summaryrefslogtreecommitdiffstats
path: root/arch/m68k/math-emu/fp_util.S
blob: a9f7f01290672feded67ad7b4bd6f662668355c8 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
/*
 * fp_util.S
 *
 * Copyright Roman Zippel, 1997.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, and the entire permission notice in its entirety,
 *    including the disclaimer of warranties.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. The name of the author may not be used to endorse or promote
 *    products derived from this software without specific prior
 *    written permission.
 *
 * ALTERNATIVELY, this product may be distributed under the terms of
 * the GNU General Public License, in which case the provisions of the GPL are
 * required INSTEAD OF the above restrictions.  (This clause is
 * necessary due to a potential bad interaction between the GPL and
 * the restrictions contained in a BSD-style copyright.)
 *
 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
 * OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include <linux/config.h>
#include "fp_emu.h"

/*
 * Here are lots of conversion and normalization functions mainly
 * used by fp_scan.S
 * Note that these functions are optimized for "normal" numbers,
 * these are handled first and exit as fast as possible, this is
 * especially important for fp_normalize_ext/fp_conv_ext2ext, as
 * it's called very often.
 * The register usage is optimized for fp_scan.S and which register
 * is currently at that time unused, be careful if you want change
 * something here. %d0 and %d1 is always usable, sometimes %d2 (or
 * only the lower half) most function have to return the %a0
 * unmodified, so that the caller can immediately reuse it.
 */

	.globl	fp_ill, fp_end

	| exits from fp_scan:
	| illegal instruction
fp_ill:
	printf	,"fp_illegal\n"
	rts
	| completed instruction
fp_end:
	tst.l	(TASK_MM-8,%a2)
	jmi	1f
	tst.l	(TASK_MM-4,%a2)
	jmi	1f
	tst.l	(TASK_MM,%a2)
	jpl	2f
1:	printf	,"oops:%p,%p,%p\n",3,%a2@(TASK_MM-8),%a2@(TASK_MM-4),%a2@(TASK_MM)
2:	clr.l	%d0
	rts

	.globl	fp_conv_long2ext, fp_conv_single2ext
	.globl	fp_conv_double2ext, fp_conv_ext2ext
	.globl	fp_normalize_ext, fp_normalize_double
	.globl	fp_normalize_single, fp_normalize_single_fast
	.globl	fp_conv_ext2double, fp_conv_ext2single
	.globl	fp_conv_ext2long, fp_conv_ext2short
	.globl	fp_conv_ext2byte
	.globl	fp_finalrounding_single, fp_finalrounding_single_fast
	.globl	fp_finalrounding_double
	.globl	fp_finalrounding, fp_finaltest, fp_final

/*
 * First several conversion functions from a source operand
 * into the extended format. Note, that only fp_conv_ext2ext
 * normalizes the number and is always called after the other
 * conversion functions, which only move the information into
 * fp_ext structure.
 */

	| fp_conv_long2ext:
	|
	| args:	%d0 = source (32-bit long)
	|	%a0 = destination (ptr to struct fp_ext)

fp_conv_long2ext:
	printf	PCONV,"l2e: %p -> %p(",2,%d0,%a0
	clr.l	%d1			| sign defaults to zero
	tst.l	%d0
	jeq	fp_l2e_zero		| is source zero?
	jpl	1f			| positive?
	moveq	#1,%d1
	neg.l	%d0
1:	swap	%d1
	move.w	#0x3fff+31,%d1
	move.l	%d1,(%a0)+		| set sign / exp
	move.l	%d0,(%a0)+		| set mantissa
	clr.l	(%a0)
	subq.l	#8,%a0			| restore %a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	rts
	| source is zero
fp_l2e_zero:
	clr.l	(%a0)+
	clr.l	(%a0)+
	clr.l	(%a0)
	subq.l	#8,%a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	rts

	| fp_conv_single2ext
	| args:	%d0 = source (single-precision fp value)
	|	%a0 = dest (struct fp_ext *)

fp_conv_single2ext:
	printf	PCONV,"s2e: %p -> %p(",2,%d0,%a0
	move.l	%d0,%d1
	lsl.l	#8,%d0			| shift mantissa
	lsr.l	#8,%d1			| exponent / sign
	lsr.l	#7,%d1
	lsr.w	#8,%d1
	jeq	fp_s2e_small		| zero / denormal?
	cmp.w	#0xff,%d1		| NaN / Inf?
	jeq	fp_s2e_large
	bset	#31,%d0			| set explizit bit
	add.w	#0x3fff-0x7f,%d1	| re-bias the exponent.
9:	move.l	%d1,(%a0)+		| fp_ext.sign, fp_ext.exp
	move.l	%d0,(%a0)+		| high lword of fp_ext.mant
	clr.l	(%a0)			| low lword = 0
	subq.l	#8,%a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	rts
	| zeros and denormalized
fp_s2e_small:
	| exponent is zero, so explizit bit is already zero too
	tst.l	%d0
	jeq	9b
	move.w	#0x4000-0x7f,%d1
	jra	9b
	| infinities and NAN
fp_s2e_large:
	bclr	#31,%d0			| clear explizit bit
	move.w	#0x7fff,%d1
	jra	9b

fp_conv_double2ext:
#ifdef FPU_EMU_DEBUG
	getuser.l %a1@(0),%d0,fp_err_ua2,%a1
	getuser.l %a1@(4),%d1,fp_err_ua2,%a1
	printf	PCONV,"d2e: %p%p -> %p(",3,%d0,%d1,%a0
#endif
	getuser.l (%a1)+,%d0,fp_err_ua2,%a1
	move.l	%d0,%d1
	lsl.l	#8,%d0			| shift high mantissa
	lsl.l	#3,%d0
	lsr.l	#8,%d1			| exponent / sign
	lsr.l	#7,%d1
	lsr.w	#5,%d1
	jeq	fp_d2e_small		| zero / denormal?
	cmp.w	#0x7ff,%d1		| NaN / Inf?
	jeq	fp_d2e_large
	bset	#31,%d0			| set explizit bit
	add.w	#0x3fff-0x3ff,%d1	| re-bias the exponent.
9:	move.l	%d1,(%a0)+		| fp_ext.sign, fp_ext.exp
	move.l	%d0,(%a0)+
	getuser.l (%a1)+,%d0,fp_err_ua2,%a1
	move.l	%d0,%d1
	lsl.l	#8,%d0
	lsl.l	#3,%d0
	move.l	%d0,(%a0)
	moveq	#21,%d0
	lsr.l	%d0,%d1
	or.l	%d1,-(%a0)
	subq.l	#4,%a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	rts
	| zeros and denormalized
fp_d2e_small:
	| exponent is zero, so explizit bit is already zero too
	tst.l	%d0
	jeq	9b
	move.w	#0x4000-0x3ff,%d1
	jra	9b
	| infinities and NAN
fp_d2e_large:
	bclr	#31,%d0			| clear explizit bit
	move.w	#0x7fff,%d1
	jra	9b

	| fp_conv_ext2ext:
	| originally used to get longdouble from userspace, now it's
	| called before arithmetic operations to make sure the number
	| is normalized [maybe rename it?].
	| args:	%a0 = dest (struct fp_ext *)
	| returns 0 in %d0 for a NaN, otherwise 1

fp_conv_ext2ext:
	printf	PCONV,"e2e: %p(",1,%a0
	printx	PCONV,%a0@
	printf	PCONV,"), "
	move.l	(%a0)+,%d0
	cmp.w	#0x7fff,%d0		| Inf / NaN?
	jeq	fp_e2e_large
	move.l	(%a0),%d0
	jpl	fp_e2e_small		| zero / denorm?
	| The high bit is set, so normalization is irrelevant.
fp_e2e_checkround:
	subq.l	#4,%a0
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	move.b	(%a0),%d0
	jne	fp_e2e_round
#endif
	printf	PCONV,"%p(",1,%a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	moveq	#1,%d0
	rts
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
fp_e2e_round:
	fp_set_sr FPSR_EXC_INEX2
	clr.b	(%a0)
	move.w	(FPD_RND,FPDATA),%d2
	jne	fp_e2e_roundother	| %d2 == 0, round to nearest
	tst.b	%d0			| test guard bit
	jpl	9f			| zero is closer
	btst	#0,(11,%a0)		| test lsb bit
	jne	fp_e2e_doroundup	| round to infinity
	lsl.b	#1,%d0			| check low bits
	jeq	9f			| round to zero
fp_e2e_doroundup:
	addq.l	#1,(8,%a0)
	jcc	9f
	addq.l	#1,(4,%a0)
	jcc	9f
	move.w	#0x8000,(4,%a0)
	addq.w	#1,(2,%a0)
9:	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
fp_e2e_roundother:
	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	1f			| %d2 > 2, round to +infinity
	tst.b	(1,%a0)			| to -inf
	jne	fp_e2e_doroundup	| negative, round to infinity
	jra	9b			| positive, round to zero
1:	tst.b	(1,%a0)			| to +inf
	jeq	fp_e2e_doroundup	| positive, round to infinity
	jra	9b			| negative, round to zero
#endif
	| zeros and subnormals:
	| try to normalize these anyway.
fp_e2e_small:
	jne	fp_e2e_small1		| high lword zero?
	move.l	(4,%a0),%d0
	jne	fp_e2e_small2
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	clr.l	%d0
	move.b	(-4,%a0),%d0
	jne	fp_e2e_small3
#endif
	| Genuine zero.
	clr.w	-(%a0)
	subq.l	#2,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	moveq	#1,%d0
	rts
	| definitely subnormal, need to shift all 64 bits
fp_e2e_small1:
	bfffo	%d0{#0,#32},%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
1:	move.w	%d2,(%a0)+
	move.w	%d1,%d2
	jeq	fp_e2e_checkround
	| fancy 64-bit double-shift begins here
	lsl.l	%d2,%d0
	move.l	%d0,(%a0)+
	move.l	(%a0),%d0
	move.l	%d0,%d1
	lsl.l	%d2,%d0
	move.l	%d0,(%a0)
	neg.w	%d2
	and.w	#0x1f,%d2
	lsr.l	%d2,%d1
	or.l	%d1,-(%a0)
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
fp_e2e_extra1:
	clr.l	%d0
	move.b	(-4,%a0),%d0
	neg.w	%d2
	add.w	#24,%d2
	jcc	1f
	clr.b	(-4,%a0)
	lsl.l	%d2,%d0
	or.l	%d0,(4,%a0)
	jra	fp_e2e_checkround
1:	addq.w	#8,%d2
	lsl.l	%d2,%d0
	move.b	%d0,(-4,%a0)
	lsr.l	#8,%d0
	or.l	%d0,(4,%a0)
#endif
	jra	fp_e2e_checkround
	| pathologically small subnormal
fp_e2e_small2:
	bfffo	%d0{#0,#32},%d1
	add.w	#32,%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Beyond pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
1:	move.w	%d2,(%a0)+
	ext.l	%d1
	jeq	fp_e2e_checkround
	clr.l	(4,%a0)
	sub.w	#32,%d2
	jcs	1f
	lsl.l	%d1,%d0			| lower lword needs only to be shifted
	move.l	%d0,(%a0)		| into the higher lword
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	clr.l	%d0
	move.b	(-4,%a0),%d0
	clr.b	(-4,%a0)
	neg.w	%d1
	add.w	#32,%d1
	bfins	%d0,(%a0){%d1,#8}
#endif
	jra	fp_e2e_checkround
1:	neg.w	%d1			| lower lword is splitted between
	bfins	%d0,(%a0){%d1,#32}	| higher and lower lword
#ifndef CONFIG_M68KFPU_EMU_EXTRAPREC
	jra	fp_e2e_checkround
#else
	move.w	%d1,%d2
	jra	fp_e2e_extra1
	| These are extremely small numbers, that will mostly end up as zero
	| anyway, so this is only important for correct rounding.
fp_e2e_small3:
	bfffo	%d0{#24,#8},%d1
	add.w	#40,%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
1:	move.w	%d2,(%a0)+
	ext.l	%d1
	jeq	fp_e2e_checkround
	cmp.w	#8,%d1
	jcs	2f
1:	clr.b	(-4,%a0)
	sub.w	#64,%d1
	jcs	1f
	add.w	#24,%d1
	lsl.l	%d1,%d0
	move.l	%d0,(%a0)
	jra	fp_e2e_checkround
1:	neg.w	%d1
	bfins	%d0,(%a0){%d1,#8}
	jra	fp_e2e_checkround
2:	lsl.l	%d1,%d0
	move.b	%d0,(-4,%a0)
	lsr.l	#8,%d0
	move.b	%d0,(7,%a0)
	jra	fp_e2e_checkround
#endif
1:	move.l	%d0,%d1			| lower lword is splitted between
	lsl.l	%d2,%d0			| higher and lower lword
	move.l	%d0,(%a0)
	move.l	%d1,%d0
	neg.w	%d2
	add.w	#32,%d2
	lsr.l	%d2,%d0
	move.l	%d0,-(%a0)
	jra	fp_e2e_checkround
	| Infinities and NaNs
fp_e2e_large:
	move.l	(%a0)+,%d0
	jne	3f
1:	tst.l	(%a0)
	jne	4f
	moveq	#1,%d0
2:	subq.l	#8,%a0
	printf	PCONV,"%p(",1,%a0
	printx	PCONV,%a0@
	printf	PCONV,")\n"
	rts
	| we have maybe a NaN, shift off the highest bit
3:	lsl.l	#1,%d0
	jeq	1b
	| we have a NaN, clear the return value
4:	clrl	%d0
	jra	2b


