gosu 0.7.15 → 0.7.16

data/dependencies/zlib/contrib/asm686/README.686

@@ -0,0 +1,34 @@
+This is a patched version of zlib, modified to use
+Pentium-Pro-optimized assembly code in the deflation algorithm. The
+files changed/added by this patch are:
+
+README.686
+match.S
+
+The speedup that this patch provides varies, depending on whether the
+compiler used to build the original version of zlib falls afoul of the
+PPro's speed traps. My own tests show a speedup of around 10-20% at
+the default compression level, and 20-30% using -9, against a version
+compiled using gcc 2.7.2.3. Your mileage may vary.
+
+Note that this code has been tailored for the PPro/PII in particular,
+and will not perform particuarly well on a Pentium.
+
+If you are using an assembler other than GNU as, you will have to
+translate match.S to use your assembler's syntax. (Have fun.)
+
+Brian Raiter
+breadbox@muppetlabs.com
+April, 1998
+
+
+Added for zlib 1.1.3:
+
+The patches come from
+http://www.muppetlabs.com/~breadbox/software/assembly.html
+
+To compile zlib with this asm file, copy match.S to the zlib directory
+then do:
+
+CFLAGS="-O3 -DASMV" ./configure
+make OBJA=match.o

data/dependencies/zlib/contrib/asm686/match.S

@@ -0,0 +1,329 @@
+/* match.s -- Pentium-Pro-optimized version of longest_match()
+ * Written for zlib 1.1.2
+ * Copyright (C) 1998 Brian Raiter <breadbox@muppetlabs.com>
+ *
+ * This is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License.
+ */
+
+#ifndef NO_UNDERLINE
+#define	match_init	_match_init
+#define	longest_match	_longest_match
+#endif
+
+#define	MAX_MATCH	(258)
+#define	MIN_MATCH	(3)
+#define	MIN_LOOKAHEAD	(MAX_MATCH + MIN_MATCH + 1)
+#define	MAX_MATCH_8	((MAX_MATCH + 7) & ~7)
+
+/* stack frame offsets */
+
+#define	chainlenwmask		0	/* high word: current chain len	*/
+					/* low word: s->wmask		*/
+#define	window			4	/* local copy of s->window	*/
+#define	windowbestlen		8	/* s->window + bestlen		*/
+#define	scanstart		16	/* first two bytes of string	*/
+#define	scanend			12	/* last two bytes of string	*/
+#define	scanalign		20	/* dword-misalignment of string	*/
+#define	nicematch		24	/* a good enough match size	*/
+#define	bestlen			28	/* size of best match so far	*/
+#define	scan			32	/* ptr to string wanting match	*/
+
+#define	LocalVarsSize		(36)
+/*	saved ebx		36 */
+/*	saved edi		40 */
+/*	saved esi		44 */
+/*	saved ebp		48 */
+/*	return address		52 */
+#define	deflatestate		56	/* the function arguments	*/
+#define	curmatch		60
+
+/* All the +zlib1222add offsets are due to the addition of fields
+ *  in zlib in the deflate_state structure since the asm code was first written
+ * (if you compile with zlib 1.0.4 or older, use "zlib1222add equ (-4)").
+ * (if you compile with zlib between 1.0.5 and 1.2.2.1, use "zlib1222add equ 0").
+ * if you compile with zlib 1.2.2.2 or later , use "zlib1222add equ 8").
