1 //===-- FastISel.cpp - Implementation of the FastISel class ---------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains the implementation of the FastISel class.
12 // "Fast" instruction selection is designed to emit very poor code quickly.
13 // Also, it is not designed to be able to do much lowering, so most illegal
14 // types (e.g. i64 on 32-bit targets) and operations are not supported. It is
15 // also not intended to be able to do much optimization, except in a few cases
16 // where doing optimizations reduces overall compile time. For example, folding
17 // constants into immediate fields is often done, because it's cheap and it
18 // reduces the number of instructions later phases have to examine.
20 // "Fast" instruction selection is able to fail gracefully and transfer
21 // control to the SelectionDAG selector for operations that it doesn't
22 // support. In many cases, this allows us to avoid duplicating a lot of
23 // the complicated lowering logic that SelectionDAG currently has.
25 // The intended use for "fast" instruction selection is "-O0" mode
26 // compilation, where the quality of the generated code is irrelevant when
27 // weighed against the speed at which the code can be generated. Also,
28 // at -O0, the LLVM optimizers are not running, and this makes the
29 // compile time of codegen a much higher portion of the overall compile
30 // time. Despite its limitations, "fast" instruction selection is able to
31 // handle enough code on its own to provide noticeable overall speedups
34 // Basic operations are supported in a target-independent way, by reading
35 // the same instruction descriptions that the SelectionDAG selector reads,
36 // and identifying simple arithmetic operations that can be directly selected
37 // from simple operators. More complicated operations currently require
38 // target-specific code.
40 //===----------------------------------------------------------------------===//
42 #include "llvm/Function.h"
43 #include "llvm/GlobalVariable.h"
44 #include "llvm/Instructions.h"
45 #include "llvm/IntrinsicInst.h"
46 #include "llvm/CodeGen/FastISel.h"
47 #include "llvm/CodeGen/FunctionLoweringInfo.h"
48 #include "llvm/CodeGen/MachineInstrBuilder.h"
49 #include "llvm/CodeGen/MachineModuleInfo.h"
50 #include "llvm/CodeGen/MachineRegisterInfo.h"
51 #include "llvm/Analysis/DebugInfo.h"
52 #include "llvm/Analysis/Loads.h"
53 #include "llvm/Target/TargetData.h"
54 #include "llvm/Target/TargetInstrInfo.h"
55 #include "llvm/Target/TargetLowering.h"
56 #include "llvm/Target/TargetMachine.h"
57 #include "llvm/Support/ErrorHandling.h"
60 bool FastISel::hasTrivialKill(const Value *V) const {
61 // Don't consider constants or arguments to have trivial kills.
62 const Instruction *I = dyn_cast<Instruction>(V);
66 // No-op casts are trivially coalesced by fast-isel.
67 if (const CastInst *Cast = dyn_cast<CastInst>(I))
68 if (Cast->isNoopCast(TD.getIntPtrType(Cast->getContext())) &&
69 !hasTrivialKill(Cast->getOperand(0)))
72 // Only instructions with a single use in the same basic block are considered
73 // to have trivial kills.
74 return I->hasOneUse() &&
75 !(I->getOpcode() == Instruction::BitCast ||
76 I->getOpcode() == Instruction::PtrToInt ||
77 I->getOpcode() == Instruction::IntToPtr) &&
78 cast<Instruction>(I->use_begin())->getParent() == I->getParent();
81 unsigned FastISel::getRegForValue(const Value *V) {
82 EVT RealVT = TLI.getValueType(V->getType(), /*AllowUnknown=*/true);
83 // Don't handle non-simple values in FastISel.
84 if (!RealVT.isSimple())
87 // Ignore illegal types. We must do this before looking up the value
88 // in ValueMap because Arguments are given virtual registers regardless
89 // of whether FastISel can handle them.
90 MVT VT = RealVT.getSimpleVT();
91 if (!TLI.isTypeLegal(VT)) {
92 // Promote MVT::i1 to a legal type though, because it's common and easy.
94 VT = TLI.getTypeToTransformTo(V->getContext(), VT).getSimpleVT();
99 // Look up the value to see if we already have a register for it. We
100 // cache values defined by Instructions across blocks, and other values
101 // only locally. This is because Instructions already have the SSA
102 // def-dominates-use requirement enforced.
103 DenseMap<const Value *, unsigned>::iterator I = FuncInfo.ValueMap.find(V);
104 if (I != FuncInfo.ValueMap.end())
106 unsigned Reg = LocalValueMap[V];
110 // In bottom-up mode, just create the virtual register which will be used
111 // to hold the value. It will be materialized later.
113 Reg = createResultReg(TLI.getRegClassFor(VT));
114 if (isa<Instruction>(V))
115 FuncInfo.ValueMap[V] = Reg;
117 LocalValueMap[V] = Reg;
121 return materializeRegForValue(V, VT);
124 /// materializeRegForValue - Helper for getRegForVale. This function is
125 /// called when the value isn't already available in a register and must
126 /// be materialized with new instructions.
127 unsigned FastISel::materializeRegForValue(const Value *V, MVT VT) {
130 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
131 if (CI->getValue().getActiveBits() <= 64)
132 Reg = FastEmit_i(VT, VT, ISD::Constant, CI->getZExtValue());
133 } else if (isa<AllocaInst>(V)) {
134 Reg = TargetMaterializeAlloca(cast<AllocaInst>(V));
135 } else if (isa<ConstantPointerNull>(V)) {
136 // Translate this as an integer zero so that it can be
137 // local-CSE'd with actual integer zeros.
