-
Notifications
You must be signed in to change notification settings - Fork 36
/
analyzer.js
1482 lines (1394 loc) · 62.6 KB
/
analyzer.js
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
//"use strict";
// Analyze intertype data. Calculates things that are necessary in order
// to do the final conversion into JavaScript later, for example,
// properties of variables, loop structures of functions, etc.
var VAR_NATIVE = 'native';
var VAR_NATIVIZED = 'nativized';
var VAR_EMULATED = 'emulated';
var ENTRY_IDENT = toNiceIdent('%0');
var ENTRY_IDENT_IDS = set(0, 1); // XXX
function recomputeLines(func) {
func.lines = func.labels.map(function(label) { return label.lines }).reduce(concatenator, []);
}
// Handy sets
var BRANCH_INVOKE = set('branch', 'invoke');
var SIDE_EFFECT_CAUSERS = set('call', 'invoke', 'atomic');
var UNUNFOLDABLE = set('value', 'type', 'phiparam');
// Analyzer
function analyzer(data, sidePass) {
var mainPass = !sidePass;
// Substrate
var substrate = new Substrate('Analyzer');
// Sorter
substrate.addActor('Sorter', {
processItem: function(item) {
item.items.sort(function (a, b) { return a.lineNum - b.lineNum });
this.forwardItem(item, 'Gatherer');
}
});
// Gatherer
substrate.addActor('Gatherer', {
processItem: function(item) {
// Single-liners
['globalVariable', 'functionStub', 'unparsedFunction', 'unparsedGlobals', 'unparsedTypes', 'alias'].forEach(function(intertype) {
var temp = splitter(item.items, function(item) { return item.intertype == intertype });
item.items = temp.leftIn;
item[intertype + 's'] = temp.splitOut;
});
var temp = splitter(item.items, function(item) { return item.intertype == 'type' });
item.items = temp.leftIn;
temp.splitOut.forEach(function(type) {
//dprint('types', 'adding defined type: ' + type.name_);
Types.types[type.name_] = type;
if (QUANTUM_SIZE === 1) {
Types.fatTypes[type.name_] = copy(type);
}
});
// Functions & labels
item.functions = [];
var currLabelFinished; // Sometimes LLVM puts a branch in the middle of a label. We need to ignore all lines after that.
item.items.sort(function(a, b) { return a.lineNum - b.lineNum });
for (var i = 0; i < item.items.length; i++) {
var subItem = item.items[i];
assert(subItem.lineNum);
if (subItem.intertype == 'function') {
item.functions.push(subItem);
subItem.endLineNum = null;
subItem.lines = []; // We will fill in the function lines after the legalizer, since it can modify them
subItem.labels = [];
subItem.forceEmulated = false;
// no explicit 'entry' label in clang on LLVM 2.8 - most of the time, but not all the time! - so we add one if necessary
if (item.items[i+1].intertype !== 'label') {
item.items.splice(i+1, 0, {
intertype: 'label',
ident: ENTRY_IDENT,
lineNum: subItem.lineNum + '.5'
});
}
} else if (subItem.intertype == 'functionEnd') {
item.functions.slice(-1)[0].endLineNum = subItem.lineNum;
} else if (subItem.intertype == 'label') {
item.functions.slice(-1)[0].labels.push(subItem);
subItem.lines = [];
currLabelFinished = false;
} else if (item.functions.length > 0 && item.functions.slice(-1)[0].endLineNum === null) {
// Internal line
if (!currLabelFinished) {
item.functions.slice(-1)[0].labels.slice(-1)[0].lines.push(subItem); // If this line fails, perhaps missing a label?
if (subItem.intertype === 'branch') {
currLabelFinished = true;
}
} else {
print('// WARNING: content after a branch in a label, line: ' + subItem.lineNum);
}
} else {
throw 'ERROR: what is this? ' + dump(subItem);
}
}
delete item.items;
this.forwardItem(item, 'Legalizer');
}
});
// Legalize LLVM unrealistic types into realistic types.
//
// With full LLVM optimizations, it can generate types like i888 which do not exist in
// any actual hardware implementation, but are useful during optimization. LLVM then
// legalizes these types into real ones during code generation. Sadly, there is no LLVM
// IR pass to legalize them, which would have been useful and nice from a design perspective.
// The LLVM community is also not interested in receiving patches to implement that
// functionality, since it would duplicate existing code from the code generation
// component. Therefore, we implement legalization here in Emscripten.
//
// Currently we just legalize completely unrealistic types into bundles of i32s, and just
// the most common instructions that can be involved with such types: load, store, shifts,
// trunc and zext.
substrate.addActor('Legalizer', {
processItem: function(data) {
// Legalization
if (USE_TYPED_ARRAYS == 2) {
function getLegalVars(base, bits) {
assert(!isNumber(base));
var ret = new Array(Math.ceil(bits/32));
var i = 0;
while (bits > 0) {
ret[i] = { ident: base + '$' + i, bits: Math.min(32, bits) };
bits -= 32;
i++;
}
return ret;
}
function getLegalLiterals(text, bits) {
var parsed = parseArbitraryInt(text, bits);
var ret = new Array(Math.ceil(bits/32));
var i = 0;
while (bits > 0) {
ret[i] = { ident: (parsed[i]|0).toString(), bits: Math.min(32, bits) }; // resign all values
bits -= 32;
i++;
}
return ret;
}
// Uses the right factor to multiply line numbers by so that they fit in between
// the line[i] and the line after it
function interpLines(lines, i, toAdd) {
var prev = i >= 0 ? lines[i].lineNum : -1;
var next = (i < lines.length-1) ? lines[i+1].lineNum : (lines[i].lineNum + 0.5);
var factor = (next - prev)/(4*toAdd.length+3);
for (var k = 0; k < toAdd.length; k++) {
toAdd[k].lineNum = prev + ((k+1)*factor);
}
}
function removeAndAdd(lines, i, toAdd) {
var item = lines[i];
interpLines(lines, i, toAdd);
Array.prototype.splice.apply(lines, [i, 1].concat(toAdd));
return toAdd.length;
}
function legalizeFunctionParameters(params) {
var i = 0;
while (i < params.length) {
var param = params[i];
if (param.intertype == 'value' && isIllegalType(param.type)) {
var toAdd = getLegalVars(param.ident, getBits(param.type)).map(function(element) {
return {
intertype: 'value',
type: 'i' + element.bits,
ident: element.ident,
byval: 0
};
});
Array.prototype.splice.apply(params, [i, 1].concat(toAdd));
i += toAdd.length;
continue;
}
i++;
}
}
function fixUnfolded(item) {
// Unfolded items may need some correction to work properly in the global scope
if (item.intertype in MATHOPS) {
item.op = item.intertype;
item.intertype = 'mathop';
}
}
data.functions.forEach(function(func) {
// Legalize function params
legalizeFunctionParameters(func.params);
// Legalize lines in labels
var tempId = 0;
func.labels.forEach(function(label) {
var i = 0, bits;
while (i < label.lines.length) {
var item = label.lines[i];
var value = item;
// Check if we need to legalize here, and do some trivial legalization along the way
var isIllegal = false;
walkInterdata(item, function(item) {
if (item.intertype == 'getelementptr' || (item.intertype == 'call' && item.ident in LLVM.INTRINSICS_32)) {
// Turn i64 args into i32
for (var i = 0; i < item.params.length; i++) {
if (item.params[i].type == 'i64') item.params[i].type = 'i32';
}
}
if (isIllegalType(item.valueType) || isIllegalType(item.type)) {
isIllegal = true;
}
});
if (!isIllegal) {
i++;
continue;
}
// Unfold this line. If we unfolded, we need to return and process the lines we just
// generated - they may need legalization too
var unfolded = [];
walkAndModifyInterdata(item, function(subItem) {
// Unfold all non-value interitems that we can, and also unfold all numbers (doing the latter
// makes it easier later since we can then assume illegal expressions are always variables
// accessible through ident$x, and not constants we need to parse then and there)
if (subItem != item && (!(subItem.intertype in UNUNFOLDABLE) ||
(subItem.intertype == 'value' && isNumber(subItem.ident) && isIllegalType(subItem.type)))) {
if (item.intertype == 'phi') {
assert(subItem.intertype == 'value', 'We can only unfold illegal constants in phis');
// we must handle this in the phi itself, if we unfold normally it will not be pushed back with the phi
} else {
var tempIdent = '$$emscripten$temp$' + (tempId++);
subItem.assignTo = tempIdent;
unfolded.unshift(subItem);
fixUnfolded(subItem);
return { intertype: 'value', ident: tempIdent, type: subItem.type };
}
} else if (subItem.intertype == 'switch' && isIllegalType(subItem.type)) {
subItem.switchLabels.forEach(function(switchLabel) {
if (switchLabel.value[0] != '$') {
var tempIdent = '$$emscripten$temp$' + (tempId++);
unfolded.unshift({
assignTo: tempIdent,
intertype: 'value',
ident: switchLabel.value,
type: subItem.type
});
switchLabel.value = tempIdent;
}
});
}
});
if (unfolded.length > 0) {
interpLines(label.lines, i-1, unfolded);
Array.prototype.splice.apply(label.lines, [i, 0].concat(unfolded));
continue; // remain at this index, to unfold newly generated lines
}
// This is an illegal-containing line, and it is unfolded. Legalize it now
dprint('legalizer', 'Legalizing ' + item.intertype + ' at line ' + item.lineNum);
var finalizer = null;
switch (item.intertype) {
case 'store': {
var toAdd = [];
bits = getBits(item.valueType);
var elements;
elements = getLegalVars(item.value.ident, bits);
var j = 0;
elements.forEach(function(element) {
var tempVar = '$st$' + i + '$' + j;
toAdd.push({
intertype: 'getelementptr',
assignTo: tempVar,
ident: item.pointer.ident,
type: '[0 x i32]*',
params: [
{ intertype: 'value', ident: item.pointer.ident, type: '[0 x i32]*' }, // technically a bitcase is needed in llvm, but not for us
{ intertype: 'value', ident: '0', type: 'i32' },
{ intertype: 'value', ident: j.toString(), type: 'i32' }
],
});
var actualSizeType = 'i' + element.bits; // The last one may be smaller than 32 bits
toAdd.push({
intertype: 'store',
valueType: actualSizeType,
value: { intertype: 'value', ident: element.ident, type: actualSizeType },
pointer: { intertype: 'value', ident: tempVar, type: actualSizeType + '*' },
ident: tempVar,
pointerType: actualSizeType + '*',
align: item.align,
});
j++;
});
Types.needAnalysis['[0 x i32]'] = 0;
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
// call, return: Return value is in an unlegalized array literal. Not fully optimal.
case 'call': {
bits = getBits(value.type);
var elements = getLegalVars(item.assignTo, bits);
var toAdd = [value];
// legalize parameters
legalizeFunctionParameters(value.params);
if (value.assignTo) {
// legalize return value
var j = 0;
toAdd = toAdd.concat(elements.map(function(element) {
return {
intertype: 'value',
assignTo: element.ident,
type: 'i' + bits,
ident: value.assignTo + '[' + (j++) + ']'
};
}));
}
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
case 'return': {
bits = getBits(item.type);
var elements = getLegalVars(item.value.ident, bits);
item.value.ident = '[' + elements.map(function(element) { return element.ident }).join(',') + ']';
i++;
continue;
}
case 'invoke': {
legalizeFunctionParameters(value.params);
// We can't add lines after this, since invoke already modifies control flow. So we handle the return in invoke
i++;
continue;
}
case 'value': {
bits = getBits(value.type);
var elements = getLegalVars(item.assignTo, bits);
var values = getLegalLiterals(item.ident, bits);
var j = 0;
var toAdd = elements.map(function(element) {
return {
intertype: 'value',
assignTo: element.ident,
type: 'i' + bits,
ident: values[j++].ident
};
});
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
case 'load': {
bits = getBits(value.valueType);
var elements = getLegalVars(item.assignTo, bits);
var j = 0;
var toAdd = [];
elements.forEach(function(element) {
var tempVar = '$st$' + i + '$' + j;
toAdd.push({
intertype: 'getelementptr',
assignTo: tempVar,
ident: value.pointer.ident,
type: '[0 x i32]*',
params: [
{ intertype: 'value', ident: value.pointer.ident, type: '[0 x i32]*' }, // technically bitcast is needed in llvm, but not for us
{ intertype: 'value', ident: '0', type: 'i32' },
{ intertype: 'value', ident: j.toString(), type: 'i32' }
]
});
var actualSizeType = 'i' + element.bits; // The last one may be smaller than 32 bits
toAdd.push({
intertype: 'load',
assignTo: element.ident,
pointerType: actualSizeType + '*',
valueType: actualSizeType,
type: actualSizeType, // XXX why is this missing from intertyper?
pointer: { intertype: 'value', ident: tempVar, type: actualSizeType + '*' },
ident: tempVar,
pointerType: actualSizeType + '*',
align: value.align
});
j++;
});
Types.needAnalysis['[0 x i32]'] = 0;
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
case 'phi': {
bits = getBits(value.type);
var toAdd = [];
var elements = getLegalVars(item.assignTo, bits);
var j = 0;
var literalValues = {}; // special handling of literals - we cannot unfold them normally
value.params.map(function(param) {
if (isNumber(param.value.ident)) {
literalValues[param.value.ident] = getLegalLiterals(param.value.ident, bits);
}
});
elements.forEach(function(element) {
toAdd.push({
intertype: 'phi',
assignTo: element.ident,
type: 'i' + element.bits,
params: value.params.map(function(param) {
return {
intertype: 'phiparam',
label: param.label,
value: {
intertype: 'value',
ident: (param.value.ident in literalValues) ? literalValues[param.value.ident][j].ident : (param.value.ident + '$' + j),
type: 'i' + element.bits,
}
};
})
});
j++;
});
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
case 'switch': {
i++;
continue; // special case, handled in makeComparison
}
case 'bitcast': {
var inType = item.type2;
var outType = item.type;
if ((inType in Runtime.INT_TYPES && outType in Runtime.FLOAT_TYPES) ||
(inType in Runtime.FLOAT_TYPES && outType in Runtime.INT_TYPES)) {
i++;
continue; // special case, handled in processMathop
}
// fall through
}
case 'inttoptr': case 'ptrtoint': case 'zext': case 'sext': case 'trunc': case 'ashr': case 'lshr': case 'shl': case 'or': case 'and': case 'xor': {
value = {
op: item.intertype,
variant: item.variant,
type: item.type,
params: item.params
};
// fall through
}
case 'mathop': {
var toAdd = [];
var sourceBits = getBits(value.params[0].type);
// All mathops can be parametrized by how many shifts we do, and how big the source is
var shifts = 0;
var targetBits = sourceBits;
var processor = null;
var signed = false;
switch (value.op) {
case 'ashr': {
signed = true;
// fall through
}
case 'lshr': {
shifts = parseInt(value.params[1].ident);
break;
}
case 'shl': {
shifts = -parseInt(value.params[1].ident);
break;
}
case 'sext': {
signed = true;
// fall through
}
case 'trunc': case 'zext': case 'ptrtoint': {
targetBits = getBits(value.params[1] ? value.params[1].ident : value.type);
break;
}
case 'inttoptr': {
targetBits = 32;
break;
}
case 'bitcast': {
if (!sourceBits) {
// we can be asked to bitcast doubles or such to integers, handle that as best we can (if it's a double that
// was an x86_fp80, this code will likely break when called)
sourceBits = targetBits = Runtime.getNativeTypeSize(value.params[0].type);
warn('legalizing non-integer bitcast on ll #' + item.lineNum);
}
break;
}
case 'select': {
sourceBits = targetBits = getBits(value.params[1].type);
var otherElementsA = getLegalVars(value.params[1].ident, sourceBits);
var otherElementsB = getLegalVars(value.params[2].ident, sourceBits);
processor = function(result, j) {
return {
intertype: 'mathop',
op: 'select',
type: 'i' + otherElementsA[j].bits,
params: [
value.params[0],
{ intertype: 'value', ident: otherElementsA[j].ident, type: 'i' + otherElementsA[j].bits },
{ intertype: 'value', ident: otherElementsB[j].ident, type: 'i' + otherElementsB[j].bits }
]
};
};
break;
}
case 'or': case 'and': case 'xor': case 'icmp': {
var otherElements = getLegalVars(value.params[1].ident, sourceBits);
processor = function(result, j) {
return {
intertype: 'mathop',
op: value.op,
variant: value.variant,
type: 'i' + otherElements[j].bits,
params: [
result,
{ intertype: 'value', ident: otherElements[j].ident, type: 'i' + otherElements[j].bits }
]
};
};
if (value.op == 'icmp') {
if (sourceBits == 64) { // handle the i64 case in processMathOp, where we handle full i64 math
i++;
continue;
}
finalizer = function() {
var ident = '';
for (var i = 0; i < targetElements.length; i++) {
if (i > 0) {
switch(value.variant) {
case 'eq': ident += '&&'; break;
case 'ne': ident += '||'; break;
default: throw 'unhandleable illegal icmp: ' + value.variant;
}
}
ident += targetElements[i].ident;
}
return {
intertype: 'value',
ident: ident,
type: 'rawJS',
assignTo: item.assignTo
};
}
}
break;
}
case 'add': case 'sub': case 'sdiv': case 'udiv': case 'mul': case 'urem': case 'srem':
case 'uitofp': case 'sitofp': {
// We cannot do these in parallel chunks of 32-bit operations. We will handle these in processMathop
i++;
continue;
}
default: throw 'Invalid mathop for legalization: ' + [value.op, item.lineNum, dump(item)];
}
// Do the legalization
var sourceElements;
if (sourceBits <= 32) {
// The input is a legal type
sourceElements = [{ ident: value.params[0].ident, bits: sourceBits }];
} else {
sourceElements = getLegalVars(value.params[0].ident, sourceBits);
}
if (!isNumber(shifts)) {
// We can't statically legalize this, do the operation at runtime TODO: optimize
assert(sourceBits == 64, 'TODO: handle nonconstant shifts on != 64 bits');
value.intertype = 'value';
value.ident = 'Runtime.bitshift64(' + sourceElements[0].ident + ', ' +
sourceElements[1].ident + ',"' + value.op + '",' + value.params[1].ident + '$0);' +
'var ' + value.assignTo + '$0 = ' + value.assignTo + '[0], ' + value.assignTo + '$1 = ' + value.assignTo + '[1];';
i++;
continue;
}
var targetElements = getLegalVars(item.assignTo, targetBits);
var sign = shifts >= 0 ? 1 : -1;
var shiftOp = shifts >= 0 ? 'shl' : 'lshr';
var shiftOpReverse = shifts >= 0 ? 'lshr' : 'shl';
var whole = shifts >= 0 ? Math.floor(shifts/32) : Math.ceil(shifts/32);
var fraction = Math.abs(shifts % 32);
if (signed) {
var signedFill = '(' + makeSignOp(sourceElements[sourceElements.length-1].ident, 'i' + sourceElements[sourceElements.length-1].bits, 're', 1, 1) + ' < 0 ? -1 : 0)';
var signedKeepAlive = { intertype: 'value', ident: sourceElements[sourceElements.length-1].ident, type: 'i32' };
}
for (var j = 0; j < targetElements.length; j++) {
var result = {
intertype: 'value',
ident: (j + whole >= 0 && j + whole < sourceElements.length) ? sourceElements[j + whole].ident : (signed ? signedFill : '0'),
params: [(signed && j + whole > sourceElements.length) ? signedKeepAlive : null],
type: 'i32',
};
if (j == 0 && isUnsignedOp(value.op) && sourceBits < 32) {
// zext sign correction
result.ident = makeSignOp(result.ident, 'i' + sourceBits, 'un', 1, 1);
}
if (fraction != 0) {
var other = {
intertype: 'value',
ident: (j + sign + whole >= 0 && j + sign + whole < sourceElements.length) ? sourceElements[j + sign + whole].ident : (signed ? signedFill : '0'),
params: [(signed && j + sign + whole > sourceElements.length) ? signedKeepAlive : null],
type: 'i32',
};
other = {
intertype: 'mathop',
op: shiftOp,
type: 'i32',
params: [
other,
{ intertype: 'value', ident: (32 - fraction).toString(), type: 'i32' }
]
};
result = {
intertype: 'mathop',
// shifting in 1s from the top is a special case
op: (signed && shifts >= 0 && j + sign + whole >= sourceElements.length) ? 'ashr' : shiftOpReverse,
type: 'i32',
params: [
result,
{ intertype: 'value', ident: fraction.toString(), type: 'i32' }
]
};
result = {
intertype: 'mathop',
op: 'or',
type: 'i32',
params: [
result,
other
]
}
}
if (targetElements[j].bits < 32 && shifts < 0) {
// truncate bits that fall off the end. This is not needed in most cases, can probably be optimized out
result = {
intertype: 'mathop',
op: 'and',
type: 'i32',
params: [
result,
{ intertype: 'value', ident: (Math.pow(2, targetElements[j].bits)-1).toString(), type: 'i32' }
]
}
}
if (processor) {
result = processor(result, j);
}
result.assignTo = targetElements[j].ident;
toAdd.push(result);
}
if (targetBits <= 32) {
// We are generating a normal legal type here
legalValue = {
intertype: 'value',
ident: targetElements[0].ident + (targetBits < 32 ? '&' + (Math.pow(2, targetBits)-1) : ''),
type: 'rawJS'
};
legalValue.assignTo = item.assignTo;
toAdd.push(legalValue);
} else if (finalizer) {
toAdd.push(finalizer());
}
i += removeAndAdd(label.lines, i, toAdd);
continue;
}
}
assert(0, 'Could not legalize illegal line: ' + [item.lineNum, dump(item)]);
}
if (dcheck('legalizer')) dprint('zz legalized: \n' + dump(label.lines));
});
});
}
// Add function lines to func.lines, after our modifications to the label lines
data.functions.forEach(function(func) {
func.labels.forEach(function(label) {
func.lines = func.lines.concat(label.lines);
});
});
this.forwardItem(data, 'Typevestigator');
}
});
function addTypeInternal(type, data) {
if (type.length == 1) return;
if (Types.types[type]) return;
if (['internal', 'hidden', 'inbounds', 'void'].indexOf(type) != -1) return;
if (Runtime.isNumberType(type)) return;
dprint('types', 'Adding type: ' + type);
// 'blocks': [14 x %struct.X] etc. If this is a pointer, we need
// to look at the underlying type - it was not defined explicitly
// anywhere else.
var nonPointing = removeAllPointing(type);
var check = /^\[(\d+)\ x\ (.*)\]$/.exec(nonPointing);
if (check && !Types.types[nonPointing]) {
var num = parseInt(check[1]);
num = Math.max(num, 1); // [0 x something] is used not for allocations and such of course, but
// for indexing - for an |array of unknown length|, basically. So we
// define the 'type' as having a single field. TODO: Ensure as a sanity
// check that we never allocate with this (either as a child structure
// in the analyzer, or in calcSize in alloca).
var subType = check[2];
addTypeInternal(subType, data); // needed for anonymous structure definitions (see below)
// Huge structural types are represented very inefficiently, both here and in generated JS. Best to avoid them - for example static char x[10*1024*1024]; is bad, while static char *x = malloc(10*1024*1024) is fine.
if (num >= 10*1024*1024) warnOnce('warning: very large fixed-size structural type: ' + type + ' - can you reduce it? (compilation may be slow)');
Types.types[nonPointing] = {
name_: nonPointing,
fields: range(num).map(function() { return subType }),
lineNum: '?'
};
// Also add a |[0 x type]| type
var zerod = '[0 x ' + subType + ']';
if (!Types.types[zerod]) {
Types.types[zerod] = {
name_: zerod,
fields: [subType, subType], // Two, so we get the flatFactor right. We care about the flatFactor, not the size here
lineNum: '?'
};
}
return;
}
// anonymous structure definition, for example |{ i32, i8*, void ()*, i32 }|
if (type[0] == '{' || type[0] == '<') {
type = nonPointing;
var packed = type[0] == '<';
Types.types[type] = {
name_: type,
fields: splitTokenList(tokenize(type.substr(2 + packed, type.length - 4 - 2*packed)).tokens).map(function(segment) {
return segment[0].text;
}),
packed: packed,
lineNum: '?'
};
return;
}
if (isPointerType(type)) return;
if (['['].indexOf(type) != -1) return;
Types.types[type] = {
name_: type,
fields: [ 'i' + (QUANTUM_SIZE*8) ], // a single quantum size
flatSize: 1,
lineNum: '?'
};
}
function addType(type, data) {
addTypeInternal(type, data);
if (QUANTUM_SIZE === 1) {
Types.flipTypes();
addTypeInternal(type, data);
Types.flipTypes();
}
}
// Typevestigator
substrate.addActor('Typevestigator', {
processItem: function(data) {
if (sidePass) { // Do not investigate in the main pass - it is only valid to start to do so in the first side pass,
// which handles type definitions, and later. Doing so before the first side pass will result in
// making bad guesses about types which are actually defined
for (var type in Types.needAnalysis) {
if (type) addType(type, data);
}
Types.needAnalysis = {};
}
this.forwardItem(data, 'Typeanalyzer');
}
});
// Type analyzer
substrate.addActor('Typeanalyzer', {
processItem: function analyzeTypes(item, fatTypes) {
var types = Types.types;
// 'fields' is the raw list of LLVM fields. However, we embed
// child structures into parent structures, basically like C.
// So { int, { int, int }, int } would be represented as
// an Array of 4 ints. getelementptr on the parent would take
// values 0, 1, 2, where 2 is the entire middle structure.
// We also need to be careful with getelementptr to child
// structures - we return a pointer to the same slab, just
// a different offset. Likewise, need to be careful for
// getelementptr of 2 (the last int) - it's real index is 4.
// The benefit of this approach is inheritance -
// { { ancestor } , etc. } = descendant
// In this case it is easy to bitcast ancestor to descendant
// pointers - nothing needs to be done. If the ancestor were
// a new slab, it would need some pointer to the outer one
// for casting in that direction.
// TODO: bitcasts of non-inheritance cases of embedding (not at start)
var more = true;
while (more) {
more = false;
for (var typeName in types) {
var type = types[typeName];
if (type.flatIndexes) continue;
var ready = true;
type.fields.forEach(function(field) {
if (isStructType(field)) {
if (!types[field]) {
addType(field, item);
ready = false;
} else {
if (!types[field].flatIndexes) {
ready = false;
}
}
}
});
if (!ready) {
more = true;
continue;
}
Runtime.calculateStructAlignment(type);
if (dcheck('types')) dprint('type (fat=' + !!fatTypes + '): ' + type.name_ + ' : ' + JSON.stringify(type.fields));
if (dcheck('types')) dprint(' has final size of ' + type.flatSize + ', flatting: ' + type.needsFlattening + ' ? ' + (type.flatFactor ? type.flatFactor : JSON.stringify(type.flatIndexes)));
}
}
if (QUANTUM_SIZE === 1 && !fatTypes) {
Types.flipTypes();
// Fake a quantum size of 4 for fat types. TODO: Might want non-4 for some reason?
var trueQuantumSize = QUANTUM_SIZE;
Runtime.QUANTUM_SIZE = 4;
analyzeTypes(item, true);
Runtime.QUANTUM_SIZE = trueQuantumSize;
Types.flipTypes();
}
if (!fatTypes) {
this.forwardItem(item, 'VariableAnalyzer');
}
}
});
// Variable analyzer
substrate.addActor('VariableAnalyzer', {
processItem: function(item) {
// Globals
var old = item.globalVariables;
item.globalVariables = {};
old.forEach(function(variable) {
variable.impl = 'emulated'; // All global variables are emulated, for now. Consider optimizing later if useful
item.globalVariables[variable.ident] = variable;
});
// Function locals
item.functions.forEach(function(func) {
func.variables = {};
// LLVM is SSA, so we always have a single assignment/write. We care about
// the reads/other uses.
// Function parameters
func.params.forEach(function(param) {
if (param.intertype !== 'varargs') {
func.variables[param.ident] = {
ident: param.ident,
type: param.type,
origin: 'funcparam',
lineNum: func.lineNum,
rawLinesIndex: -1
};
}
});
// Normal variables
func.lines.forEach(function(item, i) {
if (item.assignTo) {
var variable = func.variables[item.assignTo] = {
ident: item.assignTo,
type: item.type,
origin: item.intertype,
lineNum: item.lineNum,
rawLinesIndex: i
};
if (variable.origin === 'alloca') {
variable.allocatedNum = item.allocatedNum;
}
}
});
if (QUANTUM_SIZE === 1) {
// Second pass over variables - notice when types are crossed by bitcast
func.lines.forEach(function(item) {
if (item.assignTo && item.intertype === 'bitcast') {
// bitcasts are unique in that they convert one pointer to another. We
// sometimes need to know the original type of a pointer, so we save that.
//
// originalType is the type this variable is created from
// derivedTypes are the types that this variable is cast into
func.variables[item.assignTo].originalType = item.type2;
if (!isNumber(item.assignTo)) {
if (!func.variables[item.assignTo].derivedTypes) {
func.variables[item.assignTo].derivedTypes = [];
}
func.variables[item.assignTo].derivedTypes.push(item.type);
}
}
});
}
// Analyze variable uses
function analyzeVariableUses() {
dprint('vars', 'Analyzing variables for ' + func.ident + '\n');
for (vname in func.variables) {
var variable = func.variables[vname];
// Whether the value itself is used. For an int, always yes. For a pointer,
// we might never use the pointer's value - we might always just store to it /
// read from it. If so, then we can optimize away the pointer.
variable.hasValueTaken = false;
variable.pointingLevels = pointingLevels(variable.type);
variable.uses = 0;
}
// TODO: improve the analysis precision. bitcast, for example, means we take the value, but perhaps we only use it to load/store
var inNoop = 0;
func.lines.forEach(function(line) {
walkInterdata(line, function(item) {
if (item.intertype == 'noop') inNoop++;
if (!inNoop) {
if (item.ident in func.variables) {
func.variables[item.ident].uses++;
if (item.intertype != 'load' && item.intertype != 'store') {
func.variables[item.ident].hasValueTaken = true;
}
}
}
}, function(item) {
if (item.intertype == 'noop') inNoop--;
});
});
//if (dcheck('vars')) dprint('analyzed variables: ' + dump(func.variables));
}
analyzeVariableUses();
// Decision time
for (vname in func.variables) {
var variable = func.variables[vname];
var pointedType = pointingLevels(variable.type) > 0 ? removePointing(variable.type) : null;
if (variable.origin == 'getelementptr') {
// Use our implementation that emulates pointers etc.
// TODO Can we perhaps nativize some of these? However to do so, we need to discover their
// true types; we have '?' for them now, as they cannot be discovered in the intertyper.
variable.impl = VAR_EMULATED;
} else if (variable.origin == 'funcparam') {
variable.impl = VAR_EMULATED;
} else if (variable.type == 'i64*' && USE_TYPED_ARRAYS == 2) {
variable.impl = VAR_EMULATED;
} else if (MICRO_OPTS && variable.pointingLevels === 0) {
// A simple int value, can be implemented as a native variable
variable.impl = VAR_NATIVE;
} else if (MICRO_OPTS && variable.origin === 'alloca' && !variable.hasValueTaken &&
variable.allocatedNum === 1 &&
(Runtime.isNumberType(pointedType) || Runtime.isPointerType(pointedType))) {
// A pointer to a value which is only accessible through this pointer. Basically
// a local value on the stack, which nothing fancy is done on. So we can
// optimize away the pointing altogether, and just have a native variable
variable.impl = VAR_NATIVIZED;
} else {
variable.impl = VAR_EMULATED;
}
if (dcheck('vars')) dprint('// var ' + vname + ': ' + JSON.stringify(variable));
}
});
this.forwardItem(item, 'Signalyzer');
}
});
// Sign analyzer
//
// Analyze our variables and detect their signs. In USE_TYPED_ARRAYS == 2,
// we can read signed or unsigned values and prevent the need for signing
// corrections. If on the other hand we are doing corrections anyhow, then
// we can skip this pass.
//
// For each variable that is the result of a Load, we look a little forward
// to see where it is used. We only care about mathops, since only they
// need signs.
//
substrate.addActor('Signalyzer', {
processItem: function(item) {
this.forwardItem(item, 'QuantumFixer');
if (USE_TYPED_ARRAYS != 2 || CORRECT_SIGNS == 1) return;
function seekIdent(item, obj) {
if (item.ident === obj.ident) {
obj.found++;
}
}
function seekMathop(item, obj) {
if (item.intertype === 'mathop' && obj.found && !obj.decided) {
if (isUnsignedOp(item.op, item.variant)) {