+ */
+
+#define zlib1222add		(8)
+
+#define	dsWSize			(36+zlib1222add)
+#define	dsWMask			(44+zlib1222add)
+#define	dsWindow		(48+zlib1222add)
+#define	dsPrev			(56+zlib1222add)
+#define	dsMatchLen		(88+zlib1222add)
+#define	dsPrevMatch		(92+zlib1222add)
+#define	dsStrStart		(100+zlib1222add)
+#define	dsMatchStart		(104+zlib1222add)
+#define	dsLookahead		(108+zlib1222add)
+#define	dsPrevLen		(112+zlib1222add)
+#define	dsMaxChainLen		(116+zlib1222add)
+#define	dsGoodMatch		(132+zlib1222add)
+#define	dsNiceMatch		(136+zlib1222add)
+
+
+.file "match.S"
+
+.globl	match_init, longest_match
+
+.text
+
+/* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */
+
+longest_match:
+
+/* Save registers that the compiler may be using, and adjust %esp to	*/
+/* make room for our stack frame.					*/
+
+		pushl	%ebp
+		pushl	%edi
+		pushl	%esi
+		pushl	%ebx
+		subl	$LocalVarsSize, %esp
+
+/* Retrieve the function arguments. %ecx will hold cur_match		*/
+/* throughout the entire function. %edx will hold the pointer to the	*/
+/* deflate_state structure during the function's setup (before		*/
+/* entering the main loop).						*/
+
+		movl	deflatestate(%esp), %edx
+		movl	curmatch(%esp), %ecx
+
+/* uInt wmask = s->w_mask;						*/
+/* unsigned chain_length = s->max_chain_length;				*/
+/* if (s->prev_length >= s->good_match) {				*/
+/*     chain_length >>= 2;						*/
+/* }									*/
+
+		movl	dsPrevLen(%edx), %eax
+		movl	dsGoodMatch(%edx), %ebx
+		cmpl	%ebx, %eax
+		movl	dsWMask(%edx), %eax
+		movl	dsMaxChainLen(%edx), %ebx
+		jl	LastMatchGood
+		shrl	$2, %ebx
+LastMatchGood:
+
+/* chainlen is decremented once beforehand so that the function can	*/
+/* use the sign flag instead of the zero flag for the exit test.	*/
+/* It is then shifted into the high word, to make room for the wmask	*/
+/* value, which it will always accompany.				*/
+
+		decl	%ebx
+		shll	$16, %ebx
+		orl	%eax, %ebx
+		movl	%ebx, chainlenwmask(%esp)
+
+/* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;	*/
+
+		movl	dsNiceMatch(%edx), %eax
+		movl	dsLookahead(%edx), %ebx
+		cmpl	%eax, %ebx
+		jl	LookaheadLess
+		movl	%eax, %ebx
+LookaheadLess:	movl	%ebx, nicematch(%esp)
+
+/* register Bytef *scan = s->window + s->strstart;			*/
+
+		movl	dsWindow(%edx), %esi
+		movl	%esi, window(%esp)
+		movl	dsStrStart(%edx), %ebp
+		lea	(%esi,%ebp), %edi
+		movl	%edi, scan(%esp)
+
+/* Determine how many bytes the scan ptr is off from being		*/
+/* dword-aligned.							*/
+
+		movl	%edi, %eax
+		negl	%eax
+		andl	$3, %eax
+		movl	%eax, scanalign(%esp)
+
+/* IPos limit = s->strstart > (IPos)MAX_DIST(s) ?			*/
+/*     s->strstart - (IPos)MAX_DIST(s) : NIL;				*/
+
+		movl	dsWSize(%edx), %eax
+		subl	$MIN_LOOKAHEAD, %eax
+		subl	%eax, %ebp
+		jg	LimitPositive
+		xorl	%ebp, %ebp
+LimitPositive:
+
+/* int best_len = s->prev_length;					*/
+
+		movl	dsPrevLen(%edx), %eax
+		movl	%eax, bestlen(%esp)
+
+/* Store the sum of s->window + best_len in %esi locally, and in %esi.	*/
+
+		addl	%eax, %esi
+		movl	%esi, windowbestlen(%esp)
+
+/* register ush scan_start = *(ushf*)scan;				*/
+/* register ush scan_end   = *(ushf*)(scan+best_len-1);			*/
+/* Posf *prev = s->prev;						*/
+
+		movzwl	(%edi), %ebx
+		movl	%ebx, scanstart(%esp)
+		movzwl	-1(%edi,%eax), %ebx
+		movl	%ebx, scanend(%esp)
+		movl	dsPrev(%edx), %edi
+
+/* Jump into the main loop.						*/
+
+		movl	chainlenwmask(%esp), %edx
+		jmp	LoopEntry
+
+.balign 16
+
+/* do {
+ *     match = s->window + cur_match;
+ *     if (*(ushf*)(match+best_len-1) != scan_end ||
+ *         *(ushf*)match != scan_start) continue;
+ *     [...]
+ * } while ((cur_match = prev[cur_match & wmask]) > limit
+ *          && --chain_length != 0);
+ *
+ * Here is the inner loop of the function. The function will spend the
+ * majority of its time in this loop, and majority of that time will
+ * be spent in the first ten instructions.
+ *
+ * Within this loop:
+ * %ebx = scanend
+ * %ecx = curmatch
+ * %edx = chainlenwmask - i.e., ((chainlen << 16) | wmask)
+ * %esi = windowbestlen - i.e., (window + bestlen)
+ * %edi = prev
+ * %ebp = limit
+ */
+LookupLoop:
+		andl	%edx, %ecx
+		movzwl	(%edi,%ecx,2), %ecx
+		cmpl	%ebp, %ecx
+		jbe	LeaveNow
+		subl	$0x00010000, %edx
+		js	LeaveNow
+LoopEntry:	movzwl	-1(%esi,%ecx), %eax
+		cmpl	%ebx, %eax
+		jnz	LookupLoop
+		movl	window(%esp), %eax
+		movzwl	(%eax,%ecx), %eax
+		cmpl	scanstart(%esp), %eax
+		jnz	LookupLoop
+
+/* Store the current value of chainlen.					*/
+
+		movl	%edx, chainlenwmask(%esp)
+
+/* Point %edi to the string under scrutiny, and %esi to the string we	*/
+/* are hoping to match it up with. In actuality, %esi and %edi are	*/
+/* both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and %edx is	*/
+/* initialized to -(MAX_MATCH_8 - scanalign).				*/
+
+		movl	window(%esp), %esi
+		movl	scan(%esp), %edi
+		addl	%ecx, %esi
+		movl	scanalign(%esp), %eax
+		movl	$(-MAX_MATCH_8), %edx
+		lea	MAX_MATCH_8(%edi,%eax), %edi
+		lea	MAX_MATCH_8(%esi,%eax), %esi
+
+/* Test the strings for equality, 8 bytes at a time. At the end,
+ * adjust %edx so that it is offset to the exact byte that mismatched.
+ *
+ * We already know at this point that the first three bytes of the
+ * strings match each other, and they can be safely passed over before
+ * starting the compare loop. So what this code does is skip over 0-3
+ * bytes, as much as necessary in order to dword-align the %edi
+ * pointer. (%esi will still be misaligned three times out of four.)
+ *
+ * It should be confessed that this loop usually does not represent
+ * much of the total running time. Replacing it with a more
+ * straightforward "rep cmpsb" would not drastically degrade
+ * performance.
+ */
+LoopCmps:
+		movl	(%esi,%edx), %eax
+		xorl	(%edi,%edx), %eax
+		jnz	LeaveLoopCmps
+		movl	4(%esi,%edx), %eax
+		xorl	4(%edi,%edx), %eax
+		jnz	LeaveLoopCmps4
+		addl	$8, %edx
+		jnz	LoopCmps
+		jmp	LenMaximum
+LeaveLoopCmps4:	addl	$4, %edx
+LeaveLoopCmps:	testl	$0x0000FFFF, %eax
+		jnz	LenLower
+		addl	$2, %edx
+		shrl	$16, %eax
+LenLower:	subb	$1, %al
+		adcl	$0, %edx
+
+/* Calculate the length of the match. If it is longer than MAX_MATCH,	*/
+/* then automatically accept it as the best possible match and leave.	*/
+
+		lea	(%edi,%edx), %eax
+		movl	scan(%esp), %edi
+		subl	%edi, %eax
+		cmpl	$MAX_MATCH, %eax
+		jge	LenMaximum
+
+/* If the length of the match is not longer than the best match we	*/
+/* have so far, then forget it and return to the lookup loop.		*/
+
+		movl	deflatestate(%esp), %edx
+		movl	bestlen(%esp), %ebx
+		cmpl	%ebx, %eax
+		jg	LongerMatch
+		movl	windowbestlen(%esp), %esi
+		movl	dsPrev(%edx), %edi
+		movl	scanend(%esp), %ebx
+		movl	chainlenwmask(%esp), %edx
+		jmp	LookupLoop
+
+/*         s->match_start = cur_match;					*/
+/*         best_len = len;						*/
+/*         if (len >= nice_match) break;				*/
+/*         scan_end = *(ushf*)(scan+best_len-1);			*/
+
+LongerMatch:	movl	nicematch(%esp), %ebx
+		movl	%eax, bestlen(%esp)
+		movl	%ecx, dsMatchStart(%edx)
+		cmpl	%ebx, %eax
+		jge	LeaveNow
+		movl	window(%esp), %esi
+		addl	%eax, %esi
+		movl	%esi, windowbestlen(%esp)
+		movzwl	-1(%edi,%eax), %ebx
+		movl	dsPrev(%edx), %edi
+		movl	%ebx, scanend(%esp)
+		movl	chainlenwmask(%esp), %edx
+		jmp	LookupLoop
+
+/* Accept the current string, with the maximum possible length.		*/
+
+LenMaximum:	movl	deflatestate(%esp), %edx
+		movl	$MAX_MATCH, bestlen(%esp)
+		movl	%ecx, dsMatchStart(%edx)
+
+/* if ((uInt)best_len <= s->lookahead) return (uInt)best_len;		*/
+/* return s->lookahead;							*/
+
+LeaveNow:
+		movl	deflatestate(%esp), %edx
+		movl	bestlen(%esp), %ebx
+		movl	dsLookahead(%edx), %eax
+		cmpl	%eax, %ebx
+		jg	LookaheadRet
+		movl	%ebx, %eax
+LookaheadRet:
+
+/* Restore the stack and return from whence we came.			*/
+
+		addl	$LocalVarsSize, %esp
+		popl	%ebx
+		popl	%esi
+		popl	%edi
+		popl	%ebp
+match_init:	ret

data/dependencies/zlib/contrib/blast/Makefile

@@ -0,0 +1,8 @@
+blast: blast.c blast.h
+	cc -DTEST -o blast blast.c
+
+test: blast
+	blast < test.pk | cmp - test.txt
+
+clean:
+	rm -f blast blast.o

data/dependencies/zlib/contrib/blast/README

@@ -0,0 +1,4 @@
+Read blast.h for purpose and usage.
+
+Mark Adler
+madler@alumni.caltech.edu

data/dependencies/zlib/contrib/blast/blast.c

@@ -0,0 +1,444 @@
+/* blast.c
+ * Copyright (C) 2003 Mark Adler
+ * For conditions of distribution and use, see copyright notice in blast.h
+ * version 1.1, 16 Feb 2003
+ *
+ * blast.c decompresses data compressed by the PKWare Compression Library.
+ * This function provides functionality similar to the explode() function of
+ * the PKWare library, hence the name "blast".
+ *
+ * This decompressor is based on the excellent format description provided by
+ * Ben Rudiak-Gould in comp.compression on August 13, 2001.  Interestingly, the
+ * example Ben provided in the post is incorrect.  The distance 110001 should
+ * instead be 111000.  When corrected, the example byte stream becomes:
+ *
+ *    00 04 82 24 25 8f 80 7f
+ *
+ * which decompresses to "AIAIAIAIAIAIA" (without the quotes).
