/*
* lzms-compress.c
+ *
+ * A compressor that produces output compatible with the LZMS compression format.
*/
/*
* along with wimlib; if not, see http://www.gnu.org/licenses/.
*/
-/* This a compressor for the LZMS compression format. More details about this
- * format can be found in lzms-decompress.c.
- *
- * Also see lzx-compress.c for general information about match-finding and
- * match-choosing that also applies to this LZMS compressor.
- *
- * NOTE: this compressor currently does not code any delta matches.
- */
-
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
-#include "wimlib/assert.h"
-#include "wimlib/compiler.h"
-#include "wimlib/compressor_ops.h"
#include "wimlib/compress_common.h"
+#include "wimlib/compressor_ops.h"
#include "wimlib/endianness.h"
#include "wimlib/error.h"
#include "wimlib/lz_mf.h"
#include <limits.h>
#include <pthread.h>
-/* Stucture used for writing raw bits to the end of the LZMS-compressed data as
- * a series of 16-bit little endian coding units. */
+/* Stucture used for writing raw bits as a series of 16-bit little endian coding
+ * units. This starts at the *end* of the compressed data buffer and proceeds
+ * backwards. */
struct lzms_output_bitstream {
- /* Buffer variable containing zero or more bits that have been logically
- * written to the bitstream but not yet written to memory. This must be
- * at least as large as the coding unit size. */
- u16 bitbuf;
- /* Number of bits in @bitbuf that are valid. */
- unsigned num_free_bits;
+ /* Bits that haven't yet been written to the output buffer. */
+ u64 bitbuf;
+
+ /* Number of bits currently held in @bitbuf. */
+ unsigned bitcount;
/* Pointer to one past the next position in the compressed data buffer
* at which to output a 16-bit coding unit. */
- le16 *out;
+ le16 *next;
- /* Maximum number of 16-bit coding units that can still be output to
- * the compressed data buffer. */
- size_t num_le16_remaining;
-
- /* Set to %true if not all coding units could be output due to
- * insufficient space. */
- bool overrun;
+ /* Pointer to the beginning of the output buffer. (The "end" when
+ * writing backwards!) */
+ le16 *begin;
};
-/* Stucture used for range encoding (raw version). */
+/* Stucture used for range encoding (raw version). This starts at the
+ * *beginning* of the compressed data buffer and proceeds forward. */
struct lzms_range_encoder_raw {
/* A 33-bit variable that holds the low boundary of the current range.
* subsequent such coding units are 0xffff. */
u32 cache_size;
- /* Pointer to the next position in the compressed data buffer at which
- * to output a 16-bit coding unit. */
- le16 *out;
-
- /* Maximum number of 16-bit coding units that can still be output to
- * the compressed data buffer. */
- size_t num_le16_remaining;
+ /* Pointer to the beginning of the output buffer. */
+ le16 *begin;
- /* %true when the very first coding unit has not yet been output. */
- bool first;
+ /* Pointer to the position in the output buffer at which the next coding
+ * unit must be written. */
+ le16 *next;
- /* Set to %true if not all coding units could be output due to
- * insufficient space. */
- bool overrun;
+ /* Pointer just past the end of the output buffer. */
+ le16 *end;
};
/* Structure used for range encoding. This wraps around `struct
* lzms_range_encoder_raw' to use and maintain probability entries. */
struct lzms_range_encoder {
+
/* Pointer to the raw range encoder, which has no persistent knowledge
* of probabilities. Multiple lzms_range_encoder's share the same
* lzms_range_encoder_raw. */
u32 codewords[LZMS_MAX_NUM_SYMS];
};
+/* Internal compression parameters */
struct lzms_compressor_params {
u32 min_match_length;
u32 nice_match_length;
u32 optim_array_length;
};
-/* State of the LZMS compressor. */
+/* State of the LZMS compressor */
struct lzms_compressor {
- /* Pointer to a buffer holding the preprocessed data to compress. */
- u8 *window;
- /* Current position in @buffer. */
- u32 cur_window_pos;
+ /* Internal compression parameters */
+ struct lzms_compressor_params params;
- /* Size of the data in @buffer. */
- u32 window_size;
+ /* Data currently being compressed */
+ u8 *cur_window;
+ u32 cur_window_size;
- /* Lempel-Ziv match-finder. */
+ /* Lempel-Ziv match-finder */
struct lz_mf *mf;
- /* Temporary space to store found matches. */
+ /* Temporary space to store found matches */
struct lz_match *matches;
- /* Match-chooser data. */
+ /* Per-position data for near-optimal parsing */
struct lzms_mc_pos_data *optimum;
- unsigned optimum_cur_idx;
- unsigned optimum_end_idx;
-
- /* Maximum block size this compressor instantiation allows. This is the
- * allocated size of @window. */
- u32 max_block_size;
-
- /* Compression parameters. */
- struct lzms_compressor_params params;
+ struct lzms_mc_pos_data *optimum_end;
/* Raw range encoder which outputs to the beginning of the compressed
- * data buffer, proceeding forwards. */
+ * data buffer, proceeding forwards */
struct lzms_range_encoder_raw rc;
/* Bitstream which outputs to the end of the compressed data buffer,
- * proceeding backwards. */
+ * proceeding backwards */
struct lzms_output_bitstream os;
- /* Range encoders. */
+ /* Range encoders */
struct lzms_range_encoder main_range_encoder;
struct lzms_range_encoder match_range_encoder;
struct lzms_range_encoder lz_match_range_encoder;
struct lzms_range_encoder delta_match_range_encoder;
struct lzms_range_encoder delta_repeat_match_range_encoders[LZMS_NUM_RECENT_OFFSETS - 1];
- /* Huffman encoders. */
+ /* Huffman encoders */
struct lzms_huffman_encoder literal_encoder;
struct lzms_huffman_encoder lz_offset_encoder;
struct lzms_huffman_encoder length_encoder;
struct lzms_huffman_encoder delta_power_encoder;
struct lzms_huffman_encoder delta_offset_encoder;
- /* LRU (least-recently-used) queues for match information. */
- struct lzms_lru_queues lru;
-
- /* Used for preprocessing. */
+ /* Used for preprocessing */
s32 last_target_usages[65536];
+
+#define LZMS_NUM_FAST_LENGTHS 256
+ /* Table: length => length slot for small lengths */
+ u8 length_slot_fast[LZMS_NUM_FAST_LENGTHS];
+
+ /* Table: length => current cost for small match lengths */
+ u32 length_cost_fast[LZMS_NUM_FAST_LENGTHS];
+
+#define LZMS_NUM_FAST_OFFSETS 32768
+ /* Table: offset => offset slot for small offsets */
+ u8 offset_slot_fast[LZMS_NUM_FAST_OFFSETS];
};
+/*
+ * Match chooser position data:
+ *
+ * An array of these structures is used during the near-optimal match-choosing
+ * algorithm. They correspond to consecutive positions in the window and are
+ * used to keep track of the cost to reach each position, and the match/literal
+ * choices that need to be chosen to reach that position.
+ */
struct lzms_mc_pos_data {
+
+ /* The cost, in bits, of the lowest-cost path that has been found to
+ * reach this position. This can change as progressively lower cost
+ * paths are found to reach this position. */
u32 cost;
-#define MC_INFINITE_COST ((u32)~0UL)
- union {
- struct {
- u32 link;
- u32 match_offset;
- } prev;
- struct {
- u32 link;
- u32 match_offset;
- } next;
- };
+#define MC_INFINITE_COST UINT32_MAX
+
+ /* The match or literal that was taken to reach this position. This can
+ * change as progressively lower cost paths are found to reach this
+ * position.
+ *
+ * This variable is divided into two bitfields.
+ *
+ * Literals:
+ * Low bits are 1, high bits are the literal.
+ *
+ * Explicit offset matches:
+ * Low bits are the match length, high bits are the offset plus 2.
+ *
+ * Repeat offset matches:
+ * Low bits are the match length, high bits are the queue index.
+ */
+ u64 mc_item_data;
+#define MC_OFFSET_SHIFT 32
+#define MC_LEN_MASK (((u64)1 << MC_OFFSET_SHIFT) - 1)
+
+ /* The LZMS adaptive state that exists at this position. This is filled
+ * in lazily, only after the minimum-cost path to this position is
+ * found.
+ *
+ * Note: the way we handle this adaptive state in the "minimum-cost"
+ * parse is actually only an approximation. It's possible for the
+ * globally optimal, minimum cost path to contain a prefix, ending at a
+ * position, where that path prefix is *not* the minimum cost path to
+ * that position. This can happen if such a path prefix results in a
+ * different adaptive state which results in lower costs later. We do
+ * not solve this problem; we only consider the lowest cost to reach
+ * each position, which seems to be an acceptable approximation.
+ *
+ * Note: this adaptive state also does not include the probability
+ * entries or current Huffman codewords. Those aren't maintained
+ * per-position and are only updated occassionally. */
struct lzms_adaptive_state {
struct lzms_lz_lru_queues lru;
u8 main_state;
} state;
};
-/* Initialize the output bitstream @os to write forwards to the specified
+static void
+lzms_init_fast_slots(struct lzms_compressor *c)
+{
+ /* Create table mapping small lengths to length slots. */
+ for (unsigned slot = 0, i = 0; i < LZMS_NUM_FAST_LENGTHS; i++) {
+ while (i >= lzms_length_slot_base[slot + 1])
+ slot++;
+ c->length_slot_fast[i] = slot;
+ }
+
+ /* Create table mapping small offsets to offset slots. */
+ for (unsigned slot = 0, i = 0; i < LZMS_NUM_FAST_OFFSETS; i++) {
+ while (i >= lzms_offset_slot_base[slot + 1])
+ slot++;
+ c->offset_slot_fast[i] = slot;
+ }
+}
+
+static inline unsigned
+lzms_get_length_slot_fast(const struct lzms_compressor *c, u32 length)
+{
+ if (likely(length < LZMS_NUM_FAST_LENGTHS))
+ return c->length_slot_fast[length];
+ else
+ return lzms_get_length_slot(length);
+}
+
+static inline unsigned
+lzms_get_offset_slot_fast(const struct lzms_compressor *c, u32 offset)
+{
+ if (offset < LZMS_NUM_FAST_OFFSETS)
+ return c->offset_slot_fast[offset];
+ else
+ return lzms_get_offset_slot(offset);
+}
+
+/* Initialize the output bitstream @os to write backwards to the specified
* compressed data buffer @out that is @out_limit 16-bit integers long. */
static void
lzms_output_bitstream_init(struct lzms_output_bitstream *os,
le16 *out, size_t out_limit)
{
os->bitbuf = 0;
- os->num_free_bits = 16;
- os->out = out + out_limit;
- os->num_le16_remaining = out_limit;
- os->overrun = false;
+ os->bitcount = 0;
+ os->next = out + out_limit;
+ os->begin = out;
}
-/* Write @num_bits bits, contained in the low @num_bits bits of @bits (ordered
- * from high-order to low-order), to the output bitstream @os. */
-static void
-lzms_output_bitstream_put_bits(struct lzms_output_bitstream *os,
- u32 bits, unsigned num_bits)
+/*
+ * Write some bits, contained in the low @num_bits bits of @bits (ordered from
+ * high-order to low-order), to the output bitstream @os.
