rucc-codegen 0.11.7

Instruction selection, scheduling, block layout, frames and prologue emission.
Documentation
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
;; Lowering AArch64.
;;
;; The same claim `x86-64.rules` makes, rule by rule: an IR term and a machine term compute the same
;; thing, and `rucc-verify` proves it against `aarch64.model` before the rule may be used. The IR
;; half of every rule is the one the x86-64 file matches, since the IR is not about a target.
;;
;; The machine has registers of thirty two and sixty four bits and nothing narrower, so a value of
;; eight or sixteen bits lives in the low end of a thirty two bit register and what is above it
;; means nothing. The arithmetic whose low bits do not depend on the high ones takes the thirty two
;; bit instruction at those widths, and the verifier holds it to the bits the IR term has. The
;; rest, which is right shifts, division and comparisons, has no rule below thirty two bits yet,
;; since each of those has to widen first and that is two instructions.
;;
;; A machine term inside another one is an instruction of its own, built first into a register of its
;; own, which is how a remainder is written: the division, then a multiply subtract that takes the
;; quotient times the divisor away from the dividend. A constant too wide for one `mov` is built the
;; same way, as a `mov` of its low sixteen bits with a `movk` for each piece above them.

;; Constants. One `mov` builds anything sixteen bits wide and anything whose complement is, and
;; that covers every constant of eight and sixteen bits as the IR holds them, sign extended. A
;; wider one is built a piece at a time, unless all but one of its sixteen bit pieces are zero or
;; all of them ones, which is a single `movz` or `movn` with a shift and is what a `double` like
;; 3.0 looks like as bits. The rules for one width share a pattern and the first whose guard
;; holds is the one that fires, so a constant takes as few instructions as these rules know how
;; to give it.

(rule (lower (iconst.i1 k))
      (a64.mov_ri_32 k)
      (spec (= k (result))))

(rule (lower (iconst.i8 k))
      (a64.mov_ri_32 k)
      (spec (= k (result))))

(rule (lower (iconst.i16 k))
      (a64.mov_ri_32 k)
      (spec (= k (result))))

(rule (lower (iconst.i32 k))
      (if (and (>= k -65536) (< k 65536)))
      (a64.mov_ri_32 k)
      (spec (= k (result))))

(rule (lower (iconst.i32 k))
      (if (or (= (extract 15 0 k) 0) (= (extract 15 0 k) 65535)))
      (a64.mov_ri_32 k)
      (spec (= k (result))))

(rule (lower (iconst.i32 k))
      (a64.movk_ri_16_32 (a64.mov_ri_32 (zero_extend 16 32 (extract 15 0 k)))
                         (zero_extend 16 32 (extract 31 16 k)))
      (spec (= k (result))))

(rule (lower (iconst.i64 k))
      (if (and (>= k -65536) (< k 65536)))
      (a64.mov_ri_64 k)
      (spec (= k (result))))

(rule (lower (iconst.i64 k))
      (if (or (and (= (extract 15 0 k) 0) (= (extract 63 32 k) 0))
              (and (= (extract 31 0 k) 0) (= (extract 63 48 k) 0))
              (= (extract 47 0 k) 0)
              (and (= (extract 15 0 k) 65535) (= (extract 63 32 k) 4294967295))
              (and (= (extract 31 0 k) 4294967295) (= (extract 63 48 k) 65535))
              (= (extract 47 0 k) 281474976710655)))
      (a64.mov_ri_64 k)
      (spec (= k (result))))

(rule (lower (iconst.i64 k))
      (if (and (>= k 0) (< k 4294967296)))
      (a64.movk_ri_16_64 (a64.mov_ri_64 (zero_extend 16 64 (extract 15 0 k)))
                         (zero_extend 16 64 (extract 31 16 k)))
      (spec (= k (result))))

(rule (lower (iconst.i64 k))
      (if (= (extract 63 48 k) 0))
      (a64.movk_ri_32_64
        (a64.movk_ri_16_64 (a64.mov_ri_64 (zero_extend 16 64 (extract 15 0 k)))
                           (zero_extend 16 64 (extract 31 16 k)))
        (zero_extend 16 64 (extract 47 32 k)))
      (spec (= k (result))))

(rule (lower (iconst.i64 k))
      (a64.movk_ri_48_64
        (a64.movk_ri_32_64
          (a64.movk_ri_16_64 (a64.mov_ri_64 (zero_extend 16 64 (extract 15 0 k)))
                             (zero_extend 16 64 (extract 31 16 k)))
          (zero_extend 16 64 (extract 47 32 k)))
        (zero_extend 16 64 (extract 63 48 k)))
      (spec (= k (result))))


;; Arithmetic, register with register.

(rule (lower (add.i8 (value.i8 x) (value.i8 y)))
      (a64.add_rr_32 x y)
      (spec (= (bvadd x y) (result))))

(rule (lower (add.i16 (value.i16 x) (value.i16 y)))
      (a64.add_rr_32 x y)
      (spec (= (bvadd x y) (result))))

(rule (lower (add.i32 (value.i32 x) (value.i32 y)))
      (a64.add_rr_32 x y)
      (spec (= (bvadd x y) (result))))

(rule (lower (add.i64 (value.i64 x) (value.i64 y)))
      (a64.add_rr_64 x y)
      (spec (= (bvadd x y) (result))))

(rule (lower (sub.i8 (value.i8 x) (value.i8 y)))
      (a64.sub_rr_32 x y)
      (spec (= (bvsub x y) (result))))

(rule (lower (sub.i16 (value.i16 x) (value.i16 y)))
      (a64.sub_rr_32 x y)
      (spec (= (bvsub x y) (result))))

(rule (lower (sub.i32 (value.i32 x) (value.i32 y)))
      (a64.sub_rr_32 x y)
      (spec (= (bvsub x y) (result))))

(rule (lower (sub.i64 (value.i64 x) (value.i64 y)))
      (a64.sub_rr_64 x y)
      (spec (= (bvsub x y) (result))))

(rule (lower (and.i8 (value.i8 x) (value.i8 y)))
      (a64.and_rr_32 x y)
      (spec (= (bvand x y) (result))))

(rule (lower (and.i16 (value.i16 x) (value.i16 y)))
      (a64.and_rr_32 x y)
      (spec (= (bvand x y) (result))))

(rule (lower (and.i32 (value.i32 x) (value.i32 y)))
      (a64.and_rr_32 x y)
      (spec (= (bvand x y) (result))))

(rule (lower (and.i64 (value.i64 x) (value.i64 y)))
      (a64.and_rr_64 x y)
      (spec (= (bvand x y) (result))))

