shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/lib/Target/X86/X86InstrInfo.cpp
Simon Pilgrim 3a5cf6d99b
[X86] Rename AVX512 VEXTRACT/INSERT??x? to VEXTRACT/INSERT??X? (#116826) 
Use uppercase in the subvector description ("32x2" -> "32X4" etc.) - matches what we already do in VBROADCAST??X?, and we try to use uppercase for all x86 instruction mnemonics anyway (and lowercase just for the arg description suffix).
2024-11-20 08:25:01 +00:00

10925 lines
356 KiB
C++
Raw Blame History

//===-- X86InstrInfo.cpp - X86 Instruction Information --------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file contains the X86 implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//
#include "X86InstrInfo.h"
#include "X86.h"
#include "X86InstrBuilder.h"
#include "X86InstrFoldTables.h"
#include "X86MachineFunctionInfo.h"
#include "X86Subtarget.h"
#include "X86TargetMachine.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/Sequence.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/LivePhysRegs.h"
#include "llvm/CodeGen/LiveVariables.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Module.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCInst.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetOptions.h"
#include <optional>
using namespace llvm;
#define DEBUG_TYPE "x86-instr-info"
#define GET_INSTRINFO_CTOR_DTOR
#include "X86GenInstrInfo.inc"
static cl::opt<bool>
NoFusing("disable-spill-fusing",
cl::desc("Disable fusing of spill code into instructions"),
cl::Hidden);
static cl::opt<bool>
PrintFailedFusing("print-failed-fuse-candidates",
cl::desc("Print instructions that the allocator wants to"
" fuse, but the X86 backend currently can't"),
cl::Hidden);
static cl::opt<bool>
ReMatPICStubLoad("remat-pic-stub-load",
cl::desc("Re-materialize load from stub in PIC mode"),
cl::init(false), cl::Hidden);
static cl::opt<unsigned>
PartialRegUpdateClearance("partial-reg-update-clearance",
cl::desc("Clearance between two register writes "
"for inserting XOR to avoid partial "
"register update"),
cl::init(64), cl::Hidden);
static cl::opt<unsigned> UndefRegClearance(
"undef-reg-clearance",
cl::desc("How many idle instructions we would like before "
"certain undef register reads"),
cl::init(128), cl::Hidden);
// Pin the vtable to this file.
void X86InstrInfo::anchor() {}
X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
: X86GenInstrInfo((STI.isTarget64BitLP64() ? X86::ADJCALLSTACKDOWN64
: X86::ADJCALLSTACKDOWN32),
(STI.isTarget64BitLP64() ? X86::ADJCALLSTACKUP64
: X86::ADJCALLSTACKUP32),
X86::CATCHRET, (STI.is64Bit() ? X86::RET64 : X86::RET32)),
Subtarget(STI), RI(STI.getTargetTriple()) {}
const TargetRegisterClass *
X86InstrInfo::getRegClass(const MCInstrDesc &MCID, unsigned OpNum,
const TargetRegisterInfo *TRI,
const MachineFunction &MF) const {
auto *RC = TargetInstrInfo::getRegClass(MCID, OpNum, TRI, MF);
// If the target does not have egpr, then r16-r31 will be resereved for all
// instructions.
if (!RC || !Subtarget.hasEGPR())
return RC;
if (X86II::canUseApxExtendedReg(MCID))
return RC;
switch (RC->getID()) {
default:
return RC;
case X86::GR8RegClassID:
return &X86::GR8_NOREX2RegClass;
case X86::GR16RegClassID:
return &X86::GR16_NOREX2RegClass;
case X86::GR32RegClassID:
return &X86::GR32_NOREX2RegClass;
case X86::GR64RegClassID:
return &X86::GR64_NOREX2RegClass;
case X86::GR32_NOSPRegClassID:
return &X86::GR32_NOREX2_NOSPRegClass;
case X86::GR64_NOSPRegClassID:
return &X86::GR64_NOREX2_NOSPRegClass;
}
}
bool X86InstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
Register &SrcReg, Register &DstReg,
unsigned &SubIdx) const {
switch (MI.getOpcode()) {
default:
break;
case X86::MOVSX16rr8:
case X86::MOVZX16rr8:
case X86::MOVSX32rr8:
case X86::MOVZX32rr8:
case X86::MOVSX64rr8:
if (!Subtarget.is64Bit())
// It's not always legal to reference the low 8-bit of the larger
// register in 32-bit mode.
return false;
[[fallthrough]];
case X86::MOVSX32rr16:
case X86::MOVZX32rr16:
case X86::MOVSX64rr16:
case X86::MOVSX64rr32: {
if (MI.getOperand(0).getSubReg() || MI.getOperand(1).getSubReg())
// Be conservative.
return false;
SrcReg = MI.getOperand(1).getReg();
DstReg = MI.getOperand(0).getReg();
switch (MI.getOpcode()) {
default:
llvm_unreachable("Unreachable!");
case X86::MOVSX16rr8:
case X86::MOVZX16rr8:
case X86::MOVSX32rr8:
case X86::MOVZX32rr8:
case X86::MOVSX64rr8:
SubIdx = X86::sub_8bit;
break;
case X86::MOVSX32rr16:
case X86::MOVZX32rr16:
case X86::MOVSX64rr16:
SubIdx = X86::sub_16bit;
break;
case X86::MOVSX64rr32:
SubIdx = X86::sub_32bit;
break;
}
return true;
}
}
return false;
}
bool X86InstrInfo::isDataInvariant(MachineInstr &MI) {
if (MI.mayLoad() || MI.mayStore())
return false;
// Some target-independent operations that trivially lower to data-invariant
// instructions.
if (MI.isCopyLike() || MI.isInsertSubreg())
return true;
unsigned Opcode = MI.getOpcode();
using namespace X86;
// On x86 it is believed that imul is constant time w.r.t. the loaded data.
// However, they set flags and are perhaps the most surprisingly constant
// time operations so we call them out here separately.
if (isIMUL(Opcode))
return true;
// Bit scanning and counting instructions that are somewhat surprisingly
// constant time as they scan across bits and do other fairly complex
// operations like popcnt, but are believed to be constant time on x86.
// However, these set flags.
if (isBSF(Opcode) || isBSR(Opcode) || isLZCNT(Opcode) || isPOPCNT(Opcode) ||
isTZCNT(Opcode))
return true;
// Bit manipulation instructions are effectively combinations of basic
// arithmetic ops, and should still execute in constant time. These also
// set flags.
if (isBLCFILL(Opcode) || isBLCI(Opcode) || isBLCIC(Opcode) ||
isBLCMSK(Opcode) || isBLCS(Opcode) || isBLSFILL(Opcode) ||
isBLSI(Opcode) || isBLSIC(Opcode) || isBLSMSK(Opcode) || isBLSR(Opcode) ||
isTZMSK(Opcode))
return true;
// Bit extracting and clearing instructions should execute in constant time,
// and set flags.
if (isBEXTR(Opcode) || isBZHI(Opcode))
return true;
// Shift and rotate.
if (isROL(Opcode) || isROR(Opcode) || isSAR(Opcode) || isSHL(Opcode) ||
isSHR(Opcode) || isSHLD(Opcode) || isSHRD(Opcode))
return true;
// Basic arithmetic is constant time on the input but does set flags.
if (isADC(Opcode) || isADD(Opcode) || isAND(Opcode) || isOR(Opcode) ||
isSBB(Opcode) || isSUB(Opcode) || isXOR(Opcode))
return true;
// Arithmetic with just 32-bit and 64-bit variants and no immediates.
if (isANDN(Opcode))
return true;
// Unary arithmetic operations.
if (isDEC(Opcode) || isINC(Opcode) || isNEG(Opcode))
return true;
// Unlike other arithmetic, NOT doesn't set EFLAGS.
if (isNOT(Opcode))
return true;
// Various move instructions used to zero or sign extend things. Note that we
// intentionally don't support the _NOREX variants as we can't handle that
// register constraint anyways.
if (isMOVSX(Opcode) || isMOVZX(Opcode) || isMOVSXD(Opcode) || isMOV(Opcode))
return true;
// Arithmetic instructions that are both constant time and don't set flags.
if (isRORX(Opcode) || isSARX(Opcode) || isSHLX(Opcode) || isSHRX(Opcode))
return true;
// LEA doesn't actually access memory, and its arithmetic is constant time.
if (isLEA(Opcode))
return true;
// By default, assume that the instruction is not data invariant.
return false;
}
bool X86InstrInfo::isDataInvariantLoad(MachineInstr &MI) {
switch (MI.getOpcode()) {
default:
// By default, assume that the load will immediately leak.
return false;
// On x86 it is believed that imul is constant time w.r.t. the loaded data.
// However, they set flags and are perhaps the most surprisingly constant
// time operations so we call them out here separately.
case X86::IMUL16rm:
case X86::IMUL16rmi:
case X86::IMUL32rm:
case X86::IMUL32rmi:
case X86::IMUL64rm:
case X86::IMUL64rmi32:
// Bit scanning and counting instructions that are somewhat surprisingly
// constant time as they scan across bits and do other fairly complex
// operations like popcnt, but are believed to be constant time on x86.
// However, these set flags.
case X86::BSF16rm:
case X86::BSF32rm:
case X86::BSF64rm:
case X86::BSR16rm:
case X86::BSR32rm:
case X86::BSR64rm:
case X86::LZCNT16rm:
case X86::LZCNT32rm:
case X86::LZCNT64rm:
case X86::POPCNT16rm:
case X86::POPCNT32rm:
case X86::POPCNT64rm:
case X86::TZCNT16rm:
case X86::TZCNT32rm:
case X86::TZCNT64rm:
// Bit manipulation instructions are effectively combinations of basic
// arithmetic ops, and should still execute in constant time. These also
// set flags.
case X86::BLCFILL32rm:
case X86::BLCFILL64rm:
case X86::BLCI32rm:
case X86::BLCI64rm:
case X86::BLCIC32rm:
case X86::BLCIC64rm:
case X86::BLCMSK32rm:
case X86::BLCMSK64rm:
case X86::BLCS32rm:
case X86::BLCS64rm:
case X86::BLSFILL32rm:
case X86::BLSFILL64rm:
case X86::BLSI32rm:
case X86::BLSI64rm:
case X86::BLSIC32rm:
case X86::BLSIC64rm:
case X86::BLSMSK32rm:
case X86::BLSMSK64rm:
case X86::BLSR32rm:
case X86::BLSR64rm:
case X86::TZMSK32rm:
case X86::TZMSK64rm:
// Bit extracting and clearing instructions should execute in constant time,
// and set flags.
case X86::BEXTR32rm:
case X86::BEXTR64rm:
case X86::BEXTRI32mi:
case X86::BEXTRI64mi:
case X86::BZHI32rm:
case X86::BZHI64rm:
// Basic arithmetic is constant time on the input but does set flags.
case X86::ADC8rm:
case X86::ADC16rm:
case X86::ADC32rm:
case X86::ADC64rm:
case X86::ADD8rm:
case X86::ADD16rm:
case X86::ADD32rm:
case X86::ADD64rm:
case X86::AND8rm:
case X86::AND16rm:
case X86::AND32rm:
case X86::AND64rm:
case X86::ANDN32rm:
case X86::ANDN64rm:
case X86::OR8rm:
case X86::OR16rm:
case X86::OR32rm:
case X86::OR64rm:
case X86::SBB8rm:
case X86::SBB16rm:
case X86::SBB32rm:
case X86::SBB64rm:
case X86::SUB8rm:
case X86::SUB16rm:
case X86::SUB32rm:
case X86::SUB64rm:
case X86::XOR8rm:
case X86::XOR16rm:
case X86::XOR32rm:
case X86::XOR64rm:
// Integer multiply w/o affecting flags is still believed to be constant
// time on x86. Called out separately as this is among the most surprising
// instructions to exhibit that behavior.
case X86::MULX32rm:
case X86::MULX64rm:
// Arithmetic instructions that are both constant time and don't set flags.
case X86::RORX32mi:
case X86::RORX64mi:
case X86::SARX32rm:
case X86::SARX64rm:
case X86::SHLX32rm:
case X86::SHLX64rm:
case X86::SHRX32rm:
case X86::SHRX64rm:
// Conversions are believed to be constant time and don't set flags.
case X86::CVTTSD2SI64rm:
case X86::VCVTTSD2SI64rm:
case X86::VCVTTSD2SI64Zrm:
case X86::CVTTSD2SIrm:
case X86::VCVTTSD2SIrm:
case X86::VCVTTSD2SIZrm:
case X86::CVTTSS2SI64rm:
case X86::VCVTTSS2SI64rm:
case X86::VCVTTSS2SI64Zrm:
case X86::CVTTSS2SIrm:
case X86::VCVTTSS2SIrm:
case X86::VCVTTSS2SIZrm:
case X86::CVTSI2SDrm:
case X86::VCVTSI2SDrm:
case X86::VCVTSI2SDZrm:
case X86::CVTSI2SSrm:
case X86::VCVTSI2SSrm:
case X86::VCVTSI2SSZrm:
case X86::CVTSI642SDrm:
case X86::VCVTSI642SDrm:
case X86::VCVTSI642SDZrm:
case X86::CVTSI642SSrm:
case X86::VCVTSI642SSrm:
case X86::VCVTSI642SSZrm:
case X86::CVTSS2SDrm:
case X86::VCVTSS2SDrm:
case X86::VCVTSS2SDZrm:
case X86::CVTSD2SSrm:
case X86::VCVTSD2SSrm:
case X86::VCVTSD2SSZrm:
// AVX512 added unsigned integer conversions.
case X86::VCVTTSD2USI64Zrm:
case X86::VCVTTSD2USIZrm:
case X86::VCVTTSS2USI64Zrm:
case X86::VCVTTSS2USIZrm:
case X86::VCVTUSI2SDZrm:
case X86::VCVTUSI642SDZrm:
case X86::VCVTUSI2SSZrm:
case X86::VCVTUSI642SSZrm:
// Loads to register don't set flags.
case X86::MOV8rm:
case X86::MOV8rm_NOREX:
case X86::MOV16rm:
case X86::MOV32rm:
case X86::MOV64rm:
case X86::MOVSX16rm8:
case X86::MOVSX32rm16:
case X86::MOVSX32rm8:
case X86::MOVSX32rm8_NOREX:
case X86::MOVSX64rm16:
case X86::MOVSX64rm32:
case X86::MOVSX64rm8:
case X86::MOVZX16rm8:
case X86::MOVZX32rm16:
case X86::MOVZX32rm8:
case X86::MOVZX32rm8_NOREX:
case X86::MOVZX64rm16:
case X86::MOVZX64rm8:
return true;
}
}
int X86InstrInfo::getSPAdjust(const MachineInstr &MI) const {
const MachineFunction *MF = MI.getParent()->getParent();
const TargetFrameLowering *TFI = MF->getSubtarget().getFrameLowering();
if (isFrameInstr(MI)) {
int SPAdj = alignTo(getFrameSize(MI), TFI->getStackAlign());
SPAdj -= getFrameAdjustment(MI);
if (!isFrameSetup(MI))
SPAdj = -SPAdj;
return SPAdj;
}
// To know whether a call adjusts the stack, we need information
// that is bound to the following ADJCALLSTACKUP pseudo.
// Look for the next ADJCALLSTACKUP that follows the call.
if (MI.isCall()) {
const MachineBasicBlock *MBB = MI.getParent();
auto I = ++MachineBasicBlock::const_iterator(MI);
for (auto E = MBB->end(); I != E; ++I) {
if (I->getOpcode() == getCallFrameDestroyOpcode() || I->isCall())
break;
}
// If we could not find a frame destroy opcode, then it has already
// been simplified, so we don't care.
if (I->getOpcode() != getCallFrameDestroyOpcode())
return 0;
return -(I->getOperand(1).getImm());
}
// Handle other opcodes we reasonably expect to see in call
// sequences. Note this may include spill/restore of FP/BP.
switch (MI.getOpcode()) {
default:
assert(!(MI.modifiesRegister(X86::RSP, &RI) ||
MI.getDesc().hasImplicitDefOfPhysReg(X86::RSP)) &&
"Unhandled opcode in getSPAdjust");
return 0;
case X86::PUSH32r:
case X86::PUSH32rmm:
case X86::PUSH32rmr:
case X86::PUSH32i:
return 4;
case X86::PUSH64r:
case X86::PUSH64rmm:
case X86::PUSH64rmr:
case X86::PUSH64i32:
return 8;
case X86::POP32r:
case X86::POP32rmm:
case X86::POP32rmr:
return -4;
case X86::POP64r:
case X86::POP64rmm:
case X86::POP64rmr:
return -8;
// FIXME: (implement and) use isAddImmediate in the
// default case instead of the following ADD/SUB cases.
case X86::ADD32ri:
case X86::ADD32ri8:
case X86::ADD64ri32:
if (MI.getOperand(0).getReg() == X86::RSP &&
MI.getOperand(1).getReg() == X86::RSP)
return -MI.getOperand(2).getImm();
return 0;
case X86::SUB32ri:
case X86::SUB32ri8:
case X86::SUB64ri32:
if (MI.getOperand(0).getReg() == X86::RSP &&
MI.getOperand(1).getReg() == X86::RSP)
return MI.getOperand(2).getImm();
return 0;
}
}
/// Return true and the FrameIndex if the specified
/// operand and follow operands form a reference to the stack frame.
bool X86InstrInfo::isFrameOperand(const MachineInstr &MI, unsigned int Op,
int &FrameIndex) const {
if (MI.getOperand(Op + X86::AddrBaseReg).isFI() &&
MI.getOperand(Op + X86::AddrScaleAmt).isImm() &&
MI.getOperand(Op + X86::AddrIndexReg).isReg() &&
MI.getOperand(Op + X86::AddrDisp).isImm() &&
MI.getOperand(Op + X86::AddrScaleAmt).getImm() == 1 &&
MI.getOperand(Op + X86::AddrIndexReg).getReg() == 0 &&
MI.getOperand(Op + X86::AddrDisp).getImm() == 0) {
FrameIndex = MI.getOperand(Op + X86::AddrBaseReg).getIndex();
return true;
}
return false;
}
static bool isFrameLoadOpcode(int Opcode, unsigned &MemBytes) {
switch (Opcode) {
default:
return false;
case X86::MOV8rm:
case X86::KMOVBkm:
case X86::KMOVBkm_EVEX:
MemBytes = 1;
return true;
case X86::MOV16rm:
case X86::KMOVWkm:
case X86::KMOVWkm_EVEX:
case X86::VMOVSHZrm:
case X86::VMOVSHZrm_alt:
MemBytes = 2;
return true;
case X86::MOV32rm:
case X86::MOVSSrm:
case X86::MOVSSrm_alt:
case X86::VMOVSSrm:
case X86::VMOVSSrm_alt:
case X86::VMOVSSZrm:
case X86::VMOVSSZrm_alt:
case X86::KMOVDkm:
case X86::KMOVDkm_EVEX:
MemBytes = 4;
return true;
case X86::MOV64rm:
case X86::LD_Fp64m:
case X86::MOVSDrm:
case X86::MOVSDrm_alt:
case X86::VMOVSDrm:
case X86::VMOVSDrm_alt:
case X86::VMOVSDZrm:
case X86::VMOVSDZrm_alt:
case X86::MMX_MOVD64rm:
case X86::MMX_MOVQ64rm:
case X86::KMOVQkm:
case X86::KMOVQkm_EVEX:
MemBytes = 8;
return true;
case X86::MOVAPSrm:
case X86::MOVUPSrm:
case X86::MOVAPDrm:
case X86::MOVUPDrm:
case X86::MOVDQArm:
case X86::MOVDQUrm:
case X86::VMOVAPSrm:
case X86::VMOVUPSrm:
case X86::VMOVAPDrm:
case X86::VMOVUPDrm:
case X86::VMOVDQArm:
case X86::VMOVDQUrm:
case X86::VMOVAPSZ128rm:
case X86::VMOVUPSZ128rm:
case X86::VMOVAPSZ128rm_NOVLX:
case X86::VMOVUPSZ128rm_NOVLX:
case X86::VMOVAPDZ128rm:
case X86::VMOVUPDZ128rm:
case X86::VMOVDQU8Z128rm:
case X86::VMOVDQU16Z128rm:
case X86::VMOVDQA32Z128rm:
case X86::VMOVDQU32Z128rm:
case X86::VMOVDQA64Z128rm:
case X86::VMOVDQU64Z128rm:
MemBytes = 16;
return true;
case X86::VMOVAPSYrm:
case X86::VMOVUPSYrm:
case X86::VMOVAPDYrm:
case X86::VMOVUPDYrm:
case X86::VMOVDQAYrm:
case X86::VMOVDQUYrm:
case X86::VMOVAPSZ256rm:
case X86::VMOVUPSZ256rm:
case X86::VMOVAPSZ256rm_NOVLX:
case X86::VMOVUPSZ256rm_NOVLX:
case X86::VMOVAPDZ256rm:
case X86::VMOVUPDZ256rm:
case X86::VMOVDQU8Z256rm:
case X86::VMOVDQU16Z256rm:
case X86::VMOVDQA32Z256rm:
case X86::VMOVDQU32Z256rm:
case X86::VMOVDQA64Z256rm:
case X86::VMOVDQU64Z256rm:
MemBytes = 32;
return true;
case X86::VMOVAPSZrm:
case X86::VMOVUPSZrm:
case X86::VMOVAPDZrm:
case X86::VMOVUPDZrm:
case X86::VMOVDQU8Zrm:
case X86::VMOVDQU16Zrm:
case X86::VMOVDQA32Zrm:
case X86::VMOVDQU32Zrm:
case X86::VMOVDQA64Zrm:
case X86::VMOVDQU64Zrm:
MemBytes = 64;
return true;
}
}
static bool isFrameStoreOpcode(int Opcode, unsigned &MemBytes) {
switch (Opcode) {
default:
return false;
case X86::MOV8mr:
case X86::KMOVBmk:
case X86::KMOVBmk_EVEX:
MemBytes = 1;
return true;
case X86::MOV16mr:
case X86::KMOVWmk:
case X86::KMOVWmk_EVEX:
case X86::VMOVSHZmr:
MemBytes = 2;
return true;
case X86::MOV32mr:
case X86::MOVSSmr:
case X86::VMOVSSmr:
case X86::VMOVSSZmr:
case X86::KMOVDmk:
case X86::KMOVDmk_EVEX:
MemBytes = 4;
return true;
case X86::MOV64mr:
case X86::ST_FpP64m:
case X86::MOVSDmr:
case X86::VMOVSDmr:
case X86::VMOVSDZmr:
case X86::MMX_MOVD64mr:
case X86::MMX_MOVQ64mr:
case X86::MMX_MOVNTQmr:
case X86::KMOVQmk:
case X86::KMOVQmk_EVEX:
MemBytes = 8;
return true;
case X86::MOVAPSmr:
case X86::MOVUPSmr:
case X86::MOVAPDmr:
case X86::MOVUPDmr:
case X86::MOVDQAmr:
case X86::MOVDQUmr:
case X86::VMOVAPSmr:
case X86::VMOVUPSmr:
case X86::VMOVAPDmr:
case X86::VMOVUPDmr:
case X86::VMOVDQAmr:
case X86::VMOVDQUmr:
case X86::VMOVUPSZ128mr:
case X86::VMOVAPSZ128mr:
case X86::VMOVUPSZ128mr_NOVLX:
case X86::VMOVAPSZ128mr_NOVLX:
case X86::VMOVUPDZ128mr:
case X86::VMOVAPDZ128mr:
case X86::VMOVDQA32Z128mr:
case X86::VMOVDQU32Z128mr:
case X86::VMOVDQA64Z128mr:
case X86::VMOVDQU64Z128mr:
case X86::VMOVDQU8Z128mr:
case X86::VMOVDQU16Z128mr:
MemBytes = 16;
return true;
case X86::VMOVUPSYmr:
case X86::VMOVAPSYmr:
case X86::VMOVUPDYmr:
case X86::VMOVAPDYmr:
case X86::VMOVDQUYmr:
case X86::VMOVDQAYmr:
case X86::VMOVUPSZ256mr:
case X86::VMOVAPSZ256mr:
case X86::VMOVUPSZ256mr_NOVLX:
case X86::VMOVAPSZ256mr_NOVLX:
case X86::VMOVUPDZ256mr:
case X86::VMOVAPDZ256mr:
case X86::VMOVDQU8Z256mr:
case X86::VMOVDQU16Z256mr:
case X86::VMOVDQA32Z256mr:
case X86::VMOVDQU32Z256mr:
case X86::VMOVDQA64Z256mr:
case X86::VMOVDQU64Z256mr:
MemBytes = 32;
return true;
case X86::VMOVUPSZmr:
case X86::VMOVAPSZmr:
case X86::VMOVUPDZmr:
case X86::VMOVAPDZmr:
case X86::VMOVDQU8Zmr:
case X86::VMOVDQU16Zmr:
case X86::VMOVDQA32Zmr:
case X86::VMOVDQU32Zmr:
case X86::VMOVDQA64Zmr:
case X86::VMOVDQU64Zmr:
MemBytes = 64;
return true;
}
return false;
}
Register X86InstrInfo::isLoadFromStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Dummy;
return X86InstrInfo::isLoadFromStackSlot(MI, FrameIndex, Dummy);
}
Register X86InstrInfo::isLoadFromStackSlot(const MachineInstr &MI,
int &FrameIndex,
unsigned &MemBytes) const {
if (isFrameLoadOpcode(MI.getOpcode(), MemBytes))
if (MI.getOperand(0).getSubReg() == 0 && isFrameOperand(MI, 1, FrameIndex))
return MI.getOperand(0).getReg();
return 0;
}
Register X86InstrInfo::isLoadFromStackSlotPostFE(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Dummy;
if (isFrameLoadOpcode(MI.getOpcode(), Dummy)) {
unsigned Reg;
if ((Reg = isLoadFromStackSlot(MI, FrameIndex)))
return Reg;
// Check for post-frame index elimination operations
SmallVector<const MachineMemOperand *, 1> Accesses;
if (hasLoadFromStackSlot(MI, Accesses)) {
FrameIndex =
cast<FixedStackPseudoSourceValue>(Accesses.front()->getPseudoValue())
->getFrameIndex();
return MI.getOperand(0).getReg();
}
}
return 0;
}
Register X86InstrInfo::isStoreToStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Dummy;
return X86InstrInfo::isStoreToStackSlot(MI, FrameIndex, Dummy);
}
Register X86InstrInfo::isStoreToStackSlot(const MachineInstr &MI,
int &FrameIndex,
unsigned &MemBytes) const {
if (isFrameStoreOpcode(MI.getOpcode(), MemBytes))
if (MI.getOperand(X86::AddrNumOperands).getSubReg() == 0 &&
isFrameOperand(MI, 0, FrameIndex))
return MI.getOperand(X86::AddrNumOperands).getReg();
return 0;
}
Register X86InstrInfo::isStoreToStackSlotPostFE(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Dummy;
if (isFrameStoreOpcode(MI.getOpcode(), Dummy)) {
unsigned Reg;
if ((Reg = isStoreToStackSlot(MI, FrameIndex)))
return Reg;
// Check for post-frame index elimination operations
SmallVector<const MachineMemOperand *, 1> Accesses;
if (hasStoreToStackSlot(MI, Accesses)) {
FrameIndex =
cast<FixedStackPseudoSourceValue>(Accesses.front()->getPseudoValue())
->getFrameIndex();
return MI.getOperand(X86::AddrNumOperands).getReg();
}
}
return 0;
}
/// Return true if register is PIC base; i.e.g defined by X86::MOVPC32r.
static bool regIsPICBase(Register BaseReg, const MachineRegisterInfo &MRI) {
// Don't waste compile time scanning use-def chains of physregs.
if (!BaseReg.isVirtual())
return false;
bool isPICBase = false;
for (const MachineInstr &DefMI : MRI.def_instructions(BaseReg)) {
if (DefMI.getOpcode() != X86::MOVPC32r)
return false;
assert(!isPICBase && "More than one PIC base?");
isPICBase = true;
}
return isPICBase;
}
bool X86InstrInfo::isReallyTriviallyReMaterializable(
const MachineInstr &MI) const {
switch (MI.getOpcode()) {
default:
// This function should only be called for opcodes with the ReMaterializable
// flag set.
llvm_unreachable("Unknown rematerializable operation!");
break;
case X86::IMPLICIT_DEF:
// Defer to generic logic.
break;
case X86::LOAD_STACK_GUARD:
case X86::LD_Fp032:
case X86::LD_Fp064:
case X86::LD_Fp080:
case X86::LD_Fp132:
case X86::LD_Fp164:
case X86::LD_Fp180:
case X86::AVX1_SETALLONES:
case X86::AVX2_SETALLONES:
case X86::AVX512_128_SET0:
case X86::AVX512_256_SET0:
case X86::AVX512_512_SET0:
case X86::AVX512_512_SETALLONES:
case X86::AVX512_FsFLD0SD:
case X86::AVX512_FsFLD0SH:
case X86::AVX512_FsFLD0SS:
case X86::AVX512_FsFLD0F128:
case X86::AVX_SET0:
case X86::FsFLD0SD:
case X86::FsFLD0SS:
case X86::FsFLD0SH:
case X86::FsFLD0F128:
case X86::KSET0D:
case X86::KSET0Q:
case X86::KSET0W:
case X86::KSET1D:
case X86::KSET1Q:
case X86::KSET1W:
case X86::MMX_SET0:
case X86::MOV32ImmSExti8:
case X86::MOV32r0:
case X86::MOV32r1:
case X86::MOV32r_1:
case X86::MOV32ri64:
case X86::MOV64ImmSExti8:
case X86::V_SET0:
case X86::V_SETALLONES:
case X86::MOV16ri:
case X86::MOV32ri:
case X86::MOV64ri:
case X86::MOV64ri32:
case X86::MOV8ri:
case X86::PTILEZEROV:
return true;
case X86::MOV8rm:
case X86::MOV8rm_NOREX:
case X86::MOV16rm:
case X86::MOV32rm:
case X86::MOV64rm:
case X86::MOVSSrm:
case X86::MOVSSrm_alt:
case X86::MOVSDrm:
case X86::MOVSDrm_alt:
case X86::MOVAPSrm:
case X86::MOVUPSrm:
case X86::MOVAPDrm:
case X86::MOVUPDrm:
case X86::MOVDQArm:
case X86::MOVDQUrm:
case X86::VMOVSSrm:
case X86::VMOVSSrm_alt:
case X86::VMOVSDrm:
case X86::VMOVSDrm_alt:
case X86::VMOVAPSrm:
case X86::VMOVUPSrm:
case X86::VMOVAPDrm:
case X86::VMOVUPDrm:
case X86::VMOVDQArm:
case X86::VMOVDQUrm:
case X86::VMOVAPSYrm:
case X86::VMOVUPSYrm:
case X86::VMOVAPDYrm:
case X86::VMOVUPDYrm:
case X86::VMOVDQAYrm:
case X86::VMOVDQUYrm:
case X86::MMX_MOVD64rm:
case X86::MMX_MOVQ64rm:
case X86::VBROADCASTSSrm:
case X86::VBROADCASTSSYrm:
case X86::VBROADCASTSDYrm:
// AVX-512
case X86::VPBROADCASTBZ128rm:
case X86::VPBROADCASTBZ256rm:
case X86::VPBROADCASTBZrm:
case X86::VBROADCASTF32X2Z256rm:
case X86::VBROADCASTF32X2Zrm:
case X86::VBROADCASTI32X2Z128rm:
case X86::VBROADCASTI32X2Z256rm:
case X86::VBROADCASTI32X2Zrm:
case X86::VPBROADCASTWZ128rm:
case X86::VPBROADCASTWZ256rm:
case X86::VPBROADCASTWZrm:
case X86::VPBROADCASTDZ128rm:
case X86::VPBROADCASTDZ256rm:
case X86::VPBROADCASTDZrm:
case X86::VBROADCASTSSZ128rm:
case X86::VBROADCASTSSZ256rm:
case X86::VBROADCASTSSZrm:
case X86::VPBROADCASTQZ128rm:
case X86::VPBROADCASTQZ256rm:
case X86::VPBROADCASTQZrm:
case X86::VBROADCASTSDZ256rm:
case X86::VBROADCASTSDZrm:
case X86::VMOVSSZrm:
case X86::VMOVSSZrm_alt:
case X86::VMOVSDZrm:
case X86::VMOVSDZrm_alt:
case X86::VMOVSHZrm:
case X86::VMOVSHZrm_alt:
case X86::VMOVAPDZ128rm:
case X86::VMOVAPDZ256rm:
case X86::VMOVAPDZrm:
case X86::VMOVAPSZ128rm:
case X86::VMOVAPSZ256rm:
case X86::VMOVAPSZ128rm_NOVLX:
case X86::VMOVAPSZ256rm_NOVLX:
case X86::VMOVAPSZrm:
case X86::VMOVDQA32Z128rm:
case X86::VMOVDQA32Z256rm:
case X86::VMOVDQA32Zrm:
case X86::VMOVDQA64Z128rm:
case X86::VMOVDQA64Z256rm:
case X86::VMOVDQA64Zrm:
case X86::VMOVDQU16Z128rm:
case X86::VMOVDQU16Z256rm:
case X86::VMOVDQU16Zrm:
case X86::VMOVDQU32Z128rm:
case X86::VMOVDQU32Z256rm:
case X86::VMOVDQU32Zrm:
case X86::VMOVDQU64Z128rm:
case X86::VMOVDQU64Z256rm:
case X86::VMOVDQU64Zrm:
case X86::VMOVDQU8Z128rm:
case X86::VMOVDQU8Z256rm:
case X86::VMOVDQU8Zrm:
case X86::VMOVUPDZ128rm:
case X86::VMOVUPDZ256rm:
case X86::VMOVUPDZrm:
case X86::VMOVUPSZ128rm:
case X86::VMOVUPSZ256rm:
case X86::VMOVUPSZ128rm_NOVLX:
case X86::VMOVUPSZ256rm_NOVLX:
case X86::VMOVUPSZrm: {
// Loads from constant pools are trivially rematerializable.
if (MI.getOperand(1 + X86::AddrBaseReg).isReg() &&
MI.getOperand(1 + X86::AddrScaleAmt).isImm() &&
MI.getOperand(1 + X86::AddrIndexReg).isReg() &&
MI.getOperand(1 + X86::AddrIndexReg).getReg() == 0 &&
MI.isDereferenceableInvariantLoad()) {
Register BaseReg = MI.getOperand(1 + X86::AddrBaseReg).getReg();
if (BaseReg == 0 || BaseReg == X86::RIP)
return true;
// Allow re-materialization of PIC load.
if (!(!ReMatPICStubLoad && MI.getOperand(1 + X86::AddrDisp).isGlobal())) {
const MachineFunction &MF = *MI.getParent()->getParent();
const MachineRegisterInfo &MRI = MF.getRegInfo();
if (regIsPICBase(BaseReg, MRI))
return true;
}
}
break;
}
case X86::LEA32r:
case X86::LEA64r: {
if (MI.getOperand(1 + X86::AddrScaleAmt).isImm() &&
MI.getOperand(1 + X86::AddrIndexReg).isReg() &&
MI.getOperand(1 + X86::AddrIndexReg).getReg() == 0 &&
!MI.getOperand(1 + X86::AddrDisp).isReg()) {
// lea fi#, lea GV, etc. are all rematerializable.
if (!MI.getOperand(1 + X86::AddrBaseReg).isReg())
return true;
Register BaseReg = MI.getOperand(1 + X86::AddrBaseReg).getReg();
if (BaseReg == 0)
return true;
// Allow re-materialization of lea PICBase + x.
const MachineFunction &MF = *MI.getParent()->getParent();
const MachineRegisterInfo &MRI = MF.getRegInfo();
if (regIsPICBase(BaseReg, MRI))
return true;
}
break;
}
}
return TargetInstrInfo::isReallyTriviallyReMaterializable(MI);
}
void X86InstrInfo::reMaterialize(MachineBasicBlock &MBB,
MachineBasicBlock::iterator I,
Register DestReg, unsigned SubIdx,
const MachineInstr &Orig,
const TargetRegisterInfo &TRI) const {
bool ClobbersEFLAGS = Orig.modifiesRegister(X86::EFLAGS, &TRI);
if (ClobbersEFLAGS && MBB.computeRegisterLiveness(&TRI, X86::EFLAGS, I) !=
MachineBasicBlock::LQR_Dead) {
// The instruction clobbers EFLAGS. Re-materialize as MOV32ri to avoid side
// effects.
int Value;
switch (Orig.getOpcode()) {
case X86::MOV32r0:
Value = 0;
break;
case X86::MOV32r1:
Value = 1;
break;
case X86::MOV32r_1:
Value = -1;
break;
default:
llvm_unreachable("Unexpected instruction!");
}
const DebugLoc &DL = Orig.getDebugLoc();
BuildMI(MBB, I, DL, get(X86::MOV32ri))
.add(Orig.getOperand(0))
.addImm(Value);
} else {
MachineInstr *MI = MBB.getParent()->CloneMachineInstr(&Orig);
MBB.insert(I, MI);
}
MachineInstr &NewMI = *std::prev(I);
NewMI.substituteRegister(Orig.getOperand(0).getReg(), DestReg, SubIdx, TRI);
}
/// True if MI has a condition code def, e.g. EFLAGS, that is not marked dead.
bool X86InstrInfo::hasLiveCondCodeDef(MachineInstr &MI) const {
for (const MachineOperand &MO : MI.operands()) {
if (MO.isReg() && MO.isDef() && MO.getReg() == X86::EFLAGS &&
!MO.isDead()) {
return true;
}
}
return false;
}
/// Check whether the shift count for a machine operand is non-zero.
inline static unsigned getTruncatedShiftCount(const MachineInstr &MI,
unsigned ShiftAmtOperandIdx) {
// The shift count is six bits with the REX.W prefix and five bits without.
unsigned ShiftCountMask = (MI.getDesc().TSFlags & X86II::REX_W) ? 63 : 31;
unsigned Imm = MI.getOperand(ShiftAmtOperandIdx).getImm();
return Imm & ShiftCountMask;
}
/// Check whether the given shift count is appropriate
/// can be represented by a LEA instruction.
inline static bool isTruncatedShiftCountForLEA(unsigned ShAmt) {
// Left shift instructions can be transformed into load-effective-address
// instructions if we can encode them appropriately.
// A LEA instruction utilizes a SIB byte to encode its scale factor.
// The SIB.scale field is two bits wide which means that we can encode any
// shift amount less than 4.
return ShAmt < 4 && ShAmt > 0;
}
static bool findRedundantFlagInstr(MachineInstr &CmpInstr,
MachineInstr &CmpValDefInstr,
const MachineRegisterInfo *MRI,
MachineInstr **AndInstr,
const TargetRegisterInfo *TRI,
bool &NoSignFlag, bool &ClearsOverflowFlag) {
if (!(CmpValDefInstr.getOpcode() == X86::SUBREG_TO_REG &&
CmpInstr.getOpcode() == X86::TEST64rr) &&
!(CmpValDefInstr.getOpcode() == X86::COPY &&
CmpInstr.getOpcode() == X86::TEST16rr))
return false;
// CmpInstr is a TEST16rr/TEST64rr instruction, and
// `X86InstrInfo::analyzeCompare` guarantees that it's analyzable only if two
// registers are identical.
assert((CmpInstr.getOperand(0).getReg() == CmpInstr.getOperand(1).getReg()) &&
"CmpInstr is an analyzable TEST16rr/TEST64rr, and "
"`X86InstrInfo::analyzeCompare` requires two reg operands are the"
"same.");
// Caller (`X86InstrInfo::optimizeCompareInstr`) guarantees that
// `CmpValDefInstr` defines the value that's used by `CmpInstr`; in this case
// if `CmpValDefInstr` sets the EFLAGS, it is likely that `CmpInstr` is
// redundant.
assert(
(MRI->getVRegDef(CmpInstr.getOperand(0).getReg()) == &CmpValDefInstr) &&
"Caller guarantees that TEST64rr is a user of SUBREG_TO_REG or TEST16rr "
"is a user of COPY sub16bit.");
MachineInstr *VregDefInstr = nullptr;
if (CmpInstr.getOpcode() == X86::TEST16rr) {
if (!CmpValDefInstr.getOperand(1).getReg().isVirtual())
return false;
VregDefInstr = MRI->getVRegDef(CmpValDefInstr.getOperand(1).getReg());
if (!VregDefInstr)
return false;
// We can only remove test when AND32ri or AND64ri32 whose imm can fit 16bit
// size, others 32/64 bit ops would test higher bits which test16rr don't
// want to.
if (!((VregDefInstr->getOpcode() == X86::AND32ri ||
VregDefInstr->getOpcode() == X86::AND64ri32) &&
isUInt<16>(VregDefInstr->getOperand(2).getImm())))
return false;
}
if (CmpInstr.getOpcode() == X86::TEST64rr) {
// As seen in X86 td files, CmpValDefInstr.getOperand(1).getImm() is
// typically 0.
if (CmpValDefInstr.getOperand(1).getImm() != 0)
return false;
// As seen in X86 td files, CmpValDefInstr.getOperand(3) is typically
// sub_32bit or sub_xmm.
if (CmpValDefInstr.getOperand(3).getImm() != X86::sub_32bit)
return false;
VregDefInstr = MRI->getVRegDef(CmpValDefInstr.getOperand(2).getReg());
}
assert(VregDefInstr && "Must have a definition (SSA)");
// Requires `CmpValDefInstr` and `VregDefInstr` are from the same MBB
// to simplify the subsequent analysis.
//
// FIXME: If `VregDefInstr->getParent()` is the only predecessor of
// `CmpValDefInstr.getParent()`, this could be handled.
if (VregDefInstr->getParent() != CmpValDefInstr.getParent())
return false;
if (X86::isAND(VregDefInstr->getOpcode())) {
// Get a sequence of instructions like
// %reg = and* ... // Set EFLAGS
// ... // EFLAGS not changed
// %extended_reg = subreg_to_reg 0, %reg, %subreg.sub_32bit
// test64rr %extended_reg, %extended_reg, implicit-def $eflags
// or
// %reg = and32* ...
// ... // EFLAGS not changed.
// %src_reg = copy %reg.sub_16bit:gr32
// test16rr %src_reg, %src_reg, implicit-def $eflags
//
// If subsequent readers use a subset of bits that don't change
// after `and*` instructions, it's likely that the test64rr could
// be optimized away.
for (const MachineInstr &Instr :
make_range(std::next(MachineBasicBlock::iterator(VregDefInstr)),
MachineBasicBlock::iterator(CmpValDefInstr))) {
// There are instructions between 'VregDefInstr' and
// 'CmpValDefInstr' that modifies EFLAGS.
if (Instr.modifiesRegister(X86::EFLAGS, TRI))
return false;
}
*AndInstr = VregDefInstr;
// AND instruction will essentially update SF and clear OF, so
// NoSignFlag should be false in the sense that SF is modified by `AND`.
//
// However, the implementation artifically sets `NoSignFlag` to true
// to poison the SF bit; that is to say, if SF is looked at later, the
// optimization (to erase TEST64rr) will be disabled.
//
// The reason to poison SF bit is that SF bit value could be different
// in the `AND` and `TEST` operation; signed bit is not known for `AND`,
// and is known to be 0 as a result of `TEST64rr`.
//
// FIXME: As opposed to poisoning the SF bit directly, consider peeking into
// the AND instruction and using the static information to guide peephole
// optimization if possible. For example, it's possible to fold a
// conditional move into a copy if the relevant EFLAG bits could be deduced
// from an immediate operand of and operation.
//
NoSignFlag = true;
// ClearsOverflowFlag is true for AND operation (no surprise).
ClearsOverflowFlag = true;
return true;
}
return false;
}
bool X86InstrInfo::classifyLEAReg(MachineInstr &MI, const MachineOperand &Src,
unsigned Opc, bool AllowSP, Register &NewSrc,
bool &isKill, MachineOperand &ImplicitOp,
LiveVariables *LV, LiveIntervals *LIS) const {
MachineFunction &MF = *MI.getParent()->getParent();
const TargetRegisterClass *RC;
if (AllowSP) {
RC = Opc != X86::LEA32r ? &X86::GR64RegClass : &X86::GR32RegClass;
} else {
RC = Opc != X86::LEA32r ? &X86::GR64_NOSPRegClass : &X86::GR32_NOSPRegClass;
}
Register SrcReg = Src.getReg();
isKill = MI.killsRegister(SrcReg, /*TRI=*/nullptr);
// For both LEA64 and LEA32 the register already has essentially the right
// type (32-bit or 64-bit) we may just need to forbid SP.
if (Opc != X86::LEA64_32r) {
NewSrc = SrcReg;
assert(!Src.isUndef() && "Undef op doesn't need optimization");
if (NewSrc.isVirtual() && !MF.getRegInfo().constrainRegClass(NewSrc, RC))
return false;
return true;
}
// This is for an LEA64_32r and incoming registers are 32-bit. One way or
// another we need to add 64-bit registers to the final MI.
if (SrcReg.isPhysical()) {
ImplicitOp = Src;
ImplicitOp.setImplicit();
NewSrc = getX86SubSuperRegister(SrcReg, 64);
assert(NewSrc.isValid() && "Invalid Operand");
assert(!Src.isUndef() && "Undef op doesn't need optimization");
} else {
// Virtual register of the wrong class, we have to create a temporary 64-bit
// vreg to feed into the LEA.
NewSrc = MF.getRegInfo().createVirtualRegister(RC);
MachineInstr *Copy =
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(TargetOpcode::COPY))
.addReg(NewSrc, RegState::Define | RegState::Undef, X86::sub_32bit)
.addReg(SrcReg, getKillRegState(isKill));
// Which is obviously going to be dead after we're done with it.
isKill = true;
if (LV)
LV->replaceKillInstruction(SrcReg, MI, *Copy);
if (LIS) {
SlotIndex CopyIdx = LIS->InsertMachineInstrInMaps(*Copy);
SlotIndex Idx = LIS->getInstructionIndex(MI);
LiveInterval &LI = LIS->getInterval(SrcReg);
LiveRange::Segment *S = LI.getSegmentContaining(Idx);
if (S->end.getBaseIndex() == Idx)
S->end = CopyIdx.getRegSlot();
}
}
// We've set all the parameters without issue.
return true;
}
MachineInstr *X86InstrInfo::convertToThreeAddressWithLEA(unsigned MIOpc,
MachineInstr &MI,
LiveVariables *LV,
LiveIntervals *LIS,
bool Is8BitOp) const {
// We handle 8-bit adds and various 16-bit opcodes in the switch below.
MachineBasicBlock &MBB = *MI.getParent();
MachineRegisterInfo &RegInfo = MBB.getParent()->getRegInfo();
assert((Is8BitOp ||
RegInfo.getTargetRegisterInfo()->getRegSizeInBits(
*RegInfo.getRegClass(MI.getOperand(0).getReg())) == 16) &&
"Unexpected type for LEA transform");
// TODO: For a 32-bit target, we need to adjust the LEA variables with
// something like this:
// Opcode = X86::LEA32r;
// InRegLEA = RegInfo.createVirtualRegister(&X86::GR32_NOSPRegClass);
// OutRegLEA =
// Is8BitOp ? RegInfo.createVirtualRegister(&X86::GR32ABCD_RegClass)
// : RegInfo.createVirtualRegister(&X86::GR32RegClass);
if (!Subtarget.is64Bit())
return nullptr;
unsigned Opcode = X86::LEA64_32r;
Register InRegLEA = RegInfo.createVirtualRegister(&X86::GR64_NOSPRegClass);
Register OutRegLEA = RegInfo.createVirtualRegister(&X86::GR32RegClass);
Register InRegLEA2;
// Build and insert into an implicit UNDEF value. This is OK because
// we will be shifting and then extracting the lower 8/16-bits.
// This has the potential to cause partial register stall. e.g.
// movw (%rbp,%rcx,2), %dx
// leal -65(%rdx), %esi
// But testing has shown this *does* help performance in 64-bit mode (at
// least on modern x86 machines).
MachineBasicBlock::iterator MBBI = MI.getIterator();
Register Dest = MI.getOperand(0).getReg();
Register Src = MI.getOperand(1).getReg();
Register Src2;
bool IsDead = MI.getOperand(0).isDead();
bool IsKill = MI.getOperand(1).isKill();
unsigned SubReg = Is8BitOp ? X86::sub_8bit : X86::sub_16bit;
assert(!MI.getOperand(1).isUndef() && "Undef op doesn't need optimization");
MachineInstr *ImpDef =
BuildMI(MBB, MBBI, MI.getDebugLoc(), get(X86::IMPLICIT_DEF), InRegLEA);
MachineInstr *InsMI =
BuildMI(MBB, MBBI, MI.getDebugLoc(), get(TargetOpcode::COPY))
.addReg(InRegLEA, RegState::Define, SubReg)
.addReg(Src, getKillRegState(IsKill));
MachineInstr *ImpDef2 = nullptr;
MachineInstr *InsMI2 = nullptr;
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI, MI.getDebugLoc(), get(Opcode), OutRegLEA);
switch (MIOpc) {
default:
llvm_unreachable("Unreachable!");
case X86::SHL8ri:
case X86::SHL16ri: {
unsigned ShAmt = MI.getOperand(2).getImm();
MIB.addReg(0)
.addImm(1LL << ShAmt)
.addReg(InRegLEA, RegState::Kill)
.addImm(0)
.addReg(0);
break;
}
case X86::INC8r:
case X86::INC16r:
addRegOffset(MIB, InRegLEA, true, 1);
break;
case X86::DEC8r:
case X86::DEC16r:
addRegOffset(MIB, InRegLEA, true, -1);
break;
case X86::ADD8ri:
case X86::ADD8ri_DB:
case X86::ADD16ri:
case X86::ADD16ri_DB:
addRegOffset(MIB, InRegLEA, true, MI.getOperand(2).getImm());
break;
case X86::ADD8rr:
case X86::ADD8rr_DB:
case X86::ADD16rr:
case X86::ADD16rr_DB: {
Src2 = MI.getOperand(2).getReg();
bool IsKill2 = MI.getOperand(2).isKill();
assert(!MI.getOperand(2).isUndef() && "Undef op doesn't need optimization");
if (Src == Src2) {
// ADD8rr/ADD16rr killed %reg1028, %reg1028
// just a single insert_subreg.
addRegReg(MIB, InRegLEA, true, InRegLEA, false);
} else {
if (Subtarget.is64Bit())
InRegLEA2 = RegInfo.createVirtualRegister(&X86::GR64_NOSPRegClass);
else
InRegLEA2 = RegInfo.createVirtualRegister(&X86::GR32_NOSPRegClass);
// Build and insert into an implicit UNDEF value. This is OK because
// we will be shifting and then extracting the lower 8/16-bits.
ImpDef2 = BuildMI(MBB, &*MIB, MI.getDebugLoc(), get(X86::IMPLICIT_DEF),
InRegLEA2);
InsMI2 = BuildMI(MBB, &*MIB, MI.getDebugLoc(), get(TargetOpcode::COPY))
.addReg(InRegLEA2, RegState::Define, SubReg)
.addReg(Src2, getKillRegState(IsKill2));
addRegReg(MIB, InRegLEA, true, InRegLEA2, true);
}
if (LV && IsKill2 && InsMI2)
LV->replaceKillInstruction(Src2, MI, *InsMI2);
break;
}
}
MachineInstr *NewMI = MIB;
MachineInstr *ExtMI =
BuildMI(MBB, MBBI, MI.getDebugLoc(), get(TargetOpcode::COPY))
.addReg(Dest, RegState::Define | getDeadRegState(IsDead))
.addReg(OutRegLEA, RegState::Kill, SubReg);
if (LV) {
// Update live variables.
LV->getVarInfo(InRegLEA).Kills.push_back(NewMI);
if (InRegLEA2)
LV->getVarInfo(InRegLEA2).Kills.push_back(NewMI);
LV->getVarInfo(OutRegLEA).Kills.push_back(ExtMI);
if (IsKill)
LV->replaceKillInstruction(Src, MI, *InsMI);
if (IsDead)
LV->replaceKillInstruction(Dest, MI, *ExtMI);
}
if (LIS) {
LIS->InsertMachineInstrInMaps(*ImpDef);
SlotIndex InsIdx = LIS->InsertMachineInstrInMaps(*InsMI);
if (ImpDef2)
LIS->InsertMachineInstrInMaps(*ImpDef2);
SlotIndex Ins2Idx;
if (InsMI2)
Ins2Idx = LIS->InsertMachineInstrInMaps(*InsMI2);
SlotIndex NewIdx = LIS->ReplaceMachineInstrInMaps(MI, *NewMI);
SlotIndex ExtIdx = LIS->InsertMachineInstrInMaps(*ExtMI);
LIS->getInterval(InRegLEA);
LIS->getInterval(OutRegLEA);
if (InRegLEA2)
LIS->getInterval(InRegLEA2);
// Move the use of Src up to InsMI.
