1 //===-- SelectionDAG.cpp - Implement the SelectionDAG data structures -----===//
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 implements the SelectionDAG class.
12 //===----------------------------------------------------------------------===//
13 #include "llvm/CodeGen/SelectionDAG.h"
14 #include "llvm/Constants.h"
15 #include "llvm/Analysis/ValueTracking.h"
16 #include "llvm/GlobalAlias.h"
17 #include "llvm/GlobalVariable.h"
18 #include "llvm/Intrinsics.h"
19 #include "llvm/DerivedTypes.h"
20 #include "llvm/Assembly/Writer.h"
21 #include "llvm/CallingConv.h"
22 #include "llvm/CodeGen/MachineBasicBlock.h"
23 #include "llvm/CodeGen/MachineConstantPool.h"
24 #include "llvm/CodeGen/MachineFrameInfo.h"
25 #include "llvm/CodeGen/MachineModuleInfo.h"
26 #include "llvm/CodeGen/PseudoSourceValue.h"
27 #include "llvm/Target/TargetRegisterInfo.h"
28 #include "llvm/Target/TargetData.h"
29 #include "llvm/Target/TargetLowering.h"
30 #include "llvm/Target/TargetInstrInfo.h"
31 #include "llvm/Target/TargetMachine.h"
32 #include "llvm/Support/MathExtras.h"
33 #include "llvm/Support/raw_ostream.h"
34 #include "llvm/ADT/SetVector.h"
35 #include "llvm/ADT/SmallPtrSet.h"
36 #include "llvm/ADT/SmallSet.h"
37 #include "llvm/ADT/SmallVector.h"
38 #include "llvm/ADT/StringExtras.h"
43 /// makeVTList - Return an instance of the SDVTList struct initialized with the
44 /// specified members.
45 static SDVTList makeVTList(const MVT *VTs, unsigned NumVTs) {
46 SDVTList Res = {VTs, NumVTs};
50 static const fltSemantics *MVTToAPFloatSemantics(MVT VT) {
51 switch (VT.getSimpleVT()) {
52 default: assert(0 && "Unknown FP format");
53 case MVT::f32: return &APFloat::IEEEsingle;
54 case MVT::f64: return &APFloat::IEEEdouble;
55 case MVT::f80: return &APFloat::x87DoubleExtended;
56 case MVT::f128: return &APFloat::IEEEquad;
57 case MVT::ppcf128: return &APFloat::PPCDoubleDouble;
61 SelectionDAG::DAGUpdateListener::~DAGUpdateListener() {}
63 //===----------------------------------------------------------------------===//
64 // ConstantFPSDNode Class
65 //===----------------------------------------------------------------------===//
67 /// isExactlyValue - We don't rely on operator== working on double values, as
68 /// it returns true for things that are clearly not equal, like -0.0 and 0.0.
69 /// As such, this method can be used to do an exact bit-for-bit comparison of
70 /// two floating point values.
71 bool ConstantFPSDNode::isExactlyValue(const APFloat& V) const {
72 return Value.bitwiseIsEqual(V);
75 bool ConstantFPSDNode::isValueValidForType(MVT VT,
77 assert(VT.isFloatingPoint() && "Can only convert between FP types");
79 // PPC long double cannot be converted to any other type.
80 if (VT == MVT::ppcf128 ||
81 &Val.getSemantics() == &APFloat::PPCDoubleDouble)
84 // convert modifies in place, so make a copy.
85 APFloat Val2 = APFloat(Val);
86 return Val2.convert(*MVTToAPFloatSemantics(VT),
87 APFloat::rmNearestTiesToEven) == APFloat::opOK;
90 //===----------------------------------------------------------------------===//
92 //===----------------------------------------------------------------------===//
94 /// isBuildVectorAllOnes - Return true if the specified node is a
95 /// BUILD_VECTOR where all of the elements are ~0 or undef.
96 bool ISD::isBuildVectorAllOnes(const SDNode *N) {
97 // Look through a bit convert.
98 if (N->getOpcode() == ISD::BIT_CONVERT)
99 N = N->getOperand(0).Val;
101 if (N->getOpcode() != ISD::BUILD_VECTOR) return false;
103 unsigned i = 0, e = N->getNumOperands();
105 // Skip over all of the undef values.
106 while (i != e && N->getOperand(i).getOpcode() == ISD::UNDEF)
109 // Do not accept an all-undef vector.
110 if (i == e) return false;
112 // Do not accept build_vectors that aren't all constants or which have non-~0
114 SDValue NotZero = N->getOperand(i);
115 if (isa<ConstantSDNode>(NotZero)) {
116 if (!cast<ConstantSDNode>(NotZero)->isAllOnesValue())
118 } else if (isa<ConstantFPSDNode>(NotZero)) {
119 if (!cast<ConstantFPSDNode>(NotZero)->getValueAPF().
120 convertToAPInt().isAllOnesValue())
125 // Okay, we have at least one ~0 value, check to see if the rest match or are
127 for (++i; i != e; ++i)
128 if (N->getOperand(i) != NotZero &&
129 N->getOperand(i).getOpcode() != ISD::UNDEF)
135 /// isBuildVectorAllZeros - Return true if the specified node is a
136 /// BUILD_VECTOR where all of the elements are 0 or undef.
137 bool ISD::isBuildVectorAllZeros(const SDNode *N) {
138 // Look through a bit convert.
139 if (N->getOpcode() == ISD::BIT_CONVERT)
140 N = N->getOperand(0).Val;
142 if (N->getOpcode() != ISD::BUILD_VECTOR) return false;
144 unsigned i = 0, e = N->getNumOperands();
146 // Skip over all of the undef values.
147 while (i != e && N->getOperand(i).getOpcode() == ISD::UNDEF)
150 // Do not accept an all-undef vector.
151 if (i == e) return false;
153 // Do not accept build_vectors that aren't all constants or which have non-~0
155 SDValue Zero = N->getOperand(i);
156 if (isa<ConstantSDNode>(Zero)) {
157 if (!cast<ConstantSDNode>(Zero)->isNullValue())
159 } else if (isa<ConstantFPSDNode>(Zero)) {
160 if (!cast<ConstantFPSDNode>(Zero)->getValueAPF().isPosZero())
165 // Okay, we have at least one ~0 value, check to see if the rest match or are
167 for (++i; i != e; ++i)
168 if (N->getOperand(i) != Zero &&
169 N->getOperand(i).getOpcode() != ISD::UNDEF)
174 /// isScalarToVector - Return true if the specified node is a
175 /// ISD::SCALAR_TO_VECTOR node or a BUILD_VECTOR node where only the low
176 /// element is not an undef.
177 bool ISD::isScalarToVector(const SDNode *N) {
178 if (N->getOpcode() == ISD::SCALAR_TO_VECTOR)
181 if (N->getOpcode() != ISD::BUILD_VECTOR)
183 if (N->getOperand(0).getOpcode() == ISD::UNDEF)
185 unsigned NumElems = N->getNumOperands();
186 for (unsigned i = 1; i < NumElems; ++i) {
187 SDValue V = N->getOperand(i);
188 if (V.getOpcode() != ISD::UNDEF)
195 /// isDebugLabel - Return true if the specified node represents a debug
196 /// label (i.e. ISD::DBG_LABEL or TargetInstrInfo::DBG_LABEL node).
197 bool ISD::isDebugLabel(const SDNode *N) {
199 if (N->getOpcode() == ISD::DBG_LABEL)
201 if (N->isMachineOpcode() &&
202 N->getMachineOpcode() == TargetInstrInfo::DBG_LABEL)
207 /// getSetCCSwappedOperands - Return the operation corresponding to (Y op X)
208 /// when given the operation for (X op Y).
209 ISD::CondCode ISD::getSetCCSwappedOperands(ISD::CondCode Operation) {
210 // To perform this operation, we just need to swap the L and G bits of the
212 unsigned OldL = (Operation >> 2) & 1;
213 unsigned OldG = (Operation >> 1) & 1;
214 return ISD::CondCode((Operation & ~6) | // Keep the N, U, E bits
215 (OldL << 1) | // New G bit
216 (OldG << 2)); // New L bit.
219 /// getSetCCInverse - Return the operation corresponding to !(X op Y), where
220 /// 'op' is a valid SetCC operation.
221 ISD::CondCode ISD::getSetCCInverse(ISD::CondCode Op, bool isInteger) {
222 unsigned Operation = Op;
224 Operation ^= 7; // Flip L, G, E bits, but not U.
226 Operation ^= 15; // Flip all of the condition bits.
227 if (Operation > ISD::SETTRUE2)
228 Operation &= ~8; // Don't let N and U bits get set.
229 return ISD::CondCode(Operation);
233 /// isSignedOp - For an integer comparison, return 1 if the comparison is a
234 /// signed operation and 2 if the result is an unsigned comparison. Return zero
235 /// if the operation does not depend on the sign of the input (setne and seteq).
236 static int isSignedOp(ISD::CondCode Opcode) {
238 default: assert(0 && "Illegal integer setcc operation!");
240 case ISD::SETNE: return 0;
244 case ISD::SETGE: return 1;
248 case ISD::SETUGE: return 2;
252 /// getSetCCOrOperation - Return the result of a logical OR between different
253 /// comparisons of identical values: ((X op1 Y) | (X op2 Y)). This function
254 /// returns SETCC_INVALID if it is not possible to represent the resultant
256 ISD::CondCode ISD::getSetCCOrOperation(ISD::CondCode Op1, ISD::CondCode Op2,
258 if (isInteger && (isSignedOp(Op1) | isSignedOp(Op2)) == 3)
259 // Cannot fold a signed integer setcc with an unsigned integer setcc.
260 return ISD::SETCC_INVALID;
262 unsigned Op = Op1 | Op2; // Combine all of the condition bits.
264 // If the N and U bits get set then the resultant comparison DOES suddenly
265 // care about orderedness, and is true when ordered.
266 if (Op > ISD::SETTRUE2)
267 Op &= ~16; // Clear the U bit if the N bit is set.
269 // Canonicalize illegal integer setcc's.
270 if (isInteger && Op == ISD::SETUNE) // e.g. SETUGT | SETULT
273 return ISD::CondCode(Op);
276 /// getSetCCAndOperation - Return the result of a logical AND between different
277 /// comparisons of identical values: ((X op1 Y) & (X op2 Y)). This
278 /// function returns zero if it is not possible to represent the resultant
280 ISD::CondCode ISD::getSetCCAndOperation(ISD::CondCode Op1, ISD::CondCode Op2,
282 if (isInteger && (isSignedOp(Op1) | isSignedOp(Op2)) == 3)
283 // Cannot fold a signed setcc with an unsigned setcc.
284 return ISD::SETCC_INVALID;
286 // Combine all of the condition bits.
287 ISD::CondCode Result = ISD::CondCode(Op1 & Op2);
289 // Canonicalize illegal integer setcc's.
293 case ISD::SETUO : Result = ISD::SETFALSE; break; // SETUGT & SETULT
294 case ISD::SETOEQ: // SETEQ & SETU[LG]E
295 case ISD::SETUEQ: Result = ISD::SETEQ ; break; // SETUGE & SETULE
296 case ISD::SETOLT: Result = ISD::SETULT ; break; // SETULT & SETNE
297 case ISD::SETOGT: Result = ISD::SETUGT ; break; // SETUGT & SETNE
304 const TargetMachine &SelectionDAG::getTarget() const {
305 return TLI.getTargetMachine();
308 //===----------------------------------------------------------------------===//
309 // SDNode Profile Support
310 //===----------------------------------------------------------------------===//
312 /// AddNodeIDOpcode - Add the node opcode to the NodeID data.
314 static void AddNodeIDOpcode(FoldingSetNodeID &ID, unsigned OpC) {
318 /// AddNodeIDValueTypes - Value type lists are intern'd so we can represent them
319 /// solely with their pointer.
320 static void AddNodeIDValueTypes(FoldingSetNodeID &ID, SDVTList VTList) {
321 ID.AddPointer(VTList.VTs);
324 /// AddNodeIDOperands - Various routines for adding operands to the NodeID data.
326 static void AddNodeIDOperands(FoldingSetNodeID &ID,
327 const SDValue *Ops, unsigned NumOps) {
328 for (; NumOps; --NumOps, ++Ops) {
329 ID.AddPointer(Ops->Val);
330 ID.AddInteger(Ops->getResNo());
334 /// AddNodeIDOperands - Various routines for adding operands to the NodeID data.
336 static void AddNodeIDOperands(FoldingSetNodeID &ID,
337 const SDUse *Ops, unsigned NumOps) {
338 for (; NumOps; --NumOps, ++Ops) {
339 ID.AddPointer(Ops->getVal());
340 ID.AddInteger(Ops->getSDValue().getResNo());
344 static void AddNodeIDNode(FoldingSetNodeID &ID,
345 unsigned short OpC, SDVTList VTList,
346 const SDValue *OpList, unsigned N) {
347 AddNodeIDOpcode(ID, OpC);
348 AddNodeIDValueTypes(ID, VTList);
349 AddNodeIDOperands(ID, OpList, N);
353 /// AddNodeIDNode - Generic routine for adding a nodes info to the NodeID
355 static void AddNodeIDNode(FoldingSetNodeID &ID, const SDNode *N) {
356 AddNodeIDOpcode(ID, N->getOpcode());
357 // Add the return value info.
358 AddNodeIDValueTypes(ID, N->getVTList());
359 // Add the operand info.
360 AddNodeIDOperands(ID, N->op_begin(), N->getNumOperands());
362 // Handle SDNode leafs with special info.
363 switch (N->getOpcode()) {
364 default: break; // Normal nodes don't need extra info.
366 ID.AddInteger(cast<ARG_FLAGSSDNode>(N)->getArgFlags().getRawBits());
368 case ISD::TargetConstant:
370 ID.Add(cast<ConstantSDNode>(N)->getAPIntValue());
372 case ISD::TargetConstantFP:
373 case ISD::ConstantFP: {
374 ID.Add(cast<ConstantFPSDNode>(N)->getValueAPF());
377 case ISD::TargetGlobalAddress:
378 case ISD::GlobalAddress:
379 case ISD::TargetGlobalTLSAddress:
380 case ISD::GlobalTLSAddress: {
381 const GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(N);
382 ID.AddPointer(GA->getGlobal());
383 ID.AddInteger(GA->getOffset());
386 case ISD::BasicBlock:
387 ID.AddPointer(cast<BasicBlockSDNode>(N)->getBasicBlock());
390 ID.AddInteger(cast<RegisterSDNode>(N)->getReg());
392 case ISD::DBG_STOPPOINT: {
393 const DbgStopPointSDNode *DSP = cast<DbgStopPointSDNode>(N);
394 ID.AddInteger(DSP->getLine());
395 ID.AddInteger(DSP->getColumn());
396 ID.AddPointer(DSP->getCompileUnit());
400 ID.AddPointer(cast<SrcValueSDNode>(N)->getValue());
402 case ISD::MEMOPERAND: {
403 const MachineMemOperand &MO = cast<MemOperandSDNode>(N)->MO;
407 case ISD::FrameIndex:
408 case ISD::TargetFrameIndex:
409 ID.AddInteger(cast<FrameIndexSDNode>(N)->getIndex());
412 case ISD::TargetJumpTable:
413 ID.AddInteger(cast<JumpTableSDNode>(N)->getIndex());
415 case ISD::ConstantPool:
416 case ISD::TargetConstantPool: {
417 const ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(N);
418 ID.AddInteger(CP->getAlignment());
419 ID.AddInteger(CP->getOffset());
420 if (CP->isMachineConstantPoolEntry())
421 CP->getMachineCPVal()->AddSelectionDAGCSEId(ID);
423 ID.AddPointer(CP->getConstVal());
427 const LoadSDNode *LD = cast<LoadSDNode>(N);
428 ID.AddInteger(LD->getAddressingMode());
429 ID.AddInteger(LD->getExtensionType());
430 ID.AddInteger(LD->getMemoryVT().getRawBits());
431 ID.AddInteger(LD->getRawFlags());
435 const StoreSDNode *ST = cast<StoreSDNode>(N);
436 ID.AddInteger(ST->getAddressingMode());
437 ID.AddInteger(ST->isTruncatingStore());
438 ID.AddInteger(ST->getMemoryVT().getRawBits());
439 ID.AddInteger(ST->getRawFlags());
442 case ISD::ATOMIC_CMP_SWAP_8:
443 case ISD::ATOMIC_SWAP_8:
444 case ISD::ATOMIC_LOAD_ADD_8:
445 case ISD::ATOMIC_LOAD_SUB_8:
446 case ISD::ATOMIC_LOAD_AND_8:
447 case ISD::ATOMIC_LOAD_OR_8:
448 case ISD::ATOMIC_LOAD_XOR_8:
449 case ISD::ATOMIC_LOAD_NAND_8:
450 case ISD::ATOMIC_LOAD_MIN_8:
451 case ISD::ATOMIC_LOAD_MAX_8:
452 case ISD::ATOMIC_LOAD_UMIN_8:
453 case ISD::ATOMIC_LOAD_UMAX_8:
454 case ISD::ATOMIC_CMP_SWAP_16:
455 case ISD::ATOMIC_SWAP_16:
456 case ISD::ATOMIC_LOAD_ADD_16:
457 case ISD::ATOMIC_LOAD_SUB_16:
458 case ISD::ATOMIC_LOAD_AND_16:
459 case ISD::ATOMIC_LOAD_OR_16:
460 case ISD::ATOMIC_LOAD_XOR_16:
461 case ISD::ATOMIC_LOAD_NAND_16:
462 case ISD::ATOMIC_LOAD_MIN_16:
463 case ISD::ATOMIC_LOAD_MAX_16:
464 case ISD::ATOMIC_LOAD_UMIN_16:
465 case ISD::ATOMIC_LOAD_UMAX_16:
466 case ISD::ATOMIC_CMP_SWAP_32:
467 case ISD::ATOMIC_SWAP_32:
468 case ISD::ATOMIC_LOAD_ADD_32:
469 case ISD::ATOMIC_LOAD_SUB_32:
470 case ISD::ATOMIC_LOAD_AND_32:
471 case ISD::ATOMIC_LOAD_OR_32:
472 case ISD::ATOMIC_LOAD_XOR_32:
473 case ISD::ATOMIC_LOAD_NAND_32:
474 case ISD::ATOMIC_LOAD_MIN_32:
475 case ISD::ATOMIC_LOAD_MAX_32:
476 case ISD::ATOMIC_LOAD_UMIN_32:
477 case ISD::ATOMIC_LOAD_UMAX_32:
478 case ISD::ATOMIC_CMP_SWAP_64:
479 case ISD::ATOMIC_SWAP_64:
480 case ISD::ATOMIC_LOAD_ADD_64:
481 case ISD::ATOMIC_LOAD_SUB_64:
482 case ISD::ATOMIC_LOAD_AND_64:
483 case ISD::ATOMIC_LOAD_OR_64:
484 case ISD::ATOMIC_LOAD_XOR_64:
485 case ISD::ATOMIC_LOAD_NAND_64:
486 case ISD::ATOMIC_LOAD_MIN_64:
487 case ISD::ATOMIC_LOAD_MAX_64:
488 case ISD::ATOMIC_LOAD_UMIN_64:
489 case ISD::ATOMIC_LOAD_UMAX_64: {
490 const AtomicSDNode *AT = cast<AtomicSDNode>(N);
491 ID.AddInteger(AT->getRawFlags());
494 } // end switch (N->getOpcode())
497 /// encodeMemSDNodeFlags - Generic routine for computing a value for use in
498 /// the CSE map that carries both alignment and volatility information.
500 static unsigned encodeMemSDNodeFlags(bool isVolatile, unsigned Alignment) {
501 return isVolatile | ((Log2_32(Alignment) + 1) << 1);
504 //===----------------------------------------------------------------------===//
505 // SelectionDAG Class
506 //===----------------------------------------------------------------------===//
508 /// RemoveDeadNodes - This method deletes all unreachable nodes in the
510 void SelectionDAG::RemoveDeadNodes() {
511 // Create a dummy node (which is not added to allnodes), that adds a reference
512 // to the root node, preventing it from being deleted.
513 HandleSDNode Dummy(getRoot());
515 SmallVector<SDNode*, 128> DeadNodes;
517 // Add all obviously-dead nodes to the DeadNodes worklist.
518 for (allnodes_iterator I = allnodes_begin(), E = allnodes_end(); I != E; ++I)
520 DeadNodes.push_back(I);
522 RemoveDeadNodes(DeadNodes);
524 // If the root changed (e.g. it was a dead load, update the root).
525 setRoot(Dummy.getValue());
528 /// RemoveDeadNodes - This method deletes the unreachable nodes in the
529 /// given list, and any nodes that become unreachable as a result.
530 void SelectionDAG::RemoveDeadNodes(SmallVectorImpl<SDNode *> &DeadNodes,
531 DAGUpdateListener *UpdateListener) {
533 // Process the worklist, deleting the nodes and adding their uses to the
535 while (!DeadNodes.empty()) {
536 SDNode *N = DeadNodes.back();
537 DeadNodes.pop_back();
540 UpdateListener->NodeDeleted(N, 0);
542 // Take the node out of the appropriate CSE map.
543 RemoveNodeFromCSEMaps(N);
545 // Next, brutally remove the operand list. This is safe to do, as there are
546 // no cycles in the graph.
547 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) {
548 SDNode *Operand = I->getVal();
549 Operand->removeUser(std::distance(N->op_begin(), I), N);
551 // Now that we removed this operand, see if there are no uses of it left.
552 if (Operand->use_empty())
553 DeadNodes.push_back(Operand);
555 if (N->OperandsNeedDelete) {
556 delete[] N->OperandList;
561 // Finally, remove N itself.
562 NodeAllocator.Deallocate(AllNodes.remove(N));
566 void SelectionDAG::RemoveDeadNode(SDNode *N, DAGUpdateListener *UpdateListener){
567 SmallVector<SDNode*, 16> DeadNodes(1, N);
568 RemoveDeadNodes(DeadNodes, UpdateListener);
571 void SelectionDAG::DeleteNode(SDNode *N) {
572 assert(N->use_empty() && "Cannot delete a node that is not dead!");
574 // First take this out of the appropriate CSE map.
575 RemoveNodeFromCSEMaps(N);
577 // Finally, remove uses due to operands of this node, remove from the
578 // AllNodes list, and delete the node.
579 DeleteNodeNotInCSEMaps(N);
582 void SelectionDAG::DeleteNodeNotInCSEMaps(SDNode *N) {
584 // Drop all of the operands and decrement used nodes use counts.
585 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I)
586 I->getVal()->removeUser(std::distance(N->op_begin(), I), N);
587 if (N->OperandsNeedDelete)
588 delete[] N->OperandList;
590 assert(N != AllNodes.begin());
591 NodeAllocator.Deallocate(AllNodes.remove(N));
594 /// RemoveNodeFromCSEMaps - Take the specified node out of the CSE map that
595 /// correspond to it. This is useful when we're about to delete or repurpose
596 /// the node. We don't want future request for structurally identical nodes
597 /// to return N anymore.
598 void SelectionDAG::RemoveNodeFromCSEMaps(SDNode *N) {
600 switch (N->getOpcode()) {
601 case ISD::EntryToken:
602 assert(0 && "EntryToken should not be in CSEMaps!");
604 case ISD::HANDLENODE: return; // noop.
606 assert(CondCodeNodes[cast<CondCodeSDNode>(N)->get()] &&
607 "Cond code doesn't exist!");
608 Erased = CondCodeNodes[cast<CondCodeSDNode>(N)->get()] != 0;
609 CondCodeNodes[cast<CondCodeSDNode>(N)->get()] = 0;
611 case ISD::ExternalSymbol:
612 Erased = ExternalSymbols.erase(cast<ExternalSymbolSDNode>(N)->getSymbol());
614 case ISD::TargetExternalSymbol:
616 TargetExternalSymbols.erase(cast<ExternalSymbolSDNode>(N)->getSymbol());
618 case ISD::VALUETYPE: {
619 MVT VT = cast<VTSDNode>(N)->getVT();
620 if (VT.isExtended()) {
621 Erased = ExtendedValueTypeNodes.erase(VT);
623 Erased = ValueTypeNodes[VT.getSimpleVT()] != 0;
624 ValueTypeNodes[VT.getSimpleVT()] = 0;
629 // Remove it from the CSE Map.
630 Erased = CSEMap.RemoveNode(N);
634 // Verify that the node was actually in one of the CSE maps, unless it has a
635 // flag result (which cannot be CSE'd) or is one of the special cases that are
636 // not subject to CSE.
637 if (!Erased && N->getValueType(N->getNumValues()-1) != MVT::Flag &&
638 !N->isTargetOpcode() &&
639 N->getOpcode() != ISD::DBG_LABEL &&
640 N->getOpcode() != ISD::DBG_STOPPOINT &&
641 N->getOpcode() != ISD::EH_LABEL &&
642 N->getOpcode() != ISD::DECLARE) {
645 assert(0 && "Node is not in map!");
650 /// AddNonLeafNodeToCSEMaps - Add the specified node back to the CSE maps. It
651 /// has been taken out and modified in some way. If the specified node already
652 /// exists in the CSE maps, do not modify the maps, but return the existing node
653 /// instead. If it doesn't exist, add it and return null.
655 SDNode *SelectionDAG::AddNonLeafNodeToCSEMaps(SDNode *N) {
656 assert(N->getNumOperands() && "This is a leaf node!");
658 if (N->getValueType(0) == MVT::Flag)
659 return 0; // Never CSE anything that produces a flag.
661 switch (N->getOpcode()) {
663 case ISD::HANDLENODE:
665 case ISD::DBG_STOPPOINT:
668 return 0; // Never add these nodes.
671 // Check that remaining values produced are not flags.
672 for (unsigned i = 1, e = N->getNumValues(); i != e; ++i)
673 if (N->getValueType(i) == MVT::Flag)
674 return 0; // Never CSE anything that produces a flag.
676 SDNode *New = CSEMap.GetOrInsertNode(N);
677 if (New != N) return New; // Node already existed.
681 /// FindModifiedNodeSlot - Find a slot for the specified node if its operands
682 /// were replaced with those specified. If this node is never memoized,
683 /// return null, otherwise return a pointer to the slot it would take. If a
684 /// node already exists with these operands, the slot will be non-null.
685 SDNode *SelectionDAG::FindModifiedNodeSlot(SDNode *N, SDValue Op,
687 if (N->getValueType(0) == MVT::Flag)
688 return 0; // Never CSE anything that produces a flag.
690 switch (N->getOpcode()) {
692 case ISD::HANDLENODE:
694 case ISD::DBG_STOPPOINT:
696 return 0; // Never add these nodes.
699 // Check that remaining values produced are not flags.
700 for (unsigned i = 1, e = N->getNumValues(); i != e; ++i)
701 if (N->getValueType(i) == MVT::Flag)
702 return 0; // Never CSE anything that produces a flag.
704 SDValue Ops[] = { Op };
706 AddNodeIDNode(ID, N->getOpcode(), N->getVTList(), Ops, 1);
707 return CSEMap.FindNodeOrInsertPos(ID, InsertPos);
710 /// FindModifiedNodeSlot - Find a slot for the specified node if its operands
711 /// were replaced with those specified. If this node is never memoized,
712 /// return null, otherwise return a pointer to the slot it would take. If a
713 /// node already exists with these operands, the slot will be non-null.
714 SDNode *SelectionDAG::FindModifiedNodeSlot(SDNode *N,
715 SDValue Op1, SDValue Op2,
717 if (N->getOpcode() == ISD::HANDLENODE || N->getValueType(0) == MVT::Flag)
719 // Check that remaining values produced are not flags.
720 for (unsigned i = 1, e = N->getNumValues(); i != e; ++i)
721 if (N->getValueType(i) == MVT::Flag)
722 return 0; // Never CSE anything that produces a flag.
724 SDValue Ops[] = { Op1, Op2 };
726 AddNodeIDNode(ID, N->getOpcode(), N->getVTList(), Ops, 2);
727 return CSEMap.FindNodeOrInsertPos(ID, InsertPos);
731 /// FindModifiedNodeSlot - Find a slot for the specified node if its operands
732 /// were replaced with those specified. If this node is never memoized,
733 /// return null, otherwise return a pointer to the slot it would take. If a
734 /// node already exists with these operands, the slot will be non-null.
735 SDNode *SelectionDAG::FindModifiedNodeSlot(SDNode *N,
736 const SDValue *Ops,unsigned NumOps,
738 if (N->getValueType(0) == MVT::Flag)
739 return 0; // Never CSE anything that produces a flag.
741 switch (N->getOpcode()) {
743 case ISD::HANDLENODE:
745 case ISD::DBG_STOPPOINT:
748 return 0; // Never add these nodes.
751 // Check that remaining values produced are not flags.
752 for (unsigned i = 1, e = N->getNumValues(); i != e; ++i)
753 if (N->getValueType(i) == MVT::Flag)
754 return 0; // Never CSE anything that produces a flag.
757 AddNodeIDNode(ID, N->getOpcode(), N->getVTList(), Ops, NumOps);
759 if (const LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
760 ID.AddInteger(LD->getAddressingMode());
761 ID.AddInteger(LD->getExtensionType());
762 ID.AddInteger(LD->getMemoryVT().getRawBits());
763 ID.AddInteger(LD->getRawFlags());
764 } else if (const StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
765 ID.AddInteger(ST->getAddressingMode());
766 ID.AddInteger(ST->isTruncatingStore());
767 ID.AddInteger(ST->getMemoryVT().getRawBits());
768 ID.AddInteger(ST->getRawFlags());
771 return CSEMap.FindNodeOrInsertPos(ID, InsertPos);
774 /// VerifyNode - Sanity check the given node. Aborts if it is invalid.
775 void SelectionDAG::VerifyNode(SDNode *N) {
776 switch (N->getOpcode()) {
779 case ISD::BUILD_VECTOR: {
780 assert(N->getNumValues() == 1 && "Too many results for BUILD_VECTOR!");
781 assert(N->getValueType(0).isVector() && "Wrong BUILD_VECTOR return type!");
782 assert(N->getNumOperands() == N->getValueType(0).getVectorNumElements() &&
783 "Wrong number of BUILD_VECTOR operands!");
784 MVT EltVT = N->getValueType(0).getVectorElementType();
785 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I)
786 assert(I->getSDValue().getValueType() == EltVT &&
787 "Wrong BUILD_VECTOR operand type!");
793 /// getMVTAlignment - Compute the default alignment value for the
796 unsigned SelectionDAG::getMVTAlignment(MVT VT) const {
797 const Type *Ty = VT == MVT::iPTR ?
798 PointerType::get(Type::Int8Ty, 0) :
801 return TLI.getTargetData()->getABITypeAlignment(Ty);
804 SelectionDAG::SelectionDAG(TargetLowering &tli, FunctionLoweringInfo &fli)
805 : TLI(tli), FLI(fli),
806 EntryNode(ISD::EntryToken, getVTList(MVT::Other)),
807 Root(getEntryNode()) {
808 AllNodes.push_back(&EntryNode);
811 void SelectionDAG::init(MachineFunction &mf, MachineModuleInfo *mmi) {
816 SelectionDAG::~SelectionDAG() {
820 void SelectionDAG::allnodes_clear() {
821 assert(&*AllNodes.begin() == &EntryNode);
822 AllNodes.remove(AllNodes.begin());
823 while (!AllNodes.empty()) {
824 SDNode *N = AllNodes.remove(AllNodes.begin());
825 N->SetNextInBucket(0);
826 if (N->OperandsNeedDelete)
827 delete [] N->OperandList;
828 NodeAllocator.Deallocate(N);
832 void SelectionDAG::clear() {
834 OperandAllocator.Reset();
837 ExtendedValueTypeNodes.clear();
838 ExternalSymbols.clear();
839 TargetExternalSymbols.clear();
840 std::fill(CondCodeNodes.begin(), CondCodeNodes.end(),
841 static_cast<CondCodeSDNode*>(0));
842 std::fill(ValueTypeNodes.begin(), ValueTypeNodes.end(),
843 static_cast<SDNode*>(0));
846 AllNodes.push_back(&EntryNode);
847 Root = getEntryNode();
850 SDValue SelectionDAG::getZeroExtendInReg(SDValue Op, MVT VT) {
851 if (Op.getValueType() == VT) return Op;
852 APInt Imm = APInt::getLowBitsSet(Op.getValueSizeInBits(),
854 return getNode(ISD::AND, Op.getValueType(), Op,
855 getConstant(Imm, Op.getValueType()));
858 SDValue SelectionDAG::getConstant(uint64_t Val, MVT VT, bool isT) {
859 MVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT;
860 return getConstant(APInt(EltVT.getSizeInBits(), Val), VT, isT);
863 SDValue SelectionDAG::getConstant(const APInt &Val, MVT VT, bool isT) {
864 assert(VT.isInteger() && "Cannot create FP integer constant!");
866 MVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT;
867 assert(Val.getBitWidth() == EltVT.getSizeInBits() &&
868 "APInt size does not match type size!");
870 unsigned Opc = isT ? ISD::TargetConstant : ISD::Constant;
872 AddNodeIDNode(ID, Opc, getVTList(EltVT), 0, 0);
876 if ((N = CSEMap.FindNodeOrInsertPos(ID, IP)))
878 return SDValue(N, 0);
880 N = NodeAllocator.Allocate<ConstantSDNode>();
881 new (N) ConstantSDNode(isT, Val, EltVT);
882 CSEMap.InsertNode(N, IP);
883 AllNodes.push_back(N);
886 SDValue Result(N, 0);
888 SmallVector<SDValue, 8> Ops;
889 Ops.assign(VT.getVectorNumElements(), Result);
890 Result = getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
895 SDValue SelectionDAG::getIntPtrConstant(uint64_t Val, bool isTarget) {
896 return getConstant(Val, TLI.getPointerTy(), isTarget);
900 SDValue SelectionDAG::getConstantFP(const APFloat& V, MVT VT, bool isTarget) {
901 assert(VT.isFloatingPoint() && "Cannot create integer FP constant!");
904 VT.isVector() ? VT.getVectorElementType() : VT;
906 // Do the map lookup using the actual bit pattern for the floating point
907 // value, so that we don't have problems with 0.0 comparing equal to -0.0, and
908 // we don't have issues with SNANs.
909 unsigned Opc = isTarget ? ISD::TargetConstantFP : ISD::ConstantFP;
911 AddNodeIDNode(ID, Opc, getVTList(EltVT), 0, 0);
915 if ((N = CSEMap.FindNodeOrInsertPos(ID, IP)))
917 return SDValue(N, 0);
919 N = NodeAllocator.Allocate<ConstantFPSDNode>();
920 new (N) ConstantFPSDNode(isTarget, V, EltVT);
921 CSEMap.InsertNode(N, IP);
922 AllNodes.push_back(N);
925 SDValue Result(N, 0);
927 SmallVector<SDValue, 8> Ops;
928 Ops.assign(VT.getVectorNumElements(), Result);
929 Result = getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
934 SDValue SelectionDAG::getConstantFP(double Val, MVT VT, bool isTarget) {
936 VT.isVector() ? VT.getVectorElementType() : VT;
938 return getConstantFP(APFloat((float)Val), VT, isTarget);
940 return getConstantFP(APFloat(Val), VT, isTarget);
943 SDValue SelectionDAG::getGlobalAddress(const GlobalValue *GV,
948 const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
950 // If GV is an alias then use the aliasee for determining thread-localness.
951 if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
952 GVar = dyn_cast_or_null<GlobalVariable>(GA->resolveAliasedGlobal());
955 if (GVar && GVar->isThreadLocal())
956 Opc = isTargetGA ? ISD::TargetGlobalTLSAddress : ISD::GlobalTLSAddress;
958 Opc = isTargetGA ? ISD::TargetGlobalAddress : ISD::GlobalAddress;
961 AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
963 ID.AddInteger(Offset);
965 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
966 return SDValue(E, 0);
967 SDNode *N = NodeAllocator.Allocate<GlobalAddressSDNode>();
968 new (N) GlobalAddressSDNode(isTargetGA, GV, VT, Offset);
969 CSEMap.InsertNode(N, IP);
970 AllNodes.push_back(N);
971 return SDValue(N, 0);
974 SDValue SelectionDAG::getFrameIndex(int FI, MVT VT, bool isTarget) {
975 unsigned Opc = isTarget ? ISD::TargetFrameIndex : ISD::FrameIndex;
977 AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
980 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
981 return SDValue(E, 0);
982 SDNode *N = NodeAllocator.Allocate<FrameIndexSDNode>();
983 new (N) FrameIndexSDNode(FI, VT, isTarget);
984 CSEMap.InsertNode(N, IP);
985 AllNodes.push_back(N);
986 return SDValue(N, 0);
989 SDValue SelectionDAG::getJumpTable(int JTI, MVT VT, bool isTarget){
990 unsigned Opc = isTarget ? ISD::TargetJumpTable : ISD::JumpTable;
992 AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
995 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
996 return SDValue(E, 0);
997 SDNode *N = NodeAllocator.Allocate<JumpTableSDNode>();
998 new (N) JumpTableSDNode(JTI, VT, isTarget);
999 CSEMap.InsertNode(N, IP);
1000 AllNodes.push_back(N);
1001 return SDValue(N, 0);
1004 SDValue SelectionDAG::getConstantPool(Constant *C, MVT VT,
1005 unsigned Alignment, int Offset,
1007 unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool;
1008 FoldingSetNodeID ID;
1009 AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
1010 ID.AddInteger(Alignment);
1011 ID.AddInteger(Offset);
1014 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1015 return SDValue(E, 0);
1016 SDNode *N = NodeAllocator.Allocate<ConstantPoolSDNode>();
1017 new (N) ConstantPoolSDNode(isTarget, C, VT, Offset, Alignment);
1018 CSEMap.InsertNode(N, IP);
1019 AllNodes.push_back(N);
1020 return SDValue(N, 0);
1024 SDValue SelectionDAG::getConstantPool(MachineConstantPoolValue *C, MVT VT,
1025 unsigned Alignment, int Offset,
1027 unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool;
1028 FoldingSetNodeID ID;
1029 AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
1030 ID.AddInteger(Alignment);
1031 ID.AddInteger(Offset);
1032 C->AddSelectionDAGCSEId(ID);
1034 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1035 return SDValue(E, 0);
1036 SDNode *N = NodeAllocator.Allocate<ConstantPoolSDNode>();
1037 new (N) ConstantPoolSDNode(isTarget, C, VT, Offset, Alignment);
1038 CSEMap.InsertNode(N, IP);
1039 AllNodes.push_back(N);
1040 return SDValue(N, 0);
1044 SDValue SelectionDAG::getBasicBlock(MachineBasicBlock *MBB) {
1045 FoldingSetNodeID ID;
1046 AddNodeIDNode(ID, ISD::BasicBlock, getVTList(MVT::Other), 0, 0);
1049 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1050 return SDValue(E, 0);
1051 SDNode *N = NodeAllocator.Allocate<BasicBlockSDNode>();
1052 new (N) BasicBlockSDNode(MBB);
1053 CSEMap.InsertNode(N, IP);
1054 AllNodes.push_back(N);
1055 return SDValue(N, 0);
1058 SDValue SelectionDAG::getArgFlags(ISD::ArgFlagsTy Flags) {
1059 FoldingSetNodeID ID;
1060 AddNodeIDNode(ID, ISD::ARG_FLAGS, getVTList(MVT::Other), 0, 0);
1061 ID.AddInteger(Flags.getRawBits());
1063 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1064 return SDValue(E, 0);
1065 SDNode *N = NodeAllocator.Allocate<ARG_FLAGSSDNode>();
1066 new (N) ARG_FLAGSSDNode(Flags);
1067 CSEMap.InsertNode(N, IP);
1068 AllNodes.push_back(N);
1069 return SDValue(N, 0);
1072 SDValue SelectionDAG::getValueType(MVT VT) {
1073 if (VT.isSimple() && (unsigned)VT.getSimpleVT() >= ValueTypeNodes.size())
1074 ValueTypeNodes.resize(VT.getSimpleVT()+1);
1076 SDNode *&N = VT.isExtended() ?
1077 ExtendedValueTypeNodes[VT] : ValueTypeNodes[VT.getSimpleVT()];
1079 if (N) return SDValue(N, 0);
1080 N = NodeAllocator.Allocate<VTSDNode>();
1081 new (N) VTSDNode(VT);
1082 AllNodes.push_back(N);
1083 return SDValue(N, 0);
1086 SDValue SelectionDAG::getExternalSymbol(const char *Sym, MVT VT) {
1087 SDNode *&N = ExternalSymbols[Sym];
1088 if (N) return SDValue(N, 0);
1089 N = NodeAllocator.Allocate<ExternalSymbolSDNode>();
1090 new (N) ExternalSymbolSDNode(false, Sym, VT);
1091 AllNodes.push_back(N);
1092 return SDValue(N, 0);
1095 SDValue SelectionDAG::getTargetExternalSymbol(const char *Sym, MVT VT) {
1096 SDNode *&N = TargetExternalSymbols[Sym];
1097 if (N) return SDValue(N, 0);
1098 N = NodeAllocator.Allocate<ExternalSymbolSDNode>();
1099 new (N) ExternalSymbolSDNode(true, Sym, VT);
1100 AllNodes.push_back(N);
1101 return SDValue(N, 0);
1104 SDValue SelectionDAG::getCondCode(ISD::CondCode Cond) {
1105 if ((unsigned)Cond >= CondCodeNodes.size())
1106 CondCodeNodes.resize(Cond+1);
1108 if (CondCodeNodes[Cond] == 0) {
1109 CondCodeSDNode *N = NodeAllocator.Allocate<CondCodeSDNode>();
1110 new (N) CondCodeSDNode(Cond);
1111 CondCodeNodes[Cond] = N;
1112 AllNodes.push_back(N);
1114 return SDValue(CondCodeNodes[Cond], 0);
1117 SDValue SelectionDAG::getRegister(unsigned RegNo, MVT VT) {
1118 FoldingSetNodeID ID;
1119 AddNodeIDNode(ID, ISD::Register, getVTList(VT), 0, 0);
1120 ID.AddInteger(RegNo);
1122 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1123 return SDValue(E, 0);
1124 SDNode *N = NodeAllocator.Allocate<RegisterSDNode>();
1125 new (N) RegisterSDNode(RegNo, VT);
1126 CSEMap.InsertNode(N, IP);
1127 AllNodes.push_back(N);
1128 return SDValue(N, 0);
1131 SDValue SelectionDAG::getDbgStopPoint(SDValue Root,
1132 unsigned Line, unsigned Col,
1133 const CompileUnitDesc *CU) {
1134 SDNode *N = NodeAllocator.Allocate<DbgStopPointSDNode>();
1135 new (N) DbgStopPointSDNode(Root, Line, Col, CU);
1136 AllNodes.push_back(N);
1137 return SDValue(N, 0);
1140 SDValue SelectionDAG::getLabel(unsigned Opcode,
1143 FoldingSetNodeID ID;
1144 SDValue Ops[] = { Root };
1145 AddNodeIDNode(ID, Opcode, getVTList(MVT::Other), &Ops[0], 1);
1146 ID.AddInteger(LabelID);
1148 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1149 return SDValue(E, 0);
1150 SDNode *N = NodeAllocator.Allocate<LabelSDNode>();
1151 new (N) LabelSDNode(Opcode, Root, LabelID);
1152 CSEMap.InsertNode(N, IP);
1153 AllNodes.push_back(N);
1154 return SDValue(N, 0);
1157 SDValue SelectionDAG::getSrcValue(const Value *V) {
1158 assert((!V || isa<PointerType>(V->getType())) &&
1159 "SrcValue is not a pointer?");
1161 FoldingSetNodeID ID;
1162 AddNodeIDNode(ID, ISD::SRCVALUE, getVTList(MVT::Other), 0, 0);
1166 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1167 return SDValue(E, 0);
1169 SDNode *N = NodeAllocator.Allocate<SrcValueSDNode>();
1170 new (N) SrcValueSDNode(V);
1171 CSEMap.InsertNode(N, IP);
1172 AllNodes.push_back(N);
1173 return SDValue(N, 0);
1176 SDValue SelectionDAG::getMemOperand(const MachineMemOperand &MO) {
1177 const Value *v = MO.getValue();
1178 assert((!v || isa<PointerType>(v->getType())) &&
1179 "SrcValue is not a pointer?");
1181 FoldingSetNodeID ID;
1182 AddNodeIDNode(ID, ISD::MEMOPERAND, getVTList(MVT::Other), 0, 0);
1186 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
1187 return SDValue(E, 0);
1189 SDNode *N = NodeAllocator.Allocate<MemOperandSDNode>();
1190 new (N) MemOperandSDNode(MO);
1191 CSEMap.InsertNode(N, IP);
1192 AllNodes.push_back(N);
1193 return SDValue(N, 0);
1196 /// CreateStackTemporary - Create a stack temporary, suitable for holding the
1197 /// specified value type.
1198 SDValue SelectionDAG::CreateStackTemporary(MVT VT, unsigned minAlign) {
1199 MachineFrameInfo *FrameInfo = getMachineFunction().getFrameInfo();
1200 unsigned ByteSize = VT.getSizeInBits()/8;
1201 const Type *Ty = VT.getTypeForMVT();
1202 unsigned StackAlign =
1203 std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty), minAlign);
1205 int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign);
1206 return getFrameIndex(FrameIdx, TLI.getPointerTy());
1209 SDValue SelectionDAG::FoldSetCC(MVT VT, SDValue N1,
1210 SDValue N2, ISD::CondCode Cond) {
1211 // These setcc operations always fold.
1215 case ISD::SETFALSE2: return getConstant(0, VT);
1217 case ISD::SETTRUE2: return getConstant(1, VT);
1229 assert(!N1.getValueType().isInteger() && "Illegal setcc for integer!");
1233 if (ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val)) {
1234 const APInt &C2 = N2C->getAPIntValue();
1235 if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val)) {
1236 const APInt &C1 = N1C->getAPIntValue();
1239 default: assert(0 && "Unknown integer setcc!");
1240 case ISD::SETEQ: return getConstant(C1 == C2, VT);
1241 case ISD::SETNE: return getConstant(C1 != C2, VT);
1242 case ISD::SETULT: return getConstant(C1.ult(C2), VT);
1243 case ISD::SETUGT: return getConstant(C1.ugt(C2), VT);
1244 case ISD::SETULE: return getConstant(C1.ule(C2), VT);
1245 case ISD::SETUGE: return getConstant(C1.uge(C2), VT);
1246 case ISD::SETLT: return getConstant(C1.slt(C2), VT);
1247 case ISD::SETGT: return getConstant(C1.sgt(C2), VT);
1248 case ISD::SETLE: return getConstant(C1.sle(C2), VT);
1249 case ISD::SETGE: return getConstant(C1.sge(C2), VT);
1253 if (ConstantFPSDNode *N1C = dyn_cast<ConstantFPSDNode>(N1.Val)) {
1254 if (ConstantFPSDNode *N2C = dyn_cast<ConstantFPSDNode>(N2.Val)) {
1255 // No compile time operations on this type yet.
1256 if (N1C->getValueType(0) == MVT::ppcf128)
1259 APFloat::cmpResult R = N1C->getValueAPF().compare(N2C->getValueAPF());
1262 case ISD::SETEQ: if (R==APFloat::cmpUnordered)
1263 return getNode(ISD::UNDEF, VT);
1265 case ISD::SETOEQ: return getConstant(R==APFloat::cmpEqual, VT);
1266 case ISD::SETNE: if (R==APFloat::cmpUnordered)
1267 return getNode(ISD::UNDEF, VT);
1269 case ISD::SETONE: return getConstant(R==APFloat::cmpGreaterThan ||
1270 R==APFloat::cmpLessThan, VT);
1271 case ISD::SETLT: if (R==APFloat::cmpUnordered)
1272 return getNode(ISD::UNDEF, VT);
1274 case ISD::SETOLT: return getConstant(R==APFloat::cmpLessThan, VT);
1275 case ISD::SETGT: if (R==APFloat::cmpUnordered)
1276 return getNode(ISD::UNDEF, VT);
1278 case ISD::SETOGT: return getConstant(R==APFloat::cmpGreaterThan, VT);
1279 case ISD::SETLE: if (R==APFloat::cmpUnordered)
1280 return getNode(ISD::UNDEF, VT);
1282 case ISD::SETOLE: return getConstant(R==APFloat::cmpLessThan ||
1283 R==APFloat::cmpEqual, VT);
1284 case ISD::SETGE: if (R==APFloat::cmpUnordered)
1285 return getNode(ISD::UNDEF, VT);
1287 case ISD::SETOGE: return getConstant(R==APFloat::cmpGreaterThan ||
1288 R==APFloat::cmpEqual, VT);
1289 case ISD::SETO: return getConstant(R!=APFloat::cmpUnordered, VT);
1290 case ISD::SETUO: return getConstant(R==APFloat::cmpUnordered, VT);
1291 case ISD::SETUEQ: return getConstant(R==APFloat::cmpUnordered ||
1292 R==APFloat::cmpEqual, VT);
1293 case ISD::SETUNE: return getConstant(R!=APFloat::cmpEqual, VT);
1294 case ISD::SETULT: return getConstant(R==APFloat::cmpUnordered ||
1295 R==APFloat::cmpLessThan, VT);
1296 case ISD::SETUGT: return getConstant(R==APFloat::cmpGreaterThan ||
1297 R==APFloat::cmpUnordered, VT);
1298 case ISD::SETULE: return getConstant(R!=APFloat::cmpGreaterThan, VT);
1299 case ISD::SETUGE: return getConstant(R!=APFloat::cmpLessThan, VT);
1302 // Ensure that the constant occurs on the RHS.
1303 return getSetCC(VT, N2, N1, ISD::getSetCCSwappedOperands(Cond));
1307 // Could not fold it.
1311 /// SignBitIsZero - Return true if the sign bit of Op is known to be zero. We
1312 /// use this predicate to simplify operations downstream.
1313 bool SelectionDAG::SignBitIsZero(SDValue Op, unsigned Depth) const {
1314 unsigned BitWidth = Op.getValueSizeInBits();
1315 return MaskedValueIsZero(Op, APInt::getSignBit(BitWidth), Depth);
1318 /// MaskedValueIsZero - Return true if 'V & Mask' is known to be zero. We use
1319 /// this predicate to simplify operations downstream. Mask is known to be zero
1320 /// for bits that V cannot have.
1321 bool SelectionDAG::MaskedValueIsZero(SDValue Op, const APInt &Mask,
1322 unsigned Depth) const {
1323 APInt KnownZero, KnownOne;
1324 ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth);
1325 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1326 return (KnownZero & Mask) == Mask;
1329 /// ComputeMaskedBits - Determine which of the bits specified in Mask are
1330 /// known to be either zero or one and return them in the KnownZero/KnownOne
1331 /// bitsets. This code only analyzes bits in Mask, in order to short-circuit
1333 void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask,
1334 APInt &KnownZero, APInt &KnownOne,
1335 unsigned Depth) const {
1336 unsigned BitWidth = Mask.getBitWidth();
1337 assert(BitWidth == Op.getValueType().getSizeInBits() &&
1338 "Mask size mismatches value type size!");
1340 KnownZero = KnownOne = APInt(BitWidth, 0); // Don't know anything.
1341 if (Depth == 6 || Mask == 0)
1342 return; // Limit search depth.
1344 APInt KnownZero2, KnownOne2;
1346 switch (Op.getOpcode()) {
1348 // We know all of the bits for a constant!
1349 KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue() & Mask;
1350 KnownZero = ~KnownOne & Mask;
1353 // If either the LHS or the RHS are Zero, the result is zero.
1354 ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
1355 ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownZero,
1356 KnownZero2, KnownOne2, Depth+1);
1357 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1358 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1360 // Output known-1 bits are only known if set in both the LHS & RHS.
1361 KnownOne &= KnownOne2;
1362 // Output known-0 are known to be clear if zero in either the LHS | RHS.
1363 KnownZero |= KnownZero2;
1366 ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
1367 ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownOne,
1368 KnownZero2, KnownOne2, Depth+1);
1369 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1370 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1372 // Output known-0 bits are only known if clear in both the LHS & RHS.
1373 KnownZero &= KnownZero2;
1374 // Output known-1 are known to be set if set in either the LHS | RHS.
1375 KnownOne |= KnownOne2;
1378 ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
1379 ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero2, KnownOne2, Depth+1);
1380 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1381 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1383 // Output known-0 bits are known if clear or set in both the LHS & RHS.
1384 APInt KnownZeroOut = (KnownZero & KnownZero2) | (KnownOne & KnownOne2);
1385 // Output known-1 are known to be set if set in only one of the LHS, RHS.
1386 KnownOne = (KnownZero & KnownOne2) | (KnownOne & KnownZero2);
1387 KnownZero = KnownZeroOut;
1391 APInt Mask2 = APInt::getAllOnesValue(BitWidth);
1392 ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero, KnownOne, Depth+1);
1393 ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1);
1394 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1395 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1397 // If low bits are zero in either operand, output low known-0 bits.
1398 // Also compute a conserative estimate for high known-0 bits.
1399 // More trickiness is possible, but this is sufficient for the
1400 // interesting case of alignment computation.
1402 unsigned TrailZ = KnownZero.countTrailingOnes() +
1403 KnownZero2.countTrailingOnes();
1404 unsigned LeadZ = std::max(KnownZero.countLeadingOnes() +
1405 KnownZero2.countLeadingOnes(),
1406 BitWidth) - BitWidth;
1408 TrailZ = std::min(TrailZ, BitWidth);
1409 LeadZ = std::min(LeadZ, BitWidth);
1410 KnownZero = APInt::getLowBitsSet(BitWidth, TrailZ) |
1411 APInt::getHighBitsSet(BitWidth, LeadZ);
1416 // For the purposes of computing leading zeros we can conservatively
1417 // treat a udiv as a logical right shift by the power of 2 known to
1418 // be less than the denominator.
1419 APInt AllOnes = APInt::getAllOnesValue(BitWidth);
1420 ComputeMaskedBits(Op.getOperand(0),
1421 AllOnes, KnownZero2, KnownOne2, Depth+1);
1422 unsigned LeadZ = KnownZero2.countLeadingOnes();
1426 ComputeMaskedBits(Op.getOperand(1),
1427 AllOnes, KnownZero2, KnownOne2, Depth+1);
1428 unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
1429 if (RHSUnknownLeadingOnes != BitWidth)
1430 LeadZ = std::min(BitWidth,
1431 LeadZ + BitWidth - RHSUnknownLeadingOnes - 1);
1433 KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ) & Mask;
1437 ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero, KnownOne, Depth+1);
1438 ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero2, KnownOne2, Depth+1);
1439 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1440 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1442 // Only known if known in both the LHS and RHS.
1443 KnownOne &= KnownOne2;
1444 KnownZero &= KnownZero2;
1446 case ISD::SELECT_CC:
1447 ComputeMaskedBits(Op.getOperand(3), Mask, KnownZero, KnownOne, Depth+1);
1448 ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero2, KnownOne2, Depth+1);
1449 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1450 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1452 // Only known if known in both the LHS and RHS.
1453 KnownOne &= KnownOne2;
1454 KnownZero &= KnownZero2;
1457 // If we know the result of a setcc has the top bits zero, use this info.
1458 if (TLI.getSetCCResultContents() == TargetLowering::ZeroOrOneSetCCResult &&
1460 KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1);
1463 // (shl X, C1) & C2 == 0 iff (X & C2 >>u C1) == 0
1464 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1465 unsigned ShAmt = SA->getValue();
1467 // If the shift count is an invalid immediate, don't do anything.
1468 if (ShAmt >= BitWidth)
1471 ComputeMaskedBits(Op.getOperand(0), Mask.lshr(ShAmt),
1472 KnownZero, KnownOne, Depth+1);
1473 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1474 KnownZero <<= ShAmt;
1476 // low bits known zero.
1477 KnownZero |= APInt::getLowBitsSet(BitWidth, ShAmt);
1481 // (ushr X, C1) & C2 == 0 iff (-1 >> C1) & C2 == 0
1482 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1483 unsigned ShAmt = SA->getValue();
1485 // If the shift count is an invalid immediate, don't do anything.
1486 if (ShAmt >= BitWidth)
1489 ComputeMaskedBits(Op.getOperand(0), (Mask << ShAmt),
1490 KnownZero, KnownOne, Depth+1);
1491 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1492 KnownZero = KnownZero.lshr(ShAmt);
1493 KnownOne = KnownOne.lshr(ShAmt);
1495 APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask;
1496 KnownZero |= HighBits; // High bits known zero.
1500 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1501 unsigned ShAmt = SA->getValue();
1503 // If the shift count is an invalid immediate, don't do anything.
1504 if (ShAmt >= BitWidth)
1507 APInt InDemandedMask = (Mask << ShAmt);
1508 // If any of the demanded bits are produced by the sign extension, we also
1509 // demand the input sign bit.
1510 APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask;
1511 if (HighBits.getBoolValue())
1512 InDemandedMask |= APInt::getSignBit(BitWidth);
1514 ComputeMaskedBits(Op.getOperand(0), InDemandedMask, KnownZero, KnownOne,
1516 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1517 KnownZero = KnownZero.lshr(ShAmt);
1518 KnownOne = KnownOne.lshr(ShAmt);
1520 // Handle the sign bits.
1521 APInt SignBit = APInt::getSignBit(BitWidth);
1522 SignBit = SignBit.lshr(ShAmt); // Adjust to where it is now in the mask.
1524 if (KnownZero.intersects(SignBit)) {
1525 KnownZero |= HighBits; // New bits are known zero.
1526 } else if (KnownOne.intersects(SignBit)) {
1527 KnownOne |= HighBits; // New bits are known one.
1531 case ISD::SIGN_EXTEND_INREG: {
1532 MVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
1533 unsigned EBits = EVT.getSizeInBits();
1535 // Sign extension. Compute the demanded bits in the result that are not
1536 // present in the input.
1537 APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - EBits) & Mask;
1539 APInt InSignBit = APInt::getSignBit(EBits);
1540 APInt InputDemandedBits = Mask & APInt::getLowBitsSet(BitWidth, EBits);
1542 // If the sign extended bits are demanded, we know that the sign
1544 InSignBit.zext(BitWidth);
1545 if (NewBits.getBoolValue())
1546 InputDemandedBits |= InSignBit;
1548 ComputeMaskedBits(Op.getOperand(0), InputDemandedBits,
1549 KnownZero, KnownOne, Depth+1);
1550 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1552 // If the sign bit of the input is known set or clear, then we know the
1553 // top bits of the result.
1554 if (KnownZero.intersects(InSignBit)) { // Input sign bit known clear
1555 KnownZero |= NewBits;
1556 KnownOne &= ~NewBits;
1557 } else if (KnownOne.intersects(InSignBit)) { // Input sign bit known set
1558 KnownOne |= NewBits;
1559 KnownZero &= ~NewBits;
1560 } else { // Input sign bit unknown
1561 KnownZero &= ~NewBits;
1562 KnownOne &= ~NewBits;
1569 unsigned LowBits = Log2_32(BitWidth)+1;
1570 KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
1575 if (ISD::isZEXTLoad(Op.Val)) {
1576 LoadSDNode *LD = cast<LoadSDNode>(Op);
1577 MVT VT = LD->getMemoryVT();
1578 unsigned MemBits = VT.getSizeInBits();
1579 KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits) & Mask;
1583 case ISD::ZERO_EXTEND: {
1584 MVT InVT = Op.getOperand(0).getValueType();
1585 unsigned InBits = InVT.getSizeInBits();
1586 APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
1587 APInt InMask = Mask;
1588 InMask.trunc(InBits);
1589 KnownZero.trunc(InBits);
1590 KnownOne.trunc(InBits);
1591 ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
1592 KnownZero.zext(BitWidth);
1593 KnownOne.zext(BitWidth);
1594 KnownZero |= NewBits;
1597 case ISD::SIGN_EXTEND: {
1598 MVT InVT = Op.getOperand(0).getValueType();
1599 unsigned InBits = InVT.getSizeInBits();
1600 APInt InSignBit = APInt::getSignBit(InBits);
1601 APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
1602 APInt InMask = Mask;
1603 InMask.trunc(InBits);
1605 // If any of the sign extended bits are demanded, we know that the sign
1606 // bit is demanded. Temporarily set this bit in the mask for our callee.
1607 if (NewBits.getBoolValue())
1608 InMask |= InSignBit;
1610 KnownZero.trunc(InBits);
1611 KnownOne.trunc(InBits);
1612 ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
1614 // Note if the sign bit is known to be zero or one.
1615 bool SignBitKnownZero = KnownZero.isNegative();
1616 bool SignBitKnownOne = KnownOne.isNegative();
1617 assert(!(SignBitKnownZero && SignBitKnownOne) &&
1618 "Sign bit can't be known to be both zero and one!");
1620 // If the sign bit wasn't actually demanded by our caller, we don't
1621 // want it set in the KnownZero and KnownOne result values. Reset the
1622 // mask and reapply it to the result values.
1624 InMask.trunc(InBits);
1625 KnownZero &= InMask;
1628 KnownZero.zext(BitWidth);
1629 KnownOne.zext(BitWidth);
1631 // If the sign bit is known zero or one, the top bits match.
1632 if (SignBitKnownZero)
1633 KnownZero |= NewBits;
1634 else if (SignBitKnownOne)
1635 KnownOne |= NewBits;
1638 case ISD::ANY_EXTEND: {
1639 MVT InVT = Op.getOperand(0).getValueType();
1640 unsigned InBits = InVT.getSizeInBits();
1641 APInt InMask = Mask;
1642 InMask.trunc(InBits);
1643 KnownZero.trunc(InBits);
1644 KnownOne.trunc(InBits);
1645 ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
1646 KnownZero.zext(BitWidth);
1647 KnownOne.zext(BitWidth);
1650 case ISD::TRUNCATE: {
1651 MVT InVT = Op.getOperand(0).getValueType();
1652 unsigned InBits = InVT.getSizeInBits();
1653 APInt InMask = Mask;
1654 InMask.zext(InBits);
1655 KnownZero.zext(InBits);
1656 KnownOne.zext(InBits);
1657 ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
1658 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1659 KnownZero.trunc(BitWidth);
1660 KnownOne.trunc(BitWidth);
1663 case ISD::AssertZext: {
1664 MVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
1665 APInt InMask = APInt::getLowBitsSet(BitWidth, VT.getSizeInBits());
1666 ComputeMaskedBits(Op.getOperand(0), Mask & InMask, KnownZero,
1668 KnownZero |= (~InMask) & Mask;
1672 // All bits are zero except the low bit.
1673 KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - 1);
1677 if (ConstantSDNode *CLHS = dyn_cast<ConstantSDNode>(Op.getOperand(0))) {
1678 // We know that the top bits of C-X are clear if X contains less bits
1679 // than C (i.e. no wrap-around can happen). For example, 20-X is
1680 // positive if we can prove that X is >= 0 and < 16.
1681 if (CLHS->getAPIntValue().isNonNegative()) {
1682 unsigned NLZ = (CLHS->getAPIntValue()+1).countLeadingZeros();
1683 // NLZ can't be BitWidth with no sign bit
1684 APInt MaskV = APInt::getHighBitsSet(BitWidth, NLZ+1);
1685 ComputeMaskedBits(Op.getOperand(1), MaskV, KnownZero2, KnownOne2,
1688 // If all of the MaskV bits are known to be zero, then we know the
1689 // output top bits are zero, because we now know that the output is
1691 if ((KnownZero2 & MaskV) == MaskV) {
1692 unsigned NLZ2 = CLHS->getAPIntValue().countLeadingZeros();
1693 // Top bits known zero.
1694 KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2) & Mask;
1701 // Output known-0 bits are known if clear or set in both the low clear bits
1702 // common to both LHS & RHS. For example, 8+(X<<3) is known to have the
1703 // low 3 bits clear.
1704 APInt Mask2 = APInt::getLowBitsSet(BitWidth, Mask.countTrailingOnes());
1705 ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1);
1706 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1707 unsigned KnownZeroOut = KnownZero2.countTrailingOnes();
1709 ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero2, KnownOne2, Depth+1);
1710 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
1711 KnownZeroOut = std::min(KnownZeroOut,
1712 KnownZero2.countTrailingOnes());
1714 KnownZero |= APInt::getLowBitsSet(BitWidth, KnownZeroOut);
1718 if (ConstantSDNode *Rem = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1719 const APInt &RA = Rem->getAPIntValue();
1720 if (RA.isPowerOf2() || (-RA).isPowerOf2()) {
1721 APInt LowBits = RA.isStrictlyPositive() ? (RA - 1) : ~RA;
1722 APInt Mask2 = LowBits | APInt::getSignBit(BitWidth);
1723 ComputeMaskedBits(Op.getOperand(0), Mask2,KnownZero2,KnownOne2,Depth+1);
1725 // If the sign bit of the first operand is zero, the sign bit of
1726 // the result is zero. If the first operand has no one bits below
1727 // the second operand's single 1 bit, its sign will be zero.
1728 if (KnownZero2[BitWidth-1] || ((KnownZero2 & LowBits) == LowBits))
1729 KnownZero2 |= ~LowBits;
1731 KnownZero |= KnownZero2 & Mask;
1733 assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?");
1738 if (ConstantSDNode *Rem = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1739 const APInt &RA = Rem->getAPIntValue();
1740 if (RA.isPowerOf2()) {
1741 APInt LowBits = (RA - 1);
1742 APInt Mask2 = LowBits & Mask;
1743 KnownZero |= ~LowBits & Mask;
1744 ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero, KnownOne,Depth+1);
1745 assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?");
1750 // Since the result is less than or equal to either operand, any leading
1751 // zero bits in either operand must also exist in the result.
1752 APInt AllOnes = APInt::getAllOnesValue(BitWidth);
1753 ComputeMaskedBits(Op.getOperand(0), AllOnes, KnownZero, KnownOne,
1755 ComputeMaskedBits(Op.getOperand(1), AllOnes, KnownZero2, KnownOne2,
1758 uint32_t Leaders = std::max(KnownZero.countLeadingOnes(),
1759 KnownZero2.countLeadingOnes());
1761 KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & Mask;
1765 // Allow the target to implement this method for its nodes.
1766 if (Op.getOpcode() >= ISD::BUILTIN_OP_END) {
1767 case ISD::INTRINSIC_WO_CHAIN:
1768 case ISD::INTRINSIC_W_CHAIN:
1769 case ISD::INTRINSIC_VOID:
1770 TLI.computeMaskedBitsForTargetNode(Op, Mask, KnownZero, KnownOne, *this);
1776 /// ComputeNumSignBits - Return the number of times the sign bit of the
1777 /// register is replicated into the other bits. We know that at least 1 bit
1778 /// is always equal to the sign bit (itself), but other cases can give us
1779 /// information. For example, immediately after an "SRA X, 2", we know that
1780 /// the top 3 bits are all equal to each other, so we return 3.
1781 unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const{
1782 MVT VT = Op.getValueType();
1783 assert(VT.isInteger() && "Invalid VT!");
1784 unsigned VTBits = VT.getSizeInBits();
1786 unsigned FirstAnswer = 1;
1789 return 1; // Limit search depth.
1791 switch (Op.getOpcode()) {
1793 case ISD::AssertSext:
1794 Tmp = cast<VTSDNode>(Op.getOperand(1))->getVT().getSizeInBits();
1795 return VTBits-Tmp+1;
1796 case ISD::AssertZext:
1797 Tmp = cast<VTSDNode>(Op.getOperand(1))->getVT().getSizeInBits();
1800 case ISD::Constant: {
1801 const APInt &Val = cast<ConstantSDNode>(Op)->getAPIntValue();
1802 // If negative, return # leading ones.
1803 if (Val.isNegative())
1804 return Val.countLeadingOnes();
1806 // Return # leading zeros.
1807 return Val.countLeadingZeros();
1810 case ISD::SIGN_EXTEND:
1811 Tmp = VTBits-Op.getOperand(0).getValueType().getSizeInBits();
1812 return ComputeNumSignBits(Op.getOperand(0), Depth+1) + Tmp;
1814 case ISD::SIGN_EXTEND_INREG:
1815 // Max of the input and what this extends.
1816 Tmp = cast<VTSDNode>(Op.getOperand(1))->getVT().getSizeInBits();
1819 Tmp2 = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1820 return std::max(Tmp, Tmp2);
1823 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1824 // SRA X, C -> adds C sign bits.
1825 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1826 Tmp += C->getValue();
1827 if (Tmp > VTBits) Tmp = VTBits;
1831 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1832 // shl destroys sign bits.
1833 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1834 if (C->getValue() >= VTBits || // Bad shift.
1835 C->getValue() >= Tmp) break; // Shifted all sign bits out.
1836 return Tmp - C->getValue();
1841 case ISD::XOR: // NOT is handled here.
1842 // Logical binary ops preserve the number of sign bits at the worst.
1843 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1845 Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1);
1846 FirstAnswer = std::min(Tmp, Tmp2);
1847 // We computed what we know about the sign bits as our first
1848 // answer. Now proceed to the generic code that uses
1849 // ComputeMaskedBits, and pick whichever answer is better.
1854 Tmp = ComputeNumSignBits(Op.getOperand(1), Depth+1);
1855 if (Tmp == 1) return 1; // Early out.
1856 Tmp2 = ComputeNumSignBits(Op.getOperand(2), Depth+1);
1857 return std::min(Tmp, Tmp2);
1860 // If setcc returns 0/-1, all bits are sign bits.
1861 if (TLI.getSetCCResultContents() ==
1862 TargetLowering::ZeroOrNegativeOneSetCCResult)
1867 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
1868 unsigned RotAmt = C->getValue() & (VTBits-1);
1870 // Handle rotate right by N like a rotate left by 32-N.
1871 if (Op.getOpcode() == ISD::ROTR)
1872 RotAmt = (VTBits-RotAmt) & (VTBits-1);
1874 // If we aren't rotating out all of the known-in sign bits, return the
1875 // number that are left. This handles rotl(sext(x), 1) for example.
1876 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1877 if (Tmp > RotAmt+1) return Tmp-RotAmt;
1881 // Add can have at most one carry bit. Thus we know that the output
1882 // is, at worst, one more bit than the inputs.
1883 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1884 if (Tmp == 1) return 1; // Early out.
1886 // Special case decrementing a value (ADD X, -1):
1887 if (ConstantSDNode *CRHS = dyn_cast<ConstantSDNode>(Op.getOperand(0)))
1888 if (CRHS->isAllOnesValue()) {
1889 APInt KnownZero, KnownOne;
1890 APInt Mask = APInt::getAllOnesValue(VTBits);
1891 ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero, KnownOne, Depth+1);
1893 // If the input is known to be 0 or 1, the output is 0/-1, which is all
1895 if ((KnownZero | APInt(VTBits, 1)) == Mask)
1898 // If we are subtracting one from a positive number, there is no carry
1899 // out of the result.
1900 if (KnownZero.isNegative())
1904 Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1);
1905 if (Tmp2 == 1) return 1;
1906 return std::min(Tmp, Tmp2)-1;
1910 Tmp2 = ComputeNumSignBits(Op.getOperand(1), Depth+1);
1911 if (Tmp2 == 1) return 1;
1914 if (ConstantSDNode *CLHS = dyn_cast<ConstantSDNode>(Op.getOperand(0)))
1915 if (CLHS->isNullValue()) {
1916 APInt KnownZero, KnownOne;
1917 APInt Mask = APInt::getAllOnesValue(VTBits);
1918 ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
1919 // If the input is known to be 0 or 1, the output is 0/-1, which is all
1921 if ((KnownZero | APInt(VTBits, 1)) == Mask)
1924 // If the input is known to be positive (the sign bit is known clear),
1925 // the output of the NEG has the same number of sign bits as the input.
1926 if (KnownZero.isNegative())
1929 // Otherwise, we treat this like a SUB.
1932 // Sub can have at most one carry bit. Thus we know that the output
1933 // is, at worst, one more bit than the inputs.
1934 Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
1935 if (Tmp == 1) return 1; // Early out.
1936 return std::min(Tmp, Tmp2)-1;
1939 // FIXME: it's tricky to do anything useful for this, but it is an important
1940 // case for targets like X86.
1944 // Handle LOADX separately here. EXTLOAD case will fallthrough.
1945 if (Op.getOpcode() == ISD::LOAD) {
1946 LoadSDNode *LD = cast<LoadSDNode>(Op);
1947 unsigned ExtType = LD->getExtensionType();
1950 case ISD::SEXTLOAD: // '17' bits known
1951 Tmp = LD->getMemoryVT().getSizeInBits();
1952 return VTBits-Tmp+1;
1953 case ISD::ZEXTLOAD: // '16' bits known
1954 Tmp = LD->getMemoryVT().getSizeInBits();
1959 // Allow the target to implement this method for its nodes.
1960 if (Op.getOpcode() >= ISD::BUILTIN_OP_END ||
1961 Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
1962 Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
1963 Op.getOpcode() == ISD::INTRINSIC_VOID) {
1964 unsigned NumBits = TLI.ComputeNumSignBitsForTargetNode(Op, Depth);
1965 if (NumBits > 1) FirstAnswer = std::max(FirstAnswer, NumBits);
1968 // Finally, if we can prove that the top bits of the result are 0's or 1's,
1969 // use this information.
1970 APInt KnownZero, KnownOne;
1971 APInt Mask = APInt::getAllOnesValue(VTBits);
1972 ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth);
1974 if (KnownZero.isNegative()) { // sign bit is 0
1976 } else if (KnownOne.isNegative()) { // sign bit is 1;
1983 // Okay, we know that the sign bit in Mask is set. Use CLZ to determine
1984 // the number of identical bits in the top of the input value.
1986 Mask <<= Mask.getBitWidth()-VTBits;
1987 // Return # leading zeros. We use 'min' here in case Val was zero before
1988 // shifting. We don't want to return '64' as for an i32 "0".
1989 return std::max(FirstAnswer, std::min(VTBits, Mask.countLeadingZeros()));
1993 bool SelectionDAG::isVerifiedDebugInfoDesc(SDValue Op) const {
1994 GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op);
1995 if (!GA) return false;
1996 GlobalVariable *GV = dyn_cast<GlobalVariable>(GA->getGlobal());
1997 if (!GV) return false;
1998 MachineModuleInfo *MMI = getMachineModuleInfo();
1999 return MMI && MMI->hasDebugInfo() && MMI->isVerified(GV);
2003 /// getShuffleScalarElt - Returns the scalar element that will make up the ith
2004 /// element of the result of the vector shuffle.
2005 SDValue SelectionDAG::getShuffleScalarElt(const SDNode *N, unsigned i) {
2006 MVT VT = N->getValueType(0);
2007 SDValue PermMask = N->getOperand(2);
2008 SDValue Idx = PermMask.getOperand(i);
2009 if (Idx.getOpcode() == ISD::UNDEF)
2010 return getNode(ISD::UNDEF, VT.getVectorElementType());
2011 unsigned Index = cast<ConstantSDNode>(Idx)->getValue();
2012 unsigned NumElems = PermMask.getNumOperands();
2013 SDValue V = (Index < NumElems) ? N->getOperand(0) : N->getOperand(1);
2016 if (V.getOpcode() == ISD::BIT_CONVERT) {
2017 V = V.getOperand(0);
2018 if (V.getValueType().getVectorNumElements() != NumElems)
2021 if (V.getOpcode() == ISD::SCALAR_TO_VECTOR)
2022 return (Index == 0) ? V.getOperand(0)
2023 : getNode(ISD::UNDEF, VT.getVectorElementType());
2024 if (V.getOpcode() == ISD::BUILD_VECTOR)
2025 return V.getOperand(Index);
2026 if (V.getOpcode() == ISD::VECTOR_SHUFFLE)
2027 return getShuffleScalarElt(V.Val, Index);
2032 /// getNode - Gets or creates the specified node.
2034 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT) {
2035 FoldingSetNodeID ID;
2036 AddNodeIDNode(ID, Opcode, getVTList(VT), 0, 0);
2038 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
2039 return SDValue(E, 0);
2040 SDNode *N = NodeAllocator.Allocate<SDNode>();
2041 new (N) SDNode(Opcode, SDNode::getSDVTList(VT));
2042 CSEMap.InsertNode(N, IP);
2044 AllNodes.push_back(N);
2048 return SDValue(N, 0);
2051 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT, SDValue Operand) {
2052 // Constant fold unary operations with an integer constant operand.
2053 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Operand.Val)) {
2054 const APInt &Val = C->getAPIntValue();
2055 unsigned BitWidth = VT.getSizeInBits();
2058 case ISD::SIGN_EXTEND:
2059 return getConstant(APInt(Val).sextOrTrunc(BitWidth), VT);
2060 case ISD::ANY_EXTEND:
2061 case ISD::ZERO_EXTEND:
2063 return getConstant(APInt(Val).zextOrTrunc(BitWidth), VT);
2064 case ISD::UINT_TO_FP:
2065 case ISD::SINT_TO_FP: {
2066 const uint64_t zero[] = {0, 0};
2067 // No compile time operations on this type.
2068 if (VT==MVT::ppcf128)
2070 APFloat apf = APFloat(APInt(BitWidth, 2, zero));
2071 (void)apf.convertFromAPInt(Val,
2072 Opcode==ISD::SINT_TO_FP,
2073 APFloat::rmNearestTiesToEven);
2074 return getConstantFP(apf, VT);
2076 case ISD::BIT_CONVERT:
2077 if (VT == MVT::f32 && C->getValueType(0) == MVT::i32)
2078 return getConstantFP(Val.bitsToFloat(), VT);
2079 else if (VT == MVT::f64 && C->getValueType(0) == MVT::i64)
2080 return getConstantFP(Val.bitsToDouble(), VT);
2083 return getConstant(Val.byteSwap(), VT);
2085 return getConstant(Val.countPopulation(), VT);
2087 return getConstant(Val.countLeadingZeros(), VT);
2089 return getConstant(Val.countTrailingZeros(), VT);
2093 // Constant fold unary operations with a floating point constant operand.
2094 if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Operand.Val)) {
2095 APFloat V = C->getValueAPF(); // make copy
2096 if (VT != MVT::ppcf128 && Operand.getValueType() != MVT::ppcf128) {
2100 return getConstantFP(V, VT);
2103 return getConstantFP(V, VT);
2105 case ISD::FP_EXTEND:
2106 // This can return overflow, underflow, or inexact; we don't care.
2107 // FIXME need to be more flexible about rounding mode.
2108 (void)V.convert(*MVTToAPFloatSemantics(VT),
2109 APFloat::rmNearestTiesToEven);
2110 return getConstantFP(V, VT);
2111 case ISD::FP_TO_SINT:
2112 case ISD::FP_TO_UINT: {
2114 assert(integerPartWidth >= 64);
2115 // FIXME need to be more flexible about rounding mode.
2116 APFloat::opStatus s = V.convertToInteger(&x, 64U,
2117 Opcode==ISD::FP_TO_SINT,
2118 APFloat::rmTowardZero);
2119 if (s==APFloat::opInvalidOp) // inexact is OK, in fact usual
2121 return getConstant(x, VT);
2123 case ISD::BIT_CONVERT:
2124 if (VT == MVT::i32 && C->getValueType(0) == MVT::f32)
2125 return getConstant((uint32_t)V.convertToAPInt().getZExtValue(), VT);
2126 else if (VT == MVT::i64 && C->getValueType(0) == MVT::f64)
2127 return getConstant(V.convertToAPInt().getZExtValue(), VT);
2133 unsigned OpOpcode = Operand.Val->getOpcode();
2135 case ISD::TokenFactor:
2136 case ISD::CONCAT_VECTORS:
2137 return Operand; // Factor or concat of one node? No need.
2138 case ISD::FP_ROUND: assert(0 && "Invalid method to make FP_ROUND node");
2139 case ISD::FP_EXTEND:
2140 assert(VT.isFloatingPoint() &&
2141 Operand.getValueType().isFloatingPoint() && "Invalid FP cast!");
2142 if (Operand.getValueType() == VT) return Operand; // noop conversion.
2143 if (Operand.getOpcode() == ISD::UNDEF)
2144 return getNode(ISD::UNDEF, VT);
2146 case ISD::SIGN_EXTEND:
2147 assert(VT.isInteger() && Operand.getValueType().isInteger() &&
2148 "Invalid SIGN_EXTEND!");
2149 if (Operand.getValueType() == VT) return Operand; // noop extension
2150 assert(Operand.getValueType().bitsLT(VT)
2151 && "Invalid sext node, dst < src!");
2152 if (OpOpcode == ISD::SIGN_EXTEND || OpOpcode == ISD::ZERO_EXTEND)
2153 return getNode(OpOpcode, VT, Operand.Val->getOperand(0));
2155 case ISD::ZERO_EXTEND:
2156 assert(VT.isInteger() && Operand.getValueType().isInteger() &&
2157 "Invalid ZERO_EXTEND!");
2158 if (Operand.getValueType() == VT) return Operand; // noop extension
2159 assert(Operand.getValueType().bitsLT(VT)
2160 && "Invalid zext node, dst < src!");
2161 if (OpOpcode == ISD::ZERO_EXTEND) // (zext (zext x)) -> (zext x)
2162 return getNode(ISD::ZERO_EXTEND, VT, Operand.Val->getOperand(0));
2164 case ISD::ANY_EXTEND:
2165 assert(VT.isInteger() && Operand.getValueType().isInteger() &&
2166 "Invalid ANY_EXTEND!");
2167 if (Operand.getValueType() == VT) return Operand; // noop extension
2168 assert(Operand.getValueType().bitsLT(VT)
2169 && "Invalid anyext node, dst < src!");
2170 if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND)
2171 // (ext (zext x)) -> (zext x) and (ext (sext x)) -> (sext x)
2172 return getNode(OpOpcode, VT, Operand.Val->getOperand(0));
2175 assert(VT.isInteger() && Operand.getValueType().isInteger() &&
2176 "Invalid TRUNCATE!");
2177 if (Operand.getValueType() == VT) return Operand; // noop truncate
2178 assert(Operand.getValueType().bitsGT(VT)
2179 && "Invalid truncate node, src < dst!");
2180 if (OpOpcode == ISD::TRUNCATE)
2181 return getNode(ISD::TRUNCATE, VT, Operand.Val->getOperand(0));
2182 else if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND ||
2183 OpOpcode == ISD::ANY_EXTEND) {
2184 // If the source is smaller than the dest, we still need an extend.
2185 if (Operand.Val->getOperand(0).getValueType().bitsLT(VT))
2186 return getNode(OpOpcode, VT, Operand.Val->getOperand(0));
2187 else if (Operand.Val->getOperand(0).getValueType().bitsGT(VT))
2188 return getNode(ISD::TRUNCATE, VT, Operand.Val->getOperand(0));
2190 return Operand.Val->getOperand(0);
2193 case ISD::BIT_CONVERT:
2194 // Basic sanity checking.
2195 assert(VT.getSizeInBits() == Operand.getValueType().getSizeInBits()
2196 && "Cannot BIT_CONVERT between types of different sizes!");
2197 if (VT == Operand.getValueType()) return Operand; // noop conversion.
2198 if (OpOpcode == ISD::BIT_CONVERT) // bitconv(bitconv(x)) -> bitconv(x)
2199 return getNode(ISD::BIT_CONVERT, VT, Operand.getOperand(0));
2200 if (OpOpcode == ISD::UNDEF)
2201 return getNode(ISD::UNDEF, VT);
2203 case ISD::SCALAR_TO_VECTOR:
2204 assert(VT.isVector() && !Operand.getValueType().isVector() &&
2205 VT.getVectorElementType() == Operand.getValueType() &&
2206 "Illegal SCALAR_TO_VECTOR node!");
2207 if (OpOpcode == ISD::UNDEF)
2208 return getNode(ISD::UNDEF, VT);
2209 // scalar_to_vector(extract_vector_elt V, 0) -> V, top bits are undefined.
2210 if (OpOpcode == ISD::EXTRACT_VECTOR_ELT &&
2211 isa<ConstantSDNode>(Operand.getOperand(1)) &&
2212 Operand.getConstantOperandVal(1) == 0 &&
2213 Operand.getOperand(0).getValueType() == VT)
2214 return Operand.getOperand(0);
2217 if (OpOpcode == ISD::FSUB) // -(X-Y) -> (Y-X)
2218 return getNode(ISD::FSUB, VT, Operand.Val->getOperand(1),
2219 Operand.Val->getOperand(0));
2220 if (OpOpcode == ISD::FNEG) // --X -> X
2221 return Operand.Val->getOperand(0);
2224 if (OpOpcode == ISD::FNEG) // abs(-X) -> abs(X)
2225 return getNode(ISD::FABS, VT, Operand.Val->getOperand(0));
2230 SDVTList VTs = getVTList(VT);
2231 if (VT != MVT::Flag) { // Don't CSE flag producing nodes
2232 FoldingSetNodeID ID;
2233 SDValue Ops[1] = { Operand };
2234 AddNodeIDNode(ID, Opcode, VTs, Ops, 1);
2236 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
2237 return SDValue(E, 0);
2238 N = NodeAllocator.Allocate<UnarySDNode>();
2239 new (N) UnarySDNode(Opcode, VTs, Operand);
2240 CSEMap.InsertNode(N, IP);
2242 N = NodeAllocator.Allocate<UnarySDNode>();
2243 new (N) UnarySDNode(Opcode, VTs, Operand);
2246 AllNodes.push_back(N);
2250 return SDValue(N, 0);
2253 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
2254 SDValue N1, SDValue N2) {
2255 ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
2256 ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val);
2259 case ISD::TokenFactor:
2260 assert(VT == MVT::Other && N1.getValueType() == MVT::Other &&
2261 N2.getValueType() == MVT::Other && "Invalid token factor!");
2262 // Fold trivial token factors.
2263 if (N1.getOpcode() == ISD::EntryToken) return N2;
2264 if (N2.getOpcode() == ISD::EntryToken) return N1;
2266 case ISD::CONCAT_VECTORS:
2267 // A CONCAT_VECTOR with all operands BUILD_VECTOR can be simplified to
2268 // one big BUILD_VECTOR.
2269 if (N1.getOpcode() == ISD::BUILD_VECTOR &&
2270 N2.getOpcode() == ISD::BUILD_VECTOR) {
2271 SmallVector<SDValue, 16> Elts(N1.Val->op_begin(), N1.Val->op_end());
2272 Elts.insert(Elts.end(), N2.Val->op_begin(), N2.Val->op_end());
2273 return getNode(ISD::BUILD_VECTOR, VT, &Elts[0], Elts.size());
2277 assert(VT.isInteger() && N1.getValueType() == N2.getValueType() &&
2278 N1.getValueType() == VT && "Binary operator types must match!");
2279 // (X & 0) -> 0. This commonly occurs when legalizing i64 values, so it's
2280 // worth handling here.
2281 if (N2C && N2C->isNullValue())
2283 if (N2C && N2C->isAllOnesValue()) // X & -1 -> X
2290 assert(VT.isInteger() && N1.getValueType() == N2.getValueType() &&
2291 N1.getValueType() == VT && "Binary operator types must match!");
2292 // (X ^|+- 0) -> X. This commonly occurs when legalizing i64 values, so
2293 // it's worth handling here.
2294 if (N2C && N2C->isNullValue())
2301 assert(VT.isInteger() && "This operator does not apply to FP types!");
2311 assert(N1.getValueType() == N2.getValueType() &&
2312 N1.getValueType() == VT && "Binary operator types must match!");
2314 case ISD::FCOPYSIGN: // N1 and result must match. N1/N2 need not match.
2315 assert(N1.getValueType() == VT &&
2316 N1.getValueType().isFloatingPoint() &&
2317 N2.getValueType().isFloatingPoint() &&
2318 "Invalid FCOPYSIGN!");
2325 assert(VT == N1.getValueType() &&
2326 "Shift operators return type must be the same as their first arg");
2327 assert(VT.isInteger() && N2.getValueType().isInteger() &&
2328 "Shifts only work on integers");
2330 // Always fold shifts of i1 values so the code generator doesn't need to
2331 // handle them. Since we know the size of the shift has to be less than the
2332 // size of the value, the shift/rotate count is guaranteed to be zero.
2336 case ISD::FP_ROUND_INREG: {
2337 MVT EVT = cast<VTSDNode>(N2)->getVT();
2338 assert(VT == N1.getValueType() && "Not an inreg round!");
2339 assert(VT.isFloatingPoint() && EVT.isFloatingPoint() &&
2340 "Cannot FP_ROUND_INREG integer types");
2341 assert(EVT.bitsLE(VT) && "Not rounding down!");
2342 if (cast<VTSDNode>(N2)->getVT() == VT) return N1; // Not actually rounding.
2346 assert(VT.isFloatingPoint() &&
2347 N1.getValueType().isFloatingPoint() &&
2348 VT.bitsLE(N1.getValueType()) &&
2349 isa<ConstantSDNode>(N2) && "Invalid FP_ROUND!");
2350 if (N1.getValueType() == VT) return N1; // noop conversion.
2352 case ISD::AssertSext:
2353 case ISD::AssertZext: {
2354 MVT EVT = cast<VTSDNode>(N2)->getVT();
2355 assert(VT == N1.getValueType() && "Not an inreg extend!");
2356 assert(VT.isInteger() && EVT.isInteger() &&
2357 "Cannot *_EXTEND_INREG FP types");
2358 assert(EVT.bitsLE(VT) && "Not extending!");
2359 if (VT == EVT) return N1; // noop assertion.
2362 case ISD::SIGN_EXTEND_INREG: {
2363 MVT EVT = cast<VTSDNode>(N2)->getVT();
2364 assert(VT == N1.getValueType() && "Not an inreg extend!");
2365 assert(VT.isInteger() && EVT.isInteger() &&
2366 "Cannot *_EXTEND_INREG FP types");
2367 assert(EVT.bitsLE(VT) && "Not extending!");
2368 if (EVT == VT) return N1; // Not actually extending
2371 APInt Val = N1C->getAPIntValue();
2372 unsigned FromBits = cast<VTSDNode>(N2)->getVT().getSizeInBits();
2373 Val <<= Val.getBitWidth()-FromBits;
2374 Val = Val.ashr(Val.getBitWidth()-FromBits);
2375 return getConstant(Val, VT);
2379 case ISD::EXTRACT_VECTOR_ELT:
2380 // EXTRACT_VECTOR_ELT of an UNDEF is an UNDEF.
2381 if (N1.getOpcode() == ISD::UNDEF)
2382 return getNode(ISD::UNDEF, VT);
2384 // EXTRACT_VECTOR_ELT of CONCAT_VECTORS is often formed while lowering is
2385 // expanding copies of large vectors from registers.
2387 N1.getOpcode() == ISD::CONCAT_VECTORS &&
2388 N1.getNumOperands() > 0) {
2390 N1.getOperand(0).getValueType().getVectorNumElements();
2391 return getNode(ISD::EXTRACT_VECTOR_ELT, VT,
2392 N1.getOperand(N2C->getValue() / Factor),
2393 getConstant(N2C->getValue() % Factor, N2.getValueType()));
2396 // EXTRACT_VECTOR_ELT of BUILD_VECTOR is often formed while lowering is
2397 // expanding large vector constants.
2398 if (N2C && N1.getOpcode() == ISD::BUILD_VECTOR)
2399 return N1.getOperand(N2C->getValue());
2401 // EXTRACT_VECTOR_ELT of INSERT_VECTOR_ELT is often formed when vector
2402 // operations are lowered to scalars.
2403 if (N1.getOpcode() == ISD::INSERT_VECTOR_ELT) {
2404 if (N1.getOperand(2) == N2)
2405 return N1.getOperand(1);
2407 return getNode(ISD::EXTRACT_VECTOR_ELT, VT, N1.getOperand(0), N2);
2410 case ISD::EXTRACT_ELEMENT:
2411 assert(N2C && (unsigned)N2C->getValue() < 2 && "Bad EXTRACT_ELEMENT!");
2412 assert(!N1.getValueType().isVector() && !VT.isVector() &&
2413 (N1.getValueType().isInteger() == VT.isInteger()) &&
2414 "Wrong types for EXTRACT_ELEMENT!");
2416 // EXTRACT_ELEMENT of BUILD_PAIR is often formed while legalize is expanding
2417 // 64-bit integers into 32-bit parts. Instead of building the extract of
2418 // the BUILD_PAIR, only to have legalize rip it apart, just do it now.
2419 if (N1.getOpcode() == ISD::BUILD_PAIR)
2420 return N1.getOperand(N2C->getValue());
2422 // EXTRACT_ELEMENT of a constant int is also very common.
2423 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N1)) {
2424 unsigned ElementSize = VT.getSizeInBits();
2425 unsigned Shift = ElementSize * N2C->getValue();
2426 APInt ShiftedVal = C->getAPIntValue().lshr(Shift);
2427 return getConstant(ShiftedVal.trunc(ElementSize), VT);
2430 case ISD::EXTRACT_SUBVECTOR:
2431 if (N1.getValueType() == VT) // Trivial extraction.
2438 const APInt &C1 = N1C->getAPIntValue(), &C2 = N2C->getAPIntValue();
2440 case ISD::ADD: return getConstant(C1 + C2, VT);
2441 case ISD::SUB: return getConstant(C1 - C2, VT);
2442 case ISD::MUL: return getConstant(C1 * C2, VT);
2444 if (C2.getBoolValue()) return getConstant(C1.udiv(C2), VT);
2447 if (C2.getBoolValue()) return getConstant(C1.urem(C2), VT);
2450 if (C2.getBoolValue()) return getConstant(C1.sdiv(C2), VT);
2453 if (C2.getBoolValue()) return getConstant(C1.srem(C2), VT);
2455 case ISD::AND : return getConstant(C1 & C2, VT);
2456 case ISD::OR : return getConstant(C1 | C2, VT);
2457 case ISD::XOR : return getConstant(C1 ^ C2, VT);
2458 case ISD::SHL : return getConstant(C1 << C2, VT);
2459 case ISD::SRL : return getConstant(C1.lshr(C2), VT);
2460 case ISD::SRA : return getConstant(C1.ashr(C2), VT);
2461 case ISD::ROTL : return getConstant(C1.rotl(C2), VT);
2462 case ISD::ROTR : return getConstant(C1.rotr(C2), VT);
2465 } else { // Cannonicalize constant to RHS if commutative
2466 if (isCommutativeBinOp(Opcode)) {
2467 std::swap(N1C, N2C);
2473 // Constant fold FP operations.
2474 ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1.Val);
2475 ConstantFPSDNode *N2CFP = dyn_cast<ConstantFPSDNode>(N2.Val);
2477 if (!N2CFP && isCommutativeBinOp(Opcode)) {
2478 // Cannonicalize constant to RHS if commutative
2479 std::swap(N1CFP, N2CFP);
2481 } else if (N2CFP && VT != MVT::ppcf128) {
2482 APFloat V1 = N1CFP->getValueAPF(), V2 = N2CFP->getValueAPF();
2483 APFloat::opStatus s;
2486 s = V1.add(V2, APFloat::rmNearestTiesToEven);
2487 if (s != APFloat::opInvalidOp)
2488 return getConstantFP(V1, VT);
2491 s = V1.subtract(V2, APFloat::rmNearestTiesToEven);
2492 if (s!=APFloat::opInvalidOp)
2493 return getConstantFP(V1, VT);
2496 s = V1.multiply(V2, APFloat::rmNearestTiesToEven);
2497 if (s!=APFloat::opInvalidOp)
2498 return getConstantFP(V1, VT);
2501 s = V1.divide(V2, APFloat::rmNearestTiesToEven);
2502 if (s!=APFloat::opInvalidOp && s!=APFloat::opDivByZero)
2503 return getConstantFP(V1, VT);
2506 s = V1.mod(V2, APFloat::rmNearestTiesToEven);
2507 if (s!=APFloat::opInvalidOp && s!=APFloat::opDivByZero)
2508 return getConstantFP(V1, VT);
2510 case ISD::FCOPYSIGN:
2512 return getConstantFP(V1, VT);
2518 // Canonicalize an UNDEF to the RHS, even over a constant.
2519 if (N1.getOpcode() == ISD::UNDEF) {
2520 if (isCommutativeBinOp(Opcode)) {
2524 case ISD::FP_ROUND_INREG:
2525 case ISD::SIGN_EXTEND_INREG:
2531 return N1; // fold op(undef, arg2) -> undef
2539 return getConstant(0, VT); // fold op(undef, arg2) -> 0
2540 // For vectors, we can't easily build an all zero vector, just return
2547 // Fold a bunch of operators when the RHS is undef.
2548 if (N2.getOpcode() == ISD::UNDEF) {
2551 if (N1.getOpcode() == ISD::UNDEF)
2552 // Handle undef ^ undef -> 0 special case. This is a common
2554 return getConstant(0, VT);
2569 return N2; // fold op(arg1, undef) -> undef
2575 return getConstant(0, VT); // fold op(arg1, undef) -> 0
2576 // For vectors, we can't easily build an all zero vector, just return
2581 return getConstant(VT.getIntegerVTBitMask(), VT);
2582 // For vectors, we can't easily build an all one vector, just return
2590 // Memoize this node if possible.
2592 SDVTList VTs = getVTList(VT);
2593 if (VT != MVT::Flag) {
2594 SDValue Ops[] = { N1, N2 };
2595 FoldingSetNodeID ID;
2596 AddNodeIDNode(ID, Opcode, VTs, Ops, 2);
2598 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
2599 return SDValue(E, 0);
2600 N = NodeAllocator.Allocate<BinarySDNode>();
2601 new (N) BinarySDNode(Opcode, VTs, N1, N2);
2602 CSEMap.InsertNode(N, IP);
2604 N = NodeAllocator.Allocate<BinarySDNode>();
2605 new (N) BinarySDNode(Opcode, VTs, N1, N2);
2608 AllNodes.push_back(N);
2612 return SDValue(N, 0);
2615 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
2616 SDValue N1, SDValue N2, SDValue N3) {
2617 // Perform various simplifications.
2618 ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
2619 ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val);
2621 case ISD::CONCAT_VECTORS:
2622 // A CONCAT_VECTOR with all operands BUILD_VECTOR can be simplified to
2623 // one big BUILD_VECTOR.
2624 if (N1.getOpcode() == ISD::BUILD_VECTOR &&
2625 N2.getOpcode() == ISD::BUILD_VECTOR &&
2626 N3.getOpcode() == ISD::BUILD_VECTOR) {
2627 SmallVector<SDValue, 16> Elts(N1.Val->op_begin(), N1.Val->op_end());
2628 Elts.insert(Elts.end(), N2.Val->op_begin(), N2.Val->op_end());
2629 Elts.insert(Elts.end(), N3.Val->op_begin(), N3.Val->op_end());
2630 return getNode(ISD::BUILD_VECTOR, VT, &Elts[0], Elts.size());
2634 // Use FoldSetCC to simplify SETCC's.
2635 SDValue Simp = FoldSetCC(VT, N1, N2, cast<CondCodeSDNode>(N3)->get());
2636 if (Simp.Val) return Simp;
2641 if (N1C->getValue())
2642 return N2; // select true, X, Y -> X
2644 return N3; // select false, X, Y -> Y
2647 if (N2 == N3) return N2; // select C, X, X -> X
2651 if (N2C->getValue()) // Unconditional branch
2652 return getNode(ISD::BR, MVT::Other, N1, N3);
2654 return N1; // Never-taken branch
2657 case ISD::VECTOR_SHUFFLE:
2658 assert(VT == N1.getValueType() && VT == N2.getValueType() &&
2659 VT.isVector() && N3.getValueType().isVector() &&
2660 N3.getOpcode() == ISD::BUILD_VECTOR &&
2661 VT.getVectorNumElements() == N3.getNumOperands() &&
2662 "Illegal VECTOR_SHUFFLE node!");
2664 case ISD::BIT_CONVERT:
2665 // Fold bit_convert nodes from a type to themselves.
2666 if (N1.getValueType() == VT)
2671 // Memoize node if it doesn't produce a flag.
2673 SDVTList VTs = getVTList(VT);
2674 if (VT != MVT::Flag) {
2675 SDValue Ops[] = { N1, N2, N3 };
2676 FoldingSetNodeID ID;
2677 AddNodeIDNode(ID, Opcode, VTs, Ops, 3);
2679 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
2680 return SDValue(E, 0);
2681 N = NodeAllocator.Allocate<TernarySDNode>();
2682 new (N) TernarySDNode(Opcode, VTs, N1, N2, N3);
2683 CSEMap.InsertNode(N, IP);
2685 N = NodeAllocator.Allocate<TernarySDNode>();
2686 new (N) TernarySDNode(Opcode, VTs, N1, N2, N3);
2688 AllNodes.push_back(N);
2692 return SDValue(N, 0);
2695 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
2696 SDValue N1, SDValue N2, SDValue N3,
2698 SDValue Ops[] = { N1, N2, N3, N4 };
2699 return getNode(Opcode, VT, Ops, 4);
2702 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
2703 SDValue N1, SDValue N2, SDValue N3,
2704 SDValue N4, SDValue N5) {
2705 SDValue Ops[] = { N1, N2, N3, N4, N5 };
2706 return getNode(Opcode, VT, Ops, 5);
2709 /// getMemsetValue - Vectorized representation of the memset value
2711 static SDValue getMemsetValue(SDValue Value, MVT VT, SelectionDAG &DAG) {
2712 unsigned NumBits = VT.isVector() ?
2713 VT.getVectorElementType().getSizeInBits() : VT.getSizeInBits();
2714 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Value)) {
2715 APInt Val = APInt(NumBits, C->getValue() & 255);
2717 for (unsigned i = NumBits; i > 8; i >>= 1) {
2718 Val = (Val << Shift) | Val;
2722 return DAG.getConstant(Val, VT);
2723 return DAG.getConstantFP(APFloat(Val), VT);
2726 Value = DAG.getNode(ISD::ZERO_EXTEND, VT, Value);
2728 for (unsigned i = NumBits; i > 8; i >>= 1) {
2729 Value = DAG.getNode(ISD::OR, VT,
2730 DAG.getNode(ISD::SHL, VT, Value,
2731 DAG.getConstant(Shift, MVT::i8)), Value);
2738 /// getMemsetStringVal - Similar to getMemsetValue. Except this is only
2739 /// used when a memcpy is turned into a memset when the source is a constant
2741 static SDValue getMemsetStringVal(MVT VT, SelectionDAG &DAG,
2742 const TargetLowering &TLI,
2743 std::string &Str, unsigned Offset) {
2744 // Handle vector with all elements zero.
2747 return DAG.getConstant(0, VT);
2748 unsigned NumElts = VT.getVectorNumElements();
2749 MVT EltVT = (VT.getVectorElementType() == MVT::f32) ? MVT::i32 : MVT::i64;
2750 return DAG.getNode(ISD::BIT_CONVERT, VT,
2751 DAG.getConstant(0, MVT::getVectorVT(EltVT, NumElts)));
2754 assert(!VT.isVector() && "Can't handle vector type here!");
2755 unsigned NumBits = VT.getSizeInBits();
2756 unsigned MSB = NumBits / 8;
2758 if (TLI.isLittleEndian())
2759 Offset = Offset + MSB - 1;
2760 for (unsigned i = 0; i != MSB; ++i) {
2761 Val = (Val << 8) | (unsigned char)Str[Offset];
2762 Offset += TLI.isLittleEndian() ? -1 : 1;
2764 return DAG.getConstant(Val, VT);
2767 /// getMemBasePlusOffset - Returns base and offset node for the
2769 static SDValue getMemBasePlusOffset(SDValue Base, unsigned Offset,
2770 SelectionDAG &DAG) {
2771 MVT VT = Base.getValueType();
2772 return DAG.getNode(ISD::ADD, VT, Base, DAG.getConstant(Offset, VT));
2775 /// isMemSrcFromString - Returns true if memcpy source is a string constant.
2777 static bool isMemSrcFromString(SDValue Src, std::string &Str) {
2778 unsigned SrcDelta = 0;
2779 GlobalAddressSDNode *G = NULL;
2780 if (Src.getOpcode() == ISD::GlobalAddress)
2781 G = cast<GlobalAddressSDNode>(Src);
2782 else if (Src.getOpcode() == ISD::ADD &&
2783 Src.getOperand(0).getOpcode() == ISD::GlobalAddress &&
2784 Src.getOperand(1).getOpcode() == ISD::Constant) {
2785 G = cast<GlobalAddressSDNode>(Src.getOperand(0));
2786 SrcDelta = cast<ConstantSDNode>(Src.getOperand(1))->getValue();
2791 GlobalVariable *GV = dyn_cast<GlobalVariable>(G->getGlobal());
2792 if (GV && GetConstantStringInfo(GV, Str, SrcDelta, false))
2798 /// MeetsMaxMemopRequirement - Determines if the number of memory ops required
2799 /// to replace the memset / memcpy is below the threshold. It also returns the
2800 /// types of the sequence of memory ops to perform memset / memcpy.
2802 bool MeetsMaxMemopRequirement(std::vector<MVT> &MemOps,
2803 SDValue Dst, SDValue Src,
2804 unsigned Limit, uint64_t Size, unsigned &Align,
2805 std::string &Str, bool &isSrcStr,
2807 const TargetLowering &TLI) {
2808 isSrcStr = isMemSrcFromString(Src, Str);
2809 bool isSrcConst = isa<ConstantSDNode>(Src);
2810 bool AllowUnalign = TLI.allowsUnalignedMemoryAccesses();
2811 MVT VT= TLI.getOptimalMemOpType(Size, Align, isSrcConst, isSrcStr);
2812 if (VT != MVT::iAny) {
2813 unsigned NewAlign = (unsigned)
2814 TLI.getTargetData()->getABITypeAlignment(VT.getTypeForMVT());
2815 // If source is a string constant, this will require an unaligned load.
2816 if (NewAlign > Align && (isSrcConst || AllowUnalign)) {
2817 if (Dst.getOpcode() != ISD::FrameIndex) {
2818 // Can't change destination alignment. It requires a unaligned store.
2822 int FI = cast<FrameIndexSDNode>(Dst)->getIndex();
2823 MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
2824 if (MFI->isFixedObjectIndex(FI)) {
2825 // Can't change destination alignment. It requires a unaligned store.
2829 // Give the stack frame object a larger alignment if needed.
2830 if (MFI->getObjectAlignment(FI) < NewAlign)
2831 MFI->setObjectAlignment(FI, NewAlign);
2838 if (VT == MVT::iAny) {
2842 switch (Align & 7) {
2843 case 0: VT = MVT::i64; break;
2844 case 4: VT = MVT::i32; break;
2845 case 2: VT = MVT::i16; break;
2846 default: VT = MVT::i8; break;
2851 while (!TLI.isTypeLegal(LVT))
2852 LVT = (MVT::SimpleValueType)(LVT.getSimpleVT() - 1);
2853 assert(LVT.isInteger());
2859 unsigned NumMemOps = 0;
2861 unsigned VTSize = VT.getSizeInBits() / 8;
2862 while (VTSize > Size) {
2863 // For now, only use non-vector load / store's for the left-over pieces.
2864 if (VT.isVector()) {
2866 while (!TLI.isTypeLegal(VT))
2867 VT = (MVT::SimpleValueType)(VT.getSimpleVT() - 1);
2868 VTSize = VT.getSizeInBits() / 8;
2870 VT = (MVT::SimpleValueType)(VT.getSimpleVT() - 1);
2875 if (++NumMemOps > Limit)
2877 MemOps.push_back(VT);
2884 static SDValue getMemcpyLoadsAndStores(SelectionDAG &DAG,
2885 SDValue Chain, SDValue Dst,
2886 SDValue Src, uint64_t Size,
2887 unsigned Align, bool AlwaysInline,
2888 const Value *DstSV, uint64_t DstSVOff,
2889 const Value *SrcSV, uint64_t SrcSVOff){
2890 const TargetLowering &TLI = DAG.getTargetLoweringInfo();
2892 // Expand memcpy to a series of load and store ops if the size operand falls
2893 // below a certain threshold.
2894 std::vector<MVT> MemOps;
2895 uint64_t Limit = -1;
2897 Limit = TLI.getMaxStoresPerMemcpy();
2898 unsigned DstAlign = Align; // Destination alignment can change.
2901 if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, Limit, Size, DstAlign,
2902 Str, CopyFromStr, DAG, TLI))
2906 bool isZeroStr = CopyFromStr && Str.empty();
2907 SmallVector<SDValue, 8> OutChains;
2908 unsigned NumMemOps = MemOps.size();
2909 uint64_t SrcOff = 0, DstOff = 0;
2910 for (unsigned i = 0; i < NumMemOps; i++) {
2912 unsigned VTSize = VT.getSizeInBits() / 8;
2913 SDValue Value, Store;
2915 if (CopyFromStr && (isZeroStr || !VT.isVector())) {
2916 // It's unlikely a store of a vector immediate can be done in a single
2917 // instruction. It would require a load from a constantpool first.
2918 // We also handle store a vector with all zero's.
2919 // FIXME: Handle other cases where store of vector immediate is done in
2920 // a single instruction.
2921 Value = getMemsetStringVal(VT, DAG, TLI, Str, SrcOff);
2922 Store = DAG.getStore(Chain, Value,
2923 getMemBasePlusOffset(Dst, DstOff, DAG),
2924 DstSV, DstSVOff + DstOff, false, DstAlign);
2926 Value = DAG.getLoad(VT, Chain,
2927 getMemBasePlusOffset(Src, SrcOff, DAG),
2928 SrcSV, SrcSVOff + SrcOff, false, Align);
2929 Store = DAG.getStore(Chain, Value,
2930 getMemBasePlusOffset(Dst, DstOff, DAG),
2931 DstSV, DstSVOff + DstOff, false, DstAlign);
2933 OutChains.push_back(Store);
2938 return DAG.getNode(ISD::TokenFactor, MVT::Other,
2939 &OutChains[0], OutChains.size());
2942 static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG,
2943 SDValue Chain, SDValue Dst,
2944 SDValue Src, uint64_t Size,
2945 unsigned Align, bool AlwaysInline,
2946 const Value *DstSV, uint64_t DstSVOff,
2947 const Value *SrcSV, uint64_t SrcSVOff){
2948 const TargetLowering &TLI = DAG.getTargetLoweringInfo();
2950 // Expand memmove to a series of load and store ops if the size operand falls
2951 // below a certain threshold.
2952 std::vector<MVT> MemOps;
2953 uint64_t Limit = -1;
2955 Limit = TLI.getMaxStoresPerMemmove();
2956 unsigned DstAlign = Align; // Destination alignment can change.
2959 if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, Limit, Size, DstAlign,
2960 Str, CopyFromStr, DAG, TLI))
2963 uint64_t SrcOff = 0, DstOff = 0;
2965 SmallVector<SDValue, 8> LoadValues;
2966 SmallVector<SDValue, 8> LoadChains;
2967 SmallVector<SDValue, 8> OutChains;
2968 unsigned NumMemOps = MemOps.size();
2969 for (unsigned i = 0; i < NumMemOps; i++) {
2971 unsigned VTSize = VT.getSizeInBits() / 8;
2972 SDValue Value, Store;
2974 Value = DAG.getLoad(VT, Chain,
2975 getMemBasePlusOffset(Src, SrcOff, DAG),
2976 SrcSV, SrcSVOff + SrcOff, false, Align);
2977 LoadValues.push_back(Value);
2978 LoadChains.push_back(Value.getValue(1));
2981 Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
2982 &LoadChains[0], LoadChains.size());
2984 for (unsigned i = 0; i < NumMemOps; i++) {
2986 unsigned VTSize = VT.getSizeInBits() / 8;
2987 SDValue Value, Store;
2989 Store = DAG.getStore(Chain, LoadValues[i],
2990 getMemBasePlusOffset(Dst, DstOff, DAG),
2991 DstSV, DstSVOff + DstOff, false, DstAlign);
2992 OutChains.push_back(Store);
2996 return DAG.getNode(ISD::TokenFactor, MVT::Other,
2997 &OutChains[0], OutChains.size());
3000 static SDValue getMemsetStores(SelectionDAG &DAG,
3001 SDValue Chain, SDValue Dst,
3002 SDValue Src, uint64_t Size,
3004 const Value *DstSV, uint64_t DstSVOff) {
3005 const TargetLowering &TLI = DAG.getTargetLoweringInfo();
3007 // Expand memset to a series of load/store ops if the size operand
3008 // falls below a certain threshold.
3009 std::vector<MVT> MemOps;
3012 if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, TLI.getMaxStoresPerMemset(),
3013 Size, Align, Str, CopyFromStr, DAG, TLI))
3016 SmallVector<SDValue, 8> OutChains;
3017 uint64_t DstOff = 0;
3019 unsigned NumMemOps = MemOps.size();
3020 for (unsigned i = 0; i < NumMemOps; i++) {
3022 unsigned VTSize = VT.getSizeInBits() / 8;
3023 SDValue Value = getMemsetValue(Src, VT, DAG);
3024 SDValue Store = DAG.getStore(Chain, Value,
3025 getMemBasePlusOffset(Dst, DstOff, DAG),
3026 DstSV, DstSVOff + DstOff);
3027 OutChains.push_back(Store);
3031 return DAG.getNode(ISD::TokenFactor, MVT::Other,
3032 &OutChains[0], OutChains.size());
3035 SDValue SelectionDAG::getMemcpy(SDValue Chain, SDValue Dst,
3036 SDValue Src, SDValue Size,
3037 unsigned Align, bool AlwaysInline,
3038 const Value *DstSV, uint64_t DstSVOff,
3039 const Value *SrcSV, uint64_t SrcSVOff) {
3041 // Check to see if we should lower the memcpy to loads and stores first.
3042 // For cases within the target-specified limits, this is the best choice.
3043 ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
3045 // Memcpy with size zero? Just return the original chain.
3046 if (ConstantSize->isNullValue())
3050 getMemcpyLoadsAndStores(*this, Chain, Dst, Src, ConstantSize->getValue(),
3051 Align, false, DstSV, DstSVOff, SrcSV, SrcSVOff);
3056 // Then check to see if we should lower the memcpy with target-specific
3057 // code. If the target chooses to do this, this is the next best.
3059 TLI.EmitTargetCodeForMemcpy(*this, Chain, Dst, Src, Size, Align,
3061 DstSV, DstSVOff, SrcSV, SrcSVOff);
3065 // If we really need inline code and the target declined to provide it,
3066 // use a (potentially long) sequence of loads and stores.
3068 assert(ConstantSize && "AlwaysInline requires a constant size!");
3069 return getMemcpyLoadsAndStores(*this, Chain, Dst, Src,
3070 ConstantSize->getValue(), Align, true,
3071 DstSV, DstSVOff, SrcSV, SrcSVOff);
3074 // Emit a library call.
3075 TargetLowering::ArgListTy Args;
3076 TargetLowering::ArgListEntry Entry;
3077 Entry.Ty = TLI.getTargetData()->getIntPtrType();
3078 Entry.Node = Dst; Args.push_back(Entry);
3079 Entry.Node = Src; Args.push_back(Entry);
3080 Entry.Node = Size; Args.push_back(Entry);
3081 std::pair<SDValue,SDValue> CallResult =
3082 TLI.LowerCallTo(Chain, Type::VoidTy,
3083 false, false, false, CallingConv::C, false,
3084 getExternalSymbol("memcpy", TLI.getPointerTy()),
3086 return CallResult.second;
3089 SDValue SelectionDAG::getMemmove(SDValue Chain, SDValue Dst,
3090 SDValue Src, SDValue Size,
3092 const Value *DstSV, uint64_t DstSVOff,
3093 const Value *SrcSV, uint64_t SrcSVOff) {
3095 // Check to see if we should lower the memmove to loads and stores first.
3096 // For cases within the target-specified limits, this is the best choice.
3097 ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
3099 // Memmove with size zero? Just return the original chain.
3100 if (ConstantSize->isNullValue())
3104 getMemmoveLoadsAndStores(*this, Chain, Dst, Src, ConstantSize->getValue(),
3105 Align, false, DstSV, DstSVOff, SrcSV, SrcSVOff);
3110 // Then check to see if we should lower the memmove with target-specific
3111 // code. If the target chooses to do this, this is the next best.
3113 TLI.EmitTargetCodeForMemmove(*this, Chain, Dst, Src, Size, Align,
3114 DstSV, DstSVOff, SrcSV, SrcSVOff);
3118 // Emit a library call.
3119 TargetLowering::ArgListTy Args;
3120 TargetLowering::ArgListEntry Entry;
3121 Entry.Ty = TLI.getTargetData()->getIntPtrType();
3122 Entry.Node = Dst; Args.push_back(Entry);
3123 Entry.Node = Src; Args.push_back(Entry);
3124 Entry.Node = Size; Args.push_back(Entry);
3125 std::pair<SDValue,SDValue> CallResult =
3126 TLI.LowerCallTo(Chain, Type::VoidTy,
3127 false, false, false, CallingConv::C, false,
3128 getExternalSymbol("memmove", TLI.getPointerTy()),
3130 return CallResult.second;
3133 SDValue SelectionDAG::getMemset(SDValue Chain, SDValue Dst,
3134 SDValue Src, SDValue Size,
3136 const Value *DstSV, uint64_t DstSVOff) {
3138 // Check to see if we should lower the memset to stores first.
3139 // For cases within the target-specified limits, this is the best choice.
3140 ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
3142 // Memset with size zero? Just return the original chain.
3143 if (ConstantSize->isNullValue())
3147 getMemsetStores(*this, Chain, Dst, Src, ConstantSize->getValue(), Align,
3153 // Then check to see if we should lower the memset with target-specific
3154 // code. If the target chooses to do this, this is the next best.
3156 TLI.EmitTargetCodeForMemset(*this, Chain, Dst, Src, Size, Align,
3161 // Emit a library call.
3162 const Type *IntPtrTy = TLI.getTargetData()->getIntPtrType();
3163 TargetLowering::ArgListTy Args;
3164 TargetLowering::ArgListEntry Entry;
3165 Entry.Node = Dst; Entry.Ty = IntPtrTy;
3166 Args.push_back(Entry);
3167 // Extend or truncate the argument to be an i32 value for the call.
3168 if (Src.getValueType().bitsGT(MVT::i32))
3169 Src = getNode(ISD::TRUNCATE, MVT::i32, Src);
3171 Src = getNode(ISD::ZERO_EXTEND, MVT::i32, Src);
3172 Entry.Node = Src; Entry.Ty = Type::Int32Ty; Entry.isSExt = true;
3173 Args.push_back(Entry);
3174 Entry.Node = Size; Entry.Ty = IntPtrTy; Entry.isSExt = false;
3175 Args.push_back(Entry);
3176 std::pair<SDValue,SDValue> CallResult =
3177 TLI.LowerCallTo(Chain, Type::VoidTy,
3178 false, false, false, CallingConv::C, false,
3179 getExternalSymbol("memset", TLI.getPointerTy()),
3181 return CallResult.second;
3184 SDValue SelectionDAG::getAtomic(unsigned Opcode, SDValue Chain,
3185 SDValue Ptr, SDValue Cmp,
3186 SDValue Swp, const Value* PtrVal,
3187 unsigned Alignment) {
3188 assert((Opcode == ISD::ATOMIC_CMP_SWAP_8 ||
3189 Opcode == ISD::ATOMIC_CMP_SWAP_16 ||
3190 Opcode == ISD::ATOMIC_CMP_SWAP_32 ||
3191 Opcode == ISD::ATOMIC_CMP_SWAP_64) && "Invalid Atomic Op");
3192 assert(Cmp.getValueType() == Swp.getValueType() && "Invalid Atomic Op Types");
3194 MVT VT = Cmp.getValueType();
3196 if (Alignment == 0) // Ensure that codegen never sees alignment 0
3197 Alignment = getMVTAlignment(VT);
3199 SDVTList VTs = getVTList(VT, MVT::Other);
3200 FoldingSetNodeID ID;
3201 SDValue Ops[] = {Chain, Ptr, Cmp, Swp};
3202 AddNodeIDNode(ID, Opcode, VTs, Ops, 4);
3204 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3205 return SDValue(E, 0);
3206 SDNode* N = NodeAllocator.Allocate<AtomicSDNode>();
3207 new (N) AtomicSDNode(Opcode, VTs, Chain, Ptr, Cmp, Swp, PtrVal, Alignment);
3208 CSEMap.InsertNode(N, IP);
3209 AllNodes.push_back(N);
3210 return SDValue(N, 0);
3213 SDValue SelectionDAG::getAtomic(unsigned Opcode, SDValue Chain,
3214 SDValue Ptr, SDValue Val,
3215 const Value* PtrVal,
3216 unsigned Alignment) {
3217 assert((Opcode == ISD::ATOMIC_LOAD_ADD_8 ||
3218 Opcode == ISD::ATOMIC_LOAD_SUB_8 ||
3219 Opcode == ISD::ATOMIC_LOAD_AND_8 ||
3220 Opcode == ISD::ATOMIC_LOAD_OR_8 ||
3221 Opcode == ISD::ATOMIC_LOAD_XOR_8 ||
3222 Opcode == ISD::ATOMIC_LOAD_NAND_8 ||
3223 Opcode == ISD::ATOMIC_LOAD_MIN_8 ||
3224 Opcode == ISD::ATOMIC_LOAD_MAX_8 ||
3225 Opcode == ISD::ATOMIC_LOAD_UMIN_8 ||
3226 Opcode == ISD::ATOMIC_LOAD_UMAX_8 ||
3227 Opcode == ISD::ATOMIC_SWAP_8 ||
3228 Opcode == ISD::ATOMIC_LOAD_ADD_16 ||
3229 Opcode == ISD::ATOMIC_LOAD_SUB_16 ||
3230 Opcode == ISD::ATOMIC_LOAD_AND_16 ||
3231 Opcode == ISD::ATOMIC_LOAD_OR_16 ||
3232 Opcode == ISD::ATOMIC_LOAD_XOR_16 ||
3233 Opcode == ISD::ATOMIC_LOAD_NAND_16 ||
3234 Opcode == ISD::ATOMIC_LOAD_MIN_16 ||
3235 Opcode == ISD::ATOMIC_LOAD_MAX_16 ||
3236 Opcode == ISD::ATOMIC_LOAD_UMIN_16 ||
3237 Opcode == ISD::ATOMIC_LOAD_UMAX_16 ||
3238 Opcode == ISD::ATOMIC_SWAP_16 ||
3239 Opcode == ISD::ATOMIC_LOAD_ADD_32 ||
3240 Opcode == ISD::ATOMIC_LOAD_SUB_32 ||
3241 Opcode == ISD::ATOMIC_LOAD_AND_32 ||
3242 Opcode == ISD::ATOMIC_LOAD_OR_32 ||
3243 Opcode == ISD::ATOMIC_LOAD_XOR_32 ||
3244 Opcode == ISD::ATOMIC_LOAD_NAND_32 ||
3245 Opcode == ISD::ATOMIC_LOAD_MIN_32 ||
3246 Opcode == ISD::ATOMIC_LOAD_MAX_32 ||
3247 Opcode == ISD::ATOMIC_LOAD_UMIN_32 ||
3248 Opcode == ISD::ATOMIC_LOAD_UMAX_32 ||
3249 Opcode == ISD::ATOMIC_SWAP_32 ||
3250 Opcode == ISD::ATOMIC_LOAD_ADD_64 ||
3251 Opcode == ISD::ATOMIC_LOAD_SUB_64 ||
3252 Opcode == ISD::ATOMIC_LOAD_AND_64 ||
3253 Opcode == ISD::ATOMIC_LOAD_OR_64 ||
3254 Opcode == ISD::ATOMIC_LOAD_XOR_64 ||
3255 Opcode == ISD::ATOMIC_LOAD_NAND_64 ||
3256 Opcode == ISD::ATOMIC_LOAD_MIN_64 ||
3257 Opcode == ISD::ATOMIC_LOAD_MAX_64 ||
3258 Opcode == ISD::ATOMIC_LOAD_UMIN_64 ||
3259 Opcode == ISD::ATOMIC_LOAD_UMAX_64 ||
3260 Opcode == ISD::ATOMIC_SWAP_64) && "Invalid Atomic Op");
3262 MVT VT = Val.getValueType();
3264 if (Alignment == 0) // Ensure that codegen never sees alignment 0
3265 Alignment = getMVTAlignment(VT);
3267 SDVTList VTs = getVTList(VT, MVT::Other);
3268 FoldingSetNodeID ID;
3269 SDValue Ops[] = {Chain, Ptr, Val};
3270 AddNodeIDNode(ID, Opcode, VTs, Ops, 3);
3272 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3273 return SDValue(E, 0);
3274 SDNode* N = NodeAllocator.Allocate<AtomicSDNode>();
3275 new (N) AtomicSDNode(Opcode, VTs, Chain, Ptr, Val, PtrVal, Alignment);
3276 CSEMap.InsertNode(N, IP);
3277 AllNodes.push_back(N);
3278 return SDValue(N, 0);
3281 /// getMergeValues - Create a MERGE_VALUES node from the given operands.
3282 /// Allowed to return something different (and simpler) if Simplify is true.
3283 SDValue SelectionDAG::getMergeValues(const SDValue *Ops, unsigned NumOps,
3285 if (Simplify && NumOps == 1)
3288 SmallVector<MVT, 4> VTs;
3289 VTs.reserve(NumOps);
3290 for (unsigned i = 0; i < NumOps; ++i)
3291 VTs.push_back(Ops[i].getValueType());
3292 return getNode(ISD::MERGE_VALUES, getVTList(&VTs[0], NumOps), Ops, NumOps);
3296 SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType,
3297 MVT VT, SDValue Chain,
3298 SDValue Ptr, SDValue Offset,
3299 const Value *SV, int SVOffset, MVT EVT,
3300 bool isVolatile, unsigned Alignment) {
3301 if (Alignment == 0) // Ensure that codegen never sees alignment 0
3302 Alignment = getMVTAlignment(VT);
3305 ExtType = ISD::NON_EXTLOAD;
3306 } else if (ExtType == ISD::NON_EXTLOAD) {
3307 assert(VT == EVT && "Non-extending load from different memory type!");
3311 assert(EVT.getVectorNumElements() == VT.getVectorNumElements() &&
3312 "Invalid vector extload!");
3314 assert(EVT.bitsLT(VT) &&
3315 "Should only be an extending load, not truncating!");
3316 assert((ExtType == ISD::EXTLOAD || VT.isInteger()) &&
3317 "Cannot sign/zero extend a FP/Vector load!");
3318 assert(VT.isInteger() == EVT.isInteger() &&
3319 "Cannot convert from FP to Int or Int -> FP!");
3322 bool Indexed = AM != ISD::UNINDEXED;
3323 assert((Indexed || Offset.getOpcode() == ISD::UNDEF) &&
3324 "Unindexed load with an offset!");
3326 SDVTList VTs = Indexed ?
3327 getVTList(VT, Ptr.getValueType(), MVT::Other) : getVTList(VT, MVT::Other);
3328 SDValue Ops[] = { Chain, Ptr, Offset };
3329 FoldingSetNodeID ID;
3330 AddNodeIDNode(ID, ISD::LOAD, VTs, Ops, 3);
3332 ID.AddInteger(ExtType);
3333 ID.AddInteger(EVT.getRawBits());
3334 ID.AddInteger(encodeMemSDNodeFlags(isVolatile, Alignment));
3336 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3337 return SDValue(E, 0);
3338 SDNode *N = NodeAllocator.Allocate<LoadSDNode>();
3339 new (N) LoadSDNode(Ops, VTs, AM, ExtType, EVT, SV, SVOffset,
3340 Alignment, isVolatile);
3341 CSEMap.InsertNode(N, IP);
3342 AllNodes.push_back(N);
3343 return SDValue(N, 0);
3346 SDValue SelectionDAG::getLoad(MVT VT,
3347 SDValue Chain, SDValue Ptr,
3348 const Value *SV, int SVOffset,
3349 bool isVolatile, unsigned Alignment) {
3350 SDValue Undef = getNode(ISD::UNDEF, Ptr.getValueType());
3351 return getLoad(ISD::UNINDEXED, ISD::NON_EXTLOAD, VT, Chain, Ptr, Undef,
3352 SV, SVOffset, VT, isVolatile, Alignment);
3355 SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, MVT VT,
3356 SDValue Chain, SDValue Ptr,
3358 int SVOffset, MVT EVT,
3359 bool isVolatile, unsigned Alignment) {
3360 SDValue Undef = getNode(ISD::UNDEF, Ptr.getValueType());
3361 return getLoad(ISD::UNINDEXED, ExtType, VT, Chain, Ptr, Undef,
3362 SV, SVOffset, EVT, isVolatile, Alignment);
3366 SelectionDAG::getIndexedLoad(SDValue OrigLoad, SDValue Base,
3367 SDValue Offset, ISD::MemIndexedMode AM) {
3368 LoadSDNode *LD = cast<LoadSDNode>(OrigLoad);
3369 assert(LD->getOffset().getOpcode() == ISD::UNDEF &&
3370 "Load is already a indexed load!");
3371 return getLoad(AM, LD->getExtensionType(), OrigLoad.getValueType(),
3372 LD->getChain(), Base, Offset, LD->getSrcValue(),
3373 LD->getSrcValueOffset(), LD->getMemoryVT(),
3374 LD->isVolatile(), LD->getAlignment());
3377 SDValue SelectionDAG::getStore(SDValue Chain, SDValue Val,
3378 SDValue Ptr, const Value *SV, int SVOffset,
3379 bool isVolatile, unsigned Alignment) {
3380 MVT VT = Val.getValueType();
3382 if (Alignment == 0) // Ensure that codegen never sees alignment 0
3383 Alignment = getMVTAlignment(VT);
3385 SDVTList VTs = getVTList(MVT::Other);
3386 SDValue Undef = getNode(ISD::UNDEF, Ptr.getValueType());
3387 SDValue Ops[] = { Chain, Val, Ptr, Undef };
3388 FoldingSetNodeID ID;
3389 AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
3390 ID.AddInteger(ISD::UNINDEXED);
3391 ID.AddInteger(false);
3392 ID.AddInteger(VT.getRawBits());
3393 ID.AddInteger(encodeMemSDNodeFlags(isVolatile, Alignment));
3395 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3396 return SDValue(E, 0);
3397 SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
3398 new (N) StoreSDNode(Ops, VTs, ISD::UNINDEXED, false,
3399 VT, SV, SVOffset, Alignment, isVolatile);
3400 CSEMap.InsertNode(N, IP);
3401 AllNodes.push_back(N);
3402 return SDValue(N, 0);
3405 SDValue SelectionDAG::getTruncStore(SDValue Chain, SDValue Val,
3406 SDValue Ptr, const Value *SV,
3407 int SVOffset, MVT SVT,
3408 bool isVolatile, unsigned Alignment) {
3409 MVT VT = Val.getValueType();
3412 return getStore(Chain, Val, Ptr, SV, SVOffset, isVolatile, Alignment);
3414 assert(VT.bitsGT(SVT) && "Not a truncation?");
3415 assert(VT.isInteger() == SVT.isInteger() &&
3416 "Can't do FP-INT conversion!");
3418 if (Alignment == 0) // Ensure that codegen never sees alignment 0
3419 Alignment = getMVTAlignment(VT);
3421 SDVTList VTs = getVTList(MVT::Other);
3422 SDValue Undef = getNode(ISD::UNDEF, Ptr.getValueType());
3423 SDValue Ops[] = { Chain, Val, Ptr, Undef };
3424 FoldingSetNodeID ID;
3425 AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
3426 ID.AddInteger(ISD::UNINDEXED);
3428 ID.AddInteger(SVT.getRawBits());
3429 ID.AddInteger(encodeMemSDNodeFlags(isVolatile, Alignment));
3431 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3432 return SDValue(E, 0);
3433 SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
3434 new (N) StoreSDNode(Ops, VTs, ISD::UNINDEXED, true,
3435 SVT, SV, SVOffset, Alignment, isVolatile);
3436 CSEMap.InsertNode(N, IP);
3437 AllNodes.push_back(N);
3438 return SDValue(N, 0);
3442 SelectionDAG::getIndexedStore(SDValue OrigStore, SDValue Base,
3443 SDValue Offset, ISD::MemIndexedMode AM) {
3444 StoreSDNode *ST = cast<StoreSDNode>(OrigStore);
3445 assert(ST->getOffset().getOpcode() == ISD::UNDEF &&
3446 "Store is already a indexed store!");
3447 SDVTList VTs = getVTList(Base.getValueType(), MVT::Other);
3448 SDValue Ops[] = { ST->getChain(), ST->getValue(), Base, Offset };
3449 FoldingSetNodeID ID;
3450 AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
3452 ID.AddInteger(ST->isTruncatingStore());
3453 ID.AddInteger(ST->getMemoryVT().getRawBits());
3454 ID.AddInteger(ST->getRawFlags());
3456 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3457 return SDValue(E, 0);
3458 SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
3459 new (N) StoreSDNode(Ops, VTs, AM,
3460 ST->isTruncatingStore(), ST->getMemoryVT(),
3461 ST->getSrcValue(), ST->getSrcValueOffset(),
3462 ST->getAlignment(), ST->isVolatile());
3463 CSEMap.InsertNode(N, IP);
3464 AllNodes.push_back(N);
3465 return SDValue(N, 0);
3468 SDValue SelectionDAG::getVAArg(MVT VT,
3469 SDValue Chain, SDValue Ptr,
3471 SDValue Ops[] = { Chain, Ptr, SV };
3472 return getNode(ISD::VAARG, getVTList(VT, MVT::Other), Ops, 3);
3475 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
3476 const SDUse *Ops, unsigned NumOps) {
3478 case 0: return getNode(Opcode, VT);
3479 case 1: return getNode(Opcode, VT, Ops[0]);
3480 case 2: return getNode(Opcode, VT, Ops[0], Ops[1]);
3481 case 3: return getNode(Opcode, VT, Ops[0], Ops[1], Ops[2]);
3485 // Copy from an SDUse array into an SDValue array for use with
3486 // the regular getNode logic.
3487 SmallVector<SDValue, 8> NewOps(Ops, Ops + NumOps);
3488 return getNode(Opcode, VT, &NewOps[0], NumOps);
3491 SDValue SelectionDAG::getNode(unsigned Opcode, MVT VT,
3492 const SDValue *Ops, unsigned NumOps) {
3494 case 0: return getNode(Opcode, VT);
3495 case 1: return getNode(Opcode, VT, Ops[0]);
3496 case 2: return getNode(Opcode, VT, Ops[0], Ops[1]);
3497 case 3: return getNode(Opcode, VT, Ops[0], Ops[1], Ops[2]);
3503 case ISD::SELECT_CC: {
3504 assert(NumOps == 5 && "SELECT_CC takes 5 operands!");
3505 assert(Ops[0].getValueType() == Ops[1].getValueType() &&
3506 "LHS and RHS of condition must have same type!");
3507 assert(Ops[2].getValueType() == Ops[3].getValueType() &&
3508 "True and False arms of SelectCC must have same type!");
3509 assert(Ops[2].getValueType() == VT &&
3510 "select_cc node must be of same type as true and false value!");
3514 assert(NumOps == 5 && "BR_CC takes 5 operands!");
3515 assert(Ops[2].getValueType() == Ops[3].getValueType() &&
3516 "LHS/RHS of comparison should match types!");
3523 SDVTList VTs = getVTList(VT);
3524 if (VT != MVT::Flag) {
3525 FoldingSetNodeID ID;
3526 AddNodeIDNode(ID, Opcode, VTs, Ops, NumOps);
3528 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3529 return SDValue(E, 0);
3530 N = NodeAllocator.Allocate<SDNode>();
3531 new (N) SDNode(Opcode, VTs, Ops, NumOps);
3532 CSEMap.InsertNode(N, IP);
3534 N = NodeAllocator.Allocate<SDNode>();
3535 new (N) SDNode(Opcode, VTs, Ops, NumOps);
3537 AllNodes.push_back(N);
3541 return SDValue(N, 0);
3544 SDValue SelectionDAG::getNode(unsigned Opcode,
3545 const std::vector<MVT> &ResultTys,
3546 const SDValue *Ops, unsigned NumOps) {
3547 return getNode(Opcode, getNodeValueTypes(ResultTys), ResultTys.size(),
3551 SDValue SelectionDAG::getNode(unsigned Opcode,
3552 const MVT *VTs, unsigned NumVTs,
3553 const SDValue *Ops, unsigned NumOps) {
3555 return getNode(Opcode, VTs[0], Ops, NumOps);
3556 return getNode(Opcode, makeVTList(VTs, NumVTs), Ops, NumOps);
3559 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3560 const SDValue *Ops, unsigned NumOps) {
3561 if (VTList.NumVTs == 1)
3562 return getNode(Opcode, VTList.VTs[0], Ops, NumOps);
3565 // FIXME: figure out how to safely handle things like
3566 // int foo(int x) { return 1 << (x & 255); }
3567 // int bar() { return foo(256); }
3569 case ISD::SRA_PARTS:
3570 case ISD::SRL_PARTS:
3571 case ISD::SHL_PARTS:
3572 if (N3.getOpcode() == ISD::SIGN_EXTEND_INREG &&
3573 cast<VTSDNode>(N3.getOperand(1))->getVT() != MVT::i1)
3574 return getNode(Opcode, VT, N1, N2, N3.getOperand(0));
3575 else if (N3.getOpcode() == ISD::AND)
3576 if (ConstantSDNode *AndRHS = dyn_cast<ConstantSDNode>(N3.getOperand(1))) {
3577 // If the and is only masking out bits that cannot effect the shift,
3578 // eliminate the and.
3579 unsigned NumBits = VT.getSizeInBits()*2;
3580 if ((AndRHS->getValue() & (NumBits-1)) == NumBits-1)
3581 return getNode(Opcode, VT, N1, N2, N3.getOperand(0));
3587 // Memoize the node unless it returns a flag.
3589 if (VTList.VTs[VTList.NumVTs-1] != MVT::Flag) {
3590 FoldingSetNodeID ID;
3591 AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
3593 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
3594 return SDValue(E, 0);
3596 N = NodeAllocator.Allocate<UnarySDNode>();
3597 new (N) UnarySDNode(Opcode, VTList, Ops[0]);
3598 } else if (NumOps == 2) {
3599 N = NodeAllocator.Allocate<BinarySDNode>();
3600 new (N) BinarySDNode(Opcode, VTList, Ops[0], Ops[1]);
3601 } else if (NumOps == 3) {
3602 N = NodeAllocator.Allocate<TernarySDNode>();
3603 new (N) TernarySDNode(Opcode, VTList, Ops[0], Ops[1], Ops[2]);
3605 N = NodeAllocator.Allocate<SDNode>();
3606 new (N) SDNode(Opcode, VTList, Ops, NumOps);
3608 CSEMap.InsertNode(N, IP);
3611 N = NodeAllocator.Allocate<UnarySDNode>();
3612 new (N) UnarySDNode(Opcode, VTList, Ops[0]);
3613 } else if (NumOps == 2) {
3614 N = NodeAllocator.Allocate<BinarySDNode>();
3615 new (N) BinarySDNode(Opcode, VTList, Ops[0], Ops[1]);
3616 } else if (NumOps == 3) {
3617 N = NodeAllocator.Allocate<TernarySDNode>();
3618 new (N) TernarySDNode(Opcode, VTList, Ops[0], Ops[1], Ops[2]);
3620 N = NodeAllocator.Allocate<SDNode>();
3621 new (N) SDNode(Opcode, VTList, Ops, NumOps);
3624 AllNodes.push_back(N);
3628 return SDValue(N, 0);
3631 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList) {
3632 return getNode(Opcode, VTList, 0, 0);
3635 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3637 SDValue Ops[] = { N1 };
3638 return getNode(Opcode, VTList, Ops, 1);
3641 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3642 SDValue N1, SDValue N2) {
3643 SDValue Ops[] = { N1, N2 };
3644 return getNode(Opcode, VTList, Ops, 2);
3647 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3648 SDValue N1, SDValue N2, SDValue N3) {
3649 SDValue Ops[] = { N1, N2, N3 };
3650 return getNode(Opcode, VTList, Ops, 3);
3653 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3654 SDValue N1, SDValue N2, SDValue N3,
3656 SDValue Ops[] = { N1, N2, N3, N4 };
3657 return getNode(Opcode, VTList, Ops, 4);
3660 SDValue SelectionDAG::getNode(unsigned Opcode, SDVTList VTList,
3661 SDValue N1, SDValue N2, SDValue N3,
3662 SDValue N4, SDValue N5) {
3663 SDValue Ops[] = { N1, N2, N3, N4, N5 };
3664 return getNode(Opcode, VTList, Ops, 5);
3667 SDVTList SelectionDAG::getVTList(MVT VT) {
3668 return makeVTList(SDNode::getValueTypeList(VT), 1);
3671 SDVTList SelectionDAG::getVTList(MVT VT1, MVT VT2) {
3672 for (std::vector<SDVTList>::reverse_iterator I = VTList.rbegin(),
3673 E = VTList.rend(); I != E; ++I)
3674 if (I->NumVTs == 2 && I->VTs[0] == VT1 && I->VTs[1] == VT2)
3677 MVT *Array = Allocator.Allocate<MVT>(2);
3680 SDVTList Result = makeVTList(Array, 2);
3681 VTList.push_back(Result);
3685 SDVTList SelectionDAG::getVTList(MVT VT1, MVT VT2, MVT VT3) {
3686 for (std::vector<SDVTList>::reverse_iterator I = VTList.rbegin(),
3687 E = VTList.rend(); I != E; ++I)
3688 if (I->NumVTs == 3 && I->VTs[0] == VT1 && I->VTs[1] == VT2 &&
3692 MVT *Array = Allocator.Allocate<MVT>(3);
3696 SDVTList Result = makeVTList(Array, 3);
3697 VTList.push_back(Result);
3701 SDVTList SelectionDAG::getVTList(const MVT *VTs, unsigned NumVTs) {
3703 case 0: assert(0 && "Cannot have nodes without results!");
3704 case 1: return getVTList(VTs[0]);
3705 case 2: return getVTList(VTs[0], VTs[1]);
3706 case 3: return getVTList(VTs[0], VTs[1], VTs[2]);
3710 for (std::vector<SDVTList>::reverse_iterator I = VTList.rbegin(),
3711 E = VTList.rend(); I != E; ++I) {
3712 if (I->NumVTs != NumVTs || VTs[0] != I->VTs[0] || VTs[1] != I->VTs[1])
3715 bool NoMatch = false;
3716 for (unsigned i = 2; i != NumVTs; ++i)
3717 if (VTs[i] != I->VTs[i]) {
3725 MVT *Array = Allocator.Allocate<MVT>(NumVTs);
3726 std::copy(VTs, VTs+NumVTs, Array);
3727 SDVTList Result = makeVTList(Array, NumVTs);
3728 VTList.push_back(Result);
3733 /// UpdateNodeOperands - *Mutate* the specified node in-place to have the
3734 /// specified operands. If the resultant node already exists in the DAG,
3735 /// this does not modify the specified node, instead it returns the node that
3736 /// already exists. If the resultant node does not exist in the DAG, the
3737 /// input node is returned. As a degenerate case, if you specify the same
3738 /// input operands as the node already has, the input node is returned.
3739 SDValue SelectionDAG::UpdateNodeOperands(SDValue InN, SDValue Op) {
3740 SDNode *N = InN.Val;
3741 assert(N->getNumOperands() == 1 && "Update with wrong number of operands");
3743 // Check to see if there is no change.
3744 if (Op == N->getOperand(0)) return InN;
3746 // See if the modified node already exists.
3747 void *InsertPos = 0;
3748 if (SDNode *Existing = FindModifiedNodeSlot(N, Op, InsertPos))
3749 return SDValue(Existing, InN.getResNo());
3751 // Nope it doesn't. Remove the node from its current place in the maps.
3753 RemoveNodeFromCSEMaps(N);
3755 // Now we update the operands.
3756 N->OperandList[0].getVal()->removeUser(0, N);
3757 N->OperandList[0] = Op;
3758 N->OperandList[0].setUser(N);
3759 Op.Val->addUser(0, N);
3761 // If this gets put into a CSE map, add it.
3762 if (InsertPos) CSEMap.InsertNode(N, InsertPos);
3766 SDValue SelectionDAG::
3767 UpdateNodeOperands(SDValue InN, SDValue Op1, SDValue Op2) {
3768 SDNode *N = InN.Val;
3769 assert(N->getNumOperands() == 2 && "Update with wrong number of operands");
3771 // Check to see if there is no change.
3772 if (Op1 == N->getOperand(0) && Op2 == N->getOperand(1))
3773 return InN; // No operands changed, just return the input node.
3775 // See if the modified node already exists.
3776 void *InsertPos = 0;
3777 if (SDNode *Existing = FindModifiedNodeSlot(N, Op1, Op2, InsertPos))
3778 return SDValue(Existing, InN.getResNo());
3780 // Nope it doesn't. Remove the node from its current place in the maps.
3782 RemoveNodeFromCSEMaps(N);
3784 // Now we update the operands.
3785 if (N->OperandList[0] != Op1) {
3786 N->OperandList[0].getVal()->removeUser(0, N);
3787 N->OperandList[0] = Op1;
3788 N->OperandList[0].setUser(N);
3789 Op1.Val->addUser(0, N);
3791 if (N->OperandList[1] != Op2) {
3792 N->OperandList[1].getVal()->removeUser(1, N);
3793 N->OperandList[1] = Op2;
3794 N->OperandList[1].setUser(N);
3795 Op2.Val->addUser(1, N);
3798 // If this gets put into a CSE map, add it.
3799 if (InsertPos) CSEMap.InsertNode(N, InsertPos);
3803 SDValue SelectionDAG::
3804 UpdateNodeOperands(SDValue N, SDValue Op1, SDValue Op2, SDValue Op3) {
3805 SDValue Ops[] = { Op1, Op2, Op3 };
3806 return UpdateNodeOperands(N, Ops, 3);
3809 SDValue SelectionDAG::
3810 UpdateNodeOperands(SDValue N, SDValue Op1, SDValue Op2,
3811 SDValue Op3, SDValue Op4) {
3812 SDValue Ops[] = { Op1, Op2, Op3, Op4 };
3813 return UpdateNodeOperands(N, Ops, 4);
3816 SDValue SelectionDAG::
3817 UpdateNodeOperands(SDValue N, SDValue Op1, SDValue Op2,
3818 SDValue Op3, SDValue Op4, SDValue Op5) {
3819 SDValue Ops[] = { Op1, Op2, Op3, Op4, Op5 };
3820 return UpdateNodeOperands(N, Ops, 5);
3823 SDValue SelectionDAG::
3824 UpdateNodeOperands(SDValue InN, const SDValue *Ops, unsigned NumOps) {
3825 SDNode *N = InN.Val;
3826 assert(N->getNumOperands() == NumOps &&
3827 "Update with wrong number of operands");
3829 // Check to see if there is no change.
3830 bool AnyChange = false;
3831 for (unsigned i = 0; i != NumOps; ++i) {
3832 if (Ops[i] != N->getOperand(i)) {
3838 // No operands changed, just return the input node.
3839 if (!AnyChange) return InN;
3841 // See if the modified node already exists.
3842 void *InsertPos = 0;
3843 if (SDNode *Existing = FindModifiedNodeSlot(N, Ops, NumOps, InsertPos))
3844 return SDValue(Existing, InN.getResNo());
3846 // Nope it doesn't. Remove the node from its current place in the maps.
3848 RemoveNodeFromCSEMaps(N);
3850 // Now we update the operands.
3851 for (unsigned i = 0; i != NumOps; ++i) {
3852 if (N->OperandList[i] != Ops[i]) {
3853 N->OperandList[i].getVal()->removeUser(i, N);
3854 N->OperandList[i] = Ops[i];
3855 N->OperandList[i].setUser(N);
3856 Ops[i].Val->addUser(i, N);
3860 // If this gets put into a CSE map, add it.
3861 if (InsertPos) CSEMap.InsertNode(N, InsertPos);
3865 /// DropOperands - Release the operands and set this node to have
3867 void SDNode::DropOperands() {
3868 // Unlike the code in MorphNodeTo that does this, we don't need to
3869 // watch for dead nodes here.
3870 for (op_iterator I = op_begin(), E = op_end(); I != E; ++I)
3871 I->getVal()->removeUser(std::distance(op_begin(), I), this);
3876 /// SelectNodeTo - These are wrappers around MorphNodeTo that accept a
3879 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3881 SDVTList VTs = getVTList(VT);
3882 return SelectNodeTo(N, MachineOpc, VTs, 0, 0);
3885 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3886 MVT VT, SDValue Op1) {
3887 SDVTList VTs = getVTList(VT);
3888 SDValue Ops[] = { Op1 };
3889 return SelectNodeTo(N, MachineOpc, VTs, Ops, 1);
3892 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3893 MVT VT, SDValue Op1,
3895 SDVTList VTs = getVTList(VT);
3896 SDValue Ops[] = { Op1, Op2 };
3897 return SelectNodeTo(N, MachineOpc, VTs, Ops, 2);
3900 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3901 MVT VT, SDValue Op1,
3902 SDValue Op2, SDValue Op3) {
3903 SDVTList VTs = getVTList(VT);
3904 SDValue Ops[] = { Op1, Op2, Op3 };
3905 return SelectNodeTo(N, MachineOpc, VTs, Ops, 3);
3908 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3909 MVT VT, const SDValue *Ops,
3911 SDVTList VTs = getVTList(VT);
3912 return SelectNodeTo(N, MachineOpc, VTs, Ops, NumOps);
3915 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3916 MVT VT1, MVT VT2, const SDValue *Ops,
3918 SDVTList VTs = getVTList(VT1, VT2);
3919 return SelectNodeTo(N, MachineOpc, VTs, Ops, NumOps);
3922 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3924 SDVTList VTs = getVTList(VT1, VT2);
3925 return SelectNodeTo(N, MachineOpc, VTs, (SDValue *)0, 0);
3928 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3929 MVT VT1, MVT VT2, MVT VT3,
3930 const SDValue *Ops, unsigned NumOps) {
3931 SDVTList VTs = getVTList(VT1, VT2, VT3);
3932 return SelectNodeTo(N, MachineOpc, VTs, Ops, NumOps);
3935 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3938 SDVTList VTs = getVTList(VT1, VT2);
3939 SDValue Ops[] = { Op1 };
3940 return SelectNodeTo(N, MachineOpc, VTs, Ops, 1);
3943 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3945 SDValue Op1, SDValue Op2) {
3946 SDVTList VTs = getVTList(VT1, VT2);
3947 SDValue Ops[] = { Op1, Op2 };
3948 return SelectNodeTo(N, MachineOpc, VTs, Ops, 2);
3951 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3953 SDValue Op1, SDValue Op2,
3955 SDVTList VTs = getVTList(VT1, VT2);
3956 SDValue Ops[] = { Op1, Op2, Op3 };
3957 return SelectNodeTo(N, MachineOpc, VTs, Ops, 3);
3960 SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
3961 SDVTList VTs, const SDValue *Ops,
3963 return MorphNodeTo(N, ~MachineOpc, VTs, Ops, NumOps);
3966 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
3968 SDVTList VTs = getVTList(VT);
3969 return MorphNodeTo(N, Opc, VTs, 0, 0);
3972 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
3973 MVT VT, SDValue Op1) {
3974 SDVTList VTs = getVTList(VT);
3975 SDValue Ops[] = { Op1 };
3976 return MorphNodeTo(N, Opc, VTs, Ops, 1);
3979 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
3980 MVT VT, SDValue Op1,
3982 SDVTList VTs = getVTList(VT);
3983 SDValue Ops[] = { Op1, Op2 };
3984 return MorphNodeTo(N, Opc, VTs, Ops, 2);
3987 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
3988 MVT VT, SDValue Op1,
3989 SDValue Op2, SDValue Op3) {
3990 SDVTList VTs = getVTList(VT);
3991 SDValue Ops[] = { Op1, Op2, Op3 };
3992 return MorphNodeTo(N, Opc, VTs, Ops, 3);
3995 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
3996 MVT VT, const SDValue *Ops,
3998 SDVTList VTs = getVTList(VT);
3999 return MorphNodeTo(N, Opc, VTs, Ops, NumOps);
4002 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4003 MVT VT1, MVT VT2, const SDValue *Ops,
4005 SDVTList VTs = getVTList(VT1, VT2);
4006 return MorphNodeTo(N, Opc, VTs, Ops, NumOps);
4009 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4011 SDVTList VTs = getVTList(VT1, VT2);
4012 return MorphNodeTo(N, Opc, VTs, (SDValue *)0, 0);
4015 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4016 MVT VT1, MVT VT2, MVT VT3,
4017 const SDValue *Ops, unsigned NumOps) {
4018 SDVTList VTs = getVTList(VT1, VT2, VT3);
4019 return MorphNodeTo(N, Opc, VTs, Ops, NumOps);
4022 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4025 SDVTList VTs = getVTList(VT1, VT2);
4026 SDValue Ops[] = { Op1 };
4027 return MorphNodeTo(N, Opc, VTs, Ops, 1);
4030 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4032 SDValue Op1, SDValue Op2) {
4033 SDVTList VTs = getVTList(VT1, VT2);
4034 SDValue Ops[] = { Op1, Op2 };
4035 return MorphNodeTo(N, Opc, VTs, Ops, 2);
4038 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4040 SDValue Op1, SDValue Op2,
4042 SDVTList VTs = getVTList(VT1, VT2);
4043 SDValue Ops[] = { Op1, Op2, Op3 };
4044 return MorphNodeTo(N, Opc, VTs, Ops, 3);
4047 /// MorphNodeTo - These *mutate* the specified node to have the specified
4048 /// return type, opcode, and operands.
4050 /// Note that MorphNodeTo returns the resultant node. If there is already a
4051 /// node of the specified opcode and operands, it returns that node instead of
4052 /// the current one.
4054 /// Using MorphNodeTo is faster than creating a new node and swapping it in
4055 /// with ReplaceAllUsesWith both because it often avoids allocating a new
4056 /// node, and because it doesn't require CSE recalulation for any of
4057 /// the node's users.
4059 SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
4060 SDVTList VTs, const SDValue *Ops,
4062 // If an identical node already exists, use it.
4064 if (VTs.VTs[VTs.NumVTs-1] != MVT::Flag) {
4065 FoldingSetNodeID ID;
4066 AddNodeIDNode(ID, Opc, VTs, Ops, NumOps);
4067 if (SDNode *ON = CSEMap.FindNodeOrInsertPos(ID, IP))
4071 RemoveNodeFromCSEMaps(N);
4073 // Start the morphing.
4075 N->ValueList = VTs.VTs;
4076 N->NumValues = VTs.NumVTs;
4078 // Clear the operands list, updating used nodes to remove this from their
4079 // use list. Keep track of any operands that become dead as a result.
4080 SmallPtrSet<SDNode*, 16> DeadNodeSet;
4081 for (SDNode::op_iterator B = N->op_begin(), I = B, E = N->op_end();
4083 SDNode *Used = I->getVal();
4084 Used->removeUser(std::distance(B, I), N);
4085 if (Used->use_empty())
4086 DeadNodeSet.insert(Used);
4089 // If NumOps is larger than the # of operands we currently have, reallocate
4090 // the operand list.
4091 if (NumOps > N->NumOperands) {
4092 if (N->OperandsNeedDelete)
4093 delete[] N->OperandList;
4094 if (N->isMachineOpcode()) {
4095 // We're creating a final node that will live unmorphed for the
4096 // remainder of the current SelectionDAG iteration, so we can allocate
4097 // the operands directly out of a pool with no recycling metadata.
4098 N->OperandList = OperandAllocator.Allocate<SDUse>(NumOps);
4099 N->OperandsNeedDelete = false;
4101 N->OperandList = new SDUse[NumOps];
4102 N->OperandsNeedDelete = true;
4106 // Assign the new operands.
4107 N->NumOperands = NumOps;
4108 for (unsigned i = 0, e = NumOps; i != e; ++i) {
4109 N->OperandList[i] = Ops[i];
4110 N->OperandList[i].setUser(N);
4111 SDNode *ToUse = N->OperandList[i].getVal();
4112 ToUse->addUser(i, N);
4115 // Delete any nodes that are still dead after adding the uses for the
4117 SmallVector<SDNode *, 16> DeadNodes;
4118 for (SmallPtrSet<SDNode *, 16>::iterator I = DeadNodeSet.begin(),
4119 E = DeadNodeSet.end(); I != E; ++I)
4120 if ((*I)->use_empty())
4121 DeadNodes.push_back(*I);
4122 RemoveDeadNodes(DeadNodes);
4125 CSEMap.InsertNode(N, IP); // Memoize the new node.
4130 /// getTargetNode - These are used for target selectors to create a new node
4131 /// with specified return type(s), target opcode, and operands.
4133 /// Note that getTargetNode returns the resultant node. If there is already a
4134 /// node of the specified opcode and operands, it returns that node instead of
4135 /// the current one.
4136 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT) {
4137 return getNode(~Opcode, VT).Val;
4139 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT, SDValue Op1) {
4140 return getNode(~Opcode, VT, Op1).Val;
4142 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT,
4143 SDValue Op1, SDValue Op2) {
4144 return getNode(~Opcode, VT, Op1, Op2).Val;
4146 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT,
4147 SDValue Op1, SDValue Op2,
4149 return getNode(~Opcode, VT, Op1, Op2, Op3).Val;
4151 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT,
4152 const SDValue *Ops, unsigned NumOps) {
4153 return getNode(~Opcode, VT, Ops, NumOps).Val;
4155 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1, MVT VT2) {
4156 const MVT *VTs = getNodeValueTypes(VT1, VT2);
4158 return getNode(~Opcode, VTs, 2, &Op, 0).Val;
4160 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1,
4161 MVT VT2, SDValue Op1) {
4162 const MVT *VTs = getNodeValueTypes(VT1, VT2);
4163 return getNode(~Opcode, VTs, 2, &Op1, 1).Val;
4165 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1,
4166 MVT VT2, SDValue Op1,
4168 const MVT *VTs = getNodeValueTypes(VT1, VT2);
4169 SDValue Ops[] = { Op1, Op2 };
4170 return getNode(~Opcode, VTs, 2, Ops, 2).Val;
4172 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1,
4173 MVT VT2, SDValue Op1,
4174 SDValue Op2, SDValue Op3) {
4175 const MVT *VTs = getNodeValueTypes(VT1, VT2);
4176 SDValue Ops[] = { Op1, Op2, Op3 };
4177 return getNode(~Opcode, VTs, 2, Ops, 3).Val;
4179 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1, MVT VT2,
4180 const SDValue *Ops, unsigned NumOps) {
4181 const MVT *VTs = getNodeValueTypes(VT1, VT2);
4182 return getNode(~Opcode, VTs, 2, Ops, NumOps).Val;
4184 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1, MVT VT2, MVT VT3,
4185 SDValue Op1, SDValue Op2) {
4186 const MVT *VTs = getNodeValueTypes(VT1, VT2, VT3);
4187 SDValue Ops[] = { Op1, Op2 };
4188 return getNode(~Opcode, VTs, 3, Ops, 2).Val;
4190 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1, MVT VT2, MVT VT3,
4191 SDValue Op1, SDValue Op2,
4193 const MVT *VTs = getNodeValueTypes(VT1, VT2, VT3);
4194 SDValue Ops[] = { Op1, Op2, Op3 };
4195 return getNode(~Opcode, VTs, 3, Ops, 3).Val;
4197 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1, MVT VT2, MVT VT3,
4198 const SDValue *Ops, unsigned NumOps) {
4199 const MVT *VTs = getNodeValueTypes(VT1, VT2, VT3);
4200 return getNode(~Opcode, VTs, 3, Ops, NumOps).Val;
4202 SDNode *SelectionDAG::getTargetNode(unsigned Opcode, MVT VT1,
4203 MVT VT2, MVT VT3, MVT VT4,
4204 const SDValue *Ops, unsigned NumOps) {
4205 std::vector<MVT> VTList;
4206 VTList.push_back(VT1);
4207 VTList.push_back(VT2);
4208 VTList.push_back(VT3);
4209 VTList.push_back(VT4);
4210 const MVT *VTs = getNodeValueTypes(VTList);
4211 return getNode(~Opcode, VTs, 4, Ops, NumOps).Val;
4213 SDNode *SelectionDAG::getTargetNode(unsigned Opcode,
4214 const std::vector<MVT> &ResultTys,
4215 const SDValue *Ops, unsigned NumOps) {
4216 const MVT *VTs = getNodeValueTypes(ResultTys);
4217 return getNode(~Opcode, VTs, ResultTys.size(),
4221 /// getNodeIfExists - Get the specified node if it's already available, or
4222 /// else return NULL.
4223 SDNode *SelectionDAG::getNodeIfExists(unsigned Opcode, SDVTList VTList,
4224 const SDValue *Ops, unsigned NumOps) {
4225 if (VTList.VTs[VTList.NumVTs-1] != MVT::Flag) {
4226 FoldingSetNodeID ID;
4227 AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
4229 if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
4236 /// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead.
4237 /// This can cause recursive merging of nodes in the DAG.
4239 /// This version assumes From has a single result value.
4241 void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To,
4242 DAGUpdateListener *UpdateListener) {
4243 SDNode *From = FromN.Val;
4244 assert(From->getNumValues() == 1 && FromN.getResNo() == 0 &&
4245 "Cannot replace with this method!");
4246 assert(From != To.Val && "Cannot replace uses of with self");
4248 while (!From->use_empty()) {
4249 SDNode::use_iterator UI = From->use_begin();
4252 // This node is about to morph, remove its old self from the CSE maps.
4253 RemoveNodeFromCSEMaps(U);
4255 for (SDNode::op_iterator I = U->op_begin(), E = U->op_end();
4256 I != E; ++I, ++operandNum)
4257 if (I->getVal() == From) {
4258 From->removeUser(operandNum, U);
4261 To.Val->addUser(operandNum, U);
4264 // Now that we have modified U, add it back to the CSE maps. If it already
4265 // exists there, recursively merge the results together.
4266 if (SDNode *Existing = AddNonLeafNodeToCSEMaps(U)) {
4267 ReplaceAllUsesWith(U, Existing, UpdateListener);
4268 // U is now dead. Inform the listener if it exists and delete it.
4270 UpdateListener->NodeDeleted(U, Existing);
4271 DeleteNodeNotInCSEMaps(U);
4273 // If the node doesn't already exist, we updated it. Inform a listener if
4276 UpdateListener->NodeUpdated(U);
4281 /// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead.
4282 /// This can cause recursive merging of nodes in the DAG.
4284 /// This version assumes From/To have matching types and numbers of result
4287 void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To,
4288 DAGUpdateListener *UpdateListener) {
4289 assert(From->getVTList().VTs == To->getVTList().VTs &&
4290 From->getNumValues() == To->getNumValues() &&
4291 "Cannot use this version of ReplaceAllUsesWith!");
4293 // Handle the trivial case.
4297 while (!From->use_empty()) {
4298 SDNode::use_iterator UI = From->use_begin();
4301 // This node is about to morph, remove its old self from the CSE maps.
4302 RemoveNodeFromCSEMaps(U);
4304 for (SDNode::op_iterator I = U->op_begin(), E = U->op_end();
4305 I != E; ++I, ++operandNum)
4306 if (I->getVal() == From) {
4307 From->removeUser(operandNum, U);
4309 To->addUser(operandNum, U);
4312 // Now that we have modified U, add it back to the CSE maps. If it already
4313 // exists there, recursively merge the results together.
4314 if (SDNode *Existing = AddNonLeafNodeToCSEMaps(U)) {
4315 ReplaceAllUsesWith(U, Existing, UpdateListener);
4316 // U is now dead. Inform the listener if it exists and delete it.
4318 UpdateListener->NodeDeleted(U, Existing);
4319 DeleteNodeNotInCSEMaps(U);
4321 // If the node doesn't already exist, we updated it. Inform a listener if
4324 UpdateListener->NodeUpdated(U);
4329 /// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead.
4330 /// This can cause recursive merging of nodes in the DAG.
4332 /// This version can replace From with any result values. To must match the
4333 /// number and types of values returned by From.
4334 void SelectionDAG::ReplaceAllUsesWith(SDNode *From,
4336 DAGUpdateListener *UpdateListener) {
4337 if (From->getNumValues() == 1) // Handle the simple case efficiently.
4338 return ReplaceAllUsesWith(SDValue(From, 0), To[0], UpdateListener);
4340 while (!From->use_empty()) {
4341 SDNode::use_iterator UI = From->use_begin();
4344 // This node is about to morph, remove its old self from the CSE maps.
4345 RemoveNodeFromCSEMaps(U);
4347 for (SDNode::op_iterator I = U->op_begin(), E = U->op_end();
4348 I != E; ++I, ++operandNum)
4349 if (I->getVal() == From) {
4350 const SDValue &ToOp = To[I->getSDValue().getResNo()];
4351 From->removeUser(operandNum, U);
4354 ToOp.Val->addUser(operandNum, U);
4357 // Now that we have modified U, add it back to the CSE maps. If it already
4358 // exists there, recursively merge the results together.
4359 if (SDNode *Existing = AddNonLeafNodeToCSEMaps(U)) {
4360 ReplaceAllUsesWith(U, Existing, UpdateListener);
4361 // U is now dead. Inform the listener if it exists and delete it.
4363 UpdateListener->NodeDeleted(U, Existing);
4364 DeleteNodeNotInCSEMaps(U);
4366 // If the node doesn't already exist, we updated it. Inform a listener if
4369 UpdateListener->NodeUpdated(U);
4374 /// ReplaceAllUsesOfValueWith - Replace any uses of From with To, leaving
4375 /// uses of other values produced by From.Val alone. The Deleted vector is
4376 /// handled the same way as for ReplaceAllUsesWith.
4377 void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To,
4378 DAGUpdateListener *UpdateListener){
4379 // Handle the really simple, really trivial case efficiently.
4380 if (From == To) return;
4382 // Handle the simple, trivial, case efficiently.
4383 if (From.Val->getNumValues() == 1) {
4384 ReplaceAllUsesWith(From, To, UpdateListener);
4388 // Get all of the users of From.Val. We want these in a nice,
4389 // deterministically ordered and uniqued set, so we use a SmallSetVector.
4390 SmallSetVector<SDNode*, 16> Users(From.Val->use_begin(), From.Val->use_end());
4392 while (!Users.empty()) {
4393 // We know that this user uses some value of From. If it is the right
4394 // value, update it.
4395 SDNode *User = Users.back();
4398 // Scan for an operand that matches From.
4399 SDNode::op_iterator Op = User->op_begin(), E = User->op_end();
4400 for (; Op != E; ++Op)
4401 if (*Op == From) break;
4403 // If there are no matches, the user must use some other result of From.
4404 if (Op == E) continue;
4406 // Okay, we know this user needs to be updated. Remove its old self
4407 // from the CSE maps.
4408 RemoveNodeFromCSEMaps(User);
4410 // Update all operands that match "From" in case there are multiple uses.
4411 for (; Op != E; ++Op) {
4413 From.Val->removeUser(Op-User->op_begin(), User);
4416 To.Val->addUser(Op-User->op_begin(), User);
4420 // Now that we have modified User, add it back to the CSE maps. If it
4421 // already exists there, recursively merge the results together.
4422 SDNode *Existing = AddNonLeafNodeToCSEMaps(User);
4424 if (UpdateListener) UpdateListener->NodeUpdated(User);
4425 continue; // Continue on to next user.
4428 // If there was already an existing matching node, use ReplaceAllUsesWith
4429 // to replace the dead one with the existing one. This can cause
4430 // recursive merging of other unrelated nodes down the line.
4431 ReplaceAllUsesWith(User, Existing, UpdateListener);
4433 // User is now dead. Notify a listener if present.
4434 if (UpdateListener) UpdateListener->NodeDeleted(User, Existing);
4435 DeleteNodeNotInCSEMaps(User);
4439 /// ReplaceAllUsesOfValuesWith - Replace any uses of From with To, leaving
4440 /// uses of other values produced by From.Val alone. The same value may
4441 /// appear in both the From and To list. The Deleted vector is
4442 /// handled the same way as for ReplaceAllUsesWith.
4443 void SelectionDAG::ReplaceAllUsesOfValuesWith(const SDValue *From,
4446 DAGUpdateListener *UpdateListener){
4447 // Handle the simple, trivial case efficiently.
4449 return ReplaceAllUsesOfValueWith(*From, *To, UpdateListener);
4451 SmallVector<std::pair<SDNode *, unsigned>, 16> Users;
4452 for (unsigned i = 0; i != Num; ++i)
4453 for (SDNode::use_iterator UI = From[i].Val->use_begin(),
4454 E = From[i].Val->use_end(); UI != E; ++UI)
4455 Users.push_back(std::make_pair(*UI, i));
4457 while (!Users.empty()) {
4458 // We know that this user uses some value of From. If it is the right
4459 // value, update it.
4460 SDNode *User = Users.back().first;
4461 unsigned i = Users.back().second;
4464 // Scan for an operand that matches From.
4465 SDNode::op_iterator Op = User->op_begin(), E = User->op_end();
4466 for (; Op != E; ++Op)
4467 if (*Op == From[i]) break;
4469 // If there are no matches, the user must use some other result of From.
4470 if (Op == E) continue;
4472 // Okay, we know this user needs to be updated. Remove its old self
4473 // from the CSE maps.
4474 RemoveNodeFromCSEMaps(User);
4476 // Update all operands that match "From" in case there are multiple uses.
4477 for (; Op != E; ++Op) {
4478 if (*Op == From[i]) {
4479 From[i].Val->removeUser(Op-User->op_begin(), User);
4482 To[i].Val->addUser(Op-User->op_begin(), User);
4486 // Now that we have modified User, add it back to the CSE maps. If it
4487 // already exists there, recursively merge the results together.
4488 SDNode *Existing = AddNonLeafNodeToCSEMaps(User);
4490 if (UpdateListener) UpdateListener->NodeUpdated(User);
4491 continue; // Continue on to next user.
4494 // If there was already an existing matching node, use ReplaceAllUsesWith
4495 // to replace the dead one with the existing one. This can cause
4496 // recursive merging of other unrelated nodes down the line.
4497 ReplaceAllUsesWith(User, Existing, UpdateListener);
4499 // User is now dead. Notify a listener if present.
4500 if (UpdateListener) UpdateListener->NodeDeleted(User, Existing);
4501 DeleteNodeNotInCSEMaps(User);
4505 /// AssignTopologicalOrder - Assign a unique node id for each node in the DAG
4506 /// based on their topological order. It returns the maximum id and a vector
4507 /// of the SDNodes* in assigned order by reference.
4508 unsigned SelectionDAG::AssignTopologicalOrder(std::vector<SDNode*> &TopOrder) {
4509 unsigned DAGSize = AllNodes.size();
4510 std::vector<SDNode*> Sources;
4512 for (allnodes_iterator I = allnodes_begin(),E = allnodes_end(); I != E; ++I){
4514 unsigned Degree = N->use_size();
4515 // Temporarily use the Node Id as scratch space for the degree count.
4516 N->setNodeId(Degree);
4518 Sources.push_back(N);
4522 TopOrder.reserve(DAGSize);
4524 while (!Sources.empty()) {
4525 SDNode *N = Sources.back();
4527 TopOrder.push_back(N);
4529 for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) {
4530 SDNode *P = I->getVal();
4531 unsigned Degree = P->getNodeId();
4533 P->setNodeId(Degree);
4535 Sources.push_back(P);
4544 //===----------------------------------------------------------------------===//
4546 //===----------------------------------------------------------------------===//
4548 // Out-of-line virtual method to give class a home.
4549 void SDNode::ANCHOR() {}
4550 void UnarySDNode::ANCHOR() {}
4551 void BinarySDNode::ANCHOR() {}
4552 void TernarySDNode::ANCHOR() {}
4553 void HandleSDNode::ANCHOR() {}
4554 void ConstantSDNode::ANCHOR() {}
4555 void ConstantFPSDNode::ANCHOR() {}
4556 void GlobalAddressSDNode::ANCHOR() {}
4557 void FrameIndexSDNode::ANCHOR() {}
4558 void JumpTableSDNode::ANCHOR() {}
4559 void ConstantPoolSDNode::ANCHOR() {}
4560 void BasicBlockSDNode::ANCHOR() {}
4561 void SrcValueSDNode::ANCHOR() {}
4562 void MemOperandSDNode::ANCHOR() {}
4563 void RegisterSDNode::ANCHOR() {}
4564 void DbgStopPointSDNode::ANCHOR() {}
4565 void LabelSDNode::ANCHOR() {}
4566 void ExternalSymbolSDNode::ANCHOR() {}
4567 void CondCodeSDNode::ANCHOR() {}
4568 void ARG_FLAGSSDNode::ANCHOR() {}
4569 void VTSDNode::ANCHOR() {}
4570 void MemSDNode::ANCHOR() {}
4571 void LoadSDNode::ANCHOR() {}
4572 void StoreSDNode::ANCHOR() {}
4573 void AtomicSDNode::ANCHOR() {}
4575 HandleSDNode::~HandleSDNode() {
4579 GlobalAddressSDNode::GlobalAddressSDNode(bool isTarget, const GlobalValue *GA,
4581 : SDNode(isa<GlobalVariable>(GA) &&
4582 cast<GlobalVariable>(GA)->isThreadLocal() ?
4584 (isTarget ? ISD::TargetGlobalTLSAddress : ISD::GlobalTLSAddress) :
4586 (isTarget ? ISD::TargetGlobalAddress : ISD::GlobalAddress),
4587 getSDVTList(VT)), Offset(o) {
4588 TheGlobal = const_cast<GlobalValue*>(GA);
4591 MemSDNode::MemSDNode(unsigned Opc, SDVTList VTs, MVT memvt,
4592 const Value *srcValue, int SVO,
4593 unsigned alignment, bool vol)
4594 : SDNode(Opc, VTs), MemoryVT(memvt), SrcValue(srcValue), SVOffset(SVO),
4595 Flags(encodeMemSDNodeFlags(vol, alignment)) {
4597 assert(isPowerOf2_32(alignment) && "Alignment is not a power of 2!");
4598 assert(getAlignment() == alignment && "Alignment representation error!");
4599 assert(isVolatile() == vol && "Volatile representation error!");
4602 /// getMemOperand - Return a MachineMemOperand object describing the memory
4603 /// reference performed by this memory reference.
4604 MachineMemOperand MemSDNode::getMemOperand() const {
4606 if (isa<LoadSDNode>(this))
4607 Flags = MachineMemOperand::MOLoad;
4608 else if (isa<StoreSDNode>(this))
4609 Flags = MachineMemOperand::MOStore;
4611 assert(isa<AtomicSDNode>(this) && "Unknown MemSDNode opcode!");
4612 Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
4615 int Size = (getMemoryVT().getSizeInBits() + 7) >> 3;
4616 if (isVolatile()) Flags |= MachineMemOperand::MOVolatile;
4618 // Check if the memory reference references a frame index
4619 const FrameIndexSDNode *FI =
4620 dyn_cast<const FrameIndexSDNode>(getBasePtr().Val);
4621 if (!getSrcValue() && FI)
4622 return MachineMemOperand(PseudoSourceValue::getFixedStack(FI->getIndex()),
4623 Flags, 0, Size, getAlignment());
4625 return MachineMemOperand(getSrcValue(), Flags, getSrcValueOffset(),
4626 Size, getAlignment());
4629 /// Profile - Gather unique data for the node.
4631 void SDNode::Profile(FoldingSetNodeID &ID) const {
4632 AddNodeIDNode(ID, this);
4635 /// getValueTypeList - Return a pointer to the specified value type.
4637 const MVT *SDNode::getValueTypeList(MVT VT) {
4638 if (VT.isExtended()) {
4639 static std::set<MVT, MVT::compareRawBits> EVTs;
4640 return &(*EVTs.insert(VT).first);
4642 static MVT VTs[MVT::LAST_VALUETYPE];
4643 VTs[VT.getSimpleVT()] = VT;
4644 return &VTs[VT.getSimpleVT()];
4648 /// hasNUsesOfValue - Return true if there are exactly NUSES uses of the
4649 /// indicated value. This method ignores uses of other values defined by this
4651 bool SDNode::hasNUsesOfValue(unsigned NUses, unsigned Value) const {
4652 assert(Value < getNumValues() && "Bad value!");
4654 // TODO: Only iterate over uses of a given value of the node
4655 for (SDNode::use_iterator UI = use_begin(), E = use_end(); UI != E; ++UI) {
4656 if (UI.getUse().getSDValue().getResNo() == Value) {
4663 // Found exactly the right number of uses?
4668 /// hasAnyUseOfValue - Return true if there are any use of the indicated
4669 /// value. This method ignores uses of other values defined by this operation.
4670 bool SDNode::hasAnyUseOfValue(unsigned Value) const {
4671 assert(Value < getNumValues() && "Bad value!");
4673 for (SDNode::use_iterator UI = use_begin(), E = use_end(); UI != E; ++UI)
4674 if (UI.getUse().getSDValue().getResNo() == Value)
4681 /// isOnlyUserOf - Return true if this node is the only use of N.
4683 bool SDNode::isOnlyUserOf(SDNode *N) const {
4685 for (SDNode::use_iterator I = N->use_begin(), E = N->use_end(); I != E; ++I) {
4696 /// isOperand - Return true if this node is an operand of N.
4698 bool SDValue::isOperandOf(SDNode *N) const {
4699 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
4700 if (*this == N->getOperand(i))
4705 bool SDNode::isOperandOf(SDNode *N) const {
4706 for (unsigned i = 0, e = N->NumOperands; i != e; ++i)
4707 if (this == N->OperandList[i].getVal())
4712 /// reachesChainWithoutSideEffects - Return true if this operand (which must
4713 /// be a chain) reaches the specified operand without crossing any
4714 /// side-effecting instructions. In practice, this looks through token
4715 /// factors and non-volatile loads. In order to remain efficient, this only
4716 /// looks a couple of nodes in, it does not do an exhaustive search.
4717 bool SDValue::reachesChainWithoutSideEffects(SDValue Dest,
4718 unsigned Depth) const {
4719 if (*this == Dest) return true;
4721 // Don't search too deeply, we just want to be able to see through
4722 // TokenFactor's etc.
4723 if (Depth == 0) return false;
4725 // If this is a token factor, all inputs to the TF happen in parallel. If any
4726 // of the operands of the TF reach dest, then we can do the xform.
4727 if (getOpcode() == ISD::TokenFactor) {
4728 for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
4729 if (getOperand(i).reachesChainWithoutSideEffects(Dest, Depth-1))
4734 // Loads don't have side effects, look through them.
4735 if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(*this)) {
4736 if (!Ld->isVolatile())
4737 return Ld->getChain().reachesChainWithoutSideEffects(Dest, Depth-1);
4743 static void findPredecessor(SDNode *N, const SDNode *P, bool &found,
4744 SmallPtrSet<SDNode *, 32> &Visited) {
4745 if (found || !Visited.insert(N))
4748 for (unsigned i = 0, e = N->getNumOperands(); !found && i != e; ++i) {
4749 SDNode *Op = N->getOperand(i).Val;
4754 findPredecessor(Op, P, found, Visited);
4758 /// isPredecessorOf - Return true if this node is a predecessor of N. This node
4759 /// is either an operand of N or it can be reached by recursively traversing
4760 /// up the operands.
4761 /// NOTE: this is an expensive method. Use it carefully.
4762 bool SDNode::isPredecessorOf(SDNode *N) const {
4763 SmallPtrSet<SDNode *, 32> Visited;
4765 findPredecessor(N, this, found, Visited);
4769 uint64_t SDNode::getConstantOperandVal(unsigned Num) const {
4770 assert(Num < NumOperands && "Invalid child # of SDNode!");
4771 return cast<ConstantSDNode>(OperandList[Num])->getValue();
4774 std::string SDNode::getOperationName(const SelectionDAG *G) const {
4775 switch (getOpcode()) {
4777 if (getOpcode() < ISD::BUILTIN_OP_END)
4778 return "<<Unknown DAG Node>>";
4779 if (isMachineOpcode()) {
4781 if (const TargetInstrInfo *TII = G->getTarget().getInstrInfo())
4782 if (getMachineOpcode() < TII->getNumOpcodes())
4783 return TII->get(getMachineOpcode()).getName();
4784 return "<<Unknown Machine Node>>";
4787 TargetLowering &TLI = G->getTargetLoweringInfo();
4788 const char *Name = TLI.getTargetNodeName(getOpcode());
4789 if (Name) return Name;
4790 return "<<Unknown Target Node>>";
4792 return "<<Unknown Node>>";
4795 case ISD::DELETED_NODE:
4796 return "<<Deleted Node!>>";
4798 case ISD::PREFETCH: return "Prefetch";
4799 case ISD::MEMBARRIER: return "MemBarrier";
4800 case ISD::ATOMIC_CMP_SWAP_8: return "AtomicCmpSwap8";
4801 case ISD::ATOMIC_SWAP_8: return "AtomicSwap8";
4802 case ISD::ATOMIC_LOAD_ADD_8: return "AtomicLoadAdd8";
4803 case ISD::ATOMIC_LOAD_SUB_8: return "AtomicLoadSub8";
4804 case ISD::ATOMIC_LOAD_AND_8: return "AtomicLoadAnd8";
4805 case ISD::ATOMIC_LOAD_OR_8: return "AtomicLoadOr8";
4806 case ISD::ATOMIC_LOAD_XOR_8: return "AtomicLoadXor8";
4807 case ISD::ATOMIC_LOAD_NAND_8: return "AtomicLoadNand8";
4808 case ISD::ATOMIC_LOAD_MIN_8: return "AtomicLoadMin8";
4809 case ISD::ATOMIC_LOAD_MAX_8: return "AtomicLoadMax8";
4810 case ISD::ATOMIC_LOAD_UMIN_8: return "AtomicLoadUMin8";
4811 case ISD::ATOMIC_LOAD_UMAX_8: return "AtomicLoadUMax8";
4812 case ISD::ATOMIC_CMP_SWAP_16: return "AtomicCmpSwap16";
4813 case ISD::ATOMIC_SWAP_16: return "AtomicSwap16";
4814 case ISD::ATOMIC_LOAD_ADD_16: return "AtomicLoadAdd16";
4815 case ISD::ATOMIC_LOAD_SUB_16: return "AtomicLoadSub16";
4816 case ISD::ATOMIC_LOAD_AND_16: return "AtomicLoadAnd16";
4817 case ISD::ATOMIC_LOAD_OR_16: return "AtomicLoadOr16";
4818 case ISD::ATOMIC_LOAD_XOR_16: return "AtomicLoadXor16";
4819 case ISD::ATOMIC_LOAD_NAND_16: return "AtomicLoadNand16";
4820 case ISD::ATOMIC_LOAD_MIN_16: return "AtomicLoadMin16";
4821 case ISD::ATOMIC_LOAD_MAX_16: return "AtomicLoadMax16";
4822 case ISD::ATOMIC_LOAD_UMIN_16: return "AtomicLoadUMin16";
4823 case ISD::ATOMIC_LOAD_UMAX_16: return "AtomicLoadUMax16";
4824 case ISD::ATOMIC_CMP_SWAP_32: return "AtomicCmpSwap32";
4825 case ISD::ATOMIC_SWAP_32: return "AtomicSwap32";
4826 case ISD::ATOMIC_LOAD_ADD_32: return "AtomicLoadAdd32";
4827 case ISD::ATOMIC_LOAD_SUB_32: return "AtomicLoadSub32";
4828 case ISD::ATOMIC_LOAD_AND_32: return "AtomicLoadAnd32";
4829 case ISD::ATOMIC_LOAD_OR_32: return "AtomicLoadOr32";
4830 case ISD::ATOMIC_LOAD_XOR_32: return "AtomicLoadXor32";
4831 case ISD::ATOMIC_LOAD_NAND_32: return "AtomicLoadNand32";
4832 case ISD::ATOMIC_LOAD_MIN_32: return "AtomicLoadMin32";
4833 case ISD::ATOMIC_LOAD_MAX_32: return "AtomicLoadMax32";
4834 case ISD::ATOMIC_LOAD_UMIN_32: return "AtomicLoadUMin32";
4835 case ISD::ATOMIC_LOAD_UMAX_32: return "AtomicLoadUMax32";
4836 case ISD::ATOMIC_CMP_SWAP_64: return "AtomicCmpSwap64";
4837 case ISD::ATOMIC_SWAP_64: return "AtomicSwap64";
4838 case ISD::ATOMIC_LOAD_ADD_64: return "AtomicLoadAdd64";
4839 case ISD::ATOMIC_LOAD_SUB_64: return "AtomicLoadSub64";
4840 case ISD::ATOMIC_LOAD_AND_64: return "AtomicLoadAnd64";
4841 case ISD::ATOMIC_LOAD_OR_64: return "AtomicLoadOr64";
4842 case ISD::ATOMIC_LOAD_XOR_64: return "AtomicLoadXor64";
4843 case ISD::ATOMIC_LOAD_NAND_64: return "AtomicLoadNand64";
4844 case ISD::ATOMIC_LOAD_MIN_64: return "AtomicLoadMin64";
4845 case ISD::ATOMIC_LOAD_MAX_64: return "AtomicLoadMax64";
4846 case ISD::ATOMIC_LOAD_UMIN_64: return "AtomicLoadUMin64";
4847 case ISD::ATOMIC_LOAD_UMAX_64: return "AtomicLoadUMax64";
4848 case ISD::PCMARKER: return "PCMarker";
4849 case ISD::READCYCLECOUNTER: return "ReadCycleCounter";
4850 case ISD::SRCVALUE: return "SrcValue";
4851 case ISD::MEMOPERAND: return "MemOperand";
4852 case ISD::EntryToken: return "EntryToken";
4853 case ISD::TokenFactor: return "TokenFactor";
4854 case ISD::AssertSext: return "AssertSext";
4855 case ISD::AssertZext: return "AssertZext";
4857 case ISD::BasicBlock: return "BasicBlock";
4858 case ISD::ARG_FLAGS: return "ArgFlags";
4859 case ISD::VALUETYPE: return "ValueType";
4860 case ISD::Register: return "Register";
4862 case ISD::Constant: return "Constant";
4863 case ISD::ConstantFP: return "ConstantFP";
4864 case ISD::GlobalAddress: return "GlobalAddress";
4865 case ISD::GlobalTLSAddress: return "GlobalTLSAddress";
4866 case ISD::FrameIndex: return "FrameIndex";
4867 case ISD::JumpTable: return "JumpTable";
4868 case ISD::GLOBAL_OFFSET_TABLE: return "GLOBAL_OFFSET_TABLE";
4869 case ISD::RETURNADDR: return "RETURNADDR";
4870 case ISD::FRAMEADDR: return "FRAMEADDR";
4871 case ISD::FRAME_TO_ARGS_OFFSET: return "FRAME_TO_ARGS_OFFSET";
4872 case ISD::EXCEPTIONADDR: return "EXCEPTIONADDR";
4873 case ISD::EHSELECTION: return "EHSELECTION";
4874 case ISD::EH_RETURN: return "EH_RETURN";
4875 case ISD::ConstantPool: return "ConstantPool";
4876 case ISD::ExternalSymbol: return "ExternalSymbol";
4877 case ISD::INTRINSIC_WO_CHAIN: {
4878 unsigned IID = cast<ConstantSDNode>(getOperand(0))->getValue();
4879 return Intrinsic::getName((Intrinsic::ID)IID);
4881 case ISD::INTRINSIC_VOID:
4882 case ISD::INTRINSIC_W_CHAIN: {
4883 unsigned IID = cast<ConstantSDNode>(getOperand(1))->getValue();
4884 return Intrinsic::getName((Intrinsic::ID)IID);
4887 case ISD::BUILD_VECTOR: return "BUILD_VECTOR";
4888 case ISD::TargetConstant: return "TargetConstant";
4889 case ISD::TargetConstantFP:return "TargetConstantFP";
4890 case ISD::TargetGlobalAddress: return "TargetGlobalAddress";
4891 case ISD::TargetGlobalTLSAddress: return "TargetGlobalTLSAddress";
4892 case ISD::TargetFrameIndex: return "TargetFrameIndex";
4893 case ISD::TargetJumpTable: return "TargetJumpTable";
4894 case ISD::TargetConstantPool: return "TargetConstantPool";
4895 case ISD::TargetExternalSymbol: return "TargetExternalSymbol";
4897 case ISD::CopyToReg: return "CopyToReg";
4898 case ISD::CopyFromReg: return "CopyFromReg";
4899 case ISD::UNDEF: return "undef";
4900 case ISD::MERGE_VALUES: return "merge_values";
4901 case ISD::INLINEASM: return "inlineasm";
4902 case ISD::DBG_LABEL: return "dbg_label";
4903 case ISD::EH_LABEL: return "eh_label";
4904 case ISD::DECLARE: return "declare";
4905 case ISD::HANDLENODE: return "handlenode";
4906 case ISD::FORMAL_ARGUMENTS: return "formal_arguments";
4907 case ISD::CALL: return "call";
4910 case ISD::FABS: return "fabs";
4911 case ISD::FNEG: return "fneg";
4912 case ISD::FSQRT: return "fsqrt";
4913 case ISD::FSIN: return "fsin";
4914 case ISD::FCOS: return "fcos";
4915 case ISD::FPOWI: return "fpowi";
4916 case ISD::FPOW: return "fpow";
4917 case ISD::FTRUNC: return "ftrunc";
4918 case ISD::FFLOOR: return "ffloor";
4919 case ISD::FCEIL: return "fceil";
4920 case ISD::FRINT: return "frint";
4921 case ISD::FNEARBYINT: return "fnearbyint";
4924 case ISD::ADD: return "add";
4925 case ISD::SUB: return "sub";
4926 case ISD::MUL: return "mul";
4927 case ISD::MULHU: return "mulhu";
4928 case ISD::MULHS: return "mulhs";
4929 case ISD::SDIV: return "sdiv";
4930 case ISD::UDIV: return "udiv";
4931 case ISD::SREM: return "srem";
4932 case ISD::UREM: return "urem";
4933 case ISD::SMUL_LOHI: return "smul_lohi";
4934 case ISD::UMUL_LOHI: return "umul_lohi";
4935 case ISD::SDIVREM: return "sdivrem";
4936 case ISD::UDIVREM: return "divrem";
4937 case ISD::AND: return "and";
4938 case ISD::OR: return "or";
4939 case ISD::XOR: return "xor";
4940 case ISD::SHL: return "shl";
4941 case ISD::SRA: return "sra";
4942 case ISD::SRL: return "srl";
4943 case ISD::ROTL: return "rotl";
4944 case ISD::ROTR: return "rotr";
4945 case ISD::FADD: return "fadd";
4946 case ISD::FSUB: return "fsub";
4947 case ISD::FMUL: return "fmul";
4948 case ISD::FDIV: return "fdiv";
4949 case ISD::FREM: return "frem";
4950 case ISD::FCOPYSIGN: return "fcopysign";
4951 case ISD::FGETSIGN: return "fgetsign";
4953 case ISD::SETCC: return "setcc";
4954 case ISD::VSETCC: return "vsetcc";
4955 case ISD::SELECT: return "select";
4956 case ISD::SELECT_CC: return "select_cc";
4957 case ISD::INSERT_VECTOR_ELT: return "insert_vector_elt";
4958 case ISD::EXTRACT_VECTOR_ELT: return "extract_vector_elt";
4959 case ISD::CONCAT_VECTORS: return "concat_vectors";
4960 case ISD::EXTRACT_SUBVECTOR: return "extract_subvector";
4961 case ISD::SCALAR_TO_VECTOR: return "scalar_to_vector";
4962 case ISD::VECTOR_SHUFFLE: return "vector_shuffle";
4963 case ISD::CARRY_FALSE: return "carry_false";
4964 case ISD::ADDC: return "addc";
4965 case ISD::ADDE: return "adde";
4966 case ISD::SUBC: return "subc";
4967 case ISD::SUBE: return "sube";
4968 case ISD::SHL_PARTS: return "shl_parts";
4969 case ISD::SRA_PARTS: return "sra_parts";
4970 case ISD::SRL_PARTS: return "srl_parts";
4972 case ISD::EXTRACT_SUBREG: return "extract_subreg";
4973 case ISD::INSERT_SUBREG: return "insert_subreg";
4975 // Conversion operators.
4976 case ISD::SIGN_EXTEND: return "sign_extend";
4977 case ISD::ZERO_EXTEND: return "zero_extend";
4978 case ISD::ANY_EXTEND: return "any_extend";
4979 case ISD::SIGN_EXTEND_INREG: return "sign_extend_inreg";
4980 case ISD::TRUNCATE: return "truncate";
4981 case ISD::FP_ROUND: return "fp_round";
4982 case ISD::FLT_ROUNDS_: return "flt_rounds";
4983 case ISD::FP_ROUND_INREG: return "fp_round_inreg";
4984 case ISD::FP_EXTEND: return "fp_extend";
4986 case ISD::SINT_TO_FP: return "sint_to_fp";
4987 case ISD::UINT_TO_FP: return "uint_to_fp";
4988 case ISD::FP_TO_SINT: return "fp_to_sint";
4989 case ISD::FP_TO_UINT: return "fp_to_uint";
4990 case ISD::BIT_CONVERT: return "bit_convert";
4992 // Control flow instructions
4993 case ISD::BR: return "br";
4994 case ISD::BRIND: return "brind";
4995 case ISD::BR_JT: return "br_jt";
4996 case ISD::BRCOND: return "brcond";
4997 case ISD::BR_CC: return "br_cc";
4998 case ISD::RET: return "ret";
4999 case ISD::CALLSEQ_START: return "callseq_start";
5000 case ISD::CALLSEQ_END: return "callseq_end";
5003 case ISD::LOAD: return "load";
5004 case ISD::STORE: return "store";
5005 case ISD::VAARG: return "vaarg";
5006 case ISD::VACOPY: return "vacopy";
5007 case ISD::VAEND: return "vaend";
5008 case ISD::VASTART: return "vastart";
5009 case ISD::DYNAMIC_STACKALLOC: return "dynamic_stackalloc";
5010 case ISD::EXTRACT_ELEMENT: return "extract_element";
5011 case ISD::BUILD_PAIR: return "build_pair";
5012 case ISD::STACKSAVE: return "stacksave";
5013 case ISD::STACKRESTORE: return "stackrestore";
5014 case ISD::TRAP: return "trap";
5017 case ISD::BSWAP: return "bswap";
5018 case ISD::CTPOP: return "ctpop";
5019 case ISD::CTTZ: return "cttz";
5020 case ISD::CTLZ: return "ctlz";
5023 case ISD::DBG_STOPPOINT: return "dbg_stoppoint";
5024 case ISD::DEBUG_LOC: return "debug_loc";
5027 case ISD::TRAMPOLINE: return "trampoline";
5030 switch (cast<CondCodeSDNode>(this)->get()) {
5031 default: assert(0 && "Unknown setcc condition!");
5032 case ISD::SETOEQ: return "setoeq";
5033 case ISD::SETOGT: return "setogt";
5034 case ISD::SETOGE: return "setoge";
5035 case ISD::SETOLT: return "setolt";
5036 case ISD::SETOLE: return "setole";
5037 case ISD::SETONE: return "setone";
5039 case ISD::SETO: return "seto";
5040 case ISD::SETUO: return "setuo";
5041 case ISD::SETUEQ: return "setue";
5042 case ISD::SETUGT: return "setugt";
5043 case ISD::SETUGE: return "setuge";
5044 case ISD::SETULT: return "setult";
5045 case ISD::SETULE: return "setule";
5046 case ISD::SETUNE: return "setune";
5048 case ISD::SETEQ: return "seteq";
5049 case ISD::SETGT: return "setgt";
5050 case ISD::SETGE: return "setge";
5051 case ISD::SETLT: return "setlt";
5052 case ISD::SETLE: return "setle";
5053 case ISD::SETNE: return "setne";
5058 const char *SDNode::getIndexedModeName(ISD::MemIndexedMode AM) {
5067 return "<post-inc>";
5069 return "<post-dec>";
5073 std::string ISD::ArgFlagsTy::getArgFlagsString() {
5074 std::string S = "< ";
5088 if (getByValAlign())
5089 S += "byval-align:" + utostr(getByValAlign()) + " ";
5091 S += "orig-align:" + utostr(getOrigAlign()) + " ";
5093 S += "byval-size:" + utostr(getByValSize()) + " ";
5097 void SDNode::dump() const { dump(0); }
5098 void SDNode::dump(const SelectionDAG *G) const {
5103 void SDNode::print(raw_ostream &OS, const SelectionDAG *G) const {
5104 OS << (void*)this << ": ";
5106 for (unsigned i = 0, e = getNumValues(); i != e; ++i) {
5108 if (getValueType(i) == MVT::Other)
5111 OS << getValueType(i).getMVTString();
5113 OS << " = " << getOperationName(G);
5116 for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
5118 OS << (void*)getOperand(i).Val;
5119 if (unsigned RN = getOperand(i).getResNo())
5123 if (!isTargetOpcode() && getOpcode() == ISD::VECTOR_SHUFFLE) {
5124 SDNode *Mask = getOperand(2).Val;
5126 for (unsigned i = 0, e = Mask->getNumOperands(); i != e; ++i) {
5128 if (Mask->getOperand(i).getOpcode() == ISD::UNDEF)
5131 OS << cast<ConstantSDNode>(Mask->getOperand(i))->getValue();
5136 if (const ConstantSDNode *CSDN = dyn_cast<ConstantSDNode>(this)) {
5137 OS << '<' << CSDN->getAPIntValue() << '>';
5138 } else if (const ConstantFPSDNode *CSDN = dyn_cast<ConstantFPSDNode>(this)) {
5139 if (&CSDN->getValueAPF().getSemantics()==&APFloat::IEEEsingle)
5140 OS << '<' << CSDN->getValueAPF().convertToFloat() << '>';
5141 else if (&CSDN->getValueAPF().getSemantics()==&APFloat::IEEEdouble)
5142 OS << '<' << CSDN->getValueAPF().convertToDouble() << '>';
5145 CSDN->getValueAPF().convertToAPInt().dump();
5148 } else if (const GlobalAddressSDNode *GADN =
5149 dyn_cast<GlobalAddressSDNode>(this)) {
5150 int offset = GADN->getOffset();
5152 WriteAsOperand(OS, GADN->getGlobal());
5155 OS << " + " << offset;
5157 OS << " " << offset;
5158 } else if (const FrameIndexSDNode *FIDN = dyn_cast<FrameIndexSDNode>(this)) {
5159 OS << "<" << FIDN->getIndex() << ">";
5160 } else if (const JumpTableSDNode *JTDN = dyn_cast<JumpTableSDNode>(this)) {
5161 OS << "<" << JTDN->getIndex() << ">";
5162 } else if (const ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(this)){
5163 int offset = CP->getOffset();
5164 if (CP->isMachineConstantPoolEntry())
5165 OS << "<" << *CP->getMachineCPVal() << ">";
5167 OS << "<" << *CP->getConstVal() << ">";
5169 OS << " + " << offset;
5171 OS << " " << offset;
5172 } else if (const BasicBlockSDNode *BBDN = dyn_cast<BasicBlockSDNode>(this)) {
5174 const Value *LBB = (const Value*)BBDN->getBasicBlock()->getBasicBlock();
5176 OS << LBB->getName() << " ";
5177 OS << (const void*)BBDN->getBasicBlock() << ">";
5178 } else if (const RegisterSDNode *R = dyn_cast<RegisterSDNode>(this)) {
5179 if (G && R->getReg() &&
5180 TargetRegisterInfo::isPhysicalRegister(R->getReg())) {
5181 OS << " " << G->getTarget().getRegisterInfo()->getName(R->getReg());
5183 OS << " #" << R->getReg();
5185 } else if (const ExternalSymbolSDNode *ES =
5186 dyn_cast<ExternalSymbolSDNode>(this)) {
5187 OS << "'" << ES->getSymbol() << "'";
5188 } else if (const SrcValueSDNode *M = dyn_cast<SrcValueSDNode>(this)) {
5190 OS << "<" << M->getValue() << ">";
5193 } else if (const MemOperandSDNode *M = dyn_cast<MemOperandSDNode>(this)) {
5194 if (M->MO.getValue())
5195 OS << "<" << M->MO.getValue() << ":" << M->MO.getOffset() << ">";
5197 OS << "<null:" << M->MO.getOffset() << ">";
5198 } else if (const ARG_FLAGSSDNode *N = dyn_cast<ARG_FLAGSSDNode>(this)) {
5199 OS << N->getArgFlags().getArgFlagsString();
5200 } else if (const VTSDNode *N = dyn_cast<VTSDNode>(this)) {
5201 OS << ":" << N->getVT().getMVTString();
5203 else if (const LoadSDNode *LD = dyn_cast<LoadSDNode>(this)) {
5204 const Value *SrcValue = LD->getSrcValue();
5205 int SrcOffset = LD->getSrcValueOffset();
5211 OS << ":" << SrcOffset << ">";
5214 switch (LD->getExtensionType()) {
5215 default: doExt = false; break;
5216 case ISD::EXTLOAD: OS << " <anyext "; break;
5217 case ISD::SEXTLOAD: OS << " <sext "; break;
5218 case ISD::ZEXTLOAD: OS << " <zext "; break;
5221 OS << LD->getMemoryVT().getMVTString() << ">";
5223 const char *AM = getIndexedModeName(LD->getAddressingMode());
5226 if (LD->isVolatile())
5227 OS << " <volatile>";
5228 OS << " alignment=" << LD->getAlignment();
5229 } else if (const StoreSDNode *ST = dyn_cast<StoreSDNode>(this)) {
5230 const Value *SrcValue = ST->getSrcValue();
5231 int SrcOffset = ST->getSrcValueOffset();
5237 OS << ":" << SrcOffset << ">";
5239 if (ST->isTruncatingStore())
5240 OS << " <trunc " << ST->getMemoryVT().getMVTString() << ">";
5242 const char *AM = getIndexedModeName(ST->getAddressingMode());
5245 if (ST->isVolatile())
5246 OS << " <volatile>";
5247 OS << " alignment=" << ST->getAlignment();
5248 } else if (const AtomicSDNode* AT = dyn_cast<AtomicSDNode>(this)) {
5249 const Value *SrcValue = AT->getSrcValue();
5250 int SrcOffset = AT->getSrcValueOffset();
5256 OS << ":" << SrcOffset << ">";
5257 if (AT->isVolatile())
5258 OS << " <volatile>";
5259 OS << " alignment=" << AT->getAlignment();
5263 static void DumpNodes(const SDNode *N, unsigned indent, const SelectionDAG *G) {
5264 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
5265 if (N->getOperand(i).Val->hasOneUse())
5266 DumpNodes(N->getOperand(i).Val, indent+2, G);
5268 cerr << "\n" << std::string(indent+2, ' ')
5269 << (void*)N->getOperand(i).Val << ": <multiple use>";
5272 cerr << "\n" << std::string(indent, ' ');
5276 void SelectionDAG::dump() const {
5277 cerr << "SelectionDAG has " << AllNodes.size() << " nodes:";
5279 for (allnodes_const_iterator I = allnodes_begin(), E = allnodes_end();
5281 const SDNode *N = I;
5282 if (!N->hasOneUse() && N != getRoot().Val)
5283 DumpNodes(N, 2, this);
5286 if (getRoot().Val) DumpNodes(getRoot().Val, 2, this);
5291 const Type *ConstantPoolSDNode::getType() const {
5292 if (isMachineConstantPoolEntry())
5293 return Val.MachineCPVal->getType();
5294 return Val.ConstVal->getType();