1 //===--- BitTracker.cpp ---------------------------------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9
10 // SSA-based bit propagation.
11 //
12 // The purpose of this code is, for a given virtual register, to provide
13 // information about the value of each bit in the register. The values
14 // of bits are represented by the class BitValue, and take one of four
15 // cases: 0, 1, "ref" and "bottom". The 0 and 1 are rather clear, the
16 // "ref" value means that the bit is a copy of another bit (which itself
17 // cannot be a copy of yet another bit---such chains are not allowed).
18 // A "ref" value is associated with a BitRef structure, which indicates
19 // which virtual register, and which bit in that register is the origin
20 // of the value. For example, given an instruction
21 // vreg2 = ASL vreg1, 1
22 // assuming that nothing is known about bits of vreg1, bit 1 of vreg2
23 // will be a "ref" to (vreg1, 0). If there is a subsequent instruction
24 // vreg3 = ASL vreg2, 2
25 // then bit 3 of vreg3 will be a "ref" to (vreg1, 0) as well.
26 // The "bottom" case means that the bit's value cannot be determined,
27 // and that this virtual register actually defines it. The "bottom" case
28 // is discussed in detail in BitTracker.h. In fact, "bottom" is a "ref
29 // to self", so for the vreg1 above, the bit 0 of it will be a "ref" to
30 // (vreg1, 0), bit 1 will be a "ref" to (vreg1, 1), etc.
31 //
32 // The tracker implements the Wegman-Zadeck algorithm, originally developed
33 // for SSA-based constant propagation. Each register is represented as
34 // a sequence of bits, with the convention that bit 0 is the least signi-
35 // ficant bit. Each bit is propagated individually. The class RegisterCell
36 // implements the register's representation, and is also the subject of
37 // the lattice operations in the tracker.
38 //
39 // The intended usage of the bit tracker is to create a target-specific
40 // machine instruction evaluator, pass the evaluator to the BitTracker
41 // object, and run the tracker. The tracker will then collect the bit
42 // value information for a given machine function. After that, it can be
43 // queried for the cells for each virtual register.
44 // Sample code:
45 // const TargetSpecificEvaluator TSE(TRI, MRI);
46 // BitTracker BT(TSE, MF);
47 // BT.run();
48 // ...
49 // unsigned Reg = interestingRegister();
50 // RegisterCell RC = BT.get(Reg);
51 // if (RC[3].is(1))
52 // Reg0bit3 = 1;
53 //
54 // The code below is intended to be fully target-independent.
55
56 #include "llvm/CodeGen/MachineBasicBlock.h"
57 #include "llvm/CodeGen/MachineFunction.h"
58 #include "llvm/CodeGen/MachineInstr.h"
59 #include "llvm/CodeGen/MachineRegisterInfo.h"
60 #include "llvm/IR/Constants.h"
61 #include "llvm/Support/Debug.h"
62 #include "llvm/Support/raw_ostream.h"
63 #include "llvm/Target/TargetRegisterInfo.h"
64
65 #include "BitTracker.h"
66
67 using namespace llvm;
68
69 typedef BitTracker BT;
70
71 namespace {
72 // Local trickery to pretty print a register (without the whole "%vreg"
73 // business).
74 struct printv {
printv__anon4c835a8c0111::printv75 printv(unsigned r) : R(r) {}
76 unsigned R;
77 };
operator <<(raw_ostream & OS,const printv & PV)78 raw_ostream &operator<< (raw_ostream &OS, const printv &PV) {
79 if (PV.R)
80 OS << 'v' << TargetRegisterInfo::virtReg2Index(PV.R);
81 else
82 OS << 's';
83 return OS;
84 }
85 }
86
operator <<(raw_ostream & OS,const BT::BitValue & BV)87 raw_ostream &llvm::operator<<(raw_ostream &OS, const BT::BitValue &BV) {
88 switch (BV.Type) {
89 case BT::BitValue::Top:
90 OS << 'T';
91 break;
92 case BT::BitValue::Zero:
93 OS << '0';
94 break;
95 case BT::BitValue::One:
96 OS << '1';
97 break;
98 case BT::BitValue::Ref:
99 OS << printv(BV.RefI.Reg) << '[' << BV.RefI.Pos << ']';
100 break;
101 }
102 return OS;
103 }
104
operator <<(raw_ostream & OS,const BT::RegisterCell & RC)105 raw_ostream &llvm::operator<<(raw_ostream &OS, const BT::RegisterCell &RC) {
106 unsigned n = RC.Bits.size();
107 OS << "{ w:" << n;
108 // Instead of printing each bit value individually, try to group them
109 // into logical segments, such as sequences of 0 or 1 bits or references
110 // to consecutive bits (e.g. "bits 3-5 are same as bits 7-9 of reg xyz").
111 // "Start" will be the index of the beginning of the most recent segment.
112 unsigned Start = 0;
113 bool SeqRef = false; // A sequence of refs to consecutive bits.
114 bool ConstRef = false; // A sequence of refs to the same bit.
115
116 for (unsigned i = 1, n = RC.Bits.size(); i < n; ++i) {
117 const BT::BitValue &V = RC[i];
118 const BT::BitValue &SV = RC[Start];
119 bool IsRef = (V.Type == BT::BitValue::Ref);
120 // If the current value is the same as Start, skip to the next one.
121 if (!IsRef && V == SV)
122 continue;
123 if (IsRef && SV.Type == BT::BitValue::Ref && V.RefI.Reg == SV.RefI.Reg) {
124 if (Start+1 == i) {
125 SeqRef = (V.RefI.Pos == SV.RefI.Pos+1);
126 ConstRef = (V.RefI.Pos == SV.RefI.Pos);
127 }
128 if (SeqRef && V.RefI.Pos == SV.RefI.Pos+(i-Start))
129 continue;
130 if (ConstRef && V.RefI.Pos == SV.RefI.Pos)
131 continue;
132 }
133
134 // The current value is different. Print the previous one and reset
135 // the Start.
136 OS << " [" << Start;
137 unsigned Count = i - Start;
138 if (Count == 1) {
139 OS << "]:" << SV;
140 } else {
141 OS << '-' << i-1 << "]:";
142 if (SV.Type == BT::BitValue::Ref && SeqRef)
143 OS << printv(SV.RefI.Reg) << '[' << SV.RefI.Pos << '-'
144 << SV.RefI.Pos+(Count-1) << ']';
145 else
146 OS << SV;
147 }
148 Start = i;
149 SeqRef = ConstRef = false;
150 }
151
152 OS << " [" << Start;
153 unsigned Count = n - Start;
154 if (n-Start == 1) {
155 OS << "]:" << RC[Start];
156 } else {
157 OS << '-' << n-1 << "]:";
158 const BT::BitValue &SV = RC[Start];
159 if (SV.Type == BT::BitValue::Ref && SeqRef)
160 OS << printv(SV.RefI.Reg) << '[' << SV.RefI.Pos << '-'
161 << SV.RefI.Pos+(Count-1) << ']';
162 else
163 OS << SV;
164 }
165 OS << " }";
166
167 return OS;
168 }
169
BitTracker(const MachineEvaluator & E,MachineFunction & F)170 BitTracker::BitTracker(const MachineEvaluator &E, MachineFunction &F)
171 : Trace(false), ME(E), MF(F), MRI(F.getRegInfo()), Map(*new CellMapType) {}
172
~BitTracker()173 BitTracker::~BitTracker() {
174 delete ⤅
175 }
176
177
178 // If we were allowed to update a cell for a part of a register, the meet
179 // operation would need to be parametrized by the register number and the
180 // exact part of the register, so that the computer BitRefs correspond to
181 // the actual bits of the "self" register.
182 // While this cannot happen in the current implementation, I'm not sure
183 // if this should be ruled out in the future.
meet(const RegisterCell & RC,unsigned SelfR)184 bool BT::RegisterCell::meet(const RegisterCell &RC, unsigned SelfR) {
185 // An example when "meet" can be invoked with SelfR == 0 is a phi node
186 // with a physical register as an operand.
187 assert(SelfR == 0 || TargetRegisterInfo::isVirtualRegister(SelfR));
188 bool Changed = false;
189 for (uint16_t i = 0, n = Bits.size(); i < n; ++i) {
190 const BitValue &RCV = RC[i];
191 Changed |= Bits[i].meet(RCV, BitRef(SelfR, i));
192 }
193 return Changed;
194 }
195
196
197 // Insert the entire cell RC into the current cell at position given by M.
insert(const BT::RegisterCell & RC,const BitMask & M)198 BT::RegisterCell &BT::RegisterCell::insert(const BT::RegisterCell &RC,
199 const BitMask &M) {
200 uint16_t B = M.first(), E = M.last(), W = width();
201 // Sanity: M must be a valid mask for *this.
202 assert(B < W && E < W);
203 // Sanity: the masked part of *this must have the same number of bits
204 // as the source.
205 assert(B > E || E-B+1 == RC.width()); // B <= E => E-B+1 = |RC|.
206 assert(B <= E || E+(W-B)+1 == RC.width()); // E < B => E+(W-B)+1 = |RC|.
207 if (B <= E) {
208 for (uint16_t i = 0; i <= E-B; ++i)
209 Bits[i+B] = RC[i];
210 } else {
211 for (uint16_t i = 0; i < W-B; ++i)
212 Bits[i+B] = RC[i];
213 for (uint16_t i = 0; i <= E; ++i)
214 Bits[i] = RC[i+(W-B)];
215 }
216 return *this;
217 }
218
219
extract(const BitMask & M) const220 BT::RegisterCell BT::RegisterCell::extract(const BitMask &M) const {
221 uint16_t B = M.first(), E = M.last(), W = width();
222 assert(B < W && E < W);
223 if (B <= E) {
224 RegisterCell RC(E-B+1);
225 for (uint16_t i = B; i <= E; ++i)
226 RC.Bits[i-B] = Bits[i];
227 return RC;
228 }
229
230 RegisterCell RC(E+(W-B)+1);
231 for (uint16_t i = 0; i < W-B; ++i)
232 RC.Bits[i] = Bits[i+B];
233 for (uint16_t i = 0; i <= E; ++i)
234 RC.Bits[i+(W-B)] = Bits[i];
235 return RC;
236 }
237
238
rol(uint16_t Sh)239 BT::RegisterCell &BT::RegisterCell::rol(uint16_t Sh) {
240 // Rotate left (i.e. towards increasing bit indices).
241 // Swap the two parts: [0..W-Sh-1] [W-Sh..W-1]
242 uint16_t W = width();
243 Sh = Sh % W;
244 if (Sh == 0)
245 return *this;
246
247 RegisterCell Tmp(W-Sh);
248 // Tmp = [0..W-Sh-1].
249 for (uint16_t i = 0; i < W-Sh; ++i)
250 Tmp[i] = Bits[i];
251 // Shift [W-Sh..W-1] to [0..Sh-1].
252 for (uint16_t i = 0; i < Sh; ++i)
253 Bits[i] = Bits[W-Sh+i];
254 // Copy Tmp to [Sh..W-1].
255 for (uint16_t i = 0; i < W-Sh; ++i)
256 Bits[i+Sh] = Tmp.Bits[i];
257 return *this;
258 }
259
260
fill(uint16_t B,uint16_t E,const BitValue & V)261 BT::RegisterCell &BT::RegisterCell::fill(uint16_t B, uint16_t E,
262 const BitValue &V) {
263 assert(B <= E);
264 while (B < E)
265 Bits[B++] = V;
266 return *this;
267 }
268
269
cat(const RegisterCell & RC)270 BT::RegisterCell &BT::RegisterCell::cat(const RegisterCell &RC) {
271 // Append the cell given as the argument to the "this" cell.
272 // Bit 0 of RC becomes bit W of the result, where W is this->width().
