#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallSite.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Module.h"
#include "llvm/MC/MCSectionELF.h"
-#include "llvm/Support/CallSite.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
// TRAP can be lowered to PTX trap
setOperationAction(ISD::TRAP, MVT::Other, Legal);
+ setOperationAction(ISD::ADDC, MVT::i64, Expand);
+ setOperationAction(ISD::ADDE, MVT::i64, Expand);
+
// Register custom handling for vector loads/stores
for (int i = MVT::FIRST_VECTOR_VALUETYPE; i <= MVT::LAST_VECTOR_VALUETYPE;
++i) {
return "NVPTXISD::RET_FLAG";
case NVPTXISD::Wrapper:
return "NVPTXISD::Wrapper";
- case NVPTXISD::NVBuiltin:
- return "NVPTXISD::NVBuiltin";
case NVPTXISD::DeclareParam:
return "NVPTXISD::DeclareParam";
case NVPTXISD::DeclareScalarParam:
return "NVPTXISD::CallSeqBegin";
case NVPTXISD::CallSeqEnd:
return "NVPTXISD::CallSeqEnd";
+ case NVPTXISD::CallPrototype:
+ return "NVPTXISD::CallPrototype";
case NVPTXISD::LoadV2:
return "NVPTXISD::LoadV2";
case NVPTXISD::LoadV4:
return "NVPTXISD::StoreV2";
case NVPTXISD::StoreV4:
return "NVPTXISD::StoreV4";
+ case NVPTXISD::Tex1DFloatI32: return "NVPTXISD::Tex1DFloatI32";
+ case NVPTXISD::Tex1DFloatFloat: return "NVPTXISD::Tex1DFloatFloat";
+ case NVPTXISD::Tex1DFloatFloatLevel:
+ return "NVPTXISD::Tex1DFloatFloatLevel";
+ case NVPTXISD::Tex1DFloatFloatGrad:
+ return "NVPTXISD::Tex1DFloatFloatGrad";
+ case NVPTXISD::Tex1DI32I32: return "NVPTXISD::Tex1DI32I32";
+ case NVPTXISD::Tex1DI32Float: return "NVPTXISD::Tex1DI32Float";
+ case NVPTXISD::Tex1DI32FloatLevel:
+ return "NVPTXISD::Tex1DI32FloatLevel";
+ case NVPTXISD::Tex1DI32FloatGrad:
+ return "NVPTXISD::Tex1DI32FloatGrad";
+ case NVPTXISD::Tex1DArrayFloatI32: return "NVPTXISD::Tex2DArrayFloatI32";
+ case NVPTXISD::Tex1DArrayFloatFloat: return "NVPTXISD::Tex2DArrayFloatFloat";
+ case NVPTXISD::Tex1DArrayFloatFloatLevel:
+ return "NVPTXISD::Tex2DArrayFloatFloatLevel";
+ case NVPTXISD::Tex1DArrayFloatFloatGrad:
+ return "NVPTXISD::Tex2DArrayFloatFloatGrad";
+ case NVPTXISD::Tex1DArrayI32I32: return "NVPTXISD::Tex2DArrayI32I32";
+ case NVPTXISD::Tex1DArrayI32Float: return "NVPTXISD::Tex2DArrayI32Float";
+ case NVPTXISD::Tex1DArrayI32FloatLevel:
+ return "NVPTXISD::Tex2DArrayI32FloatLevel";
+ case NVPTXISD::Tex1DArrayI32FloatGrad:
+ return "NVPTXISD::Tex2DArrayI32FloatGrad";
+ case NVPTXISD::Tex2DFloatI32: return "NVPTXISD::Tex2DFloatI32";
+ case NVPTXISD::Tex2DFloatFloat: return "NVPTXISD::Tex2DFloatFloat";
+ case NVPTXISD::Tex2DFloatFloatLevel:
+ return "NVPTXISD::Tex2DFloatFloatLevel";
+ case NVPTXISD::Tex2DFloatFloatGrad:
+ return "NVPTXISD::Tex2DFloatFloatGrad";
+ case NVPTXISD::Tex2DI32I32: return "NVPTXISD::Tex2DI32I32";
+ case NVPTXISD::Tex2DI32Float: return "NVPTXISD::Tex2DI32Float";
+ case NVPTXISD::Tex2DI32FloatLevel:
+ return "NVPTXISD::Tex2DI32FloatLevel";
+ case NVPTXISD::Tex2DI32FloatGrad:
+ return "NVPTXISD::Tex2DI32FloatGrad";
+ case NVPTXISD::Tex2DArrayFloatI32: return "NVPTXISD::Tex2DArrayFloatI32";
+ case NVPTXISD::Tex2DArrayFloatFloat: return "NVPTXISD::Tex2DArrayFloatFloat";
+ case NVPTXISD::Tex2DArrayFloatFloatLevel:
+ return "NVPTXISD::Tex2DArrayFloatFloatLevel";
+ case NVPTXISD::Tex2DArrayFloatFloatGrad:
+ return "NVPTXISD::Tex2DArrayFloatFloatGrad";
+ case NVPTXISD::Tex2DArrayI32I32: return "NVPTXISD::Tex2DArrayI32I32";
+ case NVPTXISD::Tex2DArrayI32Float: return "NVPTXISD::Tex2DArrayI32Float";
+ case NVPTXISD::Tex2DArrayI32FloatLevel:
+ return "NVPTXISD::Tex2DArrayI32FloatLevel";
+ case NVPTXISD::Tex2DArrayI32FloatGrad:
+ return "NVPTXISD::Tex2DArrayI32FloatGrad";
+ case NVPTXISD::Tex3DFloatI32: return "NVPTXISD::Tex3DFloatI32";
+ case NVPTXISD::Tex3DFloatFloat: return "NVPTXISD::Tex3DFloatFloat";
+ case NVPTXISD::Tex3DFloatFloatLevel:
+ return "NVPTXISD::Tex3DFloatFloatLevel";
+ case NVPTXISD::Tex3DFloatFloatGrad:
+ return "NVPTXISD::Tex3DFloatFloatGrad";
+ case NVPTXISD::Tex3DI32I32: return "NVPTXISD::Tex3DI32I32";
+ case NVPTXISD::Tex3DI32Float: return "NVPTXISD::Tex3DI32Float";
+ case NVPTXISD::Tex3DI32FloatLevel:
+ return "NVPTXISD::Tex3DI32FloatLevel";
+ case NVPTXISD::Tex3DI32FloatGrad:
+ return "NVPTXISD::Tex3DI32FloatGrad";
+
+ case NVPTXISD::Suld1DI8Trap: return "NVPTXISD::Suld1DI8Trap";
