File size: 25,691 Bytes
fa5173a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""cm-bh-v2: context-mixing compressor, bit-history state machines (paq8l table).

Models (mixer inputs): exact order-1/2; hashed order-3/4/5/6/8 and two word
contexts via the 253-state paq8l bit-history FSM + 20-bit adaptive state maps
(count limit 1023); two sparse contexts; three match models (order-4/6/12);
six indirect (successor-history) contexts on orders 1/2/3/4/6 and words.
Mixing: 32 logistic weight sets (char class x match state x bit half) +
second bank keyed by high nibble, adaptive final blend.
SSE: paq8px-style series - p1=(p+3*APM(order0))/4, then blend of APMs over
order-1/2/3, match-state and word contexts.
All tables fixed-size (bounded memory regardless of input size).
Deterministic, CPU-only, pure Python stdlib.
"""
import math
from array import array

STRETCH_TABLE = [0.0] * 4097
for i in range(1, 4096):
    p = i / 4096.0
    STRETCH_TABLE[i] = math.log(p / (1 - p))
STRETCH_TABLE[0] = STRETCH_TABLE[1]
STRETCH_TABLE[4096] = STRETCH_TABLE[4095]

MATCH_ORDER = 4
MATCH_ORDER2 = 6
MATCH_ORDER3 = 12
HBITS = 25
HASH_MASK = (1 << HBITS) - 1
HASH_MASK6 = (1 << HBITS) - 1
WBITS = 22
WORD_MASK = (1 << WBITS) - 1

APM_CTX = 256
APM_CTX2 = 1024
APM_CTX3 = 64
APM_STEPS = 33

N_WSETS = 32
# ---- таблица состояний bit-history из paq8l/lpaq1 ----
N_STATES = 256
NEXT = bytearray(b'\x01\x02\x03\x05\x04\x06\x07\n\x08\x0c\t\r\x0b\x0e\x0f\x13\x10\x17\x11\x18\x12\x19\x14\x1b\x15\x1c\x16\x1d\x1a\x1e\x1f! # # # #"%"%"%"%"%"%$\'$\'$\'$\'&()+*-*-,/,/.1.10303245+69698;8;:=:=<?<?>A>A2BC7D9D9FIFIHKHKJMJMLOLO>Q>Q@RSETGTGVIVI,;,;:=:=<1<1LYLYN[N[P\\]E^W^W`-`-0c0cXeXePfgEhWhWj9j9>m>mXoXoPpqUrWrWt9t9>w>wXyXyZz{U|a|a~9~9>\x81>\x81b\x83b\x83Z\x84\x85U\x86a\x86a\x889\x889>\x8b>\x8bb\x8db\x8dZ\x8e\x8f_\x90a\x90aD9D9>Q>Qb\x93b\x93d\x94\x95_\x96k\x96kl\x97l\x97d\x98\x99_\x9akl\x9bd\x9c\x9d_\x9ekl\x9fd\xa0\xa1i\xa2kl\xa3n\xa4\xa5i\xa6uv\xa7n\xa8\xa9i\xaauv\xabn\xac\xadi\xaeuv\xafn\xb0\xb1i\xb2uv\xb3n\xb4\xb5s\xb6uv\xb7x\xb8\xb9s\xba\x7f\x80\xbbx\xbc\xbds\xbe\x7f\x80\xbfx\xc0\xc1s\xc2\x7f\x80\xc3x\xc4\xc5s\xc6\x7f\x80\xc7x\xc8\xc9s\xca\x7f\x80\xcbx\xcc\xcds\xce\x7f\x80\xcfx\xd0\xd1}\xd2\x7f\x80\xd3\x82\xd4\xd5}\xd6\x89\x8a\xd7\x82\xd8\xd9}\xda\x89\x8a\xdb\x82\xdc\xdd}\xde\x89\x8a\xdf\x82\xe0\xe1}\xe2\x89\x8a\xe3\x82\xe4\xe5}\xe6\x89\x8a\xe7\x82\xe8\xe9}\xea\x89\x8a\xeb\x82\xec\xed}\xee\x89\x8a\xef\x82\xf0\xf1}\xf2\x89\x8a\xf3\x82\xf4\xf5\x87\xf6\x89\x8a\xf7\x8c\xf8\xf9\x87\xfaEP\xfb\x8c\xfc\xf9\x87\xfaEP\xfb\x8c\xfc\x00\x00\x00\x00\x00\x00')
SM_INIT = [2048] * N_STATES
DSHIFT = bytes(2 if i < 7 else (3 if i < 31 else (4 if i < 120 else 5)) for i in range(N_STATES))

