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
|
/* ieee754-sf.S single-precision floating point support for ARM
Copyright (C) 2003, 2004 Free Software Foundation, Inc.
Contributed by Nicolas Pitre (nico@cam.org)
This file is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.
In addition to the permissions in the GNU General Public License, the
Free Software Foundation gives you unlimited permission to link the
compiled version of this file into combinations with other programs,
and to distribute those combinations without any restriction coming
from the use of this file. (The General Public License restrictions
do apply in other respects; for example, they cover modification of
the file, and distribution when not linked into a combine
executable.)
This file 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
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; see the file COPYING. If not, write to
the Free Software Foundation, 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */
/*
* Notes:
*
* The goal of this code is to be as fast as possible. This is
* not meant to be easy to understand for the casual reader.
*
* Only the default rounding mode is intended for best performances.
* Exceptions aren't supported yet, but that can be added quite easily
* if necessary without impacting performances.
*/
#ifdef L_negsf2
ARM_FUNC_START negsf2
eor r0, r0, #0x80000000 @ flip sign bit
RET
FUNC_END negsf2
#endif
#ifdef L_addsubsf3
ARM_FUNC_START subsf3
eor r1, r1, #0x80000000 @ flip sign bit of second arg
#if defined(__thumb__) && !defined(__THUMB_INTERWORK__)
b 1f @ Skip Thumb-code prologue
#endif
ARM_FUNC_START addsf3
1: @ Compare both args, return zero if equal but the sign.
eor r2, r0, r1
teq r2, #0x80000000
beq LSYM(Lad_z)
@ If first arg is 0 or -0, return second arg.
@ If second arg is 0 or -0, return first arg.
bics r2, r0, #0x80000000
moveq r0, r1
bicnes r2, r1, #0x80000000
RETc(eq)
@ Mask out exponents.
mov ip, #0xff000000
and r2, r0, ip, lsr #1
and r3, r1, ip, lsr #1
@ If either of them is 255, result will be INF or NAN
teq r2, ip, lsr #1
teqne r3, ip, lsr #1
beq LSYM(Lad_i)
@ Compute exponent difference. Make largest exponent in r2,
@ corresponding arg in r0, and positive exponent difference in r3.
subs r3, r3, r2
addgt r2, r2, r3
eorgt r1, r0, r1
eorgt r0, r1, r0
eorgt r1, r0, r1
rsblt r3, r3, #0
@ If exponent difference is too large, return largest argument
@ already in r0. We need up to 25 bit to handle proper rounding
@ of 0x1p25 - 1.1.
cmp r3, #(25 << 23)
RETc(hi)
@ Convert mantissa to signed integer.
tst r0, #0x80000000
orr r0, r0, #0x00800000
bic r0, r0, #0xff000000
rsbne r0, r0, #0
tst r1, #0x80000000
orr r1, r1, #0x00800000
bic r1, r1, #0xff000000
rsbne r1, r1, #0
@ If exponent == difference, one or both args were denormalized.
@ Since this is not common case, rescale them off line.
teq r2, r3
beq LSYM(Lad_d)
LSYM(Lad_x):
@ Scale down second arg with exponent difference.
@ Apply shift one bit left to first arg and the rest to second arg
@ to simplify things later, but only if exponent does not become 0.
movs r3, r3, lsr #23
teqne r2, #(1 << 23)
movne r0, r0, lsl #1
subne r2, r2, #(1 << 23)
subne r3, r3, #1
@ Shift second arg into ip, keep leftover bits into r1.
mov ip, r1, asr r3
rsb r3, r3, #32
mov r1, r1, lsl r3
add r0, r0, ip @ the actual addition
@ We now have a 64 bit result in r0-r1.
@ Keep absolute value in r0-r1, sign in r3.
ands r3, r0, #0x80000000
bpl LSYM(Lad_p)
rsbs r1, r1, #0
rsc r0, r0, #0
@ Determine how to normalize the result.
