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
|
//===---- ScheduleDAG.cpp - Implement the ScheduleDAG class ---------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements the ScheduleDAG class, which is a base class used by
// scheduling implementation classes.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "pre-RA-sched"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/ScheduleHazardRecognizer.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <climits>
using namespace llvm;
ScheduleDAG::ScheduleDAG(MachineFunction &mf)
: TM(mf.getTarget()),
TII(TM.getInstrInfo()),
TRI(TM.getRegisterInfo()),
TLI(TM.getTargetLowering()),
MF(mf), MRI(mf.getRegInfo()),
ConstPool(MF.getConstantPool()),
EntrySU(), ExitSU() {
}
ScheduleDAG::~ScheduleDAG() {}
/// dump - dump the schedule.
void ScheduleDAG::dumpSchedule() const {
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
if (SUnit *SU = Sequence[i])
SU->dump(this);
else
dbgs() << "**** NOOP ****\n";
}
}
/// Run - perform scheduling.
///
void ScheduleDAG::Run(MachineBasicBlock *bb,
MachineBasicBlock::iterator insertPos) {
BB = bb;
InsertPos = insertPos;
SUnits.clear();
Sequence.clear();
EntrySU = SUnit();
ExitSU = SUnit();
Schedule();
DEBUG({
dbgs() << "*** Final schedule ***\n";
dumpSchedule();
dbgs() << '\n';
});
}
/// addPred - This adds the specified edge as a pred of the current node if
/// not already. It also adds the current node as a successor of the
/// specified node.
void SUnit::addPred(const SDep &D) {
// If this node already has this depenence, don't add a redundant one.
for (SmallVector<SDep, 4>::const_iterator I = Preds.begin(), E = Preds.end();
I != E; ++I)
if (*I == D)
return;
// Now add a corresponding succ to N.
SDep P = D;
P.setSUnit(this);
SUnit *N = D.getSUnit();
// Update the bookkeeping.
if (D.getKind() == SDep::Data) {
assert(NumPreds < UINT_MAX && "NumPreds will overflow!");
assert(N->NumSuccs < UINT_MAX && "NumSuccs will overflow!");
++NumPreds;
++N->NumSuccs;
}
if (!N->isScheduled) {
assert(NumPredsLeft < UINT_MAX && "NumPredsLeft will overflow!");
++NumPredsLeft;
}
if (!isScheduled) {
assert(N->NumSuccsLeft < UINT_MAX && "NumSuccsLeft will overflow!");
++N->NumSuccsLeft;
}
Preds.push_back(D);
N->Succs.push_back(P);
if (P.getLatency() != 0) {
this->setDepthDirty();
N->setHeightDirty();
}
}
/// removePred - This removes the specified edge as a pred of the current
/// node if it exists. It also removes the current node as a successor of
/// the specified node.
void SUnit::removePred(const SDep &D) {
// Find the matching predecessor.
for (SmallVector<SDep, 4>::iterator I = Preds.begin(), E = Preds.end();
I != E; ++I)
if (*I == D) {
bool FoundSucc = false;
// Find the corresponding successor in N.
SDep P = D;
P.setSUnit(this);
SUnit *N = D.getSUnit();
for (SmallVector<SDep, 4>::iterator II = N->Succs.begin(),
EE = N->Succs.end(); II != EE; ++II)
if (*II == P) {
FoundSucc = true;
N->Succs.erase(II);
break;
}
assert(FoundSucc && "Mismatching preds / succs lists!");
Preds.erase(I);
// Update the bookkeeping.
if (P.getKind() == SDep::Data) {
assert(NumPreds > 0 && "NumPreds will underflow!");
assert(N->NumSuccs > 0 && "NumSuccs will underflow!");
--NumPreds;
--N->NumSuccs;
}
if (!N->isScheduled) {
assert(NumPredsLeft > 0 && "NumPredsLeft will underflow!");
--NumPredsLeft;
}
if (!isScheduled) {
assert(N->NumSuccsLeft > 0 && "NumSuccsLeft will underflow!");
--N->NumSuccsLeft;
}
if (P.getLatency() != 0) {
this->setDepthDirty();
N->setHeightDirty();
}
return;
}
}
void SUnit::setDepthDirty() {
if (!isDepthCurrent) return;
SmallVector<SUnit*, 8> WorkList;
WorkList.push_back(this);
do {
SUnit *SU = WorkList.pop_back_val();
SU->isDepthCurrent = false;
for (SUnit::const_succ_iterator I = SU->Succs.begin(),
E = SU->Succs.end(); I != E; ++I) {
SUnit *SuccSU = I->getSUnit();
if (SuccSU->isDepthCurrent)
WorkList.push_back(SuccSU);
}
} while (!WorkList.empty());
}
void SUnit::setHeightDirty() {
if (!isHeightCurrent) return;
SmallVector<SUnit*, 8> WorkList;
WorkList.push_back(this);
do {
SUnit *SU = WorkList.pop_back_val();
SU->isHeightCurrent = false;
for (SUnit::const_pred_iterator I = SU->Preds.begin(),
E = SU->Preds.end(); I != E; ++I) {
SUnit *PredSU = I->getSUnit();
if (PredSU->isHeightCurrent)
WorkList.push_back(PredSU);
}
} while (!WorkList.empty());
}
/// setDepthToAtLeast - Update this node's successors to reflect the
/// fact that this node's depth just increased.
///
void SUnit::setDepthToAtLeast(unsigned NewDepth) {
if (NewDepth <= getDepth())
return;
setDepthDirty();
Depth = NewDepth;
isDepthCurrent = true;
}
/// setHeightToAtLeast - Update this node's predecessors to reflect the
/// fact that this node's height just increased.
///
void SUnit::setHeightToAtLeast(unsigned NewHeight) {
if (NewHeight <= getHeight())
return;
setHeightDirty();
Height = NewHeight;
isHeightCurrent = true;
}
/// ComputeDepth - Calculate the maximal path from the node to the exit.
///
void SUnit::ComputeDepth() {
SmallVector<SUnit*, 8> WorkList;
WorkList.push_back(this);
do {
SUnit *Cur = WorkList.back();
bool Done = true;
unsigned MaxPredDepth = 0;
for (SUnit::const_pred_iterator I = Cur->Preds.begin(),
E = Cur->Preds.end(); I != E; ++I) {
SUnit *PredSU = I->getSUnit();
if (PredSU->isDepthCurrent)
MaxPredDepth = std::max(MaxPredDepth,
PredSU->Depth + I->getLatency());
else {
Done = false;
WorkList.push_back(PredSU);
}
}
if (Done) {
WorkList.pop_back();
if (MaxPredDepth != Cur->Depth) {
Cur->setDepthDirty();
Cur->Depth = MaxPredDepth;
}
Cur->isDepthCurrent = true;
}
} while (!WorkList.empty());
}
/// ComputeHeight - Calculate the maximal path from the node to the entry.
///
void SUnit::ComputeHeight() {
SmallVector<SUnit*, 8> WorkList;
WorkList.push_back(this);
do {
SUnit *Cur = WorkList.back();
bool Done = true;
unsigned MaxSuccHeight = 0;
for (SUnit::const_succ_iterator I = Cur->Succs.begin(),
E = Cur->Succs.end(); I != E; ++I) {
SUnit *SuccSU = I->getSUnit();
if (SuccSU->isHeightCurrent)
MaxSuccHeight = std::max(MaxSuccHeight,
SuccSU->Height + I->getLatency());
else {
Done = false;
WorkList.push_back(SuccSU);
}
}
if (Done) {
WorkList.pop_back();
if (MaxSuccHeight != Cur->Height) {
Cur->setHeightDirty();
Cur->Height = MaxSuccHeight;
}
Cur->isHeightCurrent = true;
}
} while (!WorkList.empty());
}
/// SUnit - Scheduling unit. It's an wrapper around either a single SDNode or
/// a group of nodes flagged together.
void SUnit::dump(const ScheduleDAG *G) const {
dbgs() << "SU(" << NodeNum << "): ";
G->dumpNode(this);
}
void SUnit::dumpAll(const ScheduleDAG *G) const {
dump(G);
dbgs() << " # preds left : " << NumPredsLeft << "\n";
dbgs() << " # succs left : " << NumSuccsLeft << "\n";
dbgs() << " Latency : " << Latency << "\n";
dbgs() << " Depth : " << Depth << "\n";
dbgs() << " Height : " << Height << "\n";
if (Preds.size() != 0) {
dbgs() << " Predecessors:\n";
for (SUnit::const_succ_iterator I = Preds.begin(), E = Preds.end();
I != E; ++I) {
dbgs() << " ";
switch (I->getKind()) {
case SDep::Data: dbgs() << "val "; break;
case SDep::Anti: dbgs() << "anti"; break;
case SDep::Output: dbgs() << "out "; break;
case SDep::Order: dbgs() << "ch "; break;
}
dbgs() << "#";
dbgs() << I->getSUnit() << " - SU(" << I->getSUnit()->NodeNum << ")";
if (I->isArtificial())
dbgs() << " *";
dbgs() << ": Latency=" << I->getLatency();
dbgs() << "\n";
}
}
if (Succs.size() != 0) {
dbgs() << " Successors:\n";
for (SUnit::const_succ_iterator I = Succs.begin(), E = Succs.end();
I != E; ++I) {
dbgs() << " ";
switch (I->getKind()) {
case SDep::Data: dbgs() << "val "; break;
case SDep::Anti: dbgs() << "anti"; break;
case SDep::Output: dbgs() << "out "; break;
case SDep::Order: dbgs() << "ch "; break;
}
dbgs() << "#";
dbgs() << I->getSUnit() << " - SU(" << I->getSUnit()->NodeNum << ")";
if (I->isArtificial())
dbgs() << " *";
dbgs() << ": Latency=" << I->getLatency();
dbgs() << "\n";
}
}
dbgs() << "\n";
}
#ifndef NDEBUG
/// VerifySchedule - Verify that all SUnits were scheduled and that
/// their state is consistent.
///
void ScheduleDAG::VerifySchedule(bool isBottomUp) {
bool AnyNotSched = false;
unsigned DeadNodes = 0;
unsigned Noops = 0;
for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
if (!SUnits[i].isScheduled) {
if (SUnits[i].NumPreds == 0 && SUnits[i].NumSuccs == 0) {
++DeadNodes;
continue;
}
if (!AnyNotSched)
dbgs() << "*** Scheduling failed! ***\n";
SUnits[i].dump(this);
dbgs() << "has not been scheduled!\n";
AnyNotSched = true;
}
if (SUnits[i].isScheduled &&
(isBottomUp ? SUnits[i].getHeight() : SUnits[i].getDepth()) >
unsigned(INT_MAX)) {
if (!AnyNotSched)
dbgs() << "*** Scheduling failed! ***\n";
SUnits[i].dump(this);
dbgs() << "has an unexpected "
<< (isBottomUp ? "Height" : "Depth") << " value!\n";
AnyNotSched = true;
}
if (isBottomUp) {
if (SUnits[i].NumSuccsLeft != 0) {
if (!AnyNotSched)
dbgs() << "*** Scheduling failed! ***\n";
SUnits[i].dump(this);
dbgs() << "has successors left!\n";
AnyNotSched = true;
}
} else {
if (SUnits[i].NumPredsLeft != 0) {
if (!AnyNotSched)
dbgs() << "*** Scheduling failed! ***\n";
SUnits[i].dump(this);
dbgs() << "has predecessors left!\n";
AnyNotSched = true;
}
}
}
for (unsigned i = 0, e = Sequence.size(); i != e; ++i)
if (!Sequence[i])
++Noops;
assert(!AnyNotSched);
assert(Sequence.size() + DeadNodes - Noops == SUnits.size() &&
"The number of nodes scheduled doesn't match the expected number!");
}
#endif
/// InitDAGTopologicalSorting - create the initial topological
/// ordering from the DAG to be scheduled.
///
/// The idea of the algorithm is taken from
/// "Online algorithms for managing the topological order of
/// a directed acyclic graph" by David J. Pearce and Paul H.J. Kelly
/// This is the MNR algorithm, which was first introduced by
/// A. Marchetti-Spaccamela, U. Nanni and H. Rohnert in
/// "Maintaining a topological order under edge insertions".
///
/// Short description of the algorithm:
///
/// Topological ordering, ord, of a DAG maps each node to a topological
/// index so that for all edges X->Y it is the case that ord(X) < ord(Y).
///
/// This means that if there is a path from the node X to the node Z,
/// then ord(X) < ord(Z).
///
/// This property can be used to check for reachability of nodes:
/// if Z is reachable from X, then an insertion of the edge Z->X would
/// create a cycle.
///
/// The algorithm first computes a topological ordering for the DAG by
/// initializing the Index2Node and Node2Index arrays and then tries to keep
/// the ordering up-to-date after edge insertions by reordering the DAG.
///
/// On insertion of the edge X->Y, the algorithm first marks by calling DFS
/// the nodes reachable from Y, and then shifts them using Shift to lie
/// immediately after X in Index2Node.
void ScheduleDAGTopologicalSort::InitDAGTopologicalSorting() {
unsigned DAGSize = SUnits.size();
std::vector<SUnit*> WorkList;
WorkList.reserve(DAGSize);
Index2Node.resize(DAGSize);
Node2Index.resize(DAGSize);
// Initialize the data structures.
for (unsigned i = 0, e = DAGSize; i != e; ++i) {
SUnit *SU = &SUnits[i];
int NodeNum = SU->NodeNum;
unsigned Degree = SU->Succs.size();
// Temporarily use the Node2Index array as scratch space for degree counts.
Node2Index[NodeNum] = Degree;
// Is it a node without dependencies?
if (Degree == 0) {
assert(SU->Succs.empty() && "SUnit should have no successors");
// Collect leaf nodes.
WorkList.push_back(SU);
}
}
int Id = DAGSize;
while (!WorkList.empty()) {
SUnit *SU = WorkList.back();
WorkList.pop_back();
Allocate(SU->NodeNum, --Id);
for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
I != E; ++I) {
SUnit *SU = I->getSUnit();
if (!--Node2Index[SU->NodeNum])
// If all dependencies of the node are processed already,
// then the node can be computed now.
WorkList.push_back(SU);
}
}
Visited.resize(DAGSize);
#ifndef NDEBUG
// Check correctness of the ordering
for (unsigned i = 0, e = DAGSize; i != e; ++i) {
SUnit *SU = &SUnits[i];
for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
I != E; ++I) {
assert(Node2Index[SU->NodeNum] > Node2Index[I->getSUnit()->NodeNum] &&
"Wrong topological sorting");
}
}
#endif
}
/// AddPred - Updates the topological ordering to accomodate an edge
/// to be added from SUnit X to SUnit Y.
void ScheduleDAGTopologicalSort::AddPred(SUnit *Y, SUnit *X) {
int UpperBound, LowerBound;
LowerBound = Node2Index[Y->NodeNum];
UpperBound = Node2Index[X->NodeNum];
bool HasLoop = false;
// Is Ord(X) < Ord(Y) ?
if (LowerBound < UpperBound) {
// Update the topological order.
Visited.reset();
DFS(Y, UpperBound, HasLoop);
assert(!HasLoop && "Inserted edge creates a loop!");
// Recompute topological indexes.
Shift(Visited, LowerBound, UpperBound);
}
}
/// RemovePred - Updates the topological ordering to accomodate an
/// an edge to be removed from the specified node N from the predecessors
/// of the current node M.
void ScheduleDAGTopologicalSort::RemovePred(SUnit *M, SUnit *N) {
// InitDAGTopologicalSorting();
}
/// DFS - Make a DFS traversal to mark all nodes reachable from SU and mark
/// all nodes affected by the edge insertion. These nodes will later get new
/// topological indexes by means of the Shift method.
void ScheduleDAGTopologicalSort::DFS(const SUnit *SU, int UpperBound,
bool& HasLoop) {
std::vector<const SUnit*> WorkList;
WorkList.reserve(SUnits.size());
WorkList.push_back(SU);
do {
SU = WorkList.back();
WorkList.pop_back();
Visited.set(SU->NodeNum);
for (int I = SU->Succs.size()-1; I >= 0; --I) {
int s = SU->Succs[I].getSUnit()->NodeNum;
if (Node2Index[s] == UpperBound) {
HasLoop = true;
return;
}
// Visit successors if not already and in affected region.
if (!Visited.test(s) && Node2Index[s] < UpperBound) {
WorkList.push_back(SU->Succs[I].getSUnit());
}
}
} while (!WorkList.empty());
}
/// Shift - Renumber the nodes so that the topological ordering is
/// preserved.
void ScheduleDAGTopologicalSort::Shift(BitVector& Visited, int LowerBound,
int UpperBound) {
std::vector<int> L;
int shift = 0;
int i;
for (i = LowerBound; i <= UpperBound; ++i) {
// w is node at topological index i.
int w = Index2Node[i];
if (Visited.test(w)) {
// Unmark.
Visited.reset(w);
L.push_back(w);
shift = shift + 1;
} else {
Allocate(w, i - shift);
}
}
for (unsigned j = 0; j < L.size(); ++j) {
Allocate(L[j], i - shift);
i = i + 1;
}
}
/// WillCreateCycle - Returns true if adding an edge from SU to TargetSU will
/// create a cycle.
bool ScheduleDAGTopologicalSort::WillCreateCycle(SUnit *SU, SUnit *TargetSU) {
if (IsReachable(TargetSU, SU))
return true;
for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
I != E; ++I)
if (I->isAssignedRegDep() &&
IsReachable(TargetSU, I->getSUnit()))
return true;
return false;
}
/// IsReachable - Checks if SU is reachable from TargetSU.
bool ScheduleDAGTopologicalSort::IsReachable(const SUnit *SU,
const SUnit *TargetSU) {
// If insertion of the edge SU->TargetSU would create a cycle
// then there is a path from TargetSU to SU.
int UpperBound, LowerBound;
LowerBound = Node2Index[TargetSU->NodeNum];
UpperBound = Node2Index[SU->NodeNum];
bool HasLoop = false;
// Is Ord(TargetSU) < Ord(SU) ?
if (LowerBound < UpperBound) {
Visited.reset();
// There may be a path from TargetSU to SU. Check for it.
DFS(TargetSU, UpperBound, HasLoop);
}
return HasLoop;
}
/// Allocate - assign the topological index to the node n.
void ScheduleDAGTopologicalSort::Allocate(int n, int index) {
Node2Index[n] = index;
Index2Node[index] = n;
}
ScheduleDAGTopologicalSort::ScheduleDAGTopologicalSort(
std::vector<SUnit> &sunits)
: SUnits(sunits) {}
ScheduleHazardRecognizer::~ScheduleHazardRecognizer() {}
|