1 files changed, 74 insertions, 56 deletions

diff --git a/C/7zCrc.c b/C/7zCrc.c
index f186324..c995a8b 100644
--- a/C/7zCrc.c
+++ b/C/7zCrc.c
@@ -1,5 +1,5 @@
-/* 7zCrc.c -- CRC32 init
-2021-04-01 : Igor Pavlov : Public domain */
+/* 7zCrc.c -- CRC32 calculation and init
+2023-04-02 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -13,22 +13,20 @@
 #else
   #define CRC_NUM_TABLES 9
 
-  #define CRC_UINT32_SWAP(v) ((v >> 24) | ((v >> 8) & 0xFF00) | ((v << 8) & 0xFF0000) | (v << 24))
-
-  UInt32 MY_FAST_CALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
-  UInt32 MY_FAST_CALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
 #endif
 
 #ifndef MY_CPU_BE
-  UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
-  UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
 #endif
 
-typedef UInt32 (MY_FAST_CALL *CRC_FUNC)(UInt32 v, const void *data, size_t size, const UInt32 *table);
-
+/*
 extern
 CRC_FUNC g_CrcUpdateT4;
 CRC_FUNC g_CrcUpdateT4;
+*/
 extern
 CRC_FUNC g_CrcUpdateT8;
 CRC_FUNC g_CrcUpdateT8;
@@ -44,20 +42,22 @@ CRC_FUNC g_CrcUpdate;
 
 UInt32 g_CrcTable[256 * CRC_NUM_TABLES];
 
-UInt32 MY_FAST_CALL CrcUpdate(UInt32 v, const void *data, size_t size)
+UInt32 Z7_FASTCALL CrcUpdate(UInt32 v, const void *data, size_t size)
 {
   return g_CrcUpdate(v, data, size, g_CrcTable);
 }
 
-UInt32 MY_FAST_CALL CrcCalc(const void *data, size_t size)
+UInt32 Z7_FASTCALL CrcCalc(const void *data, size_t size)
 {
   return g_CrcUpdate(CRC_INIT_VAL, data, size, g_CrcTable) ^ CRC_INIT_VAL;
 }
 
+#if CRC_NUM_TABLES < 4 \
+   || (CRC_NUM_TABLES == 4 && defined(MY_CPU_BE)) \
+   || (!defined(MY_CPU_LE) && !defined(MY_CPU_BE))
 #define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
-
-UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table);
-UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table)
+UInt32 Z7_FASTCALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
   const Byte *p = (const Byte *)data;
   const Byte *pEnd = p + size;
@@ -65,7 +65,7 @@ UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const U
     v = CRC_UPDATE_BYTE_2(v, *p);
   return v;
 }
-
+#endif
 
 /* ---------- hardware CRC ---------- */
 
@@ -78,16 +78,29 @@ UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const U
   #if defined(_MSC_VER)
     #if defined(MY_CPU_ARM64)
     #if (_MSC_VER >= 1910)
+    #ifndef __clang__
         #define USE_ARM64_CRC
+        #include <intrin.h>
+    #endif
     #endif
     #endif
   #elif (defined(__clang__) && (__clang_major__ >= 3)) \
      || (defined(__GNUC__) && (__GNUC__ > 4))
       #if !defined(__ARM_FEATURE_CRC32)
         #define __ARM_FEATURE_CRC32 1
-          #if (!defined(__clang__) || (__clang_major__ > 3)) // fix these numbers
+        #if defined(__clang__)
+          #if defined(MY_CPU_ARM64)
+            #define ATTRIB_CRC __attribute__((__target__("crc")))
+          #else
+            #define ATTRIB_CRC __attribute__((__target__("armv8-a,crc")))
+          #endif
+        #else
+          #if defined(MY_CPU_ARM64)
+            #define ATTRIB_CRC __attribute__((__target__("+crc")))
+          #else
             #define ATTRIB_CRC __attribute__((__target__("arch=armv8-a+crc")))
           #endif
+        #endif
       #endif
       #if defined(__ARM_FEATURE_CRC32)
         #define USE_ARM64_CRC
@@ -105,7 +118,7 @@ UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const U
 
 #pragma message("ARM64 CRC emulation")
 
-MY_FORCE_INLINE
+Z7_FORCE_INLINE
 UInt32 __crc32b(UInt32 v, UInt32 data)
 {
   const UInt32 *table = g_CrcTable;
@@ -113,7 +126,7 @@ UInt32 __crc32b(UInt32 v, UInt32 data)
   return v;
 }
 
-MY_FORCE_INLINE
+Z7_FORCE_INLINE
 UInt32 __crc32w(UInt32 v, UInt32 data)
 {
   const UInt32 *table = g_CrcTable;
@@ -124,7 +137,7 @@ UInt32 __crc32w(UInt32 v, UInt32 data)
   return v;
 }
 
-MY_FORCE_INLINE
+Z7_FORCE_INLINE
 UInt32 __crc32d(UInt32 v, UInt64 data)
 {
   const UInt32 *table = g_CrcTable;
@@ -156,9 +169,9 @@ UInt32 __crc32d(UInt32 v, UInt64 data)
 // #pragma message("USE ARM HW CRC")
 
 ATTRIB_CRC
-UInt32 MY_FAST_CALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table);
 ATTRIB_CRC
-UInt32 MY_FAST_CALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table)
+UInt32 Z7_FASTCALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
   const Byte *p = (const Byte *)data;
   UNUSED_VAR(table);
@@ -188,9 +201,9 @@ UInt32 MY_FAST_CALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, cons
 }
 