LSYM(Lad_p):
cmp r0, #0x00800000
bcc LSYM(Lad_l)
cmp r0, #0x01000000
bcc LSYM(Lad_r0)
cmp r0, #0x02000000
bcc LSYM(Lad_r1)
@ Result needs to be shifted right.
movs r0, r0, lsr #1
mov r1, r1, rrx
add r2, r2, #(1 << 23)
LSYM(Lad_r1):
movs r0, r0, lsr #1
mov r1, r1, rrx
add r2, r2, #(1 << 23)
@ Our result is now properly aligned into r0, remaining bits in r1.
@ Round with MSB of r1. If halfway between two numbers, round towards
@ LSB of r0 = 0.
LSYM(Lad_r0):
add r0, r0, r1, lsr #31
teq r1, #0x80000000
biceq r0, r0, #1
@ Rounding may have added a new MSB. Adjust exponent.
@ That MSB will be cleared when exponent is merged below.
tst r0, #0x01000000
addne r2, r2, #(1 << 23)
@ Make sure we did not bust our exponent.
cmp r2, #(254 << 23)
bhi LSYM(Lad_o)
@ Pack final result together.
LSYM(Lad_e):
bic r0, r0, #0x01800000
orr r0, r0, r2
orr r0, r0, r3
RET
@ Result must be shifted left.
@ No rounding necessary since r1 will always be 0.
LSYM(Lad_l):
#if __ARM_ARCH__ < 5
movs ip, r0, lsr #12
moveq r0, r0, lsl #12
subeq r2, r2, #(12 << 23)
tst r0, #0x00ff0000
moveq r0, r0, lsl #8
subeq r2, r2, #(8 << 23)
tst r0, #0x00f00000
moveq r0, r0, lsl #4
subeq r2, r2, #(4 << 23)
tst r0, #0x00c00000
moveq r0, r0, lsl #2
subeq r2, r2, #(2 << 23)
tst r0, #0x00800000
moveq r0, r0, lsl #1
subeq r2, r2, #(1 << 23)
cmp r2, #0
bgt LSYM(Lad_e)
#else
clz ip, r0
sub ip, ip, #8
mov r0, r0, lsl ip
subs r2, r2, ip, lsl #23
bgt LSYM(Lad_e)
#endif
@ Exponent too small, denormalize result.
mvn r2, r2, asr #23
add r2, r2, #2
orr r0, r3, r0, lsr r2
RET
@ Fixup and adjust bit position for denormalized arguments.
@ Note that r2 must not remain equal to 0.
LSYM(Lad_d):
teq r2, #0
eoreq r0, r0, #0x00800000
addeq r2, r2, #(1 << 23)
eor r1, r1, #0x00800000
subne r3, r3, #(1 << 23)
b LSYM(Lad_x)
@ Result is x - x = 0, unless x is INF or NAN.
LSYM(Lad_z):
mov ip, #0xff000000
and r2, r0, ip, lsr #1
teq r2, ip, lsr #1
moveq r0, ip, asr #2
movne r0, #0
RET
@ Overflow: return INF.
LSYM(Lad_o):
orr r0, r3, #0x7f000000
orr r0, r0, #0x00800000
RET
@ At least one of r0/r1 is INF/NAN.
@ if r0 != INF/NAN: return r1 (which is INF/NAN)
@ if r1 != INF/NAN: return r0 (which is INF/NAN)
@ if r0 or r1 is NAN: return NAN
@ if opposite sign: return NAN
@ return r0 (which is INF or -INF)
LSYM(Lad_i):
teq r2, ip, lsr #1
movne r0, r1
teqeq r3, ip, lsr #1
RETc(ne)
movs r2, r0, lsl #9
moveqs r2, r1, lsl #9
teqeq r0, r1
orrne r0, r3, #0x00400000 @ NAN
RET
FUNC_END addsf3
FUNC_END subsf3
ARM_FUNC_START floatunsisf
mov r3, #0
b 1f
ARM_FUNC_START floatsisf
ands r3, r0, #0x80000000
rsbmi r0, r0, #0
1: teq r0, #0
RETc(eq)
mov r1, #0
mov r2, #((127 + 23) << 23)
tst r0, #0xfc000000
beq LSYM(Lad_p)
@ We need to scale the value a little before branching to code above.
tst r0, #0xf0000000
movne r1, r0, lsl #28
movne r0, r0, lsr #4
addne r2, r2, #(4 << 23)
tst r0, #0x0c000000
beq LSYM(Lad_p)
mov r1, r1, lsr #2
orr r1, r1, r0, lsl #30
mov r0, r0, lsr #2
add r2, r2, #(2 << 23)
b LSYM(Lad_p)
FUNC_END floatsisf
FUNC_END floatunsisf
#endif /* L_addsubsf3 */
#ifdef L_muldivsf3
ARM_FUNC_START mulsf3
@ Mask out exponents.
mov ip, #0xff000000
and r2, r0, ip, lsr #1
and r3, r1, ip, lsr #1
@ Trap any INF/NAN.
teq r2, ip, lsr #1
teqne r3, ip, lsr #1
beq LSYM(Lml_s)
@ Trap any multiplication by 0.
bics ip, r0, #0x80000000
bicnes ip, r1, #0x80000000
beq LSYM(Lml_z)
@ Shift exponents right one bit to make room for overflow bit.
@ If either of them is 0, scale denormalized arguments off line.
@ Then add both exponents together.
movs r2, r2, lsr #1
teqne r3, #0
beq LSYM(Lml_d)
LSYM(Lml_x):
add r2, r2, r3, asr #1
@ Preserve final sign in r2 along with exponent for now.
teq r0, r1
orrmi r2, r2, #0x8000
@ Convert mantissa to unsigned integer.
bic r0, r0, #0xff000000
bic r1, r1, #0xff000000
orr r0, r0, #0x00800000
orr r1, r1, #0x00800000
#if __ARM_ARCH__ < 4
@ Well, no way to make it shorter without the umull instruction.
@ We must perform that 24 x 24 -> 48 bit multiplication by hand.
stmfd sp!, {r4, r5}
mov r4, r0, lsr #16
mov r5, r1, lsr #16
bic r0, r0, #0x00ff0000
bic r1, r1, #0x00ff0000
mul ip, r4, r5
mul r3, r0, r1
mul r0, r5, r0
mla r0, r4, r1, r0
adds r3, r3, r0, lsl #16
adc ip, ip, r0, lsr #16
ldmfd sp!, {r4, r5}
#else
umull r3, ip, r0, r1 @ The actual multiplication.
#endif
@ Put final sign in r0.
mov r0, r2, lsl #16
bic r2, r2, #0x8000
@ Adjust result if one extra MSB appeared.
@ The LSB may be lost but this never changes the result in this case.
tst ip, #(1 << 15)
addne r2, r2, #(1 << 22)
movnes ip, ip, lsr #1
movne r3, r3, rrx
@ Apply exponent bias, check range for underflow.
subs r2, r2, #(127 << 22)
ble LSYM(Lml_u)
@ Scale back to 24 bits with rounding.
@ r0 contains sign bit already.
orrs r0, r0, r3, lsr #23
adc r0, r0, ip, lsl #9
@ If halfway between two numbers, rounding should be towards LSB = 0.
mov r3, r3, lsl #9
teq r3, #0x80000000
biceq r0, r0, #1
@ Note: rounding may have produced an extra MSB here.
@ The extra bit is cleared before merging the exponent below.
tst r0, #0x01000000
addne r2, r2, #(1 << 22)
@ Check for exponent overflow
cmp r2, #(255 << 22)
bge LSYM(Lml_o)
@ Add final exponent.
bic r0, r0, #0x01800000
orr r0, r0, r2, lsl #1
RET
@ Result is 0, but determine sign anyway.
LSYM(Lml_z):
eor r0, r0, r1
bic r0, r0, #0x7fffffff
RET
@ Check if denormalized result is possible, otherwise return signed 0.
LSYM(Lml_u):
cmn r2, #(24 << 22)
RETc(le)
@ Find out proper shift value.
mvn r1, r2, asr #22
subs r1, r1, #7
bgt LSYM(Lml_ur)
@ Shift value left, round, etc.
add r1, r1, #32
orrs r0, r0, r3, lsr r1
rsb r1, r1, #32
adc r0, r0, ip, lsl r1
mov ip, r3, lsl r1
teq ip, #0x80000000
biceq r0, r0, #1
RET
@ Shift value right, round, etc.
