-
Notifications
You must be signed in to change notification settings - Fork 143
/
wiring.c
1803 lines (1694 loc) · 92 KB
/
wiring.c
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
/*
wiring.c - Partial implementation of the Wiring API
Originally part of Arduino - http://www.arduino.cc/
Copyright (c) 2005-2006 David A. Mellis
Copyright (c) 2018-2021 Spence Konde
This has been ported to modern AVRs (Arduino team did that)
Almost every part of it has since been rewritten for
megaTinyCore and DxCore. This is the megaTinyCore version, and is
part of megaTinyCore.
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General
Public License along with this library; if not, write to the
Free Software Foundation, Inc., 59 Temple Place, Suite 330,
Boston, MA 02111-1307 USA
*/
#include "wiring_private.h"
#include "util/delay.h"
void init_timers();
#ifndef F_CPU
#error "F_CPU not defined. F_CPU must always be defined as the clock frequency in Hz"
#endif
#ifndef CLOCK_SOURCE
#error "CLOCK_SOURCE not defined. Must be 0 for internal, 1 for crystal, or 2 for external clock"
#endif
/* __PeripheralControl is used to mark peripherals as being "taken over" by the user
* 0x40 = TIMERD0
* 0x10 = TIMERA0
* 0x08 = TIMERA1
* Implementation and use is not portable between cores - tradeoffs are made which
* trade generalizability for low resource use
*/
uint8_t __PeripheralControl = 0xFF;
// the prescaler is set so that timer ticks every 64 clock cycles, and the
// the overflow handler is called every 256 ticks.
/* Use prescale appropriate for system clock speed
* Detect conflict between wiring.c and timers.h if we spot them, as that indicates
* a defect in the core and would result in extremely bad behavior
*/
#if (F_CPU > 30000000) // use 256 divider when clocked over 30 MHz
#if defined(MILLIS_USE_TIMERA0) && (TIME_TRACKING_TIMER_DIVIDER != 256)
#error "wiring.c and timers.h want to set millis timer TCA0 to different divider"
#endif
#define TIMERA_PRESCALER_bm (TCA_SPLIT_CLKSEL_DIV256_gc)
#elif (F_CPU > 5000000) // use 64 divider unless it's 5 MHz or under
#if defined(MILLIS_USE_TIMERA0) && (TIME_TRACKING_TIMER_DIVIDER != 64)
#error "wiring.c and timers.h want to set millis timer TCA0 to different divider"
#endif
#define TIMERA_PRESCALER_bm (TCA_SPLIT_CLKSEL_DIV64_gc)
#elif (F_CPU > 1000000) // anything above 1 MHz
#if defined(MILLIS_USE_TIMERA0) && (TIME_TRACKING_TIMER_DIVIDER != 16)
#error "wiring.c and timers.h want to set millis timer TCA0 to different divider"
#endif
#define TIMERA_PRESCALER_bm (TCA_SPLIT_CLKSEL_DIV16_gc)
#else /* for 1 MHz and lower */
#if defined(MILLIS_USE_TIMERA0) && (TIME_TRACKING_TIMER_DIVIDER != 8)
#error "wiring.c and timers.h want to set millis timer TCA0 to different divider"
#endif
#define TIMERA_PRESCALER_bm (TCA_SPLIT_CLKSEL_DIV8_gc)
#endif
#ifndef MILLIS_USE_TIMERNONE
// volatile uint16_t microseconds_per_timer_overflow;
// volatile uint16_t microseconds_per_timer_tick;
// overflow count is tracked for all timer options, even the RTC
struct sTimer {
uint8_t intClear;
volatile uint8_t *intStatusReg;
};
#if defined(MILLIS_USE_TIMERRTC)
#define MILLIS_TIMER_VECTOR (RTC_CNT_vect)
const struct sTimer _timerS = {RTC_OVF_bm, &RTC.INTCTRL};
#else
// when TCD0 is used as millis source, this will be different from above, but 99 times out of 10, when a piece of code asks for clockCyclesPerMicrosecond(), they're asking about CLK_PER/CLK_MAIN/etc, not the unprescaled TCD0!
#if defined (MILLIS_USE_TIMERA0)
#ifndef TCA0
#error "Selected millis timer, TCA0 does not exist on this part."
#endif
#if defined(TCA_BUFFERED_3PIN)
#define MILLIS_TIMER_VECTOR (TCA0_OVF_vect)
const struct sTimer _timerS = {TCA_SINGLE_OVF_bm, &TCA0.SINGLE.INTFLAGS};
#else
#define MILLIS_TIMER_VECTOR (TCA0_HUNF_vect)
const struct sTimer _timerS = {TCA_SPLIT_HUNF_bm, &TCA0.SPLIT.INTFLAGS};
#endif
#elif defined (MILLIS_USE_TIMERA1)
#ifndef TCA1
#error "Selected millis timer, TCA1 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCA1_HUNF_vect)
const struct sTimer _timerS = {TCA_SPLIT_HUNF_bm, &TCA1.SPLIT.INTFLAGS};
#elif defined(MILLIS_USE_TIMERB0)
#ifndef TCB0
#error "Selected millis timer, TCB0 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCB0_INT_vect)
static volatile TCB_t *_timer = &TCB0;
const struct sTimer _timerS = {TCB_CAPT_bm, &TCB0.INTFLAGS};
#elif defined(MILLIS_USE_TIMERB1)
#ifndef TCB1
#error "Selected millis timer, TCB1 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCB1_INT_vect)
static volatile TCB_t *_timer = &TCB1;
const struct sTimer _timerS = {TCB_CAPT_bm, &TCB1.INTFLAGS};
#elif defined(MILLIS_USE_TIMERB2)
#ifndef TCB2
#error "Selected millis timer, TCB2 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCB2_INT_vect)
static volatile TCB_t *_timer = &TCB2;
const struct sTimer _timerS = {TCB_CAPT_bm, &TCB2.INTFLAGS};
#elif defined(MILLIS_USE_TIMERB3)
#ifndef TCB3
#error "Selected millis timer, TCB3 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCB3_INT_vect)
static volatile TCB_t *_timer = &TCB3;
const struct sTimer _timerS = {TCB_CAPT_bm, &TCB3.INTFLAGS};
#elif defined(MILLIS_USE_TIMERB4)
#ifndef TCB4
#error "Selected millis timer, TCB4 does not exist on this part."
#endif
#define MILLIS_TIMER_VECTOR (TCB4_INT_vect)
static volatile TCB_t *_timer = &TCB4;
const struct sTimer _timerS = {TCB_CAPT_bm, &TCB4.INTFLAGS};
#elif defined(MILLIS_USE_TIMERD0)
#ifndef TCD0
#error "Selected millis timer, TCD0, is only valid for 1-series tinyAVR"
#endif
#define MILLIS_TIMER_VECTOR (TCD0_OVF_vect)
const struct sTimer _timerS = {TCD_OVF_bm, &TCD0.INTFLAGS};
#else
#error "No millis timer selected, but not disabled - can't happen!".
