#include "PPCPerfectShuffle.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/VectorExtras.h"
-#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
+#include "llvm/ParameterAttributes.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
: TargetLowering(TM), PPCSubTarget(*TM.getSubtargetImpl()) {
setPow2DivIsCheap();
-
+
// Use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(true);
setUseUnderscoreLongJmp(true);
setIndexedStoreAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i64, Legal);
- setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
- setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
-
// Shortening conversions involving ppcf128 get expanded (2 regs -> 1 reg)
setConvertAction(MVT::ppcf128, MVT::f64, Expand);
setConvertAction(MVT::ppcf128, MVT::f32, Expand);
setOperationAction(ISD::FP_ROUND_INREG, MVT::ppcf128, Custom);
// PowerPC has no intrinsics for these particular operations
- setOperationAction(ISD::MEMMOVE, MVT::Other, Expand);
- setOperationAction(ISD::MEMSET, MVT::Other, Expand);
- setOperationAction(ISD::MEMCPY, MVT::Other, Expand);
-
+ setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
+
// PowerPC has no SREM/UREM instructions
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::FREM , MVT::f32, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
- setOperationAction(ISD::FLT_ROUNDS, MVT::i32, Custom);
+ setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
// If we're enabling GP optimizations, use hardware square root
if (!TM.getSubtarget<PPCSubtarget>().hasFSQRT()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
// Support label based line numbers.
- setOperationAction(ISD::LOCATION, MVT::Other, Expand);
+ setOperationAction(ISD::DBG_STOPPOINT, MVT::Other, Expand);
setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Custom);
+ setOperationAction(ISD::ATOMIC_LOAD_ADD , MVT::i32 , Custom);
+ setOperationAction(ISD::ATOMIC_CMP_SWAP , MVT::i32 , Custom);
+ setOperationAction(ISD::ATOMIC_SWAP , MVT::i32 , Custom);
+ if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
+ setOperationAction(ISD::ATOMIC_LOAD_ADD , MVT::i64 , Custom);
+ setOperationAction(ISD::ATOMIC_CMP_SWAP , MVT::i64 , Custom);
+ setOperationAction(ISD::ATOMIC_SWAP , MVT::i64 , Custom);
+ }
+
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
addRegisterClass(MVT::i64, PPC::G8RCRegisterClass);
// BUILD_PAIR can't be handled natively, and should be expanded to shl/or
setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
+ // 64-bit PowerPC wants to expand i128 shifts itself.
+ setOperationAction(ISD::SHL_PARTS, MVT::i64, Custom);
+ setOperationAction(ISD::SRA_PARTS, MVT::i64, Custom);
+ setOperationAction(ISD::SRL_PARTS, MVT::i64, Custom);
} else {
// 32-bit PowerPC wants to expand i64 shifts itself.
setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
if (TM.getSubtarget<PPCSubtarget>().hasAltivec()) {
// First set operation action for all vector types to expand. Then we
// will selectively turn on ones that can be effectively codegen'd.
- for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
+ for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
+ i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
+ MVT VT = (MVT::SimpleValueType)i;
+
// add/sub are legal for all supported vector VT's.
- setOperationAction(ISD::ADD , (MVT::ValueType)VT, Legal);
- setOperationAction(ISD::SUB , (MVT::ValueType)VT, Legal);
+ setOperationAction(ISD::ADD , VT, Legal);
+ setOperationAction(ISD::SUB , VT, Legal);
// We promote all shuffles to v16i8.
- setOperationAction(ISD::VECTOR_SHUFFLE, (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::VECTOR_SHUFFLE, (MVT::ValueType)VT, MVT::v16i8);
+ setOperationAction(ISD::VECTOR_SHUFFLE, VT, Promote);
+ AddPromotedToType (ISD::VECTOR_SHUFFLE, VT, MVT::v16i8);
// We promote all non-typed operations to v4i32.
- setOperationAction(ISD::AND , (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::AND , (MVT::ValueType)VT, MVT::v4i32);
- setOperationAction(ISD::OR , (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::OR , (MVT::ValueType)VT, MVT::v4i32);
- setOperationAction(ISD::XOR , (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::XOR , (MVT::ValueType)VT, MVT::v4i32);
- setOperationAction(ISD::LOAD , (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::LOAD , (MVT::ValueType)VT, MVT::v4i32);
- setOperationAction(ISD::SELECT, (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::SELECT, (MVT::ValueType)VT, MVT::v4i32);
- setOperationAction(ISD::STORE, (MVT::ValueType)VT, Promote);
- AddPromotedToType (ISD::STORE, (MVT::ValueType)VT, MVT::v4i32);
+ setOperationAction(ISD::AND , VT, Promote);
+ AddPromotedToType (ISD::AND , VT, MVT::v4i32);
+ setOperationAction(ISD::OR , VT, Promote);
+ AddPromotedToType (ISD::OR , VT, MVT::v4i32);
+ setOperationAction(ISD::XOR , VT, Promote);
+ AddPromotedToType (ISD::XOR , VT, MVT::v4i32);
+ setOperationAction(ISD::LOAD , VT, Promote);
+ AddPromotedToType (ISD::LOAD , VT, MVT::v4i32);
+ setOperationAction(ISD::SELECT, VT, Promote);
+ AddPromotedToType (ISD::SELECT, VT, MVT::v4i32);
+ setOperationAction(ISD::STORE, VT, Promote);
+ AddPromotedToType (ISD::STORE, VT, MVT::v4i32);
// No other operations are legal.
- setOperationAction(ISD::MUL , (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::SDIV, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::SREM, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::UDIV, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::UREM, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::FDIV, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::FNEG, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::EXTRACT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::INSERT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::BUILD_VECTOR, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::UMUL_LOHI, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::SMUL_LOHI, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::UDIVREM, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::SDIVREM, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::SCALAR_TO_VECTOR, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::FPOW, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::CTPOP, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::CTLZ, (MVT::ValueType)VT, Expand);
- setOperationAction(ISD::CTTZ, (MVT::ValueType)VT, Expand);
+ setOperationAction(ISD::MUL , VT, Expand);
+ setOperationAction(ISD::SDIV, VT, Expand);
+ setOperationAction(ISD::SREM, VT, Expand);
+ setOperationAction(ISD::UDIV, VT, Expand);
+ setOperationAction(ISD::UREM, VT, Expand);
+ setOperationAction(ISD::FDIV, VT, Expand);
+ setOperationAction(ISD::FNEG, VT, Expand);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Expand);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
+ setOperationAction(ISD::BUILD_VECTOR, VT, Expand);
+ setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+ setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+ setOperationAction(ISD::UDIVREM, VT, Expand);
+ setOperationAction(ISD::SDIVREM, VT, Expand);
+ setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Expand);
+ setOperationAction(ISD::FPOW, VT, Expand);
+ setOperationAction(ISD::CTPOP, VT, Expand);
+ setOperationAction(ISD::CTLZ, VT, Expand);
+ setOperationAction(ISD::CTTZ, VT, Expand);
}
// We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
- setSetCCResultType(MVT::i32);
setShiftAmountType(MVT::i32);
setSetCCResultContents(ZeroOrOneSetCCResult);
computeRegisterProperties();
}
+/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
+/// function arguments in the caller parameter area.
+unsigned PPCTargetLowering::getByValTypeAlignment(const Type *Ty) const {
+ TargetMachine &TM = getTargetMachine();
+ // Darwin passes everything on 4 byte boundary.
+ if (TM.getSubtarget<PPCSubtarget>().isDarwin())
+ return 4;
+ // FIXME Elf TBD
+ return 4;
+}
+
const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
default: return 0;
- case PPCISD::FSEL: return "PPCISD::FSEL";
- case PPCISD::FCFID: return "PPCISD::FCFID";
- case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ";
- case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ";
- case PPCISD::STFIWX: return "PPCISD::STFIWX";
- case PPCISD::VMADDFP: return "PPCISD::VMADDFP";
- case PPCISD::VNMSUBFP: return "PPCISD::VNMSUBFP";
- case PPCISD::VPERM: return "PPCISD::VPERM";
- case PPCISD::Hi: return "PPCISD::Hi";
- case PPCISD::Lo: return "PPCISD::Lo";
- case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC";
- case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg";
- case PPCISD::SRL: return "PPCISD::SRL";
- case PPCISD::SRA: return "PPCISD::SRA";
- case PPCISD::SHL: return "PPCISD::SHL";
- case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32";
- case PPCISD::STD_32: return "PPCISD::STD_32";
- case PPCISD::CALL_ELF: return "PPCISD::CALL_ELF";
- case PPCISD::CALL_Macho: return "PPCISD::CALL_Macho";
- case PPCISD::MTCTR: return "PPCISD::MTCTR";
- case PPCISD::BCTRL_Macho: return "PPCISD::BCTRL_Macho";
- case PPCISD::BCTRL_ELF: return "PPCISD::BCTRL_ELF";
- case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG";
- case PPCISD::MFCR: return "PPCISD::MFCR";
- case PPCISD::VCMP: return "PPCISD::VCMP";
- case PPCISD::VCMPo: return "PPCISD::VCMPo";
- case PPCISD::LBRX: return "PPCISD::LBRX";
- case PPCISD::STBRX: return "PPCISD::STBRX";
- case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH";
- case PPCISD::MFFS: return "PPCISD::MFFS";
- case PPCISD::MTFSB0: return "PPCISD::MTFSB0";
- case PPCISD::MTFSB1: return "PPCISD::MTFSB1";
- case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ";
- case PPCISD::MTFSF: return "PPCISD::MTFSF";
+ case PPCISD::FSEL: return "PPCISD::FSEL";
+ case PPCISD::FCFID: return "PPCISD::FCFID";
+ case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ";
+ case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ";
+ case PPCISD::STFIWX: return "PPCISD::STFIWX";
+ case PPCISD::VMADDFP: return "PPCISD::VMADDFP";
+ case PPCISD::VNMSUBFP: return "PPCISD::VNMSUBFP";
+ case PPCISD::VPERM: return "PPCISD::VPERM";
+ case PPCISD::Hi: return "PPCISD::Hi";
+ case PPCISD::Lo: return "PPCISD::Lo";
+ case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC";
+ case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg";
+ case PPCISD::SRL: return "PPCISD::SRL";
+ case PPCISD::SRA: return "PPCISD::SRA";
+ case PPCISD::SHL: return "PPCISD::SHL";
+ case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32";
+ case PPCISD::STD_32: return "PPCISD::STD_32";
+ case PPCISD::CALL_ELF: return "PPCISD::CALL_ELF";
+ case PPCISD::CALL_Macho: return "PPCISD::CALL_Macho";
+ case PPCISD::MTCTR: return "PPCISD::MTCTR";
+ case PPCISD::BCTRL_Macho: return "PPCISD::BCTRL_Macho";
+ case PPCISD::BCTRL_ELF: return "PPCISD::BCTRL_ELF";
+ case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG";
+ case PPCISD::MFCR: return "PPCISD::MFCR";
+ case PPCISD::VCMP: return "PPCISD::VCMP";
+ case PPCISD::VCMPo: return "PPCISD::VCMPo";
+ case PPCISD::LBRX: return "PPCISD::LBRX";
+ case PPCISD::STBRX: return "PPCISD::STBRX";
+ case PPCISD::ATOMIC_LOAD_ADD: return "PPCISD::ATOMIC_LOAD_ADD";
+ case PPCISD::ATOMIC_CMP_SWAP: return "PPCISD::ATOMIC_CMP_SWAP";
+ case PPCISD::ATOMIC_SWAP: return "PPCISD::ATOMIC_SWAP";
+ case PPCISD::LARX: return "PPCISD::LARX";
+ case PPCISD::STCX: return "PPCISD::STCX";
+ case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH";
+ case PPCISD::MFFS: return "PPCISD::MFFS";
+ case PPCISD::MTFSB0: return "PPCISD::MTFSB0";
+ case PPCISD::MTFSB1: return "PPCISD::MTFSB1";
+ case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ";
+ case PPCISD::MTFSF: return "PPCISD::MTFSF";
+ case PPCISD::TAILCALL: return "PPCISD::TAILCALL";
+ case PPCISD::TC_RETURN: return "PPCISD::TC_RETURN";
}
}
+
+MVT PPCTargetLowering::getSetCCResultType(const SDValue &) const {
+ return MVT::i32;
+}
+
+
//===----------------------------------------------------------------------===//
// Node matching predicates, for use by the tblgen matching code.
//===----------------------------------------------------------------------===//
/// isFloatingPointZero - Return true if this is 0.0 or -0.0.
-static bool isFloatingPointZero(SDOperand Op) {
+static bool isFloatingPointZero(SDValue Op) {
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op))
return CFP->getValueAPF().isZero();
else if (ISD::isEXTLoad(Op.Val) || ISD::isNON_EXTLoad(Op.Val)) {
/// isConstantOrUndef - Op is either an undef node or a ConstantSDNode. Return
/// true if Op is undef or if it matches the specified value.
-static bool isConstantOrUndef(SDOperand Op, unsigned Val) {
+static bool isConstantOrUndef(SDValue Op, unsigned Val) {
return Op.getOpcode() == ISD::UNDEF ||
cast<ConstantSDNode>(Op)->getValue() == Val;
}
// This is a splat operation if each element of the permute is the same, and
// if the value doesn't reference the second vector.
unsigned ElementBase = 0;
- SDOperand Elt = N->getOperand(0);
+ SDValue Elt = N->getOperand(0);
if (ConstantSDNode *EltV = dyn_cast<ConstantSDNode>(Elt))
ElementBase = EltV->getValue();
else
/// by using a vspltis[bhw] instruction of the specified element size, return
/// the constant being splatted. The ByteSize field indicates the number of
/// bytes of each element [124] -> [bhw].
-SDOperand PPC::get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG) {
- SDOperand OpVal(0, 0);
+SDValue PPC::get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG) {
+ SDValue OpVal(0, 0);
// If ByteSize of the splat is bigger than the element size of the
// build_vector, then we have a case where we are checking for a splat where
unsigned EltSize = 16/N->getNumOperands();
if (EltSize < ByteSize) {
unsigned Multiple = ByteSize/EltSize; // Number of BV entries per spltval.
- SDOperand UniquedVals[4];
+ SDValue UniquedVals[4];
assert(Multiple > 1 && Multiple <= 4 && "How can this happen?");
// See if all of the elements in the buildvector agree across.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
// If the element isn't a constant, bail fully out.
- if (!isa<ConstantSDNode>(N->getOperand(i))) return SDOperand();
+ if (!isa<ConstantSDNode>(N->getOperand(i))) return SDValue();
if (UniquedVals[i&(Multiple-1)].Val == 0)
UniquedVals[i&(Multiple-1)] = N->getOperand(i);
else if (UniquedVals[i&(Multiple-1)] != N->getOperand(i))
- return SDOperand(); // no match.
+ return SDValue(); // no match.
}
// Okay, if we reached this point, UniquedVals[0..Multiple-1] contains
return DAG.getTargetConstant(Val, MVT::i32);
}
- return SDOperand();
+ return SDValue();
}
// Check to see if this buildvec has a single non-undef value in its elements.
if (OpVal.Val == 0)
OpVal = N->getOperand(i);
else if (OpVal != N->getOperand(i))
- return SDOperand();
+ return SDValue();
}
- if (OpVal.Val == 0) return SDOperand(); // All UNDEF: use implicit def.
+ if (OpVal.Val == 0) return SDValue(); // All UNDEF: use implicit def.
unsigned ValSizeInBytes = 0;
uint64_t Value = 0;
if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
Value = CN->getValue();
- ValSizeInBytes = MVT::getSizeInBits(CN->getValueType(0))/8;
+ ValSizeInBytes = CN->getValueType(0).getSizeInBits()/8;
} else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal)) {
assert(CN->getValueType(0) == MVT::f32 && "Only one legal FP vector type!");
Value = FloatToBits(CN->getValueAPF().convertToFloat());
// If the splat value is larger than the element value, then we can never do
// this splat. The only case that we could fit the replicated bits into our
// immediate field for would be zero, and we prefer to use vxor for it.
- if (ValSizeInBytes < ByteSize) return SDOperand();
+ if (ValSizeInBytes < ByteSize) return SDValue();
// If the element value is larger than the splat value, cut it in half and
// check to see if the two halves are equal. Continue doing this until we
// If the top half equals the bottom half, we're still ok.
if (((Value >> (ValSizeInBytes*8)) & ((1 << (8*ValSizeInBytes))-1)) !=
(Value & ((1 << (8*ValSizeInBytes))-1)))
- return SDOperand();
+ return SDValue();
}
// Properly sign extend the value.
int MaskVal = ((int)Value << ShAmt) >> ShAmt;
// If this is zero, don't match, zero matches ISD::isBuildVectorAllZeros.
- if (MaskVal == 0) return SDOperand();
+ if (MaskVal == 0) return SDValue();
// Finally, if this value fits in a 5 bit sext field, return it
if (((MaskVal << (32-5)) >> (32-5)) == MaskVal)
return DAG.getTargetConstant(MaskVal, MVT::i32);
- return SDOperand();
+ return SDValue();
}
//===----------------------------------------------------------------------===//
else
return Imm == (int64_t)cast<ConstantSDNode>(N)->getValue();
}
-static bool isIntS16Immediate(SDOperand Op, short &Imm) {
+static bool isIntS16Immediate(SDValue Op, short &Imm) {
return isIntS16Immediate(Op.Val, Imm);
}
/// SelectAddressRegReg - Given the specified addressed, check to see if it
/// can be represented as an indexed [r+r] operation. Returns false if it
/// can be more efficiently represented with [r+imm].
-bool PPCTargetLowering::SelectAddressRegReg(SDOperand N, SDOperand &Base,
- SDOperand &Index,
+bool PPCTargetLowering::SelectAddressRegReg(SDValue N, SDValue &Base,
+ SDValue &Index,
SelectionDAG &DAG) {
short imm = 0;
if (N.getOpcode() == ISD::ADD) {
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are provably
// disjoint.
- uint64_t LHSKnownZero, LHSKnownOne;
- uint64_t RHSKnownZero, RHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
+ APInt LHSKnownZero, LHSKnownOne;
+ APInt RHSKnownZero, RHSKnownOne;
+ DAG.ComputeMaskedBits(N.getOperand(0),
+ APInt::getAllOnesValue(N.getOperand(0)
+ .getValueSizeInBits()),
+ LHSKnownZero, LHSKnownOne);
- if (LHSKnownZero) {
- DAG.ComputeMaskedBits(N.getOperand(1), ~0U, RHSKnownZero, RHSKnownOne);
+ if (LHSKnownZero.getBoolValue()) {
+ DAG.ComputeMaskedBits(N.getOperand(1),
+ APInt::getAllOnesValue(N.getOperand(1)
+ .getValueSizeInBits()),
+ RHSKnownZero, RHSKnownOne);
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
- if ((LHSKnownZero | RHSKnownZero) == ~0U) {
+ if (~(LHSKnownZero | RHSKnownZero) == 0) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
/// Returns true if the address N can be represented by a base register plus
/// a signed 16-bit displacement [r+imm], and if it is not better
/// represented as reg+reg.
