[ViewVC] Diff of: cvs/liblzf/lzf

Comparing liblzf/lzf_c.c (file contents):
Revision 1.16 by root, Mon Nov 5 23:28:40 2007 UTC vs.
Revision 1.47 by root, Sun Feb 26 03:17:50 2012 UTC

 /*
- * Copyright (c) 2000-2007 Marc Alexander Lehmann <schmorp@schmorp.de>
+ * Copyright (c) 2000-2010,2012 Marc Alexander Lehmann <schmorp@schmorp.de>
  *
  * Redistribution and use in source and binary forms, with or without modifica-
  * tion, are permitted provided that the following conditions are met:
  *
  *   1.  Redistributions of source code must retain the above copyright notice,
 #define HSIZE (1 << (HLOG))
 /*
  * don't play with this unless you benchmark!
- * decompression is not dependent on the hash function
+ * the data format is not dependent on the hash function.
- * the hashing function might seem strange, just believe me
+ * the hash function might seem strange, just believe me,
  * it works ;)
  */
 #ifndef FRST
 # define FRST(p) (((p[0]) << 8) | p[1])
 # define NEXT(v,p) (((v) << 8) | p[2])
+# if MULTIPLICATION_IS_SLOW
+#  if ULTRA_FAST
+#   define IDX(h) ((( h             >> (3*8 - HLOG)) - h  ) & (HSIZE - 1))
+#  elif VERY_FAST
+#   define IDX(h) ((( h             >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
+#  else
-# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
+#   define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
-/*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (HSIZE-1))*/
+#  endif
+# else
+/* this one was developed with sesse,
+ * and is very similar to the one in snappy.
+ * it does need a modern enough cpu with a fast multiplication.
+ */
+#  define IDX(h) (((h * 0x1e35a7bdU) >> (32 - HLOG - 8)) & (HSIZE - 1))
-#endif
+# endif
-/*
+#endif
- * IDX works because it is very similar to a multiplicative hash, e.g.
- * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
- * the latter is also quite fast on newer CPUs, and compresses similarly.
- *
- * the next one is also quite good, albeit slow ;)
- * (int)(cos(h & 0xffffff) * 1e6)
- */
 #if 0
 /* original lzv-like hash function, much worse and thus slower */
 # define FRST(p) (p[0] << 5) ^ p[1]
 # define NEXT(v,p) ((v) << 5) ^ p[2]
 #define        MAX_LIT        (1 <<  5)
 #define        MAX_OFF        (1 << 13)
 #define        MAX_REF        ((1 << 8) + (1 << 3))
-#if (__i386 || __amd64) && __GNUC__ >= 3
+#if __GNUC__ >= 3
-# define lzf_movsb(dst, src, len)                \
+# define expect(expr,value)         __builtin_expect ((expr),(value))
-   asm ("rep movsb"                              \
+# define inline                     inline
-        : "=D" (dst), "=S" (src), "=c" (len)     \
+#else
-        :  "0" (dst),  "1" (src),  "2" (len));
+# define expect(expr,value)         (expr)
+# define inline                     static
 #endif
+#define expect_false(expr) expect ((expr) != 0, 0)
+#define expect_true(expr)  expect ((expr) != 0, 1)
 /*
  * compressed format
  *
- * 000LLLLL <L+1>    ; literal
+ * 000LLLLL <L+1>    ; literal, L+1=1..33 octets
- * LLLooooo oooooooo ; backref L
+ * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
- * 111ooooo LLLLLLLL oooooooo ; backref L+7
+ * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
  *
  */
 unsigned int
 lzf_compress (const void *const in_data, unsigned int in_len,
               )
 {
 #if !LZF_STATE_ARG
   LZF_STATE htab;
 #endif
-  const u8 **hslot;
   const u8 *ip = (const u8 *)in_data;
         u8 *op = (u8 *)out_data;
   const u8 *in_end  = ip + in_len;
         u8 *out_end = op + out_len;
   const u8 *ref;
-  unsigned int hval = FRST (ip);
+  /* off requires a type wide enough to hold a general pointer difference.
+   * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only
+   * works for differences within a single object). We also assume that
+   * no bit pattern traps. Since the only platform that is both non-POSIX
+   * and fails to support both assumptions is windows 64 bit, we make a
+   * special workaround for it.
+   */
+#if defined (_WIN32) && defined (_M_X64)
+  /* workaround for missing POSIX compliance */
+  #if  #if __GNUC__
+    unsigned long long off;
+  #else
+    unsigned __int64 off;
+  #endif
+#else
   unsigned long off;
-           int lit = 0;
+#endif
+  unsigned int hval;
+  int lit;
+  if (!in_len || !out_len)
+    return 0;
 #if INIT_HTAB
   memset (htab, 0, sizeof (htab));
-# if 0
-  for (hslot = htab; hslot < htab + HSIZE; hslot++)
-    *hslot++ = ip;
-# endif
+#endif
-#endif
-  for (;;)
+  lit = 0; op++; /* start run */
+  hval = FRST (ip);
+  while (ip < in_end - 2)
     {
-      if (ip < in_end - 2)
+      LZF_HSLOT *hslot;
+      hval = NEXT (hval, ip);
+      hslot = htab + IDX (hval);
+      ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS;
+      if (1
+#if INIT_HTAB
+          && ref < ip /* the next test will actually take care of this, but this is faster */
+#endif
+          && (off = ip - ref - 1) < MAX_OFF
+          && ref > (u8 *)in_data
+          && ref[2] == ip[2]
+#if STRICT_ALIGN
+          && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0])
+#else
+          && *(u16 *)ref == *(u16 *)ip
+#endif
+        )
         {
-          hval = NEXT (hval, ip);
-          hslot = htab + IDX (hval);
-          ref = *hslot; *hslot = ip;
-          if (1
-#if INIT_HTAB && !USE_MEMCPY
-              && ref < ip /* the next test will actually take care of this, but this is faster */
-#endif
-              && (off = ip - ref - 1) < MAX_OFF
-              && ip + 4 < in_end
