#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
+#include "llvm/DerivedTypes.h"
using namespace llvm;
-static cl::opt<bool> EnablePPCPreinc("enable-ppc-preinc",
+static cl::opt<bool> EnablePPCPreinc("enable-ppc-preinc",
cl::desc("enable preincrement load/store generation on PPC (experimental)"),
cl::Hidden);
PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
: TargetLowering(TM), PPCSubTarget(*TM.getSubtargetImpl()) {
-
+
setPow2DivIsCheap();
// Use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(true);
setUseUnderscoreLongJmp(true);
-
+
// Set up the register classes.
addRegisterClass(MVT::i32, PPC::GPRCRegisterClass);
addRegisterClass(MVT::f32, PPC::F4RCRegisterClass);
addRegisterClass(MVT::f64, PPC::F8RCRegisterClass);
-
+
// PowerPC has an i16 but no i8 (or i1) SEXTLOAD
setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
-
+
// PowerPC has pre-inc load and store's.
setIndexedLoadAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i8, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i64, Legal);
- // Shortening conversions involving ppcf128 get expanded (2 regs -> 1 reg)
- setConvertAction(MVT::ppcf128, MVT::f64, Expand);
- setConvertAction(MVT::ppcf128, MVT::f32, Expand);
// This is used in the ppcf128->int sequence. Note it has different semantics
// from FP_ROUND: that rounds to nearest, this rounds to zero.
setOperationAction(ISD::FP_ROUND_INREG, MVT::ppcf128, Custom);
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
-
+
// We don't support sin/cos/sqrt/fmod/pow
setOperationAction(ISD::FSIN , MVT::f64, Expand);
setOperationAction(ISD::FCOS , MVT::f64, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
-
+
// If we're enabling GP optimizations, use hardware square root
if (!TM.getSubtarget<PPCSubtarget>().hasFSQRT()) {
setOperationAction(ISD::FSQRT, MVT::f64, Expand);
setOperationAction(ISD::FSQRT, MVT::f32, Expand);
}
-
+
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
-
+
// PowerPC does not have BSWAP, CTPOP or CTTZ
setOperationAction(ISD::BSWAP, MVT::i32 , Expand);
setOperationAction(ISD::CTPOP, MVT::i32 , Expand);
setOperationAction(ISD::BSWAP, MVT::i64 , Expand);
setOperationAction(ISD::CTPOP, MVT::i64 , Expand);
setOperationAction(ISD::CTTZ , MVT::i64 , Expand);
-
+
// PowerPC does not have ROTR
setOperationAction(ISD::ROTR, MVT::i32 , Expand);
setOperationAction(ISD::ROTR, MVT::i64 , Expand);
-
+
// PowerPC does not have Select
setOperationAction(ISD::SELECT, MVT::i32, Expand);
setOperationAction(ISD::SELECT, MVT::i64, Expand);
setOperationAction(ISD::SELECT, MVT::f32, Expand);
setOperationAction(ISD::SELECT, MVT::f64, Expand);
-
+
// PowerPC wants to turn select_cc of FP into fsel when possible.
setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
// PowerPC wants to optimize integer setcc a bit
setOperationAction(ISD::SETCC, MVT::i32, Custom);
-
+
// PowerPC does not have BRCOND which requires SetCC
setOperationAction(ISD::BRCOND, MVT::Other, Expand);
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
-
+
// PowerPC turns FP_TO_SINT into FCTIWZ and some load/stores.
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
// Support label based line numbers.
setOperationAction(ISD::DBG_STOPPOINT, MVT::Other, Expand);
setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
-
+
setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
setOperationAction(ISD::EHSELECTION, MVT::i64, Expand);
setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
setOperationAction(ISD::EHSELECTION, MVT::i32, Expand);
-
-
- // We want to legalize GlobalAddress and ConstantPool nodes into the
+
+
+ // We want to legalize GlobalAddress and ConstantPool nodes into the
// appropriate instructions to materialize the address.
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
setOperationAction(ISD::JumpTable, MVT::i64, Custom);
-
+
// RET must be custom lowered, to meet ABI requirements.
setOperationAction(ISD::RET , MVT::Other, Custom);
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
-
+
// VAARG is custom lowered with ELF 32 ABI
if (TM.getSubtarget<PPCSubtarget>().isELF32_ABI())
setOperationAction(ISD::VAARG, MVT::Other, Custom);
else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
-
+
// Use the default implementation.
setOperationAction(ISD::VACOPY , MVT::Other, Expand);
setOperationAction(ISD::VAEND , MVT::Other, Expand);
- setOperationAction(ISD::STACKSAVE , MVT::Other, Expand);
+ setOperationAction(ISD::STACKSAVE , MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE , MVT::Other, Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Custom);
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
-
+
// Comparisons that require checking two conditions.
setCondCodeAction(ISD::SETULT, MVT::f32, Expand);
setCondCodeAction(ISD::SETULT, MVT::f64, Expand);
setCondCodeAction(ISD::SETOLE, MVT::f64, Expand);
setCondCodeAction(ISD::SETONE, MVT::f32, Expand);
setCondCodeAction(ISD::SETONE, MVT::f64, Expand);
-
+
if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
// They also have instructions for converting between i64 and fp.
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
-
+ // This is just the low 32 bits of a (signed) fp->i64 conversion.
+ // We cannot do this with Promote because i64 is not a legal type.
+ setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
+
// FIXME: disable this lowered code. This generates 64-bit register values,
// and we don't model the fact that the top part is clobbered by calls. We
// need to flag these together so that the value isn't live across a call.
//setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
-
- // To take advantage of the above i64 FP_TO_SINT, promote i32 FP_TO_UINT
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Promote);
} else {
// PowerPC does not have FP_TO_UINT on 32-bit implementations.
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
// add/sub are legal for all supported vector VT's.
setOperationAction(ISD::ADD , VT, Legal);
setOperationAction(ISD::SUB , VT, Legal);
-
+
// We promote all shuffles to v16i8.
setOperationAction(ISD::VECTOR_SHUFFLE, VT, Promote);
AddPromotedToType (ISD::VECTOR_SHUFFLE, VT, MVT::v16i8);
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 , VT, Expand);
setOperationAction(ISD::SDIV, VT, Expand);
setOperationAction(ISD::LOAD , MVT::v4i32, Legal);
setOperationAction(ISD::SELECT, MVT::v4i32, Expand);
setOperationAction(ISD::STORE , MVT::v4i32, Legal);
-
+
addRegisterClass(MVT::v4f32, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v4i32, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v8i16, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v16i8, PPC::VRRCRegisterClass);
-
+
setOperationAction(ISD::MUL, MVT::v4f32, Legal);
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::MUL, MVT::v8i16, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4i32, Custom);
-
+
setOperationAction(ISD::BUILD_VECTOR, MVT::v16i8, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v8i16, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
-
+
setShiftAmountType(MVT::i32);
setBooleanContents(ZeroOrOneBooleanContent);
-
+
if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
setStackPointerRegisterToSaveRestore(PPC::X1);
setExceptionPointerRegister(PPC::X3);
setExceptionPointerRegister(PPC::R3);
setExceptionSelectorRegister(PPC::R4);
}
-
+
// We have target-specific dag combine patterns for the following nodes:
setTargetDAGCombine(ISD::SINT_TO_FP);
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::BR_CC);
setTargetDAGCombine(ISD::BSWAP);
-
+
// Darwin long double math library functions have $LDBL128 appended.
if (TM.getSubtarget<PPCSubtarget>().isDarwin()) {
setLibcallName(RTLIB::COS_PPCF128, "cosl$LDBL128");
/// 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(SDValue Op, unsigned Val) {
- return Op.getOpcode() == ISD::UNDEF ||
- cast<ConstantSDNode>(Op)->getZExtValue() == Val;
+static bool isConstantOrUndef(int Op, int Val) {
+ return Op < 0 || Op == Val;
}
/// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUHUM instruction.
-bool PPC::isVPKUHUMShuffleMask(SDNode *N, bool isUnary) {
+bool PPC::isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, bool isUnary) {
if (!isUnary) {
for (unsigned i = 0; i != 16; ++i)
- if (!isConstantOrUndef(N->getOperand(i), i*2+1))
+ if (!isConstantOrUndef(N->getMaskElt(i), i*2+1))
return false;
} else {
for (unsigned i = 0; i != 8; ++i)
- if (!isConstantOrUndef(N->getOperand(i), i*2+1) ||
- !isConstantOrUndef(N->getOperand(i+8), i*2+1))
+ if (!isConstantOrUndef(N->getMaskElt(i), i*2+1) ||
+ !isConstantOrUndef(N->getMaskElt(i+8), i*2+1))
return false;
}
return true;
/// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUWUM instruction.
-bool PPC::isVPKUWUMShuffleMask(SDNode *N, bool isUnary) {
+bool PPC::isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, bool isUnary) {
if (!isUnary) {
for (unsigned i = 0; i != 16; i += 2)
- if (!isConstantOrUndef(N->getOperand(i ), i*2+2) ||
- !isConstantOrUndef(N->getOperand(i+1), i*2+3))
+ if (!isConstantOrUndef(N->getMaskElt(i ), i*2+2) ||
+ !isConstantOrUndef(N->getMaskElt(i+1), i*2+3))
return false;
} else {
for (unsigned i = 0; i != 8; i += 2)
- if (!isConstantOrUndef(N->getOperand(i ), i*2+2) ||
- !isConstantOrUndef(N->getOperand(i+1), i*2+3) ||
- !isConstantOrUndef(N->getOperand(i+8), i*2+2) ||
- !isConstantOrUndef(N->getOperand(i+9), i*2+3))
+ if (!isConstantOrUndef(N->getMaskElt(i ), i*2+2) ||
+ !isConstantOrUndef(N->getMaskElt(i+1), i*2+3) ||
+ !isConstantOrUndef(N->getMaskElt(i+8), i*2+2) ||
+ !isConstantOrUndef(N->getMaskElt(i+9), i*2+3))
return false;
}
return true;
/// isVMerge - Common function, used to match vmrg* shuffles.
///
-static bool isVMerge(SDNode *N, unsigned UnitSize,
+static bool isVMerge(ShuffleVectorSDNode *N, unsigned UnitSize,
unsigned LHSStart, unsigned RHSStart) {
- assert(N->getOpcode() == ISD::BUILD_VECTOR &&
- N->getNumOperands() == 16 && "PPC only supports shuffles by bytes!");
+ assert(N->getValueType(0) == MVT::v16i8 &&
+ "PPC only supports shuffles by bytes!");
assert((UnitSize == 1 || UnitSize == 2 || UnitSize == 4) &&
"Unsupported merge size!");
-
+
for (unsigned i = 0; i != 8/UnitSize; ++i) // Step over units
for (unsigned j = 0; j != UnitSize; ++j) { // Step over bytes within unit
- if (!isConstantOrUndef(N->getOperand(i*UnitSize*2+j),
+ if (!isConstantOrUndef(N->getMaskElt(i*UnitSize*2+j),
LHSStart+j+i*UnitSize) ||
- !isConstantOrUndef(N->getOperand(i*UnitSize*2+UnitSize+j),
+ !isConstantOrUndef(N->getMaskElt(i*UnitSize*2+UnitSize+j),
RHSStart+j+i*UnitSize))
return false;
}
- return true;
+ return true;
}
/// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
/// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
-bool PPC::isVMRGLShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary) {
+bool PPC::isVMRGLShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
+ bool isUnary) {
if (!isUnary)
return isVMerge(N, UnitSize, 8, 24);
return isVMerge(N, UnitSize, 8, 8);
/// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
/// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
-bool PPC::isVMRGHShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary) {
+bool PPC::isVMRGHShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
+ bool isUnary) {
if (!isUnary)
return isVMerge(N, UnitSize, 0, 16);
return isVMerge(N, UnitSize, 0, 0);
/// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
/// amount, otherwise return -1.
int PPC::isVSLDOIShuffleMask(SDNode *N, bool isUnary) {
- assert(N->getOpcode() == ISD::BUILD_VECTOR &&
- N->getNumOperands() == 16 && "PPC only supports shuffles by bytes!");
+ assert(N->getValueType(0) == MVT::v16i8 &&
+ "PPC only supports shuffles by bytes!");
+
+ ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(N);
+
// Find the first non-undef value in the shuffle mask.
unsigned i;
- for (i = 0; i != 16 && N->getOperand(i).getOpcode() == ISD::UNDEF; ++i)
+ for (i = 0; i != 16 && SVOp->getMaskElt(i) < 0; ++i)
/*search*/;
-
+
if (i == 16) return -1; // all undef.
-
- // Otherwise, check to see if the rest of the elements are consequtively
+
+ // Otherwise, check to see if the rest of the elements are consecutively
// numbered from this value.
- unsigned ShiftAmt = cast<ConstantSDNode>(N->getOperand(i))->getZExtValue();
+ unsigned ShiftAmt = SVOp->getMaskElt(i);
if (ShiftAmt < i) return -1;
ShiftAmt -= i;
if (!isUnary) {
- // Check the rest of the elements to see if they are consequtive.
+ // Check the rest of the elements to see if they are consecutive.
for (++i; i != 16; ++i)
- if (!isConstantOrUndef(N->getOperand(i), ShiftAmt+i))
+ if (!isConstantOrUndef(SVOp->getMaskElt(i), ShiftAmt+i))
return -1;
} else {
- // Check the rest of the elements to see if they are consequtive.
+ // Check the rest of the elements to see if they are consecutive.
for (++i; i != 16; ++i)
- if (!isConstantOrUndef(N->getOperand(i), (ShiftAmt+i) & 15))
+ if (!isConstantOrUndef(SVOp->getMaskElt(i), (ShiftAmt+i) & 15))
return -1;
}
-
return ShiftAmt;
}
/// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
/// specifies a splat of a single element that is suitable for input to
/// VSPLTB/VSPLTH/VSPLTW.
-bool PPC::isSplatShuffleMask(SDNode *N, unsigned EltSize) {
- assert(N->getOpcode() == ISD::BUILD_VECTOR &&
- N->getNumOperands() == 16 &&
+bool PPC::isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize) {
+ assert(N->getValueType(0) == MVT::v16i8 &&
(EltSize == 1 || EltSize == 2 || EltSize == 4));
-
+
// 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;
- SDValue Elt = N->getOperand(0);
- if (ConstantSDNode *EltV = dyn_cast<ConstantSDNode>(Elt))
- ElementBase = EltV->getZExtValue();
- else
- return false; // FIXME: Handle UNDEF elements too!
-
- if (cast<ConstantSDNode>(Elt)->getZExtValue() >= 16)
- return false;
+ unsigned ElementBase = N->getMaskElt(0);
- // Check that they are consequtive.
- for (unsigned i = 1; i != EltSize; ++i) {
- if (!isa<ConstantSDNode>(N->getOperand(i)) ||
- cast<ConstantSDNode>(N->getOperand(i))->getZExtValue() != i+ElementBase)
+ // FIXME: Handle UNDEF elements too!
+ if (ElementBase >= 16)
+ return false;
+
+ // Check that the indices are consecutive, in the case of a multi-byte element
+ // splatted with a v16i8 mask.
+ for (unsigned i = 1; i != EltSize; ++i)
+ if (N->getMaskElt(i) < 0 || N->getMaskElt(i) != (int)(i+ElementBase))
return false;
- }
-
- assert(isa<ConstantSDNode>(Elt) && "Invalid VECTOR_SHUFFLE mask!");
+
for (unsigned i = EltSize, e = 16; i != e; i += EltSize) {
- if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
- assert(isa<ConstantSDNode>(N->getOperand(i)) &&
- "Invalid VECTOR_SHUFFLE mask!");
+ if (N->getMaskElt(i) < 0) continue;
for (unsigned j = 0; j != EltSize; ++j)
- if (N->getOperand(i+j) != N->getOperand(j))
+ if (N->getMaskElt(i+j) != N->getMaskElt(j))
return false;
}
-
return true;
}
/// isAllNegativeZeroVector - Returns true if all elements of build_vector
/// are -0.0.
bool PPC::isAllNegativeZeroVector(SDNode *N) {
- assert(N->getOpcode() == ISD::BUILD_VECTOR);
- if (PPC::isSplatShuffleMask(N, N->getNumOperands()))
- if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N))
+ BuildVectorSDNode *BV = cast<BuildVectorSDNode>(N);
+
+ APInt APVal, APUndef;
+ unsigned BitSize;
+ bool HasAnyUndefs;
+
+ if (BV->isConstantSplat(APVal, APUndef, BitSize, HasAnyUndefs, 32))
+ if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N->getOperand(0)))
return CFP->getValueAPF().isNegZero();
+
return false;
}
/// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
/// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
unsigned PPC::getVSPLTImmediate(SDNode *N, unsigned EltSize) {
- assert(isSplatShuffleMask(N, EltSize));
- return cast<ConstantSDNode>(N->getOperand(0))->getZExtValue() / EltSize;
+ ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(N);
+ assert(isSplatShuffleMask(SVOp, EltSize));
+ return SVOp->getMaskElt(0) / EltSize;
}
/// get_VSPLTI_elt - If this is a build_vector of constants which can be formed
unsigned Multiple = ByteSize/EltSize; // Number of BV entries per spltval.
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 SDValue();
-
+
if (UniquedVals[i&(Multiple-1)].getNode() == 0)
UniquedVals[i&(Multiple-1)] = N->getOperand(i);
else if (UniquedVals[i&(Multiple-1)] != N->getOperand(i))
return SDValue(); // no match.
}
-
+
// Okay, if we reached this point, UniquedVals[0..Multiple-1] contains
// either constant or undef values that are identical for each chunk. See
// if these chunks can form into a larger vspltis*.
-
+
// Check to see if all of the leading entries are either 0 or -1. If
// neither, then this won't fit into the immediate field.
bool LeadingZero = true;
bool LeadingOnes = true;
for (unsigned i = 0; i != Multiple-1; ++i) {
if (UniquedVals[i].getNode() == 0) continue; // Must have been undefs.
-
+
LeadingZero &= cast<ConstantSDNode>(UniquedVals[i])->isNullValue();
LeadingOnes &= cast<ConstantSDNode>(UniquedVals[i])->isAllOnesValue();
}
if (Val >= -16) // -1,-1,-1,-2 -> vspltisw(-2)
return DAG.getTargetConstant(Val, MVT::i32);
}
-
+
return SDValue();
}
-
+
// Check to see if this buildvec has a single non-undef value in its elements.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
else if (OpVal != N->getOperand(i))
return SDValue();
}
-
+
if (OpVal.getNode() == 0) return SDValue(); // All UNDEF: use implicit def.
-
- unsigned ValSizeInBytes = 0;
+
+ unsigned ValSizeInBytes = EltSize;
uint64_t Value = 0;
if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
Value = CN->getZExtValue();
- 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());
- ValSizeInBytes = 4;
}
// 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 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
// get to ByteSize. This allows us to handle 0x01010101 as 0x01.
while (ValSizeInBytes > ByteSize) {
ValSizeInBytes >>= 1;
-
+
// 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)))
// Properly sign extend the value.
int ShAmt = (4-ByteSize)*8;
int MaskVal = ((int)Value << ShAmt) >> ShAmt;
-
+
// If this is zero, don't match, zero matches ISD::isBuildVectorAllZeros.
if (MaskVal == 0) return SDValue();
static bool isIntS16Immediate(SDNode *N, short &Imm) {
if (N->getOpcode() != ISD::Constant)
return false;
-
+
Imm = (short)cast<ConstantSDNode>(N)->getZExtValue();
if (N->getValueType(0) == MVT::i32)
return Imm == (int32_t)cast<ConstantSDNode>(N)->getZExtValue();
return false; // r+i
if (N.getOperand(1).getOpcode() == PPCISD::Lo)
return false; // r+i
-
+
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
} else if (N.getOpcode() == ISD::OR) {
if (isIntS16Immediate(N.getOperand(1), imm))
return false; // r+i can fold it if we can.
-
+
// 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.
APInt::getAllOnesValue(N.getOperand(0)
.getValueSizeInBits()),
LHSKnownZero, LHSKnownOne);
-
+
if (LHSKnownZero.getBoolValue()) {
DAG.ComputeMaskedBits(N.getOperand(1),
APInt::getAllOnesValue(N.getOperand(1)
}
}
}
-
+
return false;
}
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
-
+
if (N.getOpcode() == ISD::ADD) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm)) {
}
} else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N)) {
// Loading from a constant address.
-
+
// If this address fits entirely in a 16-bit sext immediate field, codegen
// this as "d, 0"
short Imm;
if (CN->getValueType(0) == MVT::i32 ||
(int64_t)CN->getZExtValue() == (int)CN->getZExtValue()) {
int Addr = (int)CN->getZExtValue();
-
+
// Otherwise, break this down into an LIS + disp.
Disp = DAG.getTargetConstant((short)Addr, MVT::i32);
-
+
Base = DAG.getTargetConstant((Addr - (signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
Base = SDValue(DAG.getTargetNode(Opc, dl, CN->getValueType(0), Base), 0);
return true;
}
}
-
+
Disp = DAG.getTargetConstant(0, getPointerTy());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
// reg+imm, e.g. where imm = 0.
if (SelectAddressRegReg(N, Base, Index, DAG))
return true;
-
+
// If the operand is an addition, always emit this as [r+r], since this is
// better (for code size, and execution, as the memop does the add for free)
// than emitting an explicit add.
Index = N.getOperand(1);
return true;
}
-
+
// Otherwise, do it the hard way, using R0 as the base register.
Base = DAG.getRegister(PPC::R0, N.getValueType());
Index = N;
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
-
+
if (N.getOpcode() == ISD::ADD) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) && (imm & 3) == 0) {
Base = DAG.getRegister(PPC::R0, CN->getValueType(0));
return true;
}
-
+
// Fold the low-part of 32-bit absolute addresses into addr mode.
if (CN->getValueType(0) == MVT::i32 ||
(int64_t)CN->getZExtValue() == (int)CN->getZExtValue()) {
int Addr = (int)CN->getZExtValue();
-
+
// Otherwise, break this down into an LIS + disp.
Disp = DAG.getTargetConstant((short)Addr >> 2, MVT::i32);
Base = DAG.getTargetConstant((Addr-(signed short)Addr) >> 16, MVT::i32);
}
}
}
-
+
Disp = DAG.getTargetConstant(0, getPointerTy());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
SelectionDAG &DAG) const {
// Disabled by default for now.
if (!EnablePPCPreinc) return false;
-
+
SDValue Ptr;
MVT VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
VT = LD->getMemoryVT();
-
+
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
ST = ST;
Ptr = ST->getBasePtr();
// PowerPC doesn't have preinc load/store instructions for vectors.
if (VT.isVector())
return false;
-
+
// TODO: Check reg+reg first.
-
+
// LDU/STU use reg+imm*4, others use reg+imm.
if (VT != MVT::i64) {
// reg + imm
LD->getExtensionType() == ISD::SEXTLOAD &&
isa<ConstantSDNode>(Offset))
return false;
- }
-
+ }
+
AM = ISD::PRE_INC;
return true;
}
// LowerOperation implementation
//===----------------------------------------------------------------------===//
-SDValue PPCTargetLowering::LowerConstantPool(SDValue Op,
+SDValue PPCTargetLowering::LowerConstantPool(SDValue Op,
SelectionDAG &DAG) {
MVT PtrVT = Op.getValueType();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
DebugLoc dl = Op.getDebugLoc();
const TargetMachine &TM = DAG.getTarget();
-
+
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, CPI, Zero);
SDValue Lo = DAG.getNode(PPCISD::Lo, dl, PtrVT, CPI, Zero);
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
}
-
+
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, dl, PtrVT,
- DAG.getNode(PPCISD::GlobalBaseReg,
+ DAG.getNode(PPCISD::GlobalBaseReg,
DebugLoc::getUnknownLoc(), PtrVT), Hi);
}
-
+
Lo = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
return Lo;
}
SDValue Zero = DAG.getConstant(0, PtrVT);
// FIXME there isn't really any debug loc here
DebugLoc dl = Op.getDebugLoc();
-
+
const TargetMachine &TM = DAG.getTarget();
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, JTI, Zero);
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
}
-
+
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, dl, PtrVT,
- DAG.getNode(PPCISD::GlobalBaseReg,
+ DAG.getNode(PPCISD::GlobalBaseReg,
DebugLoc::getUnknownLoc(), PtrVT), Hi);
}
-
+
Lo = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
return Lo;
}
-SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
+SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
SelectionDAG &DAG) {
assert(0 && "TLS not implemented for PPC.");
return SDValue(); // Not reached
}
-SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
+SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) {
MVT PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
SDValue Zero = DAG.getConstant(0, PtrVT);
// FIXME there isn't really any debug info here
DebugLoc dl = GSDN->getDebugLoc();
-
+
const TargetMachine &TM = DAG.getTarget();
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, GA, Zero);
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
}
-
+
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, dl, PtrVT,
- DAG.getNode(PPCISD::GlobalBaseReg,
+ DAG.getNode(PPCISD::GlobalBaseReg,
DebugLoc::getUnknownLoc(), PtrVT), Hi);
}
-
+
Lo = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
-
+
if (!TM.getSubtarget<PPCSubtarget>().hasLazyResolverStub(GV))
return Lo;
-
+
// If the global is weak or external, we have to go through the lazy
// resolution stub.
