Windows: Update LZMA SDK to version 26.01

src/Common/lzma/7zTypes.h

@@ -1,5 +1,5 @@
 /* 7zTypes.h -- Basic types
-2024-01-24 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_7Z_TYPES_H
 #define ZIP7_7Z_TYPES_H
@@ -46,8 +46,9 @@ typedef int SRes;
 
 
 #ifdef _MSC_VER
+  #define MY_ALIGN_IN_STRUCT(n) __declspec(align(n))
   #if _MSC_VER > 1200
-    #define MY_ALIGN(n) __declspec(align(n))
+    #define MY_ALIGN(n) MY_ALIGN_IN_STRUCT(n)
   #else
     #define MY_ALIGN(n)
   #endif
@@ -58,6 +59,7 @@ typedef int SRes;
   #define MY_ALIGN(n) alignas(n)
   */
   #define MY_ALIGN(n) __attribute__ ((aligned(n)))
+  #define MY_ALIGN_IN_STRUCT(n) MY_ALIGN(n)
 #endif
 
 

src/Common/lzma/7zWindows.h

@@ -1,11 +1,17 @@
-/* 7zWindows.h -- StdAfx
+/* 7zWindows.h -- Windows.h and related code
-2023-04-02 : Igor Pavlov : Public domain */
+Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_7Z_WINDOWS_H
 #define ZIP7_INC_7Z_WINDOWS_H
 
 #ifdef _WIN32
 
+#if defined(_MSC_VER) && _MSC_VER >= 1950 && !defined(__clang__) // VS2026
+// <Windows.h> and some another windows files need that option
+// VS2026: wtypesbase.h: warning C4865: 'tagCLSCTX': the underlying type will change from 'int' to 'unsigned int' when '/Zc:enumTypes' is specified on the command line
+#pragma warning(disable : 4865)
+#endif
+
 #if defined(__clang__)
 # pragma clang diagnostic push
 #endif

src/Common/lzma/Alloc.c

@@ -1,5 +1,5 @@
 /* Alloc.c -- Memory allocation functions
-2024-02-18 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -24,8 +24,6 @@
 #endif
 
 // #define SZ_ALLOC_DEBUG
-/* #define SZ_ALLOC_DEBUG */
-
 /* use SZ_ALLOC_DEBUG to debug alloc/free operations */
 #ifdef SZ_ALLOC_DEBUG
 
@@ -34,9 +32,10 @@
 static int g_allocCount = 0;
 #ifdef _WIN32
 static int g_allocCountMid = 0;
+#ifdef Z7_LARGE_PAGES
 static int g_allocCountBig = 0;
 #endif
-
+#endif
 
 #define CONVERT_INT_TO_STR(charType, tempSize) \
   char temp[tempSize]; unsigned i = 0; \
@@ -140,8 +139,10 @@ static void PrintAddr(void *p)
 #else
 
 #ifdef _WIN32
+#ifdef Z7_LARGE_PAGES
 #define PRINT_ALLOC(name, cnt, size, ptr)
 #endif
+#endif
 #define PRINT_FREE(name, cnt, ptr)
 #define Print(s)
 #define PrintLn()
@@ -245,6 +246,7 @@ void MidFree(void *address)
 }
 
 #ifdef Z7_LARGE_PAGES
+// #pragma message("Z7_LARGE_PAGES")
 
 #ifdef MEM_LARGE_PAGES
   #define MY_MEM_LARGE_PAGES  MEM_LARGE_PAGES
@@ -253,32 +255,14 @@ void MidFree(void *address)
 #endif
 
 extern
-SIZE_T g_LargePageSize;
+size_t g_LargePageSize;
-SIZE_T g_LargePageSize = 0;
+size_t g_LargePageSize = 0;
-typedef SIZE_T (WINAPI *Func_GetLargePageMinimum)(VOID);
+extern
-
+size_t g_LargePageThresholdMin;
-void SetLargePageSize(void)
+size_t g_LargePageThresholdMin = 0;
-{
+extern
-  SIZE_T size;
+UInt32 g_LargePageFlags;
-#ifdef Z7_USE_DYN_GetLargePageMinimum
+UInt32 g_LargePageFlags = 0;
-Z7_DIAGNOSTIC_IGNORE_CAST_FUNCTION
-
-  const
-   Func_GetLargePageMinimum fn =
-  (Func_GetLargePageMinimum) Z7_CAST_FUNC_C GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
-       "GetLargePageMinimum");
-  if (!fn)
-    return;
-  size = fn();
-#else
-  size = GetLargePageMinimum();
-#endif
-  if (size == 0 || (size & (size - 1)) != 0)
-    return;
-  g_LargePageSize = size;
-}
-
-#endif // Z7_LARGE_PAGES
 
 void *BigAlloc(size_t size)
 {
@@ -289,12 +273,10 @@ void *BigAlloc(size_t size)
 
   #ifdef Z7_LARGE_PAGES
   {
-    SIZE_T ps = g_LargePageSize;
+    const size_t ps = g_LargePageSize - 1;
-    if (ps != 0 && ps <= (1 << 30) && size > (ps / 2))
+    if (ps < (1u << 30) && size > g_LargePageThresholdMin)
     {
-      size_t size2;
+      const size_t size2 = (size + ps) & ~ps;
-      ps--;
-      size2 = (size + ps) & ~ps;
       if (size2 >= size)
       {
         void *p = VirtualAlloc(NULL, size2, MEM_COMMIT | MY_MEM_LARGE_PAGES, PAGE_READWRITE);
@@ -303,6 +285,8 @@ void *BigAlloc(size_t size)
           PRINT_ALLOC("Alloc-BM ", g_allocCountMid, size2, p)
           return p;
         }
+        if (g_LargePageFlags & Z7_LARGE_PAGES_FLAG_FAIL_STOP)
+          return p;
       }
     }
   }
@@ -317,6 +301,7 @@ void BigFree(void *address)
   MidFree(address);
 }
 
+#endif // Z7_LARGE_PAGES
 #endif // _WIN32
 
 
@@ -327,9 +312,12 @@ const ISzAlloc g_Alloc = { SzAlloc, SzFree };
 #ifdef _WIN32
 static void *SzMidAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p)  return MidAlloc(size); }
 static void SzMidFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p)  MidFree(address); }
+const ISzAlloc g_MidAlloc = { SzMidAlloc, SzMidFree };
+#endif
+
+#if defined(Z7_LARGE_PAGES)
 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_MidAlloc = { SzMidAlloc, SzMidFree };
 const ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree };
 #endif
 
@@ -371,10 +359,16 @@ typedef
 
 #endif
 
-#if !defined(_WIN32) \
+#ifndef _WIN32
-    && (defined(Z7_ALLOC_NO_OFFSET_ALLOCATOR) \
+#include <unistd.h> // for _POSIX_ADVISORY_INFO : for some linux
-        || defined(_POSIX_C_SOURCE) && (_POSIX_C_SOURCE >= 200112L))
+#if (defined(Z7_ALLOC_NO_OFFSET_ALLOCATOR) \
+        || defined(_POSIX_C_SOURCE) && (_POSIX_C_SOURCE >= 200112L) \
+        || defined(_POSIX_ADVISORY_INFO) && (_POSIX_ADVISORY_INFO >= 200112L) \
+        || defined(__APPLE__) \
+        /* || defined(__linux__) */)
   #define USE_posix_memalign
+  // #pragma message("USE_posix_memalign")
+#endif
 #endif
 
 #ifndef USE_posix_memalign
@@ -488,6 +482,181 @@ static void SzAlignedFree(ISzAllocPtr pp, void *address)
 #endif
 }
 
+#ifndef _WIN32
+
+#ifdef Z7_LARGE_PAGES
+
+#if 0 // 1 for debug
+  #include <stdio.h>
+  #include <string.h>  // for strerror()
+  #define PRF(x) x
+#else
+  #define PRF(x)
+#endif
+
+#ifdef USE_posix_memalign
+  /* madvise():
+     glibc <= 2.19 : _BSD_SOURCE
+     glibc  > 2.19 : _DEFAULT_SOURCE
+  */
+  /* && (defined(_DEFAULT_SOURCE) || defined(_BSD_SOURCE)) */
+#if 1 && !defined(Z7_NO_MADVISE) && \
+  (defined(__linux__) || defined(__unix__) || defined(__APPLE__))
+#include <sys/mman.h> // for madvise
+// #pragma message("sys/mman.h")
+#if (defined(MADV_HUGEPAGE) && defined(MADV_NOHUGEPAGE))
+  #define Z7_USE_BIG_ALLOC_MADVISE
+  // #pragma message("Z7_USE_BIG_ALLOC_MADVISE")
+#endif
+#endif
+#endif // USE_posix_memalign
+
+#ifdef Z7_USE_BIG_ALLOC_MADVISE
+#define LARGE_PAGE_SIZE_DEFAULT (1 << 21)
+#else
+#define LARGE_PAGE_SIZE_DEFAULT 0
+#endif
+
+extern
+size_t g_LargePageSize;
+size_t g_LargePageSize = LARGE_PAGE_SIZE_DEFAULT;
+extern
+size_t g_LargePageThresholdMin;
+size_t g_LargePageThresholdMin = LARGE_PAGE_SIZE_DEFAULT / 2;
+extern
+UInt32 g_LargePageFlags;
+UInt32 g_LargePageFlags = 0;
+
+void *BigAlloc(size_t size)
+{
+  if (size == 0)
+    return NULL;
+#ifdef USE_posix_memalign
+  {
+    const size_t pageSize = g_LargePageSize;
+    void *buf = NULL; // on Linux (and other systems), posix_memalign() does not modify memptr on failure (POSIX.1-2008 TC2).
+    PRF(printf("\nBigAlloc 0x%08x=%5uMB", (unsigned)(size), (unsigned)(size >> 20));)
+    if (pageSize && size > g_LargePageThresholdMin)
+    {
+      int res;
+      const size_t mask = pageSize - 1;
+      /* we can allocate aligned size, so data at the end of buffer also will use huge page
+         if (size2 for madvise() is not aligned for huge page size)
+           { Last data block will use small pages. It reduces memory allocation,
+             but last data block with small pages can work slower.
+             It's useful, if we have very large HUGE_PAGE: 32MB or 512MB. }
+      */
+      size_t size2 = (size + mask) & ~mask;
+      if (size2 < size || (size & mask) <= g_LargePageThresholdMin)
+        size2 = size;
+      res = posix_memalign(&buf, pageSize, size2);
+      PRF(printf(" posix_memalign size=0x%08x=%5uMB align=%u",
+          (unsigned)(size2), (unsigned)(size2 >> 20), (unsigned)pageSize);)
+      PRF(printf(" buf=%p", (void *)buf);)
+      if (res == 0)
+      {
+#ifdef Z7_USE_BIG_ALLOC_MADVISE
+        if ((g_LargePageFlags & Z7_LARGE_PAGES_FLAG_NO_MADVISE) == 0)
+        {
+          // Advise the kernel to use huge pages for this memory range
+          // MADV_HUGEPAGE / MADV_NOHUGEPAGE : since Linux 2.6.38
+          // madvise() only operates on whole pages, therefore addr must be page-aligned (4KB/8KB/16KB/64KB).
+          // The value of size is rounded up to a multiple of page size.
+          PRF(printf(" madvise g_LargePageFlags=%x", (unsigned)g_LargePageFlags);)
+          res = madvise(buf, size2, (g_LargePageFlags & Z7_LARGE_PAGES_FLAG_NO_HUGEPAGE) ? MADV_NOHUGEPAGE : MADV_HUGEPAGE);
+          if (res)
+          {
+            PRF(printf("\nERROR res=%d, errno=%d=%s\n", res, (int)errno, strerror(errno));)
+            if (g_LargePageFlags & Z7_LARGE_PAGES_FLAG_FAIL_STOP)
+            {
+              free(buf);
+              return NULL;
+            }
+          }
+        }
+#endif // Z7_USE_BIG_ALLOC_MADVISE
+        PRF(printf("\n");)
+        return buf;
+      }
+      PRF(printf("\nERROR res=%d=%s\n", res, strerror(res));)
+      if (g_LargePageFlags & Z7_LARGE_PAGES_FLAG_FAIL_STOP)
+        return NULL;
+      // (res == ENOMEM) "Out of memory" is possible, if pageSize is too big.
+      // so we do second attempt with smaller alignment
+    }
+  }
+#endif // !USE_posix_memalign
+  PRF(printf(" z7_AlignedAlloc size=0x%08x=%5uMB\n", (unsigned)(size), (unsigned)(size >> 20));)
+  return z7_AlignedAlloc(size);
+}
+
+
+void BigFree(void *address)
+{
+  z7_AlignedFree(address);
+}
+#endif // Z7_LARGE_PAGES
+#endif // !_WIN32
+
+
+#ifdef Z7_LARGE_PAGES
+void z7_LargePage_Set(UInt32 flags, size_t pageSize, size_t threshold)
+{
+  g_LargePageFlags = flags;
+
+#ifdef _WIN32
+  if ((flags & Z7_LARGE_PAGES_FLAG_USE_HUGEPAGE) == 0)
+  {
+    g_LargePageSize = 0;
+    g_LargePageThresholdMin = 0;
+  }
+  else
+  {
+    if ((flags & Z7_LARGE_PAGES_FLAG_DIRECT_PAGE_SIZE) == 0)
+    {
+#ifdef Z7_USE_DYN_GetLargePageMinimum
+      Z7_DIAGNOSTIC_IGNORE_CAST_FUNCTION
+typedef SIZE_T (WINAPI *Func_GetLargePageMinimum)(VOID);
+      const
+        Func_GetLargePageMinimum fn =
+       (Func_GetLargePageMinimum) Z7_CAST_FUNC_C GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
+            "GetLargePageMinimum");
+      if (fn)
+        pageSize = fn();
+      else
+        pageSize = 0;
+#else
+      pageSize = GetLargePageMinimum();
+#endif
+      if (pageSize & (pageSize - 1))
+        pageSize = 0;
+    }
+    g_LargePageSize = pageSize;
+    if ((flags & Z7_LARGE_PAGES_FLAG_DIRECT_THRESHOLD) == 0)
+      threshold = pageSize / 2;
+    g_LargePageThresholdMin = threshold;
+  }
+
+#else // !_WIN32
+
+  if (flags & Z7_LARGE_PAGES_FLAG_NO_PAGECODE)
+  {
+    g_LargePageSize = 0;
+    g_LargePageThresholdMin = 0;
+  }
+  else
+  {
+    if ((flags & Z7_LARGE_PAGES_FLAG_DIRECT_PAGE_SIZE) == 0)
+      pageSize = LARGE_PAGE_SIZE_DEFAULT;
+    g_LargePageSize = pageSize;
+    if ((flags & Z7_LARGE_PAGES_FLAG_DIRECT_THRESHOLD) == 0)
+      threshold = pageSize / 2;
+    g_LargePageThresholdMin = threshold;
+  }
+  // PRF(printf("\ng_LargePageSize=%x g_LargePageThresholdMin = %x g_LargePageFlags = %x", (unsigned)g_LargePageSize, (unsigned)g_LargePageThresholdMin, (unsigned)g_LargePageFlags);)
+#endif // !_WIN32
+}
+#endif // Z7_LARGE_PAGES
 
 const ISzAlloc g_AlignedAlloc = { SzAlignedAlloc, SzAlignedFree };
 

src/Common/lzma/Alloc.h

@@ -1,5 +1,5 @@
 /* Alloc.h -- Memory allocation functions
-2024-01-22 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_ALLOC_H
 #define ZIP7_INC_ALLOC_H
@@ -25,40 +25,40 @@ void *MyRealloc(void *address, size_t size);
 void *z7_AlignedAlloc(size_t size);
 void  z7_AlignedFree(void *p);
 
+extern const ISzAlloc g_Alloc;
+extern const ISzAlloc g_AlignedAlloc;
+
 #ifdef _WIN32
+  void *MidAlloc(size_t size);
+  void MidFree(void *address);
+  extern const ISzAlloc g_MidAlloc;
+#else
+  #define MidAlloc(size)    z7_AlignedAlloc(size)
+  #define MidFree(address)  z7_AlignedFree(address)
+  #define g_MidAlloc g_AlignedAlloc
+#endif
 
 #ifdef Z7_LARGE_PAGES
-void SetLargePageSize(void);
-#endif
 
-void *MidAlloc(size_t size);
+#define Z7_LARGE_PAGES_FLAG_USE_HUGEPAGE  (1 << 0)  //    PAGE_ALIGNED / MADV_HUGEPAGE
-void MidFree(void *address);
+#define Z7_LARGE_PAGES_FLAG_NO_PAGECODE   (1 << 1)  // no PAGE_ALIGNED / no madvise
-void *BigAlloc(size_t size);
+#define Z7_LARGE_PAGES_FLAG_NO_MADVISE    (1 << 2)  //    PAGE_ALIGNED / no madvise : for THP=always
-void BigFree(void *address);
+#define Z7_LARGE_PAGES_FLAG_NO_HUGEPAGE   (1 << 3)  //    PAGE_ALIGNED / MADV_NOHUGEPAGE
+#define Z7_LARGE_PAGES_FLAG_FAIL_STOP     (1 << 15) // for benchmarks
+#define Z7_LARGE_PAGES_FLAG_DIRECT_PAGE_SIZE  (1 << 16)
+#define Z7_LARGE_PAGES_FLAG_DIRECT_THRESHOLD  (1 << 17)
 
-/* #define Z7_BIG_ALLOC_IS_ZERO_FILLED */
+void z7_LargePage_Set(UInt32 flags, size_t pageSize, size_t threshold);
 
+  void *BigAlloc(size_t size);
+  void BigFree(void *address);
+  extern const ISzAlloc g_BigAlloc;
 #else
-
+  #define BigAlloc(size)    MidAlloc(size)
-#define MidAlloc(size)    z7_AlignedAlloc(size)
+  #define BigFree(address)  MidFree(address)
-#define MidFree(address)  z7_AlignedFree(address)
+  #define g_BigAlloc g_MidAlloc
-#define BigAlloc(size)    z7_AlignedAlloc(size)
-#define BigFree(address)  z7_AlignedFree(address)
-
 #endif
 
-extern const ISzAlloc g_Alloc;
-
-#ifdef _WIN32
-extern const ISzAlloc g_BigAlloc;
-extern const ISzAlloc g_MidAlloc;
-#else
-#define g_BigAlloc g_AlignedAlloc
-#define g_MidAlloc g_AlignedAlloc
-#endif
-
-extern const ISzAlloc g_AlignedAlloc;
-
 
 typedef struct
 {

src/Common/lzma/Compiler.h

@@ -1,5 +1,5 @@
 /* Compiler.h : Compiler specific defines and pragmas
-2024-01-22 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_COMPILER_H
 #define ZIP7_INC_COMPILER_H
@@ -54,6 +54,12 @@
 #pragma GCC diagnostic ignored "-Wexcess-padding"
 #endif
 
+#if defined(Z7_APPLE_CLANG_VERSION) && __clang_major__ >= 21
+// warning: function MyAlloc might be an allocator wrapper
+// clang in xcode: clang 21.0.0
+#pragma GCC diagnostic ignored "-Wallocator-wrappers"
+#endif
+
 #if __clang_major__ >= 16
 #pragma GCC diagnostic ignored "-Wunsafe-buffer-usage"
 #endif
@@ -72,7 +78,7 @@
 
 #endif // __clang__
 
-#if defined(_WIN32) && defined(__clang__) && __clang_major__ >= 16
+#if defined(__clang__) && __clang_major__ >= 16
 // #pragma GCC diagnostic ignored "-Wcast-function-type-strict"
 #define Z7_DIAGNOSTIC_IGNORE_CAST_FUNCTION \
   _Pragma("GCC diagnostic ignored \"-Wcast-function-type-strict\"")
@@ -183,6 +189,16 @@ typedef void (*Z7_void_Function)(void);
   #define Z7_ATTRIB_NO_VECTORIZE
 #endif
 
+#if defined(Z7_MSC_VER_ORIGINAL) && (Z7_MSC_VER_ORIGINAL >= 1920)
+  #define Z7_PRAGMA_OPTIMIZE_FOR_CODE_SIZE _Pragma("optimize ( \"s\", on )")
+  #define Z7_PRAGMA_OPTIMIZE_DEFAULT       _Pragma("optimize ( \"\", on )")
+#else
+  #define Z7_PRAGMA_OPTIMIZE_FOR_CODE_SIZE
+  #define Z7_PRAGMA_OPTIMIZE_DEFAULT
+#endif
+
+
+
 #if defined(MY_CPU_X86_OR_AMD64) && ( \
        defined(__clang__) && (__clang_major__ >= 4) \
     || defined(__GNUC__) && (__GNUC__ >= 5))

src/Common/lzma/CpuArch.c

@@ -859,7 +859,7 @@ BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; }
 
 #if defined(__GLIBC__) && (__GLIBC__ * 100 + __GLIBC_MINOR__ >= 216)
   #define Z7_GETAUXV_AVAILABLE
-#else
+#elif !defined(__QNXNTO__)
 // #pragma message("=== is not NEW GLIBC === ")
   #if defined __has_include
   #if __has_include (<sys/auxv.h>)
@@ -877,7 +877,7 @@ BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; }
 
 #ifdef USE_HWCAP
 
-#if defined(__FreeBSD__)
+#if defined(__FreeBSD__) || defined(__OpenBSD__)
 static unsigned long MY_getauxval(int aux)
 {
   unsigned long val;

src/Common/lzma/CpuArch.h

@@ -31,7 +31,12 @@ MY_CPU_64BIT means that processor can work with 64-bit registers.
     #define MY_CPU_NAME "x32"
     #define MY_CPU_SIZEOF_POINTER 4
   #else
-    #define MY_CPU_NAME "x64"
+    #if defined(__APX_EGPR__) || defined(__EGPR__)
+      #define MY_CPU_NAME "x64-apx"
+      #define MY_CPU_AMD64_APX
+    #else
+      #define MY_CPU_NAME "x64"
+    #endif
     #define MY_CPU_SIZEOF_POINTER 8
   #endif
   #define MY_CPU_64BIT
@@ -47,6 +52,12 @@ MY_CPU_64BIT means that processor can work with 64-bit registers.
   #define MY_CPU_SIZEOF_POINTER 4
 #endif
 
+#if defined(__SSE2__) \
+    || defined(MY_CPU_AMD64) \
+    || defined(_M_IX86_FP) && (_M_IX86_FP >= 2)
+#define MY_CPU_SSE2
+#endif
+
 
 #if  defined(_M_ARM64) \
   || defined(_M_ARM64EC) \
@@ -243,11 +254,12 @@ MY_CPU_64BIT means that processor can work with 64-bit registers.
 #endif
 
 
+// _LITTLE_ENDIAN macro can be defined for big-endian platform with some compilers
+
 #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(__LITTLE_ENDIAN__) \
     || defined(__ARMEL__) \
     || defined(__THUMBEL__) \
@@ -571,10 +583,12 @@ problem-4 : performace:
 #define Z7_CONV_BE_TO_NATIVE_CONST32(v)  (v)
 #define Z7_CONV_LE_TO_NATIVE_CONST32(v)  Z7_BSWAP32_CONST(v)
 #define Z7_CONV_NATIVE_TO_BE_32(v)       (v)
+// #define Z7_GET_NATIVE16_FROM_2_BYTES(b0, b1)  ((b1) | ((b0) << 8))
 #elif defined(MY_CPU_LE)
 #define Z7_CONV_BE_TO_NATIVE_CONST32(v)  Z7_BSWAP32_CONST(v)
 #define Z7_CONV_LE_TO_NATIVE_CONST32(v)  (v)
 #define Z7_CONV_NATIVE_TO_BE_32(v)       Z7_BSWAP32(v)
+// #define Z7_GET_NATIVE16_FROM_2_BYTES(b0, b1)  ((b0) | ((b1) << 8))
 #else
 #error Stop_Compiling_Unknown_Endian_CONV
 #endif
@@ -588,8 +602,20 @@ problem-4 : performace:
 #define SetBe32a(p, v)   { *(UInt32 *)(void *)(p) = (v); }
 #define SetBe16a(p, v)   { *(UInt16 *)(void *)(p) = (v); }
 
+// gcc and clang for powerpc can transform load byte access to load reverse word access.
+// sp we can use byte access instead of word access. Z7_BSWAP64 cab be slow
+#if 1 && defined(Z7_CPU_FAST_BSWAP_SUPPORTED) && defined(MY_CPU_64BIT)
+#define GetUi64a(p)   Z7_BSWAP64 (*(const UInt64 *)(const void *)(p))
+#else
 #define GetUi64a(p)      GetUi64(p)
+#endif
+
+#if 1 && defined(Z7_CPU_FAST_BSWAP_SUPPORTED)
+#define GetUi32a(p)   Z7_BSWAP32 (*(const UInt32 *)(const void *)(p))
+#else
 #define GetUi32a(p)      GetUi32(p)
+#endif
+
 #define GetUi16a(p)      GetUi16(p)
 #define SetUi32a(p, v)   SetUi32(p, v)
 #define SetUi16a(p, v)   SetUi16(p, v)

src/Common/lzma/LzFind.c

@@ -1,5 +1,5 @@
 /* LzFind.c -- Match finder for LZ algorithms
-2024-03-01 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -404,7 +404,7 @@ int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
         const unsigned nbMax =
             (p->numHashBytes == 2 ? 16 :
             (p->numHashBytes == 3 ? 24 : 32));
-        if (numBits > nbMax)
+        if (numBits >= nbMax)
           numBits = nbMax;
         if (numBits >= 32)
           hs = (UInt32)0 - 1;
@@ -416,14 +416,14 @@ int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
           hs |= (256 << kLzHash_CrcShift_2) - 1;
         {
           const UInt32 hs2 = MatchFinder_GetHashMask2(p, historySize);
-          if (hs > hs2)
+          if (hs >= hs2)
             hs = hs2;
         }
         hsCur = hs;
         if (p->expectedDataSize < historySize)
         {
           const UInt32 hs2 = MatchFinder_GetHashMask2(p, (UInt32)p->expectedDataSize);
-          if (hsCur > hs2)
+          if (hsCur >= hs2)
             hsCur = hs2;
         }
       }
@@ -434,7 +434,7 @@ int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
         if (p->expectedDataSize < historySize)
         {
           hsCur = MatchFinder_GetHashMask(p, (UInt32)p->expectedDataSize);
-          if (hsCur > hs) // is it possible?
+          if (hsCur >= hs) // is it possible?
             hsCur = hs;
         }
       }
@@ -598,7 +598,7 @@ void MatchFinder_Init(void *_p)
 
 #ifdef MY_CPU_X86_OR_AMD64
   #if defined(__clang__) && (__clang_major__ >= 4) \
-    || defined(Z7_GCC_VERSION) && (Z7_GCC_VERSION >= 40701)
+    || defined(Z7_GCC_VERSION) && (Z7_GCC_VERSION >= 40900)
     // || defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1900)
 
       #define USE_LZFIND_SATUR_SUB_128
@@ -890,7 +890,7 @@ static UInt32 * Hc_GetMatchesSpec(size_t lenLimit, UInt32 curMatch, UInt32 pos,
       return d;
     {
       const Byte *pb = cur - delta;
-      curMatch = son[_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)];
+      curMatch = son[_cyclicBufferPos - delta + (_cyclicBufferPos < delta ? _cyclicBufferSize : 0)];
       if (pb[maxLen] == cur[maxLen] && *pb == *cur)
       {
         UInt32 len = 0;
@@ -925,7 +925,7 @@ static UInt32 * Hc_GetMatchesSpec(size_t lenLimit, UInt32 curMatch, UInt32 pos,
       break;
     {
       ptrdiff_t diff;
-      curMatch = son[_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)];
+      curMatch = son[_cyclicBufferPos - delta + (_cyclicBufferPos < delta ? _cyclicBufferSize : 0)];
       diff = (ptrdiff_t)0 - (ptrdiff_t)delta;
       if (cur[maxLen] == cur[(ptrdiff_t)maxLen + diff])
       {
@@ -972,7 +972,7 @@ UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byt
   // if (curMatch >= pos) { *ptr0 = *ptr1 = kEmptyHashValue; return NULL; }
 
   cmCheck = (UInt32)(pos - _cyclicBufferSize);
-  if ((UInt32)pos <= _cyclicBufferSize)
+  if ((UInt32)pos < _cyclicBufferSize)
     cmCheck = 0;
 
   if (cmCheck < curMatch)
@@ -980,7 +980,7 @@ UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byt
   {
     const UInt32 delta = pos - curMatch;
     {
-      CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)) << 1);
+      CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + (_cyclicBufferPos < delta ? _cyclicBufferSize : 0)) << 1);
       const Byte *pb = cur - delta;
       unsigned len = (len0 < len1 ? len0 : len1);
       const UInt32 pair0 = pair[0];
@@ -1039,7 +1039,7 @@ static void SkipMatchesSpec(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const
   UInt32 cmCheck;
 
   cmCheck = (UInt32)(pos - _cyclicBufferSize);
-  if ((UInt32)pos <= _cyclicBufferSize)
+  if ((UInt32)pos < _cyclicBufferSize)
     cmCheck = 0;
 
   if (// curMatch >= pos ||  // failure
@@ -1048,7 +1048,7 @@ static void SkipMatchesSpec(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const
   {
     const UInt32 delta = pos - curMatch;
     {
-      CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)) << 1);
+      CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + (_cyclicBufferPos < delta ? _cyclicBufferSize : 0)) << 1);
       const Byte *pb = cur - delta;
       unsigned len = (len0 < len1 ? len0 : len1);
       if (pb[len] == cur[len])
@@ -1595,7 +1595,7 @@ static void Bt5_MatchFinder_Skip(void *_p, UInt32 num)
     UInt32 pos = p->pos; \
     UInt32 num2 = num; \
     /* (p->pos == p->posLimit) is not allowed here !!! */ \
-    { const UInt32 rem = p->posLimit - pos; if (num2 > rem) num2 = rem; } \
+    { const UInt32 rem = p->posLimit - pos; if (num2 >= rem) num2 = rem; } \
     num -= num2; \
     { const UInt32 cycPos = p->cyclicBufferPos; \
       son = p->son + cycPos; \

src/Common/lzma/LzFindMt.c

@@ -1,5 +1,5 @@
 /* LzFindMt.c -- multithreaded Match finder for LZ algorithms
-2024-01-22 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -82,6 +82,8 @@ extern UInt64 g_NumIters_Bytes;
 Z7_NO_INLINE
 static void MtSync_Construct(CMtSync *p)
 {
+  p->affinityGroup = -1;
+  p->affinityInGroup = 0;
   p->affinity = 0;
   p->wasCreated = False;
   p->csWasInitialized = False;
@@ -259,6 +261,12 @@ static WRes MtSync_Create_WRes(CMtSync *p, THREAD_FUNC_TYPE startAddress, void *
   // return ERROR_TOO_MANY_POSTS; // for debug
   // return EINVAL; // for debug
 
+#ifdef _WIN32
+  if (p->affinityGroup >= 0)
+    wres = Thread_Create_With_Group(&p->thread, startAddress, obj,
+        (unsigned)(UInt32)p->affinityGroup, (CAffinityMask)p->affinityInGroup);
+  else
+#endif
   if (p->affinity != 0)
     wres = Thread_Create_With_Affinity(&p->thread, startAddress, obj, (CAffinityMask)p->affinity);
   else

src/Common/lzma/LzFindMt.h

@@ -1,5 +1,5 @@
 /* LzFindMt.h -- multithreaded Match finder for LZ algorithms
-2024-01-22 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_LZ_FIND_MT_H
 #define ZIP7_INC_LZ_FIND_MT_H
@@ -12,8 +12,10 @@ EXTERN_C_BEGIN
 typedef struct
 {
   UInt32 numProcessedBlocks;
-  CThread thread;
+  Int32 affinityGroup;
+  UInt64 affinityInGroup;
   UInt64 affinity;
+  CThread thread;
 