-              && ref > (u8 *)in_data
-#if STRICT_ALIGN
-              && ref[0] == ip[0]
-              && ref[1] == ip[1]
-              && ref[2] == ip[2]
-#else
-              && *(u16 *)ref == *(u16 *)ip
-              && ref[2] == ip[2]
-#endif
-            )
-            {
-              /* match found at *ref++ */
+          /* match found at *ref++ */
-              unsigned int len = 2;
+          unsigned int len = 2;
-              unsigned int maxlen = in_end - ip - len;
+          unsigned int maxlen = in_end - ip - len;
-              maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
+          maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
-              if (op + lit + 1 + 3 >= out_end)
+          if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */
+            if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */
-                return 0;
+              return 0;
+          op [- lit - 1] = lit - 1; /* stop run */
+          op -= !lit; /* undo run if length is zero */
+          for (;;)
+            {
+              if (expect_true (maxlen > 16))
+                {
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                }
               do
                 len++;
               while (len < maxlen && ref[len] == ip[len]);
-              if (lit)
-                {
+              break;
-                  *op++ = lit - 1;
-                  lit = -lit;
-                  do
-                    *op++ = ip[lit];
-                  while (++lit);
-                }
+            }
+          len -= 2; /* len is now #octets - 1 */
+          ip++;
+          if (len < 7)
+            {
+              *op++ = (off >> 8) + (len << 5);
+            }
+          else
+            {
+              *op++ = (off >> 8) + (  7 << 5);
-              len -= 2;
+              *op++ = len - 7;
+            }
+          *op++ = off;
+          lit = 0; op++; /* start run */
+          ip += len + 1;
+          if (expect_false (ip >= in_end - 2))
+            break;
+#if ULTRA_FAST || VERY_FAST
+          --ip;
+# if VERY_FAST && !ULTRA_FAST
+          --ip;
+# endif
+          hval = FRST (ip);
+          hval = NEXT (hval, ip);
+          htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
+          ip++;
+# if VERY_FAST && !ULTRA_FAST
+          hval = NEXT (hval, ip);
+          htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
+          ip++;
+# endif
+#else
+          ip -= len + 1;
+          do
+            {
+              hval = NEXT (hval, ip);
+              htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
               ip++;
-              if (len < 7)
-                {
-                  *op++ = (off >> 8) + (len << 5);
-                }
+            }
-              else
-                {
-                  *op++ = (off >> 8) + (  7 << 5);
-                  *op++ = len - 7;
-                }
-              *op++ = off;
-#if ULTRA_FAST || VERY_FAST
-              ip += len;
-#if VERY_FAST && !ULTRA_FAST
-              --ip;
-#endif
-              hval = FRST (ip);
-              hval = NEXT (hval, ip);
-              htab[IDX (hval)] = ip;
-              ip++;
-#if VERY_FAST && !ULTRA_FAST
-              hval = NEXT (hval, ip);
-              htab[IDX (hval)] = ip;
-              ip++;
-#endif
-#else
-              do
-                {
-                  hval = NEXT (hval, ip);
-                  htab[IDX (hval)] = ip;
-                  ip++;
-                }
-              while (len--);
+          while (len--);
 #endif
-              continue;
-            }
         }
-      else if (ip == in_end)
+      else
-        break;
-      /* one more literal byte we must copy */
-      lit++;
-      ip++;
-      if (lit == MAX_LIT)
         {
-          if (op + 1 + MAX_LIT >= out_end)
+          /* one more literal byte we must copy */
+          if (expect_false (op >= out_end))
             return 0;
-          *op++ = MAX_LIT - 1;
+          lit++; *op++ = *ip++;
-#ifdef lzf_movsb
+          if (expect_false (lit == MAX_LIT))
-          ip -= lit;
-          lzf_movsb (op, ip, lit);
-#else
-          lit = -lit;
-          do
+            {
-            *op++ = ip[lit];
+              op [- lit - 1] = lit - 1; /* stop run */
-          while (++lit);
+              lit = 0; op++; /* start run */
-#endif
+            }
         }
     }
-  if (lit)
+  if (op + 3 > out_end) /* at most 3 bytes can be missing here */
+    return 0;
+  while (ip < in_end)
     {
-      if (op + lit + 1 >= out_end)
+      lit++; *op++ = *ip++;
-	return 0;
-      *op++ = lit - 1;
+      if (expect_false (lit == MAX_LIT))
-#ifdef lzf_movsb
+        {
-      ip -= lit;
+          op [- lit - 1] = lit - 1; /* stop run */
-      lzf_movsb (op, ip, lit);
+          lit = 0; op++; /* start run */
-#else
+        }
-      lit = -lit;
-      do
-	*op++ = ip[lit];
-      while (++lit);
-#endif
     }
+  op [- lit - 1] = lit - 1; /* end run */
+  op -= !lit; /* undo run if length is zero */
-  return op - (u8 *) out_data;
+  return op - (u8 *)out_data;
 }

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing liblzf/lzf_c.c (file contents): Revision 1.16 by root, Mon Nov 5 23:28:40 2007 UTC vs. Revision 1.47 by root, Sun Feb 26 03:17:50 2012 UTC

Diff Legend

Comparing liblzf/lzf_c.c (file contents):
Revision 1.16 by root, Mon Nov 5 23:28:40 2007 UTC vs.
Revision 1.47 by root, Sun Feb 26 03:17:50 2012 UTC