blob: 1ff4356b6046137ef45ad16e3837b41d9f99d9d4 [file] [log] [blame]
//===- MLRegAllocEvictAdvisor.cpp - ML eviction advisor -------------------===//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
// Implementation of the ML eviction advisor and reward injection pass
#include "AllocationOrder.h"
#include "RegAllocEvictionAdvisor.h"
#include "RegAllocGreedy.h"
#include "llvm/Analysis/InteractiveModelRunner.h"
#include "llvm/Analysis/MLModelRunner.h"
#include "llvm/Analysis/TensorSpec.h"
#include "llvm/Analysis/ModelUnderTrainingRunner.h"
#include "llvm/Analysis/NoInferenceModelRunner.h"
#include "llvm/Analysis/Utils/TrainingLogger.h"
#include "MLRegallocEvictAdvisor.h"
#include "llvm/Analysis/ReleaseModeModelRunner.h"
#include "llvm/CodeGen/CalcSpillWeights.h"
#include "llvm/CodeGen/LiveRegMatrix.h"
#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/RegisterClassInfo.h"
#include "llvm/CodeGen/VirtRegMap.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/PassRegistry.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include <array>
#include <memory>
using namespace llvm;
#define DEBUG_TYPE "ml-regalloc"
// Generated header in release (AOT) mode
#include "RegallocEvictModel.h"
using CompiledModelType = RegallocEvictModel;
using CompiledModelType = NoopSavedModelImpl;
static cl::opt<std::string> InteractiveChannelBaseName(
"regalloc-evict-interactive-channel-base", cl::Hidden,
"Base file path for the interactive mode. The incoming filename should "
"have the name <regalloc-evict-interactive-channel-base>.in, while the "
"outgoing name should be "
// Options that only make sense in development mode
#include "RegAllocScore.h"
#include "llvm/Analysis/Utils/TFUtils.h"
static cl::opt<std::string> TrainingLog(
"regalloc-training-log", cl::Hidden,
cl::desc("Training log for the register allocator eviction model"));
static cl::opt<std::string> ModelUnderTraining(
"regalloc-model", cl::Hidden,
cl::desc("The model being trained for register allocation eviction"));
static cl::opt<bool> EnableDevelopmentFeatures(
"regalloc-enable-development-features", cl::Hidden,
cl::desc("Whether or not to enable features under development for the ML "
"regalloc advisor"));
static const bool EnableDevelopmentFeatures = false;
#endif // #ifdef LLVM_HAVE_TFLITE
/// The score injection pass.
/// This pass calculates the score for a function and inserts it in the log, but
/// this happens only in development mode. It's a no-op otherwise.
namespace llvm {
extern cl::opt<unsigned> EvictInterferenceCutoff;
class RegAllocScoring : public MachineFunctionPass {
static char ID;
RegAllocScoring() : MachineFunctionPass(ID) {
~RegAllocScoring() override = default;
StringRef getPassName() const override {
return "Register Allocation Pass Scoring";
/// RegAllocReward analysis usage.
void getAnalysisUsage(AnalysisUsage &AU) const override {
/// Performs this pass
bool runOnMachineFunction(MachineFunction &) override;
char RegAllocScoring::ID = 0;
FunctionPass *createRegAllocScoringPass() { return new RegAllocScoring(); }
} // namespace llvm
INITIALIZE_PASS(RegAllocScoring, "regallocscoringpass",
"Register Allocation Scoring Pass", false, false)
// ===================================
// Common ML Advisor declarations
// ===================================
namespace {
// The model can only accept a specified number of opcodes and will error it if
// fed an opcode it hasn't seen before. This constant sets the current cutoff.
static const int OpcodeValueCutoff = 17716;
// Most features are as described above, so we'll reuse this vector in defining
// them.
static const std::vector<int64_t> PerLiveRangeShape{1, NumberOfInterferences};
// --------------
// Features table
// --------------
// For each interfering live range (incl. the candidate) we collect a number of
// features. However, because the features are of different types (and because
// of ML best practices), we organize the tensors per feature, not per
// candidate. Each such tensor has a scalar value corresponding to the
// interferring live range at that position, in the order in AllocationOrder.
// The last position corresponds to the virt reg seeking allocation.
// Exception to all that is the progression feature, which is just a scalar (see
// its documentation for details).
// Note on naming: the "_by_max" are normalized using the largest value of that
// tensor, as observed in the current decision making stage (i.e. for the
// current call to the advisor's tryFindEvictionCandidate)
// The feature list format: type, name, shape, documentation.
// Note: we can really just use int64 and float, hence the modeling of some
// bools as int64 values.
M(int64_t, mask, PerLiveRangeShape, \
"boolean values, 0 for unavailable candidates (i.e. if a position is 0, " \
"it " \
"can't be evicted)") \
M(int64_t, is_free, PerLiveRangeShape, \
"boolean values, 1 if this phys reg is actually free (no interferences)") \
M(float, nr_urgent, PerLiveRangeShape, \
"number of 'urgent' intervals, normalized. Urgent are those that are OK " \
"to break cascades") \
M(float, nr_broken_hints, PerLiveRangeShape, \
"if this position were evicted, how many broken hints would there be") \
M(int64_t, is_hint, PerLiveRangeShape, \
"is this a preferred phys reg for the candidate") \
M(int64_t, is_local, PerLiveRangeShape, \
"is this live range local to a basic block") \
M(float, nr_rematerializable, PerLiveRangeShape, \
"nr rematerializable ranges") \
M(float, nr_defs_and_uses, PerLiveRangeShape, \
"bb freq - weighed nr defs and uses") \
M(float, weighed_reads_by_max, PerLiveRangeShape, \
"bb freq - weighed nr of reads, normalized") \
M(float, weighed_writes_by_max, PerLiveRangeShape, \
"bb feq - weighed nr of writes, normalized") \
M(float, weighed_read_writes_by_max, PerLiveRangeShape, \
"bb freq - weighed nr of uses that are both read and writes, normalized") \
M(float, weighed_indvars_by_max, PerLiveRangeShape, \
"bb freq - weighed nr of uses that are indvars, normalized") \
M(float, hint_weights_by_max, PerLiveRangeShape, \
"bb freq - weighed nr of uses that are hints, normalized") \
M(float, start_bb_freq_by_max, PerLiveRangeShape, \
"the freq in the start block, normalized") \
M(float, end_bb_freq_by_max, PerLiveRangeShape, \
"freq of end block, normalized") \
M(float, hottest_bb_freq_by_max, PerLiveRangeShape, \
"hottest BB freq, normalized") \
M(float, liverange_size, PerLiveRangeShape, \
"size (instr index diff) of the LR") \
M(float, use_def_density, PerLiveRangeShape, \
"the max weight, as computed by the manual heuristic") \
M(int64_t, max_stage, PerLiveRangeShape, \
"largest stage of an interval in this LR") \
M(int64_t, min_stage, PerLiveRangeShape, \
"lowest stage of an interval in this LR") \
M(float, progress, {1}, "ratio of current queue size to initial size")
M(int64_t, instructions, InstructionsShape, \
"Opcodes of the instructions covered by the eviction problem")
M(int64_t, instructions_mapping, InstructionsMappingShape, \
"A binary matrix mapping LRs to instruction opcodes") \
M(float, mbb_frequencies, MBBFrequencyShape, \
"A vector of machine basic block frequencies") \
M(int64_t, mbb_mapping, InstructionsShape, \
"A vector of indicies mapping instructions to MBBs")
// The model learns to pick one of the mask == 1 interferences. This is the
// name of the output tensor. The contract with the model is that the output
// will be guaranteed to be to a mask == 1 position. Using a macro here to
// avoid 'not used' warnings (and keep cond compilation to a minimum)
#define DecisionName "index_to_evict"
static const TensorSpec DecisionSpec =
TensorSpec::createSpec<int64_t>(DecisionName, {1});
// Named features index.
enum FeatureIDs {
#define _FEATURE_IDX_SIMPLE(_, name, __, ___) name
#endif // #ifdef LLVM_HAVE_TFLITE
// The ML advisor will typically have a sparse input to the evaluator, because
// various phys regs won't be available. It's easier (maintenance-wise) to
// bulk-reset the state of the evaluator each time we are about to use it
// again.
template <typename T> size_t getTotalSize(const std::vector<int64_t> &Shape) {
size_t Ret = sizeof(T);
for (const auto V : Shape)
Ret *= V;
return Ret;
void resetInputs(MLModelRunner &Runner) {
#define _RESET(TYPE, NAME, SHAPE, __) \
std::memset(Runner.getTensorUntyped(FeatureIDs::NAME), 0, \
if (EnableDevelopmentFeatures) {
#undef _RESET
// Per-live interval components that get aggregated into the feature values
// that will be passed to the evaluator.
struct LIFeatureComponents {
double R = 0;
double W = 0;
double RW = 0;
double IndVarUpdates = 0;
double HintWeights = 0.0;
int64_t NrDefsAndUses = 0;
float HottestBlockFreq = 0.0;
bool IsRemat = false;
using CandidateRegList =
std::array<std::pair<MCRegister, bool>, NumberOfInterferences>;
using FeaturesListNormalizer =
llvm::SmallVector<float, FeatureIDs::FeatureCount>;
/// The ML evictor (commonalities between release and development mode)
class MLEvictAdvisor : public RegAllocEvictionAdvisor {
MLEvictAdvisor(const MachineFunction &MF, const RAGreedy &RA,
MLModelRunner *Runner, const MachineBlockFrequencyInfo &MBFI,
const MachineLoopInfo &Loops);
const RegAllocEvictionAdvisor &getDefaultAdvisor() const {
return static_cast<const RegAllocEvictionAdvisor &>(DefaultAdvisor);
// The assumption is that if the Runner could not be constructed, we emit-ed
// error, and we shouldn't be asking for it here.
const MLModelRunner &getRunner() const { return *Runner; }
/// This just calls Evaluate on the Runner, but in the development mode
/// case, if we're just capturing the log of the default advisor, it needs
/// to call the latter instead, so we need to pass all the necessary
/// parameters for it. In the development case, it will also log.
virtual int64_t
tryFindEvictionCandidatePosition(const LiveInterval &VirtReg,
const AllocationOrder &Order,
unsigned OrderLimit, uint8_t CostPerUseLimit,
const SmallVirtRegSet &FixedRegisters) const;
/// Load the features of the given VirtReg (allocated or not) at column Pos,
/// but if that can't be evicted, return false instead.
loadInterferenceFeatures(const LiveInterval &VirtReg, MCRegister PhysReg,
bool IsHint, const SmallVirtRegSet &FixedRegisters,
llvm::SmallVectorImpl<float> &Largest, size_t Pos,
SmallVectorImpl<LRStartEndInfo> &LRPosInfo) const;
static float getInitialQueueSize(const MachineFunction &MF);
MCRegister tryFindEvictionCandidate(
const LiveInterval &VirtReg, const AllocationOrder &Order,
uint8_t CostPerUseLimit,
const SmallVirtRegSet &FixedRegisters) const override;
void extractFeatures(const SmallVectorImpl<const LiveInterval *> &Intervals,
llvm::SmallVectorImpl<float> &Largest, size_t Pos,
int64_t IsHint, int64_t LocalIntfsCount, float NrUrgent,
SmallVectorImpl<LRStartEndInfo> &LRPosInfo) const;
// Point-in-time: we didn't learn this, so we always delegate to the
// default.
bool canEvictHintInterference(
const LiveInterval &VirtReg, MCRegister PhysReg,
const SmallVirtRegSet &FixedRegisters) const override {
return getDefaultAdvisor().canEvictHintInterference(VirtReg, PhysReg,
const LIFeatureComponents &
getLIFeatureComponents(const LiveInterval &LI) const;
// Hold on to a default advisor for:
// 1) the implementation of canEvictHintInterference, because we didn't
// learn that nuance yet; 2) for bootstrapping (logging) in the development
// mode case.
const DefaultEvictionAdvisor DefaultAdvisor;
MLModelRunner *const Runner;
const MachineBlockFrequencyInfo &MBFI;
const MachineLoopInfo &Loops;
// Indices of those features we don't want to normalize.
// This could be static and shared, but its initialization is non-trivial.
std::bitset<FeatureIDs::FeatureCount> DoNotNormalize;
const float InitialQSize;
using RegID = unsigned;
mutable DenseMap<RegID, LIFeatureComponents> CachedFeatures;
#define _DECL_FEATURES(type, name, shape, _) \
TensorSpec::createSpec<type>(#name, shape),
// ===================================
// Release (AOT) - specifics
// ===================================
class ReleaseModeEvictionAdvisorAnalysis final
: public RegAllocEvictionAdvisorAnalysis {
: RegAllocEvictionAdvisorAnalysis(AdvisorMode::Release) {
if (EnableDevelopmentFeatures) {
} else {
// support for isa<> and dyn_cast.
static bool classof(const RegAllocEvictionAdvisorAnalysis *R) {
return R->getAdvisorMode() == AdvisorMode::Release;
std::vector<TensorSpec> InputFeatures;
void getAnalysisUsage(AnalysisUsage &AU) const override {
getAdvisor(const MachineFunction &MF, const RAGreedy &RA) override {
if (!Runner) {
if (InteractiveChannelBaseName.empty())
Runner = std::make_unique<ReleaseModeModelRunner<CompiledModelType>>(
MF.getFunction().getContext(), InputFeatures, DecisionName);
Runner = std::make_unique<InteractiveModelRunner>(
MF.getFunction().getContext(), InputFeatures, DecisionSpec,
InteractiveChannelBaseName + ".out",
InteractiveChannelBaseName + ".in");
return std::make_unique<MLEvictAdvisor>(
MF, RA, Runner.get(), getAnalysis<MachineBlockFrequencyInfo>(),
std::unique_ptr<MLModelRunner> Runner;
// ===================================
// Development mode-specifics
// ===================================
// Features we log
static const TensorSpec Reward = TensorSpec::createSpec<float>("reward", {1});
// Features we bind on the model. The tensor names have a prefix, and we also
// need to include some tensors that are expected to be present by the
// training algo.
// TODO: can we just get rid of these?
#define _DECL_TRAIN_FEATURES(type, name, shape, _) \
TensorSpec::createSpec<type>(std::string("action_") + #name, shape),
class DevelopmentModeEvictAdvisor : public MLEvictAdvisor {
DevelopmentModeEvictAdvisor(const MachineFunction &MF, const RAGreedy &RA,
MLModelRunner *Runner,
const MachineBlockFrequencyInfo &MBFI,
const MachineLoopInfo &Loops, Logger *Log)
: MLEvictAdvisor(MF, RA, Runner, MBFI, Loops), Log(Log) {}
int64_t tryFindEvictionCandidatePosition(
const LiveInterval &VirtReg, const AllocationOrder &Order,
unsigned OrderLimit, uint8_t CostPerUseLimit,
const SmallVirtRegSet &FixedRegisters) const override;
Logger *const Log;
class DevelopmentModeEvictionAdvisorAnalysis final
: public RegAllocEvictionAdvisorAnalysis {
: RegAllocEvictionAdvisorAnalysis(AdvisorMode::Development) {
if (EnableDevelopmentFeatures) {
TrainingInputFeatures = {
TensorSpec::createSpec<float>("action_discount", {1}),
TensorSpec::createSpec<int32_t>("action_step_type", {1}),
TensorSpec::createSpec<float>("action_reward", {1})};
} else {
TrainingInputFeatures = {
TensorSpec::createSpec<float>("action_discount", {1}),
TensorSpec::createSpec<int32_t>("action_step_type", {1}),
TensorSpec::createSpec<float>("action_reward", {1})};
// support for isa<> and dyn_cast.
static bool classof(const RegAllocEvictionAdvisorAnalysis *R) {
return R->getAdvisorMode() == AdvisorMode::Development;
void logRewardIfNeeded(const MachineFunction &MF,
llvm::function_ref<float()> GetReward) override {
if (!Log || !Log->hasAnyObservationForContext(MF.getName()))
// The function pass manager would run all the function passes for a
// function, so we assume the last context belongs to this function. If
// this invariant ever changes, we can implement at that time switching
// contexts. At this point, it'd be an error
if (Log->currentContext() != MF.getName()) {
"The training log context shouldn't have had changed.");
if (Log->hasObservationInProgress())
std::vector<TensorSpec> InputFeatures;
std::vector<TensorSpec> TrainingInputFeatures;
void getAnalysisUsage(AnalysisUsage &AU) const override {
bool doInitialization(Module &M) override {
LLVMContext &Ctx = M.getContext();
if (ModelUnderTraining.empty() && TrainingLog.empty()) {
Ctx.emitError("Regalloc development mode should be requested with at "
"least logging enabled and/or a training model");
return false;
if (ModelUnderTraining.empty())
Runner = std::make_unique<NoInferenceModelRunner>(Ctx, InputFeatures);
Runner = ModelUnderTrainingRunner::createAndEnsureValid(
Ctx, ModelUnderTraining, DecisionName, TrainingInputFeatures);
if (!Runner) {
Ctx.emitError("Regalloc: could not set up the model runner");
return false;
if (TrainingLog.empty())
return false;
std::error_code EC;
auto OS = std::make_unique<raw_fd_ostream>(TrainingLog, EC);
if (EC) {
M.getContext().emitError(EC.message() + ":" + TrainingLog);
return false;
std::vector<TensorSpec> LFS = InputFeatures;
if (auto *MUTR = dyn_cast<ModelUnderTrainingRunner>(Runner.get()))
append_range(LFS, MUTR->extraOutputsForLoggingSpecs());
// We always log the output; in particular, if we're not evaluating, we
// don't have an output spec json file. That's why we handle the
// 'normal' output separately.
Log = std::make_unique<Logger>(std::move(OS), LFS, Reward,
/*IncludeReward*/ true);
return false;
getAdvisor(const MachineFunction &MF, const RAGreedy &RA) override {
if (!Runner)
return nullptr;
if (Log)
return std::make_unique<DevelopmentModeEvictAdvisor>(
MF, RA, Runner.get(), getAnalysis<MachineBlockFrequencyInfo>(),
getAnalysis<MachineLoopInfo>(), Log.get());
std::unique_ptr<MLModelRunner> Runner;
std::unique_ptr<Logger> Log;
#endif //#ifdef LLVM_HAVE_TFLITE
} // namespace
float MLEvictAdvisor::getInitialQueueSize(const MachineFunction &MF) {
auto &MRI = MF.getRegInfo();
float Ret = 0.0;
for (unsigned I = 0, E = MRI.getNumVirtRegs(); I != E; ++I) {
Register Reg = Register::index2VirtReg(I);
if (MRI.reg_nodbg_empty(Reg))
return Ret;
MLEvictAdvisor::MLEvictAdvisor(const MachineFunction &MF, const RAGreedy &RA,
MLModelRunner *Runner,
const MachineBlockFrequencyInfo &MBFI,
const MachineLoopInfo &Loops)
: RegAllocEvictionAdvisor(MF, RA), DefaultAdvisor(MF, RA),
Runner(std::move(Runner)), MBFI(MBFI), Loops(Loops),
InitialQSize(MLEvictAdvisor::getInitialQueueSize(MF)) {
int64_t MLEvictAdvisor::tryFindEvictionCandidatePosition(
const LiveInterval &, const AllocationOrder &, unsigned, uint8_t,
const SmallVirtRegSet &) const {
int64_t Ret = Runner->evaluate<int64_t>();
assert(Ret >= 0);
assert(Ret <= CandidateVirtRegPos);
return Ret;
bool MLEvictAdvisor::loadInterferenceFeatures(
const LiveInterval &VirtReg, MCRegister PhysReg, bool IsHint,
const SmallVirtRegSet &FixedRegisters,
llvm::SmallVectorImpl<float> &Largest, size_t Pos,
llvm::SmallVectorImpl<LRStartEndInfo> &LRPosInfo) const {
// It is only possible to evict virtual register interference.
if (Matrix->checkInterference(VirtReg, PhysReg) > LiveRegMatrix::IK_VirtReg) {
// leave unavailable
return false;
const bool IsLocal = LIS->intervalIsInOneMBB(VirtReg);
int64_t LocalIntfs = 0;
float NrUrgent = 0.0f;
// The cascade tracking is the same as in the default advisor
unsigned Cascade = RA.getExtraInfo().getCascadeOrCurrentNext(VirtReg.reg());
SmallVector<const LiveInterval *, MaxInterferences> InterferingIntervals;
for (MCRegUnitIterator Units(PhysReg, TRI); Units.isValid(); ++Units) {
LiveIntervalUnion::Query &Q = Matrix->query(VirtReg, *Units);
// Different from the default heuristic, we don't make any assumptions
// about what having more than 10 results in the query may mean.
const auto &IFIntervals = Q.interferingVRegs(EvictInterferenceCutoff);
if (IFIntervals.empty() && InterferingIntervals.empty())
if (IFIntervals.size() >= EvictInterferenceCutoff)
return false;
InterferingIntervals.append(IFIntervals.begin(), IFIntervals.end());
for (const LiveInterval *Intf : reverse(IFIntervals)) {
assert(Intf->reg().isVirtual() &&
"Only expecting virtual register interference from query");
// This is the same set of legality checks as in the default case: don't
// try to evict fixed regs or 'done' ones. Also don't break cascades,
// except in the urgent case, with the same nuances used in the default
// heuristic.
// We could try sharing this between the advisors, but it may end up
// more complex than it is right now.
if (FixedRegisters.count(Intf->reg()))
return false;
if (RA.getExtraInfo().getStage(*Intf) == RