From 7c2a874e50b871d04fbd19501f7b42cff55e5abc Mon Sep 17 00:00:00 2001 From: Mark Adler Date: Fri, 9 Sep 2011 23:21:47 -0700 Subject: zlib 1.2.0 --- contrib/README.contrib | 10 +- contrib/blast/Makefile | 8 + contrib/blast/README | 4 + contrib/blast/blast.c | 444 ++++++++++++++++++ contrib/blast/blast.h | 71 +++ contrib/blast/test.pk | Bin 0 -> 8 bytes contrib/blast/test.txt | 1 + contrib/inflate86/inffast.S | 1095 +++++++++++++++++++++++++++++++++++++++++++ contrib/puff/Makefile | 8 + contrib/puff/README | 63 +++ contrib/puff/puff.c | 833 ++++++++++++++++++++++++++++++++ contrib/puff/puff.h | 31 ++ contrib/puff/zeros.raw | Bin 0 -> 1213 bytes 13 files changed, 2566 insertions(+), 2 deletions(-) create mode 100644 contrib/blast/Makefile create mode 100644 contrib/blast/README create mode 100644 contrib/blast/blast.c create mode 100644 contrib/blast/blast.h create mode 100644 contrib/blast/test.pk create mode 100644 contrib/blast/test.txt create mode 100644 contrib/inflate86/inffast.S create mode 100644 contrib/puff/Makefile create mode 100644 contrib/puff/README create mode 100644 contrib/puff/puff.c create mode 100644 contrib/puff/puff.h create mode 100644 contrib/puff/zeros.raw (limited to 'contrib') diff --git a/contrib/README.contrib b/contrib/README.contrib index 7ad191c..fcee020 100644 --- a/contrib/README.contrib +++ b/contrib/README.contrib @@ -11,12 +11,18 @@ asm586/ and asm686/ by Brian Raiter asm code for Pentium and Pentium Pro See http://www.muppetlabs.com/~breadbox/software/assembly.html -delphi/ by Bob Dellaca +blast/ by Mark Adler + Decompressor for output of PKWare Data Compression Library + +delphi/ by Bob Dellaca Support for Delphi -delphi2/ by Davide Moretti +delphi2/ by Davide Moretti Another support for C++Builder and Delphi +inflate86/ by Chris Anderson + Tuned x86 gcc asm code to replace inflate_fast() + minizip/ by Gilles Vollant Mini zip and unzip based on zlib See http://www.winimage.com/zLibDll/unzip.html diff --git a/contrib/blast/Makefile b/contrib/blast/Makefile new file mode 100644 index 0000000..9be80ba --- /dev/null +++ b/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 diff --git a/contrib/blast/README b/contrib/blast/README new file mode 100644 index 0000000..e3a60b3 --- /dev/null +++ b/contrib/blast/README @@ -0,0 +1,4 @@ +Read blast.h for purpose and usage. + +Mark Adler +madler@alumni.caltech.edu diff --git a/contrib/blast/blast.c b/contrib/blast/blast.c new file mode 100644 index 0000000..67dab4e --- /dev/null +++ b/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 /* 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 +#include + +#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 diff --git a/contrib/blast/blast.h b/contrib/blast/blast.h new file mode 100644 index 0000000..2417837 --- /dev/null +++ b/contrib/blast/blast.h @@ -0,0 +1,71 @@ +/* blast.h -- interface for blast.c + Copyright (C) 2003 Mark Adler + version 1.1, 16 Feb 2003 + + This software is provided 'as-is', without any express or implied + warranty. In no event will the author be held liable for any damages + arising from the use of this software. + + Permission is granted to anyone to use this software for any purpose, + including commercial applications, and to alter it and redistribute it + freely, subject to the following restrictions: + + 1. The origin of this software must not be misrepresented; you must not + claim that you wrote the original software. If you use this software + in a product, an acknowledgment in the product documentation would be + appreciated but is not required. + 2. Altered source versions must be plainly marked as such, and must not be + misrepresented as being the original software. + 3. This notice may not be removed or altered from any source distribution. + + Mark Adler madler@alumni.caltech.edu + */ + + +/* + * blast() decompresses the PKWare Data Compression Library (DCL) compressed + * format. It provides the same functionality as the explode() function in + * that library. (Note: PKWare overused the "implode" verb, and the format + * used by their library implode() function is completely different and + * incompatible with the implode compression method supported by PKZIP.) + */ + + +typedef unsigned (*blast_in)(void *how, unsigned char **buf); +typedef int (*blast_out)(void *how, unsigned char *buf, unsigned len); +/* Definitions for input/output functions passed to blast(). See below for + * what the provided functions need to do. + */ + + +int blast(blast_in infun, void *inhow, blast_out outfun, void *outhow); +/* Decompress input to output using the provided infun() and outfun() calls. + * On success, the return value of blast() is zero. If there is an error in + * the source data, i.e. it is not in the proper format, then a negative value + * is returned. If there is not enough input available or there is not enough + * output space, then a positive error is returned. + * + * The input function is invoked: len = infun(how, &buf), where buf is set by + * infun() to point to the input buffer, and infun() returns the number of + * available bytes there. If infun() returns zero, then blast() returns with + * an input error. (blast() only asks for input if it needs it.) inhow is for + * use by the application to pass an input descriptor to infun(), if desired. + * + * The output function is invoked: err = outfun(how, buf, len), where the bytes + * to be written are buf[0..len-1]. If err is not zero, then blast() returns + * with an output error. outfun() is always called with len <= 4096. outhow + * is for use by the application to pass an output descriptor to outfun(), if + * desired. + * + * The return codes are: + * + * 2: ran out of input before completing decompression + * 1: output error before completing decompression + * 0: successful decompression + * -1: literal flag not zero or one + * -2: dictionary size not in 4..6 + * -3: distance is too far back + * + * At the bottom of blast.c is an example program that uses blast() that can be + * compiled to produce a command-line decompression filter by defining TEST. + */ diff --git a/contrib/blast/test.pk b/contrib/blast/test.pk new file mode 100644 index 0000000..be10b2b Binary files /dev/null and b/contrib/blast/test.pk differ diff --git a/contrib/blast/test.txt b/contrib/blast/test.txt new file mode 100644 index 0000000..bfdf1c5 --- /dev/null +++ b/contrib/blast/test.txt @@ -0,0 +1 @@ +AIAIAIAIAIAIA \ No newline at end of file diff --git a/contrib/inflate86/inffast.S b/contrib/inflate86/inffast.S new file mode 100644 index 0000000..d1e80ef --- /dev/null +++ b/contrib/inflate86/inffast.S @@ -0,0 +1,1095 @@ +/* + * inffast.S is a hand tuned assembler version of: + * + * inffast.c -- fast decoding + * Copyright (C) 1995-2003 Mark Adler + * For conditions of distribution and use, see copyright notice in zlib.h + * + * Copyright (C) 2003 Chris Anderson + * Please use the copyright conditions above. + * + * This version (Jan-23-2003) of inflate_fast was coded and tested under + * GNU/Linux on a pentium 3, using the gcc-3.2 compiler distribution. On that + * machine, I found that gzip style archives decompressed about 20% faster than + * the gcc-3.2 -O3 -fomit-frame-pointer compiled version. Your results will + * depend on how large of a buffer is used for z_stream.next_in & next_out + * (8K-32K worked best for my 256K cpu cache) and how much overhead there is in + * stream processing I/O and crc32/addler32. In my case, this routine used + * 70% of the cpu time and crc32 used 20%. + * + * I am confident that this version will work in the general case, but I have + * not tested a wide variety of datasets or a wide variety of platforms. + * + * Jan-24-2003 -- Added -DUSE_MMX define for slightly faster inflating. + * It should be a runtime flag instead of compile time flag... + */ + +.file "inffast.S" + +.globl inflate_fast + +.text +.align 4,0 +.L_invalid_literal_length_code_msg: +.string "invalid literal/length code" + +.align 4,0 +.L_invalid_distance_code_msg: +.string "invalid distance code" + +.align 4,0 +.L_invalid_distance_too_far_msg: +.string "invalid distance too far back" + +#if defined( USE_MMX ) +.align 4,0 +.L_mask: /* mask[N] = ( 1 << N ) - 1 */ +.long 0 +.long 1 +.long 3 +.long 7 +.long 15 +.long 31 +.long 63 +.long 127 +.long 255 +.long 511 +.long 1023 +.long 2047 +.long 4095 +.long 8191 +.long 16383 +.long 32767 +.long 65535 +.long 131071 +.long 262143 +.long 524287 +.long 1048575 +.long 2097151 +.long 4194303 +.long 8388607 +.long 16777215 +.long 33554431 +.long 67108863 +.long 134217727 +.long 268435455 +.long 536870911 +.long 1073741823 +.long 2147483647 +.long 4294967295 +#endif + +.text + +/* + * struct z_stream offsets, in zlib.h + */ +#define next_in_strm 0 /* strm->next_in */ +#define avail_in_strm 4 /* strm->avail_in */ +#define next_out_strm 12 /* strm->next_out */ +#define avail_out_strm 16 /* strm->avail_out */ +#define msg_strm 24 /* strm->msg */ +#define state_strm 28 /* strm->state */ + +/* + * struct inflate_state offsets, in inflate.h + */ +#define mode_state 0 /* state->mode */ +#define wsize_state 32 /* state->wsize */ +#define write_state 36 /* state->write */ +#define window_state 40 /* state->window */ +#define hold_state 44 /* state->hold */ +#define bits_state 48 /* state->bits */ +#define lencode_state 64 /* state->lencode */ +#define distcode_state 68 /* state->distcode */ +#define lenbits_state 72 /* state->lenbits */ +#define distbits_state 76 /* state->distbits */ + +/* + * inflate_fast's activation record + */ +#define local_var_size 56 /* how much local space for vars */ +#define strm_sp 80 /* first arg: z_stream * (local_var_size + 24) */ +#define start_sp 84 /* second arg: unsigned int (local_var_size + 28) */ + +/* + * offsets for local vars on stack + */ +#define out 52 /* unsigned char* */ +#define window 48 /* unsigned char* */ +#define wsize 44 /* unsigned int */ +#define write 40 /* unsigned int */ +#define in 36 /* unsigned char* */ +#define beg 32 /* unsigned char* */ +#define dist 28 /* unsigned int */ +#define len 24 /* unsigned int */ +#define last 20 /* unsigned char* */ +#define end 16 /* unsigned char* */ +#define dcode 12 /* code* */ +#define lcode 8 /* code* */ +#define dmask 4 /* unsigned int */ +#define lmask 0 /* unsigned int */ + +/* + * typedef enum inflate_mode consts, in inflate.h + */ +#ifndef NO_GUNZIP +#define GUNZIP +#endif + +#ifdef GUNZIP +#define INFLATE_MODE_TYPE 11 /* state->mode flags enum-ed in inflate.h */ +#define INFLATE_MODE_BAD 26 +#else +#define INFLATE_MODE_TYPE 3 +#define INFLATE_MODE_BAD 17 +#endif + + +.align 16,0x90 +inflate_fast: + pushl %edi + pushl %esi + pushl %ebp + pushl %ebx + pushf /* save eflags (strm_sp, state_sp assumes this is 32 bits) */ + subl $local_var_size, %esp + cld +#if defined( USE_MMX ) + emms +#endif + +#define strm_r %esi +#define state_r %edi + + movl strm_sp(%esp), strm_r + movl state_strm(strm_r), state_r + + /* in = strm->next_in; + * out = strm->next_out; + * last = in + strm->avail_in - 5; + * beg = out - (start - strm->avail_out); + * end = out + (strm->avail_out - 257); + */ + movl next_in_strm(strm_r), %eax + movl next_out_strm(strm_r), %ebx + movl avail_in_strm(strm_r), %edx + movl avail_out_strm(strm_r), %ecx + movl start_sp(%esp), %ebp + + addl %eax, %edx /* avail_in += next_in */ + subl $5, %edx /* avail_in -= 5 */ + + subl %ecx, %ebp /* start -= avail_out */ + negl %ebp /* start = -start */ + addl %ebx, %ebp /* start += next_out */ + + subl $257, %ecx /* avail_out -= 257 */ + addl %ebx, %ecx /* avail_out += out */ + + movl %eax, in(%esp) + movl %ebx, out(%esp) + movl %edx, last(%esp) + movl %ebp, beg(%esp) + movl %ecx, end(%esp) + + /* wsize = state->wsize; + * write = state->write; + * window = state->window; + * hold = state->hold; + * bits = state->bits; + * lcode = state->lencode; + * dcode = state->distcode; + * lmask = ( 1 << state->lenbits ) - 1; + * dmask = ( 1 << state->distbits ) - 1; + */ + + movl lencode_state(state_r), %eax + movl distcode_state(state_r), %ecx + + movl %eax, lcode(%esp) + movl %ecx, dcode(%esp) + + movl $1, %eax + movl lenbits_state(state_r), %ecx + shll %cl, %eax + decl %eax + movl %eax, lmask(%esp) + + movl $1, %eax + movl distbits_state(state_r), %ecx + shll %cl, %eax + decl %eax + movl %eax, dmask(%esp) + + movl wsize_state(state_r), %eax + movl write_state(state_r), %ecx + movl window_state(state_r), %edx + + movl %eax, wsize(%esp) + movl %ecx, write(%esp) + movl %edx, window(%esp) + +#if ! defined( USE_MMX ) + +#define hold_r %ebp +#define bits_r %bl +#define bitslong_r %ebx + + movl hold_state(state_r), hold_r + movl bits_state(state_r), bitslong_r + +#else /* USE_MMX */ + +#define hold_mm %mm0 +#define bits_r %ebp +#define bitslong_r %ebp + + movl hold_state(state_r), %ebx + movl bits_state(state_r), bitslong_r + +#endif + +#undef strm_r +#undef state_r +#define in_r %esi +#define from_r %esi +#define out_r %edi + + movl in(%esp), in_r + +#if ! defined ( USE_MMX ) + + /* align in_r on word boundary */ + testl $1, in_r + jz .L_is_aligned + xorl %eax, %eax + movb (in_r), %al + incl in_r + movb bits_r, %cl + addb $8, bits_r + shll %cl, %eax + orl %eax, hold_r + +#else + /* align in_r on long boundary */ +.L_align_long: + testl $3, in_r + jz .L_is_aligned + xorl %eax, %eax + movb (in_r), %al + incl in_r + movl bits_r, %ecx + addl $8, bits_r + shll %cl, %eax + orl %eax, %ebx + jmp .L_align_long + +#endif + +.L_is_aligned: + movl out(%esp), out_r + +#if defined ( USE_MMX ) + +#define used_mm %mm1 +#define dmask2_mm %mm2 +#define lmask2_mm %mm3 +#define lmask_mm %mm4 +#define dmask_mm %mm5 +#define tmp_mm %mm6 + + movl out(%esp), out_r + movd lmask(%esp), lmask_mm + movq lmask_mm, lmask2_mm + movd dmask(%esp), dmask_mm + movq dmask_mm, dmask2_mm + movd %ebx, hold_mm + pxor used_mm, used_mm + movl lcode(%esp), %ebx /* ebx = lcode */ +#endif + + jmp .L_do_loop + +.align 16,0x90 + +#if ! defined ( USE_MMX ) + +.L_do_loop: + /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out + * + * do { + * if (bits < 15) { + * hold |= *((unsigned short *)in)++ << bits; + * bits += 16 + * } + * this = lcode[hold & lmask] + */ + cmpb $15, bits_r + ja .L_get_length_code /* if (15 < bits) */ + + xorl %eax, %eax + lodsw /* al = *(ushort *)in++ */ + movb bits_r, %cl /* cl = bits, needs it for shifting */ + addb $16, bits_r /* bits += 16 */ + shll %cl, %eax + orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ + +.L_get_length_code: + movl lmask(%esp), %edx /* edx = lmask */ + movl lcode(%esp), %ecx /* ecx = lcode */ + andl hold_r, %edx /* edx &= hold */ + movl (%ecx,%edx,4), %eax /* eax = lcode[hold & lmask] */ + +#else /* USE_MMX */ + +.L_do_loop: + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + + cmpl $32, bits_r + ja .L_get_length_code /* if (32 < bits) */ + + movd bits_r, tmp_mm + movd (in_r), %mm7 + addl $4, in_r + psllq tmp_mm, %mm7 + addl $32, bits_r + por %mm7, hold_mm /* hold_mm |= *((uint *)in)++ << bits */ + +.L_get_length_code: + pand hold_mm, lmask_mm + movd lmask_mm, %eax + movq lmask2_mm, lmask_mm + movl (%ebx,%eax,4), %eax /* eax = lcode[hold & lmask] */ + +#endif + +#if ! defined( USE_MMX ) + +.L_dolen: + /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out + * + * dolen: + * bits -= this.bits; + * hold >>= this.bits + */ + movb %ah, %cl /* cl = this.bits */ + subb %ah, bits_r /* bits -= this.bits */ + shrl %cl, hold_r /* hold >>= this.bits */ + + /* check if op is a literal + * if (op == 0) { + * PUP(out) = this.val; + * } + */ + testb %al, %al + jnz .L_test_for_length_base /* if (op != 0) 45.7% */ + + shrl $16, %eax /* output this.val char */ + stosb + +#else /* USE_MMX */ + +#define len_r %edx + +.L_dolen: + movzbl %ah, %ecx /* ecx = this.bits */ + movl %eax, len_r /* len = this */ + shrl $16, len_r /* len = this.val */ + movd %ecx, used_mm + subl %ecx, bits_r /* bits -= this.bits */ + + testb %al, %al + jnz .L_test_for_length_base /* if (op != 0) 45.7% */ + + movb %dl, (out_r) + incl out_r + +#endif + +.L_while_test: + /* while (in < last && out < end) + */ + cmpl out_r, end(%esp) + jbe .L_break_loop /* if (out >= end) */ + + cmpl in_r, last(%esp) + ja .L_do_loop /* if (in < last) */ + jmp .L_break_loop + +#if ! defined( USE_MMX ) + +.L_test_for_length_base: + /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out, %edx = len + * + * else if (op & 16) { + * len = this.val + * op &= 15 + * if (op) { + * if (op > bits) { + * hold |= *((unsigned short *)in)++ << bits; + * bits += 16 + * } + * len += hold & mask[op]; + * bits -= op; + * hold >>= op; + * } + */ +#define len_r %edx + movl %eax, len_r /* len = this */ + shrl $16, len_r /* len = this.val */ + movb %al, %cl + + testb $16, %al + jz .L_test_for_second_level_length /* if ((op & 16) == 0) 8% */ + andb $15, %cl /* op &= 15 */ + jz .L_save_len /* if (!op) */ + cmpb %cl, bits_r + jae .L_add_bits_to_len /* if (op <= bits) */ + + movb %cl, %ch /* stash op in ch, freeing cl */ + xorl %eax, %eax + lodsw /* al = *(ushort *)in++ */ + movb bits_r, %cl /* cl = bits, needs it for shifting */ + addb $16, bits_r /* bits += 16 */ + shll %cl, %eax + orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ + movb %ch, %cl /* move op back to ecx */ + +.L_add_bits_to_len: + movl $1, %eax + shll %cl, %eax + decl %eax + subb %cl, bits_r + andl hold_r, %eax /* eax &= hold */ + shrl %cl, hold_r + addl %eax, len_r /* len += hold & mask[op] */ + +.L_save_len: + movl len_r, len(%esp) /* save len */ +#undef len_r + +.L_decode_distance: + /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * + * if (bits < 15) { + * hold |= *((unsigned short *)in)++ << bits; + * bits += 16 + * } + * this = dcode[hold & dmask]; + * dodist: + * bits -= this.bits; + * hold >>= this.bits; + * op = this.op; + */ + + cmpb $15, bits_r + ja .L_get_distance_code /* if (15 < bits) */ + + xorl %eax, %eax + lodsw /* al = *(ushort *)in++ */ + movb bits_r, %cl /* cl = bits, needs it for shifting */ + addb $16, bits_r /* bits += 16 */ + shll %cl, %eax + orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ + +.L_get_distance_code: + movl dmask(%esp), %edx /* edx = dmask */ + movl dcode(%esp), %ecx /* ecx = dcode */ + andl hold_r, %edx /* edx &= hold */ + movl (%ecx,%edx,4), %eax /* eax = dcode[hold & dmask] */ + +#else /* USE_MMX */ + +.L_test_for_length_base: + testb $16, %al + jz .L_test_for_second_level_length /* if ((op & 16) == 0) 8% */ + andl $15, %eax /* op &= 15 */ + jz .L_decode_distance /* if (!op) */ + + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + movd %eax, used_mm + movd hold_mm, %ecx + subl %eax, bits_r + andl .L_mask(,%eax,4), %ecx + addl %ecx, len_r /* len += hold & mask[op] */ + +.L_decode_distance: + + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + + cmpl $32, bits_r + ja .L_get_dist_code /* if (32 < bits) */ + + movd bits_r, tmp_mm + movd (in_r), %mm7 + addl $4, in_r + psllq tmp_mm, %mm7 + addl $32, bits_r + por %mm7, hold_mm /* hold_mm |= *((uint *)in)++ << bits */ + +.L_get_dist_code: + movl dcode(%esp), %ebx /* ebx = dcode */ + pand hold_mm, dmask_mm + movd dmask_mm, %eax + movq dmask2_mm, dmask_mm + movl (%ebx,%eax,4), %eax /* eax = dcode[hold & lmask] */ + +#endif + +#if ! defined( USE_MMX ) + +#define dist_r %edx +.L_dodist: + movl %eax, dist_r /* dist = this */ + shrl $16, dist_r /* dist = this.val */ + movb %ah, %cl + subb %ah, bits_r /* bits -= this.bits */ + shrl %cl, hold_r /* hold >>= this.bits */ + + /* if (op & 16) { + * dist = this.val + * op &= 15 + * if (op > bits) { + * hold |= *((unsigned short *)in)++ << bits; + * bits += 16 + * } + * dist += hold & mask[op]; + * bits -= op; + * hold >>= op; + */ + movb %al, %cl /* cl = this.op */ + + testb $16, %al /* if ((op & 16) == 0) */ + jz .L_test_for_second_level_dist + andb $15, %cl /* op &= 15 */ + jz .L_check_dist_one + cmpb %cl, bits_r + jae .L_add_bits_to_dist /* if (op <= bits) 97.6% */ + + movb %cl, %ch /* stash op in ch, freeing cl */ + xorl %eax, %eax + lodsw /* al = *(ushort *)in++ */ + movb bits_r, %cl /* cl = bits, needs it for shifting */ + addb $16, bits_r /* bits += 16 */ + shll %cl, %eax + orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ + movb %ch, %cl /* move op back to ecx */ + +.L_add_bits_to_dist: + movl $1, %eax + shll %cl, %eax + decl %eax /* (1 << op) - 1 */ + subb %cl, bits_r + andl hold_r, %eax /* eax &= hold */ + shrl %cl, hold_r + addl %eax, dist_r /* dist += hold & ((1 << op) - 1) */ + jmp .L_check_window + +#else /* USE_MMX */ + +#define dist_r %ebx +.L_dodist: + movzbl %ah, %ecx /* ecx = this.bits */ + movl %eax, dist_r + shrl $16, dist_r /* dist = this.val */ + subl %ecx, bits_r /* bits -= this.bits */ + movd %ecx, used_mm + + testb $16, %al /* if ((op & 16) == 0) */ + jz .L_test_for_second_level_dist + andl $15, %eax /* op &= 15 */ + jz .L_check_dist_one + +.L_add_bits_to_dist: + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + movd %eax, used_mm /* save bit length of current op */ + movd hold_mm, %ecx /* get the next bits on input stream */ + subl %eax, bits_r /* bits -= op bits */ + andl .L_mask(,%eax,4), %ecx /* ecx = hold & mask[op] */ + addl %ecx, dist_r /* dist += hold & mask[op] */ + jmp .L_check_window + +#endif + +.align 16,0x90 + +.L_check_dist_one: + cmpl $1, dist_r + jne .L_check_window + cmpl out_r, beg(%esp) + je .L_check_window + + decl out_r +#if ! defined( USE_MMX ) + movl len(%esp), %ecx +#else + movl len_r, %ecx +#endif + movb (out_r), %al + subl $3, %ecx + + movb %al, 1(out_r) + movb %al, 2(out_r) + movb %al, 3(out_r) + addl $4, out_r + rep stosb + +#if defined( USE_MMX ) + movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ +#endif + jmp .L_while_test + +.align 16,0x90 + +.L_check_window: + /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * %ecx = nbytes + * + * nbytes = out - beg; + * if (dist <= nbytes) { + * from = out - dist; + * do { + * PUP(out) = PUP(from); + * } while (--len > 0) { + * } + */ + + movl in_r, in(%esp) /* save in so from can use it's reg */ + movl out_r, %eax + subl beg(%esp), %eax /* nbytes = out - beg */ + + cmpl dist_r, %eax + jb .L_clip_window /* if (dist > nbytes) 4.2% */ + +#if ! defined( USE_MMX ) + movl len(%esp), %ecx +#else + movl len_r, %ecx +#endif + movl out_r, from_r + subl dist_r, from_r /* from = out - dist */ + + subl $3, %ecx + movb (from_r), %al + movb %al, (out_r) + movb 1(from_r), %al + movb 2(from_r), %dl + addl $3, from_r + movb %al, 1(out_r) + movb %dl, 2(out_r) + addl $3, out_r + rep movsb + + movl in(%esp), in_r /* move in back to %esi, toss from */ +#if defined( USE_MMX ) + movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ +#endif + jmp .L_while_test + +.align 16,0x90 + +#if ! defined( USE_MMX ) + +.L_test_for_second_level_length: + /* else if ((op & 64) == 0) { + * this = lcode[this.val + (hold & mask[op])]; + * } + */ + testb $64, %al + jnz .L_test_for_end_of_block /* if ((op & 64) != 0) */ + + movl $1, %eax + shll %cl, %eax + decl %eax + andl hold_r, %eax /* eax &= hold */ + addl %edx, %eax /* eax += this.val */ + movl lcode(%esp), %edx /* edx = lcode */ + movl (%edx,%eax,4), %eax /* eax = lcode[val + (hold&mask[op])] */ + jmp .L_dolen + +#else /* USE_MMX */ + +.L_test_for_second_level_length: + testb $64, %al + jnz .L_test_for_end_of_block /* if ((op & 64) != 0) */ + + andl $15, %eax + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + movd hold_mm, %ecx + andl .L_mask(,%eax,4), %ecx + addl len_r, %ecx + movl (%ebx,%ecx,4), %eax /* eax = lcode[hold & lmask] */ + jmp .L_dolen + +#endif + +.align 16,0x90 + +#if ! defined( USE_MMX ) + +.L_test_for_second_level_dist: + /* else if ((op & 64) == 0) { + * this = dcode[this.val + (hold & mask[op])]; + * } + */ + testb $64, %al + jnz .L_invalid_distance_code /* if ((op & 64) != 0) */ + + movl $1, %eax + shll %cl, %eax + decl %eax + andl hold_r, %eax /* eax &= hold */ + addl %edx, %eax /* eax += this.val */ + movl dcode(%esp), %edx /* edx = dcode */ + movl (%edx,%eax,4), %eax /* eax = dcode[val + (hold&mask[op])] */ + jmp .L_dodist + +#else /* USE_MMX */ + +.L_test_for_second_level_dist: + testb $64, %al + jnz .L_invalid_distance_code /* if ((op & 64) != 0) */ + + andl $15, %eax + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + movd hold_mm, %ecx + andl .L_mask(,%eax,4), %ecx + movl dcode(%esp), %eax /* ecx = dcode */ + addl dist_r, %ecx + movl (%eax,%ecx,4), %eax /* eax = lcode[hold & lmask] */ + jmp .L_dodist + +#endif + +.align 16,0x90 +.L_clip_window: + /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * %ecx = nbytes + * + * else { + * if (dist > wsize) { + * invalid distance + * } + * from = window; + * nbytes = dist - nbytes; + * if (write == 0) { + * from += wsize - nbytes; + */ +#define nbytes_r %ecx + + movl %eax, nbytes_r + movl wsize(%esp), %eax /* prepare for dist compare */ + negl nbytes_r /* nbytes = -nbytes */ + movl window(%esp), from_r /* from = window */ + + cmpl dist_r, %eax + jb .L_invalid_distance_too_far /* if (dist > wsize) */ + + addl dist_r, nbytes_r /* nbytes = dist - nbytes */ + cmpl $0, write(%esp) + jne .L_wrap_around_window /* if (write != 0) */ + + subl nbytes_r, %eax + addl %eax, from_r /* from += wsize - nbytes */ + + /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * %ecx = nbytes, %eax = len + * + * if (nbytes < len) { + * len -= nbytes; + * do { + * PUP(out) = PUP(from); + * } while (--nbytes); + * from = out - dist; + * } + * } + */ + +#if ! defined( USE_MMX ) +#define len_r %eax + movl len(%esp), len_r +#endif + cmpl nbytes_r, len_r + jbe .L_do_copy1 /* if (nbytes >= len) */ + + subl nbytes_r, len_r /* len -= nbytes */ + rep movsb + movl out_r, from_r + subl dist_r, from_r /* from = out - dist */ + jmp .L_do_copy1 + + cmpl nbytes_r, len_r + jbe .L_do_copy1 /* if (nbytes >= len) */ + + subl nbytes_r, len_r /* len -= nbytes */ + rep movsb + movl out_r, from_r + subl dist_r, from_r /* from = out - dist */ + jmp .L_do_copy1 + +.L_wrap_around_window: + /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * %ecx = nbytes, %eax = write, %eax = len + * + * else if (write < nbytes) { + * from += wsize + write - nbytes; + * nbytes -= write; + * if (nbytes < len) { + * len -= nbytes; + * do { + * PUP(out) = PUP(from); + * } while (--nbytes); + * from = window; + * nbytes = write; + * if (nbytes < len) { + * len -= nbytes; + * do { + * PUP(out) = PUP(from); + * } while(--nbytes); + * from = out - dist; + * } + * } + * } + */ +#define write_r %eax + + movl write(%esp), write_r + cmpl write_r, nbytes_r + jbe .L_contiguous_in_window /* if (write >= nbytes) */ + + addl wsize(%esp), from_r + addl write_r, from_r + subl nbytes_r, from_r /* from += wsize + write - nbytes */ + subl write_r, nbytes_r /* nbytes -= write */ +#undef write_r + +#if ! defined( USE_MMX ) + movl len(%esp), len_r +#endif + cmpl nbytes_r, len_r + jbe .L_do_copy1 /* if (nbytes >= len) */ + + subl nbytes_r, len_r /* len -= nbytes */ + rep movsb + movl window(%esp), from_r /* from = window */ + movl write(%esp), nbytes_r /* nbytes = write */ + cmpl nbytes_r, len_r + jbe .L_do_copy1 /* if (nbytes >= len) */ + + subl nbytes_r, len_r /* len -= nbytes */ + rep movsb + movl out_r, from_r + subl dist_r, from_r /* from = out - dist */ + jmp .L_do_copy1 + +.L_contiguous_in_window: + /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist + * %ecx = nbytes, %eax = write, %eax = len + * + * else { + * from += write - nbytes; + * if (nbytes < len) { + * len -= nbytes; + * do { + * PUP(out) = PUP(from); + * } while (--nbytes); + * from = out - dist; + * } + * } + */ +#define write_r %eax + + addl write_r, from_r + subl nbytes_r, from_r /* from += write - nbytes */ +#undef write_r + +#if ! defined( USE_MMX ) + movl len(%esp), len_r +#endif + cmpl nbytes_r, len_r + jbe .L_do_copy1 /* if (nbytes >= len) */ + + subl nbytes_r, len_r /* len -= nbytes */ + rep movsb + movl out_r, from_r + subl dist_r, from_r /* from = out - dist */ + +.L_do_copy1: + /* regs: %esi = from, %esi = in, %ebp = hold, %bl = bits, %edi = out + * %eax = len + * + * while (len > 0) { + * PUP(out) = PUP(from); + * len--; + * } + * } + * } while (in < last && out < end); + */ +#undef nbytes_r +#define in_r %esi + + movl len_r, %ecx + rep movsb + + movl in(%esp), in_r /* move in back to %esi, toss from */ +#if defined( USE_MMX ) + movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ +#endif + jmp .L_while_test + +#undef len_r +#undef from_r +#undef dist_r + +.L_invalid_distance_code: + /* else { + * strm->msg = "invalid distance code"; + * state->mode = BAD; + * } + */ + movl $.L_invalid_distance_code_msg, %ecx + movl $INFLATE_MODE_BAD, %edx + jmp .L_update_stream_state + +.L_test_for_end_of_block: + /* else if (op & 32) { + * state->mode = TYPE; + * break; + * } + */ + testb $32, %al + jz .L_invalid_literal_length_code /* if ((op & 32) == 0) */ + + movl $0, %ecx + movl $INFLATE_MODE_TYPE, %edx + jmp .L_update_stream_state + +.L_invalid_literal_length_code: + /* else { + * strm->msg = "invalid literal/length code"; + * state->mode = BAD; + * } + */ + movl $.L_invalid_literal_length_code_msg, %ecx + movl $INFLATE_MODE_BAD, %edx + jmp .L_update_stream_state + +.L_invalid_distance_too_far: + /* strm->msg = "invalid distance too far back"; + * state->mode = BAD; + */ + movl in(%esp), in_r /* from_r has in's reg, put in back */ + movl $.L_invalid_distance_too_far_msg, %ecx + movl $INFLATE_MODE_BAD, %edx + jmp .L_update_stream_state + +.L_update_stream_state: + /* set strm->msg = %ecx, strm->state->mode = %edx */ + movl strm_sp(%esp), %eax + testl %ecx, %ecx /* if (msg != NULL) */ + jz .L_skip_msg + movl %ecx, msg_strm(%eax) /* strm->msg = msg */ +.L_skip_msg: + movl state_strm(%eax), %eax /* state = strm->state */ + movl %edx, mode_state(%eax) /* state->mode = edx (BAD | TYPE) */ + +.L_break_loop: + +#define strm_r %eax +#define state_r %edx + + /* len = bits >> 3; + * in -= len; + * bits -= len << 3; + * hold &= (1U << bits) - 1; + * state->hold = hold; + * state->bits = bits; + * strm->next_in = in; + * strm->next_out = out; + */ + movl strm_sp(%esp), strm_r + movl bitslong_r, %ecx + movl state_strm(strm_r), state_r + shrl $3, %ecx + subl %ecx, in_r + shll $3, %ecx + subl %ecx, bitslong_r + movl out_r, next_out_strm(strm_r) + movl in_r, next_in_strm(strm_r) + movl bitslong_r, bits_state(state_r) + + movl bitslong_r, %ecx + movl $1, %ebx /* overwrites bitslong_r, %bl */ + shll %cl, %ebx + decl %ebx + +#undef bits_r +#undef bitslong_r + +#if ! defined( USE_MMX ) + + andl %ebx, hold_r + movl hold_r, hold_state(state_r) + +#else /* USE_MMX */ + + psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ + movd hold_mm, %ecx + andl %ebx, %ecx + movl %ecx, hold_state(state_r) + +#endif + +#define last_r %ebx + + /* strm->avail_in = in < last ? 5 + (last - in) : 5 - (in - last) */ + movl last(%esp), last_r + cmpl in_r, last_r + jbe .L_last_is_smaller /* if (in >= last) */ + + subl in_r, last_r /* last -= in */ + addl $5, last_r /* last += 5 */ + movl last_r, avail_in_strm(strm_r) + jmp .L_fixup_out +.L_last_is_smaller: + subl last_r, in_r /* in -= last */ + negl in_r /* in = -in */ + addl $5, in_r /* in += 5 */ + movl in_r, avail_in_strm(strm_r) + +#undef last_r +#define end_r %ebx + +.L_fixup_out: + /* strm->avail_out = out < end ? 257 + (end - out) : 257 - (out - end)*/ + movl end(%esp), end_r + cmpl out_r, end_r + jbe .L_end_is_smaller /* if (out >= end) */ + + subl out_r, end_r /* end -= out */ + addl $257, end_r /* end += 257 */ + movl end_r, avail_out_strm(strm_r) + jmp .L_done +.L_end_is_smaller: + subl end_r, out_r /* out -= end */ + negl out_r /* out = -out */ + addl $257, out_r /* out += 257 */ + movl out_r, avail_out_strm(strm_r) + +#undef end_r + +.L_done: +#if defined( USE_MMX ) + emms +#endif + addl $local_var_size, %esp + popf + popl %ebx + popl %ebp + popl %esi + popl %edi + ret + +.type inflate_fast,@function +.size inflate_fast,.-inflate_fast diff --git a/contrib/puff/Makefile b/contrib/puff/Makefile new file mode 100644 index 0000000..b6b6940 --- /dev/null +++ b/contrib/puff/Makefile @@ -0,0 +1,8 @@ +puff: puff.c puff.h + cc -DTEST -o puff puff.c + +test: puff + puff zeros.raw + +clean: + rm -f puff puff.o diff --git a/contrib/puff/README b/contrib/puff/README new file mode 100644 index 0000000..59b3533 --- /dev/null +++ b/contrib/puff/README @@ -0,0 +1,63 @@ +Puff -- A Simple Inflate +3 Mar 2003 +Mark Adler +madler@alumni.caltech.edu + +What this is -- + +puff.c provides the routine puff() to decompress the deflate data format. It +does so more slowly than zlib, but the code is about one-fifth the size of the +inflate code in zlib, and written to be very easy to read. + +Why I wrote this -- + +puff.c was written to document the deflate format unambiguously, by virtue of +being working C code. It is meant to supplement RFC 1951, which formally +describes the deflate format. I have received many questions on details of the +deflate format, and I hope that reading this code will answer those questions. +puff.c is heavily commented with details of the deflate format, especially +those little nooks and cranies of the format that might not be obvious from a +specification. + +puff.c may also be useful in applications where code size or memory usage is a +very limited resource, and speed is not as important. + +How to use it -- + +Well, most likely you should just be reading puff.c and using zlib for actual +applications, but if you must ... + +Include puff.h in your code, which provides this prototype: + +int puff(unsigned char *dest, /* pointer to destination pointer */ + unsigned long *destlen, /* amount of output space */ + unsigned char *source, /* pointer to source data pointer */ + unsigned long *sourcelen); /* amount of input available */ + +Then you can call puff() to decompress a deflate stream that is in memory in +its entirety at source, to a sufficiently sized block of memory for the +decompressed data at dest. puff() is the only external symbol in puff.c The +only C library functions that puff.c needs are setjmp() and longjmp(), which +are used to simplify error checking in the code to improve readabilty. puff.c +does no memory allocation, and uses less than 2K bytes off of the stack. + +If destlen is not enough space for the uncompressed data, then inflate will +return an error without writing more than destlen bytes. Note that this means +that in order to decompress the deflate data successfully, you need to know +the size of the uncompressed data ahead of time. + +If needed, puff() can determine the size of the uncompressed data with no +output space. This is done by passing dest equal to (unsigned char *)0. Then +the initial value of *destlen is ignored and *destlen is set to the length of +the uncompressed data. So if the size of the uncompressed data is not known, +then two passes of puff() can be used--first to determine the size, and second +to do the actual inflation after allocating the appropriate memory. Not +pretty, but it works. (This is one of the reasons you should be using zlib.) + +The deflate format is self-terminating. If the deflate stream does not end +in *sourcelen bytes, puff() will return an error without reading at or past +endsource. + +On return, *sourcelen is updated to the amount of input data consumed, and +*destlen is updated to the size of the uncompressed data. See the comments +in puff.c for the possible return codes for puff(). diff --git a/contrib/puff/puff.c b/contrib/puff/puff.c new file mode 100644 index 0000000..b6039dd --- /dev/null +++ b/contrib/puff/puff.c @@ -0,0 +1,833 @@ +/* + * puff.c + * Copyright (C) 2002, 2003 Mark Adler + * For conditions of distribution and use, see copyright notice in puff.h + * version 1.7, 3 Mar 2003 + * + * puff.c is a simple inflate written to be an unambiguous way to specify the + * deflate format. It is not written for speed but rather simplicity. As a + * side benefit, this code might actually be useful when small code is more + * important than speed, such as bootstrap applications. For typical deflate + * data, zlib's inflate() is about four times as fast as puff(). zlib's + * inflate compiles to around 20K on my machine, whereas puff.c compiles to + * around 4K on my machine (a PowerPC using GNU cc). If the faster decode() + * function here is used, then puff() is only twice as slow as zlib's + * inflate(). + * + * All dynamically allocated memory comes from the stack. The stack required + * is less than 2K bytes. This code is compatible with 16-bit int's and + * assumes that long's are at least 32 bits. puff.c uses the short data type, + * assumed to be 16 bits, for arrays in order to to conserve memory. The code + * works whether integers are stored big endian or little endian. + * + * In the comments below are "Format notes" that describe the inflate process + * and document some of the less obvious aspects of the format. This source + * code is meant to supplement RFC 1951, which formally describes the deflate + * format: + * + * http://www.zlib.org/rfc-deflate.html + */ + +/* + * Change history: + * + * 1.0 10 Feb 2002 - First version + * 1.1 17 Feb 2002 - Clarifications of some comments and notes + * - Update puff() dest and source pointers on negative + * errors to facilitate debugging deflators + * - Remove longest from struct huffman -- not needed + * - Simplify offs[] index in construct() + * - Add input size and checking, using longjmp() to + * maintain easy readability + * - Use short data type for large arrays + * - Use pointers instead of long to specify source and + * destination sizes to avoid arbitrary 4 GB limits + * 1.2 17 Mar 2002 - Add faster version of decode(), doubles speed (!), + * but leave simple version for readabilty + * - Make sure invalid distances detected if pointers + * are 16 bits + * - Fix fixed codes table error + * - Provide a scanning mode for determining size of + * uncompressed data + * 1.3 20 Mar 2002 - Go back to lengths for puff() parameters [Jean-loup] + * - Add a puff.h file for the interface + * - Add braces in puff() for else do [Jean-loup] + * - Use indexes instead of pointers for readability + * 1.4 31 Mar 2002 - Simplify construct() code set check + * - Fix some comments + * - Add FIXLCODES #define + * 1.5 6 Apr 2002 - Minor comment fixes + * 1.6 7 Aug 2002 - Minor format changes + * 1.7 3 Mar 2002 - Added test code for distribution + * - Added zlib-like license + */ + +#include /* for setjmp(), longjmp(), and jmp_buf */ +#include "puff.h" /* prototype for puff() */ + +#define local static /* for local function definitions */ +#define NIL ((unsigned char *)0) /* for no output option */ + +/* + * Maximums for allocations and loops. It is not useful to change these -- + * they are fixed by the deflate format. + */ +#define MAXBITS 15 /* maximum bits in a code */ +#define MAXLCODES 286 /* maximum number of literal/length codes */ +#define MAXDCODES 30 /* maximum number of distance codes */ +#define MAXCODES (MAXLCODES+MAXDCODES) /* maximum codes lengths to read */ +#define FIXLCODES 288 /* number of fixed literal/length codes */ + +/* input and output state */ +struct state { + /* output state */ + unsigned char *out; /* output buffer */ + unsigned long outlen; /* available space at out */ + unsigned long outcnt; /* bytes written to out so far */ + + /* input state */ + unsigned char *in; /* input buffer */ + unsigned long inlen; /* available input at in */ + unsigned long incnt; /* bytes read so far */ + int bitbuf; /* bit buffer */ + int bitcnt; /* number of bits in bit buffer */ + + /* input limit error return state for bits() and decode() */ + jmp_buf env; +}; + +/* + * 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) +{ + long val; /* bit accumulator (can use up to 20 bits) */ + + /* load at least need bits into val */ + val = s->bitbuf; + while (s->bitcnt < need) { + if (s->incnt == s->inlen) longjmp(s->env, 1); /* out of input */ + val |= (long)(s->in[s->incnt++]) << s->bitcnt; /* load eight bits */ + s->bitcnt += 8; + } + + /* drop need bits and update buffer, always zero to seven bits left */ + s->bitbuf = (int)(val >> need); + s->bitcnt -= need; + + /* return need bits, zeroing the bits above that */ + return (int)(val & ((1L << need) - 1)); +} + +/* + * Process a stored block. + * + * Format notes: + * + * - After the two-bit stored block type (00), the stored block length and + * stored bytes are byte-aligned for fast copying. Therefore any leftover + * bits in the byte that has the last bit of the type, as many as seven, are + * discarded. The value of the discarded bits are not defined and should not + * be checked against any expectation. + * + * - The second inverted copy of the stored block length does not have to be + * checked, but it's probably a good idea to do so anyway. + * + * - A stored block can have zero length. This is sometimes used to byte-align + * subsets of the compressed data for random access or partial recovery. + */ +local int stored(struct state *s) +{ + unsigned len; /* length of stored block */ + + /* discard leftover bits from current byte (assumes s->bitcnt < 8) */ + s->bitbuf = 0; + s->bitcnt = 0; + + /* get length and check against its one's complement */ + if (s->incnt + 4 > s->inlen) return 2; /* not enough input */ + len = s->in[s->incnt++]; + len |= s->in[s->incnt++] << 8; + if (s->in[s->incnt++] != (~len & 0xff) || + s->in[s->incnt++] != ((~len >> 8) & 0xff)) + return -2; /* didn't match complement! */ + + /* copy len bytes from in to out */ + if (s->incnt + len > s->inlen) return 2; /* not enough input */ + if (s->out != NIL) { + if (s->outcnt + len > s->outlen) + return 1; /* not enough output space */ + while (len--) + s->out[s->outcnt++] = s->in[s->incnt++]; + } + else { /* just scanning */ + s->outcnt += len; + s->incnt += len; + } + + /* done with a valid stored block */ + return 0; +} + +/* + * 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. A table-based decoding + * scheme (as used in zlib) does not need to do this reversal. + * + * - The first code for the shortest length is all zeros. Subsequent codes of + * the same length are simply integer increments of the previous code. When + * moving up a length, a zero bit is appended to the code. For a complete + * code, the last code of the longest length will be all ones. + * + * - Incomplete codes are handled by this decoder, since they are permitted + * in the deflate format. See the format notes for fixed() and dynamic(). + */ +#ifdef SLOW +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 */ + + code = first = index = 0; + for (len = 1; len <= MAXBITS; len++) { + code |= bits(s, 1); /* get next bit */ + count = h->count[len]; + if (code < first + count) /* if length len, return symbol */ + return h->symbol[index + (code - first)]; + index += count; /* else update for next length */ + first += count; + first <<= 1; + code <<= 1; + } + return -9; /* ran out of codes */ +} + +/* + * A faster version of decode() for real applications of this code. It's not + * as readable, but it makes puff() twice as fast. And it only makes the code + * a few percent larger. + */ +#else /* !SLOW */ +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; + 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->incnt == s->inlen) longjmp(s->env, 1); /* out of input */ + bitbuf = s->in[s->incnt++]; + if (left > 8) left = 8; + } + return -9; /* ran out of codes */ +} +#endif /* SLOW */ + +/* + * 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. + * + * Not used by decode(), but used for error checking, h->count[0] is the number + * of the n symbols not in the code. So n - h->count[0] is the number of + * codes. This is useful for checking for incomplete codes that have more than + * one symbol, which is an error in a dynamic block. + * + * Assumption: for all i in 0..n-1, 0 <= length[i] <= MAXBITS + * This is assured by the construction of the length arrays in dynamic() and + * fixed() and is not verified by construct(). + * + * Format notes: + * + * - Permitted and expected examples of incomplete codes are one of the fixed + * codes and any code with a single symbol which in deflate is coded as one + * bit instead of zero bits. See the format notes for fixed() and dynamic(). + * + * - Within a given code length, the symbols are kept in ascending order for + * the code bits definition. + */ +local int construct(struct huffman *h, short *length, 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 */ + + /* 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 literal/length and distance codes until an end-of-block code. + * + * Format notes: + * + * - Compressed data that is after the block type if fixed or after the code + * description if dynamic is a combination of literals and length/distance + * pairs terminated by and end-of-block code. Literals are simply Huffman + * coded 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. + * + * - Literals, lengths, and the end-of-block code are combined into a single + * code of up to 286 symbols. They are 256 literals (0..255), 29 length + * symbols (257..285), and the end-of-block symbol (256). + * + * - There are 256 possible lengths (3..258), and so 29 symbols are not enough + * to represent all of those. Lengths 3..10 and 258 are in fact represented + * by just a length symbol. Lengths 11..257 are represented as a symbol and + * some number of extra bits that are added as an integer to the base length + * of the length symbol. The number of extra bits is determined by the base + * length symbol. These are in the static arrays below, lens[] for the base + * lengths and lext[] for the corresponding number of extra bits. + * + * - The reason that 258 gets its own symbol is that the longest length is used + * often in highly redundant files. Note that 258 can also be coded as the + * base value 227 plus the maximum extra value of 31. While a good deflate + * should never do this, it is not an error, and should be decoded properly. + * + * - If a length is decoded, including its extra bits if any, then it is + * followed a distance code. There are up to 30 distance symbols. Again + * there are many more possible distances (1..32768), so extra bits are added + * to a base value represented by the symbol. The distances 1..4 get their + * own symbol, but the rest require extra bits. The base distances and + * corresponding number of extra bits are below in the static arrays dist[] + * and dext[]. + * + * - 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 258 + * simply copies the last byte 258 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. You should not use memcpy() since its behavior is not + * defined for overlapped arrays. You should not use memmove() or bcopy() + * since though their behavior -is- defined for overlapping arrays, it is + * defined to do the wrong thing in this case. + */ +local int codes(struct state *s, + struct huffman *lencode, + struct huffman *distcode) +{ + int symbol; /* decoded symbol */ + int len; /* length for copy */ + unsigned dist; /* distance for copy */ + static const short lens[29] = { /* Size base for length codes 257..285 */ + 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, + 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258}; + static const short lext[29] = { /* Extra bits for length codes 257..285 */ + 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, + 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0}; + static const short dists[30] = { /* Offset base for distance codes 0..29 */ + 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, + 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, + 8193, 12289, 16385, 24577}; + static const short dext[30] = { /* Extra bits for distance codes 0..29 */ + 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, + 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, + 12, 12, 13, 13}; + + /* decode literals and length/distance pairs */ + do { + symbol = decode(s, lencode); + if (symbol < 0) return symbol; /* invalid symbol */ + if (symbol < 256) { /* literal: symbol is the byte */ + /* write out the literal */ + if (s->out != NIL) { + if (s->outcnt == s->outlen) return 1; + s->out[s->outcnt] = symbol; + } + s->outcnt++; + } + else if (symbol > 256) { /* length */ + /* get and compute length */ + symbol -= 257; + if (symbol >= 29) return -9; /* invalid fixed code */ + len = lens[symbol] + bits(s, lext[symbol]); + + /* get and check distance */ + symbol = decode(s, distcode); + if (symbol < 0) return symbol; /* invalid symbol */ + dist = dists[symbol] + bits(s, dext[symbol]); + if (dist > s->outcnt) + return -10; /* distance too far back */ + + /* copy length bytes from distance bytes back */ + if (s->out != NIL) { + if (s->outcnt + len > s->outlen) return 1; + while (len--) { + s->out[s->outcnt] = s->out[s->outcnt - dist]; + s->outcnt++; + } + } + else + s->outcnt += len; + } + } while (symbol != 256); /* end of block symbol */ + + /* done with a valid fixed or dynamic block */ + return 0; +} + +/* + * Process a fixed codes block. + * + * Format notes: + * + * - This block type can be useful for compressing small amounts of data for + * which the size of the code descriptions in a dynamic block exceeds the + * benefit of custom codes for that block. For fixed codes, no bits are + * spent on code descriptions. Instead the code lengths for literal/length + * codes and distance codes are fixed. The specific lengths for each symbol + * can be seen in the "for" loops below. + * + * - The literal/length code is complete, but has two symbols that are invalid + * and should result in an error if received. This cannot be implemented + * simply as an incomplete code since those two symbols are in the "middle" + * of the code. They are eight bits long and the longest literal/length\ + * code is nine bits. Therefore the code must be constructed with those + * symbols, and the invalid symbols must be detected after decoding. + * + * - The fixed distance codes also have two invalid symbols that should result + * in an error if received. Since all of the distance codes are the same + * length, this can be implemented as an incomplete code. Then the invalid + * codes are detected while decoding. + */ +local int fixed(struct state *s) +{ + static int virgin = 1; + static short lencnt[MAXBITS+1], lensym[FIXLCODES]; + static short distcnt[MAXBITS+1], distsym[MAXDCODES]; + static struct huffman lencode = {lencnt, lensym}; + static struct huffman distcode = {distcnt, distsym}; + + /* build fixed huffman tables if first call (may not be thread safe) */ + if (virgin) { + int symbol; + short lengths[FIXLCODES]; + + /* literal/length table */ + for (symbol = 0; symbol < 144; symbol++) + lengths[symbol] = 8; + for (; symbol < 256; symbol++) + lengths[symbol] = 9; + for (; symbol < 280; symbol++) + lengths[symbol] = 7; + for (; symbol < FIXLCODES; symbol++) + lengths[symbol] = 8; + construct(&lencode, lengths, FIXLCODES); + + /* distance table */ + for (symbol = 0; symbol < MAXDCODES; symbol++) + lengths[symbol] = 5; + construct(&distcode, lengths, MAXDCODES); + + /* do this just once */ + virgin = 0; + } + + /* decode data until end-of-block code */ + return codes(s, &lencode, &distcode); +} + +/* + * Process a dynamic codes block. + * + * Format notes: + * + * - A dynamic block starts with a description of the literal/length and + * distance codes for that block. New dynamic blocks allow the compressor to + * rapidly adapt to changing data with new codes optimized for that data. + * + * - The codes used by the deflate format are "canonical", which means that + * the actual bits of the codes are generated in an unambiguous way simply + * from the number of bits in each code. Therefore the code descriptions + * are simply a list of code lengths for each symbol. + * + * - The code lengths are stored in order for the symbols, so lengths are + * provided for each of the literal/length symbols, and for each of the + * distance symbols. + * + * - If a symbol is not used in the block, this is represented by a zero as + * as the code length. This does not mean a zero-length code, but rather + * that no code should be created for this symbol. There is no way in the + * deflate format to represent a zero-length code. + * + * - The maximum number of bits in a code is 15, so the possible lengths for + * any code are 1..15. + * + * - The fact that a length of zero is not permitted for a code has an + * interesting consequence. Normally if only one symbol is used for a given + * code, then in fact that code could be represented with zero bits. However + * in deflate, that code has to be at least one bit. So for example, if + * only a single distance base symbol appears in a block, then it will be + * represented by a single code of length one, in particular one 0 bit. This + * is an incomplete code, since if a 1 bit is received, it has no meaning, + * and should result in an error. So incomplete distance codes of one symbol + * should be permitted, and the receipt of invalid codes should be handled. + * + * - It is also possible to have a single literal/length code, but that code + * must be the end-of-block code, since every dynamic block has one. This + * is not the most efficient way to create an empty block (an empty fixed + * block is fewer bits), but it is allowed by the format. So incomplete + * literal/length codes of one symbol should also be permitted. + * + * - The list of up to 286 length/literal lengths and up to 30 distance lengths + * are themselves compressed using Huffman codes and run-length encoding. In + * the list of code lengths, a 0 symbol means no code, a 1..15 symbol means + * that length, and the symbols 16, 17, and 18 are run-length instructions. + * Each of 16, 17, and 18 are follwed by extra bits to define the length of + * the run. 16 copies the last length 3 to 6 times. 17 represents 3 to 10 + * zero lengths, and 18 represents 11 to 138 zero lengths. Unused symbols + * are common, hence the special coding for zero lengths. + * + * - The symbols for 0..18 are Huffman coded, and so that code must be + * described first. This is simply a sequence of up to 19 three-bit values + * representing no code (0) or the code length for that symbol (1..7). + * + * - A dynamic block starts with three fixed-size counts from which is computed + * the number of literal/length code lengths, the number of distance code + * lengths, and the number of code length code lengths (ok, you come up with + * a better name!) in the code descriptions. For the literal/length and + * distance codes, lengths after those provided are considered zero, i.e. no + * code. The code length code lengths are received in a permuted order (see + * the order[] array below) to make a short code length code length list more + * likely. As it turns out, very short and very long codes are less likely + * to be seen in a dynamic code description, hence what may appear initially + * to be a peculiar ordering. + * + * - Given the number of literal/length code lengths (nlen) and distance code + * lengths (ndist), then they are treated as one long list of nlen + ndist + * code lengths. Therefore run-length coding can and often does cross the + * boundary between the two sets of lengths. + * + * - So to summarize, the code description at the start of a dynamic block is + * three counts for the number of code lengths for the literal/length codes, + * the distance codes, and the code length codes. This is followed by the + * code length code lengths, three bits each. This is used to construct the + * code length code which is used to read the remainder of the lengths. Then + * the literal/length code lengths and distance lengths are read as a single + * set of lengths using the code length codes. Codes are constructed from + * the resulting two sets of lengths, and then finally you can start + * decoding actual compressed data in the block. + * + * - For reference, a "typical" size for the code description in a dynamic + * block is around 80 bytes. + */ +local int dynamic(struct state *s) +{ + int nlen, ndist, ncode; /* number of lengths in descriptor */ + int index; /* index of lengths[] */ + int err; /* construct() return value */ + short lengths[MAXCODES]; /* descriptor code lengths */ + short lencnt[MAXBITS+1], lensym[MAXLCODES]; /* lencode memory */ + short distcnt[MAXBITS+1], distsym[MAXDCODES]; /* distcode memory */ + struct huffman lencode = {lencnt, lensym}; /* length code */ + struct huffman distcode = {distcnt, distsym}; /* distance code */ + static const short order[19] = /* permutation of code length codes */ + {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}; + + /* get number of lengths in each table, check lengths */ + nlen = bits(s, 5) + 257; + ndist = bits(s, 5) + 1; + ncode = bits(s, 4) + 4; + if (nlen > MAXLCODES || ndist > MAXDCODES) + return -3; /* bad counts */ + + /* read code length code lengths (really), missing lengths are zero */ + for (index = 0; index < ncode; index++) + lengths[order[index]] = bits(s, 3); + for (; index < 19; index++) + lengths[order[index]] = 0; + + /* build huffman table for code lengths codes (use lencode temporarily) */ + err = construct(&lencode, lengths, 19); + if (err != 0) return -4; /* require complete code set here */ + + /* read length/literal and distance code length tables */ + index = 0; + while (index < nlen + ndist) { + int symbol; /* decoded value */ + int len; /* last length to repeat */ + + symbol = decode(s, &lencode); + if (symbol < 16) /* length in 0..15 */ + lengths[index++] = symbol; + else { /* repeat instruction */ + len = 0; /* assume repeating zeros */ + if (symbol == 16) { /* repeat last length 3..6 times */ + if (index == 0) return -5; /* no last length! */ + len = lengths[index - 1]; /* last length */ + symbol = 3 + bits(s, 2); + } + else if (symbol == 17) /* repeat zero 3..10 times */ + symbol = 3 + bits(s, 3); + else /* == 18, repeat zero 11..138 times */ + symbol = 11 + bits(s, 7); + if (index + symbol > nlen + ndist) + return -6; /* too many lengths! */ + while (symbol--) /* repeat last or zero symbol times */ + lengths[index++] = len; + } + } + + /* build huffman table for literal/length codes */ + err = construct(&lencode, lengths, nlen); + if (err < 0 || (err > 0 && nlen - lencode.count[0] != 1)) + return -7; /* only allow incomplete codes if just one code */ + + /* build huffman table for distance codes */ + err = construct(&distcode, lengths + nlen, ndist); + if (err < 0 || (err > 0 && ndist - distcode.count[0] != 1)) + return -8; /* only allow incomplete codes if just one code */ + + /* decode data until end-of-block code */ + return codes(s, &lencode, &distcode); +} + +/* + * Inflate source to dest. On return, destlen and sourcelen are updated to the + * size of the uncompressed data and the size of the deflate data respectively. + * On success, the return value of puff() is zero. If there is an error in the + * source data, i.e. it is not in the deflate format, then a negative value is + * returned. If there is not enough input available or there is not enough + * output space, then a positive error is returned. In that case, destlen and + * sourcelen are not updated to facilitate retrying from the beginning with the + * provision of more input data or more output space. In the case of invalid + * inflate data (a negative error), the dest and source pointers are updated to + * facilitate the debugging of deflators. + * + * puff() also has a mode to determine the size of the uncompressed output with + * no output written. For this dest must be (unsigned char *)0. In this case, + * the input value of *destlen is ignored, and on return *destlen is set to the + * size of the uncompressed output. + * + * The return codes are: + * + * 2: available inflate data did not terminate + * 1: output space exhausted before completing inflate + * 0: successful inflate + * -1: invalid block type (type == 3) + * -2: stored block length did not match one's complement + * -3: dynamic block code description: too many length or distance codes + * -4: dynamic block code description: code lengths codes incomplete + * -5: dynamic block code description: repeat lengths with no first length + * -6: dynamic block code description: repeat more than specified lengths + * -7: dynamic block code description: invalid literal/length code lengths + * -8: dynamic block code description: invalid distance code lengths + * -9: invalid literal/length or distance code in fixed or dynamic block + * -10: distance is too far back in fixed or dynamic block + * + * Format notes: + * + * - Three bits are read for each block to determine the kind of block and + * whether or not it is the last block. Then the block is decoded and the + * process repeated if it was not the last block. + * + * - The leftover bits in the last byte of the deflate data after the last + * block (if it was a fixed or dynamic block) are undefined and have no + * expected values to check. + */ +int puff(unsigned char *dest, /* pointer to destination pointer */ + unsigned long *destlen, /* amount of output space */ + unsigned char *source, /* pointer to source data pointer */ + unsigned long *sourcelen) /* amount of input available */ +{ + struct state s; /* input/output state */ + int last, type; /* block information */ + int err; /* return value */ + + /* initialize output state */ + s.out = dest; + s.outlen = *destlen; /* ignored if dest is NIL */ + s.outcnt = 0; + + /* initialize input state */ + s.in = source; + s.inlen = *sourcelen; + s.incnt = 0; + s.bitbuf = 0; + s.bitcnt = 0; + + /* 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 do-loop, return error */ + else { + /* process blocks until last block or error */ + do { + last = bits(&s, 1); /* one if last block */ + type = bits(&s, 2); /* block type 0..3 */ + err = type == 0 ? stored(&s) : + (type == 1 ? fixed(&s) : + (type == 2 ? dynamic(&s) : + -1)); /* type == 3, invalid */ + if (err != 0) break; /* return with error */ + } while (!last); + } + + /* update the lengths and return */ + if (err <= 0) { + *destlen = s.outcnt; + *sourcelen = s.incnt; + } + return err; +} + +#ifdef TEST +/* Example of how to use puff() */ +#include +#include +#include +#include + +local unsigned char *yank(char *name, unsigned long *len) +{ + unsigned long size; + unsigned char *buf; + FILE *in; + struct stat s; + + *len = 0; + if (stat(name, &s)) return NULL; + if ((s.st_mode & S_IFMT) != S_IFREG) return NULL; + size = (unsigned long)(s.st_size); + if (size == 0 || (off_t)size != s.st_size) return NULL; + in = fopen(name, "r"); + if (in == NULL) return NULL; + buf = malloc(size); + if (buf != NULL && fread(buf, 1, size, in) != size) { + free(buf); + buf = NULL; + } + fclose(in); + *len = size; + return buf; +} + +int main(int argc, char **argv) +{ + int ret; + unsigned char *source; + unsigned long len, sourcelen, destlen; + + if (argc < 2) return 2; + source = yank(argv[1], &len); + if (source == NULL) return 2; + sourcelen = len; + ret = puff(NIL, &destlen, source, &sourcelen); + if (ret) + printf("puff() failed with return code %d\n", ret); + else { + printf("puff() succeeded uncompressing %lu bytes\n", destlen); + if (sourcelen < len) printf("%lu compressed bytes unused\n", + len - sourcelen); + } + free(source); + return ret; +} +#endif diff --git a/contrib/puff/puff.h b/contrib/puff/puff.h new file mode 100644 index 0000000..41ea7e1 --- /dev/null +++ b/contrib/puff/puff.h @@ -0,0 +1,31 @@ +/* puff.h + Copyright (C) 2002, 2003 Mark Adler, all rights reserved + version 1.7, 3 Mar 2002 + + This software is provided 'as-is', without any express or implied + warranty. In no event will the author be held liable for any damages + arising from the use of this software. + + Permission is granted to anyone to use this software for any purpose, + including commercial applications, and to alter it and redistribute it + freely, subject to the following restrictions: + + 1. The origin of this software must not be misrepresented; you must not + claim that you wrote the original software. If you use this software + in a product, an acknowledgment in the product documentation would be + appreciated but is not required. + 2. Altered source versions must be plainly marked as such, and must not be + misrepresented as being the original software. + 3. This notice may not be removed or altered from any source distribution. + + Mark Adler madler@alumni.caltech.edu + */ + + +/* + * See puff.c for purpose and usage. + */ +int puff(unsigned char *dest, /* pointer to destination pointer */ + unsigned long *destlen, /* amount of output space */ + unsigned char *source, /* pointer to source data pointer */ + unsigned long *sourcelen); /* amount of input available */ diff --git a/contrib/puff/zeros.raw b/contrib/puff/zeros.raw new file mode 100644 index 0000000..637b7be Binary files /dev/null and b/contrib/puff/zeros.raw differ -- cgit v1.2.3-55-g6feb