*/
/*
- * Copyright (C) 2013 Eric Biggers
+ * Copyright (C) 2013, 2014 Eric Biggers
*
* This file is part of wimlib, a library for working with WIM files.
*
/* 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.
*/
# include "config.h"
#endif
-#include "wimlib.h"
#include "wimlib/assert.h"
#include "wimlib/compiler.h"
#include "wimlib/compressor_ops.h"
#include "wimlib/compress_common.h"
#include "wimlib/endianness.h"
#include "wimlib/error.h"
-#include "wimlib/lz_sarray.h"
+#include "wimlib/lz_mf.h"
+#include "wimlib/lz_repsearch.h"
#include "wimlib/lzms.h"
#include "wimlib/util.h"
#include <limits.h>
#include <pthread.h>
-struct lzms_compressor;
-struct lzms_adaptive_state {
- struct lzms_lz_lru_queues lru;
- u8 main_state;
- u8 match_state;
- u8 lz_match_state;
- u8 lz_repeat_match_state[LZMS_NUM_RECENT_OFFSETS - 1];
-};
-#define LZ_ADAPTIVE_STATE struct lzms_adaptive_state
-#define LZ_COMPRESSOR struct lzms_compressor
-#include "wimlib/lz_optimal.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. */
struct lzms_output_bitstream {
u32 codewords[LZMS_MAX_NUM_SYMS];
};
+struct lzms_compressor_params {
+ u32 min_match_length;
+ u32 nice_match_length;
+ u32 max_search_depth;
+ u32 optim_array_length;
+};
+
/* State of the LZMS compressor. */
struct lzms_compressor {
/* Pointer to a buffer holding the preprocessed data to compress. */
/* Size of the data in @buffer. */
u32 window_size;
- /* Suffix array match-finder. */
- struct lz_sarray lz_sarray;
+ /* Lempel-Ziv match-finder. */
+ struct lz_mf *mf;
/* Temporary space to store found matches. */
- struct raw_match *matches;
+ struct lz_match *matches;
- /* Match-chooser. */
- struct lz_match_chooser mc;
+ /* Match-chooser data. */
+ 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;
+
/* Raw range encoder which outputs to the beginning of the compressed
* data buffer, proceeding forwards. */
struct lzms_range_encoder_raw rc;
s32 last_target_usages[65536];
};
+struct lzms_mc_pos_data {
+ u32 cost;
+#define MC_INFINITE_COST ((u32)~0UL)
+ union {
+ struct {
+ u32 link;
+ u32 match_offset;
+ } prev;
+ struct {
+ u32 link;
+ u32 match_offset;
+ } next;
+ };
+ struct lzms_adaptive_state {
+ struct lzms_lz_lru_queues lru;
+ u8 main_state;
+ u8 match_state;
+ u8 lz_match_state;
+ u8 lz_repeat_match_state[LZMS_NUM_RECENT_OFFSETS - 1];
+ } state;
+};
+
/* Initialize the output bitstream @os to write forwards to the specified
* compressed data buffer @out that is @out_limit 16-bit integers long. */
static void
lzms_end_encode_item(ctx, length);
}
-/* Fast heuristic cost evaluation to use in the inner loop of the match-finder.
- * Unlike lzms_get_lz_match_cost(), which does a true cost evaluation, this
- * simply prioritize matches based on their offset. */
-static input_idx_t
-lzms_lz_match_cost_fast(input_idx_t length, input_idx_t offset, const void *_lru)
-{
- const struct lzms_lz_lru_queues *lru = _lru;
-
- for (input_idx_t i = 0; i < LZMS_NUM_RECENT_OFFSETS; i++)
- if (offset == lru->recent_offsets[i])
- return i;
-
- return offset;
-}
-
#define LZMS_COST_SHIFT 5
/*#define LZMS_RC_COSTS_USE_FLOATING_POINT*/
}
static u32
-lzms_length_cost(const struct lzms_huffman_encoder *enc, u32 length)
+lzms_get_length_cost(const struct lzms_huffman_encoder *enc, u32 length)
{
u32 slot;
u32 num_extra_bits;
}
static u32
-lzms_get_matches(struct lzms_compressor *ctx,
- const struct lzms_adaptive_state *state,
- struct raw_match **matches_ret)
+lzms_get_matches(struct lzms_compressor *ctx, struct lz_match **matches_ret)
{
*matches_ret = ctx->matches;
- return lz_sarray_get_matches(&ctx->lz_sarray,
- ctx->matches,
- lzms_lz_match_cost_fast,
- &state->lru);
+ return lz_mf_get_matches(ctx->mf, ctx->matches);
}
static void
-lzms_skip_bytes(struct lzms_compressor *ctx, input_idx_t n)
+lzms_skip_bytes(struct lzms_compressor *ctx, u32 n)
{
- while (n--)
- lz_sarray_skip_position(&ctx->lz_sarray);
+ lz_mf_skip_positions(ctx->mf, n);
}
static u32
-lzms_get_prev_literal_cost(struct lzms_compressor *ctx,
- struct lzms_adaptive_state *state)
+lzms_get_literal_cost(struct lzms_compressor *ctx,
+ struct lzms_adaptive_state *state, u8 literal)
{
- u8 literal = ctx->window[lz_sarray_get_pos(&ctx->lz_sarray) - 1];
u32 cost = 0;
state->lru.upcoming_offset = 0;
}
static u32
-lzms_get_lz_match_cost(struct lzms_compressor *ctx,
- struct lzms_adaptive_state *state,
- input_idx_t length, input_idx_t offset)
+lzms_get_lz_match_cost_nolen(struct lzms_compressor *ctx,
+ struct lzms_adaptive_state *state, u32 offset)
{
u32 cost = 0;
int recent_offset_idx;
state->lru.recent_offsets[i] = state->lru.recent_offsets[i + 1];
}
- cost += lzms_length_cost(&ctx->length_encoder, length);
state->lru.upcoming_offset = offset;
lzms_update_lz_lru_queues(&state->lru);
return cost;
}
-static struct raw_match
-lzms_get_near_optimal_match(struct lzms_compressor *ctx)
+static u32
+lzms_get_lz_match_cost(struct lzms_compressor *ctx,
+ struct lzms_adaptive_state *state,
+ u32 length, u32 offset)
+{
+ return lzms_get_lz_match_cost_nolen(ctx, state, offset) +
+ lzms_get_length_cost(&ctx->length_encoder, length);
+}
+
+static inline u32
+lzms_repsearch(const u8 * const strptr, const u32 bytes_remaining,
+ const struct lzms_lz_lru_queues *queue, u32 *offset_ret)
+{
+ u32 len;
+ unsigned slot = 0;
+
+ len = lz_repsearch(strptr, bytes_remaining, UINT32_MAX,
+ queue->recent_offsets, LZMS_NUM_RECENT_OFFSETS, &slot);
+ *offset_ret = queue->recent_offsets[slot];
+ return len;
+}
+
+
+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;
+
+ ctx->optimum_end_idx = cur_pos;
+
+ saved_prev_link = ctx->optimum[cur_pos].prev.link;
+ saved_prev_match_offset = ctx->optimum[cur_pos].prev.match_offset;
+
+ do {
+ prev_link = saved_prev_link;
+ prev_match_offset = saved_prev_match_offset;
+
+ saved_prev_link = ctx->optimum[prev_link].prev.link;
+ saved_prev_match_offset = ctx->optimum[prev_link].prev.match_offset;
+
+ ctx->optimum[prev_link].next.link = cur_pos;
+ ctx->optimum[prev_link].next.match_offset = prev_match_offset;
+
+ cur_pos = prev_link;
+ } while (cur_pos != 0);
