--- gvpe/src/lzf/lzf_c.c 2008/08/07 16:34:21 1.5 +++ gvpe/src/lzf/lzf_c.c 2019/04/01 03:10:26 1.7 @@ -1,16 +1,16 @@ /* - * Copyright (c) 2000-2007 Marc Alexander Lehmann - * + * Copyright (c) 2000-2010,2012 Marc Alexander Lehmann + * * 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, * this list of conditions and the following disclaimer. - * + * * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * + * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO @@ -40,29 +40,29 @@ /* * don't play with this unless you benchmark! - * decompression is not dependent on the hash function - * the hashing function might seem strange, just believe me + * the data format is not dependent on the hash function. + * 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 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)) +# 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)) +# endif # else -# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) +/* 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 -/* - * 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 */ @@ -71,10 +71,6 @@ # define IDX(h) ((h) & (HSIZE - 1)) #endif -#define MAX_LIT (1 << 5) -#define MAX_OFF (1 << 13) -#define MAX_REF ((1 << 8) + (1 << 3)) - #if __GNUC__ >= 3 # define expect(expr,value) __builtin_expect ((expr),(value)) # define inline inline @@ -89,9 +85,9 @@ /* * compressed format * - * 000LLLLL ; literal - * LLLooooo oooooooo ; backref L - * 111ooooo LLLLLLLL oooooooo ; backref L+7 + * 000LLLLL ; literal, L+1=1..33 octets + * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset + * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset * */ @@ -106,7 +102,6 @@ #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; @@ -115,13 +110,18 @@ /* 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 + * 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) - unsigned _int64 off; /* workaround for missing POSIX compliance */ +#if defined (_WIN32) && defined (_M_X64) + /* workaround for missing POSIX compliance */ + #if __GNUC__ + unsigned long long off; + #else + unsigned __int64 off; + #endif #else unsigned long off; #endif @@ -132,11 +132,7 @@ return 0; #if INIT_HTAB - memset (htab, 0, sizeof (htab)); -# if 0 - for (hslot = htab; hslot < htab + HSIZE; hslot++) - *hslot++ = ip; -# endif + memset (htab, 0, sizeof (LZF_STATE)); #endif lit = 0; op++; /* start run */ @@ -144,38 +140,38 @@ hval = FRST (ip); while (ip < in_end - 2) { + LZF_HSLOT *hslot; + hval = NEXT (hval, ip); hslot = htab + IDX (hval); - ref = *hslot; *hslot = ip; + 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 - && ip + 4 < in_end + && (off = ip - ref - 1) < LZF_MAX_OFF && ref > (u8 *)in_data -#if STRICT_ALIGN - && ref[0] == ip[0] - && ref[1] == ip[1] && ref[2] == ip[2] +#if STRICT_ALIGN + && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) #else && *(u16 *)ref == *(u16 *)ip - && ref[2] == ip[2] #endif ) { /* match found at *ref++ */ unsigned int len = 2; unsigned int maxlen = in_end - ip - len; - maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; + maxlen = maxlen > LZF_MAX_REF ? LZF_MAX_REF : maxlen; + + 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; op [- lit - 1] = lit - 1; /* stop run */ op -= !lit; /* undo run if length is zero */ - if (expect_false (op + 3 + 1 >= out_end)) - return 0; - for (;;) { if (expect_true (maxlen > 16)) @@ -208,7 +204,7 @@ break; } - len -= 2; + len -= 2; /* len is now #octets - 1 */ ip++; if (len < 7) @@ -223,33 +219,40 @@ *op++ = off; + lit = 0; op++; /* start run */ + + ip += len + 1; + + if (expect_false (ip >= in_end - 2)) + break; + #if ULTRA_FAST || VERY_FAST - ip += len; -#if VERY_FAST && !ULTRA_FAST --ip; -#endif +# if VERY_FAST && !ULTRA_FAST + --ip; +# endif hval = FRST (ip); hval = NEXT (hval, ip); - htab[IDX (hval)] = ip; + htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; ip++; -#if VERY_FAST && !ULTRA_FAST +# if VERY_FAST && !ULTRA_FAST hval = NEXT (hval, ip); - htab[IDX (hval)] = ip; + htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; ip++; -#endif +# endif #else + ip -= len + 1; + do { hval = NEXT (hval, ip); - htab[IDX (hval)] = ip; + htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; ip++; } while (len--); #endif - - lit = 0; op++; /* start run */ } else { @@ -259,7 +262,7 @@ lit++; *op++ = *ip++; - if (expect_false (lit == MAX_LIT)) + if (expect_false (lit == LZF_MAX_LIT)) { op [- lit - 1] = lit - 1; /* stop run */ lit = 0; op++; /* start run */ @@ -274,7 +277,7 @@ { lit++; *op++ = *ip++; - if (expect_false (lit == MAX_LIT)) + if (expect_false (lit == LZF_MAX_LIT)) { op [- lit - 1] = lit - 1; /* stop run */ lit = 0; op++; /* start run */