(rule (lower (or.i8 (value.i8 x) (value.i8 y)))
      (a64.orr_rr_32 x y)
      (spec (= (bvor x y) (result))))

(rule (lower (or.i16 (value.i16 x) (value.i16 y)))
      (a64.orr_rr_32 x y)
      (spec (= (bvor x y) (result))))

(rule (lower (or.i32 (value.i32 x) (value.i32 y)))
      (a64.orr_rr_32 x y)
      (spec (= (bvor x y) (result))))

(rule (lower (or.i64 (value.i64 x) (value.i64 y)))
      (a64.orr_rr_64 x y)
      (spec (= (bvor x y) (result))))

(rule (lower (xor.i8 (value.i8 x) (value.i8 y)))
      (a64.eor_rr_32 x y)
      (spec (= (bvxor x y) (result))))

(rule (lower (xor.i16 (value.i16 x) (value.i16 y)))
      (a64.eor_rr_32 x y)
      (spec (= (bvxor x y) (result))))

(rule (lower (xor.i32 (value.i32 x) (value.i32 y)))
      (a64.eor_rr_32 x y)
      (spec (= (bvxor x y) (result))))

(rule (lower (xor.i64 (value.i64 x) (value.i64 y)))
      (a64.eor_rr_64 x y)
      (spec (= (bvxor x y) (result))))

(rule (lower (mul.i8 (value.i8 x) (value.i8 y)))
      (a64.mul_rr_32 x y)
      (spec (= (bvmul x y) (result))))

(rule (lower (mul.i16 (value.i16 x) (value.i16 y)))
      (a64.mul_rr_32 x y)
      (spec (= (bvmul x y) (result))))

(rule (lower (mul.i32 (value.i32 x) (value.i32 y)))
      (a64.mul_rr_32 x y)
      (spec (= (bvmul x y) (result))))

(rule (lower (mul.i64 (value.i64 x) (value.i64 y)))
      (a64.mul_rr_64 x y)
      (spec (= (bvmul x y) (result))))


;; The three bitwise operations at one bit, which take zeros and ones to zeros and ones.

(rule (lower (and.i1 (value.i1 x) (value.i1 y)))
      (a64.and_rr_32 x y)
      (spec (= (bvand x y) (result))))

(rule (lower (or.i1 (value.i1 x) (value.i1 y)))
      (a64.orr_rr_32 x y)
      (spec (= (bvor x y) (result))))

(rule (lower (xor.i1 (value.i1 x) (value.i1 y)))
      (a64.eor_rr_32 x y)
      (spec (= (bvxor x y) (result))))


;; Arithmetic with an immediate. The instruction holds twelve bits, unsigned, and a negative one
;; would be the other instruction with the constant negated, which is a computation the verifier has
;; no way to read yet. A constant outside the range is left to the register form.

(rule (lower (add.i8 (value.i8 x) (iconst.i8 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.add_ri_32 x k)
      (spec (= (bvadd x k) (result))))

(rule (lower (add.i16 (value.i16 x) (iconst.i16 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.add_ri_32 x k)
      (spec (= (bvadd x k) (result))))

(rule (lower (add.i32 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.add_ri_32 x k)
      (spec (= (bvadd x k) (result))))

(rule (lower (add.i64 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.add_ri_64 x k)
      (spec (= (bvadd x k) (result))))

(rule (lower (sub.i8 (value.i8 x) (iconst.i8 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.sub_ri_32 x k)
      (spec (= (bvsub x k) (result))))

(rule (lower (sub.i16 (value.i16 x) (iconst.i16 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.sub_ri_32 x k)
      (spec (= (bvsub x k) (result))))

(rule (lower (sub.i32 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.sub_ri_32 x k)
      (spec (= (bvsub x k) (result))))

(rule (lower (sub.i64 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.sub_ri_64 x k)
      (spec (= (bvsub x k) (result))))


;; Negation and complement, which the IR writes as arithmetic against a constant.

(rule (lower (sub.i8 (iconst.i8 0) (value.i8 x)))
      (a64.neg_r_32 x)
      (spec (= (bvsub 0 x) (result))))

(rule (lower (sub.i16 (iconst.i16 0) (value.i16 x)))
      (a64.neg_r_32 x)
      (spec (= (bvsub 0 x) (result))))

(rule (lower (sub.i32 (iconst.i32 0) (value.i32 x)))
      (a64.neg_r_32 x)
      (spec (= (bvsub 0 x) (result))))

(rule (lower (sub.i64 (iconst.i64 0) (value.i64 x)))
      (a64.neg_r_64 x)
      (spec (= (bvsub 0 x) (result))))

(rule (lower (xor.i8 (value.i8 x) (iconst.i8 -1)))
      (a64.mvn_r_32 x)
      (spec (= (bvnot x) (result))))

(rule (lower (xor.i16 (value.i16 x) (iconst.i16 -1)))
      (a64.mvn_r_32 x)
      (spec (= (bvnot x) (result))))

(rule (lower (xor.i32 (value.i32 x) (iconst.i32 -1)))
      (a64.mvn_r_32 x)
      (spec (= (bvnot x) (result))))

(rule (lower (xor.i64 (value.i64 x) (iconst.i64 -1)))
      (a64.mvn_r_64 x)
      (spec (= (bvnot x) (result))))


;; Division. The IR never carries a division by zero, so what the machine does with one, which is
;; to give back zero rather than trap, is not something any rule here relies on.

(rule (lower (sdiv.i32 (value.i32 x) (value.i32 y)))
      (a64.sdiv_rr_32 x y)
      (spec (= (bvsdiv x y) (result))))

(rule (lower (sdiv.i64 (value.i64 x) (value.i64 y)))
      (a64.sdiv_rr_64 x y)
      (spec (= (bvsdiv x y) (result))))

(rule (lower (udiv.i32 (value.i32 x) (value.i32 y)))
      (a64.udiv_rr_32 x y)
      (spec (= (bvudiv x y) (result))))

(rule (lower (udiv.i64 (value.i64 x) (value.i64 y)))
      (a64.udiv_rr_64 x y)
      (spec (= (bvudiv x y) (result))))


;; Remainder, as what is left of the dividend once the quotient times the divisor is taken away.