   BoolInt wasCreated;
   BoolInt needStart;

src/Common/lzma/LzmaEnc.c

@@ -62,7 +62,9 @@ void LzmaEncProps_Init(CLzmaEncProps *p)
   p->lc = p->lp = p->pb = p->algo = p->fb = p->btMode = p->numHashBytes = p->numThreads = -1;
   p->numHashOutBits = 0;
   p->writeEndMark = 0;
+  p->affinityGroup = -1;
   p->affinity = 0;
+  p->affinityInGroup = 0;
 }
 
 void LzmaEncProps_Normalize(CLzmaEncProps *p)
@@ -598,6 +600,10 @@ SRes LzmaEnc_SetProps(CLzmaEncHandle p, const CLzmaEncProps *props2)
   p->multiThread = (props.numThreads > 1);
   p->matchFinderMt.btSync.affinity =
   p->matchFinderMt.hashSync.affinity = props.affinity;
+  p->matchFinderMt.btSync.affinityGroup =
+  p->matchFinderMt.hashSync.affinityGroup = props.affinityGroup;
+  p->matchFinderMt.btSync.affinityInGroup =
+  p->matchFinderMt.hashSync.affinityInGroup = props.affinityInGroup;
   #endif
 
   return SZ_OK;
@@ -2345,10 +2351,9 @@ static void LzmaEnc_Construct(CLzmaEnc *p)
 
 CLzmaEncHandle LzmaEnc_Create(ISzAllocPtr alloc)
 {
-  void *p;
+  CLzmaEncHandle p = (CLzmaEncHandle)ISzAlloc_Alloc(alloc, sizeof(CLzmaEnc));
-  p = ISzAlloc_Alloc(alloc, sizeof(CLzmaEnc));
   if (p)
-    LzmaEnc_Construct((CLzmaEnc *)p);
+    LzmaEnc_Construct(p);
   return p;
 }
 

src/Common/lzma/LzmaEnc.h

@@ -1,5 +1,5 @@
 /*  LzmaEnc.h -- LZMA Encoder
-2023-04-13 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_LZMA_ENC_H
 #define ZIP7_INC_LZMA_ENC_H
@@ -29,11 +29,13 @@ typedef struct
   int numThreads;  /* 1 or 2, default = 2 */
 
   // int _pad;
+  Int32 affinityGroup;
 
   UInt64 reduceSize; /* estimated size of data that will be compressed. default = (UInt64)(Int64)-1.
                         Encoder uses this value to reduce dictionary size */
 
   UInt64 affinity;
+  UInt64 affinityInGroup;
 } CLzmaEncProps;
 
 void LzmaEncProps_Init(CLzmaEncProps *p);

src/Common/lzma/Precomp.h

@@ -1,5 +1,5 @@
 /* Precomp.h -- precompilation file
-2024-01-25 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_PRECOMP_H
 #define ZIP7_INC_PRECOMP_H
@@ -40,18 +40,18 @@
 #endif
 */
 
+#ifndef Z7_LARGE_PAGES
+#if !defined(Z7_NO_LARGE_PAGES) && !defined(UNDER_CE)
+#define Z7_LARGE_PAGES 1
+#endif
+#endif
+
 #ifdef _WIN32
 /*
   this "Precomp.h" file must be included before <windows.h>,
   if we want to define _WIN32_WINNT before <windows.h>.
 */
 
-#ifndef Z7_LARGE_PAGES
-#ifndef Z7_NO_LARGE_PAGES
-#define Z7_LARGE_PAGES 1
-#endif
-#endif
-
 #ifndef Z7_LONG_PATH
 #ifndef Z7_NO_LONG_PATH
 #define Z7_LONG_PATH 1

src/Common/lzma/Threads.c

@@ -1,5 +1,5 @@
 /* Threads.c -- multithreading library
-2024-03-28 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -59,6 +59,111 @@ WRes Thread_Wait_Close(CThread *p)
   return (res != 0 ? res : res2);
 }
 
+typedef struct MY_PROCESSOR_NUMBER {
+    WORD  Group;
+    BYTE  Number;
+    BYTE  Reserved;
+} MY_PROCESSOR_NUMBER, *MY_PPROCESSOR_NUMBER;
+
+typedef struct MY_GROUP_AFFINITY {
+#if defined(Z7_GCC_VERSION) && (Z7_GCC_VERSION < 100000)
+    // KAFFINITY is not defined in old mingw
+    ULONG_PTR
+#else
+    KAFFINITY
+#endif
+      Mask;
+    WORD   Group;
+    WORD   Reserved[3];
+} MY_GROUP_AFFINITY, *MY_PGROUP_AFFINITY;
+
+typedef BOOL (WINAPI *Func_SetThreadGroupAffinity)(
+    HANDLE hThread,
+    CONST MY_GROUP_AFFINITY *GroupAffinity,
+    MY_PGROUP_AFFINITY PreviousGroupAffinity);
+
+typedef BOOL (WINAPI *Func_GetThreadGroupAffinity)(
+    HANDLE hThread,
+    MY_PGROUP_AFFINITY GroupAffinity);
+
+typedef BOOL (WINAPI *Func_GetProcessGroupAffinity)(
+    HANDLE hProcess,
+    PUSHORT GroupCount,
+    PUSHORT GroupArray);
+
+Z7_DIAGNOSTIC_IGNORE_CAST_FUNCTION
+
+#if 0
+#include <stdio.h>
+#define PRF(x) x
+/*
+--
+  before call of SetThreadGroupAffinity()
+    GetProcessGroupAffinity return one group.
+  after call of SetThreadGroupAffinity():
+    GetProcessGroupAffinity return more than group,
+    if SetThreadGroupAffinity() was to another group.
+--
+  GetProcessAffinityMask MS DOCs:
+  {
+    If the calling process contains threads in multiple groups,
+    the function returns zero for both affinity masks.
+  }
+  but tests in win10 with 2 groups (less than 64 cores total):
+    GetProcessAffinityMask() still returns non-zero affinity masks
+    even after SetThreadGroupAffinity() calls.
+*/
+static void PrintProcess_Info()
+{
+  {
+    const
+      Func_GetProcessGroupAffinity fn_GetProcessGroupAffinity =
+     (Func_GetProcessGroupAffinity) Z7_CAST_FUNC_C GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
+          "GetProcessGroupAffinity");
+    if (fn_GetProcessGroupAffinity)
+    {
+      unsigned i;
+      USHORT GroupCounts[64];
+      USHORT GroupCount = Z7_ARRAY_SIZE(GroupCounts);
+      BOOL boolRes = fn_GetProcessGroupAffinity(GetCurrentProcess(),
+          &GroupCount, GroupCounts);
+      printf("\n====== GetProcessGroupAffinity : "
+          "boolRes=%u GroupCounts = %u :",
+          boolRes, (unsigned)GroupCount);
+      for (i = 0; i < GroupCount; i++)
+        printf(" %u", GroupCounts[i]);
+      printf("\n");
+    }
+  }
+  {
+    DWORD_PTR processAffinityMask, systemAffinityMask;
+    if (GetProcessAffinityMask(GetCurrentProcess(), &processAffinityMask, &systemAffinityMask))
+    {
+      PRF(printf("\n====== GetProcessAffinityMask : "
+        ": processAffinityMask=%x, systemAffinityMask=%x\n",
+        (UInt32)processAffinityMask, (UInt32)systemAffinityMask);)
+    }
+    else
+      printf("\n==GetProcessAffinityMask FAIL");
+  }
+}
+#else
+#ifndef USE_THREADS_CreateThread
+// #define PRF(x)
+#endif
+#endif
+
+/* if we send (stackSize=0) to CreateThread(), it will
+   use default value PE::SizeOfStackReserve from exe file.
+   PE::SizeOfStackReserve == 1 MiB in exe file with default linker options.
+   Windows aligns specified value to the next 64 KB range. */
+static const unsigned k_StackSize_ReserveSize =
+  #ifdef UNDER_CE
+    1 << 17;
+  #else
+    1 << 20;
+  #endif
+
 WRes Thread_Create(CThread *p, THREAD_FUNC_TYPE func, LPVOID param)
 {
   /* Windows Me/98/95: threadId parameter may not be NULL in _beginthreadex/CreateThread functions */
@@ -66,12 +171,51 @@ WRes Thread_Create(CThread *p, THREAD_FUNC_TYPE func, LPVOID param)
   #ifdef USE_THREADS_CreateThread
 