LiveInterval &SrcLI = LIS->getInterval(Src);
LiveRange::Segment *SrcSeg = SrcLI.getSegmentContaining(NewIdx);
if (SrcSeg->end == NewIdx.getRegSlot())
SrcSeg->end = InsIdx.getRegSlot();
if (InsMI2) {
// Move the use of Src2 up to InsMI2.
LiveInterval &Src2LI = LIS->getInterval(Src2);
LiveRange::Segment *Src2Seg = Src2LI.getSegmentContaining(NewIdx);
if (Src2Seg->end == NewIdx.getRegSlot())
Src2Seg->end = Ins2Idx.getRegSlot();
}
// Move the definition of Dest down to ExtMI.
LiveInterval &DestLI = LIS->getInterval(Dest);
LiveRange::Segment *DestSeg =
DestLI.getSegmentContaining(NewIdx.getRegSlot());
assert(DestSeg->start == NewIdx.getRegSlot() &&
DestSeg->valno->def == NewIdx.getRegSlot());
DestSeg->start = ExtIdx.getRegSlot();
DestSeg->valno->def = ExtIdx.getRegSlot();
}
return ExtMI;
}
/// This method must be implemented by targets that
/// set the M_CONVERTIBLE_TO_3_ADDR flag. When this flag is set, the target
/// may be able to convert a two-address instruction into a true
/// three-address instruction on demand. This allows the X86 target (for
/// example) to convert ADD and SHL instructions into LEA instructions if they
/// would require register copies due to two-addressness.
///
/// This method returns a null pointer if the transformation cannot be
/// performed, otherwise it returns the new instruction.
///
MachineInstr *X86InstrInfo::convertToThreeAddress(MachineInstr &MI,
LiveVariables *LV,
LiveIntervals *LIS) const {
// The following opcodes also sets the condition code register(s). Only
// convert them to equivalent lea if the condition code register def's
// are dead!
if (hasLiveCondCodeDef(MI))
return nullptr;
MachineFunction &MF = *MI.getParent()->getParent();
// All instructions input are two-addr instructions. Get the known operands.
const MachineOperand &Dest = MI.getOperand(0);
const MachineOperand &Src = MI.getOperand(1);
// Ideally, operations with undef should be folded before we get here, but we
// can't guarantee it. Bail out because optimizing undefs is a waste of time.
// Without this, we have to forward undef state to new register operands to
// avoid machine verifier errors.
if (Src.isUndef())
return nullptr;
if (MI.getNumOperands() > 2)
if (MI.getOperand(2).isReg() && MI.getOperand(2).isUndef())
return nullptr;
MachineInstr *NewMI = nullptr;
Register SrcReg, SrcReg2;
bool Is64Bit = Subtarget.is64Bit();
bool Is8BitOp = false;
unsigned NumRegOperands = 2;
unsigned MIOpc = MI.getOpcode();
switch (MIOpc) {
default:
llvm_unreachable("Unreachable!");
case X86::SHL64ri: {
assert(MI.getNumOperands() >= 3 && "Unknown shift instruction!");
unsigned ShAmt = getTruncatedShiftCount(MI, 2);
if (!isTruncatedShiftCountForLEA(ShAmt))
return nullptr;
// LEA can't handle RSP.
if (Src.getReg().isVirtual() && !MF.getRegInfo().constrainRegClass(
Src.getReg(), &X86::GR64_NOSPRegClass))
return nullptr;
NewMI = BuildMI(MF, MI.getDebugLoc(), get(X86::LEA64r))
.add(Dest)
.addReg(0)
.addImm(1LL << ShAmt)
.add(Src)
.addImm(0)
.addReg(0);
break;
}
case X86::SHL32ri: {
assert(MI.getNumOperands() >= 3 && "Unknown shift instruction!");
unsigned ShAmt = getTruncatedShiftCount(MI, 2);
if (!isTruncatedShiftCountForLEA(ShAmt))
return nullptr;
unsigned Opc = Is64Bit ? X86::LEA64_32r : X86::LEA32r;
// LEA can't handle ESP.
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/false, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.addReg(0)
.addImm(1LL << ShAmt)
.addReg(SrcReg, getKillRegState(isKill))
.addImm(0)
.addReg(0);
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
NewMI = MIB;
// Add kills if classifyLEAReg created a new register.
if (LV && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
break;
}
case X86::SHL8ri:
Is8BitOp = true;
[[fallthrough]];
case X86::SHL16ri: {
assert(MI.getNumOperands() >= 3 && "Unknown shift instruction!");
unsigned ShAmt = getTruncatedShiftCount(MI, 2);
if (!isTruncatedShiftCountForLEA(ShAmt))
return nullptr;
return convertToThreeAddressWithLEA(MIOpc, MI, LV, LIS, Is8BitOp);
}
case X86::INC64r:
case X86::INC32r: {
assert(MI.getNumOperands() >= 2 && "Unknown inc instruction!");
unsigned Opc = MIOpc == X86::INC64r
? X86::LEA64r
: (Is64Bit ? X86::LEA64_32r : X86::LEA32r);
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/false, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.addReg(SrcReg, getKillRegState(isKill));
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
NewMI = addOffset(MIB, 1);
// Add kills if classifyLEAReg created a new register.
if (LV && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
break;
}
case X86::DEC64r:
case X86::DEC32r: {
assert(MI.getNumOperands() >= 2 && "Unknown dec instruction!");
unsigned Opc = MIOpc == X86::DEC64r
? X86::LEA64r
: (Is64Bit ? X86::LEA64_32r : X86::LEA32r);
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/false, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.addReg(SrcReg, getKillRegState(isKill));
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
NewMI = addOffset(MIB, -1);
// Add kills if classifyLEAReg created a new register.
if (LV && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
break;
}
case X86::DEC8r:
case X86::INC8r:
Is8BitOp = true;
[[fallthrough]];
case X86::DEC16r:
case X86::INC16r:
return convertToThreeAddressWithLEA(MIOpc, MI, LV, LIS, Is8BitOp);
case X86::ADD64rr:
case X86::ADD64rr_DB:
case X86::ADD32rr:
case X86::ADD32rr_DB: {
assert(MI.getNumOperands() >= 3 && "Unknown add instruction!");
unsigned Opc;
if (MIOpc == X86::ADD64rr || MIOpc == X86::ADD64rr_DB)
Opc = X86::LEA64r;
else
Opc = Is64Bit ? X86::LEA64_32r : X86::LEA32r;
const MachineOperand &Src2 = MI.getOperand(2);
bool isKill2;
MachineOperand ImplicitOp2 = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src2, Opc, /*AllowSP=*/false, SrcReg2, isKill2,
ImplicitOp2, LV, LIS))
return nullptr;
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (Src.getReg() == Src2.getReg()) {
// Don't call classify LEAReg a second time on the same register, in case
// the first call inserted a COPY from Src2 and marked it as killed.
isKill = isKill2;
SrcReg = SrcReg2;
} else {
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/true, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
}
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc)).add(Dest);
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
if (ImplicitOp2.getReg() != 0)
MIB.add(ImplicitOp2);
NewMI = addRegReg(MIB, SrcReg, isKill, SrcReg2, isKill2);
// Add kills if classifyLEAReg created a new register.
if (LV) {
if (SrcReg2 != Src2.getReg())
LV->getVarInfo(SrcReg2).Kills.push_back(NewMI);
if (SrcReg != SrcReg2 && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
}
NumRegOperands = 3;
break;
}
case X86::ADD8rr:
case X86::ADD8rr_DB:
Is8BitOp = true;
[[fallthrough]];
case X86::ADD16rr:
case X86::ADD16rr_DB:
return convertToThreeAddressWithLEA(MIOpc, MI, LV, LIS, Is8BitOp);
case X86::ADD64ri32:
case X86::ADD64ri32_DB:
assert(MI.getNumOperands() >= 3 && "Unknown add instruction!");
NewMI = addOffset(
BuildMI(MF, MI.getDebugLoc(), get(X86::LEA64r)).add(Dest).add(Src),
MI.getOperand(2));
break;
case X86::ADD32ri:
case X86::ADD32ri_DB: {
assert(MI.getNumOperands() >= 3 && "Unknown add instruction!");
unsigned Opc = Is64Bit ? X86::LEA64_32r : X86::LEA32r;
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/true, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.addReg(SrcReg, getKillRegState(isKill));
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
NewMI = addOffset(MIB, MI.getOperand(2));
// Add kills if classifyLEAReg created a new register.
if (LV && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
break;
}
case X86::ADD8ri:
case X86::ADD8ri_DB:
Is8BitOp = true;
[[fallthrough]];
case X86::ADD16ri:
case X86::ADD16ri_DB:
return convertToThreeAddressWithLEA(MIOpc, MI, LV, LIS, Is8BitOp);
case X86::SUB8ri:
case X86::SUB16ri:
/// FIXME: Support these similar to ADD8ri/ADD16ri*.
return nullptr;
case X86::SUB32ri: {
if (!MI.getOperand(2).isImm())
return nullptr;
int64_t Imm = MI.getOperand(2).getImm();
if (!isInt<32>(-Imm))
return nullptr;
assert(MI.getNumOperands() >= 3 && "Unknown add instruction!");
unsigned Opc = Is64Bit ? X86::LEA64_32r : X86::LEA32r;
bool isKill;
MachineOperand ImplicitOp = MachineOperand::CreateReg(0, false);
if (!classifyLEAReg(MI, Src, Opc, /*AllowSP=*/true, SrcReg, isKill,
ImplicitOp, LV, LIS))
return nullptr;
MachineInstrBuilder MIB = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.addReg(SrcReg, getKillRegState(isKill));
if (ImplicitOp.getReg() != 0)
MIB.add(ImplicitOp);
NewMI = addOffset(MIB, -Imm);
// Add kills if classifyLEAReg created a new register.
if (LV && SrcReg != Src.getReg())
LV->getVarInfo(SrcReg).Kills.push_back(NewMI);
break;
}
case X86::SUB64ri32: {
if (!MI.getOperand(2).isImm())
return nullptr;
int64_t Imm = MI.getOperand(2).getImm();
if (!isInt<32>(-Imm))
return nullptr;
assert(MI.getNumOperands() >= 3 && "Unknown sub instruction!");
MachineInstrBuilder MIB =
BuildMI(MF, MI.getDebugLoc(), get(X86::LEA64r)).add(Dest).add(Src);
NewMI = addOffset(MIB, -Imm);
break;
}
case X86::VMOVDQU8Z128rmk:
case X86::VMOVDQU8Z256rmk:
case X86::VMOVDQU8Zrmk:
case X86::VMOVDQU16Z128rmk:
case X86::VMOVDQU16Z256rmk:
case X86::VMOVDQU16Zrmk:
case X86::VMOVDQU32Z128rmk:
case X86::VMOVDQA32Z128rmk:
case X86::VMOVDQU32Z256rmk:
case X86::VMOVDQA32Z256rmk:
case X86::VMOVDQU32Zrmk:
case X86::VMOVDQA32Zrmk:
case X86::VMOVDQU64Z128rmk:
case X86::VMOVDQA64Z128rmk:
case X86::VMOVDQU64Z256rmk:
case X86::VMOVDQA64Z256rmk:
case X86::VMOVDQU64Zrmk:
case X86::VMOVDQA64Zrmk:
case X86::VMOVUPDZ128rmk:
case X86::VMOVAPDZ128rmk:
case X86::VMOVUPDZ256rmk:
case X86::VMOVAPDZ256rmk:
case X86::VMOVUPDZrmk:
case X86::VMOVAPDZrmk:
case X86::VMOVUPSZ128rmk:
case X86::VMOVAPSZ128rmk:
case X86::VMOVUPSZ256rmk:
case X86::VMOVAPSZ256rmk:
case X86::VMOVUPSZrmk:
case X86::VMOVAPSZrmk:
case X86::VBROADCASTSDZ256rmk:
case X86::VBROADCASTSDZrmk:
case X86::VBROADCASTSSZ128rmk:
case X86::VBROADCASTSSZ256rmk:
case X86::VBROADCASTSSZrmk:
case X86::VPBROADCASTDZ128rmk:
case X86::VPBROADCASTDZ256rmk:
case X86::VPBROADCASTDZrmk:
case X86::VPBROADCASTQZ128rmk:
case X86::VPBROADCASTQZ256rmk:
case X86::VPBROADCASTQZrmk: {
unsigned Opc;
switch (MIOpc) {
default:
llvm_unreachable("Unreachable!");
case X86::VMOVDQU8Z128rmk:
Opc = X86::VPBLENDMBZ128rmk;
break;
case X86::VMOVDQU8Z256rmk:
Opc = X86::VPBLENDMBZ256rmk;
break;
case X86::VMOVDQU8Zrmk:
Opc = X86::VPBLENDMBZrmk;
break;
case X86::VMOVDQU16Z128rmk:
Opc = X86::VPBLENDMWZ128rmk;
break;
case X86::VMOVDQU16Z256rmk:
Opc = X86::VPBLENDMWZ256rmk;
break;
case X86::VMOVDQU16Zrmk:
Opc = X86::VPBLENDMWZrmk;
break;
case X86::VMOVDQU32Z128rmk:
Opc = X86::VPBLENDMDZ128rmk;
break;
case X86::VMOVDQU32Z256rmk:
Opc = X86::VPBLENDMDZ256rmk;
break;
case X86::VMOVDQU32Zrmk:
Opc = X86::VPBLENDMDZrmk;
break;
case X86::VMOVDQU64Z128rmk:
Opc = X86::VPBLENDMQZ128rmk;
break;
case X86::VMOVDQU64Z256rmk:
Opc = X86::VPBLENDMQZ256rmk;
break;
case X86::VMOVDQU64Zrmk:
Opc = X86::VPBLENDMQZrmk;
break;
case X86::VMOVUPDZ128rmk:
Opc = X86::VBLENDMPDZ128rmk;
break;
case X86::VMOVUPDZ256rmk:
Opc = X86::VBLENDMPDZ256rmk;
break;
case X86::VMOVUPDZrmk:
Opc = X86::VBLENDMPDZrmk;
break;
case X86::VMOVUPSZ128rmk:
Opc = X86::VBLENDMPSZ128rmk;
break;
case X86::VMOVUPSZ256rmk:
Opc = X86::VBLENDMPSZ256rmk;
break;
case X86::VMOVUPSZrmk:
Opc = X86::VBLENDMPSZrmk;
break;
case X86::VMOVDQA32Z128rmk:
Opc = X86::VPBLENDMDZ128rmk;
break;
case X86::VMOVDQA32Z256rmk:
Opc = X86::VPBLENDMDZ256rmk;
break;
case X86::VMOVDQA32Zrmk:
Opc = X86::VPBLENDMDZrmk;
break;
case X86::VMOVDQA64Z128rmk:
Opc = X86::VPBLENDMQZ128rmk;
break;
case X86::VMOVDQA64Z256rmk:
Opc = X86::VPBLENDMQZ256rmk;
break;
case X86::VMOVDQA64Zrmk:
Opc = X86::VPBLENDMQZrmk;
break;
case X86::VMOVAPDZ128rmk:
Opc = X86::VBLENDMPDZ128rmk;
break;
case X86::VMOVAPDZ256rmk:
Opc = X86::VBLENDMPDZ256rmk;
break;
case X86::VMOVAPDZrmk:
Opc = X86::VBLENDMPDZrmk;
break;
case X86::VMOVAPSZ128rmk:
Opc = X86::VBLENDMPSZ128rmk;
break;
case X86::VMOVAPSZ256rmk:
Opc = X86::VBLENDMPSZ256rmk;
break;
case X86::VMOVAPSZrmk:
Opc = X86::VBLENDMPSZrmk;
break;
case X86::VBROADCASTSDZ256rmk:
Opc = X86::VBLENDMPDZ256rmbk;
break;
case X86::VBROADCASTSDZrmk:
Opc = X86::VBLENDMPDZrmbk;
break;
case X86::VBROADCASTSSZ128rmk:
Opc = X86::VBLENDMPSZ128rmbk;
break;
case X86::VBROADCASTSSZ256rmk:
Opc = X86::VBLENDMPSZ256rmbk;
break;
case X86::VBROADCASTSSZrmk:
Opc = X86::VBLENDMPSZrmbk;
break;
case X86::VPBROADCASTDZ128rmk:
Opc = X86::VPBLENDMDZ128rmbk;
break;
case X86::VPBROADCASTDZ256rmk:
Opc = X86::VPBLENDMDZ256rmbk;
break;
case X86::VPBROADCASTDZrmk:
Opc = X86::VPBLENDMDZrmbk;
break;
case X86::VPBROADCASTQZ128rmk:
Opc = X86::VPBLENDMQZ128rmbk;
break;
case X86::VPBROADCASTQZ256rmk:
Opc = X86::VPBLENDMQZ256rmbk;
break;
case X86::VPBROADCASTQZrmk:
Opc = X86::VPBLENDMQZrmbk;
break;
}
NewMI = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.add(MI.getOperand(2))
.add(Src)
.add(MI.getOperand(3))
.add(MI.getOperand(4))
.add(MI.getOperand(5))
.add(MI.getOperand(6))
.add(MI.getOperand(7));
NumRegOperands = 4;
break;
}
case X86::VMOVDQU8Z128rrk:
case X86::VMOVDQU8Z256rrk:
case X86::VMOVDQU8Zrrk:
case X86::VMOVDQU16Z128rrk:
case X86::VMOVDQU16Z256rrk:
case X86::VMOVDQU16Zrrk:
case X86::VMOVDQU32Z128rrk:
case X86::VMOVDQA32Z128rrk:
case X86::VMOVDQU32Z256rrk:
case X86::VMOVDQA32Z256rrk:
case X86::VMOVDQU32Zrrk:
case X86::VMOVDQA32Zrrk:
case X86::VMOVDQU64Z128rrk:
case X86::VMOVDQA64Z128rrk:
case X86::VMOVDQU64Z256rrk:
case X86::VMOVDQA64Z256rrk:
case X86::VMOVDQU64Zrrk:
case X86::VMOVDQA64Zrrk:
case X86::VMOVUPDZ128rrk:
case X86::VMOVAPDZ128rrk:
case X86::VMOVUPDZ256rrk:
case X86::VMOVAPDZ256rrk:
case X86::VMOVUPDZrrk:
case X86::VMOVAPDZrrk:
case X86::VMOVUPSZ128rrk:
case X86::VMOVAPSZ128rrk:
case X86::VMOVUPSZ256rrk:
case X86::VMOVAPSZ256rrk:
case X86::VMOVUPSZrrk:
case X86::VMOVAPSZrrk: {
unsigned Opc;
switch (MIOpc) {
default:
llvm_unreachable("Unreachable!");
case X86::VMOVDQU8Z128rrk:
Opc = X86::VPBLENDMBZ128rrk;
break;
case X86::VMOVDQU8Z256rrk:
Opc = X86::VPBLENDMBZ256rrk;
break;
case X86::VMOVDQU8Zrrk:
Opc = X86::VPBLENDMBZrrk;
break;
case X86::VMOVDQU16Z128rrk:
Opc = X86::VPBLENDMWZ128rrk;
break;
case X86::VMOVDQU16Z256rrk:
Opc = X86::VPBLENDMWZ256rrk;
break;
case X86::VMOVDQU16Zrrk:
Opc = X86::VPBLENDMWZrrk;
break;
case X86::VMOVDQU32Z128rrk:
Opc = X86::VPBLENDMDZ128rrk;
break;
case X86::VMOVDQU32Z256rrk:
Opc = X86::VPBLENDMDZ256rrk;
break;
case X86::VMOVDQU32Zrrk:
Opc = X86::VPBLENDMDZrrk;
break;
case X86::VMOVDQU64Z128rrk:
Opc = X86::VPBLENDMQZ128rrk;
break;
case X86::VMOVDQU64Z256rrk:
Opc = X86::VPBLENDMQZ256rrk;
break;
case X86::VMOVDQU64Zrrk:
Opc = X86::VPBLENDMQZrrk;
break;
case X86::VMOVUPDZ128rrk:
Opc = X86::VBLENDMPDZ128rrk;
break;
case X86::VMOVUPDZ256rrk:
Opc = X86::VBLENDMPDZ256rrk;
break;
case X86::VMOVUPDZrrk:
Opc = X86::VBLENDMPDZrrk;
break;
case X86::VMOVUPSZ128rrk:
Opc = X86::VBLENDMPSZ128rrk;
break;
case X86::VMOVUPSZ256rrk:
Opc = X86::VBLENDMPSZ256rrk;
break;
case X86::VMOVUPSZrrk:
Opc = X86::VBLENDMPSZrrk;
break;
case X86::VMOVDQA32Z128rrk:
Opc = X86::VPBLENDMDZ128rrk;
break;
case X86::VMOVDQA32Z256rrk:
Opc = X86::VPBLENDMDZ256rrk;
break;
case X86::VMOVDQA32Zrrk:
Opc = X86::VPBLENDMDZrrk;
break;
case X86::VMOVDQA64Z128rrk:
Opc = X86::VPBLENDMQZ128rrk;
break;
case X86::VMOVDQA64Z256rrk:
Opc = X86::VPBLENDMQZ256rrk;
break;
case X86::VMOVDQA64Zrrk:
Opc = X86::VPBLENDMQZrrk;
break;
case X86::VMOVAPDZ128rrk:
Opc = X86::VBLENDMPDZ128rrk;
break;
case X86::VMOVAPDZ256rrk:
Opc = X86::VBLENDMPDZ256rrk;
break;
case X86::VMOVAPDZrrk:
Opc = X86::VBLENDMPDZrrk;
break;
case X86::VMOVAPSZ128rrk:
Opc = X86::VBLENDMPSZ128rrk;
break;
case X86::VMOVAPSZ256rrk:
Opc = X86::VBLENDMPSZ256rrk;
break;
case X86::VMOVAPSZrrk:
Opc = X86::VBLENDMPSZrrk;
break;
}
NewMI = BuildMI(MF, MI.getDebugLoc(), get(Opc))
.add(Dest)
.add(MI.getOperand(2))
.add(Src)
.add(MI.getOperand(3));
NumRegOperands = 4;
break;
}
}
if (!NewMI)
return nullptr;
if (LV) { // Update live variables
for (unsigned I = 0; I < NumRegOperands; ++I) {
MachineOperand &Op = MI.getOperand(I);
if (Op.isReg() && (Op.isDead() || Op.isKill()))
LV->replaceKillInstruction(Op.getReg(), MI, *NewMI);
}
}
MachineBasicBlock &MBB = *MI.getParent();
MBB.insert(MI.getIterator(), NewMI); // Insert the new inst
if (LIS) {
LIS->ReplaceMachineInstrInMaps(MI, *NewMI);
if (SrcReg)
LIS->getInterval(SrcReg);
if (SrcReg2)
LIS->getInterval(SrcReg2);
}
return NewMI;
}
/// This determines which of three possible cases of a three source commute
/// the source indexes correspond to taking into account any mask operands.
/// All prevents commuting a passthru operand. Returns -1 if the commute isn't
/// possible.
/// Case 0 - Possible to commute the first and second operands.
/// Case 1 - Possible to commute the first and third operands.
/// Case 2 - Possible to commute the second and third operands.
static unsigned getThreeSrcCommuteCase(uint64_t TSFlags, unsigned SrcOpIdx1,
unsigned SrcOpIdx2) {
// Put the lowest index to SrcOpIdx1 to simplify the checks below.
if (SrcOpIdx1 > SrcOpIdx2)
std::swap(SrcOpIdx1, SrcOpIdx2);
unsigned Op1 = 1, Op2 = 2, Op3 = 3;
if (X86II::isKMasked(TSFlags)) {
Op2++;
Op3++;
}
if (SrcOpIdx1 == Op1 && SrcOpIdx2 == Op2)
return 0;
if (SrcOpIdx1 == Op1 && SrcOpIdx2 == Op3)
return 1;
if (SrcOpIdx1 == Op2 && SrcOpIdx2 == Op3)
return 2;
llvm_unreachable("Unknown three src commute case.");
}
unsigned X86InstrInfo::getFMA3OpcodeToCommuteOperands(
const MachineInstr &MI, unsigned SrcOpIdx1, unsigned SrcOpIdx2,
const X86InstrFMA3Group &FMA3Group) const {
unsigned Opc = MI.getOpcode();
// TODO: Commuting the 1st operand of FMA*_Int requires some additional
// analysis. The commute optimization is legal only if all users of FMA*_Int
// use only the lowest element of the FMA*_Int instruction. Such analysis are
// not implemented yet. So, just return 0 in that case.
// When such analysis are available this place will be the right place for
// calling it.
assert(!(FMA3Group.isIntrinsic() && (SrcOpIdx1 == 1 || SrcOpIdx2 == 1)) &&
"Intrinsic instructions can't commute operand 1");
// Determine which case this commute is or if it can't be done.
unsigned Case =
getThreeSrcCommuteCase(MI.getDesc().TSFlags, SrcOpIdx1, SrcOpIdx2);
assert(Case < 3 && "Unexpected case number!");
// Define the FMA forms mapping array that helps to map input FMA form
// to output FMA form to preserve the operation semantics after
// commuting the operands.
const unsigned Form132Index = 0;
const unsigned Form213Index = 1;
const unsigned Form231Index = 2;
static const unsigned FormMapping[][3] = {
// 0: SrcOpIdx1 == 1 && SrcOpIdx2 == 2;
// FMA132 A, C, b; ==> FMA231 C, A, b;
// FMA213 B, A, c; ==> FMA213 A, B, c;
// FMA231 C, A, b; ==> FMA132 A, C, b;
{Form231Index, Form213Index, Form132Index},
// 1: SrcOpIdx1 == 1 && SrcOpIdx2 == 3;
// FMA132 A, c, B; ==> FMA132 B, c, A;
// FMA213 B, a, C; ==> FMA231 C, a, B;
// FMA231 C, a, B; ==> FMA213 B, a, C;
{Form132Index, Form231Index, Form213Index},
// 2: SrcOpIdx1 == 2 && SrcOpIdx2 == 3;
// FMA132 a, C, B; ==> FMA213 a, B, C;
// FMA213 b, A, C; ==> FMA132 b, C, A;
// FMA231 c, A, B; ==> FMA231 c, B, A;
{Form213Index, Form132Index, Form231Index}};
unsigned FMAForms[3];
FMAForms[0] = FMA3Group.get132Opcode();
FMAForms[1] = FMA3Group.get213Opcode();
FMAForms[2] = FMA3Group.get231Opcode();
// Everything is ready, just adjust the FMA opcode and return it.
for (unsigned FormIndex = 0; FormIndex < 3; FormIndex++)
if (Opc == FMAForms[FormIndex])
return FMAForms[FormMapping[Case][FormIndex]];
llvm_unreachable("Illegal FMA3 format");
}
static void commuteVPTERNLOG(MachineInstr &MI, unsigned SrcOpIdx1,
unsigned SrcOpIdx2) {
// Determine which case this commute is or if it can't be done.
unsigned Case =
getThreeSrcCommuteCase(MI.getDesc().TSFlags, SrcOpIdx1, SrcOpIdx2);
assert(Case < 3 && "Unexpected case value!");
// For each case we need to swap two pairs of bits in the final immediate.
static const uint8_t SwapMasks[3][4] = {
{0x04, 0x10, 0x08, 0x20}, // Swap bits 2/4 and 3/5.
{0x02, 0x10, 0x08, 0x40}, // Swap bits 1/4 and 3/6.
{0x02, 0x04, 0x20, 0x40}, // Swap bits 1/2 and 5/6.
};
uint8_t Imm = MI.getOperand(MI.getNumOperands() - 1).getImm();
// Clear out the bits we are swapping.
uint8_t NewImm = Imm & ~(SwapMasks[Case][0] | SwapMasks[Case][1] |
SwapMasks[Case][2] | SwapMasks[Case][3]);
// If the immediate had a bit of the pair set, then set the opposite bit.
if (Imm & SwapMasks[Case][0])
NewImm |= SwapMasks[Case][1];
if (Imm & SwapMasks[Case][1])
NewImm |= SwapMasks[Case][0];
if (Imm & SwapMasks[Case][2])
NewImm |= SwapMasks[Case][3];
if (Imm & SwapMasks[Case][3])
NewImm |= SwapMasks[Case][2];
MI.getOperand(MI.getNumOperands() - 1).setImm(NewImm);
}
// Returns true if this is a VPERMI2 or VPERMT2 instruction that can be
// commuted.
static bool isCommutableVPERMV3Instruction(unsigned Opcode) {
#define VPERM_CASES(Suffix) \
case X86::VPERMI2##Suffix##Z128rr: \
case X86::VPERMT2##Suffix##Z128rr: \
case X86::VPERMI2##Suffix##Z256rr: \
case X86::VPERMT2##Suffix##Z256rr: \
case X86::VPERMI2##Suffix##Zrr: \
case X86::VPERMT2##Suffix##Zrr: \
case X86::VPERMI2##Suffix##Z128rm: \
case X86::VPERMT2##Suffix##Z128rm: \
case X86::VPERMI2##Suffix##Z256rm: \
case X86::VPERMT2##Suffix##Z256rm: \
case X86::VPERMI2##Suffix##Zrm: \
case X86::VPERMT2##Suffix##Zrm: \
case X86::VPERMI2##Suffix##Z128rrkz: \
case X86::VPERMT2##Suffix##Z128rrkz: \
case X86::VPERMI2##Suffix##Z256rrkz: \
case X86::VPERMT2##Suffix##Z256rrkz: \
case X86::VPERMI2##Suffix##Zrrkz: \
case X86::VPERMT2##Suffix##Zrrkz: \
case X86::VPERMI2##Suffix##Z128rmkz: \
case X86::VPERMT2##Suffix##Z128rmkz: \
case X86::VPERMI2##Suffix##Z256rmkz: \
case X86::VPERMT2##Suffix##Z256rmkz: \
case X86::VPERMI2##Suffix##Zrmkz: \
case X86::VPERMT2##Suffix##Zrmkz:
#define VPERM_CASES_BROADCAST(Suffix) \
VPERM_CASES(Suffix) \
case X86::VPERMI2##Suffix##Z128rmb: \
case X86::VPERMT2##Suffix##Z128rmb: \
case X86::VPERMI2##Suffix##Z256rmb: \
case X86::VPERMT2##Suffix##Z256rmb: \
case X86::VPERMI2##Suffix##Zrmb: \
case X86::VPERMT2##Suffix##Zrmb: \
case X86::VPERMI2##Suffix##Z128rmbkz: \
case X86::VPERMT2##Suffix##Z128rmbkz: \
case X86::VPERMI2##Suffix##Z256rmbkz: \
case X86::VPERMT2##Suffix##Z256rmbkz: \
case X86::VPERMI2##Suffix##Zrmbkz: \
case X86::VPERMT2##Suffix##Zrmbkz:
switch (Opcode) {
default:
return false;
VPERM_CASES(B)
VPERM_CASES_BROADCAST(D)
VPERM_CASES_BROADCAST(PD)
VPERM_CASES_BROADCAST(PS)
VPERM_CASES_BROADCAST(Q)
VPERM_CASES(W)
return true;
}
#undef VPERM_CASES_BROADCAST
#undef VPERM_CASES
}
// Returns commuted opcode for VPERMI2 and VPERMT2 instructions by switching
// from the I opcode to the T opcode and vice versa.
static unsigned getCommutedVPERMV3Opcode(unsigned Opcode) {
#define VPERM_CASES(Orig, New) \
case X86::Orig##Z128rr: \
return X86::New##Z128rr; \
case X86::Orig##Z128rrkz: \
return X86::New##Z128rrkz; \
case X86::Orig##Z128rm: \
return X86::New##Z128rm; \
case X86::Orig##Z128rmkz: \
return X86::New##Z128rmkz; \
case X86::Orig##Z256rr: \
return X86::New##Z256rr; \
case X86::Orig##Z256rrkz: \
return X86::New##Z256rrkz; \
case X86::Orig##Z256rm: \
return X86::New##Z256rm; \
case X86::Orig##Z256rmkz: \
return X86::New##Z256rmkz; \
case X86::Orig##Zrr: \
return X86::New##Zrr; \
case X86::Orig##Zrrkz: \
return X86::New##Zrrkz; \
case X86::Orig##Zrm: \
return X86::New##Zrm; \
case X86::Orig##Zrmkz: \
return X86::New##Zrmkz;
#define VPERM_CASES_BROADCAST(Orig, New) \
VPERM_CASES(Orig, New) \
case X86::Orig##Z128rmb: \
return X86::New##Z128rmb; \
case X86::Orig##Z128rmbkz: \
return X86::New##Z128rmbkz; \
case X86::Orig##Z256rmb: \
return X86::New##Z256rmb; \
case X86::Orig##Z256rmbkz: \
return X86::New##Z256rmbkz; \
case X86::Orig##Zrmb: \
return X86::New##Zrmb; \
case X86::Orig##Zrmbkz: \
return X86::New##Zrmbkz;
switch (Opcode) {
VPERM_CASES(VPERMI2B, VPERMT2B)
VPERM_CASES_BROADCAST(VPERMI2D, VPERMT2D)
VPERM_CASES_BROADCAST(VPERMI2PD, VPERMT2PD)
VPERM_CASES_BROADCAST(VPERMI2PS, VPERMT2PS)
VPERM_CASES_BROADCAST(VPERMI2Q, VPERMT2Q)
VPERM_CASES(VPERMI2W, VPERMT2W)
VPERM_CASES(VPERMT2B, VPERMI2B)
VPERM_CASES_BROADCAST(VPERMT2D, VPERMI2D)
VPERM_CASES_BROADCAST(VPERMT2PD, VPERMI2PD)
VPERM_CASES_BROADCAST(VPERMT2PS, VPERMI2PS)
VPERM_CASES_BROADCAST(VPERMT2Q, VPERMI2Q)
VPERM_CASES(VPERMT2W, VPERMI2W)
}
llvm_unreachable("Unreachable!");
#undef VPERM_CASES_BROADCAST
#undef VPERM_CASES
}
MachineInstr *X86InstrInfo::commuteInstructionImpl(MachineInstr &MI, bool NewMI,
unsigned OpIdx1,
unsigned OpIdx2) const {
auto CloneIfNew = [&](MachineInstr &MI) {
return std::exchange(NewMI, false)
? MI.getParent()->getParent()->CloneMachineInstr(&MI)
: &MI;
};
MachineInstr *WorkingMI = nullptr;
unsigned Opc = MI.getOpcode();
#define CASE_ND(OP) \
case X86::OP: \
case X86::OP##_ND:
switch (Opc) {
// SHLD B, C, I <-> SHRD C, B, (BitWidth - I)
CASE_ND(SHRD16rri8)
CASE_ND(SHLD16rri8)
CASE_ND(SHRD32rri8)
CASE_ND(SHLD32rri8)
CASE_ND(SHRD64rri8)
CASE_ND(SHLD64rri8) {
unsigned Size;
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
#define FROM_TO_SIZE(A, B, S) \
case X86::A: \
Opc = X86::B; \
Size = S; \
break; \
case X86::A##_ND: \
Opc = X86::B##_ND; \
Size = S; \
break; \
case X86::B: \
Opc = X86::A; \
Size = S; \
break; \
case X86::B##_ND: \
Opc = X86::A##_ND; \
Size = S; \
break;
FROM_TO_SIZE(SHRD16rri8, SHLD16rri8, 16)
FROM_TO_SIZE(SHRD32rri8, SHLD32rri8, 32)
FROM_TO_SIZE(SHRD64rri8, SHLD64rri8, 64)
#undef FROM_TO_SIZE
}
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(Opc));
WorkingMI->getOperand(3).setImm(Size - MI.getOperand(3).getImm());
break;
}
case X86::PFSUBrr:
case X86::PFSUBRrr:
// PFSUB x, y: x = x - y
// PFSUBR x, y: x = y - x
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(
get(X86::PFSUBRrr == Opc ? X86::PFSUBrr : X86::PFSUBRrr));
break;
case X86::BLENDPDrri:
case X86::BLENDPSrri:
case X86::VBLENDPDrri:
case X86::VBLENDPSrri:
// If we're optimizing for size, try to use MOVSD/MOVSS.
if (MI.getParent()->getParent()->getFunction().hasOptSize()) {
unsigned Mask = (Opc == X86::BLENDPDrri || Opc == X86::VBLENDPDrri) ? 0x03: 0x0F;
if ((MI.getOperand(3).getImm() ^ Mask) == 1) {
#define FROM_TO(FROM, TO) \
case X86::FROM: \
Opc = X86::TO; \
break;
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
FROM_TO(BLENDPDrri, MOVSDrr)
FROM_TO(BLENDPSrri, MOVSSrr)
FROM_TO(VBLENDPDrri, VMOVSDrr)
FROM_TO(VBLENDPSrri, VMOVSSrr)
}
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(Opc));
WorkingMI->removeOperand(3);
break;
}
#undef FROM_TO
}
[[fallthrough]];
case X86::PBLENDWrri:
case X86::VBLENDPDYrri:
case X86::VBLENDPSYrri:
case X86::VPBLENDDrri:
case X86::VPBLENDWrri:
case X86::VPBLENDDYrri:
case X86::VPBLENDWYrri: {
int8_t Mask;
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
case X86::BLENDPDrri:
Mask = (int8_t)0x03;
break;
case X86::BLENDPSrri:
Mask = (int8_t)0x0F;
break;
case X86::PBLENDWrri:
Mask = (int8_t)0xFF;
break;
case X86::VBLENDPDrri:
Mask = (int8_t)0x03;
break;
case X86::VBLENDPSrri:
Mask = (int8_t)0x0F;
break;
case X86::VBLENDPDYrri:
Mask = (int8_t)0x0F;
break;
case X86::VBLENDPSYrri:
Mask = (int8_t)0xFF;
break;
case X86::VPBLENDDrri:
Mask = (int8_t)0x0F;
break;
case X86::VPBLENDWrri:
Mask = (int8_t)0xFF;
break;
case X86::VPBLENDDYrri:
Mask = (int8_t)0xFF;
break;
case X86::VPBLENDWYrri:
Mask = (int8_t)0xFF;
break;
}
// Only the least significant bits of Imm are used.
// Using int8_t to ensure it will be sign extended to the int64_t that
// setImm takes in order to match isel behavior.
int8_t Imm = MI.getOperand(3).getImm() & Mask;
WorkingMI = CloneIfNew(MI);
WorkingMI->getOperand(3).setImm(Mask ^ Imm);
break;
}
case X86::INSERTPSrri:
case X86::VINSERTPSrri:
case X86::VINSERTPSZrri: {
unsigned Imm = MI.getOperand(MI.getNumOperands() - 1).getImm();
unsigned ZMask = Imm & 15;
unsigned DstIdx = (Imm >> 4) & 3;
unsigned SrcIdx = (Imm >> 6) & 3;
// We can commute insertps if we zero 2 of the elements, the insertion is
// "inline" and we don't override the insertion with a zero.
if (DstIdx == SrcIdx && (ZMask & (1 << DstIdx)) == 0 &&
llvm::popcount(ZMask) == 2) {
unsigned AltIdx = llvm::countr_zero((ZMask | (1 << DstIdx)) ^ 15);
assert(AltIdx < 4 && "Illegal insertion index");
unsigned AltImm = (AltIdx << 6) | (AltIdx << 4) | ZMask;
WorkingMI = CloneIfNew(MI);
WorkingMI->getOperand(MI.getNumOperands() - 1).setImm(AltImm);
break;
}
return nullptr;
}
case X86::MOVSDrr:
case X86::MOVSSrr:
case X86::VMOVSDrr:
case X86::VMOVSSrr: {
// On SSE41 or later we can commute a MOVSS/MOVSD to a BLENDPS/BLENDPD.
if (Subtarget.hasSSE41()) {
unsigned Mask;
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
case X86::MOVSDrr:
Opc = X86::BLENDPDrri;
Mask = 0x02;
break;
case X86::MOVSSrr:
Opc = X86::BLENDPSrri;
Mask = 0x0E;
break;
case X86::VMOVSDrr:
Opc = X86::VBLENDPDrri;
Mask = 0x02;
break;
case X86::VMOVSSrr:
Opc = X86::VBLENDPSrri;
Mask = 0x0E;
break;
}
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(Opc));
WorkingMI->addOperand(MachineOperand::CreateImm(Mask));
break;
}
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(X86::SHUFPDrri));
WorkingMI->addOperand(MachineOperand::CreateImm(0x02));
break;
}
case X86::SHUFPDrri: {
// Commute to MOVSD.
assert(MI.getOperand(3).getImm() == 0x02 && "Unexpected immediate!");
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(X86::MOVSDrr));
WorkingMI->removeOperand(3);
break;
}
case X86::PCLMULQDQrri:
case X86::VPCLMULQDQrri:
case X86::VPCLMULQDQYrri:
case X86::VPCLMULQDQZrri:
case X86::VPCLMULQDQZ128rri:
case X86::VPCLMULQDQZ256rri: {
// SRC1 64bits = Imm[0] ? SRC1[127:64] : SRC1[63:0]
// SRC2 64bits = Imm[4] ? SRC2[127:64] : SRC2[63:0]
unsigned Imm = MI.getOperand(3).getImm();
unsigned Src1Hi = Imm & 0x01;
unsigned Src2Hi = Imm & 0x10;
WorkingMI = CloneIfNew(MI);
WorkingMI->getOperand(3).setImm((Src1Hi << 4) | (Src2Hi >> 4));
break;
}
case X86::VPCMPBZ128rri:
case X86::VPCMPUBZ128rri:
case X86::VPCMPBZ256rri:
case X86::VPCMPUBZ256rri:
case X86::VPCMPBZrri:
case X86::VPCMPUBZrri:
case X86::VPCMPDZ128rri:
case X86::VPCMPUDZ128rri:
case X86::VPCMPDZ256rri:
case X86::VPCMPUDZ256rri:
case X86::VPCMPDZrri:
case X86::VPCMPUDZrri:
case X86::VPCMPQZ128rri:
case X86::VPCMPUQZ128rri:
case X86::VPCMPQZ256rri:
case X86::VPCMPUQZ256rri:
case X86::VPCMPQZrri:
case X86::VPCMPUQZrri:
case X86::VPCMPWZ128rri:
case X86::VPCMPUWZ128rri:
case X86::VPCMPWZ256rri:
case X86::VPCMPUWZ256rri:
case X86::VPCMPWZrri:
case X86::VPCMPUWZrri:
case X86::VPCMPBZ128rrik:
case X86::VPCMPUBZ128rrik:
case X86::VPCMPBZ256rrik:
case X86::VPCMPUBZ256rrik:
case X86::VPCMPBZrrik:
case X86::VPCMPUBZrrik:
case X86::VPCMPDZ128rrik:
case X86::VPCMPUDZ128rrik:
case X86::VPCMPDZ256rrik:
case X86::VPCMPUDZ256rrik:
case X86::VPCMPDZrrik:
case X86::VPCMPUDZrrik:
case X86::VPCMPQZ128rrik:
case X86::VPCMPUQZ128rrik:
case X86::VPCMPQZ256rrik:
case X86::VPCMPUQZ256rrik:
case X86::VPCMPQZrrik:
case X86::VPCMPUQZrrik:
case X86::VPCMPWZ128rrik:
case X86::VPCMPUWZ128rrik:
case X86::VPCMPWZ256rrik:
case X86::VPCMPUWZ256rrik:
case X86::VPCMPWZrrik:
case X86::VPCMPUWZrrik:
WorkingMI = CloneIfNew(MI);
// Flip comparison mode immediate (if necessary).
WorkingMI->getOperand(MI.getNumOperands() - 1)
.setImm(X86::getSwappedVPCMPImm(
MI.getOperand(MI.getNumOperands() - 1).getImm() & 0x7));
break;
case X86::VPCOMBri:
case X86::VPCOMUBri:
case X86::VPCOMDri:
case X86::VPCOMUDri:
case X86::VPCOMQri:
case X86::VPCOMUQri:
case X86::VPCOMWri:
case X86::VPCOMUWri:
WorkingMI = CloneIfNew(MI);
// Flip comparison mode immediate (if necessary).
WorkingMI->getOperand(3).setImm(
X86::getSwappedVPCOMImm(MI.getOperand(3).getImm() & 0x7));
break;
case X86::VCMPSDZrri:
case X86::VCMPSSZrri:
case X86::VCMPPDZrri:
case X86::VCMPPSZrri:
case X86::VCMPSHZrri:
case X86::VCMPPHZrri:
case X86::VCMPPHZ128rri:
case X86::VCMPPHZ256rri:
case X86::VCMPPDZ128rri:
case X86::VCMPPSZ128rri:
case X86::VCMPPDZ256rri:
case X86::VCMPPSZ256rri:
case X86::VCMPPDZrrik:
case X86::VCMPPSZrrik:
case X86::VCMPPDZ128rrik:
case X86::VCMPPSZ128rrik:
case X86::VCMPPDZ256rrik:
case X86::VCMPPSZ256rrik:
WorkingMI = CloneIfNew(MI);
WorkingMI->getOperand(MI.getNumExplicitOperands() - 1)
.setImm(X86::getSwappedVCMPImm(
MI.getOperand(MI.getNumExplicitOperands() - 1).getImm() & 0x1f));
break;
case X86::VPERM2F128rri:
case X86::VPERM2I128rri:
// Flip permute source immediate.
// Imm & 0x02: lo = if set, select Op1.lo/hi else Op0.lo/hi.
// Imm & 0x20: hi = if set, select Op1.lo/hi else Op0.lo/hi.
WorkingMI = CloneIfNew(MI);
WorkingMI->getOperand(3).setImm((MI.getOperand(3).getImm() & 0xFF) ^ 0x22);
break;
case X86::MOVHLPSrr:
case X86::UNPCKHPDrr:
case X86::VMOVHLPSrr:
case X86::VUNPCKHPDrr:
case X86::VMOVHLPSZrr:
case X86::VUNPCKHPDZ128rr:
assert(Subtarget.hasSSE2() && "Commuting MOVHLP/UNPCKHPD requires SSE2!");
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
case X86::MOVHLPSrr:
Opc = X86::UNPCKHPDrr;
break;
case X86::UNPCKHPDrr:
Opc = X86::MOVHLPSrr;
break;
case X86::VMOVHLPSrr:
Opc = X86::VUNPCKHPDrr;
break;
case X86::VUNPCKHPDrr:
Opc = X86::VMOVHLPSrr;
break;
case X86::VMOVHLPSZrr:
Opc = X86::VUNPCKHPDZ128rr;
break;
case X86::VUNPCKHPDZ128rr:
Opc = X86::VMOVHLPSZrr;
break;
}
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(Opc));
break;
CASE_ND(CMOV16rr)
CASE_ND(CMOV32rr)
CASE_ND(CMOV64rr) {
WorkingMI = CloneIfNew(MI);
unsigned OpNo = MI.getDesc().getNumOperands() - 1;
X86::CondCode CC = static_cast<X86::CondCode>(MI.getOperand(OpNo).getImm());
WorkingMI->getOperand(OpNo).setImm(X86::GetOppositeBranchCondition(CC));
break;
}
case X86::VPTERNLOGDZrri:
case X86::VPTERNLOGDZrmi:
case X86::VPTERNLOGDZ128rri:
case X86::VPTERNLOGDZ128rmi:
case X86::VPTERNLOGDZ256rri:
case X86::VPTERNLOGDZ256rmi:
case X86::VPTERNLOGQZrri:
case X86::VPTERNLOGQZrmi:
case X86::VPTERNLOGQZ128rri:
case X86::VPTERNLOGQZ128rmi:
case X86::VPTERNLOGQZ256rri:
case X86::VPTERNLOGQZ256rmi:
case X86::VPTERNLOGDZrrik:
case X86::VPTERNLOGDZ128rrik:
case X86::VPTERNLOGDZ256rrik:
case X86::VPTERNLOGQZrrik:
case X86::VPTERNLOGQZ128rrik:
case X86::VPTERNLOGQZ256rrik:
case X86::VPTERNLOGDZrrikz:
case X86::VPTERNLOGDZrmikz:
case X86::VPTERNLOGDZ128rrikz:
case X86::VPTERNLOGDZ128rmikz:
case X86::VPTERNLOGDZ256rrikz:
case X86::VPTERNLOGDZ256rmikz:
case X86::VPTERNLOGQZrrikz:
case X86::VPTERNLOGQZrmikz:
case X86::VPTERNLOGQZ128rrikz:
case X86::VPTERNLOGQZ128rmikz:
case X86::VPTERNLOGQZ256rrikz:
case X86::VPTERNLOGQZ256rmikz:
case X86::VPTERNLOGDZ128rmbi:
case X86::VPTERNLOGDZ256rmbi:
case X86::VPTERNLOGDZrmbi:
case X86::VPTERNLOGQZ128rmbi:
case X86::VPTERNLOGQZ256rmbi:
case X86::VPTERNLOGQZrmbi:
case X86::VPTERNLOGDZ128rmbikz:
case X86::VPTERNLOGDZ256rmbikz:
case X86::VPTERNLOGDZrmbikz:
case X86::VPTERNLOGQZ128rmbikz:
case X86::VPTERNLOGQZ256rmbikz:
case X86::VPTERNLOGQZrmbikz: {
WorkingMI = CloneIfNew(MI);
commuteVPTERNLOG(*WorkingMI, OpIdx1, OpIdx2);
break;
}
default:
if (isCommutableVPERMV3Instruction(Opc)) {
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(get(getCommutedVPERMV3Opcode(Opc)));
break;
}
if (auto *FMA3Group = getFMA3Group(Opc, MI.getDesc().TSFlags)) {
WorkingMI = CloneIfNew(MI);
WorkingMI->setDesc(
get(getFMA3OpcodeToCommuteOperands(MI, OpIdx1, OpIdx2, *FMA3Group)));
break;
}
}
return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}
bool X86InstrInfo::findThreeSrcCommutedOpIndices(const MachineInstr &MI,
unsigned &SrcOpIdx1,
unsigned &SrcOpIdx2,
bool IsIntrinsic) const {
uint64_t TSFlags = MI.getDesc().TSFlags;
unsigned FirstCommutableVecOp = 1;
unsigned LastCommutableVecOp = 3;
unsigned KMaskOp = -1U;
if (X86II::isKMasked(TSFlags)) {
// For k-zero-masked operations it is Ok to commute the first vector
// operand. Unless this is an intrinsic instruction.
// For regular k-masked operations a conservative choice is done as the
// elements of the first vector operand, for which the corresponding bit
// in the k-mask operand is set to 0, are copied to the result of the
// instruction.
// TODO/FIXME: The commute still may be legal if it is known that the
// k-mask operand is set to either all ones or all zeroes.
// It is also Ok to commute the 1st operand if all users of MI use only
// the elements enabled by the k-mask operand. For example,
// v4 = VFMADD213PSZrk v1, k, v2, v3; // v1[i] = k[i] ? v2[i]*v1[i]+v3[i]
// : v1[i];
// VMOVAPSZmrk <mem_addr>, k, v4; // this is the ONLY user of v4 ->
// // Ok, to commute v1 in FMADD213PSZrk.
// The k-mask operand has index = 2 for masked and zero-masked operations.
KMaskOp = 2;
// The operand with index = 1 is used as a source for those elements for
// which the corresponding bit in the k-mask is set to 0.
if (X86II::isKMergeMasked(TSFlags) || IsIntrinsic)
FirstCommutableVecOp = 3;
LastCommutableVecOp++;
} else if (IsIntrinsic) {
// Commuting the first operand of an intrinsic instruction isn't possible
// unless we can prove that only the lowest element of the result is used.
FirstCommutableVecOp = 2;
}
if (isMem(MI, LastCommutableVecOp))
LastCommutableVecOp--;
// Only the first RegOpsNum operands are commutable.
// Also, the value 'CommuteAnyOperandIndex' is valid here as it means
// that the operand is not specified/fixed.
if (SrcOpIdx1 != CommuteAnyOperandIndex &&
(SrcOpIdx1 < FirstCommutableVecOp || SrcOpIdx1 > LastCommutableVecOp ||
SrcOpIdx1 == KMaskOp))
return false;
if (SrcOpIdx2 != CommuteAnyOperandIndex &&
(SrcOpIdx2 < FirstCommutableVecOp || SrcOpIdx2 > LastCommutableVecOp ||
SrcOpIdx2 == KMaskOp))
return false;
// Look for two different register operands assumed to be commutable
// regardless of the FMA opcode. The FMA opcode is adjusted later.
if (SrcOpIdx1 == CommuteAnyOperandIndex ||
SrcOpIdx2 == CommuteAnyOperandIndex) {
unsigned CommutableOpIdx2 = SrcOpIdx2;
// At least one of operands to be commuted is not specified and
// this method is free to choose appropriate commutable operands.
if (SrcOpIdx1 == SrcOpIdx2)
// Both of operands are not fixed. By default set one of commutable
// operands to the last register operand of the instruction.
