diff options
Diffstat (limited to 'lib/Target/SystemZ/SystemZISelLowering.cpp')
| -rw-r--r-- | lib/Target/SystemZ/SystemZISelLowering.cpp | 104 |
1 files changed, 92 insertions, 12 deletions
diff --git a/lib/Target/SystemZ/SystemZISelLowering.cpp b/lib/Target/SystemZ/SystemZISelLowering.cpp index 2801141cd951..2d916d2e1521 100644 --- a/lib/Target/SystemZ/SystemZISelLowering.cpp +++ b/lib/Target/SystemZ/SystemZISelLowering.cpp @@ -101,7 +101,10 @@ SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM, addRegisterClass(MVT::f32, &SystemZ::FP32BitRegClass); addRegisterClass(MVT::f64, &SystemZ::FP64BitRegClass); } - addRegisterClass(MVT::f128, &SystemZ::FP128BitRegClass); + if (Subtarget.hasVectorEnhancements1()) + addRegisterClass(MVT::f128, &SystemZ::VR128BitRegClass); + else + addRegisterClass(MVT::f128, &SystemZ::FP128BitRegClass); if (Subtarget.hasVector()) { addRegisterClass(MVT::v16i8, &SystemZ::VR128BitRegClass); @@ -316,7 +319,10 @@ SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM, setOperationAction(ISD::AND, VT, Legal); setOperationAction(ISD::OR, VT, Legal); setOperationAction(ISD::XOR, VT, Legal); - setOperationAction(ISD::CTPOP, VT, Custom); + if (Subtarget.hasVectorEnhancements1()) + setOperationAction(ISD::CTPOP, VT, Legal); + else + setOperationAction(ISD::CTPOP, VT, Custom); setOperationAction(ISD::CTTZ, VT, Legal); setOperationAction(ISD::CTLZ, VT, Legal); @@ -414,10 +420,60 @@ SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM, setOperationAction(ISD::FROUND, MVT::v2f64, Legal); } + // The vector enhancements facility 1 has instructions for these. + if (Subtarget.hasVectorEnhancements1()) { + setOperationAction(ISD::FADD, MVT::v4f32, Legal); + setOperationAction(ISD::FNEG, MVT::v4f32, Legal); + setOperationAction(ISD::FSUB, MVT::v4f32, Legal); + setOperationAction(ISD::FMUL, MVT::v4f32, Legal); + setOperationAction(ISD::FMA, MVT::v4f32, Legal); + setOperationAction(ISD::FDIV, MVT::v4f32, Legal); + setOperationAction(ISD::FABS, MVT::v4f32, Legal); + setOperationAction(ISD::FSQRT, MVT::v4f32, Legal); + setOperationAction(ISD::FRINT, MVT::v4f32, Legal); + setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Legal); + setOperationAction(ISD::FFLOOR, MVT::v4f32, Legal); + setOperationAction(ISD::FCEIL, MVT::v4f32, Legal); + setOperationAction(ISD::FTRUNC, MVT::v4f32, Legal); + setOperationAction(ISD::FROUND, MVT::v4f32, Legal); + + setOperationAction(ISD::FMAXNUM, MVT::f64, Legal); + setOperationAction(ISD::FMAXNAN, MVT::f64, Legal); + setOperationAction(ISD::FMINNUM, MVT::f64, Legal); + setOperationAction(ISD::FMINNAN, MVT::f64, Legal); + + setOperationAction(ISD::FMAXNUM, MVT::v2f64, Legal); + setOperationAction(ISD::FMAXNAN, MVT::v2f64, Legal); + setOperationAction(ISD::FMINNUM, MVT::v2f64, Legal); + setOperationAction(ISD::FMINNAN, MVT::v2f64, Legal); + + setOperationAction(ISD::FMAXNUM, MVT::f32, Legal); + setOperationAction(ISD::FMAXNAN, MVT::f32, Legal); + setOperationAction(ISD::FMINNUM, MVT::f32, Legal); + setOperationAction(ISD::FMINNAN, MVT::f32, Legal); + + setOperationAction(ISD::FMAXNUM, MVT::v4f32, Legal); + setOperationAction(ISD::FMAXNAN, MVT::v4f32, Legal); + setOperationAction(ISD::FMINNUM, MVT::v4f32, Legal); + setOperationAction(ISD::FMINNAN, MVT::v4f32, Legal); + + setOperationAction(ISD::FMAXNUM, MVT::f128, Legal); + setOperationAction(ISD::FMAXNAN, MVT::f128, Legal); + setOperationAction(ISD::FMINNUM, MVT::f128, Legal); + setOperationAction(ISD::FMINNAN, MVT::f128, Legal); + } + // We have fused multiply-addition for f32 and f64 but not f128. setOperationAction(ISD::FMA, MVT::f32, Legal); setOperationAction(ISD::FMA, MVT::f64, Legal); - setOperationAction(ISD::FMA, MVT::f128, Expand); + if (Subtarget.hasVectorEnhancements1()) + setOperationAction(ISD::FMA, MVT::f128, Legal); + else + setOperationAction(ISD::FMA, MVT::f128, Expand); + + // We don't have a copysign instruction on vector registers. + if (Subtarget.hasVectorEnhancements1()) + setOperationAction(ISD::FCOPYSIGN, MVT::f128, Expand); // Needed so that we don't try to implement f128 constant loads using // a load-and-extend of a f80 constant (in cases where the constant @@ -425,6 +481,12 @@ SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM, for (MVT VT : MVT::fp_valuetypes()) setLoadExtAction(ISD::EXTLOAD, VT, MVT::f80, Expand); + // We don't have extending load instruction on vector registers. + if (Subtarget.hasVectorEnhancements1()) { + setLoadExtAction(ISD::EXTLOAD, MVT::f128, MVT::f32, Expand); + setLoadExtAction(ISD::EXTLOAD, MVT::f128, MVT::f64, Expand); + } + // Floating-point truncation and stores need to be done separately. setTruncStoreAction(MVT::f64, MVT::f32, Expand); setTruncStoreAction(MVT::f128, MVT::f32, Expand); @@ -489,7 +551,7 @@ bool SystemZTargetLowering::isFMAFasterThanFMulAndFAdd(EVT VT) const { case MVT::f64: return true; case MVT::f128: - return false; + return Subtarget.hasVectorEnhancements1(); default: break; } @@ -1462,21 +1524,25 @@ static bool isIntrinsicWithCC(SDValue Op, unsigned &Opcode, unsigned &CCValid) { return true; case Intrinsic::s390_vfcedbs: + case Intrinsic::s390_vfcesbs: Opcode = SystemZISD::VFCMPES; CCValid = SystemZ::CCMASK_VCMP; return true; case Intrinsic::s390_vfchdbs: + case Intrinsic::s390_vfchsbs: Opcode = SystemZISD::VFCMPHS; CCValid = SystemZ::CCMASK_VCMP; return true; case Intrinsic::s390_vfchedbs: + case Intrinsic::s390_vfchesbs: Opcode = SystemZISD::VFCMPHES; CCValid = SystemZ::CCMASK_VCMP; return true; case Intrinsic::s390_vftcidb: + case Intrinsic::s390_vftcisb: Opcode = SystemZISD::VFTCI; CCValid = SystemZ::CCMASK_VCMP; return true; @@ -2316,11 +2382,15 @@ static SDValue expandV4F32ToV2F64(SelectionDAG &DAG, int Start, const SDLoc &DL, // Build a comparison of vectors CmpOp0 and CmpOp1 using opcode Opcode, // producing a result of type VT. -static SDValue getVectorCmp(SelectionDAG &DAG, unsigned Opcode, const SDLoc &DL, - EVT VT, SDValue CmpOp0, SDValue CmpOp1) { - // There is no hardware support for v4f32, so extend the vector into - // two v2f64s and compare those. - if (CmpOp0.getValueType() == MVT::v4f32) { +SDValue SystemZTargetLowering::getVectorCmp(SelectionDAG &DAG, unsigned Opcode, + const SDLoc &DL, EVT VT, + SDValue CmpOp0, + SDValue CmpOp1) const { + // There is no hardware support for v4f32 (unless we have the vector + // enhancements facility 1), so extend the vector into two v2f64s + // and compare those. + if (CmpOp0.getValueType() == MVT::v4f32 && + !Subtarget.hasVectorEnhancements1()) { SDValue H0 = expandV4F32ToV2F64(DAG, 0, DL, CmpOp0); SDValue L0 = expandV4F32ToV2F64(DAG, 2, DL, CmpOp0); SDValue H1 = expandV4F32ToV2F64(DAG, 0, DL, CmpOp1); @@ -2334,9 +2404,11 @@ static SDValue getVectorCmp(SelectionDAG &DAG, unsigned Opcode, const SDLoc &DL, // Lower a vector comparison of type CC between CmpOp0 and CmpOp1, producing // an integer mask of type VT. -static SDValue lowerVectorSETCC(SelectionDAG &DAG, const SDLoc &DL, EVT VT, - ISD::CondCode CC, SDValue CmpOp0, - SDValue CmpOp1) { +SDValue SystemZTargetLowering::lowerVectorSETCC(SelectionDAG &DAG, + const SDLoc &DL, EVT VT, + ISD::CondCode CC, + SDValue CmpOp0, + SDValue CmpOp1) const { bool IsFP = CmpOp0.getValueType().isFloatingPoint(); bool Invert = false; SDValue Cmp; @@ -2960,6 +3032,12 @@ SDValue SystemZTargetLowering::lowerSMUL_LOHI(SDValue Op, // We define this so that it can be used for constant division. lowerMUL_LOHI32(DAG, DL, ISD::SIGN_EXTEND, Op.getOperand(0), Op.getOperand(1), Ops[1], Ops[0]); + else if (Subtarget.hasMiscellaneousExtensions2()) + // SystemZISD::SMUL_LOHI returns the low result in the odd register and + // the high result in the even register. ISD::SMUL_LOHI is defined to + // return the low half first, so the results are in reverse order. + lowerGR128Binary(DAG, DL, VT, SystemZISD::SMUL_LOHI, + Op.getOperand(0), Op.getOperand(1), Ops[1], Ops[0]); else { // Do a full 128-bit multiplication based on SystemZISD::UMUL_LOHI: // @@ -4658,6 +4736,7 @@ const char *SystemZTargetLowering::getTargetNodeName(unsigned Opcode) const { OPCODE(SELECT_CCMASK); OPCODE(ADJDYNALLOC); OPCODE(POPCNT); + OPCODE(SMUL_LOHI); OPCODE(UMUL_LOHI); OPCODE(SDIVREM); OPCODE(UDIVREM); @@ -6118,6 +6197,7 @@ MachineBasicBlock *SystemZTargetLowering::EmitInstrWithCustomInserter( case SystemZ::SelectF32: case SystemZ::SelectF64: case SystemZ::SelectF128: + case SystemZ::SelectVR128: return emitSelect(MI, MBB, 0); case SystemZ::CondStore8Mux: |