llvm-6502/lib/Target/ARM/ARMISelDAGToDAG.cpp

3351 lines
126 KiB
C++
Raw Normal View History

//===-- ARMISelDAGToDAG.cpp - A dag to dag inst selector for ARM ----------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines an instruction selector for the ARM target.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "arm-isel"
#include "ARM.h"
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
#include "ARMBaseInstrInfo.h"
#include "ARMTargetMachine.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
#include "llvm/LLVMContext.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
static cl::opt<bool>
DisableShifterOp("disable-shifter-op", cl::Hidden,
cl::desc("Disable isel of shifter-op"),
cl::init(false));
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
static cl::opt<bool>
CheckVMLxHazard("check-vmlx-hazard", cl::Hidden,
cl::desc("Check fp vmla / vmls hazard at isel time"),
This patch combines several changes from Evan Cheng for rdar://8659675. Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Enable these fp vmlx codegen changes for Cortex-A9. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129775 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 18:11:57 +00:00
cl::init(true));
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
//===--------------------------------------------------------------------===//
/// ARMDAGToDAGISel - ARM specific code to select ARM machine
/// instructions for SelectionDAG operations.
///
namespace {
enum AddrMode2Type {
AM2_BASE, // Simple AM2 (+-imm12)
AM2_SHOP // Shifter-op AM2
};
class ARMDAGToDAGISel : public SelectionDAGISel {
ARMBaseTargetMachine &TM;
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
const ARMBaseInstrInfo *TII;
/// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
/// make the right decision when generating code for different targets.
const ARMSubtarget *Subtarget;
public:
explicit ARMDAGToDAGISel(ARMBaseTargetMachine &tm,
CodeGenOpt::Level OptLevel)
: SelectionDAGISel(tm, OptLevel), TM(tm),
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
TII(static_cast<const ARMBaseInstrInfo*>(TM.getInstrInfo())),
Subtarget(&TM.getSubtarget<ARMSubtarget>()) {
}
virtual const char *getPassName() const {
return "ARM Instruction Selection";
}
/// getI32Imm - Return a target constant of type i32 with the specified
/// value.
inline SDValue getI32Imm(unsigned Imm) {
return CurDAG->getTargetConstant(Imm, MVT::i32);
}
SDNode *Select(SDNode *N);
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
bool hasNoVMLxHazardUse(SDNode *N) const;
bool isShifterOpProfitable(const SDValue &Shift,
ARM_AM::ShiftOpc ShOpcVal, unsigned ShAmt);
bool SelectRegShifterOperand(SDValue N, SDValue &A,
SDValue &B, SDValue &C,
bool CheckProfitability = true);
bool SelectImmShifterOperand(SDValue N, SDValue &A,
SDValue &B, bool CheckProfitability = true);
bool SelectShiftRegShifterOperand(SDValue N, SDValue &A,
SDValue &B, SDValue &C) {
// Don't apply the profitability check
return SelectRegShifterOperand(N, A, B, C, false);
}
bool SelectShiftImmShifterOperand(SDValue N, SDValue &A,
SDValue &B) {
// Don't apply the profitability check
return SelectImmShifterOperand(N, A, B, false);
}
bool SelectAddrModeImm12(SDValue N, SDValue &Base, SDValue &OffImm);
bool SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset, SDValue &Opc);
AddrMode2Type SelectAddrMode2Worker(SDValue N, SDValue &Base,
SDValue &Offset, SDValue &Opc);
bool SelectAddrMode2Base(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &Opc) {
return SelectAddrMode2Worker(N, Base, Offset, Opc) == AM2_BASE;
}
bool SelectAddrMode2ShOp(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &Opc) {
return SelectAddrMode2Worker(N, Base, Offset, Opc) == AM2_SHOP;
}
bool SelectAddrMode2(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &Opc) {
SelectAddrMode2Worker(N, Base, Offset, Opc);
// return SelectAddrMode2ShOp(N, Base, Offset, Opc);
// This always matches one way or another.
return true;
}
bool SelectAddrMode2OffsetReg(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc);
bool SelectAddrMode2OffsetImm(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc);
bool SelectAddrMode2OffsetImmPre(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc);
bool SelectAddrOffsetNone(SDValue N, SDValue &Base);
bool SelectAddrMode3(SDValue N, SDValue &Base,
SDValue &Offset, SDValue &Opc);
bool SelectAddrMode3Offset(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc);
bool SelectAddrMode5(SDValue N, SDValue &Base,
SDValue &Offset);
bool SelectAddrMode6(SDNode *Parent, SDValue N, SDValue &Addr,SDValue &Align);
bool SelectAddrMode6Offset(SDNode *Op, SDValue N, SDValue &Offset);
bool SelectAddrModePC(SDValue N, SDValue &Offset, SDValue &Label);
// Thumb Addressing Modes:
bool SelectThumbAddrModeRR(SDValue N, SDValue &Base, SDValue &Offset);
bool SelectThumbAddrModeRI(SDValue N, SDValue &Base, SDValue &Offset,
unsigned Scale);
bool SelectThumbAddrModeRI5S1(SDValue N, SDValue &Base, SDValue &Offset);
bool SelectThumbAddrModeRI5S2(SDValue N, SDValue &Base, SDValue &Offset);
bool SelectThumbAddrModeRI5S4(SDValue N, SDValue &Base, SDValue &Offset);
bool SelectThumbAddrModeImm5S(SDValue N, unsigned Scale, SDValue &Base,
SDValue &OffImm);
bool SelectThumbAddrModeImm5S1(SDValue N, SDValue &Base,
SDValue &OffImm);
bool SelectThumbAddrModeImm5S2(SDValue N, SDValue &Base,
SDValue &OffImm);
bool SelectThumbAddrModeImm5S4(SDValue N, SDValue &Base,
SDValue &OffImm);
bool SelectThumbAddrModeSP(SDValue N, SDValue &Base, SDValue &OffImm);
// Thumb 2 Addressing Modes:
bool SelectT2ShifterOperandReg(SDValue N,
SDValue &BaseReg, SDValue &Opc);
bool SelectT2AddrModeImm12(SDValue N, SDValue &Base, SDValue &OffImm);
bool SelectT2AddrModeImm8(SDValue N, SDValue &Base,
SDValue &OffImm);
bool SelectT2AddrModeImm8Offset(SDNode *Op, SDValue N,
SDValue &OffImm);
bool SelectT2AddrModeSoReg(SDValue N, SDValue &Base,
SDValue &OffReg, SDValue &ShImm);
inline bool is_so_imm(unsigned Imm) const {
return ARM_AM::getSOImmVal(Imm) != -1;
}
inline bool is_so_imm_not(unsigned Imm) const {
return ARM_AM::getSOImmVal(~Imm) != -1;
}
inline bool is_t2_so_imm(unsigned Imm) const {
return ARM_AM::getT2SOImmVal(Imm) != -1;
}
inline bool is_t2_so_imm_not(unsigned Imm) const {
return ARM_AM::getT2SOImmVal(~Imm) != -1;
}
// Include the pieces autogenerated from the target description.
#include "ARMGenDAGISel.inc"
private:
/// SelectARMIndexedLoad - Indexed (pre/post inc/dec) load matching code for
/// ARM.
SDNode *SelectARMIndexedLoad(SDNode *N);
SDNode *SelectT2IndexedLoad(SDNode *N);
/// SelectVLD - Select NEON load intrinsics. NumVecs should be
/// 1, 2, 3 or 4. The opcode arrays specify the instructions used for
/// loads of D registers and even subregs and odd subregs of Q registers.
/// For NumVecs <= 2, QOpcodes1 is not used.
SDNode *SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes,
const uint16_t *QOpcodes0, const uint16_t *QOpcodes1);
/// SelectVST - Select NEON store intrinsics. NumVecs should
/// be 1, 2, 3 or 4. The opcode arrays specify the instructions used for
/// stores of D registers and even subregs and odd subregs of Q registers.
/// For NumVecs <= 2, QOpcodes1 is not used.
SDNode *SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes,
const uint16_t *QOpcodes0, const uint16_t *QOpcodes1);
/// SelectVLDSTLane - Select NEON load/store lane intrinsics. NumVecs should
/// be 2, 3 or 4. The opcode arrays specify the instructions used for
/// load/store of D registers and Q registers.
SDNode *SelectVLDSTLane(SDNode *N, bool IsLoad,
bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes, const uint16_t *QOpcodes);
/// SelectVLDDup - Select NEON load-duplicate intrinsics. NumVecs
/// should be 2, 3 or 4. The opcode array specifies the instructions used
/// for loading D registers. (Q registers are not supported.)
SDNode *SelectVLDDup(SDNode *N, bool isUpdating, unsigned NumVecs,
const uint16_t *Opcodes);
/// SelectVTBL - Select NEON VTBL and VTBX intrinsics. NumVecs should be 2,
/// 3 or 4. These are custom-selected so that a REG_SEQUENCE can be
/// generated to force the table registers to be consecutive.
SDNode *SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs, unsigned Opc);
/// SelectV6T2BitfieldExtractOp - Select SBFX/UBFX instructions for ARM.
SDNode *SelectV6T2BitfieldExtractOp(SDNode *N, bool isSigned);
/// SelectCMOVOp - Select CMOV instructions for ARM.
SDNode *SelectCMOVOp(SDNode *N);
SDNode *SelectT2CMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR,
SDValue InFlag);
SDNode *SelectARMCMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR,
SDValue InFlag);
SDNode *SelectT2CMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR,
SDValue InFlag);
SDNode *SelectARMCMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR,
SDValue InFlag);
// Select special operations if node forms integer ABS pattern
SDNode *SelectABSOp(SDNode *N);
SDNode *SelectConcatVector(SDNode *N);
SDNode *SelectAtomic64(SDNode *Node, unsigned Opc);
/// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
/// inline asm expressions.
virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
char ConstraintCode,
std::vector<SDValue> &OutOps);
// Form pairs of consecutive S, D, or Q registers.
SDNode *PairSRegs(EVT VT, SDValue V0, SDValue V1);
SDNode *PairDRegs(EVT VT, SDValue V0, SDValue V1);
SDNode *PairQRegs(EVT VT, SDValue V0, SDValue V1);
// Form sequences of 4 consecutive S, D, or Q registers.
SDNode *QuadSRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
SDNode *QuadDRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
SDNode *QuadQRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
// Get the alignment operand for a NEON VLD or VST instruction.
SDValue GetVLDSTAlign(SDValue Align, unsigned NumVecs, bool is64BitVector);
};
}
/// isInt32Immediate - This method tests to see if the node is a 32-bit constant
/// operand. If so Imm will receive the 32-bit value.
static bool isInt32Immediate(SDNode *N, unsigned &Imm) {
if (N->getOpcode() == ISD::Constant && N->getValueType(0) == MVT::i32) {
Imm = cast<ConstantSDNode>(N)->getZExtValue();
return true;
}
return false;
}
// isInt32Immediate - This method tests to see if a constant operand.
// If so Imm will receive the 32 bit value.
static bool isInt32Immediate(SDValue N, unsigned &Imm) {
return isInt32Immediate(N.getNode(), Imm);
}
// isOpcWithIntImmediate - This method tests to see if the node is a specific
// opcode and that it has a immediate integer right operand.
// If so Imm will receive the 32 bit value.
static bool isOpcWithIntImmediate(SDNode *N, unsigned Opc, unsigned& Imm) {
return N->getOpcode() == Opc &&
isInt32Immediate(N->getOperand(1).getNode(), Imm);
}
/// \brief Check whether a particular node is a constant value representable as
/// (N * Scale) where (N in [\p RangeMin, \p RangeMax).
///
/// \param ScaledConstant [out] - On success, the pre-scaled constant value.
static bool isScaledConstantInRange(SDValue Node, int Scale,
int RangeMin, int RangeMax,
int &ScaledConstant) {
assert(Scale > 0 && "Invalid scale!");
// Check that this is a constant.
const ConstantSDNode *C = dyn_cast<ConstantSDNode>(Node);
if (!C)
return false;
ScaledConstant = (int) C->getZExtValue();
if ((ScaledConstant % Scale) != 0)
return false;
ScaledConstant /= Scale;
return ScaledConstant >= RangeMin && ScaledConstant < RangeMax;
}
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
/// hasNoVMLxHazardUse - Return true if it's desirable to select a FP MLA / MLS
/// node. VFP / NEON fp VMLA / VMLS instructions have special RAW hazards (at
/// least on current ARM implementations) which should be avoidded.
bool ARMDAGToDAGISel::hasNoVMLxHazardUse(SDNode *N) const {
if (OptLevel == CodeGenOpt::None)
return true;
if (!CheckVMLxHazard)
return true;
if (!Subtarget->isCortexA8() && !Subtarget->isLikeA9() &&
!Subtarget->isSwift())
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
return true;
if (!N->hasOneUse())
return false;
SDNode *Use = *N->use_begin();
if (Use->getOpcode() == ISD::CopyToReg)
return true;
if (Use->isMachineOpcode()) {
const MCInstrDesc &MCID = TII->get(Use->getMachineOpcode());
if (MCID.mayStore())
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
return true;
unsigned Opcode = MCID.getOpcode();
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
if (Opcode == ARM::VMOVRS || Opcode == ARM::VMOVRRD)
return true;
// vmlx feeding into another vmlx. We actually want to unfold
// the use later in the MLxExpansion pass. e.g.
// vmla
// vmla (stall 8 cycles)
//
// vmul (5 cycles)
// vadd (5 cycles)
// vmla
// This adds up to about 18 - 19 cycles.
//
// vmla
// vmul (stall 4 cycles)
// vadd adds up to about 14 cycles.
return TII->isFpMLxInstruction(Opcode);
}
return false;
}
bool ARMDAGToDAGISel::isShifterOpProfitable(const SDValue &Shift,
ARM_AM::ShiftOpc ShOpcVal,
unsigned ShAmt) {
if (!Subtarget->isLikeA9() && !Subtarget->isSwift())
return true;
if (Shift.hasOneUse())
return true;
// R << 2 is free.
return ShOpcVal == ARM_AM::lsl &&
(ShAmt == 2 || (Subtarget->isSwift() && ShAmt == 1));
}
bool ARMDAGToDAGISel::SelectImmShifterOperand(SDValue N,
SDValue &BaseReg,
SDValue &Opc,
bool CheckProfitability) {
if (DisableShifterOp)
return false;
ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode());
// Don't match base register only case. That is matched to a separate
// lower complexity pattern with explicit register operand.
if (ShOpcVal == ARM_AM::no_shift) return false;
BaseReg = N.getOperand(0);
unsigned ShImmVal = 0;
ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1));
if (!RHS) return false;
ShImmVal = RHS->getZExtValue() & 31;
Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal),
MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectRegShifterOperand(SDValue N,
SDValue &BaseReg,
SDValue &ShReg,
SDValue &Opc,
bool CheckProfitability) {
if (DisableShifterOp)
return false;
ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode());
// Don't match base register only case. That is matched to a separate
// lower complexity pattern with explicit register operand.
if (ShOpcVal == ARM_AM::no_shift) return false;
BaseReg = N.getOperand(0);
unsigned ShImmVal = 0;
ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1));
if (RHS) return false;
ShReg = N.getOperand(1);
if (CheckProfitability && !isShifterOpProfitable(N, ShOpcVal, ShImmVal))
return false;
Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal),
MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrModeImm12(SDValue N,
SDValue &Base,
SDValue &OffImm) {
// Match simple R + imm12 operands.
// Base only.
if (N.getOpcode() != ISD::ADD && N.getOpcode() != ISD::SUB &&
!CurDAG->isBaseWithConstantOffset(N)) {
if (N.getOpcode() == ISD::FrameIndex) {
// Match frame index.
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
if (N.getOpcode() == ARMISD::Wrapper &&
!(Subtarget->useMovt() &&
N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
Base = N.getOperand(0);
} else
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
int RHSC = (int)RHS->getZExtValue();
if (N.getOpcode() == ISD::SUB)
RHSC = -RHSC;
if (RHSC >= 0 && RHSC < 0x1000) { // 12 bits (unsigned)
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
return true;
}
}
// Base only.
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &Opc) {
if (N.getOpcode() == ISD::MUL &&
((!Subtarget->isLikeA9() && !Subtarget->isSwift()) || N.hasOneUse())) {
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
// X * [3,5,9] -> X + X * [2,4,8] etc.
int RHSC = (int)RHS->getZExtValue();
if (RHSC & 1) {
RHSC = RHSC & ~1;
ARM_AM::AddrOpc AddSub = ARM_AM::add;
if (RHSC < 0) {
AddSub = ARM_AM::sub;
RHSC = - RHSC;
}
if (isPowerOf2_32(RHSC)) {
unsigned ShAmt = Log2_32(RHSC);
Base = Offset = N.getOperand(0);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt,
ARM_AM::lsl),
MVT::i32);
return true;
}
}
}
}
if (N.getOpcode() != ISD::ADD && N.getOpcode() != ISD::SUB &&
// ISD::OR that is equivalent to an ISD::ADD.
!CurDAG->isBaseWithConstantOffset(N))
return false;
// Leave simple R +/- imm12 operands for LDRi12
if (N.getOpcode() == ISD::ADD || N.getOpcode() == ISD::OR) {
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), /*Scale=*/1,
-0x1000+1, 0x1000, RHSC)) // 12 bits.
return false;
}
// Otherwise this is R +/- [possibly shifted] R.
ARM_AM::AddrOpc AddSub = N.getOpcode() == ISD::SUB ? ARM_AM::sub:ARM_AM::add;
ARM_AM::ShiftOpc ShOpcVal =
ARM_AM::getShiftOpcForNode(N.getOperand(1).getOpcode());
unsigned ShAmt = 0;
Base = N.getOperand(0);
Offset = N.getOperand(1);
if (ShOpcVal != ARM_AM::no_shift) {
// Check to see if the RHS of the shift is a constant, if not, we can't fold
// it.
if (ConstantSDNode *Sh =
dyn_cast<ConstantSDNode>(N.getOperand(1).getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (isShifterOpProfitable(Offset, ShOpcVal, ShAmt))
Offset = N.getOperand(1).getOperand(0);
else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
// Try matching (R shl C) + (R).
if (N.getOpcode() != ISD::SUB && ShOpcVal == ARM_AM::no_shift &&
!(Subtarget->isLikeA9() || Subtarget->isSwift() ||
N.getOperand(0).hasOneUse())) {
ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0).getOpcode());
if (ShOpcVal != ARM_AM::no_shift) {
// Check to see if the RHS of the shift is a constant, if not, we can't
// fold it.
if (ConstantSDNode *Sh =
dyn_cast<ConstantSDNode>(N.getOperand(0).getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt)) {
Offset = N.getOperand(0).getOperand(0);
Base = N.getOperand(1);
} else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
}
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
MVT::i32);
return true;
}
//-----
AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
SDValue &Base,
SDValue &Offset,
SDValue &Opc) {
if (N.getOpcode() == ISD::MUL &&
(!(Subtarget->isLikeA9() || Subtarget->isSwift()) || N.hasOneUse())) {
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
// X * [3,5,9] -> X + X * [2,4,8] etc.
int RHSC = (int)RHS->getZExtValue();
if (RHSC & 1) {
RHSC = RHSC & ~1;
ARM_AM::AddrOpc AddSub = ARM_AM::add;
if (RHSC < 0) {
AddSub = ARM_AM::sub;
RHSC = - RHSC;
}
if (isPowerOf2_32(RHSC)) {
unsigned ShAmt = Log2_32(RHSC);
Base = Offset = N.getOperand(0);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt,
ARM_AM::lsl),
MVT::i32);
return AM2_SHOP;
}
}
}
}
if (N.getOpcode() != ISD::ADD && N.getOpcode() != ISD::SUB &&
// ISD::OR that is equivalent to an ADD.
!CurDAG->isBaseWithConstantOffset(N)) {
Base = N;
if (N.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
} else if (N.getOpcode() == ARMISD::Wrapper &&
!(Subtarget->useMovt() &&
N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
Base = N.getOperand(0);
}
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(ARM_AM::add, 0,
ARM_AM::no_shift),
MVT::i32);
return AM2_BASE;
}
// Match simple R +/- imm12 operands.
if (N.getOpcode() != ISD::SUB) {
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), /*Scale=*/1,
-0x1000+1, 0x1000, RHSC)) { // 12 bits.
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
Offset = CurDAG->getRegister(0, MVT::i32);
ARM_AM::AddrOpc AddSub = ARM_AM::add;
if (RHSC < 0) {
AddSub = ARM_AM::sub;
RHSC = - RHSC;
}
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, RHSC,
ARM_AM::no_shift),
MVT::i32);
return AM2_BASE;
}
}
if ((Subtarget->isLikeA9() || Subtarget->isSwift()) && !N.hasOneUse()) {
// Compute R +/- (R << N) and reuse it.
Base = N;
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(ARM_AM::add, 0,
ARM_AM::no_shift),
MVT::i32);
return AM2_BASE;
}
// Otherwise this is R +/- [possibly shifted] R.
ARM_AM::AddrOpc AddSub = N.getOpcode() != ISD::SUB ? ARM_AM::add:ARM_AM::sub;
ARM_AM::ShiftOpc ShOpcVal =
ARM_AM::getShiftOpcForNode(N.getOperand(1).getOpcode());
unsigned ShAmt = 0;
Base = N.getOperand(0);
Offset = N.getOperand(1);
if (ShOpcVal != ARM_AM::no_shift) {
// Check to see if the RHS of the shift is a constant, if not, we can't fold
// it.
if (ConstantSDNode *Sh =
dyn_cast<ConstantSDNode>(N.getOperand(1).getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (isShifterOpProfitable(Offset, ShOpcVal, ShAmt))
Offset = N.getOperand(1).getOperand(0);
else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
// Try matching (R shl C) + (R).
if (N.getOpcode() != ISD::SUB && ShOpcVal == ARM_AM::no_shift &&
!(Subtarget->isLikeA9() || Subtarget->isSwift() ||
N.getOperand(0).hasOneUse())) {
ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOperand(0).getOpcode());
if (ShOpcVal != ARM_AM::no_shift) {
// Check to see if the RHS of the shift is a constant, if not, we can't
// fold it.
if (ConstantSDNode *Sh =
dyn_cast<ConstantSDNode>(N.getOperand(0).getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (isShifterOpProfitable(N.getOperand(0), ShOpcVal, ShAmt)) {
Offset = N.getOperand(0).getOperand(0);
Base = N.getOperand(1);
} else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
}
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
MVT::i32);
return AM2_SHOP;
}
bool ARMDAGToDAGISel::SelectAddrMode2OffsetReg(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc) {
unsigned Opcode = Op->getOpcode();
ISD::MemIndexedMode AM = (Opcode == ISD::LOAD)
? cast<LoadSDNode>(Op)->getAddressingMode()
: cast<StoreSDNode>(Op)->getAddressingMode();
ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC)
? ARM_AM::add : ARM_AM::sub;
int Val;
if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val))
return false;
Offset = N;
ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode());
unsigned ShAmt = 0;
if (ShOpcVal != ARM_AM::no_shift) {
// Check to see if the RHS of the shift is a constant, if not, we can't fold
// it.
if (ConstantSDNode *Sh = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (isShifterOpProfitable(N, ShOpcVal, ShAmt))
Offset = N.getOperand(0);
else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode2OffsetImmPre(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc) {
unsigned Opcode = Op->getOpcode();
ISD::MemIndexedMode AM = (Opcode == ISD::LOAD)
? cast<LoadSDNode>(Op)->getAddressingMode()
: cast<StoreSDNode>(Op)->getAddressingMode();
ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC)
? ARM_AM::add : ARM_AM::sub;
int Val;
if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits.
if (AddSub == ARM_AM::sub) Val *= -1;
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(Val, MVT::i32);
return true;
}
return false;
}
bool ARMDAGToDAGISel::SelectAddrMode2OffsetImm(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc) {
unsigned Opcode = Op->getOpcode();
ISD::MemIndexedMode AM = (Opcode == ISD::LOAD)
? cast<LoadSDNode>(Op)->getAddressingMode()
: cast<StoreSDNode>(Op)->getAddressingMode();
ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC)
? ARM_AM::add : ARM_AM::sub;
int Val;
if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits.
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, Val,
ARM_AM::no_shift),
MVT::i32);
return true;
}
return false;
}
bool ARMDAGToDAGISel::SelectAddrOffsetNone(SDValue N, SDValue &Base) {
Base = N;
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N,
SDValue &Base, SDValue &Offset,
SDValue &Opc) {
if (N.getOpcode() == ISD::SUB) {
// X - C is canonicalize to X + -C, no need to handle it here.
Base = N.getOperand(0);
Offset = N.getOperand(1);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::sub, 0),MVT::i32);
return true;
}
if (!CurDAG->isBaseWithConstantOffset(N)) {
Base = N;
if (N.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0),MVT::i32);
return true;
}
// If the RHS is +/- imm8, fold into addr mode.
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), /*Scale=*/1,
-256 + 1, 256, RHSC)) { // 8 bits.
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
Offset = CurDAG->getRegister(0, MVT::i32);
ARM_AM::AddrOpc AddSub = ARM_AM::add;
if (RHSC < 0) {
AddSub = ARM_AM::sub;
RHSC = -RHSC;
}
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, RHSC),MVT::i32);
return true;
}
Base = N.getOperand(0);
Offset = N.getOperand(1);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0), MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode3Offset(SDNode *Op, SDValue N,
SDValue &Offset, SDValue &Opc) {
unsigned Opcode = Op->getOpcode();
ISD::MemIndexedMode AM = (Opcode == ISD::LOAD)
? cast<LoadSDNode>(Op)->getAddressingMode()
: cast<StoreSDNode>(Op)->getAddressingMode();
ARM_AM::AddrOpc AddSub = (AM == ISD::PRE_INC || AM == ISD::POST_INC)
? ARM_AM::add : ARM_AM::sub;
int Val;
if (isScaledConstantInRange(N, /*Scale=*/1, 0, 256, Val)) { // 12 bits.
Offset = CurDAG->getRegister(0, MVT::i32);
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, Val), MVT::i32);
return true;
}
Offset = N;
Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, 0), MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode5(SDValue N,
SDValue &Base, SDValue &Offset) {
if (!CurDAG->isBaseWithConstantOffset(N)) {
Base = N;
if (N.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
} else if (N.getOpcode() == ARMISD::Wrapper &&
!(Subtarget->useMovt() &&
N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
Base = N.getOperand(0);
}
Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(ARM_AM::add, 0),
MVT::i32);
return true;
}
// If the RHS is +/- imm8, fold into addr mode.
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), /*Scale=*/4,
-256 + 1, 256, RHSC)) {
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
ARM_AM::AddrOpc AddSub = ARM_AM::add;
if (RHSC < 0) {
AddSub = ARM_AM::sub;
RHSC = -RHSC;
}
Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(AddSub, RHSC),
MVT::i32);
return true;
}
Base = N;
Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(ARM_AM::add, 0),
MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode6(SDNode *Parent, SDValue N, SDValue &Addr,
SDValue &Align) {
Addr = N;
unsigned Alignment = 0;
if (LSBaseSDNode *LSN = dyn_cast<LSBaseSDNode>(Parent)) {
// This case occurs only for VLD1-lane/dup and VST1-lane instructions.
// The maximum alignment is equal to the memory size being referenced.
unsigned LSNAlign = LSN->getAlignment();
unsigned MemSize = LSN->getMemoryVT().getSizeInBits() / 8;
if (LSNAlign >= MemSize && MemSize > 1)
Alignment = MemSize;
} else {
// All other uses of addrmode6 are for intrinsics. For now just record
// the raw alignment value; it will be refined later based on the legal
// alignment operands for the intrinsic.
Alignment = cast<MemIntrinsicSDNode>(Parent)->getAlignment();
}
Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectAddrMode6Offset(SDNode *Op, SDValue N,
SDValue &Offset) {
LSBaseSDNode *LdSt = cast<LSBaseSDNode>(Op);
ISD::MemIndexedMode AM = LdSt->getAddressingMode();
if (AM != ISD::POST_INC)
return false;
Offset = N;
if (ConstantSDNode *NC = dyn_cast<ConstantSDNode>(N)) {
if (NC->getZExtValue() * 8 == LdSt->getMemoryVT().getSizeInBits())
Offset = CurDAG->getRegister(0, MVT::i32);
}
return true;
}
bool ARMDAGToDAGISel::SelectAddrModePC(SDValue N,
SDValue &Offset, SDValue &Label) {
if (N.getOpcode() == ARMISD::PIC_ADD && N.hasOneUse()) {
Offset = N.getOperand(0);
SDValue N1 = N.getOperand(1);
Label = CurDAG->getTargetConstant(cast<ConstantSDNode>(N1)->getZExtValue(),
MVT::i32);
return true;
}
return false;
}
//===----------------------------------------------------------------------===//
// Thumb Addressing Modes
//===----------------------------------------------------------------------===//
bool ARMDAGToDAGISel::SelectThumbAddrModeRR(SDValue N,
SDValue &Base, SDValue &Offset){
if (N.getOpcode() != ISD::ADD && !CurDAG->isBaseWithConstantOffset(N)) {
ConstantSDNode *NC = dyn_cast<ConstantSDNode>(N);
if (!NC || !NC->isNullValue())
return false;
Base = Offset = N;
return true;
}
Base = N.getOperand(0);
Offset = N.getOperand(1);
return true;
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeRI(SDValue N, SDValue &Base,
SDValue &Offset, unsigned Scale) {
if (Scale == 4) {
SDValue TmpBase, TmpOffImm;
if (SelectThumbAddrModeSP(N, TmpBase, TmpOffImm))
return false; // We want to select tLDRspi / tSTRspi instead.
if (N.getOpcode() == ARMISD::Wrapper &&
N.getOperand(0).getOpcode() == ISD::TargetConstantPool)
return false; // We want to select tLDRpci instead.
}
if (!CurDAG->isBaseWithConstantOffset(N))
return false;
// Thumb does not have [sp, r] address mode.
RegisterSDNode *LHSR = dyn_cast<RegisterSDNode>(N.getOperand(0));
RegisterSDNode *RHSR = dyn_cast<RegisterSDNode>(N.getOperand(1));
if ((LHSR && LHSR->getReg() == ARM::SP) ||
(RHSR && RHSR->getReg() == ARM::SP))
return false;
// FIXME: Why do we explicitly check for a match here and then return false?
// Presumably to allow something else to match, but shouldn't this be
// documented?
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), Scale, 0, 32, RHSC))
return false;
Base = N.getOperand(0);
Offset = N.getOperand(1);
return true;
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeRI5S1(SDValue N,
SDValue &Base,
SDValue &Offset) {
return SelectThumbAddrModeRI(N, Base, Offset, 1);
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeRI5S2(SDValue N,
SDValue &Base,
SDValue &Offset) {
return SelectThumbAddrModeRI(N, Base, Offset, 2);
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeRI5S4(SDValue N,
SDValue &Base,
SDValue &Offset) {
return SelectThumbAddrModeRI(N, Base, Offset, 4);
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeImm5S(SDValue N, unsigned Scale,
SDValue &Base, SDValue &OffImm) {
if (Scale == 4) {
SDValue TmpBase, TmpOffImm;
if (SelectThumbAddrModeSP(N, TmpBase, TmpOffImm))
return false; // We want to select tLDRspi / tSTRspi instead.
if (N.getOpcode() == ARMISD::Wrapper &&
N.getOperand(0).getOpcode() == ISD::TargetConstantPool)
return false; // We want to select tLDRpci instead.
}
if (!CurDAG->isBaseWithConstantOffset(N)) {
if (N.getOpcode() == ARMISD::Wrapper &&
!(Subtarget->useMovt() &&
N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
Base = N.getOperand(0);
} else {
Base = N;
}
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
RegisterSDNode *LHSR = dyn_cast<RegisterSDNode>(N.getOperand(0));
RegisterSDNode *RHSR = dyn_cast<RegisterSDNode>(N.getOperand(1));
if ((LHSR && LHSR->getReg() == ARM::SP) ||
(RHSR && RHSR->getReg() == ARM::SP)) {
ConstantSDNode *LHS = dyn_cast<ConstantSDNode>(N.getOperand(0));
ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1));
unsigned LHSC = LHS ? LHS->getZExtValue() : 0;
unsigned RHSC = RHS ? RHS->getZExtValue() : 0;
// Thumb does not have [sp, #imm5] address mode for non-zero imm5.
if (LHSC != 0 || RHSC != 0) return false;
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
// If the RHS is + imm5 * scale, fold into addr mode.
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), Scale, 0, 32, RHSC)) {
Base = N.getOperand(0);
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
return true;
}
Base = N.getOperand(0);
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeImm5S4(SDValue N, SDValue &Base,
SDValue &OffImm) {
return SelectThumbAddrModeImm5S(N, 4, Base, OffImm);
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeImm5S2(SDValue N, SDValue &Base,
SDValue &OffImm) {
return SelectThumbAddrModeImm5S(N, 2, Base, OffImm);
}
bool
ARMDAGToDAGISel::SelectThumbAddrModeImm5S1(SDValue N, SDValue &Base,
SDValue &OffImm) {
return SelectThumbAddrModeImm5S(N, 1, Base, OffImm);
}
bool ARMDAGToDAGISel::SelectThumbAddrModeSP(SDValue N,
SDValue &Base, SDValue &OffImm) {
if (N.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
if (!CurDAG->isBaseWithConstantOffset(N))
return false;
RegisterSDNode *LHSR = dyn_cast<RegisterSDNode>(N.getOperand(0));
if (N.getOperand(0).getOpcode() == ISD::FrameIndex ||
(LHSR && LHSR->getReg() == ARM::SP)) {
// If the RHS is + imm8 * scale, fold into addr mode.
int RHSC;
if (isScaledConstantInRange(N.getOperand(1), /*Scale=*/4, 0, 256, RHSC)) {
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
return true;
}
}
return false;
}
//===----------------------------------------------------------------------===//
// Thumb 2 Addressing Modes
//===----------------------------------------------------------------------===//
bool ARMDAGToDAGISel::SelectT2ShifterOperandReg(SDValue N, SDValue &BaseReg,
SDValue &Opc) {
if (DisableShifterOp)
return false;
ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode());
// Don't match base register only case. That is matched to a separate
// lower complexity pattern with explicit register operand.
if (ShOpcVal == ARM_AM::no_shift) return false;
BaseReg = N.getOperand(0);
unsigned ShImmVal = 0;
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
ShImmVal = RHS->getZExtValue() & 31;
Opc = getI32Imm(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal));
return true;
}
return false;
}
bool ARMDAGToDAGISel::SelectT2AddrModeImm12(SDValue N,
SDValue &Base, SDValue &OffImm) {
// Match simple R + imm12 operands.
// Base only.
if (N.getOpcode() != ISD::ADD && N.getOpcode() != ISD::SUB &&
!CurDAG->isBaseWithConstantOffset(N)) {
if (N.getOpcode() == ISD::FrameIndex) {
// Match frame index.
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
if (N.getOpcode() == ARMISD::Wrapper &&
!(Subtarget->useMovt() &&
N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::TargetConstantPool)
return false; // We want to select t2LDRpci instead.
} else
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
if (SelectT2AddrModeImm8(N, Base, OffImm))
// Let t2LDRi8 handle (R - imm8).
return false;
int RHSC = (int)RHS->getZExtValue();
if (N.getOpcode() == ISD::SUB)
RHSC = -RHSC;
if (RHSC >= 0 && RHSC < 0x1000) { // 12 bits (unsigned)
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
return true;
}
}
// Base only.
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i32);
return true;
}
bool ARMDAGToDAGISel::SelectT2AddrModeImm8(SDValue N,
SDValue &Base, SDValue &OffImm) {
// Match simple R - imm8 operands.
if (N.getOpcode() != ISD::ADD && N.getOpcode() != ISD::SUB &&
!CurDAG->isBaseWithConstantOffset(N))
return false;
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
int RHSC = (int)RHS->getSExtValue();
if (N.getOpcode() == ISD::SUB)
RHSC = -RHSC;
if ((RHSC >= -255) && (RHSC < 0)) { // 8 bits (always negative)
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
return true;
}
}
return false;
}
bool ARMDAGToDAGISel::SelectT2AddrModeImm8Offset(SDNode *Op, SDValue N,
SDValue &OffImm){
unsigned Opcode = Op->getOpcode();
ISD::MemIndexedMode AM = (Opcode == ISD::LOAD)
? cast<LoadSDNode>(Op)->getAddressingMode()
: cast<StoreSDNode>(Op)->getAddressingMode();
int RHSC;
if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x100, RHSC)) { // 8 bits.
OffImm = ((AM == ISD::PRE_INC) || (AM == ISD::POST_INC))
? CurDAG->getTargetConstant(RHSC, MVT::i32)
: CurDAG->getTargetConstant(-RHSC, MVT::i32);
return true;
}
return false;
}
bool ARMDAGToDAGISel::SelectT2AddrModeSoReg(SDValue N,
SDValue &Base,
SDValue &OffReg, SDValue &ShImm) {
// (R - imm8) should be handled by t2LDRi8. The rest are handled by t2LDRi12.
if (N.getOpcode() != ISD::ADD && !CurDAG->isBaseWithConstantOffset(N))
return false;
// Leave (R + imm12) for t2LDRi12, (R - imm8) for t2LDRi8.
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
int RHSC = (int)RHS->getZExtValue();
if (RHSC >= 0 && RHSC < 0x1000) // 12 bits (unsigned)
return false;
else if (RHSC < 0 && RHSC >= -255) // 8 bits
return false;
}
// Look for (R + R) or (R + (R << [1,2,3])).
unsigned ShAmt = 0;
Base = N.getOperand(0);
OffReg = N.getOperand(1);
// Swap if it is ((R << c) + R).
ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(OffReg.getOpcode());
if (ShOpcVal != ARM_AM::lsl) {
ShOpcVal = ARM_AM::getShiftOpcForNode(Base.getOpcode());
if (ShOpcVal == ARM_AM::lsl)
std::swap(Base, OffReg);
}
if (ShOpcVal == ARM_AM::lsl) {
// Check to see if the RHS of the shift is a constant, if not, we can't fold
// it.
if (ConstantSDNode *Sh = dyn_cast<ConstantSDNode>(OffReg.getOperand(1))) {
ShAmt = Sh->getZExtValue();
if (ShAmt < 4 && isShifterOpProfitable(OffReg, ShOpcVal, ShAmt))
OffReg = OffReg.getOperand(0);
else {
ShAmt = 0;
ShOpcVal = ARM_AM::no_shift;
}
} else {
ShOpcVal = ARM_AM::no_shift;
}
}
ShImm = CurDAG->getTargetConstant(ShAmt, MVT::i32);
return true;
}
//===--------------------------------------------------------------------===//
/// getAL - Returns a ARMCC::AL immediate node.
static inline SDValue getAL(SelectionDAG *CurDAG) {
return CurDAG->getTargetConstant((uint64_t)ARMCC::AL, MVT::i32);
}
SDNode *ARMDAGToDAGISel::SelectARMIndexedLoad(SDNode *N) {
LoadSDNode *LD = cast<LoadSDNode>(N);
ISD::MemIndexedMode AM = LD->getAddressingMode();
if (AM == ISD::UNINDEXED)
return NULL;
EVT LoadedVT = LD->getMemoryVT();
SDValue Offset, AMOpc;
bool isPre = (AM == ISD::PRE_INC) || (AM == ISD::PRE_DEC);
unsigned Opcode = 0;
bool Match = false;
if (LoadedVT == MVT::i32 && isPre &&
SelectAddrMode2OffsetImmPre(N, LD->getOffset(), Offset, AMOpc)) {
Opcode = ARM::LDR_PRE_IMM;
Match = true;
} else if (LoadedVT == MVT::i32 && !isPre &&
SelectAddrMode2OffsetImm(N, LD->getOffset(), Offset, AMOpc)) {
Opcode = ARM::LDR_POST_IMM;
Match = true;
} else if (LoadedVT == MVT::i32 &&
SelectAddrMode2OffsetReg(N, LD->getOffset(), Offset, AMOpc)) {
Opcode = isPre ? ARM::LDR_PRE_REG : ARM::LDR_POST_REG;
Match = true;
} else if (LoadedVT == MVT::i16 &&
SelectAddrMode3Offset(N, LD->getOffset(), Offset, AMOpc)) {
Match = true;
Opcode = (LD->getExtensionType() == ISD::SEXTLOAD)
? (isPre ? ARM::LDRSH_PRE : ARM::LDRSH_POST)
: (isPre ? ARM::LDRH_PRE : ARM::LDRH_POST);
} else if (LoadedVT == MVT::i8 || LoadedVT == MVT::i1) {
if (LD->getExtensionType() == ISD::SEXTLOAD) {
if (SelectAddrMode3Offset(N, LD->getOffset(), Offset, AMOpc)) {
Match = true;
Opcode = isPre ? ARM::LDRSB_PRE : ARM::LDRSB_POST;
}
} else {
if (isPre &&
SelectAddrMode2OffsetImmPre(N, LD->getOffset(), Offset, AMOpc)) {
Match = true;
Opcode = ARM::LDRB_PRE_IMM;
} else if (!isPre &&
SelectAddrMode2OffsetImm(N, LD->getOffset(), Offset, AMOpc)) {
Match = true;
Opcode = ARM::LDRB_POST_IMM;
} else if (SelectAddrMode2OffsetReg(N, LD->getOffset(), Offset, AMOpc)) {
Match = true;
Opcode = isPre ? ARM::LDRB_PRE_REG : ARM::LDRB_POST_REG;
}
}
}
if (Match) {
if (Opcode == ARM::LDR_PRE_IMM || Opcode == ARM::LDRB_PRE_IMM) {
SDValue Chain = LD->getChain();
SDValue Base = LD->getBasePtr();
SDValue Ops[]= { Base, AMOpc, getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32), Chain };
return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32,
MVT::i32, MVT::Other, Ops, 5);
} else {
SDValue Chain = LD->getChain();
SDValue Base = LD->getBasePtr();
SDValue Ops[]= { Base, Offset, AMOpc, getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32), Chain };
return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32,
MVT::i32, MVT::Other, Ops, 6);
}
}
return NULL;
}
SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) {
LoadSDNode *LD = cast<LoadSDNode>(N);
ISD::MemIndexedMode AM = LD->getAddressingMode();
if (AM == ISD::UNINDEXED)
return NULL;
EVT LoadedVT = LD->getMemoryVT();
bool isSExtLd = LD->getExtensionType() == ISD::SEXTLOAD;
SDValue Offset;
bool isPre = (AM == ISD::PRE_INC) || (AM == ISD::PRE_DEC);
unsigned Opcode = 0;
bool Match = false;
if (SelectT2AddrModeImm8Offset(N, LD->getOffset(), Offset)) {
switch (LoadedVT.getSimpleVT().SimpleTy) {
case MVT::i32:
Opcode = isPre ? ARM::t2LDR_PRE : ARM::t2LDR_POST;
break;
case MVT::i16:
if (isSExtLd)
Opcode = isPre ? ARM::t2LDRSH_PRE : ARM::t2LDRSH_POST;
else
Opcode = isPre ? ARM::t2LDRH_PRE : ARM::t2LDRH_POST;
break;
case MVT::i8:
case MVT::i1:
if (isSExtLd)
Opcode = isPre ? ARM::t2LDRSB_PRE : ARM::t2LDRSB_POST;
else
Opcode = isPre ? ARM::t2LDRB_PRE : ARM::t2LDRB_POST;
break;
default:
return NULL;
}
Match = true;
}
if (Match) {
SDValue Chain = LD->getChain();
SDValue Base = LD->getBasePtr();
SDValue Ops[]= { Base, Offset, getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32), Chain };
return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32, MVT::i32,
MVT::Other, Ops, 5);
}
return NULL;
}
/// PairSRegs - Form a D register from a pair of S registers.
///
SDNode *ARMDAGToDAGISel::PairSRegs(EVT VT, SDValue V0, SDValue V1) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass =
CurDAG->getTargetConstant(ARM::DPR_VFP2RegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
}
/// PairDRegs - Form a quad register from a pair of D registers.
///
SDNode *ARMDAGToDAGISel::PairDRegs(EVT VT, SDValue V0, SDValue V1) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass = CurDAG->getTargetConstant(ARM::QPRRegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
}
/// PairQRegs - Form 4 consecutive D registers from a pair of Q registers.
///
SDNode *ARMDAGToDAGISel::PairQRegs(EVT VT, SDValue V0, SDValue V1) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
}
/// QuadSRegs - Form 4 consecutive S registers.
///
SDNode *ARMDAGToDAGISel::QuadSRegs(EVT VT, SDValue V0, SDValue V1,
SDValue V2, SDValue V3) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass =
CurDAG->getTargetConstant(ARM::QPR_VFP2RegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32);
SDValue SubReg2 = CurDAG->getTargetConstant(ARM::ssub_2, MVT::i32);
SDValue SubReg3 = CurDAG->getTargetConstant(ARM::ssub_3, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
V2, SubReg2, V3, SubReg3 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
}
/// QuadDRegs - Form 4 consecutive D registers.
///
SDNode *ARMDAGToDAGISel::QuadDRegs(EVT VT, SDValue V0, SDValue V1,
SDValue V2, SDValue V3) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32);
SDValue SubReg2 = CurDAG->getTargetConstant(ARM::dsub_2, MVT::i32);
SDValue SubReg3 = CurDAG->getTargetConstant(ARM::dsub_3, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
V2, SubReg2, V3, SubReg3 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
}
/// QuadQRegs - Form 4 consecutive Q registers.
///
SDNode *ARMDAGToDAGISel::QuadQRegs(EVT VT, SDValue V0, SDValue V1,
SDValue V2, SDValue V3) {
DebugLoc dl = V0.getNode()->getDebugLoc();
SDValue RegClass = CurDAG->getTargetConstant(ARM::QQQQPRRegClassID, MVT::i32);
SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32);
SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32);
SDValue SubReg2 = CurDAG->getTargetConstant(ARM::qsub_2, MVT::i32);
SDValue SubReg3 = CurDAG->getTargetConstant(ARM::qsub_3, MVT::i32);
const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
V2, SubReg2, V3, SubReg3 };
return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
}
/// GetVLDSTAlign - Get the alignment (in bytes) for the alignment operand
/// of a NEON VLD or VST instruction. The supported values depend on the
/// number of registers being loaded.
SDValue ARMDAGToDAGISel::GetVLDSTAlign(SDValue Align, unsigned NumVecs,
bool is64BitVector) {
unsigned NumRegs = NumVecs;
if (!is64BitVector && NumVecs < 3)
NumRegs *= 2;
unsigned Alignment = cast<ConstantSDNode>(Align)->getZExtValue();
if (Alignment >= 32 && NumRegs == 4)
Alignment = 32;
else if (Alignment >= 16 && (NumRegs == 2 || NumRegs == 4))
Alignment = 16;
else if (Alignment >= 8)
Alignment = 8;
else
Alignment = 0;
return CurDAG->getTargetConstant(Alignment, MVT::i32);
}
// Get the register stride update opcode of a VLD/VST instruction that
// is otherwise equivalent to the given fixed stride updating instruction.
static unsigned getVLDSTRegisterUpdateOpcode(unsigned Opc) {
switch (Opc) {
default: break;
case ARM::VLD1d8wb_fixed: return ARM::VLD1d8wb_register;
case ARM::VLD1d16wb_fixed: return ARM::VLD1d16wb_register;
case ARM::VLD1d32wb_fixed: return ARM::VLD1d32wb_register;
case ARM::VLD1d64wb_fixed: return ARM::VLD1d64wb_register;
case ARM::VLD1q8wb_fixed: return ARM::VLD1q8wb_register;
case ARM::VLD1q16wb_fixed: return ARM::VLD1q16wb_register;
case ARM::VLD1q32wb_fixed: return ARM::VLD1q32wb_register;
case ARM::VLD1q64wb_fixed: return ARM::VLD1q64wb_register;
case ARM::VST1d8wb_fixed: return ARM::VST1d8wb_register;
case ARM::VST1d16wb_fixed: return ARM::VST1d16wb_register;
case ARM::VST1d32wb_fixed: return ARM::VST1d32wb_register;
case ARM::VST1d64wb_fixed: return ARM::VST1d64wb_register;
case ARM::VST1q8wb_fixed: return ARM::VST1q8wb_register;
case ARM::VST1q16wb_fixed: return ARM::VST1q16wb_register;
case ARM::VST1q32wb_fixed: return ARM::VST1q32wb_register;
case ARM::VST1q64wb_fixed: return ARM::VST1q64wb_register;
case ARM::VST1d64TPseudoWB_fixed: return ARM::VST1d64TPseudoWB_register;
case ARM::VST1d64QPseudoWB_fixed: return ARM::VST1d64QPseudoWB_register;
case ARM::VLD2d8wb_fixed: return ARM::VLD2d8wb_register;
case ARM::VLD2d16wb_fixed: return ARM::VLD2d16wb_register;
case ARM::VLD2d32wb_fixed: return ARM::VLD2d32wb_register;
case ARM::VLD2q8PseudoWB_fixed: return ARM::VLD2q8PseudoWB_register;
case ARM::VLD2q16PseudoWB_fixed: return ARM::VLD2q16PseudoWB_register;
case ARM::VLD2q32PseudoWB_fixed: return ARM::VLD2q32PseudoWB_register;
case ARM::VST2d8wb_fixed: return ARM::VST2d8wb_register;
case ARM::VST2d16wb_fixed: return ARM::VST2d16wb_register;
case ARM::VST2d32wb_fixed: return ARM::VST2d32wb_register;
case ARM::VST2q8PseudoWB_fixed: return ARM::VST2q8PseudoWB_register;
case ARM::VST2q16PseudoWB_fixed: return ARM::VST2q16PseudoWB_register;
case ARM::VST2q32PseudoWB_fixed: return ARM::VST2q32PseudoWB_register;
case ARM::VLD2DUPd8wb_fixed: return ARM::VLD2DUPd8wb_register;
case ARM::VLD2DUPd16wb_fixed: return ARM::VLD2DUPd16wb_register;
case ARM::VLD2DUPd32wb_fixed: return ARM::VLD2DUPd32wb_register;
}
return Opc; // If not one we handle, return it unchanged.
}
SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes,
const uint16_t *QOpcodes0,
const uint16_t *QOpcodes1) {
assert(NumVecs >= 1 && NumVecs <= 4 && "VLD NumVecs out-of-range");
DebugLoc dl = N->getDebugLoc();
SDValue MemAddr, Align;
unsigned AddrOpIdx = isUpdating ? 1 : 2;
if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
return NULL;
SDValue Chain = N->getOperand(0);
EVT VT = N->getValueType(0);
bool is64BitVector = VT.is64BitVector();
Align = GetVLDSTAlign(Align, NumVecs, is64BitVector);
unsigned OpcodeIndex;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("unhandled vld type");
// Double-register operations:
case MVT::v8i8: OpcodeIndex = 0; break;
case MVT::v4i16: OpcodeIndex = 1; break;
case MVT::v2f32:
case MVT::v2i32: OpcodeIndex = 2; break;
case MVT::v1i64: OpcodeIndex = 3; break;
// Quad-register operations:
case MVT::v16i8: OpcodeIndex = 0; break;
case MVT::v8i16: OpcodeIndex = 1; break;
case MVT::v4f32:
case MVT::v4i32: OpcodeIndex = 2; break;
case MVT::v2i64: OpcodeIndex = 3;
assert(NumVecs == 1 && "v2i64 type only supported for VLD1");
break;
}
EVT ResTy;
if (NumVecs == 1)
ResTy = VT;
else {
unsigned ResTyElts = (NumVecs == 3) ? 4 : NumVecs;
if (!is64BitVector)
ResTyElts *= 2;
ResTy = EVT::getVectorVT(*CurDAG->getContext(), MVT::i64, ResTyElts);
}
std::vector<EVT> ResTys;
ResTys.push_back(ResTy);
if (isUpdating)
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
SDValue Pred = getAL(CurDAG);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SDNode *VLd;
SmallVector<SDValue, 7> Ops;
// Double registers and VLD1/VLD2 quad registers are directly supported.
if (is64BitVector || NumVecs <= 2) {
unsigned Opc = (is64BitVector ? DOpcodes[OpcodeIndex] :
QOpcodes0[OpcodeIndex]);
Ops.push_back(MemAddr);
Ops.push_back(Align);
if (isUpdating) {
SDValue Inc = N->getOperand(AddrOpIdx + 1);
// FIXME: VLD1/VLD2 fixed increment doesn't need Reg0. Remove the reg0
// case entirely when the rest are updated to that form, too.
if ((NumVecs == 1 || NumVecs == 2) && !isa<ConstantSDNode>(Inc.getNode()))
Opc = getVLDSTRegisterUpdateOpcode(Opc);
// We use a VLD1 for v1i64 even if the pseudo says vld2/3/4, so
// check for that explicitly too. Horribly hacky, but temporary.
if ((NumVecs != 1 && NumVecs != 2 && Opc != ARM::VLD1q64wb_fixed) ||
!isa<ConstantSDNode>(Inc.getNode()))
Ops.push_back(isa<ConstantSDNode>(Inc.getNode()) ? Reg0 : Inc);
}
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
VLd = CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
} else {
// Otherwise, quad registers are loaded with two separate instructions,
// where one loads the even registers and the other loads the odd registers.
EVT AddrTy = MemAddr.getValueType();
// Load the even subregs. This is always an updating load, so that it
// provides the address to the second load for the odd subregs.
SDValue ImplDef =
SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, ResTy), 0);
const SDValue OpsA[] = { MemAddr, Align, Reg0, ImplDef, Pred, Reg0, Chain };
SDNode *VLdA = CurDAG->getMachineNode(QOpcodes0[OpcodeIndex], dl,
ResTy, AddrTy, MVT::Other, OpsA, 7);
Chain = SDValue(VLdA, 2);
// Load the odd subregs.
Ops.push_back(SDValue(VLdA, 1));
Ops.push_back(Align);
if (isUpdating) {
SDValue Inc = N->getOperand(AddrOpIdx + 1);
assert(isa<ConstantSDNode>(Inc.getNode()) &&
"only constant post-increment update allowed for VLD3/4");
(void)Inc;
Ops.push_back(Reg0);
}
Ops.push_back(SDValue(VLdA, 0));
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
VLd = CurDAG->getMachineNode(QOpcodes1[OpcodeIndex], dl, ResTys,
Ops.data(), Ops.size());
}
// Transfer memoperands.
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
cast<MachineSDNode>(VLd)->setMemRefs(MemOp, MemOp + 1);
if (NumVecs == 1)
return VLd;
// Extract out the subregisters.
SDValue SuperReg = SDValue(VLd, 0);
assert(ARM::dsub_7 == ARM::dsub_0+7 &&
ARM::qsub_3 == ARM::qsub_0+3 && "Unexpected subreg numbering");
unsigned Sub0 = (is64BitVector ? ARM::dsub_0 : ARM::qsub_0);
for (unsigned Vec = 0; Vec < NumVecs; ++Vec)
ReplaceUses(SDValue(N, Vec),
CurDAG->getTargetExtractSubreg(Sub0 + Vec, dl, VT, SuperReg));
ReplaceUses(SDValue(N, NumVecs), SDValue(VLd, 1));
if (isUpdating)
ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLd, 2));
return NULL;
}
SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes,
const uint16_t *QOpcodes0,
const uint16_t *QOpcodes1) {
assert(NumVecs >= 1 && NumVecs <= 4 && "VST NumVecs out-of-range");
DebugLoc dl = N->getDebugLoc();
SDValue MemAddr, Align;
unsigned AddrOpIdx = isUpdating ? 1 : 2;
unsigned Vec0Idx = 3; // AddrOpIdx + (isUpdating ? 2 : 1)
if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
return NULL;
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
SDValue Chain = N->getOperand(0);
EVT VT = N->getOperand(Vec0Idx).getValueType();
bool is64BitVector = VT.is64BitVector();
Align = GetVLDSTAlign(Align, NumVecs, is64BitVector);
unsigned OpcodeIndex;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("unhandled vst type");
// Double-register operations:
case MVT::v8i8: OpcodeIndex = 0; break;
case MVT::v4i16: OpcodeIndex = 1; break;
case MVT::v2f32:
case MVT::v2i32: OpcodeIndex = 2; break;
case MVT::v1i64: OpcodeIndex = 3; break;
// Quad-register operations:
case MVT::v16i8: OpcodeIndex = 0; break;
case MVT::v8i16: OpcodeIndex = 1; break;
case MVT::v4f32:
case MVT::v4i32: OpcodeIndex = 2; break;
case MVT::v2i64: OpcodeIndex = 3;
assert(NumVecs == 1 && "v2i64 type only supported for VST1");
break;
}
std::vector<EVT> ResTys;
if (isUpdating)
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
SDValue Pred = getAL(CurDAG);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SmallVector<SDValue, 7> Ops;
// Double registers and VST1/VST2 quad registers are directly supported.
if (is64BitVector || NumVecs <= 2) {
SDValue SrcReg;
if (NumVecs == 1) {
SrcReg = N->getOperand(Vec0Idx);
} else if (is64BitVector) {
// Form a REG_SEQUENCE to force register allocation.
SDValue V0 = N->getOperand(Vec0Idx + 0);
SDValue V1 = N->getOperand(Vec0Idx + 1);
if (NumVecs == 2)
SrcReg = SDValue(PairDRegs(MVT::v2i64, V0, V1), 0);
else {
SDValue V2 = N->getOperand(Vec0Idx + 2);
// If it's a vst3, form a quad D-register and leave the last part as
// an undef.
SDValue V3 = (NumVecs == 3)
? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF,dl,VT), 0)
: N->getOperand(Vec0Idx + 3);
SrcReg = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
}
} else {
// Form a QQ register.
SDValue Q0 = N->getOperand(Vec0Idx);
SDValue Q1 = N->getOperand(Vec0Idx + 1);
SrcReg = SDValue(PairQRegs(MVT::v4i64, Q0, Q1), 0);
}
unsigned Opc = (is64BitVector ? DOpcodes[OpcodeIndex] :
QOpcodes0[OpcodeIndex]);
Ops.push_back(MemAddr);
Ops.push_back(Align);
if (isUpdating) {
SDValue Inc = N->getOperand(AddrOpIdx + 1);
// FIXME: VST1/VST2 fixed increment doesn't need Reg0. Remove the reg0
// case entirely when the rest are updated to that form, too.
if (NumVecs <= 2 && !isa<ConstantSDNode>(Inc.getNode()))
Opc = getVLDSTRegisterUpdateOpcode(Opc);
// We use a VST1 for v1i64 even if the pseudo says vld2/3/4, so
// check for that explicitly too. Horribly hacky, but temporary.
if ((NumVecs > 2 && Opc != ARM::VST1q64wb_fixed) ||
!isa<ConstantSDNode>(Inc.getNode()))
Ops.push_back(isa<ConstantSDNode>(Inc.getNode()) ? Reg0 : Inc);
}
Ops.push_back(SrcReg);
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
SDNode *VSt =
CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
// Transfer memoperands.
cast<MachineSDNode>(VSt)->setMemRefs(MemOp, MemOp + 1);
return VSt;
}
// Otherwise, quad registers are stored with two separate instructions,
// where one stores the even registers and the other stores the odd registers.
// Form the QQQQ REG_SEQUENCE.
SDValue V0 = N->getOperand(Vec0Idx + 0);
SDValue V1 = N->getOperand(Vec0Idx + 1);
SDValue V2 = N->getOperand(Vec0Idx + 2);
SDValue V3 = (NumVecs == 3)
? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
: N->getOperand(Vec0Idx + 3);
SDValue RegSeq = SDValue(QuadQRegs(MVT::v8i64, V0, V1, V2, V3), 0);
// Store the even D registers. This is always an updating store, so that it
// provides the address to the second store for the odd subregs.
const SDValue OpsA[] = { MemAddr, Align, Reg0, RegSeq, Pred, Reg0, Chain };
SDNode *VStA = CurDAG->getMachineNode(QOpcodes0[OpcodeIndex], dl,
MemAddr.getValueType(),
MVT::Other, OpsA, 7);
cast<MachineSDNode>(VStA)->setMemRefs(MemOp, MemOp + 1);
Chain = SDValue(VStA, 1);
// Store the odd D registers.
Ops.push_back(SDValue(VStA, 0));
Ops.push_back(Align);
if (isUpdating) {
SDValue Inc = N->getOperand(AddrOpIdx + 1);
assert(isa<ConstantSDNode>(Inc.getNode()) &&
"only constant post-increment update allowed for VST3/4");
(void)Inc;
Ops.push_back(Reg0);
}
Ops.push_back(RegSeq);
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
SDNode *VStB = CurDAG->getMachineNode(QOpcodes1[OpcodeIndex], dl, ResTys,
Ops.data(), Ops.size());
cast<MachineSDNode>(VStB)->setMemRefs(MemOp, MemOp + 1);
return VStB;
}
SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
bool isUpdating, unsigned NumVecs,
const uint16_t *DOpcodes,
const uint16_t *QOpcodes) {
assert(NumVecs >=2 && NumVecs <= 4 && "VLDSTLane NumVecs out-of-range");
DebugLoc dl = N->getDebugLoc();
SDValue MemAddr, Align;
unsigned AddrOpIdx = isUpdating ? 1 : 2;
unsigned Vec0Idx = 3; // AddrOpIdx + (isUpdating ? 2 : 1)
if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
return NULL;
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
SDValue Chain = N->getOperand(0);
unsigned Lane =
cast<ConstantSDNode>(N->getOperand(Vec0Idx + NumVecs))->getZExtValue();
EVT VT = N->getOperand(Vec0Idx).getValueType();
bool is64BitVector = VT.is64BitVector();
unsigned Alignment = 0;
if (NumVecs != 3) {
Alignment = cast<ConstantSDNode>(Align)->getZExtValue();
unsigned NumBytes = NumVecs * VT.getVectorElementType().getSizeInBits()/8;
if (Alignment > NumBytes)
Alignment = NumBytes;
if (Alignment < 8 && Alignment < NumBytes)
Alignment = 0;
// Alignment must be a power of two; make sure of that.
Alignment = (Alignment & -Alignment);
if (Alignment == 1)
Alignment = 0;
}
Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
unsigned OpcodeIndex;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("unhandled vld/vst lane type");
// Double-register operations:
case MVT::v8i8: OpcodeIndex = 0; break;
case MVT::v4i16: OpcodeIndex = 1; break;
case MVT::v2f32:
case MVT::v2i32: OpcodeIndex = 2; break;
// Quad-register operations:
case MVT::v8i16: OpcodeIndex = 0; break;
case MVT::v4f32:
case MVT::v4i32: OpcodeIndex = 1; break;
}
std::vector<EVT> ResTys;
if (IsLoad) {
unsigned ResTyElts = (NumVecs == 3) ? 4 : NumVecs;
if (!is64BitVector)
ResTyElts *= 2;
ResTys.push_back(EVT::getVectorVT(*CurDAG->getContext(),
MVT::i64, ResTyElts));
}
if (isUpdating)
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
SDValue Pred = getAL(CurDAG);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SmallVector<SDValue, 8> Ops;
Ops.push_back(MemAddr);
Ops.push_back(Align);
if (isUpdating) {
SDValue Inc = N->getOperand(AddrOpIdx + 1);
Ops.push_back(isa<ConstantSDNode>(Inc.getNode()) ? Reg0 : Inc);
}
SDValue SuperReg;
SDValue V0 = N->getOperand(Vec0Idx + 0);
SDValue V1 = N->getOperand(Vec0Idx + 1);
if (NumVecs == 2) {
if (is64BitVector)
SuperReg = SDValue(PairDRegs(MVT::v2i64, V0, V1), 0);
else
SuperReg = SDValue(PairQRegs(MVT::v4i64, V0, V1), 0);
} else {
SDValue V2 = N->getOperand(Vec0Idx + 2);
SDValue V3 = (NumVecs == 3)
? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
: N->getOperand(Vec0Idx + 3);
if (is64BitVector)
SuperReg = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
else
SuperReg = SDValue(QuadQRegs(MVT::v8i64, V0, V1, V2, V3), 0);
}
Ops.push_back(SuperReg);
Ops.push_back(getI32Imm(Lane));
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
unsigned Opc = (is64BitVector ? DOpcodes[OpcodeIndex] :
QOpcodes[OpcodeIndex]);
SDNode *VLdLn = CurDAG->getMachineNode(Opc, dl, ResTys,
Ops.data(), Ops.size());
cast<MachineSDNode>(VLdLn)->setMemRefs(MemOp, MemOp + 1);
if (!IsLoad)
return VLdLn;
// Extract the subregisters.
SuperReg = SDValue(VLdLn, 0);
assert(ARM::dsub_7 == ARM::dsub_0+7 &&
ARM::qsub_3 == ARM::qsub_0+3 && "Unexpected subreg numbering");
unsigned Sub0 = is64BitVector ? ARM::dsub_0 : ARM::qsub_0;
for (unsigned Vec = 0; Vec < NumVecs; ++Vec)
ReplaceUses(SDValue(N, Vec),
CurDAG->getTargetExtractSubreg(Sub0 + Vec, dl, VT, SuperReg));
ReplaceUses(SDValue(N, NumVecs), SDValue(VLdLn, 1));
if (isUpdating)
ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLdLn, 2));
return NULL;
}
SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
unsigned NumVecs,
const uint16_t *Opcodes) {
assert(NumVecs >=2 && NumVecs <= 4 && "VLDDup NumVecs out-of-range");
DebugLoc dl = N->getDebugLoc();
SDValue MemAddr, Align;
if (!SelectAddrMode6(N, N->getOperand(1), MemAddr, Align))
return NULL;
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
SDValue Chain = N->getOperand(0);
EVT VT = N->getValueType(0);
unsigned Alignment = 0;
if (NumVecs != 3) {
Alignment = cast<ConstantSDNode>(Align)->getZExtValue();
unsigned NumBytes = NumVecs * VT.getVectorElementType().getSizeInBits()/8;
if (Alignment > NumBytes)
Alignment = NumBytes;
if (Alignment < 8 && Alignment < NumBytes)
Alignment = 0;
// Alignment must be a power of two; make sure of that.
Alignment = (Alignment & -Alignment);
if (Alignment == 1)
Alignment = 0;
}
Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
unsigned OpcodeIndex;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("unhandled vld-dup type");
case MVT::v8i8: OpcodeIndex = 0; break;
case MVT::v4i16: OpcodeIndex = 1; break;
case MVT::v2f32:
case MVT::v2i32: OpcodeIndex = 2; break;
}
SDValue Pred = getAL(CurDAG);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SDValue SuperReg;
unsigned Opc = Opcodes[OpcodeIndex];
SmallVector<SDValue, 6> Ops;
Ops.push_back(MemAddr);
Ops.push_back(Align);
if (isUpdating) {
// fixed-stride update instructions don't have an explicit writeback
// operand. It's implicit in the opcode itself.
SDValue Inc = N->getOperand(2);
if (!isa<ConstantSDNode>(Inc.getNode()))
Ops.push_back(Inc);
// FIXME: VLD3 and VLD4 haven't been updated to that form yet.
else if (NumVecs > 2)
Ops.push_back(Reg0);
}
Ops.push_back(Pred);
Ops.push_back(Reg0);
Ops.push_back(Chain);
unsigned ResTyElts = (NumVecs == 3) ? 4 : NumVecs;
std::vector<EVT> ResTys;
ResTys.push_back(EVT::getVectorVT(*CurDAG->getContext(), MVT::i64,ResTyElts));
if (isUpdating)
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
SDNode *VLdDup =
CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
cast<MachineSDNode>(VLdDup)->setMemRefs(MemOp, MemOp + 1);
SuperReg = SDValue(VLdDup, 0);
// Extract the subregisters.
assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
unsigned SubIdx = ARM::dsub_0;
for (unsigned Vec = 0; Vec < NumVecs; ++Vec)
ReplaceUses(SDValue(N, Vec),
CurDAG->getTargetExtractSubreg(SubIdx+Vec, dl, VT, SuperReg));
ReplaceUses(SDValue(N, NumVecs), SDValue(VLdDup, 1));
if (isUpdating)
ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLdDup, 2));
return NULL;
}
SDNode *ARMDAGToDAGISel::SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs,
unsigned Opc) {
assert(NumVecs >= 2 && NumVecs <= 4 && "VTBL NumVecs out-of-range");
DebugLoc dl = N->getDebugLoc();
EVT VT = N->getValueType(0);
unsigned FirstTblReg = IsExt ? 2 : 1;
// Form a REG_SEQUENCE to force register allocation.
SDValue RegSeq;
SDValue V0 = N->getOperand(FirstTblReg + 0);
SDValue V1 = N->getOperand(FirstTblReg + 1);
if (NumVecs == 2)
RegSeq = SDValue(PairDRegs(MVT::v16i8, V0, V1), 0);
else {
SDValue V2 = N->getOperand(FirstTblReg + 2);
// If it's a vtbl3, form a quad D-register and leave the last part as
// an undef.
SDValue V3 = (NumVecs == 3)
? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
: N->getOperand(FirstTblReg + 3);
RegSeq = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
}
SmallVector<SDValue, 6> Ops;
if (IsExt)
Ops.push_back(N->getOperand(1));
Ops.push_back(RegSeq);
Ops.push_back(N->getOperand(FirstTblReg + NumVecs));
Ops.push_back(getAL(CurDAG)); // predicate
Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // predicate register
return CurDAG->getMachineNode(Opc, dl, VT, Ops.data(), Ops.size());
}
SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
bool isSigned) {
if (!Subtarget->hasV6T2Ops())
return NULL;
unsigned Opc = isSigned ? (Subtarget->isThumb() ? ARM::t2SBFX : ARM::SBFX)
: (Subtarget->isThumb() ? ARM::t2UBFX : ARM::UBFX);
// For unsigned extracts, check for a shift right and mask
unsigned And_imm = 0;
if (N->getOpcode() == ISD::AND) {
if (isOpcWithIntImmediate(N, ISD::AND, And_imm)) {
// The immediate is a mask of the low bits iff imm & (imm+1) == 0
if (And_imm & (And_imm + 1))
return NULL;
unsigned Srl_imm = 0;
if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SRL,
Srl_imm)) {
assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!");
// Note: The width operand is encoded as width-1.
unsigned Width = CountTrailingOnes_32(And_imm) - 1;
unsigned LSB = Srl_imm;
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { N->getOperand(0).getOperand(0),
CurDAG->getTargetConstant(LSB, MVT::i32),
CurDAG->getTargetConstant(Width, MVT::i32),
getAL(CurDAG), Reg0 };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
}
}
return NULL;
}
// Otherwise, we're looking for a shift of a shift
unsigned Shl_imm = 0;
if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SHL, Shl_imm)) {
assert(Shl_imm > 0 && Shl_imm < 32 && "bad amount in shift node!");
unsigned Srl_imm = 0;
if (isInt32Immediate(N->getOperand(1), Srl_imm)) {
assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!");
// Note: The width operand is encoded as width-1.
unsigned Width = 32 - Srl_imm - 1;
int LSB = Srl_imm - Shl_imm;
if (LSB < 0)
return NULL;
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { N->getOperand(0).getOperand(0),
CurDAG->getTargetConstant(LSB, MVT::i32),
CurDAG->getTargetConstant(Width, MVT::i32),
getAL(CurDAG), Reg0 };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
}
}
return NULL;
}
SDNode *ARMDAGToDAGISel::
SelectT2CMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
SDValue CPTmp0;
SDValue CPTmp1;
if (SelectT2ShifterOperandReg(TrueVal, CPTmp0, CPTmp1)) {
unsigned SOVal = cast<ConstantSDNode>(CPTmp1)->getZExtValue();
unsigned SOShOp = ARM_AM::getSORegShOp(SOVal);
unsigned Opc = 0;
switch (SOShOp) {
case ARM_AM::lsl: Opc = ARM::t2MOVCClsl; break;
case ARM_AM::lsr: Opc = ARM::t2MOVCClsr; break;
case ARM_AM::asr: Opc = ARM::t2MOVCCasr; break;
case ARM_AM::ror: Opc = ARM::t2MOVCCror; break;
default:
llvm_unreachable("Unknown so_reg opcode!");
}
SDValue SOShImm =
CurDAG->getTargetConstant(ARM_AM::getSORegOffset(SOVal), MVT::i32);
SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, CPTmp0, SOShImm, CC, CCR, InFlag };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32,Ops, 6);
}
return 0;
}
SDNode *ARMDAGToDAGISel::
SelectARMCMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
SDValue CPTmp0;
SDValue CPTmp1;
SDValue CPTmp2;
if (SelectImmShifterOperand(TrueVal, CPTmp0, CPTmp2)) {
SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, CPTmp0, CPTmp2, CC, CCR, InFlag };
return CurDAG->SelectNodeTo(N, ARM::MOVCCsi, MVT::i32, Ops, 6);
}
if (SelectRegShifterOperand(TrueVal, CPTmp0, CPTmp1, CPTmp2)) {
SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, CPTmp0, CPTmp1, CPTmp2, CC, CCR, InFlag };
return CurDAG->SelectNodeTo(N, ARM::MOVCCsr, MVT::i32, Ops, 7);
}
return 0;
}
SDNode *ARMDAGToDAGISel::
SelectT2CMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
ConstantSDNode *T = dyn_cast<ConstantSDNode>(TrueVal);
if (!T)
return 0;
unsigned Opc = 0;
unsigned TrueImm = T->getZExtValue();
if (is_t2_so_imm(TrueImm)) {
Opc = ARM::t2MOVCCi;
} else if (TrueImm <= 0xffff) {
Opc = ARM::t2MOVCCi16;
} else if (is_t2_so_imm_not(TrueImm)) {
TrueImm = ~TrueImm;
Opc = ARM::t2MVNCCi;
} else if (TrueVal.getNode()->hasOneUse() && Subtarget->hasV6T2Ops()) {
// Large immediate.
Opc = ARM::t2MOVCCi32imm;
}
if (Opc) {
SDValue True = CurDAG->getTargetConstant(TrueImm, MVT::i32);
SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, True, CC, CCR, InFlag };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
}
return 0;
}
SDNode *ARMDAGToDAGISel::
SelectARMCMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
ConstantSDNode *T = dyn_cast<ConstantSDNode>(TrueVal);
if (!T)
return 0;
unsigned Opc = 0;
unsigned TrueImm = T->getZExtValue();
bool isSoImm = is_so_imm(TrueImm);
if (isSoImm) {
Opc = ARM::MOVCCi;
} else if (Subtarget->hasV6T2Ops() && TrueImm <= 0xffff) {
Opc = ARM::MOVCCi16;
} else if (is_so_imm_not(TrueImm)) {
TrueImm = ~TrueImm;
Opc = ARM::MVNCCi;
} else if (TrueVal.getNode()->hasOneUse() &&
(Subtarget->hasV6T2Ops() || ARM_AM::isSOImmTwoPartVal(TrueImm))) {
// Large immediate.
Opc = ARM::MOVCCi32imm;
}
if (Opc) {
SDValue True = CurDAG->getTargetConstant(TrueImm, MVT::i32);
SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, True, CC, CCR, InFlag };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
}
return 0;
}
SDNode *ARMDAGToDAGISel::SelectCMOVOp(SDNode *N) {
EVT VT = N->getValueType(0);
SDValue FalseVal = N->getOperand(0);
SDValue TrueVal = N->getOperand(1);
SDValue CC = N->getOperand(2);
SDValue CCR = N->getOperand(3);
SDValue InFlag = N->getOperand(4);
assert(CC.getOpcode() == ISD::Constant);
assert(CCR.getOpcode() == ISD::Register);
ARMCC::CondCodes CCVal =
(ARMCC::CondCodes)cast<ConstantSDNode>(CC)->getZExtValue();
if (!Subtarget->isThumb1Only() && VT == MVT::i32) {
// Pattern: (ARMcmov:i32 GPR:i32:$false, so_reg:i32:$true, (imm:i32):$cc)
// Emits: (MOVCCs:i32 GPR:i32:$false, so_reg:i32:$true, (imm:i32):$cc)
// Pattern complexity = 18 cost = 1 size = 0
if (Subtarget->isThumb()) {
SDNode *Res = SelectT2CMOVShiftOp(N, FalseVal, TrueVal,
CCVal, CCR, InFlag);
if (!Res)
Res = SelectT2CMOVShiftOp(N, TrueVal, FalseVal,
ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
if (Res)
return Res;
} else {
SDNode *Res = SelectARMCMOVShiftOp(N, FalseVal, TrueVal,
CCVal, CCR, InFlag);
if (!Res)
Res = SelectARMCMOVShiftOp(N, TrueVal, FalseVal,
ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
if (Res)
return Res;
}
// Pattern: (ARMcmov:i32 GPR:i32:$false,
// (imm:i32)<<P:Pred_so_imm>>:$true,
// (imm:i32):$cc)
// Emits: (MOVCCi:i32 GPR:i32:$false,
// (so_imm:i32 (imm:i32):$true), (imm:i32):$cc)
// Pattern complexity = 10 cost = 1 size = 0
if (Subtarget->isThumb()) {
SDNode *Res = SelectT2CMOVImmOp(N, FalseVal, TrueVal,
CCVal, CCR, InFlag);
if (!Res)
Res = SelectT2CMOVImmOp(N, TrueVal, FalseVal,
ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
if (Res)
return Res;
} else {
SDNode *Res = SelectARMCMOVImmOp(N, FalseVal, TrueVal,
CCVal, CCR, InFlag);
if (!Res)
Res = SelectARMCMOVImmOp(N, TrueVal, FalseVal,
ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
if (Res)
return Res;
}
}
// Pattern: (ARMcmov:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
// Emits: (MOVCCr:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
// Pattern complexity = 6 cost = 1 size = 0
//
// Pattern: (ARMcmov:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
// Emits: (tMOVCCr:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
// Pattern complexity = 6 cost = 11 size = 0
//
// Also VMOVScc and VMOVDcc.
SDValue Tmp2 = CurDAG->getTargetConstant(CCVal, MVT::i32);
SDValue Ops[] = { FalseVal, TrueVal, Tmp2, CCR, InFlag };
unsigned Opc = 0;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Illegal conditional move type!");
case MVT::i32:
Opc = Subtarget->isThumb()
? (Subtarget->hasThumb2() ? ARM::t2MOVCCr : ARM::tMOVCCr_pseudo)
: ARM::MOVCCr;
break;
case MVT::f32:
Opc = ARM::VMOVScc;
break;
case MVT::f64:
Opc = ARM::VMOVDcc;
break;
}
return CurDAG->SelectNodeTo(N, Opc, VT, Ops, 5);
}
/// Target-specific DAG combining for ISD::XOR.
/// Target-independent combining lowers SELECT_CC nodes of the form
/// select_cc setg[ge] X, 0, X, -X
/// select_cc setgt X, -1, X, -X
/// select_cc setl[te] X, 0, -X, X
/// select_cc setlt X, 1, -X, X
/// which represent Integer ABS into:
/// Y = sra (X, size(X)-1); xor (add (X, Y), Y)
/// ARM instruction selection detects the latter and matches it to
/// ARM::ABS or ARM::t2ABS machine node.
SDNode *ARMDAGToDAGISel::SelectABSOp(SDNode *N){
SDValue XORSrc0 = N->getOperand(0);
SDValue XORSrc1 = N->getOperand(1);
EVT VT = N->getValueType(0);
if (Subtarget->isThumb1Only())
return NULL;
if (XORSrc0.getOpcode() != ISD::ADD || XORSrc1.getOpcode() != ISD::SRA)
return NULL;
SDValue ADDSrc0 = XORSrc0.getOperand(0);
SDValue ADDSrc1 = XORSrc0.getOperand(1);
SDValue SRASrc0 = XORSrc1.getOperand(0);
SDValue SRASrc1 = XORSrc1.getOperand(1);
ConstantSDNode *SRAConstant = dyn_cast<ConstantSDNode>(SRASrc1);
EVT XType = SRASrc0.getValueType();
unsigned Size = XType.getSizeInBits() - 1;
if (ADDSrc1 == XORSrc1 && ADDSrc0 == SRASrc0 &&
XType.isInteger() && SRAConstant != NULL &&
Size == SRAConstant->getZExtValue()) {
unsigned Opcode = Subtarget->isThumb2() ? ARM::t2ABS : ARM::ABS;
return CurDAG->SelectNodeTo(N, Opcode, VT, ADDSrc0);
}
return NULL;
}
SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) {
// The only time a CONCAT_VECTORS operation can have legal types is when
// two 64-bit vectors are concatenated to a 128-bit vector.
EVT VT = N->getValueType(0);
if (!VT.is128BitVector() || N->getNumOperands() != 2)
llvm_unreachable("unexpected CONCAT_VECTORS");
return PairDRegs(VT, N->getOperand(0), N->getOperand(1));
}
SDNode *ARMDAGToDAGISel::SelectAtomic64(SDNode *Node, unsigned Opc) {
SmallVector<SDValue, 6> Ops;
Ops.push_back(Node->getOperand(1)); // Ptr
Ops.push_back(Node->getOperand(2)); // Low part of Val1
Ops.push_back(Node->getOperand(3)); // High part of Val1
if (Opc == ARM::ATOMCMPXCHG6432) {
Ops.push_back(Node->getOperand(4)); // Low part of Val2
Ops.push_back(Node->getOperand(5)); // High part of Val2
}
Ops.push_back(Node->getOperand(0)); // Chain
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemSDNode>(Node)->getMemOperand();
SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
MVT::i32, MVT::i32, MVT::Other,
Ops.data() ,Ops.size());
cast<MachineSDNode>(ResNode)->setMemRefs(MemOp, MemOp + 1);
return ResNode;
}
SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
DebugLoc dl = N->getDebugLoc();
if (N->isMachineOpcode())
return NULL; // Already selected.
switch (N->getOpcode()) {
default: break;
case ISD::XOR: {
// Select special operations if XOR node forms integer ABS pattern
SDNode *ResNode = SelectABSOp(N);
if (ResNode)
return ResNode;
// Other cases are autogenerated.
break;
}
case ISD::Constant: {
unsigned Val = cast<ConstantSDNode>(N)->getZExtValue();
bool UseCP = true;
if (Subtarget->hasThumb2())
// Thumb2-aware targets have the MOVT instruction, so all immediates can
// be done with MOV + MOVT, at worst.
UseCP = 0;
else {
if (Subtarget->isThumb()) {
UseCP = (Val > 255 && // MOV
~Val > 255 && // MOV + MVN
!ARM_AM::isThumbImmShiftedVal(Val)); // MOV + LSL
} else
UseCP = (ARM_AM::getSOImmVal(Val) == -1 && // MOV
ARM_AM::getSOImmVal(~Val) == -1 && // MVN
!ARM_AM::isSOImmTwoPartVal(Val)); // two instrs.
}
if (UseCP) {
SDValue CPIdx =
CurDAG->getTargetConstantPool(ConstantInt::get(
Type::getInt32Ty(*CurDAG->getContext()), Val),
TLI.getPointerTy());
SDNode *ResNode;
if (Subtarget->isThumb1Only()) {
SDValue Pred = getAL(CurDAG);
SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { CPIdx, Pred, PredReg, CurDAG->getEntryNode() };
ResNode = CurDAG->getMachineNode(ARM::tLDRpci, dl, MVT::i32, MVT::Other,
Ops, 4);
} else {
SDValue Ops[] = {
CPIdx,
CurDAG->getTargetConstant(0, MVT::i32),
getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32),
CurDAG->getEntryNode()
};
ResNode=CurDAG->getMachineNode(ARM::LDRcp, dl, MVT::i32, MVT::Other,
Ops, 5);
}
ReplaceUses(SDValue(N, 0), SDValue(ResNode, 0));
return NULL;
}
// Other cases are autogenerated.
break;
}
case ISD::FrameIndex: {
// Selects to ADDri FI, 0 which in turn will become ADDri SP, imm.
int FI = cast<FrameIndexSDNode>(N)->getIndex();
SDValue TFI = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
if (Subtarget->isThumb1Only()) {
SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
return CurDAG->SelectNodeTo(N, ARM::tADDrSPi, MVT::i32, Ops, 4);
} else {
unsigned Opc = ((Subtarget->isThumb() && Subtarget->hasThumb2()) ?
ARM::t2ADDri : ARM::ADDri);
SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
}
}
case ISD::SRL:
if (SDNode *I = SelectV6T2BitfieldExtractOp(N, false))
return I;
break;
case ISD::SRA:
if (SDNode *I = SelectV6T2BitfieldExtractOp(N, true))
return I;
break;
case ISD::MUL:
if (Subtarget->isThumb1Only())
break;
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
unsigned RHSV = C->getZExtValue();
if (!RHSV) break;
if (isPowerOf2_32(RHSV-1)) { // 2^n+1?
unsigned ShImm = Log2_32(RHSV-1);
if (ShImm >= 32)
break;
SDValue V = N->getOperand(0);
ShImm = ARM_AM::getSORegOpc(ARM_AM::lsl, ShImm);
SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, MVT::i32);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
if (Subtarget->isThumb()) {
SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
return CurDAG->SelectNodeTo(N, ARM::t2ADDrs, MVT::i32, Ops, 6);
} else {
SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
return CurDAG->SelectNodeTo(N, ARM::ADDrsi, MVT::i32, Ops, 7);
}
}
if (isPowerOf2_32(RHSV+1)) { // 2^n-1?
unsigned ShImm = Log2_32(RHSV+1);
if (ShImm >= 32)
break;
SDValue V = N->getOperand(0);
ShImm = ARM_AM::getSORegOpc(ARM_AM::lsl, ShImm);
SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, MVT::i32);
SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
if (Subtarget->isThumb()) {
SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
return CurDAG->SelectNodeTo(N, ARM::t2RSBrs, MVT::i32, Ops, 6);
} else {
SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
return CurDAG->SelectNodeTo(N, ARM::RSBrsi, MVT::i32, Ops, 7);
}
}
}
break;
case ISD::AND: {
// Check for unsigned bitfield extract
if (SDNode *I = SelectV6T2BitfieldExtractOp(N, false))
return I;
// (and (or x, c2), c1) and top 16-bits of c1 and c2 match, lower 16-bits
// of c1 are 0xffff, and lower 16-bit of c2 are 0. That is, the top 16-bits
// are entirely contributed by c2 and lower 16-bits are entirely contributed
// by x. That's equal to (or (and x, 0xffff), (and c1, 0xffff0000)).
// Select it to: "movt x, ((c1 & 0xffff) >> 16)
EVT VT = N->getValueType(0);
if (VT != MVT::i32)
break;
unsigned Opc = (Subtarget->isThumb() && Subtarget->hasThumb2())
? ARM::t2MOVTi16
: (Subtarget->hasV6T2Ops() ? ARM::MOVTi16 : 0);
if (!Opc)
break;
SDValue N0 = N->getOperand(0), N1 = N->getOperand(1);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (!N1C)
break;
if (N0.getOpcode() == ISD::OR && N0.getNode()->hasOneUse()) {
SDValue N2 = N0.getOperand(1);
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2);
if (!N2C)
break;
unsigned N1CVal = N1C->getZExtValue();
unsigned N2CVal = N2C->getZExtValue();
if ((N1CVal & 0xffff0000U) == (N2CVal & 0xffff0000U) &&
(N1CVal & 0xffffU) == 0xffffU &&
(N2CVal & 0xffffU) == 0x0U) {
SDValue Imm16 = CurDAG->getTargetConstant((N2CVal & 0xFFFF0000U) >> 16,
MVT::i32);
SDValue Ops[] = { N0.getOperand(0), Imm16,
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(Opc, dl, VT, Ops, 4);
}
}
break;
}
case ARMISD::VMOVRRD:
return CurDAG->getMachineNode(ARM::VMOVRRD, dl, MVT::i32, MVT::i32,
N->getOperand(0), getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32));
case ISD::UMUL_LOHI: {
if (Subtarget->isThumb1Only())
break;
if (Subtarget->isThumb()) {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(ARM::t2UMULL, dl, MVT::i32, MVT::i32,Ops,4);
} else {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
ARM::UMULL : ARM::UMULLv5,
dl, MVT::i32, MVT::i32, Ops, 5);
}
}
case ISD::SMUL_LOHI: {
if (Subtarget->isThumb1Only())
break;
if (Subtarget->isThumb()) {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(ARM::t2SMULL, dl, MVT::i32, MVT::i32,Ops,4);
} else {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
ARM::SMULL : ARM::SMULLv5,
dl, MVT::i32, MVT::i32, Ops, 5);
}
}
case ARMISD::UMLAL:{
if (Subtarget->isThumb()) {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
N->getOperand(3), getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32)};
return CurDAG->getMachineNode(ARM::t2UMLAL, dl, MVT::i32, MVT::i32, Ops, 6);
}else{
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
N->getOperand(3), getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
ARM::UMLAL : ARM::UMLALv5,
dl, MVT::i32, MVT::i32, Ops, 7);
}
}
case ARMISD::SMLAL:{
if (Subtarget->isThumb()) {
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
N->getOperand(3), getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32)};
return CurDAG->getMachineNode(ARM::t2SMLAL, dl, MVT::i32, MVT::i32, Ops, 6);
}else{
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
N->getOperand(3), getAL(CurDAG),
CurDAG->getRegister(0, MVT::i32),
CurDAG->getRegister(0, MVT::i32) };
return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
ARM::SMLAL : ARM::SMLALv5,
dl, MVT::i32, MVT::i32, Ops, 7);
}
}
case ISD::LOAD: {
SDNode *ResNode = 0;
if (Subtarget->isThumb() && Subtarget->hasThumb2())
ResNode = SelectT2IndexedLoad(N);
else
ResNode = SelectARMIndexedLoad(N);
if (ResNode)
return ResNode;
// Other cases are autogenerated.
break;
}
case ARMISD::BRCOND: {
// Pattern: (ARMbrcond:void (bb:Other):$dst, (imm:i32):$cc)
// Emits: (Bcc:void (bb:Other):$dst, (imm:i32):$cc)
// Pattern complexity = 6 cost = 1 size = 0
// Pattern: (ARMbrcond:void (bb:Other):$dst, (imm:i32):$cc)
// Emits: (tBcc:void (bb:Other):$dst, (imm:i32):$cc)
// Pattern complexity = 6 cost = 1 size = 0
// Pattern: (ARMbrcond:void (bb:Other):$dst, (imm:i32):$cc)
// Emits: (t2Bcc:void (bb:Other):$dst, (imm:i32):$cc)
// Pattern complexity = 6 cost = 1 size = 0
unsigned Opc = Subtarget->isThumb() ?
((Subtarget->hasThumb2()) ? ARM::t2Bcc : ARM::tBcc) : ARM::Bcc;
SDValue Chain = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue N2 = N->getOperand(2);
SDValue N3 = N->getOperand(3);
SDValue InFlag = N->getOperand(4);
assert(N1.getOpcode() == ISD::BasicBlock);
assert(N2.getOpcode() == ISD::Constant);
assert(N3.getOpcode() == ISD::Register);
SDValue Tmp2 = CurDAG->getTargetConstant(((unsigned)
cast<ConstantSDNode>(N2)->getZExtValue()),
MVT::i32);
SDValue Ops[] = { N1, Tmp2, N3, Chain, InFlag };
SDNode *ResNode = CurDAG->getMachineNode(Opc, dl, MVT::Other,
MVT::Glue, Ops, 5);
Chain = SDValue(ResNode, 0);
if (N->getNumValues() == 2) {
InFlag = SDValue(ResNode, 1);
ReplaceUses(SDValue(N, 1), InFlag);
}
ReplaceUses(SDValue(N, 0),
SDValue(Chain.getNode(), Chain.getResNo()));
return NULL;
}
case ARMISD::CMOV:
return SelectCMOVOp(N);
case ARMISD::VZIP: {
unsigned Opc = 0;
EVT VT = N->getValueType(0);
switch (VT.getSimpleVT().SimpleTy) {
default: return NULL;
case MVT::v8i8: Opc = ARM::VZIPd8; break;
case MVT::v4i16: Opc = ARM::VZIPd16; break;
case MVT::v2f32:
// vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
case MVT::v2i32: Opc = ARM::VTRNd32; break;
case MVT::v16i8: Opc = ARM::VZIPq8; break;
case MVT::v8i16: Opc = ARM::VZIPq16; break;
case MVT::v4f32:
case MVT::v4i32: Opc = ARM::VZIPq32; break;
}
SDValue Pred = getAL(CurDAG);
SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
}
case ARMISD::VUZP: {
unsigned Opc = 0;
EVT VT = N->getValueType(0);
switch (VT.getSimpleVT().SimpleTy) {
default: return NULL;
case MVT::v8i8: Opc = ARM::VUZPd8; break;
case MVT::v4i16: Opc = ARM::VUZPd16; break;
case MVT::v2f32:
// vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
case MVT::v2i32: Opc = ARM::VTRNd32; break;
case MVT::v16i8: Opc = ARM::VUZPq8; break;
case MVT::v8i16: Opc = ARM::VUZPq16; break;
case MVT::v4f32:
case MVT::v4i32: Opc = ARM::VUZPq32; break;
}
SDValue Pred = getAL(CurDAG);
SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
}
case ARMISD::VTRN: {
unsigned Opc = 0;
EVT VT = N->getValueType(0);
switch (VT.getSimpleVT().SimpleTy) {
default: return NULL;
case MVT::v8i8: Opc = ARM::VTRNd8; break;
case MVT::v4i16: Opc = ARM::VTRNd16; break;
case MVT::v2f32:
case MVT::v2i32: Opc = ARM::VTRNd32; break;
case MVT::v16i8: Opc = ARM::VTRNq8; break;
case MVT::v8i16: Opc = ARM::VTRNq16; break;
case MVT::v4f32:
case MVT::v4i32: Opc = ARM::VTRNq32; break;
}
SDValue Pred = getAL(CurDAG);
SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
}
case ARMISD::BUILD_VECTOR: {
EVT VecVT = N->getValueType(0);
EVT EltVT = VecVT.getVectorElementType();
unsigned NumElts = VecVT.getVectorNumElements();
if (EltVT == MVT::f64) {
assert(NumElts == 2 && "unexpected type for BUILD_VECTOR");
return PairDRegs(VecVT, N->getOperand(0), N->getOperand(1));
}
assert(EltVT == MVT::f32 && "unexpected type for BUILD_VECTOR");
if (NumElts == 2)
return PairSRegs(VecVT, N->getOperand(0), N->getOperand(1));
assert(NumElts == 4 && "unexpected type for BUILD_VECTOR");
return QuadSRegs(VecVT, N->getOperand(0), N->getOperand(1),
N->getOperand(2), N->getOperand(3));
}
case ARMISD::VLD2DUP: {
static const uint16_t Opcodes[] = { ARM::VLD2DUPd8, ARM::VLD2DUPd16,
ARM::VLD2DUPd32 };
return SelectVLDDup(N, false, 2, Opcodes);
}
case ARMISD::VLD3DUP: {
static const uint16_t Opcodes[] = { ARM::VLD3DUPd8Pseudo,
ARM::VLD3DUPd16Pseudo,
ARM::VLD3DUPd32Pseudo };
return SelectVLDDup(N, false, 3, Opcodes);
}
case ARMISD::VLD4DUP: {
static const uint16_t Opcodes[] = { ARM::VLD4DUPd8Pseudo,
ARM::VLD4DUPd16Pseudo,
ARM::VLD4DUPd32Pseudo };
return SelectVLDDup(N, false, 4, Opcodes);
}
case ARMISD::VLD2DUP_UPD: {
static const uint16_t Opcodes[] = { ARM::VLD2DUPd8wb_fixed,
ARM::VLD2DUPd16wb_fixed,
ARM::VLD2DUPd32wb_fixed };
return SelectVLDDup(N, true, 2, Opcodes);
}
case ARMISD::VLD3DUP_UPD: {
static const uint16_t Opcodes[] = { ARM::VLD3DUPd8Pseudo_UPD,
ARM::VLD3DUPd16Pseudo_UPD,
ARM::VLD3DUPd32Pseudo_UPD };
return SelectVLDDup(N, true, 3, Opcodes);
}
case ARMISD::VLD4DUP_UPD: {
static const uint16_t Opcodes[] = { ARM::VLD4DUPd8Pseudo_UPD,
ARM::VLD4DUPd16Pseudo_UPD,
ARM::VLD4DUPd32Pseudo_UPD };
return SelectVLDDup(N, true, 4, Opcodes);
}
case ARMISD::VLD1_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD1d8wb_fixed,
ARM::VLD1d16wb_fixed,
ARM::VLD1d32wb_fixed,
ARM::VLD1d64wb_fixed };
static const uint16_t QOpcodes[] = { ARM::VLD1q8wb_fixed,
ARM::VLD1q16wb_fixed,
ARM::VLD1q32wb_fixed,
ARM::VLD1q64wb_fixed };
return SelectVLD(N, true, 1, DOpcodes, QOpcodes, 0);
}
case ARMISD::VLD2_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD2d8wb_fixed,
ARM::VLD2d16wb_fixed,
ARM::VLD2d32wb_fixed,
ARM::VLD1q64wb_fixed};
static const uint16_t QOpcodes[] = { ARM::VLD2q8PseudoWB_fixed,
ARM::VLD2q16PseudoWB_fixed,
ARM::VLD2q32PseudoWB_fixed };
return SelectVLD(N, true, 2, DOpcodes, QOpcodes, 0);
}
case ARMISD::VLD3_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD3d8Pseudo_UPD,
ARM::VLD3d16Pseudo_UPD,
ARM::VLD3d32Pseudo_UPD,
ARM::VLD1q64wb_fixed};
static const uint16_t QOpcodes0[] = { ARM::VLD3q8Pseudo_UPD,
ARM::VLD3q16Pseudo_UPD,
ARM::VLD3q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VLD3q8oddPseudo_UPD,
ARM::VLD3q16oddPseudo_UPD,
ARM::VLD3q32oddPseudo_UPD };
return SelectVLD(N, true, 3, DOpcodes, QOpcodes0, QOpcodes1);
}
case ARMISD::VLD4_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD4d8Pseudo_UPD,
ARM::VLD4d16Pseudo_UPD,
ARM::VLD4d32Pseudo_UPD,
ARM::VLD1q64wb_fixed};
static const uint16_t QOpcodes0[] = { ARM::VLD4q8Pseudo_UPD,
ARM::VLD4q16Pseudo_UPD,
ARM::VLD4q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VLD4q8oddPseudo_UPD,
ARM::VLD4q16oddPseudo_UPD,
ARM::VLD4q32oddPseudo_UPD };
return SelectVLD(N, true, 4, DOpcodes, QOpcodes0, QOpcodes1);
}
case ARMISD::VLD2LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD2LNd8Pseudo_UPD,
ARM::VLD2LNd16Pseudo_UPD,
ARM::VLD2LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VLD2LNq16Pseudo_UPD,
ARM::VLD2LNq32Pseudo_UPD };
return SelectVLDSTLane(N, true, true, 2, DOpcodes, QOpcodes);
}
case ARMISD::VLD3LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD3LNd8Pseudo_UPD,
ARM::VLD3LNd16Pseudo_UPD,
ARM::VLD3LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VLD3LNq16Pseudo_UPD,
ARM::VLD3LNq32Pseudo_UPD };
return SelectVLDSTLane(N, true, true, 3, DOpcodes, QOpcodes);
}
case ARMISD::VLD4LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VLD4LNd8Pseudo_UPD,
ARM::VLD4LNd16Pseudo_UPD,
ARM::VLD4LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VLD4LNq16Pseudo_UPD,
ARM::VLD4LNq32Pseudo_UPD };
return SelectVLDSTLane(N, true, true, 4, DOpcodes, QOpcodes);
}
case ARMISD::VST1_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST1d8wb_fixed,
ARM::VST1d16wb_fixed,
ARM::VST1d32wb_fixed,
ARM::VST1d64wb_fixed };
static const uint16_t QOpcodes[] = { ARM::VST1q8wb_fixed,
ARM::VST1q16wb_fixed,
ARM::VST1q32wb_fixed,
ARM::VST1q64wb_fixed };
return SelectVST(N, true, 1, DOpcodes, QOpcodes, 0);
}
case ARMISD::VST2_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST2d8wb_fixed,
ARM::VST2d16wb_fixed,
ARM::VST2d32wb_fixed,
ARM::VST1q64wb_fixed};
static const uint16_t QOpcodes[] = { ARM::VST2q8PseudoWB_fixed,
ARM::VST2q16PseudoWB_fixed,
ARM::VST2q32PseudoWB_fixed };
return SelectVST(N, true, 2, DOpcodes, QOpcodes, 0);
}
case ARMISD::VST3_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST3d8Pseudo_UPD,
ARM::VST3d16Pseudo_UPD,
ARM::VST3d32Pseudo_UPD,
ARM::VST1d64TPseudoWB_fixed};
static const uint16_t QOpcodes0[] = { ARM::VST3q8Pseudo_UPD,
ARM::VST3q16Pseudo_UPD,
ARM::VST3q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VST3q8oddPseudo_UPD,
ARM::VST3q16oddPseudo_UPD,
ARM::VST3q32oddPseudo_UPD };
return SelectVST(N, true, 3, DOpcodes, QOpcodes0, QOpcodes1);
}
case ARMISD::VST4_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST4d8Pseudo_UPD,
ARM::VST4d16Pseudo_UPD,
ARM::VST4d32Pseudo_UPD,
ARM::VST1d64QPseudoWB_fixed};
static const uint16_t QOpcodes0[] = { ARM::VST4q8Pseudo_UPD,
ARM::VST4q16Pseudo_UPD,
ARM::VST4q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VST4q8oddPseudo_UPD,
ARM::VST4q16oddPseudo_UPD,
ARM::VST4q32oddPseudo_UPD };
return SelectVST(N, true, 4, DOpcodes, QOpcodes0, QOpcodes1);
}
case ARMISD::VST2LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST2LNd8Pseudo_UPD,
ARM::VST2LNd16Pseudo_UPD,
ARM::VST2LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VST2LNq16Pseudo_UPD,
ARM::VST2LNq32Pseudo_UPD };
return SelectVLDSTLane(N, false, true, 2, DOpcodes, QOpcodes);
}
case ARMISD::VST3LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST3LNd8Pseudo_UPD,
ARM::VST3LNd16Pseudo_UPD,
ARM::VST3LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VST3LNq16Pseudo_UPD,
ARM::VST3LNq32Pseudo_UPD };
return SelectVLDSTLane(N, false, true, 3, DOpcodes, QOpcodes);
}
case ARMISD::VST4LN_UPD: {
static const uint16_t DOpcodes[] = { ARM::VST4LNd8Pseudo_UPD,
ARM::VST4LNd16Pseudo_UPD,
ARM::VST4LNd32Pseudo_UPD };
static const uint16_t QOpcodes[] = { ARM::VST4LNq16Pseudo_UPD,
ARM::VST4LNq32Pseudo_UPD };
return SelectVLDSTLane(N, false, true, 4, DOpcodes, QOpcodes);
}
case ISD::INTRINSIC_VOID:
case ISD::INTRINSIC_W_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
switch (IntNo) {
default:
break;
case Intrinsic::arm_ldrexd: {
SDValue MemAddr = N->getOperand(2);
DebugLoc dl = N->getDebugLoc();
SDValue Chain = N->getOperand(0);
unsigned NewOpc = ARM::LDREXD;
if (Subtarget->isThumb() && Subtarget->hasThumb2())
NewOpc = ARM::t2LDREXD;
// arm_ldrexd returns a i64 value in {i32, i32}
std::vector<EVT> ResTys;
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
// place arguments in the right order
SmallVector<SDValue, 7> Ops;
Ops.push_back(MemAddr);
Ops.push_back(getAL(CurDAG));
Ops.push_back(CurDAG->getRegister(0, MVT::i32));
Ops.push_back(Chain);
SDNode *Ld = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops.data(),
Ops.size());
// Transfer memoperands.
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
cast<MachineSDNode>(Ld)->setMemRefs(MemOp, MemOp + 1);
// Until there's support for specifing explicit register constraints
// like the use of even/odd register pair, hardcode ldrexd to always
// use the pair [R0, R1] to hold the load result.
Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, ARM::R0,
SDValue(Ld, 0), SDValue(0,0));
Chain = CurDAG->getCopyToReg(Chain, dl, ARM::R1,
SDValue(Ld, 1), Chain.getValue(1));
// Remap uses.
SDValue Glue = Chain.getValue(1);
if (!SDValue(N, 0).use_empty()) {
SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
ARM::R0, MVT::i32, Glue);
Glue = Result.getValue(2);
ReplaceUses(SDValue(N, 0), Result);
}
if (!SDValue(N, 1).use_empty()) {
SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
ARM::R1, MVT::i32, Glue);
Glue = Result.getValue(2);
ReplaceUses(SDValue(N, 1), Result);
}
ReplaceUses(SDValue(N, 2), SDValue(Ld, 2));
return NULL;
}
case Intrinsic::arm_strexd: {
DebugLoc dl = N->getDebugLoc();
SDValue Chain = N->getOperand(0);
SDValue Val0 = N->getOperand(2);
SDValue Val1 = N->getOperand(3);
SDValue MemAddr = N->getOperand(4);
// Until there's support for specifing explicit register constraints
// like the use of even/odd register pair, hardcode strexd to always
// use the pair [R2, R3] to hold the i64 (i32, i32) value to be stored.
Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, ARM::R2, Val0,
SDValue(0, 0));
Chain = CurDAG->getCopyToReg(Chain, dl, ARM::R3, Val1, Chain.getValue(1));
SDValue Glue = Chain.getValue(1);
Val0 = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
ARM::R2, MVT::i32, Glue);
Glue = Val0.getValue(1);
Val1 = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
ARM::R3, MVT::i32, Glue);
// Store exclusive double return a i32 value which is the return status
// of the issued store.
std::vector<EVT> ResTys;
ResTys.push_back(MVT::i32);
ResTys.push_back(MVT::Other);
// place arguments in the right order
SmallVector<SDValue, 7> Ops;
Ops.push_back(Val0);
Ops.push_back(Val1);
Ops.push_back(MemAddr);
Ops.push_back(getAL(CurDAG));
Ops.push_back(CurDAG->getRegister(0, MVT::i32));
Ops.push_back(Chain);
unsigned NewOpc = ARM::STREXD;
if (Subtarget->isThumb() && Subtarget->hasThumb2())
NewOpc = ARM::t2STREXD;
SDNode *St = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops.data(),
Ops.size());
// Transfer memoperands.
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
cast<MachineSDNode>(St)->setMemRefs(MemOp, MemOp + 1);
return St;
}
case Intrinsic::arm_neon_vld1: {
static const uint16_t DOpcodes[] = { ARM::VLD1d8, ARM::VLD1d16,
ARM::VLD1d32, ARM::VLD1d64 };
static const uint16_t QOpcodes[] = { ARM::VLD1q8, ARM::VLD1q16,
ARM::VLD1q32, ARM::VLD1q64};
return SelectVLD(N, false, 1, DOpcodes, QOpcodes, 0);
}
case Intrinsic::arm_neon_vld2: {
static const uint16_t DOpcodes[] = { ARM::VLD2d8, ARM::VLD2d16,
ARM::VLD2d32, ARM::VLD1q64 };
static const uint16_t QOpcodes[] = { ARM::VLD2q8Pseudo, ARM::VLD2q16Pseudo,
ARM::VLD2q32Pseudo };
return SelectVLD(N, false, 2, DOpcodes, QOpcodes, 0);
}
case Intrinsic::arm_neon_vld3: {
static const uint16_t DOpcodes[] = { ARM::VLD3d8Pseudo,
ARM::VLD3d16Pseudo,
ARM::VLD3d32Pseudo,
ARM::VLD1d64TPseudo };
static const uint16_t QOpcodes0[] = { ARM::VLD3q8Pseudo_UPD,
ARM::VLD3q16Pseudo_UPD,
ARM::VLD3q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VLD3q8oddPseudo,
ARM::VLD3q16oddPseudo,
ARM::VLD3q32oddPseudo };
return SelectVLD(N, false, 3, DOpcodes, QOpcodes0, QOpcodes1);
}
case Intrinsic::arm_neon_vld4: {
static const uint16_t DOpcodes[] = { ARM::VLD4d8Pseudo,
ARM::VLD4d16Pseudo,
ARM::VLD4d32Pseudo,
ARM::VLD1d64QPseudo };
static const uint16_t QOpcodes0[] = { ARM::VLD4q8Pseudo_UPD,
ARM::VLD4q16Pseudo_UPD,
ARM::VLD4q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VLD4q8oddPseudo,
ARM::VLD4q16oddPseudo,
ARM::VLD4q32oddPseudo };
return SelectVLD(N, false, 4, DOpcodes, QOpcodes0, QOpcodes1);
}
case Intrinsic::arm_neon_vld2lane: {
static const uint16_t DOpcodes[] = { ARM::VLD2LNd8Pseudo,
ARM::VLD2LNd16Pseudo,
ARM::VLD2LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VLD2LNq16Pseudo,
ARM::VLD2LNq32Pseudo };
return SelectVLDSTLane(N, true, false, 2, DOpcodes, QOpcodes);
}
case Intrinsic::arm_neon_vld3lane: {
static const uint16_t DOpcodes[] = { ARM::VLD3LNd8Pseudo,
ARM::VLD3LNd16Pseudo,
ARM::VLD3LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VLD3LNq16Pseudo,
ARM::VLD3LNq32Pseudo };
return SelectVLDSTLane(N, true, false, 3, DOpcodes, QOpcodes);
}
case Intrinsic::arm_neon_vld4lane: {
static const uint16_t DOpcodes[] = { ARM::VLD4LNd8Pseudo,
ARM::VLD4LNd16Pseudo,
ARM::VLD4LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VLD4LNq16Pseudo,
ARM::VLD4LNq32Pseudo };
return SelectVLDSTLane(N, true, false, 4, DOpcodes, QOpcodes);
}
case Intrinsic::arm_neon_vst1: {
static const uint16_t DOpcodes[] = { ARM::VST1d8, ARM::VST1d16,
ARM::VST1d32, ARM::VST1d64 };
static const uint16_t QOpcodes[] = { ARM::VST1q8, ARM::VST1q16,
ARM::VST1q32, ARM::VST1q64 };
return SelectVST(N, false, 1, DOpcodes, QOpcodes, 0);
}
case Intrinsic::arm_neon_vst2: {
static const uint16_t DOpcodes[] = { ARM::VST2d8, ARM::VST2d16,
ARM::VST2d32, ARM::VST1q64 };
static uint16_t QOpcodes[] = { ARM::VST2q8Pseudo, ARM::VST2q16Pseudo,
ARM::VST2q32Pseudo };
return SelectVST(N, false, 2, DOpcodes, QOpcodes, 0);
}
case Intrinsic::arm_neon_vst3: {
static const uint16_t DOpcodes[] = { ARM::VST3d8Pseudo,
ARM::VST3d16Pseudo,
ARM::VST3d32Pseudo,
ARM::VST1d64TPseudo };
static const uint16_t QOpcodes0[] = { ARM::VST3q8Pseudo_UPD,
ARM::VST3q16Pseudo_UPD,
ARM::VST3q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VST3q8oddPseudo,
ARM::VST3q16oddPseudo,
ARM::VST3q32oddPseudo };
return SelectVST(N, false, 3, DOpcodes, QOpcodes0, QOpcodes1);
}
case Intrinsic::arm_neon_vst4: {
static const uint16_t DOpcodes[] = { ARM::VST4d8Pseudo,
ARM::VST4d16Pseudo,
ARM::VST4d32Pseudo,
ARM::VST1d64QPseudo };
static const uint16_t QOpcodes0[] = { ARM::VST4q8Pseudo_UPD,
ARM::VST4q16Pseudo_UPD,
ARM::VST4q32Pseudo_UPD };
static const uint16_t QOpcodes1[] = { ARM::VST4q8oddPseudo,
ARM::VST4q16oddPseudo,
ARM::VST4q32oddPseudo };
return SelectVST(N, false, 4, DOpcodes, QOpcodes0, QOpcodes1);
}
case Intrinsic::arm_neon_vst2lane: {
static const uint16_t DOpcodes[] = { ARM::VST2LNd8Pseudo,
ARM::VST2LNd16Pseudo,
ARM::VST2LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VST2LNq16Pseudo,
ARM::VST2LNq32Pseudo };
return SelectVLDSTLane(N, false, false, 2, DOpcodes, QOpcodes);
}
case Intrinsic::arm_neon_vst3lane: {
static const uint16_t DOpcodes[] = { ARM::VST3LNd8Pseudo,
ARM::VST3LNd16Pseudo,
ARM::VST3LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VST3LNq16Pseudo,
ARM::VST3LNq32Pseudo };
return SelectVLDSTLane(N, false, false, 3, DOpcodes, QOpcodes);
}
case Intrinsic::arm_neon_vst4lane: {
static const uint16_t DOpcodes[] = { ARM::VST4LNd8Pseudo,
ARM::VST4LNd16Pseudo,
ARM::VST4LNd32Pseudo };
static const uint16_t QOpcodes[] = { ARM::VST4LNq16Pseudo,
ARM::VST4LNq32Pseudo };
return SelectVLDSTLane(N, false, false, 4, DOpcodes, QOpcodes);
}
}
break;
}
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
switch (IntNo) {
default:
break;
case Intrinsic::arm_neon_vtbl2:
return SelectVTBL(N, false, 2, ARM::VTBL2);
case Intrinsic::arm_neon_vtbl3:
return SelectVTBL(N, false, 3, ARM::VTBL3Pseudo);
case Intrinsic::arm_neon_vtbl4:
return SelectVTBL(N, false, 4, ARM::VTBL4Pseudo);
case Intrinsic::arm_neon_vtbx2:
return SelectVTBL(N, true, 2, ARM::VTBX2);
case Intrinsic::arm_neon_vtbx3:
return SelectVTBL(N, true, 3, ARM::VTBX3Pseudo);
case Intrinsic::arm_neon_vtbx4:
return SelectVTBL(N, true, 4, ARM::VTBX4Pseudo);
}
break;
}
case ARMISD::VTBL1: {
DebugLoc dl = N->getDebugLoc();
EVT VT = N->getValueType(0);
SmallVector<SDValue, 6> Ops;
Ops.push_back(N->getOperand(0));
Ops.push_back(N->getOperand(1));
Ops.push_back(getAL(CurDAG)); // Predicate
Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // Predicate Register
return CurDAG->getMachineNode(ARM::VTBL1, dl, VT, Ops.data(), Ops.size());
}
case ARMISD::VTBL2: {
DebugLoc dl = N->getDebugLoc();
EVT VT = N->getValueType(0);
// Form a REG_SEQUENCE to force register allocation.
SDValue V0 = N->getOperand(0);
SDValue V1 = N->getOperand(1);
SDValue RegSeq = SDValue(PairDRegs(MVT::v16i8, V0, V1), 0);
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(N->getOperand(2));
Ops.push_back(getAL(CurDAG)); // Predicate
Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // Predicate Register
return CurDAG->getMachineNode(ARM::VTBL2, dl, VT,
Ops.data(), Ops.size());
}
case ISD::CONCAT_VECTORS:
return SelectConcatVector(N);
case ARMISD::ATOMOR64_DAG:
return SelectAtomic64(N, ARM::ATOMOR6432);
case ARMISD::ATOMXOR64_DAG:
return SelectAtomic64(N, ARM::ATOMXOR6432);
case ARMISD::ATOMADD64_DAG:
return SelectAtomic64(N, ARM::ATOMADD6432);
case ARMISD::ATOMSUB64_DAG:
return SelectAtomic64(N, ARM::ATOMSUB6432);
case ARMISD::ATOMNAND64_DAG:
return SelectAtomic64(N, ARM::ATOMNAND6432);
case ARMISD::ATOMAND64_DAG:
return SelectAtomic64(N, ARM::ATOMAND6432);
case ARMISD::ATOMSWAP64_DAG:
return SelectAtomic64(N, ARM::ATOMSWAP6432);
case ARMISD::ATOMCMPXCHG64_DAG:
return SelectAtomic64(N, ARM::ATOMCMPXCHG6432);
}
return SelectCode(N);
}
bool ARMDAGToDAGISel::
SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
std::vector<SDValue> &OutOps) {
assert(ConstraintCode == 'm' && "unexpected asm memory constraint");
// Require the address to be in a register. That is safe for all ARM
// variants and it is hard to do anything much smarter without knowing
// how the operand is used.
OutOps.push_back(Op);
return false;
}
/// createARMISelDag - This pass converts a legalized DAG into a
/// ARM-specific DAG, ready for instruction scheduling.
///
FunctionPass *llvm::createARMISelDag(ARMBaseTargetMachine &TM,
CodeGenOpt::Level OptLevel) {
return new ARMDAGToDAGISel(TM, OptLevel);
}