blob: 798e8784405f732902e2653a365e4d0a66d70888 [file] [log] [blame] [edit]
//===- MipsSEISelLowering.cpp - MipsSE DAG Lowering Interface -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// Subclass of MipsTargetLowering specialized for mips32/64.
//
//===----------------------------------------------------------------------===//
#include "MipsSEISelLowering.h"
#include "MipsMachineFunction.h"
#include "MipsRegisterInfo.h"
#include "MipsSubtarget.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Triple.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/IntrinsicsMips.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <iterator>
#include <utility>
using namespace llvm;
#define DEBUG_TYPE "mips-isel"
static cl::opt<bool>
UseMipsTailCalls("mips-tail-calls", cl::Hidden,
cl::desc("MIPS: permit tail calls."), cl::init(false));
static cl::opt<bool> NoDPLoadStore("mno-ldc1-sdc1", cl::init(false),
cl::desc("Expand double precision loads and "
"stores to their single precision "
"counterparts"));
MipsSETargetLowering::MipsSETargetLowering(const MipsTargetMachine &TM,
const MipsSubtarget &STI)
: MipsTargetLowering(TM, STI) {
// Set up the register classes
addRegisterClass(MVT::i32, &Mips::GPR32RegClass);
if (Subtarget.isGP64bit())
addRegisterClass(MVT::i64, &Mips::GPR64RegClass);
if (Subtarget.hasDSP() || Subtarget.hasMSA()) {
// Expand all truncating stores and extending loads.
for (MVT VT0 : MVT::fixedlen_vector_valuetypes()) {
for (MVT VT1 : MVT::fixedlen_vector_valuetypes()) {
setTruncStoreAction(VT0, VT1, Expand);
setLoadExtAction(ISD::SEXTLOAD, VT0, VT1, Expand);
setLoadExtAction(ISD::ZEXTLOAD, VT0, VT1, Expand);
setLoadExtAction(ISD::EXTLOAD, VT0, VT1, Expand);
}
}
}
if (Subtarget.hasDSP()) {
MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8};
for (unsigned i = 0; i < array_lengthof(VecTys); ++i) {
addRegisterClass(VecTys[i], &Mips::DSPRRegClass);
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, VecTys[i], Expand);
setOperationAction(ISD::ADD, VecTys[i], Legal);
setOperationAction(ISD::SUB, VecTys[i], Legal);
setOperationAction(ISD::LOAD, VecTys[i], Legal);
setOperationAction(ISD::STORE, VecTys[i], Legal);
setOperationAction(ISD::BITCAST, VecTys[i], Legal);
}
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::SRL);
setTargetDAGCombine(ISD::SETCC);
setTargetDAGCombine(ISD::VSELECT);
if (Subtarget.hasMips32r2()) {
setOperationAction(ISD::ADDC, MVT::i32, Legal);
setOperationAction(ISD::ADDE, MVT::i32, Legal);
}
}
if (Subtarget.hasDSPR2())
setOperationAction(ISD::MUL, MVT::v2i16, Legal);
if (Subtarget.hasMSA()) {
addMSAIntType(MVT::v16i8, &Mips::MSA128BRegClass);
addMSAIntType(MVT::v8i16, &Mips::MSA128HRegClass);
addMSAIntType(MVT::v4i32, &Mips::MSA128WRegClass);
addMSAIntType(MVT::v2i64, &Mips::MSA128DRegClass);
addMSAFloatType(MVT::v8f16, &Mips::MSA128HRegClass);
addMSAFloatType(MVT::v4f32, &Mips::MSA128WRegClass);
addMSAFloatType(MVT::v2f64, &Mips::MSA128DRegClass);
// f16 is a storage-only type, always promote it to f32.
addRegisterClass(MVT::f16, &Mips::MSA128HRegClass);
setOperationAction(ISD::SETCC, MVT::f16, Promote);
setOperationAction(ISD::BR_CC, MVT::f16, Promote);
setOperationAction(ISD::SELECT_CC, MVT::f16, Promote);
setOperationAction(ISD::SELECT, MVT::f16, Promote);
setOperationAction(ISD::FADD, MVT::f16, Promote);
setOperationAction(ISD::FSUB, MVT::f16, Promote);
setOperationAction(ISD::FMUL, MVT::f16, Promote);
setOperationAction(ISD::FDIV, MVT::f16, Promote);
setOperationAction(ISD::FREM, MVT::f16, Promote);
setOperationAction(ISD::FMA, MVT::f16, Promote);
setOperationAction(ISD::FNEG, MVT::f16, Promote);
setOperationAction(ISD::FABS, MVT::f16, Promote);
setOperationAction(ISD::FCEIL, MVT::f16, Promote);
setOperationAction(ISD::FCOPYSIGN, MVT::f16, Promote);
setOperationAction(ISD::FCOS, MVT::f16, Promote);
setOperationAction(ISD::FP_EXTEND, MVT::f16, Promote);
setOperationAction(ISD::FFLOOR, MVT::f16, Promote);
setOperationAction(ISD::FNEARBYINT, MVT::f16, Promote);
setOperationAction(ISD::FPOW, MVT::f16, Promote);
setOperationAction(ISD::FPOWI, MVT::f16, Promote);
setOperationAction(ISD::FRINT, MVT::f16, Promote);
setOperationAction(ISD::FSIN, MVT::f16, Promote);
setOperationAction(ISD::FSINCOS, MVT::f16, Promote);
setOperationAction(ISD::FSQRT, MVT::f16, Promote);
setOperationAction(ISD::FEXP, MVT::f16, Promote);
setOperationAction(ISD::FEXP2, MVT::f16, Promote);
setOperationAction(ISD::FLOG, MVT::f16, Promote);
setOperationAction(ISD::FLOG2, MVT::f16, Promote);
setOperationAction(ISD::FLOG10, MVT::f16, Promote);
setOperationAction(ISD::FROUND, MVT::f16, Promote);
setOperationAction(ISD::FTRUNC, MVT::f16, Promote);
setOperationAction(ISD::FMINNUM, MVT::f16, Promote);
setOperationAction(ISD::FMAXNUM, MVT::f16, Promote);
setOperationAction(ISD::FMINIMUM, MVT::f16, Promote);
setOperationAction(ISD::FMAXIMUM, MVT::f16, Promote);
setTargetDAGCombine(ISD::AND);
setTargetDAGCombine(ISD::OR);
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::VSELECT);
setTargetDAGCombine(ISD::XOR);
}
if (!Subtarget.useSoftFloat()) {
addRegisterClass(MVT::f32, &Mips::FGR32RegClass);
// When dealing with single precision only, use libcalls
if (!Subtarget.isSingleFloat()) {
if (Subtarget.isFP64bit())
addRegisterClass(MVT::f64, &Mips::FGR64RegClass);
else
addRegisterClass(MVT::f64, &Mips::AFGR64RegClass);
}
}
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Custom);
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Custom);
setOperationAction(ISD::MULHS, MVT::i32, Custom);
setOperationAction(ISD::MULHU, MVT::i32, Custom);
if (Subtarget.hasCnMips())
setOperationAction(ISD::MUL, MVT::i64, Legal);
else if (Subtarget.isGP64bit())
setOperationAction(ISD::MUL, MVT::i64, Custom);
if (Subtarget.isGP64bit()) {
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Custom);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Custom);
setOperationAction(ISD::MULHS, MVT::i64, Custom);
setOperationAction(ISD::MULHU, MVT::i64, Custom);
setOperationAction(ISD::SDIVREM, MVT::i64, Custom);
setOperationAction(ISD::UDIVREM, MVT::i64, Custom);
}
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i64, Custom);
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i64, Custom);
setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
setOperationAction(ISD::LOAD, MVT::i32, Custom);
setOperationAction(ISD::STORE, MVT::i32, Custom);
setTargetDAGCombine(ISD::MUL);
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
if (Subtarget.hasMips32r2() && !Subtarget.useSoftFloat() &&
!Subtarget.hasMips64()) {
setOperationAction(ISD::BITCAST, MVT::i64, Custom);
}
if (NoDPLoadStore) {
setOperationAction(ISD::LOAD, MVT::f64, Custom);
setOperationAction(ISD::STORE, MVT::f64, Custom);
}
if (Subtarget.hasMips32r6()) {
// MIPS32r6 replaces the accumulator-based multiplies with a three register
// instruction
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::MUL, MVT::i32, Legal);
setOperationAction(ISD::MULHS, MVT::i32, Legal);
setOperationAction(ISD::MULHU, MVT::i32, Legal);
// MIPS32r6 replaces the accumulator-based division/remainder with separate
// three register division and remainder instructions.
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
setOperationAction(ISD::SDIV, MVT::i32, Legal);
setOperationAction(ISD::UDIV, MVT::i32, Legal);
setOperationAction(ISD::SREM, MVT::i32, Legal);
setOperationAction(ISD::UREM, MVT::i32, Legal);
// MIPS32r6 replaces conditional moves with an equivalent that removes the
// need for three GPR read ports.
setOperationAction(ISD::SETCC, MVT::i32, Legal);
setOperationAction(ISD::SELECT, MVT::i32, Legal);
setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
setOperationAction(ISD::SETCC, MVT::f32, Legal);
setOperationAction(ISD::SELECT, MVT::f32, Legal);
setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
assert(Subtarget.isFP64bit() && "FR=1 is required for MIPS32r6");
setOperationAction(ISD::SETCC, MVT::f64, Legal);
setOperationAction(ISD::SELECT, MVT::f64, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
setOperationAction(ISD::BRCOND, MVT::Other, Legal);
// Floating point > and >= are supported via < and <=
setCondCodeAction(ISD::SETOGE, MVT::f32, Expand);
setCondCodeAction(ISD::SETOGT, MVT::f32, Expand);
setCondCodeAction(ISD::SETUGE, MVT::f32, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f32, Expand);
setCondCodeAction(ISD::SETOGE, MVT::f64, Expand);
setCondCodeAction(ISD::SETOGT, MVT::f64, Expand);
setCondCodeAction(ISD::SETUGE, MVT::f64, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f64, Expand);
}
if (Subtarget.hasMips64r6()) {
// MIPS64r6 replaces the accumulator-based multiplies with a three register
// instruction
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::MUL, MVT::i64, Legal);
setOperationAction(ISD::MULHS, MVT::i64, Legal);
setOperationAction(ISD::MULHU, MVT::i64, Legal);
// MIPS32r6 replaces the accumulator-based division/remainder with separate
// three register division and remainder instructions.
setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
setOperationAction(ISD::SDIV, MVT::i64, Legal);
setOperationAction(ISD::UDIV, MVT::i64, Legal);
setOperationAction(ISD::SREM, MVT::i64, Legal);
setOperationAction(ISD::UREM, MVT::i64, Legal);
// MIPS64r6 replaces conditional moves with an equivalent that removes the
// need for three GPR read ports.
