1 ///////////////////////////////////////////////////////////////////////////////
3 /// \file lzma2_encoder.c
4 /// \brief LZMA2 encoder
6 // Authors: Igor Pavlov
9 // This file has been put into the public domain.
10 // You can do whatever you want with this file.
12 ///////////////////////////////////////////////////////////////////////////////
14 #include "lz_encoder.h"
15 #include "lzma_encoder.h"
17 #include "lzma2_encoder.h"
25 SEQ_UNCOMPRESSED_HEADER
,
26 SEQ_UNCOMPRESSED_COPY
,
32 /// LZMA options currently in use.
33 lzma_options_lzma opt_cur
;
36 bool need_state_reset
;
37 bool need_dictionary_reset
;
39 /// Uncompressed size of a chunk
40 size_t uncompressed_size
;
42 /// Compressed size of a chunk (excluding headers); this is also used
43 /// to indicate the end of buf[] in SEQ_LZMA_COPY.
44 size_t compressed_size
;
46 /// Read position in buf[]
49 /// Buffer to hold the chunk header and LZMA compressed data
50 uint8_t buf
[LZMA2_HEADER_MAX
+ LZMA2_CHUNK_MAX
];
55 lzma2_header_lzma(lzma_lzma2_coder
*coder
)
57 assert(coder
->uncompressed_size
> 0);
58 assert(coder
->uncompressed_size
<= LZMA2_UNCOMPRESSED_MAX
);
59 assert(coder
->compressed_size
> 0);
60 assert(coder
->compressed_size
<= LZMA2_CHUNK_MAX
);
64 if (coder
->need_properties
) {
67 if (coder
->need_dictionary_reset
)
68 coder
->buf
[pos
] = 0x80 + (3 << 5);
70 coder
->buf
[pos
] = 0x80 + (2 << 5);
74 if (coder
->need_state_reset
)
75 coder
->buf
[pos
] = 0x80 + (1 << 5);
77 coder
->buf
[pos
] = 0x80;
80 // Set the start position for copying.
84 size_t size
= coder
->uncompressed_size
- 1;
85 coder
->buf
[pos
++] += size
>> 16;
86 coder
->buf
[pos
++] = (size
>> 8) & 0xFF;
87 coder
->buf
[pos
++] = size
& 0xFF;
90 size
= coder
->compressed_size
- 1;
91 coder
->buf
[pos
++] = size
>> 8;
92 coder
->buf
[pos
++] = size
& 0xFF;
94 // Properties, if needed
95 if (coder
->need_properties
)
96 lzma_lzma_lclppb_encode(&coder
->opt_cur
, coder
->buf
+ pos
);
98 coder
->need_properties
= false;
99 coder
->need_state_reset
= false;
100 coder
->need_dictionary_reset
= false;
102 // The copying code uses coder->compressed_size to indicate the end
103 // of coder->buf[], so we need add the maximum size of the header here.
104 coder
->compressed_size
+= LZMA2_HEADER_MAX
;
111 lzma2_header_uncompressed(lzma_lzma2_coder
*coder
)
113 assert(coder
->uncompressed_size
> 0);
114 assert(coder
->uncompressed_size
<= LZMA2_CHUNK_MAX
);
116 // If this is the first chunk, we need to include dictionary
118 if (coder
->need_dictionary_reset
)
123 coder
->need_dictionary_reset
= false;
126 coder
->buf
[1] = (coder
->uncompressed_size
- 1) >> 8;
127 coder
->buf
[2] = (coder
->uncompressed_size
- 1) & 0xFF;
129 // Set the start position for copying.
136 lzma2_encode(void *coder_ptr
, lzma_mf
*restrict mf
,
137 uint8_t *restrict out
, size_t *restrict out_pos
,
140 lzma_lzma2_coder
*restrict coder
= coder_ptr
;
142 while (*out_pos
< out_size
)
143 switch (coder
->sequence
) {
145 // If there's no input left and we are flushing or finishing,
146 // don't start a new chunk.
147 if (mf_unencoded(mf
) == 0) {
148 // Write end of payload marker if finishing.
149 if (mf
->action
== LZMA_FINISH
)
150 out
[(*out_pos
)++] = 0;
152 return mf
->action
== LZMA_RUN
153 ? LZMA_OK
: LZMA_STREAM_END
;
156 if (coder
->need_state_reset
)
157 return_if_error(lzma_lzma_encoder_reset(
158 coder
->lzma
, &coder
->opt_cur
));
160 coder
->uncompressed_size
= 0;
161 coder
->compressed_size
= 0;
162 coder
->sequence
= SEQ_LZMA_ENCODE
;
166 case SEQ_LZMA_ENCODE
: {
167 // Calculate how much more uncompressed data this chunk
169 const uint32_t left
= LZMA2_UNCOMPRESSED_MAX
170 - coder
->uncompressed_size
;
173 if (left
< mf
->match_len_max
) {
174 // Must flush immediately since the next LZMA symbol
175 // could make the uncompressed size of the chunk too
179 // Calculate maximum read_limit that is OK from point
180 // of view of LZMA2 chunk size.
181 limit
= mf
->read_pos
- mf
->read_ahead
182 + left
- mf
->match_len_max
;
185 // Save the start position so that we can update
186 // coder->uncompressed_size.
187 const uint32_t read_start
= mf
->read_pos
- mf
->read_ahead
;
189 // Call the LZMA encoder until the chunk is finished.
190 const lzma_ret ret
= lzma_lzma_encode(coder
->lzma
, mf
,
191 coder
->buf
+ LZMA2_HEADER_MAX
,
192 &coder
->compressed_size
,
193 LZMA2_CHUNK_MAX
, limit
);
195 coder
->uncompressed_size
+= mf
->read_pos
- mf
->read_ahead
198 assert(coder
->compressed_size
<= LZMA2_CHUNK_MAX
);
199 assert(coder
->uncompressed_size
<= LZMA2_UNCOMPRESSED_MAX
);
201 if (ret
!= LZMA_STREAM_END
)
204 // See if the chunk compressed. If it didn't, we encode it
205 // as uncompressed chunk. This saves a few bytes of space
206 // and makes decoding faster.
207 if (coder
->compressed_size
>= coder
->uncompressed_size
) {
208 coder
->uncompressed_size
+= mf
->read_ahead
;
209 assert(coder
->uncompressed_size
210 <= LZMA2_UNCOMPRESSED_MAX
);
212 lzma2_header_uncompressed(coder
);
213 coder
->need_state_reset
= true;
214 coder
->sequence
= SEQ_UNCOMPRESSED_HEADER
;
218 // The chunk did compress at least by one byte, so we store
219 // the chunk as LZMA.
220 lzma2_header_lzma(coder
);
222 coder
->sequence
= SEQ_LZMA_COPY
;
228 // Copy the compressed chunk along its headers to the
230 lzma_bufcpy(coder
->buf
, &coder
->buf_pos
,
231 coder
->compressed_size
,
232 out
, out_pos
, out_size
);
233 if (coder
->buf_pos
!= coder
->compressed_size
)
236 coder
->sequence
= SEQ_INIT
;
239 case SEQ_UNCOMPRESSED_HEADER
:
240 // Copy the three-byte header to indicate uncompressed chunk.
241 lzma_bufcpy(coder
->buf
, &coder
->buf_pos
,
242 LZMA2_HEADER_UNCOMPRESSED
,
243 out
, out_pos
, out_size
);
244 if (coder
->buf_pos
!= LZMA2_HEADER_UNCOMPRESSED
)
247 coder
->sequence
= SEQ_UNCOMPRESSED_COPY
;
251 case SEQ_UNCOMPRESSED_COPY
:
252 // Copy the uncompressed data as is from the dictionary
253 // to the output buffer.
254 mf_read(mf
, out
, out_pos
, out_size
, &coder
->uncompressed_size
);
255 if (coder
->uncompressed_size
!= 0)
258 coder
->sequence
= SEQ_INIT
;
267 lzma2_encoder_end(void *coder_ptr
, const lzma_allocator
*allocator
)
269 lzma_lzma2_coder
*coder
= coder_ptr
;
270 lzma_free(coder
->lzma
, allocator
);
271 lzma_free(coder
, allocator
);
277 lzma2_encoder_options_update(void *coder_ptr
, const lzma_filter
*filter
)
279 lzma_lzma2_coder
*coder
= coder_ptr
;
281 // New options can be set only when there is no incomplete chunk.
282 // This is the case at the beginning of the raw stream and right
283 // after LZMA_SYNC_FLUSH.
284 if (filter
->options
== NULL
|| coder
->sequence
!= SEQ_INIT
)
285 return LZMA_PROG_ERROR
;
287 // Look if there are new options. At least for now,
288 // only lc/lp/pb can be changed.
289 const lzma_options_lzma
*opt
= filter
->options
;
290 if (coder
->opt_cur
.lc
!= opt
->lc
|| coder
->opt_cur
.lp
!= opt
->lp
291 || coder
->opt_cur
.pb
!= opt
->pb
) {
292 // Validate the options.
293 if (opt
->lc
> LZMA_LCLP_MAX
|| opt
->lp
> LZMA_LCLP_MAX
294 || opt
->lc
+ opt
->lp
> LZMA_LCLP_MAX
295 || opt
->pb
> LZMA_PB_MAX
)
296 return LZMA_OPTIONS_ERROR
;
298 // The new options will be used when the encoder starts
299 // a new LZMA2 chunk.
300 coder
->opt_cur
.lc
= opt
->lc
;
301 coder
->opt_cur
.lp
= opt
->lp
;
302 coder
->opt_cur
.pb
= opt
->pb
;
303 coder
->need_properties
= true;
304 coder
->need_state_reset
= true;
312 lzma2_encoder_init(lzma_lz_encoder
*lz
, const lzma_allocator
*allocator
,
313 lzma_vli id
lzma_attribute((__unused__
)), const void *options
,
314 lzma_lz_options
*lz_options
)
317 return LZMA_PROG_ERROR
;
319 lzma_lzma2_coder
*coder
= lz
->coder
;
321 coder
= lzma_alloc(sizeof(lzma_lzma2_coder
), allocator
);
323 return LZMA_MEM_ERROR
;
326 lz
->code
= &lzma2_encode
;
327 lz
->end
= &lzma2_encoder_end
;
328 lz
->options_update
= &lzma2_encoder_options_update
;
333 coder
->opt_cur
= *(const lzma_options_lzma
*)(options
);
335 coder
->sequence
= SEQ_INIT
;
336 coder
->need_properties
= true;
337 coder
->need_state_reset
= false;
338 coder
->need_dictionary_reset
339 = coder
->opt_cur
.preset_dict
== NULL
340 || coder
->opt_cur
.preset_dict_size
== 0;
342 // Initialize LZMA encoder
343 return_if_error(lzma_lzma_encoder_create(&coder
->lzma
, allocator
,
344 LZMA_FILTER_LZMA2
, &coder
->opt_cur
, lz_options
));
346 // Make sure that we will always have enough history available in
347 // case we need to use uncompressed chunks. They are used when the
348 // compressed size of a chunk is not smaller than the uncompressed
349 // size, so we need to have at least LZMA2_COMPRESSED_MAX bytes
350 // history available.
351 if (lz_options
->before_size
+ lz_options
->dict_size
< LZMA2_CHUNK_MAX
)
352 lz_options
->before_size
353 = LZMA2_CHUNK_MAX
- lz_options
->dict_size
;
360 lzma_lzma2_encoder_init(lzma_next_coder
*next
, const lzma_allocator
*allocator
,
361 const lzma_filter_info
*filters
)
363 return lzma_lz_encoder_init(
364 next
, allocator
, filters
, &lzma2_encoder_init
);
369 lzma_lzma2_encoder_memusage(const void *options
)
371 const uint64_t lzma_mem
= lzma_lzma_encoder_memusage(options
);
372 if (lzma_mem
== UINT64_MAX
)
375 return sizeof(lzma_lzma2_coder
) + lzma_mem
;
380 lzma_lzma2_props_encode(const void *options
, uint8_t *out
)
383 return LZMA_PROG_ERROR
;
385 const lzma_options_lzma
*const opt
= options
;
386 uint32_t d
= my_max(opt
->dict_size
, LZMA_DICT_SIZE_MIN
);
388 // Round up to the next 2^n - 1 or 2^n + 2^(n - 1) - 1 depending
389 // on which one is the next:
397 // Get the highest two bits using the proper encoding:
401 out
[0] = get_dist_slot(d
+ 1) - 24;
408 lzma_lzma2_block_size(const void *options
)
410 const lzma_options_lzma
*const opt
= options
;
412 // Use at least 1 MiB to keep compression ratio better.
413 return my_max((uint64_t)(opt
->dict_size
) * 3, UINT64_C(1) << 20);