+ */
+
+/*
+ * Change history:
+ *
+ * 1.0  12 Feb 2003     - First version
+ * 1.1  16 Feb 2003     - Fixed distance check for > 4 GB uncompressed data
+ */
+
+#include <setjmp.h>             /* for setjmp(), longjmp(), and jmp_buf */
+#include "blast.h"              /* prototype for blast() */
+
+#define local static            /* for local function definitions */
+#define MAXBITS 13              /* maximum code length */
+#define MAXWIN 4096             /* maximum window size */
+
+/* input and output state */
+struct state {
+    /* input state */
+    blast_in infun;             /* input function provided by user */
+    void *inhow;                /* opaque information passed to infun() */
+    unsigned char *in;          /* next input location */
+    unsigned left;              /* available input at in */
+    int bitbuf;                 /* bit buffer */
+    int bitcnt;                 /* number of bits in bit buffer */
+
+    /* input limit error return state for bits() and decode() */
+    jmp_buf env;
+
+    /* output state */
+    blast_out outfun;           /* output function provided by user */
+    void *outhow;               /* opaque information passed to outfun() */
+    unsigned next;              /* index of next write location in out[] */
+    int first;                  /* true to check distances (for first 4K) */
+    unsigned char out[MAXWIN];  /* output buffer and sliding window */
+};
+
+/*
+ * Return need bits from the input stream.  This always leaves less than
+ * eight bits in the buffer.  bits() works properly for need == 0.
+ *
+ * Format notes:
+ *
+ * - Bits are stored in bytes from the least significant bit to the most
+ *   significant bit.  Therefore bits are dropped from the bottom of the bit
+ *   buffer, using shift right, and new bytes are appended to the top of the
+ *   bit buffer, using shift left.
+ */
+local int bits(struct state *s, int need)
+{
+    int val;            /* bit accumulator */
+
+    /* load at least need bits into val */
+    val = s->bitbuf;
+    while (s->bitcnt < need) {
+        if (s->left == 0) {
+            s->left = s->infun(s->inhow, &(s->in));
+            if (s->left == 0) longjmp(s->env, 1);       /* out of input */
+        }
+        val |= (int)(*(s->in)++) << s->bitcnt;          /* load eight bits */
+        s->left--;
+        s->bitcnt += 8;
+    }
+
+    /* drop need bits and update buffer, always zero to seven bits left */
+    s->bitbuf = val >> need;
+    s->bitcnt -= need;
+
+    /* return need bits, zeroing the bits above that */
+    return val & ((1 << need) - 1);
+}
+
+/*
+ * Huffman code decoding tables.  count[1..MAXBITS] is the number of symbols of
+ * each length, which for a canonical code are stepped through in order.
+ * symbol[] are the symbol values in canonical order, where the number of
+ * entries is the sum of the counts in count[].  The decoding process can be
+ * seen in the function decode() below.
+ */
+struct huffman {
+    short *count;       /* number of symbols of each length */
+    short *symbol;      /* canonically ordered symbols */
+};
+
+/*
+ * Decode a code from the stream s using huffman table h.  Return the symbol or
+ * a negative value if there is an error.  If all of the lengths are zero, i.e.
+ * an empty code, or if the code is incomplete and an invalid code is received,
+ * then -9 is returned after reading MAXBITS bits.
+ *
+ * Format notes:
+ *
+ * - The codes as stored in the compressed data are bit-reversed relative to
+ *   a simple integer ordering of codes of the same lengths.  Hence below the
+ *   bits are pulled from the compressed data one at a time and used to
+ *   build the code value reversed from what is in the stream in order to
+ *   permit simple integer comparisons for decoding.
+ *
+ * - The first code for the shortest length is all ones.  Subsequent codes of
+ *   the same length are simply integer decrements of the previous code.  When
+ *   moving up a length, a one bit is appended to the code.  For a complete
+ *   code, the last code of the longest length will be all zeros.  To support
+ *   this ordering, the bits pulled during decoding are inverted to apply the
+ *   more "natural" ordering starting with all zeros and incrementing.