setOperationAction(ISD::SETCC, MVT::i64, Legal);
setOperationAction(ISD::SELECT, MVT::i64, Legal);
setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
}
computeRegisterProperties(Subtarget.getRegisterInfo());
}
const MipsTargetLowering *
llvm::createMipsSETargetLowering(const MipsTargetMachine &TM,
const MipsSubtarget &STI) {
return new MipsSETargetLowering(TM, STI);
}
const TargetRegisterClass *
MipsSETargetLowering::getRepRegClassFor(MVT VT) const {
if (VT == MVT::Untyped)
return Subtarget.hasDSP() ? &Mips::ACC64DSPRegClass : &Mips::ACC64RegClass;
return TargetLowering::getRepRegClassFor(VT);
}
// Enable MSA support for the given integer type and Register class.
void MipsSETargetLowering::
addMSAIntType(MVT::SimpleValueType Ty, const TargetRegisterClass *RC) {
addRegisterClass(Ty, RC);
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, Ty, Expand);
setOperationAction(ISD::BITCAST, Ty, Legal);
setOperationAction(ISD::LOAD, Ty, Legal);
setOperationAction(ISD::STORE, Ty, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, Ty, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, Ty, Legal);
setOperationAction(ISD::BUILD_VECTOR, Ty, Custom);
setOperationAction(ISD::UNDEF, Ty, Legal);
setOperationAction(ISD::ADD, Ty, Legal);
setOperationAction(ISD::AND, Ty, Legal);
setOperationAction(ISD::CTLZ, Ty, Legal);
setOperationAction(ISD::CTPOP, Ty, Legal);
setOperationAction(ISD::MUL, Ty, Legal);
setOperationAction(ISD::OR, Ty, Legal);
setOperationAction(ISD::SDIV, Ty, Legal);
setOperationAction(ISD::SREM, Ty, Legal);
setOperationAction(ISD::SHL, Ty, Legal);
setOperationAction(ISD::SRA, Ty, Legal);
setOperationAction(ISD::SRL, Ty, Legal);
setOperationAction(ISD::SUB, Ty, Legal);
setOperationAction(ISD::SMAX, Ty, Legal);
setOperationAction(ISD::SMIN, Ty, Legal);
setOperationAction(ISD::UDIV, Ty, Legal);
setOperationAction(ISD::UREM, Ty, Legal);
setOperationAction(ISD::UMAX, Ty, Legal);
setOperationAction(ISD::UMIN, Ty, Legal);
setOperationAction(ISD::VECTOR_SHUFFLE, Ty, Custom);
setOperationAction(ISD::VSELECT, Ty, Legal);
setOperationAction(ISD::XOR, Ty, Legal);
if (Ty == MVT::v4i32 || Ty == MVT::v2i64) {
setOperationAction(ISD::FP_TO_SINT, Ty, Legal);
setOperationAction(ISD::FP_TO_UINT, Ty, Legal);
setOperationAction(ISD::SINT_TO_FP, Ty, Legal);
setOperationAction(ISD::UINT_TO_FP, Ty, Legal);
}
setOperationAction(ISD::SETCC, Ty, Legal);
setCondCodeAction(ISD::SETNE, Ty, Expand);
setCondCodeAction(ISD::SETGE, Ty, Expand);
setCondCodeAction(ISD::SETGT, Ty, Expand);
setCondCodeAction(ISD::SETUGE, Ty, Expand);
setCondCodeAction(ISD::SETUGT, Ty, Expand);
}
// Enable MSA support for the given floating-point type and Register class.
void MipsSETargetLowering::
addMSAFloatType(MVT::SimpleValueType Ty, const TargetRegisterClass *RC) {
addRegisterClass(Ty, RC);
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, Ty, Expand);
setOperationAction(ISD::LOAD, Ty, Legal);
setOperationAction(ISD::STORE, Ty, Legal);
setOperationAction(ISD::BITCAST, Ty, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, Ty, Legal);
setOperationAction(ISD::INSERT_VECTOR_ELT, Ty, Legal);
setOperationAction(ISD::BUILD_VECTOR, Ty, Custom);
if (Ty != MVT::v8f16) {
setOperationAction(ISD::FABS, Ty, Legal);
setOperationAction(ISD::FADD, Ty, Legal);
setOperationAction(ISD::FDIV, Ty, Legal);
setOperationAction(ISD::FEXP2, Ty, Legal);
setOperationAction(ISD::FLOG2, Ty, Legal);
setOperationAction(ISD::FMA, Ty, Legal);
setOperationAction(ISD::FMUL, Ty, Legal);
setOperationAction(ISD::FRINT, Ty, Legal);
setOperationAction(ISD::FSQRT, Ty, Legal);
setOperationAction(ISD::FSUB, Ty, Legal);
setOperationAction(ISD::VSELECT, Ty, Legal);
setOperationAction(ISD::SETCC, Ty, Legal);
setCondCodeAction(ISD::SETOGE, Ty, Expand);
setCondCodeAction(ISD::SETOGT, Ty, Expand);
setCondCodeAction(ISD::SETUGE, Ty, Expand);
setCondCodeAction(ISD::SETUGT, Ty, Expand);
setCondCodeAction(ISD::SETGE, Ty, Expand);
setCondCodeAction(ISD::SETGT, Ty, Expand);
}
}
SDValue MipsSETargetLowering::lowerSELECT(SDValue Op, SelectionDAG &DAG) const {
if(!Subtarget.hasMips32r6())
return MipsTargetLowering::LowerOperation(Op, DAG);
EVT ResTy = Op->getValueType(0);
SDLoc DL(Op);
// Although MTC1_D64 takes an i32 and writes an f64, the upper 32 bits of the
// floating point register are undefined. Not really an issue as sel.d, which
// is produced from an FSELECT node, only looks at bit 0.
SDValue Tmp = DAG.getNode(MipsISD::MTC1_D64, DL, MVT::f64, Op->getOperand(0));
return DAG.getNode(MipsISD::FSELECT, DL, ResTy, Tmp, Op->getOperand(1),
Op->getOperand(2));
}
bool MipsSETargetLowering::allowsMisalignedMemoryAccesses(
EVT VT, unsigned, unsigned, MachineMemOperand::Flags, bool *Fast) const {
MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy;
if (Subtarget.systemSupportsUnalignedAccess()) {
// MIPS32r6/MIPS64r6 is required to support unaligned access. It's
// implementation defined whether this is handled by hardware, software, or
// a hybrid of the two but it's expected that most implementations will
// handle the majority of cases in hardware.
if (Fast)
*Fast = true;
return true;
}
switch (SVT) {
case MVT::i64:
case MVT::i32:
if (Fast)
*Fast = true;
return true;
default:
return false;
}
}
SDValue MipsSETargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
switch(Op.getOpcode()) {
case ISD::LOAD: return lowerLOAD(Op, DAG);
case ISD::STORE: return lowerSTORE(Op, DAG);
case ISD::SMUL_LOHI: return lowerMulDiv(Op, MipsISD::Mult, true, true, DAG);
case ISD::UMUL_LOHI: return lowerMulDiv(Op, MipsISD::Multu, true, true, DAG);
case ISD::MULHS: return lowerMulDiv(Op, MipsISD::Mult, false, true, DAG);
case ISD::MULHU: return lowerMulDiv(Op, MipsISD::Multu, false, true, DAG);
case ISD::MUL: return lowerMulDiv(Op, MipsISD::Mult, true, false, DAG);
case ISD::SDIVREM: return lowerMulDiv(Op, MipsISD::DivRem, true, true, DAG);
case ISD::UDIVREM: return lowerMulDiv(Op, MipsISD::DivRemU, true, true,
DAG);
case ISD::INTRINSIC_WO_CHAIN: return lowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::INTRINSIC_W_CHAIN: return lowerINTRINSIC_W_CHAIN(Op, DAG);
case ISD::INTRINSIC_VOID: return lowerINTRINSIC_VOID(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT: return lowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::BUILD_VECTOR: return lowerBUILD_VECTOR(Op, DAG);
case ISD::VECTOR_SHUFFLE: return lowerVECTOR_SHUFFLE(Op, DAG);
case ISD::SELECT: return lowerSELECT(Op, DAG);
case ISD::BITCAST: return lowerBITCAST(Op, DAG);
}
return MipsTargetLowering::LowerOperation(Op, DAG);
}
// Fold zero extensions into MipsISD::VEXTRACT_[SZ]EXT_ELT
//
// Performs the following transformations:
// - Changes MipsISD::VEXTRACT_[SZ]EXT_ELT to zero extension if its
// sign/zero-extension is completely overwritten by the new one performed by
// the ISD::AND.
// - Removes redundant zero extensions performed by an ISD::AND.
static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (!Subtarget.hasMSA())
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
unsigned Op0Opcode = Op0->getOpcode();
// (and (MipsVExtract[SZ]Ext $a, $b, $c), imm:$d)
// where $d + 1 == 2^n and n == 32
// or $d + 1 == 2^n and n <= 32 and ZExt
// -> (MipsVExtractZExt $a, $b, $c)
if (Op0Opcode == MipsISD::VEXTRACT_SEXT_ELT ||
Op0Opcode == MipsISD::VEXTRACT_ZEXT_ELT) {
ConstantSDNode *Mask = dyn_cast<ConstantSDNode>(Op1);
if (!Mask)
return SDValue();
int32_t Log2IfPositive = (Mask->getAPIntValue() + 1).exactLogBase2();
if (Log2IfPositive <= 0)
return SDValue(); // Mask+1 is not a power of 2
SDValue Op0Op2 = Op0->getOperand(2);
EVT ExtendTy = cast<VTSDNode>(Op0Op2)->getVT();
unsigned ExtendTySize = ExtendTy.getSizeInBits();
unsigned Log2 = Log2IfPositive;
if ((Op0Opcode == MipsISD::VEXTRACT_ZEXT_ELT && Log2 >= ExtendTySize) ||
Log2 == ExtendTySize) {
SDValue Ops[] = { Op0->getOperand(0), Op0->getOperand(1), Op0Op2 };
return DAG.getNode(MipsISD::VEXTRACT_ZEXT_ELT, SDLoc(Op0),
Op0->getVTList(),
makeArrayRef(Ops, Op0->getNumOperands()));
}
}
return SDValue();
}
// Determine if the specified node is a constant vector splat.
