diff options
Diffstat (limited to '')
25 files changed, 7524 insertions, 0 deletions
diff --git a/src/liblzma/lz/Makefile.inc b/src/liblzma/lz/Makefile.inc new file mode 100644 index 0000000..75742a8 --- /dev/null +++ b/src/liblzma/lz/Makefile.inc @@ -0,0 +1,22 @@ +## +## Author: Lasse Collin +## +## This file has been put into the public domain. +## You can do whatever you want with this file. +## + +if COND_ENCODER_LZ +liblzma_la_SOURCES += \ + lz/lz_encoder.c \ + lz/lz_encoder.h \ + lz/lz_encoder_hash.h \ + lz/lz_encoder_hash_table.h \ + lz/lz_encoder_mf.c +endif + + +if COND_DECODER_LZ +liblzma_la_SOURCES += \ + lz/lz_decoder.c \ + lz/lz_decoder.h +endif diff --git a/src/liblzma/lz/lz_decoder.c b/src/liblzma/lz/lz_decoder.c new file mode 100644 index 0000000..06c95c1 --- /dev/null +++ b/src/liblzma/lz/lz_decoder.c @@ -0,0 +1,304 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lz_decoder.c +/// \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. +// +/////////////////////////////////////////////////////////////////////////////// + +// liblzma supports multiple LZ77-based filters. The LZ part is shared +// between these filters. The LZ code takes care of dictionary handling +// and passing the data between filters in the chain. The filter-specific +// part decodes from the input buffer to the dictionary. + + +#include "lz_decoder.h" + + +typedef struct { + /// Dictionary (history buffer) + lzma_dict dict; + + /// The actual LZ-based decoder e.g. LZMA + lzma_lz_decoder lz; + + /// Next filter in the chain, if any. Note that LZMA and LZMA2 are + /// only allowed as the last filter, but the long-range filter in + /// future can be in the middle of the chain. + lzma_next_coder next; + + /// True if the next filter in the chain has returned LZMA_STREAM_END. + bool next_finished; + + /// True if the LZ decoder (e.g. LZMA) has detected end of payload + /// marker. This may become true before next_finished becomes true. + bool this_finished; + + /// Temporary buffer needed when the LZ-based filter is not the last + /// filter in the chain. The output of the next filter is first + /// decoded into buffer[], which is then used as input for the actual + /// LZ-based decoder. + struct { + size_t pos; + size_t size; + uint8_t buffer[LZMA_BUFFER_SIZE]; + } temp; +} lzma_coder; + + +static void +lz_decoder_reset(lzma_coder *coder) +{ + coder->dict.pos = 0; + coder->dict.full = 0; + coder->dict.buf[coder->dict.size - 1] = '\0'; + coder->dict.need_reset = false; + return; +} + + +static lzma_ret +decode_buffer(lzma_coder *coder, + const uint8_t *restrict in, size_t *restrict in_pos, + size_t in_size, uint8_t *restrict out, + size_t *restrict out_pos, size_t out_size) +{ + while (true) { + // Wrap the dictionary if needed. + if (coder->dict.pos == coder->dict.size) + coder->dict.pos = 0; + + // Store the current dictionary position. It is needed to know + // where to start copying to the out[] buffer. + const size_t dict_start = coder->dict.pos; + + // Calculate how much we allow coder->lz.code() to decode. + // It must not decode past the end of the dictionary + // buffer, and we don't want it to decode more than is + // actually needed to fill the out[] buffer. + coder->dict.limit = coder->dict.pos + + my_min(out_size - *out_pos, + coder->dict.size - coder->dict.pos); + + // Call the coder->lz.code() to do the actual decoding. + const lzma_ret ret = coder->lz.code( + coder->lz.coder, &coder->dict, + in, in_pos, in_size); + + // Copy the decoded data from the dictionary to the out[] + // buffer. Do it conditionally because out can be NULL + // (in which case copy_size is always 0). Calling memcpy() + // with a null-pointer is undefined even if the third + // argument is 0. + const size_t copy_size = coder->dict.pos - dict_start; + assert(copy_size <= out_size - *out_pos); + + if (copy_size > 0) + memcpy(out + *out_pos, coder->dict.buf + dict_start, + copy_size); + + *out_pos += copy_size; + + // Reset the dictionary if so requested by coder->lz.code(). + if (coder->dict.need_reset) { + lz_decoder_reset(coder); + + // Since we reset dictionary, we don't check if + // dictionary became full. + if (ret != LZMA_OK || *out_pos == out_size) + return ret; + } else { + // Return if everything got decoded or an error + // occurred, or if there's no more data to decode. + // + // Note that detecting if there's something to decode + // is done by looking if dictionary become full + // instead of looking if *in_pos == in_size. This + // is because it is possible that all the input was + // consumed already but some data is pending to be + // written to the dictionary. + if (ret != LZMA_OK || *out_pos == out_size + || coder->dict.pos < coder->dict.size) + return ret; + } + } +} + + +static lzma_ret +lz_decode(void *coder_ptr, const lzma_allocator *allocator, + const uint8_t *restrict in, size_t *restrict in_pos, + size_t in_size, uint8_t *restrict out, + size_t *restrict out_pos, size_t out_size, + lzma_action action) +{ + lzma_coder *coder = coder_ptr; + + if (coder->next.code == NULL) + return decode_buffer(coder, in, in_pos, in_size, + out, out_pos, out_size); + + // We aren't the last coder in the chain, we need to decode + // our input to a temporary buffer. + while (*out_pos < out_size) { + // Fill the temporary buffer if it is empty. + if (!coder->next_finished + && coder->temp.pos == coder->temp.size) { + coder->temp.pos = 0; + coder->temp.size = 0; + + const lzma_ret ret = coder->next.code( + coder->next.coder, + allocator, in, in_pos, in_size, + coder->temp.buffer, &coder->temp.size, + LZMA_BUFFER_SIZE, action); + + if (ret == LZMA_STREAM_END) + coder->next_finished = true; + else if (ret != LZMA_OK || coder->temp.size == 0) + return ret; + } + + if (coder->this_finished) { + if (coder->temp.size != 0) + return LZMA_DATA_ERROR; + + if (coder->next_finished) + return LZMA_STREAM_END; + + return LZMA_OK; + } + + const lzma_ret ret = decode_buffer(coder, coder->temp.buffer, + &coder->temp.pos, coder->temp.size, + out, out_pos, out_size); + + if (ret == LZMA_STREAM_END) + coder->this_finished = true; + else if (ret != LZMA_OK) + return ret; + else if (coder->next_finished && *out_pos < out_size) + return LZMA_DATA_ERROR; + } + + return LZMA_OK; +} + + +static void +lz_decoder_end(void *coder_ptr, const lzma_allocator *allocator) +{ + lzma_coder *coder = coder_ptr; + + lzma_next_end(&coder->next, allocator); + lzma_free(coder->dict.buf, allocator); + + if (coder->lz.end != NULL) + coder->lz.end(coder->lz.coder, allocator); + else + lzma_free(coder->lz.coder, allocator); + + lzma_free(coder, allocator); + return; +} + + +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, + lzma_vli id, const void *options, + lzma_lz_options *lz_options)) +{ + // Allocate the base structure if it isn't already allocated. + lzma_coder *coder = next->coder; + if (coder == NULL) { + coder = lzma_alloc(sizeof(lzma_coder), allocator); + if (coder == NULL) + return LZMA_MEM_ERROR; + + next->coder = coder; + next->code = &lz_decode; + next->end = &lz_decoder_end; + + coder->dict.buf = NULL; + coder->dict.size = 0; + coder->lz = LZMA_LZ_DECODER_INIT; + coder->next = LZMA_NEXT_CODER_INIT; + } + + // Allocate and initialize the LZ-based decoder. It will also give + // us the dictionary size. + lzma_lz_options lz_options; + return_if_error(lz_init(&coder->lz, allocator, + filters[0].id, filters[0].options, &lz_options)); + + // If the dictionary size is very small, increase it to 4096 bytes. + // This is to prevent constant wrapping of the dictionary, which + // would slow things down. The downside is that since we don't check + // separately for the real dictionary size, we may happily accept + // corrupt files. + if (lz_options.dict_size < 4096) + lz_options.dict_size = 4096; + + // Make dictionary size a multiple of 16. Some LZ-based decoders like + // LZMA use the lowest bits lzma_dict.pos to know the alignment of the + // data. Aligned buffer is also good when memcpying from the + // dictionary to the output buffer, since applications are + // recommended to give aligned buffers to liblzma. + // + // Avoid integer overflow. + if (lz_options.dict_size > SIZE_MAX - 15) + return LZMA_MEM_ERROR; + + lz_options.dict_size = (lz_options.dict_size + 15) & ~((size_t)(15)); + + // Allocate and initialize the dictionary. + if (coder->dict.size != lz_options.dict_size) { + lzma_free(coder->dict.buf, allocator); + coder->dict.buf + = lzma_alloc(lz_options.dict_size, allocator); + if (coder->dict.buf == NULL) + return LZMA_MEM_ERROR; + + coder->dict.size = lz_options.dict_size; + } + + lz_decoder_reset(next->coder); + + // Use the preset dictionary if it was given to us. + if (lz_options.preset_dict != NULL + && lz_options.preset_dict_size > 0) { + // If the preset dictionary is bigger than the actual + // dictionary, copy only the tail. + const size_t copy_size = my_min(lz_options.preset_dict_size, + lz_options.dict_size); + const size_t offset = lz_options.preset_dict_size - copy_size; + memcpy(coder->dict.buf, lz_options.preset_dict + offset, + copy_size); + coder->dict.pos = copy_size; + coder->dict.full = copy_size; + } + + // Miscellaneous initializations + coder->next_finished = false; + coder->this_finished = false; + coder->temp.pos = 0; + coder->temp.size = 0; + + // Initialize the next filter in the chain, if any. + return lzma_next_filter_init(&coder->next, allocator, filters + 1); +} + + +extern uint64_t +lzma_lz_decoder_memusage(size_t dictionary_size) +{ + return sizeof(lzma_coder) + (uint64_t)(dictionary_size); +} diff --git a/src/liblzma/lz/lz_decoder.h b/src/liblzma/lz/lz_decoder.h new file mode 100644 index 0000000..ad80d4d --- /dev/null +++ b/src/liblzma/lz/lz_decoder.h @@ -0,0 +1,234 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \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 + void *coder; + + /// Function to decode from in[] to *dict + lzma_ret (*code)(void *coder, + lzma_dict *restrict dict, const uint8_t *restrict in, + size_t *restrict in_pos, size_t in_size); + + void (*reset)(void *coder, const void *options); + + /// Set the uncompressed size. If uncompressed_size == LZMA_VLI_UNKNOWN + /// then allow_eopm will always be true. + void (*set_uncompressed)(void *coder, lzma_vli uncompressed_size, + bool allow_eopm); + + /// Free allocated resources + void (*end)(void *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, + lzma_vli id, const void *options, + lzma_lz_options *lz_options)); + +extern uint64_t lzma_lz_decoder_memusage(size_t dictionary_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 diff --git a/src/liblzma/lz/lz_encoder.c b/src/liblzma/lz/lz_encoder.c new file mode 100644 index 0000000..5489085 --- /dev/null +++ b/src/liblzma/lz/lz_encoder.c @@ -0,0 +1,633 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lz_encoder.c +/// \brief LZ in window +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lz_encoder.h" +#include "lz_encoder_hash.h" + +// See lz_encoder_hash.h. This is a bit hackish but avoids making +// endianness a conditional in makefiles. +#if defined(WORDS_BIGENDIAN) && !defined(HAVE_SMALL) +# include "lz_encoder_hash_table.h" +#endif + +#include "memcmplen.h" + + +typedef struct { + /// LZ-based encoder e.g. LZMA + lzma_lz_encoder lz; + + /// History buffer and match finder + lzma_mf mf; + + /// Next coder in the chain + lzma_next_coder next; +} lzma_coder; + + +/// \brief Moves the data in the input window to free space for new data +/// +/// mf->buffer is a sliding input window, which keeps mf->keep_size_before +/// bytes of input history available all the time. Now and then we need to +/// "slide" the buffer to make space for the new data to the end of the +/// buffer. At the same time, data older than keep_size_before is dropped. +/// +static void +move_window(lzma_mf *mf) +{ + // Align the move to a multiple of 16 bytes. Some LZ-based encoders + // like LZMA use the lowest bits of mf->read_pos to know the + // alignment of the uncompressed data. We also get better speed + // for memmove() with aligned buffers. + assert(mf->read_pos > mf->keep_size_before); + const uint32_t move_offset + = (mf->read_pos - mf->keep_size_before) & ~UINT32_C(15); + + assert(mf->write_pos > move_offset); + const size_t move_size = mf->write_pos - move_offset; + + assert(move_offset + move_size <= mf->size); + + memmove(mf->buffer, mf->buffer + move_offset, move_size); + + mf->offset += move_offset; + mf->read_pos -= move_offset; + mf->read_limit -= move_offset; + mf->write_pos -= move_offset; + + return; +} + + +/// \brief Tries to fill the input window (mf->buffer) +/// +/// If we are the last encoder in the chain, our input data is in in[]. +/// Otherwise we call the next filter in the chain to process in[] and +/// write its output to mf->buffer. +/// +/// This function must not be called once it has returned LZMA_STREAM_END. +/// +static lzma_ret +fill_window(lzma_coder *coder, const lzma_allocator *allocator, + const uint8_t *in, size_t *in_pos, size_t in_size, + lzma_action action) +{ + assert(coder->mf.read_pos <= coder->mf.write_pos); + + // Move the sliding window if needed. + if (coder->mf.read_pos >= coder->mf.size - coder->mf.keep_size_after) + move_window(&coder->mf); + + // Maybe this is ugly, but lzma_mf uses uint32_t for most things + // (which I find cleanest), but we need size_t here when filling + // the history window. + size_t write_pos = coder->mf.write_pos; + lzma_ret ret; + if (coder->next.code == NULL) { + // Not using a filter, simply memcpy() as much as possible. + lzma_bufcpy(in, in_pos, in_size, coder->mf.buffer, + &write_pos, coder->mf.size); + + ret = action != LZMA_RUN && *in_pos == in_size + ? LZMA_STREAM_END : LZMA_OK; + + } else { + ret = coder->next.code(coder->next.coder, allocator, + in, in_pos, in_size, + coder->mf.buffer, &write_pos, + coder->mf.size, action); + } + + coder->mf.write_pos = write_pos; + + // Silence Valgrind. lzma_memcmplen() can read extra bytes + // and Valgrind will give warnings if those bytes are uninitialized + // because Valgrind cannot see that the values of the uninitialized + // bytes are eventually ignored. + memzero(coder->mf.buffer + write_pos, LZMA_MEMCMPLEN_EXTRA); + + // If end of stream has been reached or flushing completed, we allow + // the encoder to process all the input (that is, read_pos is allowed + // to reach write_pos). Otherwise we keep keep_size_after bytes + // available as prebuffer. + if (ret == LZMA_STREAM_END) { + assert(*in_pos == in_size); + ret = LZMA_OK; + coder->mf.action = action; + coder->mf.read_limit = coder->mf.write_pos; + + } else if (coder->mf.write_pos > coder->mf.keep_size_after) { + // This needs to be done conditionally, because if we got + // only little new input, there may be too little input + // to do any encoding yet. + coder->mf.read_limit = coder->mf.write_pos + - coder->mf.keep_size_after; + } + + // Restart the match finder after finished LZMA_SYNC_FLUSH. + if (coder->mf.pending > 0 + && coder->mf.read_pos < coder->mf.read_limit) { + // Match finder may update coder->pending and expects it to + // start from zero, so use a temporary variable. + const uint32_t pending = coder->mf.pending; + coder->mf.pending = 0; + + // Rewind read_pos so that the match finder can hash + // the pending bytes. + assert(coder->mf.read_pos >= pending); + coder->mf.read_pos -= pending; + + // Call the skip function directly instead of using + // mf_skip(), since we don't want to touch mf->read_ahead. + coder->mf.skip(&coder->mf, pending); + } + + return ret; +} + + +static lzma_ret +lz_encode(void *coder_ptr, const lzma_allocator *allocator, + const uint8_t *restrict in, size_t *restrict in_pos, + size_t in_size, + uint8_t *restrict out, size_t *restrict out_pos, + size_t out_size, lzma_action action) +{ + lzma_coder *coder = coder_ptr; + + while (*out_pos < out_size + && (*in_pos < in_size || action != LZMA_RUN)) { + // Read more data to coder->mf.buffer if needed. + if (coder->mf.action == LZMA_RUN && coder->mf.read_pos + >= coder->mf.read_limit) + return_if_error(fill_window(coder, allocator, + in, in_pos, in_size, action)); + + // Encode + const lzma_ret ret = coder->lz.code(coder->lz.coder, + &coder->mf, out, out_pos, out_size); + if (ret != LZMA_OK) { + // Setting this to LZMA_RUN for cases when we are + // flushing. It doesn't matter when finishing or if + // an error occurred. + coder->mf.action = LZMA_RUN; + return ret; + } + } + + return LZMA_OK; +} + + +static bool +lz_encoder_prepare(lzma_mf *mf, const lzma_allocator *allocator, + const lzma_lz_options *lz_options) +{ + // For now, the dictionary size is limited to 1.5 GiB. This may grow + // in the future if needed, but it needs a little more work than just + // changing this check. + if (lz_options->dict_size < LZMA_DICT_SIZE_MIN + || lz_options->dict_size + > (UINT32_C(1) << 30) + (UINT32_C(1) << 29) + || lz_options->nice_len > lz_options->match_len_max) + return true; + + mf->keep_size_before = lz_options->before_size + lz_options->dict_size; + + mf->keep_size_after = lz_options->after_size + + lz_options->match_len_max; + + // To avoid constant memmove()s, allocate some extra space. Since + // memmove()s become more expensive when the size of the buffer + // increases, we reserve more space when a large dictionary is + // used to make the memmove() calls rarer. + // + // This works with dictionaries up to about 3 GiB. If bigger + // dictionary is wanted, some extra work is needed: + // - Several variables in lzma_mf have to be changed from uint32_t + // to size_t. + // - Memory usage calculation needs something too, e.g. use uint64_t + // for mf->size. + uint32_t reserve = lz_options->dict_size / 2; + if (reserve > (UINT32_C(1) << 30)) + reserve /= 2; + + reserve += (lz_options->before_size + lz_options->match_len_max + + lz_options->after_size) / 2 + (UINT32_C(1) << 19); + + const uint32_t old_size = mf->size; + mf->size = mf->keep_size_before + reserve + mf->keep_size_after; + + // Deallocate the old history buffer if it exists but has different + // size than what is needed now. + if (mf->buffer != NULL && old_size != mf->size) { + lzma_free(mf->buffer, allocator); + mf->buffer = NULL; + } + + // Match finder options + mf->match_len_max = lz_options->match_len_max; + mf->nice_len = lz_options->nice_len; + + // cyclic_size has to stay smaller than 2 Gi. Note that this doesn't + // mean limiting dictionary size to less than 2 GiB. With a match + // finder that uses multibyte resolution (hashes start at e.g. every + // fourth byte), cyclic_size would stay below 2 Gi even when + // dictionary size is greater than 2 GiB. + // + // It would be possible to allow cyclic_size >= 2 Gi, but then we + // would need to be careful to use 64-bit types in various places + // (size_t could do since we would need bigger than 32-bit address + // space anyway). It would also require either zeroing a multigigabyte + // buffer at initialization (waste of time and RAM) or allow + // normalization in lz_encoder_mf.c to access uninitialized + // memory to keep the code simpler. The current way is simple and + // still allows pretty big dictionaries, so I don't expect these + // limits to change. + mf->cyclic_size = lz_options->dict_size + 1; + + // Validate the match finder ID and setup the function pointers. + switch (lz_options->match_finder) { +#ifdef HAVE_MF_HC3 + case LZMA_MF_HC3: + mf->find = &lzma_mf_hc3_find; + mf->skip = &lzma_mf_hc3_skip; + break; +#endif +#ifdef HAVE_MF_HC4 + case LZMA_MF_HC4: + mf->find = &lzma_mf_hc4_find; + mf->skip = &lzma_mf_hc4_skip; + break; +#endif +#ifdef HAVE_MF_BT2 + case LZMA_MF_BT2: + mf->find = &lzma_mf_bt2_find; + mf->skip = &lzma_mf_bt2_skip; + break; +#endif +#ifdef HAVE_MF_BT3 + case LZMA_MF_BT3: + mf->find = &lzma_mf_bt3_find; + mf->skip = &lzma_mf_bt3_skip; + break; +#endif +#ifdef HAVE_MF_BT4 + case LZMA_MF_BT4: + mf->find = &lzma_mf_bt4_find; + mf->skip = &lzma_mf_bt4_skip; + break; +#endif + + default: + return true; + } + + // Calculate the sizes of mf->hash and mf->son. + // + // NOTE: Since 5.3.5beta the LZMA encoder ensures that nice_len + // is big enough for the selected match finder. This makes it + // easier for applications as nice_len = 2 will always be accepted + // even though the effective value can be slightly bigger. + const uint32_t hash_bytes + = mf_get_hash_bytes(lz_options->match_finder); + assert(hash_bytes <= mf->nice_len); + + const bool is_bt = (lz_options->match_finder & 0x10) != 0; + uint32_t hs; + + if (hash_bytes == 2) { + hs = 0xFFFF; + } else { + // Round dictionary size up to the next 2^n - 1 so it can + // be used as a hash mask. + hs = lz_options->dict_size - 1; + hs |= hs >> 1; + hs |= hs >> 2; + hs |= hs >> 4; + hs |= hs >> 8; + hs >>= 1; + hs |= 0xFFFF; + + if (hs > (UINT32_C(1) << 24)) { + if (hash_bytes == 3) + hs = (UINT32_C(1) << 24) - 1; + else + hs >>= 1; + } + } + + mf->hash_mask = hs; + + ++hs; + if (hash_bytes > 2) + hs += HASH_2_SIZE; + if (hash_bytes > 3) + hs += HASH_3_SIZE; +/* + No match finder uses this at the moment. + if (mf->hash_bytes > 4) + hs += HASH_4_SIZE; +*/ + + const uint32_t old_hash_count = mf->hash_count; + const uint32_t old_sons_count = mf->sons_count; + mf->hash_count = hs; + mf->sons_count = mf->cyclic_size; + if (is_bt) + mf->sons_count *= 2; + + // Deallocate the old hash array if it exists and has different size + // than what is needed now. + if (old_hash_count != mf->hash_count + || old_sons_count != mf->sons_count) { + lzma_free(mf->hash, allocator); + mf->hash = NULL; + + lzma_free(mf->son, allocator); + mf->son = NULL; + } + + // Maximum number of match finder cycles + mf->depth = lz_options->depth; + if (mf->depth == 0) { + if (is_bt) + mf->depth = 16 + mf->nice_len / 2; + else + mf->depth = 4 + mf->nice_len / 4; + } + + return false; +} + + +static bool +lz_encoder_init(lzma_mf *mf, const lzma_allocator *allocator, + const lzma_lz_options *lz_options) +{ + // Allocate the history buffer. + if (mf->buffer == NULL) { + // lzma_memcmplen() is used for the dictionary buffer + // so we need to allocate a few extra bytes to prevent + // it from reading past the end of the buffer. + mf->buffer = lzma_alloc(mf->size + LZMA_MEMCMPLEN_EXTRA, + allocator); + if (mf->buffer == NULL) + return true; + + // Keep Valgrind happy with lzma_memcmplen() and initialize + // the extra bytes whose value may get read but which will + // effectively get ignored. + memzero(mf->buffer + mf->size, LZMA_MEMCMPLEN_EXTRA); + } + + // Use cyclic_size as initial mf->offset. This allows + // avoiding a few branches in the match finders. The downside is + // that match finder needs to be normalized more often, which may + // hurt performance with huge dictionaries. + mf->offset = mf->cyclic_size; + mf->read_pos = 0; + mf->read_ahead = 0; + mf->read_limit = 0; + mf->write_pos = 0; + mf->pending = 0; + +#if UINT32_MAX >= SIZE_MAX / 4 + // Check for integer overflow. (Huge dictionaries are not + // possible on 32-bit CPU.) + if (mf->hash_count > SIZE_MAX / sizeof(uint32_t) + || mf->sons_count > SIZE_MAX / sizeof(uint32_t)) + return true; +#endif + + // Allocate and initialize the hash table. Since EMPTY_HASH_VALUE + // is zero, we can use lzma_alloc_zero() or memzero() for mf->hash. + // + // We don't need to initialize mf->son, but not doing that may + // make Valgrind complain in normalization (see normalize() in + // lz_encoder_mf.c). Skipping the initialization is *very* good + // when big dictionary is used but only small amount of data gets + // actually compressed: most of the mf->son won't get actually + // allocated by the kernel, so we avoid wasting RAM and improve + // initialization speed a lot. + if (mf->hash == NULL) { + mf->hash = lzma_alloc_zero(mf->hash_count * sizeof(uint32_t), + allocator); + mf->son = lzma_alloc(mf->sons_count * sizeof(uint32_t), + allocator); + + if (mf->hash == NULL || mf->son == NULL) { + lzma_free(mf->hash, allocator); + mf->hash = NULL; + + lzma_free(mf->son, allocator); + mf->son = NULL; + + return true; + } + } else { +/* + for (uint32_t i = 0; i < mf->hash_count; ++i) + mf->hash[i] = EMPTY_HASH_VALUE; +*/ + memzero(mf->hash, mf->hash_count * sizeof(uint32_t)); + } + + mf->cyclic_pos = 0; + + // Handle preset dictionary. + if (lz_options->preset_dict != NULL + && lz_options->preset_dict_size > 0) { + // If the preset dictionary is bigger than the actual + // dictionary, use only the tail. + mf->write_pos = my_min(lz_options->preset_dict_size, mf->size); + memcpy(mf->buffer, lz_options->preset_dict + + lz_options->preset_dict_size - mf->write_pos, + mf->write_pos); + mf->action = LZMA_SYNC_FLUSH; + mf->skip(mf, mf->write_pos); + } + + mf->action = LZMA_RUN; + + return false; +} + + +extern uint64_t +lzma_lz_encoder_memusage(const lzma_lz_options *lz_options) +{ + // Old buffers must not exist when calling lz_encoder_prepare(). + lzma_mf mf = { + .buffer = NULL, + .hash = NULL, + .son = NULL, + .hash_count = 0, + .sons_count = 0, + }; + + // Setup the size information into mf. + if (lz_encoder_prepare(&mf, NULL, lz_options)) + return UINT64_MAX; + + // Calculate the memory usage. + return ((uint64_t)(mf.hash_count) + mf.sons_count) * sizeof(uint32_t) + + mf.size + sizeof(lzma_coder); +} + + +static void +lz_encoder_end(void *coder_ptr, const lzma_allocator *allocator) +{ + lzma_coder *coder = coder_ptr; + + lzma_next_end(&coder->next, allocator); + + lzma_free(coder->mf.son, allocator); + lzma_free(coder->mf.hash, allocator); + lzma_free(coder->mf.buffer, allocator); + + if (coder->lz.end != NULL) + coder->lz.end(coder->lz.coder, allocator); + else + lzma_free(coder->lz.coder, allocator); + + lzma_free(coder, allocator); + return; +} + + +static lzma_ret +lz_encoder_update(void *coder_ptr, const lzma_allocator *allocator, + const lzma_filter *filters_null lzma_attribute((__unused__)), + const lzma_filter *reversed_filters) +{ + lzma_coder *coder = coder_ptr; + + if (coder->lz.options_update == NULL) + return LZMA_PROG_ERROR; + + return_if_error(coder->lz.options_update( + coder->lz.coder, reversed_filters)); + + return lzma_next_filter_update( + &coder->next, allocator, reversed_filters + 1); +} + + +static lzma_ret +lz_encoder_set_out_limit(void *coder_ptr, uint64_t *uncomp_size, + uint64_t out_limit) +{ + lzma_coder *coder = coder_ptr; + + // This is supported only if there are no other filters chained. + if (coder->next.code == NULL && coder->lz.set_out_limit != NULL) + return coder->lz.set_out_limit( + coder->lz.coder, uncomp_size, out_limit); + + return LZMA_OPTIONS_ERROR; +} + + +extern lzma_ret +lzma_lz_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters, + lzma_ret (*lz_init)(lzma_lz_encoder *lz, + const lzma_allocator *allocator, + lzma_vli id, const void *options, + lzma_lz_options *lz_options)) +{ +#if defined(HAVE_SMALL) && !defined(HAVE_FUNC_ATTRIBUTE_CONSTRUCTOR) + // We need that the CRC32 table has been initialized. + lzma_crc32_init(); +#endif + + // Allocate and initialize the base data structure. + lzma_coder *coder = next->coder; + if (coder == NULL) { + coder = lzma_alloc(sizeof(lzma_coder), allocator); + if (coder == NULL) + return LZMA_MEM_ERROR; + + next->coder = coder; + next->code = &lz_encode; + next->end = &lz_encoder_end; + next->update = &lz_encoder_update; + next->set_out_limit = &lz_encoder_set_out_limit; + + coder->lz.coder = NULL; + coder->lz.code = NULL; + coder->lz.end = NULL; + + // mf.size is initialized to silence Valgrind + // when used on optimized binaries (GCC may reorder + // code in a way that Valgrind gets unhappy). + coder->mf.buffer = NULL; + coder->mf.size = 0; + coder->mf.hash = NULL; + coder->mf.son = NULL; + coder->mf.hash_count = 0; + coder->mf.sons_count = 0; + + coder->next = LZMA_NEXT_CODER_INIT; + } + + // Initialize the LZ-based encoder. + lzma_lz_options lz_options; + return_if_error(lz_init(&coder->lz, allocator, + filters[0].id, filters[0].options, &lz_options)); + + // Setup the size information into coder->mf and deallocate + // old buffers if they have wrong size. + if (lz_encoder_prepare(&coder->mf, allocator, &lz_options)) + return LZMA_OPTIONS_ERROR; + + // Allocate new buffers if needed, and do the rest of + // the initialization. + if (lz_encoder_init(&coder->mf, allocator, &lz_options)) + return LZMA_MEM_ERROR; + + // Initialize the next filter in the chain, if any. + return lzma_next_filter_init(&coder->next, allocator, filters + 1); +} + + +extern LZMA_API(lzma_bool) +lzma_mf_is_supported(lzma_match_finder mf) +{ + switch (mf) { +#ifdef HAVE_MF_HC3 + case LZMA_MF_HC3: + return true; +#endif +#ifdef HAVE_MF_HC4 + case LZMA_MF_HC4: + return true; +#endif +#ifdef HAVE_MF_BT2 + case LZMA_MF_BT2: + return true; +#endif +#ifdef HAVE_MF_BT3 + case LZMA_MF_BT3: + return true; +#endif +#ifdef HAVE_MF_BT4 + case LZMA_MF_BT4: + return true; +#endif + default: + return false; + } +} diff --git a/src/liblzma/lz/lz_encoder.h b/src/liblzma/lz/lz_encoder.h new file mode 100644 index 0000000..ffcba02 --- /dev/null +++ b/src/liblzma/lz/lz_encoder.h @@ -0,0 +1,341 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lz_encoder.h +/// \brief LZ in window and match finder API +/// +// 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_ENCODER_H +#define LZMA_LZ_ENCODER_H + +#include "common.h" + + +/// A table of these is used by the LZ-based encoder to hold +/// the length-distance pairs found by the match finder. +typedef struct { + uint32_t len; + uint32_t dist; +} lzma_match; + + +typedef struct lzma_mf_s lzma_mf; +struct lzma_mf_s { + /////////////// + // In Window // + /////////////// + + /// Pointer to buffer with data to be compressed + uint8_t *buffer; + + /// Total size of the allocated buffer (that is, including all + /// the extra space) + uint32_t size; + + /// Number of bytes that must be kept available in our input history. + /// That is, once keep_size_before bytes have been processed, + /// buffer[read_pos - keep_size_before] is the oldest byte that + /// must be available for reading. + uint32_t keep_size_before; + + /// Number of bytes that must be kept in buffer after read_pos. + /// That is, read_pos <= write_pos - keep_size_after as long as + /// action is LZMA_RUN; when action != LZMA_RUN, read_pos is allowed + /// to reach write_pos so that the last bytes get encoded too. + uint32_t keep_size_after; + + /// Match finders store locations of matches using 32-bit integers. + /// To avoid adjusting several megabytes of integers every time the + /// input window is moved with move_window, we only adjust the + /// offset of the buffer. Thus, buffer[value_in_hash_table - offset] + /// is the byte pointed by value_in_hash_table. + uint32_t offset; + + /// buffer[read_pos] is the next byte to run through the match + /// finder. This is incremented in the match finder once the byte + /// has been processed. + uint32_t read_pos; + + /// Number of bytes that have been ran through the match finder, but + /// which haven't been encoded by the LZ-based encoder yet. + uint32_t read_ahead; + + /// As long as read_pos is less than read_limit, there is enough + /// input available in buffer for at least one encoding loop. + /// + /// Because of the stateful API, read_limit may and will get greater + /// than read_pos quite often. This is taken into account when + /// calculating the value for keep_size_after. + uint32_t read_limit; + + /// buffer[write_pos] is the first byte that doesn't contain valid + /// uncompressed data; that is, the next input byte will be copied + /// to buffer[write_pos]. + uint32_t write_pos; + + /// Number of bytes not hashed before read_pos. This is needed to + /// restart the match finder after LZMA_SYNC_FLUSH. + uint32_t pending; + + ////////////////// + // Match Finder // + ////////////////// + + /// Find matches. Returns the number of distance-length pairs written + /// to the matches array. This is called only via lzma_mf_find(). + uint32_t (*find)(lzma_mf *mf, lzma_match *matches); + + /// Skips num bytes. This is like find() but doesn't make the + /// distance-length pairs available, thus being a little faster. + /// This is called only via mf_skip(). + void (*skip)(lzma_mf *mf, uint32_t num); + + uint32_t *hash; + uint32_t *son; + uint32_t cyclic_pos; + uint32_t cyclic_size; // Must be dictionary size + 1. + uint32_t hash_mask; + + /// Maximum number of loops in the match finder + uint32_t depth; + + /// Maximum length of a match that the match finder will try to find. + uint32_t nice_len; + + /// Maximum length of a match supported by the LZ-based encoder. + /// If the longest match found by the match finder is nice_len, + /// mf_find() tries to expand it up to match_len_max bytes. + uint32_t match_len_max; + + /// When running out of input, binary tree match finders need to know + /// if it is due to flushing or finishing. The action is used also + /// by the LZ-based encoders themselves. + lzma_action action; + + /// Number of elements in hash[] + uint32_t hash_count; + + /// Number of elements in son[] + uint32_t sons_count; +}; + + +typedef struct { + /// Extra amount of data to keep available before the "actual" + /// dictionary. + size_t before_size; + + /// Size of the history buffer + size_t dict_size; + + /// Extra amount of data to keep available after the "actual" + /// dictionary. + size_t after_size; + + /// Maximum length of a match that the LZ-based encoder can accept. + /// This is used to extend matches of length nice_len to the + /// maximum possible length. + size_t match_len_max; + + /// Match finder will search matches up to this length. + /// This must be less than or equal to match_len_max. + size_t nice_len; + + /// Type of the match finder to use + lzma_match_finder match_finder; + + /// Maximum search depth + uint32_t depth; + + /// TODO: Comment + const uint8_t *preset_dict; + + uint32_t preset_dict_size; + +} lzma_lz_options; + + +// The total usable buffer space at any moment outside the match finder: +// before_size + dict_size + after_size + match_len_max +// +// In reality, there's some extra space allocated to prevent the number of +// memmove() calls reasonable. The bigger the dict_size is, the bigger +// this extra buffer will be since with bigger dictionaries memmove() would +// also take longer. +// +// A single encoder loop in the LZ-based encoder may call the match finder +// (mf_find() or mf_skip()) at most after_size times. In other words, +// a single encoder loop may increment lzma_mf.read_pos at most after_size +// times. Since matches are looked up to +// lzma_mf.buffer[lzma_mf.read_pos + match_len_max - 1], the total +// amount of extra buffer needed after dict_size becomes +// after_size + match_len_max. +// +// before_size has two uses. The first one is to keep literals available +// in cases when the LZ-based encoder has made some read ahead. +// TODO: Maybe this could be changed by making the LZ-based encoders to +// store the actual literals as they do with length-distance pairs. +// +// Algorithms such as LZMA2 first try to compress a chunk, and then check +// if the encoded result is smaller than the uncompressed one. If the chunk +// was incompressible, it is better to store it in uncompressed form in +// the output stream. To do this, the whole uncompressed chunk has to be +// still available in the history buffer. before_size achieves that. + + +typedef struct { + /// Data specific to the LZ-based encoder + void *coder; + + /// Function to encode from *dict to out[] + lzma_ret (*code)(void *coder, + lzma_mf *restrict mf, uint8_t *restrict out, + size_t *restrict out_pos, size_t out_size); + + /// Free allocated resources + void (*end)(void *coder, const lzma_allocator *allocator); + + /// Update the options in the middle of the encoding. + lzma_ret (*options_update)(void *coder, const lzma_filter *filter); + + /// Set maximum allowed output size + lzma_ret (*set_out_limit)(void *coder, uint64_t *uncomp_size, + uint64_t out_limit); + +} lzma_lz_encoder; + + +// Basic steps: +// 1. Input gets copied into the dictionary. +// 2. Data in dictionary gets run through the match finder byte by byte. +// 3. The literals and matches are encoded using e.g. LZMA. +// +// The bytes that have been ran through the match finder, but not encoded yet, +// are called `read ahead'. + + +/// Get how many bytes the match finder hashes in its initial step. +/// This is also the minimum nice_len value with the match finder. +static inline uint32_t +mf_get_hash_bytes(lzma_match_finder match_finder) +{ + return (uint32_t)match_finder & 0x0F; +} + + +/// Get pointer to the first byte not ran through the match finder +static inline const uint8_t * +mf_ptr(const lzma_mf *mf) +{ + return mf->buffer + mf->read_pos; +} + + +/// Get the number of bytes that haven't been ran through the match finder yet. +static inline uint32_t +mf_avail(const lzma_mf *mf) +{ + return mf->write_pos - mf->read_pos; +} + + +/// Get the number of bytes that haven't been encoded yet (some of these +/// bytes may have been ran through the match finder though). +static inline uint32_t +mf_unencoded(const lzma_mf *mf) +{ + return mf->write_pos - mf->read_pos + mf->read_ahead; +} + + +/// Calculate the absolute offset from the beginning of the most recent +/// dictionary reset. Only the lowest four bits are important, so there's no +/// problem that we don't know the 64-bit size of the data encoded so far. +/// +/// NOTE: When moving the input window, we need to do it so that the lowest +/// bits of dict->read_pos are not modified to keep this macro working +/// as intended. +static inline uint32_t +mf_position(const lzma_mf *mf) +{ + return mf->read_pos - mf->read_ahead; +} + + +/// Since everything else begins with mf_, use it also for lzma_mf_find(). +#define mf_find lzma_mf_find + + +/// Skip the given number of bytes. This is used when a good match was found. +/// For example, if mf_find() finds a match of 200 bytes long, the first byte +/// of that match was already consumed by mf_find(), and the rest 199 bytes +/// have to be skipped with mf_skip(mf, 199). +static inline void +mf_skip(lzma_mf *mf, uint32_t amount) +{ + if (amount != 0) { + mf->skip(mf, amount); + mf->read_ahead += amount; + } +} + + +/// Copies at most *left number of bytes from the history buffer +/// to out[]. This is needed by LZMA2 to encode uncompressed chunks. +static inline void +mf_read(lzma_mf *mf, uint8_t *out, size_t *out_pos, size_t out_size, + size_t *left) +{ + const size_t out_avail = out_size - *out_pos; + const size_t copy_size = my_min(out_avail, *left); + + assert(mf->read_ahead == 0); + assert(mf->read_pos >= *left); + + memcpy(out + *out_pos, mf->buffer + mf->read_pos - *left, + copy_size); + + *out_pos += copy_size; + *left -= copy_size; + return; +} + + +extern lzma_ret lzma_lz_encoder_init( + lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters, + lzma_ret (*lz_init)(lzma_lz_encoder *lz, + const lzma_allocator *allocator, + lzma_vli id, const void *options, + lzma_lz_options *lz_options)); + + +extern uint64_t lzma_lz_encoder_memusage(const lzma_lz_options *lz_options); + + +// These are only for LZ encoder's internal use. +extern uint32_t lzma_mf_find( + lzma_mf *mf, uint32_t *count, lzma_match *matches); + +extern uint32_t lzma_mf_hc3_find(lzma_mf *dict, lzma_match *matches); +extern void lzma_mf_hc3_skip(lzma_mf *dict, uint32_t amount); + +extern uint32_t lzma_mf_hc4_find(lzma_mf *dict, lzma_match *matches); +extern void lzma_mf_hc4_skip(lzma_mf *dict, uint32_t amount); + +extern uint32_t lzma_mf_bt2_find(lzma_mf *dict, lzma_match *matches); +extern void lzma_mf_bt2_skip(lzma_mf *dict, uint32_t amount); + +extern uint32_t lzma_mf_bt3_find(lzma_mf *dict, lzma_match *matches); +extern void lzma_mf_bt3_skip(lzma_mf *dict, uint32_t amount); + +extern uint32_t lzma_mf_bt4_find(lzma_mf *dict, lzma_match *matches); +extern void lzma_mf_bt4_skip(lzma_mf *dict, uint32_t amount); + +#endif diff --git a/src/liblzma/lz/lz_encoder_hash.h b/src/liblzma/lz/lz_encoder_hash.h new file mode 100644 index 0000000..4d9971a --- /dev/null +++ b/src/liblzma/lz/lz_encoder_hash.h @@ -0,0 +1,109 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lz_encoder_hash.h +/// \brief Hash macros for match finders +// +// Author: Igor Pavlov +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#ifndef LZMA_LZ_ENCODER_HASH_H +#define LZMA_LZ_ENCODER_HASH_H + +#if defined(WORDS_BIGENDIAN) && !defined(HAVE_SMALL) + // This is to make liblzma produce the same output on big endian + // systems that it does on little endian systems. lz_encoder.c + // takes care of including the actual table. + lzma_attr_visibility_hidden + extern const uint32_t lzma_lz_hash_table[256]; +# define hash_table lzma_lz_hash_table +#else +# include "check.h" +# define hash_table lzma_crc32_table[0] +#endif + +#define HASH_2_SIZE (UINT32_C(1) << 10) +#define HASH_3_SIZE (UINT32_C(1) << 16) +#define HASH_4_SIZE (UINT32_C(1) << 20) + +#define HASH_2_MASK (HASH_2_SIZE - 1) +#define HASH_3_MASK (HASH_3_SIZE - 1) +#define HASH_4_MASK (HASH_4_SIZE - 1) + +#define FIX_3_HASH_SIZE (HASH_2_SIZE) +#define FIX_4_HASH_SIZE (HASH_2_SIZE + HASH_3_SIZE) +#define FIX_5_HASH_SIZE (HASH_2_SIZE + HASH_3_SIZE + HASH_4_SIZE) + +// Endianness doesn't matter in hash_2_calc() (no effect on the output). +#ifdef TUKLIB_FAST_UNALIGNED_ACCESS +# define hash_2_calc() \ + const uint32_t hash_value = read16ne(cur) +#else +# define hash_2_calc() \ + const uint32_t hash_value \ + = (uint32_t)(cur[0]) | ((uint32_t)(cur[1]) << 8) +#endif + +#define hash_3_calc() \ + const uint32_t temp = hash_table[cur[0]] ^ cur[1]; \ + const uint32_t hash_2_value = temp & HASH_2_MASK; \ + const uint32_t hash_value \ + = (temp ^ ((uint32_t)(cur[2]) << 8)) & mf->hash_mask + +#define hash_4_calc() \ + const uint32_t temp = hash_table[cur[0]] ^ cur[1]; \ + const uint32_t hash_2_value = temp & HASH_2_MASK; \ + const uint32_t hash_3_value \ + = (temp ^ ((uint32_t)(cur[2]) << 8)) & HASH_3_MASK; \ + const uint32_t hash_value = (temp ^ ((uint32_t)(cur[2]) << 8) \ + ^ (hash_table[cur[3]] << 5)) & mf->hash_mask + + +// The following are not currently used. + +#define hash_5_calc() \ + const uint32_t temp = hash_table[cur[0]] ^ cur[1]; \ + const uint32_t hash_2_value = temp & HASH_2_MASK; \ + const uint32_t hash_3_value \ + = (temp ^ ((uint32_t)(cur[2]) << 8)) & HASH_3_MASK; \ + uint32_t hash_4_value = (temp ^ ((uint32_t)(cur[2]) << 8) ^ \ + ^ hash_table[cur[3]] << 5); \ + const uint32_t hash_value \ + = (hash_4_value ^ (hash_table[cur[4]] << 3)) \ + & mf->hash_mask; \ + hash_4_value &= HASH_4_MASK + +/* +#define hash_zip_calc() \ + const uint32_t hash_value \ + = (((uint32_t)(cur[0]) | ((uint32_t)(cur[1]) << 8)) \ + ^ hash_table[cur[2]]) & 0xFFFF +*/ + +#define hash_zip_calc() \ + const uint32_t hash_value \ + = (((uint32_t)(cur[2]) | ((uint32_t)(cur[0]) << 8)) \ + ^ hash_table[cur[1]]) & 0xFFFF + +#define mt_hash_2_calc() \ + const uint32_t hash_2_value \ + = (hash_table[cur[0]] ^ cur[1]) & HASH_2_MASK + +#define mt_hash_3_calc() \ + const uint32_t temp = hash_table[cur[0]] ^ cur[1]; \ + const uint32_t hash_2_value = temp & HASH_2_MASK; \ + const uint32_t hash_3_value \ + = (temp ^ ((uint32_t)(cur[2]) << 8)) & HASH_3_MASK + +#define mt_hash_4_calc() \ + const uint32_t temp = hash_table[cur[0]] ^ cur[1]; \ + const uint32_t hash_2_value = temp & HASH_2_MASK; \ + const uint32_t hash_3_value \ + = (temp ^ ((uint32_t)(cur[2]) << 8)) & HASH_3_MASK; \ + const uint32_t hash_4_value = (temp ^ ((uint32_t)(cur[2]) << 8) ^ \ + (hash_table[cur[3]] << 5)) & HASH_4_MASK + +#endif diff --git a/src/liblzma/lz/lz_encoder_hash_table.h b/src/liblzma/lz/lz_encoder_hash_table.h new file mode 100644 index 0000000..8c51717 --- /dev/null +++ b/src/liblzma/lz/lz_encoder_hash_table.h @@ -0,0 +1,68 @@ +/* This file has been automatically generated by crc32_tablegen.c. */ + +const uint32_t lzma_lz_hash_table[256] = { + 0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, + 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3, + 0x0EDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, + 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91, + 0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, + 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7, + 0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, + 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5, + 0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, + 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B, + 0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, + 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59, + 0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, + 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F, + 0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, + 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D, + 0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, + 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433, + 0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, + 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01, + 0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, + 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457, + 0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, + 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65, + 0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, + 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB, + 0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, + 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9, + 0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, + 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F, + 0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, + 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD, + 0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, + 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683, + 0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, + 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1, + 0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, + 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7, + 0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, + 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5, + 0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, + 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B, + 0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, + 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79, + 0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, + 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F, + 0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, + 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D, + 0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, + 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713, + 0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, + 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21, + 0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, + 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777, + 0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, + 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45, + 0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, + 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB, + 0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, + 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9, + 0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, + 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF, + 0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, + 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D +}; diff --git a/src/liblzma/lz/lz_encoder_mf.c b/src/liblzma/lz/lz_encoder_mf.c new file mode 100644 index 0000000..1fdc2d7 --- /dev/null +++ b/src/liblzma/lz/lz_encoder_mf.c @@ -0,0 +1,745 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lz_encoder_mf.c +/// \brief Match finders +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lz_encoder.h" +#include "lz_encoder_hash.h" +#include "memcmplen.h" + + +/// \brief Find matches starting from the current byte +/// +/// \return The length of the longest match found +extern uint32_t +lzma_mf_find(lzma_mf *mf, uint32_t *count_ptr, lzma_match *matches) +{ + // Call the match finder. It returns the number of length-distance + // pairs found. + // FIXME: Minimum count is zero, what _exactly_ is the maximum? + const uint32_t count = mf->find(mf, matches); + + // Length of the longest match; assume that no matches were found + // and thus the maximum length is zero. + uint32_t len_best = 0; + + if (count > 0) { +#ifndef NDEBUG + // Validate the matches. + for (uint32_t i = 0; i < count; ++i) { + assert(matches[i].len <= mf->nice_len); + assert(matches[i].dist < mf->read_pos); + assert(memcmp(mf_ptr(mf) - 1, + mf_ptr(mf) - matches[i].dist - 2, + matches[i].len) == 0); + } +#endif + + // The last used element in the array contains + // the longest match. + len_best = matches[count - 1].len; + + // If a match of maximum search length was found, try to + // extend the match to maximum possible length. + if (len_best == mf->nice_len) { + // The limit for the match length is either the + // maximum match length supported by the LZ-based + // encoder or the number of bytes left in the + // dictionary, whichever is smaller. + uint32_t limit = mf_avail(mf) + 1; + if (limit > mf->match_len_max) + limit = mf->match_len_max; + + // Pointer to the byte we just ran through + // the match finder. + const uint8_t *p1 = mf_ptr(mf) - 1; + + // Pointer to the beginning of the match. We need -1 + // here because the match distances are zero based. + const uint8_t *p2 = p1 - matches[count - 1].dist - 1; + + len_best = lzma_memcmplen(p1, p2, len_best, limit); + } + } + + *count_ptr = count; + + // Finally update the read position to indicate that match finder was + // run for this dictionary offset. + ++mf->read_ahead; + + return len_best; +} + + +/// Hash value to indicate unused element in the hash. Since we start the +/// positions from dict_size + 1, zero is always too far to qualify +/// as usable match position. +#define EMPTY_HASH_VALUE 0 + + +/// Normalization must be done when lzma_mf.offset + lzma_mf.read_pos +/// reaches MUST_NORMALIZE_POS. +#define MUST_NORMALIZE_POS UINT32_MAX + + +/// \brief Normalizes hash values +/// +/// The hash arrays store positions of match candidates. The positions are +/// relative to an arbitrary offset that is not the same as the absolute +/// offset in the input stream. The relative position of the current byte +/// is lzma_mf.offset + lzma_mf.read_pos. The distances of the matches are +/// the differences of the current read position and the position found from +/// the hash. +/// +/// To prevent integer overflows of the offsets stored in the hash arrays, +/// we need to "normalize" the stored values now and then. During the +/// normalization, we drop values that indicate distance greater than the +/// dictionary size, thus making space for new values. +static void +normalize(lzma_mf *mf) +{ + assert(mf->read_pos + mf->offset == MUST_NORMALIZE_POS); + + // In future we may not want to touch the lowest bits, because there + // may be match finders that use larger resolution than one byte. + const uint32_t subvalue + = (MUST_NORMALIZE_POS - mf->cyclic_size); + // & ~((UINT32_C(1) << 10) - 1); + + for (uint32_t i = 0; i < mf->hash_count; ++i) { + // If the distance is greater than the dictionary size, + // we can simply mark the hash element as empty. + if (mf->hash[i] <= subvalue) + mf->hash[i] = EMPTY_HASH_VALUE; + else + mf->hash[i] -= subvalue; + } + + for (uint32_t i = 0; i < mf->sons_count; ++i) { + // Do the same for mf->son. + // + // NOTE: There may be uninitialized elements in mf->son. + // Valgrind may complain that the "if" below depends on + // an uninitialized value. In this case it is safe to ignore + // the warning. See also the comments in lz_encoder_init() + // in lz_encoder.c. + if (mf->son[i] <= subvalue) + mf->son[i] = EMPTY_HASH_VALUE; + else + mf->son[i] -= subvalue; + } + + // Update offset to match the new locations. + mf->offset -= subvalue; + + return; +} + + +/// Mark the current byte as processed from point of view of the match finder. +static void +move_pos(lzma_mf *mf) +{ + if (++mf->cyclic_pos == mf->cyclic_size) + mf->cyclic_pos = 0; + + ++mf->read_pos; + assert(mf->read_pos <= mf->write_pos); + + if (unlikely(mf->read_pos + mf->offset == UINT32_MAX)) + normalize(mf); +} + + +/// When flushing, we cannot run the match finder unless there is nice_len +/// bytes available in the dictionary. Instead, we skip running the match +/// finder (indicating that no match was found), and count how many bytes we +/// have ignored this way. +/// +/// When new data is given after the flushing was completed, the match finder +/// is restarted by rewinding mf->read_pos backwards by mf->pending. Then +/// the missed bytes are added to the hash using the match finder's skip +/// function (with small amount of input, it may start using mf->pending +/// again if flushing). +/// +/// Due to this rewinding, we don't touch cyclic_pos or test for +/// normalization. It will be done when the match finder's skip function +/// catches up after a flush. +static void +move_pending(lzma_mf *mf) +{ + ++mf->read_pos; + assert(mf->read_pos <= mf->write_pos); + ++mf->pending; +} + + +/// Calculate len_limit and determine if there is enough input to run +/// the actual match finder code. Sets up "cur" and "pos". This macro +/// is used by all find functions and binary tree skip functions. Hash +/// chain skip function doesn't need len_limit so a simpler code is used +/// in them. +#define header(is_bt, len_min, ret_op) \ + uint32_t len_limit = mf_avail(mf); \ + if (mf->nice_len <= len_limit) { \ + len_limit = mf->nice_len; \ + } else if (len_limit < (len_min) \ + || (is_bt && mf->action == LZMA_SYNC_FLUSH)) { \ + assert(mf->action != LZMA_RUN); \ + move_pending(mf); \ + ret_op; \ + } \ + const uint8_t *cur = mf_ptr(mf); \ + const uint32_t pos = mf->read_pos + mf->offset + + +/// Header for find functions. "return 0" indicates that zero matches +/// were found. +#define header_find(is_bt, len_min) \ + header(is_bt, len_min, return 0); \ + uint32_t matches_count = 0 + + +/// Header for a loop in a skip function. "continue" tells to skip the rest +/// of the code in the loop. +#define header_skip(is_bt, len_min) \ + header(is_bt, len_min, continue) + + +/// Calls hc_find_func() or bt_find_func() and calculates the total number +/// of matches found. Updates the dictionary position and returns the number +/// of matches found. +#define call_find(func, len_best) \ +do { \ + matches_count = (uint32_t)(func(len_limit, pos, cur, cur_match, \ + mf->depth, mf->son, \ + mf->cyclic_pos, mf->cyclic_size, \ + matches + matches_count, len_best) \ + - matches); \ + move_pos(mf); \ + return matches_count; \ +} while (0) + + +//////////////// +// Hash Chain // +//////////////// + +#if defined(HAVE_MF_HC3) || defined(HAVE_MF_HC4) +/// +/// +/// \param len_limit Don't look for matches longer than len_limit. +/// \param pos lzma_mf.read_pos + lzma_mf.offset +/// \param cur Pointer to current byte (mf_ptr(mf)) +/// \param cur_match Start position of the current match candidate +/// \param depth Maximum length of the hash chain +/// \param son lzma_mf.son (contains the hash chain) +/// \param cyclic_pos lzma_mf.cyclic_pos +/// \param cyclic_size lzma_mf_cyclic_size +/// \param matches Array to hold the matches. +/// \param len_best The length of the longest match found so far. +static lzma_match * +hc_find_func( + const uint32_t len_limit, + const uint32_t pos, + const uint8_t *const cur, + uint32_t cur_match, + uint32_t depth, + uint32_t *const son, + const uint32_t cyclic_pos, + const uint32_t cyclic_size, + lzma_match *matches, + uint32_t len_best) +{ + son[cyclic_pos] = cur_match; + + while (true) { + const uint32_t delta = pos - cur_match; + if (depth-- == 0 || delta >= cyclic_size) + return matches; + + const uint8_t *const pb = cur - delta; + cur_match = son[cyclic_pos - delta + + (delta > cyclic_pos ? cyclic_size : 0)]; + + if (pb[len_best] == cur[len_best] && pb[0] == cur[0]) { + uint32_t len = lzma_memcmplen(pb, cur, 1, len_limit); + + if (len_best < len) { + len_best = len; + matches->len = len; + matches->dist = delta - 1; + ++matches; + + if (len == len_limit) + return matches; + } + } + } +} + + +#define hc_find(len_best) \ + call_find(hc_find_func, len_best) + + +#define hc_skip() \ +do { \ + mf->son[mf->cyclic_pos] = cur_match; \ + move_pos(mf); \ +} while (0) + +#endif + + +#ifdef HAVE_MF_HC3 +extern uint32_t +lzma_mf_hc3_find(lzma_mf *mf, lzma_match *matches) +{ + header_find(false, 3); + + hash_3_calc(); + + const uint32_t delta2 = pos - mf->hash[hash_2_value]; + const uint32_t cur_match = mf->hash[FIX_3_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_value] = pos; + + uint32_t len_best = 2; + + if (delta2 < mf->cyclic_size && *(cur - delta2) == *cur) { + len_best = lzma_memcmplen(cur - delta2, cur, + len_best, len_limit); + + matches[0].len = len_best; + matches[0].dist = delta2 - 1; + matches_count = 1; + + if (len_best == len_limit) { + hc_skip(); + return 1; // matches_count + } + } + + hc_find(len_best); +} + + +extern void +lzma_mf_hc3_skip(lzma_mf *mf, uint32_t amount) +{ + do { + if (mf_avail(mf) < 3) { + move_pending(mf); + continue; + } + + const uint8_t *cur = mf_ptr(mf); + const uint32_t pos = mf->read_pos + mf->offset; + + hash_3_calc(); + + const uint32_t cur_match + = mf->hash[FIX_3_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_value] = pos; + + hc_skip(); + + } while (--amount != 0); +} +#endif + + +#ifdef HAVE_MF_HC4 +extern uint32_t +lzma_mf_hc4_find(lzma_mf *mf, lzma_match *matches) +{ + header_find(false, 4); + + hash_4_calc(); + + uint32_t delta2 = pos - mf->hash[hash_2_value]; + const uint32_t delta3 + = pos - mf->hash[FIX_3_HASH_SIZE + hash_3_value]; + const uint32_t cur_match = mf->hash[FIX_4_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value ] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_3_value] = pos; + mf->hash[FIX_4_HASH_SIZE + hash_value] = pos; + + uint32_t len_best = 1; + + if (delta2 < mf->cyclic_size && *(cur - delta2) == *cur) { + len_best = 2; + matches[0].len = 2; + matches[0].dist = delta2 - 1; + matches_count = 1; + } + + if (delta2 != delta3 && delta3 < mf->cyclic_size + && *(cur - delta3) == *cur) { + len_best = 3; + matches[matches_count++].dist = delta3 - 1; + delta2 = delta3; + } + + if (matches_count != 0) { + len_best = lzma_memcmplen(cur - delta2, cur, + len_best, len_limit); + + matches[matches_count - 1].len = len_best; + + if (len_best == len_limit) { + hc_skip(); + return matches_count; + } + } + + if (len_best < 3) + len_best = 3; + + hc_find(len_best); +} + + +extern void +lzma_mf_hc4_skip(lzma_mf *mf, uint32_t amount) +{ + do { + if (mf_avail(mf) < 4) { + move_pending(mf); + continue; + } + + const uint8_t *cur = mf_ptr(mf); + const uint32_t pos = mf->read_pos + mf->offset; + + hash_4_calc(); + + const uint32_t cur_match + = mf->hash[FIX_4_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_3_value] = pos; + mf->hash[FIX_4_HASH_SIZE + hash_value] = pos; + + hc_skip(); + + } while (--amount != 0); +} +#endif + + +///////////////// +// Binary Tree // +///////////////// + +#if defined(HAVE_MF_BT2) || defined(HAVE_MF_BT3) || defined(HAVE_MF_BT4) +static lzma_match * +bt_find_func( + const uint32_t len_limit, + const uint32_t pos, + const uint8_t *const cur, + uint32_t cur_match, + uint32_t depth, + uint32_t *const son, + const uint32_t cyclic_pos, + const uint32_t cyclic_size, + lzma_match *matches, + uint32_t len_best) +{ + uint32_t *ptr0 = son + (cyclic_pos << 1) + 1; + uint32_t *ptr1 = son + (cyclic_pos << 1); + + uint32_t len0 = 0; + uint32_t len1 = 0; + + while (true) { + const uint32_t delta = pos - cur_match; + if (depth-- == 0 || delta >= cyclic_size) { + *ptr0 = EMPTY_HASH_VALUE; + *ptr1 = EMPTY_HASH_VALUE; + return matches; + } + + uint32_t *const pair = son + ((cyclic_pos - delta + + (delta > cyclic_pos ? cyclic_size : 0)) + << 1); + + const uint8_t *const pb = cur - delta; + uint32_t len = my_min(len0, len1); + + if (pb[len] == cur[len]) { + len = lzma_memcmplen(pb, cur, len + 1, len_limit); + + if (len_best < len) { + len_best = len; + matches->len = len; + matches->dist = delta - 1; + ++matches; + + if (len == len_limit) { + *ptr1 = pair[0]; + *ptr0 = pair[1]; + return matches; + } + } + } + + if (pb[len] < cur[len]) { + *ptr1 = cur_match; + ptr1 = pair + 1; + cur_match = *ptr1; + len1 = len; + } else { + *ptr0 = cur_match; + ptr0 = pair; + cur_match = *ptr0; + len0 = len; + } + } +} + + +static void +bt_skip_func( + const uint32_t len_limit, + const uint32_t pos, + const uint8_t *const cur, + uint32_t cur_match, + uint32_t depth, + uint32_t *const son, + const uint32_t cyclic_pos, + const uint32_t cyclic_size) +{ + uint32_t *ptr0 = son + (cyclic_pos << 1) + 1; + uint32_t *ptr1 = son + (cyclic_pos << 1); + + uint32_t len0 = 0; + uint32_t len1 = 0; + + while (true) { + const uint32_t delta = pos - cur_match; + if (depth-- == 0 || delta >= cyclic_size) { + *ptr0 = EMPTY_HASH_VALUE; + *ptr1 = EMPTY_HASH_VALUE; + return; + } + + uint32_t *pair = son + ((cyclic_pos - delta + + (delta > cyclic_pos ? cyclic_size : 0)) + << 1); + const uint8_t *pb = cur - delta; + uint32_t len = my_min(len0, len1); + + if (pb[len] == cur[len]) { + len = lzma_memcmplen(pb, cur, len + 1, len_limit); + + if (len == len_limit) { + *ptr1 = pair[0]; + *ptr0 = pair[1]; + return; + } + } + + if (pb[len] < cur[len]) { + *ptr1 = cur_match; + ptr1 = pair + 1; + cur_match = *ptr1; + len1 = len; + } else { + *ptr0 = cur_match; + ptr0 = pair; + cur_match = *ptr0; + len0 = len; + } + } +} + + +#define bt_find(len_best) \ + call_find(bt_find_func, len_best) + +#define bt_skip() \ +do { \ + bt_skip_func(len_limit, pos, cur, cur_match, mf->depth, \ + mf->son, mf->cyclic_pos, \ + mf->cyclic_size); \ + move_pos(mf); \ +} while (0) + +#endif + + +#ifdef HAVE_MF_BT2 +extern uint32_t +lzma_mf_bt2_find(lzma_mf *mf, lzma_match *matches) +{ + header_find(true, 2); + + hash_2_calc(); + + const uint32_t cur_match = mf->hash[hash_value]; + mf->hash[hash_value] = pos; + + bt_find(1); +} + + +extern void +lzma_mf_bt2_skip(lzma_mf *mf, uint32_t amount) +{ + do { + header_skip(true, 2); + + hash_2_calc(); + + const uint32_t cur_match = mf->hash[hash_value]; + mf->hash[hash_value] = pos; + + bt_skip(); + + } while (--amount != 0); +} +#endif + + +#ifdef HAVE_MF_BT3 +extern uint32_t +lzma_mf_bt3_find(lzma_mf *mf, lzma_match *matches) +{ + header_find(true, 3); + + hash_3_calc(); + + const uint32_t delta2 = pos - mf->hash[hash_2_value]; + const uint32_t cur_match = mf->hash[FIX_3_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_value] = pos; + + uint32_t len_best = 2; + + if (delta2 < mf->cyclic_size && *(cur - delta2) == *cur) { + len_best = lzma_memcmplen( + cur, cur - delta2, len_best, len_limit); + + matches[0].len = len_best; + matches[0].dist = delta2 - 1; + matches_count = 1; + + if (len_best == len_limit) { + bt_skip(); + return 1; // matches_count + } + } + + bt_find(len_best); +} + + +extern void +lzma_mf_bt3_skip(lzma_mf *mf, uint32_t amount) +{ + do { + header_skip(true, 3); + + hash_3_calc(); + + const uint32_t cur_match + = mf->hash[FIX_3_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_value] = pos; + + bt_skip(); + + } while (--amount != 0); +} +#endif + + +#ifdef HAVE_MF_BT4 +extern uint32_t +lzma_mf_bt4_find(lzma_mf *mf, lzma_match *matches) +{ + header_find(true, 4); + + hash_4_calc(); + + uint32_t delta2 = pos - mf->hash[hash_2_value]; + const uint32_t delta3 + = pos - mf->hash[FIX_3_HASH_SIZE + hash_3_value]; + const uint32_t cur_match = mf->hash[FIX_4_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_3_value] = pos; + mf->hash[FIX_4_HASH_SIZE + hash_value] = pos; + + uint32_t len_best = 1; + + if (delta2 < mf->cyclic_size && *(cur - delta2) == *cur) { + len_best = 2; + matches[0].len = 2; + matches[0].dist = delta2 - 1; + matches_count = 1; + } + + if (delta2 != delta3 && delta3 < mf->cyclic_size + && *(cur - delta3) == *cur) { + len_best = 3; + matches[matches_count++].dist = delta3 - 1; + delta2 = delta3; + } + + if (matches_count != 0) { + len_best = lzma_memcmplen( + cur, cur - delta2, len_best, len_limit); + + matches[matches_count - 1].len = len_best; + + if (len_best == len_limit) { + bt_skip(); + return matches_count; + } + } + + if (len_best < 3) + len_best = 3; + + bt_find(len_best); +} + + +extern void +lzma_mf_bt4_skip(lzma_mf *mf, uint32_t amount) +{ + do { + header_skip(true, 4); + + hash_4_calc(); + + const uint32_t cur_match + = mf->hash[FIX_4_HASH_SIZE + hash_value]; + + mf->hash[hash_2_value] = pos; + mf->hash[FIX_3_HASH_SIZE + hash_3_value] = pos; + mf->hash[FIX_4_HASH_SIZE + hash_value] = pos; + + bt_skip(); + + } while (--amount != 0); +} +#endif diff --git a/src/liblzma/lzma/Makefile.inc b/src/liblzma/lzma/Makefile.inc new file mode 100644 index 0000000..25440d8 --- /dev/null +++ b/src/liblzma/lzma/Makefile.inc @@ -0,0 +1,44 @@ +## +## Author: Lasse Collin +## +## This file has been put into the public domain. +## You can do whatever you want with this file. +## + +EXTRA_DIST += lzma/fastpos_tablegen.c + +liblzma_la_SOURCES += \ + lzma/lzma_common.h \ + lzma/lzma_encoder_presets.c + +if COND_ENCODER_LZMA1 +liblzma_la_SOURCES += \ + lzma/fastpos.h \ + lzma/lzma_encoder.h \ + lzma/lzma_encoder.c \ + lzma/lzma_encoder_private.h \ + lzma/lzma_encoder_optimum_fast.c \ + lzma/lzma_encoder_optimum_normal.c + +if !COND_SMALL +liblzma_la_SOURCES += lzma/fastpos_table.c +endif +endif + +if COND_DECODER_LZMA1 +liblzma_la_SOURCES += \ + lzma/lzma_decoder.c \ + lzma/lzma_decoder.h +endif + +if COND_ENCODER_LZMA2 +liblzma_la_SOURCES += \ + lzma/lzma2_encoder.c \ + lzma/lzma2_encoder.h +endif + +if COND_DECODER_LZMA2 +liblzma_la_SOURCES += \ + lzma/lzma2_decoder.c \ + lzma/lzma2_decoder.h +endif diff --git a/src/liblzma/lzma/fastpos.h b/src/liblzma/lzma/fastpos.h new file mode 100644 index 0000000..dbeb16f --- /dev/null +++ b/src/liblzma/lzma/fastpos.h @@ -0,0 +1,142 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file fastpos.h +/// \brief Kind of two-bit version of bit scan reverse +/// +// 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_FASTPOS_H +#define LZMA_FASTPOS_H + +// LZMA encodes match distances by storing the highest two bits using +// a six-bit value [0, 63], and then the missing lower bits. +// Dictionary size is also stored using this encoding in the .xz +// file format header. +// +// fastpos.h provides a way to quickly find out the correct six-bit +// values. The following table gives some examples of this encoding: +// +// dist return +// 0 0 +// 1 1 +// 2 2 +// 3 3 +// 4 4 +// 5 4 +// 6 5 +// 7 5 +// 8 6 +// 11 6 +// 12 7 +// ... ... +// 15 7 +// 16 8 +// 17 8 +// ... ... +// 23 8 +// 24 9 +// 25 9 +// ... ... +// +// +// Provided functions or macros +// ---------------------------- +// +// get_dist_slot(dist) is the basic version. get_dist_slot_2(dist) +// assumes that dist >= FULL_DISTANCES, thus the result is at least +// FULL_DISTANCES_BITS * 2. Using get_dist_slot(dist) instead of +// get_dist_slot_2(dist) would give the same result, but get_dist_slot_2(dist) +// should be tiny bit faster due to the assumption being made. +// +// +// Size vs. speed +// -------------- +// +// With some CPUs that have fast BSR (bit scan reverse) instruction, the +// size optimized version is slightly faster than the bigger table based +// approach. Such CPUs include Intel Pentium Pro, Pentium II, Pentium III +// and Core 2 (possibly others). AMD K7 seems to have slower BSR, but that +// would still have speed roughly comparable to the table version. Older +// x86 CPUs like the original Pentium have very slow BSR; on those systems +// the table version is a lot faster. +// +// On some CPUs, the table version is a lot faster when using position +// dependent code, but with position independent code the size optimized +// version is slightly faster. This occurs at least on 32-bit SPARC (no +// ASM optimizations). +// +// I'm making the table version the default, because that has good speed +// on all systems I have tried. The size optimized version is sometimes +// slightly faster, but sometimes it is a lot slower. + +#ifdef HAVE_SMALL +# define get_dist_slot(dist) \ + ((dist) <= 4 ? (dist) : get_dist_slot_2(dist)) + +static inline uint32_t +get_dist_slot_2(uint32_t dist) +{ + const uint32_t i = bsr32(dist); + return (i + i) + ((dist >> (i - 1)) & 1); +} + + +#else + +#define FASTPOS_BITS 13 + +lzma_attr_visibility_hidden +extern const uint8_t lzma_fastpos[1 << FASTPOS_BITS]; + + +#define fastpos_shift(extra, n) \ + ((extra) + (n) * (FASTPOS_BITS - 1)) + +#define fastpos_limit(extra, n) \ + (UINT32_C(1) << (FASTPOS_BITS + fastpos_shift(extra, n))) + +#define fastpos_result(dist, extra, n) \ + (uint32_t)(lzma_fastpos[(dist) >> fastpos_shift(extra, n)]) \ + + 2 * fastpos_shift(extra, n) + + +static inline uint32_t +get_dist_slot(uint32_t dist) +{ + // If it is small enough, we can pick the result directly from + // the precalculated table. + if (dist < fastpos_limit(0, 0)) + return lzma_fastpos[dist]; + + if (dist < fastpos_limit(0, 1)) + return fastpos_result(dist, 0, 1); + + return fastpos_result(dist, 0, 2); +} + + +#ifdef FULL_DISTANCES_BITS +static inline uint32_t +get_dist_slot_2(uint32_t dist) +{ + assert(dist >= FULL_DISTANCES); + + if (dist < fastpos_limit(FULL_DISTANCES_BITS - 1, 0)) + return fastpos_result(dist, FULL_DISTANCES_BITS - 1, 0); + + if (dist < fastpos_limit(FULL_DISTANCES_BITS - 1, 1)) + return fastpos_result(dist, FULL_DISTANCES_BITS - 1, 1); + + return fastpos_result(dist, FULL_DISTANCES_BITS - 1, 2); +} +#endif + +#endif + +#endif diff --git a/src/liblzma/lzma/fastpos_table.c b/src/liblzma/lzma/fastpos_table.c new file mode 100644 index 0000000..6a3ceac --- /dev/null +++ b/src/liblzma/lzma/fastpos_table.c @@ -0,0 +1,519 @@ +/* This file has been automatically generated by fastpos_tablegen.c. */ + +#include "common.h" +#include "fastpos.h" + +const uint8_t lzma_fastpos[1 << FASTPOS_BITS] = { + 0, 1, 2, 3, 4, 4, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, + 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, + 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25 +}; diff --git a/src/liblzma/lzma/fastpos_tablegen.c b/src/liblzma/lzma/fastpos_tablegen.c new file mode 100644 index 0000000..57ed150 --- /dev/null +++ b/src/liblzma/lzma/fastpos_tablegen.c @@ -0,0 +1,57 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file fastpos_tablegen.c +/// \brief Generates the lzma_fastpos[] lookup table +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include <inttypes.h> +#include <stdio.h> + +#define lzma_attr_visibility_hidden +#include "fastpos.h" + + +int +main(void) +{ + uint8_t fastpos[1 << FASTPOS_BITS]; + + const uint8_t fast_slots = 2 * FASTPOS_BITS; + uint32_t c = 2; + + fastpos[0] = 0; + fastpos[1] = 1; + + for (uint8_t slot_fast = 2; slot_fast < fast_slots; ++slot_fast) { + const uint32_t k = 1 << ((slot_fast >> 1) - 1); + for (uint32_t j = 0; j < k; ++j, ++c) + fastpos[c] = slot_fast; + } + + printf("/* This file has been automatically generated " + "by fastpos_tablegen.c. */\n\n" + "#include \"common.h\"\n" + "#include \"fastpos.h\"\n\n" + "const uint8_t lzma_fastpos[1 << FASTPOS_BITS] = {"); + + for (size_t i = 0; i < (1 << FASTPOS_BITS); ++i) { + if (i % 16 == 0) + printf("\n\t"); + + printf("%3u", (unsigned int)(fastpos[i])); + + if (i != (1 << FASTPOS_BITS) - 1) + printf(","); + } + + printf("\n};\n"); + + return 0; +} diff --git a/src/liblzma/lzma/lzma2_decoder.c b/src/liblzma/lzma/lzma2_decoder.c new file mode 100644 index 0000000..567df49 --- /dev/null +++ b/src/liblzma/lzma/lzma2_decoder.c @@ -0,0 +1,311 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma2_decoder.c +/// \brief LZMA2 decoder +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lzma2_decoder.h" +#include "lz_decoder.h" +#include "lzma_decoder.h" + + +typedef struct { + enum sequence { + SEQ_CONTROL, + SEQ_UNCOMPRESSED_1, + SEQ_UNCOMPRESSED_2, + SEQ_COMPRESSED_0, + SEQ_COMPRESSED_1, + SEQ_PROPERTIES, + SEQ_LZMA, + SEQ_COPY, + } sequence; + + /// Sequence after the size fields have been decoded. + enum sequence next_sequence; + + /// LZMA decoder + lzma_lz_decoder lzma; + + /// Uncompressed size of LZMA chunk + size_t uncompressed_size; + + /// Compressed size of the chunk (naturally equals to uncompressed + /// size of uncompressed chunk) + size_t compressed_size; + + /// True if properties are needed. This is false before the + /// first LZMA chunk. + bool need_properties; + + /// True if dictionary reset is needed. This is false before the + /// first chunk (LZMA or uncompressed). + bool need_dictionary_reset; + + lzma_options_lzma options; +} lzma_lzma2_coder; + + +static lzma_ret +lzma2_decode(void *coder_ptr, lzma_dict *restrict dict, + const uint8_t *restrict in, size_t *restrict in_pos, + size_t in_size) +{ + lzma_lzma2_coder *restrict coder = coder_ptr; + + // With SEQ_LZMA it is possible that no new input is needed to do + // some progress. The rest of the sequences assume that there is + // at least one byte of input. + while (*in_pos < in_size || coder->sequence == SEQ_LZMA) + switch (coder->sequence) { + case SEQ_CONTROL: { + const uint32_t control = in[*in_pos]; + ++*in_pos; + + // End marker + if (control == 0x00) + return LZMA_STREAM_END; + + if (control >= 0xE0 || control == 1) { + // Dictionary reset implies that next LZMA chunk has + // to set new properties. + coder->need_properties = true; + coder->need_dictionary_reset = true; + } else if (coder->need_dictionary_reset) { + return LZMA_DATA_ERROR; + } + + if (control >= 0x80) { + // LZMA chunk. The highest five bits of the + // uncompressed size are taken from the control byte. + coder->uncompressed_size = (control & 0x1F) << 16; + coder->sequence = SEQ_UNCOMPRESSED_1; + + // See if there are new properties or if we need to + // reset the state. + if (control >= 0xC0) { + // When there are new properties, state reset + // is done at SEQ_PROPERTIES. + coder->need_properties = false; + coder->next_sequence = SEQ_PROPERTIES; + + } else if (coder->need_properties) { + return LZMA_DATA_ERROR; + + } else { + coder->next_sequence = SEQ_LZMA; + + // If only state reset is wanted with old + // properties, do the resetting here for + // simplicity. + if (control >= 0xA0) + coder->lzma.reset(coder->lzma.coder, + &coder->options); + } + } else { + // Invalid control values + if (control > 2) + return LZMA_DATA_ERROR; + + // It's uncompressed chunk + coder->sequence = SEQ_COMPRESSED_0; + coder->next_sequence = SEQ_COPY; + } + + if (coder->need_dictionary_reset) { + // Finish the dictionary reset and let the caller + // flush the dictionary to the actual output buffer. + coder->need_dictionary_reset = false; + dict_reset(dict); + return LZMA_OK; + } + + break; + } + + case SEQ_UNCOMPRESSED_1: + coder->uncompressed_size += (uint32_t)(in[(*in_pos)++]) << 8; + coder->sequence = SEQ_UNCOMPRESSED_2; + break; + + case SEQ_UNCOMPRESSED_2: + coder->uncompressed_size += in[(*in_pos)++] + 1U; + coder->sequence = SEQ_COMPRESSED_0; + coder->lzma.set_uncompressed(coder->lzma.coder, + coder->uncompressed_size, false); + break; + + case SEQ_COMPRESSED_0: + coder->compressed_size = (uint32_t)(in[(*in_pos)++]) << 8; + coder->sequence = SEQ_COMPRESSED_1; + break; + + case SEQ_COMPRESSED_1: + coder->compressed_size += in[(*in_pos)++] + 1U; + coder->sequence = coder->next_sequence; + break; + + case SEQ_PROPERTIES: + if (lzma_lzma_lclppb_decode(&coder->options, in[(*in_pos)++])) + return LZMA_DATA_ERROR; + + coder->lzma.reset(coder->lzma.coder, &coder->options); + + coder->sequence = SEQ_LZMA; + break; + + case SEQ_LZMA: { + // Store the start offset so that we can update + // coder->compressed_size later. + const size_t in_start = *in_pos; + + // Decode from in[] to *dict. + const lzma_ret ret = coder->lzma.code(coder->lzma.coder, + dict, in, in_pos, in_size); + + // Validate and update coder->compressed_size. + const size_t in_used = *in_pos - in_start; + if (in_used > coder->compressed_size) + return LZMA_DATA_ERROR; + + coder->compressed_size -= in_used; + + // Return if we didn't finish the chunk, or an error occurred. + if (ret != LZMA_STREAM_END) + return ret; + + // The LZMA decoder must have consumed the whole chunk now. + // We don't need to worry about uncompressed size since it + // is checked by the LZMA decoder. + if (coder->compressed_size != 0) + return LZMA_DATA_ERROR; + + coder->sequence = SEQ_CONTROL; + break; + } + + case SEQ_COPY: { + // Copy from input to the dictionary as is. + dict_write(dict, in, in_pos, in_size, &coder->compressed_size); + if (coder->compressed_size != 0) + return LZMA_OK; + + coder->sequence = SEQ_CONTROL; + break; + } + + default: + assert(0); + return LZMA_PROG_ERROR; + } + + return LZMA_OK; +} + + +static void +lzma2_decoder_end(void *coder_ptr, const lzma_allocator *allocator) +{ + lzma_lzma2_coder *coder = coder_ptr; + + assert(coder->lzma.end == NULL); + lzma_free(coder->lzma.coder, allocator); + + lzma_free(coder, allocator); + + return; +} + + +static lzma_ret +lzma2_decoder_init(lzma_lz_decoder *lz, const lzma_allocator *allocator, + lzma_vli id lzma_attribute((__unused__)), const void *opt, + lzma_lz_options *lz_options) +{ + lzma_lzma2_coder *coder = lz->coder; + if (coder == NULL) { + coder = lzma_alloc(sizeof(lzma_lzma2_coder), allocator); + if (coder == NULL) + return LZMA_MEM_ERROR; + + lz->coder = coder; + lz->code = &lzma2_decode; + lz->end = &lzma2_decoder_end; + + coder->lzma = LZMA_LZ_DECODER_INIT; + } + + const lzma_options_lzma *options = opt; + + coder->sequence = SEQ_CONTROL; + coder->need_properties = true; + coder->need_dictionary_reset = options->preset_dict == NULL + || options->preset_dict_size == 0; + + return lzma_lzma_decoder_create(&coder->lzma, + allocator, options, lz_options); +} + + +extern lzma_ret +lzma_lzma2_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters) +{ + // LZMA2 can only be the last filter in the chain. This is enforced + // by the raw_decoder initialization. + assert(filters[1].init == NULL); + + return lzma_lz_decoder_init(next, allocator, filters, + &lzma2_decoder_init); +} + + +extern uint64_t +lzma_lzma2_decoder_memusage(const void *options) +{ + return sizeof(lzma_lzma2_coder) + + lzma_lzma_decoder_memusage_nocheck(options); +} + + +extern lzma_ret +lzma_lzma2_props_decode(void **options, const lzma_allocator *allocator, + const uint8_t *props, size_t props_size) +{ + if (props_size != 1) + return LZMA_OPTIONS_ERROR; + + // Check that reserved bits are unset. + if (props[0] & 0xC0) + return LZMA_OPTIONS_ERROR; + + // Decode the dictionary size. + if (props[0] > 40) + return LZMA_OPTIONS_ERROR; + + lzma_options_lzma *opt = lzma_alloc( + sizeof(lzma_options_lzma), allocator); + if (opt == NULL) + return LZMA_MEM_ERROR; + + if (props[0] == 40) { + opt->dict_size = UINT32_MAX; + } else { + opt->dict_size = 2 | (props[0] & 1U); + opt->dict_size <<= props[0] / 2U + 11; + } + + opt->preset_dict = NULL; + opt->preset_dict_size = 0; + + *options = opt; + + return LZMA_OK; +} diff --git a/src/liblzma/lzma/lzma2_decoder.h b/src/liblzma/lzma/lzma2_decoder.h new file mode 100644 index 0000000..ef2dcbf --- /dev/null +++ b/src/liblzma/lzma/lzma2_decoder.h @@ -0,0 +1,29 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma2_decoder.h +/// \brief LZMA2 decoder +/// +// 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_LZMA2_DECODER_H +#define LZMA_LZMA2_DECODER_H + +#include "common.h" + +extern lzma_ret lzma_lzma2_decoder_init(lzma_next_coder *next, + const lzma_allocator *allocator, + const lzma_filter_info *filters); + +extern uint64_t lzma_lzma2_decoder_memusage(const void *options); + +extern lzma_ret lzma_lzma2_props_decode( + void **options, const lzma_allocator *allocator, + const uint8_t *props, size_t props_size); + +#endif diff --git a/src/liblzma/lzma/lzma2_encoder.c b/src/liblzma/lzma/lzma2_encoder.c new file mode 100644 index 0000000..4b6b231 --- /dev/null +++ b/src/liblzma/lzma/lzma2_encoder.c @@ -0,0 +1,414 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma2_encoder.c +/// \brief LZMA2 encoder +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lz_encoder.h" +#include "lzma_encoder.h" +#include "fastpos.h" +#include "lzma2_encoder.h" + + +typedef struct { + enum { + SEQ_INIT, + SEQ_LZMA_ENCODE, + SEQ_LZMA_COPY, + SEQ_UNCOMPRESSED_HEADER, + SEQ_UNCOMPRESSED_COPY, + } sequence; + + /// LZMA encoder + void *lzma; + + /// LZMA options currently in use. + lzma_options_lzma opt_cur; + + bool need_properties; + bool need_state_reset; + bool need_dictionary_reset; + + /// Uncompressed size of a chunk + size_t uncompressed_size; + + /// Compressed size of a chunk (excluding headers); this is also used + /// to indicate the end of buf[] in SEQ_LZMA_COPY. + size_t compressed_size; + + /// Read position in buf[] + size_t buf_pos; + + /// Buffer to hold the chunk header and LZMA compressed data + uint8_t buf[LZMA2_HEADER_MAX + LZMA2_CHUNK_MAX]; +} lzma_lzma2_coder; + + +static void +lzma2_header_lzma(lzma_lzma2_coder *coder) +{ + assert(coder->uncompressed_size > 0); + assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX); + assert(coder->compressed_size > 0); + assert(coder->compressed_size <= LZMA2_CHUNK_MAX); + + size_t pos; + + if (coder->need_properties) { + pos = 0; + + if (coder->need_dictionary_reset) + coder->buf[pos] = 0x80 + (3 << 5); + else + coder->buf[pos] = 0x80 + (2 << 5); + } else { + pos = 1; + + if (coder->need_state_reset) + coder->buf[pos] = 0x80 + (1 << 5); + else + coder->buf[pos] = 0x80; + } + + // Set the start position for copying. + coder->buf_pos = pos; + + // Uncompressed size + size_t size = coder->uncompressed_size - 1; + coder->buf[pos++] += size >> 16; + coder->buf[pos++] = (size >> 8) & 0xFF; + coder->buf[pos++] = size & 0xFF; + + // Compressed size + size = coder->compressed_size - 1; + coder->buf[pos++] = size >> 8; + coder->buf[pos++] = size & 0xFF; + + // Properties, if needed + if (coder->need_properties) + lzma_lzma_lclppb_encode(&coder->opt_cur, coder->buf + pos); + + coder->need_properties = false; + coder->need_state_reset = false; + coder->need_dictionary_reset = false; + + // The copying code uses coder->compressed_size to indicate the end + // of coder->buf[], so we need add the maximum size of the header here. + coder->compressed_size += LZMA2_HEADER_MAX; + + return; +} + + +static void +lzma2_header_uncompressed(lzma_lzma2_coder *coder) +{ + assert(coder->uncompressed_size > 0); + assert(coder->uncompressed_size <= LZMA2_CHUNK_MAX); + + // If this is the first chunk, we need to include dictionary + // reset indicator. + if (coder->need_dictionary_reset) + coder->buf[0] = 1; + else + coder->buf[0] = 2; + + coder->need_dictionary_reset = false; + + // "Compressed" size + coder->buf[1] = (coder->uncompressed_size - 1) >> 8; + coder->buf[2] = (coder->uncompressed_size - 1) & 0xFF; + + // Set the start position for copying. + coder->buf_pos = 0; + return; +} + + +static lzma_ret +lzma2_encode(void *coder_ptr, lzma_mf *restrict mf, + uint8_t *restrict out, size_t *restrict out_pos, + size_t out_size) +{ + lzma_lzma2_coder *restrict coder = coder_ptr; + + while (*out_pos < out_size) + switch (coder->sequence) { + case SEQ_INIT: + // If there's no input left and we are flushing or finishing, + // don't start a new chunk. + if (mf_unencoded(mf) == 0) { + // Write end of payload marker if finishing. + if (mf->action == LZMA_FINISH) + out[(*out_pos)++] = 0; + + return mf->action == LZMA_RUN + ? LZMA_OK : LZMA_STREAM_END; + } + + if (coder->need_state_reset) + return_if_error(lzma_lzma_encoder_reset( + coder->lzma, &coder->opt_cur)); + + coder->uncompressed_size = 0; + coder->compressed_size = 0; + coder->sequence = SEQ_LZMA_ENCODE; + + // Fall through + + case SEQ_LZMA_ENCODE: { + // Calculate how much more uncompressed data this chunk + // could accept. + const uint32_t left = LZMA2_UNCOMPRESSED_MAX + - coder->uncompressed_size; + uint32_t limit; + + if (left < mf->match_len_max) { + // Must flush immediately since the next LZMA symbol + // could make the uncompressed size of the chunk too + // big. + limit = 0; + } else { + // Calculate maximum read_limit that is OK from point + // of view of LZMA2 chunk size. + limit = mf->read_pos - mf->read_ahead + + left - mf->match_len_max; + } + + // Save the start position so that we can update + // coder->uncompressed_size. + const uint32_t read_start = mf->read_pos - mf->read_ahead; + + // Call the LZMA encoder until the chunk is finished. + const lzma_ret ret = lzma_lzma_encode(coder->lzma, mf, + coder->buf + LZMA2_HEADER_MAX, + &coder->compressed_size, + LZMA2_CHUNK_MAX, limit); + + coder->uncompressed_size += mf->read_pos - mf->read_ahead + - read_start; + + assert(coder->compressed_size <= LZMA2_CHUNK_MAX); + assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX); + + if (ret != LZMA_STREAM_END) + return LZMA_OK; + + // See if the chunk compressed. If it didn't, we encode it + // as uncompressed chunk. This saves a few bytes of space + // and makes decoding faster. + if (coder->compressed_size >= coder->uncompressed_size) { + coder->uncompressed_size += mf->read_ahead; + assert(coder->uncompressed_size + <= LZMA2_UNCOMPRESSED_MAX); + mf->read_ahead = 0; + lzma2_header_uncompressed(coder); + coder->need_state_reset = true; + coder->sequence = SEQ_UNCOMPRESSED_HEADER; + break; + } + + // The chunk did compress at least by one byte, so we store + // the chunk as LZMA. + lzma2_header_lzma(coder); + + coder->sequence = SEQ_LZMA_COPY; + } + + // Fall through + + case SEQ_LZMA_COPY: + // Copy the compressed chunk along its headers to the + // output buffer. + lzma_bufcpy(coder->buf, &coder->buf_pos, + coder->compressed_size, + out, out_pos, out_size); + if (coder->buf_pos != coder->compressed_size) + return LZMA_OK; + + coder->sequence = SEQ_INIT; + break; + + case SEQ_UNCOMPRESSED_HEADER: + // Copy the three-byte header to indicate uncompressed chunk. + lzma_bufcpy(coder->buf, &coder->buf_pos, + LZMA2_HEADER_UNCOMPRESSED, + out, out_pos, out_size); + if (coder->buf_pos != LZMA2_HEADER_UNCOMPRESSED) + return LZMA_OK; + + coder->sequence = SEQ_UNCOMPRESSED_COPY; + + // Fall through + + case SEQ_UNCOMPRESSED_COPY: + // Copy the uncompressed data as is from the dictionary + // to the output buffer. + mf_read(mf, out, out_pos, out_size, &coder->uncompressed_size); + if (coder->uncompressed_size != 0) + return LZMA_OK; + + coder->sequence = SEQ_INIT; + break; + } + + return LZMA_OK; +} + + +static void +lzma2_encoder_end(void *coder_ptr, const lzma_allocator *allocator) +{ + lzma_lzma2_coder *coder = coder_ptr; + lzma_free(coder->lzma, allocator); + lzma_free(coder, allocator); + return; +} + + +static lzma_ret +lzma2_encoder_options_update(void *coder_ptr, const lzma_filter *filter) +{ + lzma_lzma2_coder *coder = coder_ptr; + + // New options can be set only when there is no incomplete chunk. + // This is the case at the beginning of the raw stream and right + // after LZMA_SYNC_FLUSH. + if (filter->options == NULL || coder->sequence != SEQ_INIT) + return LZMA_PROG_ERROR; + + // Look if there are new options. At least for now, + // only lc/lp/pb can be changed. + const lzma_options_lzma *opt = filter->options; + if (coder->opt_cur.lc != opt->lc || coder->opt_cur.lp != opt->lp + || coder->opt_cur.pb != opt->pb) { + // Validate the options. + if (opt->lc > LZMA_LCLP_MAX || opt->lp > LZMA_LCLP_MAX + || opt->lc + opt->lp > LZMA_LCLP_MAX + || opt->pb > LZMA_PB_MAX) + return LZMA_OPTIONS_ERROR; + + // The new options will be used when the encoder starts + // a new LZMA2 chunk. + coder->opt_cur.lc = opt->lc; + coder->opt_cur.lp = opt->lp; + coder->opt_cur.pb = opt->pb; + coder->need_properties = true; + coder->need_state_reset = true; + } + + return LZMA_OK; +} + + +static lzma_ret +lzma2_encoder_init(lzma_lz_encoder *lz, const lzma_allocator *allocator, + lzma_vli id lzma_attribute((__unused__)), const void *options, + lzma_lz_options *lz_options) +{ + if (options == NULL) + return LZMA_PROG_ERROR; + + lzma_lzma2_coder *coder = lz->coder; + if (coder == NULL) { + coder = lzma_alloc(sizeof(lzma_lzma2_coder), allocator); + if (coder == NULL) + return LZMA_MEM_ERROR; + + lz->coder = coder; + lz->code = &lzma2_encode; + lz->end = &lzma2_encoder_end; + lz->options_update = &lzma2_encoder_options_update; + + coder->lzma = NULL; + } + + coder->opt_cur = *(const lzma_options_lzma *)(options); + + coder->sequence = SEQ_INIT; + coder->need_properties = true; + coder->need_state_reset = false; + coder->need_dictionary_reset + = coder->opt_cur.preset_dict == NULL + || coder->opt_cur.preset_dict_size == 0; + + // Initialize LZMA encoder + return_if_error(lzma_lzma_encoder_create(&coder->lzma, allocator, + LZMA_FILTER_LZMA2, &coder->opt_cur, lz_options)); + + // Make sure that we will always have enough history available in + // case we need to use uncompressed chunks. They are used when the + // compressed size of a chunk is not smaller than the uncompressed + // size, so we need to have at least LZMA2_COMPRESSED_MAX bytes + // history available. + if (lz_options->before_size + lz_options->dict_size < LZMA2_CHUNK_MAX) + lz_options->before_size + = LZMA2_CHUNK_MAX - lz_options->dict_size; + + return LZMA_OK; +} + + +extern lzma_ret +lzma_lzma2_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters) +{ + return lzma_lz_encoder_init( + next, allocator, filters, &lzma2_encoder_init); +} + + +extern uint64_t +lzma_lzma2_encoder_memusage(const void *options) +{ + const uint64_t lzma_mem = lzma_lzma_encoder_memusage(options); + if (lzma_mem == UINT64_MAX) + return UINT64_MAX; + + return sizeof(lzma_lzma2_coder) + lzma_mem; +} + + +extern lzma_ret +lzma_lzma2_props_encode(const void *options, uint8_t *out) +{ + if (options == NULL) + return LZMA_PROG_ERROR; + + const lzma_options_lzma *const opt = options; + uint32_t d = my_max(opt->dict_size, LZMA_DICT_SIZE_MIN); + + // Round up to the next 2^n - 1 or 2^n + 2^(n - 1) - 1 depending + // on which one is the next: + --d; + d |= d >> 2; + d |= d >> 3; + d |= d >> 4; + d |= d >> 8; + d |= d >> 16; + + // Get the highest two bits using the proper encoding: + if (d == UINT32_MAX) + out[0] = 40; + else + out[0] = get_dist_slot(d + 1) - 24; + + return LZMA_OK; +} + + +extern uint64_t +lzma_lzma2_block_size(const void *options) +{ + const lzma_options_lzma *const opt = options; + + // Use at least 1 MiB to keep compression ratio better. + return my_max((uint64_t)(opt->dict_size) * 3, UINT64_C(1) << 20); +} diff --git a/src/liblzma/lzma/lzma2_encoder.h b/src/liblzma/lzma/lzma2_encoder.h new file mode 100644 index 0000000..515f183 --- /dev/null +++ b/src/liblzma/lzma/lzma2_encoder.h @@ -0,0 +1,43 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma2_encoder.h +/// \brief LZMA2 encoder +/// +// 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_LZMA2_ENCODER_H +#define LZMA_LZMA2_ENCODER_H + +#include "common.h" + + +/// Maximum number of bytes of actual data per chunk (no headers) +#define LZMA2_CHUNK_MAX (UINT32_C(1) << 16) + +/// Maximum uncompressed size of LZMA chunk (no headers) +#define LZMA2_UNCOMPRESSED_MAX (UINT32_C(1) << 21) + +/// Maximum size of LZMA2 headers +#define LZMA2_HEADER_MAX 6 + +/// Size of a header for uncompressed chunk +#define LZMA2_HEADER_UNCOMPRESSED 3 + + +extern lzma_ret lzma_lzma2_encoder_init( + lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters); + +extern uint64_t lzma_lzma2_encoder_memusage(const void *options); + +extern lzma_ret lzma_lzma2_props_encode(const void *options, uint8_t *out); + +extern uint64_t lzma_lzma2_block_size(const void *options); + +#endif diff --git a/src/liblzma/lzma/lzma_common.h b/src/liblzma/lzma/lzma_common.h new file mode 100644 index 0000000..9d040d9 --- /dev/null +++ b/src/liblzma/lzma/lzma_common.h @@ -0,0 +1,225 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_common.h +/// \brief Private definitions common to LZMA encoder and decoder +/// +// 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_LZMA_COMMON_H +#define LZMA_LZMA_COMMON_H + +#include "common.h" +#include "range_common.h" + + +/////////////////// +// Miscellaneous // +/////////////////// + +/// Maximum number of position states. A position state is the lowest pos bits +/// number of bits of the current uncompressed offset. In some places there +/// are different sets of probabilities for different pos states. +#define POS_STATES_MAX (1 << LZMA_PB_MAX) + + +/// Validates lc, lp, and pb. +static inline bool +is_lclppb_valid(const lzma_options_lzma *options) +{ + return options->lc <= LZMA_LCLP_MAX && options->lp <= LZMA_LCLP_MAX + && options->lc + options->lp <= LZMA_LCLP_MAX + && options->pb <= LZMA_PB_MAX; +} + + +/////////// +// State // +/////////// + +/// This enum is used to track which events have occurred most recently and +/// in which order. This information is used to predict the next event. +/// +/// Events: +/// - Literal: One 8-bit byte +/// - Match: Repeat a chunk of data at some distance +/// - Long repeat: Multi-byte match at a recently seen distance +/// - Short repeat: One-byte repeat at a recently seen distance +/// +/// The event names are in from STATE_oldest_older_previous. REP means +/// either short or long repeated match, and NONLIT means any non-literal. +typedef enum { + STATE_LIT_LIT, + STATE_MATCH_LIT_LIT, + STATE_REP_LIT_LIT, + STATE_SHORTREP_LIT_LIT, + STATE_MATCH_LIT, + STATE_REP_LIT, + STATE_SHORTREP_LIT, + STATE_LIT_MATCH, + STATE_LIT_LONGREP, + STATE_LIT_SHORTREP, + STATE_NONLIT_MATCH, + STATE_NONLIT_REP, +} lzma_lzma_state; + + +/// Total number of states +#define STATES 12 + +/// The lowest 7 states indicate that the previous state was a literal. +#define LIT_STATES 7 + + +/// Indicate that the latest state was a literal. +#define update_literal(state) \ + state = ((state) <= STATE_SHORTREP_LIT_LIT \ + ? STATE_LIT_LIT \ + : ((state) <= STATE_LIT_SHORTREP \ + ? (state) - 3 \ + : (state) - 6)) + +/// Indicate that the latest state was a match. +#define update_match(state) \ + state = ((state) < LIT_STATES ? STATE_LIT_MATCH : STATE_NONLIT_MATCH) + +/// Indicate that the latest state was a long repeated match. +#define update_long_rep(state) \ + state = ((state) < LIT_STATES ? STATE_LIT_LONGREP : STATE_NONLIT_REP) + +/// Indicate that the latest state was a short match. +#define update_short_rep(state) \ + state = ((state) < LIT_STATES ? STATE_LIT_SHORTREP : STATE_NONLIT_REP) + +/// Test if the previous state was a literal. +#define is_literal_state(state) \ + ((state) < LIT_STATES) + + +///////////// +// Literal // +///////////// + +/// Each literal coder is divided in three sections: +/// - 0x001-0x0FF: Without match byte +/// - 0x101-0x1FF: With match byte; match bit is 0 +/// - 0x201-0x2FF: With match byte; match bit is 1 +/// +/// Match byte is used when the previous LZMA symbol was something else than +/// a literal (that is, it was some kind of match). +#define LITERAL_CODER_SIZE 0x300 + +/// Maximum number of literal coders +#define LITERAL_CODERS_MAX (1 << LZMA_LCLP_MAX) + +/// Locate the literal coder for the next literal byte. The choice depends on +/// - the lowest literal_pos_bits bits of the position of the current +/// byte; and +/// - the highest literal_context_bits bits of the previous byte. +#define literal_subcoder(probs, lc, lp_mask, pos, prev_byte) \ + ((probs)[(((pos) & (lp_mask)) << (lc)) \ + + ((uint32_t)(prev_byte) >> (8U - (lc)))]) + + +static inline void +literal_init(probability (*probs)[LITERAL_CODER_SIZE], + uint32_t lc, uint32_t lp) +{ + assert(lc + lp <= LZMA_LCLP_MAX); + + const uint32_t coders = 1U << (lc + lp); + + for (uint32_t i = 0; i < coders; ++i) + for (uint32_t j = 0; j < LITERAL_CODER_SIZE; ++j) + bit_reset(probs[i][j]); + + return; +} + + +////////////////// +// Match length // +////////////////// + +// Minimum length of a match is two bytes. +#define MATCH_LEN_MIN 2 + +// Match length is encoded with 4, 5, or 10 bits. +// +// Length Bits +// 2-9 4 = Choice=0 + 3 bits +// 10-17 5 = Choice=1 + Choice2=0 + 3 bits +// 18-273 10 = Choice=1 + Choice2=1 + 8 bits +#define LEN_LOW_BITS 3 +#define LEN_LOW_SYMBOLS (1 << LEN_LOW_BITS) +#define LEN_MID_BITS 3 +#define LEN_MID_SYMBOLS (1 << LEN_MID_BITS) +#define LEN_HIGH_BITS 8 +#define LEN_HIGH_SYMBOLS (1 << LEN_HIGH_BITS) +#define LEN_SYMBOLS (LEN_LOW_SYMBOLS + LEN_MID_SYMBOLS + LEN_HIGH_SYMBOLS) + +// Maximum length of a match is 273 which is a result of the encoding +// described above. +#define MATCH_LEN_MAX (MATCH_LEN_MIN + LEN_SYMBOLS - 1) + + +//////////////////// +// Match distance // +//////////////////// + +// Different sets of probabilities are used for match distances that have very +// short match length: Lengths of 2, 3, and 4 bytes have a separate set of +// probabilities for each length. The matches with longer length use a shared +// set of probabilities. +#define DIST_STATES 4 + +// Macro to get the index of the appropriate probability array. +#define get_dist_state(len) \ + ((len) < DIST_STATES + MATCH_LEN_MIN \ + ? (len) - MATCH_LEN_MIN \ + : DIST_STATES - 1) + +// The highest two bits of a match distance (distance slot) are encoded +// using six bits. See fastpos.h for more explanation. +#define DIST_SLOT_BITS 6 +#define DIST_SLOTS (1 << DIST_SLOT_BITS) + +// Match distances up to 127 are fully encoded using probabilities. Since +// the highest two bits (distance slot) are always encoded using six bits, +// the distances 0-3 don't need any additional bits to encode, since the +// distance slot itself is the same as the actual distance. DIST_MODEL_START +// indicates the first distance slot where at least one additional bit is +// needed. +#define DIST_MODEL_START 4 + +// Match distances greater than 127 are encoded in three pieces: +// - distance slot: the highest two bits +// - direct bits: 2-26 bits below the highest two bits +// - alignment bits: four lowest bits +// +// Direct bits don't use any probabilities. +// +// The distance slot value of 14 is for distances 128-191 (see the table in +// fastpos.h to understand why). +#define DIST_MODEL_END 14 + +// Distance slots that indicate a distance <= 127. +#define FULL_DISTANCES_BITS (DIST_MODEL_END / 2) +#define FULL_DISTANCES (1 << FULL_DISTANCES_BITS) + +// For match distances greater than 127, only the highest two bits and the +// lowest four bits (alignment) is encoded using probabilities. +#define ALIGN_BITS 4 +#define ALIGN_SIZE (1 << ALIGN_BITS) +#define ALIGN_MASK (ALIGN_SIZE - 1) + +// LZMA remembers the four most recent match distances. Reusing these distances +// tends to take less space than re-encoding the actual distance value. +#define REPS 4 + +#endif diff --git a/src/liblzma/lzma/lzma_decoder.c b/src/liblzma/lzma/lzma_decoder.c new file mode 100644 index 0000000..26c148a --- /dev/null +++ b/src/liblzma/lzma/lzma_decoder.c @@ -0,0 +1,1133 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_decoder.c +/// \brief LZMA decoder +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lz_decoder.h" +#include "lzma_common.h" +#include "lzma_decoder.h" +#include "range_decoder.h" + +// The macros unroll loops with switch statements. +// Silence warnings about missing fall-through comments. +#if TUKLIB_GNUC_REQ(7, 0) +# pragma GCC diagnostic ignored "-Wimplicit-fallthrough" +#endif + + +#ifdef HAVE_SMALL + +// Macros for (somewhat) size-optimized code. +#define seq_4(seq) seq + +#define seq_6(seq) seq + +#define seq_8(seq) seq + +#define seq_len(seq) \ + seq ## _CHOICE, \ + seq ## _CHOICE2, \ + seq ## _BITTREE + +#define len_decode(target, ld, pos_state, seq) \ +do { \ +case seq ## _CHOICE: \ + rc_if_0(ld.choice, seq ## _CHOICE) { \ + rc_update_0(ld.choice); \ + probs = ld.low[pos_state];\ + limit = LEN_LOW_SYMBOLS; \ + target = MATCH_LEN_MIN; \ + } else { \ + rc_update_1(ld.choice); \ +case seq ## _CHOICE2: \ + rc_if_0(ld.choice2, seq ## _CHOICE2) { \ + rc_update_0(ld.choice2); \ + probs = ld.mid[pos_state]; \ + limit = LEN_MID_SYMBOLS; \ + target = MATCH_LEN_MIN + LEN_LOW_SYMBOLS; \ + } else { \ + rc_update_1(ld.choice2); \ + probs = ld.high; \ + limit = LEN_HIGH_SYMBOLS; \ + target = MATCH_LEN_MIN + LEN_LOW_SYMBOLS \ + + LEN_MID_SYMBOLS; \ + } \ + } \ + symbol = 1; \ +case seq ## _BITTREE: \ + do { \ + rc_bit(probs[symbol], , , seq ## _BITTREE); \ + } while (symbol < limit); \ + target += symbol - limit; \ +} while (0) + +#else // HAVE_SMALL + +// Unrolled versions +#define seq_4(seq) \ + seq ## 0, \ + seq ## 1, \ + seq ## 2, \ + seq ## 3 + +#define seq_6(seq) \ + seq ## 0, \ + seq ## 1, \ + seq ## 2, \ + seq ## 3, \ + seq ## 4, \ + seq ## 5 + +#define seq_8(seq) \ + seq ## 0, \ + seq ## 1, \ + seq ## 2, \ + seq ## 3, \ + seq ## 4, \ + seq ## 5, \ + seq ## 6, \ + seq ## 7 + +#define seq_len(seq) \ + seq ## _CHOICE, \ + seq ## _LOW0, \ + seq ## _LOW1, \ + seq ## _LOW2, \ + seq ## _CHOICE2, \ + seq ## _MID0, \ + seq ## _MID1, \ + seq ## _MID2, \ + seq ## _HIGH0, \ + seq ## _HIGH1, \ + seq ## _HIGH2, \ + seq ## _HIGH3, \ + seq ## _HIGH4, \ + seq ## _HIGH5, \ + seq ## _HIGH6, \ + seq ## _HIGH7 + +#define len_decode(target, ld, pos_state, seq) \ +do { \ + symbol = 1; \ +case seq ## _CHOICE: \ + rc_if_0(ld.choice, seq ## _CHOICE) { \ + rc_update_0(ld.choice); \ + rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW0); \ + rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW1); \ + rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW2); \ + target = symbol - LEN_LOW_SYMBOLS + MATCH_LEN_MIN; \ + } else { \ + rc_update_1(ld.choice); \ +case seq ## _CHOICE2: \ + rc_if_0(ld.choice2, seq ## _CHOICE2) { \ + rc_update_0(ld.choice2); \ + rc_bit_case(ld.mid[pos_state][symbol], , , \ + seq ## _MID0); \ + rc_bit_case(ld.mid[pos_state][symbol], , , \ + seq ## _MID1); \ + rc_bit_case(ld.mid[pos_state][symbol], , , \ + seq ## _MID2); \ + target = symbol - LEN_MID_SYMBOLS \ + + MATCH_LEN_MIN + LEN_LOW_SYMBOLS; \ + } else { \ + rc_update_1(ld.choice2); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH0); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH1); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH2); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH3); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH4); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH5); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH6); \ + rc_bit_case(ld.high[symbol], , , seq ## _HIGH7); \ + target = symbol - LEN_HIGH_SYMBOLS \ + + MATCH_LEN_MIN \ + + LEN_LOW_SYMBOLS + LEN_MID_SYMBOLS; \ + } \ + } \ +} while (0) + +#endif // HAVE_SMALL + + +/// Length decoder probabilities; see comments in lzma_common.h. +typedef struct { + probability choice; + probability choice2; + probability low[POS_STATES_MAX][LEN_LOW_SYMBOLS]; + probability mid[POS_STATES_MAX][LEN_MID_SYMBOLS]; + probability high[LEN_HIGH_SYMBOLS]; +} lzma_length_decoder; + + +typedef struct { + /////////////////// + // Probabilities // + /////////////////// + + /// Literals; see comments in lzma_common.h. + probability literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE]; + + /// If 1, it's a match. Otherwise it's a single 8-bit literal. + probability is_match[STATES][POS_STATES_MAX]; + + /// If 1, it's a repeated match. The distance is one of rep0 .. rep3. + probability is_rep[STATES]; + + /// If 0, distance of a repeated match is rep0. + /// Otherwise check is_rep1. + probability is_rep0[STATES]; + + /// If 0, distance of a repeated match is rep1. + /// Otherwise check is_rep2. + probability is_rep1[STATES]; + + /// If 0, distance of a repeated match is rep2. Otherwise it is rep3. + probability is_rep2[STATES]; + + /// If 1, the repeated match has length of one byte. Otherwise + /// the length is decoded from rep_len_decoder. + probability is_rep0_long[STATES][POS_STATES_MAX]; + + /// Probability tree for the highest two bits of the match distance. + /// There is a separate probability tree for match lengths of + /// 2 (i.e. MATCH_LEN_MIN), 3, 4, and [5, 273]. + probability dist_slot[DIST_STATES][DIST_SLOTS]; + + /// Probability trees for additional bits for match distance when the + /// distance is in the range [4, 127]. + probability pos_special[FULL_DISTANCES - DIST_MODEL_END]; + + /// Probability tree for the lowest four bits of a match distance + /// that is equal to or greater than 128. + probability pos_align[ALIGN_SIZE]; + + /// Length of a normal match + lzma_length_decoder match_len_decoder; + + /// Length of a repeated match + lzma_length_decoder rep_len_decoder; + + /////////////////// + // Decoder state // + /////////////////// + + // Range coder + lzma_range_decoder rc; + + // Types of the most recently seen LZMA symbols + lzma_lzma_state state; + + uint32_t rep0; ///< Distance of the latest match + uint32_t rep1; ///< Distance of second latest match + uint32_t rep2; ///< Distance of third latest match + uint32_t rep3; ///< Distance of fourth latest match + + uint32_t pos_mask; // (1U << pb) - 1 + uint32_t literal_context_bits; + uint32_t literal_pos_mask; + + /// Uncompressed size as bytes, or LZMA_VLI_UNKNOWN if end of + /// payload marker is expected. + lzma_vli uncompressed_size; + + /// True if end of payload marker (EOPM) is allowed even when + /// uncompressed_size is known; false if EOPM must not be present. + /// This is ignored if uncompressed_size == LZMA_VLI_UNKNOWN. + bool allow_eopm; + + //////////////////////////////// + // State of incomplete symbol // + //////////////////////////////// + + /// Position where to continue the decoder loop + enum { + SEQ_NORMALIZE, + SEQ_IS_MATCH, + seq_8(SEQ_LITERAL), + seq_8(SEQ_LITERAL_MATCHED), + SEQ_LITERAL_WRITE, + SEQ_IS_REP, + seq_len(SEQ_MATCH_LEN), + seq_6(SEQ_DIST_SLOT), + SEQ_DIST_MODEL, + SEQ_DIRECT, + seq_4(SEQ_ALIGN), + SEQ_EOPM, + SEQ_IS_REP0, + SEQ_SHORTREP, + SEQ_IS_REP0_LONG, + SEQ_IS_REP1, + SEQ_IS_REP2, + seq_len(SEQ_REP_LEN), + SEQ_COPY, + } sequence; + + /// Base of the current probability tree + probability *probs; + + /// Symbol being decoded. This is also used as an index variable in + /// bittree decoders: probs[symbol] + uint32_t symbol; + + /// Used as a loop termination condition on bittree decoders and + /// direct bits decoder. + uint32_t limit; + + /// Matched literal decoder: 0x100 or 0 to help avoiding branches. + /// Bittree reverse decoders: Offset of the next bit: 1 << offset + uint32_t offset; + + /// If decoding a literal: match byte. + /// If decoding a match: length of the match. + uint32_t len; +} lzma_lzma1_decoder; + + +static lzma_ret +lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr, + const uint8_t *restrict in, + size_t *restrict in_pos, size_t in_size) +{ + lzma_lzma1_decoder *restrict coder = coder_ptr; + + //////////////////// + // Initialization // + //////////////////// + + { + const lzma_ret ret = rc_read_init( + &coder->rc, in, in_pos, in_size); + if (ret != LZMA_STREAM_END) + return ret; + } + + /////////////// + // Variables // + /////////////// + + // Making local copies of often-used variables improves both + // speed and readability. + + lzma_dict dict = *dictptr; + + const size_t dict_start = dict.pos; + + // Range decoder + rc_to_local(coder->rc, *in_pos); + + // State + uint32_t state = coder->state; + uint32_t rep0 = coder->rep0; + uint32_t rep1 = coder->rep1; + uint32_t rep2 = coder->rep2; + uint32_t rep3 = coder->rep3; + + const uint32_t pos_mask = coder->pos_mask; + + // These variables are actually needed only if we last time ran + // out of input in the middle of the decoder loop. + probability *probs = coder->probs; + uint32_t symbol = coder->symbol; + uint32_t limit = coder->limit; + uint32_t offset = coder->offset; + uint32_t len = coder->len; + + const uint32_t literal_pos_mask = coder->literal_pos_mask; + const uint32_t literal_context_bits = coder->literal_context_bits; + + // Temporary variables + uint32_t pos_state = dict.pos & pos_mask; + + lzma_ret ret = LZMA_OK; + + // This is true when the next LZMA symbol is allowed to be EOPM. + // That is, if this is false, then EOPM is considered + // an invalid symbol and we will return LZMA_DATA_ERROR. + // + // EOPM is always required (not just allowed) when + // the uncompressed size isn't known. When uncompressed size + // is known, eopm_is_valid may be set to true later. + bool eopm_is_valid = coder->uncompressed_size == LZMA_VLI_UNKNOWN; + + // If uncompressed size is known and there is enough output space + // to decode all the data, limit the available buffer space so that + // the main loop won't try to decode past the end of the stream. + bool might_finish_without_eopm = false; + if (coder->uncompressed_size != LZMA_VLI_UNKNOWN + && coder->uncompressed_size <= dict.limit - dict.pos) { + dict.limit = dict.pos + (size_t)(coder->uncompressed_size); + might_finish_without_eopm = true; + } + + // The main decoder loop. The "switch" is used to restart the decoder at + // correct location. Once restarted, the "switch" is no longer used. + switch (coder->sequence) + while (true) { + // Calculate new pos_state. This is skipped on the first loop + // since we already calculated it when setting up the local + // variables. + pos_state = dict.pos & pos_mask; + + case SEQ_NORMALIZE: + case SEQ_IS_MATCH: + if (unlikely(might_finish_without_eopm + && dict.pos == dict.limit)) { + // In rare cases there is a useless byte that needs + // to be read anyway. + rc_normalize(SEQ_NORMALIZE); + + // If the range decoder state is such that we can + // be at the end of the LZMA stream, then the + // decoding is finished. + if (rc_is_finished(rc)) { + ret = LZMA_STREAM_END; + goto out; + } + + // If the caller hasn't allowed EOPM to be present + // together with known uncompressed size, then the + // LZMA stream is corrupt. + if (!coder->allow_eopm) { + ret = LZMA_DATA_ERROR; + goto out; + } + + // Otherwise continue decoding with the expectation + // that the next LZMA symbol is EOPM. + eopm_is_valid = true; + } + + rc_if_0(coder->is_match[state][pos_state], SEQ_IS_MATCH) { + rc_update_0(coder->is_match[state][pos_state]); + + // It's a literal i.e. a single 8-bit byte. + + probs = literal_subcoder(coder->literal, + literal_context_bits, literal_pos_mask, + dict.pos, dict_get(&dict, 0)); + symbol = 1; + + if (is_literal_state(state)) { + // Decode literal without match byte. +#ifdef HAVE_SMALL + case SEQ_LITERAL: + do { + rc_bit(probs[symbol], , , SEQ_LITERAL); + } while (symbol < (1 << 8)); +#else + rc_bit_case(probs[symbol], , , SEQ_LITERAL0); + rc_bit_case(probs[symbol], , , SEQ_LITERAL1); + rc_bit_case(probs[symbol], , , SEQ_LITERAL2); + rc_bit_case(probs[symbol], , , SEQ_LITERAL3); + rc_bit_case(probs[symbol], , , SEQ_LITERAL4); + rc_bit_case(probs[symbol], , , SEQ_LITERAL5); + rc_bit_case(probs[symbol], , , SEQ_LITERAL6); + rc_bit_case(probs[symbol], , , SEQ_LITERAL7); +#endif + } else { + // Decode literal with match byte. + // + // We store the byte we compare against + // ("match byte") to "len" to minimize the + // number of variables we need to store + // between decoder calls. + len = (uint32_t)(dict_get(&dict, rep0)) << 1; + + // The usage of "offset" allows omitting some + // branches, which should give tiny speed + // improvement on some CPUs. "offset" gets + // set to zero if match_bit didn't match. + offset = 0x100; + +#ifdef HAVE_SMALL + case SEQ_LITERAL_MATCHED: + do { + const uint32_t match_bit + = len & offset; + const uint32_t subcoder_index + = offset + match_bit + + symbol; + + rc_bit(probs[subcoder_index], + offset &= ~match_bit, + offset &= match_bit, + SEQ_LITERAL_MATCHED); + + // It seems to be faster to do this + // here instead of putting it to the + // beginning of the loop and then + // putting the "case" in the middle + // of the loop. + len <<= 1; + + } while (symbol < (1 << 8)); +#else + // Unroll the loop. + uint32_t match_bit; + uint32_t subcoder_index; + +# define d(seq) \ + case seq: \ + match_bit = len & offset; \ + subcoder_index = offset + match_bit + symbol; \ + rc_bit(probs[subcoder_index], \ + offset &= ~match_bit, \ + offset &= match_bit, \ + seq) + + d(SEQ_LITERAL_MATCHED0); + len <<= 1; + d(SEQ_LITERAL_MATCHED1); + len <<= 1; + d(SEQ_LITERAL_MATCHED2); + len <<= 1; + d(SEQ_LITERAL_MATCHED3); + len <<= 1; + d(SEQ_LITERAL_MATCHED4); + len <<= 1; + d(SEQ_LITERAL_MATCHED5); + len <<= 1; + d(SEQ_LITERAL_MATCHED6); + len <<= 1; + d(SEQ_LITERAL_MATCHED7); +# undef d +#endif + } + + //update_literal(state); + // Use a lookup table to update to literal state, + // since compared to other state updates, this would + // need two branches. + static const lzma_lzma_state next_state[] = { + STATE_LIT_LIT, + STATE_LIT_LIT, + STATE_LIT_LIT, + STATE_LIT_LIT, + STATE_MATCH_LIT_LIT, + STATE_REP_LIT_LIT, + STATE_SHORTREP_LIT_LIT, + STATE_MATCH_LIT, + STATE_REP_LIT, + STATE_SHORTREP_LIT, + STATE_MATCH_LIT, + STATE_REP_LIT + }; + state = next_state[state]; + + case SEQ_LITERAL_WRITE: + if (unlikely(dict_put(&dict, symbol))) { + coder->sequence = SEQ_LITERAL_WRITE; + goto out; + } + + continue; + } + + // Instead of a new byte we are going to get a byte range + // (distance and length) which will be repeated from our + // output history. + + rc_update_1(coder->is_match[state][pos_state]); + + case SEQ_IS_REP: + rc_if_0(coder->is_rep[state], SEQ_IS_REP) { + // Not a repeated match + rc_update_0(coder->is_rep[state]); + update_match(state); + + // The latest three match distances are kept in + // memory in case there are repeated matches. + rep3 = rep2; + rep2 = rep1; + rep1 = rep0; + + // Decode the length of the match. + len_decode(len, coder->match_len_decoder, + pos_state, SEQ_MATCH_LEN); + + // Prepare to decode the highest two bits of the + // match distance. + probs = coder->dist_slot[get_dist_state(len)]; + symbol = 1; + +#ifdef HAVE_SMALL + case SEQ_DIST_SLOT: + do { + rc_bit(probs[symbol], , , SEQ_DIST_SLOT); + } while (symbol < DIST_SLOTS); +#else + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT0); + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT1); + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT2); + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT3); + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT4); + rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT5); +#endif + // Get rid of the highest bit that was needed for + // indexing of the probability array. + symbol -= DIST_SLOTS; + assert(symbol <= 63); + + if (symbol < DIST_MODEL_START) { + // Match distances [0, 3] have only two bits. + rep0 = symbol; + } else { + // Decode the lowest [1, 29] bits of + // the match distance. + limit = (symbol >> 1) - 1; + assert(limit >= 1 && limit <= 30); + rep0 = 2 + (symbol & 1); + + if (symbol < DIST_MODEL_END) { + // Prepare to decode the low bits for + // a distance of [4, 127]. + assert(limit <= 5); + rep0 <<= limit; + assert(rep0 <= 96); + // -1 is fine, because we start + // decoding at probs[1], not probs[0]. + // NOTE: This violates the C standard, + // since we are doing pointer + // arithmetic past the beginning of + // the array. + assert((int32_t)(rep0 - symbol - 1) + >= -1); + assert((int32_t)(rep0 - symbol - 1) + <= 82); + probs = coder->pos_special + rep0 + - symbol - 1; + symbol = 1; + offset = 0; + case SEQ_DIST_MODEL: +#ifdef HAVE_SMALL + do { + rc_bit(probs[symbol], , + rep0 += 1U << offset, + SEQ_DIST_MODEL); + } while (++offset < limit); +#else + switch (limit) { + case 5: + assert(offset == 0); + rc_bit(probs[symbol], , + rep0 += 1U, + SEQ_DIST_MODEL); + ++offset; + --limit; + case 4: + rc_bit(probs[symbol], , + rep0 += 1U << offset, + SEQ_DIST_MODEL); + ++offset; + --limit; + case 3: + rc_bit(probs[symbol], , + rep0 += 1U << offset, + SEQ_DIST_MODEL); + ++offset; + --limit; + case 2: + rc_bit(probs[symbol], , + rep0 += 1U << offset, + SEQ_DIST_MODEL); + ++offset; + --limit; + case 1: + // We need "symbol" only for + // indexing the probability + // array, thus we can use + // rc_bit_last() here to omit + // the unneeded updating of + // "symbol". + rc_bit_last(probs[symbol], , + rep0 += 1U << offset, + SEQ_DIST_MODEL); + } +#endif + } else { + // The distance is >= 128. Decode the + // lower bits without probabilities + // except the lowest four bits. + assert(symbol >= 14); + assert(limit >= 6); + limit -= ALIGN_BITS; + assert(limit >= 2); + case SEQ_DIRECT: + // Not worth manual unrolling + do { + rc_direct(rep0, SEQ_DIRECT); + } while (--limit > 0); + + // Decode the lowest four bits using + // probabilities. + rep0 <<= ALIGN_BITS; + symbol = 1; +#ifdef HAVE_SMALL + offset = 0; + case SEQ_ALIGN: + do { + rc_bit(coder->pos_align[ + symbol], , + rep0 += 1U << offset, + SEQ_ALIGN); + } while (++offset < ALIGN_BITS); +#else + case SEQ_ALIGN0: + rc_bit(coder->pos_align[symbol], , + rep0 += 1, SEQ_ALIGN0); + case SEQ_ALIGN1: + rc_bit(coder->pos_align[symbol], , + rep0 += 2, SEQ_ALIGN1); + case SEQ_ALIGN2: + rc_bit(coder->pos_align[symbol], , + rep0 += 4, SEQ_ALIGN2); + case SEQ_ALIGN3: + // Like in SEQ_DIST_MODEL, we don't + // need "symbol" for anything else + // than indexing the probability array. + rc_bit_last(coder->pos_align[symbol], , + rep0 += 8, SEQ_ALIGN3); +#endif + + if (rep0 == UINT32_MAX) { + // End of payload marker was + // found. It may only be + // present if + // - uncompressed size is + // unknown or + // - after known uncompressed + // size amount of bytes has + // been decompressed and + // caller has indicated + // that EOPM might be used + // (it's not allowed in + // LZMA2). + if (!eopm_is_valid) { + ret = LZMA_DATA_ERROR; + goto out; + } + + case SEQ_EOPM: + // LZMA1 stream with + // end-of-payload marker. + rc_normalize(SEQ_EOPM); + ret = rc_is_finished(rc) + ? LZMA_STREAM_END + : LZMA_DATA_ERROR; + goto out; + } + } + } + + // Validate the distance we just decoded. + if (unlikely(!dict_is_distance_valid(&dict, rep0))) { + ret = LZMA_DATA_ERROR; + goto out; + } + + } else { + rc_update_1(coder->is_rep[state]); + + // Repeated match + // + // The match distance is a value that we have had + // earlier. The latest four match distances are + // available as rep0, rep1, rep2 and rep3. We will + // now decode which of them is the new distance. + // + // There cannot be a match if we haven't produced + // any output, so check that first. + if (unlikely(!dict_is_distance_valid(&dict, 0))) { + ret = LZMA_DATA_ERROR; + goto out; + } + + case SEQ_IS_REP0: + rc_if_0(coder->is_rep0[state], SEQ_IS_REP0) { + rc_update_0(coder->is_rep0[state]); + // The distance is rep0. + + case SEQ_IS_REP0_LONG: + rc_if_0(coder->is_rep0_long[state][pos_state], + SEQ_IS_REP0_LONG) { + rc_update_0(coder->is_rep0_long[ + state][pos_state]); + + update_short_rep(state); + + case SEQ_SHORTREP: + if (unlikely(dict_put(&dict, dict_get( + &dict, rep0)))) { + coder->sequence = SEQ_SHORTREP; + goto out; + } + + continue; + } + + // Repeating more than one byte at + // distance of rep0. + rc_update_1(coder->is_rep0_long[ + state][pos_state]); + + } else { + rc_update_1(coder->is_rep0[state]); + + case SEQ_IS_REP1: + // The distance is rep1, rep2 or rep3. Once + // we find out which one of these three, it + // is stored to rep0 and rep1, rep2 and rep3 + // are updated accordingly. + rc_if_0(coder->is_rep1[state], SEQ_IS_REP1) { + rc_update_0(coder->is_rep1[state]); + + const uint32_t distance = rep1; + rep1 = rep0; + rep0 = distance; + + } else { + rc_update_1(coder->is_rep1[state]); + case SEQ_IS_REP2: + rc_if_0(coder->is_rep2[state], + SEQ_IS_REP2) { + rc_update_0(coder->is_rep2[ + state]); + + const uint32_t distance = rep2; + rep2 = rep1; + rep1 = rep0; + rep0 = distance; + + } else { + rc_update_1(coder->is_rep2[ + state]); + + const uint32_t distance = rep3; + rep3 = rep2; + rep2 = rep1; + rep1 = rep0; + rep0 = distance; + } + } + } + + update_long_rep(state); + + // Decode the length of the repeated match. + len_decode(len, coder->rep_len_decoder, + pos_state, SEQ_REP_LEN); + } + + ///////////////////////////////// + // Repeat from history buffer. // + ///////////////////////////////// + + // The length is always between these limits. There is no way + // to trigger the algorithm to set len outside this range. + assert(len >= MATCH_LEN_MIN); + assert(len <= MATCH_LEN_MAX); + + case SEQ_COPY: + // Repeat len bytes from distance of rep0. + if (unlikely(dict_repeat(&dict, rep0, &len))) { + coder->sequence = SEQ_COPY; + goto out; + } + } + +out: + // Save state + + // NOTE: Must not copy dict.limit. + dictptr->pos = dict.pos; + dictptr->full = dict.full; + + rc_from_local(coder->rc, *in_pos); + + coder->state = state; + coder->rep0 = rep0; + coder->rep1 = rep1; + coder->rep2 = rep2; + coder->rep3 = rep3; + + coder->probs = probs; + coder->symbol = symbol; + coder->limit = limit; + coder->offset = offset; + coder->len = len; + + // Update the remaining amount of uncompressed data if uncompressed + // size was known. + if (coder->uncompressed_size != LZMA_VLI_UNKNOWN) { + coder->uncompressed_size -= dict.pos - dict_start; + + // If we have gotten all the output but the decoder wants + // to write more output, the file is corrupt. There are + // three SEQ values where output is produced. + if (coder->uncompressed_size == 0 && ret == LZMA_OK + && (coder->sequence == SEQ_LITERAL_WRITE + || coder->sequence == SEQ_SHORTREP + || coder->sequence == SEQ_COPY)) + ret = LZMA_DATA_ERROR; + } + + if (ret == LZMA_STREAM_END) { + // Reset the range decoder so that it is ready to reinitialize + // for a new LZMA2 chunk. + rc_reset(coder->rc); + coder->sequence = SEQ_IS_MATCH; + } + + return ret; +} + + + +static void +lzma_decoder_uncompressed(void *coder_ptr, lzma_vli uncompressed_size, + bool allow_eopm) +{ + lzma_lzma1_decoder *coder = coder_ptr; + coder->uncompressed_size = uncompressed_size; + coder->allow_eopm = allow_eopm; +} + + +static void +lzma_decoder_reset(void *coder_ptr, const void *opt) +{ + lzma_lzma1_decoder *coder = coder_ptr; + const lzma_options_lzma *options = opt; + + // NOTE: We assume that lc/lp/pb are valid since they were + // successfully decoded with lzma_lzma_decode_properties(). + + // Calculate pos_mask. We don't need pos_bits as is for anything. + coder->pos_mask = (1U << options->pb) - 1; + + // Initialize the literal decoder. + literal_init(coder->literal, options->lc, options->lp); + + coder->literal_context_bits = options->lc; + coder->literal_pos_mask = (1U << options->lp) - 1; + + // State + coder->state = STATE_LIT_LIT; + coder->rep0 = 0; + coder->rep1 = 0; + coder->rep2 = 0; + coder->rep3 = 0; + coder->pos_mask = (1U << options->pb) - 1; + + // Range decoder + rc_reset(coder->rc); + + // Bit and bittree decoders + for (uint32_t i = 0; i < STATES; ++i) { + for (uint32_t j = 0; j <= coder->pos_mask; ++j) { + bit_reset(coder->is_match[i][j]); + bit_reset(coder->is_rep0_long[i][j]); + } + + bit_reset(coder->is_rep[i]); + bit_reset(coder->is_rep0[i]); + bit_reset(coder->is_rep1[i]); + bit_reset(coder->is_rep2[i]); + } + + for (uint32_t i = 0; i < DIST_STATES; ++i) + bittree_reset(coder->dist_slot[i], DIST_SLOT_BITS); + + for (uint32_t i = 0; i < FULL_DISTANCES - DIST_MODEL_END; ++i) + bit_reset(coder->pos_special[i]); + + bittree_reset(coder->pos_align, ALIGN_BITS); + + // Len decoders (also bit/bittree) + const uint32_t num_pos_states = 1U << options->pb; + bit_reset(coder->match_len_decoder.choice); + bit_reset(coder->match_len_decoder.choice2); + bit_reset(coder->rep_len_decoder.choice); + bit_reset(coder->rep_len_decoder.choice2); + + for (uint32_t pos_state = 0; pos_state < num_pos_states; ++pos_state) { + bittree_reset(coder->match_len_decoder.low[pos_state], + LEN_LOW_BITS); + bittree_reset(coder->match_len_decoder.mid[pos_state], + LEN_MID_BITS); + + bittree_reset(coder->rep_len_decoder.low[pos_state], + LEN_LOW_BITS); + bittree_reset(coder->rep_len_decoder.mid[pos_state], + LEN_MID_BITS); + } + + bittree_reset(coder->match_len_decoder.high, LEN_HIGH_BITS); + bittree_reset(coder->rep_len_decoder.high, LEN_HIGH_BITS); + + coder->sequence = SEQ_IS_MATCH; + coder->probs = NULL; + coder->symbol = 0; + coder->limit = 0; + coder->offset = 0; + coder->len = 0; + + return; +} + + +extern lzma_ret +lzma_lzma_decoder_create(lzma_lz_decoder *lz, const lzma_allocator *allocator, + const lzma_options_lzma *options, lzma_lz_options *lz_options) +{ + if (lz->coder == NULL) { + lz->coder = lzma_alloc(sizeof(lzma_lzma1_decoder), allocator); + if (lz->coder == NULL) + return LZMA_MEM_ERROR; + + lz->code = &lzma_decode; + lz->reset = &lzma_decoder_reset; + lz->set_uncompressed = &lzma_decoder_uncompressed; + } + + // All dictionary sizes are OK here. LZ decoder will take care of + // the special cases. + lz_options->dict_size = options->dict_size; + lz_options->preset_dict = options->preset_dict; + lz_options->preset_dict_size = options->preset_dict_size; + + return LZMA_OK; +} + + +/// Allocate and initialize LZMA decoder. This is used only via LZ +/// initialization (lzma_lzma_decoder_init() passes function pointer to +/// the LZ initialization). +static lzma_ret +lzma_decoder_init(lzma_lz_decoder *lz, const lzma_allocator *allocator, + lzma_vli id, const void *options, lzma_lz_options *lz_options) +{ + if (!is_lclppb_valid(options)) + return LZMA_PROG_ERROR; + + lzma_vli uncomp_size = LZMA_VLI_UNKNOWN; + bool allow_eopm = true; + + if (id == LZMA_FILTER_LZMA1EXT) { + const lzma_options_lzma *opt = options; + + // Only one flag is supported. + if (opt->ext_flags & ~LZMA_LZMA1EXT_ALLOW_EOPM) + return LZMA_OPTIONS_ERROR; + + // FIXME? Using lzma_vli instead of uint64_t is weird because + // this has nothing to do with .xz headers and variable-length + // integer encoding. On the other hand, using LZMA_VLI_UNKNOWN + // instead of UINT64_MAX is clearer when unknown size is + // meant. A problem with using lzma_vli is that now we + // allow > LZMA_VLI_MAX which is fine in this file but + // it's still confusing. Note that alone_decoder.c also + // allows > LZMA_VLI_MAX when setting uncompressed size. + uncomp_size = opt->ext_size_low + + ((uint64_t)(opt->ext_size_high) << 32); + allow_eopm = (opt->ext_flags & LZMA_LZMA1EXT_ALLOW_EOPM) != 0 + || uncomp_size == LZMA_VLI_UNKNOWN; + } + + return_if_error(lzma_lzma_decoder_create( + lz, allocator, options, lz_options)); + + lzma_decoder_reset(lz->coder, options); + lzma_decoder_uncompressed(lz->coder, uncomp_size, allow_eopm); + + return LZMA_OK; +} + + +extern lzma_ret +lzma_lzma_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters) +{ + // LZMA can only be the last filter in the chain. This is enforced + // by the raw_decoder initialization. + assert(filters[1].init == NULL); + + return lzma_lz_decoder_init(next, allocator, filters, + &lzma_decoder_init); +} + + +extern bool +lzma_lzma_lclppb_decode(lzma_options_lzma *options, uint8_t byte) +{ + if (byte > (4 * 5 + 4) * 9 + 8) + return true; + + // See the file format specification to understand this. + options->pb = byte / (9 * 5); + byte -= options->pb * 9 * 5; + options->lp = byte / 9; + options->lc = byte - options->lp * 9; + + return options->lc + options->lp > LZMA_LCLP_MAX; +} + + +extern uint64_t +lzma_lzma_decoder_memusage_nocheck(const void *options) +{ + const lzma_options_lzma *const opt = options; + return sizeof(lzma_lzma1_decoder) + + lzma_lz_decoder_memusage(opt->dict_size); +} + + +extern uint64_t +lzma_lzma_decoder_memusage(const void *options) +{ + if (!is_lclppb_valid(options)) + return UINT64_MAX; + + return lzma_lzma_decoder_memusage_nocheck(options); +} + + +extern lzma_ret +lzma_lzma_props_decode(void **options, const lzma_allocator *allocator, + const uint8_t *props, size_t props_size) +{ + if (props_size != 5) + return LZMA_OPTIONS_ERROR; + + lzma_options_lzma *opt + = lzma_alloc(sizeof(lzma_options_lzma), allocator); + if (opt == NULL) + return LZMA_MEM_ERROR; + + if (lzma_lzma_lclppb_decode(opt, props[0])) + goto error; + + // All dictionary sizes are accepted, including zero. LZ decoder + // will automatically use a dictionary at least a few KiB even if + // a smaller dictionary is requested. + opt->dict_size = read32le(props + 1); + + opt->preset_dict = NULL; + opt->preset_dict_size = 0; + + *options = opt; + + return LZMA_OK; + +error: + lzma_free(opt, allocator); + return LZMA_OPTIONS_ERROR; +} diff --git a/src/liblzma/lzma/lzma_decoder.h b/src/liblzma/lzma/lzma_decoder.h new file mode 100644 index 0000000..1427bc2 --- /dev/null +++ b/src/liblzma/lzma/lzma_decoder.h @@ -0,0 +1,53 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_decoder.h +/// \brief LZMA decoder API +/// +// 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_LZMA_DECODER_H +#define LZMA_LZMA_DECODER_H + +#include "common.h" + + +/// Allocates and initializes LZMA decoder +extern lzma_ret lzma_lzma_decoder_init(lzma_next_coder *next, + const lzma_allocator *allocator, + const lzma_filter_info *filters); + +extern uint64_t lzma_lzma_decoder_memusage(const void *options); + +extern lzma_ret lzma_lzma_props_decode( + void **options, const lzma_allocator *allocator, + const uint8_t *props, size_t props_size); + + +/// \brief Decodes the LZMA Properties byte (lc/lp/pb) +/// +/// \return true if error occurred, false on success +/// +extern bool lzma_lzma_lclppb_decode( + lzma_options_lzma *options, uint8_t byte); + + +#ifdef LZMA_LZ_DECODER_H +/// Allocate and setup function pointers only. This is used by LZMA1 and +/// LZMA2 decoders. +extern lzma_ret lzma_lzma_decoder_create( + lzma_lz_decoder *lz, const lzma_allocator *allocator, + const lzma_options_lzma *opt, lzma_lz_options *lz_options); + +/// Gets memory usage without validating lc/lp/pb. This is used by LZMA2 +/// decoder, because raw LZMA2 decoding doesn't need lc/lp/pb. +extern uint64_t lzma_lzma_decoder_memusage_nocheck(const void *options); + +#endif + +#endif diff --git a/src/liblzma/lzma/lzma_encoder.c b/src/liblzma/lzma/lzma_encoder.c new file mode 100644 index 0000000..559c63e --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder.c @@ -0,0 +1,784 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder.c +/// \brief LZMA encoder +/// +// Authors: Igor Pavlov +// Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lzma2_encoder.h" +#include "lzma_encoder_private.h" +#include "fastpos.h" + + +///////////// +// Literal // +///////////// + +static inline void +literal_matched(lzma_range_encoder *rc, probability *subcoder, + uint32_t match_byte, uint32_t symbol) +{ + uint32_t offset = 0x100; + symbol += UINT32_C(1) << 8; + + do { + match_byte <<= 1; + const uint32_t match_bit = match_byte & offset; + const uint32_t subcoder_index + = offset + match_bit + (symbol >> 8); + const uint32_t bit = (symbol >> 7) & 1; + rc_bit(rc, &subcoder[subcoder_index], bit); + + symbol <<= 1; + offset &= ~(match_byte ^ symbol); + + } while (symbol < (UINT32_C(1) << 16)); +} + + +static inline void +literal(lzma_lzma1_encoder *coder, lzma_mf *mf, uint32_t position) +{ + // Locate the literal byte to be encoded and the subcoder. + const uint8_t cur_byte = mf->buffer[ + mf->read_pos - mf->read_ahead]; + probability *subcoder = literal_subcoder(coder->literal, + coder->literal_context_bits, coder->literal_pos_mask, + position, mf->buffer[mf->read_pos - mf->read_ahead - 1]); + + if (is_literal_state(coder->state)) { + // Previous LZMA-symbol was a literal. Encode a normal + // literal without a match byte. + rc_bittree(&coder->rc, subcoder, 8, cur_byte); + } else { + // Previous LZMA-symbol was a match. Use the last byte of + // the match as a "match byte". That is, compare the bits + // of the current literal and the match byte. + const uint8_t match_byte = mf->buffer[ + mf->read_pos - coder->reps[0] - 1 + - mf->read_ahead]; + literal_matched(&coder->rc, subcoder, match_byte, cur_byte); + } + + update_literal(coder->state); +} + + +////////////////// +// Match length // +////////////////// + +static void +length_update_prices(lzma_length_encoder *lc, const uint32_t pos_state) +{ + const uint32_t table_size = lc->table_size; + lc->counters[pos_state] = table_size; + + const uint32_t a0 = rc_bit_0_price(lc->choice); + const uint32_t a1 = rc_bit_1_price(lc->choice); + const uint32_t b0 = a1 + rc_bit_0_price(lc->choice2); + const uint32_t b1 = a1 + rc_bit_1_price(lc->choice2); + uint32_t *const prices = lc->prices[pos_state]; + + uint32_t i; + for (i = 0; i < table_size && i < LEN_LOW_SYMBOLS; ++i) + prices[i] = a0 + rc_bittree_price(lc->low[pos_state], + LEN_LOW_BITS, i); + + for (; i < table_size && i < LEN_LOW_SYMBOLS + LEN_MID_SYMBOLS; ++i) + prices[i] = b0 + rc_bittree_price(lc->mid[pos_state], + LEN_MID_BITS, i - LEN_LOW_SYMBOLS); + + for (; i < table_size; ++i) + prices[i] = b1 + rc_bittree_price(lc->high, LEN_HIGH_BITS, + i - LEN_LOW_SYMBOLS - LEN_MID_SYMBOLS); + + return; +} + + +static inline void +length(lzma_range_encoder *rc, lzma_length_encoder *lc, + const uint32_t pos_state, uint32_t len, const bool fast_mode) +{ + assert(len <= MATCH_LEN_MAX); + len -= MATCH_LEN_MIN; + + if (len < LEN_LOW_SYMBOLS) { + rc_bit(rc, &lc->choice, 0); + rc_bittree(rc, lc->low[pos_state], LEN_LOW_BITS, len); + } else { + rc_bit(rc, &lc->choice, 1); + len -= LEN_LOW_SYMBOLS; + + if (len < LEN_MID_SYMBOLS) { + rc_bit(rc, &lc->choice2, 0); + rc_bittree(rc, lc->mid[pos_state], LEN_MID_BITS, len); + } else { + rc_bit(rc, &lc->choice2, 1); + len -= LEN_MID_SYMBOLS; + rc_bittree(rc, lc->high, LEN_HIGH_BITS, len); + } + } + + // Only getoptimum uses the prices so don't update the table when + // in fast mode. + if (!fast_mode) + if (--lc->counters[pos_state] == 0) + length_update_prices(lc, pos_state); +} + + +/////////// +// Match // +/////////// + +static inline void +match(lzma_lzma1_encoder *coder, const uint32_t pos_state, + const uint32_t distance, const uint32_t len) +{ + update_match(coder->state); + + length(&coder->rc, &coder->match_len_encoder, pos_state, len, + coder->fast_mode); + + const uint32_t dist_slot = get_dist_slot(distance); + const uint32_t dist_state = get_dist_state(len); + rc_bittree(&coder->rc, coder->dist_slot[dist_state], + DIST_SLOT_BITS, dist_slot); + + if (dist_slot >= DIST_MODEL_START) { + const uint32_t footer_bits = (dist_slot >> 1) - 1; + const uint32_t base = (2 | (dist_slot & 1)) << footer_bits; + const uint32_t dist_reduced = distance - base; + + if (dist_slot < DIST_MODEL_END) { + // Careful here: base - dist_slot - 1 can be -1, but + // rc_bittree_reverse starts at probs[1], not probs[0]. + rc_bittree_reverse(&coder->rc, + coder->dist_special + base - dist_slot - 1, + footer_bits, dist_reduced); + } else { + rc_direct(&coder->rc, dist_reduced >> ALIGN_BITS, + footer_bits - ALIGN_BITS); + rc_bittree_reverse( + &coder->rc, coder->dist_align, + ALIGN_BITS, dist_reduced & ALIGN_MASK); + ++coder->align_price_count; + } + } + + coder->reps[3] = coder->reps[2]; + coder->reps[2] = coder->reps[1]; + coder->reps[1] = coder->reps[0]; + coder->reps[0] = distance; + ++coder->match_price_count; +} + + +//////////////////// +// Repeated match // +//////////////////// + +static inline void +rep_match(lzma_lzma1_encoder *coder, const uint32_t pos_state, + const uint32_t rep, const uint32_t len) +{ + if (rep == 0) { + rc_bit(&coder->rc, &coder->is_rep0[coder->state], 0); + rc_bit(&coder->rc, + &coder->is_rep0_long[coder->state][pos_state], + len != 1); + } else { + const uint32_t distance = coder->reps[rep]; + rc_bit(&coder->rc, &coder->is_rep0[coder->state], 1); + + if (rep == 1) { + rc_bit(&coder->rc, &coder->is_rep1[coder->state], 0); + } else { + rc_bit(&coder->rc, &coder->is_rep1[coder->state], 1); + rc_bit(&coder->rc, &coder->is_rep2[coder->state], + rep - 2); + + if (rep == 3) + coder->reps[3] = coder->reps[2]; + + coder->reps[2] = coder->reps[1]; + } + + coder->reps[1] = coder->reps[0]; + coder->reps[0] = distance; + } + + if (len == 1) { + update_short_rep(coder->state); + } else { + length(&coder->rc, &coder->rep_len_encoder, pos_state, len, + coder->fast_mode); + update_long_rep(coder->state); + } +} + + +////////// +// Main // +////////// + +static void +encode_symbol(lzma_lzma1_encoder *coder, lzma_mf *mf, + uint32_t back, uint32_t len, uint32_t position) +{ + const uint32_t pos_state = position & coder->pos_mask; + + if (back == UINT32_MAX) { + // Literal i.e. eight-bit byte + assert(len == 1); + rc_bit(&coder->rc, + &coder->is_match[coder->state][pos_state], 0); + literal(coder, mf, position); + } else { + // Some type of match + rc_bit(&coder->rc, + &coder->is_match[coder->state][pos_state], 1); + + if (back < REPS) { + // It's a repeated match i.e. the same distance + // has been used earlier. + rc_bit(&coder->rc, &coder->is_rep[coder->state], 1); + rep_match(coder, pos_state, back, len); + } else { + // Normal match + rc_bit(&coder->rc, &coder->is_rep[coder->state], 0); + match(coder, pos_state, back - REPS, len); + } + } + + assert(mf->read_ahead >= len); + mf->read_ahead -= len; +} + + +static bool +encode_init(lzma_lzma1_encoder *coder, lzma_mf *mf) +{ + assert(mf_position(mf) == 0); + assert(coder->uncomp_size == 0); + + if (mf->read_pos == mf->read_limit) { + if (mf->action == LZMA_RUN) + return false; // We cannot do anything. + + // We are finishing (we cannot get here when flushing). + assert(mf->write_pos == mf->read_pos); + assert(mf->action == LZMA_FINISH); + } else { + // Do the actual initialization. The first LZMA symbol must + // always be a literal. + mf_skip(mf, 1); + mf->read_ahead = 0; + rc_bit(&coder->rc, &coder->is_match[0][0], 0); + rc_bittree(&coder->rc, coder->literal[0], 8, mf->buffer[0]); + ++coder->uncomp_size; + } + + // Initialization is done (except if empty file). + coder->is_initialized = true; + + return true; +} + + +static void +encode_eopm(lzma_lzma1_encoder *coder, uint32_t position) +{ + const uint32_t pos_state = position & coder->pos_mask; + rc_bit(&coder->rc, &coder->is_match[coder->state][pos_state], 1); + rc_bit(&coder->rc, &coder->is_rep[coder->state], 0); + match(coder, pos_state, UINT32_MAX, MATCH_LEN_MIN); +} + + +/// Number of bytes that a single encoding loop in lzma_lzma_encode() can +/// consume from the dictionary. This limit comes from lzma_lzma_optimum() +/// and may need to be updated if that function is significantly modified. +#define LOOP_INPUT_MAX (OPTS + 1) + + +extern lzma_ret +lzma_lzma_encode(lzma_lzma1_encoder *restrict coder, lzma_mf *restrict mf, + uint8_t *restrict out, size_t *restrict out_pos, + size_t out_size, uint32_t limit) +{ + // Initialize the stream if no data has been encoded yet. + if (!coder->is_initialized && !encode_init(coder, mf)) + return LZMA_OK; + + // Encode pending output bytes from the range encoder. + // At the start of the stream, encode_init() encodes one literal. + // Later there can be pending output only with LZMA1 because LZMA2 + // ensures that there is always enough output space. Thus when using + // LZMA2, rc_encode() calls in this function will always return false. + if (rc_encode(&coder->rc, out, out_pos, out_size)) { + // We don't get here with LZMA2. + assert(limit == UINT32_MAX); + return LZMA_OK; + } + + // If the range encoder was flushed in an earlier call to this + // function but there wasn't enough output buffer space, those + // bytes would have now been encoded by the above rc_encode() call + // and the stream has now been finished. This can only happen with + // LZMA1 as LZMA2 always provides enough output buffer space. + if (coder->is_flushed) { + assert(limit == UINT32_MAX); + return LZMA_STREAM_END; + } + + while (true) { + // With LZMA2 we need to take care that compressed size of + // a chunk doesn't get too big. + // FIXME? Check if this could be improved. + if (limit != UINT32_MAX + && (mf->read_pos - mf->read_ahead >= limit + || *out_pos + rc_pending(&coder->rc) + >= LZMA2_CHUNK_MAX + - LOOP_INPUT_MAX)) + break; + + // Check that there is some input to process. + if (mf->read_pos >= mf->read_limit) { + if (mf->action == LZMA_RUN) + return LZMA_OK; + + if (mf->read_ahead == 0) + break; + } + + // Get optimal match (repeat position and length). + // Value ranges for pos: + // - [0, REPS): repeated match + // - [REPS, UINT32_MAX): + // match at (pos - REPS) + // - UINT32_MAX: not a match but a literal + // Value ranges for len: + // - [MATCH_LEN_MIN, MATCH_LEN_MAX] + uint32_t len; + uint32_t back; + + if (coder->fast_mode) + lzma_lzma_optimum_fast(coder, mf, &back, &len); + else + lzma_lzma_optimum_normal(coder, mf, &back, &len, + (uint32_t)(coder->uncomp_size)); + + encode_symbol(coder, mf, back, len, + (uint32_t)(coder->uncomp_size)); + + // If output size limiting is active (out_limit != 0), check + // if encoding this LZMA symbol would make the output size + // exceed the specified limit. + if (coder->out_limit != 0 && rc_encode_dummy( + &coder->rc, coder->out_limit)) { + // The most recent LZMA symbol would make the output + // too big. Throw it away. + rc_forget(&coder->rc); + + // FIXME: Tell the LZ layer to not read more input as + // it would be waste of time. This doesn't matter if + // output-size-limited encoding is done with a single + // call though. + + break; + } + + // This symbol will be encoded so update the uncompressed size. + coder->uncomp_size += len; + + // Encode the LZMA symbol. + if (rc_encode(&coder->rc, out, out_pos, out_size)) { + // Once again, this can only happen with LZMA1. + assert(limit == UINT32_MAX); + return LZMA_OK; + } + } + + // Make the uncompressed size available to the application. + if (coder->uncomp_size_ptr != NULL) + *coder->uncomp_size_ptr = coder->uncomp_size; + + // LZMA2 doesn't use EOPM at LZMA level. + // + // Plain LZMA streams without EOPM aren't supported except when + // output size limiting is enabled. + if (coder->use_eopm) + encode_eopm(coder, (uint32_t)(coder->uncomp_size)); + + // Flush the remaining bytes from the range encoder. + rc_flush(&coder->rc); + + // Copy the remaining bytes to the output buffer. If there + // isn't enough output space, we will copy out the remaining + // bytes on the next call to this function. + if (rc_encode(&coder->rc, out, out_pos, out_size)) { + // This cannot happen with LZMA2. + assert(limit == UINT32_MAX); + + coder->is_flushed = true; + return LZMA_OK; + } + + return LZMA_STREAM_END; +} + + +static lzma_ret +lzma_encode(void *coder, lzma_mf *restrict mf, + uint8_t *restrict out, size_t *restrict out_pos, + size_t out_size) +{ + // Plain LZMA has no support for sync-flushing. + if (unlikely(mf->action == LZMA_SYNC_FLUSH)) + return LZMA_OPTIONS_ERROR; + + return lzma_lzma_encode(coder, mf, out, out_pos, out_size, UINT32_MAX); +} + + +static lzma_ret +lzma_lzma_set_out_limit( + void *coder_ptr, uint64_t *uncomp_size, uint64_t out_limit) +{ + // Minimum output size is 5 bytes but that cannot hold any output + // so we use 6 bytes. + if (out_limit < 6) + return LZMA_BUF_ERROR; + + lzma_lzma1_encoder *coder = coder_ptr; + coder->out_limit = out_limit; + coder->uncomp_size_ptr = uncomp_size; + coder->use_eopm = false; + return LZMA_OK; +} + + +//////////////////// +// Initialization // +//////////////////// + +static bool +is_options_valid(const lzma_options_lzma *options) +{ + // Validate some of the options. LZ encoder validates nice_len too + // but we need a valid value here earlier. + return is_lclppb_valid(options) + && options->nice_len >= MATCH_LEN_MIN + && options->nice_len <= MATCH_LEN_MAX + && (options->mode == LZMA_MODE_FAST + || options->mode == LZMA_MODE_NORMAL); +} + + +static void +set_lz_options(lzma_lz_options *lz_options, const lzma_options_lzma *options) +{ + // LZ encoder initialization does the validation for these so we + // don't need to validate here. + lz_options->before_size = OPTS; + lz_options->dict_size = options->dict_size; + lz_options->after_size = LOOP_INPUT_MAX; + lz_options->match_len_max = MATCH_LEN_MAX; + lz_options->nice_len = my_max(mf_get_hash_bytes(options->mf), + options->nice_len); + lz_options->match_finder = options->mf; + lz_options->depth = options->depth; + lz_options->preset_dict = options->preset_dict; + lz_options->preset_dict_size = options->preset_dict_size; + return; +} + + +static void +length_encoder_reset(lzma_length_encoder *lencoder, + const uint32_t num_pos_states, const bool fast_mode) +{ + bit_reset(lencoder->choice); + bit_reset(lencoder->choice2); + + for (size_t pos_state = 0; pos_state < num_pos_states; ++pos_state) { + bittree_reset(lencoder->low[pos_state], LEN_LOW_BITS); + bittree_reset(lencoder->mid[pos_state], LEN_MID_BITS); + } + + bittree_reset(lencoder->high, LEN_HIGH_BITS); + + if (!fast_mode) + for (uint32_t pos_state = 0; pos_state < num_pos_states; + ++pos_state) + length_update_prices(lencoder, pos_state); + + return; +} + + +extern lzma_ret +lzma_lzma_encoder_reset(lzma_lzma1_encoder *coder, + const lzma_options_lzma *options) +{ + if (!is_options_valid(options)) + return LZMA_OPTIONS_ERROR; + + coder->pos_mask = (1U << options->pb) - 1; + coder->literal_context_bits = options->lc; + coder->literal_pos_mask = (1U << options->lp) - 1; + + // Range coder + rc_reset(&coder->rc); + + // State + coder->state = STATE_LIT_LIT; + for (size_t i = 0; i < REPS; ++i) + coder->reps[i] = 0; + + literal_init(coder->literal, options->lc, options->lp); + + // Bit encoders + for (size_t i = 0; i < STATES; ++i) { + for (size_t j = 0; j <= coder->pos_mask; ++j) { + bit_reset(coder->is_match[i][j]); + bit_reset(coder->is_rep0_long[i][j]); + } + + bit_reset(coder->is_rep[i]); + bit_reset(coder->is_rep0[i]); + bit_reset(coder->is_rep1[i]); + bit_reset(coder->is_rep2[i]); + } + + for (size_t i = 0; i < FULL_DISTANCES - DIST_MODEL_END; ++i) + bit_reset(coder->dist_special[i]); + + // Bit tree encoders + for (size_t i = 0; i < DIST_STATES; ++i) + bittree_reset(coder->dist_slot[i], DIST_SLOT_BITS); + + bittree_reset(coder->dist_align, ALIGN_BITS); + + // Length encoders + length_encoder_reset(&coder->match_len_encoder, + 1U << options->pb, coder->fast_mode); + + length_encoder_reset(&coder->rep_len_encoder, + 1U << options->pb, coder->fast_mode); + + // Price counts are incremented every time appropriate probabilities + // are changed. price counts are set to zero when the price tables + // are updated, which is done when the appropriate price counts have + // big enough value, and lzma_mf.read_ahead == 0 which happens at + // least every OPTS (a few thousand) possible price count increments. + // + // By resetting price counts to UINT32_MAX / 2, we make sure that the + // price tables will be initialized before they will be used (since + // the value is definitely big enough), and that it is OK to increment + // price counts without risk of integer overflow (since UINT32_MAX / 2 + // is small enough). The current code doesn't increment price counts + // before initializing price tables, but it maybe done in future if + // we add support for saving the state between LZMA2 chunks. + coder->match_price_count = UINT32_MAX / 2; + coder->align_price_count = UINT32_MAX / 2; + + coder->opts_end_index = 0; + coder->opts_current_index = 0; + + return LZMA_OK; +} + + +extern lzma_ret +lzma_lzma_encoder_create(void **coder_ptr, const lzma_allocator *allocator, + lzma_vli id, const lzma_options_lzma *options, + lzma_lz_options *lz_options) +{ + assert(id == LZMA_FILTER_LZMA1 || id == LZMA_FILTER_LZMA1EXT + || id == LZMA_FILTER_LZMA2); + + // Allocate lzma_lzma1_encoder if it wasn't already allocated. + if (*coder_ptr == NULL) { + *coder_ptr = lzma_alloc(sizeof(lzma_lzma1_encoder), allocator); + if (*coder_ptr == NULL) + return LZMA_MEM_ERROR; + } + + lzma_lzma1_encoder *coder = *coder_ptr; + + // Set compression mode. Note that we haven't validated the options + // yet. Invalid options will get rejected by lzma_lzma_encoder_reset() + // call at the end of this function. + switch (options->mode) { + case LZMA_MODE_FAST: + coder->fast_mode = true; + break; + + case LZMA_MODE_NORMAL: { + coder->fast_mode = false; + + // Set dist_table_size. + // Round the dictionary size up to next 2^n. + // + // Currently the maximum encoder dictionary size + // is 1.5 GiB due to lz_encoder.c and here we need + // to be below 2 GiB to make the rounded up value + // fit in an uint32_t and avoid an infinite while-loop + // (and undefined behavior due to a too large shift). + // So do the same check as in LZ encoder, + // limiting to 1.5 GiB. + if (options->dict_size > (UINT32_C(1) << 30) + + (UINT32_C(1) << 29)) + return LZMA_OPTIONS_ERROR; + + uint32_t log_size = 0; + while ((UINT32_C(1) << log_size) < options->dict_size) + ++log_size; + + coder->dist_table_size = log_size * 2; + + // Length encoders' price table size + const uint32_t nice_len = my_max( + mf_get_hash_bytes(options->mf), + options->nice_len); + + coder->match_len_encoder.table_size + = nice_len + 1 - MATCH_LEN_MIN; + coder->rep_len_encoder.table_size + = nice_len + 1 - MATCH_LEN_MIN; + break; + } + + default: + return LZMA_OPTIONS_ERROR; + } + + // We don't need to write the first byte as literal if there is + // a non-empty preset dictionary. encode_init() wouldn't even work + // if there is a non-empty preset dictionary, because encode_init() + // assumes that position is zero and previous byte is also zero. + coder->is_initialized = options->preset_dict != NULL + && options->preset_dict_size > 0; + coder->is_flushed = false; + coder->uncomp_size = 0; + coder->uncomp_size_ptr = NULL; + + // Output size limiting is disabled by default. + coder->out_limit = 0; + + // Determine if end marker is wanted: + // - It is never used with LZMA2. + // - It is always used with LZMA_FILTER_LZMA1 (unless + // lzma_lzma_set_out_limit() is called later). + // - LZMA_FILTER_LZMA1EXT has a flag for it in the options. + coder->use_eopm = (id == LZMA_FILTER_LZMA1); + if (id == LZMA_FILTER_LZMA1EXT) { + // Check if unsupported flags are present. + if (options->ext_flags & ~LZMA_LZMA1EXT_ALLOW_EOPM) + return LZMA_OPTIONS_ERROR; + + coder->use_eopm = (options->ext_flags + & LZMA_LZMA1EXT_ALLOW_EOPM) != 0; + + // TODO? As long as there are no filters that change the size + // of the data, it is enough to look at lzma_stream.total_in + // after encoding has been finished to know the uncompressed + // size of the LZMA1 stream. But in the future there could be + // filters that change the size of the data and then total_in + // doesn't work as the LZMA1 stream size might be different + // due to another filter in the chain. The problem is simple + // to solve: Add another flag to ext_flags and then set + // coder->uncomp_size_ptr to the address stored in + // lzma_options_lzma.reserved_ptr2 (or _ptr1). + } + + set_lz_options(lz_options, options); + + return lzma_lzma_encoder_reset(coder, options); +} + + +static lzma_ret +lzma_encoder_init(lzma_lz_encoder *lz, const lzma_allocator *allocator, + lzma_vli id, const void *options, lzma_lz_options *lz_options) +{ + lz->code = &lzma_encode; + lz->set_out_limit = &lzma_lzma_set_out_limit; + return lzma_lzma_encoder_create( + &lz->coder, allocator, id, options, lz_options); +} + + +extern lzma_ret +lzma_lzma_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator, + const lzma_filter_info *filters) +{ + return lzma_lz_encoder_init( + next, allocator, filters, &lzma_encoder_init); +} + + +extern uint64_t +lzma_lzma_encoder_memusage(const void *options) +{ + if (!is_options_valid(options)) + return UINT64_MAX; + + lzma_lz_options lz_options; + set_lz_options(&lz_options, options); + + const uint64_t lz_memusage = lzma_lz_encoder_memusage(&lz_options); + if (lz_memusage == UINT64_MAX) + return UINT64_MAX; + + return (uint64_t)(sizeof(lzma_lzma1_encoder)) + lz_memusage; +} + + +extern bool +lzma_lzma_lclppb_encode(const lzma_options_lzma *options, uint8_t *byte) +{ + if (!is_lclppb_valid(options)) + return true; + + *byte = (options->pb * 5 + options->lp) * 9 + options->lc; + assert(*byte <= (4 * 5 + 4) * 9 + 8); + + return false; +} + + +#ifdef HAVE_ENCODER_LZMA1 +extern lzma_ret +lzma_lzma_props_encode(const void *options, uint8_t *out) +{ + if (options == NULL) + return LZMA_PROG_ERROR; + + const lzma_options_lzma *const opt = options; + + if (lzma_lzma_lclppb_encode(opt, out)) + return LZMA_PROG_ERROR; + + write32le(out + 1, opt->dict_size); + + return LZMA_OK; +} +#endif + + +extern LZMA_API(lzma_bool) +lzma_mode_is_supported(lzma_mode mode) +{ + return mode == LZMA_MODE_FAST || mode == LZMA_MODE_NORMAL; +} diff --git a/src/liblzma/lzma/lzma_encoder.h b/src/liblzma/lzma/lzma_encoder.h new file mode 100644 index 0000000..84d8c91 --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder.h @@ -0,0 +1,59 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder.h +/// \brief LZMA encoder API +/// +// 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_LZMA_ENCODER_H +#define LZMA_LZMA_ENCODER_H + +#include "common.h" + + +typedef struct lzma_lzma1_encoder_s lzma_lzma1_encoder; + + +extern lzma_ret lzma_lzma_encoder_init(lzma_next_coder *next, + const lzma_allocator *allocator, + const lzma_filter_info *filters); + + +extern uint64_t lzma_lzma_encoder_memusage(const void *options); + +extern lzma_ret lzma_lzma_props_encode(const void *options, uint8_t *out); + + +/// Encodes lc/lp/pb into one byte. Returns false on success and true on error. +extern bool lzma_lzma_lclppb_encode( + const lzma_options_lzma *options, uint8_t *byte); + + +#ifdef LZMA_LZ_ENCODER_H + +/// Initializes raw LZMA encoder; this is used by LZMA2. +extern lzma_ret lzma_lzma_encoder_create( + void **coder_ptr, const lzma_allocator *allocator, + lzma_vli id, const lzma_options_lzma *options, + lzma_lz_options *lz_options); + + +/// Resets an already initialized LZMA encoder; this is used by LZMA2. +extern lzma_ret lzma_lzma_encoder_reset( + lzma_lzma1_encoder *coder, const lzma_options_lzma *options); + + +extern lzma_ret lzma_lzma_encode(lzma_lzma1_encoder *restrict coder, + lzma_mf *restrict mf, uint8_t *restrict out, + size_t *restrict out_pos, size_t out_size, + uint32_t read_limit); + +#endif + +#endif diff --git a/src/liblzma/lzma/lzma_encoder_optimum_fast.c b/src/liblzma/lzma/lzma_encoder_optimum_fast.c new file mode 100644 index 0000000..6c53d2b --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder_optimum_fast.c @@ -0,0 +1,170 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder_optimum_fast.c +// +// Author: Igor Pavlov +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lzma_encoder_private.h" +#include "memcmplen.h" + + +#define change_pair(small_dist, big_dist) \ + (((big_dist) >> 7) > (small_dist)) + + +extern void +lzma_lzma_optimum_fast(lzma_lzma1_encoder *restrict coder, + lzma_mf *restrict mf, + uint32_t *restrict back_res, uint32_t *restrict len_res) +{ + const uint32_t nice_len = mf->nice_len; + + uint32_t len_main; + uint32_t matches_count; + if (mf->read_ahead == 0) { + len_main = mf_find(mf, &matches_count, coder->matches); + } else { + assert(mf->read_ahead == 1); + len_main = coder->longest_match_length; + matches_count = coder->matches_count; + } + + const uint8_t *buf = mf_ptr(mf) - 1; + const uint32_t buf_avail = my_min(mf_avail(mf) + 1, MATCH_LEN_MAX); + + if (buf_avail < 2) { + // There's not enough input left to encode a match. + *back_res = UINT32_MAX; + *len_res = 1; + return; + } + + // Look for repeated matches; scan the previous four match distances + uint32_t rep_len = 0; + uint32_t rep_index = 0; + + for (uint32_t i = 0; i < REPS; ++i) { + // Pointer to the beginning of the match candidate + const uint8_t *const buf_back = buf - coder->reps[i] - 1; + + // If the first two bytes (2 == MATCH_LEN_MIN) do not match, + // this rep is not useful. + if (not_equal_16(buf, buf_back)) + continue; + + // The first two bytes matched. + // Calculate the length of the match. + const uint32_t len = lzma_memcmplen( + buf, buf_back, 2, buf_avail); + + // If we have found a repeated match that is at least + // nice_len long, return it immediately. + if (len >= nice_len) { + *back_res = i; + *len_res = len; + mf_skip(mf, len - 1); + return; + } + + if (len > rep_len) { + rep_index = i; + rep_len = len; + } + } + + // We didn't find a long enough repeated match. Encode it as a normal + // match if the match length is at least nice_len. + if (len_main >= nice_len) { + *back_res = coder->matches[matches_count - 1].dist + REPS; + *len_res = len_main; + mf_skip(mf, len_main - 1); + return; + } + + uint32_t back_main = 0; + if (len_main >= 2) { + back_main = coder->matches[matches_count - 1].dist; + + while (matches_count > 1 && len_main == + coder->matches[matches_count - 2].len + 1) { + if (!change_pair(coder->matches[ + matches_count - 2].dist, + back_main)) + break; + + --matches_count; + len_main = coder->matches[matches_count - 1].len; + back_main = coder->matches[matches_count - 1].dist; + } + + if (len_main == 2 && back_main >= 0x80) + len_main = 1; + } + + if (rep_len >= 2) { + if (rep_len + 1 >= len_main + || (rep_len + 2 >= len_main + && back_main > (UINT32_C(1) << 9)) + || (rep_len + 3 >= len_main + && back_main > (UINT32_C(1) << 15))) { + *back_res = rep_index; + *len_res = rep_len; + mf_skip(mf, rep_len - 1); + return; + } + } + + if (len_main < 2 || buf_avail <= 2) { + *back_res = UINT32_MAX; + *len_res = 1; + return; + } + + // Get the matches for the next byte. If we find a better match, + // the current byte is encoded as a literal. + coder->longest_match_length = mf_find(mf, + &coder->matches_count, coder->matches); + + if (coder->longest_match_length >= 2) { + const uint32_t new_dist = coder->matches[ + coder->matches_count - 1].dist; + + if ((coder->longest_match_length >= len_main + && new_dist < back_main) + || (coder->longest_match_length == len_main + 1 + && !change_pair(back_main, new_dist)) + || (coder->longest_match_length > len_main + 1) + || (coder->longest_match_length + 1 >= len_main + && len_main >= 3 + && change_pair(new_dist, back_main))) { + *back_res = UINT32_MAX; + *len_res = 1; + return; + } + } + + // In contrast to LZMA SDK, dictionary could not have been moved + // between mf_find() calls, thus it is safe to just increment + // the old buf pointer instead of recalculating it with mf_ptr(). + ++buf; + + const uint32_t limit = my_max(2, len_main - 1); + + for (uint32_t i = 0; i < REPS; ++i) { + if (memcmp(buf, buf - coder->reps[i] - 1, limit) == 0) { + *back_res = UINT32_MAX; + *len_res = 1; + return; + } + } + + *back_res = back_main + REPS; + *len_res = len_main; + mf_skip(mf, len_main - 2); + return; +} diff --git a/src/liblzma/lzma/lzma_encoder_optimum_normal.c b/src/liblzma/lzma/lzma_encoder_optimum_normal.c new file mode 100644 index 0000000..101c8d4 --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder_optimum_normal.c @@ -0,0 +1,859 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder_optimum_normal.c +// +// Author: Igor Pavlov +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "lzma_encoder_private.h" +#include "fastpos.h" +#include "memcmplen.h" + + +//////////// +// Prices // +//////////// + +static uint32_t +get_literal_price(const lzma_lzma1_encoder *const coder, const uint32_t pos, + const uint32_t prev_byte, const bool match_mode, + uint32_t match_byte, uint32_t symbol) +{ + const probability *const subcoder = literal_subcoder(coder->literal, + coder->literal_context_bits, coder->literal_pos_mask, + pos, prev_byte); + + uint32_t price = 0; + + if (!match_mode) { + price = rc_bittree_price(subcoder, 8, symbol); + } else { + uint32_t offset = 0x100; + symbol += UINT32_C(1) << 8; + + do { + match_byte <<= 1; + + const uint32_t match_bit = match_byte & offset; + const uint32_t subcoder_index + = offset + match_bit + (symbol >> 8); + const uint32_t bit = (symbol >> 7) & 1; + price += rc_bit_price(subcoder[subcoder_index], bit); + + symbol <<= 1; + offset &= ~(match_byte ^ symbol); + + } while (symbol < (UINT32_C(1) << 16)); + } + + return price; +} + + +static inline uint32_t +get_len_price(const lzma_length_encoder *const lencoder, + const uint32_t len, const uint32_t pos_state) +{ + // NOTE: Unlike the other price tables, length prices are updated + // in lzma_encoder.c + return lencoder->prices[pos_state][len - MATCH_LEN_MIN]; +} + + +static inline uint32_t +get_short_rep_price(const lzma_lzma1_encoder *const coder, + const lzma_lzma_state state, const uint32_t pos_state) +{ + return rc_bit_0_price(coder->is_rep0[state]) + + rc_bit_0_price(coder->is_rep0_long[state][pos_state]); +} + + +static inline uint32_t +get_pure_rep_price(const lzma_lzma1_encoder *const coder, const uint32_t rep_index, + const lzma_lzma_state state, uint32_t pos_state) +{ + uint32_t price; + + if (rep_index == 0) { + price = rc_bit_0_price(coder->is_rep0[state]); + price += rc_bit_1_price(coder->is_rep0_long[state][pos_state]); + } else { + price = rc_bit_1_price(coder->is_rep0[state]); + + if (rep_index == 1) { + price += rc_bit_0_price(coder->is_rep1[state]); + } else { + price += rc_bit_1_price(coder->is_rep1[state]); + price += rc_bit_price(coder->is_rep2[state], + rep_index - 2); + } + } + + return price; +} + + +static inline uint32_t +get_rep_price(const lzma_lzma1_encoder *const coder, const uint32_t rep_index, + const uint32_t len, const lzma_lzma_state state, + const uint32_t pos_state) +{ + return get_len_price(&coder->rep_len_encoder, len, pos_state) + + get_pure_rep_price(coder, rep_index, state, pos_state); +} + + +static inline uint32_t +get_dist_len_price(const lzma_lzma1_encoder *const coder, const uint32_t dist, + const uint32_t len, const uint32_t pos_state) +{ + const uint32_t dist_state = get_dist_state(len); + uint32_t price; + + if (dist < FULL_DISTANCES) { + price = coder->dist_prices[dist_state][dist]; + } else { + const uint32_t dist_slot = get_dist_slot_2(dist); + price = coder->dist_slot_prices[dist_state][dist_slot] + + coder->align_prices[dist & ALIGN_MASK]; + } + + price += get_len_price(&coder->match_len_encoder, len, pos_state); + + return price; +} + + +static void +fill_dist_prices(lzma_lzma1_encoder *coder) +{ + for (uint32_t dist_state = 0; dist_state < DIST_STATES; ++dist_state) { + + uint32_t *const dist_slot_prices + = coder->dist_slot_prices[dist_state]; + + // Price to encode the dist_slot. + for (uint32_t dist_slot = 0; + dist_slot < coder->dist_table_size; ++dist_slot) + dist_slot_prices[dist_slot] = rc_bittree_price( + coder->dist_slot[dist_state], + DIST_SLOT_BITS, dist_slot); + + // For matches with distance >= FULL_DISTANCES, add the price + // of the direct bits part of the match distance. (Align bits + // are handled by fill_align_prices()). + for (uint32_t dist_slot = DIST_MODEL_END; + dist_slot < coder->dist_table_size; + ++dist_slot) + dist_slot_prices[dist_slot] += rc_direct_price( + ((dist_slot >> 1) - 1) - ALIGN_BITS); + + // Distances in the range [0, 3] are fully encoded with + // dist_slot, so they are used for coder->dist_prices + // as is. + for (uint32_t i = 0; i < DIST_MODEL_START; ++i) + coder->dist_prices[dist_state][i] + = dist_slot_prices[i]; + } + + // Distances in the range [4, 127] depend on dist_slot and + // dist_special. We do this in a loop separate from the above + // loop to avoid redundant calls to get_dist_slot(). + for (uint32_t i = DIST_MODEL_START; i < FULL_DISTANCES; ++i) { + const uint32_t dist_slot = get_dist_slot(i); + const uint32_t footer_bits = ((dist_slot >> 1) - 1); + const uint32_t base = (2 | (dist_slot & 1)) << footer_bits; + const uint32_t price = rc_bittree_reverse_price( + coder->dist_special + base - dist_slot - 1, + footer_bits, i - base); + + for (uint32_t dist_state = 0; dist_state < DIST_STATES; + ++dist_state) + coder->dist_prices[dist_state][i] + = price + coder->dist_slot_prices[ + dist_state][dist_slot]; + } + + coder->match_price_count = 0; + return; +} + + +static void +fill_align_prices(lzma_lzma1_encoder *coder) +{ + for (uint32_t i = 0; i < ALIGN_SIZE; ++i) + coder->align_prices[i] = rc_bittree_reverse_price( + coder->dist_align, ALIGN_BITS, i); + + coder->align_price_count = 0; + return; +} + + +///////////// +// Optimal // +///////////// + +static inline void +make_literal(lzma_optimal *optimal) +{ + optimal->back_prev = UINT32_MAX; + optimal->prev_1_is_literal = false; +} + + +static inline void +make_short_rep(lzma_optimal *optimal) +{ + optimal->back_prev = 0; + optimal->prev_1_is_literal = false; +} + + +#define is_short_rep(optimal) \ + ((optimal).back_prev == 0) + + +static void +backward(lzma_lzma1_encoder *restrict coder, uint32_t *restrict len_res, + uint32_t *restrict back_res, uint32_t cur) +{ + coder->opts_end_index = cur; + + uint32_t pos_mem = coder->opts[cur].pos_prev; + uint32_t back_mem = coder->opts[cur].back_prev; + + do { + if (coder->opts[cur].prev_1_is_literal) { + make_literal(&coder->opts[pos_mem]); + coder->opts[pos_mem].pos_prev = pos_mem - 1; + + if (coder->opts[cur].prev_2) { + coder->opts[pos_mem - 1].prev_1_is_literal + = false; + coder->opts[pos_mem - 1].pos_prev + = coder->opts[cur].pos_prev_2; + coder->opts[pos_mem - 1].back_prev + = coder->opts[cur].back_prev_2; + } + } + + const uint32_t pos_prev = pos_mem; + const uint32_t back_cur = back_mem; + + back_mem = coder->opts[pos_prev].back_prev; + pos_mem = coder->opts[pos_prev].pos_prev; + + coder->opts[pos_prev].back_prev = back_cur; + coder->opts[pos_prev].pos_prev = cur; + cur = pos_prev; + + } while (cur != 0); + + coder->opts_current_index = coder->opts[0].pos_prev; + *len_res = coder->opts[0].pos_prev; + *back_res = coder->opts[0].back_prev; + + return; +} + + +////////// +// Main // +////////// + +static inline uint32_t +helper1(lzma_lzma1_encoder *restrict coder, lzma_mf *restrict mf, + uint32_t *restrict back_res, uint32_t *restrict len_res, + uint32_t position) +{ + const uint32_t nice_len = mf->nice_len; + + uint32_t len_main; + uint32_t matches_count; + + if (mf->read_ahead == 0) { + len_main = mf_find(mf, &matches_count, coder->matches); + } else { + assert(mf->read_ahead == 1); + len_main = coder->longest_match_length; + matches_count = coder->matches_count; + } + + const uint32_t buf_avail = my_min(mf_avail(mf) + 1, MATCH_LEN_MAX); + if (buf_avail < 2) { + *back_res = UINT32_MAX; + *len_res = 1; + return UINT32_MAX; + } + + const uint8_t *const buf = mf_ptr(mf) - 1; + + uint32_t rep_lens[REPS]; + uint32_t rep_max_index = 0; + + for (uint32_t i = 0; i < REPS; ++i) { + const uint8_t *const buf_back = buf - coder->reps[i] - 1; + + if (not_equal_16(buf, buf_back)) { + rep_lens[i] = 0; + continue; + } + + rep_lens[i] = lzma_memcmplen(buf, buf_back, 2, buf_avail); + + if (rep_lens[i] > rep_lens[rep_max_index]) + rep_max_index = i; + } + + if (rep_lens[rep_max_index] >= nice_len) { + *back_res = rep_max_index; + *len_res = rep_lens[rep_max_index]; + mf_skip(mf, *len_res - 1); + return UINT32_MAX; + } + + + if (len_main >= nice_len) { + *back_res = coder->matches[matches_count - 1].dist + REPS; + *len_res = len_main; + mf_skip(mf, len_main - 1); + return UINT32_MAX; + } + + const uint8_t current_byte = *buf; + const uint8_t match_byte = *(buf - coder->reps[0] - 1); + + if (len_main < 2 && current_byte != match_byte + && rep_lens[rep_max_index] < 2) { + *back_res = UINT32_MAX; + *len_res = 1; + return UINT32_MAX; + } + + coder->opts[0].state = coder->state; + + const uint32_t pos_state = position & coder->pos_mask; + + coder->opts[1].price = rc_bit_0_price( + coder->is_match[coder->state][pos_state]) + + get_literal_price(coder, position, buf[-1], + !is_literal_state(coder->state), + match_byte, current_byte); + + make_literal(&coder->opts[1]); + + const uint32_t match_price = rc_bit_1_price( + coder->is_match[coder->state][pos_state]); + const uint32_t rep_match_price = match_price + + rc_bit_1_price(coder->is_rep[coder->state]); + + if (match_byte == current_byte) { + const uint32_t short_rep_price = rep_match_price + + get_short_rep_price( + coder, coder->state, pos_state); + + if (short_rep_price < coder->opts[1].price) { + coder->opts[1].price = short_rep_price; + make_short_rep(&coder->opts[1]); + } + } + + const uint32_t len_end = my_max(len_main, rep_lens[rep_max_index]); + + if (len_end < 2) { + *back_res = coder->opts[1].back_prev; + *len_res = 1; + return UINT32_MAX; + } + + coder->opts[1].pos_prev = 0; + + for (uint32_t i = 0; i < REPS; ++i) + coder->opts[0].backs[i] = coder->reps[i]; + + uint32_t len = len_end; + do { + coder->opts[len].price = RC_INFINITY_PRICE; + } while (--len >= 2); + + + for (uint32_t i = 0; i < REPS; ++i) { + uint32_t rep_len = rep_lens[i]; + if (rep_len < 2) + continue; + + const uint32_t price = rep_match_price + get_pure_rep_price( + coder, i, coder->state, pos_state); + + do { + const uint32_t cur_and_len_price = price + + get_len_price( + &coder->rep_len_encoder, + rep_len, pos_state); + + if (cur_and_len_price < coder->opts[rep_len].price) { + coder->opts[rep_len].price = cur_and_len_price; + coder->opts[rep_len].pos_prev = 0; + coder->opts[rep_len].back_prev = i; + coder->opts[rep_len].prev_1_is_literal = false; + } + } while (--rep_len >= 2); + } + + + const uint32_t normal_match_price = match_price + + rc_bit_0_price(coder->is_rep[coder->state]); + + len = rep_lens[0] >= 2 ? rep_lens[0] + 1 : 2; + if (len <= len_main) { + uint32_t i = 0; + while (len > coder->matches[i].len) + ++i; + + for(; ; ++len) { + const uint32_t dist = coder->matches[i].dist; + const uint32_t cur_and_len_price = normal_match_price + + get_dist_len_price(coder, + dist, len, pos_state); + + if (cur_and_len_price < coder->opts[len].price) { + coder->opts[len].price = cur_and_len_price; + coder->opts[len].pos_prev = 0; + coder->opts[len].back_prev = dist + REPS; + coder->opts[len].prev_1_is_literal = false; + } + + if (len == coder->matches[i].len) + if (++i == matches_count) + break; + } + } + + return len_end; +} + + +static inline uint32_t +helper2(lzma_lzma1_encoder *coder, uint32_t *reps, const uint8_t *buf, + uint32_t len_end, uint32_t position, const uint32_t cur, + const uint32_t nice_len, const uint32_t buf_avail_full) +{ + uint32_t matches_count = coder->matches_count; + uint32_t new_len = coder->longest_match_length; + uint32_t pos_prev = coder->opts[cur].pos_prev; + lzma_lzma_state state; + + if (coder->opts[cur].prev_1_is_literal) { + --pos_prev; + + if (coder->opts[cur].prev_2) { + state = coder->opts[coder->opts[cur].pos_prev_2].state; + + if (coder->opts[cur].back_prev_2 < REPS) + update_long_rep(state); + else + update_match(state); + + } else { + state = coder->opts[pos_prev].state; + } + + update_literal(state); + + } else { + state = coder->opts[pos_prev].state; + } + + if (pos_prev == cur - 1) { + if (is_short_rep(coder->opts[cur])) + update_short_rep(state); + else + update_literal(state); + } else { + uint32_t pos; + if (coder->opts[cur].prev_1_is_literal + && coder->opts[cur].prev_2) { + pos_prev = coder->opts[cur].pos_prev_2; + pos = coder->opts[cur].back_prev_2; + update_long_rep(state); + } else { + pos = coder->opts[cur].back_prev; + if (pos < REPS) + update_long_rep(state); + else + update_match(state); + } + + if (pos < REPS) { + reps[0] = coder->opts[pos_prev].backs[pos]; + + uint32_t i; + for (i = 1; i <= pos; ++i) + reps[i] = coder->opts[pos_prev].backs[i - 1]; + + for (; i < REPS; ++i) + reps[i] = coder->opts[pos_prev].backs[i]; + + } else { + reps[0] = pos - REPS; + + for (uint32_t i = 1; i < REPS; ++i) + reps[i] = coder->opts[pos_prev].backs[i - 1]; + } + } + + coder->opts[cur].state = state; + + for (uint32_t i = 0; i < REPS; ++i) + coder->opts[cur].backs[i] = reps[i]; + + const uint32_t cur_price = coder->opts[cur].price; + + const uint8_t current_byte = *buf; + const uint8_t match_byte = *(buf - reps[0] - 1); + + const uint32_t pos_state = position & coder->pos_mask; + + const uint32_t cur_and_1_price = cur_price + + rc_bit_0_price(coder->is_match[state][pos_state]) + + get_literal_price(coder, position, buf[-1], + !is_literal_state(state), match_byte, current_byte); + + bool next_is_literal = false; + + if (cur_and_1_price < coder->opts[cur + 1].price) { + coder->opts[cur + 1].price = cur_and_1_price; + coder->opts[cur + 1].pos_prev = cur; + make_literal(&coder->opts[cur + 1]); + next_is_literal = true; + } + + const uint32_t match_price = cur_price + + rc_bit_1_price(coder->is_match[state][pos_state]); + const uint32_t rep_match_price = match_price + + rc_bit_1_price(coder->is_rep[state]); + + if (match_byte == current_byte + && !(coder->opts[cur + 1].pos_prev < cur + && coder->opts[cur + 1].back_prev == 0)) { + + const uint32_t short_rep_price = rep_match_price + + get_short_rep_price(coder, state, pos_state); + + if (short_rep_price <= coder->opts[cur + 1].price) { + coder->opts[cur + 1].price = short_rep_price; + coder->opts[cur + 1].pos_prev = cur; + make_short_rep(&coder->opts[cur + 1]); + next_is_literal = true; + } + } + + if (buf_avail_full < 2) + return len_end; + + const uint32_t buf_avail = my_min(buf_avail_full, nice_len); + + if (!next_is_literal && match_byte != current_byte) { // speed optimization + // try literal + rep0 + const uint8_t *const buf_back = buf - reps[0] - 1; + const uint32_t limit = my_min(buf_avail_full, nice_len + 1); + + const uint32_t len_test = lzma_memcmplen(buf, buf_back, 1, limit) - 1; + + if (len_test >= 2) { + lzma_lzma_state state_2 = state; + update_literal(state_2); + + const uint32_t pos_state_next = (position + 1) & coder->pos_mask; + const uint32_t next_rep_match_price = cur_and_1_price + + rc_bit_1_price(coder->is_match[state_2][pos_state_next]) + + rc_bit_1_price(coder->is_rep[state_2]); + + //for (; len_test >= 2; --len_test) { + const uint32_t offset = cur + 1 + len_test; + + while (len_end < offset) + coder->opts[++len_end].price = RC_INFINITY_PRICE; + + const uint32_t cur_and_len_price = next_rep_match_price + + get_rep_price(coder, 0, len_test, + state_2, pos_state_next); + + if (cur_and_len_price < coder->opts[offset].price) { + coder->opts[offset].price = cur_and_len_price; + coder->opts[offset].pos_prev = cur + 1; + coder->opts[offset].back_prev = 0; + coder->opts[offset].prev_1_is_literal = true; + coder->opts[offset].prev_2 = false; + } + //} + } + } + + + uint32_t start_len = 2; // speed optimization + + for (uint32_t rep_index = 0; rep_index < REPS; ++rep_index) { + const uint8_t *const buf_back = buf - reps[rep_index] - 1; + if (not_equal_16(buf, buf_back)) + continue; + + uint32_t len_test = lzma_memcmplen(buf, buf_back, 2, buf_avail); + + while (len_end < cur + len_test) + coder->opts[++len_end].price = RC_INFINITY_PRICE; + + const uint32_t len_test_temp = len_test; + const uint32_t price = rep_match_price + get_pure_rep_price( + coder, rep_index, state, pos_state); + + do { + const uint32_t cur_and_len_price = price + + get_len_price(&coder->rep_len_encoder, + len_test, pos_state); + + if (cur_and_len_price < coder->opts[cur + len_test].price) { + coder->opts[cur + len_test].price = cur_and_len_price; + coder->opts[cur + len_test].pos_prev = cur; + coder->opts[cur + len_test].back_prev = rep_index; + coder->opts[cur + len_test].prev_1_is_literal = false; + } + } while (--len_test >= 2); + + len_test = len_test_temp; + + if (rep_index == 0) + start_len = len_test + 1; + + + uint32_t len_test_2 = len_test + 1; + const uint32_t limit = my_min(buf_avail_full, + len_test_2 + nice_len); + // NOTE: len_test_2 may be greater than limit so the call to + // lzma_memcmplen() must be done conditionally. + if (len_test_2 < limit) + len_test_2 = lzma_memcmplen(buf, buf_back, len_test_2, limit); + + len_test_2 -= len_test + 1; + + if (len_test_2 >= 2) { + lzma_lzma_state state_2 = state; + update_long_rep(state_2); + + uint32_t pos_state_next = (position + len_test) & coder->pos_mask; + + const uint32_t cur_and_len_literal_price = price + + get_len_price(&coder->rep_len_encoder, + len_test, pos_state) + + rc_bit_0_price(coder->is_match[state_2][pos_state_next]) + + get_literal_price(coder, position + len_test, + buf[len_test - 1], true, + buf_back[len_test], buf[len_test]); + + update_literal(state_2); + + pos_state_next = (position + len_test + 1) & coder->pos_mask; + + const uint32_t next_rep_match_price = cur_and_len_literal_price + + rc_bit_1_price(coder->is_match[state_2][pos_state_next]) + + rc_bit_1_price(coder->is_rep[state_2]); + + //for(; len_test_2 >= 2; len_test_2--) { + const uint32_t offset = cur + len_test + 1 + len_test_2; + + while (len_end < offset) + coder->opts[++len_end].price = RC_INFINITY_PRICE; + + const uint32_t cur_and_len_price = next_rep_match_price + + get_rep_price(coder, 0, len_test_2, + state_2, pos_state_next); + + if (cur_and_len_price < coder->opts[offset].price) { + coder->opts[offset].price = cur_and_len_price; + coder->opts[offset].pos_prev = cur + len_test + 1; + coder->opts[offset].back_prev = 0; + coder->opts[offset].prev_1_is_literal = true; + coder->opts[offset].prev_2 = true; + coder->opts[offset].pos_prev_2 = cur; + coder->opts[offset].back_prev_2 = rep_index; + } + //} + } + } + + + //for (uint32_t len_test = 2; len_test <= new_len; ++len_test) + if (new_len > buf_avail) { + new_len = buf_avail; + + matches_count = 0; + while (new_len > coder->matches[matches_count].len) + ++matches_count; + + coder->matches[matches_count++].len = new_len; + } + + + if (new_len >= start_len) { + const uint32_t normal_match_price = match_price + + rc_bit_0_price(coder->is_rep[state]); + + while (len_end < cur + new_len) + coder->opts[++len_end].price = RC_INFINITY_PRICE; + + uint32_t i = 0; + while (start_len > coder->matches[i].len) + ++i; + + for (uint32_t len_test = start_len; ; ++len_test) { + const uint32_t cur_back = coder->matches[i].dist; + uint32_t cur_and_len_price = normal_match_price + + get_dist_len_price(coder, + cur_back, len_test, pos_state); + + if (cur_and_len_price < coder->opts[cur + len_test].price) { + coder->opts[cur + len_test].price = cur_and_len_price; + coder->opts[cur + len_test].pos_prev = cur; + coder->opts[cur + len_test].back_prev + = cur_back + REPS; + coder->opts[cur + len_test].prev_1_is_literal = false; + } + + if (len_test == coder->matches[i].len) { + // Try Match + Literal + Rep0 + const uint8_t *const buf_back = buf - cur_back - 1; + uint32_t len_test_2 = len_test + 1; + const uint32_t limit = my_min(buf_avail_full, + len_test_2 + nice_len); + + // NOTE: len_test_2 may be greater than limit + // so the call to lzma_memcmplen() must be + // done conditionally. + if (len_test_2 < limit) + len_test_2 = lzma_memcmplen(buf, buf_back, + len_test_2, limit); + + len_test_2 -= len_test + 1; + + if (len_test_2 >= 2) { + lzma_lzma_state state_2 = state; + update_match(state_2); + uint32_t pos_state_next + = (position + len_test) & coder->pos_mask; + + const uint32_t cur_and_len_literal_price = cur_and_len_price + + rc_bit_0_price( + coder->is_match[state_2][pos_state_next]) + + get_literal_price(coder, + position + len_test, + buf[len_test - 1], + true, + buf_back[len_test], + buf[len_test]); + + update_literal(state_2); + pos_state_next = (pos_state_next + 1) & coder->pos_mask; + + const uint32_t next_rep_match_price + = cur_and_len_literal_price + + rc_bit_1_price( + coder->is_match[state_2][pos_state_next]) + + rc_bit_1_price(coder->is_rep[state_2]); + + // for(; len_test_2 >= 2; --len_test_2) { + const uint32_t offset = cur + len_test + 1 + len_test_2; + + while (len_end < offset) + coder->opts[++len_end].price = RC_INFINITY_PRICE; + + cur_and_len_price = next_rep_match_price + + get_rep_price(coder, 0, len_test_2, + state_2, pos_state_next); + + if (cur_and_len_price < coder->opts[offset].price) { + coder->opts[offset].price = cur_and_len_price; + coder->opts[offset].pos_prev = cur + len_test + 1; + coder->opts[offset].back_prev = 0; + coder->opts[offset].prev_1_is_literal = true; + coder->opts[offset].prev_2 = true; + coder->opts[offset].pos_prev_2 = cur; + coder->opts[offset].back_prev_2 + = cur_back + REPS; + } + //} + } + + if (++i == matches_count) + break; + } + } + } + + return len_end; +} + + +extern void +lzma_lzma_optimum_normal(lzma_lzma1_encoder *restrict coder, + lzma_mf *restrict mf, + uint32_t *restrict back_res, uint32_t *restrict len_res, + uint32_t position) +{ + // If we have symbols pending, return the next pending symbol. + if (coder->opts_end_index != coder->opts_current_index) { + assert(mf->read_ahead > 0); + *len_res = coder->opts[coder->opts_current_index].pos_prev + - coder->opts_current_index; + *back_res = coder->opts[coder->opts_current_index].back_prev; + coder->opts_current_index = coder->opts[ + coder->opts_current_index].pos_prev; + return; + } + + // Update the price tables. In LZMA SDK <= 4.60 (and possibly later) + // this was done in both initialization function and in the main loop. + // In liblzma they were moved into this single place. + if (mf->read_ahead == 0) { + if (coder->match_price_count >= (1 << 7)) + fill_dist_prices(coder); + + if (coder->align_price_count >= ALIGN_SIZE) + fill_align_prices(coder); + } + + // TODO: This needs quite a bit of cleaning still. But splitting + // the original function into two pieces makes it at least a little + // more readable, since those two parts don't share many variables. + + uint32_t len_end = helper1(coder, mf, back_res, len_res, position); + if (len_end == UINT32_MAX) + return; + + uint32_t reps[REPS]; + memcpy(reps, coder->reps, sizeof(reps)); + + uint32_t cur; + for (cur = 1; cur < len_end; ++cur) { + assert(cur < OPTS); + + coder->longest_match_length = mf_find( + mf, &coder->matches_count, coder->matches); + + if (coder->longest_match_length >= mf->nice_len) + break; + + len_end = helper2(coder, reps, mf_ptr(mf) - 1, len_end, + position + cur, cur, mf->nice_len, + my_min(mf_avail(mf) + 1, OPTS - 1 - cur)); + } + + backward(coder, len_res, back_res, cur); + return; +} diff --git a/src/liblzma/lzma/lzma_encoder_presets.c b/src/liblzma/lzma/lzma_encoder_presets.c new file mode 100644 index 0000000..711df02 --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder_presets.c @@ -0,0 +1,64 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder_presets.c +/// \brief Encoder presets +/// \note xz needs this even when only decoding is enabled. +// +// Author: Lasse Collin +// +// This file has been put into the public domain. +// You can do whatever you want with this file. +// +/////////////////////////////////////////////////////////////////////////////// + +#include "common.h" + + +extern LZMA_API(lzma_bool) +lzma_lzma_preset(lzma_options_lzma *options, uint32_t preset) +{ + const uint32_t level = preset & LZMA_PRESET_LEVEL_MASK; + const uint32_t flags = preset & ~LZMA_PRESET_LEVEL_MASK; + const uint32_t supported_flags = LZMA_PRESET_EXTREME; + + if (level > 9 || (flags & ~supported_flags)) + return true; + + options->preset_dict = NULL; + options->preset_dict_size = 0; + + options->lc = LZMA_LC_DEFAULT; + options->lp = LZMA_LP_DEFAULT; + options->pb = LZMA_PB_DEFAULT; + + static const uint8_t dict_pow2[] + = { 18, 20, 21, 22, 22, 23, 23, 24, 25, 26 }; + options->dict_size = UINT32_C(1) << dict_pow2[level]; + + if (level <= 3) { + options->mode = LZMA_MODE_FAST; + options->mf = level == 0 ? LZMA_MF_HC3 : LZMA_MF_HC4; + options->nice_len = level <= 1 ? 128 : 273; + static const uint8_t depths[] = { 4, 8, 24, 48 }; + options->depth = depths[level]; + } else { + options->mode = LZMA_MODE_NORMAL; + options->mf = LZMA_MF_BT4; + options->nice_len = level == 4 ? 16 : level == 5 ? 32 : 64; + options->depth = 0; + } + + if (flags & LZMA_PRESET_EXTREME) { + options->mode = LZMA_MODE_NORMAL; + options->mf = LZMA_MF_BT4; + if (level == 3 || level == 5) { + options->nice_len = 192; + options->depth = 0; + } else { + options->nice_len = 273; + options->depth = 512; + } + } + + return false; +} diff --git a/src/liblzma/lzma/lzma_encoder_private.h b/src/liblzma/lzma/lzma_encoder_private.h new file mode 100644 index 0000000..b228c57 --- /dev/null +++ b/src/liblzma/lzma/lzma_encoder_private.h @@ -0,0 +1,162 @@ +/////////////////////////////////////////////////////////////////////////////// +// +/// \file lzma_encoder_private.h +/// \brief Private definitions for LZMA encoder +/// +// 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_LZMA_ENCODER_PRIVATE_H +#define LZMA_LZMA_ENCODER_PRIVATE_H + +#include "lz_encoder.h" +#include "range_encoder.h" +#include "lzma_common.h" +#include "lzma_encoder.h" + + +// Macro to compare if the first two bytes in two buffers differ. This is +// needed in lzma_lzma_optimum_*() to test if the match is at least +// MATCH_LEN_MIN bytes. Unaligned access gives tiny gain so there's no +// reason to not use it when it is supported. +#ifdef TUKLIB_FAST_UNALIGNED_ACCESS +# define not_equal_16(a, b) (read16ne(a) != read16ne(b)) +#else +# define not_equal_16(a, b) \ + ((a)[0] != (b)[0] || (a)[1] != (b)[1]) +#endif + + +// Optimal - Number of entries in the optimum array. +#define OPTS (1 << 12) + + +typedef struct { + probability choice; + probability choice2; + probability low[POS_STATES_MAX][LEN_LOW_SYMBOLS]; + probability mid[POS_STATES_MAX][LEN_MID_SYMBOLS]; + probability high[LEN_HIGH_SYMBOLS]; + + uint32_t prices[POS_STATES_MAX][LEN_SYMBOLS]; + uint32_t table_size; + uint32_t counters[POS_STATES_MAX]; + +} lzma_length_encoder; + + +typedef struct { + lzma_lzma_state state; + + bool prev_1_is_literal; + bool prev_2; + + uint32_t pos_prev_2; + uint32_t back_prev_2; + + uint32_t price; + uint32_t pos_prev; // pos_next; + uint32_t back_prev; + + uint32_t backs[REPS]; + +} lzma_optimal; + + +struct lzma_lzma1_encoder_s { + /// Range encoder + lzma_range_encoder rc; + + /// Uncompressed size (doesn't include possible preset dictionary) + uint64_t uncomp_size; + + /// If non-zero, produce at most this much output. + /// Some input may then be missing from the output. + uint64_t out_limit; + + /// If the above out_limit is non-zero, *uncomp_size_ptr is set to + /// the amount of uncompressed data that we were able to fit + /// in the output buffer. + uint64_t *uncomp_size_ptr; + + /// State + lzma_lzma_state state; + + /// The four most recent match distances + uint32_t reps[REPS]; + + /// Array of match candidates + lzma_match matches[MATCH_LEN_MAX + 1]; + + /// Number of match candidates in matches[] + uint32_t matches_count; + + /// Variable to hold the length of the longest match between calls + /// to lzma_lzma_optimum_*(). + uint32_t longest_match_length; + + /// True if using getoptimumfast + bool fast_mode; + + /// True if the encoder has been initialized by encoding the first + /// byte as a literal. + bool is_initialized; + + /// True if the range encoder has been flushed, but not all bytes + /// have been written to the output buffer yet. + bool is_flushed; + + /// True if end of payload marker will be written. + bool use_eopm; + + uint32_t pos_mask; ///< (1 << pos_bits) - 1 + uint32_t literal_context_bits; + uint32_t literal_pos_mask; + + // These are the same as in lzma_decoder.c. See comments there. + probability literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE]; + probability is_match[STATES][POS_STATES_MAX]; + probability is_rep[STATES]; + probability is_rep0[STATES]; + probability is_rep1[STATES]; + probability is_rep2[STATES]; + probability is_rep0_long[STATES][POS_STATES_MAX]; + probability dist_slot[DIST_STATES][DIST_SLOTS]; + probability dist_special[FULL_DISTANCES - DIST_MODEL_END]; + probability dist_align[ALIGN_SIZE]; + + // These are the same as in lzma_decoder.c except that the encoders + // include also price tables. + lzma_length_encoder match_len_encoder; + lzma_length_encoder rep_len_encoder; + + // Price tables + uint32_t dist_slot_prices[DIST_STATES][DIST_SLOTS]; + uint32_t dist_prices[DIST_STATES][FULL_DISTANCES]; + uint32_t dist_table_size; + uint32_t match_price_count; + + uint32_t align_prices[ALIGN_SIZE]; + uint32_t align_price_count; + + // Optimal + uint32_t opts_end_index; + uint32_t opts_current_index; + lzma_optimal opts[OPTS]; +}; + + +extern void lzma_lzma_optimum_fast( + lzma_lzma1_encoder *restrict coder, lzma_mf *restrict mf, + uint32_t *restrict back_res, uint32_t *restrict len_res); + +extern void lzma_lzma_optimum_normal(lzma_lzma1_encoder *restrict coder, + lzma_mf *restrict mf, uint32_t *restrict back_res, + uint32_t *restrict len_res, uint32_t position); + +#endif |