32 kMaxSize = (UInt32)1 << 28;
-    if (blockSize < kMinSize) blockSize = kMinSize;
-    if (blockSize > kMaxSize) blockSize = kMaxSize;
-    if (blockSize < dictSize) blockSize = dictSize;
-    p->blockSize = (size_t)blockSize;
   }
-  if (t2 > 1)
+  else if (p->blockSize == LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO && t2 <= 1)
   {
-    UInt64 temp = p->lzmaProps.reduceSize + p->blockSize - 1;
+    /* if there is no block multi-threading, we use SOLID block */
-    if (temp > p->lzmaProps.reduceSize)
+    p->blockSize = LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID;
+  }
+  else
+  {
+    if (p->blockSize == LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO)
     {
-      UInt64 numBlocks = temp / p->blockSize;
+      const UInt32 kMinSize = (UInt32)1 << 20;
+      const UInt32 kMaxSize = (UInt32)1 << 28;
+      const UInt32 dictSize = p->lzmaProps.dictSize;
+      UInt64 blockSize = (UInt64)dictSize << 2;
+      if (blockSize < kMinSize) blockSize = kMinSize;
+      if (blockSize > kMaxSize) blockSize = kMaxSize;
+      if (blockSize < dictSize) blockSize = dictSize;
+      blockSize += (kMinSize - 1);
+      blockSize &= ~(UInt64)(kMinSize - 1);
+      p->blockSize = blockSize;
+    }
+    
+    if (t2 > 1 && fileSize != (UInt64)(Int64)-1)
+    {
+      UInt64 numBlocks = fileSize / p->blockSize;
+      if (numBlocks * p->blockSize != fileSize)
+        numBlocks++;
       if (numBlocks < (unsigned)t2)
       {
-        t2 = (unsigned)numBlocks;
+        t2r = (unsigned)numBlocks;
-        t3 = t1 * t2;
+        if (t2r == 0)
+          t2r = 1;
+        t3 = t1 * t2r;
       }
     }
   }
-  p->numBlockThreads = t2;
+  
+  p->numBlockThreads_Max = t2;
+  p->numBlockThreads_Reduced = t2r;
   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;
+  return (p && ICompressProgress_Progress(p, inSize, outSize) != SZ_OK) ? SZ_ERROR_PROGRESS : SZ_OK;
 }
 
+
 /* ---------- Lzma2 ---------- */
 
 typedef struct
 {
   Byte propEncoded;
   CLzma2EncProps props;
+  UInt64 expectedDataSize;
   
-  Byte *outBuf;
+  Byte *tempBufLzma;
 
-  ISzAlloc *alloc;
+  ISzAllocPtr alloc;
-  ISzAlloc *allocBig;
+  ISzAllocPtr allocBig;
 
-  CLzma2EncInt coders[NUM_MT_CODER_THREADS_MAX];
+  CLzma2EncInt coders[MTCODER__THREADS_MAX];
 
   #ifndef _7ZIP_ST
+  
+  ISeqOutStream *outStream;
+  Byte *outBuf;
+  size_t outBuf_Rem;   /* remainder in outBuf */
+
+  size_t outBufSize;   /* size of allocated outBufs[i] */
+  size_t outBufsDataSizes[MTCODER__BLOCKS_MAX];
+  Bool mtCoder_WasConstructed;
   CMtCoder mtCoder;
+  Byte *outBufs[MTCODER__BLOCKS_MAX];
+
   #endif
 
 } CLzma2Enc;
 
 
-/* ---------- Lzma2EncThread ---------- */
 
-static SRes Lzma2Enc_EncodeMt1(CLzma2EncInt *p, CLzma2Enc *mainEncoder,
+CLzma2EncHandle Lzma2Enc_Create(ISzAllocPtr alloc, ISzAllocPtr allocBig)
-  ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress)
 {
-  UInt64 packTotal = 0;
+  CLzma2Enc *p = (CLzma2Enc *)ISzAlloc_Alloc(alloc, sizeof(CLzma2Enc));
-  SRes res = SZ_OK;
+  if (!p)
-
-  if (mainEncoder->outBuf == 0)
-  {
-    mainEncoder->outBuf = (Byte *)IAlloc_Alloc(mainEncoder->alloc, LZMA2_CHUNK_SIZE_COMPRESSED_MAX);
-    if (mainEncoder->outBuf == 0)
-      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;
-    res = Lzma2EncInt_EncodeSubblock(p, mainEncoder->outBuf, &packSize, outStream);
-    if (res != SZ_OK)
-      break;
-    packTotal += packSize;
-    res = Progress(progress, p->srcPos, packTotal);
-    if (res != SZ_OK)
-      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
-
-typedef struct
-{
-  IMtCoderCallback funcTable;
-  CLzma2Enc *lzma2Enc;
-} CMtCallbackImp;
-
-static SRes MtCallbackImp_Code(void *pp, unsigned index, Byte *dest, size_t *destSize,
-      const Byte *src, size_t srcSize, int finished)
-{
-  CMtCallbackImp *imp = (CMtCallbackImp *)pp;
-  CLzma2Enc *mainEncoder = imp->lzma2Enc;
-  CLzma2EncInt *p = &mainEncoder->coders[index];
-
-  SRes res = SZ_OK;
-  {
-    size_t destLim = *destSize;
-    *destSize = 0;
-
-    if (srcSize != 0)
-    {
-      RINOK(Lzma2EncInt_Init(p, &mainEncoder->props));
-     
-      RINOK(LzmaEnc_MemPrepare(p->enc, src, srcSize, LZMA2_KEEP_WINDOW_SIZE,
-          mainEncoder->alloc, mainEncoder->allocBig));
-     
-      while (p->srcPos < srcSize)
-      {
-        size_t packSize = destLim - *destSize;
-        res = Lzma2EncInt_EncodeSubblock(p, dest + *destSize, &packSize, NULL);
-        if (res != SZ_OK)
-          break;
-        *destSize += packSize;
-
-        if (packSize == 0)
-        {
-          res = SZ_ERROR_FAIL;
-          break;
-        }
-
-        if (MtProgress_Set(&mainEncoder->mtCoder.mtProgress, index, p->srcPos, *destSize) != SZ_OK)
-        {
-          res = SZ_ERROR_PROGRESS;
-          break;
-        }
-      }
-      LzmaEnc_Finish(p->enc);
-      if (res != SZ_OK)
-        return res;
-    }
-    if (finished)
-    {
-      if (*destSize == destLim)
-        return SZ_ERROR_OUTPUT_EOF;
-      dest[(*destSize)++] = 0;
-    }
-  }
-  return res;
-}
-
-#endif
-
-/* ---------- Lzma2Enc ---------- */
-
-CLzma2EncHandle Lzma2Enc_Create(ISzAlloc *alloc, ISzAlloc *allocBig)
-{
-  CLzma2Enc *p = (CLzma2Enc *)alloc->Alloc(alloc, sizeof(CLzma2Enc));
-  if (p == 0)
     return NULL;
   Lzma2EncProps_Init(&p->props);
   Lzma2EncProps_Normalize(&p->props);
-  p->outBuf = 0;
+  p->expectedDataSize = (UInt64)(Int64)-1;
+  p->tempBufLzma = NULL;
   p->alloc = alloc;
   p->allocBig = allocBig;
   {
     unsigned i;
-    for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
+    for (i = 0; i < MTCODER__THREADS_MAX; i++)
-      p->coders[i].enc = 0;
+      p->coders[i].enc = NULL;
   }
+  
   #ifndef _7ZIP_ST
-  MtCoder_Construct(&p->mtCoder);
+  p->mtCoder_WasConstructed = False;
+  {
+    unsigned i;
+    for (i = 0; i < MTCODER__BLOCKS_MAX; i++)
+      p->outBufs[i] = NULL;
+    p->outBufSize = 0;
+  }
   #endif
 
   return p;
 }
 
+
+#ifndef _7ZIP_ST
+
+static void Lzma2Enc_FreeOutBufs(CLzma2Enc *p)
+{
+  unsigned i;
+  for (i = 0; i < MTCODER__BLOCKS_MAX; i++)
+    if (p->outBufs[i])
+    {
+      ISzAlloc_Free(p->alloc, p->outBufs[i]);
+      p->outBufs[i] = NULL;
+    }
+  p->outBufSize = 0;
+}
+
+#endif
+
+
 void Lzma2Enc_Destroy(CLzma2EncHandle pp)
 {
   CLzma2Enc *p = (CLzma2Enc *)pp;
   unsigned i;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
+  for (i = 0; i < MTCODER__THREADS_MAX; i++)
   {
     CLzma2EncInt *t = &p->coders[i];
     if (t->enc)
     {
       LzmaEnc_Destroy(t->enc, p->alloc, p->allocBig);
-      t->enc = 0;
+      t->enc = NULL;
     }
   }
 
+
   #ifndef _7ZIP_ST
-  MtCoder_Destruct(&p->mtCoder);
+  if (p->mtCoder_WasConstructed)
+  {
+    MtCoder_Destruct(&p->mtCoder);
+    p->mtCoder_WasConstructed = False;
+  }
+  Lzma2Enc_FreeOutBufs(p);
   #endif
 
-  IAlloc_Free(p->alloc, p->outBuf);
+  ISzAlloc_Free(p->alloc, p->tempBufLzma);
-  IAlloc_Free(p->alloc, pp);
+  p->tempBufLzma = NULL;
+
+  ISzAlloc_Free(p->alloc, pp);
 }
 
+
 SRes Lzma2Enc_SetProps(CLzma2EncHandle pp, const CLzma2EncProps *props)
 {
   CLzma2Enc *p = (CLzma2Enc *)pp;
@@ -436,6 +474,14 @@ SRes Lzma2Enc_SetProps(CLzma2EncHandle pp, const CLzma2EncProps *props)
   return SZ_OK;
 }
 
+
+void Lzma2Enc_SetDataSize(CLzmaEncHandle pp, UInt64 expectedDataSiize)
+{
+  CLzma2Enc *p = (CLzma2Enc *)pp;
+  p->expectedDataSize = expectedDataSiize;
+}
+
+
 Byte Lzma2Enc_WriteProperties(CLzma2EncHandle pp)
 {
   CLzma2Enc *p = (CLzma2Enc *)pp;
@@ -447,47 +493,311 @@ Byte Lzma2Enc_WriteProperties(CLzma2EncHandle pp)
   return (Byte)i;
 }
 
-SRes Lzma2Enc_Encode(CLzma2EncHandle pp,
-    ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress)
-{
-  CLzma2Enc *p = (CLzma2Enc *)pp;
-  int i;
 
-  for (i = 0; i < p->props.numBlockThreads; i++)
+static SRes Lzma2Enc_EncodeMt1(
+    CLzma2Enc *me,
+    CLzma2EncInt *p,
+    ISeqOutStream *outStream,
+    Byte *outBuf, size_t *outBufSize,
+    ISeqInStream *inStream,
+    const Byte *inData, size_t inDataSize,
+    int finished,
+    ICompressProgress *progress)
+{
+  UInt64 unpackTotal = 0;
+  UInt64 packTotal = 0;
+  size_t outLim = 0;
+  CLimitedSeqInStream limitedInStream;
+
+  if (outBuf)
   {
-    CLzma2EncInt *t = &p->coders[i];
+    outLim = *outBufSize;
-    if (t->enc == NULL)
+    *outBufSize = 0;
+  }
+
+  if (!p->enc)
+  {
+    p->propsAreSet = False;
+    p->enc = LzmaEnc_Create(me->alloc);
+    if (!p->enc)
+      return SZ_ERROR_MEM;
+  }
+
+  limitedInStream.realStream = inStream;
+  if (inStream)
+  {
+    limitedInStream.vt.Read = LimitedSeqInStream_Read;
+  }
+  
+  if (!outBuf)
+  {
+    // outStream version works only in one thread. So we use CLzma2Enc::tempBufLzma
+    if (!me->tempBufLzma)
     {
-      t->enc = LzmaEnc_Create(p->alloc);
+      me->tempBufLzma = (Byte *)ISzAlloc_Alloc(me->alloc, LZMA2_CHUNK_SIZE_COMPRESSED_MAX);
-      if (t->enc == NULL)
+      if (!me->tempBufLzma)
         return SZ_ERROR_MEM;
     }
   }
 
-  #ifndef _7ZIP_ST
+  RINOK(Lzma2EncInt_InitStream(p, &me->props));
-  if (p->props.numBlockThreads <= 1)
-  #endif
-    return Lzma2Enc_EncodeMt1(&p->coders[0], p, outStream, inStream, progress);
 
-  #ifndef _7ZIP_ST
+  for (;;)
+  {
+    SRes res = SZ_OK;
+    size_t inSizeCur = 0;
+
+    Lzma2EncInt_InitBlock(p);
+    
+    LimitedSeqInStream_Init(&limitedInStream);
+    limitedInStream.limit = me->props.blockSize;
+
+    if (inStream)
+    {
+      UInt64 expected = (UInt64)(Int64)-1;
+      // inStream version works only in one thread. So we use CLzma2Enc::expectedDataSize
+      if (me->expectedDataSize != (UInt64)(Int64)-1
+          && me->expectedDataSize >= unpackTotal)
+        expected = me->expectedDataSize - unpackTotal;
+      if (me->props.blockSize != LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID
+          && expected > me->props.blockSize)
+        expected = (size_t)me->props.blockSize;
+
+      LzmaEnc_SetDataSize(p->enc, expected);
+
+      RINOK(LzmaEnc_PrepareForLzma2(p->enc,
+          &limitedInStream.vt,
+          LZMA2_KEEP_WINDOW_SIZE,
+          me->alloc,
+          me->allocBig));
+    }
+    else
+    {
+      inSizeCur = inDataSize - (size_t)unpackTotal;
+      if (me->props.blockSize != LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID
+          && inSizeCur > me->props.blockSize)
+        inSizeCur = (size_t)me->props.blockSize;
+    
+      // LzmaEnc_SetDataSize(p->enc, inSizeCur);
+      
+      RINOK(LzmaEnc_MemPrepare(p->enc,
+          inData + (size_t)unpackTotal, inSizeCur,
+          LZMA2_KEEP_WINDOW_SIZE,
+          me->alloc,
+          me->allocBig));
+    }
+
+    for (;;)
+    {
+      size_t packSize = LZMA2_CHUNK_SIZE_COMPRESSED_MAX;
+      if (outBuf)
+        packSize = outLim - (size_t)packTotal;
+      
+      res = Lzma2EncInt_EncodeSubblock(p,
+          outBuf ? outBuf + (size_t)packTotal : me->tempBufLzma, &packSize,
+          outBuf ? NULL : outStream);
+      
+      if (res != SZ_OK)
+        break;
+
+      packTotal += packSize;
+      if (outBuf)
+        *outBufSize = (size_t)packTotal;
+      
+      res = Progress(progress, unpackTotal + p->srcPos, packTotal);
+      if (res != SZ_OK)
+        break;
+
+      /*
+      if (LzmaEnc_GetNumAvailableBytes(p->enc) == 0)
+        break;
+      */
+
+      if (packSize == 0)
+        break;
+    }
+    
+    LzmaEnc_Finish(p->enc);
+    
+    unpackTotal += p->srcPos;
+    
+    RINOK(res);
+
+    if (p->srcPos != (inStream ? limitedInStream.processed : inSizeCur))
+      return SZ_ERROR_FAIL;
+    
+    if (inStream ? limitedInStream.finished : (unpackTotal == inDataSize))
+    {
+      if (finished)
+      {
+        if (outBuf)
+        {
+          size_t destPos = *outBufSize;
+          if (destPos >= outLim)
+            return SZ_ERROR_OUTPUT_EOF;
+          outBuf[destPos] = 0;
+          *outBufSize = destPos + 1;
+        }
+        else
+        {
+          Byte b = 0;
+          if (ISeqOutStream_Write(outStream, &b, 1) != 1)
+            return SZ_ERROR_WRITE;
+        }
+      }
+      return SZ_OK;
+    }
+  }
+}
+
+
+
+#ifndef _7ZIP_ST
+
+static SRes Lzma2Enc_MtCallback_Code(void *pp, unsigned coderIndex, unsigned outBufIndex,
+    const Byte *src, size_t srcSize, int finished)
+{
+  CLzma2Enc *me = (CLzma2Enc *)pp;
+  size_t destSize = me->outBufSize;
+  SRes res;
+  CMtProgressThunk progressThunk;
+
+  Byte *dest = me->outBufs[outBufIndex];
+
+  me->outBufsDataSizes[outBufIndex] = 0;
+
+  if (!dest)
+  {
+    dest = (Byte *)ISzAlloc_Alloc(me->alloc, me->outBufSize);
+    if (!dest)
+      return SZ_ERROR_MEM;
+    me->outBufs[outBufIndex] = dest;
+  }
+
+  MtProgressThunk_CreateVTable(&progressThunk);
+  progressThunk.mtProgress = &me->mtCoder.mtProgress;
+  progressThunk.inSize = 0;
+  progressThunk.outSize = 0;
+
+  res = Lzma2Enc_EncodeMt1(me,
+      &me->coders[coderIndex],
+      NULL, dest, &destSize,
+      NULL, src, srcSize,
+      finished,
+      &progressThunk.vt);
+
+  me->outBufsDataSizes[outBufIndex] = destSize;
+
+  return res;
+}
+
+
+static SRes Lzma2Enc_MtCallback_Write(void *pp, unsigned outBufIndex)
+{
+  CLzma2Enc *me = (CLzma2Enc *)pp;
+  size_t size = me->outBufsDataSizes[outBufIndex];
+  const Byte *data = me->outBufs[outBufIndex];
+  
+  if (me->outStream)
+    return ISeqOutStream_Write(me->outStream, data, size) == size ? SZ_OK : SZ_ERROR_WRITE;
+  
+  if (size > me->outBuf_Rem)
+    return SZ_ERROR_OUTPUT_EOF;
+  memcpy(me->outBuf, data, size);
+  me->outBuf_Rem -= size;
+  me->outBuf += size;
+  return SZ_OK;
+}
+
+#endif
+
+
+
+SRes Lzma2Enc_Encode2(CLzma2EncHandle pp,
+    ISeqOutStream *outStream,
+    Byte *outBuf, size_t *outBufSize,
+    ISeqInStream *inStream,
+    const Byte *inData, size_t inDataSize,
+    ICompressProgress *progress)
+{
+  CLzma2Enc *p = (CLzma2Enc *)pp;
+
+  if (inStream && inData)
+    return SZ_ERROR_PARAM;
+
+  if (outStream && outBuf)
+    return SZ_ERROR_PARAM;
 
   {
-    CMtCallbackImp mtCallback;
+    unsigned i;
+    for (i = 0; i < MTCODER__THREADS_MAX; i++)
+      p->coders[i].propsAreSet = False;
+  }
 
-    mtCallback.funcTable.Code = MtCallbackImp_Code;
+  #ifndef _7ZIP_ST
-    mtCallback.lzma2Enc = p;
+  
-    
+  if (p->props.numBlockThreads_Reduced > 1)
+  {
+    IMtCoderCallback2 vt;
+
+    if (!p->mtCoder_WasConstructed)
+    {
+      p->mtCoder_WasConstructed = True;
+      MtCoder_Construct(&p->mtCoder);
+    }
+
+    vt.Code = Lzma2Enc_MtCallback_Code;
+    vt.Write = Lzma2Enc_MtCallback_Write;
+
+    p->outStream = outStream;
+    p->outBuf = NULL;
+    p->outBuf_Rem = 0;
+    if (!outStream)
+    {
+      p->outBuf = outBuf;
+      p->outBuf_Rem = *outBufSize;
+      *outBufSize = 0;
+    }
+
+    p->mtCoder.allocBig = p->allocBig;
     p->mtCoder.progress = progress;
     p->mtCoder.inStream = inStream;
-    p->mtCoder.outStream = outStream;
+    p->mtCoder.inData = inData;
-    p->mtCoder.alloc = p->alloc;
+    p->mtCoder.inDataSize = inDataSize;
-    p->mtCoder.mtCallback = &mtCallback.funcTable;
+    p->mtCoder.mtCallback = &vt;
+    p->mtCoder.mtCallbackObject = p;
 
-    p->mtCoder.blockSize = p->props.blockSize;
+    p->mtCoder.blockSize = (size_t)p->props.blockSize;
-    p->mtCoder.destBlockSize = p->props.blockSize + (p->props.blockSize >> 10) + 16;
+    if (p->mtCoder.blockSize != p->props.blockSize)
-    p->mtCoder.numThreads = p->props.numBlockThreads;
+      return SZ_ERROR_PARAM; /* SZ_ERROR_MEM */
+
+    {
+      size_t destBlockSize = p->mtCoder.blockSize + (p->mtCoder.blockSize >> 10) + 16;
+      if (destBlockSize < p->mtCoder.blockSize)
+        return SZ_ERROR_PARAM;
+      if (p->outBufSize != destBlockSize)
+        Lzma2Enc_FreeOutBufs(p);
+      p->outBufSize = destBlockSize;
+    }
+
+    p->mtCoder.numThreadsMax = p->props.numBlockThreads_Max;
+    p->mtCoder.expectedDataSize = p->expectedDataSize;
     
-    return MtCoder_Code(&p->mtCoder);
+    {
+      SRes res = MtCoder_Code(&p->mtCoder);
+      if (!outStream)
+        *outBufSize = p->outBuf - outBuf;
+      return res;
+    }
   }
+
   #endif
+
+
+  return Lzma2Enc_EncodeMt1(p,
+      &p->coders[0],
+      outStream, outBuf, outBufSize,
+      inStream, inData, inDataSize,
+      True, /* finished */
+      progress);
 }

C/Lzma2Enc.h

@@ -1,5 +1,5 @@
 /* Lzma2Enc.h -- LZMA2 Encoder
-2013-01-18 : Igor Pavlov : Public domain */
+2017-07-27 : Igor Pavlov : Public domain */
 
 #ifndef __LZMA2_ENC_H
 #define __LZMA2_ENC_H
@@ -8,11 +8,15 @@
 
 EXTERN_C_BEGIN
 
+#define LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO 0
+#define LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID ((UInt64)(Int64)-1)
+
 typedef struct
 {
   CLzmaEncProps lzmaProps;
-  size_t blockSize;
+  UInt64 blockSize;
-  int numBlockThreads;
+  int numBlockThreads_Reduced;
+  int numBlockThreads_Max;
   int numTotalThreads;
 } CLzma2EncProps;
 
@@ -22,40 +26,29 @@ void Lzma2EncProps_Normalize(CLzma2EncProps *p);
 /* ---------- CLzmaEnc2Handle Interface ---------- */
 
 /* Lzma2Enc_* functions can return the following exit codes:
-Returns:
+SRes:
   SZ_OK           - OK
   SZ_ERROR_MEM    - Memory allocation error
   SZ_ERROR_PARAM  - Incorrect paramater in props
-  SZ_ERROR_WRITE  - Write callback error
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
   SZ_ERROR_PROGRESS - some break from progress callback
-  SZ_ERROR_THREAD - errors in multithreading functions (only for Mt version)
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
 */
 
 typedef void * CLzma2EncHandle;
 
-CLzma2EncHandle Lzma2Enc_Create(ISzAlloc *alloc, ISzAlloc *allocBig);
+CLzma2EncHandle Lzma2Enc_Create(ISzAllocPtr alloc, ISzAllocPtr allocBig);
 void Lzma2Enc_Destroy(CLzma2EncHandle p);
 SRes Lzma2Enc_SetProps(CLzma2EncHandle p, const CLzma2EncProps *props);
+void Lzma2Enc_SetDataSize(CLzma2EncHandle p, UInt64 expectedDataSiize);
 Byte Lzma2Enc_WriteProperties(CLzma2EncHandle p);
-SRes Lzma2Enc_Encode(CLzma2EncHandle p,
+SRes Lzma2Enc_Encode2(CLzma2EncHandle p,
-    ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress);
+    ISeqOutStream *outStream,
-
+    Byte *outBuf, size_t *outBufSize,
-/* ---------- One Call Interface ---------- */
+    ISeqInStream *inStream,
-
+    const Byte *inData, size_t inDataSize,
-/* Lzma2Encode
+    ICompressProgress *progress);
-Return code:
-  SZ_OK               - OK
-  SZ_ERROR_MEM        - Memory allocation error
-  SZ_ERROR_PARAM      - Incorrect paramater
-  SZ_ERROR_OUTPUT_EOF - output buffer overflow
-  SZ_ERROR_THREAD     - errors in multithreading functions (only for Mt version)
-*/
-
-/*
-SRes Lzma2Encode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
-    const CLzmaEncProps *props, Byte *propsEncoded, int writeEndMark,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
-*/
 
 EXTERN_C_END
 

C/Lzma86Dec.c

@@ -1,5 +1,7 @@
 /* Lzma86Dec.c -- LZMA + x86 (BCJ) Filter Decoder
-2009-08-14 : Igor Pavlov : Public domain */
+2016-05-16 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Lzma86.h"
 
@@ -7,9 +9,6 @@
 #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); }
-
 SRes Lzma86_GetUnpackSize(const Byte *src, SizeT srcLen, UInt64 *unpackSize)
 {
   unsigned i;
@@ -23,7 +22,6 @@ SRes Lzma86_GetUnpackSize(const Byte *src, SizeT srcLen, UInt64 *unpackSize)
 
 SRes Lzma86_Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen)
 {
-  ISzAlloc g_Alloc = { SzAlloc, SzFree };
   SRes res;
   int useFilter;
   SizeT inSizePure;

C/Lzma86Enc.c

@@ -1,5 +1,7 @@
 /* Lzma86Enc.c -- LZMA + x86 (BCJ) Filter Encoder
-2009-08-14 : Igor Pavlov : Public domain */
+2016-05-16 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
 
@@ -11,13 +13,9 @@
 
 #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); }
-
 int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
     int level, UInt32 dictSize, int filterMode)
 {
-  ISzAlloc g_Alloc = { SzAlloc, SzFree };
   size_t outSize2 = *destLen;
   Byte *filteredStream;
   Bool useFilter;

C/LzmaDec.c

@@ -1,8 +1,9 @@
 /* LzmaDec.c -- LZMA Decoder
-2015-06-23 : Igor Pavlov : Public domain */
+2018-02-28 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
+/* #include "CpuArch.h" */
 #include "LzmaDec.h"
 
 #include <string.h>
@@ -24,9 +25,16 @@
 #define GET_BIT2(p, i, A0, A1) IF_BIT_0(p) \
   { UPDATE_0(p); i = (i + i); A0; } else \
   { UPDATE_1(p); i = (i + i) + 1; A1; }
-#define GET_BIT(p, i) GET_BIT2(p, i, ; , ;)
 
-#define TREE_GET_BIT(probs, i) { GET_BIT((probs + i), i); }
+#define TREE_GET_BIT(probs, i) { GET_BIT2(probs + i, i, ;, ;); }
+
+#define REV_BIT(p, i, A0, A1) IF_BIT_0(p + i) \
+  { UPDATE_0(p + i); A0; } else \
+  { UPDATE_1(p + i); A1; }
+#define REV_BIT_VAR(  p, i, m) REV_BIT(p, i, i += m; m += m, m += m; i += m; )
+#define REV_BIT_CONST(p, i, m) REV_BIT(p, i, i += m;       , i += m * 2; )
+#define REV_BIT_LAST( p, i, m) REV_BIT(p, i, i -= m        , ; )
+
 #define TREE_DECODE(probs, limit, i) \
   { i = 1; do { TREE_GET_BIT(probs, i); } while (i < limit); i -= limit; }
 
@@ -46,12 +54,15 @@
   i -= 0x40; }
 #endif
 
-#define NORMAL_LITER_DEC GET_BIT(prob + symbol, symbol)
+#define NORMAL_LITER_DEC TREE_GET_BIT(prob, symbol)
 #define MATCHED_LITER_DEC \
-  matchByte <<= 1; \
+  matchByte += matchByte; \
-  bit = (matchByte & offs); \
+  bit = offs; \
-  probLit = prob + offs + bit + symbol; \
+  offs &= matchByte; \
-  GET_BIT2(probLit, symbol, offs &= ~bit, offs &= bit)
+  probLit = prob + (offs + bit + symbol); \
+  GET_BIT2(probLit, symbol, offs ^= bit; , ;)
+
+
 
 #define NORMALIZE_CHECK if (range < kTopValue) { if (buf >= bufLimit) return DUMMY_ERROR; range <<= 8; code = (code << 8) | (*buf++); }
 
@@ -66,25 +77,28 @@
   { i = 1; do { GET_BIT_CHECK(probs + i, i) } while (i < limit); i -= limit; }
 
 
+#define REV_BIT_CHECK(p, i, m) IF_BIT_0_CHECK(p + i) \
+  { UPDATE_0_CHECK; i += m; m += m; } else \
+  { UPDATE_1_CHECK; m += m; i += m; }
+
+
 #define kNumPosBitsMax 4
 #define kNumPosStatesMax (1 << kNumPosBitsMax)
 
 #define kLenNumLowBits 3
 #define kLenNumLowSymbols (1 << kLenNumLowBits)
-#define kLenNumMidBits 3
-#define kLenNumMidSymbols (1 << kLenNumMidBits)
 #define kLenNumHighBits 8
 #define kLenNumHighSymbols (1 << kLenNumHighBits)
 
-#define LenChoice 0
+#define LenLow 0
-#define LenChoice2 (LenChoice + 1)
+#define LenHigh (LenLow + 2 * (kNumPosStatesMax << kLenNumLowBits))
-#define LenLow (LenChoice2 + 1)
-#define LenMid (LenLow + (kNumPosStatesMax << kLenNumLowBits))
-#define LenHigh (LenMid + (kNumPosStatesMax << kLenNumMidBits))
 #define kNumLenProbs (LenHigh + kLenNumHighSymbols)
 
+#define LenChoice LenLow
+#define LenChoice2 (LenLow + (1 << kLenNumLowBits))
 
 #define kNumStates 12
+#define kNumStates2 16
 #define kNumLitStates 7
 
 #define kStartPosModelIndex 4
@@ -98,54 +112,117 @@
 #define kAlignTableSize (1 << kNumAlignBits)
 
 #define kMatchMinLen 2
-#define kMatchSpecLenStart (kMatchMinLen + kLenNumLowSymbols + kLenNumMidSymbols + kLenNumHighSymbols)
+#define kMatchSpecLenStart (kMatchMinLen + kLenNumLowSymbols * 2 + kLenNumHighSymbols)
 
-#define IsMatch 0
+/* External ASM code needs same CLzmaProb array layout. So don't change it. */
-#define IsRep (IsMatch + (kNumStates << kNumPosBitsMax))
+
+/* (probs_1664) is faster and better for code size at some platforms */
+/*
+#ifdef MY_CPU_X86_OR_AMD64
+*/
+#define kStartOffset 1664
+#define GET_PROBS p->probs_1664
+/*
+#define GET_PROBS p->probs + kStartOffset
+#else
+#define kStartOffset 0
+#define GET_PROBS p->probs
+#endif
+*/
+
+#define SpecPos (-kStartOffset)
+#define IsRep0Long (SpecPos + kNumFullDistances)
+#define RepLenCoder (IsRep0Long + (kNumStates2 << kNumPosBitsMax))
+#define LenCoder (RepLenCoder + kNumLenProbs)
+#define IsMatch (LenCoder + kNumLenProbs)
+#define Align (IsMatch + (kNumStates2 << kNumPosBitsMax))
+#define IsRep (Align + kAlignTableSize)
 #define IsRepG0 (IsRep + kNumStates)
 #define IsRepG1 (IsRepG0 + kNumStates)
 #define IsRepG2 (IsRepG1 + kNumStates)
-#define IsRep0Long (IsRepG2 + kNumStates)
+#define PosSlot (IsRepG2 + kNumStates)
-#define PosSlot (IsRep0Long + (kNumStates << kNumPosBitsMax))
+#define Literal (PosSlot + (kNumLenToPosStates << kNumPosSlotBits))
-#define SpecPos (PosSlot + (kNumLenToPosStates << kNumPosSlotBits))
+#define NUM_BASE_PROBS (Literal + kStartOffset)
-#define Align (SpecPos + kNumFullDistances - kEndPosModelIndex)
-#define LenCoder (Align + kAlignTableSize)
-#define RepLenCoder (LenCoder + kNumLenProbs)
-#define Literal (RepLenCoder + kNumLenProbs)
 
-#define LZMA_BASE_SIZE 1846
+#if Align != 0 && kStartOffset != 0
-#define LZMA_LIT_SIZE 0x300
+  #error Stop_Compiling_Bad_LZMA_kAlign
-
-#if Literal != LZMA_BASE_SIZE
-StopCompilingDueBUG
 #endif
 
-#define LzmaProps_GetNumProbs(p) (Literal + ((UInt32)LZMA_LIT_SIZE << ((p)->lc + (p)->lp)))
+#if NUM_BASE_PROBS != 1984
+  #error Stop_Compiling_Bad_LZMA_PROBS
+#endif
+
+
+#define LZMA_LIT_SIZE 0x300
+
+#define LzmaProps_GetNumProbs(p) (NUM_BASE_PROBS + ((UInt32)LZMA_LIT_SIZE << ((p)->lc + (p)->lp)))
+
+
+#define CALC_POS_STATE(processedPos, pbMask) (((processedPos) & (pbMask)) << 4)
+#define COMBINED_PS_STATE (posState + state)
+#define GET_LEN_STATE (posState)
 
 #define LZMA_DIC_MIN (1 << 12)
 
-/* First LZMA-symbol is always decoded.
+/*
-And it decodes new LZMA-symbols while (buf < bufLimit), but "buf" is without last normalization
+p->remainLen : shows status of LZMA decoder:
+    < kMatchSpecLenStart : normal remain
+    = kMatchSpecLenStart : finished
+    = kMatchSpecLenStart + 1 : need init range coder
+    = kMatchSpecLenStart + 2 : need init range coder and state
+*/
+
+/* ---------- LZMA_DECODE_REAL ---------- */
+/*
+LzmaDec_DecodeReal_3() can be implemented in external ASM file.
+3 - is the code compatibility version of that function for check at link time.
+*/
+
+#define LZMA_DECODE_REAL LzmaDec_DecodeReal_3
+
+/*
+LZMA_DECODE_REAL()
+In:
+  RangeCoder is normalized
+  if (p->dicPos == limit)
+  {
+    LzmaDec_TryDummy() was called before to exclude LITERAL and MATCH-REP cases.
+    So first symbol can be only MATCH-NON-REP. And if that MATCH-NON-REP symbol
+    is not END_OF_PAYALOAD_MARKER, then function returns error code.
+  }
+
+Processing:
+  first LZMA symbol will be decoded in any case
+  All checks for limits are at the end of main loop,
+  It will decode new LZMA-symbols while (p->buf < bufLimit && dicPos < limit),
+  RangeCoder is still without last normalization when (p->buf < bufLimit) is being checked.
+
 Out:
+  RangeCoder is normalized
   Result:
     SZ_OK - OK
     SZ_ERROR_DATA - Error
   p->remainLen:
     < kMatchSpecLenStart : normal remain
     = kMatchSpecLenStart : finished
-    = 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)
-{
-  CLzmaProb *probs = p->probs;
 
-  unsigned state = p->state;
+#ifdef _LZMA_DEC_OPT
+
+int MY_FAST_CALL LZMA_DECODE_REAL(CLzmaDec *p, SizeT limit, const Byte *bufLimit);
+
+#else
+
+static
+int MY_FAST_CALL LZMA_DECODE_REAL(CLzmaDec *p, SizeT limit, const Byte *bufLimit)
+{
+  CLzmaProb *probs = GET_PROBS;
+  unsigned state = (unsigned)p->state;
   UInt32 rep0 = p->reps[0], rep1 = p->reps[1], rep2 = p->reps[2], rep3 = p->reps[3];
   unsigned pbMask = ((unsigned)1 << (p->prop.pb)) - 1;
-  unsigned lpMask = ((unsigned)1 << (p->prop.lp)) - 1;
   unsigned lc = p->prop.lc;
+  unsigned lpMask = ((unsigned)0x100 << p->prop.lp) - ((unsigned)0x100 >> lc);
 
   Byte *dic = p->dic;
   SizeT dicBufSize = p->dicBufSize;
@@ -164,17 +241,16 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
     CLzmaProb *prob;
     UInt32 bound;
     unsigned ttt;
-    unsigned posState = processedPos & pbMask;
+    unsigned posState = CALC_POS_STATE(processedPos, pbMask);
 
-    prob = probs + IsMatch + (state << kNumPosBitsMax) + posState;
+    prob = probs + IsMatch + COMBINED_PS_STATE;
     IF_BIT_0(prob)
     {
       unsigned symbol;
       UPDATE_0(prob);
       prob = probs + Literal;
       if (processedPos != 0 || checkDicSize != 0)
-        prob += ((UInt32)LZMA_LIT_SIZE * (((processedPos & lpMask) << lc) +
+        prob += (UInt32)3 * ((((processedPos << 8) + dic[(dicPos == 0 ? dicBufSize : dicPos) - 1]) & lpMask) << lc);
-            (dic[(dicPos == 0 ? dicBufSize : dicPos) - 1] >> (8 - lc))));
       processedPos++;
 
       if (state < kNumLitStates)
@@ -240,13 +316,16 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
       else
       {
         UPDATE_1(prob);
+        /*
+        // that case was checked before with kBadRepCode
         if (checkDicSize == 0 && processedPos == 0)
           return SZ_ERROR_DATA;
+        */
         prob = probs + IsRepG0 + state;
         IF_BIT_0(prob)
         {
           UPDATE_0(prob);
-          prob = probs + IsRep0Long + (state << kNumPosBitsMax) + posState;
+          prob = probs + IsRep0Long + COMBINED_PS_STATE;
           IF_BIT_0(prob)
           {
             UPDATE_0(prob);
@@ -294,14 +373,14 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
       
       #ifdef _LZMA_SIZE_OPT
       {
-        unsigned limit, offset;
+        unsigned lim, offset;
         CLzmaProb *probLen = prob + LenChoice;
         IF_BIT_0(probLen)
         {
           UPDATE_0(probLen);
-          probLen = prob + LenLow + (posState << kLenNumLowBits);
+          probLen = prob + LenLow + GET_LEN_STATE;
           offset = 0;
-          limit = (1 << kLenNumLowBits);
+          lim = (1 << kLenNumLowBits);
         }
         else
         {
@@ -310,19 +389,19 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
           IF_BIT_0(probLen)
           {
             UPDATE_0(probLen);
-            probLen = prob + LenMid + (posState << kLenNumMidBits);
+            probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits);
             offset = kLenNumLowSymbols;
-            limit = (1 << kLenNumMidBits);
+            lim = (1 << kLenNumLowBits);
           }
           else
           {
             UPDATE_1(probLen);
             probLen = prob + LenHigh;
-            offset = kLenNumLowSymbols + kLenNumMidSymbols;
+            offset = kLenNumLowSymbols * 2;
-            limit = (1 << kLenNumHighBits);
+            lim = (1 << kLenNumHighBits);
           }
         }
-        TREE_DECODE(probLen, limit, len);
+        TREE_DECODE(probLen, lim, len);
         len += offset;
       }
       #else
@@ -331,7 +410,7 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
         IF_BIT_0(probLen)
         {
           UPDATE_0(probLen);
-          probLen = prob + LenLow + (posState << kLenNumLowBits);
+          probLen = prob + LenLow + GET_LEN_STATE;
           len = 1;
           TREE_GET_BIT(probLen, len);
           TREE_GET_BIT(probLen, len);
@@ -345,7 +424,7 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
           IF_BIT_0(probLen)
           {
             UPDATE_0(probLen);
-            probLen = prob + LenMid + (posState << kLenNumMidBits);
+            probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits);
             len = 1;
             TREE_GET_BIT(probLen, len);
             TREE_GET_BIT(probLen, len);
@@ -356,7 +435,7 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
             UPDATE_1(probLen);
             probLen = prob + LenHigh;
             TREE_DECODE(probLen, (1 << kLenNumHighBits), len);
-            len += kLenNumLowSymbols + kLenNumMidSymbols;
+            len += kLenNumLowSymbols * 2;
           }
         }
       }
@@ -376,16 +455,16 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
           if (posSlot < kEndPosModelIndex)
           {
             distance <<= numDirectBits;
-            prob = probs + SpecPos + distance - posSlot - 1;
+            prob = probs + SpecPos;
             {
-              UInt32 mask = 1;
+              UInt32 m = 1;
-              unsigned i = 1;
+              distance++;
               do
               {
-                GET_BIT2(prob + i, i, ; , distance |= mask);
+                REV_BIT_VAR(prob, distance, m);
-                mask <<= 1;
               }
-              while (--numDirectBits != 0);
+              while (--numDirectBits);
+              distance -= m;
             }
           }
           else
@@ -412,19 +491,20 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
               }
               */
             }
-            while (--numDirectBits != 0);
+            while (--numDirectBits);
             prob = probs + Align;
             distance <<= kNumAlignBits;
             {
               unsigned i = 1;
-              GET_BIT2(prob + i, i, ; , distance |= 1);
+              REV_BIT_CONST(prob, i, 1);
-              GET_BIT2(prob + i, i, ; , distance |= 2);
+              REV_BIT_CONST(prob, i, 2);
-              GET_BIT2(prob + i, i, ; , distance |= 4);
+              REV_BIT_CONST(prob, i, 4);
-              GET_BIT2(prob + i, i, ; , distance |= 8);
+              REV_BIT_LAST (prob, i, 8);
+              distance |= i;
             }
             if (distance == (UInt32)0xFFFFFFFF)
             {
-              len += kMatchSpecLenStart;
+              len = kMatchSpecLenStart;
               state -= kNumStates;
               break;
             }
@@ -435,20 +515,12 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
         rep2 = rep1;
         rep1 = rep0;
         rep0 = distance + 1;
-        if (checkDicSize == 0)
+        state = (state < kNumStates + kNumLitStates) ? kNumLitStates : kNumLitStates + 3;
-        {
+        if (distance >= (checkDicSize == 0 ? processedPos: checkDicSize))
-          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;
@@ -511,6 +583,7 @@ static int MY_FAST_CALL LzmaDec_DecodeReal(CLzmaDec *p, SizeT limit, const Byte
 
   return SZ_OK;
 }
+#endif
 
 static void MY_FAST_CALL LzmaDec_WriteRem(CLzmaDec *p, SizeT limit)
 {
@@ -519,7 +592,7 @@ static void MY_FAST_CALL LzmaDec_WriteRem(CLzmaDec *p, SizeT limit)
     Byte *dic = p->dic;
     SizeT dicPos = p->dicPos;
     SizeT dicBufSize = p->dicBufSize;
-    unsigned len = p->remainLen;
+    unsigned len = (unsigned)p->remainLen;
     SizeT rep0 = p->reps[0]; /* we use SizeT to avoid the BUG of VC14 for AMD64 */
     SizeT rem = limit - dicPos;
     if (rem < len)
@@ -540,6 +613,14 @@ static void MY_FAST_CALL LzmaDec_WriteRem(CLzmaDec *p, SizeT limit)
   }
 }
 
+
+#define kRange0 0xFFFFFFFF
+#define kBound0 ((kRange0 >> kNumBitModelTotalBits) << (kNumBitModelTotalBits - 1))
+#define kBadRepCode (kBound0 + (((kRange0 - kBound0) >> kNumBitModelTotalBits) << (kNumBitModelTotalBits - 1)))
+#if kBadRepCode != (0xC0000000 - 0x400)
+  #error Stop_Compiling_Bad_LZMA_Check
+#endif
+
 static int MY_FAST_CALL LzmaDec_DecodeReal2(CLzmaDec *p, SizeT limit, const Byte *bufLimit)
 {
   do
@@ -550,9 +631,13 @@ static int MY_FAST_CALL LzmaDec_DecodeReal2(CLzmaDec *p, SizeT limit, const Byte
       UInt32 rem = p->prop.dicSize - p->processedPos;
       if (limit - p->dicPos > rem)
         limit2 = p->dicPos + rem;
+
+      if (p->processedPos == 0)
+        if (p->code >= kBadRepCode)
+          return SZ_ERROR_DATA;
     }
-    
+
-    RINOK(LzmaDec_DecodeReal(p, limit2, bufLimit));
+    RINOK(LZMA_DECODE_REAL(p, limit2, bufLimit));
     
     if (p->checkDicSize == 0 && p->processedPos >= p->prop.dicSize)
       p->checkDicSize = p->prop.dicSize;
@@ -561,9 +646,6 @@ static int MY_FAST_CALL LzmaDec_DecodeReal2(CLzmaDec *p, SizeT limit, const Byte
   }
   while (p->dicPos < limit && p->buf < bufLimit && p->remainLen < kMatchSpecLenStart);
 
-  if (p->remainLen > kMatchSpecLenStart)
-    p->remainLen = kMatchSpecLenStart;
-
   return 0;
 }
 
@@ -580,17 +662,17 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
   UInt32 range = p->range;
   UInt32 code = p->code;
   const Byte *bufLimit = buf + inSize;
-  const CLzmaProb *probs = p->probs;
+  const CLzmaProb *probs = GET_PROBS;
-  unsigned state = p->state;
+  unsigned state = (unsigned)p->state;
   ELzmaDummy res;
 
   {
     const CLzmaProb *prob;
     UInt32 bound;
     unsigned ttt;
-    unsigned posState = (p->processedPos) & ((1 << p->prop.pb) - 1);
+    unsigned posState = CALC_POS_STATE(p->processedPos, (1 << p->prop.pb) - 1);
 
-    prob = probs + IsMatch + (state << kNumPosBitsMax) + posState;
+    prob = probs + IsMatch + COMBINED_PS_STATE;
     IF_BIT_0_CHECK(prob)
     {
       UPDATE_0_CHECK
@@ -618,10 +700,11 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
         {
           unsigned bit;
           const CLzmaProb *probLit;
-          matchByte <<= 1;
+          matchByte += matchByte;
-          bit = (matchByte & offs);
+          bit = offs;
-          probLit = prob + offs + bit + symbol;
+          offs &= matchByte;
-          GET_BIT2_CHECK(probLit, symbol, offs &= ~bit, offs &= bit)
+          probLit = prob + (offs + bit + symbol);
+          GET_BIT2_CHECK(probLit, symbol, offs ^= bit; , ; )
         }
         while (symbol < 0x100);
       }
@@ -648,7 +731,7 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
         IF_BIT_0_CHECK(prob)
         {
           UPDATE_0_CHECK;
-          prob = probs + IsRep0Long + (state << kNumPosBitsMax) + posState;
+          prob = probs + IsRep0Long + COMBINED_PS_STATE;
           IF_BIT_0_CHECK(prob)
           {
             UPDATE_0_CHECK;
@@ -691,7 +774,7 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
         IF_BIT_0_CHECK(probLen)
         {
           UPDATE_0_CHECK;
-          probLen = prob + LenLow + (posState << kLenNumLowBits);
+          probLen = prob + LenLow + GET_LEN_STATE;
           offset = 0;
           limit = 1 << kLenNumLowBits;
         }
@@ -702,15 +785,15 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
           IF_BIT_0_CHECK(probLen)
           {
             UPDATE_0_CHECK;
-            probLen = prob + LenMid + (posState << kLenNumMidBits);
+            probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits);
             offset = kLenNumLowSymbols;
-            limit = 1 << kLenNumMidBits;
+            limit = 1 << kLenNumLowBits;
           }
           else
           {
             UPDATE_1_CHECK;
             probLen = prob + LenHigh;
-            offset = kLenNumLowSymbols + kLenNumMidSymbols;
+            offset = kLenNumLowSymbols * 2;
             limit = 1 << kLenNumHighBits;
           }
         }
@@ -722,7 +805,7 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
       {
         unsigned posSlot;
         prob = probs + PosSlot +
-            ((len < kNumLenToPosStates ? len : kNumLenToPosStates - 1) <<
+            ((len < kNumLenToPosStates - 1 ? len : kNumLenToPosStates - 1) <<
             kNumPosSlotBits);
         TREE_DECODE_CHECK(prob, 1 << kNumPosSlotBits, posSlot);
         if (posSlot >= kStartPosModelIndex)
@@ -733,7 +816,7 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
 
           if (posSlot < kEndPosModelIndex)
           {
-            prob = probs + SpecPos + ((2 | (posSlot & 1)) << numDirectBits) - posSlot - 1;
+            prob = probs + SpecPos + ((2 | (posSlot & 1)) << numDirectBits);
           }
           else
           {
@@ -745,17 +828,18 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
               code -= range & (((code - range) >> 31) - 1);
               /* if (code >= range) code -= range; */
             }
-            while (--numDirectBits != 0);
+            while (--numDirectBits);
             prob = probs + Align;
             numDirectBits = kNumAlignBits;
           }
           {
             unsigned i = 1;
+            unsigned m = 1;
             do
             {
-              GET_BIT_CHECK(prob + i, i);
+              REV_BIT_CHECK(prob, i, m);
             }
-            while (--numDirectBits != 0);
+            while (--numDirectBits);
           }
         }
       }
@@ -768,18 +852,17 @@ static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, SizeT inS
 
 void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState)
 {
-  p->needFlush = 1;
+  p->remainLen = kMatchSpecLenStart + 1;
-  p->remainLen = 0;
   p->tempBufSize = 0;
 
   if (initDic)
   {
     p->processedPos = 0;
     p->checkDicSize = 0;
-    p->needInitState = 1;
+    p->remainLen = kMatchSpecLenStart + 2;
   }
   if (initState)
-    p->needInitState = 1;
+    p->remainLen = kMatchSpecLenStart + 2;
 }
 
 void LzmaDec_Init(CLzmaDec *p)
@@ -788,53 +871,54 @@ void LzmaDec_Init(CLzmaDec *p)
   LzmaDec_InitDicAndState(p, True, True);
 }
 
-static void LzmaDec_InitStateReal(CLzmaDec *p)
-{
-  SizeT numProbs = LzmaProps_GetNumProbs(&p->prop);
-  SizeT i;
-  CLzmaProb *probs = p->probs;
-  for (i = 0; i < numProbs; i++)
-    probs[i] = kBitModelTotal >> 1;
-  p->reps[0] = p->reps[1] = p->reps[2] = p->reps[3] = 1;
-  p->state = 0;
-  p->needInitState = 0;
-}
 
 SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit, const Byte *src, SizeT *srcLen,
     ELzmaFinishMode finishMode, ELzmaStatus *status)
 {
   SizeT inSize = *srcLen;
   (*srcLen) = 0;
-  LzmaDec_WriteRem(p, dicLimit);
   
   *status = LZMA_STATUS_NOT_SPECIFIED;
 
+  if (p->remainLen > kMatchSpecLenStart)
+  {
+    for (; inSize > 0 && p->tempBufSize < RC_INIT_SIZE; (*srcLen)++, inSize--)
+      p->tempBuf[p->tempBufSize++] = *src++;
+    if (p->tempBufSize != 0 && p->tempBuf[0] != 0)
+      return SZ_ERROR_DATA;
+    if (p->tempBufSize < RC_INIT_SIZE)
+    {
+      *status = LZMA_STATUS_NEEDS_MORE_INPUT;
+      return SZ_OK;
+    }
+    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->tempBufSize = 0;
+
+    if (p->remainLen > kMatchSpecLenStart + 1)
+    {
+      SizeT numProbs = LzmaProps_GetNumProbs(&p->prop);
+      SizeT i;
+      CLzmaProb *probs = p->probs;
+      for (i = 0; i < numProbs; i++)
+        probs[i] = kBitModelTotal >> 1;
+      p->reps[0] = p->reps[1] = p->reps[2] = p->reps[3] = 1;
+      p->state = 0;
+    }
+
+    p->remainLen = 0;
+  }
+
+  LzmaDec_WriteRem(p, dicLimit);
+
   while (p->remainLen != kMatchSpecLenStart)
   {
-      int checkEndMarkNow;
+      int checkEndMarkNow = 0;
 
-      if (p->needFlush)
-      {
-        for (; inSize > 0 && p->tempBufSize < RC_INIT_SIZE; (*srcLen)++, inSize--)
-          p->tempBuf[p->tempBufSize++] = *src++;
-        if (p->tempBufSize < RC_INIT_SIZE)
-        {
-          *status = LZMA_STATUS_NEEDS_MORE_INPUT;
-          return SZ_OK;
-        }
-        if (p->tempBuf[0] != 0)
-          return SZ_ERROR_DATA;
-        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;
-      }
-
-      checkEndMarkNow = 0;
       if (p->dicPos >= dicLimit)
       {
         if (p->remainLen == 0 && p->code == 0)
@@ -855,9 +939,6 @@ SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit, const Byte *src, SizeT *sr
         checkEndMarkNow = 1;
       }
 
-      if (p->needInitState)
-        LzmaDec_InitStateReal(p);
-  
       if (p->tempBufSize == 0)
       {
         SizeT processed;
@@ -930,11 +1011,14 @@ SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit, const Byte *src, SizeT *sr
         p->tempBufSize = 0;
       }
   }
-  if (p->code == 0)
+  
-    *status = LZMA_STATUS_FINISHED_WITH_MARK;
+  if (p->code != 0)
-  return (p->code == 0) ? SZ_OK : SZ_ERROR_DATA;
+    return SZ_ERROR_DATA;
+  *status = LZMA_STATUS_FINISHED_WITH_MARK;
+  return SZ_OK;
 }
 
+
 SRes LzmaDec_DecodeToBuf(CLzmaDec *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
 {
   SizeT outSize = *destLen;
@@ -975,19 +1059,19 @@ SRes LzmaDec_DecodeToBuf(CLzmaDec *p, Byte *dest, SizeT *destLen, const Byte *sr
   }
 }
 
-void LzmaDec_FreeProbs(CLzmaDec *p, ISzAlloc *alloc)
+void LzmaDec_FreeProbs(CLzmaDec *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->probs);
+  ISzAlloc_Free(alloc, p->probs);
   p->probs = NULL;
 }
 
-static void LzmaDec_FreeDict(CLzmaDec *p, ISzAlloc *alloc)
+static void LzmaDec_FreeDict(CLzmaDec *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->dic);
+  ISzAlloc_Free(alloc, p->dic);
   p->dic = NULL;
 }
 
-void LzmaDec_Free(CLzmaDec *p, ISzAlloc *alloc)
+void LzmaDec_Free(CLzmaDec *p, ISzAllocPtr alloc)
 {
   LzmaDec_FreeProbs(p, alloc);
   LzmaDec_FreeDict(p, alloc);
@@ -1011,29 +1095,30 @@ SRes LzmaProps_Decode(CLzmaProps *p, const Byte *data, unsigned size)
   if (d >= (9 * 5 * 5))
     return SZ_ERROR_UNSUPPORTED;
 
-  p->lc = d % 9;
+  p->lc = (Byte)(d % 9);
   d /= 9;
-  p->pb = d / 5;
+  p->pb = (Byte)(d / 5);
-  p->lp = d % 5;
+  p->lp = (Byte)(d % 5);
 
   return SZ_OK;
 }
 
-static SRes LzmaDec_AllocateProbs2(CLzmaDec *p, const CLzmaProps *propNew, ISzAlloc *alloc)
+static SRes LzmaDec_AllocateProbs2(CLzmaDec *p, const CLzmaProps *propNew, ISzAllocPtr alloc)
 {
   UInt32 numProbs = LzmaProps_GetNumProbs(propNew);
   if (!p->probs || numProbs != p->numProbs)
   {
     LzmaDec_FreeProbs(p, alloc);
-    p->probs = (CLzmaProb *)alloc->Alloc(alloc, numProbs * sizeof(CLzmaProb));
+    p->probs = (CLzmaProb *)ISzAlloc_Alloc(alloc, numProbs * sizeof(CLzmaProb));
-    p->numProbs = numProbs;
     if (!p->probs)
       return SZ_ERROR_MEM;
+    p->probs_1664 = p->probs + 1664;
+    p->numProbs = numProbs;
   }
   return SZ_OK;
 }
 
-SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAlloc *alloc)
+SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc)
 {
   CLzmaProps propNew;
   RINOK(LzmaProps_Decode(&propNew, props, propsSize));
@@ -1042,7 +1127,7 @@ SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, I
   return SZ_OK;
 }
 
-SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAlloc *alloc)
+SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc)
 {
   CLzmaProps propNew;
   SizeT dicBufSize;
@@ -1062,7 +1147,7 @@ SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAll
   if (!p->dic || dicBufSize != p->dicBufSize)
   {
     LzmaDec_FreeDict(p, alloc);
-    p->dic = (Byte *)alloc->Alloc(alloc, dicBufSize);
+    p->dic = (Byte *)ISzAlloc_Alloc(alloc, dicBufSize);
     if (!p->dic)
     {
       LzmaDec_FreeProbs(p, alloc);
@@ -1076,7 +1161,7 @@ SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAll
 
 SRes LzmaDecode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
     const Byte *propData, unsigned propSize, ELzmaFinishMode finishMode,
-    ELzmaStatus *status, ISzAlloc *alloc)
+    ELzmaStatus *status, ISzAllocPtr alloc)
 {
   CLzmaDec p;
   SRes res;

C/LzmaDec.h

@@ -1,5 +1,5 @@
 /* LzmaDec.h -- LZMA Decoder
-2013-01-18 : Igor Pavlov : Public domain */
+2018-04-21 : Igor Pavlov : Public domain */
 
 #ifndef __LZMA_DEC_H
 #define __LZMA_DEC_H
@@ -12,11 +12,13 @@ EXTERN_C_BEGIN
 /* _LZMA_PROB32 can increase the speed on some CPUs,
    but memory usage for CLzmaDec::probs will be doubled in that case */
 
+typedef
 #ifdef _LZMA_PROB32
-#define CLzmaProb UInt32
+  UInt32
 #else
-#define CLzmaProb UInt16
+  UInt16
 #endif
+  CLzmaProb;
 
 
 /* ---------- LZMA Properties ---------- */
@@ -25,7 +27,10 @@ EXTERN_C_BEGIN
 
 typedef struct _CLzmaProps
 {
-  unsigned lc, lp, pb;
+  Byte lc;
+  Byte lp;
+  Byte pb;
+  Byte _pad_;
   UInt32 dicSize;
 } CLzmaProps;
 
@@ -47,32 +52,34 @@ SRes LzmaProps_Decode(CLzmaProps *p, const Byte *data, unsigned size);
 
 typedef struct
 {
+  /* Don't change this structure. ASM code can use it. */
   CLzmaProps prop;
   CLzmaProb *probs;
+  CLzmaProb *probs_1664;
   Byte *dic;
-  const Byte *buf;
-  UInt32 range, code;
-  SizeT dicPos;
   SizeT dicBufSize;
+  SizeT dicPos;
+  const Byte *buf;
+  UInt32 range;
+  UInt32 code;
   UInt32 processedPos;
   UInt32 checkDicSize;
-  unsigned state;
   UInt32 reps[4];
-  unsigned remainLen;
+  UInt32 state;
-  int needFlush;
+  UInt32 remainLen;
-  int needInitState;
+
   UInt32 numProbs;
   unsigned tempBufSize;
   Byte tempBuf[LZMA_REQUIRED_INPUT_MAX];
 } CLzmaDec;
 
-#define LzmaDec_Construct(p) { (p)->dic = 0; (p)->probs = 0; }
+#define LzmaDec_Construct(p) { (p)->dic = NULL; (p)->probs = NULL; }
 
 void LzmaDec_Init(CLzmaDec *p);
 
 /* There are two types of LZMA streams:
-     0) Stream with end mark. That end mark adds about 6 bytes to compressed size.
+     - Stream with end mark. That end mark adds about 6 bytes to compressed size.
-     1) Stream without end mark. You must know exact uncompressed size to decompress such stream. */
+     - Stream without end mark. You must know exact uncompressed size to decompress such stream. */
 
 typedef enum
 {
@@ -129,11 +136,11 @@ LzmaDec_Allocate* can return:
   SZ_ERROR_UNSUPPORTED - Unsupported properties
 */
    
-SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAlloc *alloc);
+SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc);
-void LzmaDec_FreeProbs(CLzmaDec *p, ISzAlloc *alloc);
+void LzmaDec_FreeProbs(CLzmaDec *p, ISzAllocPtr alloc);
 
-SRes LzmaDec_Allocate(CLzmaDec *state, const Byte *prop, unsigned propsSize, ISzAlloc *alloc);
+SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc);
-void LzmaDec_Free(CLzmaDec *state, ISzAlloc *alloc);
+void LzmaDec_Free(CLzmaDec *p, ISzAllocPtr alloc);
 
 /* ---------- Dictionary Interface ---------- */
 
@@ -142,7 +149,7 @@ void LzmaDec_Free(CLzmaDec *state, ISzAlloc *alloc);
    You must work with CLzmaDec variables directly in this interface.
 
    STEPS:
-     LzmaDec_Constr()
+     LzmaDec_Construct()
      LzmaDec_Allocate()
      for (each new stream)
      {
@@ -220,7 +227,7 @@ Returns:
 
 SRes LzmaDecode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
     const Byte *propData, unsigned propSize, ELzmaFinishMode finishMode,
-    ELzmaStatus *status, ISzAlloc *alloc);
+    ELzmaStatus *status, ISzAllocPtr alloc);
 
 EXTERN_C_END
 

C/LzmaEnc.h

@@ -1,5 +1,5 @@
 /*  LzmaEnc.h -- LZMA Encoder
-2013-01-18 : Igor Pavlov : Public domain */
+2017-07-27 : Igor Pavlov : Public domain */
 
 #ifndef __LZMA_ENC_H
 #define __LZMA_ENC_H
@@ -12,12 +12,10 @@ EXTERN_C_BEGIN
 
 typedef struct _CLzmaEncProps
 {
-  int level;       /*  0 <= level <= 9 */
+  int level;       /* 0 <= level <= 9 */
   UInt32 dictSize; /* (1 << 12) <= dictSize <= (1 << 27) for 32-bit version
-                      (1 << 12) <= dictSize <= (1 << 30) for 64-bit version
+                      (1 << 12) <= dictSize <= (3 << 29) for 64-bit version
-                       default = (1 << 24) */
+                      default = (1 << 24) */
-  UInt64 reduceSize; /* estimated size of data that will be compressed. default = 0xFFFFFFFF.
-                        Encoder uses this value to reduce dictionary size */
   int lc;          /* 0 <= lc <= 8, default = 3 */
   int lp;          /* 0 <= lp <= 4, default = 0 */
   int pb;          /* 0 <= pb <= 4, default = 2 */
@@ -25,9 +23,12 @@ typedef struct _CLzmaEncProps
   int fb;          /* 5 <= fb <= 273, default = 32 */
   int btMode;      /* 0 - hashChain Mode, 1 - binTree mode - normal, default = 1 */
   int numHashBytes; /* 2, 3 or 4, default = 4 */
-  UInt32 mc;        /* 1 <= mc <= (1 << 30), default = 32 */
+  UInt32 mc;       /* 1 <= mc <= (1 << 30), default = 32 */
   unsigned writeEndMark;  /* 0 - do not write EOPM, 1 - write EOPM, default = 0 */
   int numThreads;  /* 1 or 2, default = 2 */
+
+  UInt64 reduceSize; /* estimated size of data that will be compressed. default = (UInt64)(Int64)-1.
+                        Encoder uses this value to reduce dictionary size */
 } CLzmaEncProps;
 
 void LzmaEncProps_Init(CLzmaEncProps *p);
@@ -37,41 +38,38 @@ UInt32 LzmaEncProps_GetDictSize(const CLzmaEncProps *props2);
 
 /* ---------- CLzmaEncHandle Interface ---------- */
 
-/* LzmaEnc_* functions can return the following exit codes:
+/* LzmaEnc* functions can return the following exit codes:
-Returns:
+SRes:
   SZ_OK           - OK
   SZ_ERROR_MEM    - Memory allocation error
   SZ_ERROR_PARAM  - Incorrect paramater in props
-  SZ_ERROR_WRITE  - Write callback error.
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
   SZ_ERROR_PROGRESS - some break from progress callback
-  SZ_ERROR_THREAD - errors in multithreading functions (only for Mt version)
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
 */
 
 typedef void * CLzmaEncHandle;
 
-CLzmaEncHandle LzmaEnc_Create(ISzAlloc *alloc);
+CLzmaEncHandle LzmaEnc_Create(ISzAllocPtr alloc);
-void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAlloc *alloc, ISzAlloc *allocBig);
+void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAllocPtr alloc, ISzAllocPtr allocBig);
+
 SRes LzmaEnc_SetProps(CLzmaEncHandle p, const CLzmaEncProps *props);
+void LzmaEnc_SetDataSize(CLzmaEncHandle p, UInt64 expectedDataSiize);
 SRes LzmaEnc_WriteProperties(CLzmaEncHandle p, Byte *properties, SizeT *size);
+unsigned LzmaEnc_IsWriteEndMark(CLzmaEncHandle p);
+
 SRes LzmaEnc_Encode(CLzmaEncHandle p, ISeqOutStream *outStream, ISeqInStream *inStream,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+    ICompressProgress *progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 SRes LzmaEnc_MemEncode(CLzmaEncHandle p, Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
-    int writeEndMark, ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+    int writeEndMark, ICompressProgress *progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
+
 
 /* ---------- One Call Interface ---------- */
 
-/* LzmaEncode
-Return code:
-  SZ_OK               - OK
-  SZ_ERROR_MEM        - Memory allocation error
-  SZ_ERROR_PARAM      - Incorrect paramater
-  SZ_ERROR_OUTPUT_EOF - output buffer overflow
-  SZ_ERROR_THREAD     - errors in multithreading functions (only for Mt version)
-*/
-
 SRes LzmaEncode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
     const CLzmaEncProps *props, Byte *propsEncoded, SizeT *propsSize, int writeEndMark,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+    ICompressProgress *progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 
 EXTERN_C_END
 

C/MtCoder.c

@@ -1,170 +1,89 @@
 /* MtCoder.c -- Multi-thread Coder
-2010-09-24 : Igor Pavlov : Public domain */
+2018-02-21 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
-#include <stdio.h>
-
 #include "MtCoder.h"
 
-void LoopThread_Construct(CLoopThread *p)
+#ifndef _7ZIP_ST
-{
-  Thread_Construct(&p->thread);
-  Event_Construct(&p->startEvent);
-  Event_Construct(&p->finishedEvent);
-}
 
-void LoopThread_Close(CLoopThread *p)
+SRes MtProgressThunk_Progress(const ICompressProgress *pp, UInt64 inSize, UInt64 outSize)
 {
-  Thread_Close(&p->thread);
+  CMtProgressThunk *thunk = CONTAINER_FROM_VTBL(pp, CMtProgressThunk, vt);
-  Event_Close(&p->startEvent);
+  UInt64 inSize2 = 0;
-  Event_Close(&p->finishedEvent);
+  UInt64 outSize2 = 0;
-}
+  if (inSize != (UInt64)(Int64)-1)
-
-static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE LoopThreadFunc(void *pp)
-{
-  CLoopThread *p = (CLoopThread *)pp;
-  for (;;)
   {
-    if (Event_Wait(&p->startEvent) != 0)
+    inSize2 = inSize - thunk->inSize;
-      return SZ_ERROR_THREAD;
+    thunk->inSize = inSize;
-    if (p->stop)
+  }
-      return 0;
+  if (outSize != (UInt64)(Int64)-1)
-    p->res = p->func(p->param);
+  {
-    if (Event_Set(&p->finishedEvent) != 0)
+    outSize2 = outSize - thunk->outSize;
-      return SZ_ERROR_THREAD;
+    thunk->outSize = outSize;
+  }
+  return MtProgress_ProgressAdd(thunk->mtProgress, inSize2, outSize2);
+}
+
+
+void MtProgressThunk_CreateVTable(CMtProgressThunk *p)
+{
+  p->vt.Progress = MtProgressThunk_Progress;
+}
+
+
+
+#define RINOK_THREAD(x) { if ((x) != 0) return SZ_ERROR_THREAD; }
+
+
+static WRes ArEvent_OptCreate_And_Reset(CEvent *p)
+{
+  if (Event_IsCreated(p))
+    return Event_Reset(p);
+  return AutoResetEvent_CreateNotSignaled(p);
+}
+
+
+static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE ThreadFunc(void *pp);
+
+
+static SRes MtCoderThread_CreateAndStart(CMtCoderThread *t)
+{
+  WRes wres = ArEvent_OptCreate_And_Reset(&t->startEvent);
+  if (wres == 0)
+  {
+    t->stop = False;
+    if (!Thread_WasCreated(&t->thread))
+      wres = Thread_Create(&t->thread, ThreadFunc, t);
+    if (wres == 0)
+      wres = Event_Set(&t->startEvent);
+  }
+  if (wres == 0)
+    return SZ_OK;
+  return MY_SRes_HRESULT_FROM_WRes(wres);
+}
+
+
+static void MtCoderThread_Destruct(CMtCoderThread *t)
+{
+  if (Thread_WasCreated(&t->thread))
+  {
+    t->stop = 1;
+    Event_Set(&t->startEvent);
+    Thread_Wait(&t->thread);
+    Thread_Close(&t->thread);
+  }
+
+  Event_Close(&t->startEvent);
+
+  if (t->inBuf)
+  {
+    ISzAlloc_Free(t->mtCoder->allocBig, t->inBuf);
+    t->inBuf = NULL;
   }
 }
 
-WRes LoopThread_Create(CLoopThread *p)
-{
-  p->stop = 0;
-  RINOK(AutoResetEvent_CreateNotSignaled(&p->startEvent));
-  RINOK(AutoResetEvent_CreateNotSignaled(&p->finishedEvent));
-  return Thread_Create(&p->thread, LoopThreadFunc, p);
-}
 
-WRes LoopThread_StopAndWait(CLoopThread *p)
-{
-  p->stop = 1;
-  if (Event_Set(&p->startEvent) != 0)
-    return SZ_ERROR_THREAD;
-  return Thread_Wait(&p->thread);
-}
-
-WRes LoopThread_StartSubThread(CLoopThread *p) { return Event_Set(&p->startEvent); }
-WRes LoopThread_WaitSubThread(CLoopThread *p) { return Event_Wait(&p->finishedEvent); }
-
-static SRes Progress(ICompressProgress *p, UInt64 inSize, UInt64 outSize)
-{
-  return (p && p->Progress(p, inSize, outSize) != SZ_OK) ? SZ_ERROR_PROGRESS : SZ_OK;
-}
-
-static void MtProgress_Init(CMtProgress *p, ICompressProgress *progress)
-{
-  unsigned i;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
-    p->inSizes[i] = p->outSizes[i] = 0;
-  p->totalInSize = p->totalOutSize = 0;
-  p->progress = progress;
-  p->res = SZ_OK;
-}
-
-static void MtProgress_Reinit(CMtProgress *p, unsigned index)
-{
-  p->inSizes[index] = 0;
-  p->outSizes[index] = 0;
-}
-
-#define UPDATE_PROGRESS(size, prev, total) \
-  if (size != (UInt64)(Int64)-1) { total += size - prev; prev = size; }
-
-SRes MtProgress_Set(CMtProgress *p, unsigned index, UInt64 inSize, UInt64 outSize)
-{
-  SRes res;
-  CriticalSection_Enter(&p->cs);
-  UPDATE_PROGRESS(inSize, p->inSizes[index], p->totalInSize)
-  UPDATE_PROGRESS(outSize, p->outSizes[index], p->totalOutSize)
-  if (p->res == SZ_OK)
-    p->res = Progress(p->progress, p->totalInSize, p->totalOutSize);
-  res = p->res;
-  CriticalSection_Leave(&p->cs);
-  return res;
-}
-
-static void MtProgress_SetError(CMtProgress *p, SRes res)
-{
-  CriticalSection_Enter(&p->cs);
-  if (p->res == SZ_OK)
-    p->res = res;
-  CriticalSection_Leave(&p->cs);
-}
-
-static void MtCoder_SetError(CMtCoder* p, SRes res)
-{
-  CriticalSection_Enter(&p->cs);
-  if (p->res == SZ_OK)
-    p->res = res;
-  CriticalSection_Leave(&p->cs);
-}
-
-/* ---------- MtThread ---------- */
-
-void CMtThread_Construct(CMtThread *p, CMtCoder *mtCoder)
-{
-  p->mtCoder = mtCoder;
-  p->outBuf = 0;
-  p->inBuf = 0;
-  Event_Construct(&p->canRead);
-  Event_Construct(&p->canWrite);
-  LoopThread_Construct(&p->thread);
-}
-
-#define RINOK_THREAD(x) { if((x) != 0) return SZ_ERROR_THREAD; }
-
-static void CMtThread_CloseEvents(CMtThread *p)
-{
-  Event_Close(&p->canRead);
-  Event_Close(&p->canWrite);
-}
-
-static void CMtThread_Destruct(CMtThread *p)
-{
-  CMtThread_CloseEvents(p);
-
-  if (Thread_WasCreated(&p->thread.thread))
-  {
-    LoopThread_StopAndWait(&p->thread);
-    LoopThread_Close(&p->thread);
-  }
-
-  if (p->mtCoder->alloc)
-    IAlloc_Free(p->mtCoder->alloc, p->outBuf);
-  p->outBuf = 0;
-
-  if (p->mtCoder->alloc)
-    IAlloc_Free(p->mtCoder->alloc, p->inBuf);
-  p->inBuf = 0;
-}
-
-#define MY_BUF_ALLOC(buf, size, newSize) \
-  if (buf == 0 || size != newSize) \
-  { IAlloc_Free(p->mtCoder->alloc, buf); \
-    size = newSize; buf = (Byte *)IAlloc_Alloc(p->mtCoder->alloc, size); \
-    if (buf == 0) return SZ_ERROR_MEM; }
-
-static SRes CMtThread_Prepare(CMtThread *p)
-{
-  MY_BUF_ALLOC(p->inBuf, p->inBufSize, p->mtCoder->blockSize)
-  MY_BUF_ALLOC(p->outBuf, p->outBufSize, p->mtCoder->destBlockSize)
-
-  p->stopReading = False;
-  p->stopWriting = False;
-  RINOK_THREAD(AutoResetEvent_CreateNotSignaled(&p->canRead));
-  RINOK_THREAD(AutoResetEvent_CreateNotSignaled(&p->canWrite));
-
-  return SZ_OK;
-}
 
 static SRes FullRead(ISeqInStream *stream, Byte *data, size_t *processedSize)
 {
@@ -172,158 +91,511 @@ static SRes FullRead(ISeqInStream *stream, Byte *data, size_t *processedSize)
   *processedSize = 0;
   while (size != 0)
   {
-    size_t curSize = size;
+    size_t cur = size;
-    SRes res = stream->Read(stream, data, &curSize);
+    SRes res = ISeqInStream_Read(stream, data, &cur);
-    *processedSize += curSize;
+    *processedSize += cur;
-    data += curSize;
+    data += cur;
-    size -= curSize;
+    size -= cur;
     RINOK(res);
-    if (curSize == 0)
+    if (cur == 0)
       return SZ_OK;
   }
   return SZ_OK;
 }
 
-#define GET_NEXT_THREAD(p) &p->mtCoder->threads[p->index == p->mtCoder->numThreads  - 1 ? 0 : p->index + 1]
 
-static SRes MtThread_Process(CMtThread *p, Bool *stop)
+/*
+  ThreadFunc2() returns:
+  SZ_OK           - in all normal cases (even for stream error or memory allocation error)
+  SZ_ERROR_THREAD - in case of failure in system synch function
+*/
+
+static SRes ThreadFunc2(CMtCoderThread *t)
 {
-  CMtThread *next;
+  CMtCoder *mtc = t->mtCoder;
-  *stop = True;
-  if (Event_Wait(&p->canRead) != 0)
-    return SZ_ERROR_THREAD;
-  
-  next = GET_NEXT_THREAD(p);
-  
-  if (p->stopReading)
-  {
-    next->stopReading = True;
-    return Event_Set(&next->canRead) == 0 ? SZ_OK : SZ_ERROR_THREAD;
-  }
 
-  {
-    size_t size = p->mtCoder->blockSize;
-    size_t destSize = p->outBufSize;
-
-    RINOK(FullRead(p->mtCoder->inStream, p->inBuf, &size));
-    next->stopReading = *stop = (size != p->mtCoder->blockSize);
-    if (Event_Set(&next->canRead) != 0)
-      return SZ_ERROR_THREAD;
-
-    RINOK(p->mtCoder->mtCallback->Code(p->mtCoder->mtCallback, p->index,
-        p->outBuf, &destSize, p->inBuf, size, *stop));
-
-    MtProgress_Reinit(&p->mtCoder->mtProgress, p->index);
-
-    if (Event_Wait(&p->canWrite) != 0)
-      return SZ_ERROR_THREAD;
-    if (p->stopWriting)
-      return SZ_ERROR_FAIL;
-    if (p->mtCoder->outStream->Write(p->mtCoder->outStream, p->outBuf, destSize) != destSize)
-      return SZ_ERROR_WRITE;
-    return Event_Set(&next->canWrite) == 0 ? SZ_OK : SZ_ERROR_THREAD;
-  }
-}
-
-static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE ThreadFunc(void *pp)
-{
-  CMtThread *p = (CMtThread *)pp;
   for (;;)
   {
-    Bool stop;
+    unsigned bi;
-    CMtThread *next = GET_NEXT_THREAD(p);
+    SRes res;
-    SRes res = MtThread_Process(p, &stop);
+    SRes res2;
-    if (res != SZ_OK)
+    Bool finished;
+    unsigned bufIndex;
+    size_t size;
+    const Byte *inData;
+    UInt64 readProcessed = 0;
+    
+    RINOK_THREAD(Event_Wait(&mtc->readEvent))
+
+    /* after Event_Wait(&mtc->readEvent) we must call Event_Set(&mtc->readEvent) in any case to unlock another threads */
+
+    if (mtc->stopReading)
     {
-      MtCoder_SetError(p->mtCoder, res);
+      return Event_Set(&mtc->readEvent) == 0 ? SZ_OK : SZ_ERROR_THREAD;
-      MtProgress_SetError(&p->mtCoder->mtProgress, res);
-      next->stopReading = True;
-      next->stopWriting = True;
-      Event_Set(&next->canRead);
-      Event_Set(&next->canWrite);
-      return res;
     }
-    if (stop)
+
+    res = MtProgress_GetError(&mtc->mtProgress);
+    
+    size = 0;
+    inData = NULL;
+    finished = True;
+
+    if (res == SZ_OK)
+    {
+      size = mtc->blockSize;
+      if (mtc->inStream)
+      {
+        if (!t->inBuf)
+        {
+          t->inBuf = (Byte *)ISzAlloc_Alloc(mtc->allocBig, mtc->blockSize);
+          if (!t->inBuf)
+            res = SZ_ERROR_MEM;
+        }
+        if (res == SZ_OK)
+        {
+          res = FullRead(mtc->inStream, t->inBuf, &size);
+          readProcessed = mtc->readProcessed + size;
+          mtc->readProcessed = readProcessed;
+        }
+        if (res != SZ_OK)
+        {
+          mtc->readRes = res;
+          /* after reading error - we can stop encoding of previous blocks */
+          MtProgress_SetError(&mtc->mtProgress, res);
+        }
+        else
+          finished = (size != mtc->blockSize);
+      }
+      else
+      {
+        size_t rem;
+        readProcessed = mtc->readProcessed;
+        rem = mtc->inDataSize - (size_t)readProcessed;
+        if (size > rem)
+          size = rem;
+        inData = mtc->inData + (size_t)readProcessed;
+        readProcessed += size;
+        mtc->readProcessed = readProcessed;
+        finished = (mtc->inDataSize == (size_t)readProcessed);
+      }
+    }
+
+    /* we must get some block from blocksSemaphore before Event_Set(&mtc->readEvent) */
+
+    res2 = SZ_OK;
+
+    if (Semaphore_Wait(&mtc->blocksSemaphore) != 0)
+    {
+      res2 = SZ_ERROR_THREAD;
+      if (res == SZ_OK)
+      {
+        res = res2;
+        // MtProgress_SetError(&mtc->mtProgress, res);
+      }
+    }
+
+    bi = mtc->blockIndex;
+
+    if (++mtc->blockIndex >= mtc->numBlocksMax)
+      mtc->blockIndex = 0;
+
+    bufIndex = (unsigned)(int)-1;
+
+    if (res == SZ_OK)
+      res = MtProgress_GetError(&mtc->mtProgress);
+
+    if (res != SZ_OK)
+      finished = True;
+
+    if (!finished)
+    {
+      if (mtc->numStartedThreads < mtc->numStartedThreadsLimit
+          && mtc->expectedDataSize != readProcessed)
+      {
+        res = MtCoderThread_CreateAndStart(&mtc->threads[mtc->numStartedThreads]);
+        if (res == SZ_OK)
+          mtc->numStartedThreads++;
+        else
+        {
+          MtProgress_SetError(&mtc->mtProgress, res);
+          finished = True;
+        }
+      }
+    }
+
+    if (finished)
+      mtc->stopReading = True;
+
+    RINOK_THREAD(Event_Set(&mtc->readEvent))
+
+    if (res2 != SZ_OK)
+      return res2;
+
+    if (res == SZ_OK)
+    {
+      CriticalSection_Enter(&mtc->cs);
+      bufIndex = mtc->freeBlockHead;
+      mtc->freeBlockHead = mtc->freeBlockList[bufIndex];
+      CriticalSection_Leave(&mtc->cs);
+      
+      res = mtc->mtCallback->Code(mtc->mtCallbackObject, t->index, bufIndex,
+          mtc->inStream ? t->inBuf : inData, size, finished);
+      
+      // MtProgress_Reinit(&mtc->mtProgress, t->index);
+
+      if (res != SZ_OK)
+        MtProgress_SetError(&mtc->mtProgress, res);
+    }
+
+    {
+      CMtCoderBlock *block = &mtc->blocks[bi];
+      block->res = res;
+      block->bufIndex = bufIndex;
+      block->finished = finished;
+    }
+    
+    #ifdef MTCODER__USE_WRITE_THREAD
+      RINOK_THREAD(Event_Set(&mtc->writeEvents[bi]))
+    #else
+    {
+      unsigned wi;
+      {
+        CriticalSection_Enter(&mtc->cs);
+        wi = mtc->writeIndex;
+        if (wi == bi)
+          mtc->writeIndex = (unsigned)(int)-1;
+        else
+          mtc->ReadyBlocks[bi] = True;
+        CriticalSection_Leave(&mtc->cs);
+      }
+
+      if (wi != bi)
+      {
+        if (res != SZ_OK || finished)
+          return 0;
+        continue;
+      }
+
+      if (mtc->writeRes != SZ_OK)
+        res = mtc->writeRes;
+
+      for (;;)
+      {
+        if (res == SZ_OK && bufIndex != (unsigned)(int)-1)
+        {
+          res = mtc->mtCallback->Write(mtc->mtCallbackObject, bufIndex);
+          if (res != SZ_OK)
+          {
+            mtc->writeRes = res;
+            MtProgress_SetError(&mtc->mtProgress, res);
+          }
+        }
+
+        if (++wi >= mtc->numBlocksMax)
+          wi = 0;
+        {
+          Bool isReady;
+
+          CriticalSection_Enter(&mtc->cs);
+          
+          if (bufIndex != (unsigned)(int)-1)
+          {
+            mtc->freeBlockList[bufIndex] = mtc->freeBlockHead;
+            mtc->freeBlockHead = bufIndex;
+          }
+          
+          isReady = mtc->ReadyBlocks[wi];
+          
+          if (isReady)
+            mtc->ReadyBlocks[wi] = False;
+          else
+            mtc->writeIndex = wi;
+          
+          CriticalSection_Leave(&mtc->cs);
+
+          RINOK_THREAD(Semaphore_Release1(&mtc->blocksSemaphore))
+
+          if (!isReady)
+            break;
+        }
+
+        {
+          CMtCoderBlock *block = &mtc->blocks[wi];
+          if (res == SZ_OK && block->res != SZ_OK)
+            res = block->res;
+          bufIndex = block->bufIndex;
+          finished = block->finished;
+        }
+      }
+    }
+    #endif
+      
+    if (finished || res != SZ_OK)
       return 0;
   }
 }
 
-void MtCoder_Construct(CMtCoder* p)
+
+static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE ThreadFunc(void *pp)
+{
+  CMtCoderThread *t = (CMtCoderThread *)pp;
+  for (;;)
+  {
+    if (Event_Wait(&t->startEvent) != 0)
+      return SZ_ERROR_THREAD;
+    if (t->stop)
+      return 0;
+    {
+      SRes res = ThreadFunc2(t);
+      CMtCoder *mtc = t->mtCoder;
+      if (res != SZ_OK)
+      {
+        MtProgress_SetError(&mtc->mtProgress, res);
+      }
+      
+      #ifndef MTCODER__USE_WRITE_THREAD
+      {
+        unsigned numFinished = (unsigned)InterlockedIncrement(&mtc->numFinishedThreads);
+        if (numFinished == mtc->numStartedThreads)
+          if (Event_Set(&mtc->finishedEvent) != 0)
+            return SZ_ERROR_THREAD;
+      }
+      #endif
+    }
+  }
+}
+
+
+
+void MtCoder_Construct(CMtCoder *p)
 {
   unsigned i;
-  p->alloc = 0;
+  
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
+  p->blockSize = 0;
+  p->numThreadsMax = 0;
+  p->expectedDataSize = (UInt64)(Int64)-1;
+
+  p->inStream = NULL;
+  p->inData = NULL;
+  p->inDataSize = 0;
+
+  p->progress = NULL;
+  p->allocBig = NULL;
+
+  p->mtCallback = NULL;
+  p->mtCallbackObject = NULL;
+
+  p->allocatedBufsSize = 0;
+
+  Event_Construct(&p->readEvent);
+  Semaphore_Construct(&p->blocksSemaphore);
+
+  for (i = 0; i < MTCODER__THREADS_MAX; i++)
   {
-    CMtThread *t = &p->threads[i];
+    CMtCoderThread *t = &p->threads[i];
+    t->mtCoder = p;
     t->index = i;
-    CMtThread_Construct(t, p);
+    t->inBuf = NULL;
+    t->stop = False;
+    Event_Construct(&t->startEvent);
+    Thread_Construct(&t->thread);
   }
+
+  #ifdef MTCODER__USE_WRITE_THREAD
+    for (i = 0; i < MTCODER__BLOCKS_MAX; i++)
+      Event_Construct(&p->writeEvents[i]);
+  #else
+    Event_Construct(&p->finishedEvent);
+  #endif
+
   CriticalSection_Init(&p->cs);
   CriticalSection_Init(&p->mtProgress.cs);
 }
 
-void MtCoder_Destruct(CMtCoder* p)
+
+
+
+static void MtCoder_Free(CMtCoder *p)
 {
   unsigned i;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
+
-    CMtThread_Destruct(&p->threads[i]);
+  /*
+  p->stopReading = True;
+  if (Event_IsCreated(&p->readEvent))
+    Event_Set(&p->readEvent);
+  */
+
+  for (i = 0; i < MTCODER__THREADS_MAX; i++)
+    MtCoderThread_Destruct(&p->threads[i]);
+
+  Event_Close(&p->readEvent);
+  Semaphore_Close(&p->blocksSemaphore);
+
+  #ifdef MTCODER__USE_WRITE_THREAD
+    for (i = 0; i < MTCODER__BLOCKS_MAX; i++)
+      Event_Close(&p->writeEvents[i]);
+  #else
+    Event_Close(&p->finishedEvent);
+  #endif
+}
+
+
+void MtCoder_Destruct(CMtCoder *p)
+{
+  MtCoder_Free(p);
+
   CriticalSection_Delete(&p->cs);
   CriticalSection_Delete(&p->mtProgress.cs);
 }
 
+
 SRes MtCoder_Code(CMtCoder *p)
 {
-  unsigned i, numThreads = p->numThreads;
+  unsigned numThreads = p->numThreadsMax;
+  unsigned numBlocksMax;
+  unsigned i;
   SRes res = SZ_OK;
-  p->res = SZ_OK;
+
+  if (numThreads > MTCODER__THREADS_MAX)
+    numThreads = MTCODER__THREADS_MAX;
+  numBlocksMax = MTCODER__GET_NUM_BLOCKS_FROM_THREADS(numThreads);
+  
+  if (p->blockSize < ((UInt32)1 << 26)) numBlocksMax++;
+  if (p->blockSize < ((UInt32)1 << 24)) numBlocksMax++;
+  if (p->blockSize < ((UInt32)1 << 22)) numBlocksMax++;
+
+  if (numBlocksMax > MTCODER__BLOCKS_MAX)
+    numBlocksMax = MTCODER__BLOCKS_MAX;
+
+  if (p->blockSize != p->allocatedBufsSize)
+  {
+    for (i = 0; i < MTCODER__THREADS_MAX; i++)
+    {
+      CMtCoderThread *t = &p->threads[i];
+      if (t->inBuf)
+      {
+        ISzAlloc_Free(p->allocBig, t->inBuf);
+        t->inBuf = NULL;
+      }
+    }
+    p->allocatedBufsSize = p->blockSize;
+  }
+
+  p->readRes = SZ_OK;
 
   MtProgress_Init(&p->mtProgress, p->progress);
 
-  for (i = 0; i < numThreads; i++)
+  #ifdef MTCODER__USE_WRITE_THREAD
+    for (i = 0; i < numBlocksMax; i++)
+    {
+      RINOK_THREAD(ArEvent_OptCreate_And_Reset(&p->writeEvents[i]));
+    }
+  #else
+    RINOK_THREAD(ArEvent_OptCreate_And_Reset(&p->finishedEvent));
+  #endif
+
   {
-    RINOK(CMtThread_Prepare(&p->threads[i]));
+    RINOK_THREAD(ArEvent_OptCreate_And_Reset(&p->readEvent));
+
+    if (Semaphore_IsCreated(&p->blocksSemaphore))
+    {
+      RINOK_THREAD(Semaphore_Close(&p->blocksSemaphore));
+    }
+    RINOK_THREAD(Semaphore_Create(&p->blocksSemaphore, numBlocksMax, numBlocksMax));
   }
 
-  for (i = 0; i < numThreads; i++)
+  for (i = 0; i < MTCODER__BLOCKS_MAX - 1; i++)
+    p->freeBlockList[i] = i + 1;
+  p->freeBlockList[MTCODER__BLOCKS_MAX - 1] = (unsigned)(int)-1;
+  p->freeBlockHead = 0;
+
+  p->readProcessed = 0;
+  p->blockIndex = 0;
+  p->numBlocksMax = numBlocksMax;
+  p->stopReading = False;
+
+  #ifndef MTCODER__USE_WRITE_THREAD
+    p->writeIndex = 0;
+    p->writeRes = SZ_OK;
+    for (i = 0; i < MTCODER__BLOCKS_MAX; i++)
+      p->ReadyBlocks[i] = False;
+    p->numFinishedThreads = 0;
+  #endif
+
+  p->numStartedThreadsLimit = numThreads;
+  p->numStartedThreads = 0;
+
+  // for (i = 0; i < numThreads; i++)
   {
-    CMtThread *t = &p->threads[i];
+    CMtCoderThread *nextThread = &p->threads[p->numStartedThreads++];
-    CLoopThread *lt = &t->thread;
+    RINOK(MtCoderThread_CreateAndStart(nextThread));
+  }
 
-    if (!Thread_WasCreated(&lt->thread))
+  RINOK_THREAD(Event_Set(&p->readEvent))
+
+  #ifdef MTCODER__USE_WRITE_THREAD
+  {
+    unsigned bi = 0;
+
+    for (;; bi++)
     {
-      lt->func = ThreadFunc;
+      if (bi >= numBlocksMax)
-      lt->param = t;
+        bi = 0;
+
+      RINOK_THREAD(Event_Wait(&p->writeEvents[bi]))
 
-      if (LoopThread_Create(lt) != SZ_OK)
       {
-        res = SZ_ERROR_THREAD;
+        const CMtCoderBlock *block = &p->blocks[bi];
-        break;
+        unsigned bufIndex = block->bufIndex;
+        Bool finished = block->finished;
+        if (res == SZ_OK && block->res != SZ_OK)
+          res = block->res;
+
+        if (bufIndex != (unsigned)(int)-1)
+        {
+          if (res == SZ_OK)
+          {
+            res = p->mtCallback->Write(p->mtCallbackObject, bufIndex);
+            if (res != SZ_OK)
+              MtProgress_SetError(&p->mtProgress, res);
+          }
+          
+          CriticalSection_Enter(&p->cs);
+          {
+            p->freeBlockList[bufIndex] = p->freeBlockHead;
+            p->freeBlockHead = bufIndex;
+          }
+          CriticalSection_Leave(&p->cs);
+        }
+        
+        RINOK_THREAD(Semaphore_Release1(&p->blocksSemaphore))
+
+        if (finished)
+          break;
       }
     }
   }
+  #else
+  {
+    WRes wres = Event_Wait(&p->finishedEvent);
+    res = MY_SRes_HRESULT_FROM_WRes(wres);
+  }
+  #endif
 
   if (res == SZ_OK)
-  {
+    res = p->readRes;
-    unsigned j;
-    for (i = 0; i < numThreads; i++)
-    {
-      CMtThread *t = &p->threads[i];
-      if (LoopThread_StartSubThread(&t->thread) != SZ_OK)
-      {
-        res = SZ_ERROR_THREAD;
-        p->threads[0].stopReading = True;
-        break;
-      }
-    }
 
-    Event_Set(&p->threads[0].canWrite);
+  if (res == SZ_OK)
-    Event_Set(&p->threads[0].canRead);
+    res = p->mtProgress.res;
 
-    for (j = 0; j < i; j++)
+  #ifndef MTCODER__USE_WRITE_THREAD
-      LoopThread_WaitSubThread(&p->threads[j].thread);
+    if (res == SZ_OK)
-  }
+      res = p->writeRes;
+  #endif
 
-  for (i = 0; i < numThreads; i++)
+  if (res != SZ_OK)
-    CMtThread_CloseEvents(&p->threads[i]);
+    MtCoder_Free(p);
-  return (res == SZ_OK) ? p->res : res;
+  return res;
 }
+
+#endif

C/MtCoder.h

@@ -1,98 +1,141 @@
 /* MtCoder.h -- Multi-thread Coder
-2009-11-19 : Igor Pavlov : Public domain */
+2018-02-21 : Igor Pavlov : Public domain */
 
 #ifndef __MT_CODER_H
 #define __MT_CODER_H
 
-#include "Threads.h"
+#include "MtDec.h"
 
 EXTERN_C_BEGIN
 
-typedef struct
+/*
-{
+  if (    defined MTCODER__USE_WRITE_THREAD) : main thread writes all data blocks to output stream
-  CThread thread;
+  if (not defined MTCODER__USE_WRITE_THREAD) : any coder thread can write data blocks to output stream
-  CAutoResetEvent startEvent;
+*/
-  CAutoResetEvent finishedEvent;
+/* #define MTCODER__USE_WRITE_THREAD */
-  int stop;
-  
-  THREAD_FUNC_TYPE func;
-  LPVOID param;
-  THREAD_FUNC_RET_TYPE res;
-} CLoopThread;
-
-void LoopThread_Construct(CLoopThread *p);
-void LoopThread_Close(CLoopThread *p);
-WRes LoopThread_Create(CLoopThread *p);
-WRes LoopThread_StopAndWait(CLoopThread *p);
-WRes LoopThread_StartSubThread(CLoopThread *p);
-WRes LoopThread_WaitSubThread(CLoopThread *p);
 
 #ifndef _7ZIP_ST
-#define NUM_MT_CODER_THREADS_MAX 32
+  #define MTCODER__GET_NUM_BLOCKS_FROM_THREADS(numThreads) ((numThreads) + (numThreads) / 8 + 1)
+  #define MTCODER__THREADS_MAX 64
+  #define MTCODER__BLOCKS_MAX (MTCODER__GET_NUM_BLOCKS_FROM_THREADS(MTCODER__THREADS_MAX) + 3)
 #else
-#define NUM_MT_CODER_THREADS_MAX 1
+  #define MTCODER__THREADS_MAX 1
+  #define MTCODER__BLOCKS_MAX 1
 #endif
 
+
+#ifndef _7ZIP_ST
+
+
 typedef struct
 {
-  UInt64 totalInSize;
+  ICompressProgress vt;
-  UInt64 totalOutSize;
+  CMtProgress *mtProgress;
-  ICompressProgress *progress;
+  UInt64 inSize;
-  SRes res;
+  UInt64 outSize;
-  CCriticalSection cs;
+} CMtProgressThunk;
-  UInt64 inSizes[NUM_MT_CODER_THREADS_MAX];
+
-  UInt64 outSizes[NUM_MT_CODER_THREADS_MAX];
+void MtProgressThunk_CreateVTable(CMtProgressThunk *p);
-} CMtProgress;
+    
+#define MtProgressThunk_Init(p) { (p)->inSize = 0; (p)->outSize = 0; }
 
-SRes MtProgress_Set(CMtProgress *p, unsigned index, UInt64 inSize, UInt64 outSize);
 
 struct _CMtCoder;
 
+
 typedef struct
 {
   struct _CMtCoder *mtCoder;
-  Byte *outBuf;
-  size_t outBufSize;
-  Byte *inBuf;
-  size_t inBufSize;
   unsigned index;
-  CLoopThread thread;
+  int stop;
+  Byte *inBuf;
+
+  CAutoResetEvent startEvent;
+  CThread thread;
+} CMtCoderThread;
 
-  Bool stopReading;
-  Bool stopWriting;
-  CAutoResetEvent canRead;
-  CAutoResetEvent canWrite;
-} CMtThread;
 
 typedef struct
 {
-  SRes (*Code)(void *p, unsigned index, Byte *dest, size_t *destSize,
+  SRes (*Code)(void *p, unsigned coderIndex, unsigned outBufIndex,
       const Byte *src, size_t srcSize, int finished);
-} IMtCoderCallback;
+  SRes (*Write)(void *p, unsigned outBufIndex);
+} IMtCoderCallback2;
+
+
+typedef struct
+{
+  SRes res;
+  unsigned bufIndex;
+  Bool finished;
+} CMtCoderBlock;
+
 
 typedef struct _CMtCoder
 {
-  size_t blockSize;
+  /* input variables */
-  size_t destBlockSize;
-  unsigned numThreads;
   
-  ISeqInStream *inStream;
+  size_t blockSize;        /* size of input block */
-  ISeqOutStream *outStream;
+  unsigned numThreadsMax;
-  ICompressProgress *progress;
+  UInt64 expectedDataSize;
-  ISzAlloc *alloc;
+
+  ISeqInStream *inStream;
+  const Byte *inData;
+  size_t inDataSize;
+
+  ICompressProgress *progress;
+  ISzAllocPtr allocBig;
+
+  IMtCoderCallback2 *mtCallback;
+  void *mtCallbackObject;
+
+  
+  /* internal variables */
+  
+  size_t allocatedBufsSize;
+
+  CAutoResetEvent readEvent;
+  CSemaphore blocksSemaphore;
+
+  Bool stopReading;
+  SRes readRes;
+
+  #ifdef MTCODER__USE_WRITE_THREAD
+    CAutoResetEvent writeEvents[MTCODER__BLOCKS_MAX];
+  #else
+    CAutoResetEvent finishedEvent;
+    SRes writeRes;
+    unsigned writeIndex;
+    Byte ReadyBlocks[MTCODER__BLOCKS_MAX];
+    LONG numFinishedThreads;
+  #endif
+
+  unsigned numStartedThreadsLimit;
+  unsigned numStartedThreads;
+
+  unsigned numBlocksMax;
+  unsigned blockIndex;
+  UInt64 readProcessed;
 
-  IMtCoderCallback *mtCallback;
   CCriticalSection cs;
-  SRes res;
+
+  unsigned freeBlockHead;
+  unsigned freeBlockList[MTCODER__BLOCKS_MAX];
 
   CMtProgress mtProgress;
-  CMtThread threads[NUM_MT_CODER_THREADS_MAX];
+  CMtCoderBlock blocks[MTCODER__BLOCKS_MAX];
+  CMtCoderThread threads[MTCODER__THREADS_MAX];
 } CMtCoder;
 
-void MtCoder_Construct(CMtCoder* p);
+
-void MtCoder_Destruct(CMtCoder* p);
+void MtCoder_Construct(CMtCoder *p);
+void MtCoder_Destruct(CMtCoder *p);
 SRes MtCoder_Code(CMtCoder *p);
 
+
+#endif
+
+
 EXTERN_C_END
 
 #endif

C/MtDec.h

@@ -0,0 +1,201 @@
+/* MtDec.h -- Multi-thread Decoder
+2018-03-02 : Igor Pavlov : Public domain */
+
+#ifndef __MT_DEC_H
+#define __MT_DEC_H
+
+#include "7zTypes.h"
+
+#ifndef _7ZIP_ST
+#include "Threads.h"
+#endif
+
+EXTERN_C_BEGIN
+
+#ifndef _7ZIP_ST
+
+#ifndef _7ZIP_ST
+  #define MTDEC__THREADS_MAX 32
+#else
+  #define MTDEC__THREADS_MAX 1
+#endif
+
+
+typedef struct
+{
+  ICompressProgress *progress;
+  SRes res;
+  UInt64 totalInSize;
+  UInt64 totalOutSize;
+  CCriticalSection cs;
+} CMtProgress;
+
+void MtProgress_Init(CMtProgress *p, ICompressProgress *progress);
+SRes MtProgress_Progress_ST(CMtProgress *p);
+SRes MtProgress_ProgressAdd(CMtProgress *p, UInt64 inSize, UInt64 outSize);
+SRes MtProgress_GetError(CMtProgress *p);
+void MtProgress_SetError(CMtProgress *p, SRes res);
+
+struct _CMtDec;
+
+typedef struct
+{
+  struct _CMtDec *mtDec;
+  unsigned index;
+  void *inBuf;
+
+  size_t inDataSize_Start; // size of input data in start block
+  UInt64 inDataSize;       // total size of input data in all blocks
+
+  CThread thread;
+  CAutoResetEvent canRead;
+  CAutoResetEvent canWrite;
+  void  *allocaPtr;
+} CMtDecThread;
+
+void MtDecThread_FreeInBufs(CMtDecThread *t);
+
+
+typedef enum
+{
+  MTDEC_PARSE_CONTINUE, // continue this block with more input data
+  MTDEC_PARSE_OVERFLOW, // MT buffers overflow, need switch to single-thread
+  MTDEC_PARSE_NEW,      // new block
+  MTDEC_PARSE_END       // end of block threading. But we still can return to threading after Write(&needContinue)
+} EMtDecParseState;
+
+typedef struct
+{
+  // in
+  int startCall;
+  const Byte *src;
+  size_t srcSize;
+      // in  : (srcSize == 0) is allowed
+      // out : it's allowed to return less that actually was used ?
+  int srcFinished;
+
+  // out
+  EMtDecParseState state;
+  Bool canCreateNewThread;
+  UInt64 outPos; // check it (size_t)
+} CMtDecCallbackInfo;
+
+
+typedef struct
+{
+  void (*Parse)(void *p, unsigned coderIndex, CMtDecCallbackInfo *ci);
+  
+  // PreCode() and Code():
+  // (SRes_return_result != SZ_OK) means stop decoding, no need another blocks
+  SRes (*PreCode)(void *p, unsigned coderIndex);
+  SRes (*Code)(void *p, unsigned coderIndex,
+      const Byte *src, size_t srcSize, int srcFinished,
+      UInt64 *inCodePos, UInt64 *outCodePos, int *stop);
+  // stop - means stop another Code calls
+
+
+  /* Write() must be called, if Parse() was called
+      set (needWrite) if
+      {
+         && (was not interrupted by progress)
+         && (was not interrupted in previous block)
+      }
+
+    out:
+      if (*needContinue), decoder still need to continue decoding with new iteration,
+         even after MTDEC_PARSE_END
+      if (*canRecode), we didn't flush current block data, so we still can decode current block later.
+  */
+  SRes (*Write)(void *p, unsigned coderIndex,
+      Bool needWriteToStream,
+      const Byte *src, size_t srcSize,
+      // int srcFinished,
+      Bool *needContinue,
+      Bool *canRecode);
+} IMtDecCallback;
+
+
+
+typedef struct _CMtDec
+{
+  /* input variables */
+  
+  size_t inBufSize;        /* size of input block */
+  unsigned numThreadsMax;
+  // size_t inBlockMax;
+  unsigned numThreadsMax_2;
+
+  ISeqInStream *inStream;
+  // const Byte *inData;
+  // size_t inDataSize;
+
+  ICompressProgress *progress;
+  ISzAllocPtr alloc;
+
+  IMtDecCallback *mtCallback;
+  void *mtCallbackObject;
+
+  
+  /* internal variables */
+  
+  size_t allocatedBufsSize;
+
+  Bool exitThread;
+  WRes exitThreadWRes;
+
+  UInt64 blockIndex;
+  Bool isAllocError;
+  Bool overflow;
+  SRes threadingErrorSRes;
+
+  Bool needContinue;
+
+  // CAutoResetEvent finishedEvent;
+
+  SRes readRes;
+  SRes codeRes;
+
+  Bool wasInterrupted;
+
+  unsigned numStartedThreads_Limit;
+  unsigned numStartedThreads;
+
+  Byte *crossBlock;
+  size_t crossStart;
+  size_t crossEnd;
+  UInt64 readProcessed;
+  Bool readWasFinished;
+  UInt64 inProcessed;
+
+  unsigned filledThreadStart;
+  unsigned numFilledThreads;
+
+  #ifndef _7ZIP_ST
+  Bool needInterrupt;
+  UInt64 interruptIndex;
+  CMtProgress mtProgress;
+  CMtDecThread threads[MTDEC__THREADS_MAX];
+  #endif
+} CMtDec;
+
+
+void MtDec_Construct(CMtDec *p);
+void MtDec_Destruct(CMtDec *p);
+
+/*
+MtDec_Code() returns:
+  SZ_OK - in most cases
+  MY_SRes_HRESULT_FROM_WRes(WRes_error) - in case of unexpected error in threading function
+*/
+  
+SRes MtDec_Code(CMtDec *p);
+Byte *MtDec_GetCrossBuff(CMtDec *p);
+
+int MtDec_PrepareRead(CMtDec *p);
+const Byte *MtDec_Read(CMtDec *p, size_t *inLim);
+
+#endif
+
+EXTERN_C_END
+
+#endif

C/Ppmd.h

@@ -1,5 +1,5 @@
 /* Ppmd.h -- PPMD codec common code
-2013-01-18 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 #ifndef __PPMD_H
@@ -77,8 +77,8 @@ typedef
   CPpmd_Byte_Ref;
 
 #define PPMD_SetAllBitsIn256Bytes(p) \
-  { unsigned i; for (i = 0; i < 256 / sizeof(p[0]); i += 8) { \
+  { size_t z; for (z = 0; z < 256 / sizeof(p[0]); z += 8) { \
-  p[i+7] = p[i+6] = p[i+5] = p[i+4] = p[i+3] = p[i+2] = p[i+1] = p[i+0] = ~(size_t)0; }}
+  p[z+7] = p[z+6] = p[z+5] = p[z+4] = p[z+3] = p[z+2] = p[z+1] = p[z+0] = ~(size_t)0; }}
 
 EXTERN_C_END
  

C/Ppmd7.c

@@ -1,10 +1,10 @@
 /* Ppmd7.c -- PPMdH codec
-2010-03-12 : Igor Pavlov : Public domain
+2017-04-03 : 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"
 
@@ -15,7 +15,7 @@ static const UInt16 kInitBinEsc[] = { 0x3CDD, 0x1F3F, 0x59BF, 0x48F3, 0x64A1, 0x
 #define UNIT_SIZE 12
 
 #define U2B(nu) ((UInt32)(nu) * UNIT_SIZE)
-#define U2I(nu) (p->Units2Indx[(nu) - 1])
+#define U2I(nu) (p->Units2Indx[(size_t)(nu) - 1])
 #define I2U(indx) (p->Indx2Units[indx])
 
 #ifdef PPMD_32BIT
@@ -66,7 +66,7 @@ void Ppmd7_Construct(CPpmd7 *p)
   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;
   }
 
@@ -88,29 +88,31 @@ void Ppmd7_Construct(CPpmd7 *p)
   memset(p->HB2Flag + 0x40, 8, 0x100 - 0x40);
 }
 
-void Ppmd7_Free(CPpmd7 *p, ISzAlloc *alloc)
+void Ppmd7_Free(CPpmd7 *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->Base);
+  ISzAlloc_Free(alloc, p->Base);
   p->Size = 0;
   p->Base = 0;
 }
 
-Bool Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAlloc *alloc)
+Bool Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAllocPtr alloc)
 {
-  if (p->Base == 0 || p->Size != size)
+  if (!p->Base || p->Size != size)
   {
+    size_t size2;
     Ppmd7_Free(p, alloc);
+    size2 = 0
+      #ifndef PPMD_32BIT
+      + UNIT_SIZE
+      #endif
+      ;
     p->AlignOffset =
       #ifdef PPMD_32BIT
         (4 - size) & 3;
       #else
         4 - (size & 3);
       #endif
-    if ((p->Base = (Byte *)alloc->Alloc(alloc, p->AlignOffset + size
+    if ((p->Base = (Byte *)ISzAlloc_Alloc(alloc, p->AlignOffset + size + size2)) == 0)
-        #ifndef PPMD_32BIT
-        + UNIT_SIZE
-        #endif
-        )) == 0)
       return False;
     p->Size = size;
   }
@@ -257,7 +259,7 @@ static void *AllocUnits(CPpmd7 *p, unsigned indx)
 
 #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)
 {
@@ -513,7 +515,7 @@ static void UpdateModel(CPpmd7 *p)
         /* Expand for one UNIT */
         unsigned oldNU = ns1 >> 1;
         unsigned i = U2I(oldNU);
-        if (i != U2I(oldNU + 1))
+        if (i != U2I((size_t)oldNU + 1))
         {
           void *ptr = AllocUnits(p, i + 1);
           void *oldPtr;
@@ -639,10 +641,10 @@ CPpmd_See *Ppmd7_MakeEscFreq(CPpmd7 *p, unsigned numMasked, UInt32 *escFreq)
   unsigned nonMasked = p->MinContext->NumStats - numMasked;
   if (p->MinContext->NumStats != 256)
   {
-    see = p->See[p->NS2Indx[nonMasked - 1]] +
+    see = p->See[(unsigned)p->NS2Indx[(size_t)nonMasked - 1]] +
         (nonMasked < (unsigned)SUFFIX(p->MinContext)->NumStats - p->MinContext->NumStats) +
-        2 * (p->MinContext->SummFreq < 11 * p->MinContext->NumStats) +
+        2 * (unsigned)(p->MinContext->SummFreq < 11 * p->MinContext->NumStats) +
-        4 * (numMasked > nonMasked) +
+        4 * (unsigned)(numMasked > nonMasked) +
         p->HiBitsFlag;
     {
       unsigned r = (see->Summ >> see->Shift);

C/Ppmd7.h

@@ -1,5 +1,5 @@
 /* Ppmd7.h -- PPMdH compression codec
-2010-03-12 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 /* This code supports virtual RangeDecoder and includes the implementation
@@ -60,8 +60,8 @@ typedef struct
 } CPpmd7;
 
 void Ppmd7_Construct(CPpmd7 *p);
-Bool Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAlloc *alloc);
+Bool Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAllocPtr alloc);
-void Ppmd7_Free(CPpmd7 *p, ISzAlloc *alloc);
+void Ppmd7_Free(CPpmd7 *p, ISzAllocPtr alloc);
 void Ppmd7_Init(CPpmd7 *p, unsigned maxOrder);
 #define Ppmd7_WasAllocated(p) ((p)->Base != NULL)
 
@@ -86,10 +86,10 @@ void Ppmd7_Update2(CPpmd7 *p);
 void Ppmd7_UpdateBin(CPpmd7 *p);
 
 #define Ppmd7_GetBinSumm(p) \
-    &p->BinSumm[Ppmd7Context_OneState(p->MinContext)->Freq - 1][p->PrevSuccess + \
+    &p->BinSumm[(size_t)(unsigned)Ppmd7Context_OneState(p->MinContext)->Freq - 1][p->PrevSuccess + \
-    p->NS2BSIndx[Ppmd7_GetContext(p, p->MinContext->Suffix)->NumStats - 1] + \
+    p->NS2BSIndx[(size_t)Ppmd7_GetContext(p, p->MinContext->Suffix)->NumStats - 1] + \
     (p->HiBitsFlag = p->HB2Flag[p->FoundState->Symbol]) + \
-    2 * p->HB2Flag[Ppmd7Context_OneState(p->MinContext)->Symbol] + \
+    2 * p->HB2Flag[(unsigned)Ppmd7Context_OneState(p->MinContext)->Symbol] + \
     ((p->RunLength >> 26) & 0x20)]
 
 CPpmd_See *Ppmd7_MakeEscFreq(CPpmd7 *p, unsigned numMasked, UInt32 *scale);
@@ -97,16 +97,18 @@ CPpmd_See *Ppmd7_MakeEscFreq(CPpmd7 *p, unsigned numMasked, UInt32 *scale);
 
 /* ---------- Decode ---------- */
 
-typedef struct
+typedef struct IPpmd7_RangeDec IPpmd7_RangeDec;
+
+struct IPpmd7_RangeDec
 {
-  UInt32 (*GetThreshold)(void *p, UInt32 total);
+  UInt32 (*GetThreshold)(const IPpmd7_RangeDec *p, UInt32 total);
-  void (*Decode)(void *p, UInt32 start, UInt32 size);
+  void (*Decode)(const IPpmd7_RangeDec *p, UInt32 start, UInt32 size);
-  UInt32 (*DecodeBit)(void *p, UInt32 size0);
+  UInt32 (*DecodeBit)(const IPpmd7_RangeDec *p, UInt32 size0);
-} IPpmd7_RangeDec;
+};
 
 typedef struct
 {
-  IPpmd7_RangeDec p;
+  IPpmd7_RangeDec vt;
   UInt32 Range;
   UInt32 Code;
   IByteIn *Stream;
@@ -116,7 +118,7 @@ void Ppmd7z_RangeDec_CreateVTable(CPpmd7z_RangeDec *p);
 Bool Ppmd7z_RangeDec_Init(CPpmd7z_RangeDec *p);
 #define Ppmd7z_RangeDec_IsFinishedOK(p) ((p)->Code == 0)
 
-int Ppmd7_DecodeSymbol(CPpmd7 *p, IPpmd7_RangeDec *rc);
+int Ppmd7_DecodeSymbol(CPpmd7 *p, const IPpmd7_RangeDec *rc);
 
 
 /* ---------- Encode ---------- */

C/Ppmd7Dec.c

@@ -1,5 +1,5 @@
 /* Ppmd7Dec.c -- PPMdH Decoder
-2010-03-12 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 #include "Precomp.h"
@@ -13,44 +13,46 @@ Bool Ppmd7z_RangeDec_Init(CPpmd7z_RangeDec *p)
   unsigned i;
   p->Code = 0;
   p->Range = 0xFFFFFFFF;
-  if (p->Stream->Read((void *)p->Stream) != 0)
+  if (IByteIn_Read(p->Stream) != 0)
     return False;
   for (i = 0; i < 4; i++)
-    p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
+    p->Code = (p->Code << 8) | IByteIn_Read(p->Stream);
   return (p->Code < 0xFFFFFFFF);
 }
 
-static UInt32 Range_GetThreshold(void *pp, UInt32 total)
+#define GET_Ppmd7z_RangeDec CPpmd7z_RangeDec *p = CONTAINER_FROM_VTBL(pp, CPpmd7z_RangeDec, vt);
+ 
+static UInt32 Range_GetThreshold(const IPpmd7_RangeDec *pp, UInt32 total)
 {
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
+  GET_Ppmd7z_RangeDec
-  return (p->Code) / (p->Range /= total);
+  return p->Code / (p->Range /= total);
 }
 
 static void Range_Normalize(CPpmd7z_RangeDec *p)
 {
   if (p->Range < kTopValue)
   {
-    p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
+    p->Code = (p->Code << 8) | IByteIn_Read(p->Stream);
     p->Range <<= 8;
     if (p->Range < kTopValue)
     {
-      p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
+      p->Code = (p->Code << 8) | IByteIn_Read(p->Stream);
       p->Range <<= 8;
     }
   }
 }
 
-static void Range_Decode(void *pp, UInt32 start, UInt32 size)
+static void Range_Decode(const IPpmd7_RangeDec *pp, UInt32 start, UInt32 size)
 {
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
+  GET_Ppmd7z_RangeDec
   p->Code -= start * p->Range;
   p->Range *= size;
   Range_Normalize(p);
 }
 
-static UInt32 Range_DecodeBit(void *pp, UInt32 size0)
+static UInt32 Range_DecodeBit(const IPpmd7_RangeDec *pp, UInt32 size0)
 {
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
+  GET_Ppmd7z_RangeDec
   UInt32 newBound = (p->Range >> 14) * size0;
   UInt32 symbol;
   if (p->Code < newBound)
@@ -70,15 +72,15 @@ static UInt32 Range_DecodeBit(void *pp, UInt32 size0)
 
 void Ppmd7z_RangeDec_CreateVTable(CPpmd7z_RangeDec *p)
 {
-  p->p.GetThreshold = Range_GetThreshold;
+  p->vt.GetThreshold = Range_GetThreshold;
-  p->p.Decode = Range_Decode;
+  p->vt.Decode = Range_Decode;
-  p->p.DecodeBit = Range_DecodeBit;
+  p->vt.DecodeBit = Range_DecodeBit;
 }
 
 
 #define MASK(sym) ((signed char *)charMask)[sym]
 
-int Ppmd7_DecodeSymbol(CPpmd7 *p, IPpmd7_RangeDec *rc)
+int Ppmd7_DecodeSymbol(CPpmd7 *p, const IPpmd7_RangeDec *rc)
 {
   size_t charMask[256 / sizeof(size_t)];
   if (p->MinContext->NumStats != 1)

C/Ppmd7Enc.c

@@ -1,5 +1,5 @@
 /* Ppmd7Enc.c -- PPMdH Encoder
-2010-03-12 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
 #include "Precomp.h"
@@ -23,10 +23,10 @@ static void RangeEnc_ShiftLow(CPpmd7z_RangeEnc *p)
     Byte temp = p->Cache;
     do
     {
-      p->Stream->Write(p->Stream, (Byte)(temp + (Byte)(p->Low >> 32)));
+      IByteOut_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++;

C/Ppmd8.c

@@ -1,10 +1,10 @@
 /* Ppmd8.c -- PPMdI codec
-2013-11-12 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on PPMd var.I (2002): Dmitry Shkarin : Public domain */
 
 #include "Precomp.h"
 
-#include <memory.h>
+#include <string.h>
 
 #include "Ppmd8.h"
 
@@ -15,7 +15,7 @@ static const UInt16 kInitBinEsc[] = { 0x3CDD, 0x1F3F, 0x59BF, 0x48F3, 0x64A1, 0x
 #define UNIT_SIZE 12
 
 #define U2B(nu) ((UInt32)(nu) * UNIT_SIZE)
-#define U2I(nu) (p->Units2Indx[(nu) - 1])
+#define U2I(nu) (p->Units2Indx[(size_t)(nu) - 1])
 #define I2U(indx) (p->Indx2Units[indx])
 
 #ifdef PPMD_32BIT
@@ -67,7 +67,7 @@ void Ppmd8_Construct(CPpmd8 *p)
   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;
   }
 
@@ -86,16 +86,16 @@ void Ppmd8_Construct(CPpmd8 *p)
   }
 }
 
-void Ppmd8_Free(CPpmd8 *p, ISzAlloc *alloc)
+void Ppmd8_Free(CPpmd8 *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->Base);
+  ISzAlloc_Free(alloc, p->Base);
   p->Size = 0;
   p->Base = 0;
 }
 
-Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAlloc *alloc)
+Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAllocPtr alloc)
 {
-  if (p->Base == 0 || p->Size != size)
+  if (!p->Base || p->Size != size)
   {
     Ppmd8_Free(p, alloc);
     p->AlignOffset =
@@ -104,7 +104,7 @@ Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAlloc *alloc)
       #else
         4 - (size & 3);
       #endif
-    if ((p->Base = (Byte *)alloc->Alloc(alloc, p->AlignOffset + size)) == 0)
+    if ((p->Base = (Byte *)ISzAlloc_Alloc(alloc, p->AlignOffset + size)) == 0)
       return False;
     p->Size = size;
   }
@@ -240,8 +240,8 @@ static void *AllocUnits(CPpmd8 *p, unsigned indx)
 }
 
 #define MyMem12Cpy(dest, src, num) \
-  { UInt32 *d = (UInt32 *)dest; const UInt32 *s = (const UInt32 *)src; UInt32 n = num; \
+  { UInt32 *d = (UInt32 *)dest; const UInt32 *z = (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] = z[0]; d[1] = z[1]; d[2] = z[2]; z += 3; d += 3; } while (--n); }
 
 static void *ShrinkUnits(CPpmd8 *p, void *oldPtr, unsigned oldNU, unsigned newNU)
 {
@@ -386,7 +386,7 @@ static void RestartModel(CPpmd8 *p)
 
   for (i = m = 0; m < 24; m++)
   {
-    while (p->NS2Indx[i + 3] == m + 3)
+    while (p->NS2Indx[(size_t)i + 3] == m + 3)
       i++;
     for (k = 0; k < 32; k++)
     {
@@ -772,7 +772,7 @@ static CTX_PTR ReduceOrder(CPpmd8 *p, CPpmd_State *s1, CTX_PTR c)
   if (SUCCESSOR(s) <= upBranch)
   {
     CTX_PTR successor;
-    CPpmd_State *s1 = p->FoundState;
+    CPpmd_State *s2 = p->FoundState;
     p->FoundState = s;
 
     successor = CreateSuccessors(p, False, NULL, c);
@@ -780,7 +780,7 @@ static CTX_PTR ReduceOrder(CPpmd8 *p, CPpmd_State *s1, CTX_PTR c)
       SetSuccessor(s, 0);
     else
       SetSuccessor(s, REF(successor));
-    p->FoundState = s1;
+    p->FoundState = s2;
   }
   
   if (p->OrderFall == 1 && c1 == p->MaxContext)
@@ -905,7 +905,7 @@ static void UpdateModel(CPpmd8 *p)
         /* Expand for one UNIT */
         unsigned oldNU = (ns1 + 1) >> 1;
         unsigned i = U2I(oldNU);
-        if (i != U2I(oldNU + 1))
+        if (i != U2I((size_t)oldNU + 1))
         {
           void *ptr = AllocUnits(p, i + 1);
           void *oldPtr;
@@ -924,19 +924,19 @@ static void UpdateModel(CPpmd8 *p)
     }
     else
     {
-      CPpmd_State *s = (CPpmd_State*)AllocUnits(p, 0);
+      CPpmd_State *s2 = (CPpmd_State*)AllocUnits(p, 0);
-      if (!s)
+      if (!s2)
       {
         RESTORE_MODEL(c, CTX(fSuccessor));
         return;
       }
-      *s = *ONE_STATE(c);
+      *s2 = *ONE_STATE(c);
-      c->Stats = REF(s);
+      c->Stats = REF(s2);
-      if (s->Freq < MAX_FREQ / 4 - 1)
+      if (s2->Freq < MAX_FREQ / 4 - 1)
-        s->Freq <<= 1;
+        s2->Freq <<= 1;
       else
-        s->Freq = MAX_FREQ - 4;
+        s2->Freq = MAX_FREQ - 4;
-      c->SummFreq = (UInt16)(s->Freq + p->InitEsc + (ns > 2));
+      c->SummFreq = (UInt16)(s2->Freq + p->InitEsc + (ns > 2));
     }
     cf = 2 * fFreq * (c->SummFreq + 6);
     sf = (UInt32)s0 + c->SummFreq;
@@ -951,10 +951,10 @@ static void UpdateModel(CPpmd8 *p)
       c->SummFreq = (UInt16)(c->SummFreq + cf);
     }
     {
-      CPpmd_State *s = STATS(c) + ns1 + 1;
+      CPpmd_State *s2 = STATS(c) + ns1 + 1;
-      SetSuccessor(s, successor);
+      SetSuccessor(s2, successor);
-      s->Symbol = fSymbol;
+      s2->Symbol = fSymbol;
-      s->Freq = (Byte)cf;
+      s2->Freq = (Byte)cf;
       c->Flags |= flag;
       c->NumStats = (Byte)(ns1 + 1);
     }
@@ -1038,9 +1038,9 @@ CPpmd_See *Ppmd8_MakeEscFreq(CPpmd8 *p, unsigned numMasked1, UInt32 *escFreq)
   CPpmd_See *see;
   if (p->MinContext->NumStats != 0xFF)
   {
-    see = p->See[p->NS2Indx[p->MinContext->NumStats + 2] - 3] +
+    see = p->See[(size_t)(unsigned)p->NS2Indx[(size_t)(unsigned)p->MinContext->NumStats + 2] - 3] +
         (p->MinContext->SummFreq > 11 * ((unsigned)p->MinContext->NumStats + 1)) +
-        2 * (2 * (unsigned)p->MinContext->NumStats <
+        2 * (unsigned)(2 * (unsigned)p->MinContext->NumStats <
         ((unsigned)SUFFIX(p->MinContext)->NumStats + numMasked1)) +
         p->MinContext->Flags;
     {

C/Ppmd8.h

@@ -1,5 +1,5 @@
 /* Ppmd8.h -- PPMdI codec
-2011-01-27 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
@@ -86,8 +86,8 @@ typedef struct
 } CPpmd8;
 
 void Ppmd8_Construct(CPpmd8 *p);
-Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAlloc *alloc);
+Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAllocPtr alloc);
-void Ppmd8_Free(CPpmd8 *p, ISzAlloc *alloc);
+void Ppmd8_Free(CPpmd8 *p, ISzAllocPtr alloc);
 void Ppmd8_Init(CPpmd8 *p, unsigned maxOrder, unsigned restoreMethod);
 #define Ppmd8_WasAllocated(p) ((p)->Base != NULL)
 
@@ -112,7 +112,7 @@ void Ppmd8_Update2(CPpmd8 *p);
 void Ppmd8_UpdateBin(CPpmd8 *p);
 
 #define Ppmd8_GetBinSumm(p) \
-    &p->BinSumm[p->NS2Indx[Ppmd8Context_OneState(p->MinContext)->Freq - 1]][ \
+    &p->BinSumm[p->NS2Indx[(size_t)Ppmd8Context_OneState(p->MinContext)->Freq - 1]][ \
     p->NS2BSIndx[Ppmd8_GetContext(p, p->MinContext->Suffix)->NumStats] + \
     p->PrevSuccess + p->MinContext->Flags + ((p->RunLength >> 26) & 0x20)]
 

C/Ppmd8Dec.c

@@ -1,5 +1,5 @@
 /* Ppmd8Dec.c -- PPMdI Decoder
-2010-04-16 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
@@ -18,7 +18,7 @@ Bool Ppmd8_RangeDec_Init(CPpmd8 *p)
   p->Range = 0xFFFFFFFF;
   p->Code = 0;
   for (i = 0; i < 4; i++)
-    p->Code = (p->Code << 8) | p->Stream.In->Read(p->Stream.In);
+    p->Code = (p->Code << 8) | IByteIn_Read(p->Stream.In);
   return (p->Code < 0xFFFFFFFF);
 }
 
@@ -37,7 +37,7 @@ static void RangeDec_Decode(CPpmd8 *p, UInt32 start, UInt32 size)
   while ((p->Low ^ (p->Low + p->Range)) < kTop ||
       (p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1)))
   {
-    p->Code = (p->Code << 8) | p->Stream.In->Read(p->Stream.In);
+    p->Code = (p->Code << 8) | IByteIn_Read(p->Stream.In);
     p->Range <<= 8;
     p->Low <<= 8;
   }

C/Ppmd8Enc.c

@@ -1,5 +1,5 @@
 /* Ppmd8Enc.c -- PPMdI Encoder
-2010-04-16 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
@@ -15,7 +15,7 @@ void Ppmd8_RangeEnc_FlushData(CPpmd8 *p)
 {
   unsigned i;
   for (i = 0; i < 4; i++, p->Low <<= 8 )
-    p->Stream.Out->Write(p->Stream.Out, (Byte)(p->Low >> 24));
+    IByteOut_Write(p->Stream.Out, (Byte)(p->Low >> 24));
 }
 
 static void RangeEnc_Normalize(CPpmd8 *p)
@@ -23,7 +23,7 @@ static void RangeEnc_Normalize(CPpmd8 *p)
   while ((p->Low ^ (p->Low + p->Range)) < kTop ||
       (p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1)))
   {
-    p->Stream.Out->Write(p->Stream.Out, (Byte)(p->Low >> 24));
+    IByteOut_Write(p->Stream.Out, (Byte)(p->Low >> 24));
     p->Range <<= 8;
     p->Low <<= 8;
   }

C/Sha1.c

@@ -1,5 +1,5 @@
 /* Sha1.c -- SHA-1 Hash
-2015-05-10 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on public domain code of Steve Reid from Wei Dai's Crypto++ library. */
 
 #include "Precomp.h"
@@ -62,8 +62,8 @@ This code is based on public domain code of Steve Reid from Wei Dai's Crypto++ l
 
 #else
   
-#define RX_15  { unsigned i; for (i = 0; i < 15; i += 5) { RX_5(R0, i); } }
+#define RX_15  { size_t i; for (i = 0; i < 15; i += 5) { RX_5(R0, i); } }
-#define RX_20(rx, ii)  { unsigned i; i = ii; for (; i < ii + 20; i += 5) { RX_5(rx, i); } }
+#define RX_20(rx, ii)  { size_t i; i = ii; for (; i < ii + 20; i += 5) { RX_5(rx, i); } }
 
 #endif
 
@@ -131,7 +131,7 @@ void Sha1_UpdateBlock_Rar(CSha1 *p, UInt32 *data, int returnRes)
 
   if (returnRes)
   {
-    unsigned i;
+    size_t i;
     for (i = 0 ; i < SHA1_NUM_BLOCK_WORDS; i++)
       data[i] = W[kNumW - SHA1_NUM_BLOCK_WORDS + i];
   }
@@ -151,18 +151,23 @@ void Sha1_Update(CSha1 *p, const Byte *data, size_t size)
   
   if (pos2 != 0)
   {
-    UInt32 w = ((UInt32)data[0]) << 24;
+    UInt32 w;
-    if (--size && pos2 < 3)
+    pos2 = (3 - pos2) * 8;
+    w = ((UInt32)*data++) << pos2;
+    if (--size && pos2)
     {
-      w |= ((UInt32)data[1]) << 16;
+      pos2 -= 8;
-      if (--size && pos2 < 2)
+      w |= ((UInt32)*data++) << pos2;
+      if (--size && pos2)
       {
-        w |= ((UInt32)data[2]) << 8;
+        pos2 -= 8;
-        --size;
+        w |= ((UInt32)*data++) << pos2;
+        size--;
       }
     }
-    data += 4 - pos2;
+    p->buffer[pos] |= w;
-    p->buffer[pos++] |= (w >> (8 * pos2));
+    if (pos2 == 0)
+      pos++;
   }
 
   for (;;)
@@ -171,7 +176,7 @@ void Sha1_Update(CSha1 *p, const Byte *data, size_t size)
     {
       for (;;)
       {
-        unsigned i;
+        size_t i;
         Sha1_UpdateBlock(p);
         if (size < SHA1_BLOCK_SIZE)
           break;
@@ -207,7 +212,7 @@ void Sha1_Update(CSha1 *p, const Byte *data, size_t size)
   }
 }
 
-void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size, int rar350Mode)
+void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size /* , int rar350Mode */)
 {
   int returnRes = False;
   
@@ -233,7 +238,7 @@ void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size, int rar350Mode)
       Sha1_UpdateBlock_Rar(p, p->buffer, returnRes);
       if (returnRes)
       {
-        unsigned i;
+        size_t i;
         for (i = 0; i < SHA1_NUM_BLOCK_WORDS; i++)
         {
           UInt32 d = p->buffer[i];
@@ -241,7 +246,8 @@ void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size, int rar350Mode)
           SetUi32(prev, d);
         }
       }
-      returnRes = rar350Mode;
+      // returnRes = rar350Mode;
+      returnRes = True;
     }
   }
 }

C/Sha1.h

@@ -1,5 +1,5 @@
 /* Sha1.h -- SHA-1 Hash
-2015-03-04 : Igor Pavlov : Public domain */
+2016-05-20 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_SHA1_H
 #define __7Z_SHA1_H
@@ -27,7 +27,7 @@ void Sha1_GetBlockDigest(CSha1 *p, const UInt32 *data, UInt32 *destDigest);
 void Sha1_Update(CSha1 *p, const Byte *data, size_t size);
 void Sha1_Final(CSha1 *p, Byte *digest);
 
-void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size, int rar350Mode);
+void Sha1_Update_Rar(CSha1 *p, Byte *data, size_t size /* , int rar350Mode */);
 
 void Sha1_32_PrepareBlock(const CSha1 *p, UInt32 *block, unsigned size);
 void Sha1_32_Update(CSha1 *p, const UInt32 *data, size_t size);

C/Sha256.c

@@ -1,5 +1,5 @@
 /* Crypto/Sha256.c -- SHA-256 Hash
-2015-03-02 : Igor Pavlov : Public domain
+2017-04-03 : Igor Pavlov : Public domain
 This code is based on public domain code from Wei Dai's Crypto++ library. */
 
 #include "Precomp.h"
@@ -45,7 +45,7 @@ void Sha256_Init(CSha256 *p)
 #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)); \
+    h += S1(e) + Ch(e,f,g) + K[(i)+(size_t)(j)] + (j ? blk2(i) : blk0(i)); \
     d += h; \
     h += S0(a) + Maj(a, b, c)
 
@@ -73,7 +73,7 @@ void Sha256_Init(CSha256 *p)
 #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)); \
+    h(i) += S1(e(i)) + Ch(e(i),f(i),g(i)) + K[(i)+(size_t)(j)] + (j ? blk2(i) : blk0(i)); \
     d(i) += h(i); \
     h(i) += S0(a(i)) + Maj(a(i), b(i), c(i)) \
 
@@ -113,10 +113,26 @@ static void Sha256_WriteByteBlock(CSha256 *p)
 {
   UInt32 W[16];
   unsigned j;
-  UInt32 *state = p->state;
+  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];
@@ -126,17 +142,10 @@ static void Sha256_WriteByteBlock(CSha256 *p)
   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 < 16; j += 2)
-  {
-    W[j    ] = GetBe32(p->buffer + j * 4);
-    W[j + 1] = GetBe32(p->buffer + j * 4 + 4);
-  }
-
   for (j = 0; j < 64; j += 16)
   {
     RX_16
@@ -226,11 +235,13 @@ void Sha256_Final(CSha256 *p, Byte *digest)
   
   Sha256_WriteByteBlock(p);
 
-  for (i = 0; i < 8; i++)
+  for (i = 0; i < 8; i += 2)
   {
-    UInt32 v = p->state[i];
+    UInt32 v0 = p->state[i];
-    SetBe32(digest, v);
+    UInt32 v1 = p->state[i + 1];
-    digest += 4;
+    SetBe32(digest    , v0);
+    SetBe32(digest + 4, v1);
+    digest += 8;
   }
   
   Sha256_Init(p);

C/Threads.c

@@ -1,5 +1,5 @@
 /* Threads.c -- multithreading library
-2014-09-21 : Igor Pavlov : Public domain */
+2017-06-26 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -12,18 +12,20 @@
 static WRes GetError()
 {
   DWORD res = GetLastError();
-  return (res) ? (WRes)(res) : 1;
+  return res ? (WRes)res : 1;
 }
 
-WRes HandleToWRes(HANDLE h) { return (h != 0) ? 0 : GetError(); }
+static WRes HandleToWRes(HANDLE h) { return (h != NULL) ? 0 : GetError(); }
-WRes BOOLToWRes(BOOL v) { return v ? 0 : GetError(); }
+static WRes BOOLToWRes(BOOL v) { return v ? 0 : GetError(); }
 
 WRes HandlePtr_Close(HANDLE *p)
 {
   if (*p != NULL)
+  {
     if (!CloseHandle(*p))
       return GetError();
-  *p = NULL;
+    *p = NULL;
+  }
   return 0;
 }
 
@@ -49,7 +51,7 @@ WRes Thread_Create(CThread *p, THREAD_FUNC_TYPE func, LPVOID param)
   return HandleToWRes(*p);
 }
 
-WRes Event_Create(CEvent *p, BOOL manualReset, int signaled)
+static WRes Event_Create(CEvent *p, BOOL manualReset, int signaled)
 {
   *p = CreateEvent(NULL, manualReset, (signaled ? TRUE : FALSE), NULL);
   return HandleToWRes(*p);

C/Threads.h

@@ -1,5 +1,5 @@
 /* Threads.h -- multithreading library
-2013-11-12 : Igor Pavlov : Public domain */
+2017-06-18 : Igor Pavlov : Public domain */
 
 #ifndef __7Z_THREADS_H
 #define __7Z_THREADS_H
@@ -49,7 +49,8 @@ WRes AutoResetEvent_Create(CAutoResetEvent *p, int signaled);
 WRes AutoResetEvent_CreateNotSignaled(CAutoResetEvent *p);
 
 typedef HANDLE CSemaphore;
-#define Semaphore_Construct(p) (*p) = NULL
+#define Semaphore_Construct(p) *(p) = NULL
+#define Semaphore_IsCreated(p) (*(p) != NULL)
 #define Semaphore_Close(p) HandlePtr_Close(p)
 #define Semaphore_Wait(p) Handle_WaitObject(*(p))
 WRes Semaphore_Create(CSemaphore *p, UInt32 initCount, UInt32 maxCount);

C/Util/7z/7zMain.c

@@ -1,11 +1,13 @@
 /* 7zMain.c - Test application for 7z Decoder
-2015-08-02 : Igor Pavlov : Public domain */
+2018-04-19 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
 #include <stdio.h>
 #include <string.h>
 
+#include "../../CpuArch.h"
+
 #include "../../7z.h"
 #include "../../7zAlloc.h"
 #include "../../7zBuf.h"
@@ -23,7 +25,17 @@
 #endif
 #endif
 
-static ISzAlloc g_Alloc = { SzAlloc, SzFree };
+
+#define kInputBufSize ((size_t)1 << 18)
+
+static const ISzAlloc g_Alloc = { SzAlloc, SzFree };
+
+
+static void Print(const char *s)
+{
+  fputs(s, stdout);
+}
+
 
 static int Buf_EnsureSize(CBuf *dest, size_t size)
 {
@@ -227,6 +239,7 @@ static WRes OutFile_OpenUtf16(CSzFile *p, const UInt16 *name)
   #endif
 }
 
+
 static SRes PrintString(const UInt16 *s)
 {
   CBuf buf;
@@ -238,12 +251,12 @@ static SRes PrintString(const UInt16 *s)
       #endif
       );
   if (res == SZ_OK)
-    fputs((const char *)buf.data, stdout);
+    Print((const char *)buf.data);
   Buf_Free(&buf, &g_Alloc);
   return res;
 }
 
-static void UInt64ToStr(UInt64 value, char *s)
+static void UInt64ToStr(UInt64 value, char *s, int numDigits)
 {
   char temp[32];
   int pos = 0;
@@ -253,6 +266,10 @@ static void UInt64ToStr(UInt64 value, char *s)
     value /= 10;
   }
   while (value != 0);
+
+  for (numDigits -= pos; numDigits > 0; numDigits--)
+    *s++ = ' ';
+
   do
     *s++ = temp[--pos];
   while (pos);
@@ -266,8 +283,10 @@ static char *UIntToStr(char *s, unsigned value, int numDigits)
   do
     temp[pos++] = (char)('0' + (value % 10));
   while (value /= 10);
+
   for (numDigits -= pos; numDigits > 0; numDigits--)
     *s++ = '0';
+
   do
     *s++ = temp[--pos];
   while (pos);
@@ -310,10 +329,10 @@ static void ConvertFileTimeToString(const CNtfsFileTime *nt, char *s)
     ms[1] = 29;
   for (mon = 0;; mon++)
   {
-    unsigned s = ms[mon];
+    unsigned d = ms[mon];
-    if (v < s)
+    if (v < d)
       break;
-    v -= s;
+    v -= d;
   }
   s = UIntToStr(s, year, 4); *s++ = '-';
   UIntToStr_2(s, mon + 1); s[2] = '-'; s += 3;
@@ -323,9 +342,16 @@ static void ConvertFileTimeToString(const CNtfsFileTime *nt, char *s)
   UIntToStr_2(s, sec); s[2] = 0;
 }
 
-void PrintError(char *sz)
+static void PrintLF()
 {
-  printf("\nERROR: %s\n", sz);
+  Print("\n");
+}
+
+static void PrintError(char *s)
+{
+  Print("\nERROR: ");
+  Print(s);
+  PrintLF();
 }
 
 static void GetAttribString(UInt32 wa, Bool isDir, char *s)
@@ -343,25 +369,27 @@ static void GetAttribString(UInt32 wa, Bool isDir, char *s)
   #endif
 }
 
+
 // #define NUM_PARENTS_MAX 128
 
 int MY_CDECL main(int numargs, char *args[])
 {
-  CFileInStream archiveStream;
-  CLookToRead lookStream;
-  CSzArEx db;
-  SRes res;
   ISzAlloc allocImp;
   ISzAlloc allocTempImp;
+
+  CFileInStream archiveStream;
+  CLookToRead2 lookStream;
+  CSzArEx db;
+  SRes res;
   UInt16 *temp = NULL;
   size_t tempSize = 0;
   // UInt32 parents[NUM_PARENTS_MAX];
 
-  printf("\n7z ANSI-C Decoder " MY_VERSION_COPYRIGHT_DATE "\n\n");
+  Print("\n7z Decoder " MY_VERSION_CPU " : " MY_COPYRIGHT_DATE "\n\n");
 
   if (numargs == 1)
   {
-    printf(
+    Print(
       "Usage: 7zDec <command> <archive_name>\n\n"
       "<Commands>\n"
       "  e: Extract files from archive (without using directory names)\n"
@@ -370,7 +398,7 @@ int MY_CDECL main(int numargs, char *args[])
       "  x: eXtract files with full paths\n");
     return 0;
   }
-  
+
   if (numargs < 3)
   {
     PrintError("incorrect command");
@@ -381,11 +409,9 @@ int MY_CDECL main(int numargs, char *args[])
   g_FileCodePage = AreFileApisANSI() ? CP_ACP : CP_OEMCP;
   #endif
 
-  allocImp.Alloc = SzAlloc;
-  allocImp.Free = SzFree;
 
-  allocTempImp.Alloc = SzAllocTemp;
+  allocImp = g_Alloc;
-  allocTempImp.Free = SzFreeTemp;
+  allocTempImp = g_Alloc;
 
   #ifdef UNDER_CE
   if (InFile_OpenW(&archiveStream.file, L"\test.7z"))
@@ -398,16 +424,31 @@ int MY_CDECL main(int numargs, char *args[])
   }
 
   FileInStream_CreateVTable(&archiveStream);
-  LookToRead_CreateVTable(&lookStream, False);
+  LookToRead2_CreateVTable(&lookStream, False);
-  
+  lookStream.buf = NULL;
-  lookStream.realStream = &archiveStream.s;
-  LookToRead_Init(&lookStream);
 
+  res = SZ_OK;
+
+  {
+    lookStream.buf = ISzAlloc_Alloc(&allocImp, kInputBufSize);
+    if (!lookStream.buf)
+      res = SZ_ERROR_MEM;
+    else
+    {
+      lookStream.bufSize = kInputBufSize;
+      lookStream.realStream = &archiveStream.vt;
+      LookToRead2_Init(&lookStream);
+    }
+  }
+    
   CrcGenerateTable();
-
+    
   SzArEx_Init(&db);
-  
+    
-  res = SzArEx_Open(&db, &lookStream.s, &allocImp, &allocTempImp);
+  if (res == SZ_OK)
+  {
+    res = SzArEx_Open(&db, &lookStream.vt, &allocImp, &allocTempImp);
+  }
   
   if (res == SZ_OK)
   {
@@ -477,7 +518,7 @@ int MY_CDECL main(int numargs, char *args[])
           GetAttribString(SzBitWithVals_Check(&db.Attribs, i) ? db.Attribs.Vals[i] : 0, isDir, attr);
 
           fileSize = SzArEx_GetFileSize(&db, i);
-          UInt64ToStr(fileSize, s);
+          UInt64ToStr(fileSize, s, 10);
           
           if (SzBitWithVals_Check(&db.MTime, i))
             ConvertFileTimeToString(&db.MTime.Vals[i], t);
@@ -489,29 +530,33 @@ int MY_CDECL main(int numargs, char *args[])
             t[j] = '\0';
           }
           
-          printf("%s %s %10s  ", t, attr, s);
+          Print(t);
+          Print(" ");
+          Print(attr);
+          Print(" ");
+          Print(s);
+          Print("  ");
           res = PrintString(temp);
           if (res != SZ_OK)
             break;
           if (isDir)
-            printf("/");
+            Print("/");
-          printf("\n");
+          PrintLF();
           continue;
         }
 
-        fputs(testCommand ?
+        Print(testCommand ?
             "Testing    ":
-            "Extracting ",
+            "Extracting ");
-            stdout);
         res = PrintString(temp);
         if (res != SZ_OK)
           break;
         
         if (isDir)
-          printf("/");
+          Print("/");
         else
         {
-          res = SzArEx_Extract(&db, &lookStream.s, i,
+          res = SzArEx_Extract(&db, &lookStream.vt, i,
               &blockIndex, &outBuffer, &outBufferSize,
               &offset, &outSizeProcessed,
               &allocImp, &allocTempImp);
@@ -543,7 +588,7 @@ int MY_CDECL main(int numargs, char *args[])
           if (isDir)
           {
             MyCreateDir(destPath);
-            printf("\n");
+            PrintLF();
             continue;
           }
           else if (OutFile_OpenUtf16(&outFile, destPath))
@@ -561,6 +606,31 @@ int MY_CDECL main(int numargs, char *args[])
             res = SZ_ERROR_FAIL;
             break;
           }
+
+          #ifdef USE_WINDOWS_FILE
+          {
+            FILETIME mtime, ctime;
+            FILETIME *mtimePtr = NULL;
+            FILETIME *ctimePtr = NULL;
+
+            if (SzBitWithVals_Check(&db.MTime, i))
+            {
+              const CNtfsFileTime *t = &db.MTime.Vals[i];
+              mtime.dwLowDateTime = (DWORD)(t->Low);
+              mtime.dwHighDateTime = (DWORD)(t->High);
+              mtimePtr = &mtime;
+            }
+            if (SzBitWithVals_Check(&db.CTime, i))
+            {
+              const CNtfsFileTime *t = &db.CTime.Vals[i];
+              ctime.dwLowDateTime = (DWORD)(t->Low);
+              ctime.dwHighDateTime = (DWORD)(t->High);
+              ctimePtr = &ctime;
+            }
+            if (mtimePtr || ctimePtr)
+              SetFileTime(outFile.handle, ctimePtr, NULL, mtimePtr);
+          }
+          #endif
           
           if (File_Close(&outFile))
           {
@@ -571,23 +641,31 @@ int MY_CDECL main(int numargs, char *args[])
           
           #ifdef USE_WINDOWS_FILE
           if (SzBitWithVals_Check(&db.Attribs, i))
-            SetFileAttributesW(destPath, db.Attribs.Vals[i]);
+          {
+            UInt32 attrib = db.Attribs.Vals[i];
+            /* p7zip stores posix attributes in high 16 bits and adds 0x8000 as marker.
+               We remove posix bits, if we detect posix mode field */
+            if ((attrib & 0xF0000000) != 0)
+              attrib &= 0x7FFF;
+            SetFileAttributesW(destPath, attrib);
+          }
           #endif
         }
-        printf("\n");
+        PrintLF();
       }
-      IAlloc_Free(&allocImp, outBuffer);
+      ISzAlloc_Free(&allocImp, outBuffer);
     }
   }
 
-  SzArEx_Free(&db, &allocImp);
   SzFree(NULL, temp);
+  SzArEx_Free(&db, &allocImp);
+  ISzAlloc_Free(&allocImp, lookStream.buf);
 
   File_Close(&archiveStream.file);
   
   if (res == SZ_OK)
   {
-    printf("\nEverything is Ok\n");
+    Print("\nEverything is Ok\n");
     return 0;
   }
   
@@ -598,7 +676,11 @@ int MY_CDECL main(int numargs, char *args[])
   else if (res == SZ_ERROR_CRC)
     PrintError("CRC error");
   else
-    printf("\nERROR #%d\n", res);
+  {
+    char s[32];
+    UInt64ToStr(res, s, 0);
+    PrintError(s);
+  }
   
   return 1;
 }

C/Util/7z/makefile.gcc

@@ -73,4 +73,3 @@ Ppmd7Dec.o: ../../Ppmd7Dec.c
 
 clean:
 	-$(RM) $(PROG) $(OBJS)
-

C/Util/7zipInstall/7zipInstall.c

@@ -1,5 +1,5 @@
 /* 7zipInstall.c - 7-Zip Installer
-2015-08-04 : Igor Pavlov : Public domain */
+2017-08-28 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -21,12 +21,26 @@
 #include "../../7zFile.h"
 #include "../../7zVersion.h"
 #include "../../CpuArch.h"
+#include "../../DllSecur.h"
 
 #include "resource.h"
 
-static const WCHAR *k_7zip = L"7-Zip";
 
-static const WCHAR *k_Reg_Software_7zip = L"Software\\7-Zip";
+#define wcscat lstrcatW
+#define wcslen lstrlenW
+#define wcscpy lstrcpyW
+#define wcsncpy lstrcpynW
+
+
+#define kInputBufSize ((size_t)1 << 18)
+
+
+#define _7ZIP_CUR_VER ((MY_VER_MAJOR << 16) | MY_VER_MINOR)
+#define _7ZIP_DLL_VER_COMPAT ((16 << 16) | 3)
+
+static LPCWSTR const k_7zip = L"7-Zip";
+
+static LPCWSTR const k_Reg_Software_7zip = L"Software\\7-Zip";
 
 // #define _64BIT_INSTALLER 1
 
@@ -42,13 +56,13 @@ static const WCHAR *k_Reg_Software_7zip = L"Software\\7-Zip";
   #define k_7zip_with_Ver k_7zip_with_Ver_base
 #endif
 
-static const WCHAR *k_7zip_with_Ver_str = k_7zip_with_Ver;
+static LPCWSTR const k_7zip_with_Ver_str = k_7zip_with_Ver;
 
-static const WCHAR *k_7zip_Setup = k_7zip_with_Ver L" Setup";
+static LPCWSTR const k_7zip_Setup = k_7zip_with_Ver L" Setup";
 
-static const WCHAR *k_Reg_Path = L"Path";
+static LPCWSTR const k_Reg_Path = L"Path";
 
-static const WCHAR *k_Reg_Path32 = L"Path"
+static LPCWSTR const k_Reg_Path32 = L"Path"
   #ifdef _64BIT_INSTALLER
     L"64"
   #else
@@ -70,8 +84,8 @@ static const WCHAR *k_Reg_Path32 = L"Path"
 
 #define k_7zip_CLSID L"{23170F69-40C1-278A-1000-000100020000}"
 
-static const WCHAR *k_Reg_CLSID_7zip = L"CLSID\\" k_7zip_CLSID;
+static LPCWSTR const k_Reg_CLSID_7zip = L"CLSID\\" k_7zip_CLSID;
-static const WCHAR *k_Reg_CLSID_7zip_Inproc = L"CLSID\\" k_7zip_CLSID L"\\InprocServer32";
+static LPCWSTR const k_Reg_CLSID_7zip_Inproc = L"CLSID\\" k_7zip_CLSID L"\\InprocServer32";
 
 #define g_AllUsers True
 
@@ -106,13 +120,41 @@ static void PrintErrorMessage(const char *s)
   MessageBoxW(g_HWND, s2, k_7zip_with_Ver_str, MB_ICONERROR);
 }
 
-static WRes MyCreateDir(const WCHAR *name)
+
+static DWORD GetFileVersion(LPCWSTR s)
+{
+  DWORD size = 0;
+  BYTE *vi = NULL;
+  DWORD version = 0;
+  
+  size = GetFileVersionInfoSizeW(s, NULL);
+  if (size == 0)
+    return 0;
+  
+  vi = malloc(size);
+  if (!vi)
+    return 0;
+  
+  if (GetFileVersionInfoW(s, 0, size, vi))
+  {
+    VS_FIXEDFILEINFO *fi = NULL;
+    UINT fiLen = 0;
+    if (VerQueryValueW(vi, L"\\", (LPVOID *)&fi, &fiLen))
+      version = fi->dwFileVersionMS;
+  }
+  
+  free(vi);
+  return version;
+}
+
+
+static WRes MyCreateDir(LPCWSTR name)
 {
   return CreateDirectoryW(name, NULL) ? 0 : GetLastError();
 }
 
 #define IS_SEPAR(c) (c == WCHAR_PATH_SEPARATOR)
-#define IS_LETTER_CHAR(c) ((c) >= 'a' && (c) <= 'z' || (c) >= 'A' && (c) <= 'Z')
+#define IS_LETTER_CHAR(c) (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z'))
 #define IS_DRIVE_PATH(s) (IS_LETTER_CHAR(s[0]) && s[1] == ':' && IS_SEPAR(s[2]))
 
 static int ReverseFind_PathSepar(const wchar_t *s)
@@ -261,7 +303,7 @@ static LONG MyRegistry_CreateKeyAndVal(HKEY parentKey, LPCWSTR keyName, LPCWSTR
   if (res == ERROR_SUCCESS)
   {
     res = MyRegistry_SetString(destKey, valName, val);
-    res = RegCloseKey(destKey);
+    /* res = */ RegCloseKey(destKey);
   }
   return res;
 }
@@ -284,7 +326,7 @@ static LONG MyRegistry_CreateKeyAndVal_32(HKEY parentKey, LPCWSTR keyName, LPCWS
   if (res == ERROR_SUCCESS)
   {
     res = MyRegistry_SetString(destKey, valName, val);
-    res = RegCloseKey(destKey);
+    /* res = */ RegCloseKey(destKey);
   }
   return res;
 }
@@ -322,7 +364,7 @@ static Bool FindSignature(CSzFile *stream, UInt64 *resPos)
     processed -= k7zStartHeaderSize;
     for (pos = 0; pos <= processed; pos++)
     {
-      for (; buf[pos] != '7' && pos <= processed; pos++);
+      for (; pos <= processed && buf[pos] != '7'; pos++);
       if (pos > processed)
         break;
       if (memcmp(buf + pos, k7zSignature, k7zSignatureSize) == 0)
@@ -454,7 +496,7 @@ static wchar_t MyWCharLower_Ascii(wchar_t c)
   return c;
 }
 
-static const WCHAR *FindSubString(const WCHAR *s1, const char *s2)
+static LPCWSTR FindSubString(LPCWSTR s1, const char *s2)
 {
   for (;;)
   {
@@ -571,6 +613,8 @@ static INT_PTR CALLBACK MyDlgProc(HWND hwnd, UINT message, WPARAM wParam, LPARAM
           #endif
           break;
         }
+        
+        default: return FALSE;
       }
       break;
     
@@ -598,8 +642,8 @@ static LONG SetRegKey_Path2(HKEY parentKey)
   if (res == ERROR_SUCCESS)
   {
     res = MyRegistry_SetString(destKey, k_Reg_Path32, path);
-    res = MyRegistry_SetString(destKey, k_Reg_Path, path);
+    /* res = */ MyRegistry_SetString(destKey, k_Reg_Path, path);
-    res = RegCloseKey(destKey);
+    /* res = */ RegCloseKey(destKey);
   }
   return res;
 }
@@ -699,8 +743,8 @@ static void SetShellProgramsGroup(HWND hwndOwner)
   #endif
 }
 
-static const WCHAR *k_Shell_Approved = L"Software\\Microsoft\\Windows\\CurrentVersion\\Shell Extensions\\Approved";
+static LPCWSTR const k_Shell_Approved = L"Software\\Microsoft\\Windows\\CurrentVersion\\Shell Extensions\\Approved";
-static const WCHAR *k_7zip_ShellExtension = L"7-Zip Shell Extension";
+static LPCWSTR const k_7zip_ShellExtension = L"7-Zip Shell Extension";
 
 static void WriteCLSID()
 {
@@ -718,10 +762,10 @@ static void WriteCLSID()
     WCHAR destPath[MAX_PATH + 10];
     wcscpy(destPath, path);
     wcscat(destPath, L"7-zip32.dll");
-    res = MyRegistry_SetString(destKey, NULL, destPath);
+    /* res = */ MyRegistry_SetString(destKey, NULL, destPath);
-    res = MyRegistry_SetString(destKey, L"ThreadingModel", L"Apartment");
+    /* res = */ MyRegistry_SetString(destKey, L"ThreadingModel", L"Apartment");
     // DeleteRegValue(destKey, L"InprocServer32");
-    res = RegCloseKey(destKey);
+    /* res = */ RegCloseKey(destKey);
   }
 
   #endif
@@ -737,14 +781,14 @@ static void WriteCLSID()
     WCHAR destPath[MAX_PATH + 10];
     wcscpy(destPath, path);
     wcscat(destPath, L"7-zip.dll");
-    res = MyRegistry_SetString(destKey, NULL, destPath);
+    /* res = */ MyRegistry_SetString(destKey, NULL, destPath);
-    res = MyRegistry_SetString(destKey, L"ThreadingModel", L"Apartment");
+    /* res = */ MyRegistry_SetString(destKey, L"ThreadingModel", L"Apartment");
     // DeleteRegValue(destKey, L"InprocServer32");
-    res = RegCloseKey(destKey);
+    /* res = */ RegCloseKey(destKey);
   }
 }
 
-static const WCHAR * const k_ShellEx_Items[] =
+static LPCWSTR const k_ShellEx_Items[] =
 {
     L"*\\shellex\\ContextMenuHandlers"
   , L"Directory\\shellex\\ContextMenuHandlers"
@@ -874,6 +918,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
   UNUSED_VAR(nCmdShow)
 
   #ifndef UNDER_CE
+  LoadSecurityDlls();
   CoInitialize(NULL);
   #endif
 
@@ -912,7 +957,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
             num = s2 - s;
             if (num > MAX_PATH)
               num = MAX_PATH;
-            wcsncpy(path, s, num);
+            wcsncpy(path, s, (unsigned)num);
             RemoveQuotes(path);
           }
         }
@@ -1011,7 +1056,8 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
       BOOL bRet;
       MSG msg;
       
-      while ((bRet = GetMessage(&msg, g_HWND, 0, 0)) != 0)
+      // we need messages for all thread windows (including EDITTEXT window in dialog)
+      while ((bRet = GetMessage(&msg, NULL, 0, 0)) != 0)
       {
         if (bRet == -1)
           return retCode;
@@ -1054,6 +1100,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
   }
 }
 
+
 static Bool GetErrorMessage(DWORD errorCode, WCHAR *message)
 {
   LPVOID msgBuf;
@@ -1068,10 +1115,12 @@ static Bool GetErrorMessage(DWORD errorCode, WCHAR *message)
   return True;
 }
 
+
+
 static int Install()
 {
   CFileInStream archiveStream;
-  CLookToRead lookStream;
+  CLookToRead2 lookStream;
   CSzArEx db;
   
   SRes res = SZ_OK;
@@ -1082,8 +1131,7 @@ static int Install()
   ISzAlloc allocTempImp;
   WCHAR sfxPath[MAX_PATH + 2];
 
-  Bool needReboot = False;
+  int needRebootLevel = 0;
-  size_t pathLen;
 
   allocImp.Alloc = SzAlloc;
   allocImp.Free = SzFree;
@@ -1116,13 +1164,15 @@ static int Install()
 
 if (res == SZ_OK)
 {
+  size_t pathLen;
   if (!g_SilentMode)
   {
     GetDlgItemTextW(g_HWND, IDE_EXTRACT_PATH, path, MAX_PATH);
   }
 
   FileInStream_CreateVTable(&archiveStream);
-  LookToRead_CreateVTable(&lookStream, False);
+  LookToRead2_CreateVTable(&lookStream, False);
+  lookStream.buf = NULL;
  
   {
     // Remove post spaces
@@ -1145,25 +1195,36 @@ if (res == SZ_OK)
   winRes = CreateComplexDir();
 
   if (winRes != 0)
-    res = E_FAIL;
+    res = SZ_ERROR_FAIL;
 
   pathLen = wcslen(path);
+
+  if (res == SZ_OK)
+  {
+    lookStream.buf = ISzAlloc_Alloc(&allocImp, kInputBufSize);
+    if (!lookStream.buf)
+      res = SZ_ERROR_MEM;
+    else
+    {
+      lookStream.bufSize = kInputBufSize;
+      lookStream.realStream = &archiveStream.vt;
+      LookToRead2_Init(&lookStream);
+    }
+  }
+
   SzArEx_Init(&db);
 
   if (res == SZ_OK)
   {
-    lookStream.realStream = &archiveStream.s;
+    res = SzArEx_Open(&db, &lookStream.vt, &allocImp, &allocTempImp);
-    LookToRead_Init(&lookStream);
-    
-    res = SzArEx_Open(&db, &lookStream.s, &allocImp, &allocTempImp);
   }
     
   if (res == SZ_OK)
   {
     UInt32 i;
-    UInt32 blockIndex = 0xFFFFFFFF; /* it can have any value before first call, if(!outBuf) */
+    UInt32 blockIndex = 0xFFFFFFFF; /* it can have any value before first call, if (!outBuf) */
     Byte *outBuf = NULL; /* it must be NULL before first call for each new archive. */
-    size_t outBufSize = 0;  /* it can have any value before first call, if(!outBuf) */
+    size_t outBufSize = 0;  /* it can have any value before first call, if (!outBuf) */
     
     g_TotalSize = 0;
 
@@ -1217,7 +1278,7 @@ if (res == SZ_OK)
         SetWindowTextW(g_InfoLine_HWND, temp);
 
       {
-        res = SzArEx_Extract(&db, &lookStream.s, i,
+        res = SzArEx_Extract(&db, &lookStream.vt, i,
             &blockIndex, &outBuf, &outBufSize,
             &offset, &outSizeProcessed,
             &allocImp, &allocTempImp);
@@ -1231,6 +1292,7 @@ if (res == SZ_OK)
         size_t j;
         // size_t nameStartPos = 0;
         UInt32 tempIndex = 0;
+        int fileLevel = 1 << 2;
         WCHAR origPath[MAX_PATH * 2 + 10];
 
         for (j = 0; temp[j] != 0; j++)
@@ -1284,13 +1346,20 @@ if (res == SZ_OK)
                 break;
             }
 
-            if (tempIndex != 0
+            if (tempIndex != 0)
-                || FindSubString(temp, "7-zip.dll")
+            {
+              tempIndex++;
+              continue;
+            }
+            
+            if (FindSubString(temp, "7-zip.dll")
                 #ifdef _64BIT_INSTALLER
                 || FindSubString(temp, "7-zip32.dll")
                 #endif
                 )
             {
+              DWORD ver = GetFileVersion(path);
+              fileLevel = ((ver < _7ZIP_DLL_VER_COMPAT || ver > _7ZIP_CUR_VER) ? 2 : 1);
               tempIndex++;
               continue;
             }
@@ -1334,7 +1403,7 @@ if (res == SZ_OK)
           */
         }
   
-        // if (res = S_OK)
+        // if (res == SZ_OK)
         {
           processedSize = outSizeProcessed;
           winRes = File_Write(&outFile, outBuf + offset, &processedSize);
@@ -1383,14 +1452,14 @@ if (res == SZ_OK)
             winRes = GetLastError();
             break;
           }
-          needReboot = True;
+          needRebootLevel |= fileLevel;
           #endif
         }
 
       }
     }
 
-    IAlloc_Free(&allocImp, outBuf);
+    ISzAlloc_Free(&allocImp, outBuf);
 
     if (!g_SilentMode)
       SendMessage(g_Progress_HWND, PBM_SETPOS, i, 0);
@@ -1411,6 +1480,8 @@ if (res == SZ_OK)
 
   SzArEx_Free(&db, &allocImp);
 
+  ISzAlloc_Free(&allocImp, lookStream.buf);
+
   File_Close(&archiveStream.file);
 
 }
@@ -1420,7 +1491,7 @@ if (res == SZ_OK)
 
   if (res == SZ_OK)
   {
-    if (!g_SilentMode && needReboot)
+    if (!g_SilentMode && needRebootLevel > 1)
     {
       if (MessageBoxW(g_HWND, L"You must restart your system to complete the installation.\nRestart now?",
           k_7zip_Setup, MB_YESNO | MB_DEFBUTTON2) == IDYES)

C/Util/7zipInstall/7zipInstall.dsp

@@ -40,9 +40,10 @@ RSC=rc.exe
 # PROP Use_Debug_Libraries 0
 # PROP Output_Dir "Release"
 # PROP Intermediate_Dir "Release"
+# PROP Ignore_Export_Lib 0
 # PROP Target_Dir ""
 # ADD BASE CPP /nologo /W3 /GX /O2 /D "WIN32" /D "NDEBUG" /D "_WINDOWS" /D "_MBCS" /YX /FD /c
-# ADD CPP /nologo /MD /W4 /WX /GX /O2 /D "NDEBUG" /D "WIN32" /D "_WINDOWS" /D "_UNICODE2" /D "UNICODE2" /Yu"Precomp.h" /FD /c
+# ADD CPP /nologo /Gr /MD /W4 /WX /GX /O1 /D "NDEBUG" /D "WIN32" /D "_WINDOWS" /D "_UNICODE2" /D "UNICODE2" /Yu"Precomp.h" /FD /c
 # ADD BASE MTL /nologo /D "NDEBUG" /mktyplib203 /win32
 # ADD MTL /nologo /D "NDEBUG" /mktyplib203 /win32
 # ADD BASE RSC /l 0x419 /d "NDEBUG"
@@ -52,7 +53,7 @@ BSC32=bscmake.exe
 # ADD BSC32 /nologo
 LINK32=link.exe
 # ADD BASE LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib /nologo /subsystem:windows /machine:I386
-# ADD LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib /nologo /subsystem:windows /machine:I386
+# ADD LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib version.lib /nologo /subsystem:windows /machine:I386
 
 !ELSEIF  "$(CFG)" == "7zipInstall - Win32 Debug"
 
@@ -78,7 +79,7 @@ BSC32=bscmake.exe
 # ADD BSC32 /nologo
 LINK32=link.exe
 # ADD BASE LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib /nologo /subsystem:windows /debug /machine:I386 /pdbtype:sept
-# ADD LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib /nologo /subsystem:windows /debug /machine:I386 /pdbtype:sept
+# ADD LINK32 kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib version.lib /nologo /subsystem:windows /debug /machine:I386 /pdbtype:sept
 
 !ENDIF 
 
@@ -191,6 +192,14 @@ SOURCE=..\..\Delta.h
 # End Source File
 # Begin Source File
 
+SOURCE=..\..\DllSecur.c
+# End Source File
+# Begin Source File
+
+SOURCE=..\..\DllSecur.h
+# End Source File
+# Begin Source File
+
 SOURCE=..\..\Lzma2Dec.c
 # End Source File
 # Begin Source File

C/Util/7zipInstall/makefile

@@ -1,4 +1,7 @@
 PROG = 7zipInstall.exe
+MY_FIXED = 1
+
+LIBS = $(LIBS) version.lib
 
 !IFDEF _64BIT_INSTALLER
 CFLAGS = $(CFLAGS) -D_64BIT_INSTALLER
@@ -25,6 +28,7 @@ C_OBJS = \
   $O\7zStream.obj \
   $O\Bcj2.obj \
   $O\CpuArch.obj \
+  $O\DllSecur.obj \
   $O\LzmaDec.obj \
 
 OBJS = \

C/Util/7zipInstall/resource.h

@@ -1,6 +1,6 @@
-#define IDD_INSTALL            	100
+#define IDD_INSTALL             100
 
-#define IDT_EXTRACT_EXTRACT_TO	110
+#define IDT_EXTRACT_EXTRACT_TO  110
 #define IDE_EXTRACT_PATH        111
 #define IDB_EXTRACT_SET_PATH    112
 #define IDT_CUR_FILE            113

C/Util/7zipUninstall/7zipUninstall.c

@@ -1,5 +1,5 @@
 /* 7zipUninstall.c - 7-Zip Uninstaller
-2015-08-09 : Igor Pavlov : Public domain */
+2018-03-01 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -20,7 +20,7 @@
 #define LLL_(quote) L##quote
 #define LLL(quote) LLL_(quote)
 
-// static const WCHAR *k_7zip = L"7-Zip";
+// static const WCHAR * const k_7zip = L"7-Zip";
 
 // #define _64BIT_INSTALLER 1
 
@@ -36,13 +36,13 @@
   #define k_7zip_with_Ver k_7zip_with_Ver_base
 #endif
 
-// static const WCHAR *k_7zip_with_Ver_str = k_7zip_with_Ver;
+// static const WCHAR * const k_7zip_with_Ver_str = k_7zip_with_Ver;
 
-static const WCHAR *k_Reg_Software_7zip = L"Software\\7-Zip";
+static const WCHAR * const k_Reg_Software_7zip = L"Software\\7-Zip";
 
-static const WCHAR *k_Reg_Path = L"Path";
+static const WCHAR * const k_Reg_Path = L"Path";
  
-static const WCHAR *k_Reg_Path32 = L"Path"
+static const WCHAR * const k_Reg_Path32 = L"Path"
   #ifdef _64BIT_INSTALLER
     L"64"
   #else
@@ -64,8 +64,8 @@ static const WCHAR *k_Reg_Path32 = L"Path"
 
 #define k_7zip_CLSID L"{23170F69-40C1-278A-1000-000100020000}"
 
-static const WCHAR *k_Reg_CLSID_7zip = L"CLSID\\" k_7zip_CLSID;
+static const WCHAR * const k_Reg_CLSID_7zip = L"CLSID\\" k_7zip_CLSID;
-static const WCHAR *k_Reg_CLSID_7zip_Inproc = L"CLSID\\" k_7zip_CLSID L"\\InprocServer32";
+static const WCHAR * const k_Reg_CLSID_7zip_Inproc = L"CLSID\\" k_7zip_CLSID L"\\InprocServer32";
 
 
 #define g_AllUsers True
@@ -90,7 +90,7 @@ static WCHAR tempPath[MAX_PATH * 2 + 40];
 static WCHAR cmdLine[MAX_PATH * 3 + 40];
 static WCHAR copyPath[MAX_PATH * 2 + 40];
 
-static const WCHAR *kUninstallExe = L"Uninstall.exe";
+static const WCHAR * const kUninstallExe = L"Uninstall.exe";
 
 #define MAKE_CHAR_UPPER(c) ((((c) >= 'a' && (c) <= 'z') ? (c) -= 0x20 : (c)))
 
@@ -340,11 +340,11 @@ static const WCHAR * const k_ShellEx_Items[] =
   , L"Drive\\shellex\\DragDropHandlers"
 };
 
-static const WCHAR *k_Shell_Approved = L"Software\\Microsoft\\Windows\\CurrentVersion\\Shell Extensions\\Approved";
+static const WCHAR * const k_Shell_Approved = L"Software\\Microsoft\\Windows\\CurrentVersion\\Shell Extensions\\Approved";
 
-static const WCHAR *k_AppPaths_7zFm = L"Software\\Microsoft\\Windows\\CurrentVersion\\App Paths\\7zFM.exe";
+static const WCHAR * const k_AppPaths_7zFm = L"Software\\Microsoft\\Windows\\CurrentVersion\\App Paths\\7zFM.exe";
 #define k_REG_Uninstall L"Software\\Microsoft\\Windows\\CurrentVersion\\Uninstall\\"
-static const WCHAR *k_Uninstall_7zip = k_REG_Uninstall L"7-Zip";
+static const WCHAR * const k_Uninstall_7zip = k_REG_Uninstall L"7-Zip";
 
 
 static Bool AreEqual_Path_PrefixName(const wchar_t *s, const wchar_t *prefix, const wchar_t *name)
@@ -362,9 +362,11 @@ static void WriteCLSID()
   {
     if (AreEqual_Path_PrefixName(s, path, L"7-zip.dll"))
     {
-      LONG res = MyRegistry_DeleteKey(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip_Inproc);
+      {
-      if (res == ERROR_SUCCESS)
+        LONG res = MyRegistry_DeleteKey(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip_Inproc);
-        MyRegistry_DeleteKey(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip);
+        if (res == ERROR_SUCCESS)
+          MyRegistry_DeleteKey(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip);
+      }
 
       {
         unsigned i;
@@ -384,7 +386,7 @@ static void WriteCLSID()
         if (res == ERROR_SUCCESS)
         {
           RegDeleteValueW(destKey, k_7zip_CLSID);
-          res = RegCloseKey(destKey);
+          /* res = */ RegCloseKey(destKey);
         }
       }
     }
@@ -397,9 +399,11 @@ static void WriteCLSID()
   {
     if (AreEqual_Path_PrefixName(s, path, L"7-zip32.dll"))
     {
-      LONG res = MyRegistry_DeleteKey_32(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip_Inproc);
+      {
-      if (res == ERROR_SUCCESS)
+        LONG res = MyRegistry_DeleteKey_32(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip_Inproc);
-        MyRegistry_DeleteKey_32(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip);
+        if (res == ERROR_SUCCESS)
+          MyRegistry_DeleteKey_32(HKEY_CLASSES_ROOT, k_Reg_CLSID_7zip);
+      }
 
       {
         unsigned i;
@@ -419,7 +423,7 @@ static void WriteCLSID()
         if (res == ERROR_SUCCESS)
         {
           RegDeleteValueW(destKey, k_7zip_CLSID);
-          res = RegCloseKey(destKey);
+          /* res = */ RegCloseKey(destKey);
         }
       }
     }
@@ -542,7 +546,7 @@ static BOOL RemoveDir()
 #define k_Lang L"Lang"
 
 // NUM_LANG_TXT_FILES files are placed before en.ttt
-#define NUM_LANG_TXT_FILES 86
+#define NUM_LANG_TXT_FILES 88
 
 #ifdef _64BIT_INSTALLER
   #define NUM_EXTRA_FILES_64BIT 1
@@ -552,11 +556,11 @@ static BOOL RemoveDir()
 
 #define NUM_FILES (NUM_LANG_TXT_FILES + 1 + 13 + NUM_EXTRA_FILES_64BIT)
 
-static const char *k_Names =
+static const char * const k_Names =
   "af an ar ast az ba be bg bn br ca co cs cy da de el eo es et eu ext"
-  " fa fi fr fur fy ga gl gu he hi hr hu hy id io is it ja ka kaa kk ko ku ku-ckb ky"
+  " fa fi fr fur fy ga gl gu he hi hr hu hy id io is it ja ka kaa kab kk ko ku ku-ckb ky"
   " lij lt lv mk mn mng mng2 mr ms nb ne nl nn pa-in pl ps pt pt-br ro ru"
-  " sa si sk sl sq sr-spc sr-spl sv ta th tr tt ug uk uz va vi zh-cn zh-tw"
+  " sa si sk sl sq sr-spc sr-spl sv ta th tr tt ug uk uz va vi yo zh-cn zh-tw"
   " en.ttt"
   " descript.ion"
   " History.txt"
@@ -771,6 +775,8 @@ static INT_PTR CALLBACK MyDlgProc(HWND hwnd, UINT message, WPARAM wParam, LPARAM
           OnClose();
           break;
         }
+
+        default: return FALSE;
       }
       break;
     
@@ -1033,7 +1039,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
       BOOL bRet;
       MSG msg;
       
-      while ((bRet = GetMessage(&msg, g_HWND, 0, 0)) != 0)
+      while ((bRet = GetMessage(&msg, NULL, 0, 0)) != 0)
       {
         if (bRet == -1)
           return retCode;

C/Util/7zipUninstall/7zipUninstall.dsp

@@ -43,7 +43,7 @@ RSC=rc.exe
 # PROP Ignore_Export_Lib 0
 # PROP Target_Dir ""
 # ADD BASE CPP /nologo /W3 /GX /O2 /D "WIN32" /D "NDEBUG" /D "_WINDOWS" /D "_MBCS" /YX /FD /c
-# ADD CPP /nologo /MD /W4 /WX /GX /O2 /D "NDEBUG" /D "WIN32" /D "_WINDOWS" /D "_UNICODE2" /D "UNICODE2" /Yu"Precomp.h" /FD /c
+# ADD CPP /nologo /Gr /MD /W4 /WX /GX /O1 /D "NDEBUG" /D "WIN32" /D "_WINDOWS" /D "_UNICODE2" /D "UNICODE2" /FAcs /Yu"Precomp.h" /FD /GF /c
 # ADD BASE MTL /nologo /D "NDEBUG" /mktyplib203 /win32
 # ADD MTL /nologo /D "NDEBUG" /mktyplib203 /win32
 # ADD BASE RSC /l 0x419 /d "NDEBUG"

C/Util/7zipUninstall/makefile

@@ -1,4 +1,5 @@
 PROG = 7zipUninstall.exe
+MY_FIXED = 1
 
 !IFDEF _64BIT_INSTALLER
 CFLAGS = $(CFLAGS) -D_64BIT_INSTALLER

C/Util/7zipUninstall/resource.h

@@ -1,6 +1,6 @@
-#define IDD_INSTALL            	100
+#define IDD_INSTALL             100
 
-#define IDT_EXTRACT_EXTRACT_TO	110
+#define IDT_EXTRACT_EXTRACT_TO  110
 #define IDE_EXTRACT_PATH        111
 
 #define IDT_CUR_FILE            113

C/Util/Lzma/LzmaUtil.c

@@ -1,5 +1,5 @@
 /* LzmaUtil.c -- Test application for LZMA compression
-2015-06-13 : Igor Pavlov : Public domain */
+2017-04-27 : Igor Pavlov : Public domain */
 
 #include "../../Precomp.h"
 
@@ -7,26 +7,29 @@
 #include <stdlib.h>
 #include <string.h>
 
+#include "../../CpuArch.h"
+
 #include "../../Alloc.h"
 #include "../../7zFile.h"
 #include "../../7zVersion.h"
 #include "../../LzmaDec.h"
 #include "../../LzmaEnc.h"
 
-const char *kCantReadMessage = "Can not read input file";
+static const char * const kCantReadMessage = "Can not read input file";
-const char *kCantWriteMessage = "Can not write output file";
+static const char * const kCantWriteMessage = "Can not write output file";
-const char *kCantAllocateMessage = "Can not allocate memory";
+static const char * const kCantAllocateMessage = "Can not allocate memory";
-const char *kDataErrorMessage = "Data error";
+static const char * const kDataErrorMessage = "Data error";
 
-void PrintHelp(char *buffer)
+static void PrintHelp(char *buffer)
 {
-  strcat(buffer, "\nLZMA Utility " MY_VERSION_COPYRIGHT_DATE "\n"
+  strcat(buffer,
-      "\nUsage:  lzma <e|d> inputFile outputFile\n"
+    "\nLZMA-C " MY_VERSION_CPU " : " MY_COPYRIGHT_DATE "\n\n"
-             "  e: encode file\n"
+    "Usage:  lzma <e|d> inputFile outputFile\n"
-             "  d: decode file\n");
+    "  e: encode file\n"
+    "  d: decode file\n");
 }
 
-int PrintError(char *buffer, const char *message)
+static int PrintError(char *buffer, const char *message)
 {
   strcat(buffer, "\nError: ");
   strcat(buffer, message);
@@ -34,20 +37,22 @@ int PrintError(char *buffer, const char *message)
   return 1;
 }
 
-int PrintErrorNumber(char *buffer, SRes val)
+static int PrintErrorNumber(char *buffer, SRes val)
 {
   sprintf(buffer + strlen(buffer), "\nError code: %x\n", (unsigned)val);
   return 1;
 }
 
-int PrintUserError(char *buffer)
+static int PrintUserError(char *buffer)
 {
   return PrintError(buffer, "Incorrect command");
 }
 
+
 #define IN_BUF_SIZE (1 << 16)
 #define OUT_BUF_SIZE (1 << 16)
 
+
 static SRes Decode2(CLzmaDec *state, ISeqOutStream *outStream, ISeqInStream *inStream,
     UInt64 unpackSize)
 {
@@ -101,6 +106,7 @@ static SRes Decode2(CLzmaDec *state, ISeqOutStream *outStream, ISeqInStream *inS
   }
 }
 
+
 static SRes Decode(ISeqOutStream *outStream, ISeqInStream *inStream)
 {
   UInt64 unpackSize;
@@ -133,7 +139,7 @@ static SRes Encode(ISeqOutStream *outStream, ISeqInStream *inStream, UInt64 file
   SRes res;
   CLzmaEncProps props;
 
-  rs = rs;
+  UNUSED_VAR(rs);
 
   enc = LzmaEnc_Create(&g_Alloc);
   if (enc == 0)
@@ -163,7 +169,8 @@ static SRes Encode(ISeqOutStream *outStream, ISeqInStream *inStream, UInt64 file
   return res;
 }
 
-int main2(int numArgs, const char *args[], char *rs)
+
+static int main2(int numArgs, const char *args[], char *rs)
 {
   CFileSeqInStream inStream;
   CFileOutStream outStream;
@@ -215,11 +222,11 @@ int main2(int numArgs, const char *args[], char *rs)
   {
     UInt64 fileSize;
     File_GetLength(&inStream.file, &fileSize);
-    res = Encode(&outStream.s, &inStream.s, fileSize, rs);
+    res = Encode(&outStream.vt, &inStream.vt, fileSize, rs);
   }
   else
   {
-    res = Decode(&outStream.s, useOutFile ? &inStream.s : NULL);
+    res = Decode(&outStream.vt, useOutFile ? &inStream.vt : NULL);
   }
 
   if (useOutFile)
@@ -241,6 +248,7 @@ int main2(int numArgs, const char *args[], char *rs)
   return 0;
 }
 
+
 int MY_CDECL main(int numArgs, const char *args[])
 {
   char rs[800] = { 0 };

C/Util/Lzma/LzmaUtil.dsp

@@ -98,6 +98,10 @@ SOURCE=..\..\7zStream.c
 # End Source File
 # Begin Source File
 
+SOURCE=..\..\7zTypes.h
+# End Source File
+# Begin Source File
+
 SOURCE=..\..\7zVersion.h
 # End Source File
 # Begin Source File
@@ -160,9 +164,5 @@ SOURCE=..\..\Threads.c
 
 SOURCE=..\..\Threads.h
 # End Source File
-# Begin Source File
-
-SOURCE=..\..\Types.h
-# End Source File
 # End Target
 # End Project

C/Util/LzmaLib/LzmaLibExports.c

@@ -1,12 +1,14 @@
 /* LzmaLibExports.c -- LZMA library DLL Entry point
-2008-10-04 : Igor Pavlov : Public domain */
+2015-11-08 : Igor Pavlov : Public domain */
+
+#include "../../Precomp.h"
 
 #include <windows.h>
 
 BOOL WINAPI DllMain(HINSTANCE hInstance, DWORD dwReason, LPVOID lpReserved)
 {
-  hInstance = hInstance;
+  UNUSED_VAR(hInstance);
-  dwReason = dwReason;
+  UNUSED_VAR(dwReason);
-  lpReserved = lpReserved;
+  UNUSED_VAR(lpReserved);
   return TRUE;
 }

C/Util/SfxSetup/SfxSetup.c

@@ -1,5 +1,5 @@
 /* SfxSetup.c - 7z SFX Setup
-2015-03-25 : Igor Pavlov : Public domain */
+2017-04-04 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -20,9 +20,12 @@
 #include "../../7zCrc.h"
 #include "../../7zFile.h"
 #include "../../CpuArch.h"
+#include "../../DllSecur.h"
 
 #define k_EXE_ExtIndex 2
 
+#define kInputBufSize ((size_t)1 << 18)
+
 static const char * const kExts[] =
 {
     "bat"
@@ -88,7 +91,7 @@ static unsigned FindItem(const char * const *items, unsigned num, const wchar_t
 #ifdef _CONSOLE
 static BOOL WINAPI HandlerRoutine(DWORD ctrlType)
 {
-  ctrlType = ctrlType;
+  UNUSED_VAR(ctrlType);
   return TRUE;
 }
 #endif
@@ -144,7 +147,7 @@ static Bool FindSignature(CSzFile *stream, UInt64 *resPos)
     processed -= k7zStartHeaderSize;
     for (pos = 0; pos <= processed; pos++)
     {
-      for (; buf[pos] != '7' && pos <= processed; pos++);
+      for (; pos <= processed && buf[pos] != '7'; pos++);
       if (pos > processed)
         break;
       if (memcmp(buf + pos, k7zSignature, k7zSignatureSize) == 0)
@@ -237,7 +240,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
 #endif
 {
   CFileInStream archiveStream;
-  CLookToRead lookStream;
+  CLookToRead2 lookStream;
   CSzArEx db;
   SRes res = SZ_OK;
   ISzAlloc allocImp;
@@ -254,13 +257,15 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
   Bool useShellExecute = True;
   DWORD exitCode = 0;
 
+  LoadSecurityDlls();
+
   #ifdef _CONSOLE
   SetConsoleCtrlHandler(HandlerRoutine, TRUE);
   #else
-  hInstance = hInstance;
+  UNUSED_VAR(hInstance);
-  hPrevInstance = hPrevInstance;
+  UNUSED_VAR(hPrevInstance);
-  lpCmdLine = lpCmdLine;
+  UNUSED_VAR(lpCmdLine);
-  nCmdShow = nCmdShow;
+  UNUSED_VAR(nCmdShow);
   #endif
 
   CrcGenerateTable();
@@ -272,7 +277,8 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
   allocTempImp.Free = SzFreeTemp;
 
   FileInStream_CreateVTable(&archiveStream);
-  LookToRead_CreateVTable(&lookStream, False);
+  LookToRead2_CreateVTable(&lookStream, False);
+  lookStream.buf = NULL;
  
   winRes = GetModuleFileNameW(NULL, sfxPath, MAX_PATH);
   if (winRes == 0 || winRes > MAX_PATH)
@@ -373,14 +379,22 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
 
   if (res == SZ_OK)
   {
-    lookStream.realStream = &archiveStream.s;
+    lookStream.buf = ISzAlloc_Alloc(&allocImp, kInputBufSize);
-    LookToRead_Init(&lookStream);
+    if (!lookStream.buf)
+      res = SZ_ERROR_MEM;
+    else
+    {
+      lookStream.bufSize = kInputBufSize;
+      lookStream.realStream = &archiveStream.vt;
+      LookToRead2_Init(&lookStream);
+    }
   }
 
   SzArEx_Init(&db);
+  
   if (res == SZ_OK)
   {
-    res = SzArEx_Open(&db, &lookStream.s, &allocImp, &allocTempImp);
+    res = SzArEx_Open(&db, &lookStream.vt, &allocImp, &allocTempImp);
   }
   
   if (res == SZ_OK)
@@ -396,11 +410,9 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
     {
       size_t offset = 0;
       size_t outSizeProcessed = 0;
-      size_t len;
       WCHAR *temp;
-      len = SzArEx_GetFileNameUtf16(&db, i, NULL);
+
-      
+      if (SzArEx_GetFileNameUtf16(&db, i, NULL) >= MAX_PATH)
-      if (len >= MAX_PATH)
       {
         res = SZ_ERROR_FAIL;
         break;
@@ -410,7 +422,7 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
       
       SzArEx_GetFileNameUtf16(&db, i, temp);
       {
-        res = SzArEx_Extract(&db, &lookStream.s, i,
+        res = SzArEx_Extract(&db, &lookStream.vt, i,
           &blockIndex, &outBuffer, &outBufferSize,
           &offset, &outSizeProcessed,
           &allocImp, &allocTempImp);
@@ -521,10 +533,13 @@ int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance,
             temp[j] = CHAR_PATH_SEPARATOR;
       }
     }
-    IAlloc_Free(&allocImp, outBuffer);
+    ISzAlloc_Free(&allocImp, outBuffer);
   }
+
   SzArEx_Free(&db, &allocImp);
 
+  ISzAlloc_Free(&allocImp, lookStream.buf);
+
   File_Close(&archiveStream.file);
 
   if (res == SZ_OK)

C/Util/SfxSetup/SfxSetup.dsp

@@ -187,6 +187,14 @@ SOURCE=..\..\Delta.h
 # End Source File
 # Begin Source File
 
+SOURCE=..\..\DllSecur.c
+# End Source File
+# Begin Source File
+
+SOURCE=..\..\DllSecur.h
+# End Source File
+# Begin Source File
+
 SOURCE=..\..\Lzma2Dec.c
 # End Source File
 # Begin Source File

C/Util/SfxSetup/makefile

@@ -1,4 +1,5 @@
 PROG = 7zS2.sfx
+MY_FIXED = 1
 
 C_OBJS = \
   $O\7zAlloc.obj \
@@ -16,6 +17,7 @@ C_OBJS = \
   $O\BraIA64.obj \
   $O\CpuArch.obj \
   $O\Delta.obj \
+  $O\DllSecur.obj \
   $O\Lzma2Dec.obj \
   $O\LzmaDec.obj \
 

C/Util/SfxSetup/makefile_con

@@ -1,4 +1,5 @@
 PROG = 7zS2con.sfx
+MY_FIXED = 1
 CFLAGS = $(CFLAGS) -D_CONSOLE
 
 C_OBJS = \
@@ -17,6 +18,7 @@ C_OBJS = \
   $O\BraIA64.obj \
   $O\CpuArch.obj \
   $O\Delta.obj \
+  $O\DllSecur.obj \
   $O\Lzma2Dec.obj \
   $O\LzmaDec.obj \
 

C/Xz.c

@@ -1,5 +1,5 @@
 /* Xz.c - Xz
-2015-05-01 : Igor Pavlov : Public domain */
+2017-05-12 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -9,7 +9,7 @@
 #include "XzCrc64.h"
 
 const Byte XZ_SIG[XZ_SIG_SIZE] = { 0xFD, '7', 'z', 'X', 'Z', 0 };
-const Byte XZ_FOOTER_SIG[XZ_FOOTER_SIG_SIZE] = { 'Y', 'Z' };
+/* const Byte XZ_FOOTER_SIG[XZ_FOOTER_SIG_SIZE] = { 'Y', 'Z' }; */
 
 unsigned Xz_WriteVarInt(Byte *buf, UInt64 v)
 {
@@ -20,22 +20,22 @@ unsigned Xz_WriteVarInt(Byte *buf, UInt64 v)
     v >>= 7;
   }
   while (v != 0);
-  buf[i - 1] &= 0x7F;
+  buf[(size_t)i - 1] &= 0x7F;
   return i;
 }
 
 void Xz_Construct(CXzStream *p)
 {
-  p->numBlocks = p->numBlocksAllocated = 0;
+  p->numBlocks = 0;
-  p->blocks = 0;
+  p->blocks = NULL;
   p->flags = 0;
 }
 
-void Xz_Free(CXzStream *p, ISzAlloc *alloc)
+void Xz_Free(CXzStream *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->blocks);
+  ISzAlloc_Free(alloc, p->blocks);
-  p->numBlocks = p->numBlocksAllocated = 0;
+  p->numBlocks = 0;
-  p->blocks = 0;
+  p->blocks = NULL;
 }
 
 unsigned XzFlags_GetCheckSize(CXzStreamFlags f)

C/Xz.h

@@ -1,5 +1,5 @@
 /* Xz.h - Xz interface
-2015-05-01 : Igor Pavlov : Public domain */
+2018-02-28 : Igor Pavlov : Public domain */
 
 #ifndef __XZ_H
 #define __XZ_H
@@ -50,6 +50,7 @@ typedef struct
 #define XzBlock_GetNumFilters(p) (((p)->flags & XZ_BF_NUM_FILTERS_MASK) + 1)
 #define XzBlock_HasPackSize(p)   (((p)->flags & XZ_BF_PACK_SIZE) != 0)
 #define XzBlock_HasUnpackSize(p) (((p)->flags & XZ_BF_UNPACK_SIZE) != 0)
+#define XzBlock_HasUnsupportedFlags(p) (((p)->flags & ~(XZ_BF_NUM_FILTERS_MASK | XZ_BF_PACK_SIZE | XZ_BF_UNPACK_SIZE)) != 0)
 
 SRes XzBlock_Parse(CXzBlock *p, const Byte *header);
 SRes XzBlock_ReadHeader(CXzBlock *p, ISeqInStream *inStream, Bool *isIndex, UInt32 *headerSizeRes);
@@ -60,7 +61,13 @@ SRes XzBlock_ReadHeader(CXzBlock *p, ISeqInStream *inStream, Bool *isIndex, UInt
 #define XZ_FOOTER_SIG_SIZE 2
 
 extern const Byte XZ_SIG[XZ_SIG_SIZE];
+
+/*
 extern const Byte XZ_FOOTER_SIG[XZ_FOOTER_SIG_SIZE];
+*/
+
+#define XZ_FOOTER_SIG_0 'Y'
+#define XZ_FOOTER_SIG_1 'Z'
 
 #define XZ_STREAM_FLAGS_SIZE 2
 #define XZ_STREAM_CRC_SIZE 4
@@ -106,13 +113,12 @@ typedef struct
 {
   CXzStreamFlags flags;
   size_t numBlocks;
-  size_t numBlocksAllocated;
   CXzBlockSizes *blocks;
   UInt64 startOffset;
 } CXzStream;
 
 void Xz_Construct(CXzStream *p);
-void Xz_Free(CXzStream *p, ISzAlloc *alloc);
+void Xz_Free(CXzStream *p, ISzAllocPtr alloc);
 
 #define XZ_SIZE_OVERFLOW ((UInt64)(Int64)-1)
 
@@ -127,12 +133,15 @@ typedef struct
 } CXzs;
 
 void Xzs_Construct(CXzs *p);
-void Xzs_Free(CXzs *p, ISzAlloc *alloc);
+void Xzs_Free(CXzs *p, ISzAllocPtr alloc);
-SRes Xzs_ReadBackward(CXzs *p, ILookInStream *inStream, Int64 *startOffset, ICompressProgress *progress, ISzAlloc *alloc);
+SRes Xzs_ReadBackward(CXzs *p, ILookInStream *inStream, Int64 *startOffset, ICompressProgress *progress, ISzAllocPtr alloc);
 
 UInt64 Xzs_GetNumBlocks(const CXzs *p);
 UInt64 Xzs_GetUnpackSize(const CXzs *p);
 
+
+// ECoderStatus values are identical to ELzmaStatus values of LZMA2 decoder
+
 typedef enum
 {
   CODER_STATUS_NOT_SPECIFIED,               /* use main error code instead */
@@ -141,43 +150,55 @@ typedef enum
   CODER_STATUS_NEEDS_MORE_INPUT             /* you must provide more input bytes */
 } ECoderStatus;
 
+
+// ECoderFinishMode values are identical to ELzmaFinishMode
+
 typedef enum
 {
   CODER_FINISH_ANY,   /* finish at any point */
   CODER_FINISH_END    /* block must be finished at the end */
 } ECoderFinishMode;
 
+
 typedef struct _IStateCoder
 {
   void *p;
-  void (*Free)(void *p, ISzAlloc *alloc);
+  void (*Free)(void *p, ISzAllocPtr alloc);
-  SRes (*SetProps)(void *p, const Byte *props, size_t propSize, ISzAlloc *alloc);
+  SRes (*SetProps)(void *p, const Byte *props, size_t propSize, ISzAllocPtr alloc);
   void (*Init)(void *p);
-  SRes (*Code)(void *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
+  SRes (*Code2)(void *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
-      int srcWasFinished, ECoderFinishMode finishMode, int *wasFinished);
+      int srcWasFinished, ECoderFinishMode finishMode,
+      // int *wasFinished,
+      ECoderStatus *status);
+  SizeT (*Filter)(void *p, Byte *data, SizeT size);
 } IStateCoder;
 
+
+
 #define MIXCODER_NUM_FILTERS_MAX 4
 
 typedef struct
 {
-  ISzAlloc *alloc;
+  ISzAllocPtr alloc;
   Byte *buf;
   unsigned numCoders;
+
+  Byte *outBuf;
+  size_t outBufSize;
+  size_t outWritten; // is equal to lzmaDecoder.dicPos (in outBuf mode)
+  Bool wasFinished;
+  SRes res;
+  ECoderStatus status;
+  // Bool SingleBufMode;
+  
   int finished[MIXCODER_NUM_FILTERS_MAX - 1];
   size_t pos[MIXCODER_NUM_FILTERS_MAX - 1];
   size_t size[MIXCODER_NUM_FILTERS_MAX - 1];
   UInt64 ids[MIXCODER_NUM_FILTERS_MAX];
+  SRes results[MIXCODER_NUM_FILTERS_MAX];
   IStateCoder coders[MIXCODER_NUM_FILTERS_MAX];
 } CMixCoder;
 
-void MixCoder_Construct(CMixCoder *p, ISzAlloc *alloc);
-void MixCoder_Free(CMixCoder *p);
-void MixCoder_Init(CMixCoder *p);
-SRes MixCoder_SetFromMethod(CMixCoder *p, unsigned coderIndex, UInt64 methodId);
-SRes MixCoder_Code(CMixCoder *p, Byte *dest, SizeT *destLen,
-    const Byte *src, SizeT *srcLen, int srcWasFinished,
-    ECoderFinishMode finishMode, ECoderStatus *status);
 
 typedef enum
 {
@@ -191,6 +212,7 @@ typedef enum
   XZ_STATE_BLOCK_FOOTER
 } EXzState;
 
+
 typedef struct
 {
   EXzState state;
@@ -204,7 +226,7 @@ typedef struct
   UInt64 packSize;
   UInt64 unpackSize;
 
-  UInt64 numBlocks;
+  UInt64 numBlocks; // number of finished blocks in current stream
   UInt64 indexSize;
   UInt64 indexPos;
   UInt64 padSize;
@@ -218,14 +240,65 @@ typedef struct
   CXzBlock block;
   CXzCheck check;
   CSha256 sha;
+
+  Bool parseMode;
+  Bool headerParsedOk;
+  Bool decodeToStreamSignature;
+  unsigned decodeOnlyOneBlock;
+
+  Byte *outBuf;
+  size_t outBufSize;
+  size_t outDataWritten; // the size of data in (outBuf) that were fully unpacked
+
   Byte shaDigest[SHA256_DIGEST_SIZE];
   Byte buf[XZ_BLOCK_HEADER_SIZE_MAX];
 } CXzUnpacker;
 
-void XzUnpacker_Construct(CXzUnpacker *p, ISzAlloc *alloc);
+/* alloc : aligned for cache line allocation is better */
+void XzUnpacker_Construct(CXzUnpacker *p, ISzAllocPtr alloc);
 void XzUnpacker_Init(CXzUnpacker *p);
+void XzUnpacker_SetOutBuf(CXzUnpacker *p, Byte *outBuf, size_t outBufSize);
 void XzUnpacker_Free(CXzUnpacker *p);
 
+/*
+  XzUnpacker
+  The sequence for decoding functions:
+  {
+    XzUnpacker_Construct()
+    [Decoding_Calls]
+    XzUnpacker_Free()
+  }
+
+  [Decoding_Calls]
+
+  There are 3 types of interfaces for [Decoding_Calls] calls:
+
+  Interface-1 : Partial output buffers:
+    {
+      XzUnpacker_Init()
+      for()
+        XzUnpacker_Code();
+    }
+    
+  Interface-2 : Direct output buffer:
+    Use it, if you know exact size of decoded data, and you need
+    whole xz unpacked data in one output buffer.
+    xz unpacker doesn't allocate additional buffer for lzma2 dictionary in that mode.
+    {
+      XzUnpacker_Init()
+      XzUnpacker_SetOutBufMode(); // to set output buffer and size
+      for()
+        XzUnpacker_Code(); // (dest = NULL) in XzUnpacker_Code()
+    }
+
+  Interface-3 : Direct output buffer : One call full decoding
+    It unpacks whole input buffer to output buffer in one call.
+    It uses Interface-2 internally.
+    {
+      XzUnpacker_CodeFull()
+    }
+*/
+
 /*
 finishMode:
   It has meaning only if the decoding reaches output limit (*destLen).
@@ -255,20 +328,132 @@ Returns:
 
 
 SRes XzUnpacker_Code(CXzUnpacker *p, Byte *dest, SizeT *destLen,
-    const Byte *src, SizeT *srcLen, ECoderFinishMode finishMode,
+    const Byte *src, SizeT *srcLen, int srcFinished,
-    ECoderStatus *status);
+    ECoderFinishMode finishMode, ECoderStatus *status);
 
-Bool XzUnpacker_IsStreamWasFinished(CXzUnpacker *p);
+SRes XzUnpacker_CodeFull(CXzUnpacker *p, Byte *dest, SizeT *destLen,
+    const Byte *src, SizeT *srcLen,
+    ECoderFinishMode finishMode, ECoderStatus *status);
+
+Bool XzUnpacker_IsStreamWasFinished(const CXzUnpacker *p);
 
 /*
-Call XzUnpacker_GetExtraSize after XzUnpacker_Code function to detect real size of
+XzUnpacker_GetExtraSize() returns then number of uncofirmed bytes,
-xz stream in two cases:
+ if it's in (XZ_STATE_STREAM_HEADER) state or in (XZ_STATE_STREAM_PADDING) state.
-XzUnpacker_Code() returns:
+These bytes can be some bytes after xz archive, or
+it can be start of new xz stream.
+ 
+Call XzUnpacker_GetExtraSize() after XzUnpacker_Code() function to detect real size of
+xz stream in two cases, if XzUnpacker_Code() returns:
   res == SZ_OK && status == CODER_STATUS_NEEDS_MORE_INPUT
   res == SZ_ERROR_NO_ARCHIVE
 */
 
-UInt64 XzUnpacker_GetExtraSize(CXzUnpacker *p);
+UInt64 XzUnpacker_GetExtraSize(const CXzUnpacker *p);
+
+
+/*
+  for random block decoding:
+    XzUnpacker_Init();
+    set CXzUnpacker::streamFlags
+    XzUnpacker_PrepareToRandomBlockDecoding()
+    loop
+    {
+      XzUnpacker_Code()
+      XzUnpacker_IsBlockFinished()
+    }
+*/
+
+void XzUnpacker_PrepareToRandomBlockDecoding(CXzUnpacker *p);
+Bool XzUnpacker_IsBlockFinished(const CXzUnpacker *p);
+
+#define XzUnpacker_GetPackSizeForIndex(p) ((p)->packSize + (p)->blockHeaderSize + XzFlags_GetCheckSize((p)->streamFlags))
+
+
+
+/* ---------- Multi Threading Decoding ---------- */
+
+
+typedef struct
+{
+  size_t inBufSize_ST;
+  size_t outStep_ST;
+  Bool ignoreErrors;
+  
+  #ifndef _7ZIP_ST
+  unsigned numThreads;
+  size_t inBufSize_MT;
+  size_t memUseMax;
+  #endif
+} CXzDecMtProps;
+
+void XzDecMtProps_Init(CXzDecMtProps *p);
+
+
+typedef void * CXzDecMtHandle;
+
+/*
+  alloc    : XzDecMt uses CAlignOffsetAlloc for addresses allocated by (alloc).
+  allocMid : for big allocations, aligned allocation is better
+*/
+
+CXzDecMtHandle XzDecMt_Create(ISzAllocPtr alloc, ISzAllocPtr allocMid);
+void XzDecMt_Destroy(CXzDecMtHandle p);
+
+
+typedef struct
+{
+  Byte UnpackSize_Defined;
+  Byte NumStreams_Defined;
+  Byte NumBlocks_Defined;
+
+  Byte DataAfterEnd;
+  Byte DecodingTruncated; // Decoding was Truncated, we need only partial output data
+
+  UInt64 InSize;  // pack size processed
+  UInt64 OutSize;
+
+  UInt64 NumStreams;
+  UInt64 NumBlocks;
+
+  SRes DecodeRes;
+  SRes ReadRes;
+  SRes ProgressRes;
+  SRes CombinedRes;
+  SRes CombinedRes_Type;
+
+} CXzStatInfo;
+
+void XzStatInfo_Clear(CXzStatInfo *p);
+
+/*
+XzDecMt_Decode()
+SRes:
+  SZ_OK               - OK
+  SZ_ERROR_MEM        - Memory allocation error
+  SZ_ERROR_NO_ARCHIVE - is not xz archive
+  SZ_ERROR_ARCHIVE    - Headers error
+  SZ_ERROR_DATA       - Data Error
+  SZ_ERROR_CRC        - CRC Error
+  SZ_ERROR_INPUT_EOF  - it needs more input data
+  SZ_ERROR_WRITE      - ISeqOutStream error
+  (SZ_ERROR_READ)     - ISeqInStream errors
+  (SZ_ERROR_PROGRESS) - ICompressProgress errors
+  // SZ_ERROR_THREAD     - error in multi-threading functions
+  MY_SRes_HRESULT_FROM_WRes(WRes_error) - error in multi-threading function
+*/
+
+SRes XzDecMt_Decode(CXzDecMtHandle p,
+    const CXzDecMtProps *props,
+    const UInt64 *outDataSize, // NULL means undefined
+    int finishMode,            // 0 - partial unpacking is allowed, 1 - xz stream(s) must be finished
+    ISeqOutStream *outStream,
+    // Byte *outBuf, size_t *outBufSize,
+    ISeqInStream *inStream,
+    // const Byte *inData, size_t inDataSize,
+    CXzStatInfo *stat,
+    int *isMT,                 // 0 means that ST (Single-Thread) version was used
+    ICompressProgress *progress);
 
 EXTERN_C_END
 

C/XzCrc64.c

@@ -1,5 +1,5 @@
 /* XzCrc64.c -- CRC64 calculation
-2015-03-01 : Igor Pavlov : Public domain */
+2017-05-23 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -9,9 +9,9 @@
 #define kCrc64Poly UINT64_CONST(0xC96C5795D7870F42)
 
 #ifdef MY_CPU_LE
-  #define CRC_NUM_TABLES 4
+  #define CRC64_NUM_TABLES 4
 #else
-  #define CRC_NUM_TABLES 5
+  #define CRC64_NUM_TABLES 5
   #define CRC_UINT64_SWAP(v) \
       ((v >> 56) \
     | ((v >> 40) & ((UInt64)0xFF <<  8)) \
@@ -29,10 +29,10 @@
   UInt64 MY_FAST_CALL XzCrc64UpdateT4(UInt64 v, const void *data, size_t size, const UInt64 *table);
 #endif
 
-typedef UInt64 (MY_FAST_CALL *CRC_FUNC)(UInt64 v, const void *data, size_t size, const UInt64 *table);
+typedef UInt64 (MY_FAST_CALL *CRC64_FUNC)(UInt64 v, const void *data, size_t size, const UInt64 *table);
 
-static CRC_FUNC g_Crc64Update;
+static CRC64_FUNC g_Crc64Update;
-UInt64 g_Crc64Table[256 * CRC_NUM_TABLES];
+UInt64 g_Crc64Table[256 * CRC64_NUM_TABLES];
 
 UInt64 MY_FAST_CALL Crc64Update(UInt64 v, const void *data, size_t size)
 {
@@ -52,12 +52,12 @@ void MY_FAST_CALL Crc64GenerateTable()
     UInt64 r = i;
     unsigned j;
     for (j = 0; j < 8; j++)
-      r = (r >> 1) ^ (kCrc64Poly & ~((r & 1) - 1));
+      r = (r >> 1) ^ (kCrc64Poly & ((UInt64)0 - (r & 1)));
     g_Crc64Table[i] = r;
   }
-  for (; i < 256 * CRC_NUM_TABLES; i++)
+  for (i = 256; i < 256 * CRC64_NUM_TABLES; i++)
   {
-    UInt64 r = g_Crc64Table[i - 256];
+    UInt64 r = g_Crc64Table[(size_t)i - 256];
     g_Crc64Table[i] = g_Crc64Table[r & 0xFF] ^ (r >> 8);
   }
   
@@ -74,9 +74,9 @@ void MY_FAST_CALL Crc64GenerateTable()
     else
     #endif
     {
-      for (i = 256 * CRC_NUM_TABLES - 1; i >= 256; i--)
+      for (i = 256 * CRC64_NUM_TABLES - 1; i >= 256; i--)
       {
-        UInt64 x = g_Crc64Table[i - 256];
+        UInt64 x = g_Crc64Table[(size_t)i - 256];
         g_Crc64Table[i] = CRC_UINT64_SWAP(x);
       }
       g_Crc64Update = XzCrc64UpdateT1_BeT4;

C/XzCrc64Opt.c

@@ -1,5 +1,5 @@
 /* XzCrc64Opt.c -- CRC64 calculation
-2015-03-01 : Igor Pavlov : Public domain */
+2017-06-30 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -7,24 +7,24 @@
 
 #ifndef MY_CPU_BE
 
-#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
+#define CRC64_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
 
 UInt64 MY_FAST_CALL XzCrc64UpdateT4(UInt64 v, const void *data, size_t size, const UInt64 *table)
 {
   const Byte *p = (const Byte *)data;
   for (; size > 0 && ((unsigned)(ptrdiff_t)p & 3) != 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2(v, *p);
+    v = CRC64_UPDATE_BYTE_2(v, *p);
   for (; size >= 4; size -= 4, p += 4)
   {
     UInt32 d = (UInt32)v ^ *(const UInt32 *)p;
     v = (v >> 32)
-        ^ table[0x300 + ((d      ) & 0xFF)]
+        ^ (table + 0x300)[((d      ) & 0xFF)]
-        ^ table[0x200 + ((d >>  8) & 0xFF)]
+        ^ (table + 0x200)[((d >>  8) & 0xFF)]
-        ^ table[0x100 + ((d >> 16) & 0xFF)]
+        ^ (table + 0x100)[((d >> 16) & 0xFF)]
-        ^ table[0x000 + ((d >> 24))];
+        ^ (table + 0x000)[((d >> 24))];
   }
   for (; size > 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2(v, *p);
+    v = CRC64_UPDATE_BYTE_2(v, *p);
   return v;
 }
 
@@ -43,7 +43,7 @@ UInt64 MY_FAST_CALL XzCrc64UpdateT4(UInt64 v, const void *data, size_t size, con
     | ((v << 40) & ((UInt64)0xFF << 48)) \
     | ((v << 56)))
 
-#define CRC_UPDATE_BYTE_2_BE(crc, b) (table[(Byte)((crc) >> 56) ^ (b)] ^ ((crc) << 8))
+#define CRC64_UPDATE_BYTE_2_BE(crc, b) (table[(Byte)((crc) >> 56) ^ (b)] ^ ((crc) << 8))
 
 UInt64 MY_FAST_CALL XzCrc64UpdateT1_BeT4(UInt64 v, const void *data, size_t size, const UInt64 *table)
 {
@@ -51,18 +51,18 @@ UInt64 MY_FAST_CALL XzCrc64UpdateT1_BeT4(UInt64 v, const void *data, size_t size
   table += 0x100;
   v = CRC_UINT64_SWAP(v);
   for (; size > 0 && ((unsigned)(ptrdiff_t)p & 3) != 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+    v = CRC64_UPDATE_BYTE_2_BE(v, *p);
   for (; size >= 4; size -= 4, p += 4)
   {
     UInt32 d = (UInt32)(v >> 32) ^ *(const UInt32 *)p;
     v = (v << 32)
-        ^ table[0x000 + ((d      ) & 0xFF)]
+        ^ (table + 0x000)[((d      ) & 0xFF)]
-        ^ table[0x100 + ((d >>  8) & 0xFF)]
+        ^ (table + 0x100)[((d >>  8) & 0xFF)]
-        ^ table[0x200 + ((d >> 16) & 0xFF)]
+        ^ (table + 0x200)[((d >> 16) & 0xFF)]
-        ^ table[0x300 + ((d >> 24))];
+        ^ (table + 0x300)[((d >> 24))];
   }
   for (; size > 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+    v = CRC64_UPDATE_BYTE_2_BE(v, *p);
   return CRC_UINT64_SWAP(v);
 }
 

C/XzEnc.h

@@ -1,5 +1,5 @@
 /* XzEnc.h -- Xz Encode
-2011-02-07 : Igor Pavlov : Public domain */
+2017-06-27 : Igor Pavlov : Public domain */
 
 #ifndef __XZ_ENC_H
 #define __XZ_ENC_H
@@ -10,6 +10,11 @@
 
 EXTERN_C_BEGIN
 
+
+#define XZ_PROPS__BLOCK_SIZE__AUTO   LZMA2_ENC_PROPS__BLOCK_SIZE__AUTO
+#define XZ_PROPS__BLOCK_SIZE__SOLID  LZMA2_ENC_PROPS__BLOCK_SIZE__SOLID
+
+
 typedef struct
 {
   UInt32 id;
@@ -20,15 +25,31 @@ typedef struct
 
 void XzFilterProps_Init(CXzFilterProps *p);
 
+
 typedef struct
 {
-  const CLzma2EncProps *lzma2Props;
+  CLzma2EncProps lzma2Props;
-  const CXzFilterProps *filterProps;
+  CXzFilterProps filterProps;
   unsigned checkId;
+  UInt64 blockSize;
+  int numBlockThreads_Reduced;
+  int numBlockThreads_Max;
+  int numTotalThreads;
+  int forceWriteSizesInHeader;
+  UInt64 reduceSize;
 } CXzProps;
 
 void XzProps_Init(CXzProps *p);
 
+
+typedef void * CXzEncHandle;
+
+CXzEncHandle XzEnc_Create(ISzAllocPtr alloc, ISzAllocPtr allocBig);
+void XzEnc_Destroy(CXzEncHandle p);
+SRes XzEnc_SetProps(CXzEncHandle p, const CXzProps *props);
+void XzEnc_SetDataSize(CXzEncHandle p, UInt64 expectedDataSiize);
+SRes XzEnc_Encode(CXzEncHandle p, ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress);
+
 SRes Xz_Encode(ISeqOutStream *outStream, ISeqInStream *inStream,
     const CXzProps *props, ICompressProgress *progress);
 

C/XzIn.c

@@ -1,5 +1,5 @@
 /* XzIn.c - Xz input
-2015-04-21 : Igor Pavlov : Public domain */
+2018-02-02 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -9,6 +9,12 @@
 #include "CpuArch.h"
 #include "Xz.h"
 
+/*
+#define XZ_FOOTER_SIG_CHECK(p) (memcmp((p), XZ_FOOTER_SIG, XZ_FOOTER_SIG_SIZE) == 0)
+*/
+#define XZ_FOOTER_SIG_CHECK(p) ((p)[0] == XZ_FOOTER_SIG_0 && (p)[1] == XZ_FOOTER_SIG_1)
+
+
 SRes Xz_ReadHeader(CXzStreamFlags *p, ISeqInStream *inStream)
 {
   Byte sig[XZ_STREAM_HEADER_SIZE];
@@ -28,7 +34,7 @@ SRes XzBlock_ReadHeader(CXzBlock *p, ISeqInStream *inStream, Bool *isIndex, UInt
   unsigned headerSize;
   *headerSizeRes = 0;
   RINOK(SeqInStream_ReadByte(inStream, &header[0]));
-  headerSize = ((unsigned)header[0] << 2) + 4;
+  headerSize = (unsigned)header[0];
   if (headerSize == 0)
   {
     *headerSizeRes = 1;
@@ -37,12 +43,13 @@ SRes XzBlock_ReadHeader(CXzBlock *p, ISeqInStream *inStream, Bool *isIndex, UInt
   }
 
   *isIndex = False;
+  headerSize = (headerSize << 2) + 4;
   *headerSizeRes = headerSize;
   RINOK(SeqInStream_Read(inStream, header + 1, headerSize - 1));
   return XzBlock_Parse(p, header);
 }
 
-#define ADD_SIZE_CHECH(size, val) \
+#define ADD_SIZE_CHECK(size, val) \
   { UInt64 newSize = size + (val); if (newSize < size) return XZ_SIZE_OVERFLOW; size = newSize; }
 
 UInt64 Xz_GetUnpackSize(const CXzStream *p)
@@ -50,7 +57,7 @@ UInt64 Xz_GetUnpackSize(const CXzStream *p)
   UInt64 size = 0;
   size_t i;
   for (i = 0; i < p->numBlocks; i++)
-    ADD_SIZE_CHECH(size, p->blocks[i].unpackSize);
+    ADD_SIZE_CHECK(size, p->blocks[i].unpackSize);
   return size;
 }
 
@@ -59,7 +66,7 @@ UInt64 Xz_GetPackSize(const CXzStream *p)
   UInt64 size = 0;
   size_t i;
   for (i = 0; i < p->numBlocks; i++)
-    ADD_SIZE_CHECH(size, (p->blocks[i].totalSize + 3) & ~(UInt64)3);
+    ADD_SIZE_CHECK(size, (p->blocks[i].totalSize + 3) & ~(UInt64)3);
   return size;
 }
 
@@ -70,9 +77,9 @@ SRes XzBlock_ReadFooter(CXzBlock *p, CXzStreamFlags f, ISeqInStream *inStream)
 }
 */
 
-static SRes Xz_ReadIndex2(CXzStream *p, const Byte *buf, size_t size, ISzAlloc *alloc)
+static SRes Xz_ReadIndex2(CXzStream *p, const Byte *buf, size_t size, ISzAllocPtr alloc)
 {
-  size_t i, numBlocks, pos = 1;
+  size_t numBlocks, pos = 1;
   UInt32 crc;
 
   if (size < 5 || buf[0] != 0)
@@ -94,10 +101,10 @@ static SRes Xz_ReadIndex2(CXzStream *p, const Byte *buf, size_t size, ISzAlloc *
   Xz_Free(p, alloc);
   if (numBlocks != 0)
   {
+    size_t i;
     p->numBlocks = numBlocks;
-    p->numBlocksAllocated = numBlocks;
+    p->blocks = (CXzBlockSizes *)ISzAlloc_Alloc(alloc, sizeof(CXzBlockSizes) * numBlocks);
-    p->blocks = alloc->Alloc(alloc, sizeof(CXzBlockSizes) * numBlocks);
+    if (!p->blocks)
-    if (p->blocks == 0)
       return SZ_ERROR_MEM;
     for (i = 0; i < numBlocks; i++)
     {
@@ -114,7 +121,7 @@ static SRes Xz_ReadIndex2(CXzStream *p, const Byte *buf, size_t size, ISzAlloc *
   return (pos == size) ? SZ_OK : SZ_ERROR_ARCHIVE;
 }
 
-static SRes Xz_ReadIndex(CXzStream *p, ILookInStream *stream, UInt64 indexSize, ISzAlloc *alloc)
+static SRes Xz_ReadIndex(CXzStream *p, ILookInStream *stream, UInt64 indexSize, ISzAllocPtr alloc)
 {
   SRes res;
   size_t size;
@@ -124,13 +131,13 @@ static SRes Xz_ReadIndex(CXzStream *p, ILookInStream *stream, UInt64 indexSize,
   size = (size_t)indexSize;
   if (size != indexSize)
     return SZ_ERROR_UNSUPPORTED;
-  buf = alloc->Alloc(alloc, size);
+  buf = (Byte *)ISzAlloc_Alloc(alloc, size);
-  if (buf == 0)
+  if (!buf)
     return SZ_ERROR_MEM;
   res = LookInStream_Read2(stream, buf, size, SZ_ERROR_UNSUPPORTED);
   if (res == SZ_OK)
     res = Xz_ReadIndex2(p, buf, size, alloc);
-  alloc->Free(alloc, buf);
+  ISzAlloc_Free(alloc, buf);
   return res;
 }
 
@@ -141,7 +148,7 @@ static SRes LookInStream_SeekRead_ForArc(ILookInStream *stream, UInt64 offset, v
   /* return LookInStream_Read2(stream, buf, size, SZ_ERROR_NO_ARCHIVE); */
 }
 
-static SRes Xz_ReadBackward(CXzStream *p, ILookInStream *stream, Int64 *startOffset, ISzAlloc *alloc)
+static SRes Xz_ReadBackward(CXzStream *p, ILookInStream *stream, Int64 *startOffset, ISzAllocPtr alloc)
 {
   UInt64 indexSize;
   Byte buf[XZ_STREAM_FOOTER_SIZE];
@@ -153,7 +160,7 @@ static SRes Xz_ReadBackward(CXzStream *p, ILookInStream *stream, Int64 *startOff
   pos -= XZ_STREAM_FOOTER_SIZE;
   RINOK(LookInStream_SeekRead_ForArc(stream, pos, buf, XZ_STREAM_FOOTER_SIZE));
   
-  if (memcmp(buf + 10, XZ_FOOTER_SIG, XZ_FOOTER_SIG_SIZE) != 0)
+  if (!XZ_FOOTER_SIG_CHECK(buf + 10))
   {
     UInt32 total = 0;
     pos += XZ_STREAM_FOOTER_SIZE;
@@ -186,7 +193,7 @@ static SRes Xz_ReadBackward(CXzStream *p, ILookInStream *stream, Int64 *startOff
       return SZ_ERROR_NO_ARCHIVE;
     pos -= XZ_STREAM_FOOTER_SIZE;
     RINOK(LookInStream_SeekRead_ForArc(stream, pos, buf, XZ_STREAM_FOOTER_SIZE));
-    if (memcmp(buf + 10, XZ_FOOTER_SIG, XZ_FOOTER_SIG_SIZE) != 0)
+    if (!XZ_FOOTER_SIG_CHECK(buf + 10))
       return SZ_ERROR_NO_ARCHIVE;
   }
   
@@ -223,7 +230,7 @@ static SRes Xz_ReadBackward(CXzStream *p, ILookInStream *stream, Int64 *startOff
     SecToRead_CreateVTable(&secToRead);
     secToRead.realStream = stream;
 
-    RINOK(Xz_ReadHeader(&headerFlags, &secToRead.s));
+    RINOK(Xz_ReadHeader(&headerFlags, &secToRead.vt));
     return (p->flags == headerFlags) ? SZ_OK : SZ_ERROR_ARCHIVE;
   }
 }
@@ -237,12 +244,12 @@ void Xzs_Construct(CXzs *p)
   p->streams = 0;
 }
 
-void Xzs_Free(CXzs *p, ISzAlloc *alloc)
+void Xzs_Free(CXzs *p, ISzAllocPtr alloc)
 {
   size_t i;
   for (i = 0; i < p->num; i++)
     Xz_Free(&p->streams[i], alloc);
-  alloc->Free(alloc, p->streams);
+  ISzAlloc_Free(alloc, p->streams);
   p->num = p->numAllocated = 0;
   p->streams = 0;
 }
@@ -261,7 +268,7 @@ UInt64 Xzs_GetUnpackSize(const CXzs *p)
   UInt64 size = 0;
   size_t i;
   for (i = 0; i < p->num; i++)
-    ADD_SIZE_CHECH(size, Xz_GetUnpackSize(&p->streams[i]));
+    ADD_SIZE_CHECK(size, Xz_GetUnpackSize(&p->streams[i]));
   return size;
 }
 
@@ -271,15 +278,15 @@ UInt64 Xzs_GetPackSize(const CXzs *p)
   UInt64 size = 0;
   size_t i;
   for (i = 0; i < p->num; i++)
-    ADD_SIZE_CHECH(size, Xz_GetTotalSize(&p->streams[i]));
+    ADD_SIZE_CHECK(size, Xz_GetTotalSize(&p->streams[i]));
   return size;
 }
 */
 
-SRes Xzs_ReadBackward(CXzs *p, ILookInStream *stream, Int64 *startOffset, ICompressProgress *progress, ISzAlloc *alloc)
+SRes Xzs_ReadBackward(CXzs *p, ILookInStream *stream, Int64 *startOffset, ICompressProgress *progress, ISzAllocPtr alloc)
 {
   Int64 endOffset = 0;
-  RINOK(stream->Seek(stream, &endOffset, SZ_SEEK_END));
+  RINOK(ILookInStream_Seek(stream, &endOffset, SZ_SEEK_END));
   *startOffset = endOffset;
   for (;;)
   {
@@ -292,20 +299,20 @@ SRes Xzs_ReadBackward(CXzs *p, ILookInStream *stream, Int64 *startOffset, ICompr
     if (p->num == p->numAllocated)
     {
       size_t newNum = p->num + p->num / 4 + 1;
-      Byte *data = (Byte *)alloc->Alloc(alloc, newNum * sizeof(CXzStream));
+      Byte *data = (Byte *)ISzAlloc_Alloc(alloc, newNum * sizeof(CXzStream));
-      if (data == 0)
+      if (!data)
         return SZ_ERROR_MEM;
       p->numAllocated = newNum;
       if (p->num != 0)
         memcpy(data, p->streams, p->num * sizeof(CXzStream));
-      alloc->Free(alloc, p->streams);
+      ISzAlloc_Free(alloc, p->streams);
       p->streams = (CXzStream *)data;
     }
     p->streams[p->num++] = st;
     if (*startOffset == 0)
       break;
     RINOK(LookInStream_SeekTo(stream, *startOffset));
-    if (progress && progress->Progress(progress, endOffset - *startOffset, (UInt64)(Int64)-1) != SZ_OK)
+    if (progress && ICompressProgress_Progress(progress, endOffset - *startOffset, (UInt64)(Int64)-1) != SZ_OK)
       return SZ_ERROR_PROGRESS;
   }
   return SZ_OK;

CPP/7zip/Aes.mak

@@ -1,7 +1,7 @@
 C_OBJS = $(C_OBJS) \
   $O\Aes.obj
 
-!IF "$(CPU)" != "IA64" && "$(CPU)" != "MIPS" && "$(CPU)" != "ARM"
+!IF "$(CPU)" != "IA64" && "$(CPU)" != "MIPS" && "$(CPU)" != "ARM" && "$(CPU)" != "ARM64"
 ASM_OBJS = $(ASM_OBJS) \
   $O\AesOpt.obj
 !ENDIF

CPP/7zip/Archive/7z/7z.dsp

@@ -158,14 +158,6 @@ SOURCE=.\7zFolderInStream.h
 # End Source File
 # Begin Source File
 
-SOURCE=.\7zFolderOutStream.cpp
-# End Source File
-# Begin Source File
-
-SOURCE=.\7zFolderOutStream.h
-# End Source File
-# Begin Source File
-
 SOURCE=.\7zHandler.cpp
 # End Source File
 # Begin Source File
@@ -350,14 +342,6 @@ SOURCE=..\Common\CoderMixer2.h
 # End Source File
 # Begin Source File
 
-SOURCE=..\Common\CrossThreadProgress.cpp
-# End Source File
-# Begin Source File
-
-SOURCE=..\Common\CrossThreadProgress.h
-# End Source File
-# Begin Source File
-
 SOURCE=..\Common\HandlerOut.cpp
 # End Source File
 # Begin Source File

CPP/7zip/Archive/7z/7zCompressionMode.h

@@ -13,7 +13,9 @@ struct CMethodFull: public CMethodProps
 {
   CMethodId Id;
   UInt32 NumStreams;
+  int CodecIndex;
 
+  CMethodFull(): CodecIndex(-1) {}
   bool IsSimpleCoder() const { return NumStreams == 1; }
 };
 

CPP/7zip/Archive/7z/7zDecode.cpp

@@ -226,19 +226,23 @@ HRESULT CDecoder::Decode(
 
     , ISequentialOutStream *outStream
     , ICompressProgressInfo *compressProgress
+    
     , ISequentialInStream **
+        #ifdef USE_MIXER_ST
+        inStreamMainRes
+        #endif
 
-    #ifdef USE_MIXER_ST
+    , bool &dataAfterEnd_Error
-    inStreamMainRes
-    #endif
     
     _7Z_DECODER_CRYPRO_VARS_DECL
 
-    #if !defined(_7ZIP_ST) && !defined(_SFX)
+    #if !defined(_7ZIP_ST)
-    , bool mtMode, UInt32 numThreads
+    , bool mtMode, UInt32 numThreads, UInt64 memUsage
     #endif
     )
 {
+  dataAfterEnd_Error = false;
+
   const UInt64 *packPositions = &folders.PackPositions[folders.FoStartPackStreamIndex[folderIndex]];
   CFolderEx folderInfo;
   folders.ParseFolderEx(folderIndex, folderInfo);
@@ -308,7 +312,7 @@ HRESULT CDecoder::Decode(
       #endif
   
       CCreatedCoder cod;
-      RINOK(CreateCoder(
+      RINOK(CreateCoder_Id(
           EXTERNAL_CODECS_LOC_VARS
           coderInfo.MethodID, false, cod));
     
@@ -351,11 +355,39 @@ HRESULT CDecoder::Decode(
 
   unsigned i;
 
+  bool mt_wasUsed = false;
+
   for (i = 0; i < folderInfo.Coders.Size(); i++)
   {
     const CCoderInfo &coderInfo = folderInfo.Coders[i];
     IUnknown *decoder = _mixer->GetCoder(i).GetUnknown();
 
+    #if !defined(_7ZIP_ST)
+    if (!mt_wasUsed)
+    {
+      if (mtMode)
+      {
+        CMyComPtr<ICompressSetCoderMt> setCoderMt;
+        decoder->QueryInterface(IID_ICompressSetCoderMt, (void **)&setCoderMt);
+        if (setCoderMt)
+        {
+          mt_wasUsed = true;
+          RINOK(setCoderMt->SetNumberOfThreads(numThreads));
+        }
+      }
+      // if (memUsage != 0)
+      {
+        CMyComPtr<ICompressSetMemLimit> setMemLimit;
+        decoder->QueryInterface(IID_ICompressSetMemLimit, (void **)&setMemLimit);
+        if (setMemLimit)
+        {
+          mt_wasUsed = true;
+          RINOK(setMemLimit->SetMemLimit(memUsage));
+        }
+      }
+    }
+    #endif
+
     {
       CMyComPtr<ICompressSetDecoderProperties2> setDecoderProperties;
       decoder->QueryInterface(IID_ICompressSetDecoderProperties2, (void **)&setDecoderProperties);
@@ -372,18 +404,6 @@ HRESULT CDecoder::Decode(
       }
     }
 
-    #if !defined(_7ZIP_ST) && !defined(_SFX)
-    if (mtMode)
-    {
-      CMyComPtr<ICompressSetCoderMt> setCoderMt;
-      decoder->QueryInterface(IID_ICompressSetCoderMt, (void **)&setCoderMt);
-      if (setCoderMt)
-      {
-        RINOK(setCoderMt->SetNumberOfThreads(numThreads));
-      }
-    }
-    #endif
-
     #ifndef _NO_CRYPTO
     {
       CMyComPtr<ICryptoSetPassword> cryptoSetPassword;
@@ -404,23 +424,25 @@ HRESULT CDecoder::Decode(
           len = password.Len();
         }
         CByteBuffer buffer(len * 2);
-        for (size_t i = 0; i < len; i++)
+        for (size_t k = 0; k < len; k++)
         {
-          wchar_t c = passwordBSTR[i];
+          wchar_t c = passwordBSTR[k];
-          ((Byte *)buffer)[i * 2] = (Byte)c;
+          ((Byte *)buffer)[k * 2] = (Byte)c;
-          ((Byte *)buffer)[i * 2 + 1] = (Byte)(c >> 8);
+          ((Byte *)buffer)[k * 2 + 1] = (Byte)(c >> 8);
         }
         RINOK(cryptoSetPassword->CryptoSetPassword((const Byte *)buffer, (UInt32)buffer.Size()));
       }
     }
     #endif
 
+    bool finishMode = false;
     {
       CMyComPtr<ICompressSetFinishMode> setFinishMode;
       decoder->QueryInterface(IID_ICompressSetFinishMode, (void **)&setFinishMode);
       if (setFinishMode)
       {
-        RINOK(setFinishMode->SetFinishMode(BoolToInt(fullUnpack)));
+        finishMode = fullUnpack;
+        RINOK(setFinishMode->SetFinishMode(BoolToInt(finishMode)));
       }
     }
     
@@ -450,7 +472,7 @@ HRESULT CDecoder::Decode(
             unpackSize :
             &folders.CoderUnpackSizes[unpackStreamIndexStart + i];
     
-    _mixer->SetCoderInfo(i, unpackSizesPointer, packSizesPointers);
+    _mixer->SetCoderInfo(i, unpackSizesPointer, packSizesPointers, finishMode);
   }
 
   if (outStream)
@@ -530,7 +552,9 @@ HRESULT CDecoder::Decode(
       progress2 = new CDecProgress(compressProgress);
 
     ISequentialOutStream *outStreamPointer = outStream;
-    return _mixer->Code(inStreamPointers, &outStreamPointer, progress2 ? (ICompressProgressInfo *)progress2 : compressProgress);
+    return _mixer->Code(inStreamPointers, &outStreamPointer,
+        progress2 ? (ICompressProgressInfo *)progress2 : compressProgress,
+        dataAfterEnd_Error);
   }
   
   #ifdef USE_MIXER_ST

CPP/7zip/Archive/7z/7zDecode.h

@@ -53,12 +53,14 @@ public:
 
       , ISequentialOutStream *outStream
       , ICompressProgressInfo *compressProgress
+
       , ISequentialInStream **inStreamMainRes
+      , bool &dataAfterEnd_Error
       
       _7Z_DECODER_CRYPRO_VARS_DECL
       
-      #if !defined(_7ZIP_ST) && !defined(_SFX)
+      #if !defined(_7ZIP_ST)
-      , bool mtMode, UInt32 numThreads
+      , bool mtMode, UInt32 numThreads, UInt64 memUsage
       #endif
       );
 };

CPP/7zip/Archive/7z/7zEncode.cpp

@@ -148,15 +148,24 @@ HRESULT CEncoder::CreateMixerCoder(
 
   RINOK(_mixer->SetBindInfo(_bindInfo));
 
-  FOR_VECTOR (i, _options.Methods)
+  FOR_VECTOR (m, _options.Methods)
   {
-    const CMethodFull &methodFull = _options.Methods[i];
+    const CMethodFull &methodFull = _options.Methods[m];
 
     CCreatedCoder cod;
     
-    RINOK(CreateCoder(
+    if (methodFull.CodecIndex >= 0)
+    {
+      RINOK(CreateCoder_Index(
+        EXTERNAL_CODECS_LOC_VARS
+        methodFull.CodecIndex, true, cod));
+    }
+    else
+    {
+      RINOK(CreateCoder_Id(
         EXTERNAL_CODECS_LOC_VARS
         methodFull.Id, true, cod));
+    }
 
     if (cod.NumStreams != methodFull.NumStreams)
       return E_FAIL;
@@ -333,7 +342,7 @@ HRESULT CEncoder::Encode(
   }
 
   for (i = 0; i < numMethods; i++)
-    _mixer->SetCoderInfo(i, NULL, NULL);
+    _mixer->SetCoderInfo(i, NULL, NULL, false);
 
 
   /* inStreamSize can be used by BCJ2 to set optimal range of conversion.
@@ -370,6 +379,17 @@ HRESULT CEncoder::Encode(
       resetInitVector->ResetInitVector();
     }
 
+    {
+      CMyComPtr<ICompressSetCoderPropertiesOpt> optProps;
+      coder->QueryInterface(IID_ICompressSetCoderPropertiesOpt, (void **)&optProps);
+      if (optProps)
+      {
+        PROPID propID = NCoderPropID::kExpectedDataSize;
+        NWindows::NCOM::CPropVariant prop = (UInt64)unpackSize;
+        RINOK(optProps->SetCoderPropertiesOpt(&propID, &prop, 1));
+      }
+    }
+
     CMyComPtr<ICompressWriteCoderProperties> writeCoderProperties;
     coder->QueryInterface(IID_ICompressWriteCoderProperties, (void **)&writeCoderProperties);
 
@@ -380,7 +400,7 @@ HRESULT CEncoder::Encode(
       CDynBufSeqOutStream *outStreamSpec = new CDynBufSeqOutStream;
       CMyComPtr<ISequentialOutStream> dynOutStream(outStreamSpec);
       outStreamSpec->Init();
-      writeCoderProperties->WriteCoderProperties(dynOutStream);
+      RINOK(writeCoderProperties->WriteCoderProperties(dynOutStream));
       outStreamSpec->CopyToBuffer(props);
     }
     else
@@ -410,9 +430,9 @@ HRESULT CEncoder::Encode(
     mtOutStreamNotifySpec->_stream = outStream;
     mtOutStreamNotifySpec->_mtProgresSpec = mtProgressSpec;
     
-    FOR_VECTOR(i, tempBufferSpecs)
+    FOR_VECTOR(t, tempBufferSpecs)
     {
-      tempBufferSpecs[i]->_mtProgresSpec = mtProgressSpec;
+      tempBufferSpecs[t]->_mtProgresSpec = mtProgressSpec;
     }
   }
   
@@ -429,10 +449,12 @@ HRESULT CEncoder::Encode(
   for (i = 1; i < _bindInfo.PackStreams.Size(); i++)
     outStreamPointers.Add(tempBuffers[i - 1]);
 
+  bool dataAfterEnd_Error;
+
   RINOK(_mixer->Code(
       &inStreamPointer,
       &outStreamPointers.Front(),
-      mtProgress ? (ICompressProgressInfo *)mtProgress : compressProgress));
+      mtProgress ? (ICompressProgressInfo *)mtProgress : compressProgress, dataAfterEnd_Error));
   
   if (_bindInfo.PackStreams.Size() != 0)
     packSizes.Add(outStreamSizeCountSpec->GetSize());
@@ -591,9 +613,9 @@ HRESULT CEncoder::EncoderConstr()
         continue;
       }
       
-      int i = _bindInfo.FindStream_in_PackStreams(outIndex);
+      int si = _bindInfo.FindStream_in_PackStreams(outIndex);
-      if (i >= 0)
+      if (si >= 0)
-        _bindInfo.PackStreams.MoveToFront(i);
+        _bindInfo.PackStreams.MoveToFront(si);
       break;
     }
   }

CPP/7zip/Archive/7z/7zExtract.cpp

@@ -152,6 +152,12 @@ STDMETHODIMP CFolderOutStream::Write(const void *data, UInt32 size, UInt32 *proc
     if (_fileIsOpen)
     {
       UInt32 cur = (size < _rem ? size : (UInt32)_rem);
+      if (_calcCrc)
+      {
+        const UInt32 k_Step = (UInt32)1 << 20;
+        if (cur > k_Step)
+          cur = k_Step;
+      }
       HRESULT result = S_OK;
       if (_stream)
         result = _stream->Write(data, cur, &cur);
@@ -254,8 +260,16 @@ STDMETHODIMP CHandler::Extract(const UInt32 *indices, UInt32 numItems,
   lps->Init(extractCallback, false);
 
   CDecoder decoder(
-    #ifndef USE_MIXER_ST
+    #if !defined(USE_MIXER_MT)
       false
+    #elif !defined(USE_MIXER_ST)
+      true
+    #elif !defined(__7Z_SET_PROPERTIES)
+      #ifdef _7ZIP_ST
+        false
+      #else
+        true
+      #endif
     #else
       _useMultiThreadMixer
     #endif
@@ -311,13 +325,15 @@ STDMETHODIMP CHandler::Extract(const UInt32 *indices, UInt32 numItems,
         curUnpacked += _db.Files[k].Size;
     }
 
-    HRESULT result = folderOutStream->Init(fileIndex,
+    {
-        allFilesMode ? NULL : indices + i,
+      HRESULT result = folderOutStream->Init(fileIndex,
-        numSolidFiles);
+          allFilesMode ? NULL : indices + i,
+          numSolidFiles);
 
-    i += numSolidFiles;
+      i += numSolidFiles;
 
-    RINOK(result);
+      RINOK(result);
+    }
 
     // to test solid block with zero unpacked size we disable that code
     if (folderOutStream->WasWritingFinished())
@@ -338,6 +354,8 @@ STDMETHODIMP CHandler::Extract(const UInt32 *indices, UInt32 numItems,
       #endif
 
 
+      bool dataAfterEnd_Error = false;
+
       HRESULT result = decoder.Decode(
           EXTERNAL_CODECS_VARS
           _inStream,
@@ -348,20 +366,27 @@ STDMETHODIMP CHandler::Extract(const UInt32 *indices, UInt32 numItems,
           outStream,
           progress,
           NULL // *inStreamMainRes
+          , dataAfterEnd_Error
           
           _7Z_DECODER_CRYPRO_VARS
-          #if !defined(_7ZIP_ST) && !defined(_SFX)
+          #if !defined(_7ZIP_ST)
-            , true, _numThreads
+            , true, _numThreads, _memUsage
           #endif
           );
 
-      if (result == S_FALSE || result == E_NOTIMPL)
+      if (result == S_FALSE || result == E_NOTIMPL || dataAfterEnd_Error)
       {
         bool wasFinished = folderOutStream->WasWritingFinished();
-      
+
-        int resOp = (result == S_FALSE ?
+        int resOp = NExtract::NOperationResult::kDataError;
-            NExtract::NOperationResult::kDataError :
+        
-            NExtract::NOperationResult::kUnsupportedMethod);
+        if (result != S_FALSE)
+        {
+          if (result == E_NOTIMPL)
+            resOp = NExtract::NOperationResult::kUnsupportedMethod;
+          else if (wasFinished && dataAfterEnd_Error)
+            resOp = NExtract::NOperationResult::kDataAfterEnd;
+        }
 
         RINOK(folderOutStream->FlushCorrupted(resOp));
 

CPP/7zip/Archive/7z/7zFolderInStream.cpp

@@ -80,14 +80,17 @@ STDMETHODIMP CFolderInStream::Read(void *data, UInt32 size, UInt32 *processedSiz
   {
     if (_stream)
     {
-      UInt32 processed2;
+      UInt32 cur = size;
-      RINOK(_stream->Read(data, size, &processed2));
+      const UInt32 kMax = (UInt32)1 << 20;
-      if (processed2 != 0)
+      if (cur > kMax)
+        cur = kMax;
+      RINOK(_stream->Read(data, cur, &cur));
+      if (cur != 0)
       {
-        _crc = CrcUpdate(_crc, data, processed2);
+        _crc = CrcUpdate(_crc, data, cur);
-        _pos += processed2;
+        _pos += cur;
         if (processedSize)
-          *processedSize = processed2;
+          *processedSize = cur;
         return S_OK;
       }