//
// Returns true and sets Imm if:
// * N is a ISD::BUILD_VECTOR representing a constant splat
//
// This function is quite similar to MipsSEDAGToDAGISel::selectVSplat. The
// differences are that it assumes the MSA has already been checked and the
// arbitrary requirement for a maximum of 32-bit integers isn't applied (and
// must not be in order for binsri.d to be selectable).
static bool isVSplat(SDValue N, APInt &Imm, bool IsLittleEndian) {
BuildVectorSDNode *Node = dyn_cast<BuildVectorSDNode>(N.getNode());
if (!Node)
return false;
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (!Node->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs,
8, !IsLittleEndian))
return false;
Imm = SplatValue;
return true;
}
// Test whether the given node is an all-ones build_vector.
static bool isVectorAllOnes(SDValue N) {
// Look through bitcasts. Endianness doesn't matter because we are looking
// for an all-ones value.
if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0);
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N);
if (!BVN)
return false;
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
// Endianness doesn't matter in this context because we are looking for
// an all-ones value.
if (BVN->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs))
return SplatValue.isAllOnesValue();
return false;
}
// Test whether N is the bitwise inverse of OfNode.
static bool isBitwiseInverse(SDValue N, SDValue OfNode) {
if (N->getOpcode() != ISD::XOR)
return false;
if (isVectorAllOnes(N->getOperand(0)))
return N->getOperand(1) == OfNode;
if (isVectorAllOnes(N->getOperand(1)))
return N->getOperand(0) == OfNode;
return false;
}
// Perform combines where ISD::OR is the root node.
//
// Performs the following transformations:
// - (or (and $a, $mask), (and $b, $inv_mask)) => (vselect $mask, $a, $b)
// where $inv_mask is the bitwise inverse of $mask and the 'or' has a 128-bit
// vector type.
static SDValue performORCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (!Subtarget.hasMSA())
return SDValue();
EVT Ty = N->getValueType(0);
if (!Ty.is128BitVector())
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
if (Op0->getOpcode() == ISD::AND && Op1->getOpcode() == ISD::AND) {
SDValue Op0Op0 = Op0->getOperand(0);
SDValue Op0Op1 = Op0->getOperand(1);
SDValue Op1Op0 = Op1->getOperand(0);
SDValue Op1Op1 = Op1->getOperand(1);
bool IsLittleEndian = !Subtarget.isLittle();
SDValue IfSet, IfClr, Cond;
bool IsConstantMask = false;
APInt Mask, InvMask;
// If Op0Op0 is an appropriate mask, try to find it's inverse in either
// Op1Op0, or Op1Op1. Keep track of the Cond, IfSet, and IfClr nodes, while
// looking.
// IfClr will be set if we find a valid match.
if (isVSplat(Op0Op0, Mask, IsLittleEndian)) {
Cond = Op0Op0;
IfSet = Op0Op1;
if (isVSplat(Op1Op0, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
IfClr = Op1Op1;
else if (isVSplat(Op1Op1, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
IfClr = Op1Op0;
IsConstantMask = true;
}
// If IfClr is not yet set, and Op0Op1 is an appropriate mask, try the same
// thing again using this mask.
// IfClr will be set if we find a valid match.
if (!IfClr.getNode() && isVSplat(Op0Op1, Mask, IsLittleEndian)) {
Cond = Op0Op1;
IfSet = Op0Op0;
if (isVSplat(Op1Op0, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
IfClr = Op1Op1;
else if (isVSplat(Op1Op1, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
IfClr = Op1Op0;
IsConstantMask = true;
}
// If IfClr is not yet set, try looking for a non-constant match.
// IfClr will be set if we find a valid match amongst the eight
// possibilities.
if (!IfClr.getNode()) {
if (isBitwiseInverse(Op0Op0, Op1Op0)) {
Cond = Op1Op0;
IfSet = Op1Op1;
IfClr = Op0Op1;
} else if (isBitwiseInverse(Op0Op1, Op1Op0)) {
Cond = Op1Op0;
IfSet = Op1Op1;
IfClr = Op0Op0;
} else if (isBitwiseInverse(Op0Op0, Op1Op1)) {
Cond = Op1Op1;
IfSet = Op1Op0;
IfClr = Op0Op1;
} else if (isBitwiseInverse(Op0Op1, Op1Op1)) {
Cond = Op1Op1;
IfSet = Op1Op0;
IfClr = Op0Op0;
} else if (isBitwiseInverse(Op1Op0, Op0Op0)) {
Cond = Op0Op0;
IfSet = Op0Op1;
IfClr = Op1Op1;
} else if (isBitwiseInverse(Op1Op1, Op0Op0)) {
Cond = Op0Op0;
IfSet = Op0Op1;
IfClr = Op1Op0;
} else if (isBitwiseInverse(Op1Op0, Op0Op1)) {
Cond = Op0Op1;
IfSet = Op0Op0;
IfClr = Op1Op1;
} else if (isBitwiseInverse(Op1Op1, Op0Op1)) {
Cond = Op0Op1;
IfSet = Op0Op0;
IfClr = Op1Op0;
}
}
// At this point, IfClr will be set if we have a valid match.
if (!IfClr.getNode())
return SDValue();
assert(Cond.getNode() && IfSet.getNode());
// Fold degenerate cases.
if (IsConstantMask) {
if (Mask.isAllOnesValue())
return IfSet;
else if (Mask == 0)
return IfClr;
}
// Transform the DAG into an equivalent VSELECT.
return DAG.getNode(ISD::VSELECT, SDLoc(N), Ty, Cond, IfSet, IfClr);
}
return SDValue();
}
static bool shouldTransformMulToShiftsAddsSubs(APInt C, EVT VT,
SelectionDAG &DAG,
const MipsSubtarget &Subtarget) {
// Estimate the number of operations the below transform will turn a
// constant multiply into. The number is approximately equal to the minimal
// number of powers of two that constant can be broken down to by adding
// or subtracting them.
//
// If we have taken more than 12[1] / 8[2] steps to attempt the
// optimization for a native sized value, it is more than likely that this
// optimization will make things worse.
//
// [1] MIPS64 requires 6 instructions at most to materialize any constant,
// multiplication requires at least 4 cycles, but another cycle (or two)
// to retrieve the result from the HI/LO registers.
//
// [2] For MIPS32, more than 8 steps is expensive as the constant could be
// materialized in 2 instructions, multiplication requires at least 4
// cycles, but another cycle (or two) to retrieve the result from the
// HI/LO registers.
//
// TODO:
// - MaxSteps needs to consider the `VT` of the constant for the current
// target.
// - Consider to perform this optimization after type legalization.
// That allows to remove a workaround for types not supported natively.
// - Take in account `-Os, -Oz` flags because this optimization
// increases code size.
unsigned MaxSteps = Subtarget.isABI_O32() ? 8 : 12;
SmallVector<APInt, 16> WorkStack(1, C);
unsigned Steps = 0;
unsigned BitWidth = C.getBitWidth();
while (!WorkStack.empty()) {
APInt Val = WorkStack.pop_back_val();
if (Val == 0 || Val == 1)
continue;
if (Steps >= MaxSteps)
return false;
if (Val.isPowerOf2()) {
++Steps;
continue;
}
APInt Floor = APInt(BitWidth, 1) << Val.logBase2();
APInt Ceil = Val.isNegative() ? APInt(BitWidth, 0)
: APInt(BitWidth, 1) << C.ceilLogBase2();
if ((Val - Floor).ule(Ceil - Val)) {
WorkStack.push_back(Floor);
WorkStack.push_back(Val - Floor);
} else {
WorkStack.push_back(Ceil);
WorkStack.push_back(Ceil - Val);
}
++Steps;
}
// If the value being multiplied is not supported natively, we have to pay
// an additional legalization cost, conservatively assume an increase in the
// cost of 3 instructions per step. This values for this heuristic were
// determined experimentally.
unsigned RegisterSize = DAG.getTargetLoweringInfo()
.getRegisterType(*DAG.getContext(), VT)
.getSizeInBits();
Steps *= (VT.getSizeInBits() != RegisterSize) * 3;
if (Steps > 27)
return false;
return true;
}
static SDValue genConstMult(SDValue X, APInt C, const SDLoc &DL, EVT VT,
EVT ShiftTy, SelectionDAG &DAG) {
// Return 0.
if (C == 0)
return DAG.getConstant(0, DL, VT);
// Return x.
if (C == 1)
return X;
// If c is power of 2, return (shl x, log2(c)).
if (C.isPowerOf2())
return DAG.getNode(ISD::SHL, DL, VT, X,
DAG.getConstant(C.logBase2(), DL, ShiftTy));
unsigned BitWidth = C.getBitWidth();
APInt Floor = APInt(BitWidth, 1) << C.logBase2();
APInt Ceil = C.isNegative() ? APInt(BitWidth, 0) :
APInt(BitWidth, 1) << C.ceilLogBase2();
// If |c - floor_c| <= |c - ceil_c|,
// where floor_c = pow(2, floor(log2(c))) and ceil_c = pow(2, ceil(log2(c))),
// return (add constMult(x, floor_c), constMult(x, c - floor_c)).
if ((C - Floor).ule(Ceil - C)) {
SDValue Op0 = genConstMult(X, Floor, DL, VT, ShiftTy, DAG);
SDValue Op1 = genConstMult(X, C - Floor, DL, VT, ShiftTy, DAG);
return DAG.getNode(ISD::ADD, DL, VT, Op0, Op1);
}
// If |c - floor_c| > |c - ceil_c|,
// return (sub constMult(x, ceil_c), constMult(x, ceil_c - c)).
SDValue Op0 = genConstMult(X, Ceil, DL, VT, ShiftTy, DAG);
SDValue Op1 = genConstMult(X, Ceil - C, DL, VT, ShiftTy, DAG);
return DAG.getNode(ISD::SUB, DL, VT, Op0, Op1);
}
static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
const TargetLowering::DAGCombinerInfo &DCI,
const MipsSETargetLowering *TL,
const MipsSubtarget &Subtarget) {
EVT VT = N->getValueType(0);
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1)))
if (!VT.isVector() && shouldTransformMulToShiftsAddsSubs(
C->getAPIntValue(), VT, DAG, Subtarget))
return genConstMult(N->getOperand(0), C->getAPIntValue(), SDLoc(N), VT,
TL->getScalarShiftAmountTy(DAG.getDataLayout(), VT),
DAG);
return SDValue(N, 0);
}
static SDValue performDSPShiftCombine(unsigned Opc, SDNode *N, EVT Ty,
SelectionDAG &DAG,
const MipsSubtarget &Subtarget) {
// See if this is a vector splat immediate node.
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
unsigned EltSize = Ty.getScalarSizeInBits();
BuildVectorSDNode *BV = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
if (!Subtarget.hasDSP())
return SDValue();
if (!BV ||
!BV->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs,
EltSize, !Subtarget.isLittle()) ||
(SplatBitSize != EltSize) ||
(SplatValue.getZExtValue() >= EltSize))
return SDValue();
SDLoc DL(N);
return DAG.getNode(Opc, DL, Ty, N->getOperand(0),
DAG.getConstant(SplatValue.getZExtValue(), DL, MVT::i32));
}
static SDValue performSHLCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
EVT Ty = N->getValueType(0);
if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8))
return SDValue();
return performDSPShiftCombine(MipsISD::SHLL_DSP, N, Ty, DAG, Subtarget);
}
// Fold sign-extensions into MipsISD::VEXTRACT_[SZ]EXT_ELT for MSA and fold
// constant splats into MipsISD::SHRA_DSP for DSPr2.
//
// Performs the following transformations:
// - Changes MipsISD::VEXTRACT_[SZ]EXT_ELT to sign extension if its
// sign/zero-extension is completely overwritten by the new one performed by
// the ISD::SRA and ISD::SHL nodes.
// - Removes redundant sign extensions performed by an ISD::SRA and ISD::SHL
// sequence.
//
// See performDSPShiftCombine for more information about the transformation
// used for DSPr2.
static SDValue performSRACombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
EVT Ty = N->getValueType(0);
if (Subtarget.hasMSA()) {
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
// (sra (shl (MipsVExtract[SZ]Ext $a, $b, $c), imm:$d), imm:$d)
// where $d + sizeof($c) == 32
// or $d + sizeof($c) <= 32 and SExt
// -> (MipsVExtractSExt $a, $b, $c)
if (Op0->getOpcode() == ISD::SHL && Op1 == Op0->getOperand(1)) {
SDValue Op0Op0 = Op0->getOperand(0);
ConstantSDNode *ShAmount = dyn_cast<ConstantSDNode>(Op1);
if (!ShAmount)
return SDValue();
if (Op0Op0->getOpcode() != MipsISD::VEXTRACT_SEXT_ELT &&
Op0Op0->getOpcode() != MipsISD::VEXTRACT_ZEXT_ELT)
return SDValue();
EVT ExtendTy = cast<VTSDNode>(Op0Op0->getOperand(2))->getVT();
unsigned TotalBits = ShAmount->getZExtValue() + ExtendTy.getSizeInBits();
if (TotalBits == 32 ||
(Op0Op0->getOpcode() == MipsISD::VEXTRACT_SEXT_ELT &&
TotalBits <= 32)) {
SDValue Ops[] = { Op0Op0->getOperand(0), Op0Op0->getOperand(1),
Op0Op0->getOperand(2) };
return DAG.getNode(MipsISD::VEXTRACT_SEXT_ELT, SDLoc(Op0Op0),
Op0Op0->getVTList(),
makeArrayRef(Ops, Op0Op0->getNumOperands()));
}
}
}
if ((Ty != MVT::v2i16) && ((Ty != MVT::v4i8) || !Subtarget.hasDSPR2()))
return SDValue();
return performDSPShiftCombine(MipsISD::SHRA_DSP, N, Ty, DAG, Subtarget);
}
static SDValue performSRLCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
EVT Ty = N->getValueType(0);
if (((Ty != MVT::v2i16) || !Subtarget.hasDSPR2()) && (Ty != MVT::v4i8))
return SDValue();
return performDSPShiftCombine(MipsISD::SHRL_DSP, N, Ty, DAG, Subtarget);
}
static bool isLegalDSPCondCode(EVT Ty, ISD::CondCode CC) {
bool IsV216 = (Ty == MVT::v2i16);
switch (CC) {
case ISD::SETEQ:
case ISD::SETNE: return true;
case ISD::SETLT:
case ISD::SETLE:
case ISD::SETGT:
case ISD::SETGE: return IsV216;
case ISD::SETULT:
case ISD::SETULE:
case ISD::SETUGT:
case ISD::SETUGE: return !IsV216;
default: return false;
}
}
static SDValue performSETCCCombine(SDNode *N, SelectionDAG &DAG) {
EVT Ty = N->getValueType(0);
if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8))
return SDValue();
if (!isLegalDSPCondCode(Ty, cast<CondCodeSDNode>(N->getOperand(2))->get()))
return SDValue();
return DAG.getNode(MipsISD::SETCC_DSP, SDLoc(N), Ty, N->getOperand(0),
N->getOperand(1), N->getOperand(2));
}
static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG) {
EVT Ty = N->getValueType(0);
if (Ty == MVT::v2i16 || Ty == MVT::v4i8) {
SDValue SetCC = N->getOperand(0);
if (SetCC.getOpcode() != MipsISD::SETCC_DSP)
return SDValue();
return DAG.getNode(MipsISD::SELECT_CC_DSP, SDLoc(N), Ty,
SetCC.getOperand(0), SetCC.getOperand(1),
N->getOperand(1), N->getOperand(2), SetCC.getOperand(2));
}
return SDValue();
}
static SDValue performXORCombine(SDNode *N, SelectionDAG &DAG,
const MipsSubtarget &Subtarget) {
EVT Ty = N->getValueType(0);
if (Subtarget.hasMSA() && Ty.is128BitVector() && Ty.isInteger()) {
// Try the following combines:
// (xor (or $a, $b), (build_vector allones))
// (xor (or $a, $b), (bitcast (build_vector allones)))
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
SDValue NotOp;
if (ISD::isBuildVectorAllOnes(Op0.getNode()))
NotOp = Op1;
else if (ISD::isBuildVectorAllOnes(Op1.getNode()))
NotOp = Op0;
else
return SDValue();
if (NotOp->getOpcode() == ISD::OR)
return DAG.getNode(MipsISD::VNOR, SDLoc(N), Ty, NotOp->getOperand(0),
NotOp->getOperand(1));
}
return SDValue();
}
SDValue
MipsSETargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDValue Val;
switch (N->getOpcode()) {
case ISD::AND:
Val = performANDCombine(N, DAG, DCI, Subtarget);
break;
case ISD::OR:
Val = performORCombine(N, DAG, DCI, Subtarget);
break;
case ISD::MUL:
return performMULCombine(N, DAG, DCI, this, Subtarget);
case ISD::SHL:
Val = performSHLCombine(N, DAG, DCI, Subtarget);
break;
case ISD::SRA:
return performSRACombine(N, DAG, DCI, Subtarget);
case ISD::SRL:
return performSRLCombine(N, DAG, DCI, Subtarget);
case ISD::VSELECT:
return performVSELECTCombine(N, DAG);
case ISD::XOR:
Val = performXORCombine(N, DAG, Subtarget);
break;
case ISD::SETCC:
Val = performSETCCCombine(N, DAG);
break;
}
if (Val.getNode()) {
LLVM_DEBUG(dbgs() << "\nMipsSE DAG Combine:\n";
N->printrWithDepth(dbgs(), &DAG); dbgs() << "\n=> \n";
Val.getNode()->printrWithDepth(dbgs(), &DAG); dbgs() << "\n");
return Val;
}
return MipsTargetLowering::PerformDAGCombine(N, DCI);
}
MachineBasicBlock *
MipsSETargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
MachineBasicBlock *BB) const {
switch (MI.getOpcode()) {
default:
return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB);
case Mips::BPOSGE32_PSEUDO:
return emitBPOSGE32(MI, BB);
case Mips::SNZ_B_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_B);
case Mips::SNZ_H_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_H);
case Mips::SNZ_W_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_W);
case Mips::SNZ_D_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_D);
case Mips::SNZ_V_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_V);
case Mips::SZ_B_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_B);
case Mips::SZ_H_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_H);
case Mips::SZ_W_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_W);
case Mips::SZ_D_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_D);
case Mips::SZ_V_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_V);
case Mips::COPY_FW_PSEUDO:
return emitCOPY_FW(MI, BB);
case Mips::COPY_FD_PSEUDO:
return emitCOPY_FD(MI, BB);
case Mips::INSERT_FW_PSEUDO:
return emitINSERT_FW(MI, BB);
case Mips::INSERT_FD_PSEUDO:
return emitINSERT_FD(MI, BB);
case Mips::INSERT_B_VIDX_PSEUDO:
case Mips::INSERT_B_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 1, false);
case Mips::INSERT_H_VIDX_PSEUDO:
case Mips::INSERT_H_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 2, false);
case Mips::INSERT_W_VIDX_PSEUDO:
case Mips::INSERT_W_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 4, false);
case Mips::INSERT_D_VIDX_PSEUDO:
case Mips::INSERT_D_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 8, false);
case Mips::INSERT_FW_VIDX_PSEUDO:
case Mips::INSERT_FW_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 4, true);
case Mips::INSERT_FD_VIDX_PSEUDO:
case Mips::INSERT_FD_VIDX64_PSEUDO:
return emitINSERT_DF_VIDX(MI, BB, 8, true);
case Mips::FILL_FW_PSEUDO:
return emitFILL_FW(MI, BB);
case Mips::FILL_FD_PSEUDO:
return emitFILL_FD(MI, BB);
case Mips::FEXP2_W_1_PSEUDO:
return emitFEXP2_W_1(MI, BB);
case Mips::FEXP2_D_1_PSEUDO:
return emitFEXP2_D_1(MI, BB);
case Mips::ST_F16:
return emitST_F16_PSEUDO(MI, BB);
case Mips::LD_F16:
return emitLD_F16_PSEUDO(MI, BB);
case Mips::MSA_FP_EXTEND_W_PSEUDO:
return emitFPEXTEND_PSEUDO(MI, BB, false);
case Mips::MSA_FP_ROUND_W_PSEUDO:
return emitFPROUND_PSEUDO(MI, BB, false);
case Mips::MSA_FP_EXTEND_D_PSEUDO:
return emitFPEXTEND_PSEUDO(MI, BB, true);
case Mips::MSA_FP_ROUND_D_PSEUDO:
return emitFPROUND_PSEUDO(MI, BB, true);
}
}
bool MipsSETargetLowering::isEligibleForTailCallOptimization(
const CCState &CCInfo, unsigned NextStackOffset,
const MipsFunctionInfo &FI) const {
if (!UseMipsTailCalls)
return false;
// Exception has to be cleared with eret.
if (FI.isISR())
return false;
// Return false if either the callee or caller has a byval argument.
if (CCInfo.getInRegsParamsCount() > 0 || FI.hasByvalArg())
return false;
// Return true if the callee's argument area is no larger than the
// caller's.
return NextStackOffset <= FI.getIncomingArgSize();
}
void MipsSETargetLowering::
getOpndList(SmallVectorImpl<SDValue> &Ops,
std::deque<std::pair<unsigned, SDValue>> &RegsToPass,
bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
bool IsCallReloc, CallLoweringInfo &CLI, SDValue Callee,
SDValue Chain) const {
Ops.push_back(Callee);
MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal,
InternalLinkage, IsCallReloc, CLI, Callee,
Chain);
}
SDValue MipsSETargetLowering::lowerLOAD(SDValue Op, SelectionDAG &DAG) const {
LoadSDNode &Nd = *cast<LoadSDNode>(Op);
if (Nd.getMemoryVT() != MVT::f64 || !NoDPLoadStore)
return MipsTargetLowering::lowerLOAD(Op, DAG);
// Replace a double precision load with two i32 loads and a buildpair64.
SDLoc DL(Op);
SDValue Ptr = Nd.getBasePtr(), Chain = Nd.getChain();
EVT PtrVT = Ptr.getValueType();
// i32 load from lower address.
SDValue Lo = DAG.getLoad(MVT::i32, DL, Chain, Ptr, MachinePointerInfo(),
Nd.getAlignment(), Nd.getMemOperand()->getFlags());
// i32 load from higher address.
Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, Ptr, DAG.getConstant(4, DL, PtrVT));
SDValue Hi = DAG.getLoad(
MVT::i32, DL, Lo.getValue(1), Ptr, MachinePointerInfo(),
std::min(Nd.getAlignment(), 4U), Nd.getMemOperand()->getFlags());
if (!Subtarget.isLittle())
std::swap(Lo, Hi);
SDValue BP = DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
SDValue Ops[2] = {BP, Hi.getValue(1)};
return DAG.getMergeValues(Ops, DL);
}
SDValue MipsSETargetLowering::lowerSTORE(SDValue Op, SelectionDAG &DAG) const {
StoreSDNode &Nd = *cast<StoreSDNode>(Op);
if (Nd.getMemoryVT() != MVT::f64 || !NoDPLoadStore)
return MipsTargetLowering::lowerSTORE(Op, DAG);
// Replace a double precision store with two extractelement64s and i32 stores.
SDLoc DL(Op);
SDValue Val = Nd.getValue(), Ptr = Nd.getBasePtr(), Chain = Nd.getChain();
EVT PtrVT = Ptr.getValueType();
SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Val, DAG.getConstant(0, DL, MVT::i32));
SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Val, DAG.getConstant(1, DL, MVT::i32));
if (!Subtarget.isLittle())
std::swap(Lo, Hi);
// i32 store to lower address.
Chain =
DAG.getStore(Chain, DL, Lo, Ptr, MachinePointerInfo(), Nd.getAlignment(),
Nd.getMemOperand()->getFlags(), Nd.getAAInfo());
// i32 store to higher address.
Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, Ptr, DAG.getConstant(4, DL, PtrVT));
return DAG.getStore(Chain, DL, Hi, Ptr, MachinePointerInfo(),
std::min(Nd.getAlignment(), 4U),
Nd.getMemOperand()->getFlags(), Nd.getAAInfo());
}
SDValue MipsSETargetLowering::lowerBITCAST(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT Src = Op.getOperand(0).getValueType().getSimpleVT();
MVT Dest = Op.getValueType().getSimpleVT();
// Bitcast i64 to double.
if (Src == MVT::i64 && Dest == MVT::f64) {
SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
Op.getOperand(0), DAG.getIntPtrConstant(0, DL));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
Op.getOperand(0), DAG.getIntPtrConstant(1, DL));
return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
}
// Bitcast double to i64.
if (Src == MVT::f64 && Dest == MVT::i64) {
SDValue Lo =
DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(0),
DAG.getConstant(0, DL, MVT::i32));
SDValue Hi =
DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(0),
DAG.getConstant(1, DL, MVT::i32));
return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
}
// Skip other cases of bitcast and use default lowering.
return SDValue();
}
SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc,
bool HasLo, bool HasHi,
SelectionDAG &DAG) const {
// MIPS32r6/MIPS64r6 removed accumulator based multiplies.
assert(!Subtarget.hasMips32r6());
EVT Ty = Op.getOperand(0).getValueType();
SDLoc DL(Op);
SDValue Mult = DAG.getNode(NewOpc, DL, MVT::Untyped,
Op.getOperand(0), Op.getOperand(1));
SDValue Lo, Hi;
if (HasLo)
Lo = DAG.getNode(MipsISD::MFLO, DL, Ty, Mult);
if (HasHi)
Hi = DAG.getNode(MipsISD::MFHI, DL, Ty, Mult);
if (!HasLo || !HasHi)
return HasLo ? Lo : Hi;
SDValue Vals[] = { Lo, Hi };
return DAG.getMergeValues(Vals, DL);
}
static SDValue initAccumulator(SDValue In, const SDLoc &DL, SelectionDAG &DAG) {
SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
DAG.getConstant(0, DL, MVT::i32));
SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
DAG.getConstant(1, DL, MVT::i32));
return DAG.getNode(MipsISD::MTLOHI, DL, MVT::Untyped, InLo, InHi);
}
static SDValue extractLOHI(SDValue Op, const SDLoc &DL, SelectionDAG &DAG) {
SDValue Lo = DAG.getNode(MipsISD::MFLO, DL, MVT::i32, Op);
SDValue Hi = DAG.getNode(MipsISD::MFHI, DL, MVT::i32, Op);
return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
}
// This function expands mips intrinsic nodes which have 64-bit input operands
// or output values.
//
// out64 = intrinsic-node in64
// =>
// lo = copy (extract-element (in64, 0))
// hi = copy (extract-element (in64, 1))
// mips-specific-node
// v0 = copy lo
// v1 = copy hi
// out64 = merge-values (v0, v1)
//
static SDValue lowerDSPIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
SDLoc DL(Op);
bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other;
SmallVector<SDValue, 3> Ops;
unsigned OpNo = 0;
// See if Op has a chain input.
if (HasChainIn)
Ops.push_back(Op->getOperand(OpNo++));
// The next operand is the intrinsic opcode.
assert(Op->getOperand(OpNo).getOpcode() == ISD::TargetConstant);
// See if the next operand has type i64.
SDValue Opnd = Op->getOperand(++OpNo), In64;
if (Opnd.getValueType() == MVT::i64)
In64 = initAccumulator(Opnd, DL, DAG);
else
Ops.push_back(Opnd);
// Push the remaining operands.
for (++OpNo ; OpNo < Op->getNumOperands(); ++OpNo)
Ops.push_back(Op->getOperand(OpNo));
// Add In64 to the end of the list.
if (In64.getNode())
Ops.push_back(In64);
// Scan output.
SmallVector<EVT, 2> ResTys;
for (SDNode::value_iterator I = Op->value_begin(), E = Op->value_end();
I != E; ++I)
ResTys.push_back((*I == MVT::i64) ? MVT::Untyped : *I);
// Create node.
SDValue Val = DAG.getNode(Opc, DL, ResTys, Ops);
SDValue Out = (ResTys[0] == MVT::Untyped) ? extractLOHI(Val, DL, DAG) : Val;
if (!HasChainIn)
return Out;
assert(Val->getValueType(1) == MVT::Other);
SDValue Vals[] = { Out, SDValue(Val.getNode(), 1) };
return DAG.getMergeValues(Vals, DL);
}
// Lower an MSA copy intrinsic into the specified SelectionDAG node
static SDValue lowerMSACopyIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
SDLoc DL(Op);
SDValue Vec = Op->getOperand(1);
SDValue Idx = Op->getOperand(2);
EVT ResTy = Op->getValueType(0);
EVT EltTy = Vec->getValueType(0).getVectorElementType();
SDValue Result = DAG.getNode(Opc, DL, ResTy, Vec, Idx,
DAG.getValueType(EltTy));
return Result;
}
static SDValue lowerMSASplatZExt(SDValue Op, unsigned OpNr, SelectionDAG &DAG) {
EVT ResVecTy = Op->getValueType(0);
EVT ViaVecTy = ResVecTy;
bool BigEndian = !DAG.getSubtarget().getTargetTriple().isLittleEndian();
SDLoc DL(Op);
// When ResVecTy == MVT::v2i64, LaneA is the upper 32 bits of the lane and
// LaneB is the lower 32-bits. Otherwise LaneA and LaneB are alternating
// lanes.
SDValue LaneA = Op->getOperand(OpNr);
SDValue LaneB;
if (ResVecTy == MVT::v2i64) {
// In case of the index being passed as an immediate value, set the upper
// lane to 0 so that the splati.d instruction can be matched.
if (isa<ConstantSDNode>(LaneA))
LaneB = DAG.getConstant(0, DL, MVT::i32);
// Having the index passed in a register, set the upper lane to the same
// value as the lower - this results in the BUILD_VECTOR node not being
// expanded through stack. This way we are able to pattern match the set of
// nodes created here to splat.d.
else
LaneB = LaneA;
ViaVecTy = MVT::v4i32;
if(BigEndian)
std::swap(LaneA, LaneB);
} else
LaneB = LaneA;
SDValue Ops[16] = { LaneA, LaneB, LaneA, LaneB, LaneA, LaneB, LaneA, LaneB,
LaneA, LaneB, LaneA, LaneB, LaneA, LaneB, LaneA, LaneB };
SDValue Result = DAG.getBuildVector(
ViaVecTy, DL, makeArrayRef(Ops, ViaVecTy.getVectorNumElements()));
if (ViaVecTy != ResVecTy) {
SDValue One = DAG.getConstant(1, DL, ViaVecTy);
Result = DAG.getNode(ISD::BITCAST, DL, ResVecTy,
DAG.getNode(ISD::AND, DL, ViaVecTy, Result, One));
}
return Result;
}
static SDValue lowerMSASplatImm(SDValue Op, unsigned ImmOp, SelectionDAG &DAG,
bool IsSigned = false) {
auto *CImm = cast<ConstantSDNode>(Op->getOperand(ImmOp));
return DAG.getConstant(
APInt(Op->getValueType(0).getScalarType().getSizeInBits(),
IsSigned ? CImm->getSExtValue() : CImm->getZExtValue(), IsSigned),
SDLoc(Op), Op->getValueType(0));
}
static SDValue getBuildVectorSplat(EVT VecTy, SDValue SplatValue,
bool BigEndian, SelectionDAG &DAG) {
EVT ViaVecTy = VecTy;
SDValue SplatValueA = SplatValue;
SDValue SplatValueB = SplatValue;
SDLoc DL(SplatValue);
if (VecTy == MVT::v2i64) {
// v2i64 BUILD_VECTOR must be performed via v4i32 so split into i32's.
ViaVecTy = MVT::v4i32;
SplatValueA = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, SplatValue);
SplatValueB = DAG.getNode(ISD::SRL, DL, MVT::i64, SplatValue,
DAG.getConstant(32, DL, MVT::i32));
SplatValueB = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, SplatValueB);
}
// We currently hold the parts in little endian order. Swap them if
// necessary.
if (BigEndian)
std::swap(SplatValueA, SplatValueB);
SDValue Ops[16] = { SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB };
SDValue Result = DAG.getBuildVector(
ViaVecTy, DL, makeArrayRef(Ops, ViaVecTy.getVectorNumElements()));
if (VecTy != ViaVecTy)
Result = DAG.getNode(ISD::BITCAST, DL, VecTy, Result);
return Result;
}
static SDValue lowerMSABinaryBitImmIntr(SDValue Op, SelectionDAG &DAG,
unsigned Opc, SDValue Imm,
bool BigEndian) {
EVT VecTy = Op->getValueType(0);
SDValue Exp2Imm;
SDLoc DL(Op);
// The DAG Combiner can't constant fold bitcasted vectors yet so we must do it
// here for now.
if (VecTy == MVT::v2i64) {
if (ConstantSDNode *CImm = dyn_cast<ConstantSDNode>(Imm)) {
APInt BitImm = APInt(64, 1) << CImm->getAPIntValue();
SDValue BitImmHiOp = DAG.getConstant(BitImm.lshr(32).trunc(32), DL,
MVT::i32);
SDValue BitImmLoOp = DAG.getConstant(BitImm.trunc(32), DL, MVT::i32);
if (BigEndian)
std::swap(BitImmLoOp, BitImmHiOp);
Exp2Imm = DAG.getNode(
ISD::BITCAST, DL, MVT::v2i64,
DAG.getBuildVector(MVT::v4i32, DL,
{BitImmLoOp, BitImmHiOp, BitImmLoOp, BitImmHiOp}));
}
}
if (!Exp2Imm.getNode()) {
// We couldnt constant fold, do a vector shift instead
// Extend i32 to i64 if necessary. Sign or zero extend doesn't matter since
// only values 0-63 are valid.
if (VecTy == MVT::v2i64)
Imm = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, Imm);
Exp2Imm = getBuildVectorSplat(VecTy, Imm, BigEndian, DAG);
Exp2Imm = DAG.getNode(ISD::SHL, DL, VecTy, DAG.getConstant(1, DL, VecTy),
Exp2Imm);
}
return DAG.getNode(Opc, DL, VecTy, Op->getOperand(1), Exp2Imm);
}
static SDValue truncateVecElts(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
EVT ResTy = Op->getValueType(0);
SDValue Vec = Op->getOperand(2);
bool BigEndian = !DAG.getSubtarget().getTargetTriple().isLittleEndian();
MVT ResEltTy = ResTy == MVT::v2i64 ? MVT::i64 : MVT::i32;
SDValue ConstValue = DAG.getConstant(Vec.getScalarValueSizeInBits() - 1,
DL, ResEltTy);
SDValue SplatVec = getBuildVectorSplat(ResTy, ConstValue, BigEndian, DAG);
return DAG.getNode(ISD::AND, DL, ResTy, Vec, SplatVec);
}
static SDValue lowerMSABitClear(SDValue Op, SelectionDAG &DAG) {
EVT ResTy = Op->getValueType(0);
SDLoc DL(Op);
SDValue One = DAG.getConstant(1, DL, ResTy);
SDValue Bit = DAG.getNode(ISD::SHL, DL, ResTy, One, truncateVecElts(Op, DAG));
return DAG.getNode(ISD::AND, DL, ResTy, Op->getOperand(1),
DAG.getNOT(DL, Bit, ResTy));
}
static SDValue lowerMSABitClearImm(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
EVT ResTy = Op->getValueType(0);
APInt BitImm = APInt(ResTy.getScalarSizeInBits(), 1)
<< cast<ConstantSDNode>(Op->getOperand(2))->getAPIntValue();
SDValue BitMask = DAG.getConstant(~BitImm, DL, ResTy);
return DAG.getNode(ISD::AND, DL, ResTy, Op->getOperand(1), BitMask);
}
SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
unsigned Intrinsic = cast<ConstantSDNode>(Op->getOperand(0))->getZExtValue();
switch (Intrinsic) {
default:
return SDValue();
case Intrinsic::mips_shilo:
return lowerDSPIntr(Op, DAG, MipsISD::SHILO);
case Intrinsic::mips_dpau_h_qbl:
return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL);
case Intrinsic::mips_dpau_h_qbr:
return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR);
case Intrinsic::mips_dpsu_h_qbl:
return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL);
case Intrinsic::mips_dpsu_h_qbr:
return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR);
case Intrinsic::mips_dpa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH);
case Intrinsic::mips_dps_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH);
case Intrinsic::mips_dpax_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH);
case Intrinsic::mips_dpsx_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH);
case Intrinsic::mips_mulsa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH);
case Intrinsic::mips_mult:
return lowerDSPIntr(Op, DAG, MipsISD::Mult);
case Intrinsic::mips_multu:
return lowerDSPIntr(Op, DAG, MipsISD::Multu);
case Intrinsic::mips_madd:
return lowerDSPIntr(Op, DAG, MipsISD::MAdd);
case Intrinsic::mips_maddu:
return lowerDSPIntr(Op, DAG, MipsISD::MAddu);
case Intrinsic::mips_msub:
return lowerDSPIntr(Op, DAG, MipsISD::MSub);
case Intrinsic::mips_msubu:
return lowerDSPIntr(Op, DAG, MipsISD::MSubu);
case Intrinsic::mips_addv_b:
case Intrinsic::mips_addv_h:
case Intrinsic::mips_addv_w:
case Intrinsic::mips_addv_d:
return DAG.getNode(ISD::ADD, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_addvi_b:
case Intrinsic::mips_addvi_h:
case Intrinsic::mips_addvi_w:
case Intrinsic::mips_addvi_d:
return DAG.getNode(ISD::ADD, DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_and_v:
return DAG.getNode(ISD::AND, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_andi_b:
return DAG.getNode(ISD::AND, DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_bclr_b:
case Intrinsic::mips_bclr_h:
case Intrinsic::mips_bclr_w:
case Intrinsic::mips_bclr_d:
return lowerMSABitClear(Op, DAG);
case Intrinsic::mips_bclri_b:
case Intrinsic::mips_bclri_h:
case Intrinsic::mips_bclri_w:
case Intrinsic::mips_bclri_d:
return lowerMSABitClearImm(Op, DAG);
case Intrinsic::mips_binsli_b:
case Intrinsic::mips_binsli_h:
case Intrinsic::mips_binsli_w:
case Intrinsic::mips_binsli_d: {
// binsli_x(IfClear, IfSet, nbits) -> (vselect LBitsMask, IfSet, IfClear)
EVT VecTy = Op->getValueType(0);
EVT EltTy = VecTy.getVectorElementType();
if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
report_fatal_error("Immediate out of range");
APInt Mask = APInt::getHighBitsSet(EltTy.getSizeInBits(),
Op->getConstantOperandVal(3) + 1);
return DAG.getNode(ISD::VSELECT, DL, VecTy,
DAG.getConstant(Mask, DL, VecTy, true),
Op->getOperand(2), Op->getOperand(1));
}
case Intrinsic::mips_binsri_b:
case Intrinsic::mips_binsri_h:
case Intrinsic::mips_binsri_w:
case Intrinsic::mips_binsri_d: {
// binsri_x(IfClear, IfSet, nbits) -> (vselect RBitsMask, IfSet, IfClear)
EVT VecTy = Op->getValueType(0);
EVT EltTy = VecTy.getVectorElementType();
if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
report_fatal_error("Immediate out of range");
APInt Mask = APInt::getLowBitsSet(EltTy.getSizeInBits(),
Op->getConstantOperandVal(3) + 1);
return DAG.getNode(ISD::VSELECT, DL, VecTy,
DAG.getConstant(Mask, DL, VecTy, true),
Op->getOperand(2), Op->getOperand(1));
}
case Intrinsic::mips_bmnz_v:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0), Op->getOperand(3),
Op->getOperand(2), Op->getOperand(1));
case Intrinsic::mips_bmnzi_b:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 3, DAG), Op->getOperand(2),
Op->getOperand(1));
case Intrinsic::mips_bmz_v:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0), Op->getOperand(3),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_bmzi_b:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 3, DAG), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_bneg_b:
case Intrinsic::mips_bneg_h:
case Intrinsic::mips_bneg_w:
case Intrinsic::mips_bneg_d: {
EVT VecTy = Op->getValueType(0);
SDValue One = DAG.getConstant(1, DL, VecTy);
return DAG.getNode(ISD::XOR, DL, VecTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, DL, VecTy, One,
truncateVecElts(Op, DAG)));
}
case Intrinsic::mips_bnegi_b:
case Intrinsic::mips_bnegi_h:
case Intrinsic::mips_bnegi_w:
case Intrinsic::mips_bnegi_d:
return lowerMSABinaryBitImmIntr(Op, DAG, ISD::XOR, Op->getOperand(2),
!Subtarget.isLittle());
case Intrinsic::mips_bnz_b:
case Intrinsic::mips_bnz_h:
case Intrinsic::mips_bnz_w:
case Intrinsic::mips_bnz_d:
return DAG.getNode(MipsISD::VALL_NONZERO, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_bnz_v:
return DAG.getNode(MipsISD::VANY_NONZERO, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_bsel_v:
// bsel_v(Mask, IfClear, IfSet) -> (vselect Mask, IfSet, IfClear)
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(3),
Op->getOperand(2));
case Intrinsic::mips_bseli_b:
// bseli_v(Mask, IfClear, IfSet) -> (vselect Mask, IfSet, IfClear)
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 3, DAG),
Op->getOperand(2));
case Intrinsic::mips_bset_b:
case Intrinsic::mips_bset_h:
case Intrinsic::mips_bset_w:
case Intrinsic::mips_bset_d: {
EVT VecTy = Op->getValueType(0);
SDValue One = DAG.getConstant(1, DL, VecTy);
return DAG.getNode(ISD::OR, DL, VecTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, DL, VecTy, One,
truncateVecElts(Op, DAG)));
}
case Intrinsic::mips_bseti_b:
case Intrinsic::mips_bseti_h:
case Intrinsic::mips_bseti_w:
case Intrinsic::mips_bseti_d:
return lowerMSABinaryBitImmIntr(Op, DAG, ISD::OR, Op->getOperand(2),
!Subtarget.isLittle());
case Intrinsic::mips_bz_b:
case Intrinsic::mips_bz_h:
case Intrinsic::mips_bz_w:
case Intrinsic::mips_bz_d:
return DAG.getNode(MipsISD::VALL_ZERO, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_bz_v:
return DAG.getNode(MipsISD::VANY_ZERO, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ceq_b:
case Intrinsic::mips_ceq_h:
case Intrinsic::mips_ceq_w:
case Intrinsic::mips_ceq_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETEQ);
case Intrinsic::mips_ceqi_b:
case Intrinsic::mips_ceqi_h:
case Intrinsic::mips_ceqi_w:
case Intrinsic::mips_ceqi_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETEQ);
case Intrinsic::mips_cle_s_b:
case Intrinsic::mips_cle_s_h:
case Intrinsic::mips_cle_s_w:
case Intrinsic::mips_cle_s_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETLE);
case Intrinsic::mips_clei_s_b:
case Intrinsic::mips_clei_s_h:
case Intrinsic::mips_clei_s_w:
case Intrinsic::mips_clei_s_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETLE);
case Intrinsic::mips_cle_u_b:
case Intrinsic::mips_cle_u_h:
case Intrinsic::mips_cle_u_w:
case Intrinsic::mips_cle_u_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETULE);
case Intrinsic::mips_clei_u_b:
case Intrinsic::mips_clei_u_h:
case Intrinsic::mips_clei_u_w:
case Intrinsic::mips_clei_u_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG), ISD::SETULE);
case Intrinsic::mips_clt_s_b:
case Intrinsic::mips_clt_s_h:
case Intrinsic::mips_clt_s_w:
case Intrinsic::mips_clt_s_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETLT);
case Intrinsic::mips_clti_s_b:
case Intrinsic::mips_clti_s_h:
case Intrinsic::mips_clti_s_w:
case Intrinsic::mips_clti_s_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETLT);
case Intrinsic::mips_clt_u_b:
case Intrinsic::mips_clt_u_h:
case Intrinsic::mips_clt_u_w:
case Intrinsic::mips_clt_u_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETULT);
case Intrinsic::mips_clti_u_b:
case Intrinsic::mips_clti_u_h:
case Intrinsic::mips_clti_u_w:
case Intrinsic::mips_clti_u_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG), ISD::SETULT);
case Intrinsic::mips_copy_s_b:
case Intrinsic::mips_copy_s_h:
case Intrinsic::mips_copy_s_w:
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_SEXT_ELT);
case Intrinsic::mips_copy_s_d:
if (Subtarget.hasMips64())
// Lower directly into VEXTRACT_SEXT_ELT since i64 is legal on Mips64.
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_SEXT_ELT);
else {
// Lower into the generic EXTRACT_VECTOR_ELT node and let the type
// legalizer and EXTRACT_VECTOR_ELT lowering sort it out.
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
}
case Intrinsic::mips_copy_u_b:
case Intrinsic::mips_copy_u_h:
case Intrinsic::mips_copy_u_w:
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_ZEXT_ELT);
case Intrinsic::mips_copy_u_d:
if (Subtarget.hasMips64())
// Lower directly into VEXTRACT_ZEXT_ELT since i64 is legal on Mips64.
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_ZEXT_ELT);
else {
// Lower into the generic EXTRACT_VECTOR_ELT node and let the type
// legalizer and EXTRACT_VECTOR_ELT lowering sort it out.
// Note: When i64 is illegal, this results in copy_s.w instructions
// instead of copy_u.w instructions. This makes no difference to the
// behaviour since i64 is only illegal when the register file is 32-bit.
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
}
case Intrinsic::mips_div_s_b:
case Intrinsic::mips_div_s_h:
case Intrinsic::mips_div_s_w:
case Intrinsic::mips_div_s_d:
return DAG.getNode(ISD::SDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_div_u_b:
case Intrinsic::mips_div_u_h:
case Intrinsic::mips_div_u_w:
case Intrinsic::mips_div_u_d:
return DAG.getNode(ISD::UDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_fadd_w:
case Intrinsic::mips_fadd_d:
// TODO: If intrinsics have fast-math-flags, propagate them.
return DAG.getNode(ISD::FADD, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
// Don't lower mips_fcaf_[wd] since LLVM folds SETFALSE condcodes away
case Intrinsic::mips_fceq_w:
case Intrinsic::mips_fceq_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETOEQ);
case Intrinsic::mips_fcle_w:
case Intrinsic::mips_fcle_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETOLE);
case Intrinsic::mips_fclt_w:
case Intrinsic::mips_fclt_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETOLT);
case Intrinsic::mips_fcne_w:
case Intrinsic::mips_fcne_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETONE);
case Intrinsic::mips_fcor_w:
case Intrinsic::mips_fcor_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETO);
case Intrinsic::mips_fcueq_w:
case Intrinsic::mips_fcueq_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETUEQ);
case Intrinsic::mips_fcule_w:
case Intrinsic::mips_fcule_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETULE);
case Intrinsic::mips_fcult_w:
case Intrinsic::mips_fcult_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETULT);
case Intrinsic::mips_fcun_w:
case Intrinsic::mips_fcun_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETUO);
case Intrinsic::mips_fcune_w:
case Intrinsic::mips_fcune_d:
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2), ISD::SETUNE);
case Intrinsic::mips_fdiv_w:
case Intrinsic::mips_fdiv_d:
// TODO: If intrinsics have fast-math-flags, propagate them.
return DAG.getNode(ISD::FDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_ffint_u_w:
case Intrinsic::mips_ffint_u_d:
return DAG.getNode(ISD::UINT_TO_FP, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ffint_s_w:
case Intrinsic::mips_ffint_s_d:
return DAG.getNode(ISD::SINT_TO_FP, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_fill_b:
case Intrinsic::mips_fill_h:
case Intrinsic::mips_fill_w:
case Intrinsic::mips_fill_d: {
EVT ResTy = Op->getValueType(0);
SmallVector<SDValue, 16> Ops(ResTy.getVectorNumElements(),
Op->getOperand(1));
// If ResTy is v2i64 then the type legalizer will break this node down into
// an equivalent v4i32.
return DAG.getBuildVector(ResTy, DL, Ops);
}
case Intrinsic::mips_fexp2_w:
case Intrinsic::mips_fexp2_d: {
// TODO: If intrinsics have fast-math-flags, propagate them.
EVT ResTy = Op->getValueType(0);
return DAG.getNode(
ISD::FMUL, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::FEXP2, SDLoc(Op), ResTy, Op->getOperand(2)));
}
case Intrinsic::mips_flog2_w:
case Intrinsic::mips_flog2_d:
return DAG.getNode(ISD::FLOG2, DL, Op->getValueType(0), Op->getOperand(1));
case Intrinsic::mips_fmadd_w:
case Intrinsic::mips_fmadd_d:
return DAG.getNode(ISD::FMA, SDLoc(Op), Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3));
case Intrinsic::mips_fmul_w:
case Intrinsic::mips_fmul_d:
// TODO: If intrinsics have fast-math-flags, propagate them.
return DAG.getNode(ISD::FMUL, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_fmsub_w:
case Intrinsic::mips_fmsub_d: {
// TODO: If intrinsics have fast-math-flags, propagate them.
return DAG.getNode(MipsISD::FMS, SDLoc(Op), Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3));
}
case Intrinsic::mips_frint_w:
case Intrinsic::mips_frint_d:
return DAG.getNode(ISD::FRINT, DL, Op->getValueType(0), Op->getOperand(1));
case Intrinsic::mips_fsqrt_w:
case Intrinsic::mips_fsqrt_d:
return DAG.getNode(ISD::FSQRT, DL, Op->getValueType(0), Op->getOperand(1));
case Intrinsic::mips_fsub_w:
case Intrinsic::mips_fsub_d:
// TODO: If intrinsics have fast-math-flags, propagate them.
return DAG.getNode(ISD::FSUB, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_ftrunc_u_w:
case Intrinsic::mips_ftrunc_u_d:
return DAG.getNode(ISD::FP_TO_UINT, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ftrunc_s_w:
case Intrinsic::mips_ftrunc_s_d:
return DAG.getNode(ISD::FP_TO_SINT, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ilvev_b:
case Intrinsic::mips_ilvev_h:
case Intrinsic::mips_ilvev_w:
case Intrinsic::mips_ilvev_d:
return DAG.getNode(MipsISD::ILVEV, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvl_b:
case Intrinsic::mips_ilvl_h:
case Intrinsic::mips_ilvl_w:
case Intrinsic::mips_ilvl_d:
return DAG.getNode(MipsISD::ILVL, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvod_b:
case Intrinsic::mips_ilvod_h:
case Intrinsic::mips_ilvod_w:
case Intrinsic::mips_ilvod_d:
return DAG.getNode(MipsISD::ILVOD, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvr_b:
case Intrinsic::mips_ilvr_h:
case Intrinsic::mips_ilvr_w:
case Intrinsic::mips_ilvr_d:
return DAG.getNode(MipsISD::ILVR, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_insert_b:
case Intrinsic::mips_insert_h:
case Intrinsic::mips_insert_w:
case Intrinsic::mips_insert_d:
return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(Op), Op->getValueType(0),
Op->getOperand(1), Op->getOperand(3), Op->getOperand(2));
case Intrinsic::mips_insve_b:
case Intrinsic::mips_insve_h:
case Intrinsic::mips_insve_w:
case Intrinsic::mips_insve_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_insve_b: Max = 15; break;
case Intrinsic::mips_insve_h: Max = 7; break;
case Intrinsic::mips_insve_w: Max = 3; break;
case Intrinsic::mips_insve_d: Max = 1; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return DAG.getNode(MipsISD::INSVE, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3),
DAG.getConstant(0, DL, MVT::i32));
}
case Intrinsic::mips_ldi_b:
case Intrinsic::mips_ldi_h:
case Intrinsic::mips_ldi_w:
case Intrinsic::mips_ldi_d:
return lowerMSASplatImm(Op, 1, DAG, true);
case Intrinsic::mips_lsa:
case Intrinsic::mips_dlsa: {
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::ADD, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
case Intrinsic::mips_maddv_b:
case Intrinsic::mips_maddv_h:
case Intrinsic::mips_maddv_w:
case Intrinsic::mips_maddv_d: {
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::ADD, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::MUL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
case Intrinsic::mips_max_s_b:
case Intrinsic::mips_max_s_h:
case Intrinsic::mips_max_s_w:
case Intrinsic::mips_max_s_d:
return DAG.getNode(ISD::SMAX, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_max_u_b:
case Intrinsic::mips_max_u_h:
case Intrinsic::mips_max_u_w:
case Intrinsic::mips_max_u_d:
return DAG.getNode(ISD::UMAX, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_maxi_s_b:
case Intrinsic::mips_maxi_s_h:
case Intrinsic::mips_maxi_s_w:
case Intrinsic::mips_maxi_s_d:
return DAG.getNode(ISD::SMAX, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG, true));
case Intrinsic::mips_maxi_u_b:
case Intrinsic::mips_maxi_u_h:
case Intrinsic::mips_maxi_u_w:
case Intrinsic::mips_maxi_u_d:
return DAG.getNode(ISD::UMAX, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_min_s_b:
case Intrinsic::mips_min_s_h:
case Intrinsic::mips_min_s_w:
case Intrinsic::mips_min_s_d:
return DAG.getNode(ISD::SMIN, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_min_u_b:
case Intrinsic::mips_min_u_h:
case Intrinsic::mips_min_u_w:
case Intrinsic::mips_min_u_d:
return DAG.getNode(ISD::UMIN, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_mini_s_b:
case Intrinsic::mips_mini_s_h:
case Intrinsic::mips_mini_s_w:
case Intrinsic::mips_mini_s_d:
return DAG.getNode(ISD::SMIN, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG, true));
case Intrinsic::mips_mini_u_b:
case Intrinsic::mips_mini_u_h:
case Intrinsic::mips_mini_u_w:
case Intrinsic::mips_mini_u_d:
return DAG.getNode(ISD::UMIN, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_mod_s_b:
case Intrinsic::mips_mod_s_h:
case Intrinsic::mips_mod_s_w:
case Intrinsic::mips_mod_s_d:
return DAG.getNode(ISD::SREM, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_mod_u_b:
case Intrinsic::mips_mod_u_h:
case Intrinsic::mips_mod_u_w:
case Intrinsic::mips_mod_u_d:
return DAG.getNode(ISD::UREM, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_mulv_b:
case Intrinsic::mips_mulv_h:
case Intrinsic::mips_mulv_w:
case Intrinsic::mips_mulv_d:
return DAG.getNode(ISD::MUL, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_msubv_b:
case Intrinsic::mips_msubv_h:
case Intrinsic::mips_msubv_w:
case Intrinsic::mips_msubv_d: {
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::SUB, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::MUL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
case Intrinsic::mips_nlzc_b:
case Intrinsic::mips_nlzc_h:
case Intrinsic::mips_nlzc_w:
case Intrinsic::mips_nlzc_d:
return DAG.getNode(ISD::CTLZ, DL, Op->getValueType(0), Op->getOperand(1));
case Intrinsic::mips_nor_v: {
SDValue Res = DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
return DAG.getNOT(DL, Res, Res->getValueType(0));
}
case Intrinsic::mips_nori_b: {
SDValue Res = DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
return DAG.getNOT(DL, Res, Res->getValueType(0));
}
case Intrinsic::mips_or_v:
return DAG.getNode(ISD::OR, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_ori_b:
return DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_pckev_b:
case Intrinsic::mips_pckev_h:
case Intrinsic::mips_pckev_w:
case Intrinsic::mips_pckev_d:
return DAG.getNode(MipsISD::PCKEV, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_pckod_b:
case Intrinsic::mips_pckod_h:
case Intrinsic::mips_pckod_w:
case Intrinsic::mips_pckod_d:
return DAG.getNode(MipsISD::PCKOD, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_pcnt_b:
case Intrinsic::mips_pcnt_h:
case Intrinsic::mips_pcnt_w:
case Intrinsic::mips_pcnt_d:
return DAG.getNode(ISD::CTPOP, DL, Op->getValueType(0), Op->getOperand(1));
case Intrinsic::mips_sat_s_b:
case Intrinsic::mips_sat_s_h:
case Intrinsic::mips_sat_s_w:
case Intrinsic::mips_sat_s_d:
case Intrinsic::mips_sat_u_b:
case Intrinsic::mips_sat_u_h:
case Intrinsic::mips_sat_u_w:
case Intrinsic::mips_sat_u_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_sat_s_b:
case Intrinsic::mips_sat_u_b: Max = 7; break;
case Intrinsic::mips_sat_s_h:
case Intrinsic::mips_sat_u_h: Max = 15; break;
case Intrinsic::mips_sat_s_w:
case Intrinsic::mips_sat_u_w: Max = 31; break;
case Intrinsic::mips_sat_s_d:
case Intrinsic::mips_sat_u_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
case Intrinsic::mips_shf_b:
case Intrinsic::mips_shf_h:
case Intrinsic::mips_shf_w: {
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > 255)
report_fatal_error("Immediate out of range");
return DAG.getNode(MipsISD::SHF, DL, Op->getValueType(0),
Op->getOperand(2), Op->getOperand(1));
}
case Intrinsic::mips_sldi_b:
case Intrinsic::mips_sldi_h:
case Intrinsic::mips_sldi_w:
case Intrinsic::mips_sldi_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_sldi_b: Max = 15; break;
case Intrinsic::mips_sldi_h: Max = 7; break;
case Intrinsic::mips_sldi_w: Max = 3; break;
case Intrinsic::mips_sldi_d: Max = 1; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(3))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
case Intrinsic::mips_sll_b:
case Intrinsic::mips_sll_h:
case Intrinsic::mips_sll_w:
case Intrinsic::mips_sll_d:
return DAG.getNode(ISD::SHL, DL, Op->getValueType(0), Op->getOperand(1),
truncateVecElts(Op, DAG));
case Intrinsic::mips_slli_b:
case Intrinsic::mips_slli_h:
case Intrinsic::mips_slli_w:
case Intrinsic::mips_slli_d:
return DAG.getNode(ISD::SHL, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_splat_b:
case Intrinsic::mips_splat_h:
case Intrinsic::mips_splat_w:
case Intrinsic::mips_splat_d:
// We can't lower via VECTOR_SHUFFLE because it requires constant shuffle
// masks, nor can we lower via BUILD_VECTOR & EXTRACT_VECTOR_ELT because
// EXTRACT_VECTOR_ELT can't extract i64's on MIPS32.
// Instead we lower to MipsISD::VSHF and match from there.
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
lowerMSASplatZExt(Op, 2, DAG), Op->getOperand(1),
Op->getOperand(1));
case Intrinsic::mips_splati_b:
case Intrinsic::mips_splati_h:
case Intrinsic::mips_splati_w:
case Intrinsic::mips_splati_d:
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 2, DAG), Op->getOperand(1),
Op->getOperand(1));
case Intrinsic::mips_sra_b:
case Intrinsic::mips_sra_h:
case Intrinsic::mips_sra_w:
case Intrinsic::mips_sra_d:
return DAG.getNode(ISD::SRA, DL, Op->getValueType(0), Op->getOperand(1),
truncateVecElts(Op, DAG));
case Intrinsic::mips_srai_b:
case Intrinsic::mips_srai_h:
case Intrinsic::mips_srai_w:
case Intrinsic::mips_srai_d:
return DAG.getNode(ISD::SRA, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_srari_b:
case Intrinsic::mips_srari_h:
case Intrinsic::mips_srari_w:
case Intrinsic::mips_srari_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_srari_b: Max = 7; break;
case Intrinsic::mips_srari_h: Max = 15; break;
case Intrinsic::mips_srari_w: Max = 31; break;
case Intrinsic::mips_srari_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
case Intrinsic::mips_srl_b:
case Intrinsic::mips_srl_h:
case Intrinsic::mips_srl_w:
case Intrinsic::mips_srl_d:
return DAG.getNode(ISD::SRL, DL, Op->getValueType(0), Op->getOperand(1),
truncateVecElts(Op, DAG));
case Intrinsic::mips_srli_b:
case Intrinsic::mips_srli_h:
case Intrinsic::mips_srli_w:
case Intrinsic::mips_srli_d:
return DAG.getNode(ISD::SRL, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_srlri_b:
case Intrinsic::mips_srlri_h:
case Intrinsic::mips_srlri_w:
case Intrinsic::mips_srlri_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_srlri_b: Max = 7; break;
case Intrinsic::mips_srlri_h: Max = 15; break;
case Intrinsic::mips_srlri_w: Max = 31; break;
case Intrinsic::mips_srlri_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
case Intrinsic::mips_subv_b:
case Intrinsic::mips_subv_h:
case Intrinsic::mips_subv_w:
case Intrinsic::mips_subv_d:
return DAG.getNode(ISD::SUB, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_subvi_b:
case Intrinsic::mips_subvi_h:
case Intrinsic::mips_subvi_w:
case Intrinsic::mips_subvi_d:
return DAG.getNode(ISD::SUB, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::mips_vshf_b:
case Intrinsic::mips_vshf_h:
case Intrinsic::mips_vshf_w:
case Intrinsic::mips_vshf_d:
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3));
case Intrinsic::mips_xor_v:
return DAG.getNode(ISD::XOR, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_xori_b:
return DAG.getNode(ISD::XOR, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
case Intrinsic::thread_pointer: {
EVT PtrVT = getPointerTy(DAG.getDataLayout());
return DAG.getNode(MipsISD::ThreadPointer, DL, PtrVT);
}
}
}
static SDValue lowerMSALoadIntr(SDValue Op, SelectionDAG &DAG, unsigned Intr,
const MipsSubtarget &Subtarget) {
SDLoc DL(Op);
SDValue ChainIn = Op->getOperand(0);
SDValue Address = Op->getOperand(2);
SDValue Offset = Op->getOperand(3);
EVT ResTy = Op->getValueType(0);
EVT PtrTy = Address->getValueType(0);
// For N64 addresses have the underlying type MVT::i64. This intrinsic
// however takes an i32 signed constant offset. The actual type of the
// intrinsic is a scaled signed i10.
if (Subtarget.isABI_N64())
Offset = DAG.getNode(ISD::SIGN_EXTEND, DL, PtrTy, Offset);
Address = DAG.getNode(ISD::ADD, DL, PtrTy, Address, Offset);
return DAG.getLoad(ResTy, DL, ChainIn, Address, MachinePointerInfo(),
/* Alignment = */ 16);
}
SDValue MipsSETargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
unsigned Intr = cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue();
switch (Intr) {
default:
return SDValue();
case Intrinsic::mips_extp:
return lowerDSPIntr(Op, DAG, MipsISD::EXTP);
case Intrinsic::mips_extpdp:
return lowerDSPIntr(Op, DAG, MipsISD::EXTPDP);
case Intrinsic::mips_extr_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_W);
case Intrinsic::mips_extr_r_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W);
case Intrinsic::mips_extr_rs_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W);
case Intrinsic::mips_extr_s_h:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H);
case Intrinsic::mips_mthlip:
return lowerDSPIntr(Op, DAG, MipsISD::MTHLIP);
case Intrinsic::mips_mulsaq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH);
case Intrinsic::mips_maq_s_w_phl:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL);
case Intrinsic::mips_maq_s_w_phr:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR);
case Intrinsic::mips_maq_sa_w_phl:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL);
case Intrinsic::mips_maq_sa_w_phr:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR);
case Intrinsic::mips_dpaq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH);
case Intrinsic::mips_dpsq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH);
case Intrinsic::mips_dpaq_sa_l_w:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W);
case Intrinsic::mips_dpsq_sa_l_w:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W);
case Intrinsic::mips_dpaqx_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH);
case Intrinsic::mips_dpaqx_sa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH);
case Intrinsic::mips_dpsqx_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH);
case Intrinsic::mips_dpsqx_sa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH);
case Intrinsic::mips_ld_b: