llvm/llvm/lib/Target/AArch64/AArch64StorePairSuppress.cpp

//===--- AArch64StorePairSuppress.cpp --- Suppress store pair formation ---===//
//
// 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 pass identifies floating point stores that should not be combined into
// store pairs. Later we may do the same for floating point loads.
// ===---------------------------------------------------------------------===//

#include "AArch64InstrInfo.h"
#include "AArch64Subtarget.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineTraceMetrics.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetSchedule.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"

usingnamespacellvm;

#define DEBUG_TYPE …

#define STPSUPPRESS_PASS_NAME …

namespace {
class AArch64StorePairSuppress : public MachineFunctionPass { … };
char AArch64StorePairSuppress::ID = …;
} // anonymous

INITIALIZE_PASS(…)

FunctionPass *llvm::createAArch64StorePairSuppressPass() { … }

/// Return true if an STP can be added to this block without increasing the
/// critical resource height. STP is good to form in Ld/St limited blocks and
/// bad to form in float-point limited blocks. This is true independent of the
/// critical path. If the critical path is longer than the resource height, the
/// extra vector ops can limit physreg renaming. Otherwise, it could simply
/// oversaturate the vector units.
bool AArch64StorePairSuppress::shouldAddSTPToBlock(const MachineBasicBlock *BB) { … }

/// Return true if this is a floating-point store smaller than the V reg. On
/// cyclone, these require a vector shuffle before storing a pair.
/// Ideally we would call getMatchingPairOpcode() and have the machine model
/// tell us if it's profitable with no cpu knowledge here.
///
/// FIXME: We plan to develop a decent Target abstraction for simple loads and
/// stores. Until then use a nasty switch similar to AArch64LoadStoreOptimizer.
bool AArch64StorePairSuppress::isNarrowFPStore(const MachineInstr &MI) { … }

bool AArch64StorePairSuppress::runOnMachineFunction(MachineFunction &MF) { … }