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Lo, NULL, 0);
SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) {
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
DebugLoc dl = Op.getDebugLoc();
-
+
// If we're comparing for equality to zero, expose the fact that this is
// implented as a ctlz/srl pair on ppc, so that the dag combiner can
// fold the new nodes.
if (VT.bitsLT(MVT::i32)) {
VT = MVT::i32;
Zext = DAG.getNode(ISD::ZERO_EXTEND, dl, VT, Op.getOperand(0));
- }
+ }
unsigned Log2b = Log2_32(VT.getSizeInBits());
SDValue Clz = DAG.getNode(ISD::CTLZ, dl, VT, Zext);
SDValue Scc = DAG.getNode(ISD::SRL, dl, VT, Clz,
DAG.getConstant(Log2b, MVT::i32));
return DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Scc);
}
- // Leave comparisons against 0 and -1 alone for now, since they're usually
+ // Leave comparisons against 0 and -1 alone for now, since they're usually
// optimized. FIXME: revisit this when we can custom lower all setcc
// optimizations.
if (C->isAllOnesValue() || C->isNullValue())
return SDValue();
}
-
+
// If we have an integer seteq/setne, turn it into a compare against zero
// by xor'ing the rhs with the lhs, which is faster than setting a
// condition register, reading it back out, and masking the correct bit. The
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, dl, LHSVT, Op.getOperand(0),
+ SDValue Sub = DAG.getNode(ISD::XOR, dl, LHSVT, Op.getOperand(0),
Op.getOperand(1));
return DAG.getSetCC(dl, VT, Sub, DAG.getConstant(0, LHSVT), CC);
}
unsigned VarArgsNumGPR,
unsigned VarArgsNumFPR,
const PPCSubtarget &Subtarget) {
-
+
assert(0 && "VAARG in ELF32 ABI not implemented yet!");
return SDValue(); // Not reached
}
const Type *IntPtrTy =
DAG.getTargetLoweringInfo().getTargetData()->getIntPtrType();
- TargetLowering::ArgListTy Args;
+ TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = IntPtrTy;
Entry.Node = FPtr; Args.push_back(Entry);
Entry.Node = Nest; Args.push_back(Entry);
-
+
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
std::pair<SDValue, SDValue> CallResult =
LowerCallTo(Chain, Op.getValueType().getTypeForMVT(), false, false,
SDValue ArgGPR = DAG.getConstant(VarArgsNumGPR, MVT::i8);
SDValue ArgFPR = DAG.getConstant(VarArgsNumFPR, MVT::i8);
-
+
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-
+
SDValue StackOffsetFI = DAG.getFrameIndex(VarArgsStackOffset, PtrVT);
SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
-
+
uint64_t FrameOffset = PtrVT.getSizeInBits()/8;
SDValue ConstFrameOffset = DAG.getConstant(FrameOffset, PtrVT);
uint64_t FPROffset = 1;
SDValue ConstFPROffset = DAG.getConstant(FPROffset, PtrVT);
-
+
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
-
+
// Store first byte : number of int regs
SDValue firstStore = DAG.getStore(Op.getOperand(0), dl, ArgGPR,
Op.getOperand(1), SV, 0);
uint64_t nextOffset = FPROffset;
SDValue nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, Op.getOperand(1),
ConstFPROffset);
-
+
// Store second byte : number of float regs
SDValue secondStore =
DAG.getStore(firstStore, dl, ArgFPR, nextPtr, SV, nextOffset);
nextOffset += StackOffset;
nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, nextPtr, ConstStackOffset);
-
+
// Store second word : arguments given on stack
SDValue thirdStore =
DAG.getStore(secondStore, dl, StackOffsetFI, nextPtr, SV, nextOffset);
};
return FPR;
}
-
-
+
+
static const unsigned FPR[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
}
SDValue
-PPCTargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op,
+PPCTargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op,
SelectionDAG &DAG,
int &VarArgsFrameIndex,
int &VarArgsStackOffset,
SDValue Root = Op.getOperand(0);
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() != 0;
DebugLoc dl = Op.getDebugLoc();
-
+
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
bool isMachoABI = Subtarget.isMachoABI();
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
-
+
static const unsigned *FPR = GetFPR(Subtarget);
-
+
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
const unsigned Num_VR_Regs = array_lengthof( VR);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
-
+
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
+ // 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.
// to handle Elf here.
unsigned VecArgOffset = ArgOffset;
if (!isVarArg && !isPPC64) {
- for (unsigned ArgNo = 0, e = Op.getNode()->getNumValues()-1; ArgNo != e;
+ for (unsigned ArgNo = 0, e = Op.getNode()->getNumValues()-1; ArgNo != e;
++ArgNo) {
MVT ObjectVT = Op.getValue(ArgNo).getValueType();
unsigned ObjSize = ObjectVT.getSizeInBits()/8;
if (Flags.isByVal()) {
// ObjSize is the true size, ArgSize rounded up to multiple of regs.
ObjSize = Flags.getByValSize();
- unsigned ArgSize =
+ unsigned ArgSize =
((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
VecArgOffset += ArgSize;
continue;
// 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.
- //
+ //
// In the ELF 32 ABI, GPRs and stack are double word align: an argument
// represented with two words (long long or double) must be copied to an
// even GPR_idx value or to an even ArgOffset value.
ISD::ArgFlagsTy Flags =
cast<ARG_FLAGSSDNode>(Op.getOperand(ArgNo+3))->getArgFlags();
// See if next argument requires stack alignment in ELF
- bool Align = Flags.isSplit();
+ bool Align = Flags.isSplit();
unsigned CurArgOffset = ArgOffset;
unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
SDValue Val = DAG.getCopyFromReg(Root, dl, VReg, PtrVT);
- SDValue Store = DAG.getTruncStore(Val.getValue(1), dl, Val, FIN,
+ SDValue Store = DAG.getTruncStore(Val.getValue(1), dl, Val, FIN,
NULL, 0, ObjSize==1 ? MVT::i8 : MVT::i16 );
MemOps.push_back(Store);
++GPR_idx;
ArgSize = PtrByteSize;
}
// Stack align in ELF
- if (needsLoad && Align && isELF32_ABI)
+ if (needsLoad && Align && isELF32_ABI)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
// All int arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += PtrByteSize;
// All int arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += 8;
break;
-
+
case MVT::f32:
case MVT::f64:
// Every 4 bytes of argument space consumes one of the GPRs available for
} else {
needsLoad = true;
}
-
+
// Stack align in ELF
if (needsLoad && Align && isELF32_ABI)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
}
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.
if (needsLoad) {
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
ArgVal = DAG.getLoad(ObjectVT, dl, Root, FIN, NULL, 0);
}
-
+
ArgValues.push_back(ArgVal);
}
// 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.
if (isVarArg) {
-
+
int depth;
if (isELF32_ABI) {
VarArgsNumGPR = GPR_idx;
VarArgsNumFPR = FPR_idx;
-
+
// Make room for Num_GPR_Regs, Num_FPR_Regs and for a possible frame
// pointer.
depth = -(Num_GPR_Regs * PtrVT.getSizeInBits()/8 +
Num_FPR_Regs * MVT(MVT::f64).getSizeInBits()/8 +
PtrVT.getSizeInBits()/8);
-
+
VarArgsStackOffset = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
ArgOffset);
}
else
depth = ArgOffset;
-
+
VarArgsFrameIndex = MFI->CreateFixedObject(PtrVT.getSizeInBits()/8,
depth);
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) {
}
}
}
-
+
if (!MemOps.empty())
- Root = DAG.getNode(ISD::TokenFactor, dl,
+ Root = DAG.getNode(ISD::TokenFactor, dl,
MVT::Other, &MemOps[0], MemOps.size());
ArgValues.push_back(Root);
-
+
// Return the new list of results.
return DAG.getNode(ISD::MERGE_VALUES, dl, Op.getNode()->getVTList(),
&ArgValues[0], ArgValues.size());
static SDNode *isBLACompatibleAddress(SDValue Op, SelectionDAG &DAG) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
if (!C) return 0;
-
+
int Addr = C->getZExtValue();
if ((Addr & 3) != 0 || // Low 2 bits are implicitly zero.
(Addr << 6 >> 6) != Addr)
return 0; // Top 6 bits have to be sext of immediate.
-
+
return DAG.getConstant((int)C->getZExtValue() >> 2,
DAG.getTargetLoweringInfo().getPointerTy()).getNode();
}
}
/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
-/// by "Src" to address "Dst" of size "Size". Alignment information is
+/// 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
+static SDValue
CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
unsigned Size, DebugLoc dl) {
SDValue Callee = TheCall->getCallee();
unsigned NumOps = TheCall->getNumArgs();
DebugLoc dl = TheCall->getDebugLoc();
-
+
bool isMachoABI = Subtarget.isMachoABI();
bool isELF32_ABI = Subtarget.isELF32_ABI();
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<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
// 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.getIntPtrConstant(NumBytes, true));
SDValue CallSeqStart = Chain;
-
+
// Load the return address and frame pointer so it can be move somewhere else
// later.
SDValue LROp, FPOp;
StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
else
StackPtr = DAG.getRegister(PPC::R1, MVT::i32);
-
+
// Figure out which arguments are going to go in registers, and which in
// memory. Also, if this is a vararg function, floating point operations
// must be stored to our stack, and loaded into integer regs as well, if
// any integer regs are available for argument passing.
unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
-
+
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const unsigned *FPR = GetFPR(Subtarget);
-
+
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
const unsigned NumGPRs = array_lengthof(GPR_32);
const unsigned NumFPRs = isMachoABI ? 13 : 8;
const unsigned NumVRs = array_lengthof( VR);
-
+
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
std::vector<std::pair<unsigned, SDValue> > RegsToPass;
// PtrOff will be used to store the current argument to the stack if a
// register cannot be found for it.
SDValue PtrOff;
-
+
// Stack align in ELF 32
if (isELF32_ABI && Align)
PtrOff = DAG.getConstant(ArgOffset + ((ArgOffset/4) % 2) * PtrByteSize,
// Everything else is passed left-justified.
MVT VT = (Size==1) ? MVT::i8 : MVT::i16;
if (GPR_idx != NumGPRs) {
- SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, PtrVT, Chain, Arg,
+ SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, PtrVT, Chain, Arg,
NULL, 0, VT);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
SDValue Const = DAG.getConstant(4 - Size, PtrOff.getValueType());
SDValue AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, Const);
SDValue MemcpyCall = CreateCopyOfByValArgument(Arg, AddPtr,
- CallSeqStart.getNode()->getOperand(0),
+ CallSeqStart.getNode()->getOperand(0),
Flags, DAG, Size, dl);
// This must go outside the CALLSEQ_START..END.
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
// 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.getNode()->getOperand(0),
+ CallSeqStart.getNode()->getOperand(0),
Flags, DAG, Size, dl);
// This must go outside the CALLSEQ_START..END.
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
case MVT::v16i8:
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
+ // 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.
}
// We could elide this store in the case where the object fits
// entirely in R registers. Maybe later.
- PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
+ PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
DAG.getConstant(ArgOffset, PtrVT));
SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0);
MemOpChains.push_back(Store);
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, 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.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
- Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
+ Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
-
+
// With the ELF 32 ABI, set CR6 to true if this is a vararg call.
if (isVarArg && isELF32_ABI) {
SDValue SetCR(DAG.getTargetNode(PPC::CRSET, dl, MVT::i32), 0);
SmallVector<SDValue, 8> Ops;
unsigned CallOpc = isMachoABI? PPCISD::CALL_Macho : PPCISD::CALL_ELF;
-
+
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
// direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
// node so that legalize doesn't hack it.