/*
 * Normalization functions.  Call these on the output of general
 * FP operators, and before any conversion into the destination
 * formats. fp_normalize_ext has always to be called first, the
 * following conversion functions expect an already normalized
 * number.
 */

	| fp_normalize_ext:
	| normalize an extended in extended (unpacked) format, basically
	| it does the same as fp_conv_ext2ext, additionally it also does
	| the necessary postprocessing checks.
	| args:	%a0 (struct fp_ext *)
	| NOTE: it does _not_ modify %a0/%a1 and the upper word of %d2

fp_normalize_ext:
	printf	PNORM,"ne: %p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,"), "
	move.l	(%a0)+,%d0
	cmp.w	#0x7fff,%d0		| Inf / NaN?
	jeq	fp_ne_large
	move.l	(%a0),%d0
	jpl	fp_ne_small		| zero / denorm?
	| The high bit is set, so normalization is irrelevant.
fp_ne_checkround:
	subq.l	#4,%a0
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	move.b	(%a0),%d0
	jne	fp_ne_round
#endif
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
fp_ne_round:
	fp_set_sr FPSR_EXC_INEX2
	clr.b	(%a0)
	move.w	(FPD_RND,FPDATA),%d2
	jne	fp_ne_roundother	| %d2 == 0, round to nearest
	tst.b	%d0			| test guard bit
	jpl	9f			| zero is closer
	btst	#0,(11,%a0)		| test lsb bit
	jne	fp_ne_doroundup		| round to infinity
	lsl.b	#1,%d0			| check low bits
	jeq	9f			| round to zero
fp_ne_doroundup:
	addq.l	#1,(8,%a0)
	jcc	9f
	addq.l	#1,(4,%a0)
	jcc	9f
	addq.w	#1,(2,%a0)
	move.w	#0x8000,(4,%a0)
9:	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
fp_ne_roundother:
	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	1f			| %d2 > 2, round to +infinity
	tst.b	(1,%a0)			| to -inf
	jne	fp_ne_doroundup		| negative, round to infinity
	jra	9b			| positive, round to zero
1:	tst.b	(1,%a0)			| to +inf
	jeq	fp_ne_doroundup		| positive, round to infinity
	jra	9b			| negative, round to zero
#endif
	| Zeros and subnormal numbers
	| These are probably merely subnormal, rather than "denormalized"
	|  numbers, so we will try to make them normal again.
fp_ne_small:
	jne	fp_ne_small1		| high lword zero?
	move.l	(4,%a0),%d0
	jne	fp_ne_small2
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	clr.l	%d0
	move.b	(-4,%a0),%d0
	jne	fp_ne_small3
#endif
	| Genuine zero.
	clr.w	-(%a0)
	subq.l	#2,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
	| Subnormal.
fp_ne_small1:
	bfffo	%d0{#0,#32},%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
	fp_set_sr FPSR_EXC_UNFL
1:	move.w	%d2,(%a0)+
	move.w	%d1,%d2
	jeq	fp_ne_checkround
	| This is exactly the same 64-bit double shift as seen above.
	lsl.l	%d2,%d0
	move.l	%d0,(%a0)+
	move.l	(%a0),%d0
	move.l	%d0,%d1
	lsl.l	%d2,%d0
	move.l	%d0,(%a0)
	neg.w	%d2
	and.w	#0x1f,%d2
	lsr.l	%d2,%d1
	or.l	%d1,-(%a0)
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
fp_ne_extra1:
	clr.l	%d0
	move.b	(-4,%a0),%d0
	neg.w	%d2
	add.w	#24,%d2
	jcc	1f
	clr.b	(-4,%a0)
	lsl.l	%d2,%d0
	or.l	%d0,(4,%a0)
	jra	fp_ne_checkround
1:	addq.w	#8,%d2
	lsl.l	%d2,%d0
	move.b	%d0,(-4,%a0)
	lsr.l	#8,%d0
	or.l	%d0,(4,%a0)
#endif
	jra	fp_ne_checkround
	| May or may not be subnormal, if so, only 32 bits to shift.
fp_ne_small2:
	bfffo	%d0{#0,#32},%d1
	add.w	#32,%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Beyond pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
	fp_set_sr FPSR_EXC_UNFL
1:	move.w	%d2,(%a0)+
	ext.l	%d1
	jeq	fp_ne_checkround
	clr.l	(4,%a0)
	sub.w	#32,%d1
	jcs	1f
	lsl.l	%d1,%d0			| lower lword needs only to be shifted
	move.l	%d0,(%a0)		| into the higher lword
#ifdef CONFIG_M68KFPU_EMU_EXTRAPREC
	clr.l	%d0
	move.b	(-4,%a0),%d0
	clr.b	(-4,%a0)
	neg.w	%d1
	add.w	#32,%d1
	bfins	%d0,(%a0){%d1,#8}
#endif
	jra	fp_ne_checkround
1:	neg.w	%d1			| lower lword is splitted between
	bfins	%d0,(%a0){%d1,#32}	| higher and lower lword
#ifndef CONFIG_M68KFPU_EMU_EXTRAPREC
	jra	fp_ne_checkround
#else
	move.w	%d1,%d2
	jra	fp_ne_extra1
	| These are extremely small numbers, that will mostly end up as zero
	| anyway, so this is only important for correct rounding.
fp_ne_small3:
	bfffo	%d0{#24,#8},%d1
	add.w	#40,%d1
	move.w	-(%a0),%d2
	sub.w	%d1,%d2
	jcc	1f
	| Pathologically small, denormalize.
	add.w	%d2,%d1
	clr.w	%d2
1:	move.w	%d2,(%a0)+
	ext.l	%d1
	jeq	fp_ne_checkround
	cmp.w	#8,%d1
	jcs	2f
1:	clr.b	(-4,%a0)
	sub.w	#64,%d1
	jcs	1f
	add.w	#24,%d1
	lsl.l	%d1,%d0
	move.l	%d0,(%a0)
	jra	fp_ne_checkround
1:	neg.w	%d1
	bfins	%d0,(%a0){%d1,#8}
	jra	fp_ne_checkround
2:	lsl.l	%d1,%d0
	move.b	%d0,(-4,%a0)
	lsr.l	#8,%d0
	move.b	%d0,(7,%a0)
	jra	fp_ne_checkround
#endif
	| Infinities and NaNs, again, same as above.
fp_ne_large:
	move.l	(%a0)+,%d0
	jne	3f
1:	tst.l	(%a0)
	jne	4f
2:	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
	| we have maybe a NaN, shift off the highest bit
3:	move.l	%d0,%d1
	lsl.l	#1,%d1
	jne	4f
	clr.l	(-4,%a0)
	jra	1b
	| we have a NaN, test if it is signaling
4:	bset	#30,%d0
	jne	2b
	fp_set_sr FPSR_EXC_SNAN
	move.l	%d0,(-4,%a0)
	jra	2b

	| these next two do rounding as per the IEEE standard.
	| values for the rounding modes appear to be:
	| 0:	Round to nearest
	| 1:	Round to zero
	| 2:	Round to -Infinity
	| 3:	Round to +Infinity
	| both functions expect that fp_normalize was already
	| called (and extended argument is already normalized
	| as far as possible), these are used if there is different
	| rounding precision is selected and before converting
	| into single/double

	| fp_normalize_double:
	| normalize an extended with double (52-bit) precision
	| args:	 %a0 (struct fp_ext *)

fp_normalize_double:
	printf	PNORM,"nd: %p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,"), "
	move.l	(%a0)+,%d2
	tst.w	%d2
	jeq	fp_nd_zero		| zero / denormalized
	cmp.w	#0x7fff,%d2
	jeq	fp_nd_huge		| NaN / infinitive.
	sub.w	#0x4000-0x3ff,%d2	| will the exponent fit?
	jcs	fp_nd_small		| too small.
	cmp.w	#0x7fe,%d2
	jcc	fp_nd_large		| too big.
	addq.l	#4,%a0
	move.l	(%a0),%d0		| low lword of mantissa
	| now, round off the low 11 bits.
fp_nd_round:
	moveq	#21,%d1
	lsl.l	%d1,%d0			| keep 11 low bits.
	jne	fp_nd_checkround	| Are they non-zero?
	| nothing to do here
9:	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
	| Be careful with the X bit! It contains the lsb
	| from the shift above, it is needed for round to nearest.
fp_nd_checkround:
	fp_set_sr FPSR_EXC_INEX2	| INEX2 bit
	and.w	#0xf800,(2,%a0)		| clear bits 0-10
	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	jne	2f			| %d2 == 0, round to nearest
	tst.l	%d0			| test guard bit
	jpl	9b			| zero is closer
	| here we test the X bit by adding it to %d2
	clr.w	%d2			| first set z bit, addx only clears it
	addx.w	%d2,%d2			| test lsb bit
	| IEEE754-specified "round to even" behaviour.  If the guard
	| bit is set, then the number is odd, so rounding works like
	| in grade-school arithmetic (i.e. 1.5 rounds to 2.0)
	| Otherwise, an equal distance rounds towards zero, so as not
	| to produce an odd number.  This is strange, but it is what
	| the standard says.
	jne	fp_nd_doroundup		| round to infinity
	lsl.l	#1,%d0			| check low bits
	jeq	9b			| round to zero
fp_nd_doroundup:
	| round (the mantissa, that is) towards infinity
	add.l	#0x800,(%a0)
	jcc	9b			| no overflow, good.
	addq.l	#1,-(%a0)		| extend to high lword
	jcc	1f			| no overflow, good.
	| Yow! we have managed to overflow the mantissa.  Since this
	| only happens when %d1 was 0xfffff800, it is now zero, so
	| reset the high bit, and increment the exponent.
	move.w	#0x8000,(%a0)
	addq.w	#1,-(%a0)
	cmp.w	#0x43ff,(%a0)+		| exponent now overflown?
	jeq	fp_nd_large		| yes, so make it infinity.
1:	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
2:	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	3f			| %d2 > 2, round to +infinity
	| Round to +Inf or -Inf.  High word of %d2 contains the
	| sign of the number, by the way.
	swap	%d2			| to -inf
	tst.b	%d2
	jne	fp_nd_doroundup		| negative, round to infinity
	jra	9b			| positive, round to zero
3:	swap	%d2			| to +inf
	tst.b	%d2
	jeq	fp_nd_doroundup		| positive, round to infinity
	jra	9b			| negative, round to zero
	| Exponent underflow.  Try to make a denormal, and set it to
	| the smallest possible fraction if this fails.
fp_nd_small:
	fp_set_sr FPSR_EXC_UNFL		| set UNFL bit
	move.w	#0x3c01,(-2,%a0)	| 2**-1022
	neg.w	%d2			| degree of underflow
	cmp.w	#32,%d2			| single or double shift?
	jcc	1f
	| Again, another 64-bit double shift.
	move.l	(%a0),%d0
	move.l	%d0,%d1
	lsr.l	%d2,%d0
	move.l	%d0,(%a0)+
	move.l	(%a0),%d0
	lsr.l	%d2,%d0
	neg.w	%d2
	add.w	#32,%d2
	lsl.l	%d2,%d1
	or.l	%d1,%d0
	move.l	(%a0),%d1
	move.l	%d0,(%a0)
	| Check to see if we shifted off any significant bits
	lsl.l	%d2,%d1
	jeq	fp_nd_round		| Nope, round.
	bset	#0,%d0			| Yes, so set the "sticky bit".
	jra	fp_nd_round		| Now, round.
	| Another 64-bit single shift and store
1:	sub.w	#32,%d2
	cmp.w	#32,%d2			| Do we really need to shift?
	jcc	2f			| No, the number is too small.
	move.l	(%a0),%d0
	clr.l	(%a0)+
	move.l	%d0,%d1
	lsr.l	%d2,%d0
	neg.w	%d2
	add.w	#32,%d2
	| Again, check to see if we shifted off any significant bits.
	tst.l	(%a0)
	jeq	1f
	bset	#0,%d0			| Sticky bit.
1:	move.l	%d0,(%a0)
	lsl.l	%d2,%d1
	jeq	fp_nd_round
	bset	#0,%d0
	jra	fp_nd_round
	| Sorry, the number is just too small.
2:	clr.l	(%a0)+
	clr.l	(%a0)
	moveq	#1,%d0			| Smallest possible fraction,
	jra	fp_nd_round		| round as desired.
	| zero and denormalized
fp_nd_zero:
	tst.l	(%a0)+
	jne	1f
	tst.l	(%a0)
	jne	1f
	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts				| zero.  nothing to do.
	| These are not merely subnormal numbers, but true denormals,
	| i.e. pathologically small (exponent is 2**-16383) numbers.
	| It is clearly impossible for even a normal extended number
	| with that exponent to fit into double precision, so just
	| write these ones off as "too darn small".
1:	fp_set_sr FPSR_EXC_UNFL		| Set UNFL bit
	clr.l	(%a0)
	clr.l	-(%a0)
	move.w	#0x3c01,-(%a0)		| i.e. 2**-1022
	addq.l	#6,%a0
	moveq	#1,%d0
	jra	fp_nd_round		| round.
	| Exponent overflow.  Just call it infinity.
fp_nd_large:
	move.w	#0x7ff,%d0
	and.w	(6,%a0),%d0
	jeq	1f
	fp_set_sr FPSR_EXC_INEX2
1:	fp_set_sr FPSR_EXC_OVFL
	move.w	(FPD_RND,FPDATA),%d2
	jne	3f			| %d2 = 0 round to nearest
1:	move.w	#0x7fff,(-2,%a0)
	clr.l	(%a0)+
	clr.l	(%a0)
2:	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
3:	subq.w	#2,%d2
	jcs	5f			| %d2 < 2, round to zero
	jhi	4f			| %d2 > 2, round to +infinity
	tst.b	(-3,%a0)		| to -inf
	jne	1b
	jra	5f
4:	tst.b	(-3,%a0)		| to +inf
	jeq	1b
5:	move.w	#0x43fe,(-2,%a0)
	moveq	#-1,%d0
	move.l	%d0,(%a0)+
	move.w	#0xf800,%d0
	move.l	%d0,(%a0)
	jra	2b
	| Infinities or NaNs
fp_nd_huge:
	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts

	| fp_normalize_single:
	| normalize an extended with single (23-bit) precision
	| args:	 %a0 (struct fp_ext *)

fp_normalize_single:
	printf	PNORM,"ns: %p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,") "
	addq.l	#2,%a0
	move.w	(%a0)+,%d2
	jeq	fp_ns_zero		| zero / denormalized
	cmp.w	#0x7fff,%d2
	jeq	fp_ns_huge		| NaN / infinitive.
	sub.w	#0x4000-0x7f,%d2	| will the exponent fit?
	jcs	fp_ns_small		| too small.
	cmp.w	#0xfe,%d2
	jcc	fp_ns_large		| too big.
	move.l	(%a0)+,%d0		| get high lword of mantissa
fp_ns_round:
	tst.l	(%a0)			| check the low lword
	jeq	1f
	| Set a sticky bit if it is non-zero.  This should only
	| affect the rounding in what would otherwise be equal-
	| distance situations, which is what we want it to do.
	bset	#0,%d0
1:	clr.l	(%a0)			| zap it from memory.
	| now, round off the low 8 bits of the hi lword.
	tst.b	%d0			| 8 low bits.
	jne	fp_ns_checkround	| Are they non-zero?
	| nothing to do here
	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
fp_ns_checkround:
	fp_set_sr FPSR_EXC_INEX2	| INEX2 bit
	clr.b	-(%a0)			| clear low byte of high lword
	subq.l	#3,%a0
	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	jne	2f			| %d2 == 0, round to nearest
	tst.b	%d0			| test guard bit
	jpl	9f			| zero is closer
	btst	#8,%d0			| test lsb bit
	| round to even behaviour, see above.
	jne	fp_ns_doroundup		| round to infinity
	lsl.b	#1,%d0			| check low bits
	jeq	9f			| round to zero
fp_ns_doroundup:
	| round (the mantissa, that is) towards infinity
	add.l	#0x100,(%a0)
	jcc	9f			| no overflow, good.
	| Overflow.  This means that the %d1 was 0xffffff00, so it
	| is now zero.  We will set the mantissa to reflect this, and
	| increment the exponent (checking for overflow there too)
	move.w	#0x8000,(%a0)
	addq.w	#1,-(%a0)
	cmp.w	#0x407f,(%a0)+		| exponent now overflown?
	jeq	fp_ns_large		| yes, so make it infinity.
9:	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
	| check nondefault rounding modes
2:	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	3f			| %d2 > 2, round to +infinity
	tst.b	(-3,%a0)		| to -inf
	jne	fp_ns_doroundup		| negative, round to infinity
	jra	9b			| positive, round to zero
3:	tst.b	(-3,%a0)		| to +inf
	jeq	fp_ns_doroundup		| positive, round to infinity
	jra	9b			| negative, round to zero
	| Exponent underflow.  Try to make a denormal, and set it to
	| the smallest possible fraction if this fails.
fp_ns_small:
	fp_set_sr FPSR_EXC_UNFL		| set UNFL bit
	move.w	#0x3f81,(-2,%a0)	| 2**-126
	neg.w	%d2			| degree of underflow
	cmp.w	#32,%d2			| single or double shift?
	jcc	2f
	| a 32-bit shift.
	move.l	(%a0),%d0
	move.l	%d0,%d1
	lsr.l	%d2,%d0
	move.l	%d0,(%a0)+
	| Check to see if we shifted off any significant bits.
	neg.w	%d2
	add.w	#32,%d2
	lsl.l	%d2,%d1
	jeq	1f
	bset	#0,%d0			| Sticky bit.
	| Check the lower lword
1:	tst.l	(%a0)
	jeq	fp_ns_round
	clr	(%a0)
	bset	#0,%d0			| Sticky bit.
	jra	fp_ns_round
	| Sorry, the number is just too small.
2:	clr.l	(%a0)+
	clr.l	(%a0)
	moveq	#1,%d0			| Smallest possible fraction,
	jra	fp_ns_round		| round as desired.
	| Exponent overflow.  Just call it infinity.
fp_ns_large:
	tst.b	(3,%a0)
	jeq	1f
	fp_set_sr FPSR_EXC_INEX2
1:	fp_set_sr FPSR_EXC_OVFL
	move.w	(FPD_RND,FPDATA),%d2
	jne	3f			| %d2 = 0 round to nearest
1:	move.w	#0x7fff,(-2,%a0)
	clr.l	(%a0)+
	clr.l	(%a0)
2:	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
3:	subq.w	#2,%d2
	jcs	5f			| %d2 < 2, round to zero
	jhi	4f			| %d2 > 2, round to +infinity
	tst.b	(-3,%a0)		| to -inf
	jne	1b
	jra	5f
4:	tst.b	(-3,%a0)		| to +inf
	jeq	1b
5:	move.w	#0x407e,(-2,%a0)
	move.l	#0xffffff00,(%a0)+
	clr.l	(%a0)
	jra	2b
	| zero and denormalized
fp_ns_zero:
	tst.l	(%a0)+
	jne	1f
	tst.l	(%a0)
	jne	1f
	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts				| zero.  nothing to do.
	| These are not merely subnormal numbers, but true denormals,
	| i.e. pathologically small (exponent is 2**-16383) numbers.
	| It is clearly impossible for even a normal extended number
	| with that exponent to fit into single precision, so just
	| write these ones off as "too darn small".
1:	fp_set_sr FPSR_EXC_UNFL		| Set UNFL bit
	clr.l	(%a0)
	clr.l	-(%a0)
	move.w	#0x3f81,-(%a0)		| i.e. 2**-126
	addq.l	#6,%a0
	moveq	#1,%d0
	jra	fp_ns_round		| round.
	| Infinities or NaNs
fp_ns_huge:
	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts

	| fp_normalize_single_fast:
	| normalize an extended with single (23-bit) precision
	| this is only used by fsgldiv/fsgdlmul, where the
	| operand is not completly normalized.
	| args:	 %a0 (struct fp_ext *)

fp_normalize_single_fast:
	printf	PNORM,"nsf: %p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,") "
	addq.l	#2,%a0
	move.w	(%a0)+,%d2
	cmp.w	#0x7fff,%d2
	jeq	fp_nsf_huge		| NaN / infinitive.
	move.l	(%a0)+,%d0		| get high lword of mantissa
fp_nsf_round:
	tst.l	(%a0)			| check the low lword
	jeq	1f
	| Set a sticky bit if it is non-zero.  This should only
	| affect the rounding in what would otherwise be equal-
	| distance situations, which is what we want it to do.
	bset	#0,%d0
1:	clr.l	(%a0)			| zap it from memory.
	| now, round off the low 8 bits of the hi lword.
	tst.b	%d0			| 8 low bits.
	jne	fp_nsf_checkround	| Are they non-zero?
	| nothing to do here
	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
fp_nsf_checkround:
	fp_set_sr FPSR_EXC_INEX2	| INEX2 bit
	clr.b	-(%a0)			| clear low byte of high lword
	subq.l	#3,%a0
	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	jne	2f			| %d2 == 0, round to nearest
	tst.b	%d0			| test guard bit
	jpl	9f			| zero is closer
	btst	#8,%d0			| test lsb bit
	| round to even behaviour, see above.
	jne	fp_nsf_doroundup		| round to infinity
	lsl.b	#1,%d0			| check low bits
	jeq	9f			| round to zero
fp_nsf_doroundup:
	| round (the mantissa, that is) towards infinity
	add.l	#0x100,(%a0)
	jcc	9f			| no overflow, good.
	| Overflow.  This means that the %d1 was 0xffffff00, so it
	| is now zero.  We will set the mantissa to reflect this, and
	| increment the exponent (checking for overflow there too)
	move.w	#0x8000,(%a0)
	addq.w	#1,-(%a0)
	cmp.w	#0x407f,(%a0)+		| exponent now overflown?
	jeq	fp_nsf_large		| yes, so make it infinity.
9:	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
	| check nondefault rounding modes
2:	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	3f			| %d2 > 2, round to +infinity
	tst.b	(-3,%a0)		| to -inf
	jne	fp_nsf_doroundup	| negative, round to infinity
	jra	9b			| positive, round to zero
3:	tst.b	(-3,%a0)		| to +inf
	jeq	fp_nsf_doroundup		| positive, round to infinity
	jra	9b			| negative, round to zero
	| Exponent overflow.  Just call it infinity.
fp_nsf_large:
	tst.b	(3,%a0)
	jeq	1f
	fp_set_sr FPSR_EXC_INEX2
1:	fp_set_sr FPSR_EXC_OVFL
	move.w	(FPD_RND,FPDATA),%d2
	jne	3f			| %d2 = 0 round to nearest
1:	move.w	#0x7fff,(-2,%a0)
	clr.l	(%a0)+
	clr.l	(%a0)
2:	subq.l	#8,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts
3:	subq.w	#2,%d2
	jcs	5f			| %d2 < 2, round to zero
	jhi	4f			| %d2 > 2, round to +infinity
	tst.b	(-3,%a0)		| to -inf
	jne	1b
	jra	5f
4:	tst.b	(-3,%a0)		| to +inf
	jeq	1b
5:	move.w	#0x407e,(-2,%a0)
	move.l	#0xffffff00,(%a0)+
	clr.l	(%a0)
	jra	2b
	| Infinities or NaNs
fp_nsf_huge:
	subq.l	#4,%a0
	printf	PNORM,"%p(",1,%a0
	printx	PNORM,%a0@
	printf	PNORM,")\n"
	rts

	| conv_ext2int (macro):
	| Generates a subroutine that converts an extended value to an
	| integer of a given size, again, with the appropriate type of
	| rounding.