139 getRegForValue(Constant::getNullValue(TD.getIntPtrType(V->getContext())));
140 } else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
141 // Try to emit the constant directly.
142 Reg = FastEmit_f(VT, VT, ISD::ConstantFP, CF);
145 // Try to emit the constant by using an integer constant with a cast.
146 const APFloat &Flt = CF->getValueAPF();
147 EVT IntVT = TLI.getPointerTy();
150 uint32_t IntBitWidth = IntVT.getSizeInBits();
152 (void) Flt.convertToInteger(x, IntBitWidth, /*isSigned=*/true,
153 APFloat::rmTowardZero, &isExact);
155 APInt IntVal(IntBitWidth, 2, x);
157 unsigned IntegerReg =
158 getRegForValue(ConstantInt::get(V->getContext(), IntVal));
160 Reg = FastEmit_r(IntVT.getSimpleVT(), VT, ISD::SINT_TO_FP,
161 IntegerReg, /*Kill=*/false);
164 } else if (const Operator *Op = dyn_cast<Operator>(V)) {
165 if (!SelectOperator(Op, Op->getOpcode()))
166 if (!isa<Instruction>(Op) ||
167 !TargetSelectInstruction(cast<Instruction>(Op)))
169 Reg = lookUpRegForValue(Op);
170 } else if (isa<UndefValue>(V)) {
171 Reg = createResultReg(TLI.getRegClassFor(VT));
172 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
173 TII.get(TargetOpcode::IMPLICIT_DEF), Reg);
176 // If target-independent code couldn't handle the value, give target-specific
178 if (!Reg && isa<Constant>(V))
179 Reg = TargetMaterializeConstant(cast<Constant>(V));
181 // Don't cache constant materializations in the general ValueMap.
182 // To do so would require tracking what uses they dominate.
184 LocalValueMap[V] = Reg;
188 unsigned FastISel::lookUpRegForValue(const Value *V) {
189 // Look up the value to see if we already have a register for it. We
190 // cache values defined by Instructions across blocks, and other values
191 // only locally. This is because Instructions already have the SSA
192 // def-dominates-use requirement enforced.
193 DenseMap<const Value *, unsigned>::iterator I = FuncInfo.ValueMap.find(V);
194 if (I != FuncInfo.ValueMap.end())
196 return LocalValueMap[V];
199 /// UpdateValueMap - Update the value map to include the new mapping for this
200 /// instruction, or insert an extra copy to get the result in a previous
201 /// determined register.
202 /// NOTE: This is only necessary because we might select a block that uses
203 /// a value before we select the block that defines the value. It might be
204 /// possible to fix this by selecting blocks in reverse postorder.
205 unsigned FastISel::UpdateValueMap(const Value *I, unsigned Reg) {
206 if (!isa<Instruction>(I)) {
207 LocalValueMap[I] = Reg;
211 unsigned &AssignedReg = FuncInfo.ValueMap[I];
212 if (AssignedReg == 0)
214 else if (Reg != AssignedReg) {
215 const TargetRegisterClass *RegClass = MRI.getRegClass(Reg);
216 TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt, AssignedReg,
217 Reg, RegClass, RegClass, DL);
222 std::pair<unsigned, bool> FastISel::getRegForGEPIndex(const Value *Idx) {
223 unsigned IdxN = getRegForValue(Idx);
225 // Unhandled operand. Halt "fast" selection and bail.
226 return std::pair<unsigned, bool>(0, false);
228 bool IdxNIsKill = hasTrivialKill(Idx);
230 // If the index is smaller or larger than intptr_t, truncate or extend it.
231 MVT PtrVT = TLI.getPointerTy();
232 EVT IdxVT = EVT::getEVT(Idx->getType(), /*HandleUnknown=*/false);
233 if (IdxVT.bitsLT(PtrVT)) {
234 IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::SIGN_EXTEND,
238 else if (IdxVT.bitsGT(PtrVT)) {
239 IdxN = FastEmit_r(IdxVT.getSimpleVT(), PtrVT, ISD::TRUNCATE,
243 return std::pair<unsigned, bool>(IdxN, IdxNIsKill);
246 /// SelectBinaryOp - Select and emit code for a binary operator instruction,
247 /// which has an opcode which directly corresponds to the given ISD opcode.
249 bool FastISel::SelectBinaryOp(const User *I, unsigned ISDOpcode) {
250 EVT VT = EVT::getEVT(I->getType(), /*HandleUnknown=*/true);
251 if (VT == MVT::Other || !VT.isSimple())
252 // Unhandled type. Halt "fast" selection and bail.
255 // We only handle legal types. For example, on x86-32 the instruction
256 // selector contains all of the 64-bit instructions from x86-64,
257 // under the assumption that i64 won't be used if the target doesn't
259 if (!TLI.isTypeLegal(VT)) {
260 // MVT::i1 is special. Allow AND, OR, or XOR because they
261 // don't require additional zeroing, which makes them easy.
263 (ISDOpcode == ISD::AND || ISDOpcode == ISD::OR ||
264 ISDOpcode == ISD::XOR))
265 VT = TLI.getTypeToTransformTo(I->getContext(), VT);
270 unsigned Op0 = getRegForValue(I->getOperand(0));
272 // Unhandled operand. Halt "fast" selection and bail.
275 bool Op0IsKill = hasTrivialKill(I->getOperand(0));
277 // Check if the second operand is a constant and handle it appropriately.