Chain = DAG.getNode(PPCISD::MTCTR, dl, NodeTys, MTCTROps,
2 + (InFlag.getNode() != 0));
InFlag = Chain.getValue(1);
-
+
// Copy the callee address into R12/X12 on darwin.
if (isMachoABI) {
unsigned Reg = Callee.getValueType() == MVT::i32 ? PPC::R12 : PPC::X12;
// 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,
+ Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
// When performing tail call optimization the callee pops its arguments off
unsigned CallerCC = DAG.getMachineFunction().getFunction()->getCallingConv();
CCState CCInfo(CallerCC, isVarArg, TM, RVLocs);
CCInfo.AnalyzeCallResult(TheCall, RetCC_PPC);
-
+
// 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, dl,
+ Chain = DAG.getCopyFromReg(Chain, dl,
VA.getLocReg(), VT, InFlag).getValue(1);
ResultVals.push_back(Chain.getValue(0));
InFlag = Chain.getValue(2);
// If the function returns void, just return the chain.
if (RVLocs.empty())
return Chain;
-
+
// Otherwise, merge everything together with a MERGE_VALUES node.
ResultVals.push_back(Chain);
SDValue Res = DAG.getNode(ISD::MERGE_VALUES, dl, TheCall->getVTList(),
return Res.getValue(Op.getResNo());
}
-SDValue PPCTargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG,
+SDValue PPCTargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG,
TargetMachine &TM) {
SmallVector<CCValAssign, 16> RVLocs;
unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
DebugLoc dl = Op.getDebugLoc();
CCState CCInfo(CC, isVarArg, TM, RVLocs);
CCInfo.AnalyzeReturn(Op.getNode(), RetCC_PPC);
-
+
// If this is the first return lowered for this function, add the regs to the
// liveout set for the function.
if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
}
SDValue Flag;
-
+
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
- Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
+ Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
Op.getOperand(i*2+1), Flag);
Flag = Chain.getValue(1);
}
const PPCSubtarget &Subtarget) {
// When we pop the dynamic allocation we need to restore the SP link.
DebugLoc dl = Op.getDebugLoc();
-
+
// Get the corect type for pointers.
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Get the operands for the STACKRESTORE.
SDValue Chain = Op.getOperand(0);
SDValue SaveSP = Op.getOperand(1);
-
+
// Load the old link SP.
SDValue LoadLinkSP = DAG.getLoad(PtrVT, dl, Chain, StackPtr, NULL, 0);
-
+
// Restore the stack pointer.
Chain = DAG.getCopyToReg(LoadLinkSP.getValue(1), dl, SP, SaveSP);
-
+
// Store the old link SP.
return DAG.getStore(Chain, dl, LoadLinkSP, StackPtr, NULL, 0);
}
if (!FPSI) {
// Find out what the fix offset of the frame pointer save area.
int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64, isMachoABI);
-
+
// Allocate the frame index for frame pointer save area.
- FPSI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, FPOffset);
+ FPSI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, FPOffset);
// Save the result.
- FI->setFramePointerSaveIndex(FPSI);
+ FI->setFramePointerSaveIndex(FPSI);
}
return DAG.getFrameIndex(FPSI, PtrVT);
}
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
- DebugLoc dl = Op.getDebugLoc();
-
+ DebugLoc dl = Op.getDebugLoc();
+
// Get the corect type for pointers.
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Negate the size.
// Not FP? Not a fsel.
if (!Op.getOperand(0).getValueType().isFloatingPoint() ||
!Op.getOperand(2).getValueType().isFloatingPoint())
- return SDValue();
-
+ return Op;
+
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
-
+
// Cannot handle SETEQ/SETNE.
- if (CC == ISD::SETEQ || CC == ISD::SETNE) return SDValue();
-
+ if (CC == ISD::SETEQ || CC == ISD::SETNE) return Op;
+
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);
DebugLoc dl = Op.getDebugLoc();
-
+
// If the RHS of the comparison is a 0.0, we don't need to do the
// subtraction at all.
if (isFloatingPointZero(RHS))
return DAG.getNode(PPCISD::FSEL, dl, ResVT,
DAG.getNode(ISD::FNEG, dl, MVT::f64, LHS), TV, FV);
}
-
+
SDValue Cmp;
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, TV, FV);
}
- return SDValue();
+ return Op;
}
// FIXME: Split this code up when LegalizeDAGTypes lands.
-SDValue PPCTargetLowering::LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG,
+SDValue PPCTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
DebugLoc dl) {
assert(Op.getOperand(0).getValueType().isFloatingPoint());
SDValue Src = Op.getOperand(0);
SDValue Tmp;
switch (Op.getValueType().getSimpleVT()) {
- default: assert(0 && "Unhandled FP_TO_SINT type in custom expander!");
+ default: assert(0 && "Unhandled FP_TO_INT type in custom expander!");
case MVT::i32:
- Tmp = DAG.getNode(PPCISD::FCTIWZ, dl, MVT::f64, Src);
+ Tmp = DAG.getNode(Op.getOpcode()==ISD::FP_TO_SINT ? PPCISD::FCTIWZ :
+ PPCISD::FCTIDZ,
+ dl, MVT::f64, Src);
break;
case MVT::i64:
Tmp = DAG.getNode(PPCISD::FCTIDZ, dl, MVT::f64, Src);
return SDValue();
if (Op.getOperand(0).getValueType() == MVT::i64) {
- SDValue Bits = DAG.getNode(ISD::BIT_CONVERT, dl,
+ SDValue Bits = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::f64, Op.getOperand(0));
SDValue FP = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Bits);
if (Op.getValueType() == MVT::f32)
- FP = DAG.getNode(ISD::FP_ROUND, dl,
+ FP = DAG.getNode(ISD::FP_ROUND, dl,
MVT::f32, FP, DAG.getIntPtrConstant(0));
return FP;
}
-
+
assert(Op.getOperand(0).getValueType() == MVT::i32 &&
"Unhandled SINT_TO_FP type in custom expander!");
// Since we only generate this in 64-bit mode, we can take advantage of
int FrameIdx = FrameInfo->CreateStackObject(8, 8);
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
-
+
SDValue Ext64 = DAG.getNode(PPCISD::EXTSW_32, dl, MVT::i32,
Op.getOperand(0));
-
+
// STD the extended value into the stack slot.
MachineMemOperand MO(PseudoSourceValue::getFixedStack(FrameIdx),
MachineMemOperand::MOStore, 0, 8, 8);
DAG.getMemOperand(MO));
// Load the value as a double.
SDValue Ld = DAG.getLoad(MVT::f64, dl, Store, FIdx, NULL, 0);
-
+
// FCFID it and return it.
SDValue FP = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Ld);
if (Op.getValueType() == MVT::f32)
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.
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, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SHL, dl, VT, Hi, Amt);
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.
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, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SRL, dl, VT, Lo, Amt);
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, dl, AmtVT,
DAG.getConstant(BitWidth, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SRL, dl, VT, Lo, Amt);
// Vector related lowering.
//
-// 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
-// zero. Return true if this is not an array of constants, false if it is.
-//
-static bool GetConstantBuildVectorBits(SDNode *BV, uint64_t VectorBits[2],
- uint64_t UndefBits[2]) {
- // Start with zero'd results.
- VectorBits[0] = VectorBits[1] = UndefBits[0] = UndefBits[1] = 0;
-
- unsigned EltBitSize = BV->getOperand(0).getValueType().getSizeInBits();
- for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++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.
-
- uint64_t EltBits = 0;
- if (OpVal.getOpcode() == ISD::UNDEF) {
- uint64_t EltUndefBits = ~0U >> (32-EltBitSize);
- UndefBits[PartNo] |= EltUndefBits << (SlotNo*EltBitSize);
- continue;
- } else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
- EltBits = CN->getZExtValue() & (~0U >> (32-EltBitSize));
- } else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal)) {
- assert(CN->getValueType(0) == MVT::f32 &&
- "Only one legal FP vector type!");
- EltBits = FloatToBits(CN->getValueAPF().convertToFloat());
- } else {
- // Nonconstant element.
- return true;
- }
-
- VectorBits[PartNo] |= EltBits << (SlotNo*EltBitSize);
- }
-
- //printf("%llx %llx %llx %llx\n",
- // VectorBits[0], VectorBits[1], UndefBits[0], UndefBits[1]);
- return false;
-}
-
-// 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
-// splat of 0x01, 0x0101, and 0x01010101. We return SplatBits = 0x01 and
-// SplatSize = 1 byte.
-static bool isConstantSplat(const uint64_t Bits128[2],
- const uint64_t Undef128[2],
- unsigned &SplatBits, unsigned &SplatUndef,
- unsigned &SplatSize) {
-
- // Don't let undefs prevent splats from matching. See if the top 64-bits are
- // the same as the lower 64-bits, ignoring undefs.
- if ((Bits128[0] & ~Undef128[1]) != (Bits128[1] & ~Undef128[0]))
- return false; // Can't be a splat if two pieces don't match.
-
- uint64_t Bits64 = Bits128[0] | Bits128[1];
- uint64_t Undef64 = Undef128[0] & Undef128[1];
-
- // Check that the top 32-bits are the same as the lower 32-bits, ignoring
- // undefs.
- if ((Bits64 & (~Undef64 >> 32)) != ((Bits64 >> 32) & ~Undef64))
- return false; // Can't be a splat if two pieces don't match.
-
- uint32_t Bits32 = uint32_t(Bits64) | uint32_t(Bits64 >> 32);
- uint32_t Undef32 = uint32_t(Undef64) & uint32_t(Undef64 >> 32);
-
- // If the top 16-bits are different than the lower 16-bits, ignoring
- // undefs, we have an i32 splat.
- if ((Bits32 & (~Undef32 >> 16)) != ((Bits32 >> 16) & ~Undef32)) {
- SplatBits = Bits32;
- SplatUndef = Undef32;
- SplatSize = 4;
- return true;
- }
-
- uint16_t Bits16 = uint16_t(Bits32) | uint16_t(Bits32 >> 16);
- uint16_t Undef16 = uint16_t(Undef32) & uint16_t(Undef32 >> 16);
-
- // If the top 8-bits are different than the lower 8-bits, ignoring
- // undefs, we have an i16 splat.
- if ((Bits16 & (uint16_t(~Undef16) >> 8)) != ((Bits16 >> 8) & ~Undef16)) {
- SplatBits = Bits16;
- SplatUndef = Undef16;
- SplatSize = 2;
- return true;
- }
-
- // Otherwise, we have an 8-bit splat.
- SplatBits = uint8_t(Bits16) | uint8_t(Bits16 >> 8);
- SplatUndef = uint8_t(Undef16) & uint8_t(Undef16 >> 8);
- SplatSize = 1;
- return true;
-}
-
/// BuildSplatI - Build a canonical splati of Val with an element size of
/// SplatSize. Cast the result to VT.
static SDValue BuildSplatI(int Val, unsigned SplatSize, MVT VT,
};
MVT ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
-
+
// Force vspltis[hw] -1 to vspltisb -1 to canonicalize.
if (Val == -1)
SplatSize = 1;
-
+
MVT CanonicalVT = VTys[SplatSize-1];
-
+
// Build a canonical splat for this value.
- SDValue Elt = DAG.getConstant(Val, CanonicalVT.getVectorElementType());
+ SDValue Elt = DAG.getConstant(Val, MVT::i32);
SmallVector<SDValue, 8> Ops;
Ops.assign(CanonicalVT.getVectorNumElements(), Elt);
SDValue Res = DAG.getNode(ISD::BUILD_VECTOR, dl, CanonicalVT,
LHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, RHS);
- SDValue Ops[16];
+ int Ops[16];
for (unsigned i = 0; i != 16; ++i)
- Ops[i] = DAG.getConstant(i+Amt, MVT::i8);
- SDValue T = DAG.getNode(ISD::VECTOR_SHUFFLE, dl, MVT::v16i8, LHS, RHS,
- DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8, Ops,16));
+ Ops[i] = i + Amt;
+ SDValue T = DAG.getVectorShuffle(MVT::v16i8, dl, LHS, RHS, Ops);
return DAG.getNode(ISD::BIT_CONVERT, dl, 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.
-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
- // zero.
- uint64_t VectorBits[2];
- uint64_t UndefBits[2];
+SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- if (GetConstantBuildVectorBits(Op.getNode(), VectorBits, UndefBits))
- 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
- // splat of 0x01, 0x0101, and 0x01010101. We return SplatBits = 0x01 and
- // SplatSize = 1 byte.
- unsigned SplatBits, SplatUndef, SplatSize;
- if (isConstantSplat(VectorBits, UndefBits, SplatBits, SplatUndef, SplatSize)){
- bool HasAnyUndefs = (UndefBits[0] | UndefBits[1]) != 0;
-
- // First, handle single instruction cases.
-
- // All zeros?
- if (SplatBits == 0) {
- // Canonicalize all zero vectors to be v4i32.
- if (Op.getValueType() != MVT::v4i32 || HasAnyUndefs) {
- SDValue Z = DAG.getConstant(0, MVT::i32);
- Z = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Z, Z, Z, Z);
- Op = DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Z);
- }
- return Op;
+ BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
+ assert(BVN != 0 && "Expected a BuildVectorSDNode in LowerBUILD_VECTOR");
+
+ // Check if this is a splat of a constant value.
+ APInt APSplatBits, APSplatUndef;
+ unsigned SplatBitSize;
+ bool HasAnyUndefs;
+ if (! BVN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
+ HasAnyUndefs) || SplatBitSize > 32)
+ return SDValue();
+
+ unsigned SplatBits = APSplatBits.getZExtValue();
+ unsigned SplatUndef = APSplatUndef.getZExtValue();
+ unsigned SplatSize = SplatBitSize / 8;
+
+ // First, handle single instruction cases.
+
+ // All zeros?
+ if (SplatBits == 0) {
+ // Canonicalize all zero vectors to be v4i32.
+ if (Op.getValueType() != MVT::v4i32 || HasAnyUndefs) {
+ SDValue Z = DAG.getConstant(0, MVT::i32);
+ Z = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Z, Z, Z, Z);
+ Op = DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Z);
}
+ return Op;
+ }
+
+ // If the sign extended value is in the range [-16,15], use VSPLTI[bhw].
+ int32_t SextVal= (int32_t(SplatBits << (32-SplatBitSize)) >>
+ (32-SplatBitSize));
+ if (SextVal >= -16 && SextVal <= 15)
+ return BuildSplatI(SextVal, SplatSize, Op.getValueType(), DAG, dl);
+
+
+ // Two instruction sequences.
+
+ // 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) {
+ SDValue Res = BuildSplatI(SextVal >> 1, SplatSize, MVT::Other, DAG, dl);
+ Res = DAG.getNode(ISD::ADD, dl, Res.getValueType(), Res, Res);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
+ }
+
+ // If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
+ // 0x7FFF_FFFF x 4, turn it into not(0x8000_0000). This is important
+ // for fneg/fabs.
+ if (SplatSize == 4 && SplatBits == (0x7FFFFFFF&~SplatUndef)) {
+ // Make -1 and vspltisw -1:
+ SDValue OnesV = BuildSplatI(-1, 4, MVT::v4i32, DAG, dl);
+
+ // Make the VSLW intrinsic, computing 0x8000_0000.
+ SDValue Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
+ OnesV, DAG, dl);
+
+ // xor by OnesV to invert it.
+ Res = DAG.getNode(ISD::XOR, dl, MVT::v4i32, Res, OnesV);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
+ }
+
+ // Check to see if this is a wide variety of vsplti*, binop self cases.
+ static const signed char SplatCsts[] = {
+ -1, 1, -2, 2, -3, 3, -4, 4, -5, 5, -6, 6, -7, 7,
+ -8, 8, -9, 9, -10, 10, -11, 11, -12, 12, -13, 13, 14, -14, 15, -15, -16
+ };
- // If the sign extended value is in the range [-16,15], use VSPLTI[bhw].
- int32_t SextVal= int32_t(SplatBits << (32-8*SplatSize)) >> (32-8*SplatSize);
- if (SextVal >= -16 && SextVal <= 15)
- return BuildSplatI(SextVal, SplatSize, Op.getValueType(), DAG, dl);
-
-
- // Two instruction sequences.
-
- // 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) {
- SDValue Res = BuildSplatI(SextVal >> 1, SplatSize, MVT::Other, DAG, dl);
- Res = DAG.getNode(ISD::ADD, dl, Res.getValueType(), Res, Res);
+ for (unsigned idx = 0; idx < array_lengthof(SplatCsts); ++idx) {
+ // Indirect through the SplatCsts array so that we favor 'vsplti -1' for
+ // cases which are ambiguous (e.g. formation of 0x8000_0000). 'vsplti -1'
+ int i = SplatCsts[idx];
+
+ // Figure out what shift amount will be used by altivec if shifted by i in
+ // this splat size.
+ unsigned TypeShiftAmt = i & (SplatBitSize-1);
+
+ // vsplti + shl self.
+ if (SextVal == (i << (int)TypeShiftAmt)) {
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
+ static const unsigned IIDs[] = { // Intrinsic to use for each size.
+ Intrinsic::ppc_altivec_vslb, Intrinsic::ppc_altivec_vslh, 0,
+ Intrinsic::ppc_altivec_vslw
+ };
+ Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
}
-
- // If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
- // 0x7FFF_FFFF x 4, turn it into not(0x8000_0000). This is important
- // for fneg/fabs.
- if (SplatSize == 4 && SplatBits == (0x7FFFFFFF&~SplatUndef)) {
- // Make -1 and vspltisw -1:
- SDValue OnesV = BuildSplatI(-1, 4, MVT::v4i32, DAG, dl);
-
- // Make the VSLW intrinsic, computing 0x8000_0000.
- SDValue Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
- OnesV, DAG, dl);
-
- // xor by OnesV to invert it.
- Res = DAG.getNode(ISD::XOR, dl, MVT::v4i32, Res, OnesV);
+
+ // vsplti + srl self.
+ if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
+ static const unsigned IIDs[] = { // Intrinsic to use for each size.
+ Intrinsic::ppc_altivec_vsrb, Intrinsic::ppc_altivec_vsrh, 0,
+ Intrinsic::ppc_altivec_vsrw
+ };
+ Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
}
- // Check to see if this is a wide variety of vsplti*, binop self cases.
- unsigned SplatBitSize = SplatSize*8;
- static const signed char SplatCsts[] = {
- -1, 1, -2, 2, -3, 3, -4, 4, -5, 5, -6, 6, -7, 7,
- -8, 8, -9, 9, -10, 10, -11, 11, -12, 12, -13, 13, 14, -14, 15, -15, -16
- };
-
- for (unsigned idx = 0; idx < array_lengthof(SplatCsts); ++idx) {
- // Indirect through the SplatCsts array so that we favor 'vsplti -1' for
- // cases which are ambiguous (e.g. formation of 0x8000_0000). 'vsplti -1'
- int i = SplatCsts[idx];
-
- // Figure out what shift amount will be used by altivec if shifted by i in
- // this splat size.
- unsigned TypeShiftAmt = i & (SplatBitSize-1);
-
- // vsplti + shl self.
- if (SextVal == (i << (int)TypeShiftAmt)) {
- SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
- static const unsigned IIDs[] = { // Intrinsic to use for each size.
- Intrinsic::ppc_altivec_vslb, Intrinsic::ppc_altivec_vslh, 0,
- Intrinsic::ppc_altivec_vslw
- };
- Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
- }
-
- // vsplti + srl self.
- if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
- SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
- static const unsigned IIDs[] = { // Intrinsic to use for each size.
- Intrinsic::ppc_altivec_vsrb, Intrinsic::ppc_altivec_vsrh, 0,
- Intrinsic::ppc_altivec_vsrw
- };
- Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
- }
-
- // vsplti + sra self.
- if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
- SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
- static const unsigned IIDs[] = { // Intrinsic to use for each size.
- Intrinsic::ppc_altivec_vsrab, Intrinsic::ppc_altivec_vsrah, 0,
- Intrinsic::ppc_altivec_vsraw
- };
- Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
- }
-
- // vsplti + rol self.
- if (SextVal == (int)(((unsigned)i << TypeShiftAmt) |
- ((unsigned)i >> (SplatBitSize-TypeShiftAmt)))) {
- SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
- static const unsigned IIDs[] = { // Intrinsic to use for each size.
- Intrinsic::ppc_altivec_vrlb, Intrinsic::ppc_altivec_vrlh, 0,
- Intrinsic::ppc_altivec_vrlw
- };
- Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
- }
+ // vsplti + sra self.
+ if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
+ static const unsigned IIDs[] = { // Intrinsic to use for each size.
+ Intrinsic::ppc_altivec_vsrab, Intrinsic::ppc_altivec_vsrah, 0,
+ Intrinsic::ppc_altivec_vsraw
+ };
+ Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
+ }
- // t = vsplti c, result = vsldoi t, t, 1
- if (SextVal == ((i << 8) | (i >> (TypeShiftAmt-8)))) {
- SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
- return BuildVSLDOI(T, T, 1, Op.getValueType(), DAG, dl);
- }
- // t = vsplti c, result = vsldoi t, t, 2
- if (SextVal == ((i << 16) | (i >> (TypeShiftAmt-16)))) {
- SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
- return BuildVSLDOI(T, T, 2, Op.getValueType(), DAG, dl);
- }
- // t = vsplti c, result = vsldoi t, t, 3
- if (SextVal == ((i << 24) | (i >> (TypeShiftAmt-24)))) {
- SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
- return BuildVSLDOI(T, T, 3, Op.getValueType(), DAG, dl);
- }
+ // vsplti + rol self.
+ if (SextVal == (int)(((unsigned)i << TypeShiftAmt) |
+ ((unsigned)i >> (SplatBitSize-TypeShiftAmt)))) {
+ SDValue Res = BuildSplatI(i, SplatSize, MVT::Other, DAG, dl);
+ static const unsigned IIDs[] = { // Intrinsic to use for each size.
+ Intrinsic::ppc_altivec_vrlb, Intrinsic::ppc_altivec_vrlh, 0,
+ Intrinsic::ppc_altivec_vrlw
+ };
+ Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Res);
+ }
+
+ // t = vsplti c, result = vsldoi t, t, 1
+ if (SextVal == ((i << 8) | (i >> (TypeShiftAmt-8)))) {
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
+ return BuildVSLDOI(T, T, 1, Op.getValueType(), DAG, dl);
}
-
- // Three instruction sequences.
-
- // Odd, in range [17,31]: (vsplti C)-(vsplti -16).
- if (SextVal >= 0 && SextVal <= 31) {
- SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG, dl);
- SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
- LHS = DAG.getNode(ISD::SUB, dl, LHS.getValueType(), LHS, RHS);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), LHS);
+ // t = vsplti c, result = vsldoi t, t, 2
+ if (SextVal == ((i << 16) | (i >> (TypeShiftAmt-16)))) {
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
+ return BuildVSLDOI(T, T, 2, Op.getValueType(), DAG, dl);
}
- // Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
- if (SextVal >= -31 && SextVal <= 0) {
- SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG, dl);
- SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
- LHS = DAG.getNode(ISD::ADD, dl, LHS.getValueType(), LHS, RHS);
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), LHS);
+ // t = vsplti c, result = vsldoi t, t, 3
+ if (SextVal == ((i << 24) | (i >> (TypeShiftAmt-24)))) {
+ SDValue T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG, dl);
+ return BuildVSLDOI(T, T, 3, Op.getValueType(), DAG, dl);
}
}
-
+
+ // Three instruction sequences.
+
+ // Odd, in range [17,31]: (vsplti C)-(vsplti -16).
+ if (SextVal >= 0 && SextVal <= 31) {
+ SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG, dl);
+ SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
+ LHS = DAG.getNode(ISD::SUB, dl, LHS.getValueType(), LHS, RHS);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), LHS);
+ }
+ // Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
+ if (SextVal >= -31 && SextVal <= 0) {
+ SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG, dl);
+ SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
+ LHS = DAG.getNode(ISD::ADD, dl, LHS.getValueType(), LHS, RHS);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), LHS);
+ }
+
return SDValue();
}
/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
/// the specified operations to build the shuffle.
static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
- SDValue RHS, SelectionDAG &DAG,
+ SDValue RHS, SelectionDAG &DAG,
DebugLoc dl) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
-
+
enum {
OP_COPY = 0, // Copy, used for things like <u,u,u,3> to say it is <0,1,2,3>
OP_VMRGHW,
OP_VSLDOI8,
OP_VSLDOI12
};
-
+
if (OpNum == OP_COPY) {
if (LHSID == (1*9+2)*9+3) return LHS;
assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
return RHS;
}
-
+
SDValue OpLHS, OpRHS;
OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG, dl);
OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG, dl);
-
- unsigned ShufIdxs[16];
+
+ int ShufIdxs[16];
switch (OpNum) {
default: assert(0 && "Unknown i32 permute!");
case OP_VMRGHW:
case OP_VSLDOI12:
return BuildVSLDOI(OpLHS, OpRHS, 12, OpLHS.getValueType(), DAG, dl);
}
- SDValue Ops[16];
- for (unsigned i = 0; i != 16; ++i)
- Ops[i] = DAG.getConstant(ShufIdxs[i], MVT::i8);
-
- return DAG.getNode(ISD::VECTOR_SHUFFLE, dl, OpLHS.getValueType(),
- OpLHS, OpRHS,
- DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8, Ops, 16));
+ MVT VT = OpLHS.getValueType();
+ OpLHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, OpLHS);
+ OpRHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, OpRHS);
+ SDValue T = DAG.getVectorShuffle(MVT::v16i8, dl, OpLHS, OpRHS, ShufIdxs);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, VT, T);
}
/// LowerVECTOR_SHUFFLE - Return the code we lower for VECTOR_SHUFFLE. If this
/// 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.
-SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
- SelectionDAG &DAG) {
+SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
+ SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
- SDValue PermMask = Op.getOperand(2);
-
+ ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
+ MVT VT = Op.getValueType();
+
// Cases that are handled by instructions that take permute immediates
// (such as vsplt*) should be left as VECTOR_SHUFFLE nodes so they can be
// selected by the instruction selector.
if (V2.getOpcode() == ISD::UNDEF) {
- if (PPC::isSplatShuffleMask(PermMask.getNode(), 1) ||
- PPC::isSplatShuffleMask(PermMask.getNode(), 2) ||
- PPC::isSplatShuffleMask(PermMask.getNode(), 4) ||
- PPC::isVPKUWUMShuffleMask(PermMask.getNode(), true) ||
- PPC::isVPKUHUMShuffleMask(PermMask.getNode(), true) ||
- PPC::isVSLDOIShuffleMask(PermMask.getNode(), true) != -1 ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 1, true) ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 2, true) ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 4, true) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 1, true) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 2, true) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 4, true)) {
+ if (PPC::isSplatShuffleMask(SVOp, 1) ||
+ PPC::isSplatShuffleMask(SVOp, 2) ||
+ PPC::isSplatShuffleMask(SVOp, 4) ||
+ PPC::isVPKUWUMShuffleMask(SVOp, true) ||
+ PPC::isVPKUHUMShuffleMask(SVOp, true) ||
+ PPC::isVSLDOIShuffleMask(SVOp, true) != -1 ||
+ PPC::isVMRGLShuffleMask(SVOp, 1, true) ||
+ PPC::isVMRGLShuffleMask(SVOp, 2, true) ||
+ PPC::isVMRGLShuffleMask(SVOp, 4, true) ||
+ PPC::isVMRGHShuffleMask(SVOp, 1, true) ||
+ PPC::isVMRGHShuffleMask(SVOp, 2, true) ||
+ PPC::isVMRGHShuffleMask(SVOp, 4, true)) {
return Op;
}
}
-
+
// Altivec has a variety of "shuffle immediates" that take two vector inputs
// and produce a fixed permutation. If any of these match, do not lower to
// VPERM.
- if (PPC::isVPKUWUMShuffleMask(PermMask.getNode(), false) ||
- PPC::isVPKUHUMShuffleMask(PermMask.getNode(), false) ||
- PPC::isVSLDOIShuffleMask(PermMask.getNode(), false) != -1 ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 1, false) ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 2, false) ||
- PPC::isVMRGLShuffleMask(PermMask.getNode(), 4, false) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 1, false) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 2, false) ||
- PPC::isVMRGHShuffleMask(PermMask.getNode(), 4, false))
+ if (PPC::isVPKUWUMShuffleMask(SVOp, false) ||
+ PPC::isVPKUHUMShuffleMask(SVOp, false) ||
+ PPC::isVSLDOIShuffleMask(SVOp, false) != -1 ||
+ PPC::isVMRGLShuffleMask(SVOp, 1, false) ||
+ PPC::isVMRGLShuffleMask(SVOp, 2, false) ||
+ PPC::isVMRGLShuffleMask(SVOp, 4, false) ||
+ PPC::isVMRGHShuffleMask(SVOp, 1, false) ||
+ PPC::isVMRGHShuffleMask(SVOp, 2, false) ||
+ PPC::isVMRGHShuffleMask(SVOp, 4, false))
return Op;
-
+
// Check to see if this is a shuffle of 4-byte values. If so, we can use our
// perfect shuffle table to emit an optimal matching sequence.
+ SmallVector<int, 16> PermMask;
+ SVOp->getMask(PermMask);
+
unsigned PFIndexes[4];
bool isFourElementShuffle = true;
for (unsigned i = 0; i != 4 && isFourElementShuffle; ++i) { // Element number
unsigned EltNo = 8; // Start out undef.
for (unsigned j = 0; j != 4; ++j) { // Intra-element byte.
- if (PermMask.getOperand(i*4+j).getOpcode() == ISD::UNDEF)
+ if (PermMask[i*4+j] < 0)
continue; // Undef, ignore it.
-
- unsigned ByteSource =
- cast<ConstantSDNode>(PermMask.getOperand(i*4+j))->getZExtValue();
+
+ unsigned ByteSource = PermMask[i*4+j];
if ((ByteSource & 3) != j) {
isFourElementShuffle = false;
break;
}
-
+
if (EltNo == 8) {
EltNo = ByteSource/4;
} else if (EltNo != ByteSource/4) {
}
PFIndexes[i] = EltNo;
}
-
- // If this shuffle can be expressed as a shuffle of 4-byte elements, use the
+
+ // If this shuffle can be expressed as a shuffle of 4-byte elements, use the
// perfect shuffle vector to determine if it is cost effective to do this as
// discrete instructions, or whether we should use a vperm.
if (isFourElementShuffle) {
// Compute the index in the perfect shuffle table.
- unsigned PFTableIndex =
+ unsigned PFTableIndex =
PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
-
+
unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
unsigned Cost = (PFEntry >> 30);
-
+
// Determining when to avoid vperm is tricky. Many things affect the cost
// of vperm, particularly how many times the perm mask needs to be computed.
// For example, if the perm mask can be hoisted out of a loop or is already
// the loop requires an extra register.
//
// As a compromise, we only emit discrete instructions if the shuffle can be
- // generated in 3 or fewer operations. When we have loop information
+ // generated in 3 or fewer operations. When we have loop information
// available, if this block is within a loop, we should avoid using vperm
// for 3-operation perms and use a constant pool load instead.
- if (Cost < 3)
+ if (Cost < 3)
return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
}
-
+
// Lower this to a VPERM(V1, V2, V3) expression, where V3 is a constant
// vector that will get spilled to the constant pool.
if (V2.getOpcode() == ISD::UNDEF) V2 = V1;
-
+
// 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 EltVT = V1.getValueType().getVectorElementType();
unsigned BytesPerElement = EltVT.getSizeInBits()/8;
-
+
SmallVector<SDValue, 16> ResultMask;
- for (unsigned i = 0, e = PermMask.getNumOperands(); i != e; ++i) {
- unsigned SrcElt;
- if (PermMask.getOperand(i).getOpcode() == ISD::UNDEF)
- SrcElt = 0;
- else
- SrcElt = cast<ConstantSDNode>(PermMask.getOperand(i))->getZExtValue();
-
+ for (unsigned i = 0, e = VT.getVectorNumElements(); i != e; ++i) {
+ unsigned SrcElt = PermMask[i] < 0 ? 0 : PermMask[i];
+
for (unsigned j = 0; j != BytesPerElement; ++j)
ResultMask.push_back(DAG.getConstant(SrcElt*BytesPerElement+j,
- MVT::i8));
+ MVT::i32));
}
-
+
SDValue VPermMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8,
&ResultMask[0], ResultMask.size());
return DAG.getNode(PPCISD::VPERM, dl, V1.getValueType(), V1, V2, VPermMask);
case Intrinsic::ppc_altivec_vcmpgtub_p: CompareOpc = 518; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtuh_p: CompareOpc = 582; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtuw_p: CompareOpc = 646; isDot = 1; break;
-
+
// Normal Comparisons.
case Intrinsic::ppc_altivec_vcmpbfp: CompareOpc = 966; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpeqfp: CompareOpc = 198; isDot = 0; break;
/// LowerINTRINSIC_WO_CHAIN - If this is an intrinsic that we want to custom
/// lower, do it, otherwise return null.
-SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
+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.
bool isDot;
if (!getAltivecCompareInfo(Op, CompareOpc, isDot))
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) {
SDValue Tmp = DAG.getNode(PPCISD::VCMP, dl, Op.getOperand(2).getValueType(),
DAG.getConstant(CompareOpc, MVT::i32));
return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Tmp);
}
-
+
// Create the PPCISD altivec 'dot' comparison node.
SDValue Ops[] = {
Op.getOperand(2), // LHS
VTs.push_back(Op.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, 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.
SDValue Flags = DAG.getNode(PPCISD::MFCR, dl, MVT::i32,
DAG.getRegister(PPC::CR6, MVT::i32),
- CompNode.getValue(1));
-
+ CompNode.getValue(1));
+
// Unpack the result based on how the target uses it.
unsigned BitNo; // Bit # of CR6.
bool InvertBit; // Invert result?
BitNo = 2; InvertBit = true;
break;
}
-
+
// Shift the bit into the low position.
Flags = DAG.getNode(ISD::SRL, dl, MVT::i32, Flags,
DAG.getConstant(8-(3-BitNo), MVT::i32));
// Isolate the bit.
Flags = DAG.getNode(ISD::AND, dl, MVT::i32, Flags,
DAG.getConstant(1, MVT::i32));
-
+
// If we are supposed to, toggle the bit.
if (InvertBit)
Flags = DAG.getNode(ISD::XOR, dl, MVT::i32, Flags,
return Flags;
}
-SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
+SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
// Create a stack slot that is 16-byte aligned.
int FrameIdx = FrameInfo->CreateStackObject(16, 16);
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
-
+
// Store the input value into Value#0 of the stack slot.
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl,
Op.getOperand(0), FIdx, NULL, 0);
DebugLoc dl = Op.getDebugLoc();
if (Op.getValueType() == MVT::v4i32) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
-
+
SDValue Zero = BuildSplatI( 0, 1, MVT::v4i32, DAG, dl);
SDValue Neg16 = BuildSplatI(-16, 4, MVT::v4i32, DAG, dl);//+16 as shift amt.
-
+
SDValue RHSSwap = // = vrlw RHS, 16
BuildIntrinsicOp(Intrinsic::ppc_altivec_vrlw, RHS, Neg16, DAG, dl);
-
+
// Shrinkify inputs to v8i16.
LHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, RHS);
RHSSwap = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, RHSSwap);
-
+
// Low parts multiplied together, generating 32-bit results (we ignore the
// top parts).
SDValue LoProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmulouh,
LHS, RHS, DAG, dl, MVT::v4i32);
-
+
SDValue HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmsumuhm,
LHS, RHSSwap, Zero, DAG, dl, MVT::v4i32);
// Shift the high parts up 16 bits.
- HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, HiProd,
+ HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, HiProd,
Neg16, DAG, dl);
return DAG.getNode(ISD::ADD, dl, MVT::v4i32, LoProd, HiProd);
} else if (Op.getValueType() == MVT::v8i16) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
-
+
SDValue Zero = BuildSplatI(0, 1, MVT::v8i16, DAG, dl);
return BuildIntrinsicOp(Intrinsic::ppc_altivec_vmladduhm,
LHS, RHS, Zero, DAG, dl);
} else if (Op.getValueType() == MVT::v16i8) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
-
+
// Multiply the even 8-bit parts, producing 16-bit sums.
SDValue EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
LHS, RHS, DAG, dl, MVT::v8i16);
EvenParts = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, EvenParts);
-
+
// Multiply the odd 8-bit parts, producing 16-bit sums.
SDValue OddParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuloub,
LHS, RHS, DAG, dl, MVT::v8i16);
OddParts = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, OddParts);
-
+
// Merge the results together.
- SDValue Ops[16];
+ int 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);
+ Ops[i*2 ] = 2*i+1;
+ Ops[i*2+1] = 2*i+1+16;
}
- return DAG.getNode(ISD::VECTOR_SHUFFLE, dl, MVT::v16i8, EvenParts, OddParts,
- DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8, Ops, 16));
+ return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
} else {
assert(0 && "Unknown mul to lower!");
abort();
///
SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
- default: assert(0 && "Wasn't expecting to be able to lower this!");
+ default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SETCC: return LowerSETCC(Op, DAG);
case ISD::TRAMPOLINE: return LowerTRAMPOLINE(Op, DAG);
- case ISD::VASTART:
+ case ISD::VASTART:
return LowerVASTART(Op, DAG, VarArgsFrameIndex, VarArgsStackOffset,
VarArgsNumGPR, VarArgsNumFPR, PPCSubTarget);
-
- case ISD::VAARG:
+
+ case ISD::VAARG:
return LowerVAARG(Op, DAG, VarArgsFrameIndex, VarArgsStackOffset,
VarArgsNumGPR, VarArgsNumFPR, PPCSubTarget);
case ISD::FORMAL_ARGUMENTS:
- return LowerFORMAL_ARGUMENTS(Op, DAG, VarArgsFrameIndex,
+ return LowerFORMAL_ARGUMENTS(Op, DAG, VarArgsFrameIndex,
VarArgsStackOffset, VarArgsNumGPR,
VarArgsNumFPR, PPCSubTarget);
return LowerDYNAMIC_STACKALLOC(Op, DAG, PPCSubTarget);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
- case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG,
+ case ISD::FP_TO_UINT:
+ case ISD::FP_TO_SINT: return LowerFP_TO_INT(Op, DAG,
Op.getDebugLoc());
case ISD::SINT_TO_FP: return LowerSINT_TO_FP(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::SCALAR_TO_VECTOR: return LowerSCALAR_TO_VECTOR(Op, DAG);
case ISD::MUL: return LowerMUL(Op, DAG);
-
+
// Frame & Return address.
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::FP_ROUND_INREG: {
assert(N->getValueType(0) == MVT::ppcf128);
assert(N->getOperand(0).getValueType() == MVT::ppcf128);
- SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
+ SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
MVT::f64, N->getOperand(0),
DAG.getIntPtrConstant(0));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
// We know the low half is about to be thrown away, so just use something
// convenient.
- Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::ppcf128,
+ Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::ppcf128,
FPreg, FPreg));
return;
}
case ISD::FP_TO_SINT:
- Results.push_back(LowerFP_TO_SINT(SDValue(N, 0), DAG, dl));
+ Results.push_back(LowerFP_TO_INT(SDValue(N, 0), DAG, dl));
return;
}
}
unsigned ptrA = MI->getOperand(1).getReg();
unsigned ptrB = MI->getOperand(2).getReg();
unsigned incr = MI->getOperand(3).getReg();
+ DebugLoc dl = MI->getDebugLoc();
MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
// bne- loopMBB
// fallthrough --> exitMBB
BB = loopMBB;
- BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
.addReg(ptrA).addReg(ptrB);
if (BinOpcode)
- BuildMI(BB, TII->get(BinOpcode), TmpReg).addReg(incr).addReg(dest);
- BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ BuildMI(BB, dl, TII->get(BinOpcode), TmpReg).addReg(incr).addReg(dest);
+ BuildMI(BB, dl, 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);
+ BuildMI(BB, dl, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
BB->addSuccessor(loopMBB);
BB->addSuccessor(exitMBB);
}
MachineBasicBlock *
-PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
+PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
MachineBasicBlock *BB,
bool is8bit, // operation
unsigned BinOpcode) const {
unsigned ptrA = MI->getOperand(1).getReg();
unsigned ptrB = MI->getOperand(2).getReg();
unsigned incr = MI->getOperand(3).getReg();
+ DebugLoc dl = MI->getDebugLoc();
MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
exitMBB->transferSuccessors(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
- const TargetRegisterClass *RC =
+ const TargetRegisterClass *RC =
is64bit ? (const TargetRegisterClass *) &PPC::G8RCRegClass :
(const TargetRegisterClass *) &PPC::GPRCRegClass;
unsigned PtrReg = RegInfo.createVirtualRegister(RC);
if (ptrA!=PPC::R0) {
Ptr1Reg = RegInfo.createVirtualRegister(RC);
- BuildMI(BB, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
.addReg(ptrA).addReg(ptrB);
} else {
Ptr1Reg = ptrB;
}
- BuildMI(BB, TII->get(PPC::RLWINM), Shift1Reg).addReg(Ptr1Reg)
+ BuildMI(BB, dl, TII->get(PPC::RLWINM), Shift1Reg).addReg(Ptr1Reg)
.addImm(3).addImm(27).addImm(is8bit ? 28 : 27);
- BuildMI(BB, TII->get(is64bit ? PPC::XORI8 : PPC::XORI), ShiftReg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::XORI8 : PPC::XORI), ShiftReg)
.addReg(Shift1Reg).addImm(is8bit ? 24 : 16);
if (is64bit)
- BuildMI(BB, TII->get(PPC::RLDICR), PtrReg)
+ BuildMI(BB, dl, TII->get(PPC::RLDICR), PtrReg)
.addReg(Ptr1Reg).addImm(0).addImm(61);
else
- BuildMI(BB, TII->get(PPC::RLWINM), PtrReg)
+ BuildMI(BB, dl, TII->get(PPC::RLWINM), PtrReg)
.addReg(Ptr1Reg).addImm(0).addImm(0).addImm(29);
- BuildMI(BB, TII->get(PPC::SLW), Incr2Reg)
+ BuildMI(BB, dl, TII->get(PPC::SLW), Incr2Reg)
.addReg(incr).addReg(ShiftReg);
if (is8bit)
- BuildMI(BB, TII->get(PPC::LI), Mask2Reg).addImm(255);
+ BuildMI(BB, dl, TII->get(PPC::LI), Mask2Reg).addImm(255);
else {
- BuildMI(BB, TII->get(PPC::LI), Mask3Reg).addImm(0);
- BuildMI(BB, TII->get(PPC::ORI), Mask2Reg).addReg(Mask3Reg).addImm(65535);
+ BuildMI(BB, dl, TII->get(PPC::LI), Mask3Reg).addImm(0);
+ BuildMI(BB, dl, TII->get(PPC::ORI),Mask2Reg).addReg(Mask3Reg).addImm(65535);
}
- BuildMI(BB, TII->get(PPC::SLW), MaskReg)
+ BuildMI(BB, dl, TII->get(PPC::SLW), MaskReg)
.addReg(Mask2Reg).addReg(ShiftReg);
BB = loopMBB;
- BuildMI(BB, TII->get(PPC::LWARX), TmpDestReg)
+ BuildMI(BB, dl, TII->get(PPC::LWARX), TmpDestReg)
.addReg(PPC::R0).addReg(PtrReg);
if (BinOpcode)
- BuildMI(BB, TII->get(BinOpcode), TmpReg)
+ BuildMI(BB, dl, TII->get(BinOpcode), TmpReg)
.addReg(Incr2Reg).addReg(TmpDestReg);
- BuildMI(BB, TII->get(is64bit ? PPC::ANDC8 : PPC::ANDC), Tmp2Reg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::ANDC8 : PPC::ANDC), Tmp2Reg)
.addReg(TmpDestReg).addReg(MaskReg);
- BuildMI(BB, TII->get(is64bit ? PPC::AND8 : PPC::AND), Tmp3Reg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::AND8 : PPC::AND), Tmp3Reg)
.addReg(TmpReg).addReg(MaskReg);
- BuildMI(BB, TII->get(is64bit ? PPC::OR8 : PPC::OR), Tmp4Reg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::OR8 : PPC::OR), Tmp4Reg)
.addReg(Tmp3Reg).addReg(Tmp2Reg);
- BuildMI(BB, TII->get(PPC::STWCX))
+ BuildMI(BB, dl, TII->get(PPC::STWCX))
.addReg(Tmp4Reg).addReg(PPC::R0).addReg(PtrReg);
- BuildMI(BB, TII->get(PPC::BCC))
- .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
+ BuildMI(BB, dl, TII->get(PPC::BCC))
+ .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
BB->addSuccessor(loopMBB);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
- BuildMI(BB, TII->get(PPC::SRW), dest).addReg(TmpDestReg).addReg(ShiftReg);
+ BuildMI(BB, dl, TII->get(PPC::SRW), dest).addReg(TmpDestReg).addReg(ShiftReg);
return 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))
+ DebugLoc dl = MI->getDebugLoc();
+ BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB);
F->insert(It, copy0MBB);
F->insert(It, sinkMBB);
// 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())
+ BuildMI(BB, dl, TII->get(PPC::PHI), MI->getOperand(0).getReg())
.addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB)
.addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
}
unsigned ptrB = MI->getOperand(2).getReg();
unsigned oldval = MI->getOperand(3).getReg();
unsigned newval = MI->getOperand(4).getReg();
+ DebugLoc dl = MI->getDebugLoc();
MachineBasicBlock *loop1MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = F->CreateMachineBasicBlock(LLVM_BB);
// st[wd]cx. dest, ptr
// exitBB:
BB = loop1MBB;
- BuildMI(BB, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::LDARX : PPC::LWARX), dest)
.addReg(ptrA).addReg(ptrB);
- BuildMI(BB, TII->get(is64bit ? PPC::CMPD : PPC::CMPW), PPC::CR0)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::CMPD : PPC::CMPW), PPC::CR0)
.addReg(oldval).addReg(dest);
- BuildMI(BB, TII->get(PPC::BCC))
+ BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(midMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(midMBB);
BB = loop2MBB;
- BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
.addReg(newval).addReg(ptrA).addReg(ptrB);
- BuildMI(BB, TII->get(PPC::BCC))
+ BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loop1MBB);
- BuildMI(BB, TII->get(PPC::B)).addMBB(exitMBB);
+ BuildMI(BB, dl, TII->get(PPC::B)).addMBB(exitMBB);
BB->addSuccessor(loop1MBB);
BB->addSuccessor(exitMBB);
-
+
BB = midMBB;
- BuildMI(BB, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
.addReg(dest).addReg(ptrA).addReg(ptrB);
BB->addSuccessor(exitMBB);
unsigned ptrB = MI->getOperand(2).getReg();
unsigned oldval = MI->getOperand(3).getReg();
unsigned newval = MI->getOperand(4).getReg();
+ DebugLoc dl = MI->getDebugLoc();
MachineBasicBlock *loop1MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = F->CreateMachineBasicBlock(LLVM_BB);
exitMBB->transferSuccessors(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
- const TargetRegisterClass *RC =
+ const TargetRegisterClass *RC =
is64bit ? (const TargetRegisterClass *) &PPC::G8RCRegClass :
(const TargetRegisterClass *) &PPC::GPRCRegClass;
unsigned PtrReg = RegInfo.createVirtualRegister(RC);
// srw dest, tmpDest, shift
if (ptrA!=PPC::R0) {
Ptr1Reg = RegInfo.createVirtualRegister(RC);
- BuildMI(BB, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
.addReg(ptrA).addReg(ptrB);
} else {
Ptr1Reg = ptrB;
}
- BuildMI(BB, TII->get(PPC::RLWINM), Shift1Reg).addReg(Ptr1Reg)
+ BuildMI(BB, dl, TII->get(PPC::RLWINM), Shift1Reg).addReg(Ptr1Reg)
.addImm(3).addImm(27).addImm(is8bit ? 28 : 27);
- BuildMI(BB, TII->get(is64bit ? PPC::XORI8 : PPC::XORI), ShiftReg)
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::XORI8 : PPC::XORI), ShiftReg)
.addReg(Shift1Reg).addImm(is8bit ? 24 : 16);
if (is64bit)
- BuildMI(BB, TII->get(PPC::RLDICR), PtrReg)
+ BuildMI(BB, dl, TII->get(PPC::RLDICR), PtrReg)