(rule (lower (srem.i32 (value.i32 x) (value.i32 y)))
      (a64.msub_rrr_32 (a64.sdiv_rr_32 x y) y x)
      (spec (= (bvsrem x y) (result))))

(rule (lower (srem.i64 (value.i64 x) (value.i64 y)))
      (a64.msub_rrr_64 (a64.sdiv_rr_64 x y) y x)
      (spec (= (bvsrem x y) (result))))

(rule (lower (urem.i32 (value.i32 x) (value.i32 y)))
      (a64.msub_rrr_32 (a64.udiv_rr_32 x y) y x)
      (spec (= (bvurem x y) (result))))

(rule (lower (urem.i64 (value.i64 x) (value.i64 y)))
      (a64.msub_rrr_64 (a64.udiv_rr_64 x y) y x)
      (spec (= (bvurem x y) (result))))


;; Shifts by a constant. A left shift keeps its low bits whatever is above them, so it is the one
;; that reaches the narrow widths.

(rule (lower (shl.i8 (value.i8 x) (iconst.i8 k)))
      (if (and (>= k 0) (< k 8)))
      (a64.lsl_ri_32 x k)
      (spec (= (bvshl x k) (result))))

(rule (lower (shl.i16 (value.i16 x) (iconst.i16 k)))
      (if (and (>= k 0) (< k 16)))
      (a64.lsl_ri_32 x k)
      (spec (= (bvshl x k) (result))))

(rule (lower (shl.i32 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 32)))
      (a64.lsl_ri_32 x k)
      (spec (= (bvshl x k) (result))))

(rule (lower (shl.i64 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 64)))
      (a64.lsl_ri_64 x k)
      (spec (= (bvshl x k) (result))))

(rule (lower (lshr.i32 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 32)))
      (a64.lsr_ri_32 x k)
      (spec (= (bvlshr x k) (result))))

(rule (lower (lshr.i64 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 64)))
      (a64.lsr_ri_64 x k)
      (spec (= (bvlshr x k) (result))))

(rule (lower (ashr.i32 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 32)))
      (a64.asr_ri_32 x k)
      (spec (= (bvashr x k) (result))))

(rule (lower (ashr.i64 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 64)))
      (a64.asr_ri_64 x k)
      (spec (= (bvashr x k) (result))))


;; Shifts by a register. The machine takes the count modulo the width, which is what the IR does,
;; so there is nothing to mask.

(rule (lower (shl.i32 (value.i32 x) (value.i32 y)))
      (a64.lsl_rr_32 x y)
      (spec (= (bvshl x (bvand y 31)) (result))))

(rule (lower (shl.i64 (value.i64 x) (value.i64 y)))
      (a64.lsl_rr_64 x y)
      (spec (= (bvshl x (bvand y 63)) (result))))

(rule (lower (lshr.i32 (value.i32 x) (value.i32 y)))
      (a64.lsr_rr_32 x y)
      (spec (= (bvlshr x (bvand y 31)) (result))))

(rule (lower (lshr.i64 (value.i64 x) (value.i64 y)))
      (a64.lsr_rr_64 x y)
      (spec (= (bvlshr x (bvand y 63)) (result))))

(rule (lower (ashr.i32 (value.i32 x) (value.i32 y)))
      (a64.asr_rr_32 x y)
      (spec (= (bvashr x (bvand y 31)) (result))))

(rule (lower (ashr.i64 (value.i64 x) (value.i64 y)))
      (a64.asr_rr_64 x y)
      (spec (= (bvashr x (bvand y 63)) (result))))


;; Comparisons. Each is a compare and the `cset` behind it, one term for the reason the x86-64 file
;; gives: the flags between the two are not a value. The IR's ten predicates are the machine's ten
;; integer conditions, with the unsigned ones under the names the architecture gives them.

(rule (lower (icmp_eq.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_eq_32 x y)
      (spec (= (ite (= x y) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_ne_32 x y)
      (spec (= (ite (not (= x y)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_lt_32 x y)
      (spec (= (ite (bvslt x y) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_le_32 x y)
      (spec (= (ite (bvsle x y) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_gt_32 x y)
      (spec (= (ite (bvsgt x y) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_ge_32 x y)
      (spec (= (ite (bvsge x y) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_lo_32 x y)
      (spec (= (ite (bvult x y) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_ls_32 x y)
      (spec (= (ite (bvule x y) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_hi_32 x y)
      (spec (= (ite (bvugt x y) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i32 x) (value.i32 y)))
      (a64.cmp_set_hs_32 x y)
      (spec (= (ite (bvuge x y) 1 0) (result))))

(rule (lower (icmp_eq.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_eq_64 x y)
      (spec (= (ite (= x y) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_ne_64 x y)
      (spec (= (ite (not (= x y)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_lt_64 x y)
      (spec (= (ite (bvslt x y) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_le_64 x y)
      (spec (= (ite (bvsle x y) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_gt_64 x y)
      (spec (= (ite (bvsgt x y) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_ge_64 x y)
      (spec (= (ite (bvsge x y) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_lo_64 x y)
      (spec (= (ite (bvult x y) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_ls_64 x y)
      (spec (= (ite (bvule x y) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_hi_64 x y)
      (spec (= (ite (bvugt x y) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i64 x) (value.i64 y)))
      (a64.cmp_set_hs_64 x y)
      (spec (= (ite (bvuge x y) 1 0) (result))))


;; A byte or a half is compared as the thirty two bit value it widens to, sign extended for the
;; signed predicates and zero extended for the rest, since the bits above it in the register are
;; whatever the instruction that wrote it left there. The optimizer narrows a comparison of two
;; `char` values to one of this width, so C reaches these at -O1 and above.

(rule (lower (icmp_eq.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_eq_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (= x y) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_ne_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (not (= x y)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_lt_32 (a64.sxtb_32 x) (a64.sxtb_32 y))
      (spec (= (ite (bvslt x y) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_le_32 (a64.sxtb_32 x) (a64.sxtb_32 y))
      (spec (= (ite (bvsle x y) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_gt_32 (a64.sxtb_32 x) (a64.sxtb_32 y))
      (spec (= (ite (bvsgt x y) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_ge_32 (a64.sxtb_32 x) (a64.sxtb_32 y))
      (spec (= (ite (bvsge x y) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_lo_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (bvult x y) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_ls_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (bvule x y) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_hi_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (bvugt x y) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i8 x) (value.i8 y)))
      (a64.cmp_set_hs_32 (a64.uxtb_32 x) (a64.uxtb_32 y))
      (spec (= (ite (bvuge x y) 1 0) (result))))

(rule (lower (icmp_eq.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_eq_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (= x y) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_ne_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (not (= x y)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_lt_32 (a64.sxth_32 x) (a64.sxth_32 y))
      (spec (= (ite (bvslt x y) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_le_32 (a64.sxth_32 x) (a64.sxth_32 y))
      (spec (= (ite (bvsle x y) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_gt_32 (a64.sxth_32 x) (a64.sxth_32 y))
      (spec (= (ite (bvsgt x y) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_ge_32 (a64.sxth_32 x) (a64.sxth_32 y))
      (spec (= (ite (bvsge x y) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_lo_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (bvult x y) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_ls_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (bvule x y) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_hi_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (bvugt x y) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i16 x) (value.i16 y)))
      (a64.cmp_set_hs_32 (a64.uxth_32 x) (a64.uxth_32 y))
      (spec (= (ite (bvuge x y) 1 0) (result))))


;; The same against a constant the compare can hold, which is twelve bits.