+ case NVPTXISD::Suld1DI16Trap: return "NVPTXISD::Suld1DI16Trap";
+ case NVPTXISD::Suld1DI32Trap: return "NVPTXISD::Suld1DI32Trap";
+ case NVPTXISD::Suld1DV2I8Trap: return "NVPTXISD::Suld1DV2I8Trap";
+ case NVPTXISD::Suld1DV2I16Trap: return "NVPTXISD::Suld1DV2I16Trap";
+ case NVPTXISD::Suld1DV2I32Trap: return "NVPTXISD::Suld1DV2I32Trap";
+ case NVPTXISD::Suld1DV4I8Trap: return "NVPTXISD::Suld1DV4I8Trap";
+ case NVPTXISD::Suld1DV4I16Trap: return "NVPTXISD::Suld1DV4I16Trap";
+ case NVPTXISD::Suld1DV4I32Trap: return "NVPTXISD::Suld1DV4I32Trap";
+
+ case NVPTXISD::Suld1DArrayI8Trap: return "NVPTXISD::Suld1DArrayI8Trap";
+ case NVPTXISD::Suld1DArrayI16Trap: return "NVPTXISD::Suld1DArrayI16Trap";
+ case NVPTXISD::Suld1DArrayI32Trap: return "NVPTXISD::Suld1DArrayI32Trap";
+ case NVPTXISD::Suld1DArrayV2I8Trap: return "NVPTXISD::Suld1DArrayV2I8Trap";
+ case NVPTXISD::Suld1DArrayV2I16Trap: return "NVPTXISD::Suld1DArrayV2I16Trap";
+ case NVPTXISD::Suld1DArrayV2I32Trap: return "NVPTXISD::Suld1DArrayV2I32Trap";
+ case NVPTXISD::Suld1DArrayV4I8Trap: return "NVPTXISD::Suld1DArrayV4I8Trap";
+ case NVPTXISD::Suld1DArrayV4I16Trap: return "NVPTXISD::Suld1DArrayV4I16Trap";
+ case NVPTXISD::Suld1DArrayV4I32Trap: return "NVPTXISD::Suld1DArrayV4I32Trap";
+
+ case NVPTXISD::Suld2DI8Trap: return "NVPTXISD::Suld2DI8Trap";
+ case NVPTXISD::Suld2DI16Trap: return "NVPTXISD::Suld2DI16Trap";
+ case NVPTXISD::Suld2DI32Trap: return "NVPTXISD::Suld2DI32Trap";
+ case NVPTXISD::Suld2DV2I8Trap: return "NVPTXISD::Suld2DV2I8Trap";
+ case NVPTXISD::Suld2DV2I16Trap: return "NVPTXISD::Suld2DV2I16Trap";
+ case NVPTXISD::Suld2DV2I32Trap: return "NVPTXISD::Suld2DV2I32Trap";
+ case NVPTXISD::Suld2DV4I8Trap: return "NVPTXISD::Suld2DV4I8Trap";
+ case NVPTXISD::Suld2DV4I16Trap: return "NVPTXISD::Suld2DV4I16Trap";
+ case NVPTXISD::Suld2DV4I32Trap: return "NVPTXISD::Suld2DV4I32Trap";
+
+ case NVPTXISD::Suld2DArrayI8Trap: return "NVPTXISD::Suld2DArrayI8Trap";
+ case NVPTXISD::Suld2DArrayI16Trap: return "NVPTXISD::Suld2DArrayI16Trap";
+ case NVPTXISD::Suld2DArrayI32Trap: return "NVPTXISD::Suld2DArrayI32Trap";
+ case NVPTXISD::Suld2DArrayV2I8Trap: return "NVPTXISD::Suld2DArrayV2I8Trap";
+ case NVPTXISD::Suld2DArrayV2I16Trap: return "NVPTXISD::Suld2DArrayV2I16Trap";
+ case NVPTXISD::Suld2DArrayV2I32Trap: return "NVPTXISD::Suld2DArrayV2I32Trap";
+ case NVPTXISD::Suld2DArrayV4I8Trap: return "NVPTXISD::Suld2DArrayV4I8Trap";
+ case NVPTXISD::Suld2DArrayV4I16Trap: return "NVPTXISD::Suld2DArrayV4I16Trap";
+ case NVPTXISD::Suld2DArrayV4I32Trap: return "NVPTXISD::Suld2DArrayV4I32Trap";
+
+ case NVPTXISD::Suld3DI8Trap: return "NVPTXISD::Suld3DI8Trap";
+ case NVPTXISD::Suld3DI16Trap: return "NVPTXISD::Suld3DI16Trap";
+ case NVPTXISD::Suld3DI32Trap: return "NVPTXISD::Suld3DI32Trap";
+ case NVPTXISD::Suld3DV2I8Trap: return "NVPTXISD::Suld3DV2I8Trap";
+ case NVPTXISD::Suld3DV2I16Trap: return "NVPTXISD::Suld3DV2I16Trap";
+ case NVPTXISD::Suld3DV2I32Trap: return "NVPTXISD::Suld3DV2I32Trap";
+ case NVPTXISD::Suld3DV4I8Trap: return "NVPTXISD::Suld3DV4I8Trap";
+ case NVPTXISD::Suld3DV4I16Trap: return "NVPTXISD::Suld3DV4I16Trap";
+ case NVPTXISD::Suld3DV4I32Trap: return "NVPTXISD::Suld3DV4I32Trap";
}
}
-bool NVPTXTargetLowering::shouldSplitVectorElementType(EVT VT) const {
- return VT == MVT::i1;
+bool NVPTXTargetLowering::shouldSplitVectorType(EVT VT) const {
+ return VT.getScalarType() == MVT::i1;
}
SDValue
return DAG.getNode(NVPTXISD::Wrapper, dl, getPointerTy(), Op);
}
-/*
-std::string NVPTXTargetLowering::getPrototype(
- Type *retTy, const ArgListTy &Args,
- const SmallVectorImpl<ISD::OutputArg> &Outs, unsigned retAlignment) const {
-
- bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
-
- std::stringstream O;
- O << "prototype_" << uniqueCallSite << " : .callprototype ";
-
- if (retTy->getTypeID() == Type::VoidTyID)
- O << "()";
- else {
- O << "(";
- if (isABI) {
- if (retTy->isPrimitiveType() || retTy->isIntegerTy()) {
- unsigned size = 0;
- if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
- size = ITy->getBitWidth();
- if (size < 32)
- size = 32;
- } else {
- assert(retTy->isFloatingPointTy() &&
- "Floating point type expected here");
- size = retTy->getPrimitiveSizeInBits();
- }
-
- O << ".param .b" << size << " _";
- } else if (isa<PointerType>(retTy))
- O << ".param .b" << getPointerTy().getSizeInBits() << " _";
- else {
- if ((retTy->getTypeID() == Type::StructTyID) ||
- isa<VectorType>(retTy)) {
- SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*this, retTy, vtparts);
- unsigned totalsz = 0;
- for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
- unsigned elems = 1;
- EVT elemtype = vtparts[i];
- if (vtparts[i].isVector()) {
- elems = vtparts[i].getVectorNumElements();
- elemtype = vtparts[i].getVectorElementType();
- }
- for (unsigned j = 0, je = elems; j != je; ++j) {
- unsigned sz = elemtype.getSizeInBits();
- if (elemtype.isInteger() && (sz < 8))
- sz = 8;
- totalsz += sz / 8;
- }
- }
- O << ".param .align " << retAlignment << " .b8 _[" << totalsz << "]";
- } else {
- assert(false && "Unknown return type");
- }
- }
- } else {
- SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*this, retTy, vtparts);
- unsigned idx = 0;
- for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
- unsigned elems = 1;
- EVT elemtype = vtparts[i];
- if (vtparts[i].isVector()) {
- elems = vtparts[i].getVectorNumElements();
- elemtype = vtparts[i].getVectorElementType();
- }
-
- for (unsigned j = 0, je = elems; j != je; ++j) {
- unsigned sz = elemtype.getSizeInBits();
- if (elemtype.isInteger() && (sz < 32))
- sz = 32;
- O << ".reg .b" << sz << " _";
- if (j < je - 1)
- O << ", ";
- ++idx;
- }
- if (i < e - 1)
- O << ", ";
- }
- }
- O << ") ";
- }
- O << "_ (";
-
- bool first = true;
- MVT thePointerTy = getPointerTy();
-
- for (unsigned i = 0, e = Args.size(); i != e; ++i) {
- const Type *Ty = Args[i].Ty;
- if (!first) {
- O << ", ";
- }
- first = false;
-
- if (Outs[i].Flags.isByVal() == false) {
- unsigned sz = 0;
- if (isa<IntegerType>(Ty)) {
- sz = cast<IntegerType>(Ty)->getBitWidth();
- if (sz < 32)
- sz = 32;
- } else if (isa<PointerType>(Ty))
- sz = thePointerTy.getSizeInBits();
- else
- sz = Ty->getPrimitiveSizeInBits();
- if (isABI)
- O << ".param .b" << sz << " ";
- else
- O << ".reg .b" << sz << " ";
- O << "_";
- continue;
- }
- const PointerType *PTy = dyn_cast<PointerType>(Ty);
- assert(PTy && "Param with byval attribute should be a pointer type");
- Type *ETy = PTy->getElementType();
-
- if (isABI) {
- unsigned align = Outs[i].Flags.getByValAlign();
- unsigned sz = getDataLayout()->getTypeAllocSize(ETy);
- O << ".param .align " << align << " .b8 ";
- O << "_";
- O << "[" << sz << "]";
- continue;
- } else {
- SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*this, ETy, vtparts);
- for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
- unsigned elems = 1;
- EVT elemtype = vtparts[i];
- if (vtparts[i].isVector()) {
- elems = vtparts[i].getVectorNumElements();
- elemtype = vtparts[i].getVectorElementType();
- }
-
- for (unsigned j = 0, je = elems; j != je; ++j) {
- unsigned sz = elemtype.getSizeInBits();
- if (elemtype.isInteger() && (sz < 32))
- sz = 32;
- O << ".reg .b" << sz << " ";
- O << "_";
- if (j < je - 1)
- O << ", ";
- }
- if (i < e - 1)
- O << ", ";
- }
- continue;
- }
- }
- O << ");";
- return O.str();
-}*/
-
std::string
NVPTXTargetLowering::getPrototype(Type *retTy, const ArgListTy &Args,
const SmallVectorImpl<ISD::OutputArg> &Outs,
O << "()";
} else {
O << "(";
- if (retTy->isPrimitiveType() || retTy->isIntegerTy()) {
+ if (retTy->isFloatingPointTy() || retTy->isIntegerTy()) {
unsigned size = 0;
if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
size = ITy->getBitWidth();
OIdx += len - 1;
continue;
}
- assert(getValueType(Ty) == Outs[OIdx].VT &&
+ // i8 types in IR will be i16 types in SDAG
+ assert((getValueType(Ty) == Outs[OIdx].VT ||
+ (getValueType(Ty) == MVT::i8 && Outs[OIdx].VT == MVT::i16)) &&
"type mismatch between callee prototype and arguments");
// scalar type
unsigned sz = 0;
Type *Ty,
unsigned Idx) const {
const DataLayout *TD = getDataLayout();
- unsigned align = 0;
- GlobalAddressSDNode *Func = dyn_cast<GlobalAddressSDNode>(Callee.getNode());
+ unsigned Align = 0;
+ const Value *DirectCallee = CS->getCalledFunction();
+
+ if (!DirectCallee) {
+ // We don't have a direct function symbol, but that may be because of
+ // constant cast instructions in the call.
+ const Instruction *CalleeI = CS->getInstruction();
+ assert(CalleeI && "Call target is not a function or derived value?");
+
+ // With bitcast'd call targets, the instruction will be the call
+ if (isa<CallInst>(CalleeI)) {
+ // Check if we have call alignment metadata
+ if (llvm::getAlign(*cast<CallInst>(CalleeI), Idx, Align))
+ return Align;
+
+ const Value *CalleeV = cast<CallInst>(CalleeI)->getCalledValue();
+ // Ignore any bitcast instructions
+ while(isa<ConstantExpr>(CalleeV)) {
+ const ConstantExpr *CE = cast<ConstantExpr>(CalleeV);
+ if (!CE->isCast())
+ break;
+ // Look through the bitcast
+ CalleeV = cast<ConstantExpr>(CalleeV)->getOperand(0);
+ }
- if (Func) { // direct call
- assert(CS->getCalledFunction() &&
- "direct call cannot find callee");
- if (!llvm::getAlign(*(CS->getCalledFunction()), Idx, align))
- align = TD->getABITypeAlignment(Ty);
- }
- else { // indirect call
- const CallInst *CallI = dyn_cast<CallInst>(CS->getInstruction());
- if (!llvm::getAlign(*CallI, Idx, align))
- align = TD->getABITypeAlignment(Ty);
+ // We have now looked past all of the bitcasts. Do we finally have a
+ // Function?
+ if (isa<Function>(CalleeV))
+ DirectCallee = CalleeV;
+ }
}
- return align;
+ // Check for function alignment information if we found that the
+ // ultimate target is a Function
+ if (DirectCallee)
+ if (llvm::getAlign(*cast<Function>(DirectCallee), Idx, Align))
+ return Align;
+
+ // Call is indirect or alignment information is not available, fall back to
+ // the ABI type alignment
+ return TD->getABITypeAlignment(Ty);
}
SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
SDLoc dl = CLI.DL;
- SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
- SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
- SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
+ SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
+ SmallVectorImpl<SDValue> &OutVals = CLI.OutVals;
+ SmallVectorImpl<ISD::InputArg> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &isTailCall = CLI.IsTailCall;
sz = 8;
SDValue StVal = OutVals[OIdx];
if (elemtype.getSizeInBits() < 16) {
- StVal = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, StVal);
+ StVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, StVal);
}
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain,
Ops.push_back(StoreVal);
}
+ Ops.push_back(InFlag);
+
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
Chain = DAG.getMemIntrinsicNode(Opc, dl, CopyParamVTs, &Ops[0],
Ops.size(), MemVT,
MachinePointerInfo(), false, false, false,
0);
if (elemtype.getSizeInBits() < 16) {
- theVal = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, theVal);
+ theVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, theVal);
}
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
// .param .align 16 .b8 retval0[<size-in-bytes>], or
// .param .b<size-in-bits> retval0
unsigned resultsz = TD->getTypeAllocSizeInBits(retTy);
- if (retTy->isPrimitiveType() || retTy->isIntegerTy() ||
- retTy->isPointerTy()) {
+ if (retTy->isSingleValueType()) {
// Scalar needs to be at least 32bit wide
if (resultsz < 32)
resultsz = 32;
// proto_0 : .callprototype(.param .b32 _) _ (.param .b32 _);
// to be emitted, and the label has to used as the last arg of call
// instruction.
- // The prototype is embedded in a string and put as the operand for an
- // INLINEASM SDNode.
- SDVTList InlineAsmVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- std::string proto_string =
- getPrototype(retTy, Args, Outs, retAlignment, CS);
- const char *asmstr = nvTM->getManagedStrPool()
- ->getManagedString(proto_string.c_str())->c_str();
- SDValue InlineAsmOps[] = {
- Chain, DAG.getTargetExternalSymbol(asmstr, getPointerTy()),
- DAG.getMDNode(0), DAG.getTargetConstant(0, MVT::i32), InFlag
+ // The prototype is embedded in a string and put as the operand for a
+ // CallPrototype SDNode which will print out to the value of the string.
+ SDVTList ProtoVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+ std::string Proto = getPrototype(retTy, Args, Outs, retAlignment, CS);
+ const char *ProtoStr =
+ nvTM->getManagedStrPool()->getManagedString(Proto.c_str())->c_str();
+ SDValue ProtoOps[] = {
+ Chain, DAG.getTargetExternalSymbol(ProtoStr, MVT::i32), InFlag,
};
- Chain = DAG.getNode(ISD::INLINEASM, dl, InlineAsmVTs, InlineAsmOps, 5);
+ Chain = DAG.getNode(NVPTXISD::CallPrototype, dl, ProtoVTs, &ProtoOps[0], 3);
InFlag = Chain.getValue(1);
}
// Op to just print "call"
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParam, dl,
- DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
+ DAG.getVTList(LoadRetVTs), &LoadRetOps[0],
LoadRetOps.size(), EltVT, MachinePointerInfo());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParamV2, dl,
- DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
+ DAG.getVTList(LoadRetVTs), &LoadRetOps[0],
LoadRetOps.size(), EltVT, MachinePointerInfo());
Chain = retval.getValue(2);
InFlag = retval.getValue(3);
LoadRetOps.push_back(DAG.getConstant(Ofst, MVT::i32));
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getMemIntrinsicNode(
- Opc, dl, DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()),
+ Opc, dl, DAG.getVTList(LoadRetVTs),
&LoadRetOps[0], LoadRetOps.size(), EltVT, MachinePointerInfo());
if (VecSize == 2) {
Chain = retval.getValue(2);
LoadRetOps.push_back(InFlag);
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParam, dl,
- DAG.getVTList(&LoadRetVTs[0], LoadRetVTs.size()), &LoadRetOps[0],
+ DAG.getVTList(LoadRetVTs), &LoadRetOps[0],
LoadRetOps.size(), TheLoadType, MachinePointerInfo());
Chain = retval.getValue(1);
InFlag = retval.getValue(2);
// Since StoreV2 is a target node, we cannot rely on DAG type legalization.
// Therefore, we must ensure the type is legal. For i1 and i8, we set the
- // stored type to i16 and propogate the "real" type as the memory type.
- bool NeedSExt = false;
+ // stored type to i16 and propagate the "real" type as the memory type.
+ bool NeedExt = false;
if (EltVT.getSizeInBits() < 16)
- NeedSExt = true;
+ NeedExt = true;
switch (NumElts) {
default:
for (unsigned i = 0; i < NumElts; ++i) {
SDValue ExtVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, Val,
DAG.getIntPtrConstant(i));
- if (NeedSExt)
- ExtVal = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i16, ExtVal);
+ if (NeedExt)
+ ExtVal = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i16, ExtVal);
Ops.push_back(ExtVal);
}
SDValue
NVPTXTargetLowering::getParamSymbol(SelectionDAG &DAG, int idx, EVT v) const {
- return getExtSymb(DAG, ".PARAM", idx, v);
+ std::string ParamSym;
+ raw_string_ostream ParamStr(ParamSym);
+
+ ParamStr << DAG.getMachineFunction().getName() << "_param_" << idx;
+ ParamStr.flush();
+
+ std::string *SavedStr =
+ nvTM->getManagedStrPool()->getManagedString(ParamSym.c_str());
+ return DAG.getTargetExternalSymbol(SavedStr->c_str(), v);
}
SDValue NVPTXTargetLowering::getParamHelpSymbol(SelectionDAG &DAG, int idx) {
aggregateIsPacked ? 1
: TD->getABITypeAlignment(
partVT.getTypeForEVT(F->getContext()));
- SDValue p;
- if (Ins[InsIdx].VT.getSizeInBits() > partVT.getSizeInBits())
- p = DAG.getExtLoad(ISD::SEXTLOAD, dl, Ins[InsIdx].VT, Root, srcAddr,
+ SDValue p;
+ if (Ins[InsIdx].VT.getSizeInBits() > partVT.getSizeInBits()) {
+ ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
+ ISD::SEXTLOAD : ISD::ZEXTLOAD;
+ p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, srcAddr,
MachinePointerInfo(srcValue), partVT, false,
false, partAlign);
- else
+ } else {
p = DAG.getLoad(partVT, dl, Root, srcAddr,
MachinePointerInfo(srcValue), false, false, false,
partAlign);
+ }
if (p.getNode())
p.getNode()->setIROrder(idx + 1);
InVals.push_back(p);
P.getNode()->setIROrder(idx + 1);
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits())
- P = DAG.getNode(ISD::SIGN_EXTEND, dl, Ins[InsIdx].VT, P);
+ P = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, P);
InVals.push_back(P);
Ofst += TD->getTypeAllocSize(EltVT.getTypeForEVT(F->getContext()));
++InsIdx;
DAG.getIntPtrConstant(1));
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits()) {
- Elt0 = DAG.getNode(ISD::SIGN_EXTEND, dl, Ins[InsIdx].VT, Elt0);
- Elt1 = DAG.getNode(ISD::SIGN_EXTEND, dl, Ins[InsIdx].VT, Elt1);
+ Elt0 = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, Elt0);
+ Elt1 = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, Elt1);
}
InVals.push_back(Elt0);
SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, P,
DAG.getIntPtrConstant(j));
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits())
- Elt = DAG.getNode(ISD::SIGN_EXTEND, dl, Ins[InsIdx].VT, Elt);
+ Elt = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, Elt);
InVals.push_back(Elt);
}
Ofst += TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
- InsIdx += VecSize;
}
+ InsIdx += NumElts;
}
if (NumElts > 0)
Value *srcValue = Constant::getNullValue(PointerType::get(
ObjectVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
SDValue p;
- if (ObjectVT.getSizeInBits() < Ins[InsIdx].VT.getSizeInBits())
- p = DAG.getExtLoad(ISD::SEXTLOAD, dl, Ins[InsIdx].VT, Root, Arg,
+ if (ObjectVT.getSizeInBits() < Ins[InsIdx].VT.getSizeInBits()) {
+ ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
+ ISD::SEXTLOAD : ISD::ZEXTLOAD;
+ p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, Arg,
MachinePointerInfo(srcValue), ObjectVT, false, false,
- TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
- else
+ TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
+ } else {
p = DAG.getLoad(Ins[InsIdx].VT, dl, Root, Arg,
MachinePointerInfo(srcValue), false, false, false,
- TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
+ TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
+ }
if (p.getNode())
p.getNode()->setIROrder(idx + 1);
InVals.push_back(p);
return false;
}
+static unsigned getOpcForTextureInstr(unsigned Intrinsic) {
+ switch (Intrinsic) {
+ default:
+ return 0;
+
+ case Intrinsic::nvvm_tex_1d_v4f32_i32:
+ return NVPTXISD::Tex1DFloatI32;
+ case Intrinsic::nvvm_tex_1d_v4f32_f32:
+ return NVPTXISD::Tex1DFloatFloat;
+ case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
+ return NVPTXISD::Tex1DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
+ return NVPTXISD::Tex1DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_1d_v4i32_i32:
+ return NVPTXISD::Tex1DI32I32;
+ case Intrinsic::nvvm_tex_1d_v4i32_f32:
+ return NVPTXISD::Tex1DI32Float;
+ case Intrinsic::nvvm_tex_1d_level_v4i32_f32:
+ return NVPTXISD::Tex1DI32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_grad_v4i32_f32:
+ return NVPTXISD::Tex1DI32FloatGrad;
+
+ case Intrinsic::nvvm_tex_1d_array_v4f32_i32:
+ return NVPTXISD::Tex1DArrayFloatI32;
+ case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
+ return NVPTXISD::Tex1DArrayFloatFloat;
+ case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
+ return NVPTXISD::Tex1DArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
+ return NVPTXISD::Tex1DArrayFloatFloatGrad;
+ case Intrinsic::nvvm_tex_1d_array_v4i32_i32:
+ return NVPTXISD::Tex1DArrayI32I32;
+ case Intrinsic::nvvm_tex_1d_array_v4i32_f32:
+ return NVPTXISD::Tex1DArrayI32Float;
+ case Intrinsic::nvvm_tex_1d_array_level_v4i32_f32:
+ return NVPTXISD::Tex1DArrayI32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_array_grad_v4i32_f32:
+ return NVPTXISD::Tex1DArrayI32FloatGrad;
+
+ case Intrinsic::nvvm_tex_2d_v4f32_i32:
+ return NVPTXISD::Tex2DFloatI32;
+ case Intrinsic::nvvm_tex_2d_v4f32_f32:
+ return NVPTXISD::Tex2DFloatFloat;
+ case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
+ return NVPTXISD::Tex2DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
+ return NVPTXISD::Tex2DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_2d_v4i32_i32:
+ return NVPTXISD::Tex2DI32I32;
+ case Intrinsic::nvvm_tex_2d_v4i32_f32:
+ return NVPTXISD::Tex2DI32Float;
+ case Intrinsic::nvvm_tex_2d_level_v4i32_f32:
+ return NVPTXISD::Tex2DI32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_grad_v4i32_f32:
+ return NVPTXISD::Tex2DI32FloatGrad;
+
+ case Intrinsic::nvvm_tex_2d_array_v4f32_i32:
+ return NVPTXISD::Tex2DArrayFloatI32;
+ case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
+ return NVPTXISD::Tex2DArrayFloatFloat;
+ case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
+ return NVPTXISD::Tex2DArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
+ return NVPTXISD::Tex2DArrayFloatFloatGrad;
+ case Intrinsic::nvvm_tex_2d_array_v4i32_i32:
+ return NVPTXISD::Tex2DArrayI32I32;
+ case Intrinsic::nvvm_tex_2d_array_v4i32_f32:
+ return NVPTXISD::Tex2DArrayI32Float;
+ case Intrinsic::nvvm_tex_2d_array_level_v4i32_f32:
+ return NVPTXISD::Tex2DArrayI32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_array_grad_v4i32_f32:
+ return NVPTXISD::Tex2DArrayI32FloatGrad;
+
+ case Intrinsic::nvvm_tex_3d_v4f32_i32:
+ return NVPTXISD::Tex3DFloatI32;
+ case Intrinsic::nvvm_tex_3d_v4f32_f32:
+ return NVPTXISD::Tex3DFloatFloat;
+ case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
+ return NVPTXISD::Tex3DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_3d_grad_v4f32_f32:
+ return NVPTXISD::Tex3DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_3d_v4i32_i32:
+ return NVPTXISD::Tex3DI32I32;
+ case Intrinsic::nvvm_tex_3d_v4i32_f32:
+ return NVPTXISD::Tex3DI32Float;
+ case Intrinsic::nvvm_tex_3d_level_v4i32_f32:
+ return NVPTXISD::Tex3DI32FloatLevel;
+ case Intrinsic::nvvm_tex_3d_grad_v4i32_f32:
+ return NVPTXISD::Tex3DI32FloatGrad;
+ }
+}
+
+static unsigned getOpcForSurfaceInstr(unsigned Intrinsic) {
+ switch (Intrinsic) {
+ default:
+ return 0;
+ case Intrinsic::nvvm_suld_1d_i8_trap:
+ return NVPTXISD::Suld1DI8Trap;
+ case Intrinsic::nvvm_suld_1d_i16_trap:
+ return NVPTXISD::Suld1DI16Trap;
+ case Intrinsic::nvvm_suld_1d_i32_trap:
+ return NVPTXISD::Suld1DI32Trap;
+ case Intrinsic::nvvm_suld_1d_v2i8_trap:
+ return NVPTXISD::Suld1DV2I8Trap;
+ case Intrinsic::nvvm_suld_1d_v2i16_trap:
+ return NVPTXISD::Suld1DV2I16Trap;
+ case Intrinsic::nvvm_suld_1d_v2i32_trap:
+ return NVPTXISD::Suld1DV2I32Trap;
+ case Intrinsic::nvvm_suld_1d_v4i8_trap:
+ return NVPTXISD::Suld1DV4I8Trap;
+ case Intrinsic::nvvm_suld_1d_v4i16_trap:
+ return NVPTXISD::Suld1DV4I16Trap;
+ case Intrinsic::nvvm_suld_1d_v4i32_trap:
+ return NVPTXISD::Suld1DV4I32Trap;
+ case Intrinsic::nvvm_suld_1d_array_i8_trap:
+ return NVPTXISD::Suld1DArrayI8Trap;
+ case Intrinsic::nvvm_suld_1d_array_i16_trap:
+ return NVPTXISD::Suld1DArrayI16Trap;
+ case Intrinsic::nvvm_suld_1d_array_i32_trap:
+ return NVPTXISD::Suld1DArrayI32Trap;
+ case Intrinsic::nvvm_suld_1d_array_v2i8_trap:
+ return NVPTXISD::Suld1DArrayV2I8Trap;
+ case Intrinsic::nvvm_suld_1d_array_v2i16_trap:
+ return NVPTXISD::Suld1DArrayV2I16Trap;
+ case Intrinsic::nvvm_suld_1d_array_v2i32_trap:
+ return NVPTXISD::Suld1DArrayV2I32Trap;
+ case Intrinsic::nvvm_suld_1d_array_v4i8_trap:
+ return NVPTXISD::Suld1DArrayV4I8Trap;
+ case Intrinsic::nvvm_suld_1d_array_v4i16_trap:
+ return NVPTXISD::Suld1DArrayV4I16Trap;
+ case Intrinsic::nvvm_suld_1d_array_v4i32_trap:
+ return NVPTXISD::Suld1DArrayV4I32Trap;
+ case Intrinsic::nvvm_suld_2d_i8_trap:
+ return NVPTXISD::Suld2DI8Trap;
+ case Intrinsic::nvvm_suld_2d_i16_trap:
+ return NVPTXISD::Suld2DI16Trap;
+ case Intrinsic::nvvm_suld_2d_i32_trap:
+ return NVPTXISD::Suld2DI32Trap;
+ case Intrinsic::nvvm_suld_2d_v2i8_trap:
+ return NVPTXISD::Suld2DV2I8Trap;
+ case Intrinsic::nvvm_suld_2d_v2i16_trap:
+ return NVPTXISD::Suld2DV2I16Trap;
+ case Intrinsic::nvvm_suld_2d_v2i32_trap:
+ return NVPTXISD::Suld2DV2I32Trap;
+ case Intrinsic::nvvm_suld_2d_v4i8_trap:
+ return NVPTXISD::Suld2DV4I8Trap;
+ case Intrinsic::nvvm_suld_2d_v4i16_trap:
+ return NVPTXISD::Suld2DV4I16Trap;
+ case Intrinsic::nvvm_suld_2d_v4i32_trap:
+ return NVPTXISD::Suld2DV4I32Trap;
+ case Intrinsic::nvvm_suld_2d_array_i8_trap:
+ return NVPTXISD::Suld2DArrayI8Trap;
+ case Intrinsic::nvvm_suld_2d_array_i16_trap:
+ return NVPTXISD::Suld2DArrayI16Trap;
+ case Intrinsic::nvvm_suld_2d_array_i32_trap:
+ return NVPTXISD::Suld2DArrayI32Trap;
+ case Intrinsic::nvvm_suld_2d_array_v2i8_trap:
+ return NVPTXISD::Suld2DArrayV2I8Trap;
+ case Intrinsic::nvvm_suld_2d_array_v2i16_trap:
+ return NVPTXISD::Suld2DArrayV2I16Trap;
+ case Intrinsic::nvvm_suld_2d_array_v2i32_trap:
+ return NVPTXISD::Suld2DArrayV2I32Trap;
+ case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
+ return NVPTXISD::Suld2DArrayV4I8Trap;
+ case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
+ return NVPTXISD::Suld2DArrayV4I16Trap;
+ case Intrinsic::nvvm_suld_2d_array_v4i32_trap:
+ return NVPTXISD::Suld2DArrayV4I32Trap;
+ case Intrinsic::nvvm_suld_3d_i8_trap:
+ return NVPTXISD::Suld3DI8Trap;
+ case Intrinsic::nvvm_suld_3d_i16_trap:
+ return NVPTXISD::Suld3DI16Trap;
+ case Intrinsic::nvvm_suld_3d_i32_trap:
+ return NVPTXISD::Suld3DI32Trap;
+ case Intrinsic::nvvm_suld_3d_v2i8_trap:
+ return NVPTXISD::Suld3DV2I8Trap;
+ case Intrinsic::nvvm_suld_3d_v2i16_trap:
+ return NVPTXISD::Suld3DV2I16Trap;
+ case Intrinsic::nvvm_suld_3d_v2i32_trap:
+ return NVPTXISD::Suld3DV2I32Trap;
+ case Intrinsic::nvvm_suld_3d_v4i8_trap:
+ return NVPTXISD::Suld3DV4I8Trap;
+ case Intrinsic::nvvm_suld_3d_v4i16_trap:
+ return NVPTXISD::Suld3DV4I16Trap;
+ case Intrinsic::nvvm_suld_3d_v4i32_trap:
+ return NVPTXISD::Suld3DV4I32Trap;
+ }
+}
+
// llvm.ptx.memcpy.const and llvm.ptx.memmove.const need to be modeled as
// TgtMemIntrinsic
// because we need the information that is only available in the "Value" type
Info.align = 0;
return true;
+ case Intrinsic::nvvm_tex_1d_v4f32_i32:
+ case Intrinsic::nvvm_tex_1d_v4f32_f32:
+ case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_1d_array_v4f32_i32:
+ case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_v4f32_i32:
+ case Intrinsic::nvvm_tex_2d_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_array_v4f32_i32:
+ case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_3d_v4f32_i32:
+ case Intrinsic::nvvm_tex_3d_v4f32_f32:
+ case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_3d_grad_v4f32_f32: {
+ Info.opc = getOpcForTextureInstr(Intrinsic);
+ Info.memVT = MVT::f32;
+ Info.ptrVal = NULL;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
+ case Intrinsic::nvvm_tex_1d_v4i32_i32:
+ case Intrinsic::nvvm_tex_1d_v4i32_f32:
+ case Intrinsic::nvvm_tex_1d_level_v4i32_f32:
+ case Intrinsic::nvvm_tex_1d_grad_v4i32_f32:
+ case Intrinsic::nvvm_tex_1d_array_v4i32_i32:
+ case Intrinsic::nvvm_tex_1d_array_v4i32_f32:
+ case Intrinsic::nvvm_tex_1d_array_level_v4i32_f32:
+ case Intrinsic::nvvm_tex_1d_array_grad_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_v4i32_i32:
+ case Intrinsic::nvvm_tex_2d_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_level_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_grad_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_array_v4i32_i32:
+ case Intrinsic::nvvm_tex_2d_array_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_array_level_v4i32_f32:
+ case Intrinsic::nvvm_tex_2d_array_grad_v4i32_f32:
+ case Intrinsic::nvvm_tex_3d_v4i32_i32:
+ case Intrinsic::nvvm_tex_3d_v4i32_f32:
+ case Intrinsic::nvvm_tex_3d_level_v4i32_f32:
+ case Intrinsic::nvvm_tex_3d_grad_v4i32_f32: {