#endif /* defined(MILLIS_USE_TIMER__) */
#endif /* defined(MILLIS_USE_TIMERRTC) */
#define ClockCyclesToMicroseconds(__a__) ((__a__) / (F_CPU / 1000000L))
#define FRACT_MAX (1000)
#define FRACT_INC (ClockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF)%1000)
#define MILLIS_INC (ClockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF)/1000)
struct sTimeMillis {
#if (defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4)) // Now TCB as millis source does not need fraction
volatile uint32_t timer_millis; // That's all we need to track here
#elif defined(MILLIS_USE_TIMERRTC) // RTC
volatile uint16_t timer_overflow_count;
#else // TCAx or TCD0
//volatile uint16_t timer_fract;
//volatile uint32_t timer_millis;
//volatile uint32_t timer_overflow_count;
volatile uint32_t timer_overflow_count;
volatile uint32_t timer_millis;
volatile uint16_t timer_fract;
#endif
} timingStruct;
// Now for the ISRs. This gets a little bit more interesting now...
#if defined (MILLIS_USE_TIMERRTC)
ISR(MILLIS_TIMER_VECTOR) {
// if RTC is used as timer, we only increment the overflow count
// Overflow count isn't used for TCB's
if (RTC.INTFLAGS & RTC_OVF_bm) {
timingStruct.timer_overflow_count++;
}
RTC.INTFLAGS = RTC_OVF_bm | RTC_CMP_bm; // clear flag
}
#else
ISR(MILLIS_TIMER_VECTOR, ISR_NAKED) {
// Common Interrupt header for TCB, TCA and TCD;
// Clears the Timer Interrupt flag and pushes the CPU Registers
// 7 words / 7 clocks
__asm__ __volatile__(
"push r24" "\n\t" // Free up one more register to load values into
"ldi r24, %[CLRFL]" "\n\t" // This is the TCB interrupt clear bitmap
"sts %[PTCLR], r24" "\n\t" // write to Timer interrupt status register to clear flag. 2 clocks for sts
"in r24, 0x3F" "\n\t" // Load SREG
"push r24" "\n\t" // and push it on the Stack
"push r30" "\n\t" // First we make room for the pointer to timingStruct by pushing the Z registers
"push r31" "\n\t" //
:: [CLRFL] "M" (_timerS.intClear),
[PTCLR] "m" (*_timerS.intStatusReg));
#if (defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4))
__asm__ __volatile__(
"ld r24, Z" "\n\t" // Z points to LSB of timer_millis, load the LSB
#if (F_CPU > 2000000) // if it's 1 or 2 MHz, millis timer overflows every 2ms, intentionally sacrificing resolution for reduced time spent in ISR
"subi r24, 0xFF" "\n\t" // sub immediate 0xFF is the same as to add 1. (There is no add immediate instruction, except add immediate to word)
#else
"subi r24, 0xFE" "\n\t" // sub immediate 0xFE is the same as to add 2
#endif
"st Z, r24" "\n\t" // Store incremented value back to Z
"ldd r24, Z+1" "\n\t" // now load the next higher byte
"sbci r24, 0xFF" "\n\t" // because this is sbci, it treats carry bit like subtraction, and unless we did just roll over with the last byte,
"std Z+1, r24" "\n\t" // carry will be cleared. Thus, sbci 0xFF after a subi pressed into service to add, is the same as adc r1 after an add
"ldd r24, Z+2" "\n\t" // which is what we want.
"sbci r24, 0xFF" "\n\t" // This gets repeated...
"std Z+2, r24" "\n\t" //
"ldd r24, Z+3" "\n\t" //
"sbci r24, 0xFF" "\n\t" //
"std Z+3, r24" "\n\t" // ... until all 4 bytes were handled, at 4 clocks and 3 words per byte -> 16 clocks
:: "z" (&timingStruct.timer_millis)
); // grrr, sublime highlights this as invalid syntax because it gets confused by the ifdef's and odd syntax on inline asm
/* ISR in C:
ISR (TCBx_INT_vect) { // x depends on user configuration
#if (F_CPU > 2000000)
timer_millis += 1;
#else
timer_millis += 2;
#endif
_timer->INTFLAGS = TCB_CAPT_bm; // reset Interrupt flag of TCBx
}
*/
#else // TCA0 or TCD0, also naked
/*
__asm__ __volatile__(
// ISR prologue (overall 10 words / 10 clocks (+ loading of Z)):
"push r25" "\n\t" // one extra Register needed
// timer_fract handling (8 words / 10 clocks):
"ldd r24, Z+8" "\n\t" // lo8(timingStruct.timer_fract).
"ldd r25, Z+9" "\n\t" // hi8(timingStruct.timer_fract)
"subi r24,%[LFRINC]" "\n\t" // use (0xFFFF - FRACT_INC) and use the lower and higher byte to add by subtraction
"sbci r25,%[HFRINC]" "\n\t" // can't use adiw since FRACT_INC might be >63
"std Z+8, r24" "\n\t" // lo8(timingStruct.timer_fract)
"std Z+9, r25" "\n\t" // hi8(timingStruct.timer_fract)
"subi r24,%[LFRMAX]" "\n\t" // subtract FRACT_MAX and see if it is lower
"sbci r25,%[HFRMAX]" "\n\t" //
#if MILLIS_INC != 0 // (6 words / 4 - 5 clocks, branches were optimize to create minimal diversion)
"brlo higher" "\n\t" // if FRAC_MAX was not reached,
"ldi r24, %[MIINC]" "\n\t" // load "normal" MILLIS_INC (0x00-MILLIS_INC)
"rjmp sub4" "\n\t" // avoid overwriting r24
"higher:"
#else // (4 words, 2 - 4 clocks)
"brlo sub_end" "\n\t" // if we know at compile time that MILLIS_INC is 0,
#endif // we don't have to check it at runtime, saving two insn (tst, branch)
"std Z+8, r24" "\n\t" // Overwrite the just stored value with the decremented value
"std Z+9, r25" "\n\t" // seems counter-intuitive, but it requires less registers
"ldi r24, %[MINCD]" "\n\t" // load MILLIS_INC that was decreased by 1 (0xFF-MILLIS_INC)
// timer_millis handling (12 words / 16 clocks):
"sub4:"
"ldd r25, Z+4" "\n\t" // lo16.lo8(timingStruct.timer_millis)
"sub r25, r24" "\n\t" //
"std Z+4, r25" "\n\t" //
"ldd r25, Z+5" "\n\t" // lo16.hi8(timingStruct.timer_millis)
"sbci r25, 0xFF" "\n\t" //
"std Z+5, r25" "\n\t" //
"ldd r25, Z+6" "\n\t" // hi16.lo8(timingStruct.timer_millis)
"sbci r25, 0xFF" "\n\t" //
"std Z+6, r25" "\n\t" //
"ldd r25, Z+7" "\n\t" // hi16.hi8(timingStruct.timer_millis)
"sbci r25, 0xFF" "\n\t" //
"std Z+7, r25" "\n\t" //
"sub_end:" // only used if MILLIS_INC == 0
// timer_overflow_count handling (12 words / 16 clocks):
"ldd r25, Z+0" "\n\t" // lo16.lo8(timingStruct.timer_overflow_count)
"subi r25, 0xFF" "\n\t" //
"std Z+0, r25" "\n\t" //
"ldd r25, Z+1" "\n\t" // lo16.hi8(timingStruct.timer_overflow_count)
"sbci r25, 0xFF" "\n\t" //
"std Z+1, r25" "\n\t" //
"ldd r25, Z+2" "\n\t" // hi16.lo8(timingStruct.timer_overflow_count)
"sbci r25, 0xFF" "\n\t" //
"std Z+2, r25" "\n\t" //
"ldd r25, Z+3" "\n\t" // hi16.hi8(timingStruct.timer_overflow_count)
"sbci r25, 0xFF" "\n\t" //
"std Z+3, r25" "\n\t" //
// ISR epilogue (7 words / 15/16 clocks):
"pop r25" "\n\t" // new: total 72 - 74 clocks, 55 words / 53 - 75 clocks and 53 words with MILLIS_INC == 0
:: "z" (&timingStruct), // old: total 77 - 79 clocks total, and 58 words, vs 104-112 clocks and 84 words
[LFRINC] "M" (((0x0000 - FRACT_INC) & 0xFF)),
[HFRINC] "M" (((0x0000 - FRACT_INC)>>8 & 0xFF)),
[LFRMAX] "M" ((FRACT_MAX & 0xFF)),
[HFRMAX] "M" ((FRACT_MAX>>8 & 0xFF)),
[MIINC] "M" ((0x0000 - MILLIS_INC) & 0xFF),
[MINCD] "M" ((0xFFFF - MILLIS_INC) & 0xFF)
);
*/
__asm__ __volatile__(
// ISR prologue (overall 10 words / 10 clocks (+ loading of Z)):
"push r25" "\n\t" // second byte
// timer_overflow_count handling (4 words / 4 + (18) + 1 = 23 clocks):
"set" "\n\t" // remember to go back here
"ldi r24, 0xFF" "\n\t" // first byte to be subtracted of 4, rest will be 0xFF
"rjmp sub4" "\n\t" // jump down to sub/sbci
"ovf_end:" "\n\t" // jump back to here afterwards
"clt" "\n\t" // make sure to not jump back again
// timer_fract handling (8 words / 10 clocks) (Z += 4):
"ldd r24, Z+4" "\n\t" // lo8(timingStruct.timer_fract).
"ldd r25, Z+5" "\n\t" // hi8(timingStruct.timer_fract)
"subi r24,%[LFRINC]" "\n\t" // use (0xFFFF - FRACT_INC) and use the lower and higher byte to add by subtraction
"sbci r25,%[HFRINC]" "\n\t" // can't use adiw since FRACT_INC might be >63
"std Z+4, r24" "\n\t" // lo8(timingStruct.timer_fract)
"std Z+5, r25" "\n\t" // hi8(timingStruct.timer_fract)
"subi r24,%[LFRMAX]" "\n\t" // subtract FRACT_MAX and see if it is lower
"sbci r25,%[HFRMAX]" "\n\t" //
#if MILLIS_INC > 0 // (6 words / 4 - 5 clocks, branches were optimize to create minimal diversion)
"brlo higher" "\n\t" // if FRAC_MAX was not reached,
"ldi r24, %[MIINC]" "\n\t" // load "normal" MILLIS_INC (0x00-MILLIS_INC)
"rjmp sub4" "\n\t" // avoid overwriting r24
"higher:"
#else // (4 words, 2 - 4 clocks)
"brlo sub_end" "\n\t" // if we know at compile time that MILLIS_INC is 0,
#endif // we don't have to check it at runtime, saving two insn (tst, branch)
"std Z+4, r24" "\n\t" // Overwrite the just stored value with the decremented value
"std Z+5, r25" "\n\t" // seems counter-intuitive, but it requires less registers
"ldi r24, %[MINCD]" "\n\t" // load MILLIS_INC that was decreased by 1 (0xFF-MILLIS_INC)
// subtracting 4 bytes from a dword (13 words / 17 clocks)
"sub4:"
"ld r25, Z" "\n\t" // lo16.lo8(timingStruct.timer_millis)
"sub r25, r24" "\n\t" //
"st Z+, r25" "\n\t" //
"ld r25, Z" "\n\t" // lo16.hi8
"sbci r25, 0xFF" "\n\t" //
"st Z+, r25" "\n\t" //
"ld r25, Z" "\n\t" // hi16.lo8
"sbci r25, 0xFF" "\n\t" //
"st Z+, r25" "\n\t" //
"ld r25, Z" "\n\t" // hi16.hi8
"sbci r25, 0xFF" "\n\t" //
"st Z+, r25" "\n\t" //
"brts ovf_end" "\n\t" // If T bit is set, we need to go back up
"sub_end:"
// ISR epilogue (7 words / 15/16 clocks):
"pop r25" "\n\t"
:: "z" (&timingStruct),
[LFRINC] "M" (((0x0000 - FRACT_INC) & 0xFF)),
[HFRINC] "M" (((0x0000 - FRACT_INC)>>8 & 0xFF)),
[LFRMAX] "M" ((FRACT_MAX & 0xFF)),
[HFRMAX] "M" ((FRACT_MAX>>8 & 0xFF)),
[MIINC] "M" ((0x0000 - MILLIS_INC) & 0xFF),
[MINCD] "M" ((0xFFFF - MILLIS_INC) & 0xFF)
);
#endif /* (defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4)) */
// Common ISR Epilogue for TCA, TCB and TCD, popping register in reverse Order
// 6 words, 14 clocks
__asm__ __volatile__(
"pop r31" "\n\t"
"pop r30" "\n\t" // 6 more clocks popping registers in reverse order.
"pop r24" "\n\t" // pop r24 to get the old SREG value - 2 clock
"out 0x3F, r24" "\n\t" // restore SREG - 1 clock
"pop r24" "\n\t"
"reti" "\n\t" // and 4 clocks for reti
::
);
}
#endif /* defined (MILLIS_USE_TIMERRTC)*/
/* Both millis and micros must take great care to prevent any kind of backward time travel.
*
* These values are unsigned, and should not decrease, except when they overflow. Hence when
* we compare a value with what we recorded previously and find the new value to be lower, it
* looks the same as it would 2^32 (4.2 billion) intervals in the future. Timeouts end prematurely
* and similar undesired behaviors occur.
*
* There are three hazardous things we read here:
* timer_millis, timer_overflow_count, and the timer count itself (TCxn.CNT).
* The normal variables need only be read with interrupts disabled, in case of an
* interrupt writing to it while we were reading it. AVRs are little-endian, so this would result
* in the low byte being read before the overflow and the high byte after, and hence a value
* higher than it should be for that call. Subsequent calls would return the right value.