@ Note: r1 must not be 0 otherwise carry does not get set.
LSYM(Lml_ur):
orrs r0, r0, ip, lsr r1
adc r0, r0, #0
rsb r1, r1, #32
mov ip, ip, lsl r1
teq r3, #0
teqeq ip, #0x80000000
biceq r0, r0, #1
RET
@ One or both arguments are denormalized.
@ Scale them leftwards and preserve sign bit.
LSYM(Lml_d):
teq r2, #0
and ip, r0, #0x80000000
1: moveq r0, r0, lsl #1
tsteq r0, #0x00800000
subeq r2, r2, #(1 << 22)
beq 1b
orr r0, r0, ip
teq r3, #0
and ip, r1, #0x80000000
2: moveq r1, r1, lsl #1
tsteq r1, #0x00800000
subeq r3, r3, #(1 << 23)
beq 2b
orr r1, r1, ip
b LSYM(Lml_x)
@ One or both args are INF or NAN.
LSYM(Lml_s):
teq r0, #0x0
teqne r1, #0x0
teqne r0, #0x80000000
teqne r1, #0x80000000
beq LSYM(Lml_n) @ 0 * INF or INF * 0 -> NAN
teq r2, ip, lsr #1
bne 1f
movs r2, r0, lsl #9
bne LSYM(Lml_n) @ NAN * <anything> -> NAN
1: teq r3, ip, lsr #1
bne LSYM(Lml_i)
movs r3, r1, lsl #9
bne LSYM(Lml_n) @ <anything> * NAN -> NAN
@ Result is INF, but we need to determine its sign.
LSYM(Lml_i):
eor r0, r0, r1
@ Overflow: return INF (sign already in r0).
LSYM(Lml_o):
and r0, r0, #0x80000000
orr r0, r0, #0x7f000000
orr r0, r0, #0x00800000
RET
@ Return NAN.
LSYM(Lml_n):
mov r0, #0x7f000000
orr r0, r0, #0x00c00000
RET
FUNC_END mulsf3
ARM_FUNC_START divsf3
@ Mask out exponents.
mov ip, #0xff000000
and r2, r0, ip, lsr #1
and r3, r1, ip, lsr #1
@ Trap any INF/NAN or zeroes.
teq r2, ip, lsr #1
teqne r3, ip, lsr #1
bicnes ip, r0, #0x80000000
bicnes ip, r1, #0x80000000
beq LSYM(Ldv_s)
@ Shift exponents right one bit to make room for overflow bit.
@ If either of them is 0, scale denormalized arguments off line.
@ Then substract divisor exponent from dividend''s.
movs r2, r2, lsr #1
teqne r3, #0
beq LSYM(Ldv_d)
LSYM(Ldv_x):
sub r2, r2, r3, asr #1
@ Preserve final sign into ip.
eor ip, r0, r1
@ Convert mantissa to unsigned integer.
@ Dividend -> r3, divisor -> r1.
mov r3, #0x10000000
movs r1, r1, lsl #9
mov r0, r0, lsl #9
beq LSYM(Ldv_1)
orr r1, r3, r1, lsr #4
orr r3, r3, r0, lsr #4
@ Initialize r0 (result) with final sign bit.
and r0, ip, #0x80000000
@ Ensure result will land to known bit position.
cmp r3, r1
subcc r2, r2, #(1 << 22)
movcc r3, r3, lsl #1
@ Apply exponent bias, check range for over/underflow.
add r2, r2, #(127 << 22)
cmn r2, #(24 << 22)
RETc(le)
cmp r2, #(255 << 22)
bge LSYM(Lml_o)
@ The actual division loop.