 ATTRIB_CRC
-UInt32 MY_FAST_CALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table);
 ATTRIB_CRC
-UInt32 MY_FAST_CALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table)
+UInt32 Z7_FASTCALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
   const Byte *p = (const Byte *)data;
   UNUSED_VAR(table);
@@ -219,6 +232,9 @@ UInt32 MY_FAST_CALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, cons
   return v;
 }
 
+#undef T0_32_UNROLL_BYTES
+#undef T0_64_UNROLL_BYTES
+
 #endif // defined(USE_ARM64_CRC) || defined(USE_CRC_EMU)
 
 #endif // MY_CPU_LE
@@ -226,7 +242,7 @@ UInt32 MY_FAST_CALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, cons
 
 
 
-void MY_FAST_CALL CrcGenerateTable()
+void Z7_FASTCALL CrcGenerateTable(void)
 {
   UInt32 i;
   for (i = 0; i < 256; i++)
@@ -239,64 +255,62 @@ void MY_FAST_CALL CrcGenerateTable()
   }
   for (i = 256; i < 256 * CRC_NUM_TABLES; i++)
   {
-    UInt32 r = g_CrcTable[(size_t)i - 256];
+    const UInt32 r = g_CrcTable[(size_t)i - 256];
     g_CrcTable[i] = g_CrcTable[r & 0xFF] ^ (r >> 8);
   }
 
   #if CRC_NUM_TABLES < 4
-  
-  g_CrcUpdate = CrcUpdateT1;
-  
-  #else
- 
-  #ifdef MY_CPU_LE
-
-    g_CrcUpdateT4 = CrcUpdateT4;
-    g_CrcUpdate = CrcUpdateT4;
-
-    #if CRC_NUM_TABLES >= 8
+    g_CrcUpdate = CrcUpdateT1;
+  #elif defined(MY_CPU_LE)
+    // g_CrcUpdateT4 = CrcUpdateT4;
+    #if CRC_NUM_TABLES < 8
+      g_CrcUpdate = CrcUpdateT4;
+    #else // CRC_NUM_TABLES >= 8
       g_CrcUpdateT8 = CrcUpdateT8;
-  
+      /*
       #ifdef MY_CPU_X86_OR_AMD64
       if (!CPU_Is_InOrder())
       #endif
-        g_CrcUpdate = CrcUpdateT8;
+      */
+      g_CrcUpdate = CrcUpdateT8;
     #endif
-
   #else
   {
-    #ifndef MY_CPU_BE
+   #ifndef MY_CPU_BE
     UInt32 k = 0x01020304;
     const Byte *p = (const Byte *)&k;
     if (p[0] == 4 && p[1] == 3)
     {
-      g_CrcUpdateT4 = CrcUpdateT4;
-      g_CrcUpdate = CrcUpdateT4;
-      #if CRC_NUM_TABLES >= 8
-      g_CrcUpdateT8 = CrcUpdateT8;
-      g_CrcUpdate = CrcUpdateT8;
+      #if CRC_NUM_TABLES < 8
+        // g_CrcUpdateT4 = CrcUpdateT4;
+        g_CrcUpdate   = CrcUpdateT4;
+      #else  // CRC_NUM_TABLES >= 8
+        g_CrcUpdateT8 = CrcUpdateT8;
+        g_CrcUpdate   = CrcUpdateT8;
       #endif
     }
     else if (p[0] != 1 || p[1] != 2)
       g_CrcUpdate = CrcUpdateT1;
     else
-    #endif
+   #endif // MY_CPU_BE
     {
       for (i = 256 * CRC_NUM_TABLES - 1; i >= 256; i--)
       {
-        UInt32 x = g_CrcTable[(size_t)i - 256];
-        g_CrcTable[i] = CRC_UINT32_SWAP(x);
+        const UInt32 x = g_CrcTable[(size_t)i - 256];
+        g_CrcTable[i] = Z7_BSWAP32(x);
       }
-      g_CrcUpdateT4 = CrcUpdateT1_BeT4;
-      g_CrcUpdate = CrcUpdateT1_BeT4;
-      #if CRC_NUM_TABLES >= 8
-      g_CrcUpdateT8 = CrcUpdateT1_BeT8;
-      g_CrcUpdate = CrcUpdateT1_BeT8;
+      #if CRC_NUM_TABLES <= 4
+        g_CrcUpdate = CrcUpdateT1;
+      #elif CRC_NUM_TABLES <= 8
+        // g_CrcUpdateT4 = CrcUpdateT1_BeT4;
+        g_CrcUpdate   = CrcUpdateT1_BeT4;
+      #else  // CRC_NUM_TABLES > 8
+        g_CrcUpdateT8 = CrcUpdateT1_BeT8;
+        g_CrcUpdate   = CrcUpdateT1_BeT8;
       #endif
     }
   }
-  #endif
-  #endif
+  #endif // CRC_NUM_TABLES < 4
 
   #ifdef MY_CPU_LE
     #ifdef USE_ARM64_CRC
@@ -320,3 +334,7 @@ void MY_FAST_CALL CrcGenerateTable()
     #endif
   #endif
 }
+
+#undef kCrcPoly
+#undef CRC64_NUM_TABLES
+#undef CRC_UPDATE_BYTE_2