	| Macro arguments:
	| s:	size, as given in an assembly instruction.
	| b:	number of bits in that size.

	| Subroutine arguments:
	| %a0:	source (struct fp_ext *)

	| Returns the integer in %d0 (like it should)

.macro conv_ext2int s,b
	.set	inf,(1<<(\b-1))-1	| i.e. MAXINT
	printf	PCONV,"e2i%d: %p(",2,#\b,%a0
	printx	PCONV,%a0@
	printf	PCONV,") "
	addq.l	#2,%a0
	move.w	(%a0)+,%d2		| exponent
	jeq	fp_e2i_zero\b		| zero / denorm (== 0, here)
	cmp.w	#0x7fff,%d2
	jeq	fp_e2i_huge\b		| Inf / NaN
	sub.w	#0x3ffe,%d2
	jcs	fp_e2i_small\b
	cmp.w	#\b,%d2
	jhi	fp_e2i_large\b
	move.l	(%a0),%d0
	move.l	%d0,%d1
	lsl.l	%d2,%d1
	jne	fp_e2i_round\b
	tst.l	(4,%a0)
	jne	fp_e2i_round\b
	neg.w	%d2
	add.w	#32,%d2
	lsr.l	%d2,%d0
9:	tst.w	(-4,%a0)
	jne	1f
	tst.\s	%d0
	jmi	fp_e2i_large\b
	printf	PCONV,"-> %p\n",1,%d0
	rts
1:	neg.\s	%d0
	jeq	1f
	jpl	fp_e2i_large\b
1:	printf	PCONV,"-> %p\n",1,%d0
	rts
fp_e2i_round\b:
	fp_set_sr FPSR_EXC_INEX2	| INEX2 bit
	neg.w	%d2
	add.w	#32,%d2
	.if	\b>16
	jeq	5f
	.endif
	lsr.l	%d2,%d0
	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	jne	2f			| %d2 == 0, round to nearest
	tst.l	%d1			| test guard bit
	jpl	9b			| zero is closer
	btst	%d2,%d0			| test lsb bit (%d2 still 0)
	jne	fp_e2i_doroundup\b
	lsl.l	#1,%d1			| check low bits
	jne	fp_e2i_doroundup\b
	tst.l	(4,%a0)
	jeq	9b
fp_e2i_doroundup\b:
	addq.l	#1,%d0
	jra	9b
	| check nondefault rounding modes
2:	subq.w	#2,%d2
	jcs	9b			| %d2 < 2, round to zero
	jhi	3f			| %d2 > 2, round to +infinity
	tst.w	(-4,%a0)		| to -inf
	jne	fp_e2i_doroundup\b	| negative, round to infinity
	jra	9b			| positive, round to zero
3:	tst.w	(-4,%a0)		| to +inf
	jeq	fp_e2i_doroundup\b	| positive, round to infinity
	jra	9b	| negative, round to zero
	| we are only want -2**127 get correctly rounded here,
	| since the guard bit is in the lower lword.
	| everything else ends up anyway as overflow.
	.if	\b>16
5:	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	jne	2b			| %d2 == 0, round to nearest
	move.l	(4,%a0),%d1		| test guard bit
	jpl	9b			| zero is closer
	lsl.l	#1,%d1			| check low bits
	jne	fp_e2i_doroundup\b
	jra	9b
	.endif
fp_e2i_zero\b:
	clr.l	%d0
	tst.l	(%a0)+
	jne	1f
	tst.l	(%a0)
	jeq	3f
1:	subq.l	#4,%a0
	fp_clr_sr FPSR_EXC_UNFL		| fp_normalize_ext has set this bit
fp_e2i_small\b:
	fp_set_sr FPSR_EXC_INEX2
	clr.l	%d0
	move.w	(FPD_RND,FPDATA),%d2	| rounding mode
	subq.w	#2,%d2
	jcs	3f			| %d2 < 2, round to nearest/zero
	jhi	2f			| %d2 > 2, round to +infinity
	tst.w	(-4,%a0)		| to -inf
	jeq	3f
	subq.\s	#1,%d0
	jra	3f
2:	tst.w	(-4,%a0)		| to +inf
	jne	3f
	addq.\s	#1,%d0
3:	printf	PCONV,"-> %p\n",1,%d0
	rts
fp_e2i_large\b:
	fp_set_sr FPSR_EXC_OPERR
	move.\s	#inf,%d0
	tst.w	(-4,%a0)
	jeq	1f
	addq.\s	#1,%d0
1:	printf	PCONV,"-> %p\n",1,%d0
	rts
fp_e2i_huge\b:
	move.\s	(%a0),%d0
	tst.l	(%a0)
	jne	1f
	tst.l	(%a0)
	jeq	fp_e2i_large\b
	| fp_normalize_ext has set this bit already
	| and made the number nonsignaling
1:	fp_tst_sr FPSR_EXC_SNAN
	jne	1f
	fp_set_sr FPSR_EXC_OPERR
1:	printf	PCONV,"-> %p\n",1,%d0
	rts
.endm

fp_conv_ext2long:
	conv_ext2int l,32

fp_conv_ext2short:
	conv_ext2int w,16

fp_conv_ext2byte:
	conv_ext2int b,8

fp_conv_ext2double:
	jsr	fp_normalize_double
	printf	PCONV,"e2d: %p(",1,%a0
	printx	PCONV,%a0@
	printf	PCONV,"), "
	move.l	(%a0)+,%d2
	cmp.w	#0x7fff,%d2
	jne	1f
	move.w	#0x7ff,%d2
	move.l	(%a0)+,%d0
	jra	2f
1:	sub.w	#0x3fff-0x3ff,%d2
	move.l	(%a0)+,%d0
	jmi	2f
	clr.w	%d2
2:	lsl.w	#5,%d2
	lsl.l	#7,%d2
	lsl.l	#8,%d2
	move.l	%d0,%d1
	lsl.l	#1,%d0
	lsr.l	#4,%d0
	lsr.l	#8,%d0
	or.l	%d2,%d0
	putuser.l %d0,(%a1)+,fp_err_ua2,%a1
	moveq	#21,%d0
	lsl.l	%d0,%d1
	move.l	(%a0),%d0
	lsr.l	#4,%d0
	lsr.l	#7,%d0
	or.l	%d1,%d0
	putuser.l %d0,(%a1),fp_err_ua2,%a1
#ifdef FPU_EMU_DEBUG
	getuser.l %a1@(-4),%d0,fp_err_ua2,%a1
	getuser.l %a1@(0),%d1,fp_err_ua2,%a1
	printf	PCONV,"%p(%08x%08x)\n",3,%a1,%d0,%d1
#endif
	rts

fp_conv_ext2single:
	jsr	fp_normalize_single
	printf	PCONV,"e2s: %p(",1,%a0
	printx	PCONV,%a0@
	printf	PCONV,"), "
	move.l	(%a0)+,%d1
	cmp.w	#0x7fff,%d1
	jne	1f
	move.w	#0xff,%d1
	move.l	(%a0)+,%d0
	jra	2f
1:	sub.w	#0x3fff-0x7f,%d1
	move.l	(%a0)+,%d0
	jmi	2f
	clr.w	%d1
2:	lsl.w	#8,%d1
	lsl.l	#7,%d1
	lsl.l	#8,%d1
	bclr	#31,%d0
	lsr.l	#8,%d0
	or.l	%d1,%d0
	printf	PCONV,"%08x\n",1,%d0
	rts

	| special return addresses for instr that
	| encode the rounding precision in the opcode
	| (e.g. fsmove,fdmove)

fp_finalrounding_single:
	addq.l	#8,%sp
	jsr	fp_normalize_ext
	jsr	fp_normalize_single
	jra	fp_finaltest

fp_finalrounding_single_fast:
	addq.l	#8,%sp
	jsr	fp_normalize_ext
	jsr	fp_normalize_single_fast
	jra	fp_finaltest

fp_finalrounding_double:
	addq.l	#8,%sp
	jsr	fp_normalize_ext
	jsr	fp_normalize_double
	jra	fp_finaltest

	| fp_finaltest:
	| set the emulated status register based on the outcome of an
	| emulated instruction.

fp_finalrounding:
	addq.l	#8,%sp
|	printf	,"f: %p\n",1,%a0
	jsr	fp_normalize_ext
	move.w	(FPD_PREC,FPDATA),%d0
	subq.w	#1,%d0
	jcs	fp_finaltest
	jne	1f
	jsr	fp_normalize_single
	jra	2f
1:	jsr	fp_normalize_double
2:|	printf	,"f: %p\n",1,%a0
fp_finaltest:
	| First, we do some of the obvious tests for the exception
	| status byte and condition code bytes of fp_sr here, so that
	| they do not have to be handled individually by every
	| emulated instruction.
	clr.l	%d0
	addq.l	#1,%a0
	tst.b	(%a0)+			| sign
	jeq	1f
	bset	#FPSR_CC_NEG-24,%d0	| N bit
1:	cmp.w	#0x7fff,(%a0)+		| exponent
	jeq	2f
	| test for zero
	moveq	#FPSR_CC_Z-24,%d1
	tst.l	(%a0)+
	jne	9f
	tst.l	(%a0)
	jne	9f
	jra	8f
	| infinitiv and NAN
2:	moveq	#FPSR_CC_NAN-24,%d1
	move.l	(%a0)+,%d2
	lsl.l	#1,%d2			| ignore high bit
	jne	8f
	tst.l	(%a0)
	jne	8f
	moveq	#FPSR_CC_INF-24,%d1
8:	bset	%d1,%d0
9:	move.b	%d0,(FPD_FPSR+0,FPDATA)	| set condition test result
	| move instructions enter here
	| Here, we test things in the exception status byte, and set
	| other things in the accrued exception byte accordingly.
	| Emulated instructions can set various things in the former,
	| as defined in fp_emu.h.
fp_final:
	move.l	(FPD_FPSR,FPDATA),%d0
#if 0
	btst	#FPSR_EXC_SNAN,%d0	| EXC_SNAN
	jne	1f
	btst	#FPSR_EXC_OPERR,%d0	| EXC_OPERR
	jeq	2f
1:	bset	#FPSR_AEXC_IOP,%d0	| set IOP bit
2:	btst	#FPSR_EXC_OVFL,%d0	| EXC_OVFL
	jeq	1f
	bset	#FPSR_AEXC_OVFL,%d0	| set OVFL bit
1:	btst	#FPSR_EXC_UNFL,%d0	| EXC_UNFL
	jeq	1f
	btst	#FPSR_EXC_INEX2,%d0	| EXC_INEX2
	jeq	1f
	bset	#FPSR_AEXC_UNFL,%d0	| set UNFL bit
1:	btst	#FPSR_EXC_DZ,%d0	| EXC_INEX1
	jeq	1f
	bset	#FPSR_AEXC_DZ,%d0	| set DZ bit
1:	btst	#FPSR_EXC_OVFL,%d0	| EXC_OVFL
	jne	1f
	btst	#FPSR_EXC_INEX2,%d0	| EXC_INEX2
	jne	1f
	btst	#FPSR_EXC_INEX1,%d0	| EXC_INEX1
	jeq	2f
1:	bset	#FPSR_AEXC_INEX,%d0	| set INEX bit
2:	move.l	%d0,(FPD_FPSR,FPDATA)
#else
	| same as above, greatly optimized, but untested (yet)
	move.l	%d0,%d2
	lsr.l	#5,%d0
	move.l	%d0,%d1
	lsr.l	#4,%d1
	or.l	%d0,%d1
	and.b	#0x08,%d1
	move.l	%d2,%d0
	lsr.l	#6,%d0
	or.l	%d1,%d0
	move.l	%d2,%d1
	lsr.l	#4,%d1
	or.b	#0xdf,%d1
	and.b	%d1,%d0
	move.l	%d2,%d1
	lsr.l	#7,%d1
	and.b	#0x80,%d1
	or.b	%d1,%d0
	and.b	#0xf8,%d0
	or.b	%d0,%d2
	move.l	%d2,(FPD_FPSR,FPDATA)
#endif
	move.b	(FPD_FPSR+2,FPDATA),%d0
	and.b	(FPD_FPCR+2,FPDATA),%d0
	jeq	1f
	printf	,"send signal!!!\n"
1:	jra	fp_end