SM_LIMIT = 1023
SM_RCP = [int(65536 / (n + 1.5)) for n in range(SM_LIMIT + 1)]  # скорость StateMap по счётчику

def make_sm():
    return [p << 8 for p in SM_INIT], array('H', [0]) * N_STATES

RATE = bytes(2 if c < 2 else (3 if c < 6 else (4 if c < 30 else 5)) for c in range(256))
LR = 0.002
SHIFT = 4
APM_W1 = 3
APM_W2 = 3
APM_W3 = 2
MIX_W = 1


def squash(x):
    if x > 20:
        return 4095
    if x < -20:
        return 0
    return int(4096 / (1 + math.exp(-x)))


class APM:
    def __init__(self, n_ctx):
        self.n = n_ctx
        self.t = [0] * (n_ctx * APM_STEPS)
        for c in range(n_ctx):
            for i in range(APM_STEPS):
                s = (i - (APM_STEPS // 2)) * (8.0 / (APM_STEPS // 2))
                self.t[c * APM_STEPS + i] = squash(s) * 16
        self.last_idx = 0
        self.last_w = 0.0

    def pp(self, p, ctx):
        st = STRETCH_TABLE[p]
        st = -8.0 if st < -8.0 else (8.0 if st > 8.0 else st)
        pos = (st + 8.0) * ((APM_STEPS - 1) / 16.0)
        lo = int(pos)
        lo = APM_STEPS - 2 if lo >= APM_STEPS - 1 else lo
        w = pos - lo
        base = ctx * APM_STEPS + lo
        v = self.t[base] * (1 - w) + self.t[base + 1] * w
        self.last_idx = base
        self.last_w = w
        out = int(v / 16)
        return 1 if out < 1 else (4094 if out > 4094 else out)

    def update(self, bit):
        g = bit * 4096 * 16
        base = self.last_idx
        self.t[base] += int((g - self.t[base]) * (1 - self.last_w) / 32)
        self.t[base + 1] += int((g - self.t[base + 1]) * self.last_w / 32)


class Predictor:
    def __init__(self, data_len_hint=0):
        self.t1 = [2048] * (256 * 256)
        self.t2 = [2048] * (1 << 24)
        self.t3 = bytearray(HASH_MASK + 1)
        self.t4 = bytearray(HASH_MASK + 1)
        self.t6 = bytearray(HASH_MASK6 + 1)
        self.succ1 = bytearray(256)
        self.succ2 = bytearray(65536)
        self.succ3 = bytearray(1 << 24)
        self.succ4 = bytearray(1 << 24)
        self.succ6 = bytearray(1 << 24)
        self.succ1b = bytearray(256)
        self.succw = bytearray(1 << 22)
        self.ti1 = bytearray(1 << 22)
        self.ti2 = bytearray(1 << 22)
        self.ti3 = bytearray(1 << 22)
        self.ti4 = bytearray(1 << 22)
        self.ti6 = bytearray(1 << 22)
        self.tiw = bytearray(1 << 22)
        self.smi1 = make_sm()
        self.smi2 = make_sm()
        self.smi3 = make_sm()
        self.smi4 = make_sm()
        self.smi6 = make_sm()
        self.smiw = make_sm()
        self.sm3 = make_sm()
        self.sm4 = make_sm()
        self.sm6 = make_sm()
        self.sm5 = make_sm()
        self.sm8 = make_sm()
        self.smw = make_sm()
        self.smw1 = make_sm()
        self.tw = bytearray(WORD_MASK + 1)
        self.tw1 = bytearray(WORD_MASK + 1)
        self.ts = [2048] * (256 * 256)
        self.ts2 = [2048] * (256 * 256)
        self.t5 = bytearray(WORD_MASK + 1)
        self.t8 = bytearray(WORD_MASK + 1)
        # weight sets: o1,o2,o3,o4,o6,match4,match8,word
        base_w = [0.2, 0.2, 0.18, 0.18, 0.14, 0.3, 0.3, 0.2, 0.15, 0.1, 0.16, 0.35, 0.12, 0.12, 0.12, 0.08, 0.12, 0.12, 0.12, 0.12]
        self.wsets = [list(base_w) for _ in range(N_WSETS)]
        self.wsets_b = [list(base_w) for _ in range(16)]
        self.wfinal = [0.5, 0.5]
        self.c1 = 0
        self.c2 = 0
        self.c3 = 0
        self.c4 = 0
        self.c5 = 0
        self.c8 = 0
        self.c6 = 0
        self.word_h = 0
        self.prev_word_h = 0
        self.history = bytearray(max(data_len_hint, 1))
        self.pos = 0
        self.match_table = [0] * (1 << 22)
        self.match_ptr = -1
        self.match_len = 0
        self.match_table2 = [0] * (1 << 22)
        self.match_ptr2 = -1
        self.match_len2 = 0
        self.match_table3 = [0] * (1 << 22)
        self.match_ptr3 = -1
        self.match_len3 = 0
        self.apm0 = APM(256)
        self.apm4 = APM(4096)
        self.apm = APM(APM_CTX)
        self.apm2 = APM(APM_CTX2)
        self.apm3 = APM(APM_CTX3)

    def _try_acquire_match(self):
        if self.pos < MATCH_ORDER:
            return None
        h = 0
        for i in range(MATCH_ORDER):
            h = (h * 0x9E3779B1 + self.history[self.pos - MATCH_ORDER + i]) & 0xFFFFFFFF
        h &= 0x3FFFFF
        if self.match_ptr < 0:
            v = self.match_table[h]
            if v:
                self.match_ptr = v - 1
                self.match_len = 1
        return h

    def _try_acquire_match2(self):
        if self.pos < MATCH_ORDER2:
            return None
        h = 0
        for i in range(MATCH_ORDER2):
            h = (h * 0x85EBCA6B + self.history[self.pos - MATCH_ORDER2 + i]) & 0xFFFFFFFF
        h &= 0x3FFFFF
        if self.match_ptr2 < 0:
            v = self.match_table2[h]
            if v:
                self.match_ptr2 = v - 1
                self.match_len2 = 1
        return h

    def _try_acquire_match3(self):
        if self.pos < MATCH_ORDER3:
            return None
        h = 0
        for i in range(MATCH_ORDER3):
            h = (h * 0xC2B2AE3D + self.history[self.pos - MATCH_ORDER3 + i]) & 0xFFFFFFFF
        h &= 0x3FFFFF
        if self.match_ptr3 < 0:
            v = self.match_table3[h]
            if v:
                self.match_ptr3 = v - 1
                self.match_len3 = 1
        return h

    def step_byte(self, io_fn, byte_in=None):
        match_hash = self._try_acquire_match()
        match_hash2 = self._try_acquire_match2()
        match_hash3 = self._try_acquire_match3()
        predicted_byte = self.history[self.match_ptr] if self.match_ptr >= 0 else -1
        predicted_byte2 = self.history[self.match_ptr2] if self.match_ptr2 >= 0 else -1
        predicted_byte3 = self.history[self.match_ptr3] if self.match_ptr3 >= 0 else -1

        node = 1
        out_byte = 0
        t1, t2, t3, t4, t6, tw = self.t1, self.t2, self.t3, self.t4, self.t6, self.tw
        tw1, ts, t5, t8 = self.tw1, self.ts, self.t5, self.t8
        ts2 = self.ts2
        smp3, smn3 = self.sm3
        smp4, smn4 = self.sm4
        smp6, smn6 = self.sm6
        smp5, smn5 = self.sm5
        smp8, smn8 = self.sm8
        smpw, smnw = self.smw
        smpw1, smnw1 = self.smw1
        NX = NEXT
        RC = SM_RCP
        ti1, ti2, ti3, ti4, ti6, tiw = self.ti1, self.ti2, self.ti3, self.ti4, self.ti6, self.tiw
        smpi1, smni1 = self.smi1
        smpi2, smni2 = self.smi2
        smpi3, smni3 = self.smi3
        smpi4, smni4 = self.smi4
        smpi6, smni6 = self.smi6
        smpiw, smniw = self.smiw
        c1, c2, c3, c4, c5, c6, c8 = self.c1, self.c2, self.c3, self.c4, self.c5, self.c6, self.c8
        match_len = self.match_len
        match_len2 = self.match_len2
        match_len3 = self.match_len3
        apm0 = self.apm0
        apm4 = self.apm4
        apm = self.apm
        apm2 = self.apm2
        apm3 = self.apm3

        # mixer weight-set selection: char class of previous byte x match states
        if 97 <= c1 <= 122 or 65 <= c1 <= 90:
            cls = 0
        elif c1 == 32:
            cls = 1
        elif 48 <= c1 <= 57:
            cls = 2
        else:
            cls = 3
        msel = (1 if predicted_byte >= 0 else 0) + 2 * (1 if predicted_byte2 >= 0 else 0)
        wlo = self.wsets[cls * 4 + msel]
        whi = self.wsets[16 + cls * 4 + msel]
        wb = self.wsets_b[c1 >> 4]
        wf = self.wfinal

        # word-model context for this byte
        ci1 = (((c1 << 16) | (self.succ1[c1] << 8) | self.succ1b[c1]) * 0x9E3779B1) & 0xFFFFFFFFFFFF
        ci2 = ((c2 << 8) | self.succ2[c2]) * 0x85EBCA6B
        ci3 = ((c3 << 8) | self.succ3[c3]) * 0xC2B2AE3D
        h4s = ((c4 * 0x2545F4914F6CDD1D) >> 24) & 0xFFFFFF
        ci4 = ((c4 << 8) | self.succ4[h4s]) * 0x27D4EB2F
        h6s = ((c6 * 0x9E3779B97F4A7C15) >> 28) & 0xFFFFFF
        ci6 = (((c6 & 0xFFFFFFFF) << 8) | self.succ6[h6s]) * 0x165667B1
        hws = (((self.word_h * 31) ^ self.prev_word_h) * 0x9E3779B1 >> 10) & 0x3FFFFF
        ciw = ((self.word_h << 8) | self.succw[hws]) * 0x94D049BB
        cw = ((self.word_h * 0x2545F4914F6CDD1D) ^ (self.prev_word_h * 0x9E3779B97F4A7C15)) & 0xFFFFFFFFFFFFFFFF
        cw1 = (self.word_h * 0x9E3779B97F4A7C15) & 0xFFFFFFFFFFFFFFFF
        sb = self.history[self.pos - 2] if self.pos >= 2 else 0
        sb2 = self.history[self.pos - 3] if self.pos >= 3 else 0

        for bitpos in range(8):
            w = whi if bitpos < 4 else wlo
            idx1 = (c1 << 8) | node
            key2 = (c2 << 8) | node
            idx3 = (((c3 * 0x2545F4914F6CDD1D) & 0xFFFFFFFFFFFFFFFF) ^ node) & HASH_MASK
            idx4 = (((c4 * 0x2545F4914F6CDD1D) & 0xFFFFFFFFFFFFFFFF) ^ node) & HASH_MASK
            idx6 = (((c6 * 0x9E3779B97F4A7C15) & 0xFFFFFFFFFFFFFFFF) ^ node) & HASH_MASK6
            idx5 = (((c5 * 0xFF51AFD7ED558CCD) & 0xFFFFFFFFFFFFFFFF) ^ node) & WORD_MASK
            idx8 = (((c8 * 0xD6E8FEB86659FD93) & 0xFFFFFFFFFFFFFFFF) ^ node) & WORD_MASK
            idxw = (cw ^ node) & WORD_MASK
            idxi1 = (ci1 ^ node) & 0x3FFFFF
            idxi2 = (ci2 ^ node) & 0x3FFFFF
            idxi3 = (ci3 ^ node) & 0x3FFFFF
            idxi4 = (ci4 ^ node) & 0x3FFFFF
            idxi6 = (ci6 ^ node) & 0x3FFFFF
            idxiw = (ciw ^ node) & 0x3FFFFF
            idxw1 = (cw1 ^ node) & WORD_MASK
            idxs = (sb << 8) | node
            idxs2 = (sb2 << 8) | node

