/
IR_X86.cpp
1830 lines (1732 loc) · 49.7 KB
/
IR_X86.cpp
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
// Copyright 2013 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
/*
For a more general explanation of the IR, see IR.cpp.
X86 codegen is a backward pass followed by a forward pass.
The first pass to actually doing codegen is a liveness analysis pass.
Liveness is important for two reasons: one, it lets us do dead code
elimination, which results both from earlier folding, PPC
instructions with unused parts like srawx, and just random strangeness.
The other bit is that is allows us to identify the last instruction to
use a value: this is absolutely essential for register allocation
because it the allocator needs to be able to free unused registers.
In addition, this allows eliminating redundant mov instructions in a lot
of cases.
The register allocation is linear scan allocation.
*/
#ifdef _MSC_VER
#pragma warning(disable:4146) // unary minus operator applied to unsigned type, result still unsigned
#endif
#include "JitIL.h"
#include "../../../Common/CPUDetect.h"
#include "MathUtil.h"
#include "HW/ProcessorInterface.h"
using namespace IREmitter;
using namespace Gen;
static const unsigned int MAX_NUMBER_OF_REGS = 16;
struct RegInfo {
JitIL *Jit;
IRBuilder* Build;
InstLoc FirstI;
std::vector<unsigned> IInfo;
std::vector<InstLoc> lastUsed;
InstLoc regs[MAX_NUMBER_OF_REGS];
InstLoc fregs[MAX_NUMBER_OF_REGS];
unsigned numSpills;
unsigned numFSpills;
unsigned exitNumber;
RegInfo(JitIL* j, InstLoc f, unsigned insts) : Jit(j), FirstI(f), IInfo(insts), lastUsed(insts) {
for (unsigned i = 0; i < MAX_NUMBER_OF_REGS; i++) {
regs[i] = 0;
fregs[i] = 0;
}
numSpills = 0;
numFSpills = 0;
exitNumber = 0;
}
private:
RegInfo(RegInfo&); // DO NOT IMPLEMENT
};
static u32 regsInUse(RegInfo& R) {
#ifdef _M_X64
u32 result = 0;
for (unsigned i = 0; i < MAX_NUMBER_OF_REGS; i++)
{
if (R.regs[i] != 0)
result |= (1 << i);
if (R.fregs[i] != 0)
result |= (1 << (16 + i));
}
return result;
#else
// not needed
return 0;
#endif
}
static void regMarkUse(RegInfo& R, InstLoc I, InstLoc Op, unsigned OpNum) {
unsigned& info = R.IInfo[Op - R.FirstI];
if (info == 0) R.IInfo[I - R.FirstI] |= 1 << (OpNum + 1);
if (info < 2) info++;
R.lastUsed[Op - R.FirstI] = max(R.lastUsed[Op - R.FirstI], I);
}
static unsigned regReadUse(RegInfo& R, InstLoc I) {
return R.IInfo[I - R.FirstI] & 3;
}
static unsigned SlotSet[1000];
static u8 GC_ALIGNED16(FSlotSet[16*1000]);
static OpArg regLocForSlot(RegInfo& RI, unsigned slot) {
return M(&SlotSet[slot - 1]);
}
static unsigned regCreateSpill(RegInfo& RI, InstLoc I) {
unsigned newSpill = ++RI.numSpills;
RI.IInfo[I - RI.FirstI] |= newSpill << 16;
return newSpill;
}
static unsigned regGetSpill(RegInfo& RI, InstLoc I) {
return RI.IInfo[I - RI.FirstI] >> 16;
}
static void regSpill(RegInfo& RI, X64Reg reg) {
if (!RI.regs[reg]) return;
unsigned slot = regGetSpill(RI, RI.regs[reg]);
if (!slot) {
slot = regCreateSpill(RI, RI.regs[reg]);
RI.Jit->MOV(32, regLocForSlot(RI, slot), R(reg));
}
RI.regs[reg] = 0;
}
static OpArg fregLocForSlot(RegInfo& RI, unsigned slot) {
return M(&FSlotSet[slot*16]);
}
static unsigned fregCreateSpill(RegInfo& RI, InstLoc I) {
unsigned newSpill = ++RI.numFSpills;
RI.IInfo[I - RI.FirstI] |= newSpill << 16;
return newSpill;
}
static unsigned fregGetSpill(RegInfo& RI, InstLoc I) {
return RI.IInfo[I - RI.FirstI] >> 16;
}
static void fregSpill(RegInfo& RI, X64Reg reg) {
if (!RI.fregs[reg]) return;
unsigned slot = fregGetSpill(RI, RI.fregs[reg]);
if (!slot) {
slot = fregCreateSpill(RI, RI.fregs[reg]);
RI.Jit->MOVAPD(fregLocForSlot(RI, slot), reg);
}
RI.fregs[reg] = 0;
}
// ECX is scratch, so we don't allocate it
#ifdef _M_X64
// 64-bit - calling conventions differ between linux & windows, so...
#ifdef _WIN32
static const X64Reg RegAllocOrder[] = {RSI, RDI, R12, R13, R14, R8, R9, R10, R11};
#else
static const X64Reg RegAllocOrder[] = {RBP, R12, R13, R14, R8, R9, R10, R11};
#endif
static const int RegAllocSize = sizeof(RegAllocOrder) / sizeof(X64Reg);
static const X64Reg FRegAllocOrder[] = {XMM6, XMM7, XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, XMM2, XMM3, XMM4, XMM5};
static const int FRegAllocSize = sizeof(FRegAllocOrder) / sizeof(X64Reg);
#else
// 32-bit
static const X64Reg RegAllocOrder[] = {EDI, ESI, EBP, EBX, EDX, EAX};
static const int RegAllocSize = sizeof(RegAllocOrder) / sizeof(X64Reg);
static const X64Reg FRegAllocOrder[] = {XMM2, XMM3, XMM4, XMM5, XMM6, XMM7};
static const int FRegAllocSize = sizeof(FRegAllocOrder) / sizeof(X64Reg);
#endif
static X64Reg regFindFreeReg(RegInfo& RI) {
for (auto& reg : RegAllocOrder)
if (RI.regs[reg] == 0)
return reg;
int bestIndex = -1;
InstLoc bestEnd = 0;
for (int i = 0; i < RegAllocSize; ++i) {
const InstLoc start = RI.regs[RegAllocOrder[i]];
const InstLoc end = RI.lastUsed[start - RI.FirstI];
if (bestEnd < end) {
bestEnd = end;
bestIndex = i;
}
}
X64Reg reg = RegAllocOrder[bestIndex];
regSpill(RI, reg);
return reg;
}
static X64Reg fregFindFreeReg(RegInfo& RI) {
for (auto& reg : FRegAllocOrder)
if (RI.fregs[reg] == 0)
return reg;
int bestIndex = -1;
InstLoc bestEnd = 0;
for (int i = 0; i < FRegAllocSize; ++i) {
const InstLoc start = RI.fregs[FRegAllocOrder[i]];
const InstLoc end = RI.lastUsed[start - RI.FirstI];
if (bestEnd < end) {
bestEnd = end;
bestIndex = i;
}
}
X64Reg reg = FRegAllocOrder[bestIndex];
fregSpill(RI, reg);
return reg;
}
static OpArg regLocForInst(RegInfo& RI, InstLoc I) {
for (auto& reg : RegAllocOrder)
if (RI.regs[reg] == I)
return R(reg);
unsigned slot = regGetSpill(RI, I);
if (!slot)
PanicAlert("Retrieving unknown spill slot?!");
return regLocForSlot(RI, slot);
}
static OpArg fregLocForInst(RegInfo& RI, InstLoc I) {
for (auto& reg : FRegAllocOrder)
if (RI.fregs[reg] == I)
return R(reg);
unsigned slot = fregGetSpill(RI, I);
if (!slot)
PanicAlert("Retrieving unknown spill slot?!");
return fregLocForSlot(RI, slot);
}
static void regClearInst(RegInfo& RI, InstLoc I) {
for (auto& reg : RegAllocOrder)
if (RI.regs[reg] == I)
RI.regs[reg] = 0;
}
static void fregClearInst(RegInfo& RI, InstLoc I) {
for (auto& reg : FRegAllocOrder)
if (RI.fregs[reg] == I)
RI.fregs[reg] = 0;
}
static X64Reg regEnsureInReg(RegInfo& RI, InstLoc I) {
OpArg loc = regLocForInst(RI, I);
if (!loc.IsSimpleReg()) {
X64Reg newReg = regFindFreeReg(RI);
RI.Jit->MOV(32, R(newReg), loc);
loc = R(newReg);
}
return loc.GetSimpleReg();
}
static X64Reg fregEnsureInReg(RegInfo& RI, InstLoc I) {
OpArg loc = fregLocForInst(RI, I);
if (!loc.IsSimpleReg()) {
X64Reg newReg = fregFindFreeReg(RI);
RI.Jit->MOVAPD(newReg, loc);
loc = R(newReg);
}
return loc.GetSimpleReg();
}
static void regSpillCallerSaved(RegInfo& RI) {
#ifdef _M_IX86
// 32-bit
regSpill(RI, EDX);
regSpill(RI, ECX);
regSpill(RI, EAX);
#else
// 64-bit
regSpill(RI, RCX);
regSpill(RI, RDX);
regSpill(RI, RSI);
regSpill(RI, RDI);
regSpill(RI, R8);
regSpill(RI, R9);
regSpill(RI, R10);
regSpill(RI, R11);
#endif
}
static X64Reg regUReg(RegInfo& RI, InstLoc I) {
const OpArg loc = regLocForInst(RI, getOp1(I));
if ((RI.IInfo[I - RI.FirstI] & 4) && loc.IsSimpleReg()) {
return loc.GetSimpleReg();
}
return regFindFreeReg(RI);
}
// Recycle the register if the lifetime of op1 register ends at I.
static X64Reg fregURegWithoutMov(RegInfo& RI, InstLoc I) {
const OpArg loc = fregLocForInst(RI, getOp1(I));
if ((RI.IInfo[I - RI.FirstI] & 4) && loc.IsSimpleReg()) {
return loc.GetSimpleReg();
}
return fregFindFreeReg(RI);
}
static X64Reg fregURegWithMov(RegInfo& RI, InstLoc I) {
const OpArg loc = fregLocForInst(RI, getOp1(I));
if ((RI.IInfo[I - RI.FirstI] & 4) && loc.IsSimpleReg()) {
return loc.GetSimpleReg();
}
X64Reg reg = fregFindFreeReg(RI);
RI.Jit->MOVAPD(reg, loc);
return reg;
}
// Recycle the register if the lifetime of op1 register ends at I.
static X64Reg fregBinLHSRegWithMov(RegInfo& RI, InstLoc I) {
const OpArg loc = fregLocForInst(RI, getOp1(I));
if ((RI.IInfo[I - RI.FirstI] & 4) && loc.IsSimpleReg()) {
return loc.GetSimpleReg();
}
X64Reg reg = fregFindFreeReg(RI);
RI.Jit->MOVAPD(reg, loc);
return reg;
}
// Recycle the register if the lifetime of op2 register ends at I.
static X64Reg fregBinRHSRegWithMov(RegInfo& RI, InstLoc I) {
const OpArg loc = fregLocForInst(RI, getOp2(I));
if ((RI.IInfo[I - RI.FirstI] & 8) && loc.IsSimpleReg()) {
return loc.GetSimpleReg();
}
X64Reg reg = fregFindFreeReg(RI);
RI.Jit->MOVAPD(reg, loc);
return reg;
}
// If the lifetime of the register used by an operand ends at I,
// return the register. Otherwise return a free register.
static X64Reg regBinReg(RegInfo& RI, InstLoc I) {
// FIXME: When regLocForInst() is extracted as a local variable,
// "Retrieving unknown spill slot?!" is shown.
if ((RI.IInfo[I - RI.FirstI] & 4) &&
regLocForInst(RI, getOp1(I)).IsSimpleReg()) {
return regLocForInst(RI, getOp1(I)).GetSimpleReg();
} else if ((RI.IInfo[I - RI.FirstI] & 8) &&
regLocForInst(RI, getOp2(I)).IsSimpleReg()) {
return regLocForInst(RI, getOp2(I)).GetSimpleReg();
}
return regFindFreeReg(RI);
}
static X64Reg regBinLHSReg(RegInfo& RI, InstLoc I) {
if (RI.IInfo[I - RI.FirstI] & 4) {
return regEnsureInReg(RI, getOp1(I));
}
X64Reg reg = regFindFreeReg(RI);
RI.Jit->MOV(32, R(reg), regLocForInst(RI, getOp1(I)));
return reg;
}
static void regNormalRegClear(RegInfo& RI, InstLoc I) {
if (RI.IInfo[I - RI.FirstI] & 4)
regClearInst(RI, getOp1(I));
if (RI.IInfo[I - RI.FirstI] & 8)
regClearInst(RI, getOp2(I));
}
static void fregNormalRegClear(RegInfo& RI, InstLoc I) {
if (RI.IInfo[I - RI.FirstI] & 4)
fregClearInst(RI, getOp1(I));
if (RI.IInfo[I - RI.FirstI] & 8)
fregClearInst(RI, getOp2(I));
}
static void regEmitBinInst(RegInfo& RI, InstLoc I,
void (JitIL::*op)(int, const OpArg&,
const OpArg&),
bool commutable = false) {
X64Reg reg;
bool commuted = false;
if (RI.IInfo[I - RI.FirstI] & 4) {
reg = regEnsureInReg(RI, getOp1(I));
} else if (commutable && (RI.IInfo[I - RI.FirstI] & 8)) {
reg = regEnsureInReg(RI, getOp2(I));
commuted = true;
} else {
reg = regFindFreeReg(RI);
RI.Jit->MOV(32, R(reg), regLocForInst(RI, getOp1(I)));
}
if (isImm(*getOp2(I))) {
unsigned RHS = RI.Build->GetImmValue(getOp2(I));
if (RHS + 128 < 256) {
(RI.Jit->*op)(32, R(reg), Imm8(RHS));
} else {
(RI.Jit->*op)(32, R(reg), Imm32(RHS));
}
} else if (commuted) {
(RI.Jit->*op)(32, R(reg), regLocForInst(RI, getOp1(I)));
} else {
(RI.Jit->*op)(32, R(reg), regLocForInst(RI, getOp2(I)));
}
RI.regs[reg] = I;
regNormalRegClear(RI, I);
}
static void fregEmitBinInst(RegInfo& RI, InstLoc I,
void (JitIL::*op)(X64Reg, OpArg)) {
X64Reg reg;
if (RI.IInfo[I - RI.FirstI] & 4) {
reg = fregEnsureInReg(RI, getOp1(I));
} else {
reg = fregFindFreeReg(RI);
RI.Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
}
(RI.Jit->*op)(reg, fregLocForInst(RI, getOp2(I)));
RI.fregs[reg] = I;
fregNormalRegClear(RI, I);
}
// Mark and calculation routines for profiled load/store addresses
// Could be extended to unprofiled addresses.
static void regMarkMemAddress(RegInfo& RI, InstLoc I, InstLoc AI, unsigned OpNum) {
if (isImm(*AI)) {
unsigned addr = RI.Build->GetImmValue(AI);
if (Memory::IsRAMAddress(addr))
return;
}
if (getOpcode(*AI) == Add && isImm(*getOp2(AI))) {
regMarkUse(RI, I, getOp1(AI), OpNum);
return;
}
regMarkUse(RI, I, AI, OpNum);
}
// in 64-bit build, this returns a completely bizarre address sometimes!
static std::pair<OpArg, u32> regBuildMemAddress(RegInfo& RI, InstLoc I,
InstLoc AI, unsigned OpNum, unsigned Size, X64Reg* dest) {
if (isImm(*AI)) {
unsigned addr = RI.Build->GetImmValue(AI);
if (Memory::IsRAMAddress(addr)) {
if (dest)
*dest = regFindFreeReg(RI);
return std::make_pair(Imm32(addr), 0);
}
}
unsigned offset;
InstLoc AddrBase;