CommutableOpIdx2 = LastCommutableVecOp;
else if (SrcOpIdx2 == CommuteAnyOperandIndex)
// Only one of operands is not fixed.
CommutableOpIdx2 = SrcOpIdx1;
// CommutableOpIdx2 is well defined now. Let's choose another commutable
// operand and assign its index to CommutableOpIdx1.
Register Op2Reg = MI.getOperand(CommutableOpIdx2).getReg();
unsigned CommutableOpIdx1;
for (CommutableOpIdx1 = LastCommutableVecOp;
CommutableOpIdx1 >= FirstCommutableVecOp; CommutableOpIdx1--) {
// Just ignore and skip the k-mask operand.
if (CommutableOpIdx1 == KMaskOp)
continue;
// The commuted operands must have different registers.
// Otherwise, the commute transformation does not change anything and
// is useless then.
if (Op2Reg != MI.getOperand(CommutableOpIdx1).getReg())
break;
}
// No appropriate commutable operands were found.
if (CommutableOpIdx1 < FirstCommutableVecOp)
return false;
// Assign the found pair of commutable indices to SrcOpIdx1 and SrcOpidx2
// to return those values.
if (!fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, CommutableOpIdx1,
CommutableOpIdx2))
return false;
}
return true;
}
bool X86InstrInfo::findCommutedOpIndices(const MachineInstr &MI,
unsigned &SrcOpIdx1,
unsigned &SrcOpIdx2) const {
const MCInstrDesc &Desc = MI.getDesc();
if (!Desc.isCommutable())
return false;
switch (MI.getOpcode()) {
case X86::CMPSDrri:
case X86::CMPSSrri:
case X86::CMPPDrri:
case X86::CMPPSrri:
case X86::VCMPSDrri:
case X86::VCMPSSrri:
case X86::VCMPPDrri:
case X86::VCMPPSrri:
case X86::VCMPPDYrri:
case X86::VCMPPSYrri:
case X86::VCMPSDZrri:
case X86::VCMPSSZrri:
case X86::VCMPPDZrri:
case X86::VCMPPSZrri:
case X86::VCMPSHZrri:
case X86::VCMPPHZrri:
case X86::VCMPPHZ128rri:
case X86::VCMPPHZ256rri:
case X86::VCMPPDZ128rri:
case X86::VCMPPSZ128rri:
case X86::VCMPPDZ256rri:
case X86::VCMPPSZ256rri:
case X86::VCMPPDZrrik:
case X86::VCMPPSZrrik:
case X86::VCMPPDZ128rrik:
case X86::VCMPPSZ128rrik:
case X86::VCMPPDZ256rrik:
case X86::VCMPPSZ256rrik: {
unsigned OpOffset = X86II::isKMasked(Desc.TSFlags) ? 1 : 0;
// Float comparison can be safely commuted for
// Ordered/Unordered/Equal/NotEqual tests
unsigned Imm = MI.getOperand(3 + OpOffset).getImm() & 0x7;
switch (Imm) {
default:
// EVEX versions can be commuted.
if ((Desc.TSFlags & X86II::EncodingMask) == X86II::EVEX)
break;
return false;
case 0x00: // EQUAL
case 0x03: // UNORDERED
case 0x04: // NOT EQUAL
case 0x07: // ORDERED
break;
}
// The indices of the commutable operands are 1 and 2 (or 2 and 3
// when masked).
// Assign them to the returned operand indices here.
return fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, 1 + OpOffset,
2 + OpOffset);
}
case X86::MOVSSrr:
// X86::MOVSDrr is always commutable. MOVSS is only commutable if we can
// form sse4.1 blend. We assume VMOVSSrr/VMOVSDrr is always commutable since
// AVX implies sse4.1.
if (Subtarget.hasSSE41())
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
return false;
case X86::SHUFPDrri:
// We can commute this to MOVSD.
if (MI.getOperand(3).getImm() == 0x02)
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
return false;
case X86::MOVHLPSrr:
case X86::UNPCKHPDrr:
case X86::VMOVHLPSrr:
case X86::VUNPCKHPDrr:
case X86::VMOVHLPSZrr:
case X86::VUNPCKHPDZ128rr:
if (Subtarget.hasSSE2())
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
return false;
case X86::VPTERNLOGDZrri:
case X86::VPTERNLOGDZrmi:
case X86::VPTERNLOGDZ128rri:
case X86::VPTERNLOGDZ128rmi:
case X86::VPTERNLOGDZ256rri:
case X86::VPTERNLOGDZ256rmi:
case X86::VPTERNLOGQZrri:
case X86::VPTERNLOGQZrmi:
case X86::VPTERNLOGQZ128rri:
case X86::VPTERNLOGQZ128rmi:
case X86::VPTERNLOGQZ256rri:
case X86::VPTERNLOGQZ256rmi:
case X86::VPTERNLOGDZrrik:
case X86::VPTERNLOGDZ128rrik:
case X86::VPTERNLOGDZ256rrik:
case X86::VPTERNLOGQZrrik:
case X86::VPTERNLOGQZ128rrik:
case X86::VPTERNLOGQZ256rrik:
case X86::VPTERNLOGDZrrikz:
case X86::VPTERNLOGDZrmikz:
case X86::VPTERNLOGDZ128rrikz:
case X86::VPTERNLOGDZ128rmikz:
case X86::VPTERNLOGDZ256rrikz:
case X86::VPTERNLOGDZ256rmikz:
case X86::VPTERNLOGQZrrikz:
case X86::VPTERNLOGQZrmikz:
case X86::VPTERNLOGQZ128rrikz:
case X86::VPTERNLOGQZ128rmikz:
case X86::VPTERNLOGQZ256rrikz:
case X86::VPTERNLOGQZ256rmikz:
case X86::VPTERNLOGDZ128rmbi:
case X86::VPTERNLOGDZ256rmbi:
case X86::VPTERNLOGDZrmbi:
case X86::VPTERNLOGQZ128rmbi:
case X86::VPTERNLOGQZ256rmbi:
case X86::VPTERNLOGQZrmbi:
case X86::VPTERNLOGDZ128rmbikz:
case X86::VPTERNLOGDZ256rmbikz:
case X86::VPTERNLOGDZrmbikz:
case X86::VPTERNLOGQZ128rmbikz:
case X86::VPTERNLOGQZ256rmbikz:
case X86::VPTERNLOGQZrmbikz:
return findThreeSrcCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
case X86::VPDPWSSDYrr:
case X86::VPDPWSSDrr:
case X86::VPDPWSSDSYrr:
case X86::VPDPWSSDSrr:
case X86::VPDPWUUDrr:
case X86::VPDPWUUDYrr:
case X86::VPDPWUUDSrr:
case X86::VPDPWUUDSYrr:
case X86::VPDPBSSDSrr:
case X86::VPDPBSSDSYrr:
case X86::VPDPBSSDrr:
case X86::VPDPBSSDYrr:
case X86::VPDPBUUDSrr:
case X86::VPDPBUUDSYrr:
case X86::VPDPBUUDrr:
case X86::VPDPBUUDYrr:
case X86::VPDPBSSDSZ128r:
case X86::VPDPBSSDSZ128rk:
case X86::VPDPBSSDSZ128rkz:
case X86::VPDPBSSDSZ256r:
case X86::VPDPBSSDSZ256rk:
case X86::VPDPBSSDSZ256rkz:
case X86::VPDPBSSDSZr:
case X86::VPDPBSSDSZrk:
case X86::VPDPBSSDSZrkz:
case X86::VPDPBSSDZ128r:
case X86::VPDPBSSDZ128rk:
case X86::VPDPBSSDZ128rkz:
case X86::VPDPBSSDZ256r:
case X86::VPDPBSSDZ256rk:
case X86::VPDPBSSDZ256rkz:
case X86::VPDPBSSDZr:
case X86::VPDPBSSDZrk:
case X86::VPDPBSSDZrkz:
case X86::VPDPBUUDSZ128r:
case X86::VPDPBUUDSZ128rk:
case X86::VPDPBUUDSZ128rkz:
case X86::VPDPBUUDSZ256r:
case X86::VPDPBUUDSZ256rk:
case X86::VPDPBUUDSZ256rkz:
case X86::VPDPBUUDSZr:
case X86::VPDPBUUDSZrk:
case X86::VPDPBUUDSZrkz:
case X86::VPDPBUUDZ128r:
case X86::VPDPBUUDZ128rk:
case X86::VPDPBUUDZ128rkz:
case X86::VPDPBUUDZ256r:
case X86::VPDPBUUDZ256rk:
case X86::VPDPBUUDZ256rkz:
case X86::VPDPBUUDZr:
case X86::VPDPBUUDZrk:
case X86::VPDPBUUDZrkz:
case X86::VPDPWSSDZ128r:
case X86::VPDPWSSDZ128rk:
case X86::VPDPWSSDZ128rkz:
case X86::VPDPWSSDZ256r:
case X86::VPDPWSSDZ256rk:
case X86::VPDPWSSDZ256rkz:
case X86::VPDPWSSDZr:
case X86::VPDPWSSDZrk:
case X86::VPDPWSSDZrkz:
case X86::VPDPWSSDSZ128r:
case X86::VPDPWSSDSZ128rk:
case X86::VPDPWSSDSZ128rkz:
case X86::VPDPWSSDSZ256r:
case X86::VPDPWSSDSZ256rk:
case X86::VPDPWSSDSZ256rkz:
case X86::VPDPWSSDSZr:
case X86::VPDPWSSDSZrk:
case X86::VPDPWSSDSZrkz:
case X86::VPDPWUUDZ128r:
case X86::VPDPWUUDZ128rk:
case X86::VPDPWUUDZ128rkz:
case X86::VPDPWUUDZ256r:
case X86::VPDPWUUDZ256rk:
case X86::VPDPWUUDZ256rkz:
case X86::VPDPWUUDZr:
case X86::VPDPWUUDZrk:
case X86::VPDPWUUDZrkz:
case X86::VPDPWUUDSZ128r:
case X86::VPDPWUUDSZ128rk:
case X86::VPDPWUUDSZ128rkz:
case X86::VPDPWUUDSZ256r:
case X86::VPDPWUUDSZ256rk:
case X86::VPDPWUUDSZ256rkz:
case X86::VPDPWUUDSZr:
case X86::VPDPWUUDSZrk:
case X86::VPDPWUUDSZrkz:
case X86::VPMADD52HUQrr:
case X86::VPMADD52HUQYrr:
case X86::VPMADD52HUQZ128r:
case X86::VPMADD52HUQZ128rk:
case X86::VPMADD52HUQZ128rkz:
case X86::VPMADD52HUQZ256r:
case X86::VPMADD52HUQZ256rk:
case X86::VPMADD52HUQZ256rkz:
case X86::VPMADD52HUQZr:
case X86::VPMADD52HUQZrk:
case X86::VPMADD52HUQZrkz:
case X86::VPMADD52LUQrr:
case X86::VPMADD52LUQYrr:
case X86::VPMADD52LUQZ128r:
case X86::VPMADD52LUQZ128rk:
case X86::VPMADD52LUQZ128rkz:
case X86::VPMADD52LUQZ256r:
case X86::VPMADD52LUQZ256rk:
case X86::VPMADD52LUQZ256rkz:
case X86::VPMADD52LUQZr:
case X86::VPMADD52LUQZrk:
case X86::VPMADD52LUQZrkz:
case X86::VFMADDCPHZr:
case X86::VFMADDCPHZrk:
case X86::VFMADDCPHZrkz:
case X86::VFMADDCPHZ128r:
case X86::VFMADDCPHZ128rk:
case X86::VFMADDCPHZ128rkz:
case X86::VFMADDCPHZ256r:
case X86::VFMADDCPHZ256rk:
case X86::VFMADDCPHZ256rkz:
case X86::VFMADDCSHZr:
case X86::VFMADDCSHZrk:
case X86::VFMADDCSHZrkz: {
unsigned CommutableOpIdx1 = 2;
unsigned CommutableOpIdx2 = 3;
if (X86II::isKMasked(Desc.TSFlags)) {
// Skip the mask register.
++CommutableOpIdx1;
++CommutableOpIdx2;
}
if (!fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, CommutableOpIdx1,
CommutableOpIdx2))
return false;
if (!MI.getOperand(SrcOpIdx1).isReg() || !MI.getOperand(SrcOpIdx2).isReg())
// No idea.
return false;
return true;
}
default:
const X86InstrFMA3Group *FMA3Group =
getFMA3Group(MI.getOpcode(), MI.getDesc().TSFlags);
if (FMA3Group)
return findThreeSrcCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2,
FMA3Group->isIntrinsic());
// Handled masked instructions since we need to skip over the mask input
// and the preserved input.
if (X86II::isKMasked(Desc.TSFlags)) {
// First assume that the first input is the mask operand and skip past it.
unsigned CommutableOpIdx1 = Desc.getNumDefs() + 1;
unsigned CommutableOpIdx2 = Desc.getNumDefs() + 2;
// Check if the first input is tied. If there isn't one then we only
// need to skip the mask operand which we did above.
if ((MI.getDesc().getOperandConstraint(Desc.getNumDefs(),
MCOI::TIED_TO) != -1)) {
// If this is zero masking instruction with a tied operand, we need to
// move the first index back to the first input since this must
// be a 3 input instruction and we want the first two non-mask inputs.
// Otherwise this is a 2 input instruction with a preserved input and
// mask, so we need to move the indices to skip one more input.
if (X86II::isKMergeMasked(Desc.TSFlags)) {
++CommutableOpIdx1;
++CommutableOpIdx2;
} else {
--CommutableOpIdx1;
}
}
if (!fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, CommutableOpIdx1,
CommutableOpIdx2))
return false;
if (!MI.getOperand(SrcOpIdx1).isReg() ||
!MI.getOperand(SrcOpIdx2).isReg())
// No idea.
return false;
return true;
}
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
}
return false;
}
static bool isConvertibleLEA(MachineInstr *MI) {
unsigned Opcode = MI->getOpcode();
if (Opcode != X86::LEA32r && Opcode != X86::LEA64r &&
Opcode != X86::LEA64_32r)
return false;
const MachineOperand &Scale = MI->getOperand(1 + X86::AddrScaleAmt);
const MachineOperand &Disp = MI->getOperand(1 + X86::AddrDisp);
const MachineOperand &Segment = MI->getOperand(1 + X86::AddrSegmentReg);
if (Segment.getReg() != 0 || !Disp.isImm() || Disp.getImm() != 0 ||
Scale.getImm() > 1)
return false;
return true;
}
bool X86InstrInfo::hasCommutePreference(MachineInstr &MI, bool &Commute) const {
// Currently we're interested in following sequence only.
// r3 = lea r1, r2
// r5 = add r3, r4
// Both r3 and r4 are killed in add, we hope the add instruction has the
// operand order
// r5 = add r4, r3
// So later in X86FixupLEAs the lea instruction can be rewritten as add.
unsigned Opcode = MI.getOpcode();
if (Opcode != X86::ADD32rr && Opcode != X86::ADD64rr)
return false;
const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
Register Reg1 = MI.getOperand(1).getReg();
Register Reg2 = MI.getOperand(2).getReg();
// Check if Reg1 comes from LEA in the same MBB.
if (MachineInstr *Inst = MRI.getUniqueVRegDef(Reg1)) {
if (isConvertibleLEA(Inst) && Inst->getParent() == MI.getParent()) {
Commute = true;
return true;
}
}
// Check if Reg2 comes from LEA in the same MBB.
if (MachineInstr *Inst = MRI.getUniqueVRegDef(Reg2)) {
if (isConvertibleLEA(Inst) && Inst->getParent() == MI.getParent()) {
Commute = false;
return true;
}
}
return false;
}
int X86::getCondSrcNoFromDesc(const MCInstrDesc &MCID) {
unsigned Opcode = MCID.getOpcode();
if (!(X86::isJCC(Opcode) || X86::isSETCC(Opcode) || X86::isSETZUCC(Opcode) ||
X86::isCMOVCC(Opcode) || X86::isCFCMOVCC(Opcode) ||
X86::isCCMPCC(Opcode) || X86::isCTESTCC(Opcode)))
return -1;
// Assume that condition code is always the last use operand.
unsigned NumUses = MCID.getNumOperands() - MCID.getNumDefs();
return NumUses - 1;
}
X86::CondCode X86::getCondFromMI(const MachineInstr &MI) {
const MCInstrDesc &MCID = MI.getDesc();
int CondNo = getCondSrcNoFromDesc(MCID);
if (CondNo < 0)
return X86::COND_INVALID;
CondNo += MCID.getNumDefs();
return static_cast<X86::CondCode>(MI.getOperand(CondNo).getImm());
}
X86::CondCode X86::getCondFromBranch(const MachineInstr &MI) {
return X86::isJCC(MI.getOpcode()) ? X86::getCondFromMI(MI)
: X86::COND_INVALID;
}
X86::CondCode X86::getCondFromSETCC(const MachineInstr &MI) {
return X86::isSETCC(MI.getOpcode()) || X86::isSETZUCC(MI.getOpcode())
? X86::getCondFromMI(MI)
: X86::COND_INVALID;
}
X86::CondCode X86::getCondFromCMov(const MachineInstr &MI) {
return X86::isCMOVCC(MI.getOpcode()) ? X86::getCondFromMI(MI)
: X86::COND_INVALID;
}
X86::CondCode X86::getCondFromCFCMov(const MachineInstr &MI) {
return X86::isCFCMOVCC(MI.getOpcode()) ? X86::getCondFromMI(MI)
: X86::COND_INVALID;
}
X86::CondCode X86::getCondFromCCMP(const MachineInstr &MI) {
return X86::isCCMPCC(MI.getOpcode()) || X86::isCTESTCC(MI.getOpcode())
? X86::getCondFromMI(MI)
: X86::COND_INVALID;
}
int X86::getCCMPCondFlagsFromCondCode(X86::CondCode CC) {
// CCMP/CTEST has two conditional operands:
// - SCC: source conditonal code (same as CMOV)
// - DCF: destination conditional flags, which has 4 valid bits
//
// +----+----+----+----+
// | OF | SF | ZF | CF |
// +----+----+----+----+
//
// If SCC(source conditional code) evaluates to false, CCMP/CTEST will updates
// the conditional flags by as follows:
//
// OF = DCF.OF
// SF = DCF.SF
// ZF = DCF.ZF
// CF = DCF.CF
// PF = DCF.CF
// AF = 0 (Auxiliary Carry Flag)
//
// Otherwise, the CMP or TEST is executed and it updates the
// CSPAZO flags normally.
//
// NOTE:
// If SCC = P, then SCC evaluates to true regardless of the CSPAZO value.
// If SCC = NP, then SCC evaluates to false regardless of the CSPAZO value.
enum { CF = 1, ZF = 2, SF = 4, OF = 8, PF = CF };
switch (CC) {
default:
llvm_unreachable("Illegal condition code!");
case X86::COND_NO:
case X86::COND_NE:
case X86::COND_GE:
case X86::COND_G:
case X86::COND_AE:
case X86::COND_A:
case X86::COND_NS:
case X86::COND_NP:
return 0;
case X86::COND_O:
return OF;
case X86::COND_B:
case X86::COND_BE:
return CF;
break;
case X86::COND_E:
case X86::COND_LE:
return ZF;
case X86::COND_S:
case X86::COND_L:
return SF;
case X86::COND_P:
return PF;
}
}
#define GET_X86_NF_TRANSFORM_TABLE
#define GET_X86_ND2NONND_TABLE
#include "X86GenInstrMapping.inc"
static unsigned getNewOpcFromTable(ArrayRef<X86TableEntry> Table,
unsigned Opc) {
const auto I = llvm::lower_bound(Table, Opc);
return (I == Table.end() || I->OldOpc != Opc) ? 0U : I->NewOpc;
}
unsigned X86::getNFVariant(unsigned Opc) {
return getNewOpcFromTable(X86NFTransformTable, Opc);
}
unsigned X86::getNonNDVariant(unsigned Opc) {
return getNewOpcFromTable(X86ND2NonNDTable, Opc);
}
/// Return the inverse of the specified condition,
/// e.g. turning COND_E to COND_NE.
X86::CondCode X86::GetOppositeBranchCondition(X86::CondCode CC) {
switch (CC) {
default:
llvm_unreachable("Illegal condition code!");
case X86::COND_E:
return X86::COND_NE;
case X86::COND_NE:
return X86::COND_E;
case X86::COND_L:
return X86::COND_GE;
case X86::COND_LE:
return X86::COND_G;
case X86::COND_G:
return X86::COND_LE;
case X86::COND_GE:
return X86::COND_L;
case X86::COND_B:
return X86::COND_AE;
case X86::COND_BE:
return X86::COND_A;
case X86::COND_A:
return X86::COND_BE;
case X86::COND_AE:
return X86::COND_B;
case X86::COND_S:
return X86::COND_NS;
case X86::COND_NS:
return X86::COND_S;
case X86::COND_P:
return X86::COND_NP;
case X86::COND_NP:
return X86::COND_P;
case X86::COND_O:
return X86::COND_NO;
case X86::COND_NO:
return X86::COND_O;
case X86::COND_NE_OR_P:
return X86::COND_E_AND_NP;
case X86::COND_E_AND_NP:
return X86::COND_NE_OR_P;
}
}
/// Assuming the flags are set by MI(a,b), return the condition code if we
/// modify the instructions such that flags are set by MI(b,a).
static X86::CondCode getSwappedCondition(X86::CondCode CC) {
switch (CC) {
default:
return X86::COND_INVALID;
case X86::COND_E:
return X86::COND_E;
case X86::COND_NE:
return X86::COND_NE;
case X86::COND_L:
return X86::COND_G;
case X86::COND_LE:
return X86::COND_GE;
case X86::COND_G:
return X86::COND_L;
case X86::COND_GE:
return X86::COND_LE;
case X86::COND_B:
return X86::COND_A;
case X86::COND_BE:
return X86::COND_AE;
case X86::COND_A:
return X86::COND_B;
case X86::COND_AE:
return X86::COND_BE;
}
}
std::pair<X86::CondCode, bool>
X86::getX86ConditionCode(CmpInst::Predicate Predicate) {
X86::CondCode CC = X86::COND_INVALID;
bool NeedSwap = false;
switch (Predicate) {
default:
break;
// Floating-point Predicates
case CmpInst::FCMP_UEQ:
CC = X86::COND_E;
break;
case CmpInst::FCMP_OLT:
NeedSwap = true;
[[fallthrough]];
case CmpInst::FCMP_OGT:
CC = X86::COND_A;
break;
case CmpInst::FCMP_OLE:
NeedSwap = true;
[[fallthrough]];
case CmpInst::FCMP_OGE:
CC = X86::COND_AE;
break;
case CmpInst::FCMP_UGT:
NeedSwap = true;
[[fallthrough]];
case CmpInst::FCMP_ULT:
CC = X86::COND_B;
break;
case CmpInst::FCMP_UGE:
NeedSwap = true;
[[fallthrough]];
case CmpInst::FCMP_ULE:
CC = X86::COND_BE;
break;
case CmpInst::FCMP_ONE:
CC = X86::COND_NE;
break;
case CmpInst::FCMP_UNO:
CC = X86::COND_P;
break;
case CmpInst::FCMP_ORD:
CC = X86::COND_NP;
break;
case CmpInst::FCMP_OEQ:
[[fallthrough]];
case CmpInst::FCMP_UNE:
CC = X86::COND_INVALID;
break;
// Integer Predicates
case CmpInst::ICMP_EQ:
CC = X86::COND_E;
break;
case CmpInst::ICMP_NE:
CC = X86::COND_NE;
break;
case CmpInst::ICMP_UGT:
CC = X86::COND_A;
break;
case CmpInst::ICMP_UGE:
CC = X86::COND_AE;
break;
case CmpInst::ICMP_ULT:
CC = X86::COND_B;
break;
case CmpInst::ICMP_ULE:
CC = X86::COND_BE;
break;
case CmpInst::ICMP_SGT:
CC = X86::COND_G;
break;
case CmpInst::ICMP_SGE:
CC = X86::COND_GE;
break;
case CmpInst::ICMP_SLT:
CC = X86::COND_L;
break;
case CmpInst::ICMP_SLE:
CC = X86::COND_LE;
break;
}
return std::make_pair(CC, NeedSwap);
}
/// Return a cmov opcode for the given register size in bytes, and operand type.
unsigned X86::getCMovOpcode(unsigned RegBytes, bool HasMemoryOperand,
bool HasNDD) {
switch (RegBytes) {
default:
llvm_unreachable("Illegal register size!");
#define GET_ND_IF_ENABLED(OPC) (HasNDD ? OPC##_ND : OPC)
case 2:
return HasMemoryOperand ? GET_ND_IF_ENABLED(X86::CMOV16rm)
: GET_ND_IF_ENABLED(X86::CMOV16rr);
case 4:
return HasMemoryOperand ? GET_ND_IF_ENABLED(X86::CMOV32rm)
: GET_ND_IF_ENABLED(X86::CMOV32rr);
case 8:
return HasMemoryOperand ? GET_ND_IF_ENABLED(X86::CMOV64rm)
: GET_ND_IF_ENABLED(X86::CMOV64rr);
}
}
/// Get the VPCMP immediate for the given condition.
unsigned X86::getVPCMPImmForCond(ISD::CondCode CC) {
switch (CC) {
default:
llvm_unreachable("Unexpected SETCC condition");
case ISD::SETNE:
return 4;
case ISD::SETEQ:
return 0;
case ISD::SETULT:
case ISD::SETLT:
return 1;
case ISD::SETUGT:
case ISD::SETGT:
return 6;
case ISD::SETUGE:
case ISD::SETGE:
return 5;
case ISD::SETULE:
case ISD::SETLE:
return 2;
}
}
/// Get the VPCMP immediate if the operands are swapped.
unsigned X86::getSwappedVPCMPImm(unsigned Imm) {
switch (Imm) {
default:
llvm_unreachable("Unreachable!");
case 0x01:
Imm = 0x06;
break; // LT -> NLE
case 0x02:
Imm = 0x05;
break; // LE -> NLT
case 0x05:
Imm = 0x02;
break; // NLT -> LE
case 0x06:
Imm = 0x01;
break; // NLE -> LT
case 0x00: // EQ
case 0x03: // FALSE
case 0x04: // NE
case 0x07: // TRUE
break;
}
return Imm;
}
/// Get the VPCOM immediate if the operands are swapped.
unsigned X86::getSwappedVPCOMImm(unsigned Imm) {
switch (Imm) {
default:
llvm_unreachable("Unreachable!");
case 0x00:
Imm = 0x02;
break; // LT -> GT
case 0x01:
Imm = 0x03;
break; // LE -> GE
case 0x02:
Imm = 0x00;
break; // GT -> LT
case 0x03:
Imm = 0x01;
break; // GE -> LE
case 0x04: // EQ
case 0x05: // NE
case 0x06: // FALSE
case 0x07: // TRUE
break;
}
return Imm;
}
/// Get the VCMP immediate if the operands are swapped.
unsigned X86::getSwappedVCMPImm(unsigned Imm) {
// Only need the lower 2 bits to distinquish.
switch (Imm & 0x3) {
default:
llvm_unreachable("Unreachable!");
case 0x00:
case 0x03:
// EQ/NE/TRUE/FALSE/ORD/UNORD don't change immediate when commuted.
break;
case 0x01:
case 0x02:
// Need to toggle bits 3:0. Bit 4 stays the same.
Imm ^= 0xf;
break;
}
return Imm;
}
unsigned X86::getVectorRegisterWidth(const MCOperandInfo &Info) {
if (Info.RegClass == X86::VR128RegClassID ||
Info.RegClass == X86::VR128XRegClassID)
return 128;
if (Info.RegClass == X86::VR256RegClassID ||
Info.RegClass == X86::VR256XRegClassID)
return 256;
if (Info.RegClass == X86::VR512RegClassID)
return 512;
llvm_unreachable("Unknown register class!");
}
/// Return true if the Reg is X87 register.
static bool isX87Reg(unsigned Reg) {
return (Reg == X86::FPCW || Reg == X86::FPSW ||
(Reg >= X86::ST0 && Reg <= X86::ST7));
}
/// check if the instruction is X87 instruction
bool X86::isX87Instruction(MachineInstr &MI) {
// Call and inlineasm defs X87 register, so we special case it here because
// otherwise calls are incorrectly flagged as x87 instructions
// as a result.
if (MI.isCall() || MI.isInlineAsm())
return false;
for (const MachineOperand &MO : MI.operands()) {
if (!MO.isReg())
continue;
if (isX87Reg(MO.getReg()))
return true;
}
return false;
}
int X86::getFirstAddrOperandIdx(const MachineInstr &MI) {
auto IsMemOp = [](const MCOperandInfo &OpInfo) {
return OpInfo.OperandType == MCOI::OPERAND_MEMORY;
};
const MCInstrDesc &Desc = MI.getDesc();
// Directly invoke the MC-layer routine for real (i.e., non-pseudo)
// instructions (fast case).
if (!X86II::isPseudo(Desc.TSFlags)) {
int MemRefIdx = X86II::getMemoryOperandNo(Desc.TSFlags);
if (MemRefIdx >= 0)
return MemRefIdx + X86II::getOperandBias(Desc);
#ifdef EXPENSIVE_CHECKS
assert(none_of(Desc.operands(), IsMemOp) &&
"Got false negative from X86II::getMemoryOperandNo()!");
#endif
return -1;
}
// Otherwise, handle pseudo instructions by examining the type of their
// operands (slow case). An instruction cannot have a memory reference if it
// has fewer than AddrNumOperands (= 5) explicit operands.
unsigned NumOps = Desc.getNumOperands();
if (NumOps < X86::AddrNumOperands) {
#ifdef EXPENSIVE_CHECKS
assert(none_of(Desc.operands(), IsMemOp) &&
"Expected no operands to have OPERAND_MEMORY type!");
#endif
return -1;
}
// The first operand with type OPERAND_MEMORY indicates the start of a memory
// reference. We expect the following AddrNumOperand-1 operands to also have
// OPERAND_MEMORY type.
for (unsigned I = 0, E = NumOps - X86::AddrNumOperands; I != E; ++I) {
if (IsMemOp(Desc.operands()[I])) {
#ifdef EXPENSIVE_CHECKS
assert(std::all_of(Desc.operands().begin() + I,
Desc.operands().begin() + I + X86::AddrNumOperands,
IsMemOp) &&
"Expected all five operands in the memory reference to have "
"OPERAND_MEMORY type!");
#endif
return I;
}
}
return -1;
}
const Constant *X86::getConstantFromPool(const MachineInstr &MI,
unsigned OpNo) {
assert(MI.getNumOperands() >= (OpNo + X86::AddrNumOperands) &&
"Unexpected number of operands!");
const MachineOperand &Index = MI.getOperand(OpNo + X86::AddrIndexReg);
if (!Index.isReg() || Index.getReg() != X86::NoRegister)
return nullptr;
const MachineOperand &Disp = MI.getOperand(OpNo + X86::AddrDisp);
if (!Disp.isCPI() || Disp.getOffset() != 0)
return nullptr;
ArrayRef<MachineConstantPoolEntry> Constants =
MI.getParent()->getParent()->getConstantPool()->getConstants();
const MachineConstantPoolEntry &ConstantEntry = Constants[Disp.getIndex()];
// Bail if this is a machine constant pool entry, we won't be able to dig out
// anything useful.
if (ConstantEntry.isMachineConstantPoolEntry())
return nullptr;
return ConstantEntry.Val.ConstVal;
}
bool X86InstrInfo::isUnconditionalTailCall(const MachineInstr &MI) const {
switch (MI.getOpcode()) {
case X86::TCRETURNdi:
case X86::TCRETURNri:
case X86::TCRETURNmi:
case X86::TCRETURNdi64:
case X86::TCRETURNri64:
case X86::TCRETURNmi64:
return true;
default:
return false;
}
}
bool X86InstrInfo::canMakeTailCallConditional(
SmallVectorImpl<MachineOperand> &BranchCond,
const MachineInstr &TailCall) const {
const MachineFunction *MF = TailCall.getMF();
if (MF->getTarget().getCodeModel() == CodeModel::Kernel) {
// Kernel patches thunk calls in runtime, these should never be conditional.
const MachineOperand &Target = TailCall.getOperand(0);
if (Target.isSymbol()) {
StringRef Symbol(Target.getSymbolName());
// this is currently only relevant to r11/kernel indirect thunk.
if (Symbol == "__x86_indirect_thunk_r11")
return false;
}
}
if (TailCall.getOpcode() != X86::TCRETURNdi &&
TailCall.getOpcode() != X86::TCRETURNdi64) {
// Only direct calls can be done with a conditional branch.
return false;
}
if (Subtarget.isTargetWin64() && MF->hasWinCFI()) {
// Conditional tail calls confuse the Win64 unwinder.
return false;
}
assert(BranchCond.size() == 1);
if (BranchCond[0].getImm() > X86::LAST_VALID_COND) {
// Can't make a conditional tail call with this condition.
return false;
}
const X86MachineFunctionInfo *X86FI = MF->getInfo<X86MachineFunctionInfo>();
if (X86FI->getTCReturnAddrDelta() != 0 ||
TailCall.getOperand(1).getImm() != 0) {
// A conditional tail call cannot do any stack adjustment.
return false;
}
return true;
}
void X86InstrInfo::replaceBranchWithTailCall(
MachineBasicBlock &MBB, SmallVectorImpl<MachineOperand> &BranchCond,
const MachineInstr &TailCall) const {
assert(canMakeTailCallConditional(BranchCond, TailCall));
MachineBasicBlock::iterator I = MBB.end();
while (I != MBB.begin()) {
--I;
if (I->isDebugInstr())
continue;
if (!I->isBranch())
assert(0 && "Can't find the branch to replace!");
X86::CondCode CC = X86::getCondFromBranch(*I);
assert(BranchCond.size() == 1);
if (CC != BranchCond[0].getImm())
continue;
break;
}
unsigned Opc = TailCall.getOpcode() == X86::TCRETURNdi ? X86::TCRETURNdicc
: X86::TCRETURNdi64cc;
auto MIB = BuildMI(MBB, I, MBB.findDebugLoc(I), get(Opc));
MIB->addOperand(TailCall.getOperand(0)); // Destination.
MIB.addImm(0); // Stack offset (not used).
MIB->addOperand(BranchCond[0]); // Condition.
MIB.copyImplicitOps(TailCall); // Regmask and (imp-used) parameters.
// Add implicit uses and defs of all live regs potentially clobbered by the
// call. This way they still appear live across the call.
LivePhysRegs LiveRegs(getRegisterInfo());
LiveRegs.addLiveOuts(MBB);
SmallVector<std::pair<MCPhysReg, const MachineOperand *>, 8> Clobbers;
LiveRegs.stepForward(*MIB, Clobbers);
for (const auto &C : Clobbers) {
MIB.addReg(C.first, RegState::Implicit);
MIB.addReg(C.first, RegState::Implicit | RegState::Define);
}
I->eraseFromParent();
}
// Given a MBB and its TBB, find the FBB which was a fallthrough MBB (it may
// not be a fallthrough MBB now due to layout changes). Return nullptr if the
// fallthrough MBB cannot be identified.
static MachineBasicBlock *getFallThroughMBB(MachineBasicBlock *MBB,
MachineBasicBlock *TBB) {
// Look for non-EHPad successors other than TBB. If we find exactly one, it
// is the fallthrough MBB. If we find zero, then TBB is both the target MBB
// and fallthrough MBB. If we find more than one, we cannot identify the
// fallthrough MBB and should return nullptr.
MachineBasicBlock *FallthroughBB = nullptr;
for (MachineBasicBlock *Succ : MBB->successors()) {
if (Succ->isEHPad() || (Succ == TBB && FallthroughBB))
continue;
// Return a nullptr if we found more than one fallthrough successor.
if (FallthroughBB && FallthroughBB != TBB)
return nullptr;
FallthroughBB = Succ;
}
return FallthroughBB;
}
bool X86InstrInfo::analyzeBranchImpl(
MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
SmallVectorImpl<MachineInstr *> &CondBranches, bool AllowModify) const {
// Start from the bottom of the block and work up, examining the
// terminator instructions.
MachineBasicBlock::iterator I = MBB.end();
MachineBasicBlock::iterator UnCondBrIter = MBB.end();
while (I != MBB.begin()) {
--I;
if (I->isDebugInstr())
continue;
// Working from the bottom, when we see a non-terminator instruction, we're
// done.
if (!isUnpredicatedTerminator(*I))
break;
// A terminator that isn't a branch can't easily be handled by this
// analysis.
if (!I->isBranch())
return true;
// Handle unconditional branches.
if (I->getOpcode() == X86::JMP_1) {
UnCondBrIter = I;
if (!AllowModify) {
TBB = I->getOperand(0).getMBB();
continue;
}
// If the block has any instructions after a JMP, delete them.
MBB.erase(std::next(I), MBB.end());
Cond.clear();
FBB = nullptr;
// Delete the JMP if it's equivalent to a fall-through.
if (MBB.isLayoutSuccessor(I->getOperand(0).getMBB())) {
TBB = nullptr;
I->eraseFromParent();
I = MBB.end();
UnCondBrIter = MBB.end();
continue;
}
// TBB is used to indicate the unconditional destination.
TBB = I->getOperand(0).getMBB();
continue;
}
// Handle conditional branches.
X86::CondCode BranchCode = X86::getCondFromBranch(*I);
if (BranchCode == X86::COND_INVALID)
return true; // Can't handle indirect branch.
// In practice we should never have an undef eflags operand, if we do
// abort here as we are not prepared to preserve the flag.
if (I->findRegisterUseOperand(X86::EFLAGS, /*TRI=*/nullptr)->isUndef())
return true;
// Working from the bottom, handle the first conditional branch.
if (Cond.empty()) {
FBB = TBB;
TBB = I->getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(BranchCode));
CondBranches.push_back(&*I);
continue;
}
// Handle subsequent conditional branches. Only handle the case where all
// conditional branches branch to the same destination and their condition
// opcodes fit one of the special multi-branch idioms.
assert(Cond.size() == 1);
assert(TBB);
// If the conditions are the same, we can leave them alone.
X86::CondCode OldBranchCode = (X86::CondCode)Cond[0].getImm();
auto NewTBB = I->getOperand(0).getMBB();
if (OldBranchCode == BranchCode && TBB == NewTBB)
continue;
// If they differ, see if they fit one of the known patterns. Theoretically,
// we could handle more patterns here, but we shouldn't expect to see them
// if instruction selection has done a reasonable job.
if (TBB == NewTBB &&
((OldBranchCode == X86::COND_P && BranchCode == X86::COND_NE) ||
(OldBranchCode == X86::COND_NE && BranchCode == X86::COND_P))) {
BranchCode = X86::COND_NE_OR_P;
} else if ((OldBranchCode == X86::COND_NP && BranchCode == X86::COND_NE) ||
(OldBranchCode == X86::COND_E && BranchCode == X86::COND_P)) {
if (NewTBB != (FBB ? FBB : getFallThroughMBB(&MBB, TBB)))
return true;
// X86::COND_E_AND_NP usually has two different branch destinations.
//
// JP B1
// JE B2
// JMP B1
// B1:
// B2:
//
// Here this condition branches to B2 only if NP && E. It has another
// equivalent form:
//
// JNE B1
// JNP B2
// JMP B1
// B1:
// B2:
//
// Similarly it branches to B2 only if E && NP. That is why this condition
// is named with COND_E_AND_NP.
BranchCode = X86::COND_E_AND_NP;
} else
return true;
// Update the MachineOperand.
Cond[0].setImm(BranchCode);
CondBranches.push_back(&*I);
}
return false;
}
bool X86InstrInfo::analyzeBranch(MachineBasicBlock &MBB,
MachineBasicBlock *&TBB,
MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify) const {
SmallVector<MachineInstr *, 4> CondBranches;
return analyzeBranchImpl(MBB, TBB, FBB, Cond, CondBranches, AllowModify);
}
static int getJumpTableIndexFromAddr(const MachineInstr &MI) {
const MCInstrDesc &Desc = MI.getDesc();
int MemRefBegin = X86II::getMemoryOperandNo(Desc.TSFlags);
assert(MemRefBegin >= 0 && "instr should have memory operand");
MemRefBegin += X86II::getOperandBias(Desc);
const MachineOperand &MO = MI.getOperand(MemRefBegin + X86::AddrDisp);
if (!MO.isJTI())
return -1;
return MO.getIndex();
}
static int getJumpTableIndexFromReg(const MachineRegisterInfo &MRI,
Register Reg) {
if (!Reg.isVirtual())
return -1;
MachineInstr *MI = MRI.getUniqueVRegDef(Reg);
if (MI == nullptr)
return -1;
unsigned Opcode = MI->getOpcode();
if (Opcode != X86::LEA64r && Opcode != X86::LEA32r)
return -1;
return getJumpTableIndexFromAddr(*MI);
}
int X86InstrInfo::getJumpTableIndex(const MachineInstr &MI) const {
unsigned Opcode = MI.getOpcode();
// Switch-jump pattern for non-PIC code looks like:
// JMP64m $noreg, 8, %X, %jump-table.X, $noreg
if (Opcode == X86::JMP64m || Opcode == X86::JMP32m) {
return getJumpTableIndexFromAddr(MI);
}
// The pattern for PIC code looks like:
// %0 = LEA64r $rip, 1, $noreg, %jump-table.X
// %1 = MOVSX64rm32 %0, 4, XX, 0, $noreg
// %2 = ADD64rr %1, %0
// JMP64r %2
if (Opcode == X86::JMP64r || Opcode == X86::JMP32r) {
Register Reg = MI.getOperand(0).getReg();
if (!Reg.isVirtual())
return -1;
const MachineFunction &MF = *MI.getParent()->getParent();
const MachineRegisterInfo &MRI = MF.getRegInfo();
MachineInstr *Add = MRI.getUniqueVRegDef(Reg);
if (Add == nullptr)
return -1;
if (Add->getOpcode() != X86::ADD64rr && Add->getOpcode() != X86::ADD32rr)
return -1;
int JTI1 = getJumpTableIndexFromReg(MRI, Add->getOperand(1).getReg());
if (JTI1 >= 0)
return JTI1;
int JTI2 = getJumpTableIndexFromReg(MRI, Add->getOperand(2).getReg());
if (JTI2 >= 0)
return JTI2;
}
return -1;
}
bool X86InstrInfo::analyzeBranchPredicate(MachineBasicBlock &MBB,
MachineBranchPredicate &MBP,
bool AllowModify) const {
using namespace std::placeholders;
SmallVector<MachineOperand, 4> Cond;
SmallVector<MachineInstr *, 4> CondBranches;
if (analyzeBranchImpl(MBB, MBP.TrueDest, MBP.FalseDest, Cond, CondBranches,
AllowModify))
return true;
if (Cond.size() != 1)
return true;
assert(MBP.TrueDest && "expected!");
if (!MBP.FalseDest)
MBP.FalseDest = MBB.getNextNode();
const TargetRegisterInfo *TRI = &getRegisterInfo();
MachineInstr *ConditionDef = nullptr;
bool SingleUseCondition = true;
for (MachineInstr &MI : llvm::drop_begin(llvm::reverse(MBB))) {
if (MI.modifiesRegister(X86::EFLAGS, TRI)) {
ConditionDef = &MI;
break;
}
if (MI.readsRegister(X86::EFLAGS, TRI))
SingleUseCondition = false;
}
if (!ConditionDef)
return true;
if (SingleUseCondition) {
for (auto *Succ : MBB.successors())
if (Succ->isLiveIn(X86::EFLAGS))
SingleUseCondition = false;
}
MBP.ConditionDef = ConditionDef;
MBP.SingleUseCondition = SingleUseCondition;
// Currently we only recognize the simple pattern:
//
// test %reg, %reg
// je %label
//
const unsigned TestOpcode =
Subtarget.is64Bit() ? X86::TEST64rr : X86::TEST32rr;
if (ConditionDef->getOpcode() == TestOpcode &&
ConditionDef->getNumOperands() == 3 &&
ConditionDef->getOperand(0).isIdenticalTo(ConditionDef->getOperand(1)) &&
(Cond[0].getImm() == X86::COND_NE || Cond[0].getImm() == X86::COND_E)) {
MBP.LHS = ConditionDef->getOperand(0);
MBP.RHS = MachineOperand::CreateImm(0);
MBP.Predicate = Cond[0].getImm() == X86::COND_NE
? MachineBranchPredicate::PRED_NE
: MachineBranchPredicate::PRED_EQ;
return false;
}
return true;
}
unsigned X86InstrInfo::removeBranch(MachineBasicBlock &MBB,
int *BytesRemoved) const {
assert(!BytesRemoved && "code size not handled");
MachineBasicBlock::iterator I = MBB.end();
unsigned Count = 0;
while (I != MBB.begin()) {
--I;
if (I->isDebugInstr())
continue;
if (I->getOpcode() != X86::JMP_1 &&
X86::getCondFromBranch(*I) == X86::COND_INVALID)
break;
// Remove the branch.
I->eraseFromParent();
I = MBB.end();
++Count;
}
return Count;
}
unsigned X86InstrInfo::insertBranch(MachineBasicBlock &MBB,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
ArrayRef<MachineOperand> Cond,
const DebugLoc &DL, int *BytesAdded) const {
// Shouldn't be a fall through.
assert(TBB && "insertBranch must not be told to insert a fallthrough");
assert((Cond.size() == 1 || Cond.size() == 0) &&
"X86 branch conditions have one component!");
assert(!BytesAdded && "code size not handled");
if (Cond.empty()) {
// Unconditional branch?
assert(!FBB && "Unconditional branch with multiple successors!");
BuildMI(&MBB, DL, get(X86::JMP_1)).addMBB(TBB);
return 1;
}
// If FBB is null, it is implied to be a fall-through block.
bool FallThru = FBB == nullptr;
// Conditional branch.
unsigned Count = 0;
X86::CondCode CC = (X86::CondCode)Cond[0].getImm();
switch (CC) {
case X86::COND_NE_OR_P:
// Synthesize NE_OR_P with two branches.
BuildMI(&MBB, DL, get(X86::JCC_1)).addMBB(TBB).addImm(X86::COND_NE);
++Count;
BuildMI(&MBB, DL, get(X86::JCC_1)).addMBB(TBB).addImm(X86::COND_P);
++Count;
break;
case X86::COND_E_AND_NP:
// Use the next block of MBB as FBB if it is null.
if (FBB == nullptr) {
FBB = getFallThroughMBB(&MBB, TBB);
assert(FBB && "MBB cannot be the last block in function when the false "
"body is a fall-through.");
}
// Synthesize COND_E_AND_NP with two branches.
BuildMI(&MBB, DL, get(X86::JCC_1)).addMBB(FBB).addImm(X86::COND_NE);
++Count;
BuildMI(&MBB, DL, get(X86::JCC_1)).addMBB(TBB).addImm(X86::COND_NP);
++Count;
break;
default: {
BuildMI(&MBB, DL, get(X86::JCC_1)).addMBB(TBB).addImm(CC);
++Count;
}
}
if (!FallThru) {
// Two-way Conditional branch. Insert the second branch.
BuildMI(&MBB, DL, get(X86::JMP_1)).addMBB(FBB);
++Count;
}
return Count;
}
bool X86InstrInfo::canInsertSelect(const MachineBasicBlock &MBB,
ArrayRef<MachineOperand> Cond,
Register DstReg, Register TrueReg,
Register FalseReg, int &CondCycles,
int &TrueCycles, int &FalseCycles) const {
// Not all subtargets have cmov instructions.
if (!Subtarget.canUseCMOV())
return false;
if (Cond.size() != 1)
return false;
// We cannot do the composite conditions, at least not in SSA form.
if ((X86::CondCode)Cond[0].getImm() > X86::LAST_VALID_COND)
return false;
// Check register classes.
const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterClass *RC =
RI.getCommonSubClass(MRI.getRegClass(TrueReg), MRI.getRegClass(FalseReg));
if (!RC)
return false;
// We have cmov instructions for 16, 32, and 64 bit general purpose registers.
if (X86::GR16RegClass.hasSubClassEq(RC) ||
X86::GR32RegClass.hasSubClassEq(RC) ||
X86::GR64RegClass.hasSubClassEq(RC)) {
// This latency applies to Pentium M, Merom, Wolfdale, Nehalem, and Sandy
// Bridge. Probably Ivy Bridge as well.
CondCycles = 2;
TrueCycles = 2;
FalseCycles = 2;
return true;
}
// Can't do vectors.
return false;
}
void X86InstrInfo::insertSelect(MachineBasicBlock &MBB,
MachineBasicBlock::iterator I,
const DebugLoc &DL, Register DstReg,
ArrayRef<MachineOperand> Cond, Register TrueReg,
Register FalseReg) const {
MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterInfo &TRI = *MRI.getTargetRegisterInfo();
const TargetRegisterClass &RC = *MRI.getRegClass(DstReg);
assert(Cond.size() == 1 && "Invalid Cond array");
unsigned Opc =
X86::getCMovOpcode(TRI.getRegSizeInBits(RC) / 8,
false /*HasMemoryOperand*/, Subtarget.hasNDD());
BuildMI(MBB, I, DL, get(Opc), DstReg)
.addReg(FalseReg)
.addReg(TrueReg)
.addImm(Cond[0].getImm());
}
/// Test if the given register is a physical h register.
static bool isHReg(unsigned Reg) {
return X86::GR8_ABCD_HRegClass.contains(Reg);
}
// Try and copy between VR128/VR64 and GR64 registers.
static unsigned CopyToFromAsymmetricReg(unsigned DestReg, unsigned SrcReg,
const X86Subtarget &Subtarget) {
bool HasAVX = Subtarget.hasAVX();
bool HasAVX512 = Subtarget.hasAVX512();
bool HasEGPR = Subtarget.hasEGPR();
// SrcReg(MaskReg) -> DestReg(GR64)
// SrcReg(MaskReg) -> DestReg(GR32)
// All KMASK RegClasses hold the same k registers, can be tested against
// anyone.
if (X86::VK16RegClass.contains(SrcReg)) {
if (X86::GR64RegClass.contains(DestReg)) {
assert(Subtarget.hasBWI());
return HasEGPR ? X86::KMOVQrk_EVEX : X86::KMOVQrk;
}
if (X86::GR32RegClass.contains(DestReg))
return Subtarget.hasBWI() ? (HasEGPR ? X86::KMOVDrk_EVEX : X86::KMOVDrk)
: (HasEGPR ? X86::KMOVWrk_EVEX : X86::KMOVWrk);
}
// SrcReg(GR64) -> DestReg(MaskReg)
// SrcReg(GR32) -> DestReg(MaskReg)
// All KMASK RegClasses hold the same k registers, can be tested against
// anyone.
if (X86::VK16RegClass.contains(DestReg)) {
if (X86::GR64RegClass.contains(SrcReg)) {
assert(Subtarget.hasBWI());
return HasEGPR ? X86::KMOVQkr_EVEX : X86::KMOVQkr;
}
if (X86::GR32RegClass.contains(SrcReg))
return Subtarget.hasBWI() ? (HasEGPR ? X86::KMOVDkr_EVEX : X86::KMOVDkr)
: (HasEGPR ? X86::KMOVWkr_EVEX : X86::KMOVWkr);
}
// SrcReg(VR128) -> DestReg(GR64)
// SrcReg(VR64) -> DestReg(GR64)
// SrcReg(GR64) -> DestReg(VR128)
// SrcReg(GR64) -> DestReg(VR64)
if (X86::GR64RegClass.contains(DestReg)) {
if (X86::VR128XRegClass.contains(SrcReg))
// Copy from a VR128 register to a GR64 register.
return HasAVX512 ? X86::VMOVPQIto64Zrr
: HasAVX ? X86::VMOVPQIto64rr
: X86::MOVPQIto64rr;
if (X86::VR64RegClass.contains(SrcReg))
// Copy from a VR64 register to a GR64 register.
return X86::MMX_MOVD64from64rr;
} else if (X86::GR64RegClass.contains(SrcReg)) {
// Copy from a GR64 register to a VR128 register.
if (X86::VR128XRegClass.contains(DestReg))
return HasAVX512 ? X86::VMOV64toPQIZrr
: HasAVX ? X86::VMOV64toPQIrr
: X86::MOV64toPQIrr;
// Copy from a GR64 register to a VR64 register.
if (X86::VR64RegClass.contains(DestReg))
return X86::MMX_MOVD64to64rr;
}
// SrcReg(VR128) -> DestReg(GR32)
// SrcReg(GR32) -> DestReg(VR128)
if (X86::GR32RegClass.contains(DestReg) &&
X86::VR128XRegClass.contains(SrcReg))
// Copy from a VR128 register to a GR32 register.
return HasAVX512 ? X86::VMOVPDI2DIZrr
: HasAVX ? X86::VMOVPDI2DIrr
: X86::MOVPDI2DIrr;
if (X86::VR128XRegClass.contains(DestReg) &&
X86::GR32RegClass.contains(SrcReg))
// Copy from a VR128 register to a VR128 register.
return HasAVX512 ? X86::VMOVDI2PDIZrr
: HasAVX ? X86::VMOVDI2PDIrr
: X86::MOVDI2PDIrr;
return 0;
}
void X86InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const DebugLoc &DL, MCRegister DestReg,
MCRegister SrcReg, bool KillSrc,
bool RenamableDest, bool RenamableSrc) const {
// First deal with the normal symmetric copies.
bool HasAVX = Subtarget.hasAVX();
bool HasVLX = Subtarget.hasVLX();
bool HasEGPR = Subtarget.hasEGPR();
unsigned Opc = 0;
if (X86::GR64RegClass.contains(DestReg, SrcReg))
Opc = X86::MOV64rr;
else if (X86::GR32RegClass.contains(DestReg, SrcReg))
Opc = X86::MOV32rr;
else if (X86::GR16RegClass.contains(DestReg, SrcReg))
Opc = X86::MOV16rr;
else if (X86::GR8RegClass.contains(DestReg, SrcReg)) {
// Copying to or from a physical H register on x86-64 requires a NOREX
// move. Otherwise use a normal move.
if ((isHReg(DestReg) || isHReg(SrcReg)) && Subtarget.is64Bit()) {
Opc = X86::MOV8rr_NOREX;
// Both operands must be encodable without an REX prefix.
assert(X86::GR8_NOREXRegClass.contains(SrcReg, DestReg) &&
"8-bit H register can not be copied outside GR8_NOREX");
} else
Opc = X86::MOV8rr;
} else if (X86::VR64RegClass.contains(DestReg, SrcReg))
Opc = X86::MMX_MOVQ64rr;
else if (X86::VR128XRegClass.contains(DestReg, SrcReg)) {
if (HasVLX)
Opc = X86::VMOVAPSZ128rr;
else if (X86::VR128RegClass.contains(DestReg, SrcReg))
Opc = HasAVX ? X86::VMOVAPSrr : X86::MOVAPSrr;
else {
// If this an extended register and we don't have VLX we need to use a
// 512-bit move.
