[llvm-commits] [parallel] CVS: llvm/runtime/zlib/contrib/puff/Makefile README puff.c puff.h zeros.raw
Misha Brukman
brukman at cs.uiuc.edu
Mon Mar 1 18:19:29 PST 2004
Changes in directory llvm/runtime/zlib/contrib/puff:
Makefile added (r1.1.2.1)
README added (r1.1.2.1)
puff.c added (r1.1.2.1)
puff.h added (r1.1.2.1)
zeros.raw added (r1.1.2.1)
---
Log message:
Merge from trunk
---
Diffs of the changes: (+935 -0)
Index: llvm/runtime/zlib/contrib/puff/Makefile
diff -c /dev/null llvm/runtime/zlib/contrib/puff/Makefile:1.1.2.1
*** /dev/null Mon Mar 1 17:58:58 2004
--- llvm/runtime/zlib/contrib/puff/Makefile Mon Mar 1 17:58:48 2004
***************
*** 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
Index: llvm/runtime/zlib/contrib/puff/README
diff -c /dev/null llvm/runtime/zlib/contrib/puff/README:1.1.2.1
*** /dev/null Mon Mar 1 17:58:58 2004
--- llvm/runtime/zlib/contrib/puff/README Mon Mar 1 17:58:48 2004
***************
*** 0 ****
--- 1,63 ----
+ Puff -- A Simple Inflate
+ 3 Mar 2003
+ Mark Adler
+ madler at 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().
Index: llvm/runtime/zlib/contrib/puff/puff.c
diff -c /dev/null llvm/runtime/zlib/contrib/puff/puff.c:1.1.2.1
*** /dev/null Mon Mar 1 17:58:59 2004
--- llvm/runtime/zlib/contrib/puff/puff.c Mon Mar 1 17:58:48 2004
***************
*** 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 2003 - Added test code for distribution
+ * - Added zlib-like license
+ */
+
+ #include <setjmp.h> /* 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 <stdio.h>
+ #include <stdlib.h>
+ #include <sys/types.h>
+ #include <sys/stat.h>
+
+ 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
Index: llvm/runtime/zlib/contrib/puff/puff.h
diff -c /dev/null llvm/runtime/zlib/contrib/puff/puff.h:1.1.2.1
*** /dev/null Mon Mar 1 17:58:59 2004
--- llvm/runtime/zlib/contrib/puff/puff.h Mon Mar 1 17:58:48 2004
***************
*** 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 at 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 */
Index: llvm/runtime/zlib/contrib/puff/zeros.raw
More information about the llvm-commits
mailing list