if (getOpcode(*AI) == Add && isImm(*getOp2(AI))) {
offset = RI.Build->GetImmValue(getOp2(AI));
AddrBase = getOp1(AI);
} else {
offset = 0;
AddrBase = AI;
}
X64Reg baseReg;
// Ok, this stuff needs a comment or three :P -ector
if (RI.IInfo[I - RI.FirstI] & (2 << OpNum)) {
baseReg = regEnsureInReg(RI, AddrBase);
regClearInst(RI, AddrBase);
if (dest)
*dest = baseReg;
} else if (dest) {
X64Reg reg = regFindFreeReg(RI);
const OpArg loc = regLocForInst(RI, AddrBase);
if (!loc.IsSimpleReg()) {
RI.Jit->MOV(32, R(reg), loc);
baseReg = reg;
} else {
baseReg = loc.GetSimpleReg();
}
*dest = reg;
} else {
baseReg = regEnsureInReg(RI, AddrBase);
}
return std::make_pair(R(baseReg), offset);
}
static void regEmitMemLoad(RegInfo& RI, InstLoc I, unsigned Size) {
X64Reg reg;
auto info = regBuildMemAddress(RI, I, getOp1(I), 1, Size, ®);
RI.Jit->SafeLoadToReg(reg, info.first, Size, info.second, regsInUse(RI), false, EmuCodeBlock::SAFE_LOADSTORE_NO_FASTMEM);
if (regReadUse(RI, I))
RI.regs[reg] = I;
}
static OpArg regImmForConst(RegInfo& RI, InstLoc I, unsigned Size) {
unsigned imm = RI.Build->GetImmValue(I);
if (Size == 32) {
return Imm32(imm);
} else if (Size == 16) {
return Imm16(imm);
} else {
return Imm8(imm);
}
}
static void regEmitMemStore(RegInfo& RI, InstLoc I, unsigned Size) {
auto info = regBuildMemAddress(RI, I, getOp2(I), 2, Size, 0);
if (info.first.IsImm())
RI.Jit->MOV(32, R(ECX), info.first);
else
RI.Jit->LEA(32, ECX, MDisp(info.first.GetSimpleReg(), info.second));
regSpill(RI, EAX);
if (isImm(*getOp1(I))) {
RI.Jit->MOV(Size, R(EAX), regImmForConst(RI, getOp1(I), Size));
} else {
RI.Jit->MOV(32, R(EAX), regLocForInst(RI, getOp1(I)));
}
RI.Jit->SafeWriteRegToReg(EAX, ECX, Size, 0, regsInUse(RI), EmuCodeBlock::SAFE_LOADSTORE_NO_FASTMEM);
if (RI.IInfo[I - RI.FirstI] & 4)
regClearInst(RI, getOp1(I));
}
static void regEmitShiftInst(RegInfo& RI, InstLoc I, void (JitIL::*op)(int, OpArg, OpArg))
{
X64Reg reg = regBinLHSReg(RI, I);
if (isImm(*getOp2(I))) {
unsigned RHS = RI.Build->GetImmValue(getOp2(I));
(RI.Jit->*op)(32, R(reg), Imm8(RHS));
RI.regs[reg] = I;
return;
}
RI.Jit->MOV(32, R(ECX), regLocForInst(RI, getOp2(I)));
(RI.Jit->*op)(32, R(reg), R(ECX));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
}
static void regStoreInstToConstLoc(RegInfo& RI, unsigned width, InstLoc I, void* loc) {
if (width != 32) {
PanicAlert("Not implemented!");
return;
}
if (isImm(*I)) {
RI.Jit->MOV(32, M(loc), Imm32(RI.Build->GetImmValue(I)));
return;
}
X64Reg reg = regEnsureInReg(RI, I);
RI.Jit->MOV(32, M(loc), R(reg));
}
static void regEmitCmp(RegInfo& RI, InstLoc I) {
if (isImm(*getOp2(I))) {
unsigned RHS = RI.Build->GetImmValue(getOp2(I));
RI.Jit->CMP(32, regLocForInst(RI, getOp1(I)), Imm32(RHS));
} else {
X64Reg reg = regEnsureInReg(RI, getOp1(I));
RI.Jit->CMP(32, R(reg), regLocForInst(RI, getOp2(I)));
}
}
static void regEmitICmpInst(RegInfo& RI, InstLoc I, CCFlags flag) {
regEmitCmp(RI, I);
RI.Jit->SETcc(flag, R(ECX)); // Caution: SETCC uses 8-bit regs!