Opc = X86::VMOVAPSZrr;
const TargetRegisterInfo *TRI = &getRegisterInfo();
DestReg =
TRI->getMatchingSuperReg(DestReg, X86::sub_xmm, &X86::VR512RegClass);
SrcReg =
TRI->getMatchingSuperReg(SrcReg, X86::sub_xmm, &X86::VR512RegClass);
}
} else if (X86::VR256XRegClass.contains(DestReg, SrcReg)) {
if (HasVLX)
Opc = X86::VMOVAPSZ256rr;
else if (X86::VR256RegClass.contains(DestReg, SrcReg))
Opc = X86::VMOVAPSYrr;
else {
// If this an extended register and we don't have VLX we need to use a
// 512-bit move.
Opc = X86::VMOVAPSZrr;
const TargetRegisterInfo *TRI = &getRegisterInfo();
DestReg =
TRI->getMatchingSuperReg(DestReg, X86::sub_ymm, &X86::VR512RegClass);
SrcReg =
TRI->getMatchingSuperReg(SrcReg, X86::sub_ymm, &X86::VR512RegClass);
}
} else if (X86::VR512RegClass.contains(DestReg, SrcReg))
Opc = X86::VMOVAPSZrr;
// All KMASK RegClasses hold the same k registers, can be tested against
// anyone.
else if (X86::VK16RegClass.contains(DestReg, SrcReg))
Opc = Subtarget.hasBWI() ? (HasEGPR ? X86::KMOVQkk_EVEX : X86::KMOVQkk)
: (HasEGPR ? X86::KMOVQkk_EVEX : X86::KMOVWkk);
if (!Opc)
Opc = CopyToFromAsymmetricReg(DestReg, SrcReg, Subtarget);
if (Opc) {
BuildMI(MBB, MI, DL, get(Opc), DestReg)
.addReg(SrcReg, getKillRegState(KillSrc));
return;
}
if (SrcReg == X86::EFLAGS || DestReg == X86::EFLAGS) {
// FIXME: We use a fatal error here because historically LLVM has tried
// lower some of these physreg copies and we want to ensure we get
// reasonable bug reports if someone encounters a case no other testing
// found. This path should be removed after the LLVM 7 release.
report_fatal_error("Unable to copy EFLAGS physical register!");
}
LLVM_DEBUG(dbgs() << "Cannot copy " << RI.getName(SrcReg) << " to "
<< RI.getName(DestReg) << '\n');
report_fatal_error("Cannot emit physreg copy instruction");
}
std::optional<DestSourcePair>
X86InstrInfo::isCopyInstrImpl(const MachineInstr &MI) const {
if (MI.isMoveReg()) {
// FIXME: Dirty hack for apparent invariant that doesn't hold when
// subreg_to_reg is coalesced with ordinary copies, such that the bits that
// were asserted as 0 are now undef.
if (MI.getOperand(0).isUndef() && MI.getOperand(0).getSubReg())
return std::nullopt;
return DestSourcePair{MI.getOperand(0), MI.getOperand(1)};
}
return std::nullopt;
}
static unsigned getLoadStoreOpcodeForFP16(bool Load, const X86Subtarget &STI) {
if (STI.hasFP16())
return Load ? X86::VMOVSHZrm_alt : X86::VMOVSHZmr;
if (Load)
return STI.hasAVX512() ? X86::VMOVSSZrm
: STI.hasAVX() ? X86::VMOVSSrm
: X86::MOVSSrm;
else
return STI.hasAVX512() ? X86::VMOVSSZmr
: STI.hasAVX() ? X86::VMOVSSmr
: X86::MOVSSmr;
}
static unsigned getLoadStoreRegOpcode(Register Reg,
const TargetRegisterClass *RC,
bool IsStackAligned,
const X86Subtarget &STI, bool Load) {
bool HasAVX = STI.hasAVX();
bool HasAVX512 = STI.hasAVX512();
bool HasVLX = STI.hasVLX();
bool HasEGPR = STI.hasEGPR();
assert(RC != nullptr && "Invalid target register class");
switch (STI.getRegisterInfo()->getSpillSize(*RC)) {
default:
llvm_unreachable("Unknown spill size");
case 1:
assert(X86::GR8RegClass.hasSubClassEq(RC) && "Unknown 1-byte regclass");
if (STI.is64Bit())
// Copying to or from a physical H register on x86-64 requires a NOREX
// move. Otherwise use a normal move.
if (isHReg(Reg) || X86::GR8_ABCD_HRegClass.hasSubClassEq(RC))
return Load ? X86::MOV8rm_NOREX : X86::MOV8mr_NOREX;
return Load ? X86::MOV8rm : X86::MOV8mr;
case 2:
if (X86::VK16RegClass.hasSubClassEq(RC))
return Load ? (HasEGPR ? X86::KMOVWkm_EVEX : X86::KMOVWkm)
: (HasEGPR ? X86::KMOVWmk_EVEX : X86::KMOVWmk);
assert(X86::GR16RegClass.hasSubClassEq(RC) && "Unknown 2-byte regclass");
return Load ? X86::MOV16rm : X86::MOV16mr;
case 4:
if (X86::GR32RegClass.hasSubClassEq(RC))
return Load ? X86::MOV32rm : X86::MOV32mr;
if (X86::FR32XRegClass.hasSubClassEq(RC))
return Load ? (HasAVX512 ? X86::VMOVSSZrm_alt
: HasAVX ? X86::VMOVSSrm_alt
: X86::MOVSSrm_alt)
: (HasAVX512 ? X86::VMOVSSZmr
: HasAVX ? X86::VMOVSSmr
: X86::MOVSSmr);
if (X86::RFP32RegClass.hasSubClassEq(RC))
return Load ? X86::LD_Fp32m : X86::ST_Fp32m;
if (X86::VK32RegClass.hasSubClassEq(RC)) {
assert(STI.hasBWI() && "KMOVD requires BWI");
return Load ? (HasEGPR ? X86::KMOVDkm_EVEX : X86::KMOVDkm)
: (HasEGPR ? X86::KMOVDmk_EVEX : X86::KMOVDmk);
}
// All of these mask pair classes have the same spill size, the same kind
// of kmov instructions can be used with all of them.
if (X86::VK1PAIRRegClass.hasSubClassEq(RC) ||
X86::VK2PAIRRegClass.hasSubClassEq(RC) ||
X86::VK4PAIRRegClass.hasSubClassEq(RC) ||
X86::VK8PAIRRegClass.hasSubClassEq(RC) ||
X86::VK16PAIRRegClass.hasSubClassEq(RC))
return Load ? X86::MASKPAIR16LOAD : X86::MASKPAIR16STORE;
if (X86::FR16RegClass.hasSubClassEq(RC) ||
X86::FR16XRegClass.hasSubClassEq(RC))
return getLoadStoreOpcodeForFP16(Load, STI);
llvm_unreachable("Unknown 4-byte regclass");
case 8:
if (X86::GR64RegClass.hasSubClassEq(RC))
return Load ? X86::MOV64rm : X86::MOV64mr;
if (X86::FR64XRegClass.hasSubClassEq(RC))
return Load ? (HasAVX512 ? X86::VMOVSDZrm_alt
: HasAVX ? X86::VMOVSDrm_alt
: X86::MOVSDrm_alt)
: (HasAVX512 ? X86::VMOVSDZmr
: HasAVX ? X86::VMOVSDmr
: X86::MOVSDmr);
if (X86::VR64RegClass.hasSubClassEq(RC))
return Load ? X86::MMX_MOVQ64rm : X86::MMX_MOVQ64mr;
if (X86::RFP64RegClass.hasSubClassEq(RC))
return Load ? X86::LD_Fp64m : X86::ST_Fp64m;
if (X86::VK64RegClass.hasSubClassEq(RC)) {
assert(STI.hasBWI() && "KMOVQ requires BWI");
return Load ? (HasEGPR ? X86::KMOVQkm_EVEX : X86::KMOVQkm)
: (HasEGPR ? X86::KMOVQmk_EVEX : X86::KMOVQmk);
}
llvm_unreachable("Unknown 8-byte regclass");
case 10:
assert(X86::RFP80RegClass.hasSubClassEq(RC) && "Unknown 10-byte regclass");
return Load ? X86::LD_Fp80m : X86::ST_FpP80m;
case 16: {
if (X86::VR128XRegClass.hasSubClassEq(RC)) {
// If stack is realigned we can use aligned stores.
if (IsStackAligned)
return Load ? (HasVLX ? X86::VMOVAPSZ128rm
: HasAVX512 ? X86::VMOVAPSZ128rm_NOVLX
: HasAVX ? X86::VMOVAPSrm
: X86::MOVAPSrm)
: (HasVLX ? X86::VMOVAPSZ128mr
: HasAVX512 ? X86::VMOVAPSZ128mr_NOVLX
: HasAVX ? X86::VMOVAPSmr
: X86::MOVAPSmr);
else
return Load ? (HasVLX ? X86::VMOVUPSZ128rm
: HasAVX512 ? X86::VMOVUPSZ128rm_NOVLX
: HasAVX ? X86::VMOVUPSrm
: X86::MOVUPSrm)
: (HasVLX ? X86::VMOVUPSZ128mr
: HasAVX512 ? X86::VMOVUPSZ128mr_NOVLX
: HasAVX ? X86::VMOVUPSmr
: X86::MOVUPSmr);
}
llvm_unreachable("Unknown 16-byte regclass");
}
case 32:
assert(X86::VR256XRegClass.hasSubClassEq(RC) && "Unknown 32-byte regclass");
// If stack is realigned we can use aligned stores.
if (IsStackAligned)
return Load ? (HasVLX ? X86::VMOVAPSZ256rm
: HasAVX512 ? X86::VMOVAPSZ256rm_NOVLX
: X86::VMOVAPSYrm)
: (HasVLX ? X86::VMOVAPSZ256mr
: HasAVX512 ? X86::VMOVAPSZ256mr_NOVLX
: X86::VMOVAPSYmr);
else
return Load ? (HasVLX ? X86::VMOVUPSZ256rm
: HasAVX512 ? X86::VMOVUPSZ256rm_NOVLX
: X86::VMOVUPSYrm)
: (HasVLX ? X86::VMOVUPSZ256mr
: HasAVX512 ? X86::VMOVUPSZ256mr_NOVLX
: X86::VMOVUPSYmr);
case 64:
assert(X86::VR512RegClass.hasSubClassEq(RC) && "Unknown 64-byte regclass");
assert(STI.hasAVX512() && "Using 512-bit register requires AVX512");
if (IsStackAligned)
return Load ? X86::VMOVAPSZrm : X86::VMOVAPSZmr;
else
return Load ? X86::VMOVUPSZrm : X86::VMOVUPSZmr;
case 1024:
assert(X86::TILERegClass.hasSubClassEq(RC) && "Unknown 1024-byte regclass");
assert(STI.hasAMXTILE() && "Using 8*1024-bit register requires AMX-TILE");
#define GET_EGPR_IF_ENABLED(OPC) (STI.hasEGPR() ? OPC##_EVEX : OPC)
return Load ? GET_EGPR_IF_ENABLED(X86::TILELOADD)
: GET_EGPR_IF_ENABLED(X86::TILESTORED);
#undef GET_EGPR_IF_ENABLED
case 2048:
assert(X86::TILEPAIRRegClass.hasSubClassEq(RC) &&
"Unknown 2048-byte regclass");
assert(STI.hasAMXTILE() && "Using 2048-bit register requires AMX-TILE");
return Load ? X86::PTILEPAIRLOAD : X86::PTILEPAIRSTORE;
}
}
std::optional<ExtAddrMode>
X86InstrInfo::getAddrModeFromMemoryOp(const MachineInstr &MemI,
const TargetRegisterInfo *TRI) const {
const MCInstrDesc &Desc = MemI.getDesc();
int MemRefBegin = X86II::getMemoryOperandNo(Desc.TSFlags);
if (MemRefBegin < 0)
return std::nullopt;
MemRefBegin += X86II::getOperandBias(Desc);
auto &BaseOp = MemI.getOperand(MemRefBegin + X86::AddrBaseReg);
if (!BaseOp.isReg()) // Can be an MO_FrameIndex
return std::nullopt;
const MachineOperand &DispMO = MemI.getOperand(MemRefBegin + X86::AddrDisp);
// Displacement can be symbolic
if (!DispMO.isImm())
return std::nullopt;
ExtAddrMode AM;
AM.BaseReg = BaseOp.getReg();
AM.ScaledReg = MemI.getOperand(MemRefBegin + X86::AddrIndexReg).getReg();
AM.Scale = MemI.getOperand(MemRefBegin + X86::AddrScaleAmt).getImm();
AM.Displacement = DispMO.getImm();
return AM;
}
bool X86InstrInfo::verifyInstruction(const MachineInstr &MI,
StringRef &ErrInfo) const {
std::optional<ExtAddrMode> AMOrNone = getAddrModeFromMemoryOp(MI, nullptr);
if (!AMOrNone)
return true;
ExtAddrMode AM = *AMOrNone;
assert(AM.Form == ExtAddrMode::Formula::Basic);
if (AM.ScaledReg != X86::NoRegister) {
switch (AM.Scale) {
case 1:
case 2:
case 4:
case 8:
break;
default:
ErrInfo = "Scale factor in address must be 1, 2, 4 or 8";
return false;
}
}
if (!isInt<32>(AM.Displacement)) {
ErrInfo = "Displacement in address must fit into 32-bit signed "
"integer";
return false;
}
return true;
}
bool X86InstrInfo::getConstValDefinedInReg(const MachineInstr &MI,
const Register Reg,
int64_t &ImmVal) const {
Register MovReg = Reg;
const MachineInstr *MovMI = &MI;
// Follow use-def for SUBREG_TO_REG to find the real move immediate
// instruction. It is quite common for x86-64.
if (MI.isSubregToReg()) {
// We use following pattern to setup 64b immediate.
// %8:gr32 = MOV32r0 implicit-def dead $eflags
// %6:gr64 = SUBREG_TO_REG 0, killed %8:gr32, %subreg.sub_32bit
if (!MI.getOperand(1).isImm())
return false;
unsigned FillBits = MI.getOperand(1).getImm();
unsigned SubIdx = MI.getOperand(3).getImm();
MovReg = MI.getOperand(2).getReg();
if (SubIdx != X86::sub_32bit || FillBits != 0)
return false;
const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
MovMI = MRI.getUniqueVRegDef(MovReg);
if (!MovMI)
return false;
}
if (MovMI->getOpcode() == X86::MOV32r0 &&
MovMI->getOperand(0).getReg() == MovReg) {
ImmVal = 0;
return true;
}
if (MovMI->getOpcode() != X86::MOV32ri &&
MovMI->getOpcode() != X86::MOV64ri &&
MovMI->getOpcode() != X86::MOV32ri64 && MovMI->getOpcode() != X86::MOV8ri)
return false;
// Mov Src can be a global address.
if (!MovMI->getOperand(1).isImm() || MovMI->getOperand(0).getReg() != MovReg)
return false;
ImmVal = MovMI->getOperand(1).getImm();
return true;
}
bool X86InstrInfo::preservesZeroValueInReg(
const MachineInstr *MI, const Register NullValueReg,
const TargetRegisterInfo *TRI) const {
if (!MI->modifiesRegister(NullValueReg, TRI))
return true;
switch (MI->getOpcode()) {
// Shift right/left of a null unto itself is still a null, i.e. rax = shl rax
// X.
case X86::SHR64ri:
case X86::SHR32ri:
case X86::SHL64ri:
case X86::SHL32ri:
assert(MI->getOperand(0).isDef() && MI->getOperand(1).isUse() &&
"expected for shift opcode!");
return MI->getOperand(0).getReg() == NullValueReg &&
MI->getOperand(1).getReg() == NullValueReg;
// Zero extend of a sub-reg of NullValueReg into itself does not change the
// null value.
case X86::MOV32rr:
return llvm::all_of(MI->operands(), [&](const MachineOperand &MO) {
return TRI->isSubRegisterEq(NullValueReg, MO.getReg());
});
default:
return false;
}
llvm_unreachable("Should be handled above!");
}
bool X86InstrInfo::getMemOperandsWithOffsetWidth(
const MachineInstr &MemOp, SmallVectorImpl<const MachineOperand *> &BaseOps,
int64_t &Offset, bool &OffsetIsScalable, LocationSize &Width,
const TargetRegisterInfo *TRI) const {
const MCInstrDesc &Desc = MemOp.getDesc();
int MemRefBegin = X86II::getMemoryOperandNo(Desc.TSFlags);
if (MemRefBegin < 0)
return false;
MemRefBegin += X86II::getOperandBias(Desc);
const MachineOperand *BaseOp =
&MemOp.getOperand(MemRefBegin + X86::AddrBaseReg);
if (!BaseOp->isReg()) // Can be an MO_FrameIndex
return false;
if (MemOp.getOperand(MemRefBegin + X86::AddrScaleAmt).getImm() != 1)
return false;
if (MemOp.getOperand(MemRefBegin + X86::AddrIndexReg).getReg() !=
X86::NoRegister)
return false;
const MachineOperand &DispMO = MemOp.getOperand(MemRefBegin + X86::AddrDisp);
// Displacement can be symbolic
if (!DispMO.isImm())
return false;
Offset = DispMO.getImm();
if (!BaseOp->isReg())
return false;
OffsetIsScalable = false;
// FIXME: Relying on memoperands() may not be right thing to do here. Check
// with X86 maintainers, and fix it accordingly. For now, it is ok, since
// there is no use of `Width` for X86 back-end at the moment.
Width =
!MemOp.memoperands_empty() ? MemOp.memoperands().front()->getSize() : 0;
BaseOps.push_back(BaseOp);
return true;
}
static unsigned getStoreRegOpcode(Register SrcReg,
const TargetRegisterClass *RC,
bool IsStackAligned,
const X86Subtarget &STI) {
return getLoadStoreRegOpcode(SrcReg, RC, IsStackAligned, STI, false);
}
static unsigned getLoadRegOpcode(Register DestReg,
const TargetRegisterClass *RC,
bool IsStackAligned, const X86Subtarget &STI) {
return getLoadStoreRegOpcode(DestReg, RC, IsStackAligned, STI, true);
}
static bool isAMXOpcode(unsigned Opc) {
switch (Opc) {
default:
return false;
case X86::TILELOADD:
case X86::TILESTORED:
case X86::TILELOADD_EVEX:
case X86::TILESTORED_EVEX:
case X86::PTILEPAIRLOAD:
case X86::PTILEPAIRSTORE:
return true;
}
}
void X86InstrInfo::loadStoreTileReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned Opc, Register Reg, int FrameIdx,
bool isKill) const {
switch (Opc) {
default:
llvm_unreachable("Unexpected special opcode!");
case X86::TILESTORED:
case X86::TILESTORED_EVEX:
case X86::PTILEPAIRSTORE: {
// tilestored %tmm, (%sp, %idx)
MachineRegisterInfo &RegInfo = MBB.getParent()->getRegInfo();
Register VirtReg = RegInfo.createVirtualRegister(&X86::GR64_NOSPRegClass);
BuildMI(MBB, MI, DebugLoc(), get(X86::MOV64ri), VirtReg).addImm(64);
MachineInstr *NewMI =
addFrameReference(BuildMI(MBB, MI, DebugLoc(), get(Opc)), FrameIdx)
.addReg(Reg, getKillRegState(isKill));
MachineOperand &MO = NewMI->getOperand(X86::AddrIndexReg);
MO.setReg(VirtReg);
MO.setIsKill(true);
break;
}
case X86::TILELOADD:
case X86::TILELOADD_EVEX:
case X86::PTILEPAIRLOAD: {
// tileloadd (%sp, %idx), %tmm
MachineRegisterInfo &RegInfo = MBB.getParent()->getRegInfo();
Register VirtReg = RegInfo.createVirtualRegister(&X86::GR64_NOSPRegClass);
BuildMI(MBB, MI, DebugLoc(), get(X86::MOV64ri), VirtReg).addImm(64);
MachineInstr *NewMI = addFrameReference(
BuildMI(MBB, MI, DebugLoc(), get(Opc), Reg), FrameIdx);
MachineOperand &MO = NewMI->getOperand(1 + X86::AddrIndexReg);
MO.setReg(VirtReg);
MO.setIsKill(true);
break;
}
}
}
void X86InstrInfo::storeRegToStackSlot(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register SrcReg,
bool isKill, int FrameIdx, const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI, Register VReg) const {
const MachineFunction &MF = *MBB.getParent();
const MachineFrameInfo &MFI = MF.getFrameInfo();
assert(MFI.getObjectSize(FrameIdx) >= TRI->getSpillSize(*RC) &&
"Stack slot too small for store");
unsigned Alignment = std::max<uint32_t>(TRI->getSpillSize(*RC), 16);
bool isAligned =
(Subtarget.getFrameLowering()->getStackAlign() >= Alignment) ||
(RI.canRealignStack(MF) && !MFI.isFixedObjectIndex(FrameIdx));
unsigned Opc = getStoreRegOpcode(SrcReg, RC, isAligned, Subtarget);
if (isAMXOpcode(Opc))
loadStoreTileReg(MBB, MI, Opc, SrcReg, FrameIdx, isKill);
else
addFrameReference(BuildMI(MBB, MI, DebugLoc(), get(Opc)), FrameIdx)
.addReg(SrcReg, getKillRegState(isKill));
}
void X86InstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
Register DestReg, int FrameIdx,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI,
Register VReg) const {
const MachineFunction &MF = *MBB.getParent();
const MachineFrameInfo &MFI = MF.getFrameInfo();
assert(MFI.getObjectSize(FrameIdx) >= TRI->getSpillSize(*RC) &&
"Load size exceeds stack slot");
unsigned Alignment = std::max<uint32_t>(TRI->getSpillSize(*RC), 16);
bool isAligned =
(Subtarget.getFrameLowering()->getStackAlign() >= Alignment) ||
(RI.canRealignStack(MF) && !MFI.isFixedObjectIndex(FrameIdx));
unsigned Opc = getLoadRegOpcode(DestReg, RC, isAligned, Subtarget);
if (isAMXOpcode(Opc))
loadStoreTileReg(MBB, MI, Opc, DestReg, FrameIdx);
else
addFrameReference(BuildMI(MBB, MI, DebugLoc(), get(Opc), DestReg),
FrameIdx);
}
bool X86InstrInfo::analyzeCompare(const MachineInstr &MI, Register &SrcReg,
Register &SrcReg2, int64_t &CmpMask,
int64_t &CmpValue) const {
switch (MI.getOpcode()) {
default:
break;
case X86::CMP64ri32:
case X86::CMP32ri:
case X86::CMP16ri:
case X86::CMP8ri:
SrcReg = MI.getOperand(0).getReg();
SrcReg2 = 0;
if (MI.getOperand(1).isImm()) {
CmpMask = ~0;
CmpValue = MI.getOperand(1).getImm();
} else {
CmpMask = CmpValue = 0;
}
return true;
// A SUB can be used to perform comparison.
CASE_ND(SUB64rm)
CASE_ND(SUB32rm)
CASE_ND(SUB16rm)
CASE_ND(SUB8rm)
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = 0;
CmpMask = 0;
CmpValue = 0;
return true;
CASE_ND(SUB64rr)
CASE_ND(SUB32rr)
CASE_ND(SUB16rr)
CASE_ND(SUB8rr)
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = MI.getOperand(2).getReg();
CmpMask = 0;
CmpValue = 0;
return true;
CASE_ND(SUB64ri32)
CASE_ND(SUB32ri)
CASE_ND(SUB16ri)
CASE_ND(SUB8ri)
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = 0;
if (MI.getOperand(2).isImm()) {
CmpMask = ~0;
CmpValue = MI.getOperand(2).getImm();
} else {
CmpMask = CmpValue = 0;
}
return true;
case X86::CMP64rr:
case X86::CMP32rr:
case X86::CMP16rr:
case X86::CMP8rr:
SrcReg = MI.getOperand(0).getReg();
SrcReg2 = MI.getOperand(1).getReg();
CmpMask = 0;
CmpValue = 0;
return true;
case X86::TEST8rr:
case X86::TEST16rr:
case X86::TEST32rr:
case X86::TEST64rr:
SrcReg = MI.getOperand(0).getReg();
if (MI.getOperand(1).getReg() != SrcReg)
return false;
// Compare against zero.
SrcReg2 = 0;
CmpMask = ~0;
CmpValue = 0;
return true;
}
return false;
}
bool X86InstrInfo::isRedundantFlagInstr(const MachineInstr &FlagI,
Register SrcReg, Register SrcReg2,
int64_t ImmMask, int64_t ImmValue,
const MachineInstr &OI, bool *IsSwapped,
int64_t *ImmDelta) const {
switch (OI.getOpcode()) {
case X86::CMP64rr:
case X86::CMP32rr:
case X86::CMP16rr:
case X86::CMP8rr:
CASE_ND(SUB64rr)
CASE_ND(SUB32rr)
CASE_ND(SUB16rr)
CASE_ND(SUB8rr) {
Register OISrcReg;
Register OISrcReg2;
int64_t OIMask;
int64_t OIValue;
if (!analyzeCompare(OI, OISrcReg, OISrcReg2, OIMask, OIValue) ||
OIMask != ImmMask || OIValue != ImmValue)
return false;
if (SrcReg == OISrcReg && SrcReg2 == OISrcReg2) {
*IsSwapped = false;
return true;
}
if (SrcReg == OISrcReg2 && SrcReg2 == OISrcReg) {
*IsSwapped = true;
return true;
}
return false;
}
case X86::CMP64ri32:
case X86::CMP32ri:
case X86::CMP16ri:
case X86::CMP8ri:
CASE_ND(SUB64ri32)
CASE_ND(SUB32ri)
CASE_ND(SUB16ri)
CASE_ND(SUB8ri)
case X86::TEST64rr:
case X86::TEST32rr:
case X86::TEST16rr:
case X86::TEST8rr: {
if (ImmMask != 0) {
Register OISrcReg;
Register OISrcReg2;
int64_t OIMask;
int64_t OIValue;
if (analyzeCompare(OI, OISrcReg, OISrcReg2, OIMask, OIValue) &&
SrcReg == OISrcReg && ImmMask == OIMask) {
if (OIValue == ImmValue) {
*ImmDelta = 0;
return true;
} else if (static_cast<uint64_t>(ImmValue) ==
static_cast<uint64_t>(OIValue) - 1) {
*ImmDelta = -1;
return true;
} else if (static_cast<uint64_t>(ImmValue) ==
static_cast<uint64_t>(OIValue) + 1) {
*ImmDelta = 1;
return true;
} else {
return false;
}
}
}
return FlagI.isIdenticalTo(OI);
}
default:
return false;
}
}
/// Check whether the definition can be converted
/// to remove a comparison against zero.
inline static bool isDefConvertible(const MachineInstr &MI, bool &NoSignFlag,
bool &ClearsOverflowFlag) {
NoSignFlag = false;
ClearsOverflowFlag = false;
// "ELF Handling for Thread-Local Storage" specifies that x86-64 GOTTPOFF, and
// i386 GOTNTPOFF/INDNTPOFF relocations can convert an ADD to a LEA during
// Initial Exec to Local Exec relaxation. In these cases, we must not depend
// on the EFLAGS modification of ADD actually happening in the final binary.
if (MI.getOpcode() == X86::ADD64rm || MI.getOpcode() == X86::ADD32rm) {
unsigned Flags = MI.getOperand(5).getTargetFlags();
if (Flags == X86II::MO_GOTTPOFF || Flags == X86II::MO_INDNTPOFF ||
Flags == X86II::MO_GOTNTPOFF)
return false;
}
switch (MI.getOpcode()) {
default:
return false;
// The shift instructions only modify ZF if their shift count is non-zero.
// N.B.: The processor truncates the shift count depending on the encoding.
CASE_ND(SAR8ri)
CASE_ND(SAR16ri)
CASE_ND(SAR32ri)
CASE_ND(SAR64ri)
CASE_ND(SHR8ri)
CASE_ND(SHR16ri)
CASE_ND(SHR32ri)
CASE_ND(SHR64ri)
return getTruncatedShiftCount(MI, 2) != 0;
// Some left shift instructions can be turned into LEA instructions but only
// if their flags aren't used. Avoid transforming such instructions.
CASE_ND(SHL8ri)
CASE_ND(SHL16ri)
CASE_ND(SHL32ri)
CASE_ND(SHL64ri) {
unsigned ShAmt = getTruncatedShiftCount(MI, 2);
if (isTruncatedShiftCountForLEA(ShAmt))
return false;
return ShAmt != 0;
}
CASE_ND(SHRD16rri8)
CASE_ND(SHRD32rri8)
CASE_ND(SHRD64rri8)
CASE_ND(SHLD16rri8)
CASE_ND(SHLD32rri8)
CASE_ND(SHLD64rri8)
return getTruncatedShiftCount(MI, 3) != 0;
CASE_ND(SUB64ri32)
CASE_ND(SUB32ri)
CASE_ND(SUB16ri)
CASE_ND(SUB8ri)
CASE_ND(SUB64rr)
CASE_ND(SUB32rr)
CASE_ND(SUB16rr)
CASE_ND(SUB8rr)
CASE_ND(SUB64rm)
CASE_ND(SUB32rm)
CASE_ND(SUB16rm)
CASE_ND(SUB8rm)
CASE_ND(DEC64r)
CASE_ND(DEC32r)
CASE_ND(DEC16r)
CASE_ND(DEC8r)
CASE_ND(ADD64ri32)
CASE_ND(ADD32ri)
CASE_ND(ADD16ri)
CASE_ND(ADD8ri)
CASE_ND(ADD64rr)
CASE_ND(ADD32rr)
CASE_ND(ADD16rr)
CASE_ND(ADD8rr)
CASE_ND(ADD64rm)
CASE_ND(ADD32rm)
CASE_ND(ADD16rm)
CASE_ND(ADD8rm)
CASE_ND(INC64r)
CASE_ND(INC32r)
CASE_ND(INC16r)
CASE_ND(INC8r)
CASE_ND(ADC64ri32)
CASE_ND(ADC32ri)
CASE_ND(ADC16ri)
CASE_ND(ADC8ri)
CASE_ND(ADC64rr)
CASE_ND(ADC32rr)
CASE_ND(ADC16rr)
CASE_ND(ADC8rr)
CASE_ND(ADC64rm)
CASE_ND(ADC32rm)
CASE_ND(ADC16rm)
CASE_ND(ADC8rm)
CASE_ND(SBB64ri32)
CASE_ND(SBB32ri)
CASE_ND(SBB16ri)
CASE_ND(SBB8ri)
CASE_ND(SBB64rr)
CASE_ND(SBB32rr)
CASE_ND(SBB16rr)
CASE_ND(SBB8rr)
CASE_ND(SBB64rm)
CASE_ND(SBB32rm)
CASE_ND(SBB16rm)
CASE_ND(SBB8rm)
CASE_ND(NEG8r)
CASE_ND(NEG16r)
CASE_ND(NEG32r)
CASE_ND(NEG64r)
case X86::LZCNT16rr:
case X86::LZCNT16rm:
case X86::LZCNT32rr:
case X86::LZCNT32rm:
case X86::LZCNT64rr:
case X86::LZCNT64rm:
case X86::POPCNT16rr:
case X86::POPCNT16rm:
case X86::POPCNT32rr:
case X86::POPCNT32rm:
case X86::POPCNT64rr:
case X86::POPCNT64rm:
case X86::TZCNT16rr:
case X86::TZCNT16rm:
case X86::TZCNT32rr:
case X86::TZCNT32rm:
case X86::TZCNT64rr:
case X86::TZCNT64rm:
return true;
CASE_ND(AND64ri32)
CASE_ND(AND32ri)
CASE_ND(AND16ri)
CASE_ND(AND8ri)
CASE_ND(AND64rr)
CASE_ND(AND32rr)
CASE_ND(AND16rr)
CASE_ND(AND8rr)
CASE_ND(AND64rm)
CASE_ND(AND32rm)
CASE_ND(AND16rm)
CASE_ND(AND8rm)
CASE_ND(XOR64ri32)
CASE_ND(XOR32ri)
CASE_ND(XOR16ri)
CASE_ND(XOR8ri)
CASE_ND(XOR64rr)
CASE_ND(XOR32rr)
CASE_ND(XOR16rr)
CASE_ND(XOR8rr)
CASE_ND(XOR64rm)
CASE_ND(XOR32rm)
CASE_ND(XOR16rm)
CASE_ND(XOR8rm)
CASE_ND(OR64ri32)
CASE_ND(OR32ri)
CASE_ND(OR16ri)
CASE_ND(OR8ri)
CASE_ND(OR64rr)
CASE_ND(OR32rr)
CASE_ND(OR16rr)
CASE_ND(OR8rr)
CASE_ND(OR64rm)
CASE_ND(OR32rm)
CASE_ND(OR16rm)
CASE_ND(OR8rm)
case X86::ANDN32rr:
case X86::ANDN32rm:
case X86::ANDN64rr:
case X86::ANDN64rm:
case X86::BLSI32rr:
case X86::BLSI32rm:
case X86::BLSI64rr:
case X86::BLSI64rm:
case X86::BLSMSK32rr:
case X86::BLSMSK32rm:
case X86::BLSMSK64rr:
case X86::BLSMSK64rm:
case X86::BLSR32rr:
case X86::BLSR32rm:
case X86::BLSR64rr:
case X86::BLSR64rm:
case X86::BLCFILL32rr:
case X86::BLCFILL32rm:
case X86::BLCFILL64rr:
case X86::BLCFILL64rm:
case X86::BLCI32rr:
case X86::BLCI32rm:
case X86::BLCI64rr:
case X86::BLCI64rm:
case X86::BLCIC32rr:
case X86::BLCIC32rm:
case X86::BLCIC64rr:
case X86::BLCIC64rm:
case X86::BLCMSK32rr:
case X86::BLCMSK32rm:
case X86::BLCMSK64rr:
case X86::BLCMSK64rm:
case X86::BLCS32rr:
case X86::BLCS32rm:
case X86::BLCS64rr:
case X86::BLCS64rm:
case X86::BLSFILL32rr:
case X86::BLSFILL32rm:
case X86::BLSFILL64rr:
case X86::BLSFILL64rm:
case X86::BLSIC32rr:
case X86::BLSIC32rm:
case X86::BLSIC64rr:
case X86::BLSIC64rm:
case X86::BZHI32rr:
case X86::BZHI32rm:
case X86::BZHI64rr:
case X86::BZHI64rm:
case X86::T1MSKC32rr:
case X86::T1MSKC32rm:
case X86::T1MSKC64rr:
case X86::T1MSKC64rm:
case X86::TZMSK32rr:
case X86::TZMSK32rm:
case X86::TZMSK64rr:
case X86::TZMSK64rm:
// These instructions clear the overflow flag just like TEST.
// FIXME: These are not the only instructions in this switch that clear the
// overflow flag.
ClearsOverflowFlag = true;
return true;
case X86::BEXTR32rr:
case X86::BEXTR64rr:
case X86::BEXTR32rm:
case X86::BEXTR64rm:
case X86::BEXTRI32ri:
case X86::BEXTRI32mi:
case X86::BEXTRI64ri:
case X86::BEXTRI64mi:
// BEXTR doesn't update the sign flag so we can't use it. It does clear
// the overflow flag, but that's not useful without the sign flag.
NoSignFlag = true;
return true;
}
}
/// Check whether the use can be converted to remove a comparison against zero.
static X86::CondCode isUseDefConvertible(const MachineInstr &MI) {
switch (MI.getOpcode()) {
default:
return X86::COND_INVALID;
CASE_ND(NEG8r)
CASE_ND(NEG16r)
CASE_ND(NEG32r)
CASE_ND(NEG64r)
return X86::COND_AE;
case X86::LZCNT16rr:
case X86::LZCNT32rr:
case X86::LZCNT64rr:
return X86::COND_B;
case X86::POPCNT16rr:
case X86::POPCNT32rr:
case X86::POPCNT64rr:
return X86::COND_E;
case X86::TZCNT16rr:
case X86::TZCNT32rr:
case X86::TZCNT64rr:
return X86::COND_B;
case X86::BSF16rr:
case X86::BSF32rr:
case X86::BSF64rr:
case X86::BSR16rr:
case X86::BSR32rr:
case X86::BSR64rr:
return X86::COND_E;
case X86::BLSI32rr:
case X86::BLSI64rr:
return X86::COND_AE;
case X86::BLSR32rr:
case X86::BLSR64rr:
case X86::BLSMSK32rr:
case X86::BLSMSK64rr:
return X86::COND_B;
// TODO: TBM instructions.
}
}
/// Check if there exists an earlier instruction that
/// operates on the same source operands and sets flags in the same way as
/// Compare; remove Compare if possible.
bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, Register SrcReg,
Register SrcReg2, int64_t CmpMask,
int64_t CmpValue,
const MachineRegisterInfo *MRI) const {
// Check whether we can replace SUB with CMP.
switch (CmpInstr.getOpcode()) {
default:
break;
CASE_ND(SUB64ri32)
CASE_ND(SUB32ri)
CASE_ND(SUB16ri)
CASE_ND(SUB8ri)
CASE_ND(SUB64rm)
CASE_ND(SUB32rm)
CASE_ND(SUB16rm)
CASE_ND(SUB8rm)
CASE_ND(SUB64rr)
CASE_ND(SUB32rr)
CASE_ND(SUB16rr)
CASE_ND(SUB8rr) {
if (!MRI->use_nodbg_empty(CmpInstr.getOperand(0).getReg()))
return false;
// There is no use of the destination register, we can replace SUB with CMP.
unsigned NewOpcode = 0;
#define FROM_TO(A, B) \
CASE_ND(A) NewOpcode = X86::B; \
break;
switch (CmpInstr.getOpcode()) {
default:
llvm_unreachable("Unreachable!");
FROM_TO(SUB64rm, CMP64rm)
FROM_TO(SUB32rm, CMP32rm)
FROM_TO(SUB16rm, CMP16rm)
FROM_TO(SUB8rm, CMP8rm)
FROM_TO(SUB64rr, CMP64rr)
FROM_TO(SUB32rr, CMP32rr)
FROM_TO(SUB16rr, CMP16rr)
FROM_TO(SUB8rr, CMP8rr)
FROM_TO(SUB64ri32, CMP64ri32)
FROM_TO(SUB32ri, CMP32ri)
FROM_TO(SUB16ri, CMP16ri)
FROM_TO(SUB8ri, CMP8ri)
}
#undef FROM_TO
CmpInstr.setDesc(get(NewOpcode));
CmpInstr.removeOperand(0);
// Mutating this instruction invalidates any debug data associated with it.
CmpInstr.dropDebugNumber();
// Fall through to optimize Cmp if Cmp is CMPrr or CMPri.
if (NewOpcode == X86::CMP64rm || NewOpcode == X86::CMP32rm ||
NewOpcode == X86::CMP16rm || NewOpcode == X86::CMP8rm)
return false;
}
}
// The following code tries to remove the comparison by re-using EFLAGS
// from earlier instructions.
bool IsCmpZero = (CmpMask != 0 && CmpValue == 0);
// Transformation currently requires SSA values.
if (SrcReg2.isPhysical())
return false;
MachineInstr *SrcRegDef = MRI->getVRegDef(SrcReg);
assert(SrcRegDef && "Must have a definition (SSA)");
MachineInstr *MI = nullptr;
MachineInstr *Sub = nullptr;
MachineInstr *Movr0Inst = nullptr;
bool NoSignFlag = false;
bool ClearsOverflowFlag = false;
bool ShouldUpdateCC = false;
bool IsSwapped = false;
X86::CondCode NewCC = X86::COND_INVALID;
int64_t ImmDelta = 0;
// Search backward from CmpInstr for the next instruction defining EFLAGS.
const TargetRegisterInfo *TRI = &getRegisterInfo();
MachineBasicBlock &CmpMBB = *CmpInstr.getParent();
MachineBasicBlock::reverse_iterator From =
std::next(MachineBasicBlock::reverse_iterator(CmpInstr));
for (MachineBasicBlock *MBB = &CmpMBB;;) {
for (MachineInstr &Inst : make_range(From, MBB->rend())) {
// Try to use EFLAGS from the instruction defining %SrcReg. Example:
// %eax = addl ...
// ... // EFLAGS not changed
// testl %eax, %eax // <-- can be removed
if (&Inst == SrcRegDef) {
if (IsCmpZero &&
isDefConvertible(Inst, NoSignFlag, ClearsOverflowFlag)) {
MI = &Inst;
break;
}
// Look back for the following pattern, in which case the
// test16rr/test64rr instruction could be erased.
//
// Example for test16rr:
// %reg = and32ri %in_reg, 5
// ... // EFLAGS not changed.
// %src_reg = copy %reg.sub_16bit:gr32
// test16rr %src_reg, %src_reg, implicit-def $eflags
// Example for test64rr:
// %reg = and32ri %in_reg, 5
// ... // EFLAGS not changed.
// %src_reg = subreg_to_reg 0, %reg, %subreg.sub_index
// test64rr %src_reg, %src_reg, implicit-def $eflags
MachineInstr *AndInstr = nullptr;
if (IsCmpZero &&
findRedundantFlagInstr(CmpInstr, Inst, MRI, &AndInstr, TRI,
NoSignFlag, ClearsOverflowFlag)) {
assert(AndInstr != nullptr && X86::isAND(AndInstr->getOpcode()));
MI = AndInstr;
break;
}
// Cannot find other candidates before definition of SrcReg.
return false;
}
if (Inst.modifiesRegister(X86::EFLAGS, TRI)) {
// Try to use EFLAGS produced by an instruction reading %SrcReg.
// Example:
// %eax = ...
// ...
// popcntl %eax
// ... // EFLAGS not changed
// testl %eax, %eax // <-- can be removed
if (IsCmpZero) {
NewCC = isUseDefConvertible(Inst);
if (NewCC != X86::COND_INVALID && Inst.getOperand(1).isReg() &&
Inst.getOperand(1).getReg() == SrcReg) {
ShouldUpdateCC = true;
MI = &Inst;
break;
}
}
// Try to use EFLAGS from an instruction with similar flag results.
// Example:
// sub x, y or cmp x, y
// ... // EFLAGS not changed
// cmp x, y // <-- can be removed
if (isRedundantFlagInstr(CmpInstr, SrcReg, SrcReg2, CmpMask, CmpValue,
Inst, &IsSwapped, &ImmDelta)) {
Sub = &Inst;
break;
}
// MOV32r0 is implemented with xor which clobbers condition code. It is
// safe to move up, if the definition to EFLAGS is dead and earlier
// instructions do not read or write EFLAGS.
if (!Movr0Inst && Inst.getOpcode() == X86::MOV32r0 &&
Inst.registerDefIsDead(X86::EFLAGS, TRI)) {
Movr0Inst = &Inst;
continue;
}
// Cannot do anything for any other EFLAG changes.
return false;
}
}
if (MI || Sub)
break;
// Reached begin of basic block. Continue in predecessor if there is
// exactly one.
if (MBB->pred_size() != 1)
return false;
MBB = *MBB->pred_begin();
From = MBB->rbegin();
}
// Scan forward from the instruction after CmpInstr for uses of EFLAGS.
// It is safe to remove CmpInstr if EFLAGS is redefined or killed.
// If we are done with the basic block, we need to check whether EFLAGS is
// live-out.
bool FlagsMayLiveOut = true;
SmallVector<std::pair<MachineInstr *, X86::CondCode>, 4> OpsToUpdate;
MachineBasicBlock::iterator AfterCmpInstr =
std::next(MachineBasicBlock::iterator(CmpInstr));
for (MachineInstr &Instr : make_range(AfterCmpInstr, CmpMBB.end())) {
bool ModifyEFLAGS = Instr.modifiesRegister(X86::EFLAGS, TRI);
bool UseEFLAGS = Instr.readsRegister(X86::EFLAGS, TRI);
// We should check the usage if this instruction uses and updates EFLAGS.
if (!UseEFLAGS && ModifyEFLAGS) {
// It is safe to remove CmpInstr if EFLAGS is updated again.
FlagsMayLiveOut = false;
break;
}
if (!UseEFLAGS && !ModifyEFLAGS)
continue;
// EFLAGS is used by this instruction.
X86::CondCode OldCC = X86::getCondFromMI(Instr);
if ((MI || IsSwapped || ImmDelta != 0) && OldCC == X86::COND_INVALID)
return false;
X86::CondCode ReplacementCC = X86::COND_INVALID;
if (MI) {
switch (OldCC) {
default:
break;
case X86::COND_A:
case X86::COND_AE:
case X86::COND_B:
case X86::COND_BE:
// CF is used, we can't perform this optimization.
return false;
case X86::COND_G:
case X86::COND_GE:
case X86::COND_L:
case X86::COND_LE:
// If SF is used, but the instruction doesn't update the SF, then we
// can't do the optimization.
if (NoSignFlag)
return false;
[[fallthrough]];
case X86::COND_O:
case X86::COND_NO:
// If OF is used, the instruction needs to clear it like CmpZero does.
if (!ClearsOverflowFlag)
return false;
break;
case X86::COND_S:
case X86::COND_NS:
// If SF is used, but the instruction doesn't update the SF, then we
// can't do the optimization.
if (NoSignFlag)
return false;
break;
}
// If we're updating the condition code check if we have to reverse the
// condition.
if (ShouldUpdateCC)
switch (OldCC) {
default:
return false;
case X86::COND_E:
ReplacementCC = NewCC;
break;
case X86::COND_NE:
ReplacementCC = GetOppositeBranchCondition(NewCC);
break;
}
} else if (IsSwapped) {
// If we have SUB(r1, r2) and CMP(r2, r1), the condition code needs
// to be changed from r2 > r1 to r1 < r2, from r2 < r1 to r1 > r2, etc.
// We swap the condition code and synthesize the new opcode.
ReplacementCC = getSwappedCondition(OldCC);
if (ReplacementCC == X86::COND_INVALID)
return false;
ShouldUpdateCC = true;
} else if (ImmDelta != 0) {
unsigned BitWidth = TRI->getRegSizeInBits(*MRI->getRegClass(SrcReg));
// Shift amount for min/max constants to adjust for 8/16/32 instruction
// sizes.
switch (OldCC) {
case X86::COND_L: // x <s (C + 1) --> x <=s C
if (ImmDelta != 1 || APInt::getSignedMinValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_LE;
break;
case X86::COND_B: // x <u (C + 1) --> x <=u C
if (ImmDelta != 1 || CmpValue == 0)
return false;
ReplacementCC = X86::COND_BE;
break;
case X86::COND_GE: // x >=s (C + 1) --> x >s C
if (ImmDelta != 1 || APInt::getSignedMinValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_G;
break;
case X86::COND_AE: // x >=u (C + 1) --> x >u C
if (ImmDelta != 1 || CmpValue == 0)
return false;
ReplacementCC = X86::COND_A;
break;
case X86::COND_G: // x >s (C - 1) --> x >=s C
if (ImmDelta != -1 || APInt::getSignedMaxValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_GE;
break;
case X86::COND_A: // x >u (C - 1) --> x >=u C
if (ImmDelta != -1 || APInt::getMaxValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_AE;
break;
case X86::COND_LE: // x <=s (C - 1) --> x <s C
if (ImmDelta != -1 || APInt::getSignedMaxValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_L;
break;
case X86::COND_BE: // x <=u (C - 1) --> x <u C
if (ImmDelta != -1 || APInt::getMaxValue(BitWidth) == CmpValue)
return false;
ReplacementCC = X86::COND_B;
break;
default:
return false;
}
ShouldUpdateCC = true;
}
if (ShouldUpdateCC && ReplacementCC != OldCC) {
// Push the MachineInstr to OpsToUpdate.
// If it is safe to remove CmpInstr, the condition code of these
// instructions will be modified.
OpsToUpdate.push_back(std::make_pair(&Instr, ReplacementCC));
}
if (ModifyEFLAGS || Instr.killsRegister(X86::EFLAGS, TRI)) {
// It is safe to remove CmpInstr if EFLAGS is updated again or killed.
FlagsMayLiveOut = false;
break;
}
}
// If we have to update users but EFLAGS is live-out abort, since we cannot
// easily find all of the users.
if ((MI != nullptr || ShouldUpdateCC) && FlagsMayLiveOut) {
for (MachineBasicBlock *Successor : CmpMBB.successors())
if (Successor->isLiveIn(X86::EFLAGS))
return false;
}
// The instruction to be updated is either Sub or MI.
assert((MI == nullptr || Sub == nullptr) && "Should not have Sub and MI set");
Sub = MI != nullptr ? MI : Sub;
MachineBasicBlock *SubBB = Sub->getParent();
// Move Movr0Inst to the appropriate place before Sub.
if (Movr0Inst) {
// Only move within the same block so we don't accidentally move to a
// block with higher execution frequency.
if (&CmpMBB != SubBB)
return false;
// Look backwards until we find a def that doesn't use the current EFLAGS.
MachineBasicBlock::reverse_iterator InsertI = Sub,
InsertE = Sub->getParent()->rend();
for (; InsertI != InsertE; ++InsertI) {
MachineInstr *Instr = &*InsertI;
if (!Instr->readsRegister(X86::EFLAGS, TRI) &&
Instr->modifiesRegister(X86::EFLAGS, TRI)) {
Movr0Inst->getParent()->remove(Movr0Inst);
Instr->getParent()->insert(MachineBasicBlock::iterator(Instr),
Movr0Inst);
break;
}
}
if (InsertI == InsertE)
return false;
}
// Make sure Sub instruction defines EFLAGS and mark the def live.
MachineOperand *FlagDef =
Sub->findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
assert(FlagDef && "Unable to locate a def EFLAGS operand");
FlagDef->setIsDead(false);
CmpInstr.eraseFromParent();
// Modify the condition code of instructions in OpsToUpdate.
for (auto &Op : OpsToUpdate) {
Op.first->getOperand(Op.first->getDesc().getNumOperands() - 1)
.setImm(Op.second);
}
// Add EFLAGS to block live-ins between CmpBB and block of flags producer.
for (MachineBasicBlock *MBB = &CmpMBB; MBB != SubBB;
MBB = *MBB->pred_begin()) {
assert(MBB->pred_size() == 1 && "Expected exactly one predecessor");
if (!MBB->isLiveIn(X86::EFLAGS))
MBB->addLiveIn(X86::EFLAGS);
}
return true;
}
/// Try to remove the load by folding it to a register
/// operand at the use. We fold the load instructions if load defines a virtual
/// register, the virtual register is used once in the same BB, and the
/// instructions in-between do not load or store, and have no side effects.
MachineInstr *X86InstrInfo::optimizeLoadInstr(MachineInstr &MI,
const MachineRegisterInfo *MRI,
Register &FoldAsLoadDefReg,
MachineInstr *&DefMI) const {
// Check whether we can move DefMI here.
DefMI = MRI->getVRegDef(FoldAsLoadDefReg);
assert(DefMI);
bool SawStore = false;
if (!DefMI->isSafeToMove(SawStore))
return nullptr;
// Collect information about virtual register operands of MI.
SmallVector<unsigned, 1> SrcOperandIds;
for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI.getOperand(i);
if (!MO.isReg())
continue;
Register Reg = MO.getReg();
if (Reg != FoldAsLoadDefReg)
continue;
// Do not fold if we have a subreg use or a def.
if (MO.getSubReg() || MO.isDef())
return nullptr;
SrcOperandIds.push_back(i);
}
if (SrcOperandIds.empty())
return nullptr;
// Check whether we can fold the def into SrcOperandId.
if (MachineInstr *FoldMI = foldMemoryOperand(MI, SrcOperandIds, *DefMI)) {
FoldAsLoadDefReg = 0;
return FoldMI;
}
return nullptr;
}
/// \returns true if the instruction can be changed to COPY when imm is 0.
static bool canConvert2Copy(unsigned Opc) {
switch (Opc) {
default:
return false;
CASE_ND(ADD64ri32)
CASE_ND(SUB64ri32)
CASE_ND(OR64ri32)
CASE_ND(XOR64ri32)
CASE_ND(ADD32ri)
CASE_ND(SUB32ri)
CASE_ND(OR32ri)
CASE_ND(XOR32ri)
return true;
}
}
/// Convert an ALUrr opcode to corresponding ALUri opcode. Such as
/// ADD32rr ==> ADD32ri
static unsigned convertALUrr2ALUri(unsigned Opc) {
switch (Opc) {
default:
return 0;
#define FROM_TO(FROM, TO) \
case X86::FROM: \
return X86::TO; \
case X86::FROM##_ND: \
return X86::TO##_ND;
FROM_TO(ADD64rr, ADD64ri32)
FROM_TO(ADC64rr, ADC64ri32)
FROM_TO(SUB64rr, SUB64ri32)
FROM_TO(SBB64rr, SBB64ri32)
FROM_TO(AND64rr, AND64ri32)
FROM_TO(OR64rr, OR64ri32)
FROM_TO(XOR64rr, XOR64ri32)
FROM_TO(SHR64rCL, SHR64ri)
FROM_TO(SHL64rCL, SHL64ri)
FROM_TO(SAR64rCL, SAR64ri)
FROM_TO(ROL64rCL, ROL64ri)
FROM_TO(ROR64rCL, ROR64ri)
FROM_TO(RCL64rCL, RCL64ri)
FROM_TO(RCR64rCL, RCR64ri)
FROM_TO(ADD32rr, ADD32ri)
FROM_TO(ADC32rr, ADC32ri)
FROM_TO(SUB32rr, SUB32ri)
FROM_TO(SBB32rr, SBB32ri)
FROM_TO(AND32rr, AND32ri)
FROM_TO(OR32rr, OR32ri)
FROM_TO(XOR32rr, XOR32ri)
FROM_TO(SHR32rCL, SHR32ri)
FROM_TO(SHL32rCL, SHL32ri)
FROM_TO(SAR32rCL, SAR32ri)
FROM_TO(ROL32rCL, ROL32ri)
FROM_TO(ROR32rCL, ROR32ri)
FROM_TO(RCL32rCL, RCL32ri)
FROM_TO(RCR32rCL, RCR32ri)
#undef FROM_TO
#define FROM_TO(FROM, TO) \
case X86::FROM: \
return X86::TO;
FROM_TO(TEST64rr, TEST64ri32)
FROM_TO(CTEST64rr, CTEST64ri32)
FROM_TO(CMP64rr, CMP64ri32)
FROM_TO(CCMP64rr, CCMP64ri32)
FROM_TO(TEST32rr, TEST32ri)
FROM_TO(CTEST32rr, CTEST32ri)
FROM_TO(CMP32rr, CMP32ri)
FROM_TO(CCMP32rr, CCMP32ri)
#undef FROM_TO
}
}
/// Reg is assigned ImmVal in DefMI, and is used in UseMI.
/// If MakeChange is true, this function tries to replace Reg by ImmVal in
/// UseMI. If MakeChange is false, just check if folding is possible.
//
/// \returns true if folding is successful or possible.
bool X86InstrInfo::foldImmediateImpl(MachineInstr &UseMI, MachineInstr *DefMI,
Register Reg, int64_t ImmVal,
MachineRegisterInfo *MRI,
bool MakeChange) const {
bool Modified = false;
// 64 bit operations accept sign extended 32 bit immediates.
// 32 bit operations accept all 32 bit immediates, so we don't need to check
// them.
const TargetRegisterClass *RC = nullptr;
if (Reg.isVirtual())
RC = MRI->getRegClass(Reg);
if ((Reg.isPhysical() && X86::GR64RegClass.contains(Reg)) ||
(Reg.isVirtual() && X86::GR64RegClass.hasSubClassEq(RC))) {
if (!isInt<32>(ImmVal))
return false;
}
if (UseMI.findRegisterUseOperand(Reg, /*TRI=*/nullptr)->getSubReg())
return false;
// Immediate has larger code size than register. So avoid folding the
// immediate if it has more than 1 use and we are optimizing for size.
if (UseMI.getMF()->getFunction().hasOptSize() && Reg.isVirtual() &&
!MRI->hasOneNonDBGUse(Reg))
return false;
unsigned Opc = UseMI.getOpcode();
unsigned NewOpc;
if (Opc == TargetOpcode::COPY) {
Register ToReg = UseMI.getOperand(0).getReg();
const TargetRegisterClass *RC = nullptr;
if (ToReg.isVirtual())
RC = MRI->getRegClass(ToReg);
bool GR32Reg = (ToReg.isVirtual() && X86::GR32RegClass.hasSubClassEq(RC)) ||
(ToReg.isPhysical() && X86::GR32RegClass.contains(ToReg));
bool GR64Reg = (ToReg.isVirtual() && X86::GR64RegClass.hasSubClassEq(RC)) ||
(ToReg.isPhysical() && X86::GR64RegClass.contains(ToReg));
bool GR8Reg = (ToReg.isVirtual() && X86::GR8RegClass.hasSubClassEq(RC)) ||
(ToReg.isPhysical() && X86::GR8RegClass.contains(ToReg));
if (ImmVal == 0) {
// We have MOV32r0 only.
if (!GR32Reg)
return false;
}
if (GR64Reg) {
if (isUInt<32>(ImmVal))
NewOpc = X86::MOV32ri64;
else
NewOpc = X86::MOV64ri;
} else if (GR32Reg) {
NewOpc = X86::MOV32ri;
if (ImmVal == 0) {
// MOV32r0 clobbers EFLAGS.
const TargetRegisterInfo *TRI = &getRegisterInfo();
if (UseMI.getParent()->computeRegisterLiveness(
TRI, X86::EFLAGS, UseMI) != MachineBasicBlock::LQR_Dead)
return false;
// MOV32r0 is different than other cases because it doesn't encode the
// immediate in the instruction. So we directly modify it here.
if (!MakeChange)
return true;
UseMI.setDesc(get(X86::MOV32r0));
UseMI.removeOperand(
UseMI.findRegisterUseOperandIdx(Reg, /*TRI=*/nullptr));
UseMI.addOperand(MachineOperand::CreateReg(X86::EFLAGS, /*isDef=*/true,
/*isImp=*/true,
/*isKill=*/false,
/*isDead=*/true));
Modified = true;
}
} else if (GR8Reg)
NewOpc = X86::MOV8ri;
else
return false;
} else
NewOpc = convertALUrr2ALUri(Opc);
if (!NewOpc)
return false;
// For SUB instructions the immediate can only be the second source operand.
if ((NewOpc == X86::SUB64ri32 || NewOpc == X86::SUB32ri ||
NewOpc == X86::SBB64ri32 || NewOpc == X86::SBB32ri ||
NewOpc == X86::SUB64ri32_ND || NewOpc == X86::SUB32ri_ND ||
NewOpc == X86::SBB64ri32_ND || NewOpc == X86::SBB32ri_ND) &&
UseMI.findRegisterUseOperandIdx(Reg, /*TRI=*/nullptr) != 2)
return false;
// For CMP instructions the immediate can only be at index 1.
if (((NewOpc == X86::CMP64ri32 || NewOpc == X86::CMP32ri) ||
(NewOpc == X86::CCMP64ri32 || NewOpc == X86::CCMP32ri)) &&
UseMI.findRegisterUseOperandIdx(Reg, /*TRI=*/nullptr) != 1)
return false;
using namespace X86;
if (isSHL(Opc) || isSHR(Opc) || isSAR(Opc) || isROL(Opc) || isROR(Opc) ||
isRCL(Opc) || isRCR(Opc)) {
unsigned RegIdx = UseMI.findRegisterUseOperandIdx(Reg, /*TRI=*/nullptr);
if (RegIdx < 2)
return false;
if (!isInt<8>(ImmVal))
return false;
assert(Reg == X86::CL);
if (!MakeChange)
return true;
UseMI.setDesc(get(NewOpc));
UseMI.removeOperand(RegIdx);
UseMI.addOperand(MachineOperand::CreateImm(ImmVal));
// Reg is physical register $cl, so we don't know if DefMI is dead through
// MRI. Let the caller handle it, or pass dead-mi-elimination can delete
// the dead physical register define instruction.
return true;
}
if (!MakeChange)
return true;
if (!Modified) {
// Modify the instruction.
if (ImmVal == 0 && canConvert2Copy(NewOpc) &&
UseMI.registerDefIsDead(X86::EFLAGS, /*TRI=*/nullptr)) {
// %100 = add %101, 0
// ==>
// %100 = COPY %101
UseMI.setDesc(get(TargetOpcode::COPY));
UseMI.removeOperand(
UseMI.findRegisterUseOperandIdx(Reg, /*TRI=*/nullptr));
UseMI.removeOperand(
UseMI.findRegisterDefOperandIdx(X86::EFLAGS, /*TRI=*/nullptr));
UseMI.untieRegOperand(0);
UseMI.clearFlag(MachineInstr::MIFlag::NoSWrap);
UseMI.clearFlag(MachineInstr::MIFlag::NoUWrap);
} else {
unsigned Op1 = 1, Op2 = CommuteAnyOperandIndex;
unsigned ImmOpNum = 2;
if (!UseMI.getOperand(0).isDef()) {
Op1 = 0; // TEST, CMP, CTEST, CCMP
ImmOpNum = 1;
}
if (Opc == TargetOpcode::COPY)
ImmOpNum = 1;
if (findCommutedOpIndices(UseMI, Op1, Op2) &&
UseMI.getOperand(Op1).getReg() == Reg)
commuteInstruction(UseMI);
assert(UseMI.getOperand(ImmOpNum).getReg() == Reg);
UseMI.setDesc(get(NewOpc));
UseMI.getOperand(ImmOpNum).ChangeToImmediate(ImmVal);
}
}
if (Reg.isVirtual() && MRI->use_nodbg_empty(Reg))
DefMI->eraseFromBundle();
return true;
}
/// foldImmediate - 'Reg' is known to be defined by a move immediate
/// instruction, try to fold the immediate into the use instruction.
bool X86InstrInfo::foldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
Register Reg, MachineRegisterInfo *MRI) const {
int64_t ImmVal;
if (!getConstValDefinedInReg(DefMI, Reg, ImmVal))
return false;
return foldImmediateImpl(UseMI, &DefMI, Reg, ImmVal, MRI, true);
}
/// Expand a single-def pseudo instruction to a two-addr
/// instruction with two undef reads of the register being defined.
/// This is used for mapping:
/// %xmm4 = V_SET0
/// to:
/// %xmm4 = PXORrr undef %xmm4, undef %xmm4
///
static bool Expand2AddrUndef(MachineInstrBuilder &MIB,
const MCInstrDesc &Desc) {
assert(Desc.getNumOperands() == 3 && "Expected two-addr instruction.");
Register Reg = MIB.getReg(0);
MIB->setDesc(Desc);
// MachineInstr::addOperand() will insert explicit operands before any
// implicit operands.
MIB.addReg(Reg, RegState::Undef).addReg(Reg, RegState::Undef);
// But we don't trust that.
assert(MIB.getReg(1) == Reg && MIB.getReg(2) == Reg && "Misplaced operand");
return true;
}
/// Expand a single-def pseudo instruction to a two-addr
/// instruction with two %k0 reads.
/// This is used for mapping:
/// %k4 = K_SET1
/// to:
/// %k4 = KXNORrr %k0, %k0
static bool Expand2AddrKreg(MachineInstrBuilder &MIB, const MCInstrDesc &Desc,
Register Reg) {
assert(Desc.getNumOperands() == 3 && "Expected two-addr instruction.");
MIB->setDesc(Desc);
MIB.addReg(Reg, RegState::Undef).addReg(Reg, RegState::Undef);
return true;
}
static bool expandMOV32r1(MachineInstrBuilder &MIB, const TargetInstrInfo &TII,
bool MinusOne) {
MachineBasicBlock &MBB = *MIB->getParent();
const DebugLoc &DL = MIB->getDebugLoc();
Register Reg = MIB.getReg(0);
// Insert the XOR.
BuildMI(MBB, MIB.getInstr(), DL, TII.get(X86::XOR32rr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
// Turn the pseudo into an INC or DEC.
MIB->setDesc(TII.get(MinusOne ? X86::DEC32r : X86::INC32r));
MIB.addReg(Reg);
return true;
}
static bool ExpandMOVImmSExti8(MachineInstrBuilder &MIB,
const TargetInstrInfo &TII,
const X86Subtarget &Subtarget) {
MachineBasicBlock &MBB = *MIB->getParent();
const DebugLoc &DL = MIB->getDebugLoc();
int64_t Imm = MIB->getOperand(1).getImm();
assert(Imm != 0 && "Using push/pop for 0 is not efficient.");
MachineBasicBlock::iterator I = MIB.getInstr();
int StackAdjustment;
if (Subtarget.is64Bit()) {
assert(MIB->getOpcode() == X86::MOV64ImmSExti8 ||
MIB->getOpcode() == X86::MOV32ImmSExti8);
// Can't use push/pop lowering if the function might write to the red zone.
X86MachineFunctionInfo *X86FI =
MBB.getParent()->getInfo<X86MachineFunctionInfo>();
if (X86FI->getUsesRedZone()) {
MIB->setDesc(TII.get(MIB->getOpcode() == X86::MOV32ImmSExti8
? X86::MOV32ri
: X86::MOV64ri));
return true;
}
// 64-bit mode doesn't have 32-bit push/pop, so use 64-bit operations and
// widen the register if necessary.
StackAdjustment = 8;
BuildMI(MBB, I, DL, TII.get(X86::PUSH64i32)).addImm(Imm);
MIB->setDesc(TII.get(X86::POP64r));
MIB->getOperand(0).setReg(getX86SubSuperRegister(MIB.getReg(0), 64));
} else {
assert(MIB->getOpcode() == X86::MOV32ImmSExti8);
StackAdjustment = 4;
BuildMI(MBB, I, DL, TII.get(X86::PUSH32i)).addImm(Imm);
MIB->setDesc(TII.get(X86::POP32r));
}
MIB->removeOperand(1);
MIB->addImplicitDefUseOperands(*MBB.getParent());
// Build CFI if necessary.
MachineFunction &MF = *MBB.getParent();
const X86FrameLowering *TFL = Subtarget.getFrameLowering();
bool IsWin64Prologue = MF.getTarget().getMCAsmInfo()->usesWindowsCFI();
bool NeedsDwarfCFI = !IsWin64Prologue && MF.needsFrameMoves();
bool EmitCFI = !TFL->hasFP(MF) && NeedsDwarfCFI;
if (EmitCFI) {
TFL->BuildCFI(
MBB, I, DL,
MCCFIInstruction::createAdjustCfaOffset(nullptr, StackAdjustment));
TFL->BuildCFI(
MBB, std::next(I), DL,
MCCFIInstruction::createAdjustCfaOffset(nullptr, -StackAdjustment));
}
return true;
}
// LoadStackGuard has so far only been implemented for 64-bit MachO. Different
// code sequence is needed for other targets.
static void expandLoadStackGuard(MachineInstrBuilder &MIB,
const TargetInstrInfo &TII) {
MachineBasicBlock &MBB = *MIB->getParent();
const DebugLoc &DL = MIB->getDebugLoc();
Register Reg = MIB.getReg(0);
const GlobalValue *GV =
cast<GlobalValue>((*MIB->memoperands_begin())->getValue());
auto Flags = MachineMemOperand::MOLoad |
MachineMemOperand::MODereferenceable |
MachineMemOperand::MOInvariant;
MachineMemOperand *MMO = MBB.getParent()->getMachineMemOperand(
MachinePointerInfo::getGOT(*MBB.getParent()), Flags, 8, Align(8));
MachineBasicBlock::iterator I = MIB.getInstr();
BuildMI(MBB, I, DL, TII.get(X86::MOV64rm), Reg)
.addReg(X86::RIP)
.addImm(1)
.addReg(0)
.addGlobalAddress(GV, 0, X86II::MO_GOTPCREL)
.addReg(0)
.addMemOperand(MMO);
MIB->setDebugLoc(DL);
MIB->setDesc(TII.get(X86::MOV64rm));
MIB.addReg(Reg, RegState::Kill).addImm(1).addReg(0).addImm(0).addReg(0);
}
static bool expandXorFP(MachineInstrBuilder &MIB, const TargetInstrInfo &TII) {
MachineBasicBlock &MBB = *MIB->getParent();
MachineFunction &MF = *MBB.getParent();
const X86Subtarget &Subtarget = MF.getSubtarget<X86Subtarget>();
const X86RegisterInfo *TRI = Subtarget.getRegisterInfo();
unsigned XorOp =
MIB->getOpcode() == X86::XOR64_FP ? X86::XOR64rr : X86::XOR32rr;
MIB->setDesc(TII.get(XorOp));
MIB.addReg(TRI->getFrameRegister(MF), RegState::Undef);
return true;
}
// This is used to handle spills for 128/256-bit registers when we have AVX512,
// but not VLX. If it uses an extended register we need to use an instruction
// that loads the lower 128/256-bit, but is available with only AVX512F.
static bool expandNOVLXLoad(MachineInstrBuilder &MIB,
const TargetRegisterInfo *TRI,
const MCInstrDesc &LoadDesc,
const MCInstrDesc &BroadcastDesc, unsigned SubIdx) {
Register DestReg = MIB.getReg(0);
// Check if DestReg is XMM16-31 or YMM16-31.
if (TRI->getEncodingValue(DestReg) < 16) {
// We can use a normal VEX encoded load.
MIB->setDesc(LoadDesc);
} else {
// Use a 128/256-bit VBROADCAST instruction.
MIB->setDesc(BroadcastDesc);
// Change the destination to a 512-bit register.
DestReg = TRI->getMatchingSuperReg(DestReg, SubIdx, &X86::VR512RegClass);
MIB->getOperand(0).setReg(DestReg);
}
return true;
}
// This is used to handle spills for 128/256-bit registers when we have AVX512,
// but not VLX. If it uses an extended register we need to use an instruction
// that stores the lower 128/256-bit, but is available with only AVX512F.
static bool expandNOVLXStore(MachineInstrBuilder &MIB,
const TargetRegisterInfo *TRI,
const MCInstrDesc &StoreDesc,
const MCInstrDesc &ExtractDesc, unsigned SubIdx) {
Register SrcReg = MIB.getReg(X86::AddrNumOperands);
// Check if DestReg is XMM16-31 or YMM16-31.
if (TRI->getEncodingValue(SrcReg) < 16) {
// We can use a normal VEX encoded store.
MIB->setDesc(StoreDesc);
} else {
// Use a VEXTRACTF instruction.
MIB->setDesc(ExtractDesc);
// Change the destination to a 512-bit register.
SrcReg = TRI->getMatchingSuperReg(SrcReg, SubIdx, &X86::VR512RegClass);
MIB->getOperand(X86::AddrNumOperands).setReg(SrcReg);
MIB.addImm(0x0); // Append immediate to extract from the lower bits.
}
return true;
}
static bool expandSHXDROT(MachineInstrBuilder &MIB, const MCInstrDesc &Desc) {
MIB->setDesc(Desc);
int64_t ShiftAmt = MIB->getOperand(2).getImm();
// Temporarily remove the immediate so we can add another source register.
MIB->removeOperand(2);
// Add the register. Don't copy the kill flag if there is one.
MIB.addReg(MIB.getReg(1), getUndefRegState(MIB->getOperand(1).isUndef()));
// Add back the immediate.
MIB.addImm(ShiftAmt);
return true;
}
bool X86InstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
bool HasAVX = Subtarget.hasAVX();
MachineInstrBuilder MIB(*MI.getParent()->getParent(), MI);
switch (MI.getOpcode()) {
case X86::MOV32r0:
return Expand2AddrUndef(MIB, get(X86::XOR32rr));
case X86::MOV32r1:
return expandMOV32r1(MIB, *this, /*MinusOne=*/false);
case X86::MOV32r_1:
return expandMOV32r1(MIB, *this, /*MinusOne=*/true);
case X86::MOV32ImmSExti8:
case X86::MOV64ImmSExti8:
return ExpandMOVImmSExti8(MIB, *this, Subtarget);
case X86::SETB_C32r:
return Expand2AddrUndef(MIB, get(X86::SBB32rr));
case X86::SETB_C64r:
return Expand2AddrUndef(MIB, get(X86::SBB64rr));
case X86::MMX_SET0:
return Expand2AddrUndef(MIB, get(X86::MMX_PXORrr));
case X86::V_SET0:
case X86::FsFLD0SS:
case X86::FsFLD0SD:
case X86::FsFLD0SH:
case X86::FsFLD0F128:
return Expand2AddrUndef(MIB, get(HasAVX ? X86::VXORPSrr : X86::XORPSrr));
case X86::AVX_SET0: {
assert(HasAVX && "AVX not supported");
const TargetRegisterInfo *TRI = &getRegisterInfo();
Register SrcReg = MIB.getReg(0);
Register XReg = TRI->getSubReg(SrcReg, X86::sub_xmm);
MIB->getOperand(0).setReg(XReg);
Expand2AddrUndef(MIB, get(X86::VXORPSrr));
MIB.addReg(SrcReg, RegState::ImplicitDefine);
return true;
}
case X86::AVX512_128_SET0:
case X86::AVX512_FsFLD0SH:
case X86::AVX512_FsFLD0SS:
case X86::AVX512_FsFLD0SD:
case X86::AVX512_FsFLD0F128: {
bool HasVLX = Subtarget.hasVLX();
Register SrcReg = MIB.getReg(0);
const TargetRegisterInfo *TRI = &getRegisterInfo();
if (HasVLX || TRI->getEncodingValue(SrcReg) < 16)
return Expand2AddrUndef(MIB,
get(HasVLX ? X86::VPXORDZ128rr : X86::VXORPSrr));
// Extended register without VLX. Use a larger XOR.
SrcReg =
TRI->getMatchingSuperReg(SrcReg, X86::sub_xmm, &X86::VR512RegClass);
MIB->getOperand(0).setReg(SrcReg);
return Expand2AddrUndef(MIB, get(X86::VPXORDZrr));
}
case X86::AVX512_256_SET0:
case X86::AVX512_512_SET0: {
bool HasVLX = Subtarget.hasVLX();
Register SrcReg = MIB.getReg(0);
const TargetRegisterInfo *TRI = &getRegisterInfo();
if (HasVLX || TRI->getEncodingValue(SrcReg) < 16) {
Register XReg = TRI->getSubReg(SrcReg, X86::sub_xmm);
MIB->getOperand(0).setReg(XReg);
Expand2AddrUndef(MIB, get(HasVLX ? X86::VPXORDZ128rr : X86::VXORPSrr));
MIB.addReg(SrcReg, RegState::ImplicitDefine);
return true;
}
if (MI.getOpcode() == X86::AVX512_256_SET0) {
// No VLX so we must reference a zmm.
unsigned ZReg =
TRI->getMatchingSuperReg(SrcReg, X86::sub_ymm, &X86::VR512RegClass);
MIB->getOperand(0).setReg(ZReg);
}
return Expand2AddrUndef(MIB, get(X86::VPXORDZrr));
}
case X86::V_SETALLONES:
return Expand2AddrUndef(MIB,
get(HasAVX ? X86::VPCMPEQDrr : X86::PCMPEQDrr));
case X86::AVX2_SETALLONES:
return Expand2AddrUndef(MIB, get(X86::VPCMPEQDYrr));
case X86::AVX1_SETALLONES: {
Register Reg = MIB.getReg(0);
// VCMPPSYrri with an immediate 0xf should produce VCMPTRUEPS.
MIB->setDesc(get(X86::VCMPPSYrri));
MIB.addReg(Reg, RegState::Undef).addReg(Reg, RegState::Undef).addImm(0xf);
return true;
}
case X86::AVX512_512_SETALLONES: {
Register Reg = MIB.getReg(0);
MIB->setDesc(get(X86::VPTERNLOGDZrri));
// VPTERNLOGD needs 3 register inputs and an immediate.
// 0xff will return 1s for any input.
MIB.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef)
.addImm(0xff);
return true;
}
case X86::AVX512_512_SEXT_MASK_32:
case X86::AVX512_512_SEXT_MASK_64: {
Register Reg = MIB.getReg(0);
Register MaskReg = MIB.getReg(1);
unsigned MaskState = getRegState(MIB->getOperand(1));
unsigned Opc = (MI.getOpcode() == X86::AVX512_512_SEXT_MASK_64)
? X86::VPTERNLOGQZrrikz
: X86::VPTERNLOGDZrrikz;
MI.removeOperand(1);
MIB->setDesc(get(Opc));
// VPTERNLOG needs 3 register inputs and an immediate.
// 0xff will return 1s for any input.
MIB.addReg(Reg, RegState::Undef)
.addReg(MaskReg, MaskState)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef)
.addImm(0xff);
return true;
}
case X86::VMOVAPSZ128rm_NOVLX:
return expandNOVLXLoad(MIB, &getRegisterInfo(), get(X86::VMOVAPSrm),
get(X86::VBROADCASTF32X4Zrm), X86::sub_xmm);
case X86::VMOVUPSZ128rm_NOVLX:
return expandNOVLXLoad(MIB, &getRegisterInfo(), get(X86::VMOVUPSrm),
get(X86::VBROADCASTF32X4Zrm), X86::sub_xmm);
case X86::VMOVAPSZ256rm_NOVLX:
return expandNOVLXLoad(MIB, &getRegisterInfo(), get(X86::VMOVAPSYrm),
get(X86::VBROADCASTF64X4Zrm), X86::sub_ymm);
case X86::VMOVUPSZ256rm_NOVLX:
return expandNOVLXLoad(MIB, &getRegisterInfo(), get(X86::VMOVUPSYrm),
get(X86::VBROADCASTF64X4Zrm), X86::sub_ymm);
case X86::VMOVAPSZ128mr_NOVLX:
return expandNOVLXStore(MIB, &getRegisterInfo(), get(X86::VMOVAPSmr),
get(X86::VEXTRACTF32X4Zmri), X86::sub_xmm);
case X86::VMOVUPSZ128mr_NOVLX:
return expandNOVLXStore(MIB, &getRegisterInfo(), get(X86::VMOVUPSmr),
get(X86::VEXTRACTF32X4Zmri), X86::sub_xmm);
case X86::VMOVAPSZ256mr_NOVLX:
return expandNOVLXStore(MIB, &getRegisterInfo(), get(X86::VMOVAPSYmr),
get(X86::VEXTRACTF64X4Zmri), X86::sub_ymm);
case X86::VMOVUPSZ256mr_NOVLX:
return expandNOVLXStore(MIB, &getRegisterInfo(), get(X86::VMOVUPSYmr),
get(X86::VEXTRACTF64X4Zmri), X86::sub_ymm);
case X86::MOV32ri64: {
Register Reg = MIB.getReg(0);
Register Reg32 = RI.getSubReg(Reg, X86::sub_32bit);
MI.setDesc(get(X86::MOV32ri));
MIB->getOperand(0).setReg(Reg32);
MIB.addReg(Reg, RegState::ImplicitDefine);
return true;
}
case X86::RDFLAGS32:
case X86::RDFLAGS64: {
unsigned Is64Bit = MI.getOpcode() == X86::RDFLAGS64;
MachineBasicBlock &MBB = *MIB->getParent();
MachineInstr *NewMI = BuildMI(MBB, MI, MIB->getDebugLoc(),
get(Is64Bit ? X86::PUSHF64 : X86::PUSHF32))
.getInstr();
// Permit reads of the EFLAGS and DF registers without them being defined.
// This intrinsic exists to read external processor state in flags, such as
// the trap flag, interrupt flag, and direction flag, none of which are
// modeled by the backend.
assert(NewMI->getOperand(2).getReg() == X86::EFLAGS &&
"Unexpected register in operand! Should be EFLAGS.");
NewMI->getOperand(2).setIsUndef();
assert(NewMI->getOperand(3).getReg() == X86::DF &&
"Unexpected register in operand! Should be DF.");
NewMI->getOperand(3).setIsUndef();
MIB->setDesc(get(Is64Bit ? X86::POP64r : X86::POP32r));
return true;
}
case X86::WRFLAGS32:
case X86::WRFLAGS64: {
unsigned Is64Bit = MI.getOpcode() == X86::WRFLAGS64;
MachineBasicBlock &MBB = *MIB->getParent();
BuildMI(MBB, MI, MIB->getDebugLoc(),
get(Is64Bit ? X86::PUSH64r : X86::PUSH32r))
.addReg(MI.getOperand(0).getReg());
BuildMI(MBB, MI, MIB->getDebugLoc(),
get(Is64Bit ? X86::POPF64 : X86::POPF32));
MI.eraseFromParent();
return true;
}
// KNL does not recognize dependency-breaking idioms for mask registers,
// so kxnor %k1, %k1, %k2 has a RAW dependence on %k1.
// Using %k0 as the undef input register is a performance heuristic based
// on the assumption that %k0 is used less frequently than the other mask
// registers, since it is not usable as a write mask.
// FIXME: A more advanced approach would be to choose the best input mask
// register based on context.
case X86::KSET0W:
return Expand2AddrKreg(MIB, get(X86::KXORWkk), X86::K0);
case X86::KSET0D:
return Expand2AddrKreg(MIB, get(X86::KXORDkk), X86::K0);
case X86::KSET0Q:
return Expand2AddrKreg(MIB, get(X86::KXORQkk), X86::K0);
case X86::KSET1W:
return Expand2AddrKreg(MIB, get(X86::KXNORWkk), X86::K0);
case X86::KSET1D:
return Expand2AddrKreg(MIB, get(X86::KXNORDkk), X86::K0);
case X86::KSET1Q:
return Expand2AddrKreg(MIB, get(X86::KXNORQkk), X86::K0);
case TargetOpcode::LOAD_STACK_GUARD:
expandLoadStackGuard(MIB, *this);
return true;
case X86::XOR64_FP:
case X86::XOR32_FP:
return expandXorFP(MIB, *this);
case X86::SHLDROT32ri:
return expandSHXDROT(MIB, get(X86::SHLD32rri8));
case X86::SHLDROT64ri:
return expandSHXDROT(MIB, get(X86::SHLD64rri8));
case X86::SHRDROT32ri:
return expandSHXDROT(MIB, get(X86::SHRD32rri8));
case X86::SHRDROT64ri:
return expandSHXDROT(MIB, get(X86::SHRD64rri8));
case X86::ADD8rr_DB:
MIB->setDesc(get(X86::OR8rr));
break;
case X86::ADD16rr_DB:
MIB->setDesc(get(X86::OR16rr));
break;
case X86::ADD32rr_DB:
MIB->setDesc(get(X86::OR32rr));
break;
case X86::ADD64rr_DB:
MIB->setDesc(get(X86::OR64rr));
break;
case X86::ADD8ri_DB:
MIB->setDesc(get(X86::OR8ri));
break;
case X86::ADD16ri_DB:
MIB->setDesc(get(X86::OR16ri));
break;
case X86::ADD32ri_DB:
MIB->setDesc(get(X86::OR32ri));
break;
case X86::ADD64ri32_DB:
MIB->setDesc(get(X86::OR64ri32));
break;
}
return false;
}
/// Return true for all instructions that only update
/// the first 32 or 64-bits of the destination register and leave the rest
/// unmodified. This can be used to avoid folding loads if the instructions
/// only update part of the destination register, and the non-updated part is
/// not needed. e.g. cvtss2sd, sqrtss. Unfolding the load from these
/// instructions breaks the partial register dependency and it can improve
/// performance. e.g.:
///
/// movss (%rdi), %xmm0
/// cvtss2sd %xmm0, %xmm0
///
/// Instead of
/// cvtss2sd (%rdi), %xmm0
///
/// FIXME: This should be turned into a TSFlags.
///
static bool hasPartialRegUpdate(unsigned Opcode, const X86Subtarget &Subtarget,
bool ForLoadFold = false) {
switch (Opcode) {
case X86::CVTSI2SSrr:
case X86::CVTSI2SSrm:
case X86::CVTSI642SSrr:
case X86::CVTSI642SSrm:
case X86::CVTSI2SDrr:
case X86::CVTSI2SDrm:
case X86::CVTSI642SDrr:
case X86::CVTSI642SDrm:
// Load folding won't effect the undef register update since the input is
// a GPR.
return !ForLoadFold;
case X86::CVTSD2SSrr:
case X86::CVTSD2SSrm:
case X86::CVTSS2SDrr:
case X86::CVTSS2SDrm:
case X86::MOVHPDrm:
case X86::MOVHPSrm:
case X86::MOVLPDrm:
case X86::MOVLPSrm:
case X86::RCPSSr:
case X86::RCPSSm:
case X86::RCPSSr_Int:
case X86::RCPSSm_Int:
case X86::ROUNDSDri:
case X86::ROUNDSDmi:
case X86::ROUNDSSri:
case X86::ROUNDSSmi:
case X86::RSQRTSSr:
case X86::RSQRTSSm:
case X86::RSQRTSSr_Int:
case X86::RSQRTSSm_Int:
case X86::SQRTSSr:
case X86::SQRTSSm:
case X86::SQRTSSr_Int:
case X86::SQRTSSm_Int:
case X86::SQRTSDr:
case X86::SQRTSDm:
case X86::SQRTSDr_Int:
case X86::SQRTSDm_Int:
return true;
case X86::VFCMULCPHZ128rm:
case X86::VFCMULCPHZ128rmb:
case X86::VFCMULCPHZ128rmbkz:
case X86::VFCMULCPHZ128rmkz:
case X86::VFCMULCPHZ128rr:
case X86::VFCMULCPHZ128rrkz:
case X86::VFCMULCPHZ256rm:
case X86::VFCMULCPHZ256rmb:
case X86::VFCMULCPHZ256rmbkz:
case X86::VFCMULCPHZ256rmkz:
case X86::VFCMULCPHZ256rr:
case X86::VFCMULCPHZ256rrkz:
case X86::VFCMULCPHZrm:
case X86::VFCMULCPHZrmb:
case X86::VFCMULCPHZrmbkz:
case X86::VFCMULCPHZrmkz:
case X86::VFCMULCPHZrr:
case X86::VFCMULCPHZrrb:
case X86::VFCMULCPHZrrbkz:
case X86::VFCMULCPHZrrkz:
case X86::VFMULCPHZ128rm:
case X86::VFMULCPHZ128rmb:
case X86::VFMULCPHZ128rmbkz:
case X86::VFMULCPHZ128rmkz:
case X86::VFMULCPHZ128rr:
case X86::VFMULCPHZ128rrkz:
case X86::VFMULCPHZ256rm:
case X86::VFMULCPHZ256rmb:
case X86::VFMULCPHZ256rmbkz:
case X86::VFMULCPHZ256rmkz:
case X86::VFMULCPHZ256rr:
case X86::VFMULCPHZ256rrkz:
case X86::VFMULCPHZrm:
case X86::VFMULCPHZrmb:
case X86::VFMULCPHZrmbkz:
case X86::VFMULCPHZrmkz:
case X86::VFMULCPHZrr:
case X86::VFMULCPHZrrb:
case X86::VFMULCPHZrrbkz:
case X86::VFMULCPHZrrkz:
case X86::VFCMULCSHZrm:
case X86::VFCMULCSHZrmkz:
case X86::VFCMULCSHZrr:
case X86::VFCMULCSHZrrb:
case X86::VFCMULCSHZrrbkz:
case X86::VFCMULCSHZrrkz:
case X86::VFMULCSHZrm:
case X86::VFMULCSHZrmkz:
case X86::VFMULCSHZrr:
case X86::VFMULCSHZrrb:
case X86::VFMULCSHZrrbkz:
case X86::VFMULCSHZrrkz:
return Subtarget.hasMULCFalseDeps();
case X86::VPERMDYrm:
case X86::VPERMDYrr:
case X86::VPERMQYmi:
case X86::VPERMQYri:
case X86::VPERMPSYrm:
case X86::VPERMPSYrr:
case X86::VPERMPDYmi:
case X86::VPERMPDYri:
case X86::VPERMDZ256rm:
case X86::VPERMDZ256rmb:
case X86::VPERMDZ256rmbkz:
case X86::VPERMDZ256rmkz:
case X86::VPERMDZ256rr:
case X86::VPERMDZ256rrkz:
case X86::VPERMDZrm:
case X86::VPERMDZrmb:
case X86::VPERMDZrmbkz:
case X86::VPERMDZrmkz:
case X86::VPERMDZrr:
case X86::VPERMDZrrkz:
case X86::VPERMQZ256mbi:
case X86::VPERMQZ256mbikz:
case X86::VPERMQZ256mi:
case X86::VPERMQZ256mikz:
case X86::VPERMQZ256ri:
case X86::VPERMQZ256rikz:
case X86::VPERMQZ256rm:
case X86::VPERMQZ256rmb:
case X86::VPERMQZ256rmbkz:
case X86::VPERMQZ256rmkz:
case X86::VPERMQZ256rr:
case X86::VPERMQZ256rrkz:
case X86::VPERMQZmbi:
case X86::VPERMQZmbikz:
case X86::VPERMQZmi:
case X86::VPERMQZmikz:
case X86::VPERMQZri:
case X86::VPERMQZrikz:
case X86::VPERMQZrm:
case X86::VPERMQZrmb:
case X86::VPERMQZrmbkz:
case X86::VPERMQZrmkz:
case X86::VPERMQZrr:
case X86::VPERMQZrrkz:
case X86::VPERMPSZ256rm:
case X86::VPERMPSZ256rmb:
case X86::VPERMPSZ256rmbkz:
case X86::VPERMPSZ256rmkz:
case X86::VPERMPSZ256rr:
case X86::VPERMPSZ256rrkz:
case X86::VPERMPSZrm:
case X86::VPERMPSZrmb:
case X86::VPERMPSZrmbkz:
case X86::VPERMPSZrmkz:
case X86::VPERMPSZrr:
case X86::VPERMPSZrrkz:
case X86::VPERMPDZ256mbi:
case X86::VPERMPDZ256mbikz:
case X86::VPERMPDZ256mi:
case X86::VPERMPDZ256mikz:
case X86::VPERMPDZ256ri:
case X86::VPERMPDZ256rikz:
case X86::VPERMPDZ256rm:
case X86::VPERMPDZ256rmb:
case X86::VPERMPDZ256rmbkz:
case X86::VPERMPDZ256rmkz:
case X86::VPERMPDZ256rr:
case X86::VPERMPDZ256rrkz:
case X86::VPERMPDZmbi:
case X86::VPERMPDZmbikz:
case X86::VPERMPDZmi:
case X86::VPERMPDZmikz:
case X86::VPERMPDZri:
case X86::VPERMPDZrikz:
case X86::VPERMPDZrm:
case X86::VPERMPDZrmb:
case X86::VPERMPDZrmbkz:
case X86::VPERMPDZrmkz:
case X86::VPERMPDZrr:
case X86::VPERMPDZrrkz:
return Subtarget.hasPERMFalseDeps();
case X86::VRANGEPDZ128rmbi:
case X86::VRANGEPDZ128rmbikz:
case X86::VRANGEPDZ128rmi:
case X86::VRANGEPDZ128rmikz:
case X86::VRANGEPDZ128rri:
case X86::VRANGEPDZ128rrikz:
case X86::VRANGEPDZ256rmbi:
case X86::VRANGEPDZ256rmbikz:
case X86::VRANGEPDZ256rmi:
case X86::VRANGEPDZ256rmikz:
case X86::VRANGEPDZ256rri:
case X86::VRANGEPDZ256rrikz:
case X86::VRANGEPDZrmbi:
case X86::VRANGEPDZrmbikz:
case X86::VRANGEPDZrmi:
case X86::VRANGEPDZrmikz:
case X86::VRANGEPDZrri:
case X86::VRANGEPDZrrib:
case X86::VRANGEPDZrribkz:
case X86::VRANGEPDZrrikz:
case X86::VRANGEPSZ128rmbi:
case X86::VRANGEPSZ128rmbikz:
case X86::VRANGEPSZ128rmi:
case X86::VRANGEPSZ128rmikz:
case X86::VRANGEPSZ128rri:
case X86::VRANGEPSZ128rrikz:
case X86::VRANGEPSZ256rmbi:
case X86::VRANGEPSZ256rmbikz:
case X86::VRANGEPSZ256rmi:
case X86::VRANGEPSZ256rmikz:
case X86::VRANGEPSZ256rri:
case X86::VRANGEPSZ256rrikz:
case X86::VRANGEPSZrmbi:
case X86::VRANGEPSZrmbikz:
case X86::VRANGEPSZrmi:
case X86::VRANGEPSZrmikz:
case X86::VRANGEPSZrri:
case X86::VRANGEPSZrrib:
case X86::VRANGEPSZrribkz:
case X86::VRANGEPSZrrikz:
case X86::VRANGESDZrmi:
case X86::VRANGESDZrmikz:
case X86::VRANGESDZrri:
case X86::VRANGESDZrrib:
case X86::VRANGESDZrribkz:
case X86::VRANGESDZrrikz:
case X86::VRANGESSZrmi:
case X86::VRANGESSZrmikz:
case X86::VRANGESSZrri:
case X86::VRANGESSZrrib:
case X86::VRANGESSZrribkz:
case X86::VRANGESSZrrikz:
return Subtarget.hasRANGEFalseDeps();
case X86::VGETMANTSSZrmi:
case X86::VGETMANTSSZrmikz:
case X86::VGETMANTSSZrri:
case X86::VGETMANTSSZrrib:
case X86::VGETMANTSSZrribkz:
case X86::VGETMANTSSZrrikz:
case X86::VGETMANTSDZrmi:
case X86::VGETMANTSDZrmikz:
case X86::VGETMANTSDZrri:
case X86::VGETMANTSDZrrib:
case X86::VGETMANTSDZrribkz:
case X86::VGETMANTSDZrrikz:
case X86::VGETMANTSHZrmi:
case X86::VGETMANTSHZrmikz:
case X86::VGETMANTSHZrri:
case X86::VGETMANTSHZrrib:
case X86::VGETMANTSHZrribkz:
case X86::VGETMANTSHZrrikz:
case X86::VGETMANTPSZ128rmbi:
case X86::VGETMANTPSZ128rmbikz:
case X86::VGETMANTPSZ128rmi:
case X86::VGETMANTPSZ128rmikz:
case X86::VGETMANTPSZ256rmbi:
case X86::VGETMANTPSZ256rmbikz:
case X86::VGETMANTPSZ256rmi:
case X86::VGETMANTPSZ256rmikz:
case X86::VGETMANTPSZrmbi:
case X86::VGETMANTPSZrmbikz:
case X86::VGETMANTPSZrmi:
case X86::VGETMANTPSZrmikz:
case X86::VGETMANTPDZ128rmbi:
case X86::VGETMANTPDZ128rmbikz:
case X86::VGETMANTPDZ128rmi:
case X86::VGETMANTPDZ128rmikz:
case X86::VGETMANTPDZ256rmbi:
case X86::VGETMANTPDZ256rmbikz:
case X86::VGETMANTPDZ256rmi:
case X86::VGETMANTPDZ256rmikz:
case X86::VGETMANTPDZrmbi:
case X86::VGETMANTPDZrmbikz:
case X86::VGETMANTPDZrmi:
case X86::VGETMANTPDZrmikz:
return Subtarget.hasGETMANTFalseDeps();
case X86::VPMULLQZ128rm:
case X86::VPMULLQZ128rmb:
case X86::VPMULLQZ128rmbkz:
case X86::VPMULLQZ128rmkz:
case X86::VPMULLQZ128rr:
case X86::VPMULLQZ128rrkz:
case X86::VPMULLQZ256rm:
case X86::VPMULLQZ256rmb:
case X86::VPMULLQZ256rmbkz:
case X86::VPMULLQZ256rmkz:
case X86::VPMULLQZ256rr:
case X86::VPMULLQZ256rrkz:
case X86::VPMULLQZrm:
case X86::VPMULLQZrmb:
case X86::VPMULLQZrmbkz:
case X86::VPMULLQZrmkz:
case X86::VPMULLQZrr:
case X86::VPMULLQZrrkz:
return Subtarget.hasMULLQFalseDeps();
// GPR
case X86::POPCNT32rm:
case X86::POPCNT32rr:
case X86::POPCNT64rm:
case X86::POPCNT64rr:
return Subtarget.hasPOPCNTFalseDeps();
case X86::LZCNT32rm:
case X86::LZCNT32rr:
case X86::LZCNT64rm:
case X86::LZCNT64rr:
case X86::TZCNT32rm:
case X86::TZCNT32rr:
case X86::TZCNT64rm:
case X86::TZCNT64rr:
return Subtarget.hasLZCNTFalseDeps();
}
return false;
}
/// Inform the BreakFalseDeps pass how many idle
/// instructions we would like before a partial register update.
unsigned X86InstrInfo::getPartialRegUpdateClearance(
const MachineInstr &MI, unsigned OpNum,
const TargetRegisterInfo *TRI) const {
if (OpNum != 0 || !hasPartialRegUpdate(MI.getOpcode(), Subtarget))
return 0;
// If MI is marked as reading Reg, the partial register update is wanted.
const MachineOperand &MO = MI.getOperand(0);
Register Reg = MO.getReg();
if (Reg.isVirtual()) {
if (MO.readsReg() || MI.readsVirtualRegister(Reg))
return 0;
} else {
if (MI.readsRegister(Reg, TRI))
return 0;
}
// If any instructions in the clearance range are reading Reg, insert a
// dependency breaking instruction, which is inexpensive and is likely to
// be hidden in other instruction's cycles.
return PartialRegUpdateClearance;
}
// Return true for any instruction the copies the high bits of the first source
// operand into the unused high bits of the destination operand.
// Also returns true for instructions that have two inputs where one may
// be undef and we want it to use the same register as the other input.
static bool hasUndefRegUpdate(unsigned Opcode, unsigned OpNum,
bool ForLoadFold = false) {
// Set the OpNum parameter to the first source operand.
switch (Opcode) {
case X86::MMX_PUNPCKHBWrr:
case X86::MMX_PUNPCKHWDrr:
case X86::MMX_PUNPCKHDQrr:
case X86::MMX_PUNPCKLBWrr:
case X86::MMX_PUNPCKLWDrr:
case X86::MMX_PUNPCKLDQrr:
case X86::MOVHLPSrr:
case X86::PACKSSWBrr:
case X86::PACKUSWBrr:
case X86::PACKSSDWrr:
case X86::PACKUSDWrr:
case X86::PUNPCKHBWrr:
case X86::PUNPCKLBWrr:
case X86::PUNPCKHWDrr:
case X86::PUNPCKLWDrr:
case X86::PUNPCKHDQrr:
case X86::PUNPCKLDQrr:
case X86::PUNPCKHQDQrr:
case X86::PUNPCKLQDQrr:
case X86::SHUFPDrri:
case X86::SHUFPSrri:
// These instructions are sometimes used with an undef first or second
// source. Return true here so BreakFalseDeps will assign this source to the
// same register as the first source to avoid a false dependency.
