#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringSwitch.h"
+#include "llvm/Analysis/LibCallSemantics.h"
#include "llvm/CodeGen/IntrinsicLowering.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
setOperationAction(ISD::EH_LABEL, MVT::Other, Expand);
}
- if (Subtarget->isTarget64BitLP64()) {
- setExceptionPointerRegister(X86::RAX);
- setExceptionSelectorRegister(X86::RDX);
- } else {
- setExceptionPointerRegister(X86::EAX);
- setExceptionSelectorRegister(X86::EDX);
- }
setOperationAction(ISD::FRAME_TO_ARGS_OFFSET, MVT::i32, Custom);
setOperationAction(ISD::FRAME_TO_ARGS_OFFSET, MVT::i64, Custom);
DAG.getMachineFunction(), UnwindHelpFI),
/*isVolatile=*/true,
/*isNonTemporal=*/false, /*Alignment=*/0);
- } else {
- // Functions using Win32 EH are considered to have opaque SP adjustments
- // to force local variables to be addressed from the frame or base
- // pointers.
- MFI->setHasOpaqueSPAdjustment(true);
}
}
return V2;
}
+/// \brief Try to lower broadcast of a single - truncated - integer element,
+/// coming from a scalar_to_vector/build_vector node \p V0 with larger elements.
+///
+/// This assumes we have AVX2.
+static SDValue lowerVectorShuffleAsTruncBroadcast(SDLoc DL, MVT VT, SDValue V0,
+ int BroadcastIdx,
+ const X86Subtarget *Subtarget,
+ SelectionDAG &DAG) {
+ assert(Subtarget->hasAVX2() &&
+ "We can only lower integer broadcasts with AVX2!");
+
+ EVT EltVT = VT.getVectorElementType();
+ EVT V0VT = V0.getValueType();
+
+ assert(VT.isInteger() && "Unexpected non-integer trunc broadcast!");
+ assert(V0VT.isVector() && "Unexpected non-vector vector-sized value!");
+
+ EVT V0EltVT = V0VT.getVectorElementType();
+ if (!V0EltVT.isInteger())
+ return SDValue();
+
+ const unsigned EltSize = EltVT.getSizeInBits();
+ const unsigned V0EltSize = V0EltVT.getSizeInBits();
+
+ // This is only a truncation if the original element type is larger.
+ if (V0EltSize <= EltSize)
+ return SDValue();
+
+ assert(((V0EltSize % EltSize) == 0) &&
+ "Scalar type sizes must all be powers of 2 on x86!");
+
+ const unsigned V0Opc = V0.getOpcode();
+ const unsigned Scale = V0EltSize / EltSize;
+ const unsigned V0BroadcastIdx = BroadcastIdx / Scale;
+
+ if ((V0Opc != ISD::SCALAR_TO_VECTOR || V0BroadcastIdx != 0) &&
+ V0Opc != ISD::BUILD_VECTOR)
+ return SDValue();
+
+ SDValue Scalar = V0.getOperand(V0BroadcastIdx);
+
+ // If we're extracting non-least-significant bits, shift so we can truncate.
+ // Hopefully, we can fold away the trunc/srl/load into the broadcast.
+ // Even if we can't (and !isShuffleFoldableLoad(Scalar)), prefer
+ // vpbroadcast+vmovd+shr to vpshufb(m)+vmovd.
+ if (const int OffsetIdx = BroadcastIdx % Scale)
+ Scalar = DAG.getNode(ISD::SRL, DL, Scalar.getValueType(), Scalar,
+ DAG.getConstant(OffsetIdx * EltSize, DL, Scalar.getValueType()));
+
+ return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
+ DAG.getNode(ISD::TRUNCATE, DL, EltVT, Scalar));
+}
+
/// \brief Try to lower broadcast of a single element.
///
/// For convenience, this code also bundles all of the subtarget feature set
// First, look through bitcast: if the original value has a larger element
// type than the shuffle, the broadcast element is in essence truncated.
// Make that explicit to ease folding.
- if (V.getOpcode() == ISD::BITCAST && VT.isInteger()) {
- MVT EltVT = VT.getVectorElementType();
- SDValue V0 = V.getOperand(0);
- MVT V0VT = V0.getSimpleValueType();
-
- if (V0VT.isInteger() && V0VT.getVectorElementType().bitsGT(EltVT) &&
- ((V0.getOpcode() == ISD::BUILD_VECTOR ||
- (V0.getOpcode() == ISD::SCALAR_TO_VECTOR && BroadcastIdx == 0)))) {
- V = DAG.getNode(ISD::TRUNCATE, DL, EltVT, V0.getOperand(BroadcastIdx));
- BroadcastIdx = 0;
- }
- }
+ if (V.getOpcode() == ISD::BITCAST && VT.isInteger())
+ if (SDValue TruncBroadcast = lowerVectorShuffleAsTruncBroadcast(
+ DL, VT, V.getOperand(0), BroadcastIdx, Subtarget, DAG))
+ return TruncBroadcast;
// Also check the simpler case, where we can directly reuse the scalar.
if (V.getOpcode() == ISD::BUILD_VECTOR ||
return DAG.getIntPtrConstant(2 * RegInfo->getSlotSize(), SDLoc(Op));
}
+unsigned X86TargetLowering::getExceptionPointerRegister(
+ const Constant *PersonalityFn) const {
+ if (classifyEHPersonality(PersonalityFn) == EHPersonality::CoreCLR)
+ return Subtarget->isTarget64BitLP64() ? X86::RDX : X86::EDX;
+
+ return Subtarget->isTarget64BitLP64() ? X86::RAX : X86::EAX;
+}
+
+unsigned X86TargetLowering::getExceptionSelectorRegister(
+ const Constant *PersonalityFn) const {
+ // Funclet personalities don't use selectors (the runtime does the selection).
+ assert(!isFuncletEHPersonality(classifyEHPersonality(PersonalityFn)));
+ return Subtarget->isTarget64BitLP64() ? X86::RDX : X86::EDX;
+}
+
SDValue X86TargetLowering::LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
SDValue Offset = Op.getOperand(1);
return BB;
}
+MachineBasicBlock *
+X86TargetLowering::EmitLoweredCatchRet(MachineInstr *MI,
+ MachineBasicBlock *BB) const {
+ MachineFunction *MF = BB->getParent();
+ const Constant *PerFn = MF->getFunction()->getPersonalityFn();
+ bool IsSEH = isAsynchronousEHPersonality(classifyEHPersonality(PerFn));
+ const TargetInstrInfo &TII = *Subtarget->getInstrInfo();
+ MachineBasicBlock *TargetMBB = MI->getOperand(0).getMBB();
+ DebugLoc DL = MI->getDebugLoc();
+
+ // SEH does not outline catch bodies into funclets. Turn CATCHRETs into
+ // JMP_4s, possibly with some extra restoration code for 32-bit EH.
+ if (IsSEH) {
+ if (Subtarget->is32Bit())
+ BuildMI(*BB, MI, DL, TII.get(X86::EH_RESTORE));
+ BuildMI(*BB, MI, DL, TII.get(X86::JMP_4)).addMBB(TargetMBB);
+ MI->eraseFromParent();
+ return BB;
+ }
+
+ // Only 32-bit EH needs to worry about manually restoring stack pointers.
+ if (!Subtarget->is32Bit())
+ return BB;
+
+ // C++ EH creates a new target block to hold the restore code, and wires up
+ // the new block to the return destination with a normal JMP_4.
+ MachineBasicBlock *RestoreMBB =
+ MF->CreateMachineBasicBlock(BB->getBasicBlock());
+ MF->insert(TargetMBB->getIterator(), RestoreMBB);
+ BB->removeSuccessor(TargetMBB);
+ BB->addSuccessor(RestoreMBB);
+ RestoreMBB->addSuccessor(TargetMBB);
+ MI->getOperand(0).setMBB(RestoreMBB);
+
+ auto RestoreMBBI = RestoreMBB->begin();
+ BuildMI(*RestoreMBB, RestoreMBBI, DL, TII.get(X86::EH_RESTORE));
+ BuildMI(*RestoreMBB, RestoreMBBI, DL, TII.get(X86::JMP_4)).addMBB(TargetMBB);
+ return BB;
+}
+
MachineBasicBlock *
X86TargetLowering::EmitLoweredTLSCall(MachineInstr *MI,
MachineBasicBlock *BB) const {
return BB;
case X86::WIN_ALLOCA:
return EmitLoweredWinAlloca(MI, BB);
+ case X86::CATCHRET:
+ return EmitLoweredCatchRet(MI, BB);
case X86::SEG_ALLOCA_32:
case X86::SEG_ALLOCA_64:
return EmitLoweredSegAlloca(MI, BB);
return SDValue();
}
-static SDValue NarrowVectorLoadToElement(LoadSDNode *Load, unsigned Index,
- SelectionDAG &DAG) {
- SDLoc dl(Load);
- MVT VT = Load->getSimpleValueType(0);
- MVT EVT = VT.getVectorElementType();
- SDValue Addr = Load->getOperand(1);
- SDValue NewAddr = DAG.getNode(
- ISD::ADD, dl, Addr.getSimpleValueType(), Addr,
- DAG.getConstant(Index * EVT.getStoreSize(), dl,
- Addr.getSimpleValueType()));
-
- SDValue NewLoad =
- DAG.getLoad(EVT, dl, Load->getChain(), NewAddr,
- DAG.getMachineFunction().getMachineMemOperand(
- Load->getMemOperand(), 0, EVT.getStoreSize()));
- return NewLoad;
-}
-
-static SDValue PerformINSERTPSCombine(SDNode *N, SelectionDAG &DAG,
- const X86Subtarget *Subtarget) {
- SDLoc dl(N);
- MVT VT = N->getOperand(1)->getSimpleValueType(0);
- assert((VT == MVT::v4f32 || VT == MVT::v4i32) &&
- "X86insertps is only defined for v4x32");
-
- SDValue Ld = N->getOperand(1);
- if (MayFoldLoad(Ld)) {
- // Extract the countS bits from the immediate so we can get the proper
- // address when narrowing the vector load to a specific element.
- // When the second source op is a memory address, insertps doesn't use
- // countS and just gets an f32 from that address.
- unsigned DestIndex =
- cast<ConstantSDNode>(N->getOperand(2))->getZExtValue() >> 6;
-
- Ld = NarrowVectorLoadToElement(cast<LoadSDNode>(Ld), DestIndex, DAG);
-
- // Create this as a scalar to vector to match the instruction pattern.
- SDValue LoadScalarToVector = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Ld);
- // countS bits are ignored when loading from memory on insertps, which
- // means we don't need to explicitly set them to 0.
- return DAG.getNode(X86ISD::INSERTPS, dl, VT, N->getOperand(0),
- LoadScalarToVector, N->getOperand(2));
- }
- return SDValue();
-}
-
static SDValue PerformBLENDICombine(SDNode *N, SelectionDAG &DAG) {
SDValue V0 = N->getOperand(0);
SDValue V1 = N->getOperand(1);
}
// Transform (SINT_TO_FP (i64 ...)) into an x87 operation if we have
- // a 32-bit target where SSE doesn't support i64->FP operations.
- if (Op0.getOpcode() == ISD::LOAD) {
+ // a 32-bit target where SSE doesn't support i64->FP operations.
+ if (!Subtarget->useSoftFloat() && Op0.getOpcode() == ISD::LOAD) {
LoadSDNode *Ld = cast<LoadSDNode>(Op0.getNode());
EVT LdVT = Ld->getValueType(0);
case X86ISD::VPERM2X128:
case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, DCI,Subtarget);
case ISD::FMA: return PerformFMACombine(N, DAG, Subtarget);
- case X86ISD::INSERTPS: {
- if (getTargetMachine().getOptLevel() > CodeGenOpt::None)
- return PerformINSERTPSCombine(N, DAG, Subtarget);
- break;
- }
case X86ISD::BLENDI: return PerformBLENDICombine(N, DAG);
}
projects / oota-llvm.git / blobdiff