#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DiagnosticInfo.h"
+#include "llvm/IR/DiagnosticPrinter.h"
using namespace llvm;
+
+namespace {
+
+/// Diagnostic information for unimplemented or unsupported feature reporting.
+class DiagnosticInfoUnsupported : public DiagnosticInfo {
+private:
+ const Twine &Description;
+ const Function &Fn;
+
+ static int KindID;
+
+ static int getKindID() {
+ if (KindID == 0)
+ KindID = llvm::getNextAvailablePluginDiagnosticKind();
+ return KindID;
+ }
+
+public:
+ DiagnosticInfoUnsupported(const Function &Fn, const Twine &Desc,
+ DiagnosticSeverity Severity = DS_Error)
+ : DiagnosticInfo(getKindID(), Severity),
+ Description(Desc),
+ Fn(Fn) { }
+
+ const Function &getFunction() const { return Fn; }
+ const Twine &getDescription() const { return Description; }
+
+ void print(DiagnosticPrinter &DP) const override {
+ DP << "unsupported " << getDescription() << " in " << Fn.getName();
+ }
+
+ static bool classof(const DiagnosticInfo *DI) {
+ return DI->getKind() == getKindID();
+ }
+};
+
+int DiagnosticInfoUnsupported::KindID = 0;
+}
+
+
static bool allocateStack(unsigned ValNo, MVT ValVT, MVT LocVT,
CCValAssign::LocInfo LocInfo,
ISD::ArgFlagsTy ArgFlags, CCState &State) {
setOperationAction(ISD::STORE, MVT::f64, Promote);
AddPromotedToType(ISD::STORE, MVT::f64, MVT::i64);
+ setOperationAction(ISD::STORE, MVT::v2f64, Promote);
+ AddPromotedToType(ISD::STORE, MVT::v2f64, MVT::v2i64);
+
// Custom lowering of vector stores is required for local address space
// stores.
setOperationAction(ISD::STORE, MVT::v4i32, Custom);
// handle 64-bit stores.
setTruncStoreAction(MVT::v4i32, MVT::v4i16, Expand);
+ setTruncStoreAction(MVT::i64, MVT::i16, Expand);
+ setTruncStoreAction(MVT::i64, MVT::i8, Expand);
setTruncStoreAction(MVT::i64, MVT::i1, Expand);
setTruncStoreAction(MVT::v2i64, MVT::v2i1, Expand);
setTruncStoreAction(MVT::v4i64, MVT::v4i1, Expand);
setOperationAction(ISD::LOAD, MVT::f64, Promote);
AddPromotedToType(ISD::LOAD, MVT::f64, MVT::i64);
+ setOperationAction(ISD::LOAD, MVT::v2f64, Promote);
+ AddPromotedToType(ISD::LOAD, MVT::v2f64, MVT::v2i64);
+
setOperationAction(ISD::CONCAT_VECTORS, MVT::v4i32, Custom);
setOperationAction(ISD::CONCAT_VECTORS, MVT::v4f32, Custom);
setOperationAction(ISD::CONCAT_VECTORS, MVT::v8i32, Custom);
setOperationAction(ISD::BR_CC, MVT::i1, Expand);
+ setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
+
setOperationAction(ISD::FNEG, MVT::v2f32, Expand);
setOperationAction(ISD::FNEG, MVT::v4f32, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::MUL, MVT::i64, Expand);
+ setOperationAction(ISD::SUB, MVT::i64, Expand);
setOperationAction(ISD::UDIV, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
+ setOperationAction(ISD::UDIVREM, MVT::i64, Custom);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::VSELECT, MVT::v2f32, Expand);
setOperationAction(ISD::VSELECT, MVT::v4f32, Expand);
MVT::SimpleValueType VT = FloatTypes[x];
setOperationAction(ISD::FABS, VT, Expand);
setOperationAction(ISD::FADD, VT, Expand);
+ setOperationAction(ISD::FCOS, VT, Expand);
setOperationAction(ISD::FDIV, VT, Expand);
setOperationAction(ISD::FPOW, VT, Expand);
setOperationAction(ISD::FFLOOR, VT, Expand);
setOperationAction(ISD::FMUL, VT, Expand);
setOperationAction(ISD::FRINT, VT, Expand);
setOperationAction(ISD::FSQRT, VT, Expand);
+ setOperationAction(ISD::FSIN, VT, Expand);
setOperationAction(ISD::FSUB, VT, Expand);
setOperationAction(ISD::SELECT, VT, Expand);
}
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i1, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i1, Custom);
-
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i8, Custom);
-
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Custom);
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i16, Custom);
-
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Custom);
-
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::Other, Custom);
+ setTargetDAGCombine(ISD::MUL);
+ setTargetDAGCombine(ISD::SELECT_CC);
}
//===----------------------------------------------------------------------===//
// Target specific lowering
//===---------------------------------------------------------------------===//
+SDValue AMDGPUTargetLowering::LowerCall(CallLoweringInfo &CLI,
+ SmallVectorImpl<SDValue> &InVals) const {
+ SDValue Callee = CLI.Callee;
+ SelectionDAG &DAG = CLI.DAG;
+
+ const Function &Fn = *DAG.getMachineFunction().getFunction();
+
+ StringRef FuncName("<unknown>");
+
+ if (const ExternalSymbolSDNode *G = dyn_cast<ExternalSymbolSDNode>(Callee))
+ FuncName = G->getSymbol();
+ else if (const GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
+ FuncName = G->getGlobal()->getName();
+
+ DiagnosticInfoUnsupported NoCalls(Fn, "call to function " + FuncName);
+ DAG.getContext()->diagnose(NoCalls);
+ return SDValue();
+}
+
SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG)
const {
switch (Op.getOpcode()) {
// ReplaceNodeResults to sext_in_reg to an illegal type, so we'll just do
// nothing here and let the illegal result integer be handled normally.
return;
+ case ISD::UDIV: {
+ SDValue Op = SDValue(N, 0);
+ SDLoc DL(Op);
+ EVT VT = Op.getValueType();
+ SDValue UDIVREM = DAG.getNode(ISD::UDIVREM, DL, DAG.getVTList(VT, VT),
+ N->getOperand(0), N->getOperand(1));
+ Results.push_back(UDIVREM);
+ break;
+ }
+ case ISD::UREM: {
+ SDValue Op = SDValue(N, 0);
+ SDLoc DL(Op);
+ EVT VT = Op.getValueType();
+ SDValue UDIVREM = DAG.getNode(ISD::UDIVREM, DL, DAG.getVTList(VT, VT),
+ N->getOperand(0), N->getOperand(1));
+ Results.push_back(UDIVREM.getValue(1));
+ break;
+ }
+ case ISD::UDIVREM: {
+ SDValue Op = SDValue(N, 0);
+ SDLoc DL(Op);
+ EVT VT = Op.getValueType();
+ EVT HalfVT = VT.getHalfSizedIntegerVT(*DAG.getContext());
+
+ SDValue one = DAG.getConstant(1, HalfVT);
+ SDValue zero = DAG.getConstant(0, HalfVT);
+
+ //HiLo split
+ SDValue LHS = N->getOperand(0);
+ SDValue LHS_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, LHS, zero);
+ SDValue LHS_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, LHS, one);
+
+ SDValue RHS = N->getOperand(1);
+ SDValue RHS_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, RHS, zero);
+ SDValue RHS_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, RHS, one);
+
+ // Get Speculative values
+ SDValue DIV_Part = DAG.getNode(ISD::UDIV, DL, HalfVT, LHS_Hi, RHS_Lo);
+ SDValue REM_Part = DAG.getNode(ISD::UREM, DL, HalfVT, LHS_Hi, RHS_Lo);
+
+ SDValue REM_Hi = zero;
+ SDValue REM_Lo = DAG.getSelectCC(DL, RHS_Hi, zero, REM_Part, LHS_Hi, ISD::SETEQ);
+
+ SDValue DIV_Hi = DAG.getSelectCC(DL, RHS_Hi, zero, DIV_Part, zero, ISD::SETEQ);
+ SDValue DIV_Lo = zero;
+
+ const unsigned halfBitWidth = HalfVT.getSizeInBits();
+
+ for (unsigned i = 0; i < halfBitWidth; ++i) {
+ SDValue POS = DAG.getConstant(halfBitWidth - i - 1, HalfVT);
+ // Get Value of high bit
+ SDValue HBit;
+ if (halfBitWidth == 32 && Subtarget->hasBFE()) {
+ HBit = DAG.getNode(AMDGPUISD::BFE_U32, DL, HalfVT, LHS_Lo, POS, one);
+ } else {
+ HBit = DAG.getNode(ISD::SRL, DL, HalfVT, LHS_Lo, POS);
+ HBit = DAG.getNode(ISD::AND, DL, HalfVT, HBit, one);
+ }
+
+ SDValue Carry = DAG.getNode(ISD::SRL, DL, HalfVT, REM_Lo,
+ DAG.getConstant(halfBitWidth - 1, HalfVT));
+ REM_Hi = DAG.getNode(ISD::SHL, DL, HalfVT, REM_Hi, one);
+ REM_Hi = DAG.getNode(ISD::OR, DL, HalfVT, REM_Hi, Carry);
+
+ REM_Lo = DAG.getNode(ISD::SHL, DL, HalfVT, REM_Lo, one);
+ REM_Lo = DAG.getNode(ISD::OR, DL, HalfVT, REM_Lo, HBit);
+
+
+ SDValue REM = DAG.getNode(ISD::BUILD_PAIR, DL, VT, REM_Lo, REM_Hi);
+
+ SDValue BIT = DAG.getConstant(1 << (halfBitWidth - i - 1), HalfVT);
+ SDValue realBIT = DAG.getSelectCC(DL, REM, RHS, BIT, zero, ISD::SETGE);
+
+ DIV_Lo = DAG.getNode(ISD::OR, DL, HalfVT, DIV_Lo, realBIT);
+
+ // Update REM
+
+ SDValue REM_sub = DAG.getNode(ISD::SUB, DL, VT, REM, RHS);
+ REM = DAG.getSelectCC(DL, REM, RHS, REM_sub, REM, ISD::SETGE);
+ REM_Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, REM, zero);
+ REM_Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, HalfVT, REM, one);
+ }
+
+ SDValue REM = DAG.getNode(ISD::BUILD_PAIR, DL, VT, REM_Lo, REM_Hi);
+ SDValue DIV = DAG.getNode(ISD::BUILD_PAIR, DL, VT, DIV_Lo, DIV_Hi);
+ Results.push_back(DIV);
+ Results.push_back(REM);
+ break;
+ }
default:
return;
}
return DAG.getStore(Chain, DL, DAG.getConstant(*CI, VT), InitPtr,
MachinePointerInfo(UndefValue::get(PtrTy)), false, false,
TD->getPrefTypeAlignment(CI->getType()));
- } else if (const ConstantFP *CFP = dyn_cast<ConstantFP>(Init)) {
+ }
+
+ if (const ConstantFP *CFP = dyn_cast<ConstantFP>(Init)) {
EVT VT = EVT::getEVT(CFP->getType());
PointerType *PtrTy = PointerType::get(CFP->getType(), 0);
return DAG.getStore(Chain, DL, DAG.getConstantFP(*CFP, VT), InitPtr,
MachinePointerInfo(UndefValue::get(PtrTy)), false, false,
TD->getPrefTypeAlignment(CFP->getType()));
- } else if (Init->getType()->isAggregateType()) {
+ }
+
+ if (Init->getType()->isAggregateType()) {
EVT PtrVT = InitPtr.getValueType();
unsigned NumElements = Init->getType()->getArrayNumElements();
SmallVector<SDValue, 8> Chains;
Chains.push_back(LowerConstantInitializer(Init->getAggregateElement(i),
GV, Ptr, Chain, DAG));
}
- return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &Chains[0],
- Chains.size());
- } else {
- Init->dump();
- llvm_unreachable("Unhandled constant initializer");
+
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
}
+
+ Init->dump();
+ llvm_unreachable("Unhandled constant initializer");
}
SDValue AMDGPUTargetLowering::LowerGlobalAddress(AMDGPUMachineFunction* MFI,
for (unsigned i = 1; i < (*I)->getNumOperands(); ++i) {
Ops.push_back((*I)->getOperand(i));
}
- DAG.UpdateNodeOperands(*I, &Ops[0], Ops.size());
+ DAG.UpdateNodeOperands(*I, Ops);
}
return DAG.getZExtOrTrunc(InitPtr, SDLoc(Op),
getPointerTy(AMDGPUAS::CONSTANT_ADDRESS));
}
}
-void AMDGPUTargetLowering::ExtractVectorElements(SDValue Op, SelectionDAG &DAG,
- SmallVectorImpl<SDValue> &Args,
- unsigned Start,
- unsigned Count) const {
- EVT VT = Op.getValueType();
- for (unsigned i = Start, e = Start + Count; i != e; ++i) {
- Args.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
- VT.getVectorElementType(),
- Op, DAG.getConstant(i, MVT::i32)));
- }
-}
-
SDValue AMDGPUTargetLowering::LowerCONCAT_VECTORS(SDValue Op,
SelectionDAG &DAG) const {
SmallVector<SDValue, 8> Args;
SDValue A = Op.getOperand(0);
SDValue B = Op.getOperand(1);
- ExtractVectorElements(A, DAG, Args, 0,
- A.getValueType().getVectorNumElements());
- ExtractVectorElements(B, DAG, Args, 0,
- B.getValueType().getVectorNumElements());
+ DAG.ExtractVectorElements(A, Args);
+ DAG.ExtractVectorElements(B, Args);
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(),
- &Args[0], Args.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(), Args);
}
SDValue AMDGPUTargetLowering::LowerEXTRACT_SUBVECTOR(SDValue Op,
SelectionDAG &DAG) const {
SmallVector<SDValue, 8> Args;
- EVT VT = Op.getValueType();
unsigned Start = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
- ExtractVectorElements(Op.getOperand(0), DAG, Args, Start,
- VT.getVectorNumElements());
+ EVT VT = Op.getValueType();
+ DAG.ExtractVectorElements(Op.getOperand(0), Args, Start,
+ VT.getVectorNumElements());
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(),
- &Args[0], Args.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op), Op.getValueType(), Args);
}
SDValue AMDGPUTargetLowering::LowerFrameIndex(SDValue Op,
return DAG.getNode(AMDGPUISD::UMIN, DL, VT, Op.getOperand(1),
Op.getOperand(2));
+ case AMDGPUIntrinsic::AMDGPU_umul24:
+ return DAG.getNode(AMDGPUISD::MUL_U24, DL, VT,
+ Op.getOperand(1), Op.getOperand(2));
+
+ case AMDGPUIntrinsic::AMDGPU_imul24:
+ return DAG.getNode(AMDGPUISD::MUL_I24, DL, VT,
+ Op.getOperand(1), Op.getOperand(2));
+
case AMDGPUIntrinsic::AMDGPU_bfe_i32:
return DAG.getNode(AMDGPUISD::BFE_I32, DL, VT,
Op.getOperand(1),
///IABS(a) = SMAX(sub(0, a), a)
SDValue AMDGPUTargetLowering::LowerIntrinsicIABS(SDValue Op,
- SelectionDAG &DAG) const {
-
+ SelectionDAG &DAG) const {
SDLoc DL(Op);
EVT VT = Op.getValueType();
SDValue Neg = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
/// Linear Interpolation
/// LRP(a, b, c) = muladd(a, b, (1 - a) * c)
SDValue AMDGPUTargetLowering::LowerIntrinsicLRP(SDValue Op,
- SelectionDAG &DAG) const {
+ SelectionDAG &DAG) const {
SDLoc DL(Op);
EVT VT = Op.getValueType();
SDValue OneSubA = DAG.getNode(ISD::FSUB, DL, VT,
}
/// \brief Generate Min/Max node
-SDValue AMDGPUTargetLowering::LowerMinMax(SDValue Op,
- SelectionDAG &DAG) const {
- SDLoc DL(Op);
- EVT VT = Op.getValueType();
+SDValue AMDGPUTargetLowering::CombineMinMax(SDNode *N,
+ SelectionDAG &DAG) const {
+ SDLoc DL(N);
+ EVT VT = N->getValueType(0);
- SDValue LHS = Op.getOperand(0);
- SDValue RHS = Op.getOperand(1);
- SDValue True = Op.getOperand(2);
- SDValue False = Op.getOperand(3);
- SDValue CC = Op.getOperand(4);
+ SDValue LHS = N->getOperand(0);
+ SDValue RHS = N->getOperand(1);
+ SDValue True = N->getOperand(2);
+ SDValue False = N->getOperand(3);
+ SDValue CC = N->getOperand(4);
if (VT != MVT::f32 ||
!((LHS == True && RHS == False) || (LHS == False && RHS == True))) {
case ISD::SETOLT:
case ISD::SETLE:
case ISD::SETLT: {
- if (LHS == True)
- return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
- else
- return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
+ unsigned Opc = (LHS == True) ? AMDGPUISD::FMIN : AMDGPUISD::FMAX;
+ return DAG.getNode(Opc, DL, VT, LHS, RHS);
}
case ISD::SETGT:
case ISD::SETGE:
case ISD::SETOGE:
case ISD::SETUGT:
case ISD::SETOGT: {
- if (LHS == True)
- return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
- else
- return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
+ unsigned Opc = (LHS == True) ? AMDGPUISD::FMAX : AMDGPUISD::FMIN;
+ return DAG.getNode(Opc, DL, VT, LHS, RHS);
}
case ISD::SETCC_INVALID:
llvm_unreachable("Invalid setcc condcode!");
}
- return Op;
+ return SDValue();
}
SDValue AMDGPUTargetLowering::SplitVectorLoad(const SDValue &Op,
MemEltVT, Load->isVolatile(), Load->isNonTemporal(),
Load->getAlignment()));
}
- return DAG.getNode(ISD::BUILD_VECTOR, SL, Op.getValueType(),
- Loads.data(), Loads.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, SL, Op.getValueType(), Loads);
}
SDValue AMDGPUTargetLowering::MergeVectorStore(const SDValue &Op,
}
SDLoc DL(Op);
- const SDValue &Value = Store->getValue();
+ SDValue Value = Store->getValue();
EVT VT = Value.getValueType();
- const SDValue &Ptr = Store->getBasePtr();
+ EVT ElemVT = VT.getVectorElementType();
+ SDValue Ptr = Store->getBasePtr();
EVT MemEltVT = MemVT.getVectorElementType();
unsigned MemEltBits = MemEltVT.getSizeInBits();
unsigned MemNumElements = MemVT.getVectorNumElements();
- EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), MemVT.getSizeInBits());
- SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, PackedVT);
+ unsigned PackedSize = MemVT.getStoreSizeInBits();
+ SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, MVT::i32);
+
+ assert(Value.getValueType().getScalarSizeInBits() >= 32);
SDValue PackedValue;
for (unsigned i = 0; i < MemNumElements; ++i) {
- EVT ElemVT = VT.getVectorElementType();
SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ElemVT, Value,
DAG.getConstant(i, MVT::i32));
- Elt = DAG.getZExtOrTrunc(Elt, DL, PackedVT);
- Elt = DAG.getNode(ISD::AND, DL, PackedVT, Elt, Mask);
- SDValue Shift = DAG.getConstant(MemEltBits * i, PackedVT);
- Elt = DAG.getNode(ISD::SHL, DL, PackedVT, Elt, Shift);
+ Elt = DAG.getZExtOrTrunc(Elt, DL, MVT::i32);
+ Elt = DAG.getNode(ISD::AND, DL, MVT::i32, Elt, Mask); // getZeroExtendInReg
+
+ SDValue Shift = DAG.getConstant(MemEltBits * i, MVT::i32);
+ Elt = DAG.getNode(ISD::SHL, DL, MVT::i32, Elt, Shift);
+
if (i == 0) {
PackedValue = Elt;
} else {
- PackedValue = DAG.getNode(ISD::OR, DL, PackedVT, PackedValue, Elt);
+ PackedValue = DAG.getNode(ISD::OR, DL, MVT::i32, PackedValue, Elt);
}
}
+
+ if (PackedSize < 32) {
+ EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), PackedSize);
+ return DAG.getTruncStore(Store->getChain(), DL, PackedValue, Ptr,
+ Store->getMemOperand()->getPointerInfo(),
+ PackedVT,
+ Store->isNonTemporal(), Store->isVolatile(),
+ Store->getAlignment());
+ }
+
return DAG.getStore(Store->getChain(), DL, PackedValue, Ptr,
- MachinePointerInfo(Store->getMemOperand()->getValue()),
+ Store->getMemOperand()->getPointerInfo(),
Store->isVolatile(), Store->isNonTemporal(),
Store->getAlignment());
}
MemEltVT, Store->isVolatile(), Store->isNonTemporal(),
Store->getAlignment()));
}
- return DAG.getNode(ISD::TokenFactor, SL, MVT::Other, &Chains[0], NumElts);
+ return DAG.getNode(ISD::TokenFactor, SL, MVT::Other, Chains);
}
SDValue AMDGPUTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
return DAG.getNode(ISD::getExtForLoadExtType(ExtType), DL, VT, ExtLoad32);
}
+ if (ExtType == ISD::NON_EXTLOAD && VT.getSizeInBits() < 32) {
+ assert(VT == MVT::i1 && "Only i1 non-extloads expected");
+ // FIXME: Copied from PPC
+ // First, load into 32 bits, then truncate to 1 bit.
+
+ SDValue Chain = Load->getChain();
+ SDValue BasePtr = Load->getBasePtr();
+ MachineMemOperand *MMO = Load->getMemOperand();
+
+ SDValue NewLD = DAG.getExtLoad(ISD::EXTLOAD, DL, MVT::i32, Chain,
+ BasePtr, MVT::i8, MMO);
+ return DAG.getNode(ISD::TRUNCATE, DL, VT, NewLD);
+ }
+
// Lower loads constant address space global variable loads
if (Load->getAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS &&
- isa<GlobalVariable>(GetUnderlyingObject(Load->getPointerInfo().V))) {
+ isa<GlobalVariable>(
+ GetUnderlyingObject(Load->getMemOperand()->getValue()))) {
SDValue Ptr = DAG.getZExtOrTrunc(Load->getBasePtr(), DL,
getPointerTy(AMDGPUAS::PRIVATE_ADDRESS));
}
SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
- SelectionDAG &DAG) const {
+ SelectionDAG &DAG) const {
SDLoc DL(Op);
EVT VT = Op.getValueType();
SDValue Num = Op.getOperand(0);
SDValue Den = Op.getOperand(1);
- SmallVector<SDValue, 8> Results;
-
// RCP = URECIP(Den) = 2^32 / Den + e
// e is rounding error.
SDValue RCP = DAG.getNode(AMDGPUISD::URECIP, DL, VT, Den);
// Rem = (Remainder_GE_Zero == 0 ? Remainder_A_Den : Rem)
Rem = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
Remainder_A_Den, Rem, ISD::SETEQ);
- SDValue Ops[2];
- Ops[0] = Div;
- Ops[1] = Rem;
- return DAG.getMergeValues(Ops, 2, DL);
+ SDValue Ops[2] = {
+ Div,
+ Rem
+ };
+ return DAG.getMergeValues(Ops, DL);
}
SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
MVT VT = Op.getSimpleValueType();
MVT ScalarVT = VT.getScalarType();
- unsigned SrcBits = ExtraVT.getScalarType().getSizeInBits();
- unsigned DestBits = ScalarVT.getSizeInBits();
- unsigned BitsDiff = DestBits - SrcBits;
-
- if (!Subtarget->hasBFE())
- return ExpandSIGN_EXTEND_INREG(Op, BitsDiff, DAG);
+ if (!VT.isVector())
+ return SDValue();
SDValue Src = Op.getOperand(0);
- if (VT.isVector()) {
- SDLoc DL(Op);
- // Need to scalarize this, and revisit each of the scalars later.
- // TODO: Don't scalarize on Evergreen?
- unsigned NElts = VT.getVectorNumElements();
- SmallVector<SDValue, 8> Args;
- ExtractVectorElements(Src, DAG, Args, 0, NElts);
+ SDLoc DL(Op);
- SDValue VTOp = DAG.getValueType(ExtraVT.getScalarType());
- for (unsigned I = 0; I < NElts; ++I)
- Args[I] = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, ScalarVT, Args[I], VTOp);
+ // TODO: Don't scalarize on Evergreen?
+ unsigned NElts = VT.getVectorNumElements();
+ SmallVector<SDValue, 8> Args;
+ DAG.ExtractVectorElements(Src, Args, 0, NElts);
- return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Args.data(), Args.size());
- }
+ SDValue VTOp = DAG.getValueType(ExtraVT.getScalarType());
+ for (unsigned I = 0; I < NElts; ++I)
+ Args[I] = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, ScalarVT, Args[I], VTOp);
- if (SrcBits == 32) {
- SDLoc DL(Op);
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Args);
+}
- // If the source is 32-bits, this is really half of a 2-register pair, and
- // we need to discard the unused half of the pair.
- SDValue TruncSrc = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Src);
- return DAG.getNode(ISD::SIGN_EXTEND, DL, VT, TruncSrc);
- }
+//===----------------------------------------------------------------------===//
+// Custom DAG optimizations
+//===----------------------------------------------------------------------===//
- unsigned NElts = VT.isVector() ? VT.getVectorNumElements() : 1;
+static bool isU24(SDValue Op, SelectionDAG &DAG) {
+ APInt KnownZero, KnownOne;
+ EVT VT = Op.getValueType();
+ DAG.ComputeMaskedBits(Op, KnownZero, KnownOne);
- // TODO: Match 64-bit BFE. SI has a 64-bit BFE, but it's scalar only so it
- // might not be worth the effort, and will need to expand to shifts when
- // fixing SGPR copies.
- if (SrcBits < 32 && DestBits <= 32) {
- SDLoc DL(Op);
- MVT ExtVT = (NElts == 1) ? MVT::i32 : MVT::getVectorVT(MVT::i32, NElts);
-
- if (DestBits != 32)
- Src = DAG.getNode(ISD::ZERO_EXTEND, DL, ExtVT, Src);
-
- // FIXME: This should use TargetConstant, but that hits assertions for
- // Evergreen.
- SDValue Ext = DAG.getNode(AMDGPUISD::BFE_I32, DL, ExtVT,
- Op.getOperand(0), // Operand
- DAG.getConstant(0, ExtVT), // Offset
- DAG.getConstant(SrcBits, ExtVT)); // Width
-
- // Truncate to the original type if necessary.
- if (ScalarVT == MVT::i32)
- return Ext;
- return DAG.getNode(ISD::TRUNCATE, DL, VT, Ext);
- }
+ return (VT.getSizeInBits() - KnownZero.countLeadingOnes()) <= 24;
+}
- // For small types, extend to 32-bits first.
- if (SrcBits < 32) {
- SDLoc DL(Op);
- MVT ExtVT = (NElts == 1) ? MVT::i32 : MVT::getVectorVT(MVT::i32, NElts);
+static bool isI24(SDValue Op, SelectionDAG &DAG) {
+ EVT VT = Op.getValueType();
- SDValue TruncSrc = DAG.getNode(ISD::TRUNCATE, DL, ExtVT, Src);
- SDValue Ext32 = DAG.getNode(AMDGPUISD::BFE_I32,
- DL,
- ExtVT,
- TruncSrc, // Operand
- DAG.getConstant(0, ExtVT), // Offset
- DAG.getConstant(SrcBits, ExtVT)); // Width
+ // In order for this to be a signed 24-bit value, bit 23, must
+ // be a sign bit.
+ return VT.getSizeInBits() >= 24 && // Types less than 24-bit should be treated
+ // as unsigned 24-bit values.
+ (VT.getSizeInBits() - DAG.ComputeNumSignBits(Op)) < 24;
+}
- return DAG.getNode(ISD::SIGN_EXTEND, DL, VT, Ext32);
- }
+static void simplifyI24(SDValue Op, TargetLowering::DAGCombinerInfo &DCI) {
+
+ SelectionDAG &DAG = DCI.DAG;
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ EVT VT = Op.getValueType();
+
+ APInt Demanded = APInt::getLowBitsSet(VT.getSizeInBits(), 24);
+ APInt KnownZero, KnownOne;
+ TargetLowering::TargetLoweringOpt TLO(DAG, true, true);
+ if (TLI.SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO))
+ DCI.CommitTargetLoweringOpt(TLO);
+}
+
+SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N,
+ DAGCombinerInfo &DCI) const {
+ SelectionDAG &DAG = DCI.DAG;
+ SDLoc DL(N);
+
+ switch(N->getOpcode()) {
+ default: break;
+ case ISD::MUL: {
+ EVT VT = N->getValueType(0);
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+ SDValue Mul;
+
+ // FIXME: Add support for 24-bit multiply with 64-bit output on SI.
+ if (VT.isVector() || VT.getSizeInBits() > 32)
+ break;
+
+ if (Subtarget->hasMulU24() && isU24(N0, DAG) && isU24(N1, DAG)) {
+ N0 = DAG.getZExtOrTrunc(N0, DL, MVT::i32);
+ N1 = DAG.getZExtOrTrunc(N1, DL, MVT::i32);
+ Mul = DAG.getNode(AMDGPUISD::MUL_U24, DL, MVT::i32, N0, N1);
+ } else if (Subtarget->hasMulI24() && isI24(N0, DAG) && isI24(N1, DAG)) {
+ N0 = DAG.getSExtOrTrunc(N0, DL, MVT::i32);
+ N1 = DAG.getSExtOrTrunc(N1, DL, MVT::i32);
+ Mul = DAG.getNode(AMDGPUISD::MUL_I24, DL, MVT::i32, N0, N1);
+ } else {
+ break;
+ }
+
+ // We need to use sext even for MUL_U24, because MUL_U24 is used
+ // for signed multiply of 8 and 16-bit types.
+ SDValue Reg = DAG.getSExtOrTrunc(Mul, DL, VT);
- // For everything else, use the standard bitshift expansion.
- return ExpandSIGN_EXTEND_INREG(Op, BitsDiff, DAG);
+ return Reg;
+ }
+ case AMDGPUISD::MUL_I24:
+ case AMDGPUISD::MUL_U24: {
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+ simplifyI24(N0, DCI);
+ simplifyI24(N1, DCI);
+ return SDValue();
+ }
+ case ISD::SELECT_CC: {
+ return CombineMinMax(N, DAG);
+ }
+ }
+ return SDValue();
}
//===----------------------------------------------------------------------===//
const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
- default: return 0;
+ default: return nullptr;
// AMDIL DAG nodes
NODE_NAME_CASE(CALL);
NODE_NAME_CASE(UMUL);
NODE_NAME_CASE(BFE_I32)
NODE_NAME_CASE(BFI)
NODE_NAME_CASE(BFM)
+ NODE_NAME_CASE(MUL_U24)
+ NODE_NAME_CASE(MUL_I24)
NODE_NAME_CASE(URECIP)
NODE_NAME_CASE(DOT4)
NODE_NAME_CASE(EXPORT)