forked from gsauthof/riscv
-
Notifications
You must be signed in to change notification settings - Fork 1
/
bcd2ascii.S
688 lines (605 loc) · 19.4 KB
/
bcd2ascii.S
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
## bcd2ascii - AVX version
##
## 2020, Georg Sauthoff <[email protected]>, LGPLv3+
.intel_syntax noprefix
## Comments use 2 hash characters to protect against accidental
## clashes with C-pre-processor directives.
#ifndef BCD2ASCII_LOOP_BYTES
## loop in increments of up to 32 source bytes
#define BCD2ASCII_LOOP_BYTES 32
#endif
#ifndef BCD2ASCII_SSE4_1
#define BCD2ASCII_SSE4_1 1
#endif
## We only need to zero the upper lanes if other libraries execute
## non-VEX SIMD instructions. Apparently, as of 2020, e.g. Fedora 31, glibc
## and others don't execute non-VEX legacy SIMD instructions, anymore.
#if BCD2ASCII_ZEROUPPER
#define vzeroupper vzeroupper
#else
#define vzeroupper
#endif
.text
.balign 16
#
## void *bcd2ascii(void* dst, const void* src, size_t n);
#
## or:
#
## struct Void_Pair { void *fst; void *snd; };
## typedef struct Void_Pair Void_Pair;
## Void_Pair bcd2ascii(void* dst, const void* src, size_t n);
#
## rdi = dst, rsi = src, rdx = n
## return values in: rax, rdx
#
.global bcd2ascii
bcd2ascii:
mov r11, rdx ## backup n
lea rax, [rdi + 2 * rdx]
add rdx, rsi
cmp r11, 1 ## shortcut for tiny input
je .Lbegin1
## ## prepare lookup table
#if ! BCD2ASCII_NO_STATIC
## Yeah, loading the table from memory is cheaper than constructing it
## by issuing a bunch of instructions.
#if BCD2ASCII_WIDTH != 128
vmovdqa ymm10, [.Lookup_table]
#else
vmovdqa xmm10, [.Lookup_table]
#endif
#else
mov r8, 0x3736353433323130 ## '7', '6', ..., '0'
mov r9, 0x6665646362613938 ## 'f', 'e', ..., '8'
vmovq xmm8, r8 ## move quad-word (64 bit) into low half of xmm
#if BCD2ASCII_SSE4_1
## of course, AVX always supports this instruction
## insert general-purpose register quad-word (64 bit) with offset 1
vpinsrq xmm10, xmm8, r9, 1 ## into xmm10 and use other half of xmm8
#else
vmovq xmm9, r9 ## move quad-word (64 bit) into low half of xmm
## unpack (interleave) low quad-words (64 bit) into double-quad-words (128 bit)
## i.e. ymm10 = low(ymm9)low(ymm8)
vpunpcklqdq xmm10, xmm8, xmm9 ## i.e. lower 128 bit contain the table
#endif
#if BCD2ASCII_WIDTH != 128
## we need to copy it to the upper 128 bit, since shuffle uses
## a separate table for each 128 bit half
vinserti128 ymm10, ymm10, xmm10, 1
#endif // BCD2ASCII_WIDTH != 128
#endif // BCD2ASCII_NO_STATIC
## ## prepare low-nibble mask
#if ! BCD2ASCII_NO_STATIC
## Yes, loading the mask from memory is cheaper than constructing with
## three instructions.
#if BCD2ASCII_WIDTH != 128
vmovdqa ymm4, [.Low_nibble_mask]
#else
vmovdqa xmm4, [.Low_nibble_mask]
#endif
#else
mov rcx, 0x0f0f0f0f
## with AVX512VL/VW, vpbroadcastb also supports broadcasting a byte
## from a general purpose register:
## vpbroadcastb ymm4, rcx # set each element to 0xf, i.e. mask low nibble
## alternatively:
vmovd xmm4, rcx ## move double-word (32 bit)
#vpbroadcastb ymm4, xmm4 ## broadcast lowest xmm byte to all ymm bytes
vpbroadcastd ymm4, xmm4 ## broadcast lowest xmm double-word to all elems
#endif // BCD2ASCII_NO_STATIC
#if BCD2ASCII_LOOP_BYTES >= 32
add rsi, 32
cmp rsi, rdx
jnbe .Lskip32
#if BCD2ASCII_ALIGN_LOOP
## Align jump target to 16 bytes as recommended by Intel
.balign 16
#endif
.Loop0:
vmovdqu ymm0, [rsi-32]
add rsi, 32 ## increment src
vpsrlw ymm1, ymm0, 4 ## shift-right-logical each 16 bit word element
## interleave/unpack each byte pair of the low 64 bit half
## into 16 bit words of a full 128 bit lane, for each 128 bit lane
vpunpcklbw ymm2, ymm1, ymm0
## interleave/unpack byte pairs of the high 64 bit half into 16 bit words,
## for each 128 bit lane
vpunpckhbw ymm3, ymm1, ymm0
vpand ymm2, ymm2, ymm4 ## bit-and, i.e. mask with 0x0f0f...0f
vpand ymm3, ymm3, ymm4 ## bit-and, i.e. mask with 0x0f0f...0f
## ymm5 = concat(xmm3, xmm2)
vinserti128 ymm5, ymm2, xmm3, 1
## ymm6 = concat(hilane(ymm3), hilane(ymm2))
## i.e. permute 2 128 bit integer values
vperm2i128 ymm6, ymm2, ymm3, 0x31
## lookup each byte in table ymm10 using indices in ymm5,
## for each 128 bit lane
vpshufb ymm5, ymm10, ymm5
vpshufb ymm6, ymm10, ymm6
vmovdqu [rdi], ymm5
vmovdqu [rdi+32], ymm6
add rdi, 64 ## increment dst
cmp rsi, rdx ## compare src with end
jbe .Loop0 ## branch if below-or-equal
.Lskip32:
sub rsi, 32
#endif // BCD2ASCII_LOOP_BYTES >= 32
#if BCD2ASCII_LOOP_BYTES >= 16
add rsi, 16
cmp rsi, rdx
jnbe .Lskip16
#if BCD2ASCII_ALIGN_LOOP
## Align jump target to 16 bytes as recommended by Intel
.balign 16
#endif
.Loop1:
/* There are basically 2 ways to deal with 16 source bytes:
- low/high unpack them in just one 128 bit lane
- do a 16 bit wideining load into a full 2 lane register (i.e. 256 bit)
*/
#if BCD2ASCII_UNPACK16
/* this branch is slower */
vmovdqu xmm0, [rsi-16]
add rsi, 16 ## increment src
vpsrlw xmm1, xmm0, 4 ## shift-right-logical each 16 bit word element
## interleave/unpack each byte pair in the low part into 16 bit words
vpunpcklbw xmm2, xmm1, xmm0
vpunpckhbw xmm3, xmm1, xmm0
vpand xmm2, xmm2, xmm4 ## bit-and, i.e. mask with 0x0f0f...0f
vpand xmm3, xmm3, xmm4 ## bit-and, i.e. mask with 0x0f0f...0f
## lookup each byte in table xmm10 using indices in xmm3
vpshufb xmm2, xmm10, xmm2
vpshufb xmm3, xmm10, xmm3
vmovdqu [rdi], xmm2 ## move double-quad-word (128 bit) to dst
vmovdqu [rdi+16], xmm3 ## move double-quad-word (128 bit) to dst
add rdi, 32 ## increment dst
#else // BCD2ASCII_UNPACK16
/* this branch is faster */
## vpmovzxbw ymm0, XMMWORD PTR [rsi] # XMMWORD PTR annotation is optional
vpmovzxbw ymm0, [rsi-16] ## move and zero-extend 16 bytes into 16 16-bit words
add rsi, 16 ## increment src
vpsllw ymm2, ymm0, 8 ## shift-left-logical each 16-bit word by 8 bits
vpsrlw ymm3, ymm0, 4 ## shift-right-logical each 16-bit word by 8 bits
vpand ymm5, ymm2, ymm4 ## bit-and, i.e. mask with 0x0f0f...0f
vpor ymm6, ymm5, ymm3 ## bit-or
## shuffle each each 128 bit half of ymm6 according to each lookup table
## in ymm10
vpshufb ymm7, ymm10, ymm6
## move 2 double-quad-words (i.e. the complete register) to unaligned dst
vmovdqu [rdi], ymm7
add rdi, 32 ## increment dst
#endif // else: BCD2ASCII_UNPACK16
cmp rsi, rdx ## compare src with mid
jbe .Loop1 ## branch if bfelow-or-equal
.Lskip16:
sub rsi, 16
#endif // BCD2ASCII_LOOP_BYTES >= 16
add rsi, 8
cmp rsi, rdx
jnbe .Lskip8
#if BCD2ASCII_ALIGN_LOOP
## Align jump target to 16 bytes as recommended by Intel
.balign 16
#endif
.Loop2:
vmovq xmm0, [rsi-8] ## load quad-word (64 bit) into lower xmm half
add rsi, 8 ## increment src
vpsrlw xmm1, xmm0, 4 ## shift-right-logical each 16 bit word element
## interleave/unpack each byte pair in the low part into 16 bit words
vpunpcklbw xmm2, xmm1, xmm0
vpand xmm3, xmm2, xmm4 ## bit-and, i.e. mask with 0x0f0f...0f
## lookup each byte in table xmm10 using indices in xmm3
vpshufb xmm5, xmm10, xmm3
vmovdqu [rdi], xmm5 ## move double-quad-word (128 bit) to dst
add rdi, 16 ## increment dst
cmp rsi, rdx ## compare src with end
jbe .Loop2 ## branch if below-equal
.Lskip8:
sub rsi, 8
mov r11, rdx
sub r11, rsi
## indirectly jump to process the last remaining input bytes
jmp [r11 * 8 + .Ljump_table0]
## Doesn't make a difference:
## .balign 16
.Lbegin0:
## avoid speed-penalty in case other functions that are executed next
## contain non-VEX SIMD instructions
vzeroupper
ret
.Lbegin1:
## mov r8b, [rsi] # would yield false dependency on register half
movzx r8, BYTE PTR [rsi] ## move and zero-extend one byte
## Yes, a lookup-table is faster than other alternatives
movzx r9, WORD PTR [r8 * 2 + .Lookup_bcd_byte]
mov WORD PTR [rdi], r9w
vzeroupper
ret
## slower alternative
#if 0
movzx r8, BYTE PTR [rsi] ## move and zero-extend one byte
jmp .Ldone
#endif
.Lbegin2:
movzx r8, BYTE PTR [rsi] ## move and zero-extend one byte
movzx r10, BYTE PTR [rsi+1]
movzx r9, WORD PTR [ r8 * 2 + .Lookup_bcd_byte]
movzx r11, WORD PTR [r10 * 2 + .Lookup_bcd_byte]
mov WORD PTR [rdi], r9w
mov WORD PTR [rdi+2], r11w
vzeroupper
ret
## slower alternative
#if 0
movzx r8, WORD PTR [rsi] ## move and zero-extend 2 bytes
## shouldn't make a difference:
## movzx r8d, WORD PTR [rsi]
jmp .Ldone
#endif
.Lbegin3:
movzx r8, WORD PTR [rsi]
movzx r9, BYTE PTR [rsi+2]
shl r9, 16 ## shift-left-logical
or r8, r9 ## bit-or
jmp .Ldone
.Lbegin4:
## moving a double-word always zero-extends
mov r8d, DWORD PTR [rsi] ## i.e. 32 bit
jmp .Ldone
.Lbegin5:
mov r8d, DWORD PTR [rsi+1]
shl r8, 8
mov r8b, BYTE PTR [rsi] ## save the or instruction ...
jmp .Ldone
.Lbegin6:
mov r8d, DWORD PTR [rsi+2]
shl r8, 16
mov r8w, WORD PTR [rsi]
jmp .Ldone
.Lbegin7:
mov r8d, DWORD PTR [rsi+3]
shl r8, 16
mov r8w, WORD PTR [rsi+1]
shl r8, 8
mov r8b, BYTE PTR [rsi]
.Ldone:
vmovq xmm0, r8 ## move quad-word (64 bit) into lower xmm half
vpsrlw xmm1, xmm0, 4 ## shift-right-logical each 16 bit word element
## interleave/unpack each byte pair in the low part into 16 bit words
vpunpcklbw xmm2, xmm1, xmm0
vpand xmm3, xmm2, xmm4 ## bit-and, i.e. mask with 0x0f0f...0f
## lookup each byte in table xmm10 using indices in xmm3
vpshufb xmm5, xmm10, xmm3
jmp [r11 * 8 + .Ljump_table1]
.Lend1:
vmovd r8, xmm5 ## move low double-word (32 bit), we can't move less
mov [rdi], r8w ## only move the lower word (16 bit)
vzeroupper
ret
.Lend2:
vmovd [rdi], xmm5
vzeroupper
ret
.Lend3:
vmovq r8, xmm5 ## move low quad-word (64 bit)
mov [rdi], r8d
shr r8, 32
mov [rdi+4], r8w
vzeroupper
ret
.Lend4:
vmovq [rdi], xmm5
vzeroupper
ret
.Lend5:
vmovq [rdi], xmm5
## move high 64 bit to low 64 bit
## (i.e. move packed single-precision floats high to low)
vmovhlps xmm6, xmm5, xmm5
vmovd r9, xmm6
mov [rdi+8], r9w
vzeroupper
ret
.Lend6:
vmovq [rdi], xmm5
vmovhlps xmm6, xmm5, xmm5
vmovd [rdi+8], xmm6
vzeroupper
ret
.Lend7:
vmovq [rdi], xmm5
vmovhlps xmm6, xmm5, xmm5
vmovq r9, xmm6
mov [rdi+8], r9d
shr r9, 32
mov [rdi+12], r9w
vzeroupper
ret
.section .rodata
.balign 32
.Lookup_table:
.quad 0x3736353433323130 ## '7', '6', ..., '0'
.quad 0x6665646362613938 ## 'f', 'e', ..., '8'
.quad 0x3736353433323130 ## '7', '6', ..., '0'
.quad 0x6665646362613938 ## 'f', 'e', ..., '8'
.Low_nibble_mask:
.quad 0x0f0f0f0f0f0f0f0f
.quad 0x0f0f0f0f0f0f0f0f
.quad 0x0f0f0f0f0f0f0f0f
.quad 0x0f0f0f0f0f0f0f0f
.Ljump_table0:
.quad .Lbegin0
.quad .Lbegin1
.quad .Lbegin2
.quad .Lbegin3
.quad .Lbegin4
.quad .Lbegin5
.quad .Lbegin6
.quad .Lbegin7
.Ljump_table1:
.quad 0 ## should not happend as we already returned via .Lbegin0
.quad .Lend1
.quad .Lend2
.quad .Lend3
.quad .Lend4
.quad .Lend5
.quad .Lend6
.quad .Lend7
.Lookup_bcd_byte:
.word /* 0x00 => */ 0x3030
.word /* 0x01 => */ 0x3130
.word /* 0x02 => */ 0x3230
.word /* 0x03 => */ 0x3330
.word /* 0x04 => */ 0x3430
.word /* 0x05 => */ 0x3530
.word /* 0x06 => */ 0x3630
.word /* 0x07 => */ 0x3730
.word /* 0x08 => */ 0x3830
.word /* 0x09 => */ 0x3930
.word /* 0x0a => */ 0x6130
.word /* 0x0b => */ 0x6230
.word /* 0x0c => */ 0x6330
.word /* 0x0d => */ 0x6430
.word /* 0x0e => */ 0x6530
.word /* 0x0f => */ 0x6630
.word /* 0x10 => */ 0x3031
.word /* 0x11 => */ 0x3131
.word /* 0x12 => */ 0x3231
.word /* 0x13 => */ 0x3331
.word /* 0x14 => */ 0x3431
.word /* 0x15 => */ 0x3531
.word /* 0x16 => */ 0x3631
.word /* 0x17 => */ 0x3731
.word /* 0x18 => */ 0x3831
.word /* 0x19 => */ 0x3931
.word /* 0x1a => */ 0x6131
.word /* 0x1b => */ 0x6231
.word /* 0x1c => */ 0x6331
.word /* 0x1d => */ 0x6431
.word /* 0x1e => */ 0x6531
.word /* 0x1f => */ 0x6631
.word /* 0x20 => */ 0x3032
.word /* 0x21 => */ 0x3132
.word /* 0x22 => */ 0x3232
.word /* 0x23 => */ 0x3332
.word /* 0x24 => */ 0x3432
.word /* 0x25 => */ 0x3532
.word /* 0x26 => */ 0x3632
.word /* 0x27 => */ 0x3732
.word /* 0x28 => */ 0x3832
.word /* 0x29 => */ 0x3932
.word /* 0x2a => */ 0x6132
.word /* 0x2b => */ 0x6232
.word /* 0x2c => */ 0x6332
.word /* 0x2d => */ 0x6432
.word /* 0x2e => */ 0x6532
.word /* 0x2f => */ 0x6632
.word /* 0x30 => */ 0x3033
.word /* 0x31 => */ 0x3133
.word /* 0x32 => */ 0x3233
.word /* 0x33 => */ 0x3333
.word /* 0x34 => */ 0x3433
.word /* 0x35 => */ 0x3533
.word /* 0x36 => */ 0x3633
.word /* 0x37 => */ 0x3733
.word /* 0x38 => */ 0x3833
.word /* 0x39 => */ 0x3933
.word /* 0x3a => */ 0x6133
.word /* 0x3b => */ 0x6233
.word /* 0x3c => */ 0x6333
.word /* 0x3d => */ 0x6433
.word /* 0x3e => */ 0x6533
.word /* 0x3f => */ 0x6633
.word /* 0x40 => */ 0x3034
.word /* 0x41 => */ 0x3134
.word /* 0x42 => */ 0x3234
.word /* 0x43 => */ 0x3334
.word /* 0x44 => */ 0x3434
.word /* 0x45 => */ 0x3534
.word /* 0x46 => */ 0x3634
.word /* 0x47 => */ 0x3734
.word /* 0x48 => */ 0x3834
.word /* 0x49 => */ 0x3934
.word /* 0x4a => */ 0x6134
.word /* 0x4b => */ 0x6234
.word /* 0x4c => */ 0x6334
.word /* 0x4d => */ 0x6434
.word /* 0x4e => */ 0x6534
.word /* 0x4f => */ 0x6634
.word /* 0x50 => */ 0x3035
.word /* 0x51 => */ 0x3135
.word /* 0x52 => */ 0x3235
.word /* 0x53 => */ 0x3335
.word /* 0x54 => */ 0x3435
.word /* 0x55 => */ 0x3535
.word /* 0x56 => */ 0x3635
.word /* 0x57 => */ 0x3735
.word /* 0x58 => */ 0x3835
.word /* 0x59 => */ 0x3935
.word /* 0x5a => */ 0x6135
.word /* 0x5b => */ 0x6235
.word /* 0x5c => */ 0x6335
.word /* 0x5d => */ 0x6435
.word /* 0x5e => */ 0x6535
.word /* 0x5f => */ 0x6635
.word /* 0x60 => */ 0x3036
.word /* 0x61 => */ 0x3136
.word /* 0x62 => */ 0x3236
.word /* 0x63 => */ 0x3336
.word /* 0x64 => */ 0x3436
.word /* 0x65 => */ 0x3536
.word /* 0x66 => */ 0x3636
.word /* 0x67 => */ 0x3736
.word /* 0x68 => */ 0x3836
.word /* 0x69 => */ 0x3936
.word /* 0x6a => */ 0x6136
.word /* 0x6b => */ 0x6236
.word /* 0x6c => */ 0x6336
.word /* 0x6d => */ 0x6436
.word /* 0x6e => */ 0x6536
.word /* 0x6f => */ 0x6636
.word /* 0x70 => */ 0x3037
.word /* 0x71 => */ 0x3137
.word /* 0x72 => */ 0x3237
.word /* 0x73 => */ 0x3337
.word /* 0x74 => */ 0x3437
.word /* 0x75 => */ 0x3537
.word /* 0x76 => */ 0x3637
.word /* 0x77 => */ 0x3737
.word /* 0x78 => */ 0x3837
.word /* 0x79 => */ 0x3937
.word /* 0x7a => */ 0x6137
.word /* 0x7b => */ 0x6237
.word /* 0x7c => */ 0x6337
.word /* 0x7d => */ 0x6437
.word /* 0x7e => */ 0x6537
.word /* 0x7f => */ 0x6637
.word /* 0x80 => */ 0x3038
.word /* 0x81 => */ 0x3138
.word /* 0x82 => */ 0x3238
.word /* 0x83 => */ 0x3338
.word /* 0x84 => */ 0x3438
.word /* 0x85 => */ 0x3538
.word /* 0x86 => */ 0x3638
.word /* 0x87 => */ 0x3738
.word /* 0x88 => */ 0x3838
.word /* 0x89 => */ 0x3938
.word /* 0x8a => */ 0x6138
.word /* 0x8b => */ 0x6238
.word /* 0x8c => */ 0x6338
.word /* 0x8d => */ 0x6438
.word /* 0x8e => */ 0x6538
.word /* 0x8f => */ 0x6638
.word /* 0x90 => */ 0x3039
.word /* 0x91 => */ 0x3139
.word /* 0x92 => */ 0x3239
.word /* 0x93 => */ 0x3339
.word /* 0x94 => */ 0x3439
.word /* 0x95 => */ 0x3539
.word /* 0x96 => */ 0x3639
.word /* 0x97 => */ 0x3739
.word /* 0x98 => */ 0x3839
.word /* 0x99 => */ 0x3939
.word /* 0x9a => */ 0x6139
.word /* 0x9b => */ 0x6239
.word /* 0x9c => */ 0x6339
.word /* 0x9d => */ 0x6439
.word /* 0x9e => */ 0x6539
.word /* 0x9f => */ 0x6639
.word /* 0xa0 => */ 0x3061
.word /* 0xa1 => */ 0x3161
.word /* 0xa2 => */ 0x3261
.word /* 0xa3 => */ 0x3361
.word /* 0xa4 => */ 0x3461
.word /* 0xa5 => */ 0x3561
.word /* 0xa6 => */ 0x3661
.word /* 0xa7 => */ 0x3761
.word /* 0xa8 => */ 0x3861
.word /* 0xa9 => */ 0x3961
.word /* 0xaa => */ 0x6161
.word /* 0xab => */ 0x6261
.word /* 0xac => */ 0x6361
.word /* 0xad => */ 0x6461
.word /* 0xae => */ 0x6561
.word /* 0xaf => */ 0x6661
.word /* 0xb0 => */ 0x3062
.word /* 0xb1 => */ 0x3162
.word /* 0xb2 => */ 0x3262
.word /* 0xb3 => */ 0x3362
.word /* 0xb4 => */ 0x3462
.word /* 0xb5 => */ 0x3562
.word /* 0xb6 => */ 0x3662
.word /* 0xb7 => */ 0x3762
.word /* 0xb8 => */ 0x3862
.word /* 0xb9 => */ 0x3962
.word /* 0xba => */ 0x6162
.word /* 0xbb => */ 0x6262
.word /* 0xbc => */ 0x6362
.word /* 0xbd => */ 0x6462
.word /* 0xbe => */ 0x6562
.word /* 0xbf => */ 0x6662
.word /* 0xc0 => */ 0x3063
.word /* 0xc1 => */ 0x3163
.word /* 0xc2 => */ 0x3263
.word /* 0xc3 => */ 0x3363
.word /* 0xc4 => */ 0x3463
.word /* 0xc5 => */ 0x3563
.word /* 0xc6 => */ 0x3663
.word /* 0xc7 => */ 0x3763
.word /* 0xc8 => */ 0x3863
.word /* 0xc9 => */ 0x3963
.word /* 0xca => */ 0x6163
.word /* 0xcb => */ 0x6263
.word /* 0xcc => */ 0x6363
.word /* 0xcd => */ 0x6463
.word /* 0xce => */ 0x6563
.word /* 0xcf => */ 0x6663
.word /* 0xd0 => */ 0x3064
.word /* 0xd1 => */ 0x3164
.word /* 0xd2 => */ 0x3264
.word /* 0xd3 => */ 0x3364
.word /* 0xd4 => */ 0x3464
.word /* 0xd5 => */ 0x3564
.word /* 0xd6 => */ 0x3664
.word /* 0xd7 => */ 0x3764
.word /* 0xd8 => */ 0x3864
.word /* 0xd9 => */ 0x3964
.word /* 0xda => */ 0x6164
.word /* 0xdb => */ 0x6264
.word /* 0xdc => */ 0x6364
.word /* 0xdd => */ 0x6464
.word /* 0xde => */ 0x6564
.word /* 0xdf => */ 0x6664
.word /* 0xe0 => */ 0x3065
.word /* 0xe1 => */ 0x3165
.word /* 0xe2 => */ 0x3265
.word /* 0xe3 => */ 0x3365
.word /* 0xe4 => */ 0x3465
.word /* 0xe5 => */ 0x3565
.word /* 0xe6 => */ 0x3665
.word /* 0xe7 => */ 0x3765
.word /* 0xe8 => */ 0x3865
.word /* 0xe9 => */ 0x3965
.word /* 0xea => */ 0x6165
.word /* 0xeb => */ 0x6265
.word /* 0xec => */ 0x6365
.word /* 0xed => */ 0x6465
.word /* 0xee => */ 0x6565
.word /* 0xef => */ 0x6665
.word /* 0xf0 => */ 0x3066
.word /* 0xf1 => */ 0x3166
.word /* 0xf2 => */ 0x3266
.word /* 0xf3 => */ 0x3366
.word /* 0xf4 => */ 0x3466
.word /* 0xf5 => */ 0x3566
.word /* 0xf6 => */ 0x3666
.word /* 0xf7 => */ 0x3766
.word /* 0xf8 => */ 0x3866
.word /* 0xf9 => */ 0x3966
.word /* 0xfa => */ 0x6166
.word /* 0xfb => */ 0x6266
.word /* 0xfc => */ 0x6366
.word /* 0xfd => */ 0x6466
.word /* 0xfe => */ 0x6566
.word /* 0xff => */ 0x6666