(rule (lower (icmp_eq.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_eq_ri_32 x k)
      (spec (= (ite (= x k) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ne_ri_32 x k)
      (spec (= (ite (not (= x k)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_lt_ri_32 x k)
      (spec (= (ite (bvslt x k) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_le_ri_32 x k)
      (spec (= (ite (bvsle x k) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_gt_ri_32 x k)
      (spec (= (ite (bvsgt x k) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ge_ri_32 x k)
      (spec (= (ite (bvsge x k) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_lo_ri_32 x k)
      (spec (= (ite (bvult x k) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ls_ri_32 x k)
      (spec (= (ite (bvule x k) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_hi_ri_32 x k)
      (spec (= (ite (bvugt x k) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i32 x) (iconst.i32 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_hs_ri_32 x k)
      (spec (= (ite (bvuge x k) 1 0) (result))))

(rule (lower (icmp_eq.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_eq_ri_64 x k)
      (spec (= (ite (= x k) 1 0) (result))))

(rule (lower (icmp_ne.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ne_ri_64 x k)
      (spec (= (ite (not (= x k)) 1 0) (result))))

(rule (lower (icmp_slt.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_lt_ri_64 x k)
      (spec (= (ite (bvslt x k) 1 0) (result))))

(rule (lower (icmp_sle.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_le_ri_64 x k)
      (spec (= (ite (bvsle x k) 1 0) (result))))

(rule (lower (icmp_sgt.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_gt_ri_64 x k)
      (spec (= (ite (bvsgt x k) 1 0) (result))))

(rule (lower (icmp_sge.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ge_ri_64 x k)
      (spec (= (ite (bvsge x k) 1 0) (result))))

(rule (lower (icmp_ult.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_lo_ri_64 x k)
      (spec (= (ite (bvult x k) 1 0) (result))))

(rule (lower (icmp_ule.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_ls_ri_64 x k)
      (spec (= (ite (bvule x k) 1 0) (result))))

(rule (lower (icmp_ugt.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_hi_ri_64 x k)
      (spec (= (ite (bvugt x k) 1 0) (result))))

(rule (lower (icmp_uge.i1 (value.i64 x) (iconst.i64 k)))
      (if (and (>= k 0) (< k 4096)))
      (a64.cmp_set_hs_ri_64 x k)
      (spec (= (ite (bvuge x k) 1 0) (result))))


;; The choice between two values, which is a compare of the condition with zero and a `csel`. The
;; operands are in the order the x86-64 rules put them, the false arm first, so the two selectors
;; read one operand list the same way.

(rule (lower (select.i8 (value.i1 c) (value.i8 t) (value.i8 f)))
      (a64.sel_32 f t c)
      (spec (= (ite (not (= c 0)) t f) (result))))

(rule (lower (select.i16 (value.i1 c) (value.i16 t) (value.i16 f)))
      (a64.sel_32 f t c)
      (spec (= (ite (not (= c 0)) t f) (result))))

(rule (lower (select.i32 (value.i1 c) (value.i32 t) (value.i32 f)))
      (a64.sel_32 f t c)
      (spec (= (ite (not (= c 0)) t f) (result))))

(rule (lower (select.i64 (value.i1 c) (value.i64 t) (value.i64 f)))
      (a64.sel_64 f t c)
      (spec (= (ite (not (= c 0)) t f) (result))))


;; Conversions between widths. Widening a byte or a half to sixty four bits unsigned is the thirty
;; two bit instruction, since writing a W register clears the top half, and that is a second meaning
;; for one head, so it waits for a head of its own.

(rule (lower (sext.i8.i32 (value.i8 x)))
      (a64.sxtb_32 x)
      (spec (= (sign_extend 8 32 x) (result))))

(rule (lower (sext.i8.i64 (value.i8 x)))
      (a64.sxtb_64 x)
      (spec (= (sign_extend 8 64 x) (result))))

(rule (lower (sext.i16.i32 (value.i16 x)))
      (a64.sxth_32 x)
      (spec (= (sign_extend 16 32 x) (result))))

(rule (lower (sext.i16.i64 (value.i16 x)))
      (a64.sxth_64 x)
      (spec (= (sign_extend 16 64 x) (result))))

(rule (lower (sext.i32.i64 (value.i32 x)))
      (a64.sxtw_64 x)
      (spec (= (sign_extend 32 64 x) (result))))

(rule (lower (zext.i8.i32 (value.i8 x)))
      (a64.uxtb_32 x)
      (spec (= (zero_extend 8 32 x) (result))))

(rule (lower (zext.i16.i32 (value.i16 x)))
      (a64.uxth_32 x)
      (spec (= (zero_extend 16 32 x) (result))))

(rule (lower (zext.i32.i64 (value.i32 x)))
      (a64.uxtw_64 x)
      (spec (= (zero_extend 32 64 x) (result))))

;; A sixteen bit value lives in a W register the same as a thirty two bit one, so widening to it is
;; the thirty two bit instruction under a head held to the sixteen bits the result has.

(rule (lower (sext.i8.i16 (value.i8 x)))
      (a64.sxtb_16 x)
      (spec (= (sign_extend 8 16 x) (result))))

(rule (lower (zext.i8.i16 (value.i8 x)))
      (a64.uxtb_16 x)
      (spec (= (zero_extend 8 16 x) (result))))

(rule (lower (zext.i8.i64 (value.i8 x)))
      (a64.uxtb_64 x)
      (spec (= (zero_extend 8 64 x) (result))))

(rule (lower (zext.i16.i64 (value.i16 x)))
      (a64.uxth_64 x)
      (spec (= (zero_extend 16 64 x) (result))))

;; A one bit value is widened with an `and`, so nothing is taken on trust about the bits above it.

(rule (lower (zext.i1.i8 (value.i1 x)))
      (a64.bit_to_8 x)
      (spec (= (zero_extend 1 8 x) (result))))

(rule (lower (zext.i1.i16 (value.i1 x)))
      (a64.bit_to_16 x)
      (spec (= (zero_extend 1 16 x) (result))))

(rule (lower (zext.i1.i32 (value.i1 x)))
      (a64.bit_to_32 x)
      (spec (= (zero_extend 1 32 x) (result))))

(rule (lower (zext.i1.i64 (value.i1 x)))
      (a64.bit_to_64 x)
      (spec (= (zero_extend 1 64 x) (result))))

;; Narrowing. The narrow value is the low end of the wide register, and a `mov` of the W register
;; puts it in one of its own. Narrowing to one bit keeps that bit and clears the rest.

(rule (lower (trunc.i16.i8 (value.i16 x)))
      (a64.low_8 x)
      (spec (= (extract 7 0 x) (result))))

(rule (lower (trunc.i32.i8 (value.i32 x)))
      (a64.low_8 x)
      (spec (= (extract 7 0 x) (result))))

(rule (lower (trunc.i64.i8 (value.i64 x)))
      (a64.low_8 x)
      (spec (= (extract 7 0 x) (result))))

(rule (lower (trunc.i32.i16 (value.i32 x)))
      (a64.low_16 x)
      (spec (= (extract 15 0 x) (result))))

(rule (lower (trunc.i64.i16 (value.i64 x)))
      (a64.low_16 x)
      (spec (= (extract 15 0 x) (result))))

(rule (lower (trunc.i64.i32 (value.i64 x)))
      (a64.low_32 x)
      (spec (= (extract 31 0 x) (result))))

(rule (lower (trunc.i8.i1 (value.i8 x)))
      (a64.bit_of_32 x 1)
      (spec (= (extract 0 0 x) (result))))

(rule (lower (trunc.i16.i1 (value.i16 x)))
      (a64.bit_of_32 x 1)
      (spec (= (extract 0 0 x) (result))))

(rule (lower (trunc.i32.i1 (value.i32 x)))
      (a64.bit_of_32 x 1)
      (spec (= (extract 0 0 x) (result))))

(rule (lower (trunc.i64.i1 (value.i64 x)))
      (a64.bit_of_64 x 1)
      (spec (= (extract 0 0 x) (result))))


;; Reading memory. Two addressing modes, a register and a register plus a constant. The constant is
;; held to the nine signed bits every load can take, which is the unscaled form; a larger one that
;; the scaled form could take is left to the plain rule for now.

(rule (lower (load.i8 (value.i64 a)))
      (a64.ldr_8 (amode_base a))
      (spec (= (select (mem) a) (result))))

(rule (lower (load.i16 (value.i64 a)))
      (a64.ldr_16 (amode_base a))
      (spec (= (concat (select (mem) (bvadd a 1)) (select (mem) a)) (result))))

(rule (lower (load.i32 (value.i64 a)))
      (a64.ldr_32 (amode_base a))
      (spec (= (concat (select (mem) (bvadd a 3)) (select (mem) (bvadd a 2))
                       (select (mem) (bvadd a 1)) (select (mem) a)) (result))))

(rule (lower (load.i64 (value.i64 a)))
      (a64.ldr_64 (amode_base a))
      (spec (= (concat (select (mem) (bvadd a 7)) (select (mem) (bvadd a 6))
                       (select (mem) (bvadd a 5)) (select (mem) (bvadd a 4))
                       (select (mem) (bvadd a 3)) (select (mem) (bvadd a 2))
                       (select (mem) (bvadd a 1)) (select (mem) a)) (result))))

(rule (lower (load.i8 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_8 (amode_base_offset a k))
      (spec (= (select (mem) (bvadd a k)) (result))))

(rule (lower (load.i16 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_16 (amode_base_offset a k))
      (spec (= (concat (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k))) (result))))

(rule (lower (load.i32 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_32 (amode_base_offset a k))
      (spec (= (concat (select (mem) (bvadd (bvadd a k) 3)) (select (mem) (bvadd (bvadd a k) 2))
                       (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k))) (result))))

(rule (lower (load.i64 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_64 (amode_base_offset a k))
      (spec (= (concat (select (mem) (bvadd (bvadd a k) 7)) (select (mem) (bvadd (bvadd a k) 6))
                       (select (mem) (bvadd (bvadd a k) 5)) (select (mem) (bvadd (bvadd a k) 4))
                       (select (mem) (bvadd (bvadd a k) 3)) (select (mem) (bvadd (bvadd a k) 2))
                       (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k))) (result))))


;; Writing memory. The value comes first in the IR and last in the machine term, as it does in the
;; x86-64 file.

(rule (lower (store.i8 (value.i8 v) (value.i64 a)))
      (a64.str_8 (amode_base a) v)
      (spec (= (store (mem) a v) (result))))

(rule (lower (store.i16 (value.i16 v) (value.i64 a)))
      (a64.str_16 (amode_base a) v)
      (spec (= (store (store (mem)
                       a (extract 7 0 v))
                       (bvadd a 1) (extract 15 8 v)) (result))))

(rule (lower (store.i32 (value.i32 v) (value.i64 a)))
      (a64.str_32 (amode_base a) v)
      (spec (= (store (store (store (store (mem)
                       a (extract 7 0 v))
                       (bvadd a 1) (extract 15 8 v))
                       (bvadd a 2) (extract 23 16 v))
                       (bvadd a 3) (extract 31 24 v)) (result))))

(rule (lower (store.i64 (value.i64 v) (value.i64 a)))
      (a64.str_64 (amode_base a) v)
      (spec (= (store (store (store (store (store (store (store (store (mem)
                       a (extract 7 0 v))
                       (bvadd a 1) (extract 15 8 v))
                       (bvadd a 2) (extract 23 16 v))
                       (bvadd a 3) (extract 31 24 v))
                       (bvadd a 4) (extract 39 32 v))
                       (bvadd a 5) (extract 47 40 v))
                       (bvadd a 6) (extract 55 48 v))
                       (bvadd a 7) (extract 63 56 v)) (result))))

(rule (lower (store.i8 (value.i8 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_8 (amode_base_offset a k) v)
      (spec (= (store (mem) (bvadd a k) v) (result))))

(rule (lower (store.i16 (value.i16 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_16 (amode_base_offset a k) v)
      (spec (= (store (store (mem)
                       (bvadd a k) (extract 7 0 v))
                       (bvadd (bvadd a k) 1) (extract 15 8 v)) (result))))