// Operand 1 of these instructions is tied so they're separate from their
// VEX counterparts.
return OpNum == 2 && !ForLoadFold;
case X86::VMOVLHPSrr:
case X86::VMOVLHPSZrr:
case X86::VPACKSSWBrr:
case X86::VPACKUSWBrr:
case X86::VPACKSSDWrr:
case X86::VPACKUSDWrr:
case X86::VPACKSSWBZ128rr:
case X86::VPACKUSWBZ128rr:
case X86::VPACKSSDWZ128rr:
case X86::VPACKUSDWZ128rr:
case X86::VPERM2F128rri:
case X86::VPERM2I128rri:
case X86::VSHUFF32X4Z256rri:
case X86::VSHUFF32X4Zrri:
case X86::VSHUFF64X2Z256rri:
case X86::VSHUFF64X2Zrri:
case X86::VSHUFI32X4Z256rri:
case X86::VSHUFI32X4Zrri:
case X86::VSHUFI64X2Z256rri:
case X86::VSHUFI64X2Zrri:
case X86::VPUNPCKHBWrr:
case X86::VPUNPCKLBWrr:
case X86::VPUNPCKHBWYrr:
case X86::VPUNPCKLBWYrr:
case X86::VPUNPCKHBWZ128rr:
case X86::VPUNPCKLBWZ128rr:
case X86::VPUNPCKHBWZ256rr:
case X86::VPUNPCKLBWZ256rr:
case X86::VPUNPCKHBWZrr:
case X86::VPUNPCKLBWZrr:
case X86::VPUNPCKHWDrr:
case X86::VPUNPCKLWDrr:
case X86::VPUNPCKHWDYrr:
case X86::VPUNPCKLWDYrr:
case X86::VPUNPCKHWDZ128rr:
case X86::VPUNPCKLWDZ128rr:
case X86::VPUNPCKHWDZ256rr:
case X86::VPUNPCKLWDZ256rr:
case X86::VPUNPCKHWDZrr:
case X86::VPUNPCKLWDZrr:
case X86::VPUNPCKHDQrr:
case X86::VPUNPCKLDQrr:
case X86::VPUNPCKHDQYrr:
case X86::VPUNPCKLDQYrr:
case X86::VPUNPCKHDQZ128rr:
case X86::VPUNPCKLDQZ128rr:
case X86::VPUNPCKHDQZ256rr:
case X86::VPUNPCKLDQZ256rr:
case X86::VPUNPCKHDQZrr:
case X86::VPUNPCKLDQZrr:
case X86::VPUNPCKHQDQrr:
case X86::VPUNPCKLQDQrr:
case X86::VPUNPCKHQDQYrr:
case X86::VPUNPCKLQDQYrr:
case X86::VPUNPCKHQDQZ128rr:
case X86::VPUNPCKLQDQZ128rr:
case X86::VPUNPCKHQDQZ256rr:
case X86::VPUNPCKLQDQZ256rr:
case X86::VPUNPCKHQDQZrr:
case X86::VPUNPCKLQDQZrr:
// These instructions are sometimes used with an undef first or second
// source. Return true here so BreakFalseDeps will assign this source to the
// same register as the first source to avoid a false dependency.
return (OpNum == 1 || OpNum == 2) && !ForLoadFold;
case X86::VCVTSI2SSrr:
case X86::VCVTSI2SSrm:
case X86::VCVTSI2SSrr_Int:
case X86::VCVTSI2SSrm_Int:
case X86::VCVTSI642SSrr:
case X86::VCVTSI642SSrm:
case X86::VCVTSI642SSrr_Int:
case X86::VCVTSI642SSrm_Int:
case X86::VCVTSI2SDrr:
case X86::VCVTSI2SDrm:
case X86::VCVTSI2SDrr_Int:
case X86::VCVTSI2SDrm_Int:
case X86::VCVTSI642SDrr:
case X86::VCVTSI642SDrm:
case X86::VCVTSI642SDrr_Int:
case X86::VCVTSI642SDrm_Int:
// AVX-512
case X86::VCVTSI2SSZrr:
case X86::VCVTSI2SSZrm:
case X86::VCVTSI2SSZrr_Int:
case X86::VCVTSI2SSZrrb_Int:
case X86::VCVTSI2SSZrm_Int:
case X86::VCVTSI642SSZrr:
case X86::VCVTSI642SSZrm:
case X86::VCVTSI642SSZrr_Int:
case X86::VCVTSI642SSZrrb_Int:
case X86::VCVTSI642SSZrm_Int:
case X86::VCVTSI2SDZrr:
case X86::VCVTSI2SDZrm:
case X86::VCVTSI2SDZrr_Int:
case X86::VCVTSI2SDZrm_Int:
case X86::VCVTSI642SDZrr:
case X86::VCVTSI642SDZrm:
case X86::VCVTSI642SDZrr_Int:
case X86::VCVTSI642SDZrrb_Int:
case X86::VCVTSI642SDZrm_Int:
case X86::VCVTUSI2SSZrr:
case X86::VCVTUSI2SSZrm:
case X86::VCVTUSI2SSZrr_Int:
case X86::VCVTUSI2SSZrrb_Int:
case X86::VCVTUSI2SSZrm_Int:
case X86::VCVTUSI642SSZrr:
case X86::VCVTUSI642SSZrm:
case X86::VCVTUSI642SSZrr_Int:
case X86::VCVTUSI642SSZrrb_Int:
case X86::VCVTUSI642SSZrm_Int:
case X86::VCVTUSI2SDZrr:
case X86::VCVTUSI2SDZrm:
case X86::VCVTUSI2SDZrr_Int:
case X86::VCVTUSI2SDZrm_Int:
case X86::VCVTUSI642SDZrr:
case X86::VCVTUSI642SDZrm:
case X86::VCVTUSI642SDZrr_Int:
case X86::VCVTUSI642SDZrrb_Int:
case X86::VCVTUSI642SDZrm_Int:
case X86::VCVTSI2SHZrr:
case X86::VCVTSI2SHZrm:
case X86::VCVTSI2SHZrr_Int:
case X86::VCVTSI2SHZrrb_Int:
case X86::VCVTSI2SHZrm_Int:
case X86::VCVTSI642SHZrr:
case X86::VCVTSI642SHZrm:
case X86::VCVTSI642SHZrr_Int:
case X86::VCVTSI642SHZrrb_Int:
case X86::VCVTSI642SHZrm_Int:
case X86::VCVTUSI2SHZrr:
case X86::VCVTUSI2SHZrm:
case X86::VCVTUSI2SHZrr_Int:
case X86::VCVTUSI2SHZrrb_Int:
case X86::VCVTUSI2SHZrm_Int:
case X86::VCVTUSI642SHZrr:
case X86::VCVTUSI642SHZrm:
case X86::VCVTUSI642SHZrr_Int:
case X86::VCVTUSI642SHZrrb_Int:
case X86::VCVTUSI642SHZrm_Int:
// Load folding won't effect the undef register update since the input is
// a GPR.
return OpNum == 1 && !ForLoadFold;
case X86::VCVTSD2SSrr:
case X86::VCVTSD2SSrm:
case X86::VCVTSD2SSrr_Int:
case X86::VCVTSD2SSrm_Int:
case X86::VCVTSS2SDrr:
case X86::VCVTSS2SDrm:
case X86::VCVTSS2SDrr_Int:
case X86::VCVTSS2SDrm_Int:
case X86::VRCPSSr:
case X86::VRCPSSr_Int:
case X86::VRCPSSm:
case X86::VRCPSSm_Int:
case X86::VROUNDSDri:
case X86::VROUNDSDmi:
case X86::VROUNDSDri_Int:
case X86::VROUNDSDmi_Int:
case X86::VROUNDSSri:
case X86::VROUNDSSmi:
case X86::VROUNDSSri_Int:
case X86::VROUNDSSmi_Int:
case X86::VRSQRTSSr:
case X86::VRSQRTSSr_Int:
case X86::VRSQRTSSm:
case X86::VRSQRTSSm_Int:
case X86::VSQRTSSr:
case X86::VSQRTSSr_Int:
case X86::VSQRTSSm:
case X86::VSQRTSSm_Int:
case X86::VSQRTSDr:
case X86::VSQRTSDr_Int:
case X86::VSQRTSDm:
case X86::VSQRTSDm_Int:
// AVX-512
case X86::VCVTSD2SSZrr:
case X86::VCVTSD2SSZrr_Int:
case X86::VCVTSD2SSZrrb_Int:
case X86::VCVTSD2SSZrm:
case X86::VCVTSD2SSZrm_Int:
case X86::VCVTSS2SDZrr:
case X86::VCVTSS2SDZrr_Int:
case X86::VCVTSS2SDZrrb_Int:
case X86::VCVTSS2SDZrm:
case X86::VCVTSS2SDZrm_Int:
case X86::VGETEXPSDZr:
case X86::VGETEXPSDZrb:
case X86::VGETEXPSDZm:
case X86::VGETEXPSSZr:
case X86::VGETEXPSSZrb:
case X86::VGETEXPSSZm:
case X86::VGETMANTSDZrri:
case X86::VGETMANTSDZrrib:
case X86::VGETMANTSDZrmi:
case X86::VGETMANTSSZrri:
case X86::VGETMANTSSZrrib:
case X86::VGETMANTSSZrmi:
case X86::VRNDSCALESDZr:
case X86::VRNDSCALESDZr_Int:
case X86::VRNDSCALESDZrb_Int:
case X86::VRNDSCALESDZm:
case X86::VRNDSCALESDZm_Int:
case X86::VRNDSCALESSZr:
case X86::VRNDSCALESSZr_Int:
case X86::VRNDSCALESSZrb_Int:
case X86::VRNDSCALESSZm:
case X86::VRNDSCALESSZm_Int:
case X86::VRCP14SDZrr:
case X86::VRCP14SDZrm:
case X86::VRCP14SSZrr:
case X86::VRCP14SSZrm:
case X86::VRCPSHZrr:
case X86::VRCPSHZrm:
case X86::VRSQRTSHZrr:
case X86::VRSQRTSHZrm:
case X86::VREDUCESHZrmi:
case X86::VREDUCESHZrri:
case X86::VREDUCESHZrrib:
case X86::VGETEXPSHZr:
case X86::VGETEXPSHZrb:
case X86::VGETEXPSHZm:
case X86::VGETMANTSHZrri:
case X86::VGETMANTSHZrrib:
case X86::VGETMANTSHZrmi:
case X86::VRNDSCALESHZr:
case X86::VRNDSCALESHZr_Int:
case X86::VRNDSCALESHZrb_Int:
case X86::VRNDSCALESHZm:
case X86::VRNDSCALESHZm_Int:
case X86::VSQRTSHZr:
case X86::VSQRTSHZr_Int:
case X86::VSQRTSHZrb_Int:
case X86::VSQRTSHZm:
case X86::VSQRTSHZm_Int:
case X86::VRCP28SDZr:
case X86::VRCP28SDZrb:
case X86::VRCP28SDZm:
case X86::VRCP28SSZr:
case X86::VRCP28SSZrb:
case X86::VRCP28SSZm:
case X86::VREDUCESSZrmi:
case X86::VREDUCESSZrri:
case X86::VREDUCESSZrrib:
case X86::VRSQRT14SDZrr:
case X86::VRSQRT14SDZrm:
case X86::VRSQRT14SSZrr:
case X86::VRSQRT14SSZrm:
case X86::VRSQRT28SDZr:
case X86::VRSQRT28SDZrb:
case X86::VRSQRT28SDZm:
case X86::VRSQRT28SSZr:
case X86::VRSQRT28SSZrb:
case X86::VRSQRT28SSZm:
case X86::VSQRTSSZr:
case X86::VSQRTSSZr_Int:
case X86::VSQRTSSZrb_Int:
case X86::VSQRTSSZm:
case X86::VSQRTSSZm_Int:
case X86::VSQRTSDZr:
case X86::VSQRTSDZr_Int:
case X86::VSQRTSDZrb_Int:
case X86::VSQRTSDZm:
case X86::VSQRTSDZm_Int:
case X86::VCVTSD2SHZrr:
case X86::VCVTSD2SHZrr_Int:
case X86::VCVTSD2SHZrrb_Int:
case X86::VCVTSD2SHZrm:
case X86::VCVTSD2SHZrm_Int:
case X86::VCVTSS2SHZrr:
case X86::VCVTSS2SHZrr_Int:
case X86::VCVTSS2SHZrrb_Int:
case X86::VCVTSS2SHZrm:
case X86::VCVTSS2SHZrm_Int:
case X86::VCVTSH2SDZrr:
case X86::VCVTSH2SDZrr_Int:
case X86::VCVTSH2SDZrrb_Int:
case X86::VCVTSH2SDZrm:
case X86::VCVTSH2SDZrm_Int:
case X86::VCVTSH2SSZrr:
case X86::VCVTSH2SSZrr_Int:
case X86::VCVTSH2SSZrrb_Int:
case X86::VCVTSH2SSZrm:
case X86::VCVTSH2SSZrm_Int:
return OpNum == 1;
case X86::VMOVSSZrrk:
case X86::VMOVSDZrrk:
return OpNum == 3 && !ForLoadFold;
case X86::VMOVSSZrrkz:
case X86::VMOVSDZrrkz:
return OpNum == 2 && !ForLoadFold;
}
return false;
}
/// Inform the BreakFalseDeps pass how many idle instructions we would like
/// before certain undef register reads.
///
/// This catches the VCVTSI2SD family of instructions:
///
/// vcvtsi2sdq %rax, undef %xmm0, %xmm14
///
/// We should to be careful *not* to catch VXOR idioms which are presumably
/// handled specially in the pipeline:
///
/// vxorps undef %xmm1, undef %xmm1, %xmm1
///
/// Like getPartialRegUpdateClearance, this makes a strong assumption that the
/// high bits that are passed-through are not live.
unsigned
X86InstrInfo::getUndefRegClearance(const MachineInstr &MI, unsigned OpNum,
const TargetRegisterInfo *TRI) const {
const MachineOperand &MO = MI.getOperand(OpNum);
if (MO.getReg().isPhysical() && hasUndefRegUpdate(MI.getOpcode(), OpNum))
return UndefRegClearance;
return 0;
}
void X86InstrInfo::breakPartialRegDependency(
MachineInstr &MI, unsigned OpNum, const TargetRegisterInfo *TRI) const {
Register Reg = MI.getOperand(OpNum).getReg();
// If MI kills this register, the false dependence is already broken.
if (MI.killsRegister(Reg, TRI))
return;
if (X86::VR128RegClass.contains(Reg)) {
// These instructions are all floating point domain, so xorps is the best
// choice.
unsigned Opc = Subtarget.hasAVX() ? X86::VXORPSrr : X86::XORPSrr;
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(Opc), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
MI.addRegisterKilled(Reg, TRI, true);
} else if (X86::VR256RegClass.contains(Reg)) {
// Use vxorps to clear the full ymm register.
// It wants to read and write the xmm sub-register.
Register XReg = TRI->getSubReg(Reg, X86::sub_xmm);
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(X86::VXORPSrr), XReg)
.addReg(XReg, RegState::Undef)
.addReg(XReg, RegState::Undef)
.addReg(Reg, RegState::ImplicitDefine);
MI.addRegisterKilled(Reg, TRI, true);
} else if (X86::VR128XRegClass.contains(Reg)) {
// Only handle VLX targets.
if (!Subtarget.hasVLX())
return;
// Since vxorps requires AVX512DQ, vpxord should be the best choice.
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(X86::VPXORDZ128rr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
MI.addRegisterKilled(Reg, TRI, true);
} else if (X86::VR256XRegClass.contains(Reg) ||
X86::VR512RegClass.contains(Reg)) {
// Only handle VLX targets.
if (!Subtarget.hasVLX())
return;
// Use vpxord to clear the full ymm/zmm register.
// It wants to read and write the xmm sub-register.
Register XReg = TRI->getSubReg(Reg, X86::sub_xmm);
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(X86::VPXORDZ128rr), XReg)
.addReg(XReg, RegState::Undef)
.addReg(XReg, RegState::Undef)
.addReg(Reg, RegState::ImplicitDefine);
MI.addRegisterKilled(Reg, TRI, true);
} else if (X86::GR64RegClass.contains(Reg)) {
// Using XOR32rr because it has shorter encoding and zeros up the upper bits
// as well.
Register XReg = TRI->getSubReg(Reg, X86::sub_32bit);
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(X86::XOR32rr), XReg)
.addReg(XReg, RegState::Undef)
.addReg(XReg, RegState::Undef)
.addReg(Reg, RegState::ImplicitDefine);
MI.addRegisterKilled(Reg, TRI, true);
} else if (X86::GR32RegClass.contains(Reg)) {
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), get(X86::XOR32rr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
MI.addRegisterKilled(Reg, TRI, true);
}
}
static void addOperands(MachineInstrBuilder &MIB, ArrayRef<MachineOperand> MOs,
int PtrOffset = 0) {
unsigned NumAddrOps = MOs.size();
if (NumAddrOps < 4) {
// FrameIndex only - add an immediate offset (whether its zero or not).
for (unsigned i = 0; i != NumAddrOps; ++i)
MIB.add(MOs[i]);
addOffset(MIB, PtrOffset);
} else {
// General Memory Addressing - we need to add any offset to an existing
// offset.
assert(MOs.size() == 5 && "Unexpected memory operand list length");
for (unsigned i = 0; i != NumAddrOps; ++i) {
const MachineOperand &MO = MOs[i];
if (i == 3 && PtrOffset != 0) {
MIB.addDisp(MO, PtrOffset);
} else {
MIB.add(MO);
}
}
}
}
static void updateOperandRegConstraints(MachineFunction &MF,
MachineInstr &NewMI,
const TargetInstrInfo &TII) {
MachineRegisterInfo &MRI = MF.getRegInfo();
const TargetRegisterInfo &TRI = *MRI.getTargetRegisterInfo();
for (int Idx : llvm::seq<int>(0, NewMI.getNumOperands())) {
MachineOperand &MO = NewMI.getOperand(Idx);
// We only need to update constraints on virtual register operands.
if (!MO.isReg())
continue;
Register Reg = MO.getReg();
if (!Reg.isVirtual())
continue;
auto *NewRC = MRI.constrainRegClass(
Reg, TII.getRegClass(NewMI.getDesc(), Idx, &TRI, MF));
if (!NewRC) {
LLVM_DEBUG(
dbgs() << "WARNING: Unable to update register constraint for operand "
<< Idx << " of instruction:\n";
NewMI.dump(); dbgs() << "\n");
}
}
}
static MachineInstr *fuseTwoAddrInst(MachineFunction &MF, unsigned Opcode,
ArrayRef<MachineOperand> MOs,
MachineBasicBlock::iterator InsertPt,
MachineInstr &MI,
const TargetInstrInfo &TII) {
// Create the base instruction with the memory operand as the first part.
// Omit the implicit operands, something BuildMI can't do.
MachineInstr *NewMI =
MF.CreateMachineInstr(TII.get(Opcode), MI.getDebugLoc(), true);
MachineInstrBuilder MIB(MF, NewMI);
addOperands(MIB, MOs);
// Loop over the rest of the ri operands, converting them over.
unsigned NumOps = MI.getDesc().getNumOperands() - 2;
for (unsigned i = 0; i != NumOps; ++i) {
MachineOperand &MO = MI.getOperand(i + 2);
MIB.add(MO);
}
for (const MachineOperand &MO : llvm::drop_begin(MI.operands(), NumOps + 2))
MIB.add(MO);
updateOperandRegConstraints(MF, *NewMI, TII);
MachineBasicBlock *MBB = InsertPt->getParent();
MBB->insert(InsertPt, NewMI);
return MIB;
}
static MachineInstr *fuseInst(MachineFunction &MF, unsigned Opcode,
unsigned OpNo, ArrayRef<MachineOperand> MOs,
MachineBasicBlock::iterator InsertPt,
MachineInstr &MI, const TargetInstrInfo &TII,
int PtrOffset = 0) {
// Omit the implicit operands, something BuildMI can't do.
MachineInstr *NewMI =
MF.CreateMachineInstr(TII.get(Opcode), MI.getDebugLoc(), true);
MachineInstrBuilder MIB(MF, NewMI);
for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI.getOperand(i);
if (i == OpNo) {
assert(MO.isReg() && "Expected to fold into reg operand!");
addOperands(MIB, MOs, PtrOffset);
} else {
MIB.add(MO);
}
}
updateOperandRegConstraints(MF, *NewMI, TII);
// Copy the NoFPExcept flag from the instruction we're fusing.
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
NewMI->setFlag(MachineInstr::MIFlag::NoFPExcept);
MachineBasicBlock *MBB = InsertPt->getParent();
MBB->insert(InsertPt, NewMI);
return MIB;
}
static MachineInstr *makeM0Inst(const TargetInstrInfo &TII, unsigned Opcode,
ArrayRef<MachineOperand> MOs,
MachineBasicBlock::iterator InsertPt,
MachineInstr &MI) {
MachineInstrBuilder MIB = BuildMI(*InsertPt->getParent(), InsertPt,
MI.getDebugLoc(), TII.get(Opcode));
addOperands(MIB, MOs);
return MIB.addImm(0);
}
MachineInstr *X86InstrInfo::foldMemoryOperandCustom(
MachineFunction &MF, MachineInstr &MI, unsigned OpNum,
ArrayRef<MachineOperand> MOs, MachineBasicBlock::iterator InsertPt,
unsigned Size, Align Alignment) const {
switch (MI.getOpcode()) {
case X86::INSERTPSrri:
case X86::VINSERTPSrri:
case X86::VINSERTPSZrri:
// Attempt to convert the load of inserted vector into a fold load
// of a single float.
if (OpNum == 2) {
unsigned Imm = MI.getOperand(MI.getNumOperands() - 1).getImm();
unsigned ZMask = Imm & 15;
unsigned DstIdx = (Imm >> 4) & 3;
unsigned SrcIdx = (Imm >> 6) & 3;
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = getRegClass(MI.getDesc(), OpNum, &RI, MF);
unsigned RCSize = TRI.getRegSizeInBits(*RC) / 8;
if ((Size == 0 || Size >= 16) && RCSize >= 16 &&
(MI.getOpcode() != X86::INSERTPSrri || Alignment >= Align(4))) {
int PtrOffset = SrcIdx * 4;
unsigned NewImm = (DstIdx << 4) | ZMask;
unsigned NewOpCode =
(MI.getOpcode() == X86::VINSERTPSZrri) ? X86::VINSERTPSZrmi
: (MI.getOpcode() == X86::VINSERTPSrri) ? X86::VINSERTPSrmi
: X86::INSERTPSrmi;
MachineInstr *NewMI =
fuseInst(MF, NewOpCode, OpNum, MOs, InsertPt, MI, *this, PtrOffset);
NewMI->getOperand(NewMI->getNumOperands() - 1).setImm(NewImm);
return NewMI;
}
}
break;
case X86::MOVHLPSrr:
case X86::VMOVHLPSrr:
case X86::VMOVHLPSZrr:
// Move the upper 64-bits of the second operand to the lower 64-bits.
// To fold the load, adjust the pointer to the upper and use (V)MOVLPS.
// TODO: In most cases AVX doesn't have a 8-byte alignment requirement.
if (OpNum == 2) {
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = getRegClass(MI.getDesc(), OpNum, &RI, MF);
unsigned RCSize = TRI.getRegSizeInBits(*RC) / 8;
if ((Size == 0 || Size >= 16) && RCSize >= 16 && Alignment >= Align(8)) {
unsigned NewOpCode =
(MI.getOpcode() == X86::VMOVHLPSZrr) ? X86::VMOVLPSZ128rm
: (MI.getOpcode() == X86::VMOVHLPSrr) ? X86::VMOVLPSrm
: X86::MOVLPSrm;
MachineInstr *NewMI =
fuseInst(MF, NewOpCode, OpNum, MOs, InsertPt, MI, *this, 8);
return NewMI;
}
}
break;
case X86::UNPCKLPDrr:
// If we won't be able to fold this to the memory form of UNPCKL, use
// MOVHPD instead. Done as custom because we can't have this in the load
// table twice.
if (OpNum == 2) {
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = getRegClass(MI.getDesc(), OpNum, &RI, MF);
unsigned RCSize = TRI.getRegSizeInBits(*RC) / 8;
if ((Size == 0 || Size >= 16) && RCSize >= 16 && Alignment < Align(16)) {
MachineInstr *NewMI =
fuseInst(MF, X86::MOVHPDrm, OpNum, MOs, InsertPt, MI, *this);
return NewMI;
}
}
break;
case X86::MOV32r0:
if (auto *NewMI =
makeM0Inst(*this, (Size == 4) ? X86::MOV32mi : X86::MOV64mi32, MOs,
InsertPt, MI))
return NewMI;
break;
}
return nullptr;
}
static bool shouldPreventUndefRegUpdateMemFold(MachineFunction &MF,
MachineInstr &MI) {
if (!hasUndefRegUpdate(MI.getOpcode(), 1, /*ForLoadFold*/ true) ||
!MI.getOperand(1).isReg())
return false;
// The are two cases we need to handle depending on where in the pipeline
// the folding attempt is being made.
// -Register has the undef flag set.
// -Register is produced by the IMPLICIT_DEF instruction.
if (MI.getOperand(1).isUndef())
return true;
MachineRegisterInfo &RegInfo = MF.getRegInfo();
MachineInstr *VRegDef = RegInfo.getUniqueVRegDef(MI.getOperand(1).getReg());
return VRegDef && VRegDef->isImplicitDef();
}
unsigned X86InstrInfo::commuteOperandsForFold(MachineInstr &MI,
unsigned Idx1) const {
unsigned Idx2 = CommuteAnyOperandIndex;
if (!findCommutedOpIndices(MI, Idx1, Idx2))
return Idx1;
bool HasDef = MI.getDesc().getNumDefs();
Register Reg0 = HasDef ? MI.getOperand(0).getReg() : Register();
Register Reg1 = MI.getOperand(Idx1).getReg();
Register Reg2 = MI.getOperand(Idx2).getReg();
bool Tied1 = 0 == MI.getDesc().getOperandConstraint(Idx1, MCOI::TIED_TO);
bool Tied2 = 0 == MI.getDesc().getOperandConstraint(Idx2, MCOI::TIED_TO);
// If either of the commutable operands are tied to the destination
// then we can not commute + fold.
if ((HasDef && Reg0 == Reg1 && Tied1) || (HasDef && Reg0 == Reg2 && Tied2))
return Idx1;
return commuteInstruction(MI, false, Idx1, Idx2) ? Idx2 : Idx1;
}
static void printFailMsgforFold(const MachineInstr &MI, unsigned Idx) {
if (PrintFailedFusing && !MI.isCopy())
dbgs() << "We failed to fuse operand " << Idx << " in " << MI;
}
MachineInstr *X86InstrInfo::foldMemoryOperandImpl(
MachineFunction &MF, MachineInstr &MI, unsigned OpNum,
ArrayRef<MachineOperand> MOs, MachineBasicBlock::iterator InsertPt,
unsigned Size, Align Alignment, bool AllowCommute) const {
bool isSlowTwoMemOps = Subtarget.slowTwoMemOps();
unsigned Opc = MI.getOpcode();
// For CPUs that favor the register form of a call or push,
// do not fold loads into calls or pushes, unless optimizing for size
// aggressively.
if (isSlowTwoMemOps && !MF.getFunction().hasMinSize() &&
(Opc == X86::CALL32r || Opc == X86::CALL64r || Opc == X86::PUSH16r ||
Opc == X86::PUSH32r || Opc == X86::PUSH64r))
return nullptr;
// Avoid partial and undef register update stalls unless optimizing for size.
if (!MF.getFunction().hasOptSize() &&
(hasPartialRegUpdate(Opc, Subtarget, /*ForLoadFold*/ true) ||
shouldPreventUndefRegUpdateMemFold(MF, MI)))
return nullptr;
unsigned NumOps = MI.getDesc().getNumOperands();
bool IsTwoAddr = NumOps > 1 && OpNum < 2 && MI.getOperand(0).isReg() &&
MI.getOperand(1).isReg() &&
MI.getOperand(0).getReg() == MI.getOperand(1).getReg();
// FIXME: AsmPrinter doesn't know how to handle
// X86II::MO_GOT_ABSOLUTE_ADDRESS after folding.
if (Opc == X86::ADD32ri &&
MI.getOperand(2).getTargetFlags() == X86II::MO_GOT_ABSOLUTE_ADDRESS)
return nullptr;
// GOTTPOFF relocation loads can only be folded into add instructions.
// FIXME: Need to exclude other relocations that only support specific
// instructions.
if (MOs.size() == X86::AddrNumOperands &&
MOs[X86::AddrDisp].getTargetFlags() == X86II::MO_GOTTPOFF &&
Opc != X86::ADD64rr)
return nullptr;
// Don't fold loads into indirect calls that need a KCFI check as we'll
// have to unfold these in X86TargetLowering::EmitKCFICheck anyway.
if (MI.isCall() && MI.getCFIType())
return nullptr;
// Attempt to fold any custom cases we have.
if (auto *CustomMI = foldMemoryOperandCustom(MF, MI, OpNum, MOs, InsertPt,
Size, Alignment))
return CustomMI;
// Folding a memory location into the two-address part of a two-address
// instruction is different than folding it other places. It requires
// replacing the *two* registers with the memory location.
//
// Utilize the mapping NonNDD -> RMW for the NDD variant.
unsigned NonNDOpc = Subtarget.hasNDD() ? X86::getNonNDVariant(Opc) : 0U;
const X86FoldTableEntry *I =
IsTwoAddr ? lookupTwoAddrFoldTable(NonNDOpc ? NonNDOpc : Opc)
: lookupFoldTable(Opc, OpNum);
MachineInstr *NewMI = nullptr;
if (I) {
unsigned Opcode = I->DstOp;
if (Alignment <
Align(1ULL << ((I->Flags & TB_ALIGN_MASK) >> TB_ALIGN_SHIFT)))
return nullptr;
bool NarrowToMOV32rm = false;
if (Size) {
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = getRegClass(MI.getDesc(), OpNum, &RI, MF);
unsigned RCSize = TRI.getRegSizeInBits(*RC) / 8;
// Check if it's safe to fold the load. If the size of the object is
// narrower than the load width, then it's not.
// FIXME: Allow scalar intrinsic instructions like ADDSSrm_Int.
if ((I->Flags & TB_FOLDED_LOAD) && Size < RCSize) {
// If this is a 64-bit load, but the spill slot is 32, then we can do
// a 32-bit load which is implicitly zero-extended. This likely is
// due to live interval analysis remat'ing a load from stack slot.
if (Opcode != X86::MOV64rm || RCSize != 8 || Size != 4)
return nullptr;
if (MI.getOperand(0).getSubReg() || MI.getOperand(1).getSubReg())
return nullptr;
Opcode = X86::MOV32rm;
NarrowToMOV32rm = true;
}
// For stores, make sure the size of the object is equal to the size of
// the store. If the object is larger, the extra bits would be garbage. If
// the object is smaller we might overwrite another object or fault.
if ((I->Flags & TB_FOLDED_STORE) && Size != RCSize)
return nullptr;
}
NewMI = IsTwoAddr ? fuseTwoAddrInst(MF, Opcode, MOs, InsertPt, MI, *this)
: fuseInst(MF, Opcode, OpNum, MOs, InsertPt, MI, *this);
if (NarrowToMOV32rm) {
// If this is the special case where we use a MOV32rm to load a 32-bit
// value and zero-extend the top bits. Change the destination register
// to a 32-bit one.
Register DstReg = NewMI->getOperand(0).getReg();
if (DstReg.isPhysical())
NewMI->getOperand(0).setReg(RI.getSubReg(DstReg, X86::sub_32bit));
else
NewMI->getOperand(0).setSubReg(X86::sub_32bit);
}
return NewMI;
}
if (AllowCommute) {
// If the instruction and target operand are commutable, commute the
// instruction and try again.
unsigned CommuteOpIdx2 = commuteOperandsForFold(MI, OpNum);
if (CommuteOpIdx2 == OpNum) {
printFailMsgforFold(MI, OpNum);
return nullptr;
}
// Attempt to fold with the commuted version of the instruction.
NewMI = foldMemoryOperandImpl(MF, MI, CommuteOpIdx2, MOs, InsertPt, Size,
Alignment, /*AllowCommute=*/false);
if (NewMI)
return NewMI;
// Folding failed again - undo the commute before returning.
commuteInstruction(MI, false, OpNum, CommuteOpIdx2);
}
printFailMsgforFold(MI, OpNum);
return nullptr;
}
MachineInstr *X86InstrInfo::foldMemoryOperandImpl(
MachineFunction &MF, MachineInstr &MI, ArrayRef<unsigned> Ops,
MachineBasicBlock::iterator InsertPt, int FrameIndex, LiveIntervals *LIS,
VirtRegMap *VRM) const {
// Check switch flag
if (NoFusing)
return nullptr;
// Avoid partial and undef register update stalls unless optimizing for size.
if (!MF.getFunction().hasOptSize() &&
(hasPartialRegUpdate(MI.getOpcode(), Subtarget, /*ForLoadFold*/ true) ||
shouldPreventUndefRegUpdateMemFold(MF, MI)))
return nullptr;
// Don't fold subreg spills, or reloads that use a high subreg.
for (auto Op : Ops) {
MachineOperand &MO = MI.getOperand(Op);
auto SubReg = MO.getSubReg();
// MOV32r0 is special b/c it's used to clear a 64-bit register too.
// (See patterns for MOV32r0 in TD files).
if (MI.getOpcode() == X86::MOV32r0 && SubReg == X86::sub_32bit)
continue;
if (SubReg && (MO.isDef() || SubReg == X86::sub_8bit_hi))
return nullptr;
}
const MachineFrameInfo &MFI = MF.getFrameInfo();
unsigned Size = MFI.getObjectSize(FrameIndex);
Align Alignment = MFI.getObjectAlign(FrameIndex);
// If the function stack isn't realigned we don't want to fold instructions
// that need increased alignment.
if (!RI.hasStackRealignment(MF))
Alignment =
std::min(Alignment, Subtarget.getFrameLowering()->getStackAlign());
auto Impl = [&]() {
return foldMemoryOperandImpl(MF, MI, Ops[0],
MachineOperand::CreateFI(FrameIndex), InsertPt,
Size, Alignment, /*AllowCommute=*/true);
};
if (Ops.size() == 2 && Ops[0] == 0 && Ops[1] == 1) {
unsigned NewOpc = 0;
unsigned RCSize = 0;
unsigned Opc = MI.getOpcode();
switch (Opc) {
default:
// NDD can be folded into RMW though its Op0 and Op1 are not tied.
return (Subtarget.hasNDD() ? X86::getNonNDVariant(Opc) : 0U) ? Impl()
: nullptr;
case X86::TEST8rr:
NewOpc = X86::CMP8ri;
RCSize = 1;
break;
case X86::TEST16rr:
NewOpc = X86::CMP16ri;
RCSize = 2;
break;
case X86::TEST32rr:
NewOpc = X86::CMP32ri;
RCSize = 4;
break;
case X86::TEST64rr:
NewOpc = X86::CMP64ri32;
RCSize = 8;
break;
}
// Check if it's safe to fold the load. If the size of the object is
// narrower than the load width, then it's not.
if (Size < RCSize)
return nullptr;
// Change to CMPXXri r, 0 first.
MI.setDesc(get(NewOpc));
MI.getOperand(1).ChangeToImmediate(0);
} else if (Ops.size() != 1)
return nullptr;
return Impl();
}
/// Check if \p LoadMI is a partial register load that we can't fold into \p MI
/// because the latter uses contents that wouldn't be defined in the folded
/// version. For instance, this transformation isn't legal:
/// movss (%rdi), %xmm0
/// addps %xmm0, %xmm0
/// ->
/// addps (%rdi), %xmm0
///
/// But this one is:
/// movss (%rdi), %xmm0
/// addss %xmm0, %xmm0
/// ->
/// addss (%rdi), %xmm0
///
static bool isNonFoldablePartialRegisterLoad(const MachineInstr &LoadMI,
const MachineInstr &UserMI,
const MachineFunction &MF) {
unsigned Opc = LoadMI.getOpcode();
unsigned UserOpc = UserMI.getOpcode();
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC =
MF.getRegInfo().getRegClass(LoadMI.getOperand(0).getReg());
unsigned RegSize = TRI.getRegSizeInBits(*RC);
if ((Opc == X86::MOVSSrm || Opc == X86::VMOVSSrm || Opc == X86::VMOVSSZrm ||
Opc == X86::MOVSSrm_alt || Opc == X86::VMOVSSrm_alt ||
Opc == X86::VMOVSSZrm_alt) &&
RegSize > 32) {
// These instructions only load 32 bits, we can't fold them if the
// destination register is wider than 32 bits (4 bytes), and its user
// instruction isn't scalar (SS).
switch (UserOpc) {
case X86::CVTSS2SDrr_Int:
case X86::VCVTSS2SDrr_Int:
case X86::VCVTSS2SDZrr_Int:
case X86::VCVTSS2SDZrr_Intk:
case X86::VCVTSS2SDZrr_Intkz:
case X86::CVTSS2SIrr_Int:
case X86::CVTSS2SI64rr_Int:
case X86::VCVTSS2SIrr_Int:
case X86::VCVTSS2SI64rr_Int:
case X86::VCVTSS2SIZrr_Int:
case X86::VCVTSS2SI64Zrr_Int:
case X86::CVTTSS2SIrr_Int:
case X86::CVTTSS2SI64rr_Int:
case X86::VCVTTSS2SIrr_Int:
case X86::VCVTTSS2SI64rr_Int:
case X86::VCVTTSS2SIZrr_Int:
case X86::VCVTTSS2SI64Zrr_Int:
case X86::VCVTSS2USIZrr_Int:
case X86::VCVTSS2USI64Zrr_Int:
case X86::VCVTTSS2USIZrr_Int:
case X86::VCVTTSS2USI64Zrr_Int:
case X86::RCPSSr_Int:
case X86::VRCPSSr_Int:
case X86::RSQRTSSr_Int:
case X86::VRSQRTSSr_Int:
case X86::ROUNDSSri_Int:
case X86::VROUNDSSri_Int:
case X86::COMISSrr_Int:
case X86::VCOMISSrr_Int:
case X86::VCOMISSZrr_Int:
case X86::UCOMISSrr_Int:
case X86::VUCOMISSrr_Int:
case X86::VUCOMISSZrr_Int:
case X86::ADDSSrr_Int:
case X86::VADDSSrr_Int:
case X86::VADDSSZrr_Int:
case X86::CMPSSrri_Int:
case X86::VCMPSSrri_Int:
case X86::VCMPSSZrri_Int:
case X86::DIVSSrr_Int:
case X86::VDIVSSrr_Int:
case X86::VDIVSSZrr_Int:
case X86::MAXSSrr_Int:
case X86::VMAXSSrr_Int:
case X86::VMAXSSZrr_Int:
case X86::MINSSrr_Int:
case X86::VMINSSrr_Int:
case X86::VMINSSZrr_Int:
case X86::MULSSrr_Int:
case X86::VMULSSrr_Int:
case X86::VMULSSZrr_Int:
case X86::SQRTSSr_Int:
case X86::VSQRTSSr_Int:
case X86::VSQRTSSZr_Int:
case X86::SUBSSrr_Int:
case X86::VSUBSSrr_Int:
case X86::VSUBSSZrr_Int:
case X86::VADDSSZrr_Intk:
case X86::VADDSSZrr_Intkz:
case X86::VCMPSSZrri_Intk:
case X86::VDIVSSZrr_Intk:
case X86::VDIVSSZrr_Intkz:
case X86::VMAXSSZrr_Intk:
case X86::VMAXSSZrr_Intkz:
case X86::VMINSSZrr_Intk:
case X86::VMINSSZrr_Intkz:
case X86::VMULSSZrr_Intk:
case X86::VMULSSZrr_Intkz:
case X86::VSQRTSSZr_Intk:
case X86::VSQRTSSZr_Intkz:
case X86::VSUBSSZrr_Intk:
case X86::VSUBSSZrr_Intkz:
case X86::VFMADDSS4rr_Int:
case X86::VFNMADDSS4rr_Int:
case X86::VFMSUBSS4rr_Int:
case X86::VFNMSUBSS4rr_Int:
case X86::VFMADD132SSr_Int:
case X86::VFNMADD132SSr_Int:
case X86::VFMADD213SSr_Int:
case X86::VFNMADD213SSr_Int:
case X86::VFMADD231SSr_Int:
case X86::VFNMADD231SSr_Int:
case X86::VFMSUB132SSr_Int:
case X86::VFNMSUB132SSr_Int:
case X86::VFMSUB213SSr_Int:
case X86::VFNMSUB213SSr_Int:
case X86::VFMSUB231SSr_Int:
case X86::VFNMSUB231SSr_Int:
case X86::VFMADD132SSZr_Int:
case X86::VFNMADD132SSZr_Int:
case X86::VFMADD213SSZr_Int:
case X86::VFNMADD213SSZr_Int:
case X86::VFMADD231SSZr_Int:
case X86::VFNMADD231SSZr_Int:
case X86::VFMSUB132SSZr_Int:
case X86::VFNMSUB132SSZr_Int:
case X86::VFMSUB213SSZr_Int:
case X86::VFNMSUB213SSZr_Int:
case X86::VFMSUB231SSZr_Int:
case X86::VFNMSUB231SSZr_Int:
case X86::VFMADD132SSZr_Intk:
case X86::VFNMADD132SSZr_Intk:
case X86::VFMADD213SSZr_Intk:
case X86::VFNMADD213SSZr_Intk:
case X86::VFMADD231SSZr_Intk:
case X86::VFNMADD231SSZr_Intk:
case X86::VFMSUB132SSZr_Intk:
case X86::VFNMSUB132SSZr_Intk:
case X86::VFMSUB213SSZr_Intk:
case X86::VFNMSUB213SSZr_Intk:
case X86::VFMSUB231SSZr_Intk:
case X86::VFNMSUB231SSZr_Intk:
case X86::VFMADD132SSZr_Intkz:
case X86::VFNMADD132SSZr_Intkz:
case X86::VFMADD213SSZr_Intkz:
case X86::VFNMADD213SSZr_Intkz:
case X86::VFMADD231SSZr_Intkz:
case X86::VFNMADD231SSZr_Intkz:
case X86::VFMSUB132SSZr_Intkz:
case X86::VFNMSUB132SSZr_Intkz:
case X86::VFMSUB213SSZr_Intkz:
case X86::VFNMSUB213SSZr_Intkz:
case X86::VFMSUB231SSZr_Intkz:
case X86::VFNMSUB231SSZr_Intkz:
case X86::VFIXUPIMMSSZrri:
case X86::VFIXUPIMMSSZrrik:
case X86::VFIXUPIMMSSZrrikz:
case X86::VFPCLASSSSZri:
case X86::VFPCLASSSSZrik:
case X86::VGETEXPSSZr:
case X86::VGETEXPSSZrk:
case X86::VGETEXPSSZrkz:
case X86::VGETMANTSSZrri:
case X86::VGETMANTSSZrrik:
case X86::VGETMANTSSZrrikz:
case X86::VRANGESSZrri:
case X86::VRANGESSZrrik:
case X86::VRANGESSZrrikz:
case X86::VRCP14SSZrr:
case X86::VRCP14SSZrrk:
case X86::VRCP14SSZrrkz:
case X86::VRCP28SSZr:
case X86::VRCP28SSZrk:
case X86::VRCP28SSZrkz:
case X86::VREDUCESSZrri:
case X86::VREDUCESSZrrik:
case X86::VREDUCESSZrrikz:
case X86::VRNDSCALESSZr_Int:
case X86::VRNDSCALESSZr_Intk:
case X86::VRNDSCALESSZr_Intkz:
case X86::VRSQRT14SSZrr:
case X86::VRSQRT14SSZrrk:
case X86::VRSQRT14SSZrrkz:
case X86::VRSQRT28SSZr:
case X86::VRSQRT28SSZrk:
case X86::VRSQRT28SSZrkz:
case X86::VSCALEFSSZrr:
case X86::VSCALEFSSZrrk:
case X86::VSCALEFSSZrrkz:
return false;
default:
return true;
}
}
if ((Opc == X86::MOVSDrm || Opc == X86::VMOVSDrm || Opc == X86::VMOVSDZrm ||
Opc == X86::MOVSDrm_alt || Opc == X86::VMOVSDrm_alt ||
Opc == X86::VMOVSDZrm_alt) &&
RegSize > 64) {
// These instructions only load 64 bits, we can't fold them if the
// destination register is wider than 64 bits (8 bytes), and its user
// instruction isn't scalar (SD).
switch (UserOpc) {
case X86::CVTSD2SSrr_Int:
case X86::VCVTSD2SSrr_Int:
case X86::VCVTSD2SSZrr_Int:
case X86::VCVTSD2SSZrr_Intk:
case X86::VCVTSD2SSZrr_Intkz:
case X86::CVTSD2SIrr_Int:
case X86::CVTSD2SI64rr_Int:
case X86::VCVTSD2SIrr_Int:
case X86::VCVTSD2SI64rr_Int:
case X86::VCVTSD2SIZrr_Int:
case X86::VCVTSD2SI64Zrr_Int:
case X86::CVTTSD2SIrr_Int:
case X86::CVTTSD2SI64rr_Int:
case X86::VCVTTSD2SIrr_Int:
case X86::VCVTTSD2SI64rr_Int:
case X86::VCVTTSD2SIZrr_Int:
case X86::VCVTTSD2SI64Zrr_Int:
case X86::VCVTSD2USIZrr_Int:
case X86::VCVTSD2USI64Zrr_Int:
case X86::VCVTTSD2USIZrr_Int:
case X86::VCVTTSD2USI64Zrr_Int:
case X86::ROUNDSDri_Int:
case X86::VROUNDSDri_Int:
case X86::COMISDrr_Int:
case X86::VCOMISDrr_Int:
case X86::VCOMISDZrr_Int:
case X86::UCOMISDrr_Int:
case X86::VUCOMISDrr_Int:
case X86::VUCOMISDZrr_Int:
case X86::ADDSDrr_Int:
case X86::VADDSDrr_Int:
case X86::VADDSDZrr_Int:
case X86::CMPSDrri_Int:
case X86::VCMPSDrri_Int:
case X86::VCMPSDZrri_Int:
case X86::DIVSDrr_Int:
case X86::VDIVSDrr_Int:
case X86::VDIVSDZrr_Int:
case X86::MAXSDrr_Int:
case X86::VMAXSDrr_Int:
case X86::VMAXSDZrr_Int:
case X86::MINSDrr_Int:
case X86::VMINSDrr_Int:
case X86::VMINSDZrr_Int:
case X86::MULSDrr_Int:
case X86::VMULSDrr_Int:
case X86::VMULSDZrr_Int:
case X86::SQRTSDr_Int:
case X86::VSQRTSDr_Int:
case X86::VSQRTSDZr_Int:
case X86::SUBSDrr_Int:
case X86::VSUBSDrr_Int:
case X86::VSUBSDZrr_Int:
case X86::VADDSDZrr_Intk:
case X86::VADDSDZrr_Intkz:
case X86::VCMPSDZrri_Intk:
case X86::VDIVSDZrr_Intk:
case X86::VDIVSDZrr_Intkz:
case X86::VMAXSDZrr_Intk:
case X86::VMAXSDZrr_Intkz:
case X86::VMINSDZrr_Intk:
case X86::VMINSDZrr_Intkz:
case X86::VMULSDZrr_Intk:
case X86::VMULSDZrr_Intkz:
case X86::VSQRTSDZr_Intk:
case X86::VSQRTSDZr_Intkz:
case X86::VSUBSDZrr_Intk:
case X86::VSUBSDZrr_Intkz:
case X86::VFMADDSD4rr_Int:
case X86::VFNMADDSD4rr_Int:
case X86::VFMSUBSD4rr_Int:
case X86::VFNMSUBSD4rr_Int:
case X86::VFMADD132SDr_Int:
case X86::VFNMADD132SDr_Int:
case X86::VFMADD213SDr_Int:
case X86::VFNMADD213SDr_Int:
case X86::VFMADD231SDr_Int:
case X86::VFNMADD231SDr_Int:
case X86::VFMSUB132SDr_Int:
case X86::VFNMSUB132SDr_Int:
case X86::VFMSUB213SDr_Int:
case X86::VFNMSUB213SDr_Int:
case X86::VFMSUB231SDr_Int:
case X86::VFNMSUB231SDr_Int:
case X86::VFMADD132SDZr_Int:
case X86::VFNMADD132SDZr_Int:
case X86::VFMADD213SDZr_Int:
case X86::VFNMADD213SDZr_Int:
case X86::VFMADD231SDZr_Int:
case X86::VFNMADD231SDZr_Int:
case X86::VFMSUB132SDZr_Int:
case X86::VFNMSUB132SDZr_Int:
case X86::VFMSUB213SDZr_Int:
case X86::VFNMSUB213SDZr_Int:
case X86::VFMSUB231SDZr_Int:
case X86::VFNMSUB231SDZr_Int:
case X86::VFMADD132SDZr_Intk:
case X86::VFNMADD132SDZr_Intk:
case X86::VFMADD213SDZr_Intk:
case X86::VFNMADD213SDZr_Intk:
case X86::VFMADD231SDZr_Intk:
case X86::VFNMADD231SDZr_Intk:
case X86::VFMSUB132SDZr_Intk:
case X86::VFNMSUB132SDZr_Intk:
case X86::VFMSUB213SDZr_Intk:
case X86::VFNMSUB213SDZr_Intk:
case X86::VFMSUB231SDZr_Intk:
case X86::VFNMSUB231SDZr_Intk:
case X86::VFMADD132SDZr_Intkz:
case X86::VFNMADD132SDZr_Intkz:
case X86::VFMADD213SDZr_Intkz:
case X86::VFNMADD213SDZr_Intkz:
case X86::VFMADD231SDZr_Intkz:
case X86::VFNMADD231SDZr_Intkz:
case X86::VFMSUB132SDZr_Intkz:
case X86::VFNMSUB132SDZr_Intkz:
case X86::VFMSUB213SDZr_Intkz:
case X86::VFNMSUB213SDZr_Intkz:
case X86::VFMSUB231SDZr_Intkz:
case X86::VFNMSUB231SDZr_Intkz:
case X86::VFIXUPIMMSDZrri:
case X86::VFIXUPIMMSDZrrik:
case X86::VFIXUPIMMSDZrrikz:
case X86::VFPCLASSSDZri:
case X86::VFPCLASSSDZrik:
case X86::VGETEXPSDZr:
case X86::VGETEXPSDZrk:
case X86::VGETEXPSDZrkz:
case X86::VGETMANTSDZrri:
case X86::VGETMANTSDZrrik:
case X86::VGETMANTSDZrrikz:
case X86::VRANGESDZrri:
case X86::VRANGESDZrrik:
case X86::VRANGESDZrrikz:
case X86::VRCP14SDZrr:
case X86::VRCP14SDZrrk:
case X86::VRCP14SDZrrkz:
case X86::VRCP28SDZr:
case X86::VRCP28SDZrk:
case X86::VRCP28SDZrkz:
case X86::VREDUCESDZrri:
case X86::VREDUCESDZrrik:
case X86::VREDUCESDZrrikz:
case X86::VRNDSCALESDZr_Int:
case X86::VRNDSCALESDZr_Intk:
case X86::VRNDSCALESDZr_Intkz:
case X86::VRSQRT14SDZrr:
case X86::VRSQRT14SDZrrk:
case X86::VRSQRT14SDZrrkz:
case X86::VRSQRT28SDZr:
case X86::VRSQRT28SDZrk:
case X86::VRSQRT28SDZrkz:
case X86::VSCALEFSDZrr:
case X86::VSCALEFSDZrrk:
case X86::VSCALEFSDZrrkz:
return false;
default:
return true;
}
}
if ((Opc == X86::VMOVSHZrm || Opc == X86::VMOVSHZrm_alt) && RegSize > 16) {
// These instructions only load 16 bits, we can't fold them if the
// destination register is wider than 16 bits (2 bytes), and its user
// instruction isn't scalar (SH).
switch (UserOpc) {
case X86::VADDSHZrr_Int:
case X86::VCMPSHZrri_Int:
case X86::VDIVSHZrr_Int:
case X86::VMAXSHZrr_Int:
case X86::VMINSHZrr_Int:
case X86::VMULSHZrr_Int:
case X86::VSUBSHZrr_Int:
case X86::VADDSHZrr_Intk:
case X86::VADDSHZrr_Intkz:
case X86::VCMPSHZrri_Intk:
case X86::VDIVSHZrr_Intk:
case X86::VDIVSHZrr_Intkz:
case X86::VMAXSHZrr_Intk:
case X86::VMAXSHZrr_Intkz:
case X86::VMINSHZrr_Intk:
case X86::VMINSHZrr_Intkz:
case X86::VMULSHZrr_Intk:
case X86::VMULSHZrr_Intkz:
case X86::VSUBSHZrr_Intk:
case X86::VSUBSHZrr_Intkz:
case X86::VFMADD132SHZr_Int:
case X86::VFNMADD132SHZr_Int:
case X86::VFMADD213SHZr_Int:
case X86::VFNMADD213SHZr_Int:
case X86::VFMADD231SHZr_Int:
case X86::VFNMADD231SHZr_Int:
case X86::VFMSUB132SHZr_Int:
case X86::VFNMSUB132SHZr_Int:
case X86::VFMSUB213SHZr_Int:
case X86::VFNMSUB213SHZr_Int:
case X86::VFMSUB231SHZr_Int:
case X86::VFNMSUB231SHZr_Int:
case X86::VFMADD132SHZr_Intk:
case X86::VFNMADD132SHZr_Intk:
case X86::VFMADD213SHZr_Intk:
case X86::VFNMADD213SHZr_Intk:
case X86::VFMADD231SHZr_Intk:
case X86::VFNMADD231SHZr_Intk:
case X86::VFMSUB132SHZr_Intk:
case X86::VFNMSUB132SHZr_Intk:
case X86::VFMSUB213SHZr_Intk:
case X86::VFNMSUB213SHZr_Intk:
case X86::VFMSUB231SHZr_Intk:
case X86::VFNMSUB231SHZr_Intk:
case X86::VFMADD132SHZr_Intkz:
case X86::VFNMADD132SHZr_Intkz:
case X86::VFMADD213SHZr_Intkz:
case X86::VFNMADD213SHZr_Intkz:
case X86::VFMADD231SHZr_Intkz:
case X86::VFNMADD231SHZr_Intkz:
case X86::VFMSUB132SHZr_Intkz:
case X86::VFNMSUB132SHZr_Intkz:
case X86::VFMSUB213SHZr_Intkz:
case X86::VFNMSUB213SHZr_Intkz:
case X86::VFMSUB231SHZr_Intkz:
case X86::VFNMSUB231SHZr_Intkz:
return false;
default:
return true;
}
}
return false;
}
MachineInstr *X86InstrInfo::foldMemoryOperandImpl(
MachineFunction &MF, MachineInstr &MI, ArrayRef<unsigned> Ops,
MachineBasicBlock::iterator InsertPt, MachineInstr &LoadMI,
LiveIntervals *LIS) const {
// TODO: Support the case where LoadMI loads a wide register, but MI
// only uses a subreg.
for (auto Op : Ops) {
if (MI.getOperand(Op).getSubReg())
return nullptr;
}
// If loading from a FrameIndex, fold directly from the FrameIndex.
unsigned NumOps = LoadMI.getDesc().getNumOperands();
int FrameIndex;
if (isLoadFromStackSlot(LoadMI, FrameIndex)) {
if (isNonFoldablePartialRegisterLoad(LoadMI, MI, MF))
return nullptr;
return foldMemoryOperandImpl(MF, MI, Ops, InsertPt, FrameIndex, LIS);
}
// Check switch flag
if (NoFusing)
return nullptr;
// Avoid partial and undef register update stalls unless optimizing for size.
if (!MF.getFunction().hasOptSize() &&
(hasPartialRegUpdate(MI.getOpcode(), Subtarget, /*ForLoadFold*/ true) ||
shouldPreventUndefRegUpdateMemFold(MF, MI)))
return nullptr;
// Determine the alignment of the load.
Align Alignment;
unsigned LoadOpc = LoadMI.getOpcode();
if (LoadMI.hasOneMemOperand())
Alignment = (*LoadMI.memoperands_begin())->getAlign();
else
switch (LoadOpc) {
case X86::AVX512_512_SET0:
case X86::AVX512_512_SETALLONES:
Alignment = Align(64);
break;
case X86::AVX2_SETALLONES:
case X86::AVX1_SETALLONES:
case X86::AVX_SET0:
case X86::AVX512_256_SET0:
Alignment = Align(32);
break;
case X86::V_SET0:
case X86::V_SETALLONES:
case X86::AVX512_128_SET0:
case X86::FsFLD0F128:
case X86::AVX512_FsFLD0F128:
Alignment = Align(16);
break;
case X86::MMX_SET0:
case X86::FsFLD0SD:
case X86::AVX512_FsFLD0SD:
Alignment = Align(8);
break;
case X86::FsFLD0SS:
case X86::AVX512_FsFLD0SS:
Alignment = Align(4);
break;
case X86::FsFLD0SH:
case X86::AVX512_FsFLD0SH:
Alignment = Align(2);
break;
default:
return nullptr;
}
if (Ops.size() == 2 && Ops[0] == 0 && Ops[1] == 1) {
unsigned NewOpc = 0;
switch (MI.getOpcode()) {
default:
return nullptr;
case X86::TEST8rr:
NewOpc = X86::CMP8ri;
break;
case X86::TEST16rr:
NewOpc = X86::CMP16ri;
break;
case X86::TEST32rr:
NewOpc = X86::CMP32ri;
break;
case X86::TEST64rr:
NewOpc = X86::CMP64ri32;
break;
}
// Change to CMPXXri r, 0 first.
MI.setDesc(get(NewOpc));
MI.getOperand(1).ChangeToImmediate(0);
} else if (Ops.size() != 1)
return nullptr;
// Make sure the subregisters match.
// Otherwise we risk changing the size of the load.
if (LoadMI.getOperand(0).getSubReg() != MI.getOperand(Ops[0]).getSubReg())
return nullptr;
SmallVector<MachineOperand, X86::AddrNumOperands> MOs;
switch (LoadOpc) {
case X86::MMX_SET0:
case X86::V_SET0:
case X86::V_SETALLONES:
case X86::AVX2_SETALLONES:
case X86::AVX1_SETALLONES:
case X86::AVX_SET0:
case X86::AVX512_128_SET0:
case X86::AVX512_256_SET0:
case X86::AVX512_512_SET0:
case X86::AVX512_512_SETALLONES:
case X86::FsFLD0SH:
case X86::AVX512_FsFLD0SH:
case X86::FsFLD0SD:
case X86::AVX512_FsFLD0SD:
case X86::FsFLD0SS:
case X86::AVX512_FsFLD0SS:
case X86::FsFLD0F128:
case X86::AVX512_FsFLD0F128: {
// Folding a V_SET0 or V_SETALLONES as a load, to ease register pressure.
// Create a constant-pool entry and operands to load from it.
// Large code model can't fold loads this way.
if (MF.getTarget().getCodeModel() == CodeModel::Large)
return nullptr;
// x86-32 PIC requires a PIC base register for constant pools.
unsigned PICBase = 0;
// Since we're using Small or Kernel code model, we can always use
// RIP-relative addressing for a smaller encoding.
if (Subtarget.is64Bit()) {
PICBase = X86::RIP;
} else if (MF.getTarget().isPositionIndependent()) {
// FIXME: PICBase = getGlobalBaseReg(&MF);
// This doesn't work for several reasons.
// 1. GlobalBaseReg may have been spilled.
// 2. It may not be live at MI.
return nullptr;
}
// Create a constant-pool entry.
MachineConstantPool &MCP = *MF.getConstantPool();
Type *Ty;
bool IsAllOnes = false;
switch (LoadOpc) {
case X86::FsFLD0SS:
case X86::AVX512_FsFLD0SS:
Ty = Type::getFloatTy(MF.getFunction().getContext());
break;
case X86::FsFLD0SD:
case X86::AVX512_FsFLD0SD:
Ty = Type::getDoubleTy(MF.getFunction().getContext());
break;
case X86::FsFLD0F128:
case X86::AVX512_FsFLD0F128:
Ty = Type::getFP128Ty(MF.getFunction().getContext());
break;
case X86::FsFLD0SH:
case X86::AVX512_FsFLD0SH:
Ty = Type::getHalfTy(MF.getFunction().getContext());
break;
case X86::AVX512_512_SETALLONES:
IsAllOnes = true;
[[fallthrough]];
case X86::AVX512_512_SET0:
Ty = FixedVectorType::get(Type::getInt32Ty(MF.getFunction().getContext()),
16);
break;
case X86::AVX1_SETALLONES:
case X86::AVX2_SETALLONES:
IsAllOnes = true;
[[fallthrough]];
case X86::AVX512_256_SET0:
case X86::AVX_SET0:
Ty = FixedVectorType::get(Type::getInt32Ty(MF.getFunction().getContext()),
8);
break;
case X86::MMX_SET0:
Ty = FixedVectorType::get(Type::getInt32Ty(MF.getFunction().getContext()),
2);
break;
case X86::V_SETALLONES:
IsAllOnes = true;
[[fallthrough]];
case X86::V_SET0:
case X86::AVX512_128_SET0:
Ty = FixedVectorType::get(Type::getInt32Ty(MF.getFunction().getContext()),
4);
break;
}
const Constant *C =
IsAllOnes ? Constant::getAllOnesValue(Ty) : Constant::getNullValue(Ty);
unsigned CPI = MCP.getConstantPoolIndex(C, Alignment);
// Create operands to load from the constant pool entry.
MOs.push_back(MachineOperand::CreateReg(PICBase, false));
MOs.push_back(MachineOperand::CreateImm(1));
MOs.push_back(MachineOperand::CreateReg(0, false));
MOs.push_back(MachineOperand::CreateCPI(CPI, 0));
MOs.push_back(MachineOperand::CreateReg(0, false));
break;
}
case X86::VPBROADCASTBZ128rm:
case X86::VPBROADCASTBZ256rm:
case X86::VPBROADCASTBZrm:
case X86::VBROADCASTF32X2Z256rm:
case X86::VBROADCASTF32X2Zrm:
case X86::VBROADCASTI32X2Z128rm:
case X86::VBROADCASTI32X2Z256rm:
case X86::VBROADCASTI32X2Zrm:
// No instructions currently fuse with 8bits or 32bits x 2.
return nullptr;
#define FOLD_BROADCAST(SIZE) \
MOs.append(LoadMI.operands_begin() + NumOps - X86::AddrNumOperands, \
LoadMI.operands_begin() + NumOps); \
return foldMemoryBroadcast(MF, MI, Ops[0], MOs, InsertPt, /*Size=*/SIZE, \
/*AllowCommute=*/true);
case X86::VPBROADCASTWZ128rm:
case X86::VPBROADCASTWZ256rm:
case X86::VPBROADCASTWZrm:
FOLD_BROADCAST(16);
case X86::VPBROADCASTDZ128rm:
case X86::VPBROADCASTDZ256rm:
case X86::VPBROADCASTDZrm:
case X86::VBROADCASTSSZ128rm:
case X86::VBROADCASTSSZ256rm:
case X86::VBROADCASTSSZrm:
FOLD_BROADCAST(32);
case X86::VPBROADCASTQZ128rm:
case X86::VPBROADCASTQZ256rm:
case X86::VPBROADCASTQZrm:
case X86::VBROADCASTSDZ256rm:
case X86::VBROADCASTSDZrm:
FOLD_BROADCAST(64);
default: {
if (isNonFoldablePartialRegisterLoad(LoadMI, MI, MF))
return nullptr;
// Folding a normal load. Just copy the load's address operands.
MOs.append(LoadMI.operands_begin() + NumOps - X86::AddrNumOperands,
LoadMI.operands_begin() + NumOps);
break;
}
}
return foldMemoryOperandImpl(MF, MI, Ops[0], MOs, InsertPt,
/*Size=*/0, Alignment, /*AllowCommute=*/true);
}
MachineInstr *
X86InstrInfo::foldMemoryBroadcast(MachineFunction &MF, MachineInstr &MI,
unsigned OpNum, ArrayRef<MachineOperand> MOs,
MachineBasicBlock::iterator InsertPt,
unsigned BitsSize, bool AllowCommute) const {
if (auto *I = lookupBroadcastFoldTable(MI.getOpcode(), OpNum))
return matchBroadcastSize(*I, BitsSize)
? fuseInst(MF, I->DstOp, OpNum, MOs, InsertPt, MI, *this)
: nullptr;
if (AllowCommute) {
// If the instruction and target operand are commutable, commute the
// instruction and try again.
unsigned CommuteOpIdx2 = commuteOperandsForFold(MI, OpNum);
if (CommuteOpIdx2 == OpNum) {
printFailMsgforFold(MI, OpNum);
return nullptr;
}
MachineInstr *NewMI =
foldMemoryBroadcast(MF, MI, CommuteOpIdx2, MOs, InsertPt, BitsSize,
/*AllowCommute=*/false);
if (NewMI)
return NewMI;
// Folding failed again - undo the commute before returning.
commuteInstruction(MI, false, OpNum, CommuteOpIdx2);
}
printFailMsgforFold(MI, OpNum);
return nullptr;
}
static SmallVector<MachineMemOperand *, 2>
extractLoadMMOs(ArrayRef<MachineMemOperand *> MMOs, MachineFunction &MF) {
SmallVector<MachineMemOperand *, 2> LoadMMOs;
for (MachineMemOperand *MMO : MMOs) {
if (!MMO->isLoad())
continue;
if (!MMO->isStore()) {
// Reuse the MMO.
LoadMMOs.push_back(MMO);
} else {
// Clone the MMO and unset the store flag.
LoadMMOs.push_back(MF.getMachineMemOperand(
MMO, MMO->getFlags() & ~MachineMemOperand::MOStore));
}
}
return LoadMMOs;
}
static SmallVector<MachineMemOperand *, 2>
extractStoreMMOs(ArrayRef<MachineMemOperand *> MMOs, MachineFunction &MF) {
SmallVector<MachineMemOperand *, 2> StoreMMOs;
for (MachineMemOperand *MMO : MMOs) {
if (!MMO->isStore())
continue;
if (!MMO->isLoad()) {
// Reuse the MMO.
StoreMMOs.push_back(MMO);
} else {
// Clone the MMO and unset the load flag.
StoreMMOs.push_back(MF.getMachineMemOperand(
MMO, MMO->getFlags() & ~MachineMemOperand::MOLoad));
}
}
return StoreMMOs;
}
static unsigned getBroadcastOpcode(const X86FoldTableEntry *I,
const TargetRegisterClass *RC,
const X86Subtarget &STI) {
assert(STI.hasAVX512() && "Expected at least AVX512!");
unsigned SpillSize = STI.getRegisterInfo()->getSpillSize(*RC);
assert((SpillSize == 64 || STI.hasVLX()) &&
"Can't broadcast less than 64 bytes without AVX512VL!");
#define CASE_BCAST_TYPE_OPC(TYPE, OP16, OP32, OP64) \
case TYPE: \
switch (SpillSize) { \
default: \
llvm_unreachable("Unknown spill size"); \
case 16: \
return X86::OP16; \
case 32: \
return X86::OP32; \
case 64: \
return X86::OP64; \
} \
break;
switch (I->Flags & TB_BCAST_MASK) {
default:
llvm_unreachable("Unexpected broadcast type!");
CASE_BCAST_TYPE_OPC(TB_BCAST_W, VPBROADCASTWZ128rm, VPBROADCASTWZ256rm,
VPBROADCASTWZrm)
CASE_BCAST_TYPE_OPC(TB_BCAST_D, VPBROADCASTDZ128rm, VPBROADCASTDZ256rm,
VPBROADCASTDZrm)
CASE_BCAST_TYPE_OPC(TB_BCAST_Q, VPBROADCASTQZ128rm, VPBROADCASTQZ256rm,
VPBROADCASTQZrm)
CASE_BCAST_TYPE_OPC(TB_BCAST_SH, VPBROADCASTWZ128rm, VPBROADCASTWZ256rm,
VPBROADCASTWZrm)
CASE_BCAST_TYPE_OPC(TB_BCAST_SS, VBROADCASTSSZ128rm, VBROADCASTSSZ256rm,
VBROADCASTSSZrm)
CASE_BCAST_TYPE_OPC(TB_BCAST_SD, VMOVDDUPZ128rm, VBROADCASTSDZ256rm,
VBROADCASTSDZrm)
}
}
bool X86InstrInfo::unfoldMemoryOperand(
MachineFunction &MF, MachineInstr &MI, unsigned Reg, bool UnfoldLoad,
bool UnfoldStore, SmallVectorImpl<MachineInstr *> &NewMIs) const {
const X86FoldTableEntry *I = lookupUnfoldTable(MI.getOpcode());
if (I == nullptr)
return false;
unsigned Opc = I->DstOp;
unsigned Index = I->Flags & TB_INDEX_MASK;
bool FoldedLoad = I->Flags & TB_FOLDED_LOAD;
bool FoldedStore = I->Flags & TB_FOLDED_STORE;
if (UnfoldLoad && !FoldedLoad)
return false;
UnfoldLoad &= FoldedLoad;
if (UnfoldStore && !FoldedStore)
return false;
UnfoldStore &= FoldedStore;
const MCInstrDesc &MCID = get(Opc);
const TargetRegisterClass *RC = getRegClass(MCID, Index, &RI, MF);
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
// TODO: Check if 32-byte or greater accesses are slow too?
if (!MI.hasOneMemOperand() && RC == &X86::VR128RegClass &&
Subtarget.isUnalignedMem16Slow())
// Without memoperands, loadRegFromAddr and storeRegToStackSlot will
// conservatively assume the address is unaligned. That's bad for
// performance.
return false;
SmallVector<MachineOperand, X86::AddrNumOperands> AddrOps;
SmallVector<MachineOperand, 2> BeforeOps;
SmallVector<MachineOperand, 2> AfterOps;
SmallVector<MachineOperand, 4> ImpOps;
for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
MachineOperand &Op = MI.getOperand(i);
if (i >= Index && i < Index + X86::AddrNumOperands)
AddrOps.push_back(Op);
else if (Op.isReg() && Op.isImplicit())
ImpOps.push_back(Op);
else if (i < Index)
BeforeOps.push_back(Op);
else if (i > Index)
AfterOps.push_back(Op);
}
// Emit the load or broadcast instruction.
if (UnfoldLoad) {
auto MMOs = extractLoadMMOs(MI.memoperands(), MF);
unsigned Opc;
if (I->Flags & TB_BCAST_MASK) {
Opc = getBroadcastOpcode(I, RC, Subtarget);
} else {
unsigned Alignment = std::max<uint32_t>(TRI.getSpillSize(*RC), 16);
bool isAligned = !MMOs.empty() && MMOs.front()->getAlign() >= Alignment;
Opc = getLoadRegOpcode(Reg, RC, isAligned, Subtarget);
}
DebugLoc DL;
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), Reg);
for (const MachineOperand &AddrOp : AddrOps)
MIB.add(AddrOp);
MIB.setMemRefs(MMOs);
NewMIs.push_back(MIB);
if (UnfoldStore) {
// Address operands cannot be marked isKill.
for (unsigned i = 1; i != 1 + X86::AddrNumOperands; ++i) {
MachineOperand &MO = NewMIs[0]->getOperand(i);
if (MO.isReg())
MO.setIsKill(false);
}
}
}
// Emit the data processing instruction.
MachineInstr *DataMI = MF.CreateMachineInstr(MCID, MI.getDebugLoc(), true);
MachineInstrBuilder MIB(MF, DataMI);
if (FoldedStore)
MIB.addReg(Reg, RegState::Define);
for (MachineOperand &BeforeOp : BeforeOps)
MIB.add(BeforeOp);
if (FoldedLoad)
MIB.addReg(Reg);
for (MachineOperand &AfterOp : AfterOps)
MIB.add(AfterOp);
for (MachineOperand &ImpOp : ImpOps) {
MIB.addReg(ImpOp.getReg(), getDefRegState(ImpOp.isDef()) |
RegState::Implicit |
getKillRegState(ImpOp.isKill()) |
getDeadRegState(ImpOp.isDead()) |
getUndefRegState(ImpOp.isUndef()));
}
// Change CMP32ri r, 0 back to TEST32rr r, r, etc.
switch (DataMI->getOpcode()) {
default:
break;
case X86::CMP64ri32:
case X86::CMP32ri:
case X86::CMP16ri:
case X86::CMP8ri: {
MachineOperand &MO0 = DataMI->getOperand(0);
MachineOperand &MO1 = DataMI->getOperand(1);
if (MO1.isImm() && MO1.getImm() == 0) {
unsigned NewOpc;
switch (DataMI->getOpcode()) {
default:
llvm_unreachable("Unreachable!");
case X86::CMP64ri32:
NewOpc = X86::TEST64rr;
break;
case X86::CMP32ri:
NewOpc = X86::TEST32rr;
break;
case X86::CMP16ri:
NewOpc = X86::TEST16rr;
break;
case X86::CMP8ri:
NewOpc = X86::TEST8rr;
break;
}
DataMI->setDesc(get(NewOpc));
MO1.ChangeToRegister(MO0.getReg(), false);
}
}
}
NewMIs.push_back(DataMI);
// Emit the store instruction.
if (UnfoldStore) {
const TargetRegisterClass *DstRC = getRegClass(MCID, 0, &RI, MF);
auto MMOs = extractStoreMMOs(MI.memoperands(), MF);
unsigned Alignment = std::max<uint32_t>(TRI.getSpillSize(*DstRC), 16);
bool isAligned = !MMOs.empty() && MMOs.front()->getAlign() >= Alignment;
unsigned Opc = getStoreRegOpcode(Reg, DstRC, isAligned, Subtarget);
DebugLoc DL;
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc));
for (const MachineOperand &AddrOp : AddrOps)
MIB.add(AddrOp);
MIB.addReg(Reg, RegState::Kill);
MIB.setMemRefs(MMOs);
NewMIs.push_back(MIB);
}
return true;
}
bool X86InstrInfo::unfoldMemoryOperand(
SelectionDAG &DAG, SDNode *N, SmallVectorImpl<SDNode *> &NewNodes) const {
if (!N->isMachineOpcode())
return false;
const X86FoldTableEntry *I = lookupUnfoldTable(N->getMachineOpcode());
if (I == nullptr)
return false;
unsigned Opc = I->DstOp;
unsigned Index = I->Flags & TB_INDEX_MASK;
bool FoldedLoad = I->Flags & TB_FOLDED_LOAD;
bool FoldedStore = I->Flags & TB_FOLDED_STORE;
const MCInstrDesc &MCID = get(Opc);
MachineFunction &MF = DAG.getMachineFunction();
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = getRegClass(MCID, Index, &RI, MF);
unsigned NumDefs = MCID.NumDefs;
std::vector<SDValue> AddrOps;
std::vector<SDValue> BeforeOps;
std::vector<SDValue> AfterOps;
SDLoc dl(N);
unsigned NumOps = N->getNumOperands();
for (unsigned i = 0; i != NumOps - 1; ++i) {
SDValue Op = N->getOperand(i);
if (i >= Index - NumDefs && i < Index - NumDefs + X86::AddrNumOperands)
AddrOps.push_back(Op);
else if (i < Index - NumDefs)
BeforeOps.push_back(Op);
else if (i > Index - NumDefs)
AfterOps.push_back(Op);
}
SDValue Chain = N->getOperand(NumOps - 1);
AddrOps.push_back(Chain);
// Emit the load instruction.
SDNode *Load = nullptr;
if (FoldedLoad) {
EVT VT = *TRI.legalclasstypes_begin(*RC);
auto MMOs = extractLoadMMOs(cast<MachineSDNode>(N)->memoperands(), MF);
if (MMOs.empty() && RC == &X86::VR128RegClass &&
Subtarget.isUnalignedMem16Slow())
// Do not introduce a slow unaligned load.
return false;
// FIXME: If a VR128 can have size 32, we should be checking if a 32-byte
// memory access is slow above.
unsigned Opc;
if (I->Flags & TB_BCAST_MASK) {
Opc = getBroadcastOpcode(I, RC, Subtarget);
} else {
unsigned Alignment = std::max<uint32_t>(TRI.getSpillSize(*RC), 16);
bool isAligned = !MMOs.empty() && MMOs.front()->getAlign() >= Alignment;
Opc = getLoadRegOpcode(0, RC, isAligned, Subtarget);
}
Load = DAG.getMachineNode(Opc, dl, VT, MVT::Other, AddrOps);
NewNodes.push_back(Load);
// Preserve memory reference information.
DAG.setNodeMemRefs(cast<MachineSDNode>(Load), MMOs);
}
// Emit the data processing instruction.
std::vector<EVT> VTs;
const TargetRegisterClass *DstRC = nullptr;
if (MCID.getNumDefs() > 0) {
DstRC = getRegClass(MCID, 0, &RI, MF);
VTs.push_back(*TRI.legalclasstypes_begin(*DstRC));
}
for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
EVT VT = N->getValueType(i);
if (VT != MVT::Other && i >= (unsigned)MCID.getNumDefs())
VTs.push_back(VT);
}
if (Load)
BeforeOps.push_back(SDValue(Load, 0));
llvm::append_range(BeforeOps, AfterOps);
// Change CMP32ri r, 0 back to TEST32rr r, r, etc.
switch (Opc) {
default:
break;
case X86::CMP64ri32:
case X86::CMP32ri:
case X86::CMP16ri:
case X86::CMP8ri:
if (isNullConstant(BeforeOps[1])) {
switch (Opc) {
default:
llvm_unreachable("Unreachable!");
case X86::CMP64ri32:
Opc = X86::TEST64rr;
break;
case X86::CMP32ri:
Opc = X86::TEST32rr;
break;
case X86::CMP16ri:
Opc = X86::TEST16rr;
break;
case X86::CMP8ri:
Opc = X86::TEST8rr;
break;
}
BeforeOps[1] = BeforeOps[0];
}
}
SDNode *NewNode = DAG.getMachineNode(Opc, dl, VTs, BeforeOps);
NewNodes.push_back(NewNode);
// Emit the store instruction.
if (FoldedStore) {
AddrOps.pop_back();
AddrOps.push_back(SDValue(NewNode, 0));
AddrOps.push_back(Chain);
auto MMOs = extractStoreMMOs(cast<MachineSDNode>(N)->memoperands(), MF);
if (MMOs.empty() && RC == &X86::VR128RegClass &&
Subtarget.isUnalignedMem16Slow())
// Do not introduce a slow unaligned store.
return false;
// FIXME: If a VR128 can have size 32, we should be checking if a 32-byte
// memory access is slow above.
unsigned Alignment = std::max<uint32_t>(TRI.getSpillSize(*RC), 16);
bool isAligned = !MMOs.empty() && MMOs.front()->getAlign() >= Alignment;
SDNode *Store =
DAG.getMachineNode(getStoreRegOpcode(0, DstRC, isAligned, Subtarget),
dl, MVT::Other, AddrOps);
NewNodes.push_back(Store);
// Preserve memory reference information.
DAG.setNodeMemRefs(cast<MachineSDNode>(Store), MMOs);
}
return true;
}
unsigned
X86InstrInfo::getOpcodeAfterMemoryUnfold(unsigned Opc, bool UnfoldLoad,
bool UnfoldStore,
unsigned *LoadRegIndex) const {
const X86FoldTableEntry *I = lookupUnfoldTable(Opc);
if (I == nullptr)
return 0;
bool FoldedLoad = I->Flags & TB_FOLDED_LOAD;
bool FoldedStore = I->Flags & TB_FOLDED_STORE;
if (UnfoldLoad && !FoldedLoad)
return 0;
if (UnfoldStore && !FoldedStore)
return 0;
if (LoadRegIndex)
*LoadRegIndex = I->Flags & TB_INDEX_MASK;
return I->DstOp;
}
bool X86InstrInfo::areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
int64_t &Offset1,
int64_t &Offset2) const {
if (!Load1->isMachineOpcode() || !Load2->isMachineOpcode())
return false;
auto IsLoadOpcode = [&](unsigned Opcode) {
switch (Opcode) {
default:
return false;
case X86::MOV8rm:
case X86::MOV16rm:
case X86::MOV32rm:
case X86::MOV64rm:
case X86::LD_Fp32m:
case X86::LD_Fp64m:
case X86::LD_Fp80m:
case X86::MOVSSrm:
case X86::MOVSSrm_alt:
case X86::MOVSDrm:
case X86::MOVSDrm_alt:
case X86::MMX_MOVD64rm:
case X86::MMX_MOVQ64rm:
case X86::MOVAPSrm:
case X86::MOVUPSrm:
case X86::MOVAPDrm:
case X86::MOVUPDrm:
case X86::MOVDQArm:
case X86::MOVDQUrm:
// AVX load instructions
case X86::VMOVSSrm:
case X86::VMOVSSrm_alt:
case X86::VMOVSDrm:
case X86::VMOVSDrm_alt:
case X86::VMOVAPSrm:
case X86::VMOVUPSrm:
case X86::VMOVAPDrm:
case X86::VMOVUPDrm:
case X86::VMOVDQArm:
case X86::VMOVDQUrm:
case X86::VMOVAPSYrm:
case X86::VMOVUPSYrm:
case X86::VMOVAPDYrm:
case X86::VMOVUPDYrm:
case X86::VMOVDQAYrm:
case X86::VMOVDQUYrm:
// AVX512 load instructions
case X86::VMOVSSZrm:
case X86::VMOVSSZrm_alt:
case X86::VMOVSDZrm:
case X86::VMOVSDZrm_alt:
case X86::VMOVAPSZ128rm:
case X86::VMOVUPSZ128rm:
case X86::VMOVAPSZ128rm_NOVLX:
case X86::VMOVUPSZ128rm_NOVLX:
case X86::VMOVAPDZ128rm:
case X86::VMOVUPDZ128rm:
case X86::VMOVDQU8Z128rm:
case X86::VMOVDQU16Z128rm:
case X86::VMOVDQA32Z128rm:
case X86::VMOVDQU32Z128rm:
case X86::VMOVDQA64Z128rm:
case X86::VMOVDQU64Z128rm:
case X86::VMOVAPSZ256rm:
case X86::VMOVUPSZ256rm:
case X86::VMOVAPSZ256rm_NOVLX:
case X86::VMOVUPSZ256rm_NOVLX:
case X86::VMOVAPDZ256rm:
case X86::VMOVUPDZ256rm:
case X86::VMOVDQU8Z256rm:
case X86::VMOVDQU16Z256rm:
case X86::VMOVDQA32Z256rm:
case X86::VMOVDQU32Z256rm:
case X86::VMOVDQA64Z256rm:
case X86::VMOVDQU64Z256rm:
case X86::VMOVAPSZrm:
case X86::VMOVUPSZrm:
case X86::VMOVAPDZrm:
case X86::VMOVUPDZrm:
case X86::VMOVDQU8Zrm:
case X86::VMOVDQU16Zrm:
case X86::VMOVDQA32Zrm:
case X86::VMOVDQU32Zrm:
case X86::VMOVDQA64Zrm:
case X86::VMOVDQU64Zrm:
case X86::KMOVBkm:
case X86::KMOVBkm_EVEX:
case X86::KMOVWkm:
case X86::KMOVWkm_EVEX:
case X86::KMOVDkm:
case X86::KMOVDkm_EVEX:
case X86::KMOVQkm:
case X86::KMOVQkm_EVEX:
return true;
}
};
if (!IsLoadOpcode(Load1->getMachineOpcode()) ||
!IsLoadOpcode(Load2->getMachineOpcode()))
return false;
// Lambda to check if both the loads have the same value for an operand index.
auto HasSameOp = [&](int I) {
return Load1->getOperand(I) == Load2->getOperand(I);
};
// All operands except the displacement should match.
if (!HasSameOp(X86::AddrBaseReg) || !HasSameOp(X86::AddrScaleAmt) ||
!HasSameOp(X86::AddrIndexReg) || !HasSameOp(X86::AddrSegmentReg))
return false;
// Chain Operand must be the same.
if (!HasSameOp(5))
return false;
// Now let's examine if the displacements are constants.
auto Disp1 = dyn_cast<ConstantSDNode>(Load1->getOperand(X86::AddrDisp));
auto Disp2 = dyn_cast<ConstantSDNode>(Load2->getOperand(X86::AddrDisp));
if (!Disp1 || !Disp2)
return false;
Offset1 = Disp1->getSExtValue();
Offset2 = Disp2->getSExtValue();
return true;
}
bool X86InstrInfo::shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
int64_t Offset1, int64_t Offset2,
unsigned NumLoads) const {
assert(Offset2 > Offset1);
if ((Offset2 - Offset1) / 8 > 64)
return false;
unsigned Opc1 = Load1->getMachineOpcode();
unsigned Opc2 = Load2->getMachineOpcode();
if (Opc1 != Opc2)
return false; // FIXME: overly conservative?
switch (Opc1) {
default:
break;
case X86::LD_Fp32m:
case X86::LD_Fp64m:
case X86::LD_Fp80m:
case X86::MMX_MOVD64rm:
case X86::MMX_MOVQ64rm:
return false;
}
EVT VT = Load1->getValueType(0);
switch (VT.getSimpleVT().SimpleTy) {
default:
// XMM registers. In 64-bit mode we can be a bit more aggressive since we
// have 16 of them to play with.
if (Subtarget.is64Bit()) {
if (NumLoads >= 3)
return false;
} else if (NumLoads) {
return false;
}
break;
case MVT::i8:
case MVT::i16:
case MVT::i32:
case MVT::i64:
case MVT::f32:
case MVT::f64:
if (NumLoads)
return false;
break;
}
return true;
}
bool X86InstrInfo::isSchedulingBoundary(const MachineInstr &MI,
const MachineBasicBlock *MBB,
const MachineFunction &MF) const {
// ENDBR instructions should not be scheduled around.
unsigned Opcode = MI.getOpcode();
if (Opcode == X86::ENDBR64 || Opcode == X86::ENDBR32 ||
Opcode == X86::PLDTILECFGV)
return true;
// Frame setup and destory can't be scheduled around.
if (MI.getFlag(MachineInstr::FrameSetup) ||
MI.getFlag(MachineInstr::FrameDestroy))
return true;
return TargetInstrInfo::isSchedulingBoundary(MI, MBB, MF);
}
bool X86InstrInfo::reverseBranchCondition(
SmallVectorImpl<MachineOperand> &Cond) const {
assert(Cond.size() == 1 && "Invalid X86 branch condition!");
X86::CondCode CC = static_cast<X86::CondCode>(Cond[0].getImm());
Cond[0].setImm(GetOppositeBranchCondition(CC));
return false;
}
bool X86InstrInfo::isSafeToMoveRegClassDefs(
const TargetRegisterClass *RC) const {
// FIXME: Return false for x87 stack register classes for now. We can't
// allow any loads of these registers before FpGet_ST0_80.
return !(RC == &X86::CCRRegClass || RC == &X86::DFCCRRegClass ||
RC == &X86::RFP32RegClass || RC == &X86::RFP64RegClass ||
RC == &X86::RFP80RegClass);
}
/// Return a virtual register initialized with the
/// the global base register value. Output instructions required to
/// initialize the register in the function entry block, if necessary.
///
/// TODO: Eliminate this and move the code to X86MachineFunctionInfo.
///
unsigned X86InstrInfo::getGlobalBaseReg(MachineFunction *MF) const {
X86MachineFunctionInfo *X86FI = MF->getInfo<X86MachineFunctionInfo>();
Register GlobalBaseReg = X86FI->getGlobalBaseReg();
if (GlobalBaseReg != 0)
return GlobalBaseReg;
// Create the register. The code to initialize it is inserted
// later, by the CGBR pass (below).
MachineRegisterInfo &RegInfo = MF->getRegInfo();
GlobalBaseReg = RegInfo.createVirtualRegister(
Subtarget.is64Bit() ? &X86::GR64_NOSPRegClass : &X86::GR32_NOSPRegClass);
X86FI->setGlobalBaseReg(GlobalBaseReg);
return GlobalBaseReg;
}
// FIXME: Some shuffle and unpack instructions have equivalents in different
// domains, but they require a bit more work than just switching opcodes.
static const uint16_t *lookup(unsigned opcode, unsigned domain,
ArrayRef<uint16_t[3]> Table) {
for (const uint16_t(&Row)[3] : Table)
if (Row[domain - 1] == opcode)
return Row;
return nullptr;
}
static const uint16_t *lookupAVX512(unsigned opcode, unsigned domain,
ArrayRef<uint16_t[4]> Table) {
// If this is the integer domain make sure to check both integer columns.
for (const uint16_t(&Row)[4] : Table)
if (Row[domain - 1] == opcode || (domain == 3 && Row[3] == opcode))
return Row;
return nullptr;
}
// Helper to attempt to widen/narrow blend masks.
static bool AdjustBlendMask(unsigned OldMask, unsigned OldWidth,
unsigned NewWidth, unsigned *pNewMask = nullptr) {
assert(((OldWidth % NewWidth) == 0 || (NewWidth % OldWidth) == 0) &&
"Illegal blend mask scale");
unsigned NewMask = 0;
if ((OldWidth % NewWidth) == 0) {
unsigned Scale = OldWidth / NewWidth;
unsigned SubMask = (1u << Scale) - 1;
for (unsigned i = 0; i != NewWidth; ++i) {
unsigned Sub = (OldMask >> (i * Scale)) & SubMask;
if (Sub == SubMask)
NewMask |= (1u << i);
else if (Sub != 0x0)
return false;
}
} else {
unsigned Scale = NewWidth / OldWidth;
unsigned SubMask = (1u << Scale) - 1;
for (unsigned i = 0; i != OldWidth; ++i) {
if (OldMask & (1 << i)) {
NewMask |= (SubMask << (i * Scale));
}
}
}
if (pNewMask)
*pNewMask = NewMask;
return true;
}
uint16_t X86InstrInfo::getExecutionDomainCustom(const MachineInstr &MI) const {
unsigned Opcode = MI.getOpcode();
unsigned NumOperands = MI.getDesc().getNumOperands();
auto GetBlendDomains = [&](unsigned ImmWidth, bool Is256) {
uint16_t validDomains = 0;
if (MI.getOperand(NumOperands - 1).isImm()) {
unsigned Imm = MI.getOperand(NumOperands - 1).getImm();
if (AdjustBlendMask(Imm, ImmWidth, Is256 ? 8 : 4))
validDomains |= 0x2; // PackedSingle
if (AdjustBlendMask(Imm, ImmWidth, Is256 ? 4 : 2))
validDomains |= 0x4; // PackedDouble
if (!Is256 || Subtarget.hasAVX2())
validDomains |= 0x8; // PackedInt
}
return validDomains;
};
switch (Opcode) {
case X86::BLENDPDrmi:
case X86::BLENDPDrri:
case X86::VBLENDPDrmi:
case X86::VBLENDPDrri:
return GetBlendDomains(2, false);
case X86::VBLENDPDYrmi:
case X86::VBLENDPDYrri:
return GetBlendDomains(4, true);
case X86::BLENDPSrmi:
case X86::BLENDPSrri:
case X86::VBLENDPSrmi:
case X86::VBLENDPSrri:
case X86::VPBLENDDrmi:
case X86::VPBLENDDrri:
return GetBlendDomains(4, false);
case X86::VBLENDPSYrmi:
case X86::VBLENDPSYrri:
case X86::VPBLENDDYrmi:
case X86::VPBLENDDYrri:
return GetBlendDomains(8, true);
case X86::PBLENDWrmi:
case X86::PBLENDWrri:
case X86::VPBLENDWrmi:
case X86::VPBLENDWrri:
// Treat VPBLENDWY as a 128-bit vector as it repeats the lo/hi masks.
case X86::VPBLENDWYrmi:
case X86::VPBLENDWYrri:
return GetBlendDomains(8, false);
case X86::VPANDDZ128rr:
case X86::VPANDDZ128rm:
case X86::VPANDDZ256rr:
case X86::VPANDDZ256rm:
case X86::VPANDQZ128rr:
case X86::VPANDQZ128rm:
case X86::VPANDQZ256rr:
case X86::VPANDQZ256rm:
case X86::VPANDNDZ128rr:
case X86::VPANDNDZ128rm:
case X86::VPANDNDZ256rr:
case X86::VPANDNDZ256rm:
case X86::VPANDNQZ128rr:
case X86::VPANDNQZ128rm:
case X86::VPANDNQZ256rr:
case X86::VPANDNQZ256rm:
case X86::VPORDZ128rr:
case X86::VPORDZ128rm:
case X86::VPORDZ256rr:
case X86::VPORDZ256rm:
case X86::VPORQZ128rr:
case X86::VPORQZ128rm:
case X86::VPORQZ256rr:
case X86::VPORQZ256rm:
case X86::VPXORDZ128rr:
case X86::VPXORDZ128rm:
case X86::VPXORDZ256rr:
case X86::VPXORDZ256rm:
case X86::VPXORQZ128rr:
case X86::VPXORQZ128rm:
case X86::VPXORQZ256rr:
case X86::VPXORQZ256rm:
// If we don't have DQI see if we can still switch from an EVEX integer
// instruction to a VEX floating point instruction.
if (Subtarget.hasDQI())
return 0;
if (RI.getEncodingValue(MI.getOperand(0).getReg()) >= 16)
return 0;
if (RI.getEncodingValue(MI.getOperand(1).getReg()) >= 16)
return 0;
// Register forms will have 3 operands. Memory form will have more.
if (NumOperands == 3 &&
RI.getEncodingValue(MI.getOperand(2).getReg()) >= 16)
return 0;
// All domains are valid.
return 0xe;
case X86::MOVHLPSrr:
// We can swap domains when both inputs are the same register.
// FIXME: This doesn't catch all the cases we would like. If the input
// register isn't KILLed by the instruction, the two address instruction
// pass puts a COPY on one input. The other input uses the original
// register. This prevents the same physical register from being used by
// both inputs.
if (MI.getOperand(1).getReg() == MI.getOperand(2).getReg() &&
MI.getOperand(0).getSubReg() == 0 &&
MI.getOperand(1).getSubReg() == 0 && MI.getOperand(2).getSubReg() == 0)
return 0x6;
return 0;
case X86::SHUFPDrri:
return 0x6;
}
return 0;
}
#include "X86ReplaceableInstrs.def"
bool X86InstrInfo::setExecutionDomainCustom(MachineInstr &MI,
unsigned Domain) const {
assert(Domain > 0 && Domain < 4 && "Invalid execution domain");
uint16_t dom = (MI.getDesc().TSFlags >> X86II::SSEDomainShift) & 3;
assert(dom && "Not an SSE instruction");
unsigned Opcode = MI.getOpcode();
unsigned NumOperands = MI.getDesc().getNumOperands();
auto SetBlendDomain = [&](unsigned ImmWidth, bool Is256) {
if (MI.getOperand(NumOperands - 1).isImm()) {
unsigned Imm = MI.getOperand(NumOperands - 1).getImm() & 255;
Imm = (ImmWidth == 16 ? ((Imm << 8) | Imm) : Imm);
unsigned NewImm = Imm;
const uint16_t *table = lookup(Opcode, dom, ReplaceableBlendInstrs);
if (!table)
table = lookup(Opcode, dom, ReplaceableBlendAVX2Instrs);
if (Domain == 1) { // PackedSingle
AdjustBlendMask(Imm, ImmWidth, Is256 ? 8 : 4, &NewImm);
} else if (Domain == 2) { // PackedDouble
AdjustBlendMask(Imm, ImmWidth, Is256 ? 4 : 2, &NewImm);
} else if (Domain == 3) { // PackedInt
if (Subtarget.hasAVX2()) {
// If we are already VPBLENDW use that, else use VPBLENDD.
if ((ImmWidth / (Is256 ? 2 : 1)) != 8) {
table = lookup(Opcode, dom, ReplaceableBlendAVX2Instrs);
AdjustBlendMask(Imm, ImmWidth, Is256 ? 8 : 4, &NewImm);
}
} else {
assert(!Is256 && "128-bit vector expected");
AdjustBlendMask(Imm, ImmWidth, 8, &NewImm);
}
}
assert(table && table[Domain - 1] && "Unknown domain op");
MI.setDesc(get(table[Domain - 1]));
MI.getOperand(NumOperands - 1).setImm(NewImm & 255);
}
return true;
};
switch (Opcode) {
case X86::BLENDPDrmi:
case X86::BLENDPDrri:
case X86::VBLENDPDrmi:
case X86::VBLENDPDrri:
return SetBlendDomain(2, false);
case X86::VBLENDPDYrmi:
case X86::VBLENDPDYrri:
return SetBlendDomain(4, true);
case X86::BLENDPSrmi:
case X86::BLENDPSrri:
case X86::VBLENDPSrmi:
case X86::VBLENDPSrri:
case X86::VPBLENDDrmi:
case X86::VPBLENDDrri:
return SetBlendDomain(4, false);
case X86::VBLENDPSYrmi:
case X86::VBLENDPSYrri:
case X86::VPBLENDDYrmi:
case X86::VPBLENDDYrri:
return SetBlendDomain(8, true);
case X86::PBLENDWrmi:
case X86::PBLENDWrri:
case X86::VPBLENDWrmi:
case X86::VPBLENDWrri:
return SetBlendDomain(8, false);
case X86::VPBLENDWYrmi:
case X86::VPBLENDWYrri:
return SetBlendDomain(16, true);
case X86::VPANDDZ128rr:
case X86::VPANDDZ128rm:
case X86::VPANDDZ256rr:
case X86::VPANDDZ256rm:
case X86::VPANDQZ128rr:
case X86::VPANDQZ128rm:
case X86::VPANDQZ256rr:
case X86::VPANDQZ256rm:
case X86::VPANDNDZ128rr:
case X86::VPANDNDZ128rm:
case X86::VPANDNDZ256rr:
case X86::VPANDNDZ256rm:
case X86::VPANDNQZ128rr:
case X86::VPANDNQZ128rm:
case X86::VPANDNQZ256rr:
case X86::VPANDNQZ256rm:
case X86::VPORDZ128rr:
case X86::VPORDZ128rm:
case X86::VPORDZ256rr:
case X86::VPORDZ256rm:
case X86::VPORQZ128rr:
case X86::VPORQZ128rm:
case X86::VPORQZ256rr:
case X86::VPORQZ256rm:
case X86::VPXORDZ128rr:
case X86::VPXORDZ128rm:
case X86::VPXORDZ256rr:
case X86::VPXORDZ256rm:
case X86::VPXORQZ128rr:
case X86::VPXORQZ128rm:
case X86::VPXORQZ256rr:
case X86::VPXORQZ256rm: {
// Without DQI, convert EVEX instructions to VEX instructions.
if (Subtarget.hasDQI())
return false;
const uint16_t *table =
lookupAVX512(MI.getOpcode(), dom, ReplaceableCustomAVX512LogicInstrs);
assert(table && "Instruction not found in table?");
// Don't change integer Q instructions to D instructions and
// use D intructions if we started with a PS instruction.
if (Domain == 3 && (dom == 1 || table[3] == MI.getOpcode()))
Domain = 4;
MI.setDesc(get(table[Domain - 1]));
return true;
}
case X86::UNPCKHPDrr:
case X86::MOVHLPSrr:
// We just need to commute the instruction which will switch the domains.
if (Domain != dom && Domain != 3 &&
MI.getOperand(1).getReg() == MI.getOperand(2).getReg() &&
MI.getOperand(0).getSubReg() == 0 &&
MI.getOperand(1).getSubReg() == 0 &&
MI.getOperand(2).getSubReg() == 0) {
commuteInstruction(MI, false);
return true;
}
// We must always return true for MOVHLPSrr.
if (Opcode == X86::MOVHLPSrr)
return true;
break;
case X86::SHUFPDrri: {
if (Domain == 1) {
unsigned Imm = MI.getOperand(3).getImm();
unsigned NewImm = 0x44;
if (Imm & 1)
NewImm |= 0x0a;
if (Imm & 2)
NewImm |= 0xa0;
MI.getOperand(3).setImm(NewImm);
MI.setDesc(get(X86::SHUFPSrri));
}
return true;
}
}
return false;
}
std::pair<uint16_t, uint16_t>
X86InstrInfo::getExecutionDomain(const MachineInstr &MI) const {
uint16_t domain = (MI.getDesc().TSFlags >> X86II::SSEDomainShift) & 3;
unsigned opcode = MI.getOpcode();
uint16_t validDomains = 0;
if (domain) {
// Attempt to match for custom instructions.
validDomains = getExecutionDomainCustom(MI);
if (validDomains)
return std::make_pair(domain, validDomains);
if (lookup(opcode, domain, ReplaceableInstrs)) {
validDomains = 0xe;
} else if (lookup(opcode, domain, ReplaceableInstrsAVX2)) {
validDomains = Subtarget.hasAVX2() ? 0xe : 0x6;
} else if (lookup(opcode, domain, ReplaceableInstrsFP)) {
validDomains = 0x6;
} else if (lookup(opcode, domain, ReplaceableInstrsAVX2InsertExtract)) {
// Insert/extract instructions should only effect domain if AVX2
// is enabled.
if (!Subtarget.hasAVX2())
return std::make_pair(0, 0);
validDomains = 0xe;
} else if (lookupAVX512(opcode, domain, ReplaceableInstrsAVX512)) {
validDomains = 0xe;
} else if (Subtarget.hasDQI() &&
lookupAVX512(opcode, domain, ReplaceableInstrsAVX512DQ)) {
validDomains = 0xe;
} else if (Subtarget.hasDQI()) {
if (const uint16_t *table =
lookupAVX512(opcode, domain, ReplaceableInstrsAVX512DQMasked)) {
if (domain == 1 || (domain == 3 && table[3] == opcode))
validDomains = 0xa;
else
validDomains = 0xc;
}
}
}
return std::make_pair(domain, validDomains);
}
void X86InstrInfo::setExecutionDomain(MachineInstr &MI, unsigned Domain) const {
assert(Domain > 0 && Domain < 4 && "Invalid execution domain");
uint16_t dom = (MI.getDesc().TSFlags >> X86II::SSEDomainShift) & 3;
assert(dom && "Not an SSE instruction");
// Attempt to match for custom instructions.
if (setExecutionDomainCustom(MI, Domain))
return;
const uint16_t *table = lookup(MI.getOpcode(), dom, ReplaceableInstrs);
if (!table) { // try the other table
assert((Subtarget.hasAVX2() || Domain < 3) &&
"256-bit vector operations only available in AVX2");
table = lookup(MI.getOpcode(), dom, ReplaceableInstrsAVX2);
}
if (!table) { // try the FP table
table = lookup(MI.getOpcode(), dom, ReplaceableInstrsFP);
assert((!table || Domain < 3) &&
"Can only select PackedSingle or PackedDouble");
}
if (!table) { // try the other table
assert(Subtarget.hasAVX2() &&
"256-bit insert/extract only available in AVX2");
table = lookup(MI.getOpcode(), dom, ReplaceableInstrsAVX2InsertExtract);
}
if (!table) { // try the AVX512 table
assert(Subtarget.hasAVX512() && "Requires AVX-512");
table = lookupAVX512(MI.getOpcode(), dom, ReplaceableInstrsAVX512);
// Don't change integer Q instructions to D instructions.
if (table && Domain == 3 && table[3] == MI.getOpcode())
Domain = 4;
}
if (!table) { // try the AVX512DQ table
assert((Subtarget.hasDQI() || Domain >= 3) && "Requires AVX-512DQ");
table = lookupAVX512(MI.getOpcode(), dom, ReplaceableInstrsAVX512DQ);
// Don't change integer Q instructions to D instructions and
// use D instructions if we started with a PS instruction.
if (table && Domain == 3 && (dom == 1 || table[3] == MI.getOpcode()))
Domain = 4;
}
if (!table) { // try the AVX512DQMasked table
assert((Subtarget.hasDQI() || Domain >= 3) && "Requires AVX-512DQ");
table = lookupAVX512(MI.getOpcode(), dom, ReplaceableInstrsAVX512DQMasked);
if (table && Domain == 3 && (dom == 1 || table[3] == MI.getOpcode()))
Domain = 4;
}
assert(table && "Cannot change domain");
MI.setDesc(get(table[Domain - 1]));
}
void X86InstrInfo::insertNoop(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI) const {
DebugLoc DL;
BuildMI(MBB, MI, DL, get(X86::NOOP));
}
/// Return the noop instruction to use for a noop.
MCInst X86InstrInfo::getNop() const {
MCInst Nop;
Nop.setOpcode(X86::NOOP);
return Nop;
}
bool X86InstrInfo::isHighLatencyDef(int opc) const {
switch (opc) {
default:
return false;
case X86::DIVPDrm:
case X86::DIVPDrr:
case X86::DIVPSrm:
case X86::DIVPSrr:
case X86::DIVSDrm:
case X86::DIVSDrm_Int:
case X86::DIVSDrr:
case X86::DIVSDrr_Int:
case X86::DIVSSrm:
case X86::DIVSSrm_Int:
case X86::DIVSSrr:
case X86::DIVSSrr_Int:
case X86::SQRTPDm:
case X86::SQRTPDr:
case X86::SQRTPSm:
case X86::SQRTPSr:
case X86::SQRTSDm:
case X86::SQRTSDm_Int:
case X86::SQRTSDr:
case X86::SQRTSDr_Int:
case X86::SQRTSSm:
case X86::SQRTSSm_Int:
case X86::SQRTSSr:
case X86::SQRTSSr_Int:
// AVX instructions with high latency
case X86::VDIVPDrm:
case X86::VDIVPDrr:
case X86::VDIVPDYrm:
case X86::VDIVPDYrr:
case X86::VDIVPSrm:
case X86::VDIVPSrr:
case X86::VDIVPSYrm:
case X86::VDIVPSYrr:
case X86::VDIVSDrm:
case X86::VDIVSDrm_Int:
case X86::VDIVSDrr:
case X86::VDIVSDrr_Int:
case X86::VDIVSSrm:
case X86::VDIVSSrm_Int:
case X86::VDIVSSrr:
case X86::VDIVSSrr_Int:
case X86::VSQRTPDm:
case X86::VSQRTPDr:
case X86::VSQRTPDYm:
case X86::VSQRTPDYr:
case X86::VSQRTPSm:
case X86::VSQRTPSr:
case X86::VSQRTPSYm:
case X86::VSQRTPSYr:
case X86::VSQRTSDm:
case X86::VSQRTSDm_Int:
case X86::VSQRTSDr:
case X86::VSQRTSDr_Int:
case X86::VSQRTSSm:
case X86::VSQRTSSm_Int:
case X86::VSQRTSSr:
case X86::VSQRTSSr_Int:
// AVX512 instructions with high latency
case X86::VDIVPDZ128rm:
case X86::VDIVPDZ128rmb:
case X86::VDIVPDZ128rmbk:
case X86::VDIVPDZ128rmbkz:
case X86::VDIVPDZ128rmk:
case X86::VDIVPDZ128rmkz:
case X86::VDIVPDZ128rr:
case X86::VDIVPDZ128rrk:
case X86::VDIVPDZ128rrkz:
case X86::VDIVPDZ256rm:
case X86::VDIVPDZ256rmb:
case X86::VDIVPDZ256rmbk:
case X86::VDIVPDZ256rmbkz:
case X86::VDIVPDZ256rmk:
case X86::VDIVPDZ256rmkz:
case X86::VDIVPDZ256rr:
case X86::VDIVPDZ256rrk:
case X86::VDIVPDZ256rrkz:
case X86::VDIVPDZrrb:
case X86::VDIVPDZrrbk:
case X86::VDIVPDZrrbkz:
case X86::VDIVPDZrm:
case X86::VDIVPDZrmb:
case X86::VDIVPDZrmbk:
case X86::VDIVPDZrmbkz:
case X86::VDIVPDZrmk:
case X86::VDIVPDZrmkz:
case X86::VDIVPDZrr:
case X86::VDIVPDZrrk:
case X86::VDIVPDZrrkz:
case X86::VDIVPSZ128rm:
case X86::VDIVPSZ128rmb:
case X86::VDIVPSZ128rmbk:
case X86::VDIVPSZ128rmbkz:
case X86::VDIVPSZ128rmk:
case X86::VDIVPSZ128rmkz:
case X86::VDIVPSZ128rr:
case X86::VDIVPSZ128rrk:
case X86::VDIVPSZ128rrkz:
case X86::VDIVPSZ256rm:
case X86::VDIVPSZ256rmb:
case X86::VDIVPSZ256rmbk:
case X86::VDIVPSZ256rmbkz:
case X86::VDIVPSZ256rmk:
case X86::VDIVPSZ256rmkz:
case X86::VDIVPSZ256rr:
case X86::VDIVPSZ256rrk:
case X86::VDIVPSZ256rrkz:
case X86::VDIVPSZrrb:
case X86::VDIVPSZrrbk:
case X86::VDIVPSZrrbkz:
case X86::VDIVPSZrm:
case X86::VDIVPSZrmb:
case X86::VDIVPSZrmbk:
case X86::VDIVPSZrmbkz:
case X86::VDIVPSZrmk:
case X86::VDIVPSZrmkz:
case X86::VDIVPSZrr:
case X86::VDIVPSZrrk:
case X86::VDIVPSZrrkz:
case X86::VDIVSDZrm:
case X86::VDIVSDZrr:
case X86::VDIVSDZrm_Int:
case X86::VDIVSDZrm_Intk:
case X86::VDIVSDZrm_Intkz:
case X86::VDIVSDZrr_Int:
case X86::VDIVSDZrr_Intk:
case X86::VDIVSDZrr_Intkz:
case X86::VDIVSDZrrb_Int:
case X86::VDIVSDZrrb_Intk:
case X86::VDIVSDZrrb_Intkz:
case X86::VDIVSSZrm:
case X86::VDIVSSZrr:
case X86::VDIVSSZrm_Int:
case X86::VDIVSSZrm_Intk:
case X86::VDIVSSZrm_Intkz:
case X86::VDIVSSZrr_Int:
case X86::VDIVSSZrr_Intk:
case X86::VDIVSSZrr_Intkz:
case X86::VDIVSSZrrb_Int:
case X86::VDIVSSZrrb_Intk:
case X86::VDIVSSZrrb_Intkz:
case X86::VSQRTPDZ128m:
case X86::VSQRTPDZ128mb:
case X86::VSQRTPDZ128mbk:
case X86::VSQRTPDZ128mbkz:
case X86::VSQRTPDZ128mk:
case X86::VSQRTPDZ128mkz:
case X86::VSQRTPDZ128r:
case X86::VSQRTPDZ128rk:
case X86::VSQRTPDZ128rkz:
case X86::VSQRTPDZ256m:
case X86::VSQRTPDZ256mb:
case X86::VSQRTPDZ256mbk:
case X86::VSQRTPDZ256mbkz:
case X86::VSQRTPDZ256mk:
case X86::VSQRTPDZ256mkz:
case X86::VSQRTPDZ256r:
case X86::VSQRTPDZ256rk:
case X86::VSQRTPDZ256rkz:
case X86::VSQRTPDZm:
case X86::VSQRTPDZmb:
case X86::VSQRTPDZmbk:
case X86::VSQRTPDZmbkz:
case X86::VSQRTPDZmk:
case X86::VSQRTPDZmkz:
case X86::VSQRTPDZr:
case X86::VSQRTPDZrb:
case X86::VSQRTPDZrbk:
case X86::VSQRTPDZrbkz:
case X86::VSQRTPDZrk:
case X86::VSQRTPDZrkz:
case X86::VSQRTPSZ128m:
case X86::VSQRTPSZ128mb:
case X86::VSQRTPSZ128mbk:
case X86::VSQRTPSZ128mbkz:
case X86::VSQRTPSZ128mk:
case X86::VSQRTPSZ128mkz:
case X86::VSQRTPSZ128r:
case X86::VSQRTPSZ128rk:
case X86::VSQRTPSZ128rkz:
case X86::VSQRTPSZ256m:
case X86::VSQRTPSZ256mb:
case X86::VSQRTPSZ256mbk:
case X86::VSQRTPSZ256mbkz:
case X86::VSQRTPSZ256mk:
case X86::VSQRTPSZ256mkz:
case X86::VSQRTPSZ256r:
case X86::VSQRTPSZ256rk:
case X86::VSQRTPSZ256rkz:
case X86::VSQRTPSZm:
case X86::VSQRTPSZmb:
case X86::VSQRTPSZmbk:
case X86::VSQRTPSZmbkz:
case X86::VSQRTPSZmk:
case X86::VSQRTPSZmkz:
case X86::VSQRTPSZr:
case X86::VSQRTPSZrb:
case X86::VSQRTPSZrbk:
case X86::VSQRTPSZrbkz:
case X86::VSQRTPSZrk:
case X86::VSQRTPSZrkz:
case X86::VSQRTSDZm:
case X86::VSQRTSDZm_Int:
case X86::VSQRTSDZm_Intk:
case X86::VSQRTSDZm_Intkz:
case X86::VSQRTSDZr:
case X86::VSQRTSDZr_Int:
case X86::VSQRTSDZr_Intk:
case X86::VSQRTSDZr_Intkz:
case X86::VSQRTSDZrb_Int:
case X86::VSQRTSDZrb_Intk:
case X86::VSQRTSDZrb_Intkz:
case X86::VSQRTSSZm:
case X86::VSQRTSSZm_Int:
case X86::VSQRTSSZm_Intk:
case X86::VSQRTSSZm_Intkz:
case X86::VSQRTSSZr:
case X86::VSQRTSSZr_Int:
case X86::VSQRTSSZr_Intk:
case X86::VSQRTSSZr_Intkz:
case X86::VSQRTSSZrb_Int:
case X86::VSQRTSSZrb_Intk:
case X86::VSQRTSSZrb_Intkz:
case X86::VGATHERDPDYrm:
case X86::VGATHERDPDZ128rm:
case X86::VGATHERDPDZ256rm:
case X86::VGATHERDPDZrm:
case X86::VGATHERDPDrm:
case X86::VGATHERDPSYrm:
case X86::VGATHERDPSZ128rm:
case X86::VGATHERDPSZ256rm:
case X86::VGATHERDPSZrm:
case X86::VGATHERDPSrm:
case X86::VGATHERPF0DPDm:
case X86::VGATHERPF0DPSm:
case X86::VGATHERPF0QPDm:
case X86::VGATHERPF0QPSm:
case X86::VGATHERPF1DPDm:
case X86::VGATHERPF1DPSm:
case X86::VGATHERPF1QPDm:
case X86::VGATHERPF1QPSm:
case X86::VGATHERQPDYrm:
case X86::VGATHERQPDZ128rm:
case X86::VGATHERQPDZ256rm:
case X86::VGATHERQPDZrm:
case X86::VGATHERQPDrm:
case X86::VGATHERQPSYrm:
case X86::VGATHERQPSZ128rm:
case X86::VGATHERQPSZ256rm:
case X86::VGATHERQPSZrm:
case X86::VGATHERQPSrm:
case X86::VPGATHERDDYrm:
case X86::VPGATHERDDZ128rm:
case X86::VPGATHERDDZ256rm:
case X86::VPGATHERDDZrm:
case X86::VPGATHERDDrm:
case X86::VPGATHERDQYrm:
case X86::VPGATHERDQZ128rm:
case X86::VPGATHERDQZ256rm:
case X86::VPGATHERDQZrm:
case X86::VPGATHERDQrm:
case X86::VPGATHERQDYrm:
case X86::VPGATHERQDZ128rm:
case X86::VPGATHERQDZ256rm:
case X86::VPGATHERQDZrm:
case X86::VPGATHERQDrm:
case X86::VPGATHERQQYrm:
case X86::VPGATHERQQZ128rm:
case X86::VPGATHERQQZ256rm:
case X86::VPGATHERQQZrm:
case X86::VPGATHERQQrm:
case X86::VSCATTERDPDZ128mr:
case X86::VSCATTERDPDZ256mr:
case X86::VSCATTERDPDZmr:
case X86::VSCATTERDPSZ128mr:
case X86::VSCATTERDPSZ256mr:
case X86::VSCATTERDPSZmr:
case X86::VSCATTERPF0DPDm:
case X86::VSCATTERPF0DPSm:
case X86::VSCATTERPF0QPDm:
case X86::VSCATTERPF0QPSm:
case X86::VSCATTERPF1DPDm:
case X86::VSCATTERPF1DPSm:
case X86::VSCATTERPF1QPDm:
case X86::VSCATTERPF1QPSm:
case X86::VSCATTERQPDZ128mr:
case X86::VSCATTERQPDZ256mr:
case X86::VSCATTERQPDZmr:
case X86::VSCATTERQPSZ128mr:
case X86::VSCATTERQPSZ256mr:
case X86::VSCATTERQPSZmr:
case X86::VPSCATTERDDZ128mr:
case X86::VPSCATTERDDZ256mr:
case X86::VPSCATTERDDZmr:
case X86::VPSCATTERDQZ128mr:
case X86::VPSCATTERDQZ256mr:
case X86::VPSCATTERDQZmr:
case X86::VPSCATTERQDZ128mr:
case X86::VPSCATTERQDZ256mr:
case X86::VPSCATTERQDZmr:
case X86::VPSCATTERQQZ128mr:
case X86::VPSCATTERQQZ256mr:
case X86::VPSCATTERQQZmr:
return true;
}
}
bool X86InstrInfo::hasHighOperandLatency(const TargetSchedModel &SchedModel,
const MachineRegisterInfo *MRI,
const MachineInstr &DefMI,
unsigned DefIdx,
const MachineInstr &UseMI,
unsigned UseIdx) const {
return isHighLatencyDef(DefMI.getOpcode());
}
bool X86InstrInfo::hasReassociableOperands(const MachineInstr &Inst,
const MachineBasicBlock *MBB) const {
assert(Inst.getNumExplicitOperands() == 3 && Inst.getNumExplicitDefs() == 1 &&
Inst.getNumDefs() <= 2 && "Reassociation needs binary operators");
// Integer binary math/logic instructions have a third source operand:
// the EFLAGS register. That operand must be both defined here and never
// used; ie, it must be dead. If the EFLAGS operand is live, then we can
// not change anything because rearranging the operands could affect other
// instructions that depend on the exact status flags (zero, sign, etc.)
// that are set by using these particular operands with this operation.
const MachineOperand *FlagDef =
Inst.findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
assert((Inst.getNumDefs() == 1 || FlagDef) && "Implicit def isn't flags?");
if (FlagDef && !FlagDef->isDead())
return false;
return TargetInstrInfo::hasReassociableOperands(Inst, MBB);
}
// TODO: There are many more machine instruction opcodes to match:
// 1. Other data types (integer, vectors)
// 2. Other math / logic operations (xor, or)
// 3. Other forms of the same operation (intrinsics and other variants)
bool X86InstrInfo::isAssociativeAndCommutative(const MachineInstr &Inst,
bool Invert) const {
if (Invert)
return false;
switch (Inst.getOpcode()) {
CASE_ND(ADD8rr)
CASE_ND(ADD16rr)
CASE_ND(ADD32rr)
CASE_ND(ADD64rr)
CASE_ND(AND8rr)
CASE_ND(AND16rr)
CASE_ND(AND32rr)
CASE_ND(AND64rr)
CASE_ND(OR8rr)
CASE_ND(OR16rr)
CASE_ND(OR32rr)
CASE_ND(OR64rr)
CASE_ND(XOR8rr)
CASE_ND(XOR16rr)
CASE_ND(XOR32rr)
CASE_ND(XOR64rr)
CASE_ND(IMUL16rr)
CASE_ND(IMUL32rr)
CASE_ND(IMUL64rr)
case X86::PANDrr:
case X86::PORrr:
case X86::PXORrr:
case X86::ANDPDrr:
case X86::ANDPSrr:
case X86::ORPDrr:
case X86::ORPSrr:
case X86::XORPDrr:
case X86::XORPSrr:
case X86::PADDBrr:
case X86::PADDWrr:
case X86::PADDDrr:
case X86::PADDQrr:
case X86::PMULLWrr:
case X86::PMULLDrr:
case X86::PMAXSBrr:
case X86::PMAXSDrr:
case X86::PMAXSWrr:
case X86::PMAXUBrr:
case X86::PMAXUDrr:
case X86::PMAXUWrr:
case X86::PMINSBrr:
case X86::PMINSDrr:
case X86::PMINSWrr:
case X86::PMINUBrr:
case X86::PMINUDrr:
case X86::PMINUWrr:
case X86::VPANDrr:
case X86::VPANDYrr:
case X86::VPANDDZ128rr:
case X86::VPANDDZ256rr:
case X86::VPANDDZrr:
case X86::VPANDQZ128rr:
case X86::VPANDQZ256rr:
case X86::VPANDQZrr:
case X86::VPORrr:
case X86::VPORYrr:
case X86::VPORDZ128rr:
case X86::VPORDZ256rr:
case X86::VPORDZrr:
case X86::VPORQZ128rr:
case X86::VPORQZ256rr:
case X86::VPORQZrr:
case X86::VPXORrr:
case X86::VPXORYrr:
case X86::VPXORDZ128rr:
case X86::VPXORDZ256rr:
case X86::VPXORDZrr:
case X86::VPXORQZ128rr:
case X86::VPXORQZ256rr:
case X86::VPXORQZrr:
case X86::VANDPDrr:
case X86::VANDPSrr:
case X86::VANDPDYrr:
case X86::VANDPSYrr:
case X86::VANDPDZ128rr:
case X86::VANDPSZ128rr:
case X86::VANDPDZ256rr:
case X86::VANDPSZ256rr:
case X86::VANDPDZrr:
case X86::VANDPSZrr:
case X86::VORPDrr:
case X86::VORPSrr:
case X86::VORPDYrr:
case X86::VORPSYrr:
case X86::VORPDZ128rr:
case X86::VORPSZ128rr:
case X86::VORPDZ256rr:
case X86::VORPSZ256rr:
case X86::VORPDZrr:
case X86::VORPSZrr:
case X86::VXORPDrr:
case X86::VXORPSrr:
case X86::VXORPDYrr:
case X86::VXORPSYrr:
case X86::VXORPDZ128rr:
case X86::VXORPSZ128rr:
case X86::VXORPDZ256rr:
case X86::VXORPSZ256rr:
case X86::VXORPDZrr:
case X86::VXORPSZrr:
case X86::KADDBkk:
case X86::KADDWkk:
case X86::KADDDkk:
case X86::KADDQkk:
case X86::KANDBkk:
case X86::KANDWkk:
case X86::KANDDkk:
case X86::KANDQkk:
case X86::KORBkk:
case X86::KORWkk:
case X86::KORDkk:
case X86::KORQkk:
case X86::KXORBkk:
case X86::KXORWkk:
case X86::KXORDkk:
case X86::KXORQkk:
case X86::VPADDBrr:
case X86::VPADDWrr:
case X86::VPADDDrr:
case X86::VPADDQrr:
case X86::VPADDBYrr:
case X86::VPADDWYrr:
case X86::VPADDDYrr:
case X86::VPADDQYrr:
case X86::VPADDBZ128rr:
case X86::VPADDWZ128rr:
case X86::VPADDDZ128rr:
case X86::VPADDQZ128rr:
case X86::VPADDBZ256rr:
case X86::VPADDWZ256rr:
case X86::VPADDDZ256rr:
case X86::VPADDQZ256rr:
case X86::VPADDBZrr:
case X86::VPADDWZrr:
case X86::VPADDDZrr:
case X86::VPADDQZrr:
case X86::VPMULLWrr:
case X86::VPMULLWYrr:
case X86::VPMULLWZ128rr:
case X86::VPMULLWZ256rr:
case X86::VPMULLWZrr:
case X86::VPMULLDrr:
case X86::VPMULLDYrr:
case X86::VPMULLDZ128rr:
case X86::VPMULLDZ256rr:
case X86::VPMULLDZrr:
case X86::VPMULLQZ128rr:
case X86::VPMULLQZ256rr:
case X86::VPMULLQZrr:
case X86::VPMAXSBrr:
case X86::VPMAXSBYrr:
case X86::VPMAXSBZ128rr:
case X86::VPMAXSBZ256rr:
case X86::VPMAXSBZrr:
case X86::VPMAXSDrr:
case X86::VPMAXSDYrr:
case X86::VPMAXSDZ128rr:
case X86::VPMAXSDZ256rr:
case X86::VPMAXSDZrr:
case X86::VPMAXSQZ128rr:
case X86::VPMAXSQZ256rr:
case X86::VPMAXSQZrr:
case X86::VPMAXSWrr:
case X86::VPMAXSWYrr:
case X86::VPMAXSWZ128rr:
case X86::VPMAXSWZ256rr:
case X86::VPMAXSWZrr:
case X86::VPMAXUBrr:
case X86::VPMAXUBYrr:
case X86::VPMAXUBZ128rr:
case X86::VPMAXUBZ256rr:
case X86::VPMAXUBZrr:
case X86::VPMAXUDrr:
case X86::VPMAXUDYrr:
case X86::VPMAXUDZ128rr:
case X86::VPMAXUDZ256rr:
case X86::VPMAXUDZrr:
case X86::VPMAXUQZ128rr:
case X86::VPMAXUQZ256rr:
case X86::VPMAXUQZrr:
case X86::VPMAXUWrr:
case X86::VPMAXUWYrr:
case X86::VPMAXUWZ128rr:
case X86::VPMAXUWZ256rr:
case X86::VPMAXUWZrr:
case X86::VPMINSBrr:
case X86::VPMINSBYrr:
case X86::VPMINSBZ128rr:
case X86::VPMINSBZ256rr:
case X86::VPMINSBZrr:
case X86::VPMINSDrr:
case X86::VPMINSDYrr:
case X86::VPMINSDZ128rr:
case X86::VPMINSDZ256rr:
case X86::VPMINSDZrr:
case X86::VPMINSQZ128rr:
case X86::VPMINSQZ256rr:
case X86::VPMINSQZrr:
case X86::VPMINSWrr:
case X86::VPMINSWYrr:
case X86::VPMINSWZ128rr:
case X86::VPMINSWZ256rr:
case X86::VPMINSWZrr:
case X86::VPMINUBrr:
case X86::VPMINUBYrr:
case X86::VPMINUBZ128rr:
case X86::VPMINUBZ256rr:
case X86::VPMINUBZrr:
case X86::VPMINUDrr:
case X86::VPMINUDYrr:
case X86::VPMINUDZ128rr:
case X86::VPMINUDZ256rr:
case X86::VPMINUDZrr:
case X86::VPMINUQZ128rr:
case X86::VPMINUQZ256rr:
case X86::VPMINUQZrr:
case X86::VPMINUWrr:
case X86::VPMINUWYrr:
case X86::VPMINUWZ128rr:
case X86::VPMINUWZ256rr:
case X86::VPMINUWZrr:
// Normal min/max instructions are not commutative because of NaN and signed
// zero semantics, but these are. Thus, there's no need to check for global
// relaxed math; the instructions themselves have the properties we need.
case X86::MAXCPDrr:
case X86::MAXCPSrr:
case X86::MAXCSDrr:
case X86::MAXCSSrr:
case X86::MINCPDrr:
case X86::MINCPSrr:
case X86::MINCSDrr:
case X86::MINCSSrr:
case X86::VMAXCPDrr:
case X86::VMAXCPSrr:
case X86::VMAXCPDYrr:
case X86::VMAXCPSYrr:
case X86::VMAXCPDZ128rr:
case X86::VMAXCPSZ128rr:
case X86::VMAXCPDZ256rr:
case X86::VMAXCPSZ256rr:
case X86::VMAXCPDZrr:
case X86::VMAXCPSZrr:
case X86::VMAXCSDrr:
case X86::VMAXCSSrr:
case X86::VMAXCSDZrr:
case X86::VMAXCSSZrr:
case X86::VMINCPDrr:
case X86::VMINCPSrr:
case X86::VMINCPDYrr:
case X86::VMINCPSYrr:
case X86::VMINCPDZ128rr:
case X86::VMINCPSZ128rr:
case X86::VMINCPDZ256rr:
case X86::VMINCPSZ256rr:
case X86::VMINCPDZrr:
case X86::VMINCPSZrr:
case X86::VMINCSDrr:
case X86::VMINCSSrr:
case X86::VMINCSDZrr:
case X86::VMINCSSZrr:
case X86::VMAXCPHZ128rr:
case X86::VMAXCPHZ256rr:
case X86::VMAXCPHZrr:
case X86::VMAXCSHZrr:
case X86::VMINCPHZ128rr:
case X86::VMINCPHZ256rr:
case X86::VMINCPHZrr:
case X86::VMINCSHZrr:
return true;
case X86::ADDPDrr:
case X86::ADDPSrr:
case X86::ADDSDrr:
case X86::ADDSSrr:
case X86::MULPDrr:
case X86::MULPSrr:
case X86::MULSDrr:
case X86::MULSSrr:
case X86::VADDPDrr:
case X86::VADDPSrr:
case X86::VADDPDYrr:
case X86::VADDPSYrr:
case X86::VADDPDZ128rr:
case X86::VADDPSZ128rr:
case X86::VADDPDZ256rr:
case X86::VADDPSZ256rr:
case X86::VADDPDZrr:
case X86::VADDPSZrr:
case X86::VADDSDrr:
case X86::VADDSSrr:
case X86::VADDSDZrr:
case X86::VADDSSZrr:
case X86::VMULPDrr:
case X86::VMULPSrr:
case X86::VMULPDYrr:
case X86::VMULPSYrr:
case X86::VMULPDZ128rr:
case X86::VMULPSZ128rr:
case X86::VMULPDZ256rr:
case X86::VMULPSZ256rr:
case X86::VMULPDZrr:
case X86::VMULPSZrr:
case X86::VMULSDrr:
case X86::VMULSSrr:
case X86::VMULSDZrr:
case X86::VMULSSZrr:
case X86::VADDPHZ128rr:
case X86::VADDPHZ256rr:
case X86::VADDPHZrr:
case X86::VADDSHZrr:
case X86::VMULPHZ128rr:
case X86::VMULPHZ256rr:
case X86::VMULPHZrr:
case X86::VMULSHZrr:
return Inst.getFlag(MachineInstr::MIFlag::FmReassoc) &&
Inst.getFlag(MachineInstr::MIFlag::FmNsz);
default:
return false;
}
}
/// If \p DescribedReg overlaps with the MOVrr instruction's destination
/// register then, if possible, describe the value in terms of the source
/// register.
static std::optional<ParamLoadedValue>
describeMOVrrLoadedValue(const MachineInstr &MI, Register DescribedReg,
const TargetRegisterInfo *TRI) {
Register DestReg = MI.getOperand(0).getReg();
Register SrcReg = MI.getOperand(1).getReg();
auto Expr = DIExpression::get(MI.getMF()->getFunction().getContext(), {});
// If the described register is the destination, just return the source.
if (DestReg == DescribedReg)
return ParamLoadedValue(MachineOperand::CreateReg(SrcReg, false), Expr);
// If the described register is a sub-register of the destination register,
// then pick out the source register's corresponding sub-register.
if (unsigned SubRegIdx = TRI->getSubRegIndex(DestReg, DescribedReg)) {
Register SrcSubReg = TRI->getSubReg(SrcReg, SubRegIdx);
return ParamLoadedValue(MachineOperand::CreateReg(SrcSubReg, false), Expr);
}
// The remaining case to consider is when the described register is a
// super-register of the destination register. MOV8rr and MOV16rr does not
// write to any of the other bytes in the register, meaning that we'd have to
// describe the value using a combination of the source register and the
// non-overlapping bits in the described register, which is not currently
// possible.
if (MI.getOpcode() == X86::MOV8rr || MI.getOpcode() == X86::MOV16rr ||
!TRI->isSuperRegister(DestReg, DescribedReg))
return std::nullopt;
assert(MI.getOpcode() == X86::MOV32rr && "Unexpected super-register case");
return ParamLoadedValue(MachineOperand::CreateReg(SrcReg, false), Expr);
}
std::optional<ParamLoadedValue>
X86InstrInfo::describeLoadedValue(const MachineInstr &MI, Register Reg) const {
const MachineOperand *Op = nullptr;
DIExpression *Expr = nullptr;
const TargetRegisterInfo *TRI = &getRegisterInfo();
switch (MI.getOpcode()) {
case X86::LEA32r:
case X86::LEA64r:
case X86::LEA64_32r: {
// We may need to describe a 64-bit parameter with a 32-bit LEA.
if (!TRI->isSuperRegisterEq(MI.getOperand(0).getReg(), Reg))
return std::nullopt;
// Operand 4 could be global address. For now we do not support
// such situation.
if (!MI.getOperand(4).isImm() || !MI.getOperand(2).isImm())
return std::nullopt;
const MachineOperand &Op1 = MI.getOperand(1);
const MachineOperand &Op2 = MI.getOperand(3);
assert(Op2.isReg() &&
(Op2.getReg() == X86::NoRegister || Op2.getReg().isPhysical()));
// Omit situations like:
// %rsi = lea %rsi, 4, ...
if ((Op1.isReg() && Op1.getReg() == MI.getOperand(0).getReg()) ||
Op2.getReg() == MI.getOperand(0).getReg())
return std::nullopt;
else if ((Op1.isReg() && Op1.getReg() != X86::NoRegister &&
TRI->regsOverlap(Op1.getReg(), MI.getOperand(0).getReg())) ||
(Op2.getReg() != X86::NoRegister &&
TRI->regsOverlap(Op2.getReg(), MI.getOperand(0).getReg())))
return std::nullopt;
int64_t Coef = MI.getOperand(2).getImm();
int64_t Offset = MI.getOperand(4).getImm();
SmallVector<uint64_t, 8> Ops;
if ((Op1.isReg() && Op1.getReg() != X86::NoRegister)) {
Op = &Op1;
} else if (Op1.isFI())
Op = &Op1;
if (Op && Op->isReg() && Op->getReg() == Op2.getReg() && Coef > 0) {
Ops.push_back(dwarf::DW_OP_constu);
Ops.push_back(Coef + 1);
Ops.push_back(dwarf::DW_OP_mul);
} else {
if (Op && Op2.getReg() != X86::NoRegister) {
int dwarfReg = TRI->getDwarfRegNum(Op2.getReg(), false);
if (dwarfReg < 0)
return std::nullopt;
else if (dwarfReg < 32) {
Ops.push_back(dwarf::DW_OP_breg0 + dwarfReg);
Ops.push_back(0);
} else {
Ops.push_back(dwarf::DW_OP_bregx);
Ops.push_back(dwarfReg);
Ops.push_back(0);
}
} else if (!Op) {
assert(Op2.getReg() != X86::NoRegister);
Op = &Op2;
}
if (Coef > 1) {
assert(Op2.getReg() != X86::NoRegister);
Ops.push_back(dwarf::DW_OP_constu);
Ops.push_back(Coef);
Ops.push_back(dwarf::DW_OP_mul);
}
if (((Op1.isReg() && Op1.getReg() != X86::NoRegister) || Op1.isFI()) &&
Op2.getReg() != X86::NoRegister) {
Ops.push_back(dwarf::DW_OP_plus);
}
}
DIExpression::appendOffset(Ops, Offset);
Expr = DIExpression::get(MI.getMF()->getFunction().getContext(), Ops);
return ParamLoadedValue(*Op, Expr);
}
case X86::MOV8ri:
case X86::MOV16ri:
// TODO: Handle MOV8ri and MOV16ri.
return std::nullopt;
case X86::MOV32ri:
case X86::MOV64ri:
case X86::MOV64ri32:
// MOV32ri may be used for producing zero-extended 32-bit immediates in
// 64-bit parameters, so we need to consider super-registers.
if (!TRI->isSuperRegisterEq(MI.getOperand(0).getReg(), Reg))
return std::nullopt;
return ParamLoadedValue(MI.getOperand(1), Expr);
case X86::MOV8rr:
case X86::MOV16rr:
case X86::MOV32rr:
case X86::MOV64rr:
return describeMOVrrLoadedValue(MI, Reg, TRI);
case X86::XOR32rr: {
// 64-bit parameters are zero-materialized using XOR32rr, so also consider
// super-registers.
if (!TRI->isSuperRegisterEq(MI.getOperand(0).getReg(), Reg))
return std::nullopt;
if (MI.getOperand(1).getReg() == MI.getOperand(2).getReg())
return ParamLoadedValue(MachineOperand::CreateImm(0), Expr);
return std::nullopt;
}
case X86::MOVSX64rr32: {
// We may need to describe the lower 32 bits of the MOVSX; for example, in
// cases like this:
//
// $ebx = [...]
// $rdi = MOVSX64rr32 $ebx
// $esi = MOV32rr $edi
if (!TRI->isSubRegisterEq(MI.getOperand(0).getReg(), Reg))
return std::nullopt;
Expr = DIExpression::get(MI.getMF()->getFunction().getContext(), {});
// If the described register is the destination register we need to
// sign-extend the source register from 32 bits. The other case we handle
// is when the described register is the 32-bit sub-register of the
// destination register, in case we just need to return the source
// register.
if (Reg == MI.getOperand(0).getReg())
Expr = DIExpression::appendExt(Expr, 32, 64, true);
else
assert(X86MCRegisterClasses[X86::GR32RegClassID].contains(Reg) &&
"Unhandled sub-register case for MOVSX64rr32");
return ParamLoadedValue(MI.getOperand(1), Expr);
}
default:
assert(!MI.isMoveImmediate() && "Unexpected MoveImm instruction");
return TargetInstrInfo::describeLoadedValue(MI, Reg);
}
}
/// This is an architecture-specific helper function of reassociateOps.
/// Set special operand attributes for new instructions after reassociation.
void X86InstrInfo::setSpecialOperandAttr(MachineInstr &OldMI1,
MachineInstr &OldMI2,
MachineInstr &NewMI1,
MachineInstr &NewMI2) const {
// Integer instructions may define an implicit EFLAGS dest register operand.
MachineOperand *OldFlagDef1 =
OldMI1.findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
MachineOperand *OldFlagDef2 =
OldMI2.findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
assert(!OldFlagDef1 == !OldFlagDef2 &&
"Unexpected instruction type for reassociation");
if (!OldFlagDef1 || !OldFlagDef2)
return;
assert(OldFlagDef1->isDead() && OldFlagDef2->isDead() &&
"Must have dead EFLAGS operand in reassociable instruction");
MachineOperand *NewFlagDef1 =
NewMI1.findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
MachineOperand *NewFlagDef2 =
NewMI2.findRegisterDefOperand(X86::EFLAGS, /*TRI=*/nullptr);
assert(NewFlagDef1 && NewFlagDef2 &&
"Unexpected operand in reassociable instruction");
// Mark the new EFLAGS operands as dead to be helpful to subsequent iterations
// of this pass or other passes. The EFLAGS operands must be dead in these new
// instructions because the EFLAGS operands in the original instructions must
// be dead in order for reassociation to occur.
NewFlagDef1->setIsDead();
NewFlagDef2->setIsDead();
}
std::pair<unsigned, unsigned>
X86InstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const {
return std::make_pair(TF, 0u);
}
ArrayRef<std::pair<unsigned, const char *>>
X86InstrInfo::getSerializableDirectMachineOperandTargetFlags() const {
using namespace X86II;
static const std::pair<unsigned, const char *> TargetFlags[] = {
{MO_GOT_ABSOLUTE_ADDRESS, "x86-got-absolute-address"},
{MO_PIC_BASE_OFFSET, "x86-pic-base-offset"},
{MO_GOT, "x86-got"},
{MO_GOTOFF, "x86-gotoff"},
{MO_GOTPCREL, "x86-gotpcrel"},
{MO_GOTPCREL_NORELAX, "x86-gotpcrel-norelax"},
{MO_PLT, "x86-plt"},
{MO_TLSGD, "x86-tlsgd"},
{MO_TLSLD, "x86-tlsld"},
{MO_TLSLDM, "x86-tlsldm"},
{MO_GOTTPOFF, "x86-gottpoff"},
{MO_INDNTPOFF, "x86-indntpoff"},
{MO_TPOFF, "x86-tpoff"},
{MO_DTPOFF, "x86-dtpoff"},
{MO_NTPOFF, "x86-ntpoff"},
{MO_GOTNTPOFF, "x86-gotntpoff"},
{MO_DLLIMPORT, "x86-dllimport"},
{MO_DARWIN_NONLAZY, "x86-darwin-nonlazy"},
{MO_DARWIN_NONLAZY_PIC_BASE, "x86-darwin-nonlazy-pic-base"},
{MO_TLVP, "x86-tlvp"},
{MO_TLVP_PIC_BASE, "x86-tlvp-pic-base"},
{MO_SECREL, "x86-secrel"},
{MO_COFFSTUB, "x86-coffstub"}};
return ArrayRef(TargetFlags);
}
namespace {
/// Create Global Base Reg pass. This initializes the PIC
/// global base register for x86-32.
struct CGBR : public MachineFunctionPass {
static char ID;
CGBR() : MachineFunctionPass(ID) {}
bool runOnMachineFunction(MachineFunction &MF) override {
const X86TargetMachine *TM =
static_cast<const X86TargetMachine *>(&MF.getTarget());
const X86Subtarget &STI = MF.getSubtarget<X86Subtarget>();
// Only emit a global base reg in PIC mode.
if (!TM->isPositionIndependent())
return false;
X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
Register GlobalBaseReg = X86FI->getGlobalBaseReg();
// If we didn't need a GlobalBaseReg, don't insert code.
if (GlobalBaseReg == 0)
return false;
// Insert the set of GlobalBaseReg into the first MBB of the function
MachineBasicBlock &FirstMBB = MF.front();
MachineBasicBlock::iterator MBBI = FirstMBB.begin();
DebugLoc DL = FirstMBB.findDebugLoc(MBBI);
MachineRegisterInfo &RegInfo = MF.getRegInfo();
const X86InstrInfo *TII = STI.getInstrInfo();
Register PC;
if (STI.isPICStyleGOT())
PC = RegInfo.createVirtualRegister(&X86::GR32RegClass);
else
PC = GlobalBaseReg;
if (STI.is64Bit()) {
if (TM->getCodeModel() == CodeModel::Large) {
// In the large code model, we are aiming for this code, though the
// register allocation may vary:
// leaq .LN$pb(%rip), %rax
// movq $_GLOBAL_OFFSET_TABLE_ - .LN$pb, %rcx
// addq %rcx, %rax
// RAX now holds address of _GLOBAL_OFFSET_TABLE_.
Register PBReg = RegInfo.createVirtualRegister(&X86::GR64RegClass);
Register GOTReg = RegInfo.createVirtualRegister(&X86::GR64RegClass);
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::LEA64r), PBReg)
.addReg(X86::RIP)
.addImm(0)
.addReg(0)
.addSym(MF.getPICBaseSymbol())
.addReg(0);
std::prev(MBBI)->setPreInstrSymbol(MF, MF.getPICBaseSymbol());
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOV64ri), GOTReg)
.addExternalSymbol("_GLOBAL_OFFSET_TABLE_",
X86II::MO_PIC_BASE_OFFSET);
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::ADD64rr), PC)
.addReg(PBReg, RegState::Kill)
.addReg(GOTReg, RegState::Kill);
} else {
// In other code models, use a RIP-relative LEA to materialize the
// GOT.
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::LEA64r), PC)
.addReg(X86::RIP)
.addImm(0)
.addReg(0)
.addExternalSymbol("_GLOBAL_OFFSET_TABLE_")
.addReg(0);
}
} else {
// Operand of MovePCtoStack is completely ignored by asm printer. It's
// only used in JIT code emission as displacement to pc.
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOVPC32r), PC).addImm(0);
// If we're using vanilla 'GOT' PIC style, we should use relative
// addressing not to pc, but to _GLOBAL_OFFSET_TABLE_ external.
if (STI.isPICStyleGOT()) {
// Generate addl $__GLOBAL_OFFSET_TABLE_ + [.-piclabel],
// %some_register
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::ADD32ri), GlobalBaseReg)
.addReg(PC)
.addExternalSymbol("_GLOBAL_OFFSET_TABLE_",
X86II::MO_GOT_ABSOLUTE_ADDRESS);
}
}
return true;
}
StringRef getPassName() const override {
return "X86 PIC Global Base Reg Initialization";
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
}
};
} // namespace
char CGBR::ID = 0;
FunctionPass *llvm::createX86GlobalBaseRegPass() { return new CGBR(); }
namespace {
struct LDTLSCleanup : public MachineFunctionPass {
static char ID;
LDTLSCleanup() : MachineFunctionPass(ID) {}
bool runOnMachineFunction(MachineFunction &MF) override {
if (skipFunction(MF.getFunction()))
return false;
X86MachineFunctionInfo *MFI = MF.getInfo<X86MachineFunctionInfo>();
if (MFI->getNumLocalDynamicTLSAccesses() < 2) {
// No point folding accesses if there isn't at least two.
return false;
}
MachineDominatorTree *DT =
&getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
return VisitNode(DT->getRootNode(), 0);
}
// Visit the dominator subtree rooted at Node in pre-order.
// If TLSBaseAddrReg is non-null, then use that to replace any
// TLS_base_addr instructions. Otherwise, create the register
// when the first such instruction is seen, and then use it
// as we encounter more instructions.
bool VisitNode(MachineDomTreeNode *Node, unsigned TLSBaseAddrReg) {
MachineBasicBlock *BB = Node->getBlock();
bool Changed = false;
// Traverse the current block.
for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;
++I) {
switch (I->getOpcode()) {
case X86::TLS_base_addr32:
case X86::TLS_base_addr64:
if (TLSBaseAddrReg)
I = ReplaceTLSBaseAddrCall(*I, TLSBaseAddrReg);
else
I = SetRegister(*I, &TLSBaseAddrReg);
Changed = true;
break;
default:
break;
}
}
// Visit the children of this block in the dominator tree.
for (auto &I : *Node) {
Changed |= VisitNode(I, TLSBaseAddrReg);
}
return Changed;
}
// Replace the TLS_base_addr instruction I with a copy from
// TLSBaseAddrReg, returning the new instruction.
MachineInstr *ReplaceTLSBaseAddrCall(MachineInstr &I,
unsigned TLSBaseAddrReg) {
MachineFunction *MF = I.getParent()->getParent();
const X86Subtarget &STI = MF->getSubtarget<X86Subtarget>();
const bool is64Bit = STI.is64Bit();
const X86InstrInfo *TII = STI.getInstrInfo();
// Insert a Copy from TLSBaseAddrReg to RAX/EAX.
MachineInstr *Copy =
BuildMI(*I.getParent(), I, I.getDebugLoc(),
TII->get(TargetOpcode::COPY), is64Bit ? X86::RAX : X86::EAX)
.addReg(TLSBaseAddrReg);
// Erase the TLS_base_addr instruction.
I.eraseFromParent();
return Copy;
}
// Create a virtual register in *TLSBaseAddrReg, and populate it by
// inserting a copy instruction after I. Returns the new instruction.
MachineInstr *SetRegister(MachineInstr &I, unsigned *TLSBaseAddrReg) {
MachineFunction *MF = I.getParent()->getParent();
const X86Subtarget &STI = MF->getSubtarget<X86Subtarget>();
const bool is64Bit = STI.is64Bit();
const X86InstrInfo *TII = STI.getInstrInfo();
// Create a virtual register for the TLS base address.
MachineRegisterInfo &RegInfo = MF->getRegInfo();
*TLSBaseAddrReg = RegInfo.createVirtualRegister(
is64Bit ? &X86::GR64RegClass : &X86::GR32RegClass);
// Insert a copy from RAX/EAX to TLSBaseAddrReg.
MachineInstr *Next = I.getNextNode();
MachineInstr *Copy = BuildMI(*I.getParent(), Next, I.getDebugLoc(),
TII->get(TargetOpcode::COPY), *TLSBaseAddrReg)
.addReg(is64Bit ? X86::RAX : X86::EAX);
return Copy;
}
StringRef getPassName() const override {
return "Local Dynamic TLS Access Clean-up";
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
AU.addRequired<MachineDominatorTreeWrapperPass>();
MachineFunctionPass::getAnalysisUsage(AU);
}
};
} // namespace
char LDTLSCleanup::ID = 0;
FunctionPass *llvm::createCleanupLocalDynamicTLSPass() {
return new LDTLSCleanup();
}
/// Constants defining how certain sequences should be outlined.
///
/// \p MachineOutlinerDefault implies that the function is called with a call
/// instruction, and a return must be emitted for the outlined function frame.
///
/// That is,
///
/// I1 OUTLINED_FUNCTION:
/// I2 --> call OUTLINED_FUNCTION I1
/// I3 I2
/// I3
/// ret
///
/// * Call construction overhead: 1 (call instruction)
/// * Frame construction overhead: 1 (return instruction)
///
/// \p MachineOutlinerTailCall implies that the function is being tail called.
/// A jump is emitted instead of a call, and the return is already present in
/// the outlined sequence. That is,
///
/// I1 OUTLINED_FUNCTION:
/// I2 --> jmp OUTLINED_FUNCTION I1
/// ret I2
/// ret
///
/// * Call construction overhead: 1 (jump instruction)
/// * Frame construction overhead: 0 (don't need to return)
///
enum MachineOutlinerClass { MachineOutlinerDefault, MachineOutlinerTailCall };
std::optional<std::unique_ptr<outliner::OutlinedFunction>>
X86InstrInfo::getOutliningCandidateInfo(
const MachineModuleInfo &MMI,
std::vector<outliner::Candidate> &RepeatedSequenceLocs,
unsigned MinRepeats) const {
unsigned SequenceSize = 0;
for (auto &MI : RepeatedSequenceLocs[0]) {
// FIXME: x86 doesn't implement getInstSizeInBytes, so
// we can't tell the cost. Just assume each instruction
// is one byte.
if (MI.isDebugInstr() || MI.isKill())
continue;
SequenceSize += 1;
}
// We check to see if CFI Instructions are present, and if they are
// we find the number of CFI Instructions in the candidates.
unsigned CFICount = 0;
for (auto &I : RepeatedSequenceLocs[0]) {
if (I.isCFIInstruction())
CFICount++;
}
// We compare the number of found CFI Instructions to the number of CFI
// instructions in the parent function for each candidate. We must check this
// since if we outline one of the CFI instructions in a function, we have to
// outline them all for correctness. If we do not, the address offsets will be
// incorrect between the two sections of the program.
for (outliner::Candidate &C : RepeatedSequenceLocs) {
std::vector<MCCFIInstruction> CFIInstructions =
C.getMF()->getFrameInstructions();
if (CFICount > 0 && CFICount != CFIInstructions.size())
return std::nullopt;
}
// FIXME: Use real size in bytes for call and ret instructions.
if (RepeatedSequenceLocs[0].back().isTerminator()) {
for (outliner::Candidate &C : RepeatedSequenceLocs)
C.setCallInfo(MachineOutlinerTailCall, 1);
return std::make_unique<outliner::OutlinedFunction>(
RepeatedSequenceLocs, SequenceSize,
0, // Number of bytes to emit frame.
MachineOutlinerTailCall // Type of frame.
);
}
if (CFICount > 0)
return std::nullopt;
for (outliner::Candidate &C : RepeatedSequenceLocs)
C.setCallInfo(MachineOutlinerDefault, 1);
return std::make_unique<outliner::OutlinedFunction>(
RepeatedSequenceLocs, SequenceSize, 1, MachineOutlinerDefault);
}
bool X86InstrInfo::isFunctionSafeToOutlineFrom(
MachineFunction &MF, bool OutlineFromLinkOnceODRs) const {
const Function &F = MF.getFunction();
// Does the function use a red zone? If it does, then we can't risk messing
// with the stack.
if (Subtarget.getFrameLowering()->has128ByteRedZone(MF)) {
// It could have a red zone. If it does, then we don't want to touch it.
const X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
if (!X86FI || X86FI->getUsesRedZone())
return false;
}
// If we *don't* want to outline from things that could potentially be deduped
// then return false.
if (!OutlineFromLinkOnceODRs && F.hasLinkOnceODRLinkage())
return false;
// This function is viable for outlining, so return true.
return true;
}
outliner::InstrType
X86InstrInfo::getOutliningTypeImpl(const MachineModuleInfo &MMI,
MachineBasicBlock::iterator &MIT,
unsigned Flags) const {
MachineInstr &MI = *MIT;
// Is this a terminator for a basic block?
if (MI.isTerminator())
// TargetInstrInfo::getOutliningType has already filtered out anything
// that would break this, so we can allow it here.
return outliner::InstrType::Legal;
// Don't outline anything that modifies or reads from the stack pointer.
//
// FIXME: There are instructions which are being manually built without
// explicit uses/defs so we also have to check the MCInstrDesc. We should be
// able to remove the extra checks once those are fixed up. For example,
// sometimes we might get something like %rax = POP64r 1. This won't be
// caught by modifiesRegister or readsRegister even though the instruction
// really ought to be formed so that modifiesRegister/readsRegister would
// catch it.
if (MI.modifiesRegister(X86::RSP, &RI) || MI.readsRegister(X86::RSP, &RI) ||
MI.getDesc().hasImplicitUseOfPhysReg(X86::RSP) ||
MI.getDesc().hasImplicitDefOfPhysReg(X86::RSP))
return outliner::InstrType::Illegal;
// Outlined calls change the instruction pointer, so don't read from it.
if (MI.readsRegister(X86::RIP, &RI) ||
MI.getDesc().hasImplicitUseOfPhysReg(X86::RIP) ||
MI.getDesc().hasImplicitDefOfPhysReg(X86::RIP))
return outliner::InstrType::Illegal;
// Don't outline CFI instructions.
if (MI.isCFIInstruction())
return outliner::InstrType::Illegal;
return outliner::InstrType::Legal;
}
void X86InstrInfo::buildOutlinedFrame(
MachineBasicBlock &MBB, MachineFunction &MF,
const outliner::OutlinedFunction &OF) const {
// If we're a tail call, we already have a return, so don't do anything.
if (OF.FrameConstructionID == MachineOutlinerTailCall)
return;
// We're a normal call, so our sequence doesn't have a return instruction.
// Add it in.
MachineInstr *retq = BuildMI(MF, DebugLoc(), get(X86::RET64));
MBB.insert(MBB.end(), retq);
}
MachineBasicBlock::iterator X86InstrInfo::insertOutlinedCall(
Module &M, MachineBasicBlock &MBB, MachineBasicBlock::iterator &It,
MachineFunction &MF, outliner::Candidate &C) const {
// Is it a tail call?
if (C.CallConstructionID == MachineOutlinerTailCall) {
// Yes, just insert a JMP.
It = MBB.insert(It, BuildMI(MF, DebugLoc(), get(X86::TAILJMPd64))
.addGlobalAddress(M.getNamedValue(MF.getName())));
} else {
// No, insert a call.
It = MBB.insert(It, BuildMI(MF, DebugLoc(), get(X86::CALL64pcrel32))
.addGlobalAddress(M.getNamedValue(MF.getName())));
}
return It;
}
void X86InstrInfo::buildClearRegister(Register Reg, MachineBasicBlock &MBB,
MachineBasicBlock::iterator Iter,
DebugLoc &DL,
bool AllowSideEffects) const {
const MachineFunction &MF = *MBB.getParent();
const X86Subtarget &ST = MF.getSubtarget<X86Subtarget>();
const TargetRegisterInfo &TRI = getRegisterInfo();
if (ST.hasMMX() && X86::VR64RegClass.contains(Reg))
// FIXME: Should we ignore MMX registers?
return;
if (TRI.isGeneralPurposeRegister(MF, Reg)) {
// Convert register to the 32-bit version. Both 'movl' and 'xorl' clear the
// upper bits of a 64-bit register automagically.
Reg = getX86SubSuperRegister(Reg, 32);
if (!AllowSideEffects)
// XOR affects flags, so use a MOV instead.
BuildMI(MBB, Iter, DL, get(X86::MOV32ri), Reg).addImm(0);
else
BuildMI(MBB, Iter, DL, get(X86::XOR32rr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
} else if (X86::VR128RegClass.contains(Reg)) {
// XMM#
if (!ST.hasSSE1())
return;
// PXOR is safe to use because it doesn't affect flags.
BuildMI(MBB, Iter, DL, get(X86::PXORrr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
} else if (X86::VR256RegClass.contains(Reg)) {
// YMM#
if (!ST.hasAVX())
return;
// VPXOR is safe to use because it doesn't affect flags.
BuildMI(MBB, Iter, DL, get(X86::VPXORrr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
} else if (X86::VR512RegClass.contains(Reg)) {
// ZMM#
if (!ST.hasAVX512())
return;
// VPXORY is safe to use because it doesn't affect flags.
BuildMI(MBB, Iter, DL, get(X86::VPXORYrr), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
} else if (X86::VK1RegClass.contains(Reg) || X86::VK2RegClass.contains(Reg) ||
X86::VK4RegClass.contains(Reg) || X86::VK8RegClass.contains(Reg) ||
X86::VK16RegClass.contains(Reg)) {
if (!ST.hasVLX())
return;
// KXOR is safe to use because it doesn't affect flags.
unsigned Op = ST.hasBWI() ? X86::KXORQkk : X86::KXORWkk;
BuildMI(MBB, Iter, DL, get(Op), Reg)
.addReg(Reg, RegState::Undef)
.addReg(Reg, RegState::Undef);
}
}
bool X86InstrInfo::getMachineCombinerPatterns(
MachineInstr &Root, SmallVectorImpl<unsigned> &Patterns,
bool DoRegPressureReduce) const {
unsigned Opc = Root.getOpcode();
switch (Opc) {
case X86::VPDPWSSDrr:
case X86::VPDPWSSDrm:
case X86::VPDPWSSDYrr:
case X86::VPDPWSSDYrm: {
if (!Subtarget.hasFastDPWSSD()) {
Patterns.push_back(X86MachineCombinerPattern::DPWSSD);
return true;
}
break;
}
case X86::VPDPWSSDZ128r:
case X86::VPDPWSSDZ128m:
case X86::VPDPWSSDZ256r:
case X86::VPDPWSSDZ256m:
case X86::VPDPWSSDZr:
case X86::VPDPWSSDZm: {
if (Subtarget.hasBWI() && !Subtarget.hasFastDPWSSD()) {
Patterns.push_back(X86MachineCombinerPattern::DPWSSD);
return true;
}
break;
}
}
return TargetInstrInfo::getMachineCombinerPatterns(Root,
Patterns, DoRegPressureReduce);
}
static void
genAlternativeDpCodeSequence(MachineInstr &Root, const TargetInstrInfo &TII,
SmallVectorImpl<MachineInstr *> &InsInstrs,
SmallVectorImpl<MachineInstr *> &DelInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) {
MachineFunction *MF = Root.getMF();
MachineRegisterInfo &RegInfo = MF->getRegInfo();
unsigned Opc = Root.getOpcode();
unsigned AddOpc = 0;
unsigned MaddOpc = 0;
switch (Opc) {
default:
assert(false && "It should not reach here");
break;
// vpdpwssd xmm2,xmm3,xmm1
// -->
// vpmaddwd xmm3,xmm3,xmm1
// vpaddd xmm2,xmm2,xmm3
case X86::VPDPWSSDrr:
MaddOpc = X86::VPMADDWDrr;
AddOpc = X86::VPADDDrr;
break;
case X86::VPDPWSSDrm:
MaddOpc = X86::VPMADDWDrm;
AddOpc = X86::VPADDDrr;
break;
case X86::VPDPWSSDZ128r:
MaddOpc = X86::VPMADDWDZ128rr;
AddOpc = X86::VPADDDZ128rr;
break;
case X86::VPDPWSSDZ128m:
MaddOpc = X86::VPMADDWDZ128rm;
AddOpc = X86::VPADDDZ128rr;
break;
// vpdpwssd ymm2,ymm3,ymm1
// -->
// vpmaddwd ymm3,ymm3,ymm1
// vpaddd ymm2,ymm2,ymm3
case X86::VPDPWSSDYrr:
MaddOpc = X86::VPMADDWDYrr;
AddOpc = X86::VPADDDYrr;
break;
case X86::VPDPWSSDYrm:
MaddOpc = X86::VPMADDWDYrm;
AddOpc = X86::VPADDDYrr;
break;
case X86::VPDPWSSDZ256r:
MaddOpc = X86::VPMADDWDZ256rr;
AddOpc = X86::VPADDDZ256rr;
break;
case X86::VPDPWSSDZ256m:
MaddOpc = X86::VPMADDWDZ256rm;
AddOpc = X86::VPADDDZ256rr;
break;
// vpdpwssd zmm2,zmm3,zmm1
// -->
// vpmaddwd zmm3,zmm3,zmm1
// vpaddd zmm2,zmm2,zmm3
case X86::VPDPWSSDZr:
MaddOpc = X86::VPMADDWDZrr;
AddOpc = X86::VPADDDZrr;
break;
case X86::VPDPWSSDZm:
MaddOpc = X86::VPMADDWDZrm;
AddOpc = X86::VPADDDZrr;
break;
}
// Create vpmaddwd.
const TargetRegisterClass *RC =
RegInfo.getRegClass(Root.getOperand(0).getReg());
Register NewReg = RegInfo.createVirtualRegister(RC);
MachineInstr *Madd = Root.getMF()->CloneMachineInstr(&Root);
Madd->setDesc(TII.get(MaddOpc));
Madd->untieRegOperand(1);
Madd->removeOperand(1);
Madd->getOperand(0).setReg(NewReg);
InstrIdxForVirtReg.insert(std::make_pair(NewReg, 0));
// Create vpaddd.
Register DstReg = Root.getOperand(0).getReg();
bool IsKill = Root.getOperand(1).isKill();
MachineInstr *Add =
BuildMI(*MF, MIMetadata(Root), TII.get(AddOpc), DstReg)
.addReg(Root.getOperand(1).getReg(), getKillRegState(IsKill))
.addReg(Madd->getOperand(0).getReg(), getKillRegState(true));
InsInstrs.push_back(Madd);
InsInstrs.push_back(Add);
DelInstrs.push_back(&Root);
}
void X86InstrInfo::genAlternativeCodeSequence(
MachineInstr &Root, unsigned Pattern,
SmallVectorImpl<MachineInstr *> &InsInstrs,
SmallVectorImpl<MachineInstr *> &DelInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) const {
switch (Pattern) {
default:
// Reassociate instructions.
TargetInstrInfo::genAlternativeCodeSequence(Root, Pattern, InsInstrs,
DelInstrs, InstrIdxForVirtReg);
return;
case X86MachineCombinerPattern::DPWSSD:
genAlternativeDpCodeSequence(Root, *this, InsInstrs, DelInstrs,
InstrIdxForVirtReg);
return;
}
}
// See also: X86DAGToDAGISel::SelectInlineAsmMemoryOperand().
void X86InstrInfo::getFrameIndexOperands(SmallVectorImpl<MachineOperand> &Ops,
int FI) const {
X86AddressMode M;
M.BaseType = X86AddressMode::FrameIndexBase;
M.Base.FrameIndex = FI;
M.getFullAddress(Ops);
}
#define GET_INSTRINFO_HELPERS
#include "X86GenInstrInfo.inc"
Reference in New Issue View Git Blame Copy Permalink