/
cod2.c
5275 lines (4844 loc) · 170 KB
/
cod2.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
// Copyright (C) 1984-1998 by Symantec
// Copyright (C) 2000-2015 by Digital Mars
// All Rights Reserved
// http://www.digitalmars.com
// Written by Walter Bright
/*
* This source file is made available for personal use
* only. The license is in backendlicense.txt
* For any other uses, please contact Digital Mars.
*/
#if !SPP
#include <stdio.h>
#include <string.h>
#include <time.h>
#include "cc.h"
#include "oper.h"
#include "el.h"
#include "code.h"
#include "global.h"
#include "type.h"
#if SCPP
#include "exh.h"
#endif
static char __file__[] = __FILE__; /* for tassert.h */
#include "tassert.h"
int cdcmp_flag;
extern signed char regtorm[8];
// from divcoeff.c
extern bool choose_multiplier(int N, targ_ullong d, int prec, targ_ullong *pm, int *pshpost);
extern bool udiv_coefficients(int N, targ_ullong d, int *pshpre, targ_ullong *pm, int *pshpost);
/*******************************************
* !=0 if cannot use this EA in anything other than a MOV instruction.
*/
int movOnly(elem *e)
{
#if TARGET_OSX
if (I64 && config.flags3 & CFG3pic && e->Eoper == OPvar)
{ symbol *s = e->EV.sp.Vsym;
// Fixups for these can only be done with a MOV
if (s->Sclass == SCglobal || s->Sclass == SCextern ||
s->Sclass == SCcomdat || s->Sclass == SCcomdef)
return 1;
}
#endif
return 0;
}
/********************************
* Return mask of index registers used by addressing mode.
* Index is rm of modregrm field.
*/
regm_t idxregm(code *c)
{
static const unsigned char idxsib[8] = { mAX,mCX,mDX,mBX,0,mBP,mSI,mDI };
static const unsigned char idxrm[8] = {mBX|mSI,mBX|mDI,mSI,mDI,mSI,mDI,0,mBX};
unsigned rm = c->Irm;
regm_t idxm = 0;
if ((rm & 0xC0) != 0xC0) /* if register is not the destination */
{
if (I16)
idxm = idxrm[rm & 7];
else
{
if ((rm & 7) == 4) /* if sib byte */
{
unsigned sib = c->Isib;
unsigned idxreg = (sib >> 3) & 7;
if (c->Irex & REX_X)
{ idxreg |= 8;
idxm = mask[idxreg]; // scaled index reg
}
else
idxm = idxsib[idxreg]; // scaled index reg
if ((sib & 7) == 5 && (rm & 0xC0) == 0)
;
else
{ unsigned base = sib & 7;
if (c->Irex & REX_B)
idxm |= mask[base | 8];
else
idxm |= idxsib[base];
}
}
else
{ unsigned base = rm & 7;
if (c->Irex & REX_B)
idxm |= mask[base | 8];
else
idxm |= idxsib[base];
}
}
}
return idxm;
}
#if TARGET_WINDOS
/***************************
* Gen code for call to floating point routine.
*/
code *opdouble(elem *e,regm_t *pretregs,unsigned clib)
{
regm_t retregs1,retregs2;
code *cl, *cr, *c;
if (config.inline8087)
return orth87(e,pretregs);
if (tybasic(e->E1->Ety) == TYfloat)
{
clib += CLIBfadd - CLIBdadd; /* convert to float operation */
retregs1 = FLOATREGS;
retregs2 = FLOATREGS2;
}
else
{
if (I32)
{ retregs1 = DOUBLEREGS_32;
retregs2 = DOUBLEREGS2_32;
}
else
{ retregs1 = mSTACK;
retregs2 = DOUBLEREGS_16;
}
}
cl = codelem(e->E1, &retregs1,FALSE);
if (retregs1 & mSTACK)
cgstate.stackclean++;
cr = scodelem(e->E2, &retregs2, retregs1 & ~mSTACK, FALSE);
if (retregs1 & mSTACK)
cgstate.stackclean--;
c = callclib(e, clib, pretregs, 0);
return cat3(cl, cr, c);
}
#endif
/*****************************
* Handle operators which are more or less orthogonal
* ( + - & | ^ )
*/
code *cdorth(elem *e,regm_t *pretregs)
{ tym_t ty1;
regm_t retregs,rretregs,posregs;
unsigned reg,rreg,op1,op2,mode;
int rval;
code *c,*cg,*cl;
targ_size_t i;
elem *e1,*e2;
int numwords; /* # of words to be operated on */
static int nest;
//printf("cdorth(e = %p, *pretregs = %s)\n",e,regm_str(*pretregs));
e1 = e->E1;
e2 = e->E2;
if (*pretregs == 0) /* if don't want result */
{ c = codelem(e1,pretregs,FALSE); /* eval left leaf */
*pretregs = 0; /* in case they got set */
return cat(c,codelem(e2,pretregs,FALSE));
}
ty1 = tybasic(e1->Ety);
if (tyfloating(ty1))
{
if (*pretregs & XMMREGS || tyvector(ty1))
return orthxmm(e,pretregs);
#if TARGET_LINUX || TARGET_OSX || TARGET_FREEBSD || TARGET_OPENBSD || TARGET_SOLARIS
return orth87(e,pretregs);
#else
return opdouble(e,pretregs,(e->Eoper == OPadd) ? CLIBdadd
: CLIBdsub);
#endif
}
if (tyxmmreg(ty1))
return orthxmm(e,pretregs);
tym_t ty2 = tybasic(e2->Ety);
int e2oper = e2->Eoper;
tym_t ty = tybasic(e->Ety);
unsigned sz = tysize[ty];
unsigned byte = (sz == 1);
unsigned char word = (!I16 && sz == SHORTSIZE) ? CFopsize : 0;
unsigned test = FALSE; // assume we destroyed lvalue
code cs;
cs.Iflags = 0;
cs.Irex = 0;
code *cr = CNIL;
switch (e->Eoper)
{ case OPadd: mode = 0;
op1 = 0x03; op2 = 0x13; break; /* ADD, ADC */
case OPmin: mode = 5;
op1 = 0x2B; op2 = 0x1B; break; /* SUB, SBB */
case OPor: mode = 1;
op1 = 0x0B; op2 = 0x0B; break; /* OR , OR */
case OPxor: mode = 6;
op1 = 0x33; op2 = 0x33; break; /* XOR, XOR */
case OPand: mode = 4;
op1 = 0x23; op2 = 0x23; /* AND, AND */
if (tyreg(ty1) &&
*pretregs == mPSW) /* if flags only */
{ test = TRUE;
op1 = 0x85; /* TEST */
mode = 0;
}
break;
default:
assert(0);
}
op1 ^= byte; /* if byte operation */
/* Compute number of words to operate on. */
numwords = 1;
if (!I16)
{ /* Cannot operate on longs and then do a 'paint' to a far */
/* pointer, because far pointers are 48 bits and longs are 32. */
/* Therefore, numwords can never be 2. */
assert(!(tyfv(ty1) && tyfv(ty2)));
if (sz == 2 * REGSIZE)
{
numwords++;
}
}
else
{ /* If ty is a TYfptr, but both operands are long, treat the */
/* operation as a long. */
#if TARGET_SEGMENTED
if ((tylong(ty1) || ty1 == TYhptr) &&
(tylong(ty2) || ty2 == TYhptr))
numwords++;
#else
if (tylong(ty1) && tylong(ty2))
numwords++;
#endif
}
// Special cases where only flags are set
if (test && tysize[ty1] <= REGSIZE &&
(e1->Eoper == OPvar || (e1->Eoper == OPind && !e1->Ecount))
#if TARGET_OSX
&& !movOnly(e1)
#endif
)
{
// Handle the case of (var & const)
if (e2->Eoper == OPconst && el_signx32(e2))
{
c = getlvalue(&cs,e1,0);
targ_size_t value = e2->EV.Vpointer;
if (sz == 2)
value &= 0xFFFF;
else if (sz == 4)
value &= 0xFFFFFFFF;
if (reghasvalue(byte ? BYTEREGS : ALLREGS,value,®))
goto L11;
if (sz == 8 && !I64)
{
assert(value == (int)value); // sign extend imm32
}
op1 = 0xF7;
cs.IEV2.Vint = value;
cs.IFL2 = FLconst;
goto L10;
}
// Handle (exp & reg)
if (isregvar(e2,&retregs,®))
{
c = getlvalue(&cs,e1,0);
L11:
code_newreg(&cs, reg);
if (I64 && byte && reg >= 4)
cs.Irex |= REX;
L10:
cs.Iop = op1 ^ byte;
cs.Iflags |= word | CFpsw;
freenode(e1);
freenode(e2);
return gen(c,&cs);
}
}
// Look for possible uses of LEA
if (e->Eoper == OPadd &&
!(*pretregs & mPSW) && /* flags aren't set by LEA */
!nest && // could cause infinite recursion if e->Ecount
(sz == REGSIZE || (I64 && sz == 4))) // far pointers aren't handled
{
unsigned rex = (sz == 8) ? REX_W : 0;
// Handle the case of (e + &var)
int e1oper = e1->Eoper;
if ((e2oper == OPrelconst && (config.target_cpu >= TARGET_Pentium || (!e2->Ecount && stackfl[el_fl(e2)])))
|| // LEA costs too much for simple EAs on older CPUs
(e2oper == OPconst && (e1->Eoper == OPcall || e1->Eoper == OPcallns) && !(*pretregs & mAX)) ||
(!I16 && (isscaledindex(e1) || isscaledindex(e2))) ||
(!I16 && e1oper == OPvar && e1->EV.sp.Vsym->Sfl == FLreg && (e2oper == OPconst || (e2oper == OPvar && e2->EV.sp.Vsym->Sfl == FLreg))) ||
(e2oper == OPconst && e1oper == OPeq && e1->E1->Eoper == OPvar) ||
(!I16 && (e2oper == OPrelconst || e2oper == OPconst) && !e1->Ecount &&
(e1oper == OPmul || e1oper == OPshl) &&
e1->E2->Eoper == OPconst &&
ssindex(e1oper,e1->E2->EV.Vuns)
) ||
(!I16 && e1->Ecount)
)
{
int inc = e->Ecount != 0;
nest += inc;
c = getlvalue(&cs,e,0);
nest -= inc;
unsigned reg;
c = cat(c,allocreg(pretregs,®,ty));
cs.Iop = 0x8D;
code_newreg(&cs, reg);
c = gen(c,&cs); // LEA reg,EA
if (rex)
code_orrex(c, rex);
return c;
}
// Handle the case of ((e + c) + e2)
if (!I16 &&
e1oper == OPadd &&
(e1->E2->Eoper == OPconst && el_signx32(e1->E2) ||
e2oper == OPconst && el_signx32(e2)) &&
!e1->Ecount
)
{ elem *e11;
elem *ebase;
elem *edisp;
int ss;
int ss2;
unsigned reg1,reg2;
code *c1,*c2,*c3;
if (e2oper == OPconst && el_signx32(e2))
{ edisp = e2;
ebase = e1->E2;
}
else
{ edisp = e1->E2;
ebase = e2;
}
e11 = e1->E1;
retregs = *pretregs & ALLREGS;
if (!retregs)
retregs = ALLREGS;
ss = 0;
ss2 = 0;
// Handle the case of (((e * c1) + c2) + e2)
// Handle the case of (((e << c1) + c2) + e2)
if ((e11->Eoper == OPmul || e11->Eoper == OPshl) &&
e11->E2->Eoper == OPconst &&
!e11->Ecount
)
{
targ_size_t co1 = el_tolong(e11->E2);
if (e11->Eoper == OPshl)
{
if (co1 > 3)
goto L13;
ss = co1;
}
else
{
ss2 = 1;
switch (co1)
{
case 6: ss = 1; break;
case 12: ss = 1; ss2 = 2; break;
case 24: ss = 1; ss2 = 3; break;
case 10: ss = 2; break;
case 20: ss = 2; ss2 = 2; break;
case 40: ss = 2; ss2 = 3; break;
case 18: ss = 3; break;
case 36: ss = 3; ss2 = 2; break;
case 72: ss = 3; ss2 = 3; break;
default:
ss2 = 0;
goto L13;
}
}
freenode(e11->E2);
freenode(e11);
e11 = e11->E1;
goto L13;
}
else
{
L13:
regm_t regm;
if (e11->Eoper == OPvar && isregvar(e11,®m,®1))
{
if (tysize[tybasic(e11->Ety)]<= REGSIZE)
retregs = mask[reg1]; // only want the LSW
else
retregs = regm;
c1 = NULL;
freenode(e11);
}
else
c1 = codelem(e11,&retregs,FALSE);
}
rretregs = ALLREGS & ~retregs & ~mBP;
c2 = scodelem(ebase,&rretregs,retregs,TRUE);
{
regm_t sregs = *pretregs & ~rretregs;
if (!sregs)
sregs = ALLREGS & ~rretregs;
c3 = allocreg(&sregs,®,ty);
}
assert((retregs & (retregs - 1)) == 0); // must be only one register
assert((rretregs & (rretregs - 1)) == 0); // must be only one register
reg1 = findreg(retregs);
reg2 = findreg(rretregs);
if (ss2)
{
assert(reg != reg2);
if ((reg1 & 7) == BP)
{ static unsigned imm32[4] = {1+1,2+1,4+1,8+1};
// IMUL reg,imm32
c = genc2(CNIL,0x69,modregxrmx(3,reg,reg1),imm32[ss]);
}
else
{ // LEA reg,[reg1*ss][reg1]
c = gen2sib(CNIL,0x8D,modregxrm(0,reg,4),modregrm(ss,reg1 & 7,reg1 & 7));
if (reg1 & 8)
code_orrex(c, REX_X | REX_B);
}
if (rex)
code_orrex(c, rex);
reg1 = reg;
ss = ss2; // use *2 for scale
}
else
c = NULL;
c = cat4(c1,c2,c3,c);
cs.Iop = 0x8D; // LEA reg,c[reg1*ss][reg2]
cs.Irm = modregrm(2,reg & 7,4);
cs.Isib = modregrm(ss,reg1 & 7,reg2 & 7);
assert(reg2 != BP);
cs.Iflags = CFoff;
cs.Irex = rex;
if (reg & 8)
cs.Irex |= REX_R;
if (reg1 & 8)
cs.Irex |= REX_X;
if (reg2 & 8)
cs.Irex |= REX_B;
cs.IFL1 = FLconst;
cs.IEV1.Vsize_t = edisp->EV.Vuns;
freenode(edisp);
freenode(e1);
c = gen(c,&cs);
return cat(c,fixresult(e,mask[reg],pretregs));
}
}
posregs = (byte) ? BYTEREGS : (mES | ALLREGS | mBP);
retregs = *pretregs & posregs;
if (retregs == 0) /* if no return regs speced */
/* (like if wanted flags only) */
retregs = ALLREGS & posregs; // give us some
if (tysize[ty1] > REGSIZE && numwords == 1)
{ /* The only possibilities are (TYfptr + tyword) or (TYfptr - tyword) */
#if DEBUG
if (tysize[ty2] != REGSIZE)
{ printf("e = %p, e->Eoper = ",e);
WROP(e->Eoper);
printf(" e1->Ety = ");
WRTYxx(ty1);
printf(" e2->Ety = ");
WRTYxx(ty2);
printf("\n");
elem_print(e);
}
#endif
assert(tysize[ty2] == REGSIZE);
#if TARGET_SEGMENTED
/* Watch out for the case here where you are going to OP reg,EA */
/* and both the reg and EA use ES! Prevent this by forcing */
/* reg into the regular registers. */
if ((e2oper == OPind ||
(e2oper == OPvar && el_fl(e2) == FLfardata)) &&
!e2->Ecount)
{
retregs = ALLREGS;
}
#endif
cl = codelem(e1,&retregs,test);
reg = findreglsw(retregs); /* reg is the register with the offset*/
}
#if TARGET_SEGMENTED
else if (ty1 == TYhptr || ty2 == TYhptr)
{ /* Generate code for add/subtract of huge pointers.
No attempt is made to generate very good code.
*/
unsigned mreg,lreg;
unsigned lrreg;
retregs = (retregs & mLSW) | mDX;
if (ty1 == TYhptr)
{ // hptr +- long
rretregs = mLSW & ~(retregs | regcon.mvar);
if (!rretregs)
rretregs = mLSW;
rretregs |= mCX;
cl = codelem(e1,&rretregs,0);
retregs &= ~rretregs;
if (!(retregs & mLSW))
retregs |= mLSW & ~rretregs;
cr = scodelem(e2,&retregs,rretregs,TRUE);
}
else
{ // long + hptr
cl = codelem(e1,&retregs,0);
rretregs = (mLSW | mCX) & ~retregs;
if (!(rretregs & mLSW))
rretregs |= mLSW;
cr = scodelem(e2,&rretregs,retregs,TRUE);
}
cg = getregs(rretregs | retregs);
mreg = DX;
lreg = findreglsw(retregs);
c = CNIL;
if (e->Eoper == OPmin)
{ // negate retregs
c = gen2(c,0xF7,modregrm(3,3,mreg)); // NEG mreg
gen2(c,0xF7,modregrm(3,3,lreg)); // NEG lreg
code_orflag(c,CFpsw);
genc2(c,0x81,modregrm(3,3,mreg),0); // SBB mreg,0
}
lrreg = findreglsw(rretregs);
c = genregs(c,0x03,lreg,lrreg); // ADD lreg,lrreg
code_orflag(c,CFpsw);
genmovreg(c,lrreg,CX); // MOV lrreg,CX
genc2(c,0x81,modregrm(3,2,mreg),0); // ADC mreg,0
genshift(c); // MOV CX,offset __AHSHIFT
gen2(c,0xD3,modregrm(3,4,mreg)); // SHL mreg,CL
genregs(c,0x03,mreg,lrreg); // ADD mreg,MSREG(h)
goto L5;
}
#endif
else
{ regm_t regm;
/* if (tyword + TYfptr) */
if (tysize[ty1] == REGSIZE && tysize[ty2] > REGSIZE)
{ retregs = ~*pretregs & ALLREGS;
/* if retregs doesn't have any regs in it that aren't reg vars */
if ((retregs & ~regcon.mvar) == 0)
retregs |= mAX;
}
else if (numwords == 2 && retregs & mES)
retregs = (retregs | mMSW) & ALLREGS;
// Determine if we should swap operands, because
// mov EAX,x
// add EAX,reg
// is faster than:
// mov EAX,reg
// add EAX,x
else if (e2oper == OPvar &&
e1->Eoper == OPvar &&
e->Eoper != OPmin &&
isregvar(e1,®m,NULL) &&
regm != retregs &&
tysize[ty1] == tysize[ty2])
{
elem *es = e1;
e1 = e2;
e2 = es;
}
cl = codelem(e1,&retregs,test); /* eval left leaf */
reg = findreg(retregs);
}
switch (e2oper)
{
case OPind: /* if addressing mode */
if (!e2->Ecount) /* if not CSE */
goto L1; /* try OP reg,EA */
/* FALL-THROUGH */
default: /* operator node */
L2:
rretregs = ALLREGS & ~retregs;
/* Be careful not to do arithmetic on ES */
if (tysize[ty1] == REGSIZE && tysize[ty2] > REGSIZE && *pretregs != mPSW)
rretregs = *pretregs & (mES | ALLREGS | mBP) & ~retregs;
else if (byte)
rretregs &= BYTEREGS;
cr = scodelem(e2,&rretregs,retregs,TRUE); /* get rvalue */
rreg = (tysize[ty2] > REGSIZE) ? findreglsw(rretregs) : findreg(rretregs);
c = CNIL;
if (numwords == 1) /* ADD reg,rreg */
{
/* reverse operands to avoid moving around the segment value */
if (tysize[ty2] > REGSIZE)
{ c = cat(c,getregs(rretregs));
c = genregs(c,op1,rreg,reg);
retregs = rretregs; /* reverse operands */
}
else
{ c = genregs(c,op1,reg,rreg);
if (!I16 && *pretregs & mPSW)
c->Iflags |= word;
}
if (I64 && sz == 8)
code_orrex(c, REX_W);
if (I64 && byte && (reg >= 4 || rreg >= 4))
code_orrex(c, REX);
}
else /* numwords == 2 */ /* ADD lsreg,lsrreg */
{
reg = findreglsw(retregs);
rreg = findreglsw(rretregs);
c = genregs(c,op1,reg,rreg);
if (e->Eoper == OPadd || e->Eoper == OPmin)
code_orflag(c,CFpsw);
reg = findregmsw(retregs);
rreg = findregmsw(rretregs);
if (!(e2oper == OPu16_32 && // if second operand is 0
(op2 == 0x0B || op2 == 0x33)) // and OR or XOR
)
genregs(c,op2,reg,rreg); // ADC msreg,msrreg
}
break;
case OPrelconst:
if (sz != REGSIZE)
goto L2;
if (segfl[el_fl(e2)] != 3) /* if not in data segment */
goto L2;
if (evalinregister(e2))
goto L2;
cs.IEVoffset2 = e2->EV.sp.Voffset;
cs.IEVsym2 = e2->EV.sp.Vsym;
cs.Iflags |= CFoff;
i = 0; /* no INC or DEC opcode */
rval = 0;
goto L3;
case OPconst:
if (tyfv(ty2))
goto L2;
if (numwords == 1)
{
if (!el_signx32(e2))
goto L2;
i = e2->EV.Vpointer;
if (word)
{
if (!(*pretregs & mPSW) &&
config.flags4 & CFG4speed &&
(e->Eoper == OPor || e->Eoper == OPxor || test ||
(e1->Eoper != OPvar && e1->Eoper != OPind)))
{ word = 0;
i &= 0xFFFF;
}
}
rval = reghasvalue(byte ? BYTEREGS : ALLREGS,i,&rreg);
cs.IEV2.Vsize_t = i;
L3:
op1 ^= byte;
cs.Iflags |= word;
if (rval)
{ cs.Iop = op1 ^ 2;
mode = rreg;
}
else
cs.Iop = 0x81;
cs.Irm = modregrm(3,mode&7,reg&7);
if (mode & 8)
cs.Irex |= REX_R;
if (reg & 8)
cs.Irex |= REX_B;
if (I64 && sz == 8)
cs.Irex |= REX_W;
if (I64 && byte && (reg >= 4 || (rval && rreg >= 4)))
cs.Irex |= REX;
cs.IFL2 = (e2->Eoper == OPconst) ? FLconst : el_fl(e2);
/* Modify instruction for special cases */
switch (e->Eoper)
{ case OPadd:
{ int iop;
if (i == 1)
iop = 0; /* INC reg */
else if (i == -1)
iop = 8; /* DEC reg */
else
break;
cs.Iop = (0x40 | iop | reg) ^ byte;
if ((byte && *pretregs & mPSW) || I64)
{ cs.Irm = modregrm(3,0,reg & 7) | iop;
cs.Iop = 0xFF;
}
break;
}
case OPand:
if (test)
cs.Iop = rval ? op1 : 0xF7; // TEST
break;
}
if (*pretregs & mPSW)
cs.Iflags |= CFpsw;
cs.Iop ^= byte;
c = gen(CNIL,&cs);
cs.Iflags &= ~CFpsw;
}
else if (numwords == 2)
{ unsigned lsreg;
targ_int msw;
c = getregs(retregs);
reg = findregmsw(retregs);
lsreg = findreglsw(retregs);
cs.Iop = 0x81;
cs.Irm = modregrm(3,mode,lsreg);
cs.IFL2 = FLconst;
msw = MSREG(e2->EV.Vllong);
cs.IEV2.Vint = e2->EV.Vlong;
switch (e->Eoper)
{ case OPadd:
case OPmin:
cs.Iflags |= CFpsw;
break;
}
c = gen(c,&cs);
cs.Iflags &= ~CFpsw;
cs.Irm = (cs.Irm & modregrm(3,7,0)) | reg;
cs.IEV2.Vint = msw;
if (e->Eoper == OPadd)
cs.Irm |= modregrm(0,2,0); /* ADC */
c = gen(c,&cs);
}
else
assert(0);
freenode(e2);
break;
case OPvar:
#if TARGET_OSX
if (movOnly(e2))
goto L2;
#endif
L1:
if (tyfv(ty2))
goto L2;
c = loadea(e2,&cs,op1,
((numwords == 2) ? findreglsw(retregs) : reg),
0,retregs,retregs);
if (!I16 && word)
{ if (*pretregs & mPSW)
code_orflag(c,word);
else
{
code *ce = code_last(c);
ce->Iflags &= ~word;
}
}
else if (numwords == 2)
{
if (e->Eoper == OPadd || e->Eoper == OPmin)
code_orflag(c,CFpsw);
reg = findregmsw(retregs);
if (EOP(e2))
{ getlvalue_msw(&cs);
cs.Iop = op2;
NEWREG(cs.Irm,reg);
c = gen(c,&cs); /* ADC reg,data+2 */
}
else
c = cat(c,loadea(e2,&cs,op2,reg,REGSIZE,retregs,0));
}
else if (I64 && sz == 8)
code_orrex(c, REX_W);
freenode(e2);
break;
}
if (sz <= REGSIZE && *pretregs & mPSW)
{
/* If the expression is (_tls_array + ...), then the flags are not set
* since the linker may rewrite these instructions into something else.
*/
if (I64 && e->Eoper == OPadd && e1->Eoper == OPvar)
{
symbol *s = e1->EV.sp.Vsym;
if (s->Sident[0] == '_' && memcmp(s->Sident + 1,"tls_array",10) == 0)
{
goto L7; // don't assume flags are set
}
}
code_orflag(c,CFpsw);
*pretregs &= ~mPSW; /* flags already set */
L7: ;
}
if (test)
cg = NULL; /* didn't destroy any */
else
cg = getregs(retregs); /* we will trash these regs */
L5:
c = cat(c,fixresult(e,retregs,pretregs));
return cat4(cl,cr,cg,c);
}
/*****************************
* Handle multiply, divide, modulo and remquo.
* Note that modulo isn't defined for doubles.
*/
code *cdmul(elem *e,regm_t *pretregs)
{ unsigned rreg,op,oper,lib,byte;
regm_t resreg,retregs,rretregs;
regm_t keepregs;
tym_t uns; // 1 if unsigned operation, 0 if not
tym_t tyml;
code *c,*cg,*cl,*cr,cs;
elem *e1,*e2;
int sz;
targ_size_t e2factor;
targ_size_t d;
bool neg;
int opunslng;
int pow2;
if (*pretregs == 0) // if don't want result
{ c = codelem(e->E1,pretregs,FALSE); // eval left leaf
*pretregs = 0; // in case they got set
return cat(c,codelem(e->E2,pretregs,FALSE));
}
//printf("cdmul(e = %p, *pretregs = %s)\n", e, regm_str(*pretregs));
keepregs = 0;
cs.Iflags = 0;
cs.Irex = 0;
c = cg = cr = CNIL; // initialize
e2 = e->E2;
e1 = e->E1;
tyml = tybasic(e1->Ety);
sz = tysize[tyml];
byte = tybyte(e->Ety) != 0;
uns = tyuns(tyml) || tyuns(e2->Ety);
oper = e->Eoper;
unsigned rex = (I64 && sz == 8) ? REX_W : 0;
unsigned grex = rex << 16;
if (tyfloating(tyml))
{
if (*pretregs & XMMREGS && oper != OPmod && tyxmmreg(tyml))
return orthxmm(e,pretregs);
#if TARGET_LINUX || TARGET_OSX || TARGET_FREEBSD || TARGET_OPENBSD || TARGET_SOLARIS
return orth87(e,pretregs);
#else
return opdouble(e,pretregs,(oper == OPmul) ? CLIBdmul : CLIBddiv);
#endif
}
if (tyxmmreg(tyml))
return orthxmm(e,pretregs);
opunslng = I16 ? OPu16_32 : OPu32_64;
switch (oper)
{
case OPmul:
resreg = mAX;
op = 5 - uns;
lib = CLIBlmul;
break;
case OPdiv:
resreg = mAX;
op = 7 - uns;
lib = uns ? CLIBuldiv : CLIBldiv;
if (I32)
keepregs |= mSI | mDI;
break;
case OPmod:
resreg = mDX;
op = 7 - uns;
lib = uns ? CLIBulmod : CLIBlmod;
if (I32)
keepregs |= mSI | mDI;
break;
case OPremquo:
resreg = mDX | mAX;
op = 7 - uns;
lib = uns ? CLIBuldiv : CLIBldiv;
if (I32)
keepregs |= mSI | mDI;
break;
default:
assert(0);
}
if (sz <= REGSIZE) // dedicated regs for mul & div
{ retregs = mAX;
/* pick some other regs */
rretregs = byte ? BYTEREGS & ~mAX
: ALLREGS & ~(mAX|mDX);
}
else
{
assert(sz <= 2 * REGSIZE);
retregs = mDX | mAX;
rretregs = mCX | mBX; // second arg
}
switch (e2->Eoper)
{
case OPu16_32:
case OPs16_32:
case OPu32_64:
case OPs32_64:
if (sz != 2 * REGSIZE || oper != OPmul || e1->Eoper != e2->Eoper ||
e1->Ecount || e2->Ecount)
goto L2;
op = (e2->Eoper == opunslng) ? 4 : 5;
retregs = mAX;
cl = codelem(e1->E1,&retregs,FALSE); /* eval left leaf */
if (e2->E1->Eoper == OPvar ||
(e2->E1->Eoper == OPind && !e2->E1->Ecount)
)
{
cr = loadea(e2->E1,&cs,0xF7,op,0,mAX,mAX | mDX);
}
else
{
rretregs = ALLREGS & ~mAX;
cr = scodelem(e2->E1,&rretregs,retregs,TRUE); // get rvalue
cg = getregs(mAX | mDX);
rreg = findreg(rretregs);
cg = gen2(cg,0xF7,grex | modregrmx(3,op,rreg)); // OP AX,rreg
}
freenode(e->E1);
freenode(e2);
c = fixresult(e,mAX | mDX,pretregs);
break;
case OPconst:
e2factor = el_tolong(e2);
neg = false;
d = e2factor;
if (!uns && (targ_llong)e2factor < 0)
{ neg = true;
d = -d;
}
// Multiply by a constant
if (oper == OPmul && I32 && sz == REGSIZE * 2)
{
/* IMUL EDX,EDX,lsw
IMUL reg,EAX,msw
ADD reg,EDX
MOV EDX,lsw
MUL EDX
ADD EDX,reg
if (msw == 0)
IMUL reg,EDX,lsw
MOV EDX,lsw
MUL EDX
ADD EDX,reg
*/
cl = codelem(e1,&retregs,FALSE); // eval left leaf
regm_t scratch = allregs & ~(mAX | mDX);
unsigned reg;
cr = allocreg(&scratch,®,TYint);
cg = getregs(mDX | mAX);
targ_int lsw = e2factor & ((1LL << (REGSIZE * 8)) - 1);
targ_int msw = e2factor >> (REGSIZE * 8);