(rule (lower (store.i32 (value.i32 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_32 (amode_base_offset a k) v)
      (spec (= (store (store (store (store (mem)
                       (bvadd a k) (extract 7 0 v))
                       (bvadd (bvadd a k) 1) (extract 15 8 v))
                       (bvadd (bvadd a k) 2) (extract 23 16 v))
                       (bvadd (bvadd a k) 3) (extract 31 24 v)) (result))))

(rule (lower (store.i64 (value.i64 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_64 (amode_base_offset a k) v)
      (spec (= (store (store (store (store (store (store (store (store (mem)
                       (bvadd a k) (extract 7 0 v))
                       (bvadd (bvadd a k) 1) (extract 15 8 v))
                       (bvadd (bvadd a k) 2) (extract 23 16 v))
                       (bvadd (bvadd a k) 3) (extract 31 24 v))
                       (bvadd (bvadd a k) 4) (extract 39 32 v))
                       (bvadd (bvadd a k) 5) (extract 47 40 v))
                       (bvadd (bvadd a k) 6) (extract 55 48 v))
                       (bvadd (bvadd a k) 7) (extract 63 56 v)) (result))))


;; Giving a value back. What is below thirty two bits comes back in the low end of `w0`, and the
;; convention says the caller extends it, so all four of those are the thirty two bit pseudo.

(rule (lower (ret.i1 (value.i1 v)))
      (a64.ret_val_32 v)
      (spec (= v (result))))

(rule (lower (ret.i8 (value.i8 v)))
      (a64.ret_val_32 v)
      (spec (= v (result))))

(rule (lower (ret.i16 (value.i16 v)))
      (a64.ret_val_32 v)
      (spec (= v (result))))

(rule (lower (ret.i32 (value.i32 v)))
      (a64.ret_val_32 v)
      (spec (= v (result))))

(rule (lower (ret.i64 (value.i64 v)))
      (a64.ret_val_64 v)
      (spec (= v (result))))


;; Branching on a truth value, which the layout turns into a compare and a `b.ne` or a `b.eq`.

(rule (lower (brif.i1 (value.i1 c)))
      (a64.br_cond_32 c)
      (spec (= c (result))))


;; Floats, at the two formats C has registers for. A `long double` is a quad here, and every
;; operation on one is a call into the runtime rather than an instruction.

(rule (lower (fadd.f32 (value.f32 x) (value.f32 y)))
      (a64.fadd_f32 x y)
      (spec (= (fp.add x y) (result))))

(rule (lower (fadd.f64 (value.f64 x) (value.f64 y)))
      (a64.fadd_f64 x y)
      (spec (= (fp.add x y) (result))))

(rule (lower (fsub.f32 (value.f32 x) (value.f32 y)))
      (a64.fsub_f32 x y)
      (spec (= (fp.sub x y) (result))))

(rule (lower (fsub.f64 (value.f64 x) (value.f64 y)))
      (a64.fsub_f64 x y)
      (spec (= (fp.sub x y) (result))))

(rule (lower (fmul.f32 (value.f32 x) (value.f32 y)))
      (a64.fmul_f32 x y)
      (spec (= (fp.mul x y) (result))))

(rule (lower (fmul.f64 (value.f64 x) (value.f64 y)))
      (a64.fmul_f64 x y)
      (spec (= (fp.mul x y) (result))))

(rule (lower (fdiv.f32 (value.f32 x) (value.f32 y)))
      (a64.fdiv_f32 x y)
      (spec (= (fp.div x y) (result))))

(rule (lower (fdiv.f64 (value.f64 x) (value.f64 y)))
      (a64.fdiv_f64 x y)
      (spec (= (fp.div x y) (result))))

(rule (lower (ret.f32 (value.f32 v)))
      (a64.ret_val_f32 v)
      (spec (= v (result))))

(rule (lower (ret.f64 (value.f64 v)))
      (a64.ret_val_f64 v)
      (spec (= v (result))))

(rule (lower (ret.f128 (value.f128 v)))
      (a64.ret_val_f128 v)
      (spec (= v (result))))


;; Comparing two floats, which is an `fcmp` and the `cset` behind it as one term, for the reason the
;; integer comparisons are. The machine has a condition for each of the IR's fourteen predicates but
;; two, and those two take a second instruction, so no rule needs its operands the other way round.

(rule (lower (fcmp_ogt.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_gt_f32 x y)
      (spec (= (ite (fp.gt x y) 1 0) (result))))

(rule (lower (fcmp_oge.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ge_f32 x y)
      (spec (= (ite (fp.geq x y) 1 0) (result))))

(rule (lower (fcmp_olt.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_lt_f32 x y)
      (spec (= (ite (fp.lt x y) 1 0) (result))))

(rule (lower (fcmp_ole.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_le_f32 x y)
      (spec (= (ite (fp.leq x y) 1 0) (result))))

(rule (lower (fcmp_one.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_one_f32 x y)
      (spec (= (ite (or (fp.lt x y) (fp.gt x y)) 1 0) (result))))

(rule (lower (fcmp_ord.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ord_f32 x y)
      (spec (= (ite (and (not (fp.isNaN x)) (not (fp.isNaN y))) 1 0) (result))))

(rule (lower (fcmp_uno.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_uno_f32 x y)
      (spec (= (ite (or (fp.isNaN x) (fp.isNaN y)) 1 0) (result))))

(rule (lower (fcmp_ueq.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ueq_f32 x y)
      (spec (= (ite (not (or (fp.lt x y) (fp.gt x y))) 1 0) (result))))

(rule (lower (fcmp_ult.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ult_f32 x y)
      (spec (= (ite (not (fp.geq x y)) 1 0) (result))))

(rule (lower (fcmp_ule.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ule_f32 x y)
      (spec (= (ite (not (fp.gt x y)) 1 0) (result))))

(rule (lower (fcmp_ugt.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ugt_f32 x y)
      (spec (= (ite (not (fp.leq x y)) 1 0) (result))))

(rule (lower (fcmp_uge.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_uge_f32 x y)
      (spec (= (ite (not (fp.lt x y)) 1 0) (result))))

(rule (lower (fcmp_oeq.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_eq_f32 x y)
      (spec (= (ite (fp.eq x y) 1 0) (result))))

(rule (lower (fcmp_une.f32.i1 (value.f32 x) (value.f32 y)))
      (a64.fcmp_set_ne_f32 x y)
      (spec (= (ite (not (fp.eq x y)) 1 0) (result))))

(rule (lower (fcmp_ogt.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_gt_f64 x y)
      (spec (= (ite (fp.gt x y) 1 0) (result))))

(rule (lower (fcmp_oge.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ge_f64 x y)
      (spec (= (ite (fp.geq x y) 1 0) (result))))

(rule (lower (fcmp_olt.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_lt_f64 x y)
      (spec (= (ite (fp.lt x y) 1 0) (result))))

(rule (lower (fcmp_ole.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_le_f64 x y)
      (spec (= (ite (fp.leq x y) 1 0) (result))))

(rule (lower (fcmp_one.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_one_f64 x y)
      (spec (= (ite (or (fp.lt x y) (fp.gt x y)) 1 0) (result))))

(rule (lower (fcmp_ord.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ord_f64 x y)
      (spec (= (ite (and (not (fp.isNaN x)) (not (fp.isNaN y))) 1 0) (result))))

(rule (lower (fcmp_uno.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_uno_f64 x y)
      (spec (= (ite (or (fp.isNaN x) (fp.isNaN y)) 1 0) (result))))

(rule (lower (fcmp_ueq.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ueq_f64 x y)
      (spec (= (ite (not (or (fp.lt x y) (fp.gt x y))) 1 0) (result))))

(rule (lower (fcmp_ult.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ult_f64 x y)
      (spec (= (ite (not (fp.geq x y)) 1 0) (result))))

(rule (lower (fcmp_ule.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ule_f64 x y)
      (spec (= (ite (not (fp.gt x y)) 1 0) (result))))

(rule (lower (fcmp_ugt.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ugt_f64 x y)
      (spec (= (ite (not (fp.leq x y)) 1 0) (result))))

(rule (lower (fcmp_uge.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_uge_f64 x y)
      (spec (= (ite (not (fp.lt x y)) 1 0) (result))))

(rule (lower (fcmp_oeq.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_eq_f64 x y)
      (spec (= (ite (fp.eq x y) 1 0) (result))))

(rule (lower (fcmp_une.f64.i1 (value.f64 x) (value.f64 y)))
      (a64.fcmp_set_ne_f64 x y)
      (spec (= (ite (not (fp.eq x y)) 1 0) (result))))


;; Reading a float out of memory and writing one back, with the same two addressing modes.

(rule (lower (load.f32 (value.i64 a)))
      (a64.ldr_f32 (amode_base a))
      (spec (= (float_from_bits 32 (concat (select (mem) (bvadd a 3)) (select (mem) (bvadd a 2))
                       (select (mem) (bvadd a 1)) (select (mem) a))) (result))))

(rule (lower (load.f32 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_f32 (amode_base_offset a k))
      (spec (= (float_from_bits 32 (concat (select (mem) (bvadd (bvadd a k) 3)) (select (mem) (bvadd (bvadd a k) 2))
                       (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k)))) (result))))

(rule (lower (load.f64 (value.i64 a)))
      (a64.ldr_f64 (amode_base a))
      (spec (= (float_from_bits 64 (concat (select (mem) (bvadd a 7)) (select (mem) (bvadd a 6))
                       (select (mem) (bvadd a 5)) (select (mem) (bvadd a 4))
                       (select (mem) (bvadd a 3)) (select (mem) (bvadd a 2))
                       (select (mem) (bvadd a 1)) (select (mem) a))) (result))))

(rule (lower (load.f64 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_f64 (amode_base_offset a k))
      (spec (= (float_from_bits 64 (concat (select (mem) (bvadd (bvadd a k) 7)) (select (mem) (bvadd (bvadd a k) 6))
                       (select (mem) (bvadd (bvadd a k) 5)) (select (mem) (bvadd (bvadd a k) 4))
                       (select (mem) (bvadd (bvadd a k) 3)) (select (mem) (bvadd (bvadd a k) 2))
                       (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k)))) (result))))

(rule (lower (store.f32 (value.f32 v) (value.i64 a)))
      (a64.str_f32 (amode_base a) v)
      (spec (= (store (store (store (store (mem)
                       a (extract 7 0 (bits_from_float 32 v)))
                       (bvadd a 1) (extract 15 8 (bits_from_float 32 v)))
                       (bvadd a 2) (extract 23 16 (bits_from_float 32 v)))
                       (bvadd a 3) (extract 31 24 (bits_from_float 32 v))) (result))))

(rule (lower (store.f32 (value.f32 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_f32 (amode_base_offset a k) v)
      (spec (= (store (store (store (store (mem)
                       (bvadd a k) (extract 7 0 (bits_from_float 32 v)))
                       (bvadd (bvadd a k) 1) (extract 15 8 (bits_from_float 32 v)))
                       (bvadd (bvadd a k) 2) (extract 23 16 (bits_from_float 32 v)))
                       (bvadd (bvadd a k) 3) (extract 31 24 (bits_from_float 32 v))) (result))))

(rule (lower (store.f64 (value.f64 v) (value.i64 a)))
      (a64.str_f64 (amode_base a) v)
      (spec (= (store (store (store (store (store (store (store (store (mem)
                       a (extract 7 0 (bits_from_float 64 v)))
                       (bvadd a 1) (extract 15 8 (bits_from_float 64 v)))
                       (bvadd a 2) (extract 23 16 (bits_from_float 64 v)))
                       (bvadd a 3) (extract 31 24 (bits_from_float 64 v)))
                       (bvadd a 4) (extract 39 32 (bits_from_float 64 v)))
                       (bvadd a 5) (extract 47 40 (bits_from_float 64 v)))
                       (bvadd a 6) (extract 55 48 (bits_from_float 64 v)))
                       (bvadd a 7) (extract 63 56 (bits_from_float 64 v))) (result))))

(rule (lower (store.f64 (value.f64 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_f64 (amode_base_offset a k) v)
      (spec (= (store (store (store (store (store (store (store (store (mem)
                       (bvadd a k) (extract 7 0 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 1) (extract 15 8 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 2) (extract 23 16 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 3) (extract 31 24 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 4) (extract 39 32 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 5) (extract 47 40 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 6) (extract 55 48 (bits_from_float 64 v)))
                       (bvadd (bvadd a k) 7) (extract 63 56 (bits_from_float 64 v))) (result))))


;; A quad is sixteen bytes in one vector register, and `ldr q` and `str q` move all of it.

(rule (lower (load.f128 (value.i64 a)))
      (a64.ldr_f128 (amode_base a))
      (spec (= (float_from_bits 128 (concat (select (mem) (bvadd a 15)) (select (mem) (bvadd a 14))
                       (select (mem) (bvadd a 13)) (select (mem) (bvadd a 12))
                       (select (mem) (bvadd a 11)) (select (mem) (bvadd a 10))
                       (select (mem) (bvadd a 9)) (select (mem) (bvadd a 8))
                       (select (mem) (bvadd a 7)) (select (mem) (bvadd a 6))
                       (select (mem) (bvadd a 5)) (select (mem) (bvadd a 4))
                       (select (mem) (bvadd a 3)) (select (mem) (bvadd a 2))
                       (select (mem) (bvadd a 1)) (select (mem) a))) (result))))

(rule (lower (load.f128 (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.ldr_f128 (amode_base_offset a k))
      (spec (= (float_from_bits 128 (concat (select (mem) (bvadd (bvadd a k) 15)) (select (mem) (bvadd (bvadd a k) 14))
                       (select (mem) (bvadd (bvadd a k) 13)) (select (mem) (bvadd (bvadd a k) 12))
                       (select (mem) (bvadd (bvadd a k) 11)) (select (mem) (bvadd (bvadd a k) 10))
                       (select (mem) (bvadd (bvadd a k) 9)) (select (mem) (bvadd (bvadd a k) 8))
                       (select (mem) (bvadd (bvadd a k) 7)) (select (mem) (bvadd (bvadd a k) 6))
                       (select (mem) (bvadd (bvadd a k) 5)) (select (mem) (bvadd (bvadd a k) 4))
                       (select (mem) (bvadd (bvadd a k) 3)) (select (mem) (bvadd (bvadd a k) 2))
                       (select (mem) (bvadd (bvadd a k) 1)) (select (mem) (bvadd a k)))) (result))))

(rule (lower (store.f128 (value.f128 v) (value.i64 a)))
      (a64.str_f128 (amode_base a) v)
      (spec (= (store (store (store (store (store (store (store (store (store (store (store (store (store (store (store (store (mem)
                       a (extract 7 0 (bits_from_float 128 v)))
                       (bvadd a 1) (extract 15 8 (bits_from_float 128 v)))
                       (bvadd a 2) (extract 23 16 (bits_from_float 128 v)))
                       (bvadd a 3) (extract 31 24 (bits_from_float 128 v)))
                       (bvadd a 4) (extract 39 32 (bits_from_float 128 v)))
                       (bvadd a 5) (extract 47 40 (bits_from_float 128 v)))
                       (bvadd a 6) (extract 55 48 (bits_from_float 128 v)))
                       (bvadd a 7) (extract 63 56 (bits_from_float 128 v)))
                       (bvadd a 8) (extract 71 64 (bits_from_float 128 v)))
                       (bvadd a 9) (extract 79 72 (bits_from_float 128 v)))
                       (bvadd a 10) (extract 87 80 (bits_from_float 128 v)))
                       (bvadd a 11) (extract 95 88 (bits_from_float 128 v)))
                       (bvadd a 12) (extract 103 96 (bits_from_float 128 v)))
                       (bvadd a 13) (extract 111 104 (bits_from_float 128 v)))
                       (bvadd a 14) (extract 119 112 (bits_from_float 128 v)))
                       (bvadd a 15) (extract 127 120 (bits_from_float 128 v))) (result))))

(rule (lower (store.f128 (value.f128 v) (add.i64 (value.i64 a) (iconst.i64 k))))
      (if (and (>= k -256) (< k 256)))
      (a64.str_f128 (amode_base_offset a k) v)
      (spec (= (store (store (store (store (store (store (store (store (store (store (store (store (store (store (store (store (mem)
                       (bvadd a k) (extract 7 0 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 1) (extract 15 8 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 2) (extract 23 16 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 3) (extract 31 24 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 4) (extract 39 32 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 5) (extract 47 40 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 6) (extract 55 48 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 7) (extract 63 56 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 8) (extract 71 64 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 9) (extract 79 72 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 10) (extract 87 80 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 11) (extract 95 88 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 12) (extract 103 96 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 13) (extract 111 104 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 14) (extract 119 112 (bits_from_float 128 v)))
                       (bvadd (bvadd a k) 15) (extract 127 120 (bits_from_float 128 v))) (result))))

;; Conversions with a float on one side. The instruction names put the destination format second
;; for `fcvt` and the source first for the rest, which is how `rucc_target::aarch64` spells them.

(rule (lower (fpext.f32.f64 (value.f32 x)))
      (a64.fcvt_f32_f64 x)
      (spec (= (float_from_float 32 64 x) (result))))

(rule (lower (fptrunc.f64.f32 (value.f64 x)))
      (a64.fcvt_f64_f32 x)
      (spec (= (float_from_float 64 32 x) (result))))

(rule (lower (fptosi.f32.i32 (value.f32 x)))
      (a64.fcvtzs_f32_32 x)
      (spec (= (signed_from_float 32 32 x) (result))))

(rule (lower (fptosi.f32.i64 (value.f32 x)))
      (a64.fcvtzs_f32_64 x)
      (spec (= (signed_from_float 32 64 x) (result))))

(rule (lower (fptosi.f64.i32 (value.f64 x)))
      (a64.fcvtzs_f64_32 x)
      (spec (= (signed_from_float 64 32 x) (result))))

(rule (lower (fptosi.f64.i64 (value.f64 x)))
      (a64.fcvtzs_f64_64 x)
      (spec (= (signed_from_float 64 64 x) (result))))

(rule (lower (sitofp.i32.f32 (value.i32 x)))
      (a64.scvtf_32_f32 x)
      (spec (= (float_from_signed 32 32 x) (result))))

(rule (lower (sitofp.i32.f64 (value.i32 x)))
      (a64.scvtf_32_f64 x)
      (spec (= (float_from_signed 32 64 x) (result))))

(rule (lower (sitofp.i64.f32 (value.i64 x)))
      (a64.scvtf_64_f32 x)
      (spec (= (float_from_signed 64 32 x) (result))))

(rule (lower (sitofp.i64.f64 (value.i64 x)))
      (a64.scvtf_64_f64 x)
      (spec (= (float_from_signed 64 64 x) (result))))


;; Moving a register from one file to the other, which is what a bitcast is.

(rule (lower (bitcast.f32.i32 (value.f32 x)))
      (a64.fmov_from_f32 x)
      (spec (= (bits_from_float 32 x) (result))))

(rule (lower (bitcast.i32.f32 (value.i32 x)))
      (a64.fmov_to_f32 x)
      (spec (= (float_from_bits 32 x) (result))))

(rule (lower (bitcast.f64.i64 (value.f64 x)))
      (a64.fmov_from_f64 x)
      (spec (= (bits_from_float 64 x) (result))))

(rule (lower (bitcast.i64.f64 (value.i64 x)))
      (a64.fmov_to_f64 x)
      (spec (= (float_from_bits 64 x) (result))))