X64Reg reg = regBinReg(RI, I);
RI.Jit->MOVZX(32, 8, reg, R(ECX));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
}
static void regWriteExit(RegInfo& RI, InstLoc dest) {
if (isImm(*dest)) {
RI.Jit->WriteExit(RI.Build->GetImmValue(dest), RI.exitNumber++);
} else {
RI.Jit->WriteExitDestInOpArg(regLocForInst(RI, dest));
}
}
// Helper function to check floating point exceptions
static double GC_ALIGNED16(isSNANTemp[2][2]);
static bool checkIsSNAN() {
return MathUtil::IsSNAN(isSNANTemp[0][0]) || MathUtil::IsSNAN(isSNANTemp[1][0]);
}
static void DoWriteCode(IRBuilder* ibuild, JitIL* Jit) {
//printf("Writing block: %x\n", js.blockStart);
RegInfo RI(Jit, ibuild->getFirstInst(), ibuild->getNumInsts());
RI.Build = ibuild;
// Pass to compute liveness
ibuild->StartBackPass();
for (unsigned int index = (unsigned int)RI.IInfo.size() - 1; index != -1U; --index) {
InstLoc I = ibuild->ReadBackward();
unsigned int op = getOpcode(*I);
bool thisUsed = regReadUse(RI, I) ? true : false;
switch (op) {
default:
PanicAlert("Unexpected inst!");
case Nop:
case CInt16:
case CInt32:
case LoadGReg:
case LoadLink:
case LoadCR:
case LoadCarry:
case LoadCTR:
case LoadMSR:
case LoadFReg:
case LoadFRegDENToZero:
case LoadGQR:
case BlockEnd:
case BlockStart:
case InterpreterFallback:
case SystemCall:
case RFIExit:
case InterpreterBranch:
case ShortIdleLoop:
case FPExceptionCheck:
case DSIExceptionCheck:
case ISIException:
case ExtExceptionCheck:
case BreakPointCheck:
case Int3:
case Tramp:
// No liveness effects
break;
case SExt8:
case SExt16:
case BSwap32:
case BSwap16:
case Cntlzw:
case Not:
case DupSingleToMReg:
case DoubleToSingle:
case ExpandPackedToMReg:
case CompactMRegToPacked:
case FPNeg:
case FPDup0:
case FPDup1:
case FSNeg:
case FDNeg:
if (thisUsed)
regMarkUse(RI, I, getOp1(I), 1);
break;
case Load8:
case Load16:
case Load32:
regMarkMemAddress(RI, I, getOp1(I), 1);
break;
case LoadDouble:
case LoadSingle:
case LoadPaired:
if (thisUsed)
regMarkUse(RI, I, getOp1(I), 1);
break;
case StoreCR:
case StoreCarry:
case StoreFPRF:
regMarkUse(RI, I, getOp1(I), 1);
break;
case StoreGReg:
case StoreLink:
case StoreCTR:
case StoreMSR:
case StoreGQR:
case StoreSRR:
case StoreFReg:
if (!isImm(*getOp1(I)))
regMarkUse(RI, I, getOp1(I), 1);
break;
case Add:
case Sub:
case And:
case Or:
case Xor:
case Mul:
case MulHighUnsigned:
case Rol:
case Shl:
case Shrl:
case Sarl:
case ICmpCRUnsigned:
case ICmpCRSigned:
case ICmpEq:
case ICmpNe:
case ICmpUgt:
case ICmpUlt:
case ICmpUge:
case ICmpUle:
case ICmpSgt:
case ICmpSlt:
case ICmpSge:
case ICmpSle:
case FSMul:
case FSAdd:
case FSSub:
case FSRSqrt:
case FDMul:
case FDAdd:
case FDSub:
case FPAdd:
case FPMul:
case FPSub:
case FPMerge00:
case FPMerge01:
case FPMerge10:
case FPMerge11:
case FDCmpCR:
case InsertDoubleInMReg:
if (thisUsed) {
regMarkUse(RI, I, getOp1(I), 1);
if (!isImm(*getOp2(I)))
regMarkUse(RI, I, getOp2(I), 2);
}
break;
case Store8:
case Store16:
case Store32:
if (!isImm(*getOp1(I)))
regMarkUse(RI, I, getOp1(I), 1);
regMarkMemAddress(RI, I, getOp2(I), 2);
break;
case StoreSingle:
case StoreDouble:
case StorePaired:
regMarkUse(RI, I, getOp1(I), 1);
regMarkUse(RI, I, getOp2(I), 2);
break;
case BranchUncond:
if (!isImm(*getOp1(I)))
regMarkUse(RI, I, getOp1(I), 1);
break;
case IdleBranch:
regMarkUse(RI, I, getOp1(getOp1(I)), 1);
break;
case BranchCond: {
if (isICmp(*getOp1(I)) &&
isImm(*getOp2(getOp1(I)))) {
regMarkUse(RI, I, getOp1(getOp1(I)), 1);
} else {
regMarkUse(RI, I, getOp1(I), 1);
}
if (!isImm(*getOp2(I)))
regMarkUse(RI, I, getOp2(I), 2);
break;
}
}
}
ibuild->StartForwardPass();
for (unsigned i = 0; i != RI.IInfo.size(); i++) {
InstLoc I = ibuild->ReadForward();
bool thisUsed = regReadUse(RI, I) ? true : false;
if (thisUsed) {
// Needed for IR Writer
ibuild->SetMarkUsed(I);
}
switch (getOpcode(*I)) {
case InterpreterFallback: {
unsigned InstCode = ibuild->GetImmValue(getOp1(I));
unsigned InstLoc = ibuild->GetImmValue(getOp2(I));
// There really shouldn't be anything live across an
// interpreter call at the moment, but optimizing interpreter
// calls isn't completely out of the question...
regSpillCallerSaved(RI);
Jit->MOV(32, M(&PC), Imm32(InstLoc));
Jit->MOV(32, M(&NPC), Imm32(InstLoc+4));
Jit->ABI_CallFunctionC((void*)GetInterpreterOp(InstCode),
InstCode);
break;
}
case LoadGReg: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
unsigned ppcreg = *I >> 8;
Jit->MOV(32, R(reg), M(&PowerPC::ppcState.gpr[ppcreg]));
RI.regs[reg] = I;
break;
}
case LoadCR: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
unsigned ppcreg = *I >> 8;
Jit->MOVZX(32, 8, reg, M(&PowerPC::ppcState.cr_fast[ppcreg]));
RI.regs[reg] = I;
break;
}
case LoadCTR: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
Jit->MOV(32, R(reg), M(&CTR));
RI.regs[reg] = I;
break;
}
case LoadLink: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
Jit->MOV(32, R(reg), M(&LR));
RI.regs[reg] = I;
break;
}
case LoadMSR: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
Jit->MOV(32, R(reg), M(&MSR));
RI.regs[reg] = I;
break;
}
case LoadGQR: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
unsigned gqr = *I >> 8;
Jit->MOV(32, R(reg), M(&GQR(gqr)));
RI.regs[reg] = I;
break;
}
case LoadCarry: {
if (!thisUsed) break;
X64Reg reg = regFindFreeReg(RI);
Jit->MOV(32, R(reg), M(&PowerPC::ppcState.spr[SPR_XER]));
Jit->SHR(32, R(reg), Imm8(29));
Jit->AND(32, R(reg), Imm8(1));
RI.regs[reg] = I;
break;
}
case StoreGReg: {
unsigned ppcreg = *I >> 16;
regStoreInstToConstLoc(RI, 32, getOp1(I),
&PowerPC::ppcState.gpr[ppcreg]);
regNormalRegClear(RI, I);
break;
}
case StoreCR: {
Jit->MOV(32, R(ECX), regLocForInst(RI, getOp1(I)));
unsigned ppcreg = *I >> 16;
// CAUTION: uses 8-bit reg!
Jit->MOV(8, M(&PowerPC::ppcState.cr_fast[ppcreg]), R(ECX));
regNormalRegClear(RI, I);
break;
}
case StoreLink: {
regStoreInstToConstLoc(RI, 32, getOp1(I), &LR);
regNormalRegClear(RI, I);
break;
}
case StoreCTR: {
regStoreInstToConstLoc(RI, 32, getOp1(I), &CTR);
regNormalRegClear(RI, I);
break;
}
case StoreMSR: {
unsigned InstLoc = ibuild->GetImmValue(getOp2(I));
regStoreInstToConstLoc(RI, 32, getOp1(I), &MSR);
regNormalRegClear(RI, I);
// If some exceptions are pending and EE are now enabled, force checking
// external exceptions when going out of mtmsr in order to execute delayed
// interrupts as soon as possible.