-bool PPCTargetLowering::SelectAddressRegImm(SDOperand N, SDOperand &Disp,
- SDOperand &Base, SelectionDAG &DAG){
+bool PPCTargetLowering::SelectAddressRegImm(SDValue N, SDValue &Disp,
+ SDValue &Base, SelectionDAG &DAG){
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
- uint64_t LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
- if ((LHSKnownZero|~(unsigned)imm) == ~0U) {
+ APInt LHSKnownZero, LHSKnownOne;
+ DAG.ComputeMaskedBits(N.getOperand(0),
+ APInt::getAllOnesValue(N.getOperand(0)
+ .getValueSizeInBits()),
+ LHSKnownZero, LHSKnownOne);
+
+ if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
Base = N.getOperand(0);
Base = DAG.getTargetConstant((Addr - (signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
- Base = SDOperand(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
+ Base = SDValue(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
return true;
}
}
/// SelectAddressRegRegOnly - Given the specified addressed, force it to be
/// represented as an indexed [r+r] operation.
-bool PPCTargetLowering::SelectAddressRegRegOnly(SDOperand N, SDOperand &Base,
- SDOperand &Index,
+bool PPCTargetLowering::SelectAddressRegRegOnly(SDValue N, SDValue &Base,
+ SDValue &Index,
SelectionDAG &DAG) {
// Check to see if we can easily represent this as an [r+r] address. This
// will fail if it thinks that the address is more profitably represented as
/// SelectAddressRegImmShift - Returns true if the address N can be
/// represented by a base register plus a signed 14-bit displacement
/// [r+imm*4]. Suitable for use by STD and friends.
-bool PPCTargetLowering::SelectAddressRegImmShift(SDOperand N, SDOperand &Disp,
- SDOperand &Base,
+bool PPCTargetLowering::SelectAddressRegImmShift(SDValue N, SDValue &Disp,
+ SDValue &Base,
SelectionDAG &DAG) {
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
- uint64_t LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
- if ((LHSKnownZero|~(unsigned)imm) == ~0U) {
+ APInt LHSKnownZero, LHSKnownOne;
+ DAG.ComputeMaskedBits(N.getOperand(0),
+ APInt::getAllOnesValue(N.getOperand(0)
+ .getValueSizeInBits()),
+ LHSKnownZero, LHSKnownOne);
+ if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
Base = N.getOperand(0);
Base = DAG.getTargetConstant((Addr-(signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
- Base = SDOperand(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
+ Base = SDValue(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
return true;
}
}
/// getPreIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if the node's address
/// can be legally represented as pre-indexed load / store address.
-bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDOperand &Base,
- SDOperand &Offset,
+bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
+ SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) {
// Disabled by default for now.
if (!EnablePPCPreinc) return false;
- SDOperand Ptr;
- MVT::ValueType VT;
+ SDValue Ptr;
+ MVT VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
- VT = LD->getLoadedVT();
+ VT = LD->getMemoryVT();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
ST = ST;
Ptr = ST->getBasePtr();
- VT = ST->getStoredVT();
+ VT = ST->getMemoryVT();
} else
return false;
// PowerPC doesn't have preinc load/store instructions for vectors.
- if (MVT::isVector(VT))
+ if (VT.isVector())
return false;
// TODO: Check reg+reg first.
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
// PPC64 doesn't have lwau, but it does have lwaux. Reject preinc load of
// sext i32 to i64 when addr mode is r+i.
- if (LD->getValueType(0) == MVT::i64 && LD->getLoadedVT() == MVT::i32 &&
+ if (LD->getValueType(0) == MVT::i64 && LD->getMemoryVT() == MVT::i32 &&
LD->getExtensionType() == ISD::SEXTLOAD &&
isa<ConstantSDNode>(Offset))
return false;
// LowerOperation implementation
//===----------------------------------------------------------------------===//
-static SDOperand LowerConstantPool(SDOperand Op, SelectionDAG &DAG) {
- MVT::ValueType PtrVT = Op.getValueType();
+SDValue PPCTargetLowering::LowerConstantPool(SDValue Op,
+ SelectionDAG &DAG) {
+ MVT PtrVT = Op.getValueType();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
- SDOperand CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
- SDOperand Zero = DAG.getConstant(0, PtrVT);
+ SDValue CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
+ SDValue Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
- SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, CPI, Zero);
- SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, CPI, Zero);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, PtrVT, CPI, Zero);
+ SDValue Lo = DAG.getNode(PPCISD::Lo, PtrVT, CPI, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
return Lo;
}
-static SDOperand LowerJumpTable(SDOperand Op, SelectionDAG &DAG) {
- MVT::ValueType PtrVT = Op.getValueType();
+SDValue PPCTargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) {
+ MVT PtrVT = Op.getValueType();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
- SDOperand JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
- SDOperand Zero = DAG.getConstant(0, PtrVT);
+ SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
+ SDValue Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
- SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, JTI, Zero);
- SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, JTI, Zero);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, PtrVT, JTI, Zero);
+ SDValue Lo = DAG.getNode(PPCISD::Lo, PtrVT, JTI, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
return Lo;
}
-static SDOperand LowerGlobalTLSAddress(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
+ SelectionDAG &DAG) {
assert(0 && "TLS not implemented for PPC.");
+ return SDValue(); // Not reached
}
-static SDOperand LowerGlobalAddress(SDOperand Op, SelectionDAG &DAG) {
- MVT::ValueType PtrVT = Op.getValueType();
+SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
+ SelectionDAG &DAG) {
+ MVT PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
- SDOperand GA = DAG.getTargetGlobalAddress(GV, PtrVT, GSDN->getOffset());
- SDOperand Zero = DAG.getConstant(0, PtrVT);
+ SDValue GA = DAG.getTargetGlobalAddress(GV, PtrVT, GSDN->getOffset());
+ // If it's a debug information descriptor, don't mess with it.
+ if (DAG.isVerifiedDebugInfoDesc(Op))
+ return GA;
+ SDValue Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
- SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, GA, Zero);
- SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, GA, Zero);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, PtrVT, GA, Zero);
+ SDValue Lo = DAG.getNode(PPCISD::Lo, PtrVT, GA, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
return DAG.getLoad(PtrVT, DAG.getEntryNode(), Lo, NULL, 0);
}
-static SDOperand LowerSETCC(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) {
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
// If we're comparing for equality to zero, expose the fact that this is
// fold the new nodes.
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
if (C->isNullValue() && CC == ISD::SETEQ) {
- MVT::ValueType VT = Op.getOperand(0).getValueType();
- SDOperand Zext = Op.getOperand(0);
- if (VT < MVT::i32) {
+ MVT VT = Op.getOperand(0).getValueType();
+ SDValue Zext = Op.getOperand(0);
+ if (VT.bitsLT(MVT::i32)) {
VT = MVT::i32;
Zext = DAG.getNode(ISD::ZERO_EXTEND, VT, Op.getOperand(0));
}
- unsigned Log2b = Log2_32(MVT::getSizeInBits(VT));
- SDOperand Clz = DAG.getNode(ISD::CTLZ, VT, Zext);
- SDOperand Scc = DAG.getNode(ISD::SRL, VT, Clz,
+ unsigned Log2b = Log2_32(VT.getSizeInBits());
+ SDValue Clz = DAG.getNode(ISD::CTLZ, VT, Zext);
+ SDValue Scc = DAG.getNode(ISD::SRL, VT, Clz,
DAG.getConstant(Log2b, MVT::i32));
return DAG.getNode(ISD::TRUNCATE, MVT::i32, Scc);
}
// optimized. FIXME: revisit this when we can custom lower all setcc
// optimizations.
if (C->isAllOnesValue() || C->isNullValue())
- return SDOperand();
+ return SDValue();
}
// If we have an integer seteq/setne, turn it into a compare against zero
// condition register, reading it back out, and masking the correct bit. The
// normal approach here uses sub to do this instead of xor. Using xor exposes
// the result to other bit-twiddling opportunities.
- MVT::ValueType LHSVT = Op.getOperand(0).getValueType();
- if (MVT::isInteger(LHSVT) && (CC == ISD::SETEQ || CC == ISD::SETNE)) {
- MVT::ValueType VT = Op.getValueType();
- SDOperand Sub = DAG.getNode(ISD::XOR, LHSVT, Op.getOperand(0),
+ MVT LHSVT = Op.getOperand(0).getValueType();
+ if (LHSVT.isInteger() && (CC == ISD::SETEQ || CC == ISD::SETNE)) {
+ MVT VT = Op.getValueType();
+ SDValue Sub = DAG.getNode(ISD::XOR, LHSVT, Op.getOperand(0),
Op.getOperand(1));
return DAG.getSetCC(VT, Sub, DAG.getConstant(0, LHSVT), CC);
}
- return SDOperand();
+ return SDValue();
}
-static SDOperand LowerVAARG(SDOperand Op, SelectionDAG &DAG,
+SDValue PPCTargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG,
int VarArgsFrameIndex,
int VarArgsStackOffset,
unsigned VarArgsNumGPR,
const PPCSubtarget &Subtarget) {
assert(0 && "VAARG in ELF32 ABI not implemented yet!");
+ return SDValue(); // Not reached
}
-static SDOperand LowerVASTART(SDOperand Op, SelectionDAG &DAG,
+SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG,
int VarArgsFrameIndex,
int VarArgsStackOffset,
unsigned VarArgsNumGPR,
if (Subtarget.isMachoABI()) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- SDOperand FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
- SrcValueSDNode *SV = cast<SrcValueSDNode>(Op.getOperand(2));
- return DAG.getStore(Op.getOperand(0), FR, Op.getOperand(1), SV->getValue(),
- SV->getOffset());
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
+ const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
+ return DAG.getStore(Op.getOperand(0), FR, Op.getOperand(1), SV, 0);
}
// For ELF 32 ABI we follow the layout of the va_list struct.
// } va_list[1];
- SDOperand ArgGPR = DAG.getConstant(VarArgsNumGPR, MVT::i8);
- SDOperand ArgFPR = DAG.getConstant(VarArgsNumFPR, MVT::i8);
+ SDValue ArgGPR = DAG.getConstant(VarArgsNumGPR, MVT::i8);
+ SDValue ArgFPR = DAG.getConstant(VarArgsNumFPR, MVT::i8);
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- SDOperand StackOffset = DAG.getFrameIndex(VarArgsStackOffset, PtrVT);
- SDOperand FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
+ SDValue StackOffsetFI = DAG.getFrameIndex(VarArgsStackOffset, PtrVT);
+ SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
- SDOperand ConstFrameOffset = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8,
- PtrVT);
- SDOperand ConstStackOffset = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8 - 1,
- PtrVT);
- SDOperand ConstFPROffset = DAG.getConstant(1, PtrVT);
+ uint64_t FrameOffset = PtrVT.getSizeInBits()/8;
+ SDValue ConstFrameOffset = DAG.getConstant(FrameOffset, PtrVT);
+
+ uint64_t StackOffset = PtrVT.getSizeInBits()/8 - 1;
+ SDValue ConstStackOffset = DAG.getConstant(StackOffset, PtrVT);
+
+ uint64_t FPROffset = 1;
+ SDValue ConstFPROffset = DAG.getConstant(FPROffset, PtrVT);
- SrcValueSDNode *SV = cast<SrcValueSDNode>(Op.getOperand(2));
+ const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// Store first byte : number of int regs
- SDOperand firstStore = DAG.getStore(Op.getOperand(0), ArgGPR,
- Op.getOperand(1), SV->getValue(),
- SV->getOffset());
- SDOperand nextPtr = DAG.getNode(ISD::ADD, PtrVT, Op.getOperand(1),
+ SDValue firstStore = DAG.getStore(Op.getOperand(0), ArgGPR,
+ Op.getOperand(1), SV, 0);
+ uint64_t nextOffset = FPROffset;
+ SDValue nextPtr = DAG.getNode(ISD::ADD, PtrVT, Op.getOperand(1),
ConstFPROffset);
// Store second byte : number of float regs
- SDOperand secondStore = DAG.getStore(firstStore, ArgFPR, nextPtr,
- SV->getValue(), SV->getOffset());
+ SDValue secondStore =
+ DAG.getStore(firstStore, ArgFPR, nextPtr, SV, nextOffset);
+ nextOffset += StackOffset;
nextPtr = DAG.getNode(ISD::ADD, PtrVT, nextPtr, ConstStackOffset);
// Store second word : arguments given on stack
- SDOperand thirdStore = DAG.getStore(secondStore, StackOffset, nextPtr,
- SV->getValue(), SV->getOffset());
+ SDValue thirdStore =
+ DAG.getStore(secondStore, StackOffsetFI, nextPtr, SV, nextOffset);
+ nextOffset += FrameOffset;
nextPtr = DAG.getNode(ISD::ADD, PtrVT, nextPtr, ConstFrameOffset);
// Store third word : arguments given in registers
- return DAG.getStore(thirdStore, FR, nextPtr, SV->getValue(),
- SV->getOffset());
+ return DAG.getStore(thirdStore, FR, nextPtr, SV, nextOffset);
}
return FPR;
}
-static SDOperand LowerFORMAL_ARGUMENTS(SDOperand Op, SelectionDAG &DAG,
- int &VarArgsFrameIndex,
- int &VarArgsStackOffset,
- unsigned &VarArgsNumGPR,
- unsigned &VarArgsNumFPR,
- const PPCSubtarget &Subtarget) {
+/// CalculateStackSlotSize - Calculates the size reserved for this argument on
+/// the stack.
+static unsigned CalculateStackSlotSize(SDValue Arg, SDValue Flag,
+ bool isVarArg, unsigned PtrByteSize) {
+ MVT ArgVT = Arg.getValueType();
+ ISD::ArgFlagsTy Flags = cast<ARG_FLAGSSDNode>(Flag)->getArgFlags();
+ unsigned ArgSize =ArgVT.getSizeInBits()/8;
+ if (Flags.isByVal())
+ ArgSize = Flags.getByValSize();
+ ArgSize = ((ArgSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
+
+ return ArgSize;
+}
+
+SDValue
+PPCTargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op,
+ SelectionDAG &DAG,
+ int &VarArgsFrameIndex,
+ int &VarArgsStackOffset,
+ unsigned &VarArgsNumGPR,
+ unsigned &VarArgsNumFPR,
+ const PPCSubtarget &Subtarget) {
// TODO: add description of PPC stack frame format, or at least some docs.
//
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
- SmallVector<SDOperand, 8> ArgValues;
- SDOperand Root = Op.getOperand(0);
+ SmallVector<SDValue, 8> ArgValues;
+ SDValue Root = Op.getOperand(0);
+ bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
bool isMachoABI = Subtarget.isMachoABI();
bool isELF32_ABI = Subtarget.isELF32_ABI();
+ // Potential tail calls could cause overwriting of argument stack slots.
+ unsigned CC = MF.getFunction()->getCallingConv();
+ bool isImmutable = !(PerformTailCallOpt && (CC==CallingConv::Fast));
unsigned PtrByteSize = isPPC64 ? 8 : 4;
unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
-
+ // Area that is at least reserved in caller of this function.
+ unsigned MinReservedArea = ArgOffset;
+
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
+ // In 32-bit non-varargs functions, the stack space for vectors is after the
+ // stack space for non-vectors. We do not use this space unless we have
+ // too many vectors to fit in registers, something that only occurs in
+ // constructed examples:), but we have to walk the arglist to figure
+ // that out...for the pathological case, compute VecArgOffset as the
+ // start of the vector parameter area. Computing VecArgOffset is the
+ // entire point of the following loop.
+ // Altivec is not mentioned in the ppc32 Elf Supplement, so I'm not trying
+ // to handle Elf here.
+ unsigned VecArgOffset = ArgOffset;
+ if (!isVarArg && !isPPC64) {
+ for (unsigned ArgNo = 0, e = Op.Val->getNumValues()-1; ArgNo != e;
+ ++ArgNo) {
+ MVT ObjectVT = Op.getValue(ArgNo).getValueType();
+ unsigned ObjSize = ObjectVT.getSizeInBits()/8;
+ ISD::ArgFlagsTy Flags =
+ cast<ARG_FLAGSSDNode>(Op.getOperand(ArgNo+3))->getArgFlags();
+
+ if (Flags.isByVal()) {
+ // ObjSize is the true size, ArgSize rounded up to multiple of regs.
+ ObjSize = Flags.getByValSize();
+ unsigned ArgSize =
+ ((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
+ VecArgOffset += ArgSize;
+ continue;
+ }
+
+ switch(ObjectVT.getSimpleVT()) {
+ default: assert(0 && "Unhandled argument type!");
+ case MVT::i32:
+ case MVT::f32:
+ VecArgOffset += isPPC64 ? 8 : 4;
+ break;
+ case MVT::i64: // PPC64
+ case MVT::f64:
+ VecArgOffset += 8;
+ break;
+ case MVT::v4f32:
+ case MVT::v4i32:
+ case MVT::v8i16:
+ case MVT::v16i8:
+ // Nothing to do, we're only looking at Nonvector args here.
+ break;
+ }
+ }
+ }
+ // We've found where the vector parameter area in memory is. Skip the
+ // first 12 parameters; these don't use that memory.
+ VecArgOffset = ((VecArgOffset+15)/16)*16;
+ VecArgOffset += 12*16;
+
// Add DAG nodes to load the arguments or copy them out of registers. On
// entry to a function on PPC, the arguments start after the linkage area,
// although the first ones are often in registers.
// represented with two words (long long or double) must be copied to an
// even GPR_idx value or to an even ArgOffset value.
+ SmallVector<SDValue, 8> MemOps;
+ unsigned nAltivecParamsAtEnd = 0;
for (unsigned ArgNo = 0, e = Op.Val->getNumValues()-1; ArgNo != e; ++ArgNo) {
- SDOperand ArgVal;
+ SDValue ArgVal;
bool needsLoad = false;
- MVT::ValueType ObjectVT = Op.getValue(ArgNo).getValueType();
- unsigned ObjSize = MVT::getSizeInBits(ObjectVT)/8;
+ MVT ObjectVT = Op.getValue(ArgNo).getValueType();
+ unsigned ObjSize = ObjectVT.getSizeInBits()/8;
unsigned ArgSize = ObjSize;
- unsigned Flags = cast<ConstantSDNode>(Op.getOperand(ArgNo+3))->getValue();
- unsigned AlignFlag = 1 << ISD::ParamFlags::OrigAlignmentOffs;
+ ISD::ArgFlagsTy Flags =
+ cast<ARG_FLAGSSDNode>(Op.getOperand(ArgNo+3))->getArgFlags();
// See if next argument requires stack alignment in ELF
- bool Expand = (ObjectVT == MVT::f64) || ((ArgNo + 1 < e) &&
- (cast<ConstantSDNode>(Op.getOperand(ArgNo+4))->getValue() & AlignFlag) &&
- (!(Flags & AlignFlag)));
+ bool Align = Flags.isSplit();
unsigned CurArgOffset = ArgOffset;
- switch (ObjectVT) {
+
+ // Varargs or 64 bit Altivec parameters are padded to a 16 byte boundary.
+ if (ObjectVT==MVT::v4f32 || ObjectVT==MVT::v4i32 ||
+ ObjectVT==MVT::v8i16 || ObjectVT==MVT::v16i8) {
+ if (isVarArg || isPPC64) {
+ MinReservedArea = ((MinReservedArea+15)/16)*16;
+ MinReservedArea += CalculateStackSlotSize(Op.getValue(ArgNo),
+ Op.getOperand(ArgNo+3),
+ isVarArg,
+ PtrByteSize);
+ } else nAltivecParamsAtEnd++;
+ } else
+ // Calculate min reserved area.
+ MinReservedArea += CalculateStackSlotSize(Op.getValue(ArgNo),
+ Op.getOperand(ArgNo+3),
+ isVarArg,
+ PtrByteSize);
+
+ // FIXME alignment for ELF may not be right
+ // FIXME the codegen can be much improved in some cases.
+ // We do not have to keep everything in memory.
+ if (Flags.isByVal()) {
+ // ObjSize is the true size, ArgSize rounded up to multiple of registers.
+ ObjSize = Flags.getByValSize();
+ ArgSize = ((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
+ // Double word align in ELF
+ if (Align && isELF32_ABI) GPR_idx += (GPR_idx % 2);
+ // Objects of size 1 and 2 are right justified, everything else is
+ // left justified. This means the memory address is adjusted forwards.
+ if (ObjSize==1 || ObjSize==2) {
+ CurArgOffset = CurArgOffset + (4 - ObjSize);
+ }
+ // The value of the object is its address.
+ int FI = MFI->CreateFixedObject(ObjSize, CurArgOffset);
+ SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
+ ArgValues.push_back(FIN);
+ if (ObjSize==1 || ObjSize==2) {
+ if (GPR_idx != Num_GPR_Regs) {
+ unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
+ RegInfo.addLiveIn(GPR[GPR_idx], VReg);
+ SDValue Val = DAG.getCopyFromReg(Root, VReg, PtrVT);
+ SDValue Store = DAG.getTruncStore(Val.getValue(1), Val, FIN,
+ NULL, 0, ObjSize==1 ? MVT::i8 : MVT::i16 );
+ MemOps.push_back(Store);
+ ++GPR_idx;
+ if (isMachoABI) ArgOffset += PtrByteSize;
+ } else {
+ ArgOffset += PtrByteSize;
+ }
+ continue;
+ }
+ for (unsigned j = 0; j < ArgSize; j += PtrByteSize) {
+ // Store whatever pieces of the object are in registers
+ // to memory. ArgVal will be address of the beginning of
+ // the object.
+ if (GPR_idx != Num_GPR_Regs) {
+ unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
+ RegInfo.addLiveIn(GPR[GPR_idx], VReg);
+ int FI = MFI->CreateFixedObject(PtrByteSize, ArgOffset);
+ SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
+ SDValue Val = DAG.getCopyFromReg(Root, VReg, PtrVT);
+ SDValue Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
+ MemOps.push_back(Store);
+ ++GPR_idx;
+ if (isMachoABI) ArgOffset += PtrByteSize;
+ } else {
+ ArgOffset += ArgSize - (ArgOffset-CurArgOffset);
+ break;
+ }
+ }
+ continue;
+ }
+
+ switch (ObjectVT.getSimpleVT()) {
default: assert(0 && "Unhandled argument type!");
case MVT::i32:
- // Double word align in ELF
- if (Expand && isELF32_ABI) GPR_idx += (GPR_idx % 2);
- if (GPR_idx != Num_GPR_Regs) {
- unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
- RegInfo.addLiveIn(GPR[GPR_idx], VReg);
- ArgVal = DAG.getCopyFromReg(Root, VReg, MVT::i32);
- ++GPR_idx;
- } else {
- needsLoad = true;
- ArgSize = PtrByteSize;
+ if (!isPPC64) {
+ // Double word align in ELF
+ if (Align && isELF32_ABI) GPR_idx += (GPR_idx % 2);
+
+ if (GPR_idx != Num_GPR_Regs) {
+ unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
+ RegInfo.addLiveIn(GPR[GPR_idx], VReg);
+ ArgVal = DAG.getCopyFromReg(Root, VReg, MVT::i32);
+ ++GPR_idx;
+ } else {
+ needsLoad = true;
+ ArgSize = PtrByteSize;
+ }
+ // Stack align in ELF
+ if (needsLoad && Align && isELF32_ABI)
+ ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
+ // All int arguments reserve stack space in Macho ABI.
+ if (isMachoABI || needsLoad) ArgOffset += PtrByteSize;
+ break;
}
- // Stack align in ELF
- if (needsLoad && Expand && isELF32_ABI)
- ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
- // All int arguments reserve stack space in Macho ABI.
- if (isMachoABI || needsLoad) ArgOffset += PtrByteSize;
- break;
-
+ // FALLTHROUGH
case MVT::i64: // PPC64
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, MVT::i64);
+
+ if (ObjectVT == MVT::i32) {
+ // PPC64 passes i8, i16, and i32 values in i64 registers. Promote
+ // value to MVT::i64 and then truncate to the correct register size.
+ if (Flags.isSExt())
+ ArgVal = DAG.getNode(ISD::AssertSext, MVT::i64, ArgVal,
+ DAG.getValueType(ObjectVT));
+ else if (Flags.isZExt())
+ ArgVal = DAG.getNode(ISD::AssertZext, MVT::i64, ArgVal,
+ DAG.getValueType(ObjectVT));
+
+ ArgVal = DAG.getNode(ISD::TRUNCATE, MVT::i32, ArgVal);
+ }
+
++GPR_idx;
} else {
needsLoad = true;
+ ArgSize = PtrByteSize;
}
// All int arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += 8;
}
// Stack align in ELF
- if (needsLoad && Expand && isELF32_ABI)
+ if (needsLoad && Align && isELF32_ABI)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
// All FP arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += isPPC64 ? 8 : ObjSize;
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
- // Note that vector arguments in registers don't reserve stack space.
+ // Note that vector arguments in registers don't reserve stack space,
+ // except in varargs functions.
if (VR_idx != Num_VR_Regs) {
unsigned VReg = RegInfo.createVirtualRegister(&PPC::VRRCRegClass);
RegInfo.addLiveIn(VR[VR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, ObjectVT);
+ if (isVarArg) {
+ while ((ArgOffset % 16) != 0) {
+ ArgOffset += PtrByteSize;
+ if (GPR_idx != Num_GPR_Regs)
+ GPR_idx++;
+ }
+ ArgOffset += 16;
+ GPR_idx = std::min(GPR_idx+4, Num_GPR_Regs);
+ }
++VR_idx;
} else {
- // This should be simple, but requires getting 16-byte aligned stack
- // values.
- assert(0 && "Loading VR argument not implemented yet!");
+ if (!isVarArg && !isPPC64) {
+ // Vectors go after all the nonvectors.
+ CurArgOffset = VecArgOffset;
+ VecArgOffset += 16;
+ } else {
+ // Vectors are aligned.
+ ArgOffset = ((ArgOffset+15)/16)*16;
+ CurArgOffset = ArgOffset;
+ ArgOffset += 16;
+ }
needsLoad = true;
}
break;
}
// We need to load the argument to a virtual register if we determined above
- // that we ran out of physical registers of the appropriate type
+ // that we ran out of physical registers of the appropriate type.
if (needsLoad) {
- // If the argument is actually used, emit a load from the right stack
- // slot.
- if (!Op.Val->hasNUsesOfValue(0, ArgNo)) {
- int FI = MFI->CreateFixedObject(ObjSize,
- CurArgOffset + (ArgSize - ObjSize));
- SDOperand FIN = DAG.getFrameIndex(FI, PtrVT);
- ArgVal = DAG.getLoad(ObjectVT, Root, FIN, NULL, 0);
- } else {
- // Don't emit a dead load.
- ArgVal = DAG.getNode(ISD::UNDEF, ObjectVT);
- }
+ int FI = MFI->CreateFixedObject(ObjSize,
+ CurArgOffset + (ArgSize - ObjSize),
+ isImmutable);
+ SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
+ ArgVal = DAG.getLoad(ObjectVT, Root, FIN, NULL, 0);
}
ArgValues.push_back(ArgVal);
}
-
+
+ // Set the size that is at least reserved in caller of this function. Tail
+ // call optimized function's reserved stack space needs to be aligned so that
+ // taking the difference between two stack areas will result in an aligned
+ // stack.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ // Add the Altivec parameters at the end, if needed.
+ if (nAltivecParamsAtEnd) {
+ MinReservedArea = ((MinReservedArea+15)/16)*16;
+ MinReservedArea += 16*nAltivecParamsAtEnd;
+ }
+ MinReservedArea =
+ std::max(MinReservedArea,
+ PPCFrameInfo::getMinCallFrameSize(isPPC64, isMachoABI));
+ unsigned TargetAlign = DAG.getMachineFunction().getTarget().getFrameInfo()->
+ getStackAlignment();
+ unsigned AlignMask = TargetAlign-1;
+ MinReservedArea = (MinReservedArea + AlignMask) & ~AlignMask;
+ FI->setMinReservedArea(MinReservedArea);
+
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
- bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
if (isVarArg) {
int depth;
// Make room for Num_GPR_Regs, Num_FPR_Regs and for a possible frame
// pointer.
- depth = -(Num_GPR_Regs * MVT::getSizeInBits(PtrVT)/8 +
- Num_FPR_Regs * MVT::getSizeInBits(MVT::f64)/8 +
- MVT::getSizeInBits(PtrVT)/8);
+ depth = -(Num_GPR_Regs * PtrVT.getSizeInBits()/8 +
+ Num_FPR_Regs * MVT(MVT::f64).getSizeInBits()/8 +
+ PtrVT.getSizeInBits()/8);
- VarArgsStackOffset = MFI->CreateFixedObject(MVT::getSizeInBits(PtrVT)/8,
+ VarArgsStackOffset = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
ArgOffset);
}
else
depth = ArgOffset;
- VarArgsFrameIndex = MFI->CreateFixedObject(MVT::getSizeInBits(PtrVT)/8,
+ VarArgsFrameIndex = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
depth);
- SDOperand FIN = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
-
- SmallVector<SDOperand, 8> MemOps;
+ SDValue FIN = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
// In ELF 32 ABI, the fixed integer arguments of a variadic function are
// stored to the VarArgsFrameIndex on the stack.
if (isELF32_ABI) {
for (GPR_idx = 0; GPR_idx != VarArgsNumGPR; ++GPR_idx) {
- SDOperand Val = DAG.getRegister(GPR[GPR_idx], PtrVT);
- SDOperand Store = DAG.getStore(Root, Val, FIN, NULL, 0);
+ SDValue Val = DAG.getRegister(GPR[GPR_idx], PtrVT);
+ SDValue Store = DAG.getStore(Root, Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
- SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8, PtrVT);
+ SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
}
VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
- SDOperand Val = DAG.getCopyFromReg(Root, VReg, PtrVT);
- SDOperand Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
+ SDValue Val = DAG.getCopyFromReg(Root, VReg, PtrVT);
+ SDValue Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
- SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8, PtrVT);
+ SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
// on the stack.
if (isELF32_ABI) {
for (FPR_idx = 0; FPR_idx != VarArgsNumFPR; ++FPR_idx) {
- SDOperand Val = DAG.getRegister(FPR[FPR_idx], MVT::f64);
- SDOperand Store = DAG.getStore(Root, Val, FIN, NULL, 0);
+ SDValue Val = DAG.getRegister(FPR[FPR_idx], MVT::f64);
+ SDValue Store = DAG.getStore(Root, Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
- SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(MVT::f64)/8,
+ SDValue PtrOff = DAG.getConstant(MVT(MVT::f64).getSizeInBits()/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
VReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
RegInfo.addLiveIn(FPR[FPR_idx], VReg);
- SDOperand Val = DAG.getCopyFromReg(Root, VReg, MVT::f64);
- SDOperand Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
+ SDValue Val = DAG.getCopyFromReg(Root, VReg, MVT::f64);
+ SDValue Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
- SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(MVT::f64)/8,
+ SDValue PtrOff = DAG.getConstant(MVT(MVT::f64).getSizeInBits()/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
}
-
- if (!MemOps.empty())
- Root = DAG.getNode(ISD::TokenFactor, MVT::Other,&MemOps[0],MemOps.size());
}
+ if (!MemOps.empty())
+ Root = DAG.getNode(ISD::TokenFactor, MVT::Other,&MemOps[0],MemOps.size());
+
ArgValues.push_back(Root);
// Return the new list of results.
- std::vector<MVT::ValueType> RetVT(Op.Val->value_begin(),
- Op.Val->value_end());
- return DAG.getNode(ISD::MERGE_VALUES, RetVT, &ArgValues[0], ArgValues.size());
+ return DAG.getMergeValues(Op.Val->getVTList(), &ArgValues[0],
+ ArgValues.size());
+}
+
+/// CalculateParameterAndLinkageAreaSize - Get the size of the paramter plus
+/// linkage area.
+static unsigned
+CalculateParameterAndLinkageAreaSize(SelectionDAG &DAG,
+ bool isPPC64,
+ bool isMachoABI,
+ bool isVarArg,
+ unsigned CC,
+ SDValue Call,
+ unsigned &nAltivecParamsAtEnd) {
+ // Count how many bytes are to be pushed on the stack, including the linkage
+ // area, and parameter passing area. We start with 24/48 bytes, which is
+ // prereserved space for [SP][CR][LR][3 x unused].
+ unsigned NumBytes = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
+ unsigned NumOps = (Call.getNumOperands() - 5) / 2;
+ unsigned PtrByteSize = isPPC64 ? 8 : 4;
+
+ // Add up all the space actually used.
+ // In 32-bit non-varargs calls, Altivec parameters all go at the end; usually
+ // they all go in registers, but we must reserve stack space for them for
+ // possible use by the caller. In varargs or 64-bit calls, parameters are
+ // assigned stack space in order, with padding so Altivec parameters are
+ // 16-byte aligned.
+ nAltivecParamsAtEnd = 0;
+ for (unsigned i = 0; i != NumOps; ++i) {
+ SDValue Arg = Call.getOperand(5+2*i);
+ SDValue Flag = Call.getOperand(5+2*i+1);
+ MVT ArgVT = Arg.getValueType();
+ // Varargs Altivec parameters are padded to a 16 byte boundary.
+ if (ArgVT==MVT::v4f32 || ArgVT==MVT::v4i32 ||
+ ArgVT==MVT::v8i16 || ArgVT==MVT::v16i8) {
+ if (!isVarArg && !isPPC64) {
+ // Non-varargs Altivec parameters go after all the non-Altivec
+ // parameters; handle those later so we know how much padding we need.
+ nAltivecParamsAtEnd++;
+ continue;
+ }
+ // Varargs and 64-bit Altivec parameters are padded to 16 byte boundary.
+ NumBytes = ((NumBytes+15)/16)*16;
+ }
+ NumBytes += CalculateStackSlotSize(Arg, Flag, isVarArg, PtrByteSize);
+ }
+
+ // Allow for Altivec parameters at the end, if needed.
+ if (nAltivecParamsAtEnd) {
+ NumBytes = ((NumBytes+15)/16)*16;
+ NumBytes += 16*nAltivecParamsAtEnd;
+ }
+
+ // The prolog code of the callee may store up to 8 GPR argument registers to
+ // the stack, allowing va_start to index over them in memory if its varargs.
+ // Because we cannot tell if this is needed on the caller side, we have to
+ // conservatively assume that it is needed. As such, make sure we have at
+ // least enough stack space for the caller to store the 8 GPRs.
+ NumBytes = std::max(NumBytes,
+ PPCFrameInfo::getMinCallFrameSize(isPPC64, isMachoABI));
+
+ // Tail call needs the stack to be aligned.
+ if (CC==CallingConv::Fast && PerformTailCallOpt) {
+ unsigned TargetAlign = DAG.getMachineFunction().getTarget().getFrameInfo()->
+ getStackAlignment();
+ unsigned AlignMask = TargetAlign-1;
+ NumBytes = (NumBytes + AlignMask) & ~AlignMask;
+ }
+
+ return NumBytes;
+}
+
+/// CalculateTailCallSPDiff - Get the amount the stack pointer has to be
+/// adjusted to accomodate the arguments for the tailcall.
+static int CalculateTailCallSPDiff(SelectionDAG& DAG, bool IsTailCall,
+ unsigned ParamSize) {
+
+ if (!IsTailCall) return 0;
+
+ PPCFunctionInfo *FI = DAG.getMachineFunction().getInfo<PPCFunctionInfo>();
+ unsigned CallerMinReservedArea = FI->getMinReservedArea();
+ int SPDiff = (int)CallerMinReservedArea - (int)ParamSize;
+ // Remember only if the new adjustement is bigger.
+ if (SPDiff < FI->getTailCallSPDelta())
+ FI->setTailCallSPDelta(SPDiff);
+
+ return SPDiff;
+}
+
+/// IsEligibleForTailCallElimination - Check to see whether the next instruction
+/// following the call is a return. A function is eligible if caller/callee
+/// calling conventions match, currently only fastcc supports tail calls, and
+/// the function CALL is immediatly followed by a RET.
+bool
+PPCTargetLowering::IsEligibleForTailCallOptimization(SDValue Call,
+ SDValue Ret,
+ SelectionDAG& DAG) const {
+ // Variable argument functions are not supported.
+ if (!PerformTailCallOpt ||
+ cast<ConstantSDNode>(Call.getOperand(2))->getValue() != 0) return false;
+
+ if (CheckTailCallReturnConstraints(Call, Ret)) {
+ MachineFunction &MF = DAG.getMachineFunction();
+ unsigned CallerCC = MF.getFunction()->getCallingConv();
+ unsigned CalleeCC = cast<ConstantSDNode>(Call.getOperand(1))->getValue();
+ if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC) {
+ // Functions containing by val parameters are not supported.
+ for (unsigned i = 0; i != ((Call.getNumOperands()-5)/2); i++) {
+ ISD::ArgFlagsTy Flags = cast<ARG_FLAGSSDNode>(Call.getOperand(5+2*i+1))
+ ->getArgFlags();
+ if (Flags.isByVal()) return false;
+ }
+
+ SDValue Callee = Call.getOperand(4);
+ // Non PIC/GOT tail calls are supported.
+ if (getTargetMachine().getRelocationModel() != Reloc::PIC_)
+ return true;
+
+ // At the moment we can only do local tail calls (in same module, hidden
+ // or protected) if we are generating PIC.
+ if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
+ return G->getGlobal()->hasHiddenVisibility()
+ || G->getGlobal()->hasProtectedVisibility();
+ }
+ }
+
+ return false;
}
/// isCallCompatibleAddress - Return the immediate to use if the specified
/// 32-bit value is representable in the immediate field of a BxA instruction.
-static SDNode *isBLACompatibleAddress(SDOperand Op, SelectionDAG &DAG) {
+static SDNode *isBLACompatibleAddress(SDValue Op, SelectionDAG &DAG) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
if (!C) return 0;
DAG.getTargetLoweringInfo().getPointerTy()).Val;
}
+namespace {
+
+struct TailCallArgumentInfo {
+ SDValue Arg;
+ SDValue FrameIdxOp;
+ int FrameIdx;
+
+ TailCallArgumentInfo() : FrameIdx(0) {}
+};
+
+}
+
+/// StoreTailCallArgumentsToStackSlot - Stores arguments to their stack slot.
+static void
+StoreTailCallArgumentsToStackSlot(SelectionDAG &DAG,
+ SDValue Chain,
+ const SmallVector<TailCallArgumentInfo, 8> &TailCallArgs,
+ SmallVector<SDValue, 8> &MemOpChains) {
+ for (unsigned i = 0, e = TailCallArgs.size(); i != e; ++i) {
+ SDValue Arg = TailCallArgs[i].Arg;
+ SDValue FIN = TailCallArgs[i].FrameIdxOp;
+ int FI = TailCallArgs[i].FrameIdx;
+ // Store relative to framepointer.
+ MemOpChains.push_back(DAG.getStore(Chain, Arg, FIN,
+ PseudoSourceValue::getFixedStack(FI),
+ 0));
+ }
+}
+
+/// EmitTailCallStoreFPAndRetAddr - Move the frame pointer and return address to
+/// the appropriate stack slot for the tail call optimized function call.
+static SDValue EmitTailCallStoreFPAndRetAddr(SelectionDAG &DAG,
+ MachineFunction &MF,
+ SDValue Chain,
+ SDValue OldRetAddr,
+ SDValue OldFP,
+ int SPDiff,
+ bool isPPC64,
+ bool isMachoABI) {
+ if (SPDiff) {
+ // Calculate the new stack slot for the return address.
+ int SlotSize = isPPC64 ? 8 : 4;
+ int NewRetAddrLoc = SPDiff + PPCFrameInfo::getReturnSaveOffset(isPPC64,
+ isMachoABI);
+ int NewRetAddr = MF.getFrameInfo()->CreateFixedObject(SlotSize,
+ NewRetAddrLoc);
+ int NewFPLoc = SPDiff + PPCFrameInfo::getFramePointerSaveOffset(isPPC64,
+ isMachoABI);
+ int NewFPIdx = MF.getFrameInfo()->CreateFixedObject(SlotSize, NewFPLoc);
+
+ MVT VT = isPPC64 ? MVT::i64 : MVT::i32;
+ SDValue NewRetAddrFrIdx = DAG.getFrameIndex(NewRetAddr, VT);
+ Chain = DAG.getStore(Chain, OldRetAddr, NewRetAddrFrIdx,
+ PseudoSourceValue::getFixedStack(NewRetAddr), 0);
+ SDValue NewFramePtrIdx = DAG.getFrameIndex(NewFPIdx, VT);
+ Chain = DAG.getStore(Chain, OldFP, NewFramePtrIdx,
+ PseudoSourceValue::getFixedStack(NewFPIdx), 0);
+ }
+ return Chain;
+}
+
+/// CalculateTailCallArgDest - Remember Argument for later processing. Calculate
+/// the position of the argument.
+static void
+CalculateTailCallArgDest(SelectionDAG &DAG, MachineFunction &MF, bool isPPC64,
+ SDValue Arg, int SPDiff, unsigned ArgOffset,
+ SmallVector<TailCallArgumentInfo, 8>& TailCallArguments) {
+ int Offset = ArgOffset + SPDiff;
+ uint32_t OpSize = (Arg.getValueType().getSizeInBits()+7)/8;
+ int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset);
+ MVT VT = isPPC64 ? MVT::i64 : MVT::i32;
+ SDValue FIN = DAG.getFrameIndex(FI, VT);
+ TailCallArgumentInfo Info;
+ Info.Arg = Arg;
+ Info.FrameIdxOp = FIN;
+ Info.FrameIdx = FI;
+ TailCallArguments.push_back(Info);
+}
+
+/// EmitTCFPAndRetAddrLoad - Emit load from frame pointer and return address
+/// stack slot. Returns the chain as result and the loaded frame pointers in
+/// LROpOut/FPOpout. Used when tail calling.
+SDValue PPCTargetLowering::EmitTailCallLoadFPAndRetAddr(SelectionDAG & DAG,
+ int SPDiff,
+ SDValue Chain,
+ SDValue &LROpOut,
+ SDValue &FPOpOut) {
+ if (SPDiff) {
+ // Load the LR and FP stack slot for later adjusting.
+ MVT VT = PPCSubTarget.isPPC64() ? MVT::i64 : MVT::i32;
+ LROpOut = getReturnAddrFrameIndex(DAG);
+ LROpOut = DAG.getLoad(VT, Chain, LROpOut, NULL, 0);
+ Chain = SDValue(LROpOut.Val, 1);
+ FPOpOut = getFramePointerFrameIndex(DAG);
+ FPOpOut = DAG.getLoad(VT, Chain, FPOpOut, NULL, 0);
+ Chain = SDValue(FPOpOut.Val, 1);
+ }
+ return Chain;
+}
-static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG,
- const PPCSubtarget &Subtarget) {
- SDOperand Chain = Op.getOperand(0);
+/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
+/// by "Src" to address "Dst" of size "Size". Alignment information is
+/// specified by the specific parameter attribute. The copy will be passed as
+/// a byval function parameter.
+/// Sometimes what we are copying is the end of a larger object, the part that
+/// does not fit in registers.
+static SDValue
+CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
+ ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
+ unsigned Size) {
+ SDValue SizeNode = DAG.getConstant(Size, MVT::i32);
+ return DAG.getMemcpy(Chain, Dst, Src, SizeNode, Flags.getByValAlign(), false,
+ NULL, 0, NULL, 0);
+}
+
+/// LowerMemOpCallTo - Store the argument to the stack or remember it in case of
+/// tail calls.
+static void
+LowerMemOpCallTo(SelectionDAG &DAG, MachineFunction &MF, SDValue Chain,
+ SDValue Arg, SDValue PtrOff, int SPDiff,
+ unsigned ArgOffset, bool isPPC64, bool isTailCall,
+ bool isVector, SmallVector<SDValue, 8> &MemOpChains,
+ SmallVector<TailCallArgumentInfo, 8>& TailCallArguments) {
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ if (!isTailCall) {
+ if (isVector) {
+ SDValue StackPtr;
+ if (isPPC64)
+ StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
+ else
+ StackPtr = DAG.getRegister(PPC::R1, MVT::i32);
+ PtrOff = DAG.getNode(ISD::ADD, PtrVT, StackPtr,
+ DAG.getConstant(ArgOffset, PtrVT));
+ }
+ MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
+ // Calculate and remember argument location.
+ } else CalculateTailCallArgDest(DAG, MF, isPPC64, Arg, SPDiff, ArgOffset,
+ TailCallArguments);
+}
+
+SDValue PPCTargetLowering::LowerCALL(SDValue Op, SelectionDAG &DAG,
+ const PPCSubtarget &Subtarget,
+ TargetMachine &TM) {
+ SDValue Chain = Op.getOperand(0);
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
- SDOperand Callee = Op.getOperand(4);
+ unsigned CC = cast<ConstantSDNode>(Op.getOperand(1))->getValue();
+ bool isTailCall = cast<ConstantSDNode>(Op.getOperand(3))->getValue() != 0 &&
+ CC == CallingConv::Fast && PerformTailCallOpt;
+ SDValue Callee = Op.getOperand(4);
unsigned NumOps = (Op.getNumOperands() - 5) / 2;
bool isMachoABI = Subtarget.isMachoABI();
bool isELF32_ABI = Subtarget.isELF32_ABI();
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
unsigned PtrByteSize = isPPC64 ? 8 : 4;
+ MachineFunction &MF = DAG.getMachineFunction();
+
// args_to_use will accumulate outgoing args for the PPCISD::CALL case in
// SelectExpr to use to put the arguments in the appropriate registers.
- std::vector<SDOperand> args_to_use;
+ std::vector<SDValue> args_to_use;
+ // Mark this function as potentially containing a function that contains a
+ // tail call. As a consequence the frame pointer will be used for dynamicalloc
+ // and restoring the callers stack pointer in this functions epilog. This is
+ // done because by tail calling the called function might overwrite the value
+ // in this function's (MF) stack pointer stack slot 0(SP).
+ if (PerformTailCallOpt && CC==CallingConv::Fast)
+ MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
+
+ unsigned nAltivecParamsAtEnd = 0;
+
// Count how many bytes are to be pushed on the stack, including the linkage
// area, and parameter passing area. We start with 24/48 bytes, which is
// prereserved space for [SP][CR][LR][3 x unused].
- unsigned NumBytes = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
-
- // Add up all the space actually used.
- for (unsigned i = 0; i != NumOps; ++i) {
- unsigned ArgSize =MVT::getSizeInBits(Op.getOperand(5+2*i).getValueType())/8;
- ArgSize = std::max(ArgSize, PtrByteSize);
- NumBytes += ArgSize;
- }
+ unsigned NumBytes =
+ CalculateParameterAndLinkageAreaSize(DAG, isPPC64, isMachoABI, isVarArg, CC,
+ Op, nAltivecParamsAtEnd);
- // The prolog code of the callee may store up to 8 GPR argument registers to
- // the stack, allowing va_start to index over them in memory if its varargs.
- // Because we cannot tell if this is needed on the caller side, we have to
- // conservatively assume that it is needed. As such, make sure we have at
- // least enough stack space for the caller to store the 8 GPRs.
- NumBytes = std::max(NumBytes,
- PPCFrameInfo::getMinCallFrameSize(isPPC64, isMachoABI));
+ // Calculate by how many bytes the stack has to be adjusted in case of tail
+ // call optimization.
+ int SPDiff = CalculateTailCallSPDiff(DAG, isTailCall, NumBytes);
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
Chain = DAG.getCALLSEQ_START(Chain,
DAG.getConstant(NumBytes, PtrVT));
+ SDValue CallSeqStart = Chain;
+ // Load the return address and frame pointer so it can be move somewhere else
+ // later.
+ SDValue LROp, FPOp;
+ Chain = EmitTailCallLoadFPAndRetAddr(DAG, SPDiff, Chain, LROp, FPOp);
+
// Set up a copy of the stack pointer for use loading and storing any
// arguments that may not fit in the registers available for argument
// passing.
- SDOperand StackPtr;
+ SDValue StackPtr;
if (isPPC64)
StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
else
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
- std::vector<std::pair<unsigned, SDOperand> > RegsToPass;
- SmallVector<SDOperand, 8> MemOpChains;
+ std::vector<std::pair<unsigned, SDValue> > RegsToPass;
+ SmallVector<TailCallArgumentInfo, 8> TailCallArguments;
+
+ SmallVector<SDValue, 8> MemOpChains;
for (unsigned i = 0; i != NumOps; ++i) {
bool inMem = false;
- SDOperand Arg = Op.getOperand(5+2*i);
- unsigned Flags = cast<ConstantSDNode>(Op.getOperand(5+2*i+1))->getValue();
- unsigned AlignFlag = 1 << ISD::ParamFlags::OrigAlignmentOffs;
+ SDValue Arg = Op.getOperand(5+2*i);
+ ISD::ArgFlagsTy Flags =
+ cast<ARG_FLAGSSDNode>(Op.getOperand(5+2*i+1))->getArgFlags();
// See if next argument requires stack alignment in ELF
- unsigned next = 5+2*(i+1)+1;
- bool Expand = (Arg.getValueType() == MVT::f64) || ((i + 1 < NumOps) &&
- (cast<ConstantSDNode>(Op.getOperand(next))->getValue() & AlignFlag) &&
- (!(Flags & AlignFlag)));
+ bool Align = Flags.isSplit();
// PtrOff will be used to store the current argument to the stack if a
// register cannot be found for it.
- SDOperand PtrOff;
+ SDValue PtrOff;
// Stack align in ELF 32
- if (isELF32_ABI && Expand)
+ if (isELF32_ABI && Align)
PtrOff = DAG.getConstant(ArgOffset + ((ArgOffset/4) % 2) * PtrByteSize,
StackPtr.getValueType());
else
// On PPC64, promote integers to 64-bit values.
if (isPPC64 && Arg.getValueType() == MVT::i32) {
- unsigned ExtOp = (Flags & 1) ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
-
+ // FIXME: Should this use ANY_EXTEND if neither sext nor zext?
+ unsigned ExtOp = Flags.isSExt() ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
Arg = DAG.getNode(ExtOp, MVT::i64, Arg);
}
-
- switch (Arg.getValueType()) {
+
+ // FIXME Elf untested, what are alignment rules?
+ // FIXME memcpy is used way more than necessary. Correctness first.
+ if (Flags.isByVal()) {
+ unsigned Size = Flags.getByValSize();
+ if (isELF32_ABI && Align) GPR_idx += (GPR_idx % 2);
+ if (Size==1 || Size==2) {
+ // Very small objects are passed right-justified.
+ // Everything else is passed left-justified.
+ MVT VT = (Size==1) ? MVT::i8 : MVT::i16;
+ if (GPR_idx != NumGPRs) {
+ SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, PtrVT, Chain, Arg,
+ NULL, 0, VT);
+ MemOpChains.push_back(Load.getValue(1));
+ RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
+ if (isMachoABI)
+ ArgOffset += PtrByteSize;
+ } else {
+ SDValue Const = DAG.getConstant(4 - Size, PtrOff.getValueType());
+ SDValue AddPtr = DAG.getNode(ISD::ADD, PtrVT, PtrOff, Const);
+ SDValue MemcpyCall = CreateCopyOfByValArgument(Arg, AddPtr,
+ CallSeqStart.Val->getOperand(0),
+ Flags, DAG, Size);
+ // This must go outside the CALLSEQ_START..END.
+ SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
+ CallSeqStart.Val->getOperand(1));
+ DAG.ReplaceAllUsesWith(CallSeqStart.Val, NewCallSeqStart.Val);
+ Chain = CallSeqStart = NewCallSeqStart;
+ ArgOffset += PtrByteSize;
+ }
+ continue;
+ }
+ // Copy entire object into memory. There are cases where gcc-generated
+ // code assumes it is there, even if it could be put entirely into
+ // registers. (This is not what the doc says.)
+ SDValue MemcpyCall = CreateCopyOfByValArgument(Arg, PtrOff,
+ CallSeqStart.Val->getOperand(0),
+ Flags, DAG, Size);
+ // This must go outside the CALLSEQ_START..END.
+ SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
+ CallSeqStart.Val->getOperand(1));
+ DAG.ReplaceAllUsesWith(CallSeqStart.Val, NewCallSeqStart.Val);
+ Chain = CallSeqStart = NewCallSeqStart;
+ // And copy the pieces of it that fit into registers.
+ for (unsigned j=0; j<Size; j+=PtrByteSize) {
+ SDValue Const = DAG.getConstant(j, PtrOff.getValueType());
+ SDValue AddArg = DAG.getNode(ISD::ADD, PtrVT, Arg, Const);
+ if (GPR_idx != NumGPRs) {
+ SDValue Load = DAG.getLoad(PtrVT, Chain, AddArg, NULL, 0);
+ MemOpChains.push_back(Load.getValue(1));
+ RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
+ if (isMachoABI)
+ ArgOffset += PtrByteSize;
+ } else {
+ ArgOffset += ((Size - j + PtrByteSize-1)/PtrByteSize)*PtrByteSize;
+ break;
+ }
+ }
+ continue;
+ }
+
+ switch (Arg.getValueType().getSimpleVT()) {
default: assert(0 && "Unexpected ValueType for argument!");
case MVT::i32:
case MVT::i64:
// Double word align in ELF
- if (isELF32_ABI && Expand) GPR_idx += (GPR_idx % 2);
+ if (isELF32_ABI && Align) GPR_idx += (GPR_idx % 2);
if (GPR_idx != NumGPRs) {
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Arg));
} else {
- MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
+ LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
+ isPPC64, isTailCall, false, MemOpChains,
+ TailCallArguments);
inMem = true;
}
if (inMem || isMachoABI) {
// Stack align in ELF
- if (isELF32_ABI && Expand)
+ if (isELF32_ABI && Align)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
ArgOffset += PtrByteSize;
break;
case MVT::f32:
case MVT::f64:
- if (isVarArg) {
- // Float varargs need to be promoted to double.
- if (Arg.getValueType() == MVT::f32)
- Arg = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Arg);
- }
-
if (FPR_idx != NumFPRs) {
RegsToPass.push_back(std::make_pair(FPR[FPR_idx++], Arg));
if (isVarArg) {
- SDOperand Store = DAG.getStore(Chain, Arg, PtrOff, NULL, 0);
+ SDValue Store = DAG.getStore(Chain, Arg, PtrOff, NULL, 0);
MemOpChains.push_back(Store);
// Float varargs are always shadowed in available integer registers
if (GPR_idx != NumGPRs) {
- SDOperand Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
MemOpChains.push_back(Load.getValue(1));
if (isMachoABI) RegsToPass.push_back(std::make_pair(GPR[GPR_idx++],
Load));
}
if (GPR_idx != NumGPRs && Arg.getValueType() == MVT::f64 && !isPPC64){
- SDOperand ConstFour = DAG.getConstant(4, PtrOff.getValueType());
+ SDValue ConstFour = DAG.getConstant(4, PtrOff.getValueType());
PtrOff = DAG.getNode(ISD::ADD, PtrVT, PtrOff, ConstFour);
- SDOperand Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
+ SDValue Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
MemOpChains.push_back(Load.getValue(1));
if (isMachoABI) RegsToPass.push_back(std::make_pair(GPR[GPR_idx++],
Load));
}
}
} else {
- MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
+ LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
+ isPPC64, isTailCall, false, MemOpChains,
+ TailCallArguments);
inMem = true;
}
if (inMem || isMachoABI) {
// Stack align in ELF
- if (isELF32_ABI && Expand)
+ if (isELF32_ABI && Align)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
if (isPPC64)
ArgOffset += 8;
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
- assert(!isVarArg && "Don't support passing vectors to varargs yet!");
- assert(VR_idx != NumVRs &&
- "Don't support passing more than 12 vector args yet!");
- RegsToPass.push_back(std::make_pair(VR[VR_idx++], Arg));
+ if (isVarArg) {
+ // These go aligned on the stack, or in the corresponding R registers
+ // when within range. The Darwin PPC ABI doc claims they also go in
+ // V registers; in fact gcc does this only for arguments that are
+ // prototyped, not for those that match the ... We do it for all
+ // arguments, seems to work.
+ while (ArgOffset % 16 !=0) {
+ ArgOffset += PtrByteSize;
+ if (GPR_idx != NumGPRs)
+ GPR_idx++;
+ }
+ // We could elide this store in the case where the object fits
+ // entirely in R registers. Maybe later.
+ PtrOff = DAG.getNode(ISD::ADD, PtrVT, StackPtr,
+ DAG.getConstant(ArgOffset, PtrVT));
+ SDValue Store = DAG.getStore(Chain, Arg, PtrOff, NULL, 0);
+ MemOpChains.push_back(Store);
+ if (VR_idx != NumVRs) {
+ SDValue Load = DAG.getLoad(MVT::v4f32, Store, PtrOff, NULL, 0);
+ MemOpChains.push_back(Load.getValue(1));
+ RegsToPass.push_back(std::make_pair(VR[VR_idx++], Load));
+ }
+ ArgOffset += 16;
+ for (unsigned i=0; i<16; i+=PtrByteSize) {
+ if (GPR_idx == NumGPRs)
+ break;
+ SDValue Ix = DAG.getNode(ISD::ADD, PtrVT, PtrOff,
+ DAG.getConstant(i, PtrVT));
+ SDValue Load = DAG.getLoad(PtrVT, Store, Ix, NULL, 0);
+ MemOpChains.push_back(Load.getValue(1));
+ RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
+ }
+ break;
+ }
+
+ // Non-varargs Altivec params generally go in registers, but have
+ // stack space allocated at the end.
+ if (VR_idx != NumVRs) {
+ // Doesn't have GPR space allocated.
+ RegsToPass.push_back(std::make_pair(VR[VR_idx++], Arg));
+ } else if (nAltivecParamsAtEnd==0) {
+ // We are emitting Altivec params in order.
+ LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
+ isPPC64, isTailCall, true, MemOpChains,
+ TailCallArguments);
+ ArgOffset += 16;
+ }
break;
}
}
+ // If all Altivec parameters fit in registers, as they usually do,
+ // they get stack space following the non-Altivec parameters. We
+ // don't track this here because nobody below needs it.
+ // If there are more Altivec parameters than fit in registers emit
+ // the stores here.
+ if (!isVarArg && nAltivecParamsAtEnd > NumVRs) {
+ unsigned j = 0;
+ // Offset is aligned; skip 1st 12 params which go in V registers.
+ ArgOffset = ((ArgOffset+15)/16)*16;
+ ArgOffset += 12*16;
+ for (unsigned i = 0; i != NumOps; ++i) {
+ SDValue Arg = Op.getOperand(5+2*i);
+ MVT ArgType = Arg.getValueType();
+ if (ArgType==MVT::v4f32 || ArgType==MVT::v4i32 ||
+ ArgType==MVT::v8i16 || ArgType==MVT::v16i8) {
+ if (++j > NumVRs) {
+ SDValue PtrOff;
+ // We are emitting Altivec params in order.
+ LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
+ isPPC64, isTailCall, true, MemOpChains,
+ TailCallArguments);
+ ArgOffset += 16;
+ }
+ }
+ }
+ }
+
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
- SDOperand InFlag;
+ SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, RegsToPass[i].first, RegsToPass[i].second,
InFlag);
// With the ELF 32 ABI, set CR6 to true if this is a vararg call.
if (isVarArg && isELF32_ABI) {
- SDOperand SetCR(DAG.getTargetNode(PPC::SETCR, MVT::i32), 0);
- Chain = DAG.getCopyToReg(Chain, PPC::CR6, SetCR, InFlag);
+ SDValue SetCR(DAG.getTargetNode(PPC::CRSET, MVT::i32), 0);
+ Chain = DAG.getCopyToReg(Chain, PPC::CR1EQ, SetCR, InFlag);
InFlag = Chain.getValue(1);
}
- std::vector<MVT::ValueType> NodeTys;
+ // Emit a sequence of copyto/copyfrom virtual registers for arguments that
+ // might overwrite each other in case of tail call optimization.
+ if (isTailCall) {
+ SmallVector<SDValue, 8> MemOpChains2;
+ // Do not flag preceeding copytoreg stuff together with the following stuff.
+ InFlag = SDValue();
+ StoreTailCallArgumentsToStackSlot(DAG, Chain, TailCallArguments,
+ MemOpChains2);
+ if (!MemOpChains2.empty())
+ Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
+ &MemOpChains2[0], MemOpChains2.size());
+
+ // Store the return address to the appropriate stack slot.
+ Chain = EmitTailCallStoreFPAndRetAddr(DAG, MF, Chain, LROp, FPOp, SPDiff,
+ isPPC64, isMachoABI);
+ }
+
+ // Emit callseq_end just before tailcall node.
+ if (isTailCall) {
+ SmallVector<SDValue, 8> CallSeqOps;
+ SDVTList CallSeqNodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
+ CallSeqOps.push_back(Chain);
+ CallSeqOps.push_back(DAG.getIntPtrConstant(NumBytes));
+ CallSeqOps.push_back(DAG.getIntPtrConstant(0));
+ if (InFlag.Val)
+ CallSeqOps.push_back(InFlag);
+ Chain = DAG.getNode(ISD::CALLSEQ_END, CallSeqNodeTys, &CallSeqOps[0],
+ CallSeqOps.size());
+ InFlag = Chain.getValue(1);
+ }
+
+ std::vector<MVT> NodeTys;
NodeTys.push_back(MVT::Other); // Returns a chain
NodeTys.push_back(MVT::Flag); // Returns a flag for retval copy to use.
- SmallVector<SDOperand, 8> Ops;
+ SmallVector<SDValue, 8> Ops;
unsigned CallOpc = isMachoABI? PPCISD::CALL_Macho : PPCISD::CALL_ELF;
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), Callee.getValueType());
else if (SDNode *Dest = isBLACompatibleAddress(Callee, DAG))
// If this is an absolute destination address, use the munged value.
- Callee = SDOperand(Dest, 0);
+ Callee = SDValue(Dest, 0);
else {
// Otherwise, this is an indirect call. We have to use a MTCTR/BCTRL pair
// to do the call, we can't use PPCISD::CALL.
- SDOperand MTCTROps[] = {Chain, Callee, InFlag};
+ SDValue MTCTROps[] = {Chain, Callee, InFlag};
Chain = DAG.getNode(PPCISD::MTCTR, NodeTys, MTCTROps, 2+(InFlag.Val!=0));
InFlag = Chain.getValue(1);
- // Copy the callee address into R12 on darwin.
+ // Copy the callee address into R12/X12 on darwin.
if (isMachoABI) {
- Chain = DAG.getCopyToReg(Chain, PPC::R12, Callee, InFlag);
+ unsigned Reg = Callee.getValueType() == MVT::i32 ? PPC::R12 : PPC::X12;
+ Chain = DAG.getCopyToReg(Chain, Reg, Callee, InFlag);
InFlag = Chain.getValue(1);
}
Ops.push_back(Chain);
CallOpc = isMachoABI ? PPCISD::BCTRL_Macho : PPCISD::BCTRL_ELF;
Callee.Val = 0;
+ // Add CTR register as callee so a bctr can be emitted later.
+ if (isTailCall)
+ Ops.push_back(DAG.getRegister(PPC::CTR, getPointerTy()));
}
// If this is a direct call, pass the chain and the callee.
Ops.push_back(Chain);
Ops.push_back(Callee);
}
-
+ // If this is a tail call add stack pointer delta.
+ if (isTailCall)
+ Ops.push_back(DAG.getConstant(SPDiff, MVT::i32));
+
// Add argument registers to the end of the list so that they are known live
// into the call.
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
-
+
+ // When performing tail call optimization the callee pops its arguments off
+ // the stack. Account for this here so these bytes can be pushed back on in
+ // PPCRegisterInfo::eliminateCallFramePseudoInstr.
+ int BytesCalleePops =
+ (CC==CallingConv::Fast && PerformTailCallOpt) ? NumBytes : 0;
+
if (InFlag.Val)
Ops.push_back(InFlag);
+
+ // Emit tail call.
+ if (isTailCall) {
+ assert(InFlag.Val &&
+ "Flag must be set. Depend on flag being set in LowerRET");
+ Chain = DAG.getNode(PPCISD::TAILCALL,
+ Op.Val->getVTList(), &Ops[0], Ops.size());
+ return SDValue(Chain.Val, Op.ResNo);
+ }
+
Chain = DAG.getNode(CallOpc, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getConstant(NumBytes, PtrVT),
- DAG.getConstant(0, PtrVT),
+ DAG.getConstant(BytesCalleePops, PtrVT),
InFlag);
if (Op.Val->getValueType(0) != MVT::Other)
InFlag = Chain.getValue(1);
- SDOperand ResultVals[3];
- unsigned NumResults = 0;
- NodeTys.clear();
+ SmallVector<SDValue, 16> ResultVals;
+ SmallVector<CCValAssign, 16> RVLocs;
+ unsigned CallerCC = DAG.getMachineFunction().getFunction()->getCallingConv();
+ CCState CCInfo(CallerCC, isVarArg, TM, RVLocs);
+ CCInfo.AnalyzeCallResult(Op.Val, RetCC_PPC);
- // If the call has results, copy the values out of the ret val registers.
- switch (Op.Val->getValueType(0)) {
- default: assert(0 && "Unexpected ret value!");
- case MVT::Other: break;
- case MVT::i32:
- if (Op.Val->getValueType(1) == MVT::i32) {
- Chain = DAG.getCopyFromReg(Chain, PPC::R3, MVT::i32, InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- Chain = DAG.getCopyFromReg(Chain, PPC::R4, MVT::i32,
- Chain.getValue(2)).getValue(1);
- ResultVals[1] = Chain.getValue(0);
- NumResults = 2;
- NodeTys.push_back(MVT::i32);
- } else {
- Chain = DAG.getCopyFromReg(Chain, PPC::R3, MVT::i32, InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- NumResults = 1;
- }
- NodeTys.push_back(MVT::i32);
- break;
- case MVT::i64:
- Chain = DAG.getCopyFromReg(Chain, PPC::X3, MVT::i64, InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- NumResults = 1;
- NodeTys.push_back(MVT::i64);
- break;
- case MVT::f64:
- if (Op.Val->getValueType(1) == MVT::f64) {
- Chain = DAG.getCopyFromReg(Chain, PPC::F1, MVT::f64, InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- Chain = DAG.getCopyFromReg(Chain, PPC::F2, MVT::f64,
- Chain.getValue(2)).getValue(1);
- ResultVals[1] = Chain.getValue(0);
- NumResults = 2;
- NodeTys.push_back(MVT::f64);
- NodeTys.push_back(MVT::f64);
- break;
- }
- // else fall through
- case MVT::f32:
- Chain = DAG.getCopyFromReg(Chain, PPC::F1, Op.Val->getValueType(0),
- InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- NumResults = 1;
- NodeTys.push_back(Op.Val->getValueType(0));
- break;
- case MVT::v4f32:
- case MVT::v4i32:
- case MVT::v8i16:
- case MVT::v16i8:
- Chain = DAG.getCopyFromReg(Chain, PPC::V2, Op.Val->getValueType(0),
- InFlag).getValue(1);
- ResultVals[0] = Chain.getValue(0);
- NumResults = 1;
- NodeTys.push_back(Op.Val->getValueType(0));
- break;
+ // Copy all of the result registers out of their specified physreg.
+ for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
+ CCValAssign &VA = RVLocs[i];
+ MVT VT = VA.getValVT();
+ assert(VA.isRegLoc() && "Can only return in registers!");
+ Chain = DAG.getCopyFromReg(Chain, VA.getLocReg(), VT, InFlag).getValue(1);
+ ResultVals.push_back(Chain.getValue(0));
+ InFlag = Chain.getValue(2);
}
-
- NodeTys.push_back(MVT::Other);
-
+
// If the function returns void, just return the chain.
- if (NumResults == 0)
+ if (RVLocs.empty())
return Chain;
// Otherwise, merge everything together with a MERGE_VALUES node.
- ResultVals[NumResults++] = Chain;
- SDOperand Res = DAG.getNode(ISD::MERGE_VALUES, NodeTys,
- ResultVals, NumResults);
+ ResultVals.push_back(Chain);
+ SDValue Res = DAG.getMergeValues(Op.Val->getVTList(), &ResultVals[0],
+ ResultVals.size());
return Res.getValue(Op.ResNo);
}
-static SDOperand LowerRET(SDOperand Op, SelectionDAG &DAG, TargetMachine &TM) {
+SDValue PPCTargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG,
+ TargetMachine &TM) {
SmallVector<CCValAssign, 16> RVLocs;
unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
- SDOperand Chain = Op.getOperand(0);
- SDOperand Flag;
+ SDValue Chain = Op.getOperand(0);
+
+ Chain = GetPossiblePreceedingTailCall(Chain, PPCISD::TAILCALL);
+ if (Chain.getOpcode() == PPCISD::TAILCALL) {
+ SDValue TailCall = Chain;
+ SDValue TargetAddress = TailCall.getOperand(1);
+ SDValue StackAdjustment = TailCall.getOperand(2);
+
+ assert(((TargetAddress.getOpcode() == ISD::Register &&
+ cast<RegisterSDNode>(TargetAddress)->getReg() == PPC::CTR) ||
+ TargetAddress.getOpcode() == ISD::TargetExternalSymbol ||
+ TargetAddress.getOpcode() == ISD::TargetGlobalAddress ||
+ isa<ConstantSDNode>(TargetAddress)) &&
+ "Expecting an global address, external symbol, absolute value or register");
+
+ assert(StackAdjustment.getOpcode() == ISD::Constant &&
+ "Expecting a const value");
+
+ SmallVector<SDValue,8> Operands;
+ Operands.push_back(Chain.getOperand(0));
+ Operands.push_back(TargetAddress);
+ Operands.push_back(StackAdjustment);
+ // Copy registers used by the call. Last operand is a flag so it is not
+ // copied.
+ for (unsigned i=3; i < TailCall.getNumOperands()-1; i++) {
+ Operands.push_back(Chain.getOperand(i));
+ }
+ return DAG.getNode(PPCISD::TC_RETURN, MVT::Other, &Operands[0],
+ Operands.size());
+ }
+
+ SDValue Flag;
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
return DAG.getNode(PPCISD::RET_FLAG, MVT::Other, Chain);
}
-static SDOperand LowerSTACKRESTORE(SDOperand Op, SelectionDAG &DAG,
+SDValue PPCTargetLowering::LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
// When we pop the dynamic allocation we need to restore the SP link.
// Get the corect type for pointers.
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Construct the stack pointer operand.
bool IsPPC64 = Subtarget.isPPC64();
unsigned SP = IsPPC64 ? PPC::X1 : PPC::R1;
- SDOperand StackPtr = DAG.getRegister(SP, PtrVT);
+ SDValue StackPtr = DAG.getRegister(SP, PtrVT);
// Get the operands for the STACKRESTORE.
- SDOperand Chain = Op.getOperand(0);
- SDOperand SaveSP = Op.getOperand(1);
+ SDValue Chain = Op.getOperand(0);
+ SDValue SaveSP = Op.getOperand(1);
// Load the old link SP.
- SDOperand LoadLinkSP = DAG.getLoad(PtrVT, Chain, StackPtr, NULL, 0);
+ SDValue LoadLinkSP = DAG.getLoad(PtrVT, Chain, StackPtr, NULL, 0);
// Restore the stack pointer.
Chain = DAG.getCopyToReg(LoadLinkSP.getValue(1), SP, SaveSP);
return DAG.getStore(Chain, LoadLinkSP, StackPtr, NULL, 0);
}
-static SDOperand LowerDYNAMIC_STACKALLOC(SDOperand Op, SelectionDAG &DAG,
- const PPCSubtarget &Subtarget) {
+
+
+SDValue
+PPCTargetLowering::getReturnAddrFrameIndex(SelectionDAG & DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
- bool IsPPC64 = Subtarget.isPPC64();
- bool isMachoABI = Subtarget.isMachoABI();
+ bool IsPPC64 = PPCSubTarget.isPPC64();
+ bool isMachoABI = PPCSubTarget.isMachoABI();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+
+ // Get current frame pointer save index. The users of this index will be
+ // primarily DYNALLOC instructions.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ int RASI = FI->getReturnAddrSaveIndex();
+
+ // If the frame pointer save index hasn't been defined yet.
+ if (!RASI) {
+ // Find out what the fix offset of the frame pointer save area.
+ int LROffset = PPCFrameInfo::getReturnSaveOffset(IsPPC64, isMachoABI);
+ // Allocate the frame index for frame pointer save area.
+ RASI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, LROffset);
+ // Save the result.
+ FI->setReturnAddrSaveIndex(RASI);
+ }
+ return DAG.getFrameIndex(RASI, PtrVT);
+}
+
+SDValue
+PPCTargetLowering::getFramePointerFrameIndex(SelectionDAG & DAG) const {
+ MachineFunction &MF = DAG.getMachineFunction();
+ bool IsPPC64 = PPCSubTarget.isPPC64();
+ bool isMachoABI = PPCSubTarget.isMachoABI();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
int FPSI = FI->getFramePointerSaveIndex();
-
+
// If the frame pointer save index hasn't been defined yet.
if (!FPSI) {
// Find out what the fix offset of the frame pointer save area.
// Save the result.
FI->setFramePointerSaveIndex(FPSI);
}
+ return DAG.getFrameIndex(FPSI, PtrVT);
+}
+SDValue PPCTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
+ SelectionDAG &DAG,
+ const PPCSubtarget &Subtarget) {
// Get the inputs.
- SDOperand Chain = Op.getOperand(0);
- SDOperand Size = Op.getOperand(1);
+ SDValue Chain = Op.getOperand(0);
+ SDValue Size = Op.getOperand(1);
// Get the corect type for pointers.
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Negate the size.
- SDOperand NegSize = DAG.getNode(ISD::SUB, PtrVT,
+ SDValue NegSize = DAG.getNode(ISD::SUB, PtrVT,
DAG.getConstant(0, PtrVT), Size);
// Construct a node for the frame pointer save index.
- SDOperand FPSIdx = DAG.getFrameIndex(FPSI, PtrVT);
+ SDValue FPSIdx = getFramePointerFrameIndex(DAG);
// Build a DYNALLOC node.
- SDOperand Ops[3] = { Chain, NegSize, FPSIdx };
+ SDValue Ops[3] = { Chain, NegSize, FPSIdx };
SDVTList VTs = DAG.getVTList(PtrVT, MVT::Other);
return DAG.getNode(PPCISD::DYNALLOC, VTs, Ops, 3);
}
+SDValue PPCTargetLowering::LowerAtomicLOAD_ADD(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.Val->getValueType(0);
+ SDValue Chain = Op.getOperand(0);
+ SDValue Ptr = Op.getOperand(1);
+ SDValue Incr = Op.getOperand(2);
+
+ SDVTList VTs = DAG.getVTList(VT, MVT::Other);
+ SDValue Ops[] = {
+ Chain,
+ Ptr,
+ Incr,
+ };
+ return DAG.getNode(PPCISD::ATOMIC_LOAD_ADD, VTs, Ops, 3);
+}
+
+SDValue PPCTargetLowering::LowerAtomicCMP_SWAP(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.Val->getValueType(0);
+ SDValue Chain = Op.getOperand(0);
+ SDValue Ptr = Op.getOperand(1);
+ SDValue NewVal = Op.getOperand(2);
+ SDValue OldVal = Op.getOperand(3);
+
+ SDVTList VTs = DAG.getVTList(VT, MVT::Other);
+ SDValue Ops[] = {
+ Chain,
+ Ptr,
+ OldVal,
+ NewVal,
+ };
+ return DAG.getNode(PPCISD::ATOMIC_CMP_SWAP, VTs, Ops, 4);
+}
+
+SDValue PPCTargetLowering::LowerAtomicSWAP(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.Val->getValueType(0);
+ SDValue Chain = Op.getOperand(0);
+ SDValue Ptr = Op.getOperand(1);
+ SDValue NewVal = Op.getOperand(2);
+
+ SDVTList VTs = DAG.getVTList(VT, MVT::Other);
+ SDValue Ops[] = {
+ Chain,
+ Ptr,
+ NewVal,
+ };
+ return DAG.getNode(PPCISD::ATOMIC_SWAP, VTs, Ops, 3);
+}
/// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when
/// possible.
-static SDOperand LowerSELECT_CC(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) {
// Not FP? Not a fsel.
- if (!MVT::isFloatingPoint(Op.getOperand(0).getValueType()) ||
- !MVT::isFloatingPoint(Op.getOperand(2).getValueType()))
- return SDOperand();
+ if (!Op.getOperand(0).getValueType().isFloatingPoint() ||
+ !Op.getOperand(2).getValueType().isFloatingPoint())
+ return SDValue();
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
// Cannot handle SETEQ/SETNE.
- if (CC == ISD::SETEQ || CC == ISD::SETNE) return SDOperand();
+ if (CC == ISD::SETEQ || CC == ISD::SETNE) return SDValue();
- MVT::ValueType ResVT = Op.getValueType();
- MVT::ValueType CmpVT = Op.getOperand(0).getValueType();
- SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
- SDOperand TV = Op.getOperand(2), FV = Op.getOperand(3);
+ MVT ResVT = Op.getValueType();
+ MVT CmpVT = Op.getOperand(0).getValueType();
+ SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+ SDValue TV = Op.getOperand(2), FV = Op.getOperand(3);
// If the RHS of the comparison is a 0.0, we don't need to do the
// subtraction at all.
DAG.getNode(ISD::FNEG, MVT::f64, LHS), TV, FV);
}
- SDOperand Cmp;
+ SDValue Cmp;
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
case ISD::SETULT:
Cmp = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, ResVT, Cmp, TV, FV);
}
- return SDOperand();
+ return SDValue();
}
// FIXME: Split this code up when LegalizeDAGTypes lands.
-static SDOperand LowerFP_TO_SINT(SDOperand Op, SelectionDAG &DAG) {
- assert(MVT::isFloatingPoint(Op.getOperand(0).getValueType()));
- SDOperand Src = Op.getOperand(0);
+SDValue PPCTargetLowering::LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) {
+ assert(Op.getOperand(0).getValueType().isFloatingPoint());
+ SDValue Src = Op.getOperand(0);
if (Src.getValueType() == MVT::f32)
Src = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Src);
-
- SDOperand Tmp;
- switch (Op.getValueType()) {
+
+ SDValue Tmp;
+ switch (Op.getValueType().getSimpleVT()) {
default: assert(0 && "Unhandled FP_TO_SINT type in custom expander!");
case MVT::i32:
Tmp = DAG.getNode(PPCISD::FCTIWZ, MVT::f64, Src);
Tmp = DAG.getNode(PPCISD::FCTIDZ, MVT::f64, Src);
break;
}
-
+
// Convert the FP value to an int value through memory.
- SDOperand FIPtr = DAG.CreateStackTemporary(MVT::f64);
-
+ SDValue FIPtr = DAG.CreateStackTemporary(MVT::f64);
+
// Emit a store to the stack slot.
- SDOperand Chain = DAG.getStore(DAG.getEntryNode(), Tmp, FIPtr, NULL, 0);
+ SDValue Chain = DAG.getStore(DAG.getEntryNode(), Tmp, FIPtr, NULL, 0);
// Result is a load from the stack slot. If loading 4 bytes, make sure to
// add in a bias.
return DAG.getLoad(Op.getValueType(), Chain, FIPtr, NULL, 0);
}
-static SDOperand LowerFP_ROUND_INREG(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerFP_ROUND_INREG(SDValue Op,
+ SelectionDAG &DAG) {
assert(Op.getValueType() == MVT::ppcf128);
SDNode *Node = Op.Val;
assert(Node->getOperand(0).getValueType() == MVT::ppcf128);
assert(Node->getOperand(0).Val->getOpcode() == ISD::BUILD_PAIR);
- SDOperand Lo = Node->getOperand(0).Val->getOperand(0);
- SDOperand Hi = Node->getOperand(0).Val->getOperand(1);
+ SDValue Lo = Node->getOperand(0).Val->getOperand(0);
+ SDValue Hi = Node->getOperand(0).Val->getOperand(1);
// This sequence changes FPSCR to do round-to-zero, adds the two halves
// of the long double, and puts FPSCR back the way it was. We do not
// actually model FPSCR.
- std::vector<MVT::ValueType> NodeTys;
- SDOperand Ops[4], Result, MFFSreg, InFlag, FPreg;
+ std::vector<MVT> NodeTys;
+ SDValue Ops[4], Result, MFFSreg, InFlag, FPreg;
NodeTys.push_back(MVT::f64); // Return register
NodeTys.push_back(MVT::Flag); // Returns a flag for later insns
return DAG.getNode(ISD::BUILD_PAIR, Lo.getValueType(), FPreg, FPreg);
}
-static SDOperand LowerSINT_TO_FP(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
+ // Don't handle ppc_fp128 here; let it be lowered to a libcall.
+ if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
+ return SDValue();
+
if (Op.getOperand(0).getValueType() == MVT::i64) {
- SDOperand Bits = DAG.getNode(ISD::BIT_CONVERT, MVT::f64, Op.getOperand(0));
- SDOperand FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Bits);
+ SDValue Bits = DAG.getNode(ISD::BIT_CONVERT, MVT::f64, Op.getOperand(0));
+ SDValue FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Bits);
if (Op.getValueType() == MVT::f32)
FP = DAG.getNode(ISD::FP_ROUND, MVT::f32, FP, DAG.getIntPtrConstant(0));
return FP;
// then lfd it and fcfid it.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = FrameInfo->CreateStackObject(8, 8);
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- SDOperand FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
- SDOperand Ext64 = DAG.getNode(PPCISD::EXTSW_32, MVT::i32,
+ SDValue Ext64 = DAG.getNode(PPCISD::EXTSW_32, MVT::i32,
Op.getOperand(0));
// STD the extended value into the stack slot.
- SDOperand Store = DAG.getNode(PPCISD::STD_32, MVT::Other,
+ MachineMemOperand MO(PseudoSourceValue::getFixedStack(FrameIdx),
+ MachineMemOperand::MOStore, 0, 8, 8);
+ SDValue Store = DAG.getNode(PPCISD::STD_32, MVT::Other,
DAG.getEntryNode(), Ext64, FIdx,
- DAG.getSrcValue(NULL));
+ DAG.getMemOperand(MO));
// Load the value as a double.
- SDOperand Ld = DAG.getLoad(MVT::f64, Store, FIdx, NULL, 0);
+ SDValue Ld = DAG.getLoad(MVT::f64, Store, FIdx, NULL, 0);
// FCFID it and return it.
- SDOperand FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Ld);
+ SDValue FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Ld);
if (Op.getValueType() == MVT::f32)
FP = DAG.getNode(ISD::FP_ROUND, MVT::f32, FP, DAG.getIntPtrConstant(0));
return FP;
}
-static SDOperand LowerFLT_ROUNDS(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) {
/*
The rounding mode is in bits 30:31 of FPSR, and has the following
settings:
*/
MachineFunction &MF = DAG.getMachineFunction();
- MVT::ValueType VT = Op.getValueType();
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- std::vector<MVT::ValueType> NodeTys;
- SDOperand MFFSreg, InFlag;
+ MVT VT = Op.getValueType();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ std::vector<MVT> NodeTys;
+ SDValue MFFSreg, InFlag;
// Save FP Control Word to register
NodeTys.push_back(MVT::f64); // return register
NodeTys.push_back(MVT::Flag); // unused in this context
- SDOperand Chain = DAG.getNode(PPCISD::MFFS, NodeTys, &InFlag, 0);
+ SDValue Chain = DAG.getNode(PPCISD::MFFS, NodeTys, &InFlag, 0);
// Save FP register to stack slot
int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
- SDOperand StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
- SDOperand Store = DAG.getStore(DAG.getEntryNode(), Chain,
+ SDValue StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
+ SDValue Store = DAG.getStore(DAG.getEntryNode(), Chain,
StackSlot, NULL, 0);
// Load FP Control Word from low 32 bits of stack slot.
- SDOperand Four = DAG.getConstant(4, PtrVT);
- SDOperand Addr = DAG.getNode(ISD::ADD, PtrVT, StackSlot, Four);
- SDOperand CWD = DAG.getLoad(MVT::i32, Store, Addr, NULL, 0);
+ SDValue Four = DAG.getConstant(4, PtrVT);
+ SDValue Addr = DAG.getNode(ISD::ADD, PtrVT, StackSlot, Four);
+ SDValue CWD = DAG.getLoad(MVT::i32, Store, Addr, NULL, 0);
// Transform as necessary
- SDOperand CWD1 =
+ SDValue CWD1 =
DAG.getNode(ISD::AND, MVT::i32,
CWD, DAG.getConstant(3, MVT::i32));
- SDOperand CWD2 =
+ SDValue CWD2 =
DAG.getNode(ISD::SRL, MVT::i32,
DAG.getNode(ISD::AND, MVT::i32,
DAG.getNode(ISD::XOR, MVT::i32,
DAG.getConstant(3, MVT::i32)),
DAG.getConstant(1, MVT::i8));
- SDOperand RetVal =
+ SDValue RetVal =
DAG.getNode(ISD::XOR, MVT::i32, CWD1, CWD2);
- return DAG.getNode((MVT::getSizeInBits(VT) < 16 ?
+ return DAG.getNode((VT.getSizeInBits() < 16 ?
ISD::TRUNCATE : ISD::ZERO_EXTEND), VT, RetVal);
}
-static SDOperand LowerSHL_PARTS(SDOperand Op, SelectionDAG &DAG) {
- assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
- Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SHL!");
+SDValue PPCTargetLowering::LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.getValueType();
+ unsigned BitWidth = VT.getSizeInBits();
+ assert(Op.getNumOperands() == 3 &&
+ VT == Op.getOperand(1).getValueType() &&
+ "Unexpected SHL!");
// Expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
- SDOperand Lo = Op.getOperand(0);
- SDOperand Hi = Op.getOperand(1);
- SDOperand Amt = Op.getOperand(2);
-
- SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
- DAG.getConstant(32, MVT::i32), Amt);
- SDOperand Tmp2 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Amt);
- SDOperand Tmp3 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Tmp1);
- SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
- SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
- DAG.getConstant(-32U, MVT::i32));
- SDOperand Tmp6 = DAG.getNode(PPCISD::SHL, MVT::i32, Lo, Tmp5);
- SDOperand OutHi = DAG.getNode(ISD::OR, MVT::i32, Tmp4, Tmp6);
- SDOperand OutLo = DAG.getNode(PPCISD::SHL, MVT::i32, Lo, Amt);
- SDOperand OutOps[] = { OutLo, OutHi };
- return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
- OutOps, 2);
-}
-
-static SDOperand LowerSRL_PARTS(SDOperand Op, SelectionDAG &DAG) {
- assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
- Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SRL!");
-
- // Otherwise, expand into a bunch of logical ops. Note that these ops
+ SDValue Lo = Op.getOperand(0);
+ SDValue Hi = Op.getOperand(1);
+ SDValue Amt = Op.getOperand(2);
+ MVT AmtVT = Amt.getValueType();
+
+ SDValue Tmp1 = DAG.getNode(ISD::SUB, AmtVT,
+ DAG.getConstant(BitWidth, AmtVT), Amt);
+ SDValue Tmp2 = DAG.getNode(PPCISD::SHL, VT, Hi, Amt);
+ SDValue Tmp3 = DAG.getNode(PPCISD::SRL, VT, Lo, Tmp1);
+ SDValue Tmp4 = DAG.getNode(ISD::OR , VT, Tmp2, Tmp3);
+ SDValue Tmp5 = DAG.getNode(ISD::ADD, AmtVT, Amt,
+ DAG.getConstant(-BitWidth, AmtVT));
+ SDValue Tmp6 = DAG.getNode(PPCISD::SHL, VT, Lo, Tmp5);
+ SDValue OutHi = DAG.getNode(ISD::OR, VT, Tmp4, Tmp6);
+ SDValue OutLo = DAG.getNode(PPCISD::SHL, VT, Lo, Amt);
+ SDValue OutOps[] = { OutLo, OutHi };
+ return DAG.getMergeValues(OutOps, 2);
+}
+
+SDValue PPCTargetLowering::LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.getValueType();
+ unsigned BitWidth = VT.getSizeInBits();
+ assert(Op.getNumOperands() == 3 &&
+ VT == Op.getOperand(1).getValueType() &&
+ "Unexpected SRL!");
+
+ // Expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
- SDOperand Lo = Op.getOperand(0);
- SDOperand Hi = Op.getOperand(1);
- SDOperand Amt = Op.getOperand(2);
-
- SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
- DAG.getConstant(32, MVT::i32), Amt);
- SDOperand Tmp2 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Amt);
- SDOperand Tmp3 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Tmp1);
- SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
- SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
- DAG.getConstant(-32U, MVT::i32));
- SDOperand Tmp6 = DAG.getNode(PPCISD::SRL, MVT::i32, Hi, Tmp5);
- SDOperand OutLo = DAG.getNode(ISD::OR, MVT::i32, Tmp4, Tmp6);
- SDOperand OutHi = DAG.getNode(PPCISD::SRL, MVT::i32, Hi, Amt);
- SDOperand OutOps[] = { OutLo, OutHi };
- return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
- OutOps, 2);
-}
-
-static SDOperand LowerSRA_PARTS(SDOperand Op, SelectionDAG &DAG) {
- assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
- Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SRA!");
-
- // Otherwise, expand into a bunch of logical ops, followed by a select_cc.
- SDOperand Lo = Op.getOperand(0);
- SDOperand Hi = Op.getOperand(1);
- SDOperand Amt = Op.getOperand(2);
-
- SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
- DAG.getConstant(32, MVT::i32), Amt);
- SDOperand Tmp2 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Amt);
- SDOperand Tmp3 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Tmp1);
- SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
- SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
- DAG.getConstant(-32U, MVT::i32));
- SDOperand Tmp6 = DAG.getNode(PPCISD::SRA, MVT::i32, Hi, Tmp5);
- SDOperand OutHi = DAG.getNode(PPCISD::SRA, MVT::i32, Hi, Amt);
- SDOperand OutLo = DAG.getSelectCC(Tmp5, DAG.getConstant(0, MVT::i32),
+ SDValue Lo = Op.getOperand(0);
+ SDValue Hi = Op.getOperand(1);
+ SDValue Amt = Op.getOperand(2);
+ MVT AmtVT = Amt.getValueType();
+
+ SDValue Tmp1 = DAG.getNode(ISD::SUB, AmtVT,
+ DAG.getConstant(BitWidth, AmtVT), Amt);
+ SDValue Tmp2 = DAG.getNode(PPCISD::SRL, VT, Lo, Amt);
+ SDValue Tmp3 = DAG.getNode(PPCISD::SHL, VT, Hi, Tmp1);
+ SDValue Tmp4 = DAG.getNode(ISD::OR , VT, Tmp2, Tmp3);
+ SDValue Tmp5 = DAG.getNode(ISD::ADD, AmtVT, Amt,
+ DAG.getConstant(-BitWidth, AmtVT));
+ SDValue Tmp6 = DAG.getNode(PPCISD::SRL, VT, Hi, Tmp5);
+ SDValue OutLo = DAG.getNode(ISD::OR, VT, Tmp4, Tmp6);
+ SDValue OutHi = DAG.getNode(PPCISD::SRL, VT, Hi, Amt);
+ SDValue OutOps[] = { OutLo, OutHi };
+ return DAG.getMergeValues(OutOps, 2);
+}
+
+SDValue PPCTargetLowering::LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) {
+ MVT VT = Op.getValueType();
+ unsigned BitWidth = VT.getSizeInBits();
+ assert(Op.getNumOperands() == 3 &&
+ VT == Op.getOperand(1).getValueType() &&
+ "Unexpected SRA!");
+
+ // Expand into a bunch of logical ops, followed by a select_cc.
+ SDValue Lo = Op.getOperand(0);
+ SDValue Hi = Op.getOperand(1);
+ SDValue Amt = Op.getOperand(2);
+ MVT AmtVT = Amt.getValueType();
+
+ SDValue Tmp1 = DAG.getNode(ISD::SUB, AmtVT,
+ DAG.getConstant(BitWidth, AmtVT), Amt);
+ SDValue Tmp2 = DAG.getNode(PPCISD::SRL, VT, Lo, Amt);
+ SDValue Tmp3 = DAG.getNode(PPCISD::SHL, VT, Hi, Tmp1);
+ SDValue Tmp4 = DAG.getNode(ISD::OR , VT, Tmp2, Tmp3);
+ SDValue Tmp5 = DAG.getNode(ISD::ADD, AmtVT, Amt,
+ DAG.getConstant(-BitWidth, AmtVT));
+ SDValue Tmp6 = DAG.getNode(PPCISD::SRA, VT, Hi, Tmp5);
+ SDValue OutHi = DAG.getNode(PPCISD::SRA, VT, Hi, Amt);
+ SDValue OutLo = DAG.getSelectCC(Tmp5, DAG.getConstant(0, AmtVT),
Tmp4, Tmp6, ISD::SETLE);
- SDOperand OutOps[] = { OutLo, OutHi };
- return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
- OutOps, 2);
+ SDValue OutOps[] = { OutLo, OutHi };
+ return DAG.getMergeValues(OutOps, 2);
}
//===----------------------------------------------------------------------===//
// Start with zero'd results.
