//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "xcore-lower"
-
#include "XCoreISelLowering.h"
#include "XCore.h"
#include "XCoreMachineFunctionInfo.h"
using namespace llvm;
+#define DEBUG_TYPE "xcore-lower"
+
const char *XCoreTargetLowering::
getTargetNodeName(unsigned Opcode) const
{
case XCoreISD::PCRelativeWrapper : return "XCoreISD::PCRelativeWrapper";
case XCoreISD::DPRelativeWrapper : return "XCoreISD::DPRelativeWrapper";
case XCoreISD::CPRelativeWrapper : return "XCoreISD::CPRelativeWrapper";
+ case XCoreISD::LDWSP : return "XCoreISD::LDWSP";
case XCoreISD::STWSP : return "XCoreISD::STWSP";
case XCoreISD::RETSP : return "XCoreISD::RETSP";
case XCoreISD::LADD : return "XCoreISD::LADD";
case XCoreISD::FRAME_TO_ARGS_OFFSET : return "XCoreISD::FRAME_TO_ARGS_OFFSET";
case XCoreISD::EH_RETURN : return "XCoreISD::EH_RETURN";
case XCoreISD::MEMBARRIER : return "XCoreISD::MEMBARRIER";
- default : return NULL;
+ default : return nullptr;
}
}
// We have target-specific dag combine patterns for the following nodes:
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::ADD);
+ setTargetDAGCombine(ISD::INTRINSIC_VOID);
+ setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
setMinFunctionAlignment(1);
+ setPrefFunctionAlignment(2);
}
bool XCoreTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
const GlobalValue *UnderlyingGV = GV;
// If GV is an alias then use the aliasee to determine the wrapper type
if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
- UnderlyingGV = GA->resolveAliasedGlobal();
+ UnderlyingGV = GA->getAliasedGlobal();
if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(UnderlyingGV)) {
- if (GVar->isConstant())
+ if ( ( GVar->isConstant() &&
+ UnderlyingGV->isLocalLinkage(GV->getLinkage()) )
+ || ( GVar->hasSection() &&
+ StringRef(GVar->getSection()).startswith(".cp.") ) )
return DAG.getNode(XCoreISD::CPRelativeWrapper, dl, MVT::i32, GA);
return DAG.getNode(XCoreISD::DPRelativeWrapper, dl, MVT::i32, GA);
}
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Low.getValue(1),
High.getValue(1));
SDValue Ops[] = { Result, Chain };
- return DAG.getMergeValues(Ops, 2, DL);
+ return DAG.getMergeValues(Ops, DL);
}
static bool isWordAligned(SDValue Value, SelectionDAG &DAG)
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Low.getValue(1),
High.getValue(1));
SDValue Ops[] = { Result, Chain };
- return DAG.getMergeValues(Ops, 2, DL);
+ return DAG.getMergeValues(Ops, DL);
}
// Lower to a call to __misaligned_load(BasePtr).
SDValue Ops[] =
{ CallResult.first, CallResult.second };
- return DAG.getMergeValues(Ops, 2, DL);
+ return DAG.getMergeValues(Ops, DL);
}
SDValue XCoreTargetLowering::
LHS, RHS);
SDValue Lo(Hi.getNode(), 1);
SDValue Ops[] = { Lo, Hi };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
SDValue XCoreTargetLowering::
Zero, Zero);
SDValue Lo(Hi.getNode(), 1);
SDValue Ops[] = { Lo, Hi };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
/// isADDADDMUL - Return whether Op is in a form that is equivalent to
if (N->getOpcode() == ISD::ADD) {
SDValue Result = TryExpandADDWithMul(N, DAG);
- if (Result.getNode() != 0)
+ if (Result.getNode())
return Result;
}
DAG.getCopyToReg(Chain, dl, HandlerReg, Handler)
};
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 2);
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
return DAG.getNode(XCoreISD::EH_RETURN, dl, MVT::Other, Chain,
DAG.getRegister(StackReg, MVT::i32),
MachinePointerInfo(TrmpAddr, 16), false, false,
0);
- return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 5);
+ return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
}
SDValue XCoreTargetLowering::
Op.getOperand(1), Op.getOperand(2) , Op.getOperand(3));
SDValue Crc(Data.getNode(), 1);
SDValue Results[] = { Crc, Data };
- return DAG.getMergeValues(Results, 2, DL);
+ return DAG.getMergeValues(Results, DL);
}
return SDValue();
}
}
}
+/// LowerCallResult - Lower the result values of a call into the
+/// appropriate copies out of appropriate physical registers / memory locations.
+static SDValue
+LowerCallResult(SDValue Chain, SDValue InFlag,
+ const SmallVectorImpl<CCValAssign> &RVLocs,
+ SDLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+ SmallVector<std::pair<int, unsigned>, 4> ResultMemLocs;
+ // Copy results out of physical registers.
+ for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
+ const CCValAssign &VA = RVLocs[i];
+ if (VA.isRegLoc()) {
+ Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getValVT(),
+ InFlag).getValue(1);
+ InFlag = Chain.getValue(2);
+ InVals.push_back(Chain.getValue(0));
+ } else {
+ assert(VA.isMemLoc());
+ ResultMemLocs.push_back(std::make_pair(VA.getLocMemOffset(),
+ InVals.size()));
+ // Reserve space for this result.
+ InVals.push_back(SDValue());
+ }
+ }
+
+ // Copy results out of memory.
+ SmallVector<SDValue, 4> MemOpChains;
+ for (unsigned i = 0, e = ResultMemLocs.size(); i != e; ++i) {
+ int offset = ResultMemLocs[i].first;
+ unsigned index = ResultMemLocs[i].second;
+ SDVTList VTs = DAG.getVTList(MVT::i32, MVT::Other);
+ SDValue Ops[] = { Chain, DAG.getConstant(offset / 4, MVT::i32) };
+ SDValue load = DAG.getNode(XCoreISD::LDWSP, dl, VTs, Ops);
+ InVals[index] = load;
+ MemOpChains.push_back(load.getValue(1));
+ }
+
+ // Transform all loads nodes into one single node because
+ // all load nodes are independent of each other.
+ if (!MemOpChains.empty())
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
+
+ return Chain;
+}
+
/// LowerCCCCallTo - functions arguments are copied from virtual
/// regs to (physical regs)/(stack frame), CALLSEQ_START and
/// CALLSEQ_END are emitted.
CCInfo.AnalyzeCallOperands(Outs, CC_XCore);
+ SmallVector<CCValAssign, 16> RVLocs;
+ // Analyze return values to determine the number of bytes of stack required.
+ CCState RetCCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+ getTargetMachine(), RVLocs, *DAG.getContext());
+ RetCCInfo.AllocateStack(CCInfo.getNextStackOffset(), 4);
+ RetCCInfo.AnalyzeCallResult(Ins, RetCC_XCore);
+
// Get a count of how many bytes are to be pushed on the stack.
- unsigned NumBytes = CCInfo.getNextStackOffset();
+ unsigned NumBytes = RetCCInfo.getNextStackOffset();
Chain = DAG.getCALLSEQ_START(Chain,DAG.getConstant(NumBytes,
getPointerTy(), true), dl);
// Transform all store nodes into one single node because
// all store nodes are independent of each other.
if (!MemOpChains.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
- &MemOpChains[0], MemOpChains.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
// Build a sequence of copy-to-reg nodes chained together with token
// chain and flag operands which copy the outgoing args into registers.
if (InFlag.getNode())
Ops.push_back(InFlag);
- Chain = DAG.getNode(XCoreISD::BL, dl, NodeTys, &Ops[0], Ops.size());
+ Chain = DAG.getNode(XCoreISD::BL, dl, NodeTys, Ops);
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
// Handle result values, copying them out of physregs into vregs that we
// return.
- return LowerCallResult(Chain, InFlag, CallConv, isVarArg,
- Ins, dl, DAG, InVals);
-}
-
-/// LowerCallResult - Lower the result values of a call into the
-/// appropriate copies out of appropriate physical registers.
-SDValue
-XCoreTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
- CallingConv::ID CallConv, bool isVarArg,
- const SmallVectorImpl<ISD::InputArg> &Ins,
- SDLoc dl, SelectionDAG &DAG,
- SmallVectorImpl<SDValue> &InVals) const {
-
- // Assign locations to each value returned by this call.
- SmallVector<CCValAssign, 16> RVLocs;
- CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext());
-
- CCInfo.AnalyzeCallResult(Ins, RetCC_XCore);
-
- // Copy all of the result registers out of their specified physreg.
- for (unsigned i = 0; i != RVLocs.size(); ++i) {
- Chain = DAG.getCopyFromReg(Chain, dl, RVLocs[i].getLocReg(),
- RVLocs[i].getValVT(), InFlag).getValue(1);
- InFlag = Chain.getValue(2);
- InVals.push_back(Chain.getValue(0));
- }
-
- return Chain;
+ return LowerCallResult(Chain, InFlag, RVLocs, dl, DAG, InVals);
}
//===----------------------------------------------------------------------===//
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
+ XCoreFunctionInfo *XFI = MF.getInfo<XCoreFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
unsigned LRSaveSize = StackSlotSize;
+ if (!isVarArg)
+ XFI->setReturnStackOffset(CCInfo.getNextStackOffset() + LRSaveSize);
+
// All getCopyFromReg ops must precede any getMemcpys to prevent the
// scheduler clobbering a register before it has been copied.
// The stages are:
errs() << "LowerFormalArguments Unhandled argument type: "
<< RegVT.getSimpleVT().SimpleTy << "\n";
#endif
- llvm_unreachable(0);
+ llvm_unreachable(nullptr);
}
case MVT::i32:
unsigned VReg = RegInfo.createVirtualRegister(&XCore::GRRegsRegClass);
// 1b. CopyFromReg vararg registers.
if (isVarArg) {
// Argument registers
- static const uint16_t ArgRegs[] = {
+ static const MCPhysReg ArgRegs[] = {
XCore::R0, XCore::R1, XCore::R2, XCore::R3
};
XCoreFunctionInfo *XFI = MF.getInfo<XCoreFunctionInfo>();
// 2. chain CopyFromReg nodes into a TokenFactor.
if (!CFRegNode.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &CFRegNode[0],
- CFRegNode.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, CFRegNode);
// 3. Memcpy 'byVal' args & push final InVals.
// Aggregates passed "byVal" need to be copied by the callee.
// 4, chain mem ops nodes into a TokenFactor.
if (!MemOps.empty()) {
MemOps.push_back(Chain);
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &MemOps[0],
- MemOps.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
}
return Chain;
LLVMContext &Context) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, MF, getTargetMachine(), RVLocs, Context);
- return CCInfo.CheckReturn(Outs, RetCC_XCore);
+ if (!CCInfo.CheckReturn(Outs, RetCC_XCore))
+ return false;
+ if (CCInfo.getNextStackOffset() != 0 && isVarArg)
+ return false;
+ return true;
}
SDValue
const SmallVectorImpl<SDValue> &OutVals,
SDLoc dl, SelectionDAG &DAG) const {
+ XCoreFunctionInfo *XFI =
+ DAG.getMachineFunction().getInfo<XCoreFunctionInfo>();
+ MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+
// CCValAssign - represent the assignment of
// the return value to a location
SmallVector<CCValAssign, 16> RVLocs;
getTargetMachine(), RVLocs, *DAG.getContext());
// Analyze return values.
+ if (!isVarArg)
+ CCInfo.AllocateStack(XFI->getReturnStackOffset(), 4);
+
CCInfo.AnalyzeReturn(Outs, RetCC_XCore);
SDValue Flag;
// Return on XCore is always a "retsp 0"
RetOps.push_back(DAG.getConstant(0, MVT::i32));
- // Copy the result values into the output registers.
- for (unsigned i = 0; i != RVLocs.size(); ++i) {
+ SmallVector<SDValue, 4> MemOpChains;
+ // Handle return values that must be copied to memory.
+ for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
- assert(VA.isRegLoc() && "Can only return in registers!");
+ if (VA.isRegLoc())
+ continue;
+ assert(VA.isMemLoc());
+ if (isVarArg) {
+ report_fatal_error("Can't return value from vararg function in memory");
+ }
+
+ int Offset = VA.getLocMemOffset();
+ unsigned ObjSize = VA.getLocVT().getSizeInBits() / 8;
+ // Create the frame index object for the memory location.
+ int FI = MFI->CreateFixedObject(ObjSize, Offset, false);
+
+ // Create a SelectionDAG node corresponding to a store
+ // to this memory location.
+ SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
+ MemOpChains.push_back(DAG.getStore(Chain, dl, OutVals[i], FIN,
+ MachinePointerInfo::getFixedStack(FI), false, false,
+ 0));
+ }
- Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
- OutVals[i], Flag);
+ // Transform all store nodes into one single node because
+ // all stores are independent of each other.
+ if (!MemOpChains.empty())
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
+
+ // Now handle return values copied to registers.
+ for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
+ CCValAssign &VA = RVLocs[i];
+ if (!VA.isRegLoc())
+ continue;
+ // Copy the result values into the output registers.
+ Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), OutVals[i], Flag);
// guarantee that all emitted copies are
// stuck together, avoiding something bad
if (Flag.getNode())
RetOps.push_back(Flag);
- return DAG.getNode(XCoreISD::RETSP, dl, MVT::Other,
- &RetOps[0], RetOps.size());
+ return DAG.getNode(XCoreISD::RETSP, dl, MVT::Other, RetOps);
}
//===----------------------------------------------------------------------===//
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
// Next, add the true and fallthrough blocks as its successors.
SDLoc dl(N);
switch (N->getOpcode()) {
default: break;
+ case ISD::INTRINSIC_VOID:
+ switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
+ case Intrinsic::xcore_outt:
+ case Intrinsic::xcore_outct:
+ case Intrinsic::xcore_chkct: {
+ SDValue OutVal = N->getOperand(3);
+ // These instructions ignore the high bits.
+ if (OutVal.hasOneUse()) {
+ unsigned BitWidth = OutVal.getValueSizeInBits();
+ APInt DemandedMask = APInt::getLowBitsSet(BitWidth, 8);
+ APInt KnownZero, KnownOne;
+ TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+ !DCI.isBeforeLegalizeOps());
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ if (TLO.ShrinkDemandedConstant(OutVal, DemandedMask) ||
+ TLI.SimplifyDemandedBits(OutVal, DemandedMask, KnownZero, KnownOne,
+ TLO))
+ DCI.CommitTargetLoweringOpt(TLO);
+ }
+ break;
+ }
+ case Intrinsic::xcore_setpt: {
+ SDValue Time = N->getOperand(3);
+ // This instruction ignores the high bits.
+ if (Time.hasOneUse()) {
+ unsigned BitWidth = Time.getValueSizeInBits();
+ APInt DemandedMask = APInt::getLowBitsSet(BitWidth, 16);
+ APInt KnownZero, KnownOne;
+ TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+ !DCI.isBeforeLegalizeOps());
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ if (TLO.ShrinkDemandedConstant(Time, DemandedMask) ||
+ TLI.SimplifyDemandedBits(Time, DemandedMask, KnownZero, KnownOne,
+ TLO))
+ DCI.CommitTargetLoweringOpt(TLO);
+ }
+ break;
+ }
+ }
+ break;
case XCoreISD::LADD: {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue Result = DAG.getNode(ISD::AND, dl, VT, N2,
DAG.getConstant(1, VT));
SDValue Ops[] = { Result, Carry };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
// fold (ladd x, 0, y) -> 0, add x, y iff carry is unused and y has only the
SDValue Carry = DAG.getConstant(0, VT);
SDValue Result = DAG.getNode(ISD::ADD, dl, VT, N0, N2);
SDValue Ops[] = { Result, Carry };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
}
}
SDValue Result = DAG.getNode(ISD::SUB, dl, VT,
DAG.getConstant(0, VT), N2);
SDValue Ops[] = { Result, Borrow };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
}
SDValue Borrow = DAG.getConstant(0, VT);
SDValue Result = DAG.getNode(ISD::SUB, dl, VT, N0, N2);
SDValue Ops[] = { Result, Borrow };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
}
}
if (N->hasNUsesOfValue(0, 0)) {
SDValue Lo = DAG.getNode(ISD::ADD, dl, VT, N2, N3);
SDValue Ops[] = { Lo, Lo };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
// Otherwise fold to ladd(a, b, 0)
SDValue Result =
DAG.getNode(XCoreISD::LADD, dl, DAG.getVTList(VT, VT), N2, N3, N1);
SDValue Carry(Result.getNode(), 1);
SDValue Ops[] = { Carry, Result };
- return DAG.getMergeValues(Ops, 2, dl);
+ return DAG.getMergeValues(Ops, dl);
}
}
break;
KnownZero.getBitWidth() - 1);
}
break;
+ case ISD::INTRINSIC_W_CHAIN:
+ {
+ unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+ switch (IntNo) {
+ case Intrinsic::xcore_getts:
+ // High bits are known to be zero.
+ KnownZero = APInt::getHighBitsSet(KnownZero.getBitWidth(),
+ KnownZero.getBitWidth() - 16);
+ break;
+ case Intrinsic::xcore_int:
+ case Intrinsic::xcore_inct:
+ // High bits are known to be zero.
+ KnownZero = APInt::getHighBitsSet(KnownZero.getBitWidth(),
+ KnownZero.getBitWidth() - 8);
+ break;
+ case Intrinsic::xcore_testct:
+ // Result is either 0 or 1.
+ KnownZero = APInt::getHighBitsSet(KnownZero.getBitWidth(),
+ KnownZero.getBitWidth() - 1);
+ break;
+ case Intrinsic::xcore_testwct:
+ // Result is in the range 0 - 4.
+ KnownZero = APInt::getHighBitsSet(KnownZero.getBitWidth(),
+ KnownZero.getBitWidth() - 3);
+ break;
+ }
+ }
+ break;
}
}