Jit->MOV(32, R(EAX), M(&MSR));
Jit->TEST(32, R(EAX), Imm32(0x8000));
FixupBranch eeDisabled = Jit->J_CC(CC_Z);
Jit->MOV(32, R(EAX), M((void*)&PowerPC::ppcState.Exceptions));
Jit->TEST(32, R(EAX), R(EAX));
FixupBranch noExceptionsPending = Jit->J_CC(CC_Z);
Jit->MOV(32, M(&PC), Imm32(InstLoc + 4));
Jit->WriteExceptionExit(); // TODO: Implement WriteExternalExceptionExit for JitIL
Jit->SetJumpTarget(eeDisabled);
Jit->SetJumpTarget(noExceptionsPending);
break;
}
case StoreGQR: {
unsigned gqr = *I >> 16;
regStoreInstToConstLoc(RI, 32, getOp1(I), &GQR(gqr));
regNormalRegClear(RI, I);
break;
}
case StoreSRR: {
unsigned srr = *I >> 16;
regStoreInstToConstLoc(RI, 32, getOp1(I),
&PowerPC::ppcState.spr[SPR_SRR0+srr]);
regNormalRegClear(RI, I);
break;
}
case StoreCarry: {
Jit->CMP(32, regLocForInst(RI, getOp1(I)), Imm8(0));
FixupBranch nocarry = Jit->J_CC(CC_Z);
Jit->JitSetCA();
FixupBranch cont = Jit->J();
Jit->SetJumpTarget(nocarry);
Jit->JitClearCA();
Jit->SetJumpTarget(cont);
regNormalRegClear(RI, I);
break;
}
case StoreFPRF: {
Jit->MOV(32, R(ECX), regLocForInst(RI, getOp1(I)));
Jit->AND(32, R(ECX), Imm8(0x1F));
Jit->SHL(32, R(ECX), Imm8(12));
Jit->AND(32, M(&FPSCR), Imm32(~(0x1F << 12)));
Jit->OR(32, M(&FPSCR), R(ECX));
regNormalRegClear(RI, I);
break;
}
case Load8: {
regEmitMemLoad(RI, I, 8);
break;
}
case Load16: {
regEmitMemLoad(RI, I, 16);
break;
}
case Load32: {
regEmitMemLoad(RI, I, 32);
break;
}
case Store8: {
regEmitMemStore(RI, I, 8);
break;
}
case Store16: {
regEmitMemStore(RI, I, 16);
break;
}
case Store32: {
regEmitMemStore(RI, I, 32);
break;
}
case SExt8: {
if (!thisUsed) break;
X64Reg reg = regUReg(RI, I);
Jit->MOV(32, R(ECX), regLocForInst(RI, getOp1(I)));
Jit->MOVSX(32, 8, reg, R(ECX));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
break;
}
case SExt16: {
if (!thisUsed) break;
X64Reg reg = regUReg(RI, I);
Jit->MOVSX(32, 16, reg, regLocForInst(RI, getOp1(I)));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
break;
}
case Cntlzw: {
if (!thisUsed) break;
X64Reg reg = regUReg(RI, I);
Jit->MOV(32, R(ECX), Imm32(63));
Jit->BSR(32, reg, regLocForInst(RI, getOp1(I)));
Jit->CMOVcc(32, reg, R(ECX), CC_Z);
Jit->XOR(32, R(reg), Imm8(31));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
break;
}
case Not: {
if (!thisUsed) break;
X64Reg reg = regBinLHSReg(RI, I);
Jit->NOT(32, R(reg));
RI.regs[reg] = I;
regNormalRegClear(RI, I);
break;
}
case And: {
if (!thisUsed) break;
regEmitBinInst(RI, I, &JitIL::AND, true);
break;
}
case Xor: {
if (!thisUsed) break;
regEmitBinInst(RI, I, &JitIL::XOR, true);
break;
}
case Sub: {
if (!thisUsed) break;
regEmitBinInst(RI, I, &JitIL::SUB);
break;
}
case Or: {
if (!thisUsed) break;
regEmitBinInst(RI, I, &JitIL::OR, true);
break;
}
case Add: {
if (!thisUsed) break;
regEmitBinInst(RI, I, &JitIL::ADD, true);
break;
}
case Mul: {
if (!thisUsed) break;
// FIXME: Use three-address capability of IMUL!
X64Reg reg = regBinLHSReg(RI, I);
if (isImm(*getOp2(I))) {
unsigned RHS = RI.Build->GetImmValue(getOp2(I));
if (RHS + 128 < 256) {
Jit->IMUL(32, reg, Imm8(RHS));
} else {
Jit->IMUL(32, reg, Imm32(RHS));
}
} else {
Jit->IMUL(32, reg, regLocForInst(RI, getOp2(I)));
}
RI.regs[reg] = I;
regNormalRegClear(RI, I);
break;
}
case MulHighUnsigned: {
if (!thisUsed) break;
regSpill(RI, EAX);
regSpill(RI, EDX);
X64Reg reg = regBinReg(RI, I);
if (isImm(*getOp2(I))) {
unsigned RHS = RI.Build->GetImmValue(getOp2(I));