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
| //===-- WebAssemblyCFGSort.cpp - CFG Sorting ------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
///
/// \file
/// This file implements a CFG sorting pass.
///
/// This pass reorders the blocks in a function to put them into topological
/// order, ignoring loop backedges, and without any loop or exception being
/// interrupted by a block not dominated by the its header, with special care
/// to keep the order as similar as possible to the original order.
///
////===----------------------------------------------------------------------===//
#include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
#include "WebAssembly.h"
#include "WebAssemblyExceptionInfo.h"
#include "WebAssemblySubtarget.h"
#include "WebAssemblyUtilities.h"
#include "llvm/ADT/PriorityQueue.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "wasm-cfg-sort"
// Option to disable EH pad first sorting. Only for testing unwind destination
// mismatches in CFGStackify.
static cl::opt<bool> WasmDisableEHPadSort(
"wasm-disable-ehpad-sort", cl::ReallyHidden,
cl::desc(
"WebAssembly: Disable EH pad-first sort order. Testing purpose only."),
cl::init(false));
namespace {
// Wrapper for loops and exceptions
class Region {
public:
virtual ~Region() = default;
virtual MachineBasicBlock *getHeader() const = 0;
virtual bool contains(const MachineBasicBlock *MBB) const = 0;
virtual unsigned getNumBlocks() const = 0;
using block_iterator = typename ArrayRef<MachineBasicBlock *>::const_iterator;
virtual iterator_range<block_iterator> blocks() const = 0;
virtual bool isLoop() const = 0;
};
template <typename T> class ConcreteRegion : public Region {
const T *Region;
public:
ConcreteRegion(const T *Region) : Region(Region) {}
MachineBasicBlock *getHeader() const override { return Region->getHeader(); }
bool contains(const MachineBasicBlock *MBB) const override {
return Region->contains(MBB);
}
unsigned getNumBlocks() const override { return Region->getNumBlocks(); }
iterator_range<block_iterator> blocks() const override {
return Region->blocks();
}
bool isLoop() const override { return false; }
};
template <> bool ConcreteRegion<MachineLoop>::isLoop() const { return true; }
// This class has information of nested Regions; this is analogous to what
// LoopInfo is for loops.
class RegionInfo {
const MachineLoopInfo &MLI;
const WebAssemblyExceptionInfo &WEI;
std::vector<const Region *> Regions;
DenseMap<const MachineLoop *, std::unique_ptr<Region>> LoopMap;
DenseMap<const WebAssemblyException *, std::unique_ptr<Region>> ExceptionMap;
public:
RegionInfo(const MachineLoopInfo &MLI, const WebAssemblyExceptionInfo &WEI)
: MLI(MLI), WEI(WEI) {}
// Returns a smallest loop or exception that contains MBB
const Region *getRegionFor(const MachineBasicBlock *MBB) {
const auto *ML = MLI.getLoopFor(MBB);
const auto *WE = WEI.getExceptionFor(MBB);
if (!ML && !WE)
return nullptr;
if ((ML && !WE) || (ML && WE && ML->getNumBlocks() < WE->getNumBlocks())) {
// If the smallest region containing MBB is a loop
if (LoopMap.count(ML))
return LoopMap[ML].get();
LoopMap[ML] = std::make_unique<ConcreteRegion<MachineLoop>>(ML);
return LoopMap[ML].get();
} else {
// If the smallest region containing MBB is an exception
if (ExceptionMap.count(WE))
return ExceptionMap[WE].get();
ExceptionMap[WE] =
std::make_unique<ConcreteRegion<WebAssemblyException>>(WE);
return ExceptionMap[WE].get();
}
}
};
class WebAssemblyCFGSort final : public MachineFunctionPass {
StringRef getPassName() const override { return "WebAssembly CFG Sort"; }
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
AU.addRequired<MachineDominatorTree>();
AU.addPreserved<MachineDominatorTree>();
AU.addRequired<MachineLoopInfo>();
AU.addPreserved<MachineLoopInfo>();
AU.addRequired<WebAssemblyExceptionInfo>();
AU.addPreserved<WebAssemblyExceptionInfo>();
MachineFunctionPass::getAnalysisUsage(AU);
}
bool runOnMachineFunction(MachineFunction &MF) override;
public:
static char ID; // Pass identification, replacement for typeid
WebAssemblyCFGSort() : MachineFunctionPass(ID) {}
};
} // end anonymous namespace
char WebAssemblyCFGSort::ID = 0;
INITIALIZE_PASS(WebAssemblyCFGSort, DEBUG_TYPE,
"Reorders blocks in topological order", false, false)
FunctionPass *llvm::createWebAssemblyCFGSort() {
return new WebAssemblyCFGSort();
}
static void maybeUpdateTerminator(MachineBasicBlock *MBB) {
#ifndef NDEBUG
bool AnyBarrier = false;
#endif
bool AllAnalyzable = true;
for (const MachineInstr &Term : MBB->terminators()) {
#ifndef NDEBUG
AnyBarrier |= Term.isBarrier();
#endif
AllAnalyzable &= Term.isBranch() && !Term.isIndirectBranch();
}
assert((AnyBarrier || AllAnalyzable) &&
"AnalyzeBranch needs to analyze any block with a fallthrough");
if (AllAnalyzable)
MBB->updateTerminator();
}
namespace {
// EH pads are selected first regardless of the block comparison order.
// When only one of the BBs is an EH pad, we give a higher priority to it, to
// prevent common mismatches between possibly throwing calls and ehpads they
// unwind to, as in the example below:
//
// bb0:
// call @foo // If this throws, unwind to bb2
// bb1:
// call @bar // If this throws, unwind to bb3
// bb2 (ehpad):
// handler_bb2
// bb3 (ehpad):
// handler_bb3
// continuing code
//
// Because this pass tries to preserve the original BB order, this order will
// not change. But this will result in this try-catch structure in CFGStackify,
// resulting in a mismatch:
// try
// try
// call @foo
// call @bar // This should unwind to bb3, not bb2!
// catch
// handler_bb2
// end
// catch
// handler_bb3
// end
// continuing code
//
// If we give a higher priority to an EH pad whenever it is ready in this
// example, when both bb1 and bb2 are ready, we would pick up bb2 first.
/// Sort blocks by their number.
struct CompareBlockNumbers {
bool operator()(const MachineBasicBlock *A,
const MachineBasicBlock *B) const {
if (!WasmDisableEHPadSort) {
if (A->isEHPad() && !B->isEHPad())
return false;
if (!A->isEHPad() && B->isEHPad())
return true;
}
return A->getNumber() > B->getNumber();
}
};
/// Sort blocks by their number in the opposite order..
struct CompareBlockNumbersBackwards {
bool operator()(const MachineBasicBlock *A,
const MachineBasicBlock *B) const {
if (!WasmDisableEHPadSort) {
if (A->isEHPad() && !B->isEHPad())
return false;
if (!A->isEHPad() && B->isEHPad())
return true;
}
return A->getNumber() < B->getNumber();
}
};
/// Bookkeeping for a region to help ensure that we don't mix blocks not
/// dominated by the its header among its blocks.
struct Entry {
const Region *TheRegion;
unsigned NumBlocksLeft;
/// List of blocks not dominated by Loop's header that are deferred until
/// after all of Loop's blocks have been seen.
std::vector<MachineBasicBlock *> Deferred;
explicit Entry(const class Region *R)
: TheRegion(R), NumBlocksLeft(R->getNumBlocks()) {}
};
} // end anonymous namespace
/// Sort the blocks, taking special care to make sure that regions are not
/// interrupted by blocks not dominated by their header.
/// TODO: There are many opportunities for improving the heuristics here.
/// Explore them.
static void sortBlocks(MachineFunction &MF, const MachineLoopInfo &MLI,
const WebAssemblyExceptionInfo &WEI,
const MachineDominatorTree &MDT) {
// Prepare for a topological sort: Record the number of predecessors each
// block has, ignoring loop backedges.
MF.RenumberBlocks();
SmallVector<unsigned, 16> NumPredsLeft(MF.getNumBlockIDs(), 0);
for (MachineBasicBlock &MBB : MF) {
unsigned N = MBB.pred_size();
if (MachineLoop *L = MLI.getLoopFor(&MBB))
if (L->getHeader() == &MBB)
for (const MachineBasicBlock *Pred : MBB.predecessors())
if (L->contains(Pred))
--N;
NumPredsLeft[MBB.getNumber()] = N;
}
// Topological sort the CFG, with additional constraints:
// - Between a region header and the last block in the region, there can be
// no blocks not dominated by its header.
// - It's desirable to preserve the original block order when possible.
// We use two ready lists; Preferred and Ready. Preferred has recently
// processed successors, to help preserve block sequences from the original
// order. Ready has the remaining ready blocks. EH blocks are picked first
// from both queues.
PriorityQueue<MachineBasicBlock *, std::vector<MachineBasicBlock *>,
CompareBlockNumbers>
Preferred;
PriorityQueue<MachineBasicBlock *, std::vector<MachineBasicBlock *>,
CompareBlockNumbersBackwards>
Ready;
RegionInfo RI(MLI, WEI);
SmallVector<Entry, 4> Entries;
for (MachineBasicBlock *MBB = &MF.front();;) {
const Region *R = RI.getRegionFor(MBB);
if (R) {
// If MBB is a region header, add it to the active region list. We can't
// put any blocks that it doesn't dominate until we see the end of the
// region.
if (R->getHeader() == MBB)
Entries.push_back(Entry(R));
// For each active region the block is in, decrement the count. If MBB is
// the last block in an active region, take it off the list and pick up
// any blocks deferred because the header didn't dominate them.
for (Entry &E : Entries)
if (E.TheRegion->contains(MBB) && --E.NumBlocksLeft == 0)
for (auto DeferredBlock : E.Deferred)
Ready.push(DeferredBlock);
while (!Entries.empty() && Entries.back().NumBlocksLeft == 0)
Entries.pop_back();
}
// The main topological sort logic.
for (MachineBasicBlock *Succ : MBB->successors()) {
// Ignore backedges.
if (MachineLoop *SuccL = MLI.getLoopFor(Succ))
if (SuccL->getHeader() == Succ && SuccL->contains(MBB))
continue;
// Decrement the predecessor count. If it's now zero, it's ready.
if (--NumPredsLeft[Succ->getNumber()] == 0)
Preferred.push(Succ);
}
// Determine the block to follow MBB. First try to find a preferred block,
// to preserve the original block order when possible.
MachineBasicBlock *Next = nullptr;
while (!Preferred.empty()) {
Next = Preferred.top();
Preferred.pop();
// If X isn't dominated by the top active region header, defer it until
// that region is done.
if (!Entries.empty() &&
!MDT.dominates(Entries.back().TheRegion->getHeader(), Next)) {
Entries.back().Deferred.push_back(Next);
Next = nullptr;
continue;
}
// If Next was originally ordered before MBB, and it isn't because it was
// loop-rotated above the header, it's not preferred.
if (Next->getNumber() < MBB->getNumber() &&
(WasmDisableEHPadSort || !Next->isEHPad()) &&
(!R || !R->contains(Next) ||
R->getHeader()->getNumber() < Next->getNumber())) {
Ready.push(Next);
Next = nullptr;
continue;
}
break;
}
// If we didn't find a suitable block in the Preferred list, check the
// general Ready list.
if (!Next) {
// If there are no more blocks to process, we're done.
if (Ready.empty()) {
maybeUpdateTerminator(MBB);
break;
}
for (;;) {
Next = Ready.top();
Ready.pop();
// If Next isn't dominated by the top active region header, defer it
// until that region is done.
if (!Entries.empty() &&
!MDT.dominates(Entries.back().TheRegion->getHeader(), Next)) {
Entries.back().Deferred.push_back(Next);
continue;
}
break;
}
}
// Move the next block into place and iterate.
Next->moveAfter(MBB);
maybeUpdateTerminator(MBB);
MBB = Next;
}
assert(Entries.empty() && "Active sort region list not finished");
MF.RenumberBlocks();
#ifndef NDEBUG
SmallSetVector<const Region *, 8> OnStack;
// Insert a sentinel representing the degenerate loop that starts at the
// function entry block and includes the entire function as a "loop" that
// executes once.
OnStack.insert(nullptr);
for (auto &MBB : MF) {
assert(MBB.getNumber() >= 0 && "Renumbered blocks should be non-negative.");
const Region *Region = RI.getRegionFor(&MBB);
if (Region && &MBB == Region->getHeader()) {
if (Region->isLoop()) {
// Loop header. The loop predecessor should be sorted above, and the
// other predecessors should be backedges below.
for (auto Pred : MBB.predecessors())
assert(
(Pred->getNumber() < MBB.getNumber() || Region->contains(Pred)) &&
"Loop header predecessors must be loop predecessors or "
"backedges");
} else {
// Not a loop header. All predecessors should be sorted above.
for (auto Pred : MBB.predecessors())
assert(Pred->getNumber() < MBB.getNumber() &&
"Non-loop-header predecessors should be topologically sorted");
}
assert(OnStack.insert(Region) &&
"Regions should be declared at most once.");
} else {
// Not a loop header. All predecessors should be sorted above.
for (auto Pred : MBB.predecessors())
assert(Pred->getNumber() < MBB.getNumber() &&
"Non-loop-header predecessors should be topologically sorted");
assert(OnStack.count(RI.getRegionFor(&MBB)) &&
"Blocks must be nested in their regions");
}
while (OnStack.size() > 1 && &MBB == WebAssembly::getBottom(OnStack.back()))
OnStack.pop_back();
}
assert(OnStack.pop_back_val() == nullptr &&
"The function entry block shouldn't actually be a region header");
assert(OnStack.empty() &&
"Control flow stack pushes and pops should be balanced.");
#endif
}
bool WebAssemblyCFGSort::runOnMachineFunction(MachineFunction &MF) {
LLVM_DEBUG(dbgs() << "********** CFG Sorting **********\n"
"********** Function: "
<< MF.getName() << '\n');
const auto &MLI = getAnalysis<MachineLoopInfo>();
const auto &WEI = getAnalysis<WebAssemblyExceptionInfo>();
auto &MDT = getAnalysis<MachineDominatorTree>();
// Liveness is not tracked for VALUE_STACK physreg.
MF.getRegInfo().invalidateLiveness();
// Sort the blocks, with contiguous sort regions.
sortBlocks(MF, MLI, WEI, MDT);
return true;
}
|