mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 03:02:36 +01:00
67f2592469
We currently have problems with the way that low overhead loops are specified, with LR being spilled between the t2LoopDec and the t2LoopEnd forcing the entire loop to be reverted late in the backend. As they will eventually become a single instruction, this patch introduces a t2LoopEndDec which is the combination of the two, combined before registry allocation to make sure this does not fail. Unfortunately this instruction is a terminator that produces a value (and also branches - it only produces the value around the branching edge). So this needs some adjustment to phi elimination and the register allocator to make sure that we do not spill this LR def around the loop (needing to put a spill after the terminator). We treat the loop very carefully, making sure that there is nothing else like calls that would break it's ability to use LR. For that, this adds a isUnspillableTerminator to opt in the new behaviour. There is a chance that this could cause problems, and so I have added an escape option incase. But I have not seen any problems in the testing that I've tried, and not reverting Low overhead loops is important for our performance. If this does work then we can hopefully do the same for t2WhileLoopStart and t2DoLoopStart instructions. This patch also contains the code needed to convert or revert the t2LoopEndDec in the backend (which just needs a subs; bne) and the code pre-ra to create them. Differential Revision: https://reviews.llvm.org/D91358
311 lines
10 KiB
C++
311 lines
10 KiB
C++
//===- CalcSpillWeights.cpp -----------------------------------------------===//
|
|
//
|
|
// 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
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "llvm/CodeGen/CalcSpillWeights.h"
|
|
#include "llvm/ADT/SmallPtrSet.h"
|
|
#include "llvm/CodeGen/LiveInterval.h"
|
|
#include "llvm/CodeGen/LiveIntervals.h"
|
|
#include "llvm/CodeGen/MachineFunction.h"
|
|
#include "llvm/CodeGen/MachineInstr.h"
|
|
#include "llvm/CodeGen/MachineLoopInfo.h"
|
|
#include "llvm/CodeGen/MachineOperand.h"
|
|
#include "llvm/CodeGen/MachineRegisterInfo.h"
|
|
#include "llvm/CodeGen/TargetInstrInfo.h"
|
|
#include "llvm/CodeGen/TargetRegisterInfo.h"
|
|
#include "llvm/CodeGen/TargetSubtargetInfo.h"
|
|
#include "llvm/CodeGen/VirtRegMap.h"
|
|
#include "llvm/Support/Debug.h"
|
|
#include "llvm/Support/raw_ostream.h"
|
|
#include <cassert>
|
|
#include <tuple>
|
|
|
|
using namespace llvm;
|
|
|
|
#define DEBUG_TYPE "calcspillweights"
|
|
|
|
void VirtRegAuxInfo::calculateSpillWeightsAndHints() {
|
|
LLVM_DEBUG(dbgs() << "********** Compute Spill Weights **********\n"
|
|
<< "********** Function: " << MF.getName() << '\n');
|
|
|
|
MachineRegisterInfo &MRI = MF.getRegInfo();
|
|
for (unsigned I = 0, E = MRI.getNumVirtRegs(); I != E; ++I) {
|
|
unsigned Reg = Register::index2VirtReg(I);
|
|
if (MRI.reg_nodbg_empty(Reg))
|
|
continue;
|
|
calculateSpillWeightAndHint(LIS.getInterval(Reg));
|
|
}
|
|
}
|
|
|
|
// Return the preferred allocation register for reg, given a COPY instruction.
