-
Notifications
You must be signed in to change notification settings - Fork 149
/
mir-gen-aarch64.c
2678 lines (2475 loc) · 108 KB
/
mir-gen-aarch64.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* This file is a part of MIR project.
Copyright (C) 2020-2024 Vladimir Makarov <[email protected]>.
*/
static void fancy_abort (int code) {
if (!code) abort ();
}
#undef gen_assert
#define gen_assert(c) fancy_abort (c)
#include <limits.h>
#include "mir-aarch64.h"
static const MIR_reg_t FP_HARD_REG = R29_HARD_REG;
static const MIR_reg_t LINK_HARD_REG = R30_HARD_REG;
static inline MIR_reg_t target_nth_loc (MIR_reg_t loc, MIR_type_t type MIR_UNUSED, int n) {
return loc + n;
}
static inline int target_call_used_hard_reg_p (MIR_reg_t hard_reg, MIR_type_t type) {
assert (hard_reg <= MAX_HARD_REG);
if (hard_reg <= SP_HARD_REG) return !(hard_reg >= R19_HARD_REG && hard_reg <= R28_HARD_REG);
return type == MIR_T_LD || !(hard_reg >= V8_HARD_REG && hard_reg <= V15_HARD_REG);
}
/* Stack layout (sp refers to the last reserved stack slot address)
from higher address to lower address memory:
| ... | prev func stack (start aligned to 16 bytes)
|---------------|
| gr save area | 64 bytes optional area for vararg func integer reg save area (absent for
APPLE)
|---------------|
| vr save area | 128 bytes optional area for vararg func fp reg save area (absent for APPLE)
|---------------|
| saved regs | callee saved regs used in the func (known only after RA), rounded 16 bytes
|---------------|
| slots assigned| can be absent for small functions (known only after RA), rounded 16 bytes
| to pseudos |
|---------------|
| previous | 16-bytes setup in prolog, used only for varag func or args passed on stack
| stack start | to move args and to setup va_start on machinize pass
|---------------|
| LR | sp before prologue and after saving LR = start sp
|---------------|
| old FP | frame pointer for previous func stack frame; new FP refers for here
| | it has lowest address as 12-bit offsets are only positive
|---------------|
| small aggr |
| save area | optional
|---------------|
| alloca areas | optional
|---------------|
| slots for | dynamically allocated/deallocated by caller
| passing args |
size of slots and saved regs is multiple of 16 bytes
*/
#if !defined(__APPLE__)
static const int int_reg_save_area_size = 8 * 8;
static const int reg_save_area_size = 8 * 8 + 8 * 16;
#endif
static const MIR_insn_code_t target_io_dup_op_insn_codes[] = {MIR_INSN_BOUND};
static MIR_insn_code_t get_ext_code (MIR_type_t type) {
switch (type) {
case MIR_T_I8: return MIR_EXT8;
case MIR_T_U8: return MIR_UEXT8;
case MIR_T_I16: return MIR_EXT16;
case MIR_T_U16: return MIR_UEXT16;
case MIR_T_I32: return MIR_EXT32;
case MIR_T_U32: return MIR_UEXT32;
default: return MIR_INVALID_INSN;
}
}
static MIR_reg_t get_arg_reg (MIR_type_t arg_type, size_t *int_arg_num, size_t *fp_arg_num,
MIR_insn_code_t *mov_code) {
MIR_reg_t arg_reg;
if (arg_type == MIR_T_F || arg_type == MIR_T_D || arg_type == MIR_T_LD) {
switch (*fp_arg_num) {
case 0:
case 1:
case 2:
case 3:
case 4:
case 5:
case 6:
case 7: arg_reg = V0_HARD_REG + *fp_arg_num; break;
default: arg_reg = MIR_NON_VAR; break;
}
(*fp_arg_num)++;
*mov_code = arg_type == MIR_T_F ? MIR_FMOV : arg_type == MIR_T_D ? MIR_DMOV : MIR_LDMOV;
} else { /* including BLK, RBLK: */
switch (*int_arg_num) {
case 0:
case 1:
case 2:
case 3:
case 4:
case 5:
case 6:
case 7: arg_reg = R0_HARD_REG + *int_arg_num; break;
default: arg_reg = MIR_NON_VAR; break;
}
(*int_arg_num)++;
*mov_code = MIR_MOV;
}
return arg_reg;
}
static void mir_blk_mov (uint64_t *to, uint64_t *from, uint64_t nwords) {
for (; nwords > 0; nwords--) *to++ = *from++;
}
static MIR_insn_t gen_mov (gen_ctx_t gen_ctx, MIR_insn_t anchor, MIR_insn_code_t code,
MIR_op_t dst_op, MIR_op_t src_op) {
MIR_insn_t insn = MIR_new_insn (gen_ctx->ctx, code, dst_op, src_op);
gen_add_insn_before (gen_ctx, anchor, insn);
return insn;
}
static MIR_reg_t target_get_stack_slot_base_reg (gen_ctx_t gen_ctx MIR_UNUSED) {
return FP_HARD_REG;
}
static int target_valid_mem_offset_p (gen_ctx_t gen_ctx, MIR_type_t type, MIR_disp_t offset);
static MIR_op_t new_mem_op (gen_ctx_t gen_ctx, MIR_insn_t anchor, MIR_type_t type, MIR_disp_t disp,
MIR_reg_t base) {
MIR_context_t ctx = gen_ctx->ctx;
if (target_valid_mem_offset_p (gen_ctx, type, disp))
return _MIR_new_var_mem_op (ctx, type, disp, base, MIR_NON_VAR, 1);
MIR_reg_t temp_reg = gen_new_temp_reg (gen_ctx, MIR_T_I64, curr_func_item->u.func);
MIR_op_t temp_reg_op = _MIR_new_var_op (ctx, temp_reg);
gen_mov (gen_ctx, anchor, MIR_MOV, temp_reg_op, MIR_new_int_op (ctx, disp));
gen_add_insn_before (gen_ctx, anchor,
MIR_new_insn (ctx, MIR_ADD, temp_reg_op, temp_reg_op,
_MIR_new_var_op (ctx, base)));
return _MIR_new_var_mem_op (ctx, type, 0, temp_reg, MIR_NON_VAR, 1);
}
static MIR_op_t get_new_hard_reg_mem_op (gen_ctx_t gen_ctx, MIR_type_t type, MIR_disp_t disp,
MIR_reg_t base, MIR_insn_t *insn1, MIR_insn_t *insn2) {
MIR_context_t ctx = gen_ctx->ctx;
*insn1 = *insn2 = NULL;
if (target_valid_mem_offset_p (gen_ctx, type, disp))
return _MIR_new_var_mem_op (ctx, type, disp, base, MIR_NON_VAR, 1);
MIR_op_t temp_reg_op = _MIR_new_var_op (ctx, TEMP_INT_HARD_REG2);
*insn1 = MIR_new_insn (ctx, MIR_MOV, temp_reg_op, MIR_new_int_op (ctx, disp));
*insn2 = MIR_new_insn (ctx, MIR_ADD, temp_reg_op, temp_reg_op, _MIR_new_var_op (ctx, base));
return _MIR_new_var_mem_op (ctx, type, 0, TEMP_INT_HARD_REG2, MIR_NON_VAR, 1);
}
static MIR_op_t new_hard_reg_mem_op (gen_ctx_t gen_ctx, MIR_insn_t anchor, MIR_type_t type,
MIR_disp_t disp, MIR_reg_t base) {
MIR_insn_t insn1, insn2;
MIR_op_t op = get_new_hard_reg_mem_op (gen_ctx, type, disp, base, &insn1, &insn2);
if (insn1 != NULL) gen_add_insn_before (gen_ctx, anchor, insn1);
if (insn2 != NULL) gen_add_insn_before (gen_ctx, anchor, insn2);
return op;
}
static const char *BLK_MOV = "mir.blk_mov";
static const char *BLK_MOV_P = "mir.blk_mov.p";
static void gen_blk_mov (gen_ctx_t gen_ctx, MIR_insn_t anchor, size_t to_disp,
MIR_reg_t to_base_hard_reg, size_t from_disp, MIR_reg_t from_base_reg,
size_t qwords, int save_regs) {
MIR_context_t ctx = gen_ctx->ctx;
MIR_func_t func = curr_func_item->u.func;
MIR_item_t proto_item, func_import_item;
MIR_insn_t new_insn;
MIR_op_t ops[5], freg_op, treg_op, treg_op2, treg_op3;
treg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
if (qwords <= 16) {
for (; qwords > 0; qwords--, to_disp += 8, from_disp += 8) {
gen_mov (gen_ctx, anchor, MIR_MOV, treg_op,
new_mem_op (gen_ctx, anchor, MIR_T_I64, from_disp, from_base_reg));
gen_mov (gen_ctx, anchor, MIR_MOV,
new_hard_reg_mem_op (gen_ctx, anchor, MIR_T_I64, to_disp, to_base_hard_reg),
treg_op);
}
return;
}
treg_op2 = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
treg_op3 = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
/* Save arg regs: */
if (save_regs > 0)
gen_mov (gen_ctx, anchor, MIR_MOV, treg_op, _MIR_new_var_op (ctx, R0_HARD_REG));
if (save_regs > 1)
gen_mov (gen_ctx, anchor, MIR_MOV, treg_op2, _MIR_new_var_op (ctx, R1_HARD_REG));
if (save_regs > 2)
gen_mov (gen_ctx, anchor, MIR_MOV, treg_op3, _MIR_new_var_op (ctx, R2_HARD_REG));
/* call blk move: */
proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, BLK_MOV_P, 0, NULL, 3, MIR_T_I64,
"to", MIR_T_I64, "from", MIR_T_I64, "nwords");
func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, BLK_MOV, mir_blk_mov);
freg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
new_insn = MIR_new_insn (ctx, MIR_MOV, freg_op, MIR_new_ref_op (ctx, func_import_item));
gen_add_insn_before (gen_ctx, anchor, new_insn);
gen_add_insn_before (gen_ctx, anchor,
MIR_new_insn (gen_ctx->ctx, MIR_ADD, _MIR_new_var_op (ctx, R0_HARD_REG),
_MIR_new_var_op (ctx, to_base_hard_reg),
MIR_new_int_op (ctx, to_disp)));
gen_add_insn_before (gen_ctx, anchor,
MIR_new_insn (gen_ctx->ctx, MIR_ADD, _MIR_new_var_op (ctx, R1_HARD_REG),
_MIR_new_var_op (ctx, from_base_reg),
MIR_new_int_op (ctx, from_disp)));
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R2_HARD_REG),
MIR_new_int_op (ctx, qwords));
ops[0] = MIR_new_ref_op (ctx, proto_item);
ops[1] = freg_op;
ops[2] = _MIR_new_var_op (ctx, R0_HARD_REG);
ops[3] = _MIR_new_var_op (ctx, R1_HARD_REG);
ops[4] = _MIR_new_var_op (ctx, R2_HARD_REG);
new_insn = MIR_new_insn_arr (ctx, MIR_CALL, 5, ops);
gen_add_insn_before (gen_ctx, anchor, new_insn);
/* Restore arg regs: */
if (save_regs > 0)
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R0_HARD_REG), treg_op);
if (save_regs > 1)
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R1_HARD_REG), treg_op2);
if (save_regs > 2)
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R2_HARD_REG), treg_op3);
}
static void machinize_call (gen_ctx_t gen_ctx, MIR_insn_t call_insn) {
MIR_context_t ctx = gen_ctx->ctx;
MIR_func_t func = curr_func_item->u.func;
MIR_proto_t proto = call_insn->ops[0].u.ref->u.proto;
size_t nargs, nops = MIR_insn_nops (ctx, call_insn), start = proto->nres + 2;
size_t int_arg_num = 0, fp_arg_num = 0, mem_size = 0, blk_offset = 0, qwords;
MIR_type_t type, mem_type;
MIR_op_mode_t mode;
MIR_var_t *arg_vars = NULL;
MIR_reg_t arg_reg;
MIR_op_t arg_op, temp_op, arg_reg_op, ret_reg_op, mem_op;
MIR_insn_code_t new_insn_code, ext_code;
MIR_insn_t new_insn, prev_insn, next_insn, ext_insn, insn1, insn2;
MIR_insn_t prev_call_insn = DLIST_PREV (MIR_insn_t, call_insn);
MIR_insn_t curr_prev_call_insn = prev_call_insn;
uint32_t n_iregs, n_vregs;
if (call_insn->code == MIR_INLINE) call_insn->code = MIR_CALL;
if (proto->args == NULL) {
nargs = 0;
} else {
gen_assert (nops >= VARR_LENGTH (MIR_var_t, proto->args)
&& (proto->vararg_p || nops - start == VARR_LENGTH (MIR_var_t, proto->args)));
nargs = VARR_LENGTH (MIR_var_t, proto->args);
arg_vars = VARR_ADDR (MIR_var_t, proto->args);
}
if (call_insn->ops[1].mode != MIR_OP_VAR) {
// ??? to optimize (can be immediate operand for func call)
temp_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
new_insn = MIR_new_insn (ctx, MIR_MOV, temp_op, call_insn->ops[1]);
call_insn->ops[1] = temp_op;
gen_add_insn_before (gen_ctx, call_insn, new_insn);
}
for (size_t i = start; i < nops; i++) { /* calculate offset for blk params */
if (i - start < nargs) {
type = arg_vars[i - start].type;
} else if (call_insn->ops[i].mode == MIR_OP_VAR_MEM) {
type = call_insn->ops[i].u.var_mem.type;
gen_assert (MIR_all_blk_type_p (type));
} else {
mode = call_insn->ops[i].value_mode; // ??? smaller ints
gen_assert (mode == MIR_OP_INT || mode == MIR_OP_UINT || mode == MIR_OP_FLOAT
|| mode == MIR_OP_DOUBLE || mode == MIR_OP_LDOUBLE);
if (mode == MIR_OP_FLOAT)
(*MIR_get_error_func (ctx)) (MIR_call_op_error,
"passing float variadic arg (should be passed as double)");
if (mode == MIR_OP_LDOUBLE && __SIZEOF_LONG_DOUBLE__ == 8) mode = MIR_OP_DOUBLE;
type = mode == MIR_OP_DOUBLE ? MIR_T_D : mode == MIR_OP_LDOUBLE ? MIR_T_LD : MIR_T_I64;
}
gen_assert (!MIR_all_blk_type_p (type) || call_insn->ops[i].mode == MIR_OP_VAR_MEM);
if (type == MIR_T_RBLK && i == start) continue; /* hidden arg */
#if defined(__APPLE__) /* all varargs are passed on stack */
if (i - start == nargs) int_arg_num = fp_arg_num = 8;
#endif
if (MIR_blk_type_p (type) && (qwords = (call_insn->ops[i].u.var_mem.disp + 7) / 8) <= 2) {
if (int_arg_num + qwords > 8) blk_offset += qwords * 8;
int_arg_num += qwords;
} else if (get_arg_reg (type, &int_arg_num, &fp_arg_num, &new_insn_code) == MIR_NON_VAR) {
if (type == MIR_T_LD && __SIZEOF_LONG_DOUBLE__ == 16 && blk_offset % 16 != 0)
blk_offset = (blk_offset + 15) / 16 * 16;
blk_offset += type == MIR_T_LD && __SIZEOF_LONG_DOUBLE__ == 16 ? 16 : 8;
}
}
blk_offset = (blk_offset + 15) / 16 * 16;
int_arg_num = fp_arg_num = 0;
for (size_t i = start; i < nops; i++) {
#if defined(__APPLE__) /* all varargs are passed on stack */
if (i - start == nargs) int_arg_num = fp_arg_num = 8;
#endif
arg_op = call_insn->ops[i];
gen_assert (arg_op.mode == MIR_OP_VAR
|| (arg_op.mode == MIR_OP_VAR_MEM && MIR_all_blk_type_p (arg_op.u.var_mem.type)));
if (i - start < nargs) {
type = arg_vars[i - start].type;
} else if (call_insn->ops[i].mode == MIR_OP_VAR_MEM) {
type = call_insn->ops[i].u.var_mem.type;
gen_assert (MIR_all_blk_type_p (type));
} else {
mode = call_insn->ops[i].value_mode; // ??? smaller ints
if (mode == MIR_OP_LDOUBLE && __SIZEOF_LONG_DOUBLE__ == 8) mode = MIR_OP_DOUBLE;
type = mode == MIR_OP_DOUBLE ? MIR_T_D : mode == MIR_OP_LDOUBLE ? MIR_T_LD : MIR_T_I64;
}
ext_insn = NULL;
if ((ext_code = get_ext_code (type)) != MIR_INVALID_INSN) { /* extend arg if necessary */
temp_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
ext_insn = MIR_new_insn (ctx, ext_code, temp_op, arg_op);
call_insn->ops[i] = arg_op = temp_op;
}
gen_assert (!MIR_all_blk_type_p (type)
|| (arg_op.mode == MIR_OP_VAR_MEM && arg_op.u.var_mem.disp >= 0
&& arg_op.u.var_mem.index == MIR_NON_VAR));
if (type == MIR_T_RBLK && i == start) { /* hidden arg */
arg_reg_op = _MIR_new_var_op (ctx, R8_HARD_REG);
gen_mov (gen_ctx, call_insn, MIR_MOV, arg_reg_op,
_MIR_new_var_op (ctx, arg_op.u.var_mem.base));
call_insn->ops[i] = arg_reg_op;
continue;
} else if (MIR_blk_type_p (type)) {
qwords = (arg_op.u.var_mem.disp + 7) / 8;
if (qwords <= 2) {
arg_reg = R0_HARD_REG + int_arg_num;
if (int_arg_num + qwords <= 8) {
/* A trick to keep arg regs live: */
call_insn->ops[i] = _MIR_new_var_mem_op (ctx, MIR_T_UNDEF, 0, int_arg_num,
qwords < 2 ? MIR_NON_VAR : int_arg_num + 1, 1);
if (qwords == 0) continue;
new_insn = MIR_new_insn (ctx, MIR_MOV, _MIR_new_var_op (ctx, R0_HARD_REG + int_arg_num++),
_MIR_new_var_mem_op (ctx, MIR_T_I64, 0, arg_op.u.var_mem.base,
MIR_NON_VAR, 1));
gen_add_insn_before (gen_ctx, call_insn, new_insn);
if (qwords == 2) {
new_insn
= MIR_new_insn (ctx, MIR_MOV, _MIR_new_var_op (ctx, R0_HARD_REG + int_arg_num++),
_MIR_new_var_mem_op (ctx, MIR_T_I64, 8, arg_op.u.var_mem.base,
MIR_NON_VAR, 1));
gen_add_insn_before (gen_ctx, call_insn, new_insn);
}
} else { /* pass on stack w/o address: */
gen_blk_mov (gen_ctx, call_insn, mem_size, SP_HARD_REG, 0, arg_op.u.var_mem.base, qwords,
int_arg_num);
call_insn->ops[i]
= _MIR_new_var_mem_op (ctx, MIR_T_UNDEF,
mem_size, /* we don't care about valid mem disp here */
SP_HARD_REG, MIR_NON_VAR, 1);
mem_size += qwords * 8;
blk_offset += qwords * 8;
int_arg_num += qwords;
}
continue;
}
gen_blk_mov (gen_ctx, call_insn, blk_offset, SP_HARD_REG, 0, arg_op.u.var_mem.base, qwords,
int_arg_num);
arg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
gen_assert (curr_prev_call_insn
!= NULL); /* call_insn should not be 1st after simplification */
new_insn = MIR_new_insn (gen_ctx->ctx, MIR_ADD, arg_op, _MIR_new_var_op (ctx, SP_HARD_REG),
MIR_new_int_op (ctx, blk_offset));
gen_add_insn_after (gen_ctx, curr_prev_call_insn, new_insn);
curr_prev_call_insn = DLIST_NEXT (MIR_insn_t, new_insn);
blk_offset += qwords * 8;
}
if ((arg_reg = get_arg_reg (type, &int_arg_num, &fp_arg_num, &new_insn_code)) != MIR_NON_VAR) {
/* put arguments to argument hard regs */
if (ext_insn != NULL) gen_add_insn_before (gen_ctx, call_insn, ext_insn);
arg_reg_op = _MIR_new_var_op (ctx, arg_reg);
if (type != MIR_T_RBLK) {
new_insn = MIR_new_insn (ctx, new_insn_code, arg_reg_op, arg_op);
} else {
assert (arg_op.mode == MIR_OP_VAR_MEM);
new_insn
= MIR_new_insn (ctx, MIR_MOV, arg_reg_op, _MIR_new_var_op (ctx, arg_op.u.var_mem.base));
arg_reg_op
= _MIR_new_var_mem_op (ctx, MIR_T_RBLK,
arg_op.u.var_mem.disp, /* we don't care about valid disp here */
arg_reg, MIR_NON_VAR, 1);
}
gen_add_insn_before (gen_ctx, call_insn, new_insn);
call_insn->ops[i] = arg_reg_op;
} else { /* put arguments on the stack */
if (type == MIR_T_LD && __SIZEOF_LONG_DOUBLE__ == 16 && mem_size % 16 != 0)
mem_size = (mem_size + 15) / 16 * 16;
mem_type = type == MIR_T_F || type == MIR_T_D || type == MIR_T_LD ? type : MIR_T_I64;
new_insn_code = (type == MIR_T_F ? MIR_FMOV
: type == MIR_T_D ? MIR_DMOV
: type == MIR_T_LD ? MIR_LDMOV
: MIR_MOV);
mem_op = get_new_hard_reg_mem_op (gen_ctx, mem_type, mem_size, SP_HARD_REG, &insn1, &insn2);
if (type != MIR_T_RBLK) {
new_insn = MIR_new_insn (ctx, new_insn_code, mem_op, arg_op);
} else {
assert (arg_op.mode == MIR_OP_VAR_MEM);
new_insn
= MIR_new_insn (ctx, new_insn_code, mem_op, _MIR_new_var_op (ctx, arg_op.u.var_mem.base));
}
gen_assert (curr_prev_call_insn
!= NULL); /* call_insn should not be 1st after simplification */
MIR_insert_insn_after (ctx, curr_func_item, curr_prev_call_insn, new_insn);
if (insn2 != NULL) MIR_insert_insn_after (ctx, curr_func_item, curr_prev_call_insn, insn2);
if (insn1 != NULL) MIR_insert_insn_after (ctx, curr_func_item, curr_prev_call_insn, insn1);
prev_insn = curr_prev_call_insn;
next_insn = DLIST_NEXT (MIR_insn_t, new_insn);
create_new_bb_insns (gen_ctx, prev_insn, next_insn, call_insn);
call_insn->ops[i] = mem_op;
mem_size += type == MIR_T_LD && __SIZEOF_LONG_DOUBLE__ == 16 ? 16 : 8;
if (ext_insn != NULL) gen_add_insn_after (gen_ctx, curr_prev_call_insn, ext_insn);
curr_prev_call_insn = new_insn;
}
}
blk_offset = (blk_offset + 15) / 16 * 16;
if (blk_offset != 0) mem_size = blk_offset;
n_iregs = n_vregs = 0;
for (size_t i = 0; i < proto->nres; i++) {
int float_p;
ret_reg_op = call_insn->ops[i + 2];
gen_assert (ret_reg_op.mode == MIR_OP_VAR);
type = proto->res_types[i];
float_p = type == MIR_T_F || type == MIR_T_D || type == MIR_T_LD;
if (float_p && n_vregs < 8) {
new_insn = MIR_new_insn (ctx,
type == MIR_T_F ? MIR_FMOV
: type == MIR_T_D ? MIR_DMOV
: MIR_LDMOV,
ret_reg_op, _MIR_new_var_op (ctx, V0_HARD_REG + n_vregs));
n_vregs++;
} else if (!float_p && n_iregs < 8) {
new_insn
= MIR_new_insn (ctx, MIR_MOV, ret_reg_op, _MIR_new_var_op (ctx, R0_HARD_REG + n_iregs));
n_iregs++;
} else {
(*MIR_get_error_func (ctx)) (MIR_ret_error,
"aarch64 can not handle this combination of return values");
}
MIR_insert_insn_after (ctx, curr_func_item, call_insn, new_insn);
call_insn->ops[i + 2] = new_insn->ops[1];
if ((ext_code = get_ext_code (type)) != MIR_INVALID_INSN) {
MIR_insert_insn_after (ctx, curr_func_item, new_insn,
MIR_new_insn (ctx, ext_code, ret_reg_op, ret_reg_op));
new_insn = DLIST_NEXT (MIR_insn_t, new_insn);
}
create_new_bb_insns (gen_ctx, call_insn, DLIST_NEXT (MIR_insn_t, new_insn), call_insn);
}
if (mem_size != 0) { /* allocate/deallocate stack for args passed on stack */
mem_size = (mem_size + 15) / 16 * 16; /* make it of several 16 bytes */
new_insn = MIR_new_insn (ctx, MIR_SUB, _MIR_new_var_op (ctx, SP_HARD_REG),
_MIR_new_var_op (ctx, SP_HARD_REG), MIR_new_int_op (ctx, mem_size));
MIR_insert_insn_after (ctx, curr_func_item, prev_call_insn, new_insn);
next_insn = DLIST_NEXT (MIR_insn_t, new_insn);
create_new_bb_insns (gen_ctx, prev_call_insn, next_insn, call_insn);
new_insn = MIR_new_insn (ctx, MIR_ADD, _MIR_new_var_op (ctx, SP_HARD_REG),
_MIR_new_var_op (ctx, SP_HARD_REG), MIR_new_int_op (ctx, mem_size));
MIR_insert_insn_after (ctx, curr_func_item, call_insn, new_insn);
next_insn = DLIST_NEXT (MIR_insn_t, new_insn);
create_new_bb_insns (gen_ctx, call_insn, next_insn, call_insn);
}
}
static long double mir_i2ld (int64_t i) { return i; }
static const char *I2LD = "mir.i2ld";
static const char *I2LD_P = "mir.i2ld.p";
static long double mir_ui2ld (uint64_t i) { return i; }
static const char *UI2LD = "mir.ui2ld";
static const char *UI2LD_P = "mir.ui2ld.p";
static long double mir_f2ld (float f) { return f; }
static const char *F2LD = "mir.f2ld";
static const char *F2LD_P = "mir.f2ld.p";
static long double mir_d2ld (double d) { return d; }
static const char *D2LD = "mir.d2ld";
static const char *D2LD_P = "mir.d2ld.p";
static int64_t mir_ld2i (long double ld) { return ld; }
static const char *LD2I = "mir.ld2i";
static const char *LD2I_P = "mir.ld2i.p";
static float mir_ld2f (long double ld) { return ld; }
static const char *LD2F = "mir.ld2f";
static const char *LD2F_P = "mir.ld2f.p";
static double mir_ld2d (long double ld) { return ld; }
static const char *LD2D = "mir.ld2d";
static const char *LD2D_P = "mir.ld2d.p";
static long double mir_ldadd (long double d1, long double d2) { return d1 + d2; }
static const char *LDADD = "mir.ldadd";
static const char *LDADD_P = "mir.ldadd.p";
static long double mir_ldsub (long double d1, long double d2) { return d1 - d2; }
static const char *LDSUB = "mir.ldsub";
static const char *LDSUB_P = "mir.ldsub.p";
static long double mir_ldmul (long double d1, long double d2) { return d1 * d2; }
static const char *LDMUL = "mir.ldmul";
static const char *LDMUL_P = "mir.ldmul.p";
static long double mir_lddiv (long double d1, long double d2) { return d1 / d2; }
static const char *LDDIV = "mir.lddiv";
static const char *LDDIV_P = "mir.lddiv.p";
static long double mir_ldneg (long double d) { return -d; }
static const char *LDNEG = "mir.ldneg";
static const char *LDNEG_P = "mir.ldneg.p";
static const char *VA_ARG_P = "mir.va_arg.p";
static const char *VA_ARG = "mir.va_arg";
static const char *VA_BLOCK_ARG_P = "mir.va_block_arg.p";
static const char *VA_BLOCK_ARG = "mir.va_block_arg";
static int64_t mir_ldeq (long double d1, long double d2) { return d1 == d2; }
static const char *LDEQ = "mir.ldeq";
static const char *LDEQ_P = "mir.ldeq.p";
static int64_t mir_ldne (long double d1, long double d2) { return d1 != d2; }
static const char *LDNE = "mir.ldne";
static const char *LDNE_P = "mir.ldne.p";
static int64_t mir_ldlt (long double d1, long double d2) { return d1 < d2; }
static const char *LDLT = "mir.ldlt";
static const char *LDLT_P = "mir.ldlt.p";
static int64_t mir_ldge (long double d1, long double d2) { return d1 >= d2; }
static const char *LDGE = "mir.ldge";
static const char *LDGE_P = "mir.ldge.p";
static int64_t mir_ldgt (long double d1, long double d2) { return d1 > d2; }
static const char *LDGT = "mir.ldgt";
static const char *LDGT_P = "mir.ldgt.p";
static int64_t mir_ldle (long double d1, long double d2) { return d1 <= d2; }
static const char *LDLE = "mir.ldle";
static const char *LDLE_P = "mir.ldle.p";
static int get_builtin (gen_ctx_t gen_ctx, MIR_insn_code_t code, MIR_item_t *proto_item,
MIR_item_t *func_import_item) {
MIR_context_t ctx = gen_ctx->ctx;
MIR_type_t res_type;
*func_import_item = *proto_item = NULL; /* to remove uninitialized warning */
switch (code) {
case MIR_I2LD:
res_type = MIR_T_LD;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, I2LD_P, 1, &res_type, 1, MIR_T_I64, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, I2LD, mir_i2ld);
return 1;
case MIR_UI2LD:
res_type = MIR_T_LD;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, UI2LD_P, 1, &res_type, 1, MIR_T_I64, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, UI2LD, mir_ui2ld);
return 1;
case MIR_F2LD:
res_type = MIR_T_LD;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, F2LD_P, 1, &res_type, 1, MIR_T_F, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, F2LD, mir_f2ld);
return 1;
case MIR_D2LD:
res_type = MIR_T_LD;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, D2LD_P, 1, &res_type, 1, MIR_T_D, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, D2LD, mir_d2ld);
return 1;
case MIR_LD2I:
res_type = MIR_T_I64;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, LD2I_P, 1, &res_type, 1, MIR_T_LD, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LD2I, mir_ld2i);
return 1;
case MIR_LD2F:
res_type = MIR_T_F;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, LD2F_P, 1, &res_type, 1, MIR_T_LD, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LD2F, mir_ld2f);
return 1;
case MIR_LD2D:
res_type = MIR_T_D;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, LD2D_P, 1, &res_type, 1, MIR_T_LD, "v");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LD2D, mir_ld2d);
return 1;
case MIR_LDADD:
res_type = MIR_T_LD;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDADD_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDADD, mir_ldadd);
return 2;
case MIR_LDSUB:
res_type = MIR_T_LD;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDSUB_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDSUB, mir_ldsub);
return 2;
case MIR_LDMUL:
res_type = MIR_T_LD;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDMUL_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDMUL, mir_ldmul);
return 2;
case MIR_LDDIV:
res_type = MIR_T_LD;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDDIV_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDDIV, mir_lddiv);
return 2;
case MIR_LDNEG:
res_type = MIR_T_LD;
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, LDNEG_P, 1, &res_type, 1, MIR_T_LD, "d");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDNEG, mir_ldneg);
return 1;
case MIR_LDEQ:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDEQ_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDEQ, mir_ldeq);
return 2;
case MIR_LDNE:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDNE_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDNE, mir_ldne);
return 2;
case MIR_LDLT:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDLT_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDLT, mir_ldlt);
return 2;
case MIR_LDGE:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDGE_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDGE, mir_ldge);
return 2;
case MIR_LDGT:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDGT_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDGT, mir_ldgt);
return 2;
case MIR_LDLE:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, LDLE_P, 1, &res_type, 2,
MIR_T_LD, "d1", MIR_T_LD, "d2");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, LDLE, mir_ldle);
return 2;
case MIR_VA_ARG:
res_type = MIR_T_I64;
*proto_item = _MIR_builtin_proto (ctx, curr_func_item->module, VA_ARG_P, 1, &res_type, 2,
MIR_T_I64, "va", MIR_T_I64, "type");
*func_import_item = _MIR_builtin_func (ctx, curr_func_item->module, VA_ARG, va_arg_builtin);
return 2;
case MIR_VA_BLOCK_ARG:
*proto_item
= _MIR_builtin_proto (ctx, curr_func_item->module, VA_BLOCK_ARG_P, 0, NULL, 4, MIR_T_I64,
"res", MIR_T_I64, "va", MIR_T_I64, "size", MIR_T_I64, "ncase");
*func_import_item
= _MIR_builtin_func (ctx, curr_func_item->module, VA_BLOCK_ARG, va_block_arg_builtin);
return 4;
default: return 0;
}
}
struct insn_pattern_info {
int start, num;
};
typedef struct insn_pattern_info insn_pattern_info_t;
DEF_VARR (insn_pattern_info_t);
struct label_ref {
int abs_addr_p, short_p;
size_t label_val_disp;
union {
MIR_label_t label;
void *jump_addr; /* absolute addr for BBV */
} u;
};
typedef struct label_ref label_ref_t;
DEF_VARR (label_ref_t);
struct target_ctx {
unsigned char alloca_p, block_arg_func_p, leaf_p, short_bb_branch_p;
size_t small_aggregate_save_area;
MIR_insn_t temp_jump;
const char *temp_jump_replacement;
VARR (int) * pattern_indexes;
VARR (insn_pattern_info_t) * insn_pattern_info;
VARR (uint8_t) * result_code;
VARR (label_ref_t) * label_refs;
VARR (uint64_t) * abs_address_locs;
VARR (MIR_code_reloc_t) * relocs;
};
#define alloca_p gen_ctx->target_ctx->alloca_p
#define block_arg_func_p gen_ctx->target_ctx->block_arg_func_p
#define leaf_p gen_ctx->target_ctx->leaf_p
#define short_bb_branch_p gen_ctx->target_ctx->short_bb_branch_p
#define small_aggregate_save_area gen_ctx->target_ctx->small_aggregate_save_area
#define temp_jump gen_ctx->target_ctx->temp_jump
#define temp_jump_replacement gen_ctx->target_ctx->temp_jump_replacement
#define pattern_indexes gen_ctx->target_ctx->pattern_indexes
#define insn_pattern_info gen_ctx->target_ctx->insn_pattern_info
#define result_code gen_ctx->target_ctx->result_code
#define label_refs gen_ctx->target_ctx->label_refs
#define abs_address_locs gen_ctx->target_ctx->abs_address_locs
#define relocs gen_ctx->target_ctx->relocs
static MIR_disp_t target_get_stack_slot_offset (gen_ctx_t gen_ctx, MIR_type_t type MIR_UNUSED,
MIR_reg_t slot) {
/* slot is 0, 1, ... */
size_t offset = curr_func_item->u.func->vararg_p || block_arg_func_p ? 32 : 16;
return ((MIR_disp_t) slot * 8 + offset);
}
static int target_valid_mem_offset_p (gen_ctx_t gen_ctx MIR_UNUSED, MIR_type_t type,
MIR_disp_t offset) {
int scale;
switch (type) {
case MIR_T_I8:
case MIR_T_U8: scale = 1; break;
case MIR_T_I16:
case MIR_T_U16: scale = 2; break;
case MIR_T_I32:
case MIR_T_U32:
#if MIR_PTR32
case MIR_T_P:
#endif
case MIR_T_F: scale = 4; break;
case MIR_T_LD: scale = 16; break;
default: scale = 8; break;
}
return offset >= 0 && offset % scale == 0 && offset / scale < (1 << 12);
}
static void target_machinize (gen_ctx_t gen_ctx) {
MIR_context_t ctx = gen_ctx->ctx;
MIR_func_t func;
MIR_type_t type, mem_type, res_type;
MIR_insn_code_t code, new_insn_code;
MIR_insn_t insn, next_insn, new_insn, anchor;
MIR_var_t var;
MIR_reg_t ret_reg, arg_reg;
MIR_op_t ret_reg_op, arg_reg_op, mem_op, temp_op;
size_t i, int_arg_num, fp_arg_num, mem_size, qwords;
assert (curr_func_item->item_type == MIR_func_item);
func = curr_func_item->u.func;
block_arg_func_p = FALSE;
anchor = DLIST_HEAD (MIR_insn_t, func->insns);
small_aggregate_save_area = 0;
for (i = int_arg_num = fp_arg_num = mem_size = 0; i < func->nargs; i++) {
/* Argument extensions is already done in simplify */
/* Prologue: generate arg_var = hard_reg|stack mem|stack addr ... */
var = VARR_GET (MIR_var_t, func->vars, i);
type = var.type;
if (type == MIR_T_RBLK && i == 0) { /* hidden arg */
arg_reg_op = _MIR_new_var_op (ctx, R8_HARD_REG);
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, i + MAX_HARD_REG + 1), arg_reg_op);
continue;
} else if (MIR_blk_type_p (type) && (qwords = (var.size + 7) / 8) <= 2) {
if (int_arg_num + qwords <= 8) {
small_aggregate_save_area += qwords * 8;
new_insn = MIR_new_insn (ctx, MIR_SUB, _MIR_new_var_op (ctx, i + MAX_HARD_REG + 1),
_MIR_new_var_op (ctx, FP_HARD_REG),
MIR_new_int_op (ctx, small_aggregate_save_area));
gen_add_insn_before (gen_ctx, anchor, new_insn);
if (qwords == 0) continue;
gen_mov (gen_ctx, anchor, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 0, i + MAX_HARD_REG + 1, MIR_NON_VAR, 1),
_MIR_new_var_op (ctx, int_arg_num));
if (qwords == 2)
gen_mov (gen_ctx, anchor, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 8, i + MAX_HARD_REG + 1, MIR_NON_VAR, 1),
_MIR_new_var_op (ctx, int_arg_num + 1));
} else { /* pass on stack w/o address: */
if (!block_arg_func_p) {
block_arg_func_p = TRUE;
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R8_HARD_REG),
_MIR_new_var_mem_op (ctx, MIR_T_I64, 16, FP_HARD_REG, MIR_NON_VAR, 1));
}
gen_add_insn_before (gen_ctx, anchor,
MIR_new_insn (ctx, MIR_ADD,
_MIR_new_var_op (ctx, i + MAX_HARD_REG + 1),
_MIR_new_var_op (ctx, R8_HARD_REG),
MIR_new_int_op (ctx, mem_size)));
mem_size += qwords * 8;
}
int_arg_num += qwords;
continue;
}
arg_reg = get_arg_reg (type, &int_arg_num, &fp_arg_num, &new_insn_code);
if (arg_reg != MIR_NON_VAR) {
arg_reg_op = _MIR_new_var_op (ctx, arg_reg);
gen_mov (gen_ctx, anchor, new_insn_code, _MIR_new_var_op (ctx, i + MAX_HARD_REG + 1),
arg_reg_op);
} else {
/* arg is on the stack */
if (!block_arg_func_p) {
block_arg_func_p = TRUE;
gen_mov (gen_ctx, anchor, MIR_MOV, _MIR_new_var_op (ctx, R8_HARD_REG),
_MIR_new_var_mem_op (ctx, MIR_T_I64, 16, FP_HARD_REG, MIR_NON_VAR, 1));
}
mem_type = type == MIR_T_F || type == MIR_T_D || type == MIR_T_LD ? type : MIR_T_I64;
if (type == MIR_T_LD) mem_size = (mem_size + 15) / 16 * 16;
new_insn_code = (type == MIR_T_F ? MIR_FMOV
: type == MIR_T_D ? MIR_DMOV
: type == MIR_T_LD ? MIR_LDMOV
: MIR_MOV);
mem_op = new_hard_reg_mem_op (gen_ctx, anchor, mem_type, mem_size, R8_HARD_REG);
gen_mov (gen_ctx, anchor, new_insn_code, _MIR_new_var_op (ctx, i + MAX_HARD_REG + 1), mem_op);
mem_size += type == MIR_T_LD ? 16 : 8;
}
}
alloca_p = FALSE;
leaf_p = TRUE;
for (insn = DLIST_HEAD (MIR_insn_t, func->insns); insn != NULL; insn = next_insn) {
MIR_item_t proto_item, func_import_item;
int nargs;
next_insn = DLIST_NEXT (MIR_insn_t, insn);
code = insn->code;
if (code == MIR_LDBEQ || code == MIR_LDBNE || code == MIR_LDBLT || code == MIR_LDBGE
|| code == MIR_LDBGT || code == MIR_LDBLE) {
temp_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
code = (code == MIR_LDBEQ ? MIR_LDEQ
: code == MIR_LDBNE ? MIR_LDNE
: code == MIR_LDBLT ? MIR_LDLT
: code == MIR_LDBGE ? MIR_LDGE
: code == MIR_LDBGT ? MIR_LDGT
: MIR_LDLE);
new_insn = MIR_new_insn (ctx, code, temp_op, insn->ops[1], insn->ops[2]);
gen_add_insn_before (gen_ctx, insn, new_insn);
next_insn = MIR_new_insn (ctx, MIR_BT, insn->ops[0], temp_op);
gen_add_insn_after (gen_ctx, new_insn, next_insn);
gen_delete_insn (gen_ctx, insn);
insn = new_insn;
}
if ((nargs = get_builtin (gen_ctx, code, &proto_item, &func_import_item)) > 0) {
if (code == MIR_VA_ARG || code == MIR_VA_BLOCK_ARG) {
/* Use a builtin func call:
mov func_reg, func ref; [mov reg3, type;] call proto, func_reg, res_reg, va_reg,
reg3 */
MIR_op_t ops[6], func_reg_op, reg_op3;
MIR_op_t res_reg_op = insn->ops[0], va_reg_op = insn->ops[1], op3 = insn->ops[2];
assert (res_reg_op.mode == MIR_OP_VAR && va_reg_op.mode == MIR_OP_VAR
&& op3.mode == (code == MIR_VA_ARG ? MIR_OP_VAR_MEM : MIR_OP_VAR));
func_reg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
reg_op3 = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
next_insn = new_insn
= MIR_new_insn (ctx, MIR_MOV, func_reg_op, MIR_new_ref_op (ctx, func_import_item));
gen_add_insn_before (gen_ctx, insn, new_insn);
if (code == MIR_VA_ARG) {
new_insn = MIR_new_insn (ctx, MIR_MOV, reg_op3,
MIR_new_int_op (ctx, (int64_t) op3.u.var_mem.type));
op3 = reg_op3;
gen_add_insn_before (gen_ctx, insn, new_insn);
}
ops[0] = MIR_new_ref_op (ctx, proto_item);
ops[1] = func_reg_op;
ops[2] = res_reg_op;
ops[3] = va_reg_op;
ops[4] = op3;
if (code == MIR_VA_BLOCK_ARG) ops[5] = insn->ops[3];
new_insn = MIR_new_insn_arr (ctx, MIR_CALL, code == MIR_VA_ARG ? 5 : 6, ops);
gen_add_insn_before (gen_ctx, insn, new_insn);
gen_delete_insn (gen_ctx, insn);
} else { /* Use builtin: mov freg, func ref; call proto, freg, res_reg, op_reg[, op_reg2] */
MIR_op_t freg_op, res_reg_op = insn->ops[0], op_reg_op = insn->ops[1], ops[5];
assert (res_reg_op.mode == MIR_OP_VAR && op_reg_op.mode == MIR_OP_VAR);
freg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
next_insn = new_insn
= MIR_new_insn (ctx, MIR_MOV, freg_op, MIR_new_ref_op (ctx, func_import_item));
gen_add_insn_before (gen_ctx, insn, new_insn);
ops[0] = MIR_new_ref_op (ctx, proto_item);
ops[1] = freg_op;
ops[2] = res_reg_op;
ops[3] = op_reg_op;
if (nargs == 2) ops[4] = insn->ops[2];
new_insn = MIR_new_insn_arr (ctx, MIR_CALL, nargs + 3, ops);
gen_add_insn_before (gen_ctx, insn, new_insn);
gen_delete_insn (gen_ctx, insn);
}
} else if (code == MIR_VA_START) {
#if !defined(__APPLE__)
MIR_op_t treg_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
#endif
MIR_op_t prev_sp_op = _MIR_new_var_op (ctx, gen_new_temp_reg (gen_ctx, MIR_T_I64, func));
MIR_op_t va_op = insn->ops[0];
MIR_reg_t va_reg;
int gp_offset, fp_offset;
assert (func->vararg_p && va_op.mode == MIR_OP_VAR);
gp_offset = (int_arg_num >= 8 ? 0 : 8 * int_arg_num - 64);
fp_offset = (fp_arg_num >= 8 ? 0 : 16 * fp_arg_num - 128);
va_reg = va_op.u.var;
/* Insns can be not simplified as soon as they match a machine insn. */
#if !defined(__APPLE__)
/* mem32[va_reg].__gr_offset = gp_offset; mem32[va_reg].__vr_offset = fp_offset */
gen_mov (gen_ctx, insn, MIR_MOV, treg_op, MIR_new_int_op (ctx, gp_offset));
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_U32, 24, va_reg, MIR_NON_VAR, 1), treg_op);
gen_mov (gen_ctx, insn, MIR_MOV, treg_op, MIR_new_int_op (ctx, fp_offset));
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_U32, 28, va_reg, MIR_NON_VAR, 1), treg_op);
#endif
/* __stack: prev_sp = mem64[fp + 16] */
gen_mov (gen_ctx, insn, MIR_MOV, prev_sp_op,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 16, FP_HARD_REG, MIR_NON_VAR, 1));
#if defined(__APPLE__)
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 0, va_reg, MIR_NON_VAR, 1), prev_sp_op);
#else
/* mem64[va_reg].__stack = prev_sp + mem_size */
new_insn = MIR_new_insn (ctx, MIR_ADD, treg_op, prev_sp_op, MIR_new_int_op (ctx, mem_size));
gen_add_insn_before (gen_ctx, insn, new_insn);
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 0, va_reg, MIR_NON_VAR, 1), treg_op);
/* __gr_top: mem64[va_reg].__gr_top = prev_sp */
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 8, va_reg, MIR_NON_VAR, 1), prev_sp_op);
/* __vr_top: treg = prev_sp - int_reg_save_area; mem64[va_reg].__vr_top = treg */
new_insn = MIR_new_insn (ctx, MIR_SUB, treg_op, prev_sp_op,
MIR_new_int_op (ctx, int_reg_save_area_size));
gen_add_insn_before (gen_ctx, insn, new_insn);
gen_mov (gen_ctx, insn, MIR_MOV,
_MIR_new_var_mem_op (ctx, MIR_T_I64, 16, va_reg, MIR_NON_VAR, 1), treg_op);
#endif
gen_delete_insn (gen_ctx, insn);
} else if (code == MIR_VA_END) { /* do nothing */
gen_delete_insn (gen_ctx, insn);
} else if (MIR_call_code_p (code)) {
machinize_call (gen_ctx, insn);
leaf_p = FALSE;
} else if (code == MIR_ALLOCA) {
alloca_p = TRUE;
} else if (code == MIR_FBLT) { /* don't use blt/ble for correct nan processing: */
SWAP (insn->ops[1], insn->ops[2], temp_op);
insn->code = MIR_FBGT;
} else if (code == MIR_FBLE) {
SWAP (insn->ops[1], insn->ops[2], temp_op);
insn->code = MIR_FBGE;
} else if (code == MIR_DBLT) {
SWAP (insn->ops[1], insn->ops[2], temp_op);
insn->code = MIR_DBGT;
} else if (code == MIR_DBLE) {
SWAP (insn->ops[1], insn->ops[2], temp_op);
insn->code = MIR_DBGE;
} else if (code == MIR_RET) {
/* In simplify we already transformed code for one return insn
and added extension insn (if any). */
uint32_t n_iregs = 0, n_vregs = 0;
assert (func->nres == MIR_insn_nops (ctx, insn));
for (i = 0; i < func->nres; i++) {
assert (insn->ops[i].mode == MIR_OP_VAR);
res_type = func->res_types[i];
if ((res_type == MIR_T_F || res_type == MIR_T_D || res_type == MIR_T_LD) && n_vregs < 8) {
new_insn_code = res_type == MIR_T_F ? MIR_FMOV
: res_type == MIR_T_D ? MIR_DMOV
: MIR_LDMOV;
ret_reg = V0_HARD_REG + n_vregs++;
} else if (n_iregs < 8) {
new_insn_code = MIR_MOV;
ret_reg = R0_HARD_REG + n_iregs++;
} else {
(*MIR_get_error_func (ctx)) (MIR_ret_error,
"aarch64 can not handle this combination of return values");
}
ret_reg_op = _MIR_new_var_op (ctx, ret_reg);
gen_mov (gen_ctx, insn, new_insn_code, ret_reg_op, insn->ops[i]);
insn->ops[i] = ret_reg_op;
}
}
}
}
#if !defined(__APPLE__)
static void isave (gen_ctx_t gen_ctx, MIR_insn_t anchor, int disp, MIR_reg_t base,