.addReg(Ptr1Reg).addImm(0).addImm(61);
else
- BuildMI(BB, TII->get(PPC::RLWINM), PtrReg)
+ BuildMI(BB, dl, TII->get(PPC::RLWINM), PtrReg)
.addReg(Ptr1Reg).addImm(0).addImm(0).addImm(29);
- BuildMI(BB, TII->get(PPC::SLW), NewVal2Reg)
+ BuildMI(BB, dl, TII->get(PPC::SLW), NewVal2Reg)
.addReg(newval).addReg(ShiftReg);
- BuildMI(BB, TII->get(PPC::SLW), OldVal2Reg)
+ BuildMI(BB, dl, TII->get(PPC::SLW), OldVal2Reg)
.addReg(oldval).addReg(ShiftReg);
if (is8bit)
- BuildMI(BB, TII->get(PPC::LI), Mask2Reg).addImm(255);
+ BuildMI(BB, dl, TII->get(PPC::LI), Mask2Reg).addImm(255);
else {
- BuildMI(BB, TII->get(PPC::LI), Mask3Reg).addImm(0);
- BuildMI(BB, TII->get(PPC::ORI), Mask2Reg).addReg(Mask3Reg).addImm(65535);
+ BuildMI(BB, dl, TII->get(PPC::LI), Mask3Reg).addImm(0);
+ BuildMI(BB, dl, TII->get(PPC::ORI), Mask2Reg)
+ .addReg(Mask3Reg).addImm(65535);
}
- BuildMI(BB, TII->get(PPC::SLW), MaskReg)
+ BuildMI(BB, dl, TII->get(PPC::SLW), MaskReg)
.addReg(Mask2Reg).addReg(ShiftReg);
- BuildMI(BB, TII->get(PPC::AND), NewVal3Reg)
+ BuildMI(BB, dl, TII->get(PPC::AND), NewVal3Reg)
.addReg(NewVal2Reg).addReg(MaskReg);
- BuildMI(BB, TII->get(PPC::AND), OldVal3Reg)
+ BuildMI(BB, dl, TII->get(PPC::AND), OldVal3Reg)
.addReg(OldVal2Reg).addReg(MaskReg);
BB = loop1MBB;
- BuildMI(BB, TII->get(PPC::LWARX), TmpDestReg)
+ BuildMI(BB, dl, TII->get(PPC::LWARX), TmpDestReg)
.addReg(PPC::R0).addReg(PtrReg);
- BuildMI(BB, TII->get(PPC::AND),TmpReg).addReg(TmpDestReg).addReg(MaskReg);
- BuildMI(BB, TII->get(PPC::CMPW), PPC::CR0)
+ BuildMI(BB, dl, TII->get(PPC::AND),TmpReg)
+ .addReg(TmpDestReg).addReg(MaskReg);
+ BuildMI(BB, dl, TII->get(PPC::CMPW), PPC::CR0)
.addReg(TmpReg).addReg(OldVal3Reg);
- BuildMI(BB, TII->get(PPC::BCC))
+ BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(midMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(midMBB);
BB = loop2MBB;
- BuildMI(BB, TII->get(PPC::ANDC),Tmp2Reg).addReg(TmpDestReg).addReg(MaskReg);
- BuildMI(BB, TII->get(PPC::OR),Tmp4Reg).addReg(Tmp2Reg).addReg(NewVal3Reg);
- BuildMI(BB, TII->get(PPC::STWCX)).addReg(Tmp4Reg)
+ BuildMI(BB, dl, TII->get(PPC::ANDC),Tmp2Reg)
+ .addReg(TmpDestReg).addReg(MaskReg);
+ BuildMI(BB, dl, TII->get(PPC::OR),Tmp4Reg)
+ .addReg(Tmp2Reg).addReg(NewVal3Reg);
+ BuildMI(BB, dl, TII->get(PPC::STWCX)).addReg(Tmp4Reg)
.addReg(PPC::R0).addReg(PtrReg);
- BuildMI(BB, TII->get(PPC::BCC))
+ BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loop1MBB);
- BuildMI(BB, TII->get(PPC::B)).addMBB(exitMBB);
+ BuildMI(BB, dl, TII->get(PPC::B)).addMBB(exitMBB);
BB->addSuccessor(loop1MBB);
BB->addSuccessor(exitMBB);
-
+
BB = midMBB;
- BuildMI(BB, TII->get(PPC::STWCX)).addReg(TmpDestReg)
+ BuildMI(BB, dl, TII->get(PPC::STWCX)).addReg(TmpDestReg)
.addReg(PPC::R0).addReg(PtrReg);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
- BuildMI(BB, TII->get(PPC::SRW),dest).addReg(TmpReg).addReg(ShiftReg);
+ BuildMI(BB, dl, TII->get(PPC::SRW),dest).addReg(TmpReg).addReg(ShiftReg);
} else {
assert(0 && "Unexpected instr type to insert");
}
return N->getOperand(0);
}
break;
-
+
case ISD::SINT_TO_FP:
if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
if (N->getOperand(0).getOpcode() == ISD::FP_TO_SINT) {
Val = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Val);
DCI.AddToWorklist(Val.getNode());
}
-
+
Val = DAG.getNode(PPCISD::FCTIDZ, dl, MVT::f64, Val);
DCI.AddToWorklist(Val.getNode());
Val = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Val);
DCI.AddToWorklist(Val.getNode());
if (N->getValueType(0) == MVT::f32) {
- Val = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Val,
+ Val = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Val,
DAG.getIntPtrConstant(0));
DCI.AddToWorklist(Val.getNode());
}
DCI.AddToWorklist(Val.getNode());
return Val;
}
-
+
// Turn STORE (BSWAP) -> sthbrx/stwbrx.
if (N->getOperand(1).getOpcode() == ISD::BSWAP &&
N->getOperand(1).getNode()->hasOneUse() &&
};
SDValue BSLoad = DAG.getNode(PPCISD::LBRX, dl, VTs, Ops, 4);
- // If this is an i16 load, insert the truncate.
+ // If this is an i16 load, insert the truncate.
SDValue ResVal = BSLoad;
if (N->getValueType(0) == MVT::i16)
ResVal = DAG.getNode(ISD::TRUNCATE, dl, MVT::i16, BSLoad);
-
+
// First, combine the bswap away. This makes the value produced by the
// load dead.
DCI.CombineTo(N, ResVal);
// Next, combine the load away, we give it a bogus result value but a real
// chain result. The result value is dead because the bswap is dead.
DCI.CombineTo(Load.getNode(), ResVal, BSLoad.getValue(1));
-
+
// Return N so it doesn't get rechecked!
return SDValue(N, 0);
}
-
+
break;
case PPCISD::VCMP: {
// If a VCMPo node already exists with exactly the same operands as this
if (!N->getOperand(0).hasOneUse() &&
!N->getOperand(1).hasOneUse() &&
!N->getOperand(2).hasOneUse()) {
-
+
// Scan all of the users of the LHS, looking for VCMPo's that match.
SDNode *VCMPoNode = 0;
-
+
SDNode *LHSN = N->getOperand(0).getNode();
for (SDNode::use_iterator UI = LHSN->use_begin(), E = LHSN->use_end();
UI != E; ++UI)
VCMPoNode = *UI;
break;
}
-
+
// If there is no VCMPo node, or if the flag value has a single use, don't
// transform this.
if (!VCMPoNode || VCMPoNode->hasNUsesOfValue(0, 1))
break;
-
- // Look at the (necessarily single) use of the flag value. If it has a
+
+ // Look at the (necessarily single) use of the flag value. If it has a
// chain, this transformation is more complex. Note that multiple things
// could use the value result, which we should ignore.
SDNode *FlagUser = 0;
- for (SDNode::use_iterator UI = VCMPoNode->use_begin();
+ for (SDNode::use_iterator UI = VCMPoNode->use_begin();
FlagUser == 0; ++UI) {
assert(UI != VCMPoNode->use_end() && "Didn't find user!");
SDNode *User = *UI;
}
}
}
-
+
// If the user is a MFCR instruction, we know this is safe. Otherwise we
// give up for right now.
if (FlagUser->getOpcode() == PPCISD::MFCR)
SDValue LHS = N->getOperand(2), RHS = N->getOperand(3);
int CompareOpc;
bool isDot;
-
+
if (LHS.getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
isa<ConstantSDNode>(RHS) && (CC == ISD::SETEQ || CC == ISD::SETNE) &&
getAltivecCompareInfo(LHS, CompareOpc, isDot)) {
assert(isDot && "Can't compare against a vector result!");
-
+
// If this is a comparison against something other than 0/1, then we know
// that the condition is never/always true.
unsigned Val = cast<ConstantSDNode>(RHS)->getZExtValue();
return DAG.getNode(ISD::BR, dl, MVT::Other,
N->getOperand(0), N->getOperand(4));
}
-
+
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
-
+
// Create the PPCISD altivec 'dot' comparison node.
std::vector<MVT> VTs;
SDValue Ops[] = {
VTs.push_back(LHS.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
-
+
// Unpack the result based on how the target uses it.
PPC::Predicate CompOpc;
switch (cast<ConstantSDNode>(LHS.getOperand(1))->getZExtValue()) {
break;
}
}
-
+
return SDValue();
}
void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
const APInt &Mask,
- APInt &KnownZero,
+ APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
case Intrinsic::ppc_altivec_vcmpgtuw_p:
KnownZero = ~1U; // All bits but the low one are known to be zero.
break;
- }
+ }
}
}
}
/// getConstraintType - Given a constraint, return the type of
/// constraint it is for this target.
-PPCTargetLowering::ConstraintType
+PPCTargetLowering::ConstraintType
PPCTargetLowering::getConstraintType(const std::string &Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
return TargetLowering::getConstraintType(Constraint);
}
-std::pair<unsigned, const TargetRegisterClass*>
+std::pair<unsigned, const TargetRegisterClass*>
PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
MVT VT) const {
if (Constraint.size() == 1) {
else if (VT == MVT::f64)
return std::make_pair(0U, PPC::F8RCRegisterClass);
break;
- case 'v':
+ case 'v':
return std::make_pair(0U, PPC::VRRCRegisterClass);
case 'y': // crrc
return std::make_pair(0U, PPC::CRRCRegisterClass);
}
}
-
+
return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
}
if ((int)Value > 0 && isPowerOf2_32(Value))
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
- case 'O': // "O" is the constant zero.
+ case 'O': // "O" is the constant zero.
if (Value == 0)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
break;
}
}
-
+
if (Result.getNode()) {
Ops.push_back(Result);
return;
}
-
+
// Handle standard constraint letters.
TargetLowering::LowerAsmOperandForConstraint(Op, Letter, hasMemory, Ops, DAG);
}
// isLegalAddressingMode - Return true if the addressing mode represented
// by AM is legal for this target, for a load/store of the specified type.
-bool PPCTargetLowering::isLegalAddressingMode(const AddrMode &AM,
+bool PPCTargetLowering::isLegalAddressingMode(const AddrMode &AM,
const Type *Ty) const {
// FIXME: PPC does not allow r+i addressing modes for vectors!
-
+
// PPC allows a sign-extended 16-bit immediate field.
if (AM.BaseOffs <= -(1LL << 16) || AM.BaseOffs >= (1LL << 16)-1)
return false;
-
+
// No global is ever allowed as a base.
if (AM.BaseGV)
return false;
-
- // PPC only support r+r,
+
+ // PPC only support r+r,
switch (AM.Scale) {
case 0: // "r+i" or just "i", depending on HasBaseReg.
break;
// No other scales are supported.
return false;
}
-
+
return true;
}
}
bool PPCTargetLowering::isLegalAddressImmediate(llvm::GlobalValue* GV) const {
- return false;
+ return false;
}
SDValue PPCTargetLowering::LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- // Depths > 0 not supported yet!
+ // Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() > 0)
return SDValue();
// Make sure the function really does not optimize away the store of the RA
// to the stack.
FuncInfo->setLRStoreRequired();
- return DAG.getLoad(getPointerTy(), dl,
+ return DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), RetAddrFI, NULL, 0);
}
SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- // Depths > 0 not supported yet!
+ // Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() > 0)
return SDValue();
-
+
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
-
+
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
- bool is31 = (NoFramePointerElim || MFI->hasVarSizedObjects())
+ bool is31 = (NoFramePointerElim || MFI->hasVarSizedObjects())
&& MFI->getStackSize();
if (isPPC64)