|
|
static Register copyHint(const MachineInstr *MI, unsigned Reg,
|
|
const TargetRegisterInfo &TRI,
|
|
const MachineRegisterInfo &MRI) {
|
|
unsigned Sub, HSub;
|
|
Register HReg;
|
|
if (MI->getOperand(0).getReg() == Reg) {
|
|
Sub = MI->getOperand(0).getSubReg();
|
|
HReg = MI->getOperand(1).getReg();
|
|
HSub = MI->getOperand(1).getSubReg();
|
|
} else {
|
|
Sub = MI->getOperand(1).getSubReg();
|
|
HReg = MI->getOperand(0).getReg();
|
|
HSub = MI->getOperand(0).getSubReg();
|
|
}
|
|
|
|
if (!HReg)
|
|
return 0;
|
|
|
|
if (Register::isVirtualRegister(HReg))
|
|
return Sub == HSub ? HReg : Register();
|
|
|
|
const TargetRegisterClass *rc = MRI.getRegClass(Reg);
|
|
MCRegister CopiedPReg = HSub ? TRI.getSubReg(HReg, HSub) : HReg.asMCReg();
|
|
if (rc->contains(CopiedPReg))
|
|
return CopiedPReg;
|
|
|
|
// Check if reg:sub matches so that a super register could be hinted.
|
|
if (Sub)
|
|
return TRI.getMatchingSuperReg(CopiedPReg, Sub, rc);
|
|
|
|
return 0;
|
|
}
|
|
|
|
// Check if all values in LI are rematerializable
|
|
static bool isRematerializable(const LiveInterval &LI, const LiveIntervals &LIS,
|
|
const VirtRegMap &VRM,
|
|
const TargetInstrInfo &TII) {
|
|
unsigned Reg = LI.reg();
|
|
unsigned Original = VRM.getOriginal(Reg);
|
|
for (LiveInterval::const_vni_iterator I = LI.vni_begin(), E = LI.vni_end();
|
|
I != E; ++I) {
|
|
const VNInfo *VNI = *I;
|
|
if (VNI->isUnused())
|
|
continue;
|
|
if (VNI->isPHIDef())
|
|
return false;
|
|
|
|
MachineInstr *MI = LIS.getInstructionFromIndex(VNI->def);
|
|
assert(MI && "Dead valno in interval");
|
|
|
|
// Trace copies introduced by live range splitting. The inline
|
|
// spiller can rematerialize through these copies, so the spill
|
|
// weight must reflect this.
|
|
while (MI->isFullCopy()) {
|
|
// The copy destination must match the interval register.
|
|
if (MI->getOperand(0).getReg() != Reg)
|
|
return false;
|
|
|
|
// Get the source register.
|
|
Reg = MI->getOperand(1).getReg();
|
|
|
|
// If the original (pre-splitting) registers match this
|
|
// copy came from a split.
|
|
if (!Register::isVirtualRegister(Reg) || VRM.getOriginal(Reg) != Original)
|
|
return false;
|
|
|
|
// Follow the copy live-in value.
|
|
const LiveInterval &SrcLI = LIS.getInterval(Reg);
|
|
LiveQueryResult SrcQ = SrcLI.Query(VNI->def);
|
|
VNI = SrcQ.valueIn();
|
|
assert(VNI && "Copy from non-existing value");
|
|
if (VNI->isPHIDef())
|
|
return false;
|
|
MI = LIS.getInstructionFromIndex(VNI->def);
|
|
assert(MI && "Dead valno in interval");
|
|
}
|
|
|
|
if (!TII.isTriviallyReMaterializable(*MI, LIS.getAliasAnalysis()))
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
void VirtRegAuxInfo::calculateSpillWeightAndHint(LiveInterval &LI) {
|
|
float Weight = weightCalcHelper(LI);
|
|
// Check if unspillable.
|
|
if (Weight < 0)
|
|
return;
|
|
LI.setWeight(Weight);
|
|
}
|
|
|
|
float VirtRegAuxInfo::futureWeight(LiveInterval &LI, SlotIndex Start,
|
|
SlotIndex End) {
|
|
return weightCalcHelper(LI, &Start, &End);
|
|
}
|
|
|
|
float VirtRegAuxInfo::weightCalcHelper(LiveInterval &LI, SlotIndex *Start,
|
|
SlotIndex *End) {
|
|
MachineRegisterInfo &MRI = MF.getRegInfo();
|
|
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
|
|
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
|
|
MachineBasicBlock *MBB = nullptr;
|
|
MachineLoop *Loop = nullptr;
|
|
bool IsExiting = false;
|
|
float TotalWeight = 0;
|
|
unsigned NumInstr = 0; // Number of instructions using LI
|
|
SmallPtrSet<MachineInstr *, 8> Visited;
|
|
|
|
std::pair<Register, Register> TargetHint = MRI.getRegAllocationHint(LI.reg());
|
|
|
|
if (LI.isSpillable()) {
|
|
Register Reg = LI.reg();
|
|
Register Original = VRM.getOriginal(Reg);
|
|
const LiveInterval &OrigInt = LIS.getInterval(Original);
|
|
// li comes from a split of OrigInt. If OrigInt was marked
|
|
// as not spillable, make sure the new interval is marked
|
|
// as not spillable as well.
|
|
if (!OrigInt.isSpillable())
|
|
LI.markNotSpillable();
|
|
}
|
|
|
|
// Don't recompute spill weight for an unspillable register.
|
|
bool IsSpillable = LI.isSpillable();
|
|
|
|
bool IsLocalSplitArtifact = Start && End;
|
|
|
|
// Do not update future local split artifacts.
|
|
bool ShouldUpdateLI = !IsLocalSplitArtifact;
|
|
|
|
if (IsLocalSplitArtifact) {
|
|
MachineBasicBlock *localMBB = LIS.getMBBFromIndex(*End);
|
|
assert(localMBB == LIS.getMBBFromIndex(*Start) &&
|
|
"start and end are expected to be in the same basic block");
|
|
|
|
// Local split artifact will have 2 additional copy instructions and they
|
|
// will be in the same BB.
|
|
// localLI = COPY other
|
|
// ...
|
|
// other = COPY localLI
|
|
TotalWeight += LiveIntervals::getSpillWeight(true, false, &MBFI, localMBB);
|
|
TotalWeight += LiveIntervals::getSpillWeight(false, true, &MBFI, localMBB);
|
|
|
|
NumInstr += 2;
|
|
}
|
|
|
|
// CopyHint is a sortable hint derived from a COPY instruction.
|
|
struct CopyHint {
|
|
const Register Reg;
|
|
const float Weight;
|
|
CopyHint(Register R, float W) : Reg(R), Weight(W) {}
|
|
bool operator<(const CopyHint &Rhs) const {
|
|
// Always prefer any physreg hint.
|
|
if (Reg.isPhysical() != Rhs.Reg.isPhysical())
|
|
return Reg.isPhysical();
|
|
if (Weight != Rhs.Weight)
|
|
return (Weight > Rhs.Weight);
|
|
return Reg.id() < Rhs.Reg.id(); // Tie-breaker.
|
|
}
|
|
};
|
|
|
|
std::set<CopyHint> CopyHints;
|
|
DenseMap<unsigned, float> Hint;
|
|
for (MachineRegisterInfo::reg_instr_nodbg_iterator
|
|
I = MRI.reg_instr_nodbg_begin(LI.reg()),
|
|
E = MRI.reg_instr_nodbg_end();
|
|
I != E;) {
|
|
MachineInstr *MI = &*(I++);
|
|
|
|
// For local split artifacts, we are interested only in instructions between
|
|
// the expected start and end of the range.
|
|
SlotIndex SI = LIS.getInstructionIndex(*MI);
|
|
if (IsLocalSplitArtifact && ((SI < *Start) || (SI > *End)))
|
|
continue;
|
|
|
|
NumInstr++;
|
|
if (MI->isIdentityCopy() || MI->isImplicitDef())
|
|
continue;
|
|
if (!Visited.insert(MI).second)
|
|
continue;
|
|
|
|
// For terminators that produce values, ask the backend if the register is
|
|
// not spillable.
|
|
if (TII.isUnspillableTerminator(MI) && MI->definesRegister(LI.reg())) {
|
|
LI.markNotSpillable();
|
|
return -1.0f;
|
|
}
|
|
|
|
float Weight = 1.0f;
|
|
if (IsSpillable) {
|
|
// Get loop info for mi.
|
|
if (MI->getParent() != MBB) {
|
|
MBB = MI->getParent();
|
|
Loop = Loops.getLoopFor(MBB);
|
|
IsExiting = Loop ? Loop->isLoopExiting(MBB) : false;
|
|
}
|
|
|
|
// Calculate instr weight.
|
|
bool Reads, Writes;
|
|
std::tie(Reads, Writes) = MI->readsWritesVirtualRegister(LI.reg());
|
|
Weight = LiveIntervals::getSpillWeight(Writes, Reads, &MBFI, *MI);
|
|
|
|
// Give extra weight to what looks like a loop induction variable update.
|
|
if (Writes && IsExiting && LIS.isLiveOutOfMBB(LI, MBB))
|
|
Weight *= 3;
|
|
|
|
TotalWeight += Weight;
|
|
}
|
|
|
|
// Get allocation hints from copies.
|
|
if (!MI->isCopy())
|
|
continue;
|
|
Register HintReg = copyHint(MI, LI.reg(), TRI, MRI);
|
|
if (!HintReg)
|
|
continue;
|
|
// Force hweight onto the stack so that x86 doesn't add hidden precision,
|
|
// making the comparison incorrectly pass (i.e., 1 > 1 == true??).
|
|
//
|
|
// FIXME: we probably shouldn't use floats at all.
|
|
volatile float HWeight = Hint[HintReg] += Weight;
|
|
if (HintReg.isVirtual() || MRI.isAllocatable(HintReg))
|
|
CopyHints.insert(CopyHint(HintReg, HWeight));
|
|
}
|
|
|
|
// Pass all the sorted copy hints to mri.
|
|
if (ShouldUpdateLI && CopyHints.size()) {
|
|
// Remove a generic hint if previously added by target.
|
|
if (TargetHint.first == 0 && TargetHint.second)
|
|
MRI.clearSimpleHint(LI.reg());
|
|
|
|
std::set<Register> HintedRegs;
|
|
for (auto &Hint : CopyHints) {
|
|
if (!HintedRegs.insert(Hint.Reg).second ||
|
|
(TargetHint.first != 0 && Hint.Reg == TargetHint.second))
|
|
// Don't add the same reg twice or the target-type hint again.
|
|
continue;
|
|
MRI.addRegAllocationHint(LI.reg(), Hint.Reg);
|
|
}
|
|
|
|
// Weakly boost the spill weight of hinted registers.
|
|
TotalWeight *= 1.01F;
|
|
}
|
|
|
|
// If the live interval was already unspillable, leave it that way.
|
|
if (!IsSpillable)
|
|
return -1.0;
|
|
|
|
// Mark li as unspillable if all live ranges are tiny and the interval
|
|
// is not live at any reg mask. If the interval is live at a reg mask
|
|
// spilling may be required.
|
|
if (ShouldUpdateLI && LI.isZeroLength(LIS.getSlotIndexes()) &&
|
|
!LI.isLiveAtIndexes(LIS.getRegMaskSlots())) {
|
|
LI.markNotSpillable();
|
|
return -1.0;
|
|
}
|
|
|
|
// If all of the definitions of the interval are re-materializable,
|
|
// it is a preferred candidate for spilling.
|
|
// FIXME: this gets much more complicated once we support non-trivial
|
|
// re-materialization.
|
|
if (isRematerializable(LI, LIS, VRM, *MF.getSubtarget().getInstrInfo()))
|
|
TotalWeight *= 0.5F;
|
|
|
|
if (IsLocalSplitArtifact)
|
|
return normalize(TotalWeight, Start->distance(*End), NumInstr);
|
|
return normalize(TotalWeight, LI.getSize(), NumInstr);
|
|
}
|