            p1 = t1[idx1]
            p2 = t2[key2]
            s3 = t3[idx3]
            P3 = smp3[s3]
            p3 = P3 >> 8
            s4 = t4[idx4]
            P4 = smp4[s4]
            p4 = P4 >> 8
            s6 = t6[idx6]
            P6 = smp6[s6]
            p6 = P6 >> 8
            s5 = t5[idx5]
            P5 = smp5[s5]
            p5 = P5 >> 8
            s8 = t8[idx8]
            P8 = smp8[s8]
            p8 = P8 >> 8
            sw = tw[idxw]
            Pw = smpw[sw]
            pw = Pw >> 8
            si1 = ti1[idxi1]
            Pi1 = smpi1[si1]
            pi1 = Pi1 >> 8
            si2 = ti2[idxi2]
            Pi2 = smpi2[si2]
            pi2 = Pi2 >> 8
            si3 = ti3[idxi3]
            Pi3 = smpi3[si3]
            pi3 = Pi3 >> 8
            si4 = ti4[idxi4]
            Pi4 = smpi4[si4]
            pi4 = Pi4 >> 8
            si6 = ti6[idxi6]
            Pi6i = smpi6[si6]
            pi6i = Pi6i >> 8
            siw = tiw[idxiw]
            Piw = smpiw[siw]
            piw = Piw >> 8
            sw1 = tw1[idxw1]
            Pw1 = smpw1[sw1]
            pw1 = Pw1 >> 8
            psp = ts[idxs]
            psp2 = ts2[idxs2]
            st1 = STRETCH_TABLE[p1]
            st2 = STRETCH_TABLE[p2]
            st3 = STRETCH_TABLE[p3]
            st4 = STRETCH_TABLE[p4]
            st6 = STRETCH_TABLE[p6]
            st5 = STRETCH_TABLE[p5]
            st8 = STRETCH_TABLE[p8]
            stw = STRETCH_TABLE[pw]
            sti1 = STRETCH_TABLE[pi1]
            sti2 = STRETCH_TABLE[pi2]
            sti3 = STRETCH_TABLE[pi3]
            sti4 = STRETCH_TABLE[pi4]
            sti6 = STRETCH_TABLE[pi6i]
            stiw = STRETCH_TABLE[piw]
            stw1 = STRETCH_TABLE[pw1]
            stsp = STRETCH_TABLE[psp]
            stsp2 = STRETCH_TABLE[psp2]

            if predicted_byte >= 0:
                pred_bit = (predicted_byte >> (7 - bitpos)) & 1
                conf = min(3.5, 0.3 + match_len * 0.25)
                stm = conf if pred_bit == 1 else -conf
            else:
                stm = 0.0

            if predicted_byte2 >= 0:
                pred_bit2 = (predicted_byte2 >> (7 - bitpos)) & 1
                conf2 = min(5.0, 0.5 + match_len2 * 0.35)
                stm2 = conf2 if pred_bit2 == 1 else -conf2
            else:
                stm2 = 0.0

            if predicted_byte3 >= 0:
                pred_bit3 = (predicted_byte3 >> (7 - bitpos)) & 1
                conf3 = min(6.5, 1.0 + match_len3 * 0.5)
                stm3 = conf3 if pred_bit3 == 1 else -conf3
            else:
                stm3 = 0.0

