1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-26 06:22:56 +02:00
llvm-mirror/include/llvm/CodeGen/LiveRangeEdit.h
Chandler Carruth d7003090ac [PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.

This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:

- FunctionAAResults is a type-erasing alias analysis results aggregation
  interface to walk a single query across a range of results from
  different alias analyses. Currently this is function-specific as we
  always assume that aliasing queries are *within* a function.

- AAResultBase is a CRTP utility providing stub implementations of
  various parts of the alias analysis result concept, notably in several
  cases in terms of other more general parts of the interface. This can
  be used to implement only a narrow part of the interface rather than
  the entire interface. This isn't really ideal, this logic should be
  hoisted into FunctionAAResults as currently it will cause
  a significant amount of redundant work, but it faithfully models the
  behavior of the prior infrastructure.

- All the alias analysis passes are ported to be wrapper passes for the
  legacy PM and new-style analysis passes for the new PM with a shared
  result object. In some cases (most notably CFL), this is an extremely
  naive approach that we should revisit when we can specialize for the
  new pass manager.

- BasicAA has been restructured to reflect that it is much more
  fundamentally a function analysis because it uses dominator trees and
  loop info that need to be constructed for each function.

All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.

The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.

This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.

Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.

One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.

Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.

Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.

Differential Revision: http://reviews.llvm.org/D12080

llvm-svn: 247167
2015-09-09 17:55:00 +00:00

234 lines
9.1 KiB
C++

//===---- LiveRangeEdit.h - Basic tools for split and spill -----*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// The LiveRangeEdit class represents changes done to a virtual register when it
// is spilled or split.
//
// The parent register is never changed. Instead, a number of new virtual
// registers are created and added to the newRegs vector.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_CODEGEN_LIVERANGEEDIT_H
#define LLVM_CODEGEN_LIVERANGEEDIT_H
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveInterval.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetSubtargetInfo.h"
namespace llvm {
class LiveIntervals;
class MachineBlockFrequencyInfo;
class MachineLoopInfo;
class VirtRegMap;
class LiveRangeEdit : private MachineRegisterInfo::Delegate {
public:
/// Callback methods for LiveRangeEdit owners.
class Delegate {
virtual void anchor();
public:
/// Called immediately before erasing a dead machine instruction.
virtual void LRE_WillEraseInstruction(MachineInstr *MI) {}
/// Called when a virtual register is no longer used. Return false to defer
/// its deletion from LiveIntervals.
virtual bool LRE_CanEraseVirtReg(unsigned) { return true; }
/// Called before shrinking the live range of a virtual register.
virtual void LRE_WillShrinkVirtReg(unsigned) {}
/// Called after cloning a virtual register.
/// This is used for new registers representing connected components of Old.
virtual void LRE_DidCloneVirtReg(unsigned New, unsigned Old) {}
virtual ~Delegate() {}
};
private:
LiveInterval *Parent;
SmallVectorImpl<unsigned> &NewRegs;
MachineRegisterInfo &MRI;
LiveIntervals &LIS;
VirtRegMap *VRM;
const TargetInstrInfo &TII;
Delegate *const TheDelegate;
/// FirstNew - Index of the first register added to NewRegs.
const unsigned FirstNew;
/// ScannedRemattable - true when remattable values have been identified.
bool ScannedRemattable;
/// Remattable - Values defined by remattable instructions as identified by
/// tii.isTriviallyReMaterializable().
SmallPtrSet<const VNInfo*,4> Remattable;
/// Rematted - Values that were actually rematted, and so need to have their
/// live range trimmed or entirely removed.
SmallPtrSet<const VNInfo*,4> Rematted;
/// scanRemattable - Identify the Parent values that may rematerialize.
void scanRemattable(AliasAnalysis *aa);
/// allUsesAvailableAt - Return true if all registers used by OrigMI at
/// OrigIdx are also available with the same value at UseIdx.
bool allUsesAvailableAt(const MachineInstr *OrigMI, SlotIndex OrigIdx,
SlotIndex UseIdx) const;
/// foldAsLoad - If LI has a single use and a single def that can be folded as
/// a load, eliminate the register by folding the def into the use.
bool foldAsLoad(LiveInterval *LI, SmallVectorImpl<MachineInstr*> &Dead);
typedef SetVector<LiveInterval*,
SmallVector<LiveInterval*, 8>,
SmallPtrSet<LiveInterval*, 8> > ToShrinkSet;
/// Helper for eliminateDeadDefs.
void eliminateDeadDef(MachineInstr *MI, ToShrinkSet &ToShrink);
/// MachineRegisterInfo callback to notify when new virtual
/// registers are created.
void MRI_NoteNewVirtualRegister(unsigned VReg) override;
/// \brief Check if MachineOperand \p MO is a last use/kill either in the
/// main live range of \p LI or in one of the matching subregister ranges.
bool useIsKill(const LiveInterval &LI, const MachineOperand &MO) const;
public:
/// Create a LiveRangeEdit for breaking down parent into smaller pieces.
/// @param parent The register being spilled or split.
/// @param newRegs List to receive any new registers created. This needn't be
/// empty initially, any existing registers are ignored.
/// @param MF The MachineFunction the live range edit is taking place in.
/// @param lis The collection of all live intervals in this function.
/// @param vrm Map of virtual registers to physical registers for this
/// function. If NULL, no virtual register map updates will
/// be done. This could be the case if called before Regalloc.
LiveRangeEdit(LiveInterval *parent, SmallVectorImpl<unsigned> &newRegs,
MachineFunction &MF, LiveIntervals &lis, VirtRegMap *vrm,
Delegate *delegate = nullptr)
: Parent(parent), NewRegs(newRegs), MRI(MF.getRegInfo()), LIS(lis),
VRM(vrm), TII(*MF.getSubtarget().getInstrInfo()),
TheDelegate(delegate), FirstNew(newRegs.size()),
ScannedRemattable(false) {
MRI.setDelegate(this);
}
~LiveRangeEdit() override { MRI.resetDelegate(this); }
LiveInterval &getParent() const {
assert(Parent && "No parent LiveInterval");
return *Parent;
}
unsigned getReg() const { return getParent().reg; }
/// Iterator for accessing the new registers added by this edit.
typedef SmallVectorImpl<unsigned>::const_iterator iterator;
iterator begin() const { return NewRegs.begin()+FirstNew; }
iterator end() const { return NewRegs.end(); }
unsigned size() const { return NewRegs.size()-FirstNew; }
bool empty() const { return size() == 0; }
unsigned get(unsigned idx) const { return NewRegs[idx+FirstNew]; }
ArrayRef<unsigned> regs() const {
return makeArrayRef(NewRegs).slice(FirstNew);
}
/// createEmptyIntervalFrom - Create a new empty interval based on OldReg.
LiveInterval &createEmptyIntervalFrom(unsigned OldReg);
/// createFrom - Create a new virtual register based on OldReg.
unsigned createFrom(unsigned OldReg);
/// create - Create a new register with the same class and original slot as
/// parent.
LiveInterval &createEmptyInterval() {
return createEmptyIntervalFrom(getReg());
}
unsigned create() {
return createFrom(getReg());
}
/// anyRematerializable - Return true if any parent values may be
/// rematerializable.
/// This function must be called before any rematerialization is attempted.
bool anyRematerializable(AliasAnalysis*);
/// checkRematerializable - Manually add VNI to the list of rematerializable
/// values if DefMI may be rematerializable.
bool checkRematerializable(VNInfo *VNI, const MachineInstr *DefMI,
AliasAnalysis*);
/// Remat - Information needed to rematerialize at a specific location.
struct Remat {
VNInfo *ParentVNI; // parent_'s value at the remat location.
MachineInstr *OrigMI; // Instruction defining ParentVNI.
explicit Remat(VNInfo *ParentVNI) : ParentVNI(ParentVNI), OrigMI(nullptr) {}
};
/// canRematerializeAt - Determine if ParentVNI can be rematerialized at
/// UseIdx. It is assumed that parent_.getVNINfoAt(UseIdx) == ParentVNI.
/// When cheapAsAMove is set, only cheap remats are allowed.
bool canRematerializeAt(Remat &RM,
SlotIndex UseIdx,
bool cheapAsAMove);
/// rematerializeAt - Rematerialize RM.ParentVNI into DestReg by inserting an
/// instruction into MBB before MI. The new instruction is mapped, but
/// liveness is not updated.
/// Return the SlotIndex of the new instruction.
SlotIndex rematerializeAt(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned DestReg,
const Remat &RM,
const TargetRegisterInfo&,
bool Late = false);
/// markRematerialized - explicitly mark a value as rematerialized after doing
/// it manually.
void markRematerialized(const VNInfo *ParentVNI) {
Rematted.insert(ParentVNI);
}
/// didRematerialize - Return true if ParentVNI was rematerialized anywhere.
bool didRematerialize(const VNInfo *ParentVNI) const {
return Rematted.count(ParentVNI);
}
/// eraseVirtReg - Notify the delegate that Reg is no longer in use, and try
/// to erase it from LIS.
void eraseVirtReg(unsigned Reg);
/// eliminateDeadDefs - Try to delete machine instructions that are now dead
/// (allDefsAreDead returns true). This may cause live intervals to be trimmed
/// and further dead efs to be eliminated.
/// RegsBeingSpilled lists registers currently being spilled by the register
/// allocator. These registers should not be split into new intervals
/// as currently those new intervals are not guaranteed to spill.
void eliminateDeadDefs(SmallVectorImpl<MachineInstr*> &Dead,
ArrayRef<unsigned> RegsBeingSpilled = None);
/// calculateRegClassAndHint - Recompute register class and hint for each new
/// register.
void calculateRegClassAndHint(MachineFunction&,
const MachineLoopInfo&,
const MachineBlockFrequencyInfo&);
};
}
#endif