+ */
+local int decode(struct state *s, struct huffman *h)
+{
+    int len;            /* current number of bits in code */
+    int code;           /* len bits being decoded */
+    int first;          /* first code of length len */
+    int count;          /* number of codes of length len */
+    int index;          /* index of first code of length len in symbol table */
+    int bitbuf;         /* bits from stream */
+    int left;           /* bits left in next or left to process */
+    short *next;        /* next number of codes */
+
+    bitbuf = s->bitbuf;
+    left = s->bitcnt;
+    code = first = index = 0;
+    len = 1;
+    next = h->count + 1;
+    while (1) {
+        while (left--) {
+            code |= (bitbuf & 1) ^ 1;   /* invert code */
+            bitbuf >>= 1;
+            count = *next++;
+            if (code < first + count) { /* if length len, return symbol */
+                s->bitbuf = bitbuf;
+                s->bitcnt = (s->bitcnt - len) & 7;
+                return h->symbol[index + (code - first)];
+            }
+            index += count;             /* else update for next length */
+            first += count;
+            first <<= 1;
+            code <<= 1;
+            len++;
+        }
+        left = (MAXBITS+1) - len;
+        if (left == 0) break;
+        if (s->left == 0) {
+            s->left = s->infun(s->inhow, &(s->in));
+            if (s->left == 0) longjmp(s->env, 1);       /* out of input */
+        }
+        bitbuf = *(s->in)++;
+        s->left--;
+        if (left > 8) left = 8;
+    }
+    return -9;                          /* ran out of codes */
+}
+
+/*
+ * Given a list of repeated code lengths rep[0..n-1], where each byte is a
+ * count (high four bits + 1) and a code length (low four bits), generate the
+ * list of code lengths.  This compaction reduces the size of the object code.
+ * Then given the list of code lengths length[0..n-1] representing a canonical
+ * Huffman code for n symbols, construct the tables required to decode those
+ * codes.  Those tables are the number of codes of each length, and the symbols
+ * sorted by length, retaining their original order within each length.  The
+ * return value is zero for a complete code set, negative for an over-
+ * subscribed code set, and positive for an incomplete code set.  The tables
+ * can be used if the return value is zero or positive, but they cannot be used
+ * if the return value is negative.  If the return value is zero, it is not
+ * possible for decode() using that table to return an error--any stream of
+ * enough bits will resolve to a symbol.  If the return value is positive, then
+ * it is possible for decode() using that table to return an error for received
+ * codes past the end of the incomplete lengths.
+ */
+local int construct(struct huffman *h, const unsigned char *rep, int n)
+{
+    int symbol;         /* current symbol when stepping through length[] */
+    int len;            /* current length when stepping through h->count[] */
+    int left;           /* number of possible codes left of current length */
+    short offs[MAXBITS+1];      /* offsets in symbol table for each length */
+    short length[256];  /* code lengths */
+
+    /* convert compact repeat counts into symbol bit length list */
+    symbol = 0;
+    do {
+        len = *rep++;
+        left = (len >> 4) + 1;
+        len &= 15;
+        do {
+            length[symbol++] = len;
+        } while (--left);
+    } while (--n);
+    n = symbol;
+
+    /* count number of codes of each length */
+    for (len = 0; len <= MAXBITS; len++)
+        h->count[len] = 0;
+    for (symbol = 0; symbol < n; symbol++)
+        (h->count[length[symbol]])++;   /* assumes lengths are within bounds */
+    if (h->count[0] == n)               /* no codes! */
+        return 0;                       /* complete, but decode() will fail */
+
+    /* check for an over-subscribed or incomplete set of lengths */
+    left = 1;                           /* one possible code of zero length */
+    for (len = 1; len <= MAXBITS; len++) {
+        left <<= 1;                     /* one more bit, double codes left */
+        left -= h->count[len];          /* deduct count from possible codes */
+        if (left < 0) return left;      /* over-subscribed--return negative */
+    }                                   /* left > 0 means incomplete */
+
+    /* generate offsets into symbol table for each length for sorting */
+    offs[1] = 0;
+    for (len = 1; len < MAXBITS; len++)
+        offs[len + 1] = offs[len] + h->count[len];
+
+    /*
+     * put symbols in table sorted by length, by symbol order within each
+     * length
+     */
+    for (symbol = 0; symbol < n; symbol++)
+        if (length[symbol] != 0)
+            h->symbol[offs[length[symbol]]++] = symbol;
+
+    /* return zero for complete set, positive for incomplete set */
+    return left;
+}
+
+/*
+ * Decode PKWare Compression Library stream.
+ *
+ * Format notes:
+ *
+ * - First byte is 0 if literals are uncoded or 1 if they are coded.  Second
+ *   byte is 4, 5, or 6 for the number of extra bits in the distance code.
+ *   This is the base-2 logarithm of the dictionary size minus six.
+ *
+ * - Compressed data is a combination of literals and length/distance pairs
+ *   terminated by an end code.  Literals are either Huffman coded or
+ *   uncoded bytes.  A length/distance pair is a coded length followed by a
+ *   coded distance to represent a string that occurs earlier in the
+ *   uncompressed data that occurs again at the current location.
+ *
+ * - A bit preceding a literal or length/distance pair indicates which comes
+ *   next, 0 for literals, 1 for length/distance.
+ *
+ * - If literals are uncoded, then the next eight bits are the literal, in the
+ *   normal bit order in th stream, i.e. no bit-reversal is needed. Similarly,
+ *   no bit reversal is needed for either the length extra bits or the distance
+ *   extra bits.
+ *
+ * - Literal bytes are simply written to the output.  A length/distance pair is
+ *   an instruction to copy previously uncompressed bytes to the output.  The
+ *   copy is from distance bytes back in the output stream, copying for length
+ *   bytes.
+ *
+ * - Distances pointing before the beginning of the output data are not
+ *   permitted.
+ *
+ * - Overlapped copies, where the length is greater than the distance, are
+ *   allowed and common.  For example, a distance of one and a length of 518
+ *   simply copies the last byte 518 times.  A distance of four and a length of
+ *   twelve copies the last four bytes three times.  A simple forward copy
+ *   ignoring whether the length is greater than the distance or not implements
+ *   this correctly.
+ */
+local int decomp(struct state *s)
+{
+    int lit;            /* true if literals are coded */
+    int dict;           /* log2(dictionary size) - 6 */
+    int symbol;         /* decoded symbol, extra bits for distance */
+    int len;            /* length for copy */
+    int dist;           /* distance for copy */
+    int copy;           /* copy counter */
+    unsigned char *from, *to;   /* copy pointers */
+    static int virgin = 1;                              /* build tables once */
+    static short litcnt[MAXBITS+1], litsym[256];        /* litcode memory */
+    static short lencnt[MAXBITS+1], lensym[16];         /* lencode memory */
+    static short distcnt[MAXBITS+1], distsym[64];       /* distcode memory */
+    static struct huffman litcode = {litcnt, litsym};   /* length code */
+    static struct huffman lencode = {lencnt, lensym};   /* length code */
+    static struct huffman distcode = {distcnt, distsym};/* distance code */
+        /* bit lengths of literal codes */
+    static const unsigned char litlen[] = {
+        11, 124, 8, 7, 28, 7, 188, 13, 76, 4, 10, 8, 12, 10, 12, 10, 8, 23, 8,
+        9, 7, 6, 7, 8, 7, 6, 55, 8, 23, 24, 12, 11, 7, 9, 11, 12, 6, 7, 22, 5,
+        7, 24, 6, 11, 9, 6, 7, 22, 7, 11, 38, 7, 9, 8, 25, 11, 8, 11, 9, 12,
+        8, 12, 5, 38, 5, 38, 5, 11, 7, 5, 6, 21, 6, 10, 53, 8, 7, 24, 10, 27,
+        44, 253, 253, 253, 252, 252, 252, 13, 12, 45, 12, 45, 12, 61, 12, 45,
+        44, 173};
+        /* bit lengths of length codes 0..15 */
+    static const unsigned char lenlen[] = {2, 35, 36, 53, 38, 23};
+        /* bit lengths of distance codes 0..63 */
+    static const unsigned char distlen[] = {2, 20, 53, 230, 247, 151, 248};
+    static const short base[16] = {     /* base for length codes */
+        3, 2, 4, 5, 6, 7, 8, 9, 10, 12, 16, 24, 40, 72, 136, 264};
+    static const char extra[16] = {     /* extra bits for length codes */
+        0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8};
+
+    /* set up decoding tables (once--might not be thread-safe) */
+    if (virgin) {
+        construct(&litcode, litlen, sizeof(litlen));
