17#ifndef LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
18#define LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
36 const char *Names[] = {
"SGPR",
"VGPR",
"AGPR",
"AVGPR"};
61 if (UnifiedVGPRFile) {
79 return alignTo(NumArchVGPRs + NumAVGPRs,
99 unsigned DynamicVGPRBlockSize)
const {
100 return std::min(ST.getOccupancyWithNumSGPRs(
getSGPRNum()),
101 ST.getOccupancyWithNumVGPRs(
getVGPRNum(ST.hasGFX90AInsts()),
102 DynamicVGPRBlockSize));
106 unsigned AGPRThreshold,
unsigned CombinedThreshold) {
108 if (!ST.hasGFX90AInsts())
114 unsigned ArchSpill = ArchPressure > ArchVGPRThreshold
115 ? (ArchPressure - ArchVGPRThreshold)
118 AGPRPressure > AGPRThreshold ? (AGPRPressure - AGPRThreshold) : 0;
121 unsigned UnifiedSpill = UnifiedPressure > CombinedThreshold
122 ? (UnifiedPressure - CombinedThreshold)
125 return std::max(UnifiedSpill, ArchSpill + AGPRSpill);
134 unsigned DynamicVGPRBlockSize)
const {
136 O.getOccupancy(ST, DynamicVGPRBlockSize);
152 unsigned MaxOccupancy = std::numeric_limits<unsigned>::max())
const;
157 return !(*
this == O);
161 for (
unsigned I = 0;
I < ValueArraySize; ++
I)
162 Value[
I] +=
RHS.Value[
I];
167 for (
unsigned I = 0;
I < ValueArraySize; ++
I)
168 Value[
I] -=
RHS.Value[
I];
181 static constexpr unsigned ValueArraySize =
TOTAL_KINDS * 2;
185 std::array<unsigned, ValueArraySize>
Value;
194 unsigned DynamicVGPRBlockSize);
199 for (
unsigned I = 0;
I < GCNRegPressure::ValueArraySize; ++
I)
200 Res.Value[
I] = std::max(P1.Value[
I], P2.Value[
I]);
243 void setTarget(
unsigned NumSGPRs,
unsigned NumVGPRs);
264 assert(!RP.less(MF, SaveRP) &&
"saving beyond current RP");
276 return UnifiedRF ? MaxUnifiedVGPRs : MaxVGPRs;
279#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
281 OS <<
"Actual/Target: " <<
Target.RP.getSGPRNum() <<
'/' <<
Target.MaxSGPRs
282 <<
" SGPRs, " <<
Target.RP.getArchVGPRNum() <<
'/' <<
Target.MaxVGPRs
283 <<
" ArchVGPRs, " <<
Target.RP.getAGPRNum() <<
'/' <<
Target.MaxVGPRs
286 if (
Target.MaxUnifiedVGPRs) {
287 OS <<
", " <<
Target.RP.getVGPRNum(
true) <<
'/' <<
Target.MaxUnifiedVGPRs
288 <<
" VGPRs (unified)";
296 const bool UnifiedRF;
307 unsigned MaxUnifiedVGPRs;
310 : MF(MF), UnifiedRF(MF.getSubtarget<
GCNSubtarget>().hasGFX90AInsts()),
356 const MachineRegisterInfo &
MRI,
376 reset(
MBB.getParent()->getRegInfo(), MBBLastSlot);
381 reset(
MI.getMF()->getRegInfo(),
LIS.getInstructionIndex(
MI).getDeadSlot());
441 bool UseInternalIterator =
true);
450 bool UseInternalIterator =
true);
482 const LiveIntervals &LIS,
483 const MachineRegisterInfo &
MRI,
487 const MachineRegisterInfo &
MRI,
495template <
typename Range>
496DenseMap<MachineInstr*, GCNRPTracker::LiveRegSet>
498 std::vector<SlotIndex> Indexes;
502 auto SI = SII.getInstructionIndex(*
I);
503 Indexes.push_back(After ?
SI.getDeadSlot() :
SI.getBaseIndex());
507 auto &
MRI = (*R.begin())->getMF()->getRegInfo();
510 for (
unsigned I = 0,
E =
MRI.getNumVirtRegs();
I !=
E; ++
I) {
516 if (!LI.findIndexesLiveAt(Indexes, std::back_inserter(LiveIdxs)))
518 if (!LI.hasSubRanges()) {
519 for (
auto SI : LiveIdxs)
520 LiveRegMap[SII.getInstructionFromIndex(
SI)][
Reg] =
521 MRI.getMaxLaneMaskForVReg(
Reg);
523 for (
const auto &S : LI.subranges()) {
526 S.findIndexesLiveAt(LiveIdxs, std::back_inserter(SRLiveIdxs));
527 for (
auto SI : SRLiveIdxs)
528 LiveRegMap[SII.getInstructionFromIndex(
SI)][
Reg] |= S.LaneMask;
537 MI.getMF()->getRegInfo());
543 MI.getMF()->getRegInfo());
546template <
typename Range>
558Printable
print(
const GCNRegPressure &RP,
const GCNSubtarget *ST =
nullptr,
559 unsigned DynamicVGPRBlockSize = 0);
562 const MachineRegisterInfo &
MRI);
566 const TargetRegisterInfo *
TRI, StringRef Pfx =
" ");
586 const MachineLoopInfo *MLI);
unsigned const MachineRegisterInfo * MRI
assert(UImm &&(UImm !=~static_cast< T >(0)) &&"Invalid immediate!")
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
AMD GCN specific subclass of TargetSubtarget.
Register const TargetRegisterInfo * TRI
ConstantRange Range(APInt(BitWidth, Low), APInt(BitWidth, High))
StandardInstrumentations SI(Mod->getContext(), Debug, VerifyEach)
Represent the analysis usage information of a pass.
AnalysisUsage & addRequired()
void setPreservesAll()
Set by analyses that do not transform their input at all.
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
GCNRegPressure moveMaxPressure()
return MaxPressure and clear it.
bool advanceBeforeNext(MachineInstr *MI=nullptr, bool UseInternalIterator=true)
Move to the state right before the next MI or after the end of MBB.
bool advance(MachineInstr *MI=nullptr, bool UseInternalIterator=true)
Move to the state at the next MI.
GCNRegPressure bumpDownwardPressure(const MachineInstr *MI, const SIRegisterInfo *TRI) const
Mostly copy/paste from CodeGen/RegisterPressure.cpp Calculate the impact MI will have on CurPressure ...
MachineBasicBlock::const_iterator getNext() const
GCNDownwardRPTracker(const LiveIntervals &LIS_)
bool reset(const MachineInstr &MI, const LiveRegSet *LiveRegs=nullptr)
Reset tracker to the point before the MI filling LiveRegs upon this point using LIS.
void advanceToNext(MachineInstr *MI=nullptr, bool UseInternalIterator=true)
Move to the state at the MI, advanceBeforeNext has to be called first.
GCNRPTarget(const MachineFunction &MF, const GCNRegPressure &RP)
Sets up the target such that the register pressure starting at RP does not show register spilling on ...
unsigned getMaxVGPRs() const
bool isSaveBeneficial(Register Reg) const
Determines whether saving virtual register Reg will be beneficial towards achieving the RP target.
void saveReg(Register Reg, LaneBitmask Mask, const MachineRegisterInfo &MRI)
Saves virtual register Reg with lanemask Mask.
bool hasVectorRegisterExcess() const
void setRP(const GCNRegPressure &NewRP)
bool satisfied() const
Whether the current RP is at or below the defined pressure target.
unsigned getMaxSGPRs() const
const GCNRegPressure & getCurrentRP() const
void setTarget(unsigned NumSGPRs, unsigned NumVGPRs)
Changes the target (same semantics as constructor).
friend raw_ostream & operator<<(raw_ostream &OS, const GCNRPTarget &Target)
void saveRP(const GCNRegPressure &SaveRP)
Saves a total pressure of SaveRP.
unsigned getNumRegsBenefit(const GCNRegPressure &SaveRP) const
Returns the benefit towards achieving the RP target that saving SaveRP represents,...
GCNRegPressure getPressure() const
const decltype(LiveRegs) & getLiveRegs() const
const MachineInstr * LastTrackedMI
decltype(LiveRegs) moveLiveRegs()
GCNRegPressure CurPressure
DenseMap< unsigned, LaneBitmask > LiveRegSet
LaneBitmask getLastUsedLanes(Register Reg, SlotIndex Pos) const
Mostly copy/paste from CodeGen/RegisterPressure.cpp.
GCNRPTracker(const LiveIntervals &LIS_)
GCNRegPressure MaxPressure
void reset(const MachineInstr &MI, const LiveRegSet *LiveRegsCopy, bool After)
void bumpDeadDefs(ArrayRef< VRegMaskOrUnit > DeadDefs)
Mostly copy/paste from CodeGen/RegisterPressure.cpp.
const MachineInstr * getLastTrackedMI() const
const MachineRegisterInfo * MRI
const LiveIntervals & LIS
GCNUpwardRPTracker(const LiveIntervals &LIS_)
GCNRegPressure getMaxPressureAndReset()
void reset(const MachineRegisterInfo &MRI, SlotIndex SI)
reset tracker at the specified slot index SI.
void recede(const MachineInstr &MI)
Move to the state of RP just before the MI .
const GCNRegPressure & getMaxPressure() const
void reset(const MachineBasicBlock &MBB)
reset tracker to the end of the MBB.
bool isValid() const
returns whether the tracker's state after receding MI corresponds to reported by LIS.
void reset(const MachineInstr &MI)
reset tracker to the point just after MI (in program order).
bool hasInterval(Register Reg) const
SlotIndexes * getSlotIndexes() const
SlotIndex getInstructionIndex(const MachineInstr &Instr) const
Returns the base index of the given instruction.
LiveInterval & getInterval(Register Reg)
A set of live virtual registers and physical register units.
MachineInstrBundleIterator< const MachineInstr > const_iterator
MachineFunctionPass(char &ID)
void getAnalysisUsage(AnalysisUsage &AU) const override
getAnalysisUsage - Subclasses that override getAnalysisUsage must call this.
const TargetSubtargetInfo & getSubtarget() const
getSubtarget - Return the subtarget for which this machine code is being compiled.
Representation of each machine instruction.
MachineRegisterInfo - Keep track of information for virtual and physical registers,...
Simple wrapper around std::function<void(raw_ostream&)>.
Wrapper class representing virtual and physical registers.
static Register index2VirtReg(unsigned Index)
Convert a 0-based index to a virtual register number.
SlotIndex - An opaque wrapper around machine indexes.
SlotIndex getDeadSlot() const
Returns the dead def kill slot for the current instruction.
SlotIndex getBaseIndex() const
Returns the base index for associated with this index.
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
TargetRegisterInfo base class - We assume that the target defines a static array of TargetRegisterDes...
Target - Wrapper for Target specific information.
LLVM Value Representation.
This class implements an extremely fast bulk output stream that can only output to a stream.
unsigned getArchVGPRAllocGranule()
For subtargets with a unified VGPR file and mixed ArchVGPR/AGPR usage, returns the allocation granule...
This is an optimization pass for GlobalISel generic memory operations.
GCNRegPressure max(const GCNRegPressure &P1, const GCNRegPressure &P2)
LaneBitmask getLiveLaneMask(unsigned Reg, SlotIndex SI, const LiveIntervals &LIS, const MachineRegisterInfo &MRI, LaneBitmask LaneMaskFilter=LaneBitmask::getAll())
bool isEqual(const GCNRPTracker::LiveRegSet &S1, const GCNRPTracker::LiveRegSet &S2)
Printable print(const GCNRegPressure &RP, const GCNSubtarget *ST=nullptr, unsigned DynamicVGPRBlockSize=0)
auto size(R &&Range, std::enable_if_t< std::is_base_of< std::random_access_iterator_tag, typename std::iterator_traits< decltype(Range.begin())>::iterator_category >::value, void > *=nullptr)
Get the size of a range.
GCNRPTracker::LiveRegSet getLiveRegs(SlotIndex SI, const LiveIntervals &LIS, const MachineRegisterInfo &MRI, GCNRegPressure::RegKind RegKind=GCNRegPressure::TOTAL_KINDS)
GCNRegPressure getRegPressure(const MachineRegisterInfo &MRI, Range &&LiveRegs)
GCNRPTracker::LiveRegSet getLiveRegsAfter(const MachineInstr &MI, const LiveIntervals &LIS)
void sort(IteratorTy Start, IteratorTy End)
uint64_t alignTo(uint64_t Size, Align A)
Returns a multiple of A needed to store Size bytes.
DenseMap< MachineInstr *, GCNRPTracker::LiveRegSet > getLiveRegMap(Range &&R, bool After, LiveIntervals &LIS)
creates a map MachineInstr -> LiveRegSet R - range of iterators on instructions After - upon entry or...
APInt operator+(APInt a, const APInt &b)
GCNRPTracker::LiveRegSet getLiveRegsBefore(const MachineInstr &MI, const LiveIntervals &LIS)
LLVM_ABI void dumpMaxRegPressure(MachineFunction &MF, GCNRegPressure::RegKind Kind, LiveIntervals &LIS, const MachineLoopInfo *MLI)
Printable reportMismatch(const GCNRPTracker::LiveRegSet &LISLR, const GCNRPTracker::LiveRegSet &TrackedL, const TargetRegisterInfo *TRI, StringRef Pfx=" ")
void getAnalysisUsage(AnalysisUsage &AU) const override
getAnalysisUsage - This function should be overriden by passes that need analysis information to do t...
bool runOnMachineFunction(MachineFunction &MF) override
runOnMachineFunction - This method must be overloaded to perform the desired machine code transformat...
static RegKind getRegKind(unsigned Reg, const MachineRegisterInfo &MRI)
static constexpr const char * getName(RegKind Kind)
bool operator!=(const GCNRegPressure &O) const
GCNRegPressure & operator+=(const GCNRegPressure &RHS)
unsigned getNumRegs(RegKind Kind) const
unsigned getVGPRTuplesWeight() const
GCNRegPressure & operator-=(const GCNRegPressure &RHS)
unsigned getVGPRSpills(MachineFunction &MF, unsigned ArchVGPRThreshold, unsigned AGPRThreshold, unsigned CombinedThreshold)
unsigned getVGPRNum(bool UnifiedVGPRFile) const
friend Printable print(const GCNRegPressure &RP, const GCNSubtarget *ST, unsigned DynamicVGPRBlockSize)
unsigned getOccupancy(const GCNSubtarget &ST, unsigned DynamicVGPRBlockSize) const
friend GCNRegPressure max(const GCNRegPressure &P1, const GCNRegPressure &P2)
void inc(unsigned Reg, LaneBitmask PrevMask, LaneBitmask NewMask, const MachineRegisterInfo &MRI)
bool higherOccupancy(const GCNSubtarget &ST, const GCNRegPressure &O, unsigned DynamicVGPRBlockSize) const
unsigned getArchVGPRNum() const
unsigned getAGPRNum() const
unsigned getSGPRNum() const
unsigned getSGPRTuplesWeight() const
bool operator==(const GCNRegPressure &O) const
static unsigned getUnifiedVGPRNum(unsigned NumArchVGPRs, unsigned NumAGPRs, unsigned NumAVGPRs)
Returns the aggregated VGPR pressure, assuming NumArchVGPRs ArchVGPRs NumAGPRs AGPRS,...
unsigned getAVGPRNum() const
bool less(const MachineFunction &MF, const GCNRegPressure &O, unsigned MaxOccupancy=std::numeric_limits< unsigned >::max()) const
Compares this GCNRegpressure to O, returning true if this is less.
static constexpr LaneBitmask getAll()
static constexpr LaneBitmask getNone()