mov ip, #0x00800000
1: cmp r3, r1
subcs r3, r3, r1
orrcs r0, r0, ip
cmp r3, r1, lsr #1
subcs r3, r3, r1, lsr #1
orrcs r0, r0, ip, lsr #1
cmp r3, r1, lsr #2
subcs r3, r3, r1, lsr #2
orrcs r0, r0, ip, lsr #2
cmp r3, r1, lsr #3
subcs r3, r3, r1, lsr #3
orrcs r0, r0, ip, lsr #3
movs r3, r3, lsl #4
movnes ip, ip, lsr #4
bne 1b
@ Check if denormalized result is needed.
cmp r2, #0
ble LSYM(Ldv_u)
@ Apply proper rounding.
cmp r3, r1
addcs r0, r0, #1
biceq r0, r0, #1
@ Add exponent to result.
bic r0, r0, #0x00800000
orr r0, r0, r2, lsl #1
RET
@ Division by 0x1p*: let''s shortcut a lot of code.
LSYM(Ldv_1):
and ip, ip, #0x80000000
orr r0, ip, r0, lsr #9
add r2, r2, #(127 << 22)
cmp r2, #(255 << 22)
bge LSYM(Lml_o)
cmp r2, #0
orrgt r0, r0, r2, lsl #1
RETc(gt)
cmn r2, #(24 << 22)
movle r0, ip
RETc(le)
orr r0, r0, #0x00800000
mov r3, #0
@ Result must be denormalized: prepare parameters to use code above.
@ r3 already contains remainder for rounding considerations.
LSYM(Ldv_u):
bic ip, r0, #0x80000000
and r0, r0, #0x80000000
mvn r1, r2, asr #22
add r1, r1, #2
b LSYM(Lml_ur)
@ One or both arguments are denormalized.
@ Scale them leftwards and preserve sign bit.
LSYM(Ldv_d):
teq r2, #0
and ip, r0, #0x80000000
1: moveq r0, r0, lsl #1
tsteq r0, #0x00800000
subeq r2, r2, #(1 << 22)
beq 1b
orr r0, r0, ip
teq r3, #0
and ip, r1, #0x80000000
2: moveq r1, r1, lsl #1
tsteq r1, #0x00800000
subeq r3, r3, #(1 << 23)
beq 2b
orr r1, r1, ip
b LSYM(Ldv_x)
@ One or both arguments is either INF, NAN or zero.
LSYM(Ldv_s):
mov ip, #0xff000000
teq r2, ip, lsr #1
teqeq r3, ip, lsr #1
beq LSYM(Lml_n) @ INF/NAN / INF/NAN -> NAN
teq r2, ip, lsr #1
bne 1f
movs r2, r0, lsl #9
bne LSYM(Lml_n) @ NAN / <anything> -> NAN
b LSYM(Lml_i) @ INF / <anything> -> INF
1: teq r3, ip, lsr #1
bne 2f
movs r3, r1, lsl #9
bne LSYM(Lml_n) @ <anything> / NAN -> NAN
b LSYM(Lml_z) @ <anything> / INF -> 0
2: @ One or both arguments are 0.
bics r2, r0, #0x80000000
bne LSYM(Lml_i) @ <non_zero> / 0 -> INF
bics r3, r1, #0x80000000
bne LSYM(Lml_z) @ 0 / <non_zero> -> 0
b LSYM(Lml_n) @ 0 / 0 -> NAN
FUNC_END divsf3
#endif /* L_muldivsf3 */
#ifdef L_cmpsf2
ARM_FUNC_START gtsf2
ARM_FUNC_ALIAS gesf2 gtsf2
mov r3, #-1
b 1f
ARM_FUNC_START ltsf2
ARM_FUNC_ALIAS lesf2 ltsf2
mov r3, #1
b 1f
ARM_FUNC_START cmpsf2
ARM_FUNC_ALIAS nesf2 cmpsf2
ARM_FUNC_ALIAS eqsf2 cmpsf2
mov r3, #1 @ how should we specify unordered here?
1: @ Trap any INF/NAN first.
mov ip, #0xff000000
and r2, r1, ip, lsr #1
teq r2, ip, lsr #1
and r2, r0, ip, lsr #1
teqne r2, ip, lsr #1
beq 3f
@ Test for equality.
@ Note that 0.0 is equal to -0.0.