+ *
+ * @max_num_bits is a compile-time constant that specifies the maximum number of
+ * bits that can ever be written at this call site.
+ */
+static inline void
+lzms_output_bitstream_put_varbits(struct lzms_output_bitstream *os,
+ u32 bits, unsigned num_bits,
+ unsigned max_num_bits)
{
- bits &= (1U << num_bits) - 1;
+ LZMS_ASSERT(num_bits <= 48);
- while (num_bits > os->num_free_bits) {
+ /* Add the bits to the bit buffer variable. */
+ os->bitcount += num_bits;
+ os->bitbuf = (os->bitbuf << num_bits) | bits;
- if (unlikely(os->num_le16_remaining == 0)) {
- os->overrun = true;
- return;
- }
+ /* Check whether any coding units need to be written. */
+ while (os->bitcount >= 16) {
- unsigned num_fill_bits = os->num_free_bits;
+ os->bitcount -= 16;
- os->bitbuf <<= num_fill_bits;
- os->bitbuf |= bits >> (num_bits - num_fill_bits);
+ /* Write a coding unit, unless it would underflow the buffer. */
+ if (os->next != os->begin)
+ *--os->next = cpu_to_le16(os->bitbuf >> os->bitcount);
- *--os->out = cpu_to_le16(os->bitbuf);
- --os->num_le16_remaining;
-
- os->num_free_bits = 16;
- num_bits -= num_fill_bits;
- bits &= (1U << num_bits) - 1;
+ /* Optimization for call sites that never write more than 16
+ * bits at once. */
+ if (max_num_bits <= 16)
+ break;
}
- os->bitbuf <<= num_bits;
- os->bitbuf |= bits;
- os->num_free_bits -= num_bits;
+}
+
+/* Use when @num_bits is a compile-time constant. Otherwise use
+ * lzms_output_bitstream_put_bits(). */
+static inline void
+lzms_output_bitstream_put_bits(struct lzms_output_bitstream *os,
+ u32 bits, unsigned num_bits)
+{
+ lzms_output_bitstream_put_varbits(os, bits, num_bits, num_bits);
}
/* Flush the output bitstream, ensuring that all bits written to it have been
- * written to memory. Returns %true if all bits were output successfully, or
- * %false if an overrun occurred. */
+ * written to memory. Returns %true if all bits have been output successfully,
+ * or %false if an overrun occurred. */
static bool
lzms_output_bitstream_flush(struct lzms_output_bitstream *os)
{
- if (os->num_free_bits != 16)
- lzms_output_bitstream_put_bits(os, 0, os->num_free_bits + 1);
- return !os->overrun;
+ if (os->next == os->begin)
+ return false;
+
+ if (os->bitcount != 0)
+ *--os->next = cpu_to_le16(os->bitbuf << (16 - os->bitcount));
+
+ return true;
}
/* Initialize the range encoder @rc to write forwards to the specified
rc->range = 0xffffffff;
rc->cache = 0;
rc->cache_size = 1;
- rc->out = out;
- rc->num_le16_remaining = out_limit;
- rc->first = true;
- rc->overrun = false;
+ rc->begin = out;
+ rc->next = out - 1;
+ rc->end = out + out_limit;
}
/*
static void
lzms_range_encoder_raw_shift_low(struct lzms_range_encoder_raw *rc)
{
- LZMS_DEBUG("low=%"PRIx64", cache=%"PRIx64", cache_size=%u",
- rc->low, rc->cache, rc->cache_size);
if ((u32)(rc->low) < 0xffff0000 ||
(u32)(rc->low >> 32) != 0)
{
/* Carry not needed (rc->low < 0xffff0000), or carry occurred
* ((rc->low >> 32) != 0, a.k.a. the carry bit is 1). */
do {
- if (!rc->first) {
- if (rc->num_le16_remaining == 0) {
- rc->overrun = true;
- return;
- }
- *rc->out++ = cpu_to_le16(rc->cache +
- (u16)(rc->low >> 32));
- --rc->num_le16_remaining;
+ if (likely(rc->next >= rc->begin)) {
+ if (rc->next != rc->end)
+ *rc->next++ = cpu_to_le16(rc->cache +
+ (u16)(rc->low >> 32));
} else {
- rc->first = false;
+ rc->next++;
}
-
rc->cache = 0xffff;
} while (--rc->cache_size != 0);
{
for (unsigned i = 0; i < 4; i++)
lzms_range_encoder_raw_shift_low(rc);
- return !rc->overrun;
+ return rc->next != rc->end;
}
/* Encode the next bit using the range encoder (raw version).
*
* @prob is the chance out of LZMS_PROBABILITY_MAX that the next bit is 0. */
-static void
-lzms_range_encoder_raw_encode_bit(struct lzms_range_encoder_raw *rc, int bit,
- u32 prob)
+static inline void
+lzms_range_encoder_raw_encode_bit(struct lzms_range_encoder_raw *rc,
+ int bit, u32 prob)
{
lzms_range_encoder_raw_normalize(rc);
/* Encode a bit using the specified range encoder. This wraps around
* lzms_range_encoder_raw_encode_bit() to handle using and updating the
- * appropriate probability table. */
+ * appropriate state and probability entry. */
static void
lzms_range_encode_bit(struct lzms_range_encoder *enc, int bit)
{
/* Load the probability entry corresponding to the current state. */
prob_entry = &enc->prob_entries[enc->state];
- /* Treat the number of zero bits in the most recently encoded
- * LZMS_PROBABILITY_MAX bits with this probability entry as the chance,
- * out of LZMS_PROBABILITY_MAX, that the next bit will be a 0. However,
- * don't allow 0% or 100% probabilities. */
- prob = prob_entry->num_recent_zero_bits;
- if (prob == 0)
- prob = 1;
- else if (prob == LZMS_PROBABILITY_MAX)
- prob = LZMS_PROBABILITY_MAX - 1;
-
- /* Encode the next bit. */
+ /* Update the state based on the next bit. */
+ enc->state = ((enc->state << 1) | bit) & enc->mask;
+
+ /* Get the probability that the bit is 0. */
+ prob = lzms_get_probability(prob_entry);
+
+ /* Update the probability entry. */
+ lzms_update_probability_entry(prob_entry, bit);
+
+ /* Encode the bit. */
lzms_range_encoder_raw_encode_bit(enc->rc, bit, prob);
+}
- /* Update the state based on the newly encoded bit. */
- enc->state = ((enc->state << 1) | bit) & enc->mask;
+/* Called when an adaptive Huffman code needs to be rebuilt. */
+static void
+lzms_rebuild_huffman_code(struct lzms_huffman_encoder *enc)
+{
+ make_canonical_huffman_code(enc->num_syms,
+ LZMS_MAX_CODEWORD_LEN,
+ enc->sym_freqs,
+ enc->lens,
+ enc->codewords);
- /* Update the recent bits, including the cached count of 0's. */
- BUILD_BUG_ON(LZMS_PROBABILITY_MAX > sizeof(prob_entry->recent_bits) * 8);
- if (bit == 0) {
- if (prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1))) {
- /* Replacing 1 bit with 0 bit; increment the zero count.
- */
- prob_entry->num_recent_zero_bits++;
- }
- } else {
- if (!(prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1)))) {
- /* Replacing 0 bit with 1 bit; decrement the zero count.
- */
- prob_entry->num_recent_zero_bits--;
- }
+ /* Dilute the frequencies. */
+ for (unsigned i = 0; i < enc->num_syms; i++) {
+ enc->sym_freqs[i] >>= 1;
+ enc->sym_freqs[i] += 1;
}
- prob_entry->recent_bits = (prob_entry->recent_bits << 1) | bit;
+ enc->num_syms_written = 0;
}
/* Encode a symbol using the specified Huffman encoder. */
-static void
-lzms_huffman_encode_symbol(struct lzms_huffman_encoder *enc, u32 sym)
+static inline void
+lzms_huffman_encode_symbol(struct lzms_huffman_encoder *enc, unsigned sym)
{
- LZMS_ASSERT(sym < enc->num_syms);
- lzms_output_bitstream_put_bits(enc->os,
- enc->codewords[sym],
- enc->lens[sym]);
+ lzms_output_bitstream_put_varbits(enc->os,
+ enc->codewords[sym],
+ enc->lens[sym],
+ LZMS_MAX_CODEWORD_LEN);
++enc->sym_freqs[sym];
- if (++enc->num_syms_written == enc->rebuild_freq) {
- /* Adaptive code needs to be rebuilt. */
- LZMS_DEBUG("Rebuilding code (num_syms=%u)", enc->num_syms);
- make_canonical_huffman_code(enc->num_syms,
- LZMS_MAX_CODEWORD_LEN,
- enc->sym_freqs,
- enc->lens,
- enc->codewords);
-
- /* Dilute the frequencies. */
- for (unsigned i = 0; i < enc->num_syms; i++) {
- enc->sym_freqs[i] >>= 1;
- enc->sym_freqs[i] += 1;
- }
- enc->num_syms_written = 0;
- }
+ if (++enc->num_syms_written == enc->rebuild_freq)
+ lzms_rebuild_huffman_code(enc);
}
static void
-lzms_encode_length(struct lzms_huffman_encoder *enc, u32 length)
+lzms_update_fast_length_costs(struct lzms_compressor *c);
+
+/* Encode a match length. */
+static void
+lzms_encode_length(struct lzms_compressor *c, u32 length)
{
unsigned slot;
unsigned num_extra_bits;
u32 extra_bits;
- slot = lzms_get_length_slot(length);
+ slot = lzms_get_length_slot_fast(c, length);
+ extra_bits = length - lzms_length_slot_base[slot];
num_extra_bits = lzms_extra_length_bits[slot];
- extra_bits = length - lzms_length_slot_base[slot];
+ lzms_huffman_encode_symbol(&c->length_encoder, slot);
+ if (c->length_encoder.num_syms_written == 0)
+ lzms_update_fast_length_costs(c);
- lzms_huffman_encode_symbol(enc, slot);
- lzms_output_bitstream_put_bits(enc->os, extra_bits, num_extra_bits);
+ lzms_output_bitstream_put_varbits(c->length_encoder.os,
+ extra_bits, num_extra_bits, 30);
}
+/* Encode an LZ match offset. */
static void
-lzms_encode_offset(struct lzms_huffman_encoder *enc, u32 offset)
+lzms_encode_lz_offset(struct lzms_compressor *c, u32 offset)
{
unsigned slot;
unsigned num_extra_bits;
u32 extra_bits;
- slot = lzms_get_position_slot(offset);
-
- num_extra_bits = lzms_extra_position_bits[slot];
+ slot = lzms_get_offset_slot_fast(c, offset);
- extra_bits = offset - lzms_position_slot_base[slot];
+ extra_bits = offset - lzms_offset_slot_base[slot];
+ num_extra_bits = lzms_extra_offset_bits[slot];
- lzms_huffman_encode_symbol(enc, slot);
- lzms_output_bitstream_put_bits(enc->os, extra_bits, num_extra_bits);
-}
-
-static void
-lzms_begin_encode_item(struct lzms_compressor *ctx)
-{
- ctx->lru.lz.upcoming_offset = 0;
- ctx->lru.delta.upcoming_offset = 0;
- ctx->lru.delta.upcoming_power = 0;
-}
-
-static void
-lzms_end_encode_item(struct lzms_compressor *ctx, u32 length)
-{
- LZMS_ASSERT(ctx->window_size - ctx->cur_window_pos >= length);
- ctx->cur_window_pos += length;
- lzms_update_lru_queues(&ctx->lru);
+ lzms_huffman_encode_symbol(&c->lz_offset_encoder, slot);
+ lzms_output_bitstream_put_varbits(c->lz_offset_encoder.os,
+ extra_bits, num_extra_bits, 30);
}
/* Encode a literal byte. */
static void
-lzms_encode_literal(struct lzms_compressor *ctx, u8 literal)
+lzms_encode_literal(struct lzms_compressor *c, unsigned literal)
{
- LZMS_DEBUG("Position %u: Encoding literal 0x%02x ('%c')",
- ctx->cur_window_pos, literal, literal);
-
- lzms_begin_encode_item(ctx);
-
/* Main bit: 0 = a literal, not a match. */
- lzms_range_encode_bit(&ctx->main_range_encoder, 0);
+ lzms_range_encode_bit(&c->main_range_encoder, 0);
/* Encode the literal using the current literal Huffman code. */
- lzms_huffman_encode_symbol(&ctx->literal_encoder, literal);
-
- lzms_end_encode_item(ctx, 1);
+ lzms_huffman_encode_symbol(&c->literal_encoder, literal);
}
-/* Encode a (length, offset) pair (LZ match). */
+/* Encode an LZ repeat offset match. */
static void
-lzms_encode_lz_match(struct lzms_compressor *ctx, u32 length, u32 offset)
+lzms_encode_lz_repeat_offset_match(struct lzms_compressor *c,
+ u32 length, unsigned rep_index)
{
- int recent_offset_idx;
-
- LZMS_DEBUG("Position %u: Encoding LZ match {length=%u, offset=%u}",
- ctx->cur_window_pos, length, offset);
-
- LZMS_ASSERT(length <= ctx->window_size - ctx->cur_window_pos);
- LZMS_ASSERT(offset <= ctx->cur_window_pos);
- LZMS_ASSERT(!memcmp(&ctx->window[ctx->cur_window_pos],
- &ctx->window[ctx->cur_window_pos - offset],
- length));
-
- lzms_begin_encode_item(ctx);
+ unsigned i;
/* Main bit: 1 = a match, not a literal. */
- lzms_range_encode_bit(&ctx->main_range_encoder, 1);
+ lzms_range_encode_bit(&c->main_range_encoder, 1);
/* Match bit: 0 = an LZ match, not a delta match. */
- lzms_range_encode_bit(&ctx->match_range_encoder, 0);
+ lzms_range_encode_bit(&c->match_range_encoder, 0);
- /* Determine if the offset can be represented as a recent offset. */
- for (recent_offset_idx = 0;
- recent_offset_idx < LZMS_NUM_RECENT_OFFSETS;
- recent_offset_idx++)
- if (offset == ctx->lru.lz.recent_offsets[recent_offset_idx])
- break;
+ /* LZ match bit: 1 = repeat offset, not an explicit offset. */
+ lzms_range_encode_bit(&c->lz_match_range_encoder, 1);
- if (recent_offset_idx == LZMS_NUM_RECENT_OFFSETS) {
- /* Explicit offset. */
+ /* Encode the repeat offset index. A 1 bit is encoded for each index
+ * passed up. This sequence of 1 bits is terminated by a 0 bit, or
+ * automatically when (LZMS_NUM_RECENT_OFFSETS - 1) 1 bits have been
+ * encoded. */
+ for (i = 0; i < rep_index; i++)
+ lzms_range_encode_bit(&c->lz_repeat_match_range_encoders[i], 1);
- /* LZ match bit: 0 = explicit offset, not a recent offset. */
- lzms_range_encode_bit(&ctx->lz_match_range_encoder, 0);
+ if (i < LZMS_NUM_RECENT_OFFSETS - 1)
+ lzms_range_encode_bit(&c->lz_repeat_match_range_encoders[i], 0);
- /* Encode the match offset. */
- lzms_encode_offset(&ctx->lz_offset_encoder, offset);
- } else {
- int i;
-
- /* Recent offset. */
+ /* Encode the match length. */
+ lzms_encode_length(c, length);
+}
- /* LZ match bit: 1 = recent offset, not an explicit offset. */
- lzms_range_encode_bit(&ctx->lz_match_range_encoder, 1);
+/* Encode an LZ explicit offset match. */
+static void
+lzms_encode_lz_explicit_offset_match(struct lzms_compressor *c,
+ u32 length, u32 offset)
+{
+ /* Main bit: 1 = a match, not a literal. */
+ lzms_range_encode_bit(&c->main_range_encoder, 1);
- /* Encode the recent offset index. A 1 bit is encoded for each
- * index passed up. This sequence of 1 bits is terminated by a
- * 0 bit, or automatically when (LZMS_NUM_RECENT_OFFSETS - 1) 1
- * bits have been encoded. */
- for (i = 0; i < recent_offset_idx; i++)
- lzms_range_encode_bit(&ctx->lz_repeat_match_range_encoders[i], 1);
+ /* Match bit: 0 = an LZ match, not a delta match. */
+ lzms_range_encode_bit(&c->match_range_encoder, 0);
- if (i < LZMS_NUM_RECENT_OFFSETS - 1)
- lzms_range_encode_bit(&ctx->lz_repeat_match_range_encoders[i], 0);
+ /* LZ match bit: 0 = explicit offset, not a repeat offset. */
+ lzms_range_encode_bit(&c->lz_match_range_encoder, 0);
- /* Initial update of the LZ match offset LRU queue. */
- for (; i < LZMS_NUM_RECENT_OFFSETS; i++)
- ctx->lru.lz.recent_offsets[i] = ctx->lru.lz.recent_offsets[i + 1];
- }
+ /* Encode the match offset. */
+ lzms_encode_lz_offset(c, offset);
/* Encode the match length. */
- lzms_encode_length(&ctx->length_encoder, length);
+ lzms_encode_length(c, length);
+}
- /* Save the match offset for later insertion at the front of the LZ
- * match offset LRU queue. */
- ctx->lru.lz.upcoming_offset = offset;
+static void
+lzms_encode_item(struct lzms_compressor *c, u64 mc_item_data)
+{
+ u32 len = mc_item_data & MC_LEN_MASK;
+ u32 offset_data = mc_item_data >> MC_OFFSET_SHIFT;
- lzms_end_encode_item(ctx, length);
+ if (len == 1)
+ lzms_encode_literal(c, offset_data);
+ else if (offset_data < LZMS_NUM_RECENT_OFFSETS)
+ lzms_encode_lz_repeat_offset_match(c, len, offset_data);
+ else
+ lzms_encode_lz_explicit_offset_match(c, len, offset_data - LZMS_OFFSET_OFFSET);
}
-#define LZMS_COST_SHIFT 5
+/* Encode a list of matches and literals chosen by the parsing algorithm. */
+static void
+lzms_encode_item_list(struct lzms_compressor *c,
+ struct lzms_mc_pos_data *cur_optimum_ptr)
+{
+ struct lzms_mc_pos_data *end_optimum_ptr;
+ u64 saved_item;
+ u64 item;
+
+ /* The list is currently in reverse order (last item to first item).
+ * Reverse it. */
+ end_optimum_ptr = cur_optimum_ptr;
+ saved_item = cur_optimum_ptr->mc_item_data;
+ do {
+ item = saved_item;
+ cur_optimum_ptr -= item & MC_LEN_MASK;
+ saved_item = cur_optimum_ptr->mc_item_data;
+ cur_optimum_ptr->mc_item_data = item;
+ } while (cur_optimum_ptr != c->optimum);
+
+ /* Walk the list of items from beginning to end, encoding each item. */
+ do {
+ lzms_encode_item(c, cur_optimum_ptr->mc_item_data);
+ cur_optimum_ptr += (cur_optimum_ptr->mc_item_data) & MC_LEN_MASK;
+ } while (cur_optimum_ptr != end_optimum_ptr);
+}
+
+/* Each bit costs 1 << LZMS_COST_SHIFT units. */
+#define LZMS_COST_SHIFT 6
/*#define LZMS_RC_COSTS_USE_FLOATING_POINT*/
pthread_once(&once, lzms_do_init_rc_costs);
}
-/*
- * Return the cost to range-encode the specified bit when in the specified
- * state.
- *
- * @enc The range encoder to use.
- * @cur_state Current state, which indicates the probability entry to choose.
- * Updated by this function.
- * @bit The bit to encode (0 or 1).
- */
-static u32
-lzms_rc_bit_cost(const struct lzms_range_encoder *enc, u8 *cur_state, int bit)
+/* Return the cost to range-encode the specified bit from the specified state.*/
+static inline u32
+lzms_rc_bit_cost(const struct lzms_range_encoder *enc, u8 cur_state, int bit)
{
u32 prob_zero;
u32 prob_correct;
- prob_zero = enc->prob_entries[*cur_state & enc->mask].num_recent_zero_bits;
-
- *cur_state = (*cur_state << 1) | bit;
+ prob_zero = enc->prob_entries[cur_state].num_recent_zero_bits;
if (bit == 0)
prob_correct = prob_zero;
return lzms_rc_costs[prob_correct];
}
-static u32
-lzms_huffman_symbol_cost(const struct lzms_huffman_encoder *enc, u32 sym)
+/* Return the cost to Huffman-encode the specified symbol. */
+static inline u32
+lzms_huffman_symbol_cost(const struct lzms_huffman_encoder *enc, unsigned sym)
{
- return enc->lens[sym] << LZMS_COST_SHIFT;
+ return (u32)enc->lens[sym] << LZMS_COST_SHIFT;
}
-static u32
-lzms_offset_cost(const struct lzms_huffman_encoder *enc, u32 offset)
+/* Return the cost to encode the specified literal byte. */
+static inline u32
+lzms_literal_cost(const struct lzms_compressor *c, unsigned literal,
+ const struct lzms_adaptive_state *state)
{
- u32 slot;
- u32 num_extra_bits;
- u32 cost = 0;
-
- slot = lzms_get_position_slot(offset);
-
- cost += lzms_huffman_symbol_cost(enc, slot);
-
- num_extra_bits = lzms_extra_position_bits[slot];
-
- cost += num_extra_bits << LZMS_COST_SHIFT;
-
- return cost;
+ return lzms_rc_bit_cost(&c->main_range_encoder, state->main_state, 0) +
+ lzms_huffman_symbol_cost(&c->literal_encoder, literal);
}
-static u32
-lzms_get_length_cost(const struct lzms_huffman_encoder *enc, u32 length)
+/* Update the table that directly provides the costs for small lengths. */
+static void
+lzms_update_fast_length_costs(struct lzms_compressor *c)
{
- u32 slot;
- u32 num_extra_bits;
+ u32 len;
+ int slot = -1;
u32 cost = 0;
- slot = lzms_get_length_slot(length);
+ for (len = 1; len < LZMS_NUM_FAST_LENGTHS; len++) {
- cost += lzms_huffman_symbol_cost(enc, slot);
-
- num_extra_bits = lzms_extra_length_bits[slot];
-
- cost += num_extra_bits << LZMS_COST_SHIFT;
+ while (len >= lzms_length_slot_base[slot + 1]) {
+ slot++;
+ cost = (u32)(c->length_encoder.lens[slot] +
+ lzms_extra_length_bits[slot]) << LZMS_COST_SHIFT;
+ }
- return cost;
+ c->length_cost_fast[len] = cost;
+ }
}
-static u32
-lzms_get_matches(struct lzms_compressor *ctx, struct lz_match **matches_ret)
+/* Return the cost to encode the specified match length, which must be less than
+ * LZMS_NUM_FAST_LENGTHS. */
+static inline u32
+lzms_fast_length_cost(const struct lzms_compressor *c, u32 length)
{
- *matches_ret = ctx->matches;
- return lz_mf_get_matches(ctx->mf, ctx->matches);
+ LZMS_ASSERT(length < LZMS_NUM_FAST_LENGTHS);
+ return c->length_cost_fast[length];
}
-static void
-lzms_skip_bytes(struct lzms_compressor *ctx, u32 n)
+/* Return the cost to encode the specified LZ match offset. */
+static inline u32
+lzms_lz_offset_cost(const struct lzms_compressor *c, u32 offset)
{
- lz_mf_skip_positions(ctx->mf, n);
+ unsigned slot = lzms_get_offset_slot_fast(c, offset);
+
+ return (u32)(c->lz_offset_encoder.lens[slot] +
+ lzms_extra_offset_bits[slot]) << LZMS_COST_SHIFT;
}
-static u32
-lzms_get_literal_cost(struct lzms_compressor *ctx,
- struct lzms_adaptive_state *state, u8 literal)
+/*
+ * Consider coding the match at repeat offset index @rep_idx. Consider each
+ * length from the minimum (2) to the full match length (@rep_len).
+ */
+static inline void
+lzms_consider_lz_repeat_offset_match(const struct lzms_compressor *c,
+ struct lzms_mc_pos_data *cur_optimum_ptr,
+ u32 rep_len, unsigned rep_idx)
{
- u32 cost = 0;
-
- state->lru.upcoming_offset = 0;
- lzms_update_lz_lru_queues(&state->lru);
+ u32 len;
+ u32 base_cost;
+ u32 cost;
+ unsigned i;
- cost += lzms_rc_bit_cost(&ctx->main_range_encoder,
- &state->main_state, 0);
+ base_cost = cur_optimum_ptr->cost;
- cost += lzms_huffman_symbol_cost(&ctx->literal_encoder, literal);
+ base_cost += lzms_rc_bit_cost(&c->main_range_encoder,
+ cur_optimum_ptr->state.main_state, 1);
- return cost;
-}
+ base_cost += lzms_rc_bit_cost(&c->match_range_encoder,
+ cur_optimum_ptr->state.match_state, 0);
-static u32
-lzms_get_lz_match_cost_nolen(struct lzms_compressor *ctx,
- struct lzms_adaptive_state *state, u32 offset)
-{
- u32 cost = 0;
- int recent_offset_idx;
+ base_cost += lzms_rc_bit_cost(&c->lz_match_range_encoder,
+ cur_optimum_ptr->state.lz_match_state, 1);
- cost += lzms_rc_bit_cost(&ctx->main_range_encoder,
- &state->main_state, 1);
- cost += lzms_rc_bit_cost(&ctx->match_range_encoder,
- &state->match_state, 0);
+ for (i = 0; i < rep_idx; i++)
+ base_cost += lzms_rc_bit_cost(&c->lz_repeat_match_range_encoders[i],
+ cur_optimum_ptr->state.lz_repeat_match_state[i], 1);
- for (recent_offset_idx = 0;
- recent_offset_idx < LZMS_NUM_RECENT_OFFSETS;
- recent_offset_idx++)
- if (offset == state->lru.recent_offsets[recent_offset_idx])
- break;
+ if (i < LZMS_NUM_RECENT_OFFSETS - 1)
+ base_cost += lzms_rc_bit_cost(&c->lz_repeat_match_range_encoders[i],
+ cur_optimum_ptr->state.lz_repeat_match_state[i], 0);
- if (recent_offset_idx == LZMS_NUM_RECENT_OFFSETS) {
- /* Explicit offset. */
- cost += lzms_rc_bit_cost(&ctx->lz_match_range_encoder,
- &state->lz_match_state, 0);
+ len = 2;
+ do {
+ cost = base_cost + lzms_fast_length_cost(c, len);
+ if (cost < (cur_optimum_ptr + len)->cost) {
+ (cur_optimum_ptr + len)->mc_item_data =
+ ((u64)rep_idx << MC_OFFSET_SHIFT) | len;
+ (cur_optimum_ptr + len)->cost = cost;
+ }
+ } while (++len <= rep_len);
+}
- cost += lzms_offset_cost(&ctx->lz_offset_encoder, offset);
- } else {
- int i;
+/*
+ * Consider coding each match in @matches as an explicit offset match.
+ *
+ * @matches must be sorted by strictly increasing length and strictly increasing
+ * offset. This is guaranteed by the match-finder.
+ *
+ * We consider each length from the minimum (2) to the longest
+ * (matches[num_matches - 1].len). For each length, we consider only the
+ * smallest offset for which that length is available. Although this is not
+ * guaranteed to be optimal due to the possibility of a larger offset costing
+ * less than a smaller offset to code, this is a very useful heuristic.
+ */
+static inline void
+lzms_consider_lz_explicit_offset_matches(const struct lzms_compressor *c,
+ struct lzms_mc_pos_data *cur_optimum_ptr,
+ const struct lz_match matches[],
+ u32 num_matches)
+{
+ u32 len;
+ u32 i;
+ u32 base_cost;
+ u32 position_cost;
+ u32 cost;
- /* Recent offset. */
- cost += lzms_rc_bit_cost(&ctx->lz_match_range_encoder,
- &state->lz_match_state, 1);
+ base_cost = cur_optimum_ptr->cost;
- for (i = 0; i < recent_offset_idx; i++)
- cost += lzms_rc_bit_cost(&ctx->lz_repeat_match_range_encoders[i],
- &state->lz_repeat_match_state[i], 0);
+ base_cost += lzms_rc_bit_cost(&c->main_range_encoder,
+ cur_optimum_ptr->state.main_state, 1);
- if (i < LZMS_NUM_RECENT_OFFSETS - 1)
- cost += lzms_rc_bit_cost(&ctx->lz_repeat_match_range_encoders[i],
- &state->lz_repeat_match_state[i], 1);
+ base_cost += lzms_rc_bit_cost(&c->match_range_encoder,
+ cur_optimum_ptr->state.match_state, 0);
+ base_cost += lzms_rc_bit_cost(&c->lz_match_range_encoder,
+ cur_optimum_ptr->state.lz_match_state, 0);
+ len = 2;
+ i = 0;
+ do {
+ position_cost = base_cost + lzms_lz_offset_cost(c,
+ matches[i].offset);
+ do {
+ cost = position_cost + lzms_fast_length_cost(c, len);
+ if (cost < (cur_optimum_ptr + len)->cost) {
+ (cur_optimum_ptr + len)->mc_item_data =
+ ((u64)(matches[i].offset + LZMS_OFFSET_OFFSET)
+ << MC_OFFSET_SHIFT) | len;
+ (cur_optimum_ptr + len)->cost = cost;
+ }
+ } while (++len <= matches[i].len);
+ } while (++i != num_matches);
+}
- /* Initial update of the LZ match offset LRU queue. */
- for (; i < LZMS_NUM_RECENT_OFFSETS; i++)
- state->lru.recent_offsets[i] = state->lru.recent_offsets[i + 1];
- }
+static void
+lzms_init_adaptive_state(struct lzms_adaptive_state *state)
+{
+ unsigned i;
+
+ lzms_init_lz_lru_queues(&state->lru);
+ state->main_state = 0;
+ state->match_state = 0;
+ state->lz_match_state = 0;
+ for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
+ state->lz_repeat_match_state[i] = 0;
+}
+static inline void
+lzms_update_main_state(struct lzms_adaptive_state *state, int is_match)
+{
+ state->main_state = ((state->main_state << 1) | is_match) % LZMS_NUM_MAIN_STATES;
+}
- state->lru.upcoming_offset = offset;
- lzms_update_lz_lru_queues(&state->lru);
+static inline void
+lzms_update_match_state(struct lzms_adaptive_state *state, int is_delta)
+{
+ state->match_state = ((state->match_state << 1) | is_delta) % LZMS_NUM_MATCH_STATES;
+}
- return cost;
+static inline void
+lzms_update_lz_match_state(struct lzms_adaptive_state *state, int is_repeat_offset)
+{
+ state->lz_match_state = ((state->lz_match_state << 1) | is_repeat_offset) % LZMS_NUM_LZ_MATCH_STATES;
}
-static u32
-lzms_get_lz_match_cost(struct lzms_compressor *ctx,
- struct lzms_adaptive_state *state,
- u32 length, u32 offset)
+static inline void
+lzms_update_lz_repeat_match_state(struct lzms_adaptive_state *state, int rep_idx)
{
- return lzms_get_lz_match_cost_nolen(ctx, state, offset) +
- lzms_get_length_cost(&ctx->length_encoder, length);
+ int i;
+
+ for (i = 0; i < rep_idx; i++)
+ state->lz_repeat_match_state[i] =
+ ((state->lz_repeat_match_state[i] << 1) | 1) %
+ LZMS_NUM_LZ_REPEAT_MATCH_STATES;
+
+ if (i < LZMS_NUM_RECENT_OFFSETS - 1)
+ state->lz_repeat_match_state[i] =
+ ((state->lz_repeat_match_state[i] << 1) | 0) %
+ LZMS_NUM_LZ_REPEAT_MATCH_STATES;
}
-static inline u32
-lzms_repsearch(const u8 * const strptr, const u32 bytes_remaining,
- const struct lzms_lz_lru_queues *queue, u32 *offset_ret)
+/*
+ * The main near-optimal parsing routine.
+ *
+ * Briefly, the algorithm does an approximate minimum-cost path search to find a
+ * "near-optimal" sequence of matches and literals to output, based on the
+ * current cost model. The algorithm steps forward, position by position (byte
+ * by byte), and updates the minimum cost path to reach each later position that
+ * can be reached using a match or literal from the current position. This is
+ * essentially Dijkstra's algorithm in disguise: the graph nodes are positions,
+ * the graph edges are possible matches/literals to code, and the cost of each
+ * edge is the estimated number of bits that will be required to output the
+ * corresponding match or literal. But one difference is that we actually
+ * compute the lowest-cost path in pieces, where each piece is terminated when
+ * there are no choices to be made.
+ *
+ * Notes:
+ *
+ * - This does not output any delta matches.
+ *
+ * - The costs of literals and matches are estimated using the range encoder
+ * states and the semi-adaptive Huffman codes. Except for range encoding
+ * states, costs are assumed to be constant throughout a single run of the
+ * parsing algorithm, which can parse up to @optim_array_length bytes of data.
+ * This introduces a source of inaccuracy because the probabilities and
+ * Huffman codes can change over this part of the data.
+ */
+static void
+lzms_near_optimal_parse(struct lzms_compressor *c)
{
+ const u8 *window_ptr;
+ const u8 *window_end;
+ struct lzms_mc_pos_data *cur_optimum_ptr;
+ struct lzms_mc_pos_data *end_optimum_ptr;
+ u32 num_matches;
+ u32 longest_len;
+ u32 rep_max_len;
+ unsigned rep_max_idx;
+ unsigned literal;
+ unsigned i;
+ u32 cost;
u32 len;
- unsigned slot = 0;
+ u32 offset_data;
- len = lz_repsearch(strptr, bytes_remaining, UINT32_MAX,
- queue->recent_offsets, LZMS_NUM_RECENT_OFFSETS, &slot);
- *offset_ret = queue->recent_offsets[slot];
- return len;
-}
+ window_ptr = c->cur_window;
+ window_end = window_ptr + c->cur_window_size;
+ lzms_init_adaptive_state(&c->optimum[0].state);
-static struct lz_match
-lzms_match_chooser_reverse_list(struct lzms_compressor *ctx, unsigned cur_pos)
-{
- unsigned prev_link, saved_prev_link;
- unsigned prev_match_offset, saved_prev_match_offset;
+begin:
+ /* Start building a new list of items, which will correspond to the next
+ * piece of the overall minimum-cost path. */
- ctx->optimum_end_idx = cur_pos;
+ cur_optimum_ptr = c->optimum;
+ cur_optimum_ptr->cost = 0;
+ end_optimum_ptr = cur_optimum_ptr;
- saved_prev_link = ctx->optimum[cur_pos].prev.link;
- saved_prev_match_offset = ctx->optimum[cur_pos].prev.match_offset;
+ /* States should currently be consistent with the encoders. */
+ LZMS_ASSERT(cur_optimum_ptr->state.main_state == c->main_range_encoder.state);
+ LZMS_ASSERT(cur_optimum_ptr->state.match_state == c->match_range_encoder.state);
+ LZMS_ASSERT(cur_optimum_ptr->state.lz_match_state == c->lz_match_range_encoder.state);
+ for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
+ LZMS_ASSERT(cur_optimum_ptr->state.lz_repeat_match_state[i] ==
+ c->lz_repeat_match_range_encoders[i].state);
- do {
- prev_link = saved_prev_link;
- prev_match_offset = saved_prev_match_offset;
+ if (window_ptr == window_end)
+ return;
- saved_prev_link = ctx->optimum[prev_link].prev.link;
- saved_prev_match_offset = ctx->optimum[prev_link].prev.match_offset;
+ /* The following loop runs once for each per byte in the window, except
+ * in a couple shortcut cases. */
+ for (;;) {
- ctx->optimum[prev_link].next.link = cur_pos;
- ctx->optimum[prev_link].next.match_offset = prev_match_offset;
+ /* Find explicit offset matches with the current position. */
+ num_matches = lz_mf_get_matches(c->mf, c->matches);
- cur_pos = prev_link;
- } while (cur_pos != 0);
+ if (num_matches) {
+ /*
+ * Find the longest repeat offset match with the current
+ * position.
+ *
+ * Heuristics:
+ *
+ * - Only search for repeat offset matches if the
+ * match-finder already found at least one match.
+ *
+ * - Only consider the longest repeat offset match. It
+ * seems to be rare for the optimal parse to include a
+ * repeat offset match that doesn't have the longest
+ * length (allowing for the possibility that not all
+ * of that length is actually used).
+ */
+ if (likely(window_ptr - c->cur_window >= LZMS_MAX_INIT_RECENT_OFFSET)) {
+ BUILD_BUG_ON(LZMS_NUM_RECENT_OFFSETS != 3);
+ rep_max_len = lz_repsearch3(window_ptr,
+ window_end - window_ptr,
+ cur_optimum_ptr->state.lru.recent_offsets,
+ &rep_max_idx);
+ } else {
+ rep_max_len = 0;
+ }
- ctx->optimum_cur_idx = ctx->optimum[0].next.link;
+ if (rep_max_len) {
+ /* If there's a very long repeat offset match,
+ * choose it immediately. */
+ if (rep_max_len >= c->params.nice_match_length) {
- return (struct lz_match)
- { .len = ctx->optimum_cur_idx,
- .offset = ctx->optimum[0].next.match_offset,
- };
-}
+ lz_mf_skip_positions(c->mf, rep_max_len - 1);
+ window_ptr += rep_max_len;
-/* This is similar to lzx_choose_near_optimal_item() in lzx-compress.c.
- * Read that one if you want to understand it. */
-static struct lz_match
-lzms_get_near_optimal_item(struct lzms_compressor *ctx)
-{
- u32 num_matches;
- struct lz_match *matches;
- struct lz_match match;
- u32 longest_len;
- u32 longest_rep_len;
- u32 longest_rep_offset;
- unsigned cur_pos;
- unsigned end_pos;
- struct lzms_adaptive_state initial_state;
-
- if (ctx->optimum_cur_idx != ctx->optimum_end_idx) {
- match.len = ctx->optimum[ctx->optimum_cur_idx].next.link -
- ctx->optimum_cur_idx;
- match.offset = ctx->optimum[ctx->optimum_cur_idx].next.match_offset;
-
- ctx->optimum_cur_idx = ctx->optimum[ctx->optimum_cur_idx].next.link;
- return match;
- }
+ if (cur_optimum_ptr != c->optimum)
+ lzms_encode_item_list(c, cur_optimum_ptr);
- ctx->optimum_cur_idx = 0;
- ctx->optimum_end_idx = 0;
+ lzms_encode_lz_repeat_offset_match(c, rep_max_len,
+ rep_max_idx);
- if (lz_mf_get_position(ctx->mf) >= LZMS_MAX_INIT_RECENT_OFFSET) {
- longest_rep_len = lzms_repsearch(lz_mf_get_window_ptr(ctx->mf),
- lz_mf_get_bytes_remaining(ctx->mf),
- &ctx->lru.lz, &longest_rep_offset);
- } else {
- longest_rep_len = 0;
- }
+ c->optimum[0].state = cur_optimum_ptr->state;
- if (longest_rep_len >= ctx->params.nice_match_length) {
- lzms_skip_bytes(ctx, longest_rep_len);
- return (struct lz_match) {
- .len = longest_rep_len,
- .offset = longest_rep_offset,
- };
- }
+ lzms_update_main_state(&c->optimum[0].state, 1);
+ lzms_update_match_state(&c->optimum[0].state, 0);
+ lzms_update_lz_match_state(&c->optimum[0].state, 1);
+ lzms_update_lz_repeat_match_state(&c->optimum[0].state,
+ rep_max_idx);
- num_matches = lzms_get_matches(ctx, &matches);
+ c->optimum[0].state.lru.upcoming_offset =
+ c->optimum[0].state.lru.recent_offsets[rep_max_idx];
- if (num_matches) {
- longest_len = matches[num_matches - 1].len;
- if (longest_len >= ctx->params.nice_match_length) {
- lzms_skip_bytes(ctx, longest_len - 1);
- return matches[num_matches - 1];
- }
- } else {
- longest_len = 1;
- }
+ for (i = rep_max_idx; i < LZMS_NUM_RECENT_OFFSETS; i++)
+ c->optimum[0].state.lru.recent_offsets[i] =
+ c->optimum[0].state.lru.recent_offsets[i + 1];
- initial_state.lru = ctx->lru.lz;
- initial_state.main_state = ctx->main_range_encoder.state;
- initial_state.match_state = ctx->match_range_encoder.state;
- initial_state.lz_match_state = ctx->lz_match_range_encoder.state;
- for (int i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
- initial_state.lz_repeat_match_state[i] = ctx->lz_repeat_match_range_encoders[i].state;
+ lzms_update_lz_lru_queue(&c->optimum[0].state.lru);
+ goto begin;
+ }
- ctx->optimum[1].state = initial_state;
- ctx->optimum[1].cost = lzms_get_literal_cost(ctx,
- &ctx->optimum[1].state,
- *(lz_mf_get_window_ptr(ctx->mf) - 1));
- ctx->optimum[1].prev.link = 0;
+ /* If reaching any positions for the first time,
+ * initialize their costs to "infinity". */
+ while (end_optimum_ptr < cur_optimum_ptr + rep_max_len)
+ (++end_optimum_ptr)->cost = MC_INFINITE_COST;
- for (u32 i = 0, len = 2; i < num_matches; i++) {
- u32 offset = matches[i].offset;
- struct lzms_adaptive_state state;
- u32 position_cost;
+ /* Consider coding a repeat offset match. */
+ lzms_consider_lz_repeat_offset_match(c, cur_optimum_ptr,
+ rep_max_len, rep_max_idx);
+ }
- state = initial_state;
- position_cost = 0;
- position_cost += lzms_get_lz_match_cost_nolen(ctx, &state, offset);
+ longest_len = c->matches[num_matches - 1].len;
- do {
- u32 cost;
+ /* If there's a very long explicit offset match, choose
+ * it immediately. */
+ if (longest_len >= c->params.nice_match_length) {
- cost = position_cost;
- cost += lzms_get_length_cost(&ctx->length_encoder, len);
+ lz_mf_skip_positions(c->mf, longest_len - 1);
+ window_ptr += longest_len;
- ctx->optimum[len].state = state;
- ctx->optimum[len].prev.link = 0;
- ctx->optimum[len].prev.match_offset = offset;
- ctx->optimum[len].cost = cost;
- } while (++len <= matches[i].len);
- }
- end_pos = longest_len;
-
- if (longest_rep_len) {
- struct lzms_adaptive_state state;
- u32 cost;
-
- while (end_pos < longest_rep_len)
- ctx->optimum[++end_pos].cost = MC_INFINITE_COST;
-
- state = initial_state;
- cost = lzms_get_lz_match_cost(ctx,
- &state,
- longest_rep_len,
- longest_rep_offset);
- if (cost <= ctx->optimum[longest_rep_len].cost) {
- ctx->optimum[longest_rep_len].state = state;
- ctx->optimum[longest_rep_len].prev.link = 0;
- ctx->optimum[longest_rep_len].prev.match_offset = longest_rep_offset;
- ctx->optimum[longest_rep_len].cost = cost;
- }
- }
+ if (cur_optimum_ptr != c->optimum)
+ lzms_encode_item_list(c, cur_optimum_ptr);
- cur_pos = 0;
- for (;;) {
- u32 cost;
- struct lzms_adaptive_state state;
+ lzms_encode_lz_explicit_offset_match(c, longest_len,
+ c->matches[num_matches - 1].offset);
- cur_pos++;
+ c->optimum[0].state = cur_optimum_ptr->state;
- if (cur_pos == end_pos || cur_pos == ctx->params.optim_array_length)
- return lzms_match_chooser_reverse_list(ctx, cur_pos);
+ lzms_update_main_state(&c->optimum[0].state, 1);
+ lzms_update_match_state(&c->optimum[0].state, 0);
+ lzms_update_lz_match_state(&c->optimum[0].state, 0);
- if (lz_mf_get_position(ctx->mf) >= LZMS_MAX_INIT_RECENT_OFFSET) {
- longest_rep_len = lzms_repsearch(lz_mf_get_window_ptr(ctx->mf),
- lz_mf_get_bytes_remaining(ctx->mf),
- &ctx->optimum[cur_pos].state.lru,
- &longest_rep_offset);
- } else {
- longest_rep_len = 0;
- }
+ c->optimum[0].state.lru.upcoming_offset =
+ c->matches[num_matches - 1].offset;
- if (longest_rep_len >= ctx->params.nice_match_length) {
- match = lzms_match_chooser_reverse_list(ctx, cur_pos);
+ lzms_update_lz_lru_queue(&c->optimum[0].state.lru);
+ goto begin;
+ }
- ctx->optimum[cur_pos].next.match_offset = longest_rep_offset;
- ctx->optimum[cur_pos].next.link = cur_pos + longest_rep_len;
- ctx->optimum_end_idx = cur_pos + longest_rep_len;
+ /* If reaching any positions for the first time,
+ * initialize their costs to "infinity". */
+ while (end_optimum_ptr < cur_optimum_ptr + longest_len)
+ (++end_optimum_ptr)->cost = MC_INFINITE_COST;
- lzms_skip_bytes(ctx, longest_rep_len);
+ /* Consider coding an explicit offset match. */
+ lzms_consider_lz_explicit_offset_matches(c, cur_optimum_ptr,
+ c->matches, num_matches);
+ } else {
+ /* No matches found. The only choice at this position
+ * is to code a literal. */
- return match;
+ if (end_optimum_ptr == cur_optimum_ptr)
+ (++end_optimum_ptr)->cost = MC_INFINITE_COST;
}
- num_matches = lzms_get_matches(ctx, &matches);
+ /* Consider coding a literal.
- if (num_matches) {
- longest_len = matches[num_matches - 1].len;
- if (longest_len >= ctx->params.nice_match_length) {
- match = lzms_match_chooser_reverse_list(ctx, cur_pos);
+ * To avoid an extra unpredictable brench, actually checking the
+ * preferability of coding a literal is integrated into the
+ * adaptive state update code below. */
+ literal = *window_ptr++;
+ cost = cur_optimum_ptr->cost +
+ lzms_literal_cost(c, literal, &cur_optimum_ptr->state);
- ctx->optimum[cur_pos].next.match_offset =
- matches[num_matches - 1].offset;
- ctx->optimum[cur_pos].next.link = cur_pos + longest_len;
- ctx->optimum_end_idx = cur_pos + longest_len;
+ /* Advance to the next position. */
+ cur_optimum_ptr++;
- lzms_skip_bytes(ctx, longest_len - 1);
+ /* The lowest-cost path to the current position is now known.
+ * Finalize the adaptive state that results from taking this
+ * lowest-cost path. */
- return match;
- }
- } else {
- longest_len = 1;
- }
+ if (cost < cur_optimum_ptr->cost) {
+ /* Literal */
+ cur_optimum_ptr->cost = cost;
+ cur_optimum_ptr->mc_item_data = ((u64)literal << MC_OFFSET_SHIFT) | 1;
- while (end_pos < cur_pos + longest_len)
- ctx->optimum[++end_pos].cost = MC_INFINITE_COST;
-
- state = ctx->optimum[cur_pos].state;
- cost = ctx->optimum[cur_pos].cost +
- lzms_get_literal_cost(ctx,
- &state,
- *(lz_mf_get_window_ptr(ctx->mf) - 1));
- if (cost < ctx->optimum[cur_pos + 1].cost) {
- ctx->optimum[cur_pos + 1].state = state;
- ctx->optimum[cur_pos + 1].cost = cost;
- ctx->optimum[cur_pos + 1].prev.link = cur_pos;
- }
+ cur_optimum_ptr->state = (cur_optimum_ptr - 1)->state;
- for (u32 i = 0, len = 2; i < num_matches; i++) {
- u32 offset = matches[i].offset;
- struct lzms_adaptive_state state;
- u32 position_cost;
+ lzms_update_main_state(&cur_optimum_ptr->state, 0);
- state = ctx->optimum[cur_pos].state;
- position_cost = ctx->optimum[cur_pos].cost;
- position_cost += lzms_get_lz_match_cost_nolen(ctx, &state, offset);
+ cur_optimum_ptr->state.lru.upcoming_offset = 0;
+ } else {
+ /* LZ match */
+ len = cur_optimum_ptr->mc_item_data & MC_LEN_MASK;
+ offset_data = cur_optimum_ptr->mc_item_data >> MC_OFFSET_SHIFT;
- do {
- u32 cost;
+ cur_optimum_ptr->state = (cur_optimum_ptr - len)->state;
- cost = position_cost;
- cost += lzms_get_length_cost(&ctx->length_encoder, len);
+ lzms_update_main_state(&cur_optimum_ptr->state, 1);
+ lzms_update_match_state(&cur_optimum_ptr->state, 0);
- if (cost < ctx->optimum[cur_pos + len].cost) {
- ctx->optimum[cur_pos + len].state = state;
- ctx->optimum[cur_pos + len].prev.link = cur_pos;
- ctx->optimum[cur_pos + len].prev.match_offset = offset;
- ctx->optimum[cur_pos + len].cost = cost;
- }
- } while (++len <= matches[i].len);
- }
+ if (offset_data >= LZMS_NUM_RECENT_OFFSETS) {
- if (longest_rep_len >= ctx->params.min_match_length) {
-
- while (end_pos < cur_pos + longest_rep_len)
- ctx->optimum[++end_pos].cost = MC_INFINITE_COST;
-
- state = ctx->optimum[cur_pos].state;
-
- cost = ctx->optimum[cur_pos].cost +
- lzms_get_lz_match_cost(ctx,
- &state,
- longest_rep_len,
- longest_rep_offset);
- if (cost <= ctx->optimum[cur_pos + longest_rep_len].cost) {
- ctx->optimum[cur_pos + longest_rep_len].state =
- state;
- ctx->optimum[cur_pos + longest_rep_len].prev.link =
- cur_pos;
- ctx->optimum[cur_pos + longest_rep_len].prev.match_offset =
- longest_rep_offset;
- ctx->optimum[cur_pos + longest_rep_len].cost =
- cost;
- }
- }
- }
-}
+ /* Explicit offset LZ match */
-/*
- * The main loop for the LZMS compressor.
- *
- * Notes:
- *
- * - This does not output any delta matches.
- *
- * - The costs of literals and matches are estimated using the range encoder
- * states and the semi-adaptive Huffman codes. Except for range encoding
- * states, costs are assumed to be constant throughout a single run of the
- * parsing algorithm, which can parse up to @optim_array_length bytes of data.
- * This introduces a source of inaccuracy because the probabilities and
- * Huffman codes can change over this part of the data.
- */
-static void
-lzms_encode(struct lzms_compressor *ctx)
-{
- struct lz_match item;
+ lzms_update_lz_match_state(&cur_optimum_ptr->state, 0);
+
+ cur_optimum_ptr->state.lru.upcoming_offset =
+ offset_data - LZMS_OFFSET_OFFSET;
+ } else {
+ /* Repeat offset LZ match */
- /* Load window into the match-finder. */
- lz_mf_load_window(ctx->mf, ctx->window, ctx->window_size);
+ lzms_update_lz_match_state(&cur_optimum_ptr->state, 1);
+ lzms_update_lz_repeat_match_state(&cur_optimum_ptr->state,
+ offset_data);
- /* Reset the match-chooser. */
- ctx->optimum_cur_idx = 0;
- ctx->optimum_end_idx = 0;
+ cur_optimum_ptr->state.lru.upcoming_offset =
+ cur_optimum_ptr->state.lru.recent_offsets[offset_data];
- while (ctx->cur_window_pos != ctx->window_size) {
- item = lzms_get_near_optimal_item(ctx);
- if (item.len <= 1)
- lzms_encode_literal(ctx, ctx->window[ctx->cur_window_pos]);
- else
- lzms_encode_lz_match(ctx, item.len, item.offset);
+ for (i = offset_data; i < LZMS_NUM_RECENT_OFFSETS; i++)
+ cur_optimum_ptr->state.lru.recent_offsets[i] =
+ cur_optimum_ptr->state.lru.recent_offsets[i + 1];
+ }
+ }
+
+ lzms_update_lz_lru_queue(&cur_optimum_ptr->state.lru);
+
+ /*
+ * This loop will terminate when either of the following
+ * conditions is true:
+ *
+ * (1) cur_optimum_ptr == end_optimum_ptr
+ *
+ * There are no paths that extend beyond the current
+ * position. In this case, any path to a later position
+ * must pass through the current position, so we can go
+ * ahead and choose the list of items that led to this
+ * position.
+ *
+ * (2) cur_optimum_ptr == c->optimum_end
+ *
+ * This bounds the number of times the algorithm can step
+ * forward before it is guaranteed to start choosing items.
+ * This limits the memory usage. It also guarantees that
+ * the parser will not go too long without updating the
+ * probability tables.
+ *
+ * Note: no check for end-of-block is needed because
+ * end-of-block will trigger condition (1).
+ */
+ if (cur_optimum_ptr == end_optimum_ptr ||
+ cur_optimum_ptr == c->optimum_end)
+ {
+ c->optimum[0].state = cur_optimum_ptr->state;
+ break;
+ }
}
+
+ /* Output the current list of items that constitute the minimum-cost
+ * path to the current position. */
+ lzms_encode_item_list(c, cur_optimum_ptr);
+ goto begin;
}
static void
{
enc->rc = rc;
enc->state = 0;
+ LZMS_ASSERT(is_power_of_2(num_states));
enc->mask = num_states - 1;
for (u32 i = 0; i < num_states; i++) {
enc->prob_entries[i].num_recent_zero_bits = LZMS_INITIAL_PROBABILITY;
enc->codewords);
}
-/* Initialize the LZMS compressor. */
+/* Prepare the LZMS compressor for compressing a block of data. */
static void
-lzms_init_compressor(struct lzms_compressor *ctx, const u8 *udata, u32 ulen,
- le16 *cdata, u32 clen16)
+lzms_prepare_compressor(struct lzms_compressor *c, const u8 *udata, u32 ulen,
+ le16 *cdata, u32 clen16)
{
- unsigned num_position_slots;
+ unsigned num_offset_slots;
- /* Copy the uncompressed data into the @ctx->window buffer. */
- memcpy(ctx->window, udata, ulen);
- ctx->cur_window_pos = 0;
- ctx->window_size = ulen;
+ /* Copy the uncompressed data into the @c->cur_window buffer. */
+ memcpy(c->cur_window, udata, ulen);
+ c->cur_window_size = ulen;
/* Initialize the raw range encoder (writing forwards). */
- lzms_range_encoder_raw_init(&ctx->rc, cdata, clen16);
+ lzms_range_encoder_raw_init(&c->rc, cdata, clen16);
/* Initialize the output bitstream for Huffman symbols and verbatim bits
* (writing backwards). */
- lzms_output_bitstream_init(&ctx->os, cdata, clen16);
-
- /* Calculate the number of position slots needed for this compressed
- * block. */
- num_position_slots = lzms_get_position_slot(ulen - 1) + 1;
+ lzms_output_bitstream_init(&c->os, cdata, clen16);
- LZMS_DEBUG("Using %u position slots", num_position_slots);
+ /* Calculate the number of offset slots required. */
+ num_offset_slots = lzms_get_offset_slot(ulen - 1) + 1;
- /* Initialize Huffman encoders for each alphabet used in the compressed
- * representation. */
- lzms_init_huffman_encoder(&ctx->literal_encoder, &ctx->os,
+ /* Initialize a Huffman encoder for each alphabet. */
+ lzms_init_huffman_encoder(&c->literal_encoder, &c->os,
LZMS_NUM_LITERAL_SYMS,
LZMS_LITERAL_CODE_REBUILD_FREQ);
- lzms_init_huffman_encoder(&ctx->lz_offset_encoder, &ctx->os,
- num_position_slots,
+ lzms_init_huffman_encoder(&c->lz_offset_encoder, &c->os,
+ num_offset_slots,
LZMS_LZ_OFFSET_CODE_REBUILD_FREQ);
- lzms_init_huffman_encoder(&ctx->length_encoder, &ctx->os,
+ lzms_init_huffman_encoder(&c->length_encoder, &c->os,
LZMS_NUM_LEN_SYMS,
LZMS_LENGTH_CODE_REBUILD_FREQ);
- lzms_init_huffman_encoder(&ctx->delta_offset_encoder, &ctx->os,
- num_position_slots,
+ lzms_init_huffman_encoder(&c->delta_offset_encoder, &c->os,
+ num_offset_slots,
LZMS_DELTA_OFFSET_CODE_REBUILD_FREQ);
- lzms_init_huffman_encoder(&ctx->delta_power_encoder, &ctx->os,
+ lzms_init_huffman_encoder(&c->delta_power_encoder, &c->os,
LZMS_NUM_DELTA_POWER_SYMS,
LZMS_DELTA_POWER_CODE_REBUILD_FREQ);
/* Initialize range encoders, all of which wrap around the same
* lzms_range_encoder_raw. */
- lzms_init_range_encoder(&ctx->main_range_encoder,
- &ctx->rc, LZMS_NUM_MAIN_STATES);
+ lzms_init_range_encoder(&c->main_range_encoder,
+ &c->rc, LZMS_NUM_MAIN_STATES);
- lzms_init_range_encoder(&ctx->match_range_encoder,
- &ctx->rc, LZMS_NUM_MATCH_STATES);
+ lzms_init_range_encoder(&c->match_range_encoder,
+ &c->rc, LZMS_NUM_MATCH_STATES);
- lzms_init_range_encoder(&ctx->lz_match_range_encoder,
- &ctx->rc, LZMS_NUM_LZ_MATCH_STATES);
+ lzms_init_range_encoder(&c->lz_match_range_encoder,
+ &c->rc, LZMS_NUM_LZ_MATCH_STATES);
- for (size_t i = 0; i < ARRAY_LEN(ctx->lz_repeat_match_range_encoders); i++)
- lzms_init_range_encoder(&ctx->lz_repeat_match_range_encoders[i],
- &ctx->rc, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
+ for (unsigned i = 0; i < ARRAY_LEN(c->lz_repeat_match_range_encoders); i++)
+ lzms_init_range_encoder(&c->lz_repeat_match_range_encoders[i],
+ &c->rc, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
- lzms_init_range_encoder(&ctx->delta_match_range_encoder,
- &ctx->rc, LZMS_NUM_DELTA_MATCH_STATES);
+ lzms_init_range_encoder(&c->delta_match_range_encoder,
+ &c->rc, LZMS_NUM_DELTA_MATCH_STATES);
- for (size_t i = 0; i < ARRAY_LEN(ctx->delta_repeat_match_range_encoders); i++)
- lzms_init_range_encoder(&ctx->delta_repeat_match_range_encoders[i],
- &ctx->rc, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
+ for (unsigned i = 0; i < ARRAY_LEN(c->delta_repeat_match_range_encoders); i++)
+ lzms_init_range_encoder(&c->delta_repeat_match_range_encoders[i],
+ &c->rc, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
- /* Initialize LRU match information. */
- lzms_init_lru_queues(&ctx->lru);
+ /* Set initial length costs for lengths < LZMS_NUM_FAST_LENGTHS. */
+ lzms_update_fast_length_costs(c);
}
/* Flush the output streams, prepare the final compressed data, and return its
* A return value of 0 indicates that the data could not be compressed to fit in
* the available space. */
static size_t
-lzms_finalize(struct lzms_compressor *ctx, u8 *cdata, size_t csize_avail)
+lzms_finalize(struct lzms_compressor *c, u8 *cdata, size_t csize_avail)
{
size_t num_forwards_bytes;
size_t num_backwards_bytes;
- size_t compressed_size;
/* Flush both the forwards and backwards streams, and make sure they
* didn't cross each other and start overwriting each other's data. */
- if (!lzms_output_bitstream_flush(&ctx->os)) {
- LZMS_DEBUG("Backwards bitstream overrun.");
+ if (!lzms_output_bitstream_flush(&c->os))
return 0;
- }
- if (!lzms_range_encoder_raw_flush(&ctx->rc)) {
- LZMS_DEBUG("Forwards bitstream overrun.");
+ if (!lzms_range_encoder_raw_flush(&c->rc))
return 0;
- }
- if (ctx->rc.out > ctx->os.out) {
- LZMS_DEBUG("Two bitstreams crossed.");
+ if (c->rc.next > c->os.next)
return 0;
- }
/* Now the compressed buffer contains the data output by the forwards
* bitstream, then empty space, then data output by the backwards
* bitstream. Move the data output by the backwards bitstream to be
* adjacent to the data output by the forward bitstream, and calculate
* the compressed size that this results in. */
- num_forwards_bytes = (u8*)ctx->rc.out - (u8*)cdata;
- num_backwards_bytes = ((u8*)cdata + csize_avail) - (u8*)ctx->os.out;
+ num_forwards_bytes = (u8*)c->rc.next - (u8*)cdata;
+ num_backwards_bytes = ((u8*)cdata + csize_avail) - (u8*)c->os.next;
- memmove(cdata + num_forwards_bytes, ctx->os.out, num_backwards_bytes);
+ memmove(cdata + num_forwards_bytes, c->os.next, num_backwards_bytes);
- compressed_size = num_forwards_bytes + num_backwards_bytes;
- LZMS_DEBUG("num_forwards_bytes=%zu, num_backwards_bytes=%zu, "
- "compressed_size=%zu",
- num_forwards_bytes, num_backwards_bytes, compressed_size);
- LZMS_ASSERT(compressed_size % 2 == 0);
- return compressed_size;
+ return num_forwards_bytes + num_backwards_bytes;
}
-
+/* Set internal compression parameters for the specified compression level and
+ * maximum window size. */
static void
lzms_build_params(unsigned int compression_level,
struct lzms_compressor_params *lzms_params)
{
- lzms_params->min_match_length = (compression_level >= 50) ? 2 : 3;
- lzms_params->nice_match_length = max(((u64)compression_level * 32) / 50,
- lzms_params->min_match_length);
- lzms_params->max_search_depth = ((u64)compression_level * 50) / 50;
- lzms_params->optim_array_length = 224 + compression_level * 16;
+ /* Allow length 2 matches if the compression level is sufficiently high.
+ */
+ if (compression_level >= 45)
+ lzms_params->min_match_length = 2;
+ else
+ lzms_params->min_match_length = 3;
+
+ /* Scale nice_match_length and max_search_depth with the compression
+ * level. But to allow an optimization on length cost calculations,
+ * don't allow nice_match_length to exceed LZMS_NUM_FAST_LENGTH. */
+ lzms_params->nice_match_length = ((u64)compression_level * 32) / 50;
+ if (lzms_params->nice_match_length < lzms_params->min_match_length)
+ lzms_params->nice_match_length = lzms_params->min_match_length;
+ if (lzms_params->nice_match_length > LZMS_NUM_FAST_LENGTHS)
+ lzms_params->nice_match_length = LZMS_NUM_FAST_LENGTHS;
+ lzms_params->max_search_depth = compression_level;
+
+ lzms_params->optim_array_length = 1024;
}
+/* Given the internal compression parameters and maximum window size, build the
+ * Lempel-Ziv match-finder parameters. */
static void
lzms_build_mf_params(const struct lzms_compressor_params *lzms_params,
u32 max_window_size, struct lz_mf_params *mf_params)
{
memset(mf_params, 0, sizeof(*mf_params));
- mf_params->algorithm = LZ_MF_DEFAULT;
+ /* Choose an appropriate match-finding algorithm. */
+ if (max_window_size <= 2097152)
+ mf_params->algorithm = LZ_MF_BINARY_TREES;
+ else if (max_window_size <= 33554432)
+ mf_params->algorithm = LZ_MF_LCP_INTERVAL_TREE;
+ else
+ mf_params->algorithm = LZ_MF_LINKED_SUFFIX_ARRAY;
+
mf_params->max_window_size = max_window_size;
mf_params->min_match_len = lzms_params->min_match_length;
mf_params->max_search_depth = lzms_params->max_search_depth;
}
static void
-lzms_free_compressor(void *_ctx);
+lzms_free_compressor(void *_c);
static u64
lzms_get_needed_memory(size_t max_block_size, unsigned int compression_level)
{
struct lzms_compressor_params params;
+ struct lz_mf_params mf_params;
u64 size = 0;
if (max_block_size >= INT32_MAX)
return 0;
lzms_build_params(compression_level, ¶ms);
+ lzms_build_mf_params(¶ms, max_block_size, &mf_params);
size += sizeof(struct lzms_compressor);
+
+ /* cur_window */
size += max_block_size;
- size += lz_mf_get_needed_memory(LZ_MF_DEFAULT, max_block_size);
- size += params.max_search_depth * sizeof(struct lz_match);
+
+ /* mf */
+ size += lz_mf_get_needed_memory(mf_params.algorithm, max_block_size);
+
+ /* matches */
+ size += min(params.max_search_depth, params.nice_match_length) *
+ sizeof(struct lz_match);
+
+ /* optimum */
size += (params.optim_array_length + params.nice_match_length) *
sizeof(struct lzms_mc_pos_data);
lzms_create_compressor(size_t max_block_size, unsigned int compression_level,
void **ctx_ret)
{
- struct lzms_compressor *ctx;
+ struct lzms_compressor *c;
struct lzms_compressor_params params;
struct lz_mf_params mf_params;
if (!lz_mf_params_valid(&mf_params))
return WIMLIB_ERR_INVALID_PARAM;
- ctx = CALLOC(1, sizeof(struct lzms_compressor));
- if (!ctx)
+ c = CALLOC(1, sizeof(struct lzms_compressor));
+ if (!c)
goto oom;
- ctx->params = params;
- ctx->max_block_size = max_block_size;
+ c->params = params;
- ctx->window = MALLOC(max_block_size);
- if (!ctx->window)
+ c->cur_window = MALLOC(max_block_size);
+ if (!c->cur_window)
goto oom;
- ctx->mf = lz_mf_alloc(&mf_params);
- if (!ctx->mf)
+ c->mf = lz_mf_alloc(&mf_params);
+ if (!c->mf)
goto oom;
- ctx->matches = MALLOC(params.max_search_depth * sizeof(struct lz_match));
- if (!ctx->matches)
+ c->matches = MALLOC(min(params.max_search_depth,
+ params.nice_match_length) *
+ sizeof(struct lz_match));
+ if (!c->matches)
goto oom;
- ctx->optimum = MALLOC((params.optim_array_length +
- params.nice_match_length) *
- sizeof(struct lzms_mc_pos_data));
- if (!ctx->optimum)
+ c->optimum = MALLOC((params.optim_array_length +
+ params.nice_match_length) *
+ sizeof(struct lzms_mc_pos_data));
+ if (!c->optimum)
goto oom;
+ c->optimum_end = &c->optimum[params.optim_array_length];
- /* Initialize position and length slot data if not done already. */
lzms_init_slots();
- /* Initialize range encoding cost table if not done already. */
lzms_init_rc_costs();
- *ctx_ret = ctx;
+ lzms_init_fast_slots(c);
+
+ *ctx_ret = c;
return 0;
oom:
- lzms_free_compressor(ctx);
+ lzms_free_compressor(c);
return WIMLIB_ERR_NOMEM;
}
static size_t
lzms_compress(const void *uncompressed_data, size_t uncompressed_size,
- void *compressed_data, size_t compressed_size_avail, void *_ctx)
+ void *compressed_data, size_t compressed_size_avail, void *_c)
{
- struct lzms_compressor *ctx = _ctx;
- size_t compressed_size;
-
- LZMS_DEBUG("uncompressed_size=%zu, compressed_size_avail=%zu",
- uncompressed_size, compressed_size_avail);
+ struct lzms_compressor *c = _c;
/* Don't bother compressing extremely small inputs. */
- if (uncompressed_size < 4) {
- LZMS_DEBUG("Input too small to bother compressing.");
+ if (uncompressed_size < 4)
return 0;
- }
/* Cap the available compressed size to a 32-bit integer and round it
* down to the nearest multiple of 2. */
compressed_size_avail--;
/* Initialize the compressor structures. */
- lzms_init_compressor(ctx, uncompressed_data, uncompressed_size,
- compressed_data, compressed_size_avail / 2);
+ lzms_prepare_compressor(c, uncompressed_data, uncompressed_size,
+ compressed_data, compressed_size_avail / 2);
/* Preprocess the uncompressed data. */
- lzms_x86_filter(ctx->window, ctx->window_size,
- ctx->last_target_usages, false);
+ lzms_x86_filter(c->cur_window, c->cur_window_size,
+ c->last_target_usages, false);
+
+ /* Load the window into the match-finder. */
+ lz_mf_load_window(c->mf, c->cur_window, c->cur_window_size);
/* Compute and encode a literal/match sequence that decompresses to the
* preprocessed data. */
- lzms_encode(ctx);
-
- /* Get and return the compressed data size. */
- compressed_size = lzms_finalize(ctx, compressed_data,
- compressed_size_avail);
-
- if (compressed_size == 0) {
- LZMS_DEBUG("Data did not compress to requested size or less.");
- return 0;
- }
-
- LZMS_DEBUG("Compressed %zu => %zu bytes",
- uncompressed_size, compressed_size);
+ lzms_near_optimal_parse(c);
- return compressed_size;
+ /* Return the compressed data size or 0. */
+ return lzms_finalize(c, compressed_data, compressed_size_avail);
}
static void
-lzms_free_compressor(void *_ctx)
+lzms_free_compressor(void *_c)
{
- struct lzms_compressor *ctx = _ctx;
-
- if (ctx) {
- FREE(ctx->window);
- lz_mf_free(ctx->mf);
- FREE(ctx->matches);
- FREE(ctx->optimum);
- FREE(ctx);
+ struct lzms_compressor *c = _c;
+
+ if (c) {
+ FREE(c->cur_window);
+ lz_mf_free(c->mf);
+ FREE(c->matches);
+ FREE(c->optimum);
+ FREE(c);
}
}
git://wimlib.net / wimlib / blobdiff