+ Info.opc = getOpcForTextureInstr(Intrinsic);
+ Info.memVT = MVT::i32;
+ Info.ptrVal = NULL;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
+ case Intrinsic::nvvm_suld_1d_i8_trap:
+ case Intrinsic::nvvm_suld_1d_v2i8_trap:
+ case Intrinsic::nvvm_suld_1d_v4i8_trap:
+ case Intrinsic::nvvm_suld_1d_array_i8_trap:
+ case Intrinsic::nvvm_suld_1d_array_v2i8_trap:
+ case Intrinsic::nvvm_suld_1d_array_v4i8_trap:
+ case Intrinsic::nvvm_suld_2d_i8_trap:
+ case Intrinsic::nvvm_suld_2d_v2i8_trap:
+ case Intrinsic::nvvm_suld_2d_v4i8_trap:
+ case Intrinsic::nvvm_suld_2d_array_i8_trap:
+ case Intrinsic::nvvm_suld_2d_array_v2i8_trap:
+ case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
+ case Intrinsic::nvvm_suld_3d_i8_trap:
+ case Intrinsic::nvvm_suld_3d_v2i8_trap:
+ case Intrinsic::nvvm_suld_3d_v4i8_trap: {
+ Info.opc = getOpcForSurfaceInstr(Intrinsic);
+ Info.memVT = MVT::i8;
+ Info.ptrVal = NULL;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
+ case Intrinsic::nvvm_suld_1d_i16_trap:
+ case Intrinsic::nvvm_suld_1d_v2i16_trap:
+ case Intrinsic::nvvm_suld_1d_v4i16_trap:
+ case Intrinsic::nvvm_suld_1d_array_i16_trap:
+ case Intrinsic::nvvm_suld_1d_array_v2i16_trap:
+ case Intrinsic::nvvm_suld_1d_array_v4i16_trap:
+ case Intrinsic::nvvm_suld_2d_i16_trap:
+ case Intrinsic::nvvm_suld_2d_v2i16_trap:
+ case Intrinsic::nvvm_suld_2d_v4i16_trap:
+ case Intrinsic::nvvm_suld_2d_array_i16_trap:
+ case Intrinsic::nvvm_suld_2d_array_v2i16_trap:
+ case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
+ case Intrinsic::nvvm_suld_3d_i16_trap:
+ case Intrinsic::nvvm_suld_3d_v2i16_trap:
+ case Intrinsic::nvvm_suld_3d_v4i16_trap: {
+ Info.opc = getOpcForSurfaceInstr(Intrinsic);
+ Info.memVT = MVT::i16;
+ Info.ptrVal = NULL;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
+ case Intrinsic::nvvm_suld_1d_i32_trap:
+ case Intrinsic::nvvm_suld_1d_v2i32_trap:
+ case Intrinsic::nvvm_suld_1d_v4i32_trap:
+ case Intrinsic::nvvm_suld_1d_array_i32_trap:
+ case Intrinsic::nvvm_suld_1d_array_v2i32_trap:
+ case Intrinsic::nvvm_suld_1d_array_v4i32_trap:
+ case Intrinsic::nvvm_suld_2d_i32_trap:
+ case Intrinsic::nvvm_suld_2d_v2i32_trap:
+ case Intrinsic::nvvm_suld_2d_v4i32_trap:
+ case Intrinsic::nvvm_suld_2d_array_i32_trap:
+ case Intrinsic::nvvm_suld_2d_array_v2i32_trap:
+ case Intrinsic::nvvm_suld_2d_array_v4i32_trap:
+ case Intrinsic::nvvm_suld_3d_i32_trap:
+ case Intrinsic::nvvm_suld_3d_v2i32_trap:
+ case Intrinsic::nvvm_suld_3d_v4i32_trap: {
+ Info.opc = getOpcForSurfaceInstr(Intrinsic);
+ Info.memVT = MVT::i32;
+ Info.ptrVal = NULL;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
+
}
return false;
}
// Since LoadV2 is a target node, we cannot rely on DAG type legalization.
// Therefore, we must ensure the type is legal. For i1 and i8, we set the
- // loaded type to i16 and propogate the "real" type as the memory type.
+ // loaded type to i16 and propagate the "real" type as the memory type.
bool NeedTrunc = false;
if (EltVT.getSizeInBits() < 16) {
EltVT = MVT::i16;
case 4: {
Opcode = NVPTXISD::LoadV4;
EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
- LdResVTs = DAG.getVTList(ListVTs, 5);
+ LdResVTs = DAG.getVTList(ListVTs);
break;
}
}
// Since LDU/LDG are target nodes, we cannot rely on DAG type
// legalization.
// Therefore, we must ensure the type is legal. For i1 and i8, we set the
- // loaded type to i16 and propogate the "real" type as the memory type.
+ // loaded type to i16 and propagate the "real" type as the memory type.
bool NeedTrunc = false;
if (EltVT.getSizeInBits() < 16) {
EltVT = MVT::i16;
break;
}
EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
- LdResVTs = DAG.getVTList(ListVTs, 5);
+ LdResVTs = DAG.getVTList(ListVTs);
break;
}
}
return;
}
}
+
+// Pin NVPTXSection's and NVPTXTargetObjectFile's vtables to this file.
+void NVPTXSection::anchor() {}
+
+NVPTXTargetObjectFile::~NVPTXTargetObjectFile() {
+ delete TextSection;
+ delete DataSection;
+ delete BSSSection;
+ delete ReadOnlySection;
+
+ delete StaticCtorSection;
+ delete StaticDtorSection;
+ delete LSDASection;
+ delete EHFrameSection;
+ delete DwarfAbbrevSection;
+ delete DwarfInfoSection;
+ delete DwarfLineSection;
+ delete DwarfFrameSection;
+ delete DwarfPubTypesSection;
+ delete DwarfDebugInlineSection;
+ delete DwarfStrSection;
+ delete DwarfLocSection;
+ delete DwarfARangesSection;
+ delete DwarfRangesSection;
+ delete DwarfMacroInfoSection;
+}
projects / oota-llvm.git / blobdiff