            dot = (w[0] * st1 + w[1] * st2 + w[2] * st3 + w[3] * st4 + w[4] * st6
                   + w[5] * stm + w[6] * stm2 + w[7] * stw + w[8] * stw1 + w[9] * stsp + w[10] * st5 + w[11] * stm3 + w[12] * st8 + w[13] * sti1 + w[14] * sti2 + w[15] * stsp2 + w[16] * sti3 + w[17] * sti4 + w[18] * sti6 + w[19] * stiw)
            dot_b = (wb[0] * st1 + wb[1] * st2 + wb[2] * st3 + wb[3] * st4 + wb[4] * st6
                   + wb[5] * stm + wb[6] * stm2 + wb[7] * stw + wb[8] * stw1 + wb[9] * stsp + wb[10] * st5 + wb[11] * stm3 + wb[12] * st8 + wb[13] * sti1 + wb[14] * sti2 + wb[15] * stsp2 + wb[16] * sti3 + wb[17] * sti4 + wb[18] * sti6 + wb[19] * stiw)
            fdot = wf[0] * dot + wf[1] * dot_b
            mixed = squash(fdot)
            mixed = 1 if mixed < 1 else (4094 if mixed > 4094 else mixed)
            m1 = (mixed + 3 * apm0.pp(mixed, node)) >> 2
            m1 = 1 if m1 < 1 else (4094 if m1 > 4094 else m1)
            r1 = apm.pp(m1, c1)
            r2 = apm2.pp(m1, c2 & (APM_CTX2 - 1))
            mstate = (0 if match_len2 == 0 else (1 if match_len2 < 8 else (2 if match_len2 < 24 else 3)))
            r3 = apm3.pp(m1, (mstate << 2) | cls)
            r4 = apm4.pp(m1, ((c3 * 0x9E3779B1) >> 20) & 4095)
            final_p = (r1 + 2 * r2 + r3 + 2 * r4) // 6
            final_p = 1 if final_p < 1 else (4094 if final_p > 4094 else final_p)

            bit = io_fn(final_p, byte_in, bitpos)

            err = (bit * 4096 - mixed) / 4096.0
            lr = LR
            w[0] += lr * err * st1
            w[1] += lr * err * st2
            w[2] += lr * err * st3
            w[3] += lr * err * st4
            w[4] += lr * err * st6
            w[5] += lr * err * stm
            w[6] += lr * err * stm2
            w[7] += lr * err * stw
            w[8] += lr * err * stw1
            w[9] += lr * err * stsp
            w[10] += lr * err * st5
            w[11] += lr * err * stm3
            w[12] += lr * err * st8
            wb[0] += lr * err * st1
            wb[1] += lr * err * st2
            wb[2] += lr * err * st3
            wb[3] += lr * err * st4
            wb[4] += lr * err * st6
            wb[5] += lr * err * stm
            wb[6] += lr * err * stm2
            wb[7] += lr * err * stw
            wb[8] += lr * err * stw1
            wb[9] += lr * err * stsp
            wb[10] += lr * err * st5
            wb[11] += lr * err * stm3
            wb[12] += lr * err * st8
            w[13] += lr * err * sti1
            w[14] += lr * err * sti2
            wb[13] += lr * err * sti1
            wb[14] += lr * err * sti2
            w[15] += lr * err * stsp2
            w[16] += lr * err * sti3
            wb[15] += lr * err * stsp2
            wb[16] += lr * err * sti3
            w[17] += lr * err * sti4
            wb[17] += lr * err * sti4
            w[18] += lr * err * sti6
            wb[18] += lr * err * sti6
            w[19] += lr * err * stiw
            wb[19] += lr * err * stiw
            wf[0] += 0.0005 * err * dot
            wf[1] += 0.0005 * err * dot_b
            apm0.update(bit)
            apm4.update(bit)
            apm.update(bit)
            apm2.update(bit)
            apm3.update(bit)

            tgt = bit * 4096
            tgth = bit * 1048320
            t1[idx1] = p1 + ((tgt - p1) >> SHIFT)
            t2[key2] = p2 + ((tgt - p2) >> SHIFT)
            c = smn3[s3]
            smp3[s3] = P3 + (((tgth - P3) * RC[c]) >> 16)
            if c < SM_LIMIT: smn3[s3] = c + 1
            t3[idx3] = NX[(s3 << 1) | bit]
            c = smn4[s4]
            smp4[s4] = P4 + (((tgth - P4) * RC[c]) >> 16)
            if c < SM_LIMIT: smn4[s4] = c + 1
            t4[idx4] = NX[(s4 << 1) | bit]
            c = smn6[s6]
            smp6[s6] = P6 + (((tgth - P6) * RC[c]) >> 16)
            if c < SM_LIMIT: smn6[s6] = c + 1
            t6[idx6] = NX[(s6 << 1) | bit]
            c = smn5[s5]
            smp5[s5] = P5 + (((tgth - P5) * RC[c]) >> 16)
            if c < SM_LIMIT: smn5[s5] = c + 1
            t5[idx5] = NX[(s5 << 1) | bit]
            c = smn8[s8]
            smp8[s8] = P8 + (((tgth - P8) * RC[c]) >> 16)
            if c < SM_LIMIT: smn8[s8] = c + 1
            t8[idx8] = NX[(s8 << 1) | bit]
            c = smni1[si1]
            smpi1[si1] = Pi1 + (((tgth - Pi1) * RC[c]) >> 16)
            if c < SM_LIMIT: smni1[si1] = c + 1
            ti1[idxi1] = NX[(si1 << 1) | bit]
            c = smni2[si2]
            smpi2[si2] = Pi2 + (((tgth - Pi2) * RC[c]) >> 16)
            if c < SM_LIMIT: smni2[si2] = c + 1
            ti2[idxi2] = NX[(si2 << 1) | bit]
            c = smni3[si3]
            smpi3[si3] = Pi3 + (((tgth - Pi3) * RC[c]) >> 16)
            if c < SM_LIMIT: smni3[si3] = c + 1
            ti3[idxi3] = NX[(si3 << 1) | bit]
            c = smni4[si4]
            smpi4[si4] = Pi4 + (((tgth - Pi4) * RC[c]) >> 16)
            if c < SM_LIMIT: smni4[si4] = c + 1
            ti4[idxi4] = NX[(si4 << 1) | bit]
            c = smni6[si6]
            smpi6[si6] = Pi6i + (((tgth - Pi6i) * RC[c]) >> 16)
            if c < SM_LIMIT: smni6[si6] = c + 1
            ti6[idxi6] = NX[(si6 << 1) | bit]
            c = smniw[siw]
            smpiw[siw] = Piw + (((tgth - Piw) * RC[c]) >> 16)
            if c < SM_LIMIT: smniw[siw] = c + 1
            tiw[idxiw] = NX[(siw << 1) | bit]
            c = smnw[sw]
            smpw[sw] = Pw + (((tgth - Pw) * RC[c]) >> 16)
            if c < SM_LIMIT: smnw[sw] = c + 1
            tw[idxw] = NX[(sw << 1) | bit]
            c = smnw1[sw1]
            smpw1[sw1] = Pw1 + (((tgth - Pw1) * RC[c]) >> 16)
            if c < SM_LIMIT: smnw1[sw1] = c + 1
            tw1[idxw1] = NX[(sw1 << 1) | bit]
            ts[idxs] = psp + ((tgt - psp) >> SHIFT)
            ts2[idxs2] = psp2 + ((tgt - psp2) >> SHIFT)

            if predicted_byte >= 0 and bit != pred_bit:
                predicted_byte = -1
            if predicted_byte2 >= 0 and bit != pred_bit2:
                predicted_byte2 = -1
            if predicted_byte3 >= 0 and bit != pred_bit3:
                predicted_byte3 = -1

            node = (node << 1) | bit
            out_byte = (out_byte << 1) | bit

        if self.pos >= len(self.history):
            self.history.append(out_byte)
        else:
            self.history[self.pos] = out_byte

        if self.match_ptr >= 0 and self.match_ptr < self.pos and self.history[self.match_ptr] == out_byte:
            self.match_ptr += 1
            self.match_len += 1
        else:
            self.match_ptr = -1
            self.match_len = 0
        if match_hash is not None:
            self.match_table[match_hash] = self.pos + 1

        if self.match_ptr2 >= 0 and self.match_ptr2 < self.pos and self.history[self.match_ptr2] == out_byte:
            self.match_ptr2 += 1
            self.match_len2 += 1
        else:
            self.match_ptr2 = -1
            self.match_len2 = 0
        if match_hash2 is not None:
            self.match_table2[match_hash2] = self.pos + 1

        if self.match_ptr3 >= 0 and self.match_ptr3 < self.pos and self.history[self.match_ptr3] == out_byte:
            self.match_ptr3 += 1
            self.match_len3 += 1
        else:
            self.match_ptr3 = -1
            self.match_len3 = 0
        if match_hash3 is not None:
            self.match_table3[match_hash3] = self.pos + 1

        # word model state
        b = out_byte
        if 97 <= b <= 122 or 65 <= b <= 90:
            self.word_h = (self.word_h * 0x9E3779B1 + (b | 0x20)) & 0xFFFFFFFFFFFFFFFF
        elif self.word_h:
            self.prev_word_h = self.word_h
            self.word_h = 0

        self.succ1b[c1] = self.succ1[c1]
        self.succ1[c1] = out_byte
        self.succ2[c2] = out_byte
        self.succ3[c3] = out_byte
        self.succ4[h4s] = out_byte
        self.succ6[h6s] = out_byte
        self.succw[hws] = out_byte
        self.c2 = ((c1 << 8) | out_byte) & 0xFFFF
        self.c3 = ((c3 << 8) | out_byte) & 0xFFFFFF
        self.c4 = ((c4 << 8) | out_byte) & 0xFFFFFFFF
        self.c5 = ((c5 << 8) | out_byte) & 0xFFFFFFFFFF
        self.c8 = ((c8 << 8) | out_byte) & 0xFFFFFFFFFFFFFFFF
        self.c6 = ((c6 << 8) | out_byte) & 0xFFFFFFFFFFFF
        self.c1 = out_byte
        self.pos += 1
        return out_byte


TOP = 1 << 24
BOT = 1 << 16
MASK32 = (1 << 32) - 1


class RangeEncoderBit:
    def __init__(self):
        self.low = 0
        self.range = MASK32
        self.out = bytearray()

    def encode_bit(self, p1, bit):
        mid = (self.range >> 12) * p1
        if bit:
            self.range = mid
        else:
            self.low = (self.low + mid) & MASK32
            self.range -= mid
        while True:
            if (self.low ^ (self.low + self.range)) < TOP:
                pass
            elif self.range < BOT:
                self.range = (-self.low) & (BOT - 1)
            else:
                break
            self.out.append((self.low >> 24) & 0xFF)
            self.low = (self.low << 8) & MASK32
            self.range = (self.range << 8) & MASK32

    def finish(self):
        for _ in range(4):
            self.out.append((self.low >> 24) & 0xFF)
            self.low = (self.low << 8) & MASK32
        return bytes(self.out)


class RangeDecoderBit:
    def __init__(self, data):
        self.data = data
        self.pos = 0
        self.low = 0
        self.range = MASK32
        self.code = 0
        for _ in range(4):
            self.code = ((self.code << 8) | self._byte()) & MASK32

    def _byte(self):
        if self.pos < len(self.data):
            b = self.data[self.pos]
            self.pos += 1
            return b
        return 0

    def decode_bit(self, p1):
        mid = (self.range >> 12) * p1
        if (self.code - self.low) & MASK32 < mid:
            bit = 1
            self.range = mid
        else:
            bit = 0
            self.low = (self.low + mid) & MASK32
            self.range -= mid
        while True:
            if (self.low ^ (self.low + self.range)) < TOP:
                pass
            elif self.range < BOT:
                self.range = (-self.low) & (BOT - 1)
            else:
                break
            self.code = ((self.code << 8) | self._byte()) & MASK32
            self.low = (self.low << 8) & MASK32
            self.range = (self.range << 8) & MASK32
        return bit


def compress(data: bytes) -> bytes:
    import gc
    gc.disable()
    pred = Predictor(len(data))
    enc = RangeEncoderBit()

    def io_fn(p, byte_in, bitpos):
        bit = (byte_in >> (7 - bitpos)) & 1
        enc.encode_bit(p, bit)
        return bit

    for b in data:
        pred.step_byte(io_fn, b)
    body = enc.finish()
    return len(data).to_bytes(8, "big") + body


def decompress(blob: bytes) -> bytes:
    import gc
    gc.disable()
    n = int.from_bytes(blob[:8], "big")
    dec = RangeDecoderBit(blob[8:])
    pred = Predictor(n)
    out = bytearray()

    def io_fn(p, byte_in, bitpos):
        return dec.decode_bit(p)

    for _ in range(n):
        b = pred.step_byte(io_fn)
        out.append(b)
    return bytes(out)