+        construct(&lencode, lenlen, sizeof(lenlen));
+        construct(&distcode, distlen, sizeof(distlen));
+        virgin = 0;
+    }
+
+    /* read header */
+    lit = bits(s, 8);
+    if (lit > 1) return -1;
+    dict = bits(s, 8);
+    if (dict < 4 || dict > 6) return -2;
+
+    /* decode literals and length/distance pairs */
+    do {
+        if (bits(s, 1)) {
+            /* get length */
+            symbol = decode(s, &lencode);
+            len = base[symbol] + bits(s, extra[symbol]);
+            if (len == 519) break;              /* end code */
+
+            /* get distance */
+            symbol = len == 2 ? 2 : dict;
+            dist = decode(s, &distcode) << symbol;
+            dist += bits(s, symbol);
+            dist++;
+            if (s->first && dist > s->next)
+                return -3;              /* distance too far back */
+
+            /* copy length bytes from distance bytes back */
+            do {
+                to = s->out + s->next;
+                from = to - dist;
+                copy = MAXWIN;
+                if (s->next < dist) {
+                    from += copy;
+                    copy = dist;
+                }
+                copy -= s->next;
+                if (copy > len) copy = len;
+                len -= copy;
+                s->next += copy;
+                do {
+                    *to++ = *from++;
+                } while (--copy);
+                if (s->next == MAXWIN) {
+                    if (s->outfun(s->outhow, s->out, s->next)) return 1;
+                    s->next = 0;
+                    s->first = 0;
+                }
+            } while (len != 0);
+        }
+        else {
+            /* get literal and write it */
+            symbol = lit ? decode(s, &litcode) : bits(s, 8);
+            s->out[s->next++] = symbol;
+            if (s->next == MAXWIN) {
+                if (s->outfun(s->outhow, s->out, s->next)) return 1;
+                s->next = 0;
+                s->first = 0;
+            }
+        }
+    } while (1);
+    return 0;
+}
+
+/* See comments in blast.h */
+int blast(blast_in infun, void *inhow, blast_out outfun, void *outhow)
+{
+    struct state s;             /* input/output state */
+    int err;                    /* return value */
+
+    /* initialize input state */
+    s.infun = infun;
+    s.inhow = inhow;
+    s.left = 0;
+    s.bitbuf = 0;
+    s.bitcnt = 0;
+
+    /* initialize output state */
+    s.outfun = outfun;
+    s.outhow = outhow;
+    s.next = 0;
+    s.first = 1;
+
+    /* return if bits() or decode() tries to read past available input */
+    if (setjmp(s.env) != 0)             /* if came back here via longjmp(), */
+        err = 2;                        /*  then skip decomp(), return error */
+    else
+        err = decomp(&s);               /* decompress */
+
+    /* write any leftover output and update the error code if needed */
+    if (err != 1 && s.next && s.outfun(s.outhow, s.out, s.next) && err == 0)
+        err = 1;
+    return err;
+}
+
+#ifdef TEST
+/* Example of how to use blast() */
+#include <stdio.h>
+#include <stdlib.h>
+
+#define CHUNK 16384
+
+local unsigned inf(void *how, unsigned char **buf)
+{
+    static unsigned char hold[CHUNK];
+
+    *buf = hold;
+    return fread(hold, 1, CHUNK, (FILE *)how);
+}
+
+local int outf(void *how, unsigned char *buf, unsigned len)
+{
+    return fwrite(buf, 1, len, (FILE *)how) != len;
+}
+
+/* Decompress a PKWare Compression Library stream from stdin to stdout */
+int main(void)
+{
+    int ret, n;
+
+    /* decompress to stdout */
+    ret = blast(inf, stdin, outf, stdout);
+    if (ret != 0) fprintf(stderr, "blast error: %d\n", ret);
+
+    /* see if there are any leftover bytes */
+    n = 0;
+    while (getchar() != EOF) n++;
+    if (n) fprintf(stderr, "blast warning: %d unused bytes of input\n", n);
+
+    /* return blast() error code */
+    return ret;
+}
+#endif