   DWORD threadId;
-  *p = CreateThread(NULL, 0, func, param, 0, &threadId);
+  *p = CreateThread(NULL, k_StackSize_ReserveSize, func, param, STACK_SIZE_PARAM_IS_A_RESERVATION, &threadId);
   
   #else
 
+#define CALL_beginthreadex(func2, param2, flags, threadIdPtr) \
+    ((HANDLE)(_beginthreadex(NULL, k_StackSize_ReserveSize, func2, param2, (flags) | STACK_SIZE_PARAM_IS_A_RESERVATION, threadIdPtr)))
+  
   unsigned threadId;
-  *p = (HANDLE)(_beginthreadex(NULL, 0, func, param, 0, &threadId));
+  *p = CALL_beginthreadex(func, param, 0, &threadId);
+
+#if 0 // 1 : for debug
+  {
+      DWORD_PTR prevMask;
+      DWORD_PTR affinity = 1 << 0;
+      prevMask = SetThreadAffinityMask(*p, (DWORD_PTR)affinity);
+      prevMask = prevMask;
+  }
+#endif
+#if 0 // 1 : for debug
+  {
+      /* win10: new thread will be created in same group that is assigned to parent thread
+                but affinity mask will contain all allowed threads of that group,
+                even if affinity mask of parent group is not full
+         win11: what group it will be created, if we have set
+                affinity of parent thread with ThreadGroupAffinity?
+      */
+      const
+         Func_GetThreadGroupAffinity fn =
+        (Func_GetThreadGroupAffinity) Z7_CAST_FUNC_C GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
+             "GetThreadGroupAffinity");
+      if (fn)
+      {
+        // BOOL wres2;
+        MY_GROUP_AFFINITY groupAffinity;
+        memset(&groupAffinity, 0, sizeof(groupAffinity));
+        /* wres2 = */ fn(*p, &groupAffinity);
+        PRF(printf("\n==Thread_Create cur = %6u GetThreadGroupAffinity(): "
+            "wres2_BOOL = %u, group=%u mask=%x\n",
+            GetCurrentThreadId(),
+            wres2,
+            groupAffinity.Group,
+            (UInt32)groupAffinity.Mask);)
+      }
+  }
+#endif
 
   #endif
 
@@ -93,7 +237,7 @@ WRes Thread_Create_With_Affinity(CThread *p, THREAD_FUNC_TYPE func, LPVOID param
   HANDLE h;
   WRes wres;
   unsigned threadId;
-  h = (HANDLE)(_beginthreadex(NULL, 0, func, param, CREATE_SUSPENDED, &threadId));
+  h = CALL_beginthreadex(func, param, CREATE_SUSPENDED, &threadId);
   *p = h;
   wres = HandleToWRes(h);
   if (h)
@@ -110,7 +254,84 @@ WRes Thread_Create_With_Affinity(CThread *p, THREAD_FUNC_TYPE func, LPVOID param
       */
     }
     {
-      DWORD prevSuspendCount = ResumeThread(h);
+      const DWORD prevSuspendCount = ResumeThread(h);
+      /* ResumeThread() returns:
+         0 : was_not_suspended
+         1 : was_resumed
+        -1 : error
+      */
+      if (prevSuspendCount == (DWORD)-1)
+        wres = GetError();
+    }
+  }
+
+  /* maybe we must use errno here, but probably GetLastError() is also OK. */
+  return wres;
+
+  #endif
+}
+
+
+WRes Thread_Create_With_Group(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, unsigned group, CAffinityMask affinityMask)
+{
+#ifdef USE_THREADS_CreateThread
+
+  UNUSED_VAR(group)
+  UNUSED_VAR(affinityMask)
+  return Thread_Create(p, func, param);
+
+#else
+
+  /* Windows Me/98/95: threadId parameter may not be NULL in _beginthreadex/CreateThread functions */
+  HANDLE h;
+  WRes wres;
+  unsigned threadId;
+  h = CALL_beginthreadex(func, param, CREATE_SUSPENDED, &threadId);
+  *p = h;
+  wres = HandleToWRes(h);
+  if (h)
+  {
+    // PrintProcess_Info();
+    {
+      const
+         Func_SetThreadGroupAffinity fn =
+        (Func_SetThreadGroupAffinity) Z7_CAST_FUNC_C GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
+             "SetThreadGroupAffinity");
+      if (fn)
+      {
+        // WRes wres2;
+        MY_GROUP_AFFINITY groupAffinity, prev_groupAffinity;
+        memset(&groupAffinity, 0, sizeof(groupAffinity));
+        // groupAffinity.Mask must use only bits that supported by current group
+        // (groupAffinity.Mask = 0) means all allowed bits
+        groupAffinity.Mask = affinityMask;
+        groupAffinity.Group = (WORD)group;
+        // wres2 =
+        fn(h, &groupAffinity, &prev_groupAffinity);
+        /*
+        if (groupAffinity.Group == prev_groupAffinity.Group)
+          wres2 = wres2;
+        else
+          wres2 = wres2;
+        if (wres2 == 0)
+        {
+          wres2 = GetError();
+          PRF(printf("\n==SetThreadGroupAffinity error: %u\n", wres2);)
+        }
+        else
+        {
+          PRF(printf("\n==Thread_Create_With_Group::SetThreadGroupAffinity()"
+            " threadId = %6u"
+            " group=%u mask=%x\n",
+            threadId,
+            prev_groupAffinity.Group,
+            (UInt32)prev_groupAffinity.Mask);)
+        }
+        */
+      }
+    }
+    {
+      const DWORD prevSuspendCount = ResumeThread(h);
       /* ResumeThread() returns:
          0 : was_not_suspended
          1 : was_resumed
@@ -297,6 +518,13 @@ WRes Thread_Create(CThread *p, THREAD_FUNC_TYPE func, LPVOID param)
   return Thread_Create_With_CpuSet(p, func, param, NULL);
 }
 
+/*
+WRes Thread_Create_With_Group(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, unsigned group, CAffinityMask affinity)
+{
+  UNUSED_VAR(group)
+  return Thread_Create_With_Affinity(p, func, param, affinity);
+}
+*/
 
 WRes Thread_Create_With_Affinity(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, CAffinityMask affinity)
 {
@@ -577,5 +805,22 @@ WRes AutoResetEvent_OptCreate_And_Reset(CAutoResetEvent *p)
   return AutoResetEvent_CreateNotSignaled(p);
 }
 
+void ThreadNextGroup_Init(CThreadNextGroup *p, UInt32 numGroups, UInt32 startGroup)
+{
+  // printf("\n====== ThreadNextGroup_Init numGroups = %x: startGroup=%x\n", numGroups, startGroup);
+  if (numGroups == 0)
+      numGroups = 1;
+  p->NumGroups = numGroups;
+  p->NextGroup = startGroup % numGroups;
+}
+
+
+UInt32 ThreadNextGroup_GetNext(CThreadNextGroup *p)
+{
+  const UInt32 next = p->NextGroup;
+  p->NextGroup = (next + 1) % p->NumGroups;
+  return next;
+}
+
 #undef PRF
 #undef Print

src/Common/lzma/Threads.h

@@ -1,5 +1,5 @@
 /* Threads.h -- multithreading library
-2024-03-28 : Igor Pavlov : Public domain */
+: Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_THREADS_H
 #define ZIP7_INC_THREADS_H
@@ -140,12 +140,22 @@ WRes Thread_Create_With_Affinity(CThread *p, THREAD_FUNC_TYPE func, LPVOID param
 WRes Thread_Wait_Close(CThread *p);
 
 #ifdef _WIN32
+WRes Thread_Create_With_Group(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, unsigned group, CAffinityMask affinityMask);
 #define Thread_Create_With_CpuSet(p, func, param, cs) \
   Thread_Create_With_Affinity(p, func, param, *cs)
 #else
 WRes Thread_Create_With_CpuSet(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, const CCpuSet *cpuSet);
 #endif
 
+typedef struct
+{
+  unsigned NumGroups;
+  unsigned NextGroup;
+} CThreadNextGroup;
+
+void ThreadNextGroup_Init(CThreadNextGroup *p, unsigned numGroups, unsigned startGroup);
+unsigned ThreadNextGroup_GetNext(CThreadNextGroup *p);
+
 
 #ifdef _WIN32
 

src/Common/lzma/lzma-history.txt

@@ -1,6 +1,47 @@
 HISTORY of the LZMA SDK
 -----------------------
 
+26.01          2026-04-27
+-------------------------
+- linux version of 7-Zip can use huge pages (2 MB pages). It can increase compression
+  speed for 10% for 7z/xz/LZMA/LZMA2 compression.
+- new -spo[d|c|r] switch specifies the path generation mode for the output directory
+  for archive extraction. The output directory path is generated from the path specified
+  in the -o{dir_path} switch and the name of the archive being unpacked.
+    -spod : for Linux/Posix/macOS: -o{dir_path} specifies the direct path to the output directory.
+            The asterisk (*) character in {dir_path} will not be replaced by the archive name.
+    -spoc : 7-Zip will concatenate the path specified in -o{dir_path} with the archive name
+            to form the final path to the output directory.
+    -spor : 7-Zip will replace asterisk (*) character in the path specified in the -o{dir_path}
+            with the archive name. This is the default option.
+- The 7zdec.exe program (a lightweight 7z archive decoder) has been modified for security purposes.
+  Now 7zdec.exe extracts files only to the current folder and its subfolders.
+- some bugs were fixed.
+
+
+
+26.00          2026-02-12
+-------------------------
+- some bugs were fixed.
+
+
+25.01          2025-08-03
+-------------------------
+- The code for handling symbolic links has been changed
+  to provide greater security when extracting files from archives.
+  Command line switch -snld20 can be used to bypass default security
+  checks when creating symbolic links.
+
+
+25.00          2025-07-05
+-------------------------
+- 7-Zip for Windows can now use more than 64 CPU threads for compression
+  to zip/7z/xz archives and for the 7-Zip benchmark.
+  If there are more than one processor group in Windows (on systems with more than
+  64 cpu threads), 7-Zip distributes running CPU threads across different processor groups.
+- fixed some bugs and vulnerabilities.
+
+
 24.09          2024-11-29
 -------------------------
 - The default dictionary size values for LZMA/LZMA2 compression methods were increased:

src/Common/lzma/lzma-sdk.txt

@@ -1,4 +1,4 @@
-LZMA SDK 24.09
+LZMA SDK 26.01
 --------------
 
 LZMA SDK provides the documentation, samples, header files,
@@ -59,6 +59,34 @@ LZMA SDK Contents
     - console programs for lzma / 7z / xz compression and decompression
     - SFX modules for installers.
 
+How to compile with makefile in Windows
+---------------------------------------
+
+Some macronames can be defined for compiling with makefile:
+
+PLATFORM
+  with possible values: x64, x86, arm64, arm, ia64
+
+OLD_COMPILER
+  for old VC compiler, like MSCV 6.0.
+
+MY_DYNAMIC_LINK
+  for dynamic linking to the run-time library (msvcrt.dll).
+  The default makefile option is static linking to the run-time library.
+
+To compile 7zr.exe file for x64 with Visual Studio 2022,
+use the following command sequence:
+
+  cd SRC\CPP\7zip\Bundles\Alone7z\
+  %comspec% /k "C:\Program Files\VS2022\VC\Auxiliary\Build\vcvars64.bat"
+  nmake
+
+You can use other "vcvars*.bat" files from the "VS2022\VC\Auxiliary\Build" directory
+to compile for other platforms:
+  vcvars64.bat
+  vcvarsamd64_arm64.bat
+  vcvarsamd64_x86.bat
+
 
 UNIX/Linux version 
 ------------------