*
* In the case of the timer value, it is more complicated.
* Here, the hardware at first glance seems to protect us (see "reading 16-bit registers" in the
* datasheet). But the register gets read in the interrupt, so we still need those disabled.
* There is an additional risk though that we get a value from after the timer has overflowed
* and since we disabled interrupts, the interrupt hasn't updated the overflow. We check the
* interrupt flag, and if it's set, we check whether the timer value we read was near overflow
* (the specifics vary by the timer - they're very different timers). If it isn't close to overflow
* but the flag is set, we must have read it after the overflow, so we compensate for the missed
* interrupt. If interrupts are disabled for long enough, this heuristic will be wrong, but in
* that case it is the user's fault, as this limitation is widely known and documentedm, as well
* as unavoidable. Failure to compensate looks like the inverse of the above case.
*
* (note that only micros reads the timer, and hence, only micros can experience backwards time
* travel due to interrupts being left disabled for too long, millis will just stop increasing.
*
* Both of these cause severe breakage everywhere. The first type is simple to avoid, but if
* missed can be more subtle, since it makes a big difference only if the byte where the read
* was interrupted rolled over. The second type is more obvious, potentially happening on every timer
* overflow, instead of just every 256th timer overflow, and when it does happen, anything waiting
* for a specific number of microseconds to pass that gets that value will do so.
* Though (see delay below) each incidence only short-circuits one ms of delay(), not the whole
* thing.
*
* All time time travel except for glitchs from disabling millis for too long should no longer
* be possible. If they are, that is a critical bug.
*/
unsigned long millis() {
// return timer_overflow_count; // for debugging timekeeping issues where these variables are out of scope from the sketch
unsigned long m;
// disable interrupts while we read timer_millis or we might get an
// inconsistent value (e.g. in the middle of a write to timer_millis)
uint8_t oldSREG = SREG;
cli();
#if defined(MILLIS_USE_TIMERRTC)
uint16_t rtccount = RTC.CNT;
m = timingStruct.timer_overflow_count;
if (RTC.INTFLAGS & RTC_OVF_bm) {
/* There has just been an overflow that hasn't been accounted for by the interrupt. Check if the high bit of counter is set.
* We just basically need to make sure that it didn't JUST roll over at the last couple of clocks. But this merthod is
* implemented very efficiently (just an sbrs) so it is more efficient than other approaches. If user code is leaving
* interrupts off nearly 30 seconds, they shouldn't be surprised. */
if (!(rtccount & 0x8000)) m++;
}
SREG = oldSREG;
m = (m << 16);
m += rtccount;
m = m - (m >> 5) + (m >> 7);
/* the compiler is incorrigible - it cannot be convinced not to copy m twice, shifting one 7 times and the other 5 times
* and wasting 35 clock cycles and several precious instruction words.
* What you want is for it to make one copy of m, shift it right 5 places, subtract, then rightshift it 2 more.
*/
/* Would this work?
uint32_t temp;
uint8_t cnt;
__asm__ __volatile__ (
"movw %A1, %A0 \n\t"
"movw %C1, %C0 \n\t" //make our copy
"ldi %2, 5" "\n\t"
"lsr %D1" "\n\t"
"ror %C1" "\n\t"
"ror %B1" "\n\t"
"ror %A1" "\n\t"
"dec %2" "\n\t"
"brne .-12" "\n\t"
"sub %A0, %A1" "\n\t"
"subc %B0, %B1" "\n\t"
"subc %C0, %C1" "\n\t"
"subc %D0, %D1" "\n\t"
"ldi %2, 2" "\n\t"
"lsr %D1" "\n\t"
"ror %C1" "\n\t"
"ror %B1" "\n\t"
"ror %A1" "\n\t"
"dec %2" "\n\t"
"brne .-12" "\n\t"
"add %A0, %A1" "\n\t"
"adc %B0, %B1" "\n\t"
"adc %C0, %C1" "\n\t"
"adc %D0, %D1" "\n\t"
: "+r" (m), "+r" (temp), "+d" (cnt)
);
*/
#else
m = timingStruct.timer_millis;
SREG = oldSREG;
#endif
return m;
}
#if !defined(MILLIS_USE_TIMERRTC)
unsigned long micros() {
uint32_t overflows, microseconds;
#if (defined(MILLIS_USE_TIMERD0) || (defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4)))
uint16_t ticks;
#else /* TCA */
uint8_t ticks;
#endif
uint8_t flags;
/* Save current state and disable interrupts */
uint8_t oldSREG = SREG;
cli(); /* INTERRUPTS OFF */
#if defined(MILLIS_USE_TIMERA0)
ticks = TCA0.SPLIT.HCNT;
flags = TCA0.SPLIT.INTFLAGS;
#elif defined(MILLIS_USE_TIMERD0)
TCD0.CTRLE = TCD_SCAPTUREA_bm;
while (!(TCD0.STATUS & TCD_CMDRDY_bm)); // wait for sync - should be only one iteration of this loop
flags = TCD0.INTFLAGS;
ticks = TCD0.CAPTUREA;
#else
ticks = _timer->CNT;
flags = _timer->INTFLAGS;
#endif // end getting ticks
/* If the timer overflow flag is raised, and the ticks we read are low, then the timer has rolled over but
* ISR has not fired. If we already read a high value of ticks, either we read it just before the overflow,
* so we shouldn't increment overflows, or interrupts are disabled and micros isn't expected to work so it
* doesn't matter.
* Get current number of overflows and timer count */
#if !((defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4)))
overflows = timingStruct.timer_overflow_count;
#else
overflows = timingStruct.timer_millis;
#endif
/* Turn interrupts back on, assuming they were on when micros was called. */
SREG = oldSREG; /* INTERRUPTS ON */
#if defined(MILLIS_USE_TIMERD0)
if ((flags & TCD_OVF_bm) && (ticks < 0x07)) {
#elif defined(MILLIS_USE_TIMERA0)
ticks = (TIME_TRACKING_TIMER_PERIOD) - ticks;
if ((flags & TCA_SPLIT_HUNF_bm) && (ticks < 0x04)) {
#else // timerb
if ((flags & TCB_CAPT_bm) && !(ticks & 0xFF00)) {
#endif
#if ((defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4)) && !(F_CPU > 2000000UL))
overflows +=2;
#else
overflows++;
#endif
} // end getting ticks
#if defined(MILLIS_USE_TIMERD0)
#if (F_CPU == 20000000UL || F_CPU == 10000000UL || F_CPU == 5000000UL)
uint8_t ticks_l = ticks >> 1;
ticks = ticks + ticks_l + ((ticks_l >> 2) - (ticks_l >> 4) + (ticks_l >> 7));
// + ticks +(ticks>>1)+(ticks>>3)-(ticks>>5)+(ticks>>8))
// speed optimization via doing math with smaller datatypes, since we know high byte is 1 or 0.
microseconds = overflows * (TIME_TRACKING_CYCLES_PER_OVF / 20) + ticks; // ticks value corrected above.
#else
microseconds = ((overflows * (TIME_TRACKING_CYCLES_PER_OVF / 16))
+ (ticks * (TIME_TRACKING_CYCLES_PER_OVF / 16 / TIME_TRACKING_TIMER_PERIOD)));
#endif
#if defined(CLOCK_TUNE_INTERNAL) && !(F_CPU == 16000000UL || F_CPU == 20000000UL || F_CPU == 8000000UL || F_CPU == 10000000UL || F_CPU == 4000000UL || F_CPU == 5000000UL)
#warning "TCD is not supported as a millis timing source when the oscillator is tuned to a frequency other than 16 or 20 MHz. Timing results will be wrong - use TCA0 or a TCB."
#endif
#elif (defined(MILLIS_USE_TIMERB0) || defined(MILLIS_USE_TIMERB1) || defined(MILLIS_USE_TIMERB2) || defined(MILLIS_USE_TIMERB3) || defined(MILLIS_USE_TIMERB4))
/* Ersatz Division for TCBs - now with inline assembly!
*
* It's well known that division is an operator you want to avoid like the plague on AVR.
* Not only is it slow, the execution time isn't even easy to analyze - it depends on the
* two opperands, particularly the divisor... so you can't just look at the generated
* assembly and count clock cycles, you've got to either time it expoerimentally with
* a representative set of sample data, or know how many times it will pass through the
* loops and then count clock cycles. If the operands aren't constant (if they are, you
* can probably manage to get it optimized away at compile time) your best hope is likely
* simulation, assuming you know enough about the values it will end up having to divide.
*
* Anyway. You don't want to be doing division. But that's what you need in order to
* appropriately scale the ick count from the prescaler-deprived TCBs. Since many users
* reconfigure the TCA for advanced PWM, using the TCA-prescaled clock was a non-starter
* particularly since many parts have only one TCA. But division can be approximated
* very closely if the divisor is constant using a series of bitshifts and addition/subtraction.
*
* The series of shifts was determined numerically with a spreadsheet that calculated the results for
* each value that could come from the initial round of rightshifts for any combination of
* bitshifts and provided a number of statistics to select based on. Backwards time travel must
* never happenb, or if it does, it must be a shorter backward step than micros runtime - 1 us
* otherwise delay() will break and timeouts can instantly expire when it is hit. Similarly,
* one wants to avoid large jumps forward, and cases where more consecutive "actual" times
* than absolutely necessary return the same value (time should flow at a constant rate).
* Finally, the artifacts of the calculation that are unavoidable should be distributed uniformly.
* Undershooting or overshooting 999 endpoint at the counter's maximum value is the usual
* source of large jumps (at the overflow point) in either direction. Terms should, as much as
* possible alternate between positive and negative to minimize artifacts.
*
* The most popular/important speeds are hand-implemented in assembly because the compiler
* was doing a miserable job of it - wasting 20-30% of the execution time and it's one of the few
* Arduino API functions that users will be surprised and dismayed to find running slowiy.
* Not only does it run faster on "normal" boards (the 16 MHz clock eliminates the need to divide
* DxCore offers many speeds where the math doesn't all optimize away to nothing like it does at
* 1/2/4/8/16/32.
*
* Do notice that we are replacing a smaller number of terms, and it's still much faster
* The 10's went from 5 term ersatz-division to 6, while 12's went from 5 terms to 9, yet still
* got a lot faster. The terrible twelves are the frequency most difficult to do this with.
* Ironically, one of the the two that are is easiest is 36, which is good enough with 3 and
* effectively exact (That "middle 12" is closer than the other 12's get with 9!)
* 25 also matches it. Maybe it's something 25 and 36 being perfect squares?
*
* The three problems were that:
* 1. Compiler generated code stubbornly insisted doing repeated shift operation in a loop
* with 3 cycle per iteration (the shift itself took only 2)
* 2. Compiler could not be convinced to do things that we know will always be < 255 as
* bytes. Sure, it wouldn't know that - it's not legal for it to do that on it's own.
* But even when I cast everything to uint8_t, it would shift a 16-bit value around
* unnecessarily.
* 3. It would distribute the ticks >> 4. That is, it wouldn't shift the value of
* ticks in place, even though it wasn't referenced after this because I was assigning
* the result to ticks, and hence the old value was "dead"
* Instead, it would copy it, shift the copy 3 or 4 places. Then when it needed the
* ticks >> 2, it would make a copy of the ORIGINAL and shift that 6 places,
* instead of copying the copy and shifting just 2 places.
*
* A general trend seems to be that the compiler is not smart enough to "reuse" an
* existing value that has already been shifted such that it's closer to the target.
* at least for multi-byte variables. This is not the worst example of it I've run into
* but the micros() function is a little bit sensitive to the execution time.
* Apparently people sometimes want to *do something* in response to the value it
* returns - and they seem to want to do that in a timely manner, otherwise they'd
* have not bothered to record a time so accurately...
*
* general algorithm in the assembly implementations is:
* start with ticks in a register pair, copy to r0, r1.
* rightshift it until we have the 0th term (closest power of 2).
* copy it to back to original location..
* continue rightshifting it, adding or subtracting from the original when we reach
* the appropriate terms.
* As soon as we've rightshifted the original enough times that we know it's < 256,
* we switch from lsr r1 ror r0 to just lsr r0. At the next term that we want to add
* we copy it to r1. Subsequent subtractions or additions are single-byte until we've got the last term.
* this time, we add r1 to r0 instead of the other way around.
* we will need to clear r1 anyway, but we do it now, since we need a known 0 to do the carry.
* we addthat to the ticks intermediate value to get the final ticks value, and drop back into C
* where we calculate overflows * 1000, the (now 0-999) ticks to it, and return it.
*
*/
// Oddball clock speeds
#if (F_CPU == 44000000UL) // Extreme overclocking
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks - /* (ticks >> 1) + */ (ticks >> 2) - (ticks >> 5) + /* (ticks >> 6) - */ (ticks >> 7)); // + (ticks >> 10)
#elif (F_CPU == 36000000UL) // 50% overclock!
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks - (ticks >> 3) + (ticks >> 6)); // - (ticks >> 9) + (ticks >> 10) // with 5 terms it is DEAD ON
#elif (F_CPU == 28000000UL) // Not supported by DxCore - nobody wants it.
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) - (ticks >> 3) + (ticks >> 5) - (ticks >> 6)); // + (ticks >> 8) - (ticks >> 9)
#elif (F_CPU == 14000000UL) // Not supported by DxCore - nobody wants it.
ticks = ticks >> 3;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) - (ticks >> 3) + (ticks >> 5) - (ticks >> 6)); // + (ticks >> 8) - (ticks >> 9)
#elif (F_CPU == 30000000UL) // Easy overclock
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks + (ticks >> 3) - (ticks >> 4) + (ticks >> 7) - (ticks >> 8)); // 5 terms is the optimal. Good but not as close as we get for most.
#elif (F_CPU == 27000000UL) // You'd think this one would be a flaming bitch right?
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) - (ticks >> 4) - (ticks >> 9)); // +0.1 average error with only 4 terms, minimal scatter... that's just not supposed to happen!
#elif (F_CPU == 25000000UL) // Barely overclocked.
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks + /* (ticks >> 1) -*/ (ticks >> 2) + /* (ticks >> 4) -*/ (ticks >> 5)); // DEAD ON with 5 terms
/* The Terrible Twelves (or threes) - Twelve may be a great number in a lot of ways... but here, it's actually 3 in disguise.
* NINE TERMS in the damned bitshift division expansion. And the result isn't even amazing. - it's worse than what can be done
* with just 5 terms for dividing by 36 or 25, or a mere 3 terms with 27... where you're dividing by 9, 12.5, and 13.5 respectively,
* or after the initial shifts, by 0.78125, 1.25 or 1.18, and comparable to the best series for division by 1.375 (44 MHz) or 0.9375 (30 MHz) which each have 7 terms,
* though it's better than the best possible for the division by 0.875 associated with 28 MHs clocks which is also a 7 term one.
* This is division by 0.75, which sounds like it should be the easiest out of the lot.
*
* This does the following:
* ticks = ticks >> (1, 2, 3, 4, or 6 for 3 MHz, 6 MHz, 12 MHz, 24 MHz, or 48 MHz)
* ticks = ticks + (ticks >> 1) - (ticks >> 2) + (ticks >> 3) - (ticks >> 4) + (ticks >> 5) - (ticks >> 6) + (ticks >> 7) - (ticks >> 9)
*
* Equivalent to :
* ticks = ticks / (1.5, 3, 6, 12, or 24)
*
* Division is way too slow, but we need to convert current timer ticks, which
* are are 0-2999, 0-5999, 0-11999, or 0-23999 into the 3 least significant digits
* of the number of microseconds so that it can be added to overflows * 1000.
*
* Runtime of the assembly is 28, 30, 32, or 34 clocks
* 3 and 6 MHz not a supported speed.
* 57 replaced with 30 save 27 clocks @ 12 = 2 us saved
* 67 replaced with 32 save 35 clocks @ 24 = 1.5us saved
* 77 replaced with 34 save 43 clocks @ 48 = 1 us saved
*/
#elif (F_CPU == 48000000UL || F_CPU == 24000000UL || F_CPU == 12000000UL || F_CPU == 6000000UL || F_CPU == 3000000UL)
__asm__ __volatile__(
"movw r0,%A0" "\n\t" // we copy ticks to r0 (temp_reg) and r1 (zero_reg) so we don't need to allocate more registers.
"lsr r1" "\n\t" // notice how at first, each shift takes insns. Compiler wants to use an upper register, ldi number of shifts
"ror r0" "\n\t" // into it, then lsr, ror, dec, breq (4 insn + 5 clocks per shift, and including the ldi, it's 5 insns + 5*shiftcount clocks)
#if (F_CPU != 3000000UL)
"lsr r1" "\n\t"
"ror r0" "\n\t"
#endif
#if (F_CPU == 12000000UL || F_CPU == 24000000UL || F_CPU == 48000000UL)
"lsr r1" "\n\t" // sacrifice 1 word for 9 clocks on the 12 MHz configuration
"ror r0" "\n\t"
#endif
#if (F_CPU == 24000000UL || F_CPU == 48000000UL)
"lsr r1" "\n\t" // sacrifice 3 words for 12 clocks on the 24 MHz configuration
"ror r0" "\n\t"
#endif
#if (F_CPU == 48000000UL)
"lsr r1" "\n\t" // sacrifice 5 words for 15 clocks on the 48 MHz configuration.
"ror r0" "\n\t"
#endif
"movw %A0,r0" "\n\t" // This is the value we call ticks, because that's what it was in old code.
"lsr r1" "\n\t" // we just copied the now shifted value back to original location.
"ror r0" "\n\t" // 2 words per shift still
"add %A0, r0" "\n\t" // we now have ticks >> 1, add it to original.
"adc %B0, r1" "\n\t" //
"lsr r1" "\n\t" //
"ror r0" "\n\t" // we now have ticks >> 2. Now it's under 250, and r1 is 0
"mov r1,r0" "\n\t" // so we copy the remaining value into r1.
"lsr r1 " "\n\t" // now it's only 1 insn/shift!
"sub r0,r1" "\n\t" // - ticks >> 3
"lsr r1" "\n\t"
"add r0,r1" "\n\t" // + ticks >> 4
"lsr r1" "\n\t"
"sub r0,r1" "\n\t" // - ticks >> 5
"lsr r1" "\n\t"
"add r0,r1" "\n\t" // + ticks >> 6
"lsr r1" "\n\t"
"sub r0,r1" "\n\t" // - ticks >> 7
"lsr r1" "\n\t"
"lsr r1" "\n\t"
"add r0,r1" "\n\t" // + ticks >> 9
"eor r1,r1" "\n\t" // clear out r1
"sub %A0,r0" "\n\t" // Add the sum of terms that fit in a byte to what was ticks in old code.
"sbc %B0,r1" "\n" // carry - see,this is why AVR needs a known zero.
: "+r" (ticks)); // Do the rest in C. ticks is a read/write operand.
microseconds = overflows * 1000 + ticks; // nice and clean.
/* The Troublesome Tens - I initially fumbled this after the **now** r1 is 0 line
* I did several dumb things - at first I thought it was my pointless moving and
* adding. But the real problem was that on that line, I'd just deleted the
* now unnecessary lsr r1, leaving the next as ror instead of lsr. So instead of pushing
* that bit into the void, it came back as the new high bit, causing the device to travel
* back in time. Unfortunately, a few hundred milliseconds isn't far back enough to
* snag a winning ticket for todays lotto, but more than than the execution time
* of micros is far enough back to thoroughly break delay() Even if I could just go back
* just far enough to tell myself where the bug was, I'd take it...
*
* This does the following:
* ticks = ticks >> (1, 2, 3, or 4 for 5 MHz, 10 MHz, 20 MHz, or 40 MHz)
* ticks = ticks - (ticks >> 2) + (ticks >> 4) - (ticks >> 6) + (ticks >> 8)
*
* Equivalent to:
* ticks = tick / (2.5, 5, 10, or 20)
* Division is way too slow, but we need to convert current timer ticks, which
* are 0-2499, 0-4999, 0-9999, or 0-19999, into the 3 least significant digits
* of the number of microseconds so that it can be added to overflows * 1000.
*
* Runtime is 23,25,27, or 29 clocks, savings vs the best I could do in C
*
* 33 replaced with 23 save 10 clocks @ 5 = 2 us saved
* 46 replaced with 25 save 21 clocks @ 10 = 2.5 us saved
* 56 replaced with 27 save 29 clocks @ 20 = 1.5 us saved
* 66 replaced with 29 save 37 clocks @ 40 = 1 us saved
*/
#elif (F_CPU == 40000000UL || F_CPU == 20000000UL || F_CPU == 10000000UL || F_CPU == 5000000UL)
__asm__ __volatile__(
"movw r0,%A0" "\n\t" // no savings until after the initial rightshifts at 5 MHz
"lsr r1" "\n\t"
"ror r0" "\n\t"
#if (F_CPU == 10000000UL || F_CPU == 20000000UL || F_CPU == 40000000UL)
"lsr r1" "\n\t" // sacrifice 1 word for 9 clocks at 10 MHz
"ror r0" "\n\t"
#endif
#if (F_CPU == 20000000UL || F_CPU == 40000000UL)
"lsr r1" "\n\t" // sacrifice 3 words for 12 clocks at 20 MHz
"ror r0" "\n\t"
#endif
#if (F_CPU == 40000000UL)
"lsr r1" "\n\t" // sacrifice 5 words for 15 clocks at 40 MHz
"ror r0" "\n\t"
#endif
"movw %A0,r0" "\n\t" // ticks
"lsr r1" "\n\t"
"ror r0" "\n\t"
"lsr r1" "\n\t"
"ror r0" "\n\t" // ticks >> 2.
"sub %A0, r0" "\n\t" // - ticks >> 2
"sbc %B0, r1" "\n\t" // It could be 312 so we can't do what we did for the 12's
"lsr r1" "\n\t"
"ror r0" "\n\t" // **now** r1 is 0.
"lsr r0" "\n\t"
"mov r1,r0" "\n\t" // + ticks >> 4
"lsr r1" "\n\t"
"lsr r1" "\n\t"
"sub r0,r1" "\n\t" // - ticks >> 6
"lsr r1" "\n\t"
"lsr r1" "\n\t"
"add r0,r1" "\n\t" // + ticks >> 8
"eor r1,r1" "\n\t" // restore zero_reg
"add %A0,r0" "\n\t" // add to the shifted ticks
"adc %B0,r1" "\n" // carry
: "+r" (ticks)); // Do the rest in C. ticks is a read/write operand.
microseconds = overflows * 1000 + ticks;
/* replaces:
#elif (F_CPU == 48000000UL) // Extreme overclocking
ticks = ticks >> 5;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) + (ticks >> 3) - (ticks >> 5)); // - (ticks >> 7)
#elif (F_CPU == 24000000UL) // max rated speed
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) + (ticks >> 3) - (ticks >> 5)); // - (ticks >> 7)
#elif (F_CPU == 12000000UL)
ticks = ticks >> 3;
microseconds = overflows * 1000 + (ticks + (ticks >> 2) + (ticks >> 3) - (ticks >> 5)); // - (ticks >> 7)
// Never was an implementation for 3 or 6, but it's obvious what the old style implementation would be,
#elif (F_CPU == 40000000UL) // overclocked aggressively
ticks = ticks >> 4;
microseconds = overflows * 1000 + (ticks - (ticks >> 2) + (ticks >> 4) - (ticks >> 6)); // + (ticks >> 8)
#elif (F_CPU == 20000000UL)
ticks = ticks >> 3;
microseconds = overflows * 1000 + (ticks - (ticks >> 2) + (ticks >> 4) - (ticks >> 6)); // + (ticks >> 8)
#elif (F_CPU == 10000000UL)
ticks = ticks >> 2;
microseconds = overflows * 1000 + (ticks - (ticks >> 2) + (ticks >> 4) - (ticks >> 6)); // + (ticks >> 8)
#elif (F_CPU == 5000000UL)
ticks = ticks >> 1;
microseconds = overflows * 1000 + (ticks - (ticks >> 2) + (ticks >> 4) - (ticks >> 6)); // + (ticks >> 8)
*/
// powers of 2 - and a catchall for parts without dedicated implementations. It gives wrong results, but
// it also doesn't take forever like doing division would.
#elif (F_CPU == 32000000UL || F_CPU > 24000000UL)
microseconds = overflows * 1000 + (ticks >> 4);
#elif (F_CPU == 16000000UL || F_CPU > 12000000UL)
microseconds = overflows * 1000 + (ticks >> 3);
#elif (F_CPU == 8000000UL || F_CPU > 6000000UL)
microseconds = overflows * 1000 + (ticks >> 2);
#elif (F_CPU == 4000000UL || F_CPU >= 3000000UL)
microseconds = overflows * 1000 + (ticks >> 1);
#else //(F_CPU == 1000000UL || F_CPU == 2000000UL) - here clock is running at system clock instead of half system clock.
// and hence overflows only once per 2ms. On 2 MHz
// also works at 2MHz, since we use CLKPER for 1MHz vs CLKPER/2 for all others.
microseconds = overflows * 1000 + ticks;
#endif
#if !((F_CPU == 48000000UL || F_CPU == 36000000UL || F_CPU == 24000000UL || F_CPU == 12000000UL || /* multiples of 12 */ \
F_CPU == 40000000UL || F_CPU == 30000000UL || F_CPU == 20000000UL || F_CPU == 10000000UL || /* multiples of 10 */ \
F_CPU == 32000000UL || F_CPU == 16000000UL || F_CPU == 8000000UL || F_CPU == 4000000UL || /* powers of 2 */ \
F_CPU == 2000000UL || F_CPU == 1000000UL || F_CPU == 25000000UL || F_CPU == 5000000UL || /* powers of 2 cont, 25, 5 */ \
F_CPU == 44000000UL || F_CPU == 28000000UL || F_CPU == 14000000UL || F_CPU == 3000000UL || /* oddball frequencies */ \
F_CPU == 27000000UL)&& /* warn fools who messed with the timers.h file too and expected that the core would sort out how */ \
((TIME_TRACKING_TIMER_DIVIDER == 2 && TIME_TRACKING_TICKS_PER_OVF == F_CPU/2000) || /*how to make the timer work correctly*/ \
(TIME_TRACKING_TIMER_DIVIDER == 1 && (TIME_TRACKING_TICKS_PER_OVF == F_CPU/500 && F_CPU == 1000000) || (TIME_TRACKING_TICKS_PER_OVF == F_CPU/1000 && F_CPU == 2000000))))
/* without them implementing it. No such luck */
#warning "Millis timer (TCBn) at this frequency and/or configuration unsupported, micros() will return totally bogus values."
#endif
#else // Done with TCB
#if (F_CPU == 30000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ ((ticks * 2) + ((uint16_t)(ticks >> 3)));
#elif (F_CPU == 28000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ ((ticks * 2) + ((uint16_t)(ticks >> 2) + (ticks >> 5)));
#elif (F_CPU == 25000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 2 + ((uint16_t)(ticks >> 1) + (ticks >> 4)));
#elif (F_CPU == 24000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 3 - ((uint16_t)(ticks >> 2) - (ticks >> 4) - (ticks >> 5)));
#elif (F_CPU == 20000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 3 + ((uint16_t)(ticks >> 2) - (ticks >> 4)));
#elif (F_CPU == 28000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 4 + ((uint16_t)(ticks >> 1) + (ticks >> 4) + (ticks >> 5)));
#elif (F_CPU == 12000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 5 + ((uint16_t)(ticks >> 2) + (ticks >> 4) + (ticks >> 5)));
#elif (F_CPU == 10000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 64)
microseconds = (overflows * clockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ ((ticks << 3) - ((uint16_t)(ticks << 1) + (ticks >> 1) - (ticks >> 3)));
#elif (F_CPU == 5000000UL && TIME_TRACKING_TICKS_PER_OVF == 255 && TIME_TRACKING_TIMER_DIVIDER == 16)
microseconds = (overflows * millisClockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * 3 + ((uint16_t)(ticks >> 2) - (ticks >> 4)));
#else
#if (TIME_TRACKING_TIMER_DIVIDER%(F_CPU/1000000))
#warning "Millis timer (TCA0) at this frequency unsupported, micros() will return bogus values."
#endif
microseconds = ((overflows * millisClockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF))
+ (ticks * (millisClockCyclesToMicroseconds(TIME_TRACKING_CYCLES_PER_OVF) / TIME_TRACKING_TIMER_PERIOD)));
#endif
#endif // end of timer-specific part of micros calculations
return microseconds;
}
#else // MILLIS_USE_TIMERRTC is defined, so we don't have micros
/* We do not have a timebase sufficiently accurate to give microsecond timing. In fact, we barely have millisecond timing available
* The microsecond delay counts clock cycles, and so it does still work. It is planned that a future library will switch the millis
* pause millis before sleeping and turn on the RTC, tracking the passage of time to a much coarser resolution with that, and turn
* it back on when waking from sleep, so people can keep time while sleeping without sacrificing micros().
* In any event, as of 2.4.3 we now provide the stub below, which we hope is more useful than being told that micros() isn't defined.
*/
unsigned long micros() {
badCall("microsecond timekeeping is not supported when the RTC is used as the sole timekeeping timer (though delayMicroseconds() is)");
return -1;
}
#endif
#else // MILLIS_USE_TIMERNONE defined - we have neither of these functions.
/* Uses should not call millis() or micros() if the core timekeeping has been disabled. Usually, encountering this error either means
* that they disabled millis earlier for some other sketch, and the preferences were saved with that - or that they are using a library
* with a dependence on the timekeeping facilities. Sometimes these are meaningful, other times it is only for a feature that isn't
* being used, or to catch a particular corner case (see: tinyNeoPixel, very end of show() for an example).
* As of 2.4.3 we provide the stubs below, which we hope is more useful than being told that millis or micros isn't defined.
*/
unsigned long micros() {
badCall("micros() is not available because it has been disabled through the tools -> millis()/micros() menu");
return -1;
}
unsigned long millis() {
badCall("millis() is not available because it has been disabled through the tools -> millis()/micros() menu");
return -1;
}
#endif // MILLIS_USE_TIMERNONE code
/* delay()
* So what do you WANT in a good delay function?
* First, obviously you want it to delay things. You do not want it to block interrupts (then a long one would throw off
* timekeeping, miss inputs, and so on). And you want the compiled size to not be prohibitive for the part.
* The reason it's so important wrt. interrupts is that in Arduino standard delay(), if an interrupt fires in the middle,
* will still end at the same time - it is "interrupt insensitive". Whenever a delay is using the builtin _delay_ms()
* if that is interrupted it has no way of knowing time has passed. Now hopefully you're not spending so much time in
* an ISR that this is significant, but it is still undesirable.
*
* For the unfortunate souls using small-flash parts, the flash usage becomes a major problem - why is it such a space-hog?
* Because it has to pull in micros(), which is bulky even with the division turned into bitshifts... RTC has same problem
* with millis(), the conversion of 1024ths of a second to 1000ths is a killer, even with the bitshift tricks,
* and the compiler seems really stupid about how it handles it; I can't keep it from making an extra copy of the 32-bit
* value, which ALSO requires 4 more push and pop operations to get registers it can use.
*
* Now we will use one of three delay() implementations:
* If you have 16k+ your delay is the standard one, it pulls in micros(), yes, but you may well already have grabbed
* that for your sketch already, and the delay is more accurate and fully interrupt insensitive, and you can afford
* the memory. For RTC users they will get the analogous implementation that is based on millis.
* Users with millis disabled, or with less than 16k flash and using RTC will get the implementation based on _delay_ms().
* Everyone else (flash under 16k but millis enabled via non-RTC timer) will get the light version which calls _delay_ms()
* if the delay is under 16 ms to get less flash usage, and calculates the delay using **millis** not micros otherwise,
* saving over 100b of flash. The reason for the split is that the limited granularity of millis introduces an error in
* the delay duration of up to 1ms. That doesn't matter much when you call delay(1000) on an internal clock that's within
* 1% on a good day. It matters greatly when you call delay(1); */
#if defined(MILLIS_USE_TIMERNONE) || (PROGMEM_SIZE < 16384 && defined(MILLIS_USE_TIMERRTC))
void delay(uint32_t ms) { /* Interrupts will prolong this delay */
if (__builtin_constant_p(ms)) {
_delay_ms(ms);
} else {
while (ms--) {
_delay_ms(1);
}
}
}
#elif (PROGMEM_SIZE >= 16384 && !defined(MILLIS_USE_TIMERRTC))
void delay(uint32_t ms) { /* Interrupts will not prolong this less flash-efficient delay */
uint16_t start = (uint16_t) micros();
while (ms > 0) {
while (((uint16_t) micros() - start) >= 1000 && ms) {
ms-- ;
start += 1000;
}
}
}
#else
void delay(uint32_t ms) {
if (__builtin_constant_p(ms) && ms < 16) {
_delay_ms(ms);
} else if (ms < 16) {
while(ms--) {
_delay_ms(1);
}
} else {
uint32_t start = millis();
while (millis() - start < ms);
}
}
#endif
inline __attribute__((always_inline)) void delayMicroseconds(unsigned int us) {
// This function gets optimized away, but to what depends on whether us is constant.
if (__builtin_constant_p(us)) {
_delay_us(us); // Constant microseconds use the avr-libc _delay_us() which is highly accurate for all values and efficient!
} else { // If it is not, we have to use the Arduino style implementation.
_delayMicroseconds(us);
}
}
/* delayMicroseconds() when delay is not a compile-time known constant.
* Delay for the given number of microseconds. This is UGLY AS SIN and explicitly depends on function call
* overhead for very short delays.
* High clock speeds shouldn't return immediately for a 1us delay - we can instead only drop a fraction of a us
* 48, 44, 40, and 32 drop 1/2 us, and 36 drops 2/3rds.
* Note that us ceases to be in units of microseconds as soon as the function is entered; it gets turned into the loop counter.