aboutsummaryrefslogtreecommitdiff
path: root/src/liblzma/common/stream_encoder_mt.c
blob: cd7b291dec9cf135d53fccf119cba27749bb3e04 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
///////////////////////////////////////////////////////////////////////////////
//
/// \file       stream_encoder_mt.c
/// \brief      Multithreaded .xz Stream encoder
//
//  Author:     Lasse Collin
//
//  This file has been put into the public domain.
//  You can do whatever you want with this file.
//
///////////////////////////////////////////////////////////////////////////////

#include "filter_encoder.h"
#include "easy_preset.h"
#include "block_encoder.h"
#include "block_buffer_encoder.h"
#include "index_encoder.h"
#include "outqueue.h"


/// Maximum supported block size. This makes it simpler to prevent integer
/// overflows if we are given unusually large block size.
#define BLOCK_SIZE_MAX (UINT64_MAX / LZMA_THREADS_MAX)


typedef enum {
	/// Waiting for work.
	THR_IDLE,

	/// Encoding is in progress.
	THR_RUN,

	/// Encoding is in progress but no more input data will
	/// be read.
	THR_FINISH,

	/// The main thread wants the thread to stop whatever it was doing
	/// but not exit.
	THR_STOP,

	/// The main thread wants the thread to exit. We could use
	/// cancellation but since there's stopped anyway, this is lazier.
	THR_EXIT,

} worker_state;


typedef struct worker_thread_s worker_thread;
struct worker_thread_s {
	worker_state state;

	/// Input buffer of coder->block_size bytes. The main thread will
	/// put new input into this and update in_size accordingly. Once
	/// no more input is coming, state will be set to THR_FINISH.
	uint8_t *in;

	/// Amount of data available in the input buffer. This is modified
	/// only by the main thread.
	size_t in_size;

	/// Output buffer for this thread. This is set by the main
	/// thread every time a new Block is started with this thread
	/// structure.
	lzma_outbuf *outbuf;

	/// Pointer to the main structure is needed when putting this
	/// thread back to the stack of free threads.
	lzma_coder *coder;

	/// The allocator is set by the main thread. Since a copy of the
	/// pointer is kept here, the application must not change the
	/// allocator before calling lzma_end().
	const lzma_allocator *allocator;

	/// Amount of uncompressed data that has already been compressed.
	uint64_t progress_in;

	/// Amount of compressed data that is ready.
	uint64_t progress_out;

	/// Block encoder
	lzma_next_coder block_encoder;

	/// Compression options for this Block
	lzma_block block_options;

	/// Next structure in the stack of free worker threads.
	worker_thread *next;

	pthread_mutex_t mutex;
	pthread_cond_t cond;

	/// The ID of this thread is used to join the thread
	/// when it's not needed anymore.
	pthread_t thread_id;
};


struct lzma_coder_s {
	enum {
		SEQ_STREAM_HEADER,
		SEQ_BLOCK,
		SEQ_INDEX,
		SEQ_STREAM_FOOTER,
	} sequence;

	/// Start a new Block every block_size bytes of input unless
	/// LZMA_FULL_FLUSH or LZMA_FULL_BARRIER is used earlier.
	size_t block_size;

	/// The filter chain currently in use
	lzma_filter filters[LZMA_FILTERS_MAX + 1];


	/// Index to hold sizes of the Blocks
	lzma_index *index;

	/// Index encoder
	lzma_next_coder index_encoder;


	/// Stream Flags for encoding the Stream Header and Stream Footer.
	lzma_stream_flags stream_flags;

	/// Buffer to hold Stream Header and Stream Footer.
	uint8_t header[LZMA_STREAM_HEADER_SIZE];

	/// Read position in header[]
	size_t header_pos;


	/// Output buffer queue for compressed data
	lzma_outq outq;


	/// True if wait_max is used.
	bool has_timeout;

	/// Maximum wait time if cannot use all the input and cannot
	/// fill the output buffer.
	struct timespec wait_max;


	/// Error code from a worker thread
	lzma_ret thread_error;

	/// Array of allocated thread-specific structures
	worker_thread *threads;

	/// Number of structures in "threads" above. This is also the
	/// number of threads that will be created at maximum.
	uint32_t threads_max;

	/// Number of thread structures that have been initialized, and
	/// thus the number of worker threads actually created so far.
	uint32_t threads_initialized;

	/// Stack of free threads. When a thread finishes, it puts itself
	/// back into this stack. This starts as empty because threads
	/// are created only when actually needed.
	worker_thread *threads_free;

	/// The most recent worker thread to which the main thread writes
	/// the new input from the application.
	worker_thread *thr;


	/// Amount of uncompressed data in Blocks that have already
	/// been finished.
	uint64_t progress_in;

	/// Amount of compressed data in Stream Header + Blocks that
	/// have already been finished.
	uint64_t progress_out;


	pthread_mutex_t mutex;
	mythread_cond cond;
};


/// Tell the main thread that something has gone wrong.
static void
worker_error(worker_thread *thr, lzma_ret ret)
{
	assert(ret != LZMA_OK);
	assert(ret != LZMA_STREAM_END);

	mythread_sync(thr->coder->mutex) {
		if (thr->coder->thread_error == LZMA_OK)
			thr->coder->thread_error = ret;

		mythread_cond_signal(&thr->coder->cond);
	}

	return;
}


static worker_state
worker_encode(worker_thread *thr, worker_state state)
{
	assert(thr->progress_in == 0);
	assert(thr->progress_out == 0);

	// Set the Block options.
	thr->block_options = (lzma_block){
		.version = 0,
		.check = thr->coder->stream_flags.check,
		.compressed_size = thr->coder->outq.buf_size_max,
		.uncompressed_size = thr->coder->block_size,

		// TODO: To allow changing the filter chain, the filters
		// array must be copied to each worker_thread.
		.filters = thr->coder->filters,
	};

	// Calculate maximum size of the Block Header. This amount is
	// reserved in the beginning of the buffer so that Block Header
	// along with Compressed Size and Uncompressed Size can be
	// written there.
	lzma_ret ret = lzma_block_header_size(&thr->block_options);
	if (ret != LZMA_OK) {
		worker_error(thr, ret);
		return THR_STOP;
	}

	// Initialize the Block encoder.
	ret = lzma_block_encoder_init(&thr->block_encoder,
			thr->allocator, &thr->block_options);
	if (ret != LZMA_OK) {
		worker_error(thr, ret);
		return THR_STOP;
	}

	size_t in_pos = 0;
	size_t in_size = 0;

	thr->outbuf->size = thr->block_options.header_size;
	const size_t out_size = thr->coder->outq.buf_size_max;

	do {
		mythread_sync(thr->mutex) {
			// Store in_pos and out_pos into *thr so that
			// an application may read them via
			// lzma_get_progress() to get progress information.
			//
			// NOTE: These aren't updated when the encoding
			// finishes. Instead, the final values are taken
			// later from thr->outbuf.
			thr->progress_in = in_pos;
			thr->progress_out = thr->outbuf->size;

			while (in_size == thr->in_size
					&& thr->state == THR_RUN)
				pthread_cond_wait(&thr->cond, &thr->mutex);

			state = thr->state;
			in_size = thr->in_size;
		}

		// Return if we were asked to stop or exit.
		if (state >= THR_STOP)
			return state;

		lzma_action action = state == THR_FINISH
				? LZMA_FINISH : LZMA_RUN;

		// Limit the amount of input given to the Block encoder
		// at once. This way this thread can react fairly quickly
		// if the main thread wants us to stop or exit.
		static const size_t in_chunk_max = 16384;
		size_t in_limit = in_size;
		if (in_size - in_pos > in_chunk_max) {
			in_limit = in_pos + in_chunk_max;
			action = LZMA_RUN;
		}

		ret = thr->block_encoder.code(
				thr->block_encoder.coder, thr->allocator,
				thr->in, &in_pos, in_limit, thr->outbuf->buf,
				&thr->outbuf->size, out_size, action);
	} while (ret == LZMA_OK && thr->outbuf->size < out_size);

	switch (ret) {
	case LZMA_STREAM_END:
		assert(state == THR_FINISH);

		// Encode the Block Header. By doing it after
		// the compression, we can store the Compressed Size
		// and Uncompressed Size fields.
		ret = lzma_block_header_encode(&thr->block_options,
				thr->outbuf->buf);
		if (ret != LZMA_OK) {
			worker_error(thr, ret);
			return THR_STOP;
		}

		break;

	case LZMA_OK:
		// The data was incompressible. Encode it using uncompressed
		// LZMA2 chunks.
		//
		// First wait that we have gotten all the input.
		mythread_sync(thr->mutex) {
			while (thr->state == THR_RUN)
				pthread_cond_wait(&thr->cond, &thr->mutex);

			state = thr->state;
			in_size = thr->in_size;
		}

		if (state >= THR_STOP)
			return state;

		// Do the encoding. This takes care of the Block Header too.
		thr->outbuf->size = 0;
		ret = lzma_block_uncomp_encode(&thr->block_options,
				thr->in, in_size, thr->outbuf->buf,
				&thr->outbuf->size, out_size);

		// It shouldn't fail.
		if (ret != LZMA_OK) {
			worker_error(thr, LZMA_PROG_ERROR);
			return THR_STOP;
		}

		break;

	default:
		worker_error(thr, ret);
		return THR_STOP;
	}

	// Set the size information that will be read by the main thread
	// to write the Index field.
	thr->outbuf->unpadded_size
			= lzma_block_unpadded_size(&thr->block_options);
	assert(thr->outbuf->unpadded_size != 0);
	thr->outbuf->uncompressed_size = thr->block_options.uncompressed_size;

	return THR_FINISH;
}


static void *
worker_start(void *thr_ptr)
{
	worker_thread *thr = thr_ptr;
	worker_state state = THR_IDLE; // Init to silence a warning

	while (true) {
		// Wait for work.
		mythread_sync(thr->mutex) {
			while (true) {
				// The thread is already idle so if we are
				// requested to stop, just set the state.
				if (thr->state == THR_STOP) {
					thr->state = THR_IDLE;
					pthread_cond_signal(&thr->cond);
				}

				state = thr->state;
				if (state != THR_IDLE)
					break;

				pthread_cond_wait(&thr->cond, &thr->mutex);
			}
		}

		assert(state != THR_IDLE);
		assert(state != THR_STOP);

		if (state <= THR_FINISH)
			state = worker_encode(thr, state);

		if (state == THR_EXIT)
			break;

		// Mark the thread as idle unless the main thread has
		// told us to exit. Signal is needed for the case
		// where the main thread is waiting for the threads to stop.
		mythread_sync(thr->mutex) {
			if (thr->state != THR_EXIT) {
				thr->state = THR_IDLE;
				pthread_cond_signal(&thr->cond);
			}
		}

		mythread_sync(thr->coder->mutex) {
			// Mark the output buffer as finished if
			// no errors occurred.
			thr->outbuf->finished = state == THR_FINISH;

			// Update the main progress info.
			thr->coder->progress_in
					+= thr->outbuf->uncompressed_size;
			thr->coder->progress_out += thr->outbuf->size;
			thr->progress_in = 0;
			thr->progress_out = 0;

			// Return this thread to the stack of free threads.
			thr->next = thr->coder->threads_free;
			thr->coder->threads_free = thr;

			mythread_cond_signal(&thr->coder->cond);
		}
	}

	// Exiting, free the resources.
	pthread_mutex_destroy(&thr->mutex);
	pthread_cond_destroy(&thr->cond);

	lzma_next_end(&thr->block_encoder, thr->allocator);
	lzma_free(thr->in, thr->allocator);
	return NULL;
}


/// Make the threads stop but not exit. Optionally wait for them to stop.
static void
threads_stop(lzma_coder *coder, bool wait)
{
	// Tell the threads to stop.
	for (uint32_t i = 0; i < coder->threads_initialized; ++i) {
		mythread_sync(coder->threads[i].mutex) {
			coder->threads[i].state = THR_STOP;
			pthread_cond_signal(&coder->threads[i].cond);
		}
	}

	if (!wait)
		return;

	// Wait for the threads to settle in the idle state.
	for (uint32_t i = 0; i < coder->threads_initialized; ++i) {
		mythread_sync(coder->threads[i].mutex) {
			while (coder->threads[i].state != THR_IDLE)
				pthread_cond_wait(&coder->threads[i].cond,
						&coder->threads[i].mutex);
		}
	}

	return;
}


/// Stop the threads and free the resources associated with them.
/// Wait until the threads have exited.
static void
threads_end(lzma_coder *coder, const lzma_allocator *allocator)
{
	for (uint32_t i = 0; i < coder->threads_initialized; ++i) {
		mythread_sync(coder->threads[i].mutex) {
			coder->threads[i].state = THR_EXIT;
			pthread_cond_signal(&coder->threads[i].cond);
		}
	}

	for (uint32_t i = 0; i < coder->threads_initialized; ++i) {
		int ret = pthread_join(coder->threads[i].thread_id, NULL);
		assert(ret == 0);
		(void)ret;
	}

	lzma_free(coder->threads, allocator);
	return;
}


/// Initialize a new worker_thread structure and create a new thread.
static lzma_ret
initialize_new_thread(lzma_coder *coder, const lzma_allocator *allocator)
{
	worker_thread *thr = &coder->threads[coder->threads_initialized];

	thr->in = lzma_alloc(coder->block_size, allocator);
	if (thr->in == NULL)
		return LZMA_MEM_ERROR;

	if (pthread_mutex_init(&thr->mutex, NULL))
		goto error_mutex;

	if (pthread_cond_init(&thr->cond, NULL))
		goto error_cond;

	thr->state = THR_IDLE;
	thr->allocator = allocator;
	thr->coder = coder;
	thr->progress_in = 0;
	thr->progress_out = 0;
	thr->block_encoder = LZMA_NEXT_CODER_INIT;

	if (mythread_create(&thr->thread_id, &worker_start, thr))
		goto error_thread;

	++coder->threads_initialized;
	coder->thr = thr;

	return LZMA_OK;

error_thread:
	pthread_cond_destroy(&thr->cond);

error_cond:
	pthread_mutex_destroy(&thr->mutex);

error_mutex:
	lzma_free(thr->in, allocator);
	return LZMA_MEM_ERROR;
}


static lzma_ret
get_thread(lzma_coder *coder, const lzma_allocator *allocator)
{
	// If there are no free output subqueues, there is no
	// point to try getting a thread.
	if (!lzma_outq_has_buf(&coder->outq))
		return LZMA_OK;

	// If there is a free structure on the stack, use it.
	mythread_sync(coder->mutex) {
		if (coder->threads_free != NULL) {
			coder->thr = coder->threads_free;
			coder->threads_free = coder->threads_free->next;
		}
	}

	if (coder->thr == NULL) {
		// If there are no uninitialized structures left, return.
		if (coder->threads_initialized == coder->threads_max)
			return LZMA_OK;

		// Initialize a new thread.
		return_if_error(initialize_new_thread(coder, allocator));
	}

	// Reset the parts of the thread state that have to be done
	// in the main thread.
	mythread_sync(coder->thr->mutex) {
		coder->thr->state = THR_RUN;
		coder->thr->in_size = 0;
		coder->thr->outbuf = lzma_outq_get_buf(&coder->outq);
		pthread_cond_signal(&coder->thr->cond);
	}

	return LZMA_OK;
}


static lzma_ret
stream_encode_in(lzma_coder *coder, const lzma_allocator *allocator,
		const uint8_t *restrict in, size_t *restrict in_pos,
		size_t in_size, lzma_action action)
{
	while (*in_pos < in_size
			|| (coder->thr != NULL && action != LZMA_RUN)) {
		if (coder->thr == NULL) {
			// Get a new thread.
			const lzma_ret ret = get_thread(coder, allocator);
			if (coder->thr == NULL)
				return ret;
		}

		// Copy the input data to thread's buffer.
		size_t thr_in_size = coder->thr->in_size;
		lzma_bufcpy(in, in_pos, in_size, coder->thr->in,
				&thr_in_size, coder->block_size);

		// Tell the Block encoder to finish if
		//  - it has got block_size bytes of input; or
		//  - all input was used and LZMA_FINISH, LZMA_FULL_FLUSH,
		//    or LZMA_FULL_BARRIER was used.
		//
		// TODO: LZMA_SYNC_FLUSH and LZMA_SYNC_BARRIER.
		const bool finish = thr_in_size == coder->block_size
				|| (*in_pos == in_size && action != LZMA_RUN);

		bool block_error = false;

		mythread_sync(coder->thr->mutex) {
			if (coder->thr->state == THR_IDLE) {
				// Something has gone wrong with the Block
				// encoder. It has set coder->thread_error
				// which we will read a few lines later.
				block_error = true;
			} else {
				// Tell the Block encoder its new amount
				// of input and update the state if needed.
				coder->thr->in_size = thr_in_size;

				if (finish)
					coder->thr->state = THR_FINISH;

				pthread_cond_signal(&coder->thr->cond);
			}
		}

		if (block_error) {
			lzma_ret ret;

			mythread_sync(coder->mutex) {
				ret = coder->thread_error;
			}

			return ret;
		}

		if (finish)
			coder->thr = NULL;
	}

	return LZMA_OK;
}


/// Wait until more input can be consumed, more output can be read, or
/// an optional timeout is reached.
static bool
wait_for_work(lzma_coder *coder, struct timespec *wait_abs,
		bool *has_blocked, bool has_input)
{
	if (coder->has_timeout && !*has_blocked) {
		// Every time when stream_encode_mt() is called via
		// lzma_code(), *has_block starts as false. We set it
		// to true here and calculate the absolute time when
		// we must return if there's nothing to do.
		//
		// The idea of *has_blocked is to avoid unneeded calls
		// to mythread_cond_abstime(), which may do a syscall
		// depending on the operating system.
		*has_blocked = true;
		*wait_abs = coder->wait_max;
		mythread_cond_abstime(&coder->cond, wait_abs);
	}

	bool timed_out = false;

	mythread_sync(coder->mutex) {
		// There are four things that we wait. If one of them
		// becomes possible, we return.
		//  - If there is input left, we need to get a free
		//    worker thread and an output buffer for it.
		//  - Data ready to be read from the output queue.
		//  - A worker thread indicates an error.
		//  - Time out occurs.
		while ((!has_input || coder->threads_free == NULL
					|| !lzma_outq_has_buf(&coder->outq))
				&& !lzma_outq_is_readable(&coder->outq)
				&& coder->thread_error == LZMA_OK
				&& !timed_out) {
			if (coder->has_timeout)
				timed_out = mythread_cond_timedwait(
						&coder->cond, &coder->mutex,
						wait_abs) != 0;
			else
				mythread_cond_wait(&coder->cond,
						&coder->mutex);
		}
	}

	return timed_out;
}


static lzma_ret
stream_encode_mt(lzma_coder *coder, const lzma_allocator *allocator,
		const uint8_t *restrict in, size_t *restrict in_pos,
		size_t in_size, uint8_t *restrict out,
		size_t *restrict out_pos, size_t out_size, lzma_action action)
{
	switch (coder->sequence) {
	case SEQ_STREAM_HEADER:
		lzma_bufcpy(coder->header, &coder->header_pos,
				sizeof(coder->header),
				out, out_pos, out_size);
		if (coder->header_pos < sizeof(coder->header))
			return LZMA_OK;

		coder->header_pos = 0;
		coder->sequence = SEQ_BLOCK;

	// Fall through

	case SEQ_BLOCK: {
		// Initialized to silence warnings.
		lzma_vli unpadded_size = 0;
		lzma_vli uncompressed_size = 0;
		lzma_ret ret = LZMA_OK;

		// These are for wait_for_work().
		bool has_blocked = false;
		struct timespec wait_abs;

		while (true) {
			mythread_sync(coder->mutex) {
				// Check for Block encoder errors.
				ret = coder->thread_error;
				if (ret != LZMA_OK) {
					assert(ret != LZMA_STREAM_END);
					break;
				}

				// Try to read compressed data to out[].
				ret = lzma_outq_read(&coder->outq,
						out, out_pos, out_size,
						&unpadded_size,
						&uncompressed_size);
			}

			if (ret == LZMA_STREAM_END) {
				// End of Block. Add it to the Index.
				ret = lzma_index_append(coder->index,
						allocator, unpadded_size,
						uncompressed_size);

				// If we didn't fill the output buffer yet,
				// try to read more data. Maybe the next
				// outbuf has been finished already too.
				if (*out_pos < out_size)
					continue;
			}

			if (ret != LZMA_OK) {
				// coder->thread_error was set or
				// lzma_index_append() failed.
				threads_stop(coder, false);
				return ret;
			}

			// Check if the last Block was finished.
			if (action == LZMA_FINISH
					&& *in_pos == in_size
					&& lzma_outq_is_empty(
						&coder->outq))
				break;

			// Try to give uncompressed data to a worker thread.
			ret = stream_encode_in(coder, allocator,
					in, in_pos, in_size, action);
			if (ret != LZMA_OK) {
				threads_stop(coder, false);
				return ret;
			}

			// Return if
			//  - we have used all the input and expect to
			//    get more input; or
			//  - the output buffer has been filled.
			//
			// TODO: Support flushing.
			if ((*in_pos == in_size && action != LZMA_FINISH)
					|| *out_pos == out_size)
				return LZMA_OK;

			// Neither in nor out has been used completely.
			// Wait until there's something we can do.
			if (wait_for_work(coder, &wait_abs, &has_blocked,
					*in_pos < in_size))
				return LZMA_TIMED_OUT;
		}

		// All Blocks have been encoded and the threads have stopped.
		// Prepare to encode the Index field.
		return_if_error(lzma_index_encoder_init(
				&coder->index_encoder, allocator,
				coder->index));
		coder->sequence = SEQ_INDEX;

		// Update the progress info to take the Index and
		// Stream Footer into account. Those are very fast to encode
		// so in terms of progress information they can be thought
		// to be ready to be copied out.
		coder->progress_out += lzma_index_size(coder->index)
				+ LZMA_STREAM_HEADER_SIZE;
	}

	// Fall through

	case SEQ_INDEX: {
		// Call the Index encoder. It doesn't take any input, so
		// those pointers can be NULL.
		const lzma_ret ret = coder->index_encoder.code(
				coder->index_encoder.coder, allocator,
				NULL, NULL, 0,
				out, out_pos, out_size, LZMA_RUN);
		if (ret != LZMA_STREAM_END)
			return ret;

		// Encode the Stream Footer into coder->buffer.
		coder->stream_flags.backward_size
				= lzma_index_size(coder->index);
		if (lzma_stream_footer_encode(&coder->stream_flags,
				coder->header) != LZMA_OK)
			return LZMA_PROG_ERROR;

		coder->sequence = SEQ_STREAM_FOOTER;
	}

	// Fall through

	case SEQ_STREAM_FOOTER:
		lzma_bufcpy(coder->header, &coder->header_pos,
				sizeof(coder->header),
				out, out_pos, out_size);
		return coder->header_pos < sizeof(coder->header)
				? LZMA_OK : LZMA_STREAM_END;
	}

	assert(0);
	return LZMA_PROG_ERROR;
}


static void
stream_encoder_mt_end(lzma_coder *coder, const lzma_allocator *allocator)
{
	// Threads must be killed before the output queue can be freed.
	threads_end(coder, allocator);
	lzma_outq_end(&coder->outq, allocator);

	for (size_t i = 0; coder->filters[i].id != LZMA_VLI_UNKNOWN; ++i)
		lzma_free(coder->filters[i].options, allocator);

	lzma_next_end(&coder->index_encoder, allocator);
	lzma_index_end(coder->index, allocator);

	mythread_cond_destroy(&coder->cond);
	pthread_mutex_destroy(&coder->mutex);

	lzma_free(coder, allocator);
	return;
}


/// Options handling for lzma_stream_encoder_mt_init() and
/// lzma_stream_encoder_mt_memusage()
static lzma_ret
get_options(const lzma_mt *options, lzma_options_easy *opt_easy,
		const lzma_filter **filters, uint64_t *block_size,
		uint64_t *outbuf_size_max)
{
	// Validate some of the options.
	if (options == NULL)
		return LZMA_PROG_ERROR;

	if (options->flags != 0 || options->threads == 0
			|| options->threads > LZMA_THREADS_MAX)
		return LZMA_OPTIONS_ERROR;

	if (options->filters != NULL) {
		// Filter chain was given, use it as is.
		*filters = options->filters;
	} else {
		// Use a preset.
		if (lzma_easy_preset(opt_easy, options->preset))
			return LZMA_OPTIONS_ERROR;

		*filters = opt_easy->filters;
	}

	// Block size
	if (options->block_size > 0) {
		if (options->block_size > BLOCK_SIZE_MAX)
			return LZMA_OPTIONS_ERROR;

		*block_size = options->block_size;
	} else {
		// Determine the Block size from the filter chain.
		*block_size = lzma_mt_block_size(*filters);
		if (*block_size == 0)
			return LZMA_OPTIONS_ERROR;

		assert(*block_size <= BLOCK_SIZE_MAX);
	}

	// Calculate the maximum amount output that a single output buffer
	// may need to hold. This is the same as the maximum total size of
	// a Block.
	*outbuf_size_max = lzma_block_buffer_bound64(*block_size);
	if (*outbuf_size_max == 0)
		return LZMA_MEM_ERROR;

	return LZMA_OK;
}


static void
get_progress(lzma_coder *coder, uint64_t *progress_in, uint64_t *progress_out)
{
	// Lock coder->mutex to prevent finishing threads from moving their
	// progress info from the worker_thread structure to lzma_coder.
	mythread_sync(coder->mutex) {
		*progress_in = coder->progress_in;
		*progress_out = coder->progress_out;

		for (size_t i = 0; i < coder->threads_initialized; ++i) {
			mythread_sync(coder->threads[i].mutex) {
				*progress_in += coder->threads[i].progress_in;
				*progress_out += coder->threads[i]
						.progress_out;
			}
		}
	}

	return;
}


static lzma_ret
stream_encoder_mt_init(lzma_next_coder *next, const lzma_allocator *allocator,
		const lzma_mt *options)
{
	lzma_next_coder_init(&stream_encoder_mt_init, next, allocator);

	// Get the filter chain.
	lzma_options_easy easy;
	const lzma_filter *filters;
	uint64_t block_size;
	uint64_t outbuf_size_max;
	return_if_error(get_options(options, &easy, &filters,
			&block_size, &outbuf_size_max));

#if SIZE_MAX < UINT64_MAX
	if (block_size > SIZE_MAX)
		return LZMA_MEM_ERROR;
#endif

	// FIXME TODO: Validate the filter chain so that we can give
	// an error in this function instead of delaying it to the first
	// call to lzma_code().

	// Validate the Check ID.
	if ((unsigned int)(options->check) > LZMA_CHECK_ID_MAX)
		return LZMA_PROG_ERROR;

	if (!lzma_check_is_supported(options->check))
		return LZMA_UNSUPPORTED_CHECK;

	// Allocate and initialize the base structure if needed.
	if (next->coder == NULL) {
		next->coder = lzma_alloc(sizeof(lzma_coder), allocator);
		if (next->coder == NULL)
			return LZMA_MEM_ERROR;

		// For the mutex and condition variable initializations
		// the error handling has to be done here because
		// stream_encoder_mt_end() doesn't know if they have
		// already been initialized or not.
		if (pthread_mutex_init(&next->coder->mutex, NULL)) {
			lzma_free(next->coder, allocator);
			next->coder = NULL;
			return LZMA_MEM_ERROR;
		}

		if (mythread_cond_init(&next->coder->cond)) {
			pthread_mutex_destroy(&next->coder->mutex);
			lzma_free(next->coder, allocator);
			next->coder = NULL;
			return LZMA_MEM_ERROR;
		}

		next->code = &stream_encode_mt;
		next->end = &stream_encoder_mt_end;
		next->get_progress = &get_progress;
// 		next->update = &stream_encoder_mt_update;

		next->coder->filters[0].id = LZMA_VLI_UNKNOWN;
		next->coder->index_encoder = LZMA_NEXT_CODER_INIT;
		next->coder->index = NULL;
		memzero(&next->coder->outq, sizeof(next->coder->outq));
		next->coder->threads = NULL;
		next->coder->threads_max = 0;
		next->coder->threads_initialized = 0;
	}

	// Basic initializations
	next->coder->sequence = SEQ_STREAM_HEADER;
	next->coder->block_size = (size_t)(block_size);
	next->coder->thread_error = LZMA_OK;
	next->coder->thr = NULL;

	// Allocate the thread-specific base structures.
	assert(options->threads > 0);
	if (next->coder->threads_max != options->threads) {
		threads_end(next->coder, allocator);

		next->coder->threads = NULL;
		next->coder->threads_max = 0;

		next->coder->threads_initialized = 0;
		next->coder->threads_free = NULL;

		next->coder->threads = lzma_alloc(
				options->threads * sizeof(worker_thread),
				allocator);
		if (next->coder->threads == NULL)
			return LZMA_MEM_ERROR;

		next->coder->threads_max = options->threads;
	} else {
		// Reuse the old structures and threads. Tell the running
		// threads to stop and wait until they have stopped.
		threads_stop(next->coder, true);
	}

	// Output queue
	return_if_error(lzma_outq_init(&next->coder->outq, allocator,
			outbuf_size_max, options->threads));

	// Timeout
	if (options->timeout > 0) {
		next->coder->wait_max.tv_sec = options->timeout / 1000;
		next->coder->wait_max.tv_nsec
				= (options->timeout % 1000) * 1000000L;
		next->coder->has_timeout = true;
	} else {
		next->coder->has_timeout = false;
	}

	// Free the old filter chain and copy the new one.
	for (size_t i = 0; next->coder->filters[i].id != LZMA_VLI_UNKNOWN; ++i)
		lzma_free(next->coder->filters[i].options, allocator);

	return_if_error(lzma_filters_copy(options->filters,
			next->coder->filters, allocator));

	// Index
	lzma_index_end(next->coder->index, allocator);
	next->coder->index = lzma_index_init(allocator);
	if (next->coder->index == NULL)
		return LZMA_MEM_ERROR;

	// Stream Header
	next->coder->stream_flags.version = 0;
	next->coder->stream_flags.check = options->check;
	return_if_error(lzma_stream_header_encode(
			&next->coder->stream_flags, next->coder->header));

	next->coder->header_pos = 0;

	// Progress info
	next->coder->progress_in = 0;
	next->coder->progress_out = LZMA_STREAM_HEADER_SIZE;

	return LZMA_OK;
}


extern LZMA_API(lzma_ret)
lzma_stream_encoder_mt(lzma_stream *strm, const lzma_mt *options)
{
	lzma_next_strm_init(stream_encoder_mt_init, strm, options);

	strm->internal->supported_actions[LZMA_RUN] = true;
// 	strm->internal->supported_actions[LZMA_SYNC_FLUSH] = true;
// 	strm->internal->supported_actions[LZMA_FULL_FLUSH] = true;
// 	strm->internal->supported_actions[LZMA_FULL_BARRIER] = true;
	strm->internal->supported_actions[LZMA_FINISH] = true;

	return LZMA_OK;
}


// This function name is a monster but it's consistent with the older
// monster names. :-( 31 chars is the max that C99 requires so in that
// sense it's not too long. ;-)
extern LZMA_API(uint64_t)
lzma_stream_encoder_mt_memusage(const lzma_mt *options)
{
	lzma_options_easy easy;
	const lzma_filter *filters;
	uint64_t block_size;
	uint64_t outbuf_size_max;

	if (get_options(options, &easy, &filters, &block_size,
			&outbuf_size_max) != LZMA_OK)
		return UINT64_MAX;

	// Memory usage of the input buffers
	const uint64_t inbuf_memusage = options->threads * block_size;

	// Memory usage of the filter encoders
	uint64_t filters_memusage
			= lzma_raw_encoder_memusage(options->filters);
	if (filters_memusage == UINT64_MAX)
		return UINT64_MAX;

	filters_memusage *= options->threads;

	// Memory usage of the output queue
	const uint64_t outq_memusage = lzma_outq_memusage(
			outbuf_size_max, options->threads);
	if (outq_memusage == UINT64_MAX)
		return UINT64_MAX;

	// Sum them with overflow checking.
	uint64_t total_memusage = LZMA_MEMUSAGE_BASE + sizeof(lzma_coder)
			+ options->threads * sizeof(worker_thread);

	if (UINT64_MAX - total_memusage < inbuf_memusage)
		return UINT64_MAX;

	total_memusage += inbuf_memusage;

	if (UINT64_MAX - total_memusage < filters_memusage)
		return UINT64_MAX;

	total_memusage += filters_memusage;

	if (UINT64_MAX - total_memusage < outq_memusage)
		return UINT64_MAX;

	return total_memusage + outq_memusage;
}