VectorBits[0] = VectorBits[1] = UndefBits[0] = UndefBits[1] = 0;
- unsigned EltBitSize = MVT::getSizeInBits(BV->getOperand(0).getValueType());
+ unsigned EltBitSize = BV->getOperand(0).getValueType().getSizeInBits();
for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) {
- SDOperand OpVal = BV->getOperand(i);
+ SDValue OpVal = BV->getOperand(i);
unsigned PartNo = i >= e/2; // In the upper 128 bits?
unsigned SlotNo = e/2 - (i & (e/2-1))-1; // Which subpiece of the uint64_t.
/// BuildSplatI - Build a canonical splati of Val with an element size of
/// SplatSize. Cast the result to VT.
-static SDOperand BuildSplatI(int Val, unsigned SplatSize, MVT::ValueType VT,
+static SDValue BuildSplatI(int Val, unsigned SplatSize, MVT VT,
SelectionDAG &DAG) {
assert(Val >= -16 && Val <= 15 && "vsplti is out of range!");
- static const MVT::ValueType VTys[] = { // canonical VT to use for each size.
+ static const MVT VTys[] = { // canonical VT to use for each size.
MVT::v16i8, MVT::v8i16, MVT::Other, MVT::v4i32
};
- MVT::ValueType ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
+ MVT ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
// Force vspltis[hw] -1 to vspltisb -1 to canonicalize.
if (Val == -1)
SplatSize = 1;
- MVT::ValueType CanonicalVT = VTys[SplatSize-1];
+ MVT CanonicalVT = VTys[SplatSize-1];
// Build a canonical splat for this value.
- SDOperand Elt = DAG.getConstant(Val, MVT::getVectorElementType(CanonicalVT));
- SmallVector<SDOperand, 8> Ops;
- Ops.assign(MVT::getVectorNumElements(CanonicalVT), Elt);
- SDOperand Res = DAG.getNode(ISD::BUILD_VECTOR, CanonicalVT,
+ SDValue Elt = DAG.getConstant(Val, CanonicalVT.getVectorElementType());
+ SmallVector<SDValue, 8> Ops;
+ Ops.assign(CanonicalVT.getVectorNumElements(), Elt);
+ SDValue Res = DAG.getNode(ISD::BUILD_VECTOR, CanonicalVT,
&Ops[0], Ops.size());
return DAG.getNode(ISD::BIT_CONVERT, ReqVT, Res);
}
/// BuildIntrinsicOp - Return a binary operator intrinsic node with the
/// specified intrinsic ID.
-static SDOperand BuildIntrinsicOp(unsigned IID, SDOperand LHS, SDOperand RHS,
+static SDValue BuildIntrinsicOp(unsigned IID, SDValue LHS, SDValue RHS,
SelectionDAG &DAG,
- MVT::ValueType DestVT = MVT::Other) {
+ MVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = LHS.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DestVT,
DAG.getConstant(IID, MVT::i32), LHS, RHS);
/// BuildIntrinsicOp - Return a ternary operator intrinsic node with the
/// specified intrinsic ID.
-static SDOperand BuildIntrinsicOp(unsigned IID, SDOperand Op0, SDOperand Op1,
- SDOperand Op2, SelectionDAG &DAG,
- MVT::ValueType DestVT = MVT::Other) {
+static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op0, SDValue Op1,
+ SDValue Op2, SelectionDAG &DAG,
+ MVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op0.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DestVT,
DAG.getConstant(IID, MVT::i32), Op0, Op1, Op2);
/// BuildVSLDOI - Return a VECTOR_SHUFFLE that is a vsldoi of the specified
/// amount. The result has the specified value type.
-static SDOperand BuildVSLDOI(SDOperand LHS, SDOperand RHS, unsigned Amt,
- MVT::ValueType VT, SelectionDAG &DAG) {
+static SDValue BuildVSLDOI(SDValue LHS, SDValue RHS, unsigned Amt,
+ MVT VT, SelectionDAG &DAG) {
// Force LHS/RHS to be the right type.
LHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, RHS);
-
- SDOperand Ops[16];
+
+ SDValue Ops[16];
for (unsigned i = 0; i != 16; ++i)
- Ops[i] = DAG.getConstant(i+Amt, MVT::i32);
- SDOperand T = DAG.getNode(ISD::VECTOR_SHUFFLE, MVT::v16i8, LHS, RHS,
+ Ops[i] = DAG.getConstant(i+Amt, MVT::i8);
+ SDValue T = DAG.getNode(ISD::VECTOR_SHUFFLE, MVT::v16i8, LHS, RHS,
DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8, Ops,16));
return DAG.getNode(ISD::BIT_CONVERT, VT, T);
}
// selects to a single instruction, return Op. Otherwise, if we can codegen
// this case more efficiently than a constant pool load, lower it to the
// sequence of ops that should be used.
-static SDOperand LowerBUILD_VECTOR(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
+ SelectionDAG &DAG) {
// If this is a vector of constants or undefs, get the bits. A bit in
// UndefBits is set if the corresponding element of the vector is an
// ISD::UNDEF value. For undefs, the corresponding VectorBits values are
uint64_t VectorBits[2];
uint64_t UndefBits[2];
if (GetConstantBuildVectorBits(Op.Val, VectorBits, UndefBits))
- return SDOperand(); // Not a constant vector.
+ return SDValue(); // Not a constant vector.
// If this is a splat (repetition) of a value across the whole vector, return
// the smallest size that splats it. For example, "0x01010101010101..." is a
if (SplatBits == 0) {
// Canonicalize all zero vectors to be v4i32.
if (Op.getValueType() != MVT::v4i32 || HasAnyUndefs) {
- SDOperand Z = DAG.getConstant(0, MVT::i32);
+ SDValue Z = DAG.getConstant(0, MVT::i32);
Z = DAG.getNode(ISD::BUILD_VECTOR, MVT::v4i32, Z, Z, Z, Z);
Op = DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Z);
}
// If this value is in the range [-32,30] and is even, use:
// tmp = VSPLTI[bhw], result = add tmp, tmp
if (SextVal >= -32 && SextVal <= 30 && (SextVal & 1) == 0) {
- Op = BuildSplatI(SextVal >> 1, SplatSize, Op.getValueType(), DAG);
- return DAG.getNode(ISD::ADD, Op.getValueType(), Op, Op);
+ SDValue Res = BuildSplatI(SextVal >> 1, SplatSize, MVT::Other, DAG);
+ Res = DAG.getNode(ISD::ADD, Res.getValueType(), Res, Res);
+ return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
// for fneg/fabs.
if (SplatSize == 4 && SplatBits == (0x7FFFFFFF&~SplatUndef)) {
// Make -1 and vspltisw -1:
- SDOperand OnesV = BuildSplatI(-1, 4, MVT::v4i32, DAG);
+ SDValue OnesV = BuildSplatI(-1, 4, MVT::v4i32, DAG);
// Make the VSLW intrinsic, computing 0x8000_0000.
- SDOperand Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
+ SDValue Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
OnesV, DAG);
// xor by OnesV to invert it.
// vsplti + shl self.
if (SextVal == (i << (int)TypeShiftAmt)) {
- SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vslb, Intrinsic::ppc_altivec_vslh, 0,
Intrinsic::ppc_altivec_vslw
// vsplti + srl self.
if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
- SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vsrb, Intrinsic::ppc_altivec_vsrh, 0,
Intrinsic::ppc_altivec_vsrw
// vsplti + sra self.
if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
- SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vsrab, Intrinsic::ppc_altivec_vsrah, 0,
Intrinsic::ppc_altivec_vsraw
// vsplti + rol self.
if (SextVal == (int)(((unsigned)i << TypeShiftAmt) |
((unsigned)i >> (SplatBitSize-TypeShiftAmt)))) {
- SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vrlb, Intrinsic::ppc_altivec_vrlh, 0,
Intrinsic::ppc_altivec_vrlw
// t = vsplti c, result = vsldoi t, t, 1
if (SextVal == ((i << 8) | (i >> (TypeShiftAmt-8)))) {
- SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 1, Op.getValueType(), DAG);
}
// t = vsplti c, result = vsldoi t, t, 2
if (SextVal == ((i << 16) | (i >> (TypeShiftAmt-16)))) {
- SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 2, Op.getValueType(), DAG);
}
// t = vsplti c, result = vsldoi t, t, 3
if (SextVal == ((i << 24) | (i >> (TypeShiftAmt-24)))) {
- SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 3, Op.getValueType(), DAG);
}
}
// Odd, in range [17,31]: (vsplti C)-(vsplti -16).
if (SextVal >= 0 && SextVal <= 31) {
- SDOperand LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG);
- SDOperand RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
+ SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG);
+ SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
LHS = DAG.getNode(ISD::SUB, LHS.getValueType(), LHS, RHS);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), LHS);
}
// Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
if (SextVal >= -31 && SextVal <= 0) {
- SDOperand LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG);
- SDOperand RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
+ SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG);
+ SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
LHS = DAG.getNode(ISD::ADD, LHS.getValueType(), LHS, RHS);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), LHS);
}
}
- return SDOperand();
+ return SDValue();
}
/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
/// the specified operations to build the shuffle.
-static SDOperand GeneratePerfectShuffle(unsigned PFEntry, SDOperand LHS,
- SDOperand RHS, SelectionDAG &DAG) {
+static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
+ SDValue RHS, SelectionDAG &DAG) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
return RHS;
}
- SDOperand OpLHS, OpRHS;
+ SDValue OpLHS, OpRHS;
OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG);
OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG);
case OP_VSLDOI12:
return BuildVSLDOI(OpLHS, OpRHS, 12, OpLHS.getValueType(), DAG);
}
- SDOperand Ops[16];
+ SDValue Ops[16];
for (unsigned i = 0; i != 16; ++i)
- Ops[i] = DAG.getConstant(ShufIdxs[i], MVT::i32);
+ Ops[i] = DAG.getConstant(ShufIdxs[i], MVT::i8);
return DAG.getNode(ISD::VECTOR_SHUFFLE, OpLHS.getValueType(), OpLHS, OpRHS,
DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8, Ops, 16));
/// is a shuffle we can handle in a single instruction, return it. Otherwise,
/// return the code it can be lowered into. Worst case, it can always be
/// lowered into a vperm.
-static SDOperand LowerVECTOR_SHUFFLE(SDOperand Op, SelectionDAG &DAG) {
- SDOperand V1 = Op.getOperand(0);
- SDOperand V2 = Op.getOperand(1);
- SDOperand PermMask = Op.getOperand(2);
+SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
+ SelectionDAG &DAG) {
+ SDValue V1 = Op.getOperand(0);
+ SDValue V2 = Op.getOperand(1);
+ SDValue PermMask = Op.getOperand(2);
// Cases that are handled by instructions that take permute immediates
// (such as vsplt*) should be left as VECTOR_SHUFFLE nodes so they can be
// The SHUFFLE_VECTOR mask is almost exactly what we want for vperm, except
// that it is in input element units, not in bytes. Convert now.
- MVT::ValueType EltVT = MVT::getVectorElementType(V1.getValueType());
- unsigned BytesPerElement = MVT::getSizeInBits(EltVT)/8;
+ MVT EltVT = V1.getValueType().getVectorElementType();
+ unsigned BytesPerElement = EltVT.getSizeInBits()/8;
- SmallVector<SDOperand, 16> ResultMask;
+ SmallVector<SDValue, 16> ResultMask;
for (unsigned i = 0, e = PermMask.getNumOperands(); i != e; ++i) {
unsigned SrcElt;
if (PermMask.getOperand(i).getOpcode() == ISD::UNDEF)
MVT::i8));
}
- SDOperand VPermMask = DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8,
+ SDValue VPermMask = DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8,
&ResultMask[0], ResultMask.size());
return DAG.getNode(PPCISD::VPERM, V1.getValueType(), V1, V2, VPermMask);
}
/// getAltivecCompareInfo - Given an intrinsic, return false if it is not an
/// altivec comparison. If it is, return true and fill in Opc/isDot with
/// information about the intrinsic.
-static bool getAltivecCompareInfo(SDOperand Intrin, int &CompareOpc,
+static bool getAltivecCompareInfo(SDValue Intrin, int &CompareOpc,
bool &isDot) {
unsigned IntrinsicID = cast<ConstantSDNode>(Intrin.getOperand(0))->getValue();
CompareOpc = -1;
/// LowerINTRINSIC_WO_CHAIN - If this is an intrinsic that we want to custom
/// lower, do it, otherwise return null.
-static SDOperand LowerINTRINSIC_WO_CHAIN(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
+ SelectionDAG &DAG) {
// If this is a lowered altivec predicate compare, CompareOpc is set to the
// opcode number of the comparison.
int CompareOpc;
bool isDot;
if (!getAltivecCompareInfo(Op, CompareOpc, isDot))
- return SDOperand(); // Don't custom lower most intrinsics.
+ return SDValue(); // Don't custom lower most intrinsics.
// If this is a non-dot comparison, make the VCMP node and we are done.
if (!isDot) {
- SDOperand Tmp = DAG.getNode(PPCISD::VCMP, Op.getOperand(2).getValueType(),
+ SDValue Tmp = DAG.getNode(PPCISD::VCMP, Op.getOperand(2).getValueType(),
Op.getOperand(1), Op.getOperand(2),
DAG.getConstant(CompareOpc, MVT::i32));
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Tmp);
}
// Create the PPCISD altivec 'dot' comparison node.
- SDOperand Ops[] = {
+ SDValue Ops[] = {
Op.getOperand(2), // LHS
Op.getOperand(3), // RHS
DAG.getConstant(CompareOpc, MVT::i32)
};
- std::vector<MVT::ValueType> VTs;
+ std::vector<MVT> VTs;
VTs.push_back(Op.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
- SDOperand CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
+ SDValue CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
// Now that we have the comparison, emit a copy from the CR to a GPR.
// This is flagged to the above dot comparison.
- SDOperand Flags = DAG.getNode(PPCISD::MFCR, MVT::i32,
+ SDValue Flags = DAG.getNode(PPCISD::MFCR, MVT::i32,
DAG.getRegister(PPC::CR6, MVT::i32),
CompNode.getValue(1));
return Flags;
}
-static SDOperand LowerSCALAR_TO_VECTOR(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
+ SelectionDAG &DAG) {
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = FrameInfo->CreateStackObject(16, 16);
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- SDOperand FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
// Store the input value into Value#0 of the stack slot.
- SDOperand Store = DAG.getStore(DAG.getEntryNode(),
+ SDValue Store = DAG.getStore(DAG.getEntryNode(),
Op.getOperand(0), FIdx, NULL, 0);
// Load it out.
return DAG.getLoad(Op.getValueType(), Store, FIdx, NULL, 0);
}
-static SDOperand LowerMUL(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) {
if (Op.getValueType() == MVT::v4i32) {
- SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+ SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
- SDOperand Zero = BuildSplatI( 0, 1, MVT::v4i32, DAG);
- SDOperand Neg16 = BuildSplatI(-16, 4, MVT::v4i32, DAG); // +16 as shift amt.
+ SDValue Zero = BuildSplatI( 0, 1, MVT::v4i32, DAG);
+ SDValue Neg16 = BuildSplatI(-16, 4, MVT::v4i32, DAG); // +16 as shift amt.
- SDOperand RHSSwap = // = vrlw RHS, 16
+ SDValue RHSSwap = // = vrlw RHS, 16
BuildIntrinsicOp(Intrinsic::ppc_altivec_vrlw, RHS, Neg16, DAG);
// Shrinkify inputs to v8i16.
// Low parts multiplied together, generating 32-bit results (we ignore the
// top parts).
- SDOperand LoProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmulouh,
+ SDValue LoProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmulouh,
LHS, RHS, DAG, MVT::v4i32);
- SDOperand HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmsumuhm,
+ SDValue HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmsumuhm,
LHS, RHSSwap, Zero, DAG, MVT::v4i32);
// Shift the high parts up 16 bits.
HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, HiProd, Neg16, DAG);
return DAG.getNode(ISD::ADD, MVT::v4i32, LoProd, HiProd);
} else if (Op.getValueType() == MVT::v8i16) {
- SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+ SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
- SDOperand Zero = BuildSplatI(0, 1, MVT::v8i16, DAG);
+ SDValue Zero = BuildSplatI(0, 1, MVT::v8i16, DAG);
return BuildIntrinsicOp(Intrinsic::ppc_altivec_vmladduhm,
LHS, RHS, Zero, DAG);
} else if (Op.getValueType() == MVT::v16i8) {
- SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+ SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
// Multiply the even 8-bit parts, producing 16-bit sums.
- SDOperand EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
+ SDValue EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
LHS, RHS, DAG, MVT::v8i16);
EvenParts = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, EvenParts);
// Multiply the odd 8-bit parts, producing 16-bit sums.
- SDOperand OddParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuloub,
+ SDValue OddParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuloub,
LHS, RHS, DAG, MVT::v8i16);
OddParts = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, OddParts);
// Merge the results together.
- SDOperand Ops[16];
+ SDValue Ops[16];
for (unsigned i = 0; i != 8; ++i) {
Ops[i*2 ] = DAG.getConstant(2*i+1, MVT::i8);
Ops[i*2+1] = DAG.getConstant(2*i+1+16, MVT::i8);
/// LowerOperation - Provide custom lowering hooks for some operations.
///
-SDOperand PPCTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
VarArgsStackOffset, VarArgsNumGPR,
VarArgsNumFPR, PPCSubTarget);
- case ISD::CALL: return LowerCALL(Op, DAG, PPCSubTarget);
+ case ISD::CALL: return LowerCALL(Op, DAG, PPCSubTarget,
+ getTargetMachine());
case ISD::RET: return LowerRET(Op, DAG, getTargetMachine());
case ISD::STACKRESTORE: return LowerSTACKRESTORE(Op, DAG, PPCSubTarget);
case ISD::DYNAMIC_STACKALLOC:
return LowerDYNAMIC_STACKALLOC(Op, DAG, PPCSubTarget);
+
+ case ISD::ATOMIC_LOAD_ADD: return LowerAtomicLOAD_ADD(Op, DAG);
+ case ISD::ATOMIC_CMP_SWAP: return LowerAtomicCMP_SWAP(Op, DAG);
+ case ISD::ATOMIC_SWAP: return LowerAtomicSWAP(Op, DAG);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
case ISD::SINT_TO_FP: return LowerSINT_TO_FP(Op, DAG);
case ISD::FP_ROUND_INREG: return LowerFP_ROUND_INREG(Op, DAG);
- case ISD::FLT_ROUNDS: return LowerFLT_ROUNDS(Op, DAG);
+ case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
// Lower 64-bit shifts.
case ISD::SHL_PARTS: return LowerSHL_PARTS(Op, DAG);
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
}
- return SDOperand();
+ return SDValue();
}
-SDNode *PPCTargetLowering::ExpandOperationResult(SDNode *N, SelectionDAG &DAG) {
+SDNode *PPCTargetLowering::ReplaceNodeResults(SDNode *N, SelectionDAG &DAG) {
switch (N->getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
- case ISD::FP_TO_SINT: return LowerFP_TO_SINT(SDOperand(N, 0), DAG).Val;
+ case ISD::FP_TO_SINT: {
+ SDValue Res = LowerFP_TO_SINT(SDValue(N, 0), DAG);
+ // Use MERGE_VALUES to drop the chain result value and get a node with one
+ // result. This requires turning off getMergeValues simplification, since
+ // otherwise it will give us Res back.
+ return DAG.getMergeValues(&Res, 1, false).Val;
+ }
}
}
//===----------------------------------------------------------------------===//
MachineBasicBlock *
-PPCTargetLowering::InsertAtEndOfBasicBlock(MachineInstr *MI,
- MachineBasicBlock *BB) {
+PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
+ MachineBasicBlock *BB) {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
- assert((MI->getOpcode() == PPC::SELECT_CC_I4 ||
- MI->getOpcode() == PPC::SELECT_CC_I8 ||
- MI->getOpcode() == PPC::SELECT_CC_F4 ||
- MI->getOpcode() == PPC::SELECT_CC_F8 ||
- MI->getOpcode() == PPC::SELECT_CC_VRRC) &&
- "Unexpected instr type to insert");
-
- // To "insert" a SELECT_CC instruction, we actually have to insert the diamond
- // control-flow pattern. The incoming instruction knows the destination vreg
- // to set, the condition code register to branch on, the true/false values to
- // select between, and a branch opcode to use.
+
+ // To "insert" these instructions we actually have to insert their
+ // control-flow patterns.
const BasicBlock *LLVM_BB = BB->getBasicBlock();
- ilist<MachineBasicBlock>::iterator It = BB;
+ MachineFunction::iterator It = BB;
++It;
-
- // thisMBB:
- // ...
- // TrueVal = ...
- // cmpTY ccX, r1, r2
- // bCC copy1MBB
- // fallthrough --> copy0MBB
- MachineBasicBlock *thisMBB = BB;
- MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB);
- MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB);
- unsigned SelectPred = MI->getOperand(4).getImm();
- BuildMI(BB, TII->get(PPC::BCC))
- .addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB);
+
MachineFunction *F = BB->getParent();
- F->getBasicBlockList().insert(It, copy0MBB);
- F->getBasicBlockList().insert(It, sinkMBB);
- // Update machine-CFG edges by first adding all successors of the current
- // block to the new block which will contain the Phi node for the select.
- for(MachineBasicBlock::succ_iterator i = BB->succ_begin(),
- e = BB->succ_end(); i != e; ++i)
- sinkMBB->addSuccessor(*i);
- // Next, remove all successors of the current block, and add the true
- // and fallthrough blocks as its successors.
- while(!BB->succ_empty())
- BB->removeSuccessor(BB->succ_begin());
- BB->addSuccessor(copy0MBB);
- BB->addSuccessor(sinkMBB);
-
- // copy0MBB:
- // %FalseValue = ...
- // # fallthrough to sinkMBB
- BB = copy0MBB;
-
- // Update machine-CFG edges
- BB->addSuccessor(sinkMBB);
-
- // sinkMBB:
- // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
- // ...
- BB = sinkMBB;
- BuildMI(BB, TII->get(PPC::PHI), MI->getOperand(0).getReg())
- .addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB)
- .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
-
- delete MI; // The pseudo instruction is gone now.
+
+ if (MI->getOpcode() == PPC::SELECT_CC_I4 ||
+ MI->getOpcode() == PPC::SELECT_CC_I8 ||
+ MI->getOpcode() == PPC::SELECT_CC_F4 ||
+ MI->getOpcode() == PPC::SELECT_CC_F8 ||
+ MI->getOpcode() == PPC::SELECT_CC_VRRC) {
+
+ // The incoming instruction knows the destination vreg to set, the
+ // condition code register to branch on, the true/false values to
+ // select between, and a branch opcode to use.
+
+ // thisMBB:
+ // ...
+ // TrueVal = ...
+ // cmpTY ccX, r1, r2
+ // bCC copy1MBB
+ // fallthrough --> copy0MBB
+ MachineBasicBlock *thisMBB = BB;
+ MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ unsigned SelectPred = MI->getOperand(4).getImm();
+ BuildMI(BB, TII->get(PPC::BCC))
+ .addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB);
+ F->insert(It, copy0MBB);
+ F->insert(It, sinkMBB);
+ // Update machine-CFG edges by transferring all successors of the current
+ // block to the new block which will contain the Phi node for the select.
+ sinkMBB->transferSuccessors(BB);
+ // Next, add the true and fallthrough blocks as its successors.
+ BB->addSuccessor(copy0MBB);
+ BB->addSuccessor(sinkMBB);
+
+ // copy0MBB:
+ // %FalseValue = ...
+ // # fallthrough to sinkMBB
+ BB = copy0MBB;
+
+ // Update machine-CFG edges
+ BB->addSuccessor(sinkMBB);
+
+ // sinkMBB:
+ // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
+ // ...
+ BB = sinkMBB;
+ BuildMI(BB, TII->get(PPC::PHI), MI->getOperand(0).getReg())
+ .addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB)
+ .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
+ }
+ else if (MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I32 ||
+ MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I64) {
+ bool is64bit = MI->getOpcode() == PPC::ATOMIC_LOAD_ADD_I64;
+
+ unsigned dest = MI->getOperand(0).getReg();
+ unsigned ptrA = MI->getOperand(1).getReg();
+ unsigned ptrB = MI->getOperand(2).getReg();
+ unsigned incr = MI->getOperand(3).getReg();
+
+ MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ F->insert(It, loopMBB);
+ F->insert(It, exitMBB);
+ exitMBB->transferSuccessors(BB);
+
+ MachineRegisterInfo &RegInfo = F->getRegInfo();
+ unsigned TmpReg = RegInfo.createVirtualRegister(
+ is64bit ? (const TargetRegisterClass *) &PPC::GPRCRegClass :
+ (const TargetRegisterClass *) &PPC::G8RCRegClass);
+
+ // thisMBB:
+ // ...
+ // fallthrough --> loopMBB
+ BB->addSuccessor(loopMBB);
+
+ // loopMBB:
+ // l[wd]arx dest, ptr
+ // add r0, dest, incr
+ // st[wd]cx. r0, ptr
+ // bne- loopMBB
+ // fallthrough --> exitMBB
+ BB = loopMBB;
+ BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
+ .addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(is64bit ? PPC::ADD4 : PPC::ADD8), TmpReg)
+ .addReg(incr).addReg(dest);
+ BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ .addReg(TmpReg).addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
+ BB->addSuccessor(loopMBB);
+ BB->addSuccessor(exitMBB);
+
+ // exitMBB:
+ // ...
+ BB = exitMBB;
+ }
+ else if (MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I32 ||
+ MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I64) {
+ bool is64bit = MI->getOpcode() == PPC::ATOMIC_CMP_SWAP_I64;
+
+ unsigned dest = MI->getOperand(0).getReg();
+ unsigned ptrA = MI->getOperand(1).getReg();
+ unsigned ptrB = MI->getOperand(2).getReg();
+ unsigned oldval = MI->getOperand(3).getReg();
+ unsigned newval = MI->getOperand(4).getReg();
+
+ MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ F->insert(It, loopMBB);
+ F->insert(It, exitMBB);
+ exitMBB->transferSuccessors(BB);
+
+ // thisMBB:
+ // ...
+ // fallthrough --> loopMBB
+ BB->addSuccessor(loopMBB);
+
+ // loopMBB:
+ // l[wd]arx dest, ptr
+ // cmp[wd] dest, oldval
+ // bne- exitMBB
+ // st[wd]cx. newval, ptr
+ // bne- loopMBB
+ // fallthrough --> exitMBB
+ BB = loopMBB;
+ BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
+ .addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(is64bit ? PPC::CMPD : PPC::CMPW), PPC::CR0)
+ .addReg(oldval).addReg(dest);
+ BuildMI(BB, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(exitMBB);
+ BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ .addReg(newval).addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
+ BB->addSuccessor(loopMBB);
+ BB->addSuccessor(exitMBB);
+
+ // exitMBB:
+ // ...
+ BB = exitMBB;
+ }
+ else if (MI->getOpcode() == PPC::ATOMIC_SWAP_I32 ||
+ MI->getOpcode() == PPC::ATOMIC_SWAP_I64) {
+ bool is64bit = MI->getOpcode() == PPC::ATOMIC_SWAP_I64;
+
+ unsigned dest = MI->getOperand(0).getReg();
+ unsigned ptrA = MI->getOperand(1).getReg();
+ unsigned ptrB = MI->getOperand(2).getReg();
+ unsigned newval = MI->getOperand(3).getReg();
+
+ MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ F->insert(It, loopMBB);
+ F->insert(It, exitMBB);
+ exitMBB->transferSuccessors(BB);
+
+ // thisMBB:
+ // ...
+ // fallthrough --> loopMBB
+ BB->addSuccessor(loopMBB);
+
+ // loopMBB:
+ // l[wd]arx dest, ptr
+ // st[wd]cx. newval, ptr
+ // bne- loopMBB
+ // fallthrough --> exitMBB
+ BB = loopMBB;
+ BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
+ .addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ .addReg(newval).addReg(ptrA).addReg(ptrB);
+ BuildMI(BB, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
+ BB->addSuccessor(loopMBB);
+ BB->addSuccessor(exitMBB);
+
+ // exitMBB:
+ // ...
+ BB = exitMBB;
+ }
+ else {
+ assert(0 && "Unexpected instr type to insert");
+ }
+
+ F->DeleteMachineInstr(MI); // The pseudo instruction is gone now.
return BB;
}
// Target Optimization Hooks
//===----------------------------------------------------------------------===//
-SDOperand PPCTargetLowering::PerformDAGCombine(SDNode *N,
+SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
TargetMachine &TM = getTargetMachine();
SelectionDAG &DAG = DCI.DAG;
// type must be i64.
if (N->getOperand(0).getValueType() == MVT::i64 &&
N->getOperand(0).getOperand(0).getValueType() != MVT::ppcf128) {
- SDOperand Val = N->getOperand(0).getOperand(0);
+ SDValue Val = N->getOperand(0).getOperand(0);
if (Val.getValueType() == MVT::f32) {
Val = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
N->getOperand(1).getOpcode() == ISD::FP_TO_SINT &&
N->getOperand(1).getValueType() == MVT::i32 &&
N->getOperand(1).getOperand(0).getValueType() != MVT::ppcf128) {
- SDOperand Val = N->getOperand(1).getOperand(0);
+ SDValue Val = N->getOperand(1).getOperand(0);
if (Val.getValueType() == MVT::f32) {
Val = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
N->getOperand(1).Val->hasOneUse() &&
(N->getOperand(1).getValueType() == MVT::i32 ||
N->getOperand(1).getValueType() == MVT::i16)) {
- SDOperand BSwapOp = N->getOperand(1).getOperand(0);
+ SDValue BSwapOp = N->getOperand(1).getOperand(0);
// Do an any-extend to 32-bits if this is a half-word input.
if (BSwapOp.getValueType() == MVT::i16)
BSwapOp = DAG.getNode(ISD::ANY_EXTEND, MVT::i32, BSwapOp);
if (ISD::isNON_EXTLoad(N->getOperand(0).Val) &&
N->getOperand(0).hasOneUse() &&
(N->getValueType(0) == MVT::i32 || N->getValueType(0) == MVT::i16)) {
- SDOperand Load = N->getOperand(0);
+ SDValue Load = N->getOperand(0);
LoadSDNode *LD = cast<LoadSDNode>(Load);
// Create the byte-swapping load.
- std::vector<MVT::ValueType> VTs;
+ std::vector<MVT> VTs;
VTs.push_back(MVT::i32);
VTs.push_back(MVT::Other);
- SDOperand SV = DAG.getSrcValue(LD->getSrcValue(), LD->getSrcValueOffset());
- SDOperand Ops[] = {
+ SDValue MO = DAG.getMemOperand(LD->getMemOperand());
+ SDValue Ops[] = {
LD->getChain(), // Chain
LD->getBasePtr(), // Ptr
- SV, // SrcValue
+ MO, // MemOperand
DAG.getValueType(N->getValueType(0)) // VT
};
- SDOperand BSLoad = DAG.getNode(PPCISD::LBRX, VTs, Ops, 4);
+ SDValue BSLoad = DAG.getNode(PPCISD::LBRX, VTs, Ops, 4);
// If this is an i16 load, insert the truncate.
- SDOperand ResVal = BSLoad;
+ SDValue ResVal = BSLoad;
if (N->getValueType(0) == MVT::i16)
ResVal = DAG.getNode(ISD::TRUNCATE, MVT::i16, BSLoad);
DCI.CombineTo(Load.Val, ResVal, BSLoad.getValue(1));
// Return N so it doesn't get rechecked!
- return SDOperand(N, 0);
+ return SDValue(N, 0);
}
break;
SDNode *LHSN = N->getOperand(0).Val;
for (SDNode::use_iterator UI = LHSN->use_begin(), E = LHSN->use_end();
UI != E; ++UI)
- if ((*UI)->getOpcode() == PPCISD::VCMPo &&
- (*UI)->getOperand(1) == N->getOperand(1) &&
- (*UI)->getOperand(2) == N->getOperand(2) &&
- (*UI)->getOperand(0) == N->getOperand(0)) {
+ if (UI->getOpcode() == PPCISD::VCMPo &&
+ UI->getOperand(1) == N->getOperand(1) &&
+ UI->getOperand(2) == N->getOperand(2) &&
+ UI->getOperand(0) == N->getOperand(0)) {
VCMPoNode = *UI;
break;
}
assert(UI != VCMPoNode->use_end() && "Didn't find user!");
SDNode *User = *UI;
for (unsigned i = 0, e = User->getNumOperands(); i != e; ++i) {
- if (User->getOperand(i) == SDOperand(VCMPoNode, 1)) {
+ if (User->getOperand(i) == SDValue(VCMPoNode, 1)) {
FlagUser = User;
break;
}
// If the user is a MFCR instruction, we know this is safe. Otherwise we
// give up for right now.
if (FlagUser->getOpcode() == PPCISD::MFCR)
- return SDOperand(VCMPoNode, 0);
+ return SDValue(VCMPoNode, 0);
}
break;
}
// lowering is done pre-legalize, because the legalizer lowers the predicate
// compare down to code that is difficult to reassemble.
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
- SDOperand LHS = N->getOperand(2), RHS = N->getOperand(3);
+ SDValue LHS = N->getOperand(2), RHS = N->getOperand(3);
int CompareOpc;
bool isDot;
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
// Create the PPCISD altivec 'dot' comparison node.
- std::vector<MVT::ValueType> VTs;
- SDOperand Ops[] = {
+ std::vector<MVT> VTs;
+ SDValue Ops[] = {
LHS.getOperand(2), // LHS of compare
LHS.getOperand(3), // RHS of compare
DAG.getConstant(CompareOpc, MVT::i32)
};
VTs.push_back(LHS.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
- SDOperand CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
+ SDValue CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
// Unpack the result based on how the target uses it.
PPC::Predicate CompOpc;
}
}
- return SDOperand();
+ return SDValue();
}
//===----------------------------------------------------------------------===//
// Inline Assembly Support
//===----------------------------------------------------------------------===//
-void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDOperand Op,
- uint64_t Mask,
- uint64_t &KnownZero,
- uint64_t &KnownOne,
+void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
+ const APInt &Mask,
+ APInt &KnownZero,
+ APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
- KnownZero = 0;
- KnownOne = 0;
+ KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
switch (Op.getOpcode()) {
default: break;
case PPCISD::LBRX: {
std::pair<unsigned, const TargetRegisterClass*>
PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
- MVT::ValueType VT) const {
+ MVT VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector. If it is invalid, don't add anything to Ops.
-void PPCTargetLowering::LowerAsmOperandForConstraint(SDOperand Op, char Letter,
- std::vector<SDOperand>&Ops,
- SelectionDAG &DAG) {
- SDOperand Result(0,0);
+void PPCTargetLowering::LowerAsmOperandForConstraint(SDValue Op, char Letter,
+ std::vector<SDValue>&Ops,
+ SelectionDAG &DAG) const {
+ SDValue Result(0,0);
switch (Letter) {
default: break;
case 'I':
return false;
}
-SDOperand PPCTargetLowering::LowerRETURNADDR(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) {
// Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getValue() > 0)
- return SDOperand();
+ return SDValue();
MachineFunction &MF = DAG.getMachineFunction();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
- int RAIdx = FuncInfo->getReturnAddrSaveIndex();
- if (RAIdx == 0) {
- bool isPPC64 = PPCSubTarget.isPPC64();
- int Offset =
- PPCFrameInfo::getReturnSaveOffset(isPPC64, PPCSubTarget.isMachoABI());
- // Set up a frame object for the return address.
- RAIdx = MF.getFrameInfo()->CreateFixedObject(isPPC64 ? 8 : 4, Offset);
-
- // Remember it for next time.
- FuncInfo->setReturnAddrSaveIndex(RAIdx);
-
- // Make sure the function really does not optimize away the store of the RA
- // to the stack.
- FuncInfo->setLRStoreRequired();
- }
-
// Just load the return address off the stack.
- SDOperand RetAddrFI = DAG.getFrameIndex(RAIdx, getPointerTy());
+ SDValue RetAddrFI = getReturnAddrFrameIndex(DAG);
+
+ // Make sure the function really does not optimize away the store of the RA
+ // to the stack.
+ FuncInfo->setLRStoreRequired();
return DAG.getLoad(getPointerTy(), DAG.getEntryNode(), RetAddrFI, NULL, 0);
}
-SDOperand PPCTargetLowering::LowerFRAMEADDR(SDOperand Op, SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) {
// Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getValue() > 0)
- return SDOperand();
+ return SDValue();
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
MachineFunction &MF = DAG.getMachineFunction();
projects / oota-llvm.git / blobdiff