+
+ ctx->optimum_cur_idx = ctx->optimum[0].next.link;
+
+ return (struct lz_match)
+ { .len = ctx->optimum_cur_idx,
+ .offset = ctx->optimum[0].next.match_offset,
+ };
+}
+
+/* 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;
+ }
+
+ ctx->optimum_cur_idx = 0;
+ ctx->optimum_end_idx = 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->lru.lz, &longest_rep_offset);
+ } else {
+ longest_rep_len = 0;
+ }
+
+ 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,
+ };
+ }
+
+ num_matches = lzms_get_matches(ctx, &matches);
+
+ 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;
+ }
+
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;
- return lz_get_near_optimal_match(&ctx->mc,
- lzms_get_matches,
- lzms_skip_bytes,
- lzms_get_prev_literal_cost,
- lzms_get_lz_match_cost,
- ctx,
- &initial_state);
+ initial_state.lz_repeat_match_state[i] = ctx->lz_repeat_match_range_encoders[i].state;
+
+ 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;
+
+ for (u32 i = 0, len = 2; i < num_matches; i++) {
+ u32 offset = matches[i].offset;
+ struct lzms_adaptive_state state;
+ u32 position_cost;
+
+ state = initial_state;
+ position_cost = 0;
+ position_cost += lzms_get_lz_match_cost_nolen(ctx, &state, offset);
+
+ do {
+ u32 cost;
+
+ cost = position_cost;
+ cost += lzms_get_length_cost(&ctx->length_encoder, 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;
+ }
+ }
+
+ cur_pos = 0;
+ for (;;) {
+ u32 cost;
+ struct lzms_adaptive_state state;
+
+ cur_pos++;
+
+ if (cur_pos == end_pos || cur_pos == ctx->params.optim_array_length)
+ return lzms_match_chooser_reverse_list(ctx, cur_pos);
+
+ 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;
+ }
+
+ if (longest_rep_len >= ctx->params.nice_match_length) {
+ match = lzms_match_chooser_reverse_list(ctx, cur_pos);
+
+ 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;
+
+ lzms_skip_bytes(ctx, longest_rep_len);
+
+ return match;
+ }
+
+ num_matches = lzms_get_matches(ctx, &matches);
+
+ 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);
+
+ 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;
+
+ lzms_skip_bytes(ctx, longest_len - 1);
+
+ return match;
+ }
+ } else {
+ longest_len = 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;
+ }
+
+ for (u32 i = 0, len = 2; i < num_matches; i++) {
+ u32 offset = matches[i].offset;
+ struct lzms_adaptive_state state;
+ u32 position_cost;
+
+ state = ctx->optimum[cur_pos].state;
+ position_cost = ctx->optimum[cur_pos].cost;
+ position_cost += lzms_get_lz_match_cost_nolen(ctx, &state, offset);
+
+ do {
+ u32 cost;
+
+ cost = position_cost;
+ cost += lzms_get_length_cost(&ctx->length_encoder, len);
+
+ 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 (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;
+ }
+ }
+ }
}
/*
*
* Notes:
*
- * - This uses near-optimal LZ parsing backed by a suffix-array match-finder.
- * More details can be found in the corresponding files (lz_optimal.h,
- * lz_sarray.{h,c}).
- *
- * - This does not output any delta matches. It would take a specialized
- * algorithm to find them, then more code in lz_optimal.h and here to handle
- * evaluating and outputting them.
+ * - 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 (from the
- * `struct wimlib_lzms_compressor_params') bytes of data. This introduces a
- * source of inaccuracy because the probabilities and Huffman codes can change
- * over this part of the data.
+ * 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 raw_match match;
+ struct lz_match item;
- /* Load window into suffix array match-finder. */
- lz_sarray_load_window(&ctx->lz_sarray, ctx->window, ctx->window_size);
+ /* Load window into the match-finder. */
+ lz_mf_load_window(ctx->mf, ctx->window, ctx->window_size);
/* Reset the match-chooser. */
- lz_match_chooser_begin(&ctx->mc);
+ ctx->optimum_cur_idx = 0;
+ ctx->optimum_end_idx = 0;
while (ctx->cur_window_pos != ctx->window_size) {
- match = lzms_get_near_optimal_match(ctx);
- if (match.len <= 1)
+ 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, match.len, match.offset);
+ lzms_encode_lz_match(ctx, item.len, item.offset);
}
}
return compressed_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;
+}
+
+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;
+ 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;
+ mf_params->nice_match_len = lzms_params->nice_match_length;
+}
+
+static void
+lzms_free_compressor(void *_ctx);
+
+static u64
+lzms_get_needed_memory(size_t max_block_size, unsigned int compression_level)
+{
+ struct lzms_compressor_params params;
+ u64 size = 0;
+
+ if (max_block_size >= INT32_MAX)
+ return 0;
+
+ lzms_build_params(compression_level, ¶ms);
+
+ size += sizeof(struct lzms_compressor);
+ 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);
+ size += (params.optim_array_length + params.nice_match_length) *
+ sizeof(struct lzms_mc_pos_data);
+
+ return size;
+}
+
+static int
+lzms_create_compressor(size_t max_block_size, unsigned int compression_level,
+ void **ctx_ret)
+{
+ struct lzms_compressor *ctx;
+ struct lzms_compressor_params params;
+ struct lz_mf_params mf_params;
+
+ if (max_block_size >= INT32_MAX)
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ lzms_build_params(compression_level, ¶ms);
+ lzms_build_mf_params(¶ms, max_block_size, &mf_params);
+ if (!lz_mf_params_valid(&mf_params))
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ ctx = CALLOC(1, sizeof(struct lzms_compressor));
+ if (!ctx)
+ goto oom;
+
+ ctx->params = params;
+ ctx->max_block_size = max_block_size;
+
+ ctx->window = MALLOC(max_block_size);
+ if (!ctx->window)
+ goto oom;
+
+ ctx->mf = lz_mf_alloc(&mf_params);
+ if (!ctx->mf)
+ goto oom;
+
+ ctx->matches = MALLOC(params.max_search_depth * sizeof(struct lz_match));
+ if (!ctx->matches)
+ goto oom;
+
+ ctx->optimum = MALLOC((params.optim_array_length +
+ params.nice_match_length) *
+ sizeof(struct lzms_mc_pos_data));
+ if (!ctx->optimum)
+ goto oom;
+
+ /* 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;
+ return 0;
+
+oom:
+ lzms_free_compressor(ctx);
+ 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)
LZMS_DEBUG("uncompressed_size=%zu, compressed_size_avail=%zu",
uncompressed_size, compressed_size_avail);
- /* Make sure the uncompressed size is compatible with this compressor.
- */
- if (uncompressed_size > ctx->max_block_size) {
- LZMS_DEBUG("Can't compress %zu bytes: LZMS context "
- "only supports %u bytes",
- uncompressed_size, ctx->max_block_size);
- return 0;
- }
-
/* Don't bother compressing extremely small inputs. */
if (uncompressed_size < 4) {
LZMS_DEBUG("Input too small to bother compressing.");
LZMS_DEBUG("Compressed %zu => %zu bytes",
uncompressed_size, compressed_size);
-#if defined(ENABLE_VERIFY_COMPRESSION) || defined(ENABLE_LZMS_DEBUG)
- /* Verify that we really get the same thing back when decompressing. */
- {
- struct wimlib_decompressor *decompressor;
-
- LZMS_DEBUG("Verifying LZMS compression.");
-
- if (0 == wimlib_create_decompressor(WIMLIB_COMPRESSION_TYPE_LZMS,
- ctx->max_block_size,
- NULL,
- &decompressor))
- {
- int ret;
- ret = wimlib_decompress(compressed_data,
- compressed_size,
- ctx->window,
- uncompressed_size,
- decompressor);
- wimlib_free_decompressor(decompressor);
-
- if (ret) {
- ERROR("Failed to decompress data we "
- "compressed using LZMS algorithm");
- wimlib_assert(0);
- return 0;
- }
- if (memcmp(uncompressed_data, ctx->window,
- uncompressed_size))
- {
- ERROR("Data we compressed using LZMS algorithm "
- "didn't decompress to original");
- wimlib_assert(0);
- return 0;
- }
- } else {
- WARNING("Failed to create decompressor for "
- "data verification!");
- }
- }
-#endif /* ENABLE_LZMS_DEBUG || ENABLE_VERIFY_COMPRESSION */
-
return compressed_size;
}
if (ctx) {
FREE(ctx->window);
+ lz_mf_free(ctx->mf);
FREE(ctx->matches);
- lz_sarray_destroy(&ctx->lz_sarray);
- lz_match_chooser_destroy(&ctx->mc);
+ FREE(ctx->optimum);
FREE(ctx);
}
}
-static const struct wimlib_lzms_compressor_params lzms_default = {
- .hdr = {
- .size = sizeof(struct wimlib_lzms_compressor_params),
- },
- .min_match_length = 2,
- .max_match_length = UINT32_MAX,
- .nice_match_length = 32,
- .max_search_depth = 50,
- .max_matches_per_pos = 3,
- .optim_array_length = 1024,
-};
-
-static bool
-lzms_params_valid(const struct wimlib_compressor_params_header *);
-
-static const struct wimlib_lzms_compressor_params *
-lzms_get_params(const struct wimlib_compressor_params_header *_params)
-{
- const struct wimlib_lzms_compressor_params *params =
- (const struct wimlib_lzms_compressor_params*)_params;
-
- if (params == NULL)
- params = &lzms_default;
-
- LZMS_ASSERT(lzms_params_valid(¶ms->hdr));
-
- return params;
-}
-
-static int
-lzms_create_compressor(size_t max_block_size,
- const struct wimlib_compressor_params_header *_params,
- void **ctx_ret)
-{
- struct lzms_compressor *ctx;
- const struct wimlib_lzms_compressor_params *params = lzms_get_params(_params);
-
- if (max_block_size == 0 || max_block_size >= INT32_MAX) {
- LZMS_DEBUG("Invalid max_block_size (%u)", max_block_size);
- return WIMLIB_ERR_INVALID_PARAM;
- }
-
- ctx = CALLOC(1, sizeof(struct lzms_compressor));
- if (ctx == NULL)
- goto oom;
-
- ctx->window = MALLOC(max_block_size);
- if (ctx->window == NULL)
- goto oom;
-
- ctx->matches = MALLOC(min(params->max_match_length -
- params->min_match_length + 1,
- params->max_matches_per_pos) *
- sizeof(ctx->matches[0]));
- if (ctx->matches == NULL)
- goto oom;
-
- if (!lz_sarray_init(&ctx->lz_sarray, max_block_size,
- params->min_match_length,
- min(params->max_match_length, LZ_SARRAY_LEN_MAX),
- params->max_search_depth,
- params->max_matches_per_pos))
- goto oom;
-
- if (!lz_match_chooser_init(&ctx->mc,
- params->optim_array_length,
- params->nice_match_length,
- params->max_match_length))
- goto oom;
-
- /* 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->max_block_size = max_block_size;
-
- *ctx_ret = ctx;
- return 0;
-
-oom:
- lzms_free_compressor(ctx);
- return WIMLIB_ERR_NOMEM;
-}
-
-static u64
-lzms_get_needed_memory(size_t max_block_size,
- const struct wimlib_compressor_params_header *_params)
-{
- const struct wimlib_lzms_compressor_params *params = lzms_get_params(_params);
-
- u64 size = 0;
-
- size += max_block_size;
- size += sizeof(struct lzms_compressor);
- size += lz_sarray_get_needed_memory(max_block_size);
- size += lz_match_chooser_get_needed_memory(params->optim_array_length,
- params->nice_match_length,
- params->max_match_length);
- size += min(params->max_match_length -
- params->min_match_length + 1,
- params->max_matches_per_pos) *
- sizeof(((struct lzms_compressor*)0)->matches[0]);
- return size;
-}
-
-static bool
-lzms_params_valid(const struct wimlib_compressor_params_header *_params)
-{
- const struct wimlib_lzms_compressor_params *params =
- (const struct wimlib_lzms_compressor_params*)_params;
-
- if (params->hdr.size != sizeof(*params) ||
- params->max_match_length < params->min_match_length ||
- params->min_match_length < 2 ||
- params->optim_array_length == 0 ||
- min(params->max_match_length, params->nice_match_length) > 65536)
- return false;
-
- return true;
-}
-
const struct compressor_ops lzms_compressor_ops = {
- .params_valid = lzms_params_valid,
.get_needed_memory = lzms_get_needed_memory,
.create_compressor = lzms_create_compressor,
.compress = lzms_compress,