273 uint16_t W = width(), WRC = RC.width();
274 Bits.resize(W+WRC);
275 for (uint16_t i = 0; i < WRC; ++i)
276 Bits[i+W] = RC.Bits[i];
277 return *this;
278 }
279
280
ct(bool B) const281 uint16_t BT::RegisterCell::ct(bool B) const {
282 uint16_t W = width();
283 uint16_t C = 0;
284 BitValue V = B;
285 while (C < W && Bits[C] == V)
286 C++;
287 return C;
288 }
289
290
cl(bool B) const291 uint16_t BT::RegisterCell::cl(bool B) const {
292 uint16_t W = width();
293 uint16_t C = 0;
294 BitValue V = B;
295 while (C < W && Bits[W-(C+1)] == V)
296 C++;
297 return C;
298 }
299
300
operator ==(const RegisterCell & RC) const301 bool BT::RegisterCell::operator== (const RegisterCell &RC) const {
302 uint16_t W = Bits.size();
303 if (RC.Bits.size() != W)
304 return false;
305 for (uint16_t i = 0; i < W; ++i)
306 if (Bits[i] != RC[i])
307 return false;
308 return true;
309 }
310
311
getRegBitWidth(const RegisterRef & RR) const312 uint16_t BT::MachineEvaluator::getRegBitWidth(const RegisterRef &RR) const {
313 // The general problem is with finding a register class that corresponds
314 // to a given reference reg:sub. There can be several such classes, and
315 // since we only care about the register size, it does not matter which
316 // such class we would find.
317 // The easiest way to accomplish what we want is to
318 // 1. find a physical register PhysR from the same class as RR.Reg,
319 // 2. find a physical register PhysS that corresponds to PhysR:RR.Sub,
320 // 3. find a register class that contains PhysS.
321 unsigned PhysR;
322 if (TargetRegisterInfo::isVirtualRegister(RR.Reg)) {
323 const TargetRegisterClass *VC = MRI.getRegClass(RR.Reg);
324 assert(VC->begin() != VC->end() && "Empty register class");
325 PhysR = *VC->begin();
326 } else {
327 assert(TargetRegisterInfo::isPhysicalRegister(RR.Reg));
328 PhysR = RR.Reg;
329 }
330
331 unsigned PhysS = (RR.Sub == 0) ? PhysR : TRI.getSubReg(PhysR, RR.Sub);
332 const TargetRegisterClass *RC = TRI.getMinimalPhysRegClass(PhysS);
333 uint16_t BW = RC->getSize()*8;
334 return BW;
335 }
336
337
getCell(const RegisterRef & RR,const CellMapType & M) const338 BT::RegisterCell BT::MachineEvaluator::getCell(const RegisterRef &RR,
339 const CellMapType &M) const {
340 uint16_t BW = getRegBitWidth(RR);
341
342 // Physical registers are assumed to be present in the map with an unknown
343 // value. Don't actually insert anything in the map, just return the cell.
344 if (TargetRegisterInfo::isPhysicalRegister(RR.Reg))
345 return RegisterCell::self(0, BW);
346
347 assert(TargetRegisterInfo::isVirtualRegister(RR.Reg));
348 // For virtual registers that belong to a class that is not tracked,
349 // generate an "unknown" value as well.
350 const TargetRegisterClass *C = MRI.getRegClass(RR.Reg);
351 if (!track(C))
352 return RegisterCell::self(0, BW);
353
354 CellMapType::const_iterator F = M.find(RR.Reg);
355 if (F != M.end()) {
356 if (!RR.Sub)
357 return F->second;
358 BitMask M = mask(RR.Reg, RR.Sub);
359 return F->second.extract(M);
360 }
361 // If not found, create a "top" entry, but do not insert it in the map.
362 return RegisterCell::top(BW);
363 }
364
365
putCell(const RegisterRef & RR,RegisterCell RC,CellMapType & M) const366 void BT::MachineEvaluator::putCell(const RegisterRef &RR, RegisterCell RC,
367 CellMapType &M) const {
368 // While updating the cell map can be done in a meaningful way for
369 // a part of a register, it makes little sense to implement it as the
370 // SSA representation would never contain such "partial definitions".
371 if (!TargetRegisterInfo::isVirtualRegister(RR.Reg))
372 return;
373 assert(RR.Sub == 0 && "Unexpected sub-register in definition");
374 // Eliminate all ref-to-reg-0 bit values: replace them with "self".
375 for (unsigned i = 0, n = RC.width(); i < n; ++i) {
376 const BitValue &V = RC[i];
377 if (V.Type == BitValue::Ref && V.RefI.Reg == 0)
378 RC[i].RefI = BitRef(RR.Reg, i);
379 }
380 M[RR.Reg] = RC;
381 }
382
383
384 // Check if the cell represents a compile-time integer value.
isInt(const RegisterCell & A) const385 bool BT::MachineEvaluator::isInt(const RegisterCell &A) const {
386 uint16_t W = A.width();
387 for (uint16_t i = 0; i < W; ++i)
388 if (!A[i].is(0) && !A[i].is(1))
389 return false;
390 return true;
391 }
392
393
394 // Convert a cell to the integer value. The result must fit in uint64_t.
toInt(const RegisterCell & A) const395 uint64_t BT::MachineEvaluator::toInt(const RegisterCell &A) const {
396 assert(isInt(A));
397 uint64_t Val = 0;
398 uint16_t W = A.width();
399 for (uint16_t i = 0; i < W; ++i) {
400 Val <<= 1;
401 Val |= A[i].is(1);
402 }
403 return Val;
404 }
405
406
407 // Evaluator helper functions. These implement some common operation on
408 // register cells that can be used to implement target-specific instructions
409 // in a target-specific evaluator.
410
eIMM(int64_t V,uint16_t W) const411 BT::RegisterCell BT::MachineEvaluator::eIMM(int64_t V, uint16_t W) const {
412 RegisterCell Res(W);
413 // For bits beyond the 63rd, this will generate the sign bit of V.
414 for (uint16_t i = 0; i < W; ++i) {
415 Res[i] = BitValue(V & 1);
416 V >>= 1;
417 }
418 return Res;
419 }
420
421
eIMM(const ConstantInt * CI) const422 BT::RegisterCell BT::MachineEvaluator::eIMM(const ConstantInt *CI) const {
423 APInt A = CI->getValue();
424 uint16_t BW = A.getBitWidth();
425 assert((unsigned)BW == A.getBitWidth() && "BitWidth overflow");
426 RegisterCell Res(BW);
427 for (uint16_t i = 0; i < BW; ++i)
428 Res[i] = A[i];
429 return Res;
430 }
431
432
eADD(const RegisterCell & A1,const RegisterCell & A2) const433 BT::RegisterCell BT::MachineEvaluator::eADD(const RegisterCell &A1,
434 const RegisterCell &A2) const {
435 uint16_t W = A1.width();
436 assert(W == A2.width());
437 RegisterCell Res(W);
438 bool Carry = false;
439 uint16_t I;
440 for (I = 0; I < W; ++I) {
441 const BitValue &V1 = A1[I];
442 const BitValue &V2 = A2[I];
443 if (!V1.num() || !V2.num())
444 break;
445 unsigned S = bool(V1) + bool(V2) + Carry;
446 Res[I] = BitValue(S & 1);
447 Carry = (S > 1);
448 }
449 for (; I < W; ++I) {
450 const BitValue &V1 = A1[I];
451 const BitValue &V2 = A2[I];
452 // If the next bit is same as Carry, the result will be 0 plus the
453 // other bit. The Carry bit will remain unchanged.
454 if (V1.is(Carry))
455 Res[I] = BitValue::ref(V2);
456 else if (V2.is(Carry))
457 Res[I] = BitValue::ref(V1);
458 else
459 break;
460 }
461 for (; I < W; ++I)
462 Res[I] = BitValue::self();
463 return Res;
464 }
465
466
eSUB(const RegisterCell & A1,const RegisterCell & A2) const467 BT::RegisterCell BT::MachineEvaluator::eSUB(const RegisterCell &A1,
468 const RegisterCell &A2) const {
469 uint16_t W = A1.width();
470 assert(W == A2.width());
471 RegisterCell Res(W);
472 bool Borrow = false;
473 uint16_t I;
474 for (I = 0; I < W; ++I) {
475 const BitValue &V1 = A1[I];
476 const BitValue &V2 = A2[I];
477 if (!V1.num() || !V2.num())
478 break;
479 unsigned S = bool(V1) - bool(V2) - Borrow;
480 Res[I] = BitValue(S & 1);
481 Borrow = (S > 1);
482 }
483 for (; I < W; ++I) {
484 const BitValue &V1 = A1[I];
485 const BitValue &V2 = A2[I];
486 if (V1.is(Borrow)) {
487 Res[I] = BitValue::ref(V2);
488 break;
489 }
490 if (V2.is(Borrow))
491 Res[I] = BitValue::ref(V1);
492 else
493 break;
494 }
495 for (; I < W; ++I)
496 Res[I] = BitValue::self();
497 return Res;
498 }
499
500
eMLS(const RegisterCell & A1,const RegisterCell & A2) const501 BT::RegisterCell BT::MachineEvaluator::eMLS(const RegisterCell &A1,
502 const RegisterCell &A2) const {
503 uint16_t W = A1.width() + A2.width();
504 uint16_t Z = A1.ct(0) + A2.ct(0);
505 RegisterCell Res(W);
506 Res.fill(0, Z, BitValue::Zero);
507 Res.fill(Z, W, BitValue::self());
508 return Res;
509 }
510
511
eMLU(const RegisterCell & A1,const RegisterCell & A2) const512 BT::RegisterCell BT::MachineEvaluator::eMLU(const RegisterCell &A1,
513 const RegisterCell &A2) const {
514 uint16_t W = A1.width() + A2.width();
515 uint16_t Z = A1.ct(0) + A2.ct(0);
516 RegisterCell Res(W);
517 Res.fill(0, Z, BitValue::Zero);
518 Res.fill(Z, W, BitValue::self());
519 return Res;
520 }
521
522
eASL(const RegisterCell & A1,uint16_t Sh) const523 BT::RegisterCell BT::MachineEvaluator::eASL(const RegisterCell &A1,
524 uint16_t Sh) const {
525 assert(Sh <= A1.width());
526 RegisterCell Res = RegisterCell::ref(A1);
527 Res.rol(Sh);
528 Res.fill(0, Sh, BitValue::Zero);
529 return Res;
530 }
531
532
eLSR(const RegisterCell & A1,uint16_t Sh) const533 BT::RegisterCell BT::MachineEvaluator::eLSR(const RegisterCell &A1,
534 uint16_t Sh) const {
535 uint16_t W = A1.width();
536 assert(Sh <= W);
537 RegisterCell Res = RegisterCell::ref(A1);
538 Res.rol(W-Sh);
539 Res.fill(W-Sh, W, BitValue::Zero);
540 return Res;
541 }
542
543
eASR(const RegisterCell & A1,uint16_t Sh) const544 BT::RegisterCell BT::MachineEvaluator::eASR(const RegisterCell &A1,
545 uint16_t Sh) const {
546 uint16_t W = A1.width();
547 assert(Sh <= W);
548 RegisterCell Res = RegisterCell::ref(A1);
549 BitValue Sign = Res[W-1];
550 Res.rol(W-Sh);
551 Res.fill(W-Sh, W, Sign);
552 return Res;
553 }
554
555
eAND(const RegisterCell & A1,const RegisterCell & A2) const556 BT::RegisterCell BT::MachineEvaluator::eAND(const RegisterCell &A1,
557 const RegisterCell &A2) const {
558 uint16_t W = A1.width();
559 assert(W == A2.width());
560 RegisterCell Res(W);
561 for (uint16_t i = 0; i < W; ++i) {
562 const BitValue &V1 = A1[i];
563 const BitValue &V2 = A2[i];
564 if (V1.is(1))
565 Res[i] = BitValue::ref(V2);
566 else if (V2.is(1))
567 Res[i] = BitValue::ref(V1);
568 else if (V1.is(0) || V2.is(0))
569 Res[i] = BitValue::Zero;
570 else if (V1 == V2)
571 Res[i] = V1;
572 else
573 Res[i] = BitValue::self();
574 }
575 return Res;
576 }
577
578
eORL(const RegisterCell & A1,const RegisterCell & A2) const579 BT::RegisterCell BT::MachineEvaluator::eORL(const RegisterCell &A1,
580 const RegisterCell &A2) const {
581 uint16_t W = A1.width();
582 assert(W == A2.width());
583 RegisterCell Res(W);
584 for (uint16_t i = 0; i < W; ++i) {
585 const BitValue &V1 = A1[i];
586 const BitValue &V2 = A2[i];
587 if (V1.is(1) || V2.is(1))
588 Res[i] = BitValue::One;
589 else if (V1.is(0))
590 Res[i] = BitValue::ref(V2);
591 else if (V2.is(0))
592 Res[i] = BitValue::ref(V1);
593 else if (V1 == V2)
594 Res[i] = V1;
595 else
596 Res[i] = BitValue::self();
597 }
598 return Res;
599 }
600
601
eXOR(const RegisterCell & A1,const RegisterCell & A2) const602 BT::RegisterCell BT::MachineEvaluator::eXOR(const RegisterCell &A1,
603 const RegisterCell &A2) const {
604 uint16_t W = A1.width();
605 assert(W == A2.width());
606 RegisterCell Res(W);
607 for (uint16_t i = 0; i < W; ++i) {
608 const BitValue &V1 = A1[i];
609 const BitValue &V2 = A2[i];
610 if (V1.is(0))
611 Res[i] = BitValue::ref(V2);
612 else if (V2.is(0))
613 Res[i] = BitValue::ref(V1);
614 else if (V1 == V2)
615 Res[i] = BitValue::Zero;
616 else
617 Res[i] = BitValue::self();
618 }
619 return Res;
620 }
621
622
eNOT(const RegisterCell & A1) const623 BT::RegisterCell BT::MachineEvaluator::eNOT(const RegisterCell &A1) const {
624 uint16_t W = A1.width();
625 RegisterCell Res(W);
626 for (uint16_t i = 0; i < W; ++i) {
627 const BitValue &V = A1[i];
628 if (V.is(0))
629 Res[i] = BitValue::One;
630 else if (V.is(1))
631 Res[i] = BitValue::Zero;
632 else
633 Res[i] = BitValue::self();
634 }
635 return Res;
636 }
637
638
eSET(const RegisterCell & A1,uint16_t BitN) const639 BT::RegisterCell BT::MachineEvaluator::eSET(const RegisterCell &A1,
640 uint16_t BitN) const {
641 assert(BitN < A1.width());
642 RegisterCell Res = RegisterCell::ref(A1);
643 Res[BitN] = BitValue::One;
644 return Res;
645 }
646
647
eCLR(const RegisterCell & A1,uint16_t BitN) const648 BT::RegisterCell BT::MachineEvaluator::eCLR(const RegisterCell &A1,
649 uint16_t BitN) const {
650 assert(BitN < A1.width());
651 RegisterCell Res = RegisterCell::ref(A1);
652 Res[BitN] = BitValue::Zero;
653 return Res;
654 }
655
656
eCLB(const RegisterCell & A1,bool B,uint16_t W) const657 BT::RegisterCell BT::MachineEvaluator::eCLB(const RegisterCell &A1, bool B,
658 uint16_t W) const {
659 uint16_t C = A1.cl(B), AW = A1.width();
660 // If the last leading non-B bit is not a constant, then we don't know
661 // the real count.
662 if ((C < AW && A1[AW-1-C].num()) || C == AW)
663 return eIMM(C, W);
664 return RegisterCell::self(0, W);
665 }
666
667
eCTB(const RegisterCell & A1,bool B,uint16_t W) const668 BT::RegisterCell BT::MachineEvaluator::eCTB(const RegisterCell &A1, bool B,
669 uint16_t W) const {
670 uint16_t C = A1.ct(B), AW = A1.width();
671 // If the last trailing non-B bit is not a constant, then we don't know
672 // the real count.
673 if ((C < AW && A1[C].num()) || C == AW)
674 return eIMM(C, W);
675 return RegisterCell::self(0, W);
676 }
677
678
eSXT(const RegisterCell & A1,uint16_t FromN) const679 BT::RegisterCell BT::MachineEvaluator::eSXT(const RegisterCell &A1,
680 uint16_t FromN) const {
681 uint16_t W = A1.width();
682 assert(FromN <= W);
683 RegisterCell Res = RegisterCell::ref(A1);
684 BitValue Sign = Res[FromN-1];
685 // Sign-extend "inreg".
686 Res.fill(FromN, W, Sign);
687 return Res;
688 }
689
690
eZXT(const RegisterCell & A1,uint16_t FromN) const691 BT::RegisterCell BT::MachineEvaluator::eZXT(const RegisterCell &A1,
692 uint16_t FromN) const {
693 uint16_t W = A1.width();
694 assert(FromN <= W);
695 RegisterCell Res = RegisterCell::ref(A1);
696 Res.fill(FromN, W, BitValue::Zero);
697 return Res;
698 }
699
700
eXTR(const RegisterCell & A1,uint16_t B,uint16_t E) const701 BT::RegisterCell BT::MachineEvaluator::eXTR(const RegisterCell &A1,
702 uint16_t B, uint16_t E) const {
703 uint16_t W = A1.width();
704 assert(B < W && E <= W);
705 if (B == E)
706 return RegisterCell(0);
707 uint16_t Last = (E > 0) ? E-1 : W-1;
708 RegisterCell Res = RegisterCell::ref(A1).extract(BT::BitMask(B, Last));
709 // Return shorter cell.
710 return Res;
711 }
712
713
eINS(const RegisterCell & A1,const RegisterCell & A2,uint16_t AtN) const714 BT::RegisterCell BT::MachineEvaluator::eINS(const RegisterCell &A1,
715 const RegisterCell &A2, uint16_t AtN) const {
716 uint16_t W1 = A1.width(), W2 = A2.width();
717 (void)W1;
718 assert(AtN < W1 && AtN+W2 <= W1);
719 // Copy bits from A1, insert A2 at position AtN.
720 RegisterCell Res = RegisterCell::ref(A1);
721 if (W2 > 0)
722 Res.insert(RegisterCell::ref(A2), BT::BitMask(AtN, AtN+W2-1));
723 return Res;
724 }
725
726
mask(unsigned Reg,unsigned Sub) const727 BT::BitMask BT::MachineEvaluator::mask(unsigned Reg, unsigned Sub) const {
728 assert(Sub == 0 && "Generic BitTracker::mask called for Sub != 0");
729 uint16_t W = getRegBitWidth(Reg);
730 assert(W > 0 && "Cannot generate mask for empty register");
731 return BitMask(0, W-1);
732 }
733
734
evaluate(const MachineInstr * MI,const CellMapType & Inputs,CellMapType & Outputs) const735 bool BT::MachineEvaluator::evaluate(const MachineInstr *MI,
736 const CellMapType &Inputs, CellMapType &Outputs) const {
737 unsigned Opc = MI->getOpcode();
738 switch (Opc) {
739 case TargetOpcode::REG_SEQUENCE: {
740 RegisterRef RD = MI->getOperand(0);
741 assert(RD.Sub == 0);
742 RegisterRef RS = MI->getOperand(1);
743 unsigned SS = MI->getOperand(2).getImm();
744 RegisterRef RT = MI->getOperand(3);
745 unsigned ST = MI->getOperand(4).getImm();
746 assert(SS != ST);
747
748 uint16_t W = getRegBitWidth(RD);
749 RegisterCell Res(W);
750 Res.insert(RegisterCell::ref(getCell(RS, Inputs)), mask(RD.Reg, SS));
751 Res.insert(RegisterCell::ref(getCell(RT, Inputs)), mask(RD.Reg, ST));
752 putCell(RD, Res, Outputs);
753 break;
754 }
755
756 case TargetOpcode::COPY: {
757 // COPY can transfer a smaller register into a wider one.
758 // If that is the case, fill the remaining high bits with 0.
759 RegisterRef RD = MI->getOperand(0);
760 RegisterRef RS = MI->getOperand(1);
761 assert(RD.Sub == 0);
762 uint16_t WD = getRegBitWidth(RD);
763 uint16_t WS = getRegBitWidth(RS);
764 assert(WD >= WS);
765 RegisterCell Src = getCell(RS, Inputs);
766 RegisterCell Res(WD);
767 Res.insert(Src, BitMask(0, WS-1));
768 Res.fill(WS, WD, BitValue::Zero);
769 putCell(RD, Res, Outputs);
770 break;
771 }
772
773 default:
774 return false;
775 }
776
777 return true;
778 }
779
780
781 // Main W-Z implementation.
782
visitPHI(const MachineInstr * PI)783 void BT::visitPHI(const MachineInstr *PI) {
784 int ThisN = PI->getParent()->getNumber();
785 if (Trace)
786 dbgs() << "Visit FI(BB#" << ThisN << "): " << *PI;
787
788 const MachineOperand &MD = PI->getOperand(0);
789 assert(MD.getSubReg() == 0 && "Unexpected sub-register in definition");
790 RegisterRef DefRR(MD);
791 uint16_t DefBW = ME.getRegBitWidth(DefRR);
792
793 RegisterCell DefC = ME.getCell(DefRR, Map);
794 if (DefC == RegisterCell::self(DefRR.Reg, DefBW)) // XXX slow
795 return;
796
797 bool Changed = false;
798
799 for (unsigned i = 1, n = PI->getNumOperands(); i < n; i += 2) {
800 const MachineBasicBlock *PB = PI->getOperand(i+1).getMBB();
801 int PredN = PB->getNumber();
802 if (Trace)
803 dbgs() << " edge BB#" << PredN << "->BB#" << ThisN;
804 if (!EdgeExec.count(CFGEdge(PredN, ThisN))) {
805 if (Trace)
806 dbgs() << " not executable\n";
807 continue;
808 }
809
810 RegisterRef RU = PI->getOperand(i);
811 RegisterCell ResC = ME.getCell(RU, Map);
812 if (Trace)
813 dbgs() << " input reg: " << PrintReg(RU.Reg, &ME.TRI, RU.Sub)
814 << " cell: " << ResC << "\n";
815 Changed |= DefC.meet(ResC, DefRR.Reg);
816 }
817
818 if (Changed) {
819 if (Trace)
820 dbgs() << "Output: " << PrintReg(DefRR.Reg, &ME.TRI, DefRR.Sub)
821 << " cell: " << DefC << "\n";
822 ME.putCell(DefRR, DefC, Map);
823 visitUsesOf(DefRR.Reg);
824 }
825 }
826
827
visitNonBranch(const MachineInstr * MI)828 void BT::visitNonBranch(const MachineInstr *MI) {
829 if (Trace) {
830 int ThisN = MI->getParent()->getNumber();
831 dbgs() << "Visit MI(BB#" << ThisN << "): " << *MI;
832 }
833 if (MI->isDebugValue())
834 return;
835 assert(!MI->isBranch() && "Unexpected branch instruction");
836
837 CellMapType ResMap;
838 bool Eval = ME.evaluate(MI, Map, ResMap);
839
840 if (Trace && Eval) {
841 for (unsigned i = 0, n = MI->getNumOperands(); i < n; ++i) {
842 const MachineOperand &MO = MI->getOperand(i);
843 if (!MO.isReg() || !MO.isUse())
844 continue;
845 RegisterRef RU(MO);
846 dbgs() << " input reg: " << PrintReg(RU.Reg, &ME.TRI, RU.Sub)
847 << " cell: " << ME.getCell(RU, Map) << "\n";
848 }
849 dbgs() << "Outputs:\n";
850 for (CellMapType::iterator I = ResMap.begin(), E = ResMap.end();
851 I != E; ++I) {
852 RegisterRef RD(I->first);
853 dbgs() << " " << PrintReg(I->first, &ME.TRI) << " cell: "
854 << ME.getCell(RD, ResMap) << "\n";
855 }
856 }
857
858 // Iterate over all definitions of the instruction, and update the
859 // cells accordingly.
860 for (unsigned i = 0, n = MI->getNumOperands(); i < n; ++i) {
861 const MachineOperand &MO = MI->getOperand(i);
862 // Visit register defs only.
863 if (!MO.isReg() || !MO.isDef())
864 continue;
865 RegisterRef RD(MO);
866 assert(RD.Sub == 0 && "Unexpected sub-register in definition");
867 if (!TargetRegisterInfo::isVirtualRegister(RD.Reg))
868 continue;
869
870 bool Changed = false;
871 if (!Eval || ResMap.count(RD.Reg) == 0) {
872 // Set to "ref" (aka "bottom").
873 uint16_t DefBW = ME.getRegBitWidth(RD);
874 RegisterCell RefC = RegisterCell::self(RD.Reg, DefBW);
875 if (RefC != ME.getCell(RD, Map)) {
876 ME.putCell(RD, RefC, Map);
877 Changed = true;
878 }
879 } else {
880 RegisterCell DefC = ME.getCell(RD, Map);
881 RegisterCell ResC = ME.getCell(RD, ResMap);
882 // This is a non-phi instruction, so the values of the inputs come
883 // from the same registers each time this instruction is evaluated.
884 // During the propagation, the values of the inputs can become lowered
885 // in the sense of the lattice operation, which may cause different
886 // results to be calculated in subsequent evaluations. This should
887 // not cause the bottoming of the result in the map, since the new
888 // result is already reflecting the lowered inputs.
889 for (uint16_t i = 0, w = DefC.width(); i < w; ++i) {
890 BitValue &V = DefC[i];
891 // Bits that are already "bottom" should not be updated.
892 if (V.Type == BitValue::Ref && V.RefI.Reg == RD.Reg)
893 continue;
894 // Same for those that are identical in DefC and ResC.
895 if (V == ResC[i])
896 continue;
897 V = ResC[i];
898 Changed = true;
899 }
900 if (Changed)
901 ME.putCell(RD, DefC, Map);
902 }
903 if (Changed)
904 visitUsesOf(RD.Reg);
905 }
906 }
907
908
visitBranchesFrom(const MachineInstr * BI)909 void BT::visitBranchesFrom(const MachineInstr *BI) {
910 const MachineBasicBlock &B = *BI->getParent();
911 MachineBasicBlock::const_iterator It = BI, End = B.end();
912 BranchTargetList Targets, BTs;
913 bool FallsThrough = true, DefaultToAll = false;
914 int ThisN = B.getNumber();
915
916 do {
917 BTs.clear();
918 const MachineInstr *MI = &*It;
919 if (Trace)
920 dbgs() << "Visit BR(BB#" << ThisN << "): " << *MI;
921 assert(MI->isBranch() && "Expecting branch instruction");
922 InstrExec.insert(MI);
923 bool Eval = ME.evaluate(MI, Map, BTs, FallsThrough);
924 if (!Eval) {
925 // If the evaluation failed, we will add all targets. Keep going in
926 // the loop to mark all executable branches as such.
927 DefaultToAll = true;
928 FallsThrough = true;
929 if (Trace)
930 dbgs() << " failed to evaluate: will add all CFG successors\n";
931 } else if (!DefaultToAll) {
932 // If evaluated successfully add the targets to the cumulative list.
933 if (Trace) {
934 dbgs() << " adding targets:";
935 for (unsigned i = 0, n = BTs.size(); i < n; ++i)
936 dbgs() << " BB#" << BTs[i]->getNumber();
937 if (FallsThrough)
938 dbgs() << "\n falls through\n";
939 else
940 dbgs() << "\n does not fall through\n";
941 }
942 Targets.insert(BTs.begin(), BTs.end());
943 }
944 ++It;
945 } while (FallsThrough && It != End);
946
947 typedef MachineBasicBlock::const_succ_iterator succ_iterator;
948 if (!DefaultToAll) {
949 // Need to add all CFG successors that lead to EH landing pads.
950 // There won't be explicit branches to these blocks, but they must
951 // be processed.
952 for (succ_iterator I = B.succ_begin(), E = B.succ_end(); I != E; ++I) {
953 const MachineBasicBlock *SB = *I;
954 if (SB->isEHPad())
955 Targets.insert(SB);
956 }
957 if (FallsThrough) {
958 MachineFunction::const_iterator BIt = B.getIterator();
959 MachineFunction::const_iterator Next = std::next(BIt);
960 if (Next != MF.end())
961 Targets.insert(&*Next);
962 }
963 } else {
964 for (succ_iterator I = B.succ_begin(), E = B.succ_end(); I != E; ++I)
965 Targets.insert(*I);
966 }
967
968 for (unsigned i = 0, n = Targets.size(); i < n; ++i) {
969 int TargetN = Targets[i]->getNumber();
970 FlowQ.push(CFGEdge(ThisN, TargetN));
971 }
972 }
973
974
visitUsesOf(unsigned Reg)975 void BT::visitUsesOf(unsigned Reg) {
976 if (Trace)
977 dbgs() << "visiting uses of " << PrintReg(Reg, &ME.TRI) << "\n";
978
979 typedef MachineRegisterInfo::use_nodbg_iterator use_iterator;
980 use_iterator End = MRI.use_nodbg_end();
981 for (use_iterator I = MRI.use_nodbg_begin(Reg); I != End; ++I) {
982 MachineInstr *UseI = I->getParent();
983 if (!InstrExec.count(UseI))
984 continue;
985 if (UseI->isPHI())
986 visitPHI(UseI);
987 else if (!UseI->isBranch())
988 visitNonBranch(UseI);
989 else
990 visitBranchesFrom(UseI);
991 }
992 }
993
994
get(RegisterRef RR) const995 BT::RegisterCell BT::get(RegisterRef RR) const {
996 return ME.getCell(RR, Map);
997 }
998
999
put(RegisterRef RR,const RegisterCell & RC)1000 void BT::put(RegisterRef RR, const RegisterCell &RC) {
1001 ME.putCell(RR, RC, Map);
1002 }
1003
1004
1005 // Replace all references to bits from OldRR with the corresponding bits
1006 // in NewRR.
subst(RegisterRef OldRR,RegisterRef NewRR)1007 void BT::subst(RegisterRef OldRR, RegisterRef NewRR) {
1008 assert(Map.count(OldRR.Reg) > 0 && "OldRR not present in map");
1009 BitMask OM = ME.mask(OldRR.Reg, OldRR.Sub);
1010 BitMask NM = ME.mask(NewRR.Reg, NewRR.Sub);
1011 uint16_t OMB = OM.first(), OME = OM.last();
1012 uint16_t NMB = NM.first(), NME = NM.last();
1013 (void)NME;
1014 assert((OME-OMB == NME-NMB) &&
1015 "Substituting registers of different lengths");
1016 for (CellMapType::iterator I = Map.begin(), E = Map.end(); I != E; ++I) {
1017 RegisterCell &RC = I->second;
1018 for (uint16_t i = 0, w = RC.width(); i < w; ++i) {
1019 BitValue &V = RC[i];
1020 if (V.Type != BitValue::Ref || V.RefI.Reg != OldRR.Reg)
1021 continue;
1022 if (V.RefI.Pos < OMB || V.RefI.Pos > OME)
1023 continue;
1024 V.RefI.Reg = NewRR.Reg;
1025 V.RefI.Pos += NMB-OMB;
1026 }
1027 }
1028 }
1029
1030
1031 // Check if the block has been "executed" during propagation. (If not, the
1032 // block is dead, but it may still appear to be reachable.)
reached(const MachineBasicBlock * B) const1033 bool BT::reached(const MachineBasicBlock *B) const {
1034 int BN = B->getNumber();
1035 assert(BN >= 0);
1036 for (EdgeSetType::iterator I = EdgeExec.begin(), E = EdgeExec.end();
1037 I != E; ++I) {
1038 if (I->second == BN)
1039 return true;
1040 }
1041 return false;
1042 }
1043
1044
reset()1045 void BT::reset() {
1046 EdgeExec.clear();
1047 InstrExec.clear();
1048 Map.clear();
1049 }
1050
1051
run()1052 void BT::run() {
1053 reset();
1054 assert(FlowQ.empty());
1055
1056 typedef GraphTraits<const MachineFunction*> MachineFlowGraphTraits;
1057 const MachineBasicBlock *Entry = MachineFlowGraphTraits::getEntryNode(&MF);
1058
1059 unsigned MaxBN = 0;
1060 for (MachineFunction::const_iterator I = MF.begin(), E = MF.end();
1061 I != E; ++I) {
1062 assert(I->getNumber() >= 0 && "Disconnected block");
1063 unsigned BN = I->getNumber();
1064 if (BN > MaxBN)
1065 MaxBN = BN;
1066 }
1067
1068 // Keep track of visited blocks.
1069 BitVector BlockScanned(MaxBN+1);
1070
1071 int EntryN = Entry->getNumber();
1072 // Generate a fake edge to get something to start with.
1073 FlowQ.push(CFGEdge(-1, EntryN));
1074
1075 while (!FlowQ.empty()) {
1076 CFGEdge Edge = FlowQ.front();
1077 FlowQ.pop();
1078
1079 if (EdgeExec.count(Edge))
1080 continue;
1081 EdgeExec.insert(Edge);
1082
1083 const MachineBasicBlock &B = *MF.getBlockNumbered(Edge.second);
1084 MachineBasicBlock::const_iterator It = B.begin(), End = B.end();
1085 // Visit PHI nodes first.
1086 while (It != End && It->isPHI()) {
1087 const MachineInstr *PI = &*It++;
1088 InstrExec.insert(PI);
1089 visitPHI(PI);
1090 }
1091
1092 // If this block has already been visited through a flow graph edge,
1093 // then the instructions have already been processed. Any updates to
1094 // the cells would now only happen through visitUsesOf...
1095 if (BlockScanned[Edge.second])
1096 continue;
1097 BlockScanned[Edge.second] = true;
1098
1099 // Visit non-branch instructions.
1100 while (It != End && !It->isBranch()) {
1101 const MachineInstr *MI = &*It++;
1102 InstrExec.insert(MI);
1103 visitNonBranch(MI);
1104 }
1105 // If block end has been reached, add the fall-through edge to the queue.
1106 if (It == End) {
1107 MachineFunction::const_iterator BIt = B.getIterator();
1108 MachineFunction::const_iterator Next = std::next(BIt);
1109 if (Next != MF.end() && B.isSuccessor(&*Next)) {
1110 int ThisN = B.getNumber();
1111 int NextN = Next->getNumber();
1112 FlowQ.push(CFGEdge(ThisN, NextN));
1113 }
1114 } else {
1115 // Handle the remaining sequence of branches. This function will update
1116 // the work queue.
1117 visitBranchesFrom(It);
1118 }
1119 } // while (!FlowQ->empty())
1120
1121 if (Trace) {
1122 dbgs() << "Cells after propagation:\n";
1123 for (CellMapType::iterator I = Map.begin(), E = Map.end(); I != E; ++I)
1124 dbgs() << PrintReg(I->first, &ME.TRI) << " -> " << I->second << "\n";
1125 }
1126 }
1127
1128