]> pd.if.org Git - zpackage/blobdiff - lzma/lz/lz_decoder.h
integrate lzma
[zpackage] / lzma / lz / lz_decoder.h
diff --git a/lzma/lz/lz_decoder.h b/lzma/lz/lz_decoder.h
new file mode 100644 (file)
index 0000000..277900a
--- /dev/null
@@ -0,0 +1,235 @@
+///////////////////////////////////////////////////////////////////////////////
+//
+/// \file       lz_decoder.h
+/// \brief      LZ out window
+///
+//  Authors:    Igor Pavlov
+//              Lasse Collin
+//
+//  This file has been put into the public domain.
+//  You can do whatever you want with this file.
+//
+///////////////////////////////////////////////////////////////////////////////
+
+#ifndef LZMA_LZ_DECODER_H
+#define LZMA_LZ_DECODER_H
+
+#include "common.h"
+
+
+typedef struct {
+       /// Pointer to the dictionary buffer. It can be an allocated buffer
+       /// internal to liblzma, or it can a be a buffer given by the
+       /// application when in single-call mode (not implemented yet).
+       uint8_t *buf;
+
+       /// Write position in dictionary. The next byte will be written to
+       /// buf[pos].
+       size_t pos;
+
+       /// Indicates how full the dictionary is. This is used by
+       /// dict_is_distance_valid() to detect corrupt files that would
+       /// read beyond the beginning of the dictionary.
+       size_t full;
+
+       /// Write limit
+       size_t limit;
+
+       /// Size of the dictionary
+       size_t size;
+
+       /// True when dictionary should be reset before decoding more data.
+       bool need_reset;
+
+} lzma_dict;
+
+
+typedef struct {
+       size_t dict_size;
+       const uint8_t *preset_dict;
+       size_t preset_dict_size;
+} lzma_lz_options;
+
+
+typedef struct {
+       /// Data specific to the LZ-based decoder
+       lzma_coder *coder;
+
+       /// Function to decode from in[] to *dict
+       lzma_ret (*code)(lzma_coder *restrict coder,
+                       lzma_dict *restrict dict, const uint8_t *restrict in,
+                       size_t *restrict in_pos, size_t in_size);
+
+       void (*reset)(lzma_coder *coder, const void *options);
+
+       /// Set the uncompressed size
+       void (*set_uncompressed)(lzma_coder *coder,
+                       lzma_vli uncompressed_size);
+
+       /// Free allocated resources
+       void (*end)(lzma_coder *coder, const lzma_allocator *allocator);
+
+} lzma_lz_decoder;
+
+
+#define LZMA_LZ_DECODER_INIT \
+       (lzma_lz_decoder){ \
+               .coder = NULL, \
+               .code = NULL, \
+               .reset = NULL, \
+               .set_uncompressed = NULL, \
+               .end = NULL, \
+       }
+
+
+extern lzma_ret lzma_lz_decoder_init(lzma_next_coder *next,
+               const lzma_allocator *allocator,
+               const lzma_filter_info *filters,
+               lzma_ret (*lz_init)(lzma_lz_decoder *lz,
+                       const lzma_allocator *allocator, const void *options,
+                       lzma_lz_options *lz_options));
+
+extern uint64_t lzma_lz_decoder_memusage(size_t dictionary_size);
+
+extern void lzma_lz_decoder_uncompressed(
+               lzma_coder *coder, lzma_vli uncompressed_size);
+
+
+//////////////////////
+// Inline functions //
+//////////////////////
+
+/// Get a byte from the history buffer.
+static inline uint8_t
+dict_get(const lzma_dict *const dict, const uint32_t distance)
+{
+       return dict->buf[dict->pos - distance - 1
+                       + (distance < dict->pos ? 0 : dict->size)];
+}
+
+
+/// Test if dictionary is empty.
+static inline bool
+dict_is_empty(const lzma_dict *const dict)
+{
+       return dict->full == 0;
+}
+
+
+/// Validate the match distance
+static inline bool
+dict_is_distance_valid(const lzma_dict *const dict, const size_t distance)
+{
+       return dict->full > distance;
+}
+
+
+/// Repeat *len bytes at distance.
+static inline bool
+dict_repeat(lzma_dict *dict, uint32_t distance, uint32_t *len)
+{
+       // Don't write past the end of the dictionary.
+       const size_t dict_avail = dict->limit - dict->pos;
+       uint32_t left = my_min(dict_avail, *len);
+       *len -= left;
+
+       // Repeat a block of data from the history. Because memcpy() is faster
+       // than copying byte by byte in a loop, the copying process gets split
+       // into three cases.
+       if (distance < left) {
+               // Source and target areas overlap, thus we can't use
+               // memcpy() nor even memmove() safely.
+               do {
+                       dict->buf[dict->pos] = dict_get(dict, distance);
+                       ++dict->pos;
+               } while (--left > 0);
+
+       } else if (distance < dict->pos) {
+               // The easiest and fastest case
+               memcpy(dict->buf + dict->pos,
+                               dict->buf + dict->pos - distance - 1,
+                               left);
+               dict->pos += left;
+
+       } else {
+               // The bigger the dictionary, the more rare this
+               // case occurs. We need to "wrap" the dict, thus
+               // we might need two memcpy() to copy all the data.
+               assert(dict->full == dict->size);
+               const uint32_t copy_pos
+                               = dict->pos - distance - 1 + dict->size;
+               uint32_t copy_size = dict->size - copy_pos;
+
+               if (copy_size < left) {
+                       memmove(dict->buf + dict->pos, dict->buf + copy_pos,
+                                       copy_size);
+                       dict->pos += copy_size;
+                       copy_size = left - copy_size;
+                       memcpy(dict->buf + dict->pos, dict->buf, copy_size);
+                       dict->pos += copy_size;
+               } else {
+                       memmove(dict->buf + dict->pos, dict->buf + copy_pos,
+                                       left);
+                       dict->pos += left;
+               }
+       }
+
+       // Update how full the dictionary is.
+       if (dict->full < dict->pos)
+               dict->full = dict->pos;
+
+       return unlikely(*len != 0);
+}
+
+
+/// Puts one byte into the dictionary. Returns true if the dictionary was
+/// already full and the byte couldn't be added.
+static inline bool
+dict_put(lzma_dict *dict, uint8_t byte)
+{
+       if (unlikely(dict->pos == dict->limit))
+               return true;
+
+       dict->buf[dict->pos++] = byte;
+
+       if (dict->pos > dict->full)
+               dict->full = dict->pos;
+
+       return false;
+}
+
+
+/// Copies arbitrary amount of data into the dictionary.
+static inline void
+dict_write(lzma_dict *restrict dict, const uint8_t *restrict in,
+               size_t *restrict in_pos, size_t in_size,
+               size_t *restrict left)
+{
+       // NOTE: If we are being given more data than the size of the
+       // dictionary, it could be possible to optimize the LZ decoder
+       // so that not everything needs to go through the dictionary.
+       // This shouldn't be very common thing in practice though, and
+       // the slowdown of one extra memcpy() isn't bad compared to how
+       // much time it would have taken if the data were compressed.
+
+       if (in_size - *in_pos > *left)
+               in_size = *in_pos + *left;
+
+       *left -= lzma_bufcpy(in, in_pos, in_size,
+                       dict->buf, &dict->pos, dict->limit);
+
+       if (dict->pos > dict->full)
+               dict->full = dict->pos;
+
+       return;
+}
+
+
+static inline void
+dict_reset(lzma_dict *dict)
+{
+       dict->need_reset = true;
+       return;
+}
+
+#endif