2: orr r3, r0, r1
bics r3, r3, #0x80000000 @ either 0.0 or -0.0
teqne r0, r1 @ or both the same
moveq r0, #0
RETc(eq)
@ Check for sign difference. The N flag is set if it is the case.
@ If so, return sign of r0.
movmi r0, r0, asr #31
orrmi r0, r0, #1
RETc(mi)
@ Compare exponents.
and r3, r1, ip, lsr #1
cmp r2, r3
@ Compare mantissa if exponents are equal
moveq r0, r0, lsl #9
cmpeq r0, r1, lsl #9
movcs r0, r1, asr #31
mvncc r0, r1, asr #31
orr r0, r0, #1
RET
@ Look for a NAN.
3: and r2, r1, ip, lsr #1
teq r2, ip, lsr #1
bne 4f
movs r2, r1, lsl #9
bne 5f @ r1 is NAN
4: and r2, r0, ip, lsr #1
teq r2, ip, lsr #1
bne 2b
movs ip, r0, lsl #9
beq 2b @ r0 is not NAN
5: mov r0, r3 @ return unordered code from r3.
RET
FUNC_END gesf2
FUNC_END gtsf2
FUNC_END lesf2
FUNC_END ltsf2
FUNC_END nesf2
FUNC_END eqsf2
FUNC_END cmpsf2
#endif /* L_cmpsf2 */
#ifdef L_unordsf2
ARM_FUNC_START unordsf2
mov ip, #0xff000000
and r2, r1, ip, lsr #1
teq r2, ip, lsr #1
bne 1f
movs r2, r1, lsl #9
bne 3f @ r1 is NAN
1: and r2, r0, ip, lsr #1
teq r2, ip, lsr #1
bne 2f
movs r2, r0, lsl #9
bne 3f @ r0 is NAN
2: mov r0, #0 @ arguments are ordered.
RET
3: mov r0, #1 @ arguments are unordered.
RET
FUNC_END unordsf2
#endif /* L_unordsf2 */
#ifdef L_fixsfsi
ARM_FUNC_START fixsfsi
movs r0, r0, lsl #1
RETc(eq) @ value is 0.
mov r1, r1, rrx @ preserve C flag (the actual sign)
@ check exponent range.
and r2, r0, #0xff000000
cmp r2, #(127 << 24)
movcc r0, #0 @ value is too small
RETc(cc)
cmp r2, #((127 + 31) << 24)
bcs 1f @ value is too large
mov r0, r0, lsl #7
orr r0, r0, #0x80000000
mov r2, r2, lsr #24
rsb r2, r2, #(127 + 31)
tst r1, #0x80000000 @ the sign bit
mov r0, r0, lsr r2
rsbne r0, r0, #0
RET
1: teq r2, #0xff000000
bne 2f
movs r0, r0, lsl #8
bne 3f @ r0 is NAN.
2: ands r0, r1, #0x80000000 @ the sign bit
moveq r0, #0x7fffffff @ the maximum signed positive si
RET
3: mov r0, #0 @ What should we convert NAN to?
RET
FUNC_END fixsfsi
#endif /* L_fixsfsi */
#ifdef L_fixunssfsi
ARM_FUNC_START fixunssfsi
movs r0, r0, lsl #1
movcss r0, #0 @ value is negative...
RETc(eq) @ ... or 0.
@ check exponent range.
and r2, r0, #0xff000000
cmp r2, #(127 << 24)
movcc r0, #0 @ value is too small
RETc(cc)
cmp r2, #((127 + 32) << 24)
bcs 1f @ value is too large
mov r0, r0, lsl #7
orr r0, r0, #0x80000000
mov r2, r2, lsr #24
rsb r2, r2, #(127 + 31)
mov r0, r0, lsr r2
RET
1: teq r2, #0xff000000
bne 2f
movs r0, r0, lsl #8
bne 3f @ r0 is NAN.
2: mov r0, #0xffffffff @ maximum unsigned si
RET
3: mov r0, #0 @ What should we convert NAN to?
RET
FUNC_END fixunssfsi
#endif /* L_fixunssfsi */
|