278 if (ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(1))) {
279 unsigned ResultReg = FastEmit_ri(VT.getSimpleVT(), VT.getSimpleVT(),
280 ISDOpcode, Op0, Op0IsKill,
282 if (ResultReg != 0) {
283 // We successfully emitted code for the given LLVM Instruction.
284 UpdateValueMap(I, ResultReg);
289 // Check if the second operand is a constant float.
290 if (ConstantFP *CF = dyn_cast<ConstantFP>(I->getOperand(1))) {
291 unsigned ResultReg = FastEmit_rf(VT.getSimpleVT(), VT.getSimpleVT(),
292 ISDOpcode, Op0, Op0IsKill, CF);
293 if (ResultReg != 0) {
294 // We successfully emitted code for the given LLVM Instruction.
295 UpdateValueMap(I, ResultReg);
300 unsigned Op1 = getRegForValue(I->getOperand(1));
302 // Unhandled operand. Halt "fast" selection and bail.
305 bool Op1IsKill = hasTrivialKill(I->getOperand(1));
307 // Now we have both operands in registers. Emit the instruction.
308 unsigned ResultReg = FastEmit_rr(VT.getSimpleVT(), VT.getSimpleVT(),
313 // Target-specific code wasn't able to find a machine opcode for
314 // the given ISD opcode and type. Halt "fast" selection and bail.
317 // We successfully emitted code for the given LLVM Instruction.
318 UpdateValueMap(I, ResultReg);
322 bool FastISel::SelectGetElementPtr(const User *I) {
323 unsigned N = getRegForValue(I->getOperand(0));
325 // Unhandled operand. Halt "fast" selection and bail.
328 bool NIsKill = hasTrivialKill(I->getOperand(0));
330 const Type *Ty = I->getOperand(0)->getType();
331 MVT VT = TLI.getPointerTy();
332 for (GetElementPtrInst::const_op_iterator OI = I->op_begin()+1,
333 E = I->op_end(); OI != E; ++OI) {
334 const Value *Idx = *OI;
335 if (const StructType *StTy = dyn_cast<StructType>(Ty)) {
336 unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
339 uint64_t Offs = TD.getStructLayout(StTy)->getElementOffset(Field);
340 // FIXME: This can be optimized by combining the add with a
342 N = FastEmit_ri_(VT, ISD::ADD, N, NIsKill, Offs, VT);
344 // Unhandled operand. Halt "fast" selection and bail.
348 Ty = StTy->getElementType(Field);
350 Ty = cast<SequentialType>(Ty)->getElementType();
352 // If this is a constant subscript, handle it quickly.
353 if (const ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
354 if (CI->isZero()) continue;
356 TD.getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
357 N = FastEmit_ri_(VT, ISD::ADD, N, NIsKill, Offs, VT);
359 // Unhandled operand. Halt "fast" selection and bail.
365 // N = N + Idx * ElementSize;
366 uint64_t ElementSize = TD.getTypeAllocSize(Ty);
367 std::pair<unsigned, bool> Pair = getRegForGEPIndex(Idx);
368 unsigned IdxN = Pair.first;
369 bool IdxNIsKill = Pair.second;
371 // Unhandled operand. Halt "fast" selection and bail.
374 if (ElementSize != 1) {
375 IdxN = FastEmit_ri_(VT, ISD::MUL, IdxN, IdxNIsKill, ElementSize, VT);
377 // Unhandled operand. Halt "fast" selection and bail.
381 N = FastEmit_rr(VT, VT, ISD::ADD, N, NIsKill, IdxN, IdxNIsKill);
383 // Unhandled operand. Halt "fast" selection and bail.
388 // We successfully emitted code for the given LLVM Instruction.
389 UpdateValueMap(I, N);
393 bool FastISel::SelectCall(const User *I) {
394 const Function *F = cast<CallInst>(I)->getCalledFunction();
395 if (!F) return false;
397 // Handle selected intrinsic function calls.
398 unsigned IID = F->getIntrinsicID();
401 case Intrinsic::dbg_declare: {
402 const DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
403 if (!DIVariable(DI->getVariable()).Verify() ||
404 !FuncInfo.MF->getMMI().hasDebugInfo())
407 const Value *Address = DI->getAddress();
410 if (isa<UndefValue>(Address))
412 const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
413 // Don't handle byval struct arguments or VLAs, for example.
414 // Note that if we have a byval struct argument, fast ISel is turned off;
415 // those are handled in SelectionDAGBuilder.
417 DenseMap<const AllocaInst*, int>::iterator SI =
418 FuncInfo.StaticAllocaMap.find(AI);
419 if (SI == FuncInfo.StaticAllocaMap.end()) break; // VLAs.
421 if (!DI->getDebugLoc().isUnknown())
422 FuncInfo.MF->getMMI().setVariableDbgInfo(DI->getVariable(),
423 FI, DI->getDebugLoc());
425 // Building the map above is target independent. Generating DBG_VALUE
426 // inline is target dependent; do this now.
427 (void)TargetSelectInstruction(cast<Instruction>(I));
430 case Intrinsic::dbg_value: {
431 // This form of DBG_VALUE is target-independent.
432 const DbgValueInst *DI = cast<DbgValueInst>(I);
433 const TargetInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
434 const Value *V = DI->getValue();
436 // Currently the optimizer can produce this; insert an undef to
437 // help debugging. Probably the optimizer should not do this.
438 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
439 .addReg(0U).addImm(DI->getOffset())
440 .addMetadata(DI->getVariable());
441 } else if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
442 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
443 .addImm(CI->getZExtValue()).addImm(DI->getOffset())
444 .addMetadata(DI->getVariable());
445 } else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
446 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
447 .addFPImm(CF).addImm(DI->getOffset())
448 .addMetadata(DI->getVariable());
449 } else if (unsigned Reg = lookUpRegForValue(V)) {
450 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
451 .addReg(Reg, RegState::Debug).addImm(DI->getOffset())
452 .addMetadata(DI->getVariable());
454 // We can't yet handle anything else here because it would require
455 // generating code, thus altering codegen because of debug info.
456 // Insert an undef so we can see what we dropped.
457 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
458 .addReg(0U).addImm(DI->getOffset())
459 .addMetadata(DI->getVariable());
463 case Intrinsic::eh_exception: {
464 EVT VT = TLI.getValueType(I->getType());
465 switch (TLI.getOperationAction(ISD::EXCEPTIONADDR, VT)) {
467 case TargetLowering::Expand: {
468 assert(FuncInfo.MBB->isLandingPad() &&
469 "Call to eh.exception not in landing pad!");
470 unsigned Reg = TLI.getExceptionAddressRegister();
471 const TargetRegisterClass *RC = TLI.getRegClassFor(VT);
472 unsigned ResultReg = createResultReg(RC);
473 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
474 ResultReg, Reg, RC, RC, DL);
475 assert(InsertedCopy && "Can't copy address registers!");
476 InsertedCopy = InsertedCopy;
477 UpdateValueMap(I, ResultReg);
483 case Intrinsic::eh_selector: {
484 EVT VT = TLI.getValueType(I->getType());
485 switch (TLI.getOperationAction(ISD::EHSELECTION, VT)) {
487 case TargetLowering::Expand: {
488 if (FuncInfo.MBB->isLandingPad())
489 AddCatchInfo(*cast<CallInst>(I), &FuncInfo.MF->getMMI(), FuncInfo.MBB);
492 FuncInfo.CatchInfoLost.insert(cast<CallInst>(I));
494 // FIXME: Mark exception selector register as live in. Hack for PR1508.
495 unsigned Reg = TLI.getExceptionSelectorRegister();
496 if (Reg) FuncInfo.MBB->addLiveIn(Reg);
499 unsigned Reg = TLI.getExceptionSelectorRegister();
500 EVT SrcVT = TLI.getPointerTy();
501 const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
502 unsigned ResultReg = createResultReg(RC);
503 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
504 ResultReg, Reg, RC, RC, DL);
505 assert(InsertedCopy && "Can't copy address registers!");
506 InsertedCopy = InsertedCopy;
508 bool ResultRegIsKill = hasTrivialKill(I);
510 // Cast the register to the type of the selector.
511 if (SrcVT.bitsGT(MVT::i32))
512 ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
513 ResultReg, ResultRegIsKill);
514 else if (SrcVT.bitsLT(MVT::i32))
515 ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
516 ISD::SIGN_EXTEND, ResultReg, ResultRegIsKill);
518 // Unhandled operand. Halt "fast" selection and bail.
521 UpdateValueMap(I, ResultReg);
530 // An arbitrary call. Bail.
534 bool FastISel::SelectCast(const User *I, unsigned Opcode) {
535 EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
536 EVT DstVT = TLI.getValueType(I->getType());
538 if (SrcVT == MVT::Other || !SrcVT.isSimple() ||
539 DstVT == MVT::Other || !DstVT.isSimple())
540 // Unhandled type. Halt "fast" selection and bail.
543 // Check if the destination type is legal. Or as a special case,
544 // it may be i1 if we're doing a truncate because that's
545 // easy and somewhat common.
546 if (!TLI.isTypeLegal(DstVT))
547 if (DstVT != MVT::i1 || Opcode != ISD::TRUNCATE)
548 // Unhandled type. Halt "fast" selection and bail.
551 // Check if the source operand is legal. Or as a special case,
552 // it may be i1 if we're doing zero-extension because that's
553 // easy and somewhat common.
554 if (!TLI.isTypeLegal(SrcVT))
555 if (SrcVT != MVT::i1 || Opcode != ISD::ZERO_EXTEND)
556 // Unhandled type. Halt "fast" selection and bail.
559 unsigned InputReg = getRegForValue(I->getOperand(0));
561 // Unhandled operand. Halt "fast" selection and bail.
564 bool InputRegIsKill = hasTrivialKill(I->getOperand(0));
566 // If the operand is i1, arrange for the high bits in the register to be zero.
567 if (SrcVT == MVT::i1) {
568 SrcVT = TLI.getTypeToTransformTo(I->getContext(), SrcVT);
569 InputReg = FastEmitZExtFromI1(SrcVT.getSimpleVT(), InputReg, InputRegIsKill);
572 InputRegIsKill = true;
574 // If the result is i1, truncate to the target's type for i1 first.
575 if (DstVT == MVT::i1)
576 DstVT = TLI.getTypeToTransformTo(I->getContext(), DstVT);
578 unsigned ResultReg = FastEmit_r(SrcVT.getSimpleVT(),
581 InputReg, InputRegIsKill);
585 UpdateValueMap(I, ResultReg);
589 bool FastISel::SelectBitCast(const User *I) {
590 // If the bitcast doesn't change the type, just use the operand value.
591 if (I->getType() == I->getOperand(0)->getType()) {
592 unsigned Reg = getRegForValue(I->getOperand(0));
595 UpdateValueMap(I, Reg);
599 // Bitcasts of other values become reg-reg copies or BIT_CONVERT operators.
600 EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
601 EVT DstVT = TLI.getValueType(I->getType());
603 if (SrcVT == MVT::Other || !SrcVT.isSimple() ||
604 DstVT == MVT::Other || !DstVT.isSimple() ||
605 !TLI.isTypeLegal(SrcVT) || !TLI.isTypeLegal(DstVT))
606 // Unhandled type. Halt "fast" selection and bail.
609 unsigned Op0 = getRegForValue(I->getOperand(0));
611 // Unhandled operand. Halt "fast" selection and bail.
614 bool Op0IsKill = hasTrivialKill(I->getOperand(0));
616 // First, try to perform the bitcast by inserting a reg-reg copy.
617 unsigned ResultReg = 0;
618 if (SrcVT.getSimpleVT() == DstVT.getSimpleVT()) {
619 TargetRegisterClass* SrcClass = TLI.getRegClassFor(SrcVT);
620 TargetRegisterClass* DstClass = TLI.getRegClassFor(DstVT);
621 ResultReg = createResultReg(DstClass);
623 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
625 DstClass, SrcClass, DL);
630 // If the reg-reg copy failed, select a BIT_CONVERT opcode.
632 ResultReg = FastEmit_r(SrcVT.getSimpleVT(), DstVT.getSimpleVT(),
633 ISD::BIT_CONVERT, Op0, Op0IsKill);
638 UpdateValueMap(I, ResultReg);
643 FastISel::SelectInstruction(const Instruction *I) {
644 // Just before the terminator instruction, insert instructions to
645 // feed PHI nodes in successor blocks.
646 if (isa<TerminatorInst>(I))
647 if (!HandlePHINodesInSuccessorBlocks(I->getParent()))
650 DL = I->getDebugLoc();
652 // First, try doing target-independent selection.
653 if (SelectOperator(I, I->getOpcode())) {
658 // Next, try calling the target to attempt to handle the instruction.
659 if (TargetSelectInstruction(I)) {
668 /// FastEmitBranch - Emit an unconditional branch to the given block,
669 /// unless it is the immediate (fall-through) successor, and update
672 FastISel::FastEmitBranch(MachineBasicBlock *MSucc, DebugLoc DL) {
673 if (FuncInfo.MBB->isLayoutSuccessor(MSucc)) {
674 // The unconditional fall-through case, which needs no instructions.
676 // The unconditional branch case.
677 TII.InsertBranch(*FuncInfo.MBB, MSucc, NULL,
678 SmallVector<MachineOperand, 0>(), DL);
680 FuncInfo.MBB->addSuccessor(MSucc);
683 /// SelectFNeg - Emit an FNeg operation.
686 FastISel::SelectFNeg(const User *I) {
687 unsigned OpReg = getRegForValue(BinaryOperator::getFNegArgument(I));
688 if (OpReg == 0) return false;
690 bool OpRegIsKill = hasTrivialKill(I);
692 // If the target has ISD::FNEG, use it.
693 EVT VT = TLI.getValueType(I->getType());
694 unsigned ResultReg = FastEmit_r(VT.getSimpleVT(), VT.getSimpleVT(),
695 ISD::FNEG, OpReg, OpRegIsKill);
696 if (ResultReg != 0) {
697 UpdateValueMap(I, ResultReg);
701 // Bitcast the value to integer, twiddle the sign bit with xor,
702 // and then bitcast it back to floating-point.
703 if (VT.getSizeInBits() > 64) return false;
704 EVT IntVT = EVT::getIntegerVT(I->getContext(), VT.getSizeInBits());
705 if (!TLI.isTypeLegal(IntVT))
708 unsigned IntReg = FastEmit_r(VT.getSimpleVT(), IntVT.getSimpleVT(),
709 ISD::BIT_CONVERT, OpReg, OpRegIsKill);
713 unsigned IntResultReg = FastEmit_ri_(IntVT.getSimpleVT(), ISD::XOR,
714 IntReg, /*Kill=*/true,
715 UINT64_C(1) << (VT.getSizeInBits()-1),
716 IntVT.getSimpleVT());
717 if (IntResultReg == 0)
720 ResultReg = FastEmit_r(IntVT.getSimpleVT(), VT.getSimpleVT(),
721 ISD::BIT_CONVERT, IntResultReg, /*Kill=*/true);
725 UpdateValueMap(I, ResultReg);
730 FastISel::SelectLoad(const User *I) {
731 LoadInst *LI = const_cast<LoadInst *>(cast<LoadInst>(I));
733 // For a load from an alloca, make a limited effort to find the value
734 // already available in a register, avoiding redundant loads.
735 if (!LI->isVolatile() && isa<AllocaInst>(LI->getPointerOperand())) {
736 BasicBlock::iterator ScanFrom = LI;
737 if (const Value *V = FindAvailableLoadedValue(LI->getPointerOperand(),
738 LI->getParent(), ScanFrom)) {
739 unsigned ResultReg = getRegForValue(V);
740 if (ResultReg != 0) {
741 UpdateValueMap(I, ResultReg);
751 FastISel::SelectOperator(const User *I, unsigned Opcode) {
753 case Instruction::Load:
754 return SelectLoad(I);
755 case Instruction::Add:
756 return SelectBinaryOp(I, ISD::ADD);
757 case Instruction::FAdd:
758 return SelectBinaryOp(I, ISD::FADD);
759 case Instruction::Sub:
760 return SelectBinaryOp(I, ISD::SUB);
761 case Instruction::FSub:
762 // FNeg is currently represented in LLVM IR as a special case of FSub.
763 if (BinaryOperator::isFNeg(I))
764 return SelectFNeg(I);
765 return SelectBinaryOp(I, ISD::FSUB);
766 case Instruction::Mul:
767 return SelectBinaryOp(I, ISD::MUL);
768 case Instruction::FMul:
769 return SelectBinaryOp(I, ISD::FMUL);
770 case Instruction::SDiv:
771 return SelectBinaryOp(I, ISD::SDIV);
772 case Instruction::UDiv:
773 return SelectBinaryOp(I, ISD::UDIV);
774 case Instruction::FDiv:
775 return SelectBinaryOp(I, ISD::FDIV);
776 case Instruction::SRem:
777 return SelectBinaryOp(I, ISD::SREM);
778 case Instruction::URem:
779 return SelectBinaryOp(I, ISD::UREM);
780 case Instruction::FRem:
781 return SelectBinaryOp(I, ISD::FREM);
782 case Instruction::Shl:
783 return SelectBinaryOp(I, ISD::SHL);
784 case Instruction::LShr:
785 return SelectBinaryOp(I, ISD::SRL);
786 case Instruction::AShr:
787 return SelectBinaryOp(I, ISD::SRA);
788 case Instruction::And:
789 return SelectBinaryOp(I, ISD::AND);
790 case Instruction::Or:
791 return SelectBinaryOp(I, ISD::OR);
792 case Instruction::Xor:
793 return SelectBinaryOp(I, ISD::XOR);
795 case Instruction::GetElementPtr:
796 return SelectGetElementPtr(I);
798 case Instruction::Br: {
799 const BranchInst *BI = cast<BranchInst>(I);
801 if (BI->isUnconditional()) {
802 const BasicBlock *LLVMSucc = BI->getSuccessor(0);
803 MachineBasicBlock *MSucc = FuncInfo.MBBMap[LLVMSucc];
804 FastEmitBranch(MSucc, BI->getDebugLoc());
808 // Conditional branches are not handed yet.
809 // Halt "fast" selection and bail.
813 case Instruction::Unreachable:
817 case Instruction::Alloca:
818 // FunctionLowering has the static-sized case covered.
819 if (FuncInfo.StaticAllocaMap.count(cast<AllocaInst>(I)))
822 // Dynamic-sized alloca is not handled yet.
825 case Instruction::Call:
826 return SelectCall(I);
828 case Instruction::BitCast:
829 return SelectBitCast(I);
831 case Instruction::FPToSI:
832 return SelectCast(I, ISD::FP_TO_SINT);
833 case Instruction::ZExt:
834 return SelectCast(I, ISD::ZERO_EXTEND);
835 case Instruction::SExt:
836 return SelectCast(I, ISD::SIGN_EXTEND);
837 case Instruction::Trunc:
838 return SelectCast(I, ISD::TRUNCATE);
839 case Instruction::SIToFP:
840 return SelectCast(I, ISD::SINT_TO_FP);
842 case Instruction::IntToPtr: // Deliberate fall-through.
843 case Instruction::PtrToInt: {
844 EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
845 EVT DstVT = TLI.getValueType(I->getType());
846 if (DstVT.bitsGT(SrcVT))
847 return SelectCast(I, ISD::ZERO_EXTEND);
848 if (DstVT.bitsLT(SrcVT))
849 return SelectCast(I, ISD::TRUNCATE);
850 unsigned Reg = getRegForValue(I->getOperand(0));
851 if (Reg == 0) return false;
852 UpdateValueMap(I, Reg);
856 case Instruction::PHI:
857 llvm_unreachable("FastISel shouldn't visit PHI nodes!");
860 // Unhandled instruction. Halt "fast" selection and bail.
865 FastISel::FastISel(FunctionLoweringInfo &funcInfo)
866 : FuncInfo(funcInfo),
867 MRI(FuncInfo.MF->getRegInfo()),
868 MFI(*FuncInfo.MF->getFrameInfo()),
869 MCP(*FuncInfo.MF->getConstantPool()),
870 TM(FuncInfo.MF->getTarget()),
871 TD(*TM.getTargetData()),
872 TII(*TM.getInstrInfo()),
873 TLI(*TM.getTargetLowering()),
874 TRI(*TM.getRegisterInfo()),
878 FastISel::~FastISel() {}
880 unsigned FastISel::FastEmit_(MVT, MVT,
885 unsigned FastISel::FastEmit_r(MVT, MVT,
887 unsigned /*Op0*/, bool /*Op0IsKill*/) {
891 unsigned FastISel::FastEmit_rr(MVT, MVT,
893 unsigned /*Op0*/, bool /*Op0IsKill*/,
894 unsigned /*Op1*/, bool /*Op1IsKill*/) {
898 unsigned FastISel::FastEmit_i(MVT, MVT, unsigned, uint64_t /*Imm*/) {
902 unsigned FastISel::FastEmit_f(MVT, MVT,
903 unsigned, const ConstantFP * /*FPImm*/) {
907 unsigned FastISel::FastEmit_ri(MVT, MVT,
909 unsigned /*Op0*/, bool /*Op0IsKill*/,
914 unsigned FastISel::FastEmit_rf(MVT, MVT,
916 unsigned /*Op0*/, bool /*Op0IsKill*/,
917 const ConstantFP * /*FPImm*/) {
921 unsigned FastISel::FastEmit_rri(MVT, MVT,
923 unsigned /*Op0*/, bool /*Op0IsKill*/,
924 unsigned /*Op1*/, bool /*Op1IsKill*/,
929 /// FastEmit_ri_ - This method is a wrapper of FastEmit_ri. It first tries
930 /// to emit an instruction with an immediate operand using FastEmit_ri.
931 /// If that fails, it materializes the immediate into a register and try
932 /// FastEmit_rr instead.
933 unsigned FastISel::FastEmit_ri_(MVT VT, unsigned Opcode,
934 unsigned Op0, bool Op0IsKill,
935 uint64_t Imm, MVT ImmType) {
936 // First check if immediate type is legal. If not, we can't use the ri form.
937 unsigned ResultReg = FastEmit_ri(VT, VT, Opcode, Op0, Op0IsKill, Imm);
940 unsigned MaterialReg = FastEmit_i(ImmType, ImmType, ISD::Constant, Imm);
941 if (MaterialReg == 0)
943 return FastEmit_rr(VT, VT, Opcode,
945 MaterialReg, /*Kill=*/true);
948 /// FastEmit_rf_ - This method is a wrapper of FastEmit_ri. It first tries
949 /// to emit an instruction with a floating-point immediate operand using
950 /// FastEmit_rf. If that fails, it materializes the immediate into a register
951 /// and try FastEmit_rr instead.
952 unsigned FastISel::FastEmit_rf_(MVT VT, unsigned Opcode,
953 unsigned Op0, bool Op0IsKill,
954 const ConstantFP *FPImm, MVT ImmType) {
955 // First check if immediate type is legal. If not, we can't use the rf form.
956 unsigned ResultReg = FastEmit_rf(VT, VT, Opcode, Op0, Op0IsKill, FPImm);
960 // Materialize the constant in a register.
961 unsigned MaterialReg = FastEmit_f(ImmType, ImmType, ISD::ConstantFP, FPImm);
962 if (MaterialReg == 0) {
963 // If the target doesn't have a way to directly enter a floating-point
964 // value into a register, use an alternate approach.
965 // TODO: The current approach only supports floating-point constants
966 // that can be constructed by conversion from integer values. This should
967 // be replaced by code that creates a load from a constant-pool entry,
968 // which will require some target-specific work.
969 const APFloat &Flt = FPImm->getValueAPF();
970 EVT IntVT = TLI.getPointerTy();
973 uint32_t IntBitWidth = IntVT.getSizeInBits();
975 (void) Flt.convertToInteger(x, IntBitWidth, /*isSigned=*/true,
976 APFloat::rmTowardZero, &isExact);
979 APInt IntVal(IntBitWidth, 2, x);
981 unsigned IntegerReg = FastEmit_i(IntVT.getSimpleVT(), IntVT.getSimpleVT(),
982 ISD::Constant, IntVal.getZExtValue());
985 MaterialReg = FastEmit_r(IntVT.getSimpleVT(), VT,
986 ISD::SINT_TO_FP, IntegerReg, /*Kill=*/true);
987 if (MaterialReg == 0)
990 return FastEmit_rr(VT, VT, Opcode,
992 MaterialReg, /*Kill=*/true);
995 unsigned FastISel::createResultReg(const TargetRegisterClass* RC) {
996 return MRI.createVirtualRegister(RC);
999 unsigned FastISel::FastEmitInst_(unsigned MachineInstOpcode,
1000 const TargetRegisterClass* RC) {
1001 unsigned ResultReg = createResultReg(RC);
1002 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1004 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg);
1008 unsigned FastISel::FastEmitInst_r(unsigned MachineInstOpcode,
1009 const TargetRegisterClass *RC,
1010 unsigned Op0, bool Op0IsKill) {
1011 unsigned ResultReg = createResultReg(RC);
1012 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1014 if (II.getNumDefs() >= 1)
1015 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1016 .addReg(Op0, Op0IsKill * RegState::Kill);
1018 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1019 .addReg(Op0, Op0IsKill * RegState::Kill);
1020 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1021 ResultReg, II.ImplicitDefs[0],
1030 unsigned FastISel::FastEmitInst_rr(unsigned MachineInstOpcode,
1031 const TargetRegisterClass *RC,
1032 unsigned Op0, bool Op0IsKill,
1033 unsigned Op1, bool Op1IsKill) {
1034 unsigned ResultReg = createResultReg(RC);
1035 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1037 if (II.getNumDefs() >= 1)
1038 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1039 .addReg(Op0, Op0IsKill * RegState::Kill)
1040 .addReg(Op1, Op1IsKill * RegState::Kill);
1042 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1043 .addReg(Op0, Op0IsKill * RegState::Kill)
1044 .addReg(Op1, Op1IsKill * RegState::Kill);
1045 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1046 ResultReg, II.ImplicitDefs[0],
1054 unsigned FastISel::FastEmitInst_ri(unsigned MachineInstOpcode,
1055 const TargetRegisterClass *RC,
1056 unsigned Op0, bool Op0IsKill,
1058 unsigned ResultReg = createResultReg(RC);
1059 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1061 if (II.getNumDefs() >= 1)
1062 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1063 .addReg(Op0, Op0IsKill * RegState::Kill)
1066 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1067 .addReg(Op0, Op0IsKill * RegState::Kill)
1069 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1070 ResultReg, II.ImplicitDefs[0],
1078 unsigned FastISel::FastEmitInst_rf(unsigned MachineInstOpcode,
1079 const TargetRegisterClass *RC,
1080 unsigned Op0, bool Op0IsKill,
1081 const ConstantFP *FPImm) {
1082 unsigned ResultReg = createResultReg(RC);
1083 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1085 if (II.getNumDefs() >= 1)
1086 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1087 .addReg(Op0, Op0IsKill * RegState::Kill)
1090 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1091 .addReg(Op0, Op0IsKill * RegState::Kill)
1093 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1094 ResultReg, II.ImplicitDefs[0],
1102 unsigned FastISel::FastEmitInst_rri(unsigned MachineInstOpcode,
1103 const TargetRegisterClass *RC,
1104 unsigned Op0, bool Op0IsKill,
1105 unsigned Op1, bool Op1IsKill,
1107 unsigned ResultReg = createResultReg(RC);
1108 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1110 if (II.getNumDefs() >= 1)
1111 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1112 .addReg(Op0, Op0IsKill * RegState::Kill)
1113 .addReg(Op1, Op1IsKill * RegState::Kill)
1116 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1117 .addReg(Op0, Op0IsKill * RegState::Kill)
1118 .addReg(Op1, Op1IsKill * RegState::Kill)
1120 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1121 ResultReg, II.ImplicitDefs[0],
1129 unsigned FastISel::FastEmitInst_i(unsigned MachineInstOpcode,
1130 const TargetRegisterClass *RC,
1132 unsigned ResultReg = createResultReg(RC);
1133 const TargetInstrDesc &II = TII.get(MachineInstOpcode);
1135 if (II.getNumDefs() >= 1)
1136 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg).addImm(Imm);
1138 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II).addImm(Imm);
1139 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1140 ResultReg, II.ImplicitDefs[0],
1148 unsigned FastISel::FastEmitInst_extractsubreg(MVT RetVT,
1149 unsigned Op0, bool Op0IsKill,
1151 const TargetRegisterClass* RC = MRI.getRegClass(Op0);
1153 unsigned ResultReg = createResultReg(TLI.getRegClassFor(RetVT));
1154 const TargetInstrDesc &II = TII.get(TargetOpcode::EXTRACT_SUBREG);
1156 if (II.getNumDefs() >= 1)
1157 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
1158 .addReg(Op0, Op0IsKill * RegState::Kill)
1161 BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
1162 .addReg(Op0, Op0IsKill * RegState::Kill)
1164 bool InsertedCopy = TII.copyRegToReg(*FuncInfo.MBB, FuncInfo.InsertPt,
1165 ResultReg, II.ImplicitDefs[0],
1173 /// FastEmitZExtFromI1 - Emit MachineInstrs to compute the value of Op
1174 /// with all but the least significant bit set to zero.
1175 unsigned FastISel::FastEmitZExtFromI1(MVT VT, unsigned Op0, bool Op0IsKill) {
1176 return FastEmit_ri(VT, VT, ISD::AND, Op0, Op0IsKill, 1);
1179 /// HandlePHINodesInSuccessorBlocks - Handle PHI nodes in successor blocks.
1180 /// Emit code to ensure constants are copied into registers when needed.
1181 /// Remember the virtual registers that need to be added to the Machine PHI
1182 /// nodes as input. We cannot just directly add them, because expansion
1183 /// might result in multiple MBB's for one BB. As such, the start of the
1184 /// BB might correspond to a different MBB than the end.
1185 bool FastISel::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) {
1186 const TerminatorInst *TI = LLVMBB->getTerminator();
1188 SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
1189 unsigned OrigNumPHINodesToUpdate = FuncInfo.PHINodesToUpdate.size();
1191 // Check successor nodes' PHI nodes that expect a constant to be available
1193 for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
1194 const BasicBlock *SuccBB = TI->getSuccessor(succ);
1195 if (!isa<PHINode>(SuccBB->begin())) continue;
1196 MachineBasicBlock *SuccMBB = FuncInfo.MBBMap[SuccBB];
1198 // If this terminator has multiple identical successors (common for
1199 // switches), only handle each succ once.
1200 if (!SuccsHandled.insert(SuccMBB)) continue;
1202 MachineBasicBlock::iterator MBBI = SuccMBB->begin();
1204 // At this point we know that there is a 1-1 correspondence between LLVM PHI
1205 // nodes and Machine PHI nodes, but the incoming operands have not been
1207 for (BasicBlock::const_iterator I = SuccBB->begin();
1208 const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1210 // Ignore dead phi's.
1211 if (PN->use_empty()) continue;
1213 // Only handle legal types. Two interesting things to note here. First,
1214 // by bailing out early, we may leave behind some dead instructions,
1215 // since SelectionDAG's HandlePHINodesInSuccessorBlocks will insert its
1216 // own moves. Second, this check is necessary becuase FastISel doesn't
1217 // use CreateRegs to create registers, so it always creates
1218 // exactly one register for each non-void instruction.
1219 EVT VT = TLI.getValueType(PN->getType(), /*AllowUnknown=*/true);
1220 if (VT == MVT::Other || !TLI.isTypeLegal(VT)) {
1223 VT = TLI.getTypeToTransformTo(LLVMBB->getContext(), VT);
1225 FuncInfo.PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
1230 const Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
1232 // Set the DebugLoc for the copy. Prefer the location of the operand
1233 // if there is one; use the location of the PHI otherwise.
1234 DL = PN->getDebugLoc();
1235 if (const Instruction *Inst = dyn_cast<Instruction>(PHIOp))
1236 DL = Inst->getDebugLoc();
1238 unsigned Reg = getRegForValue(PHIOp);
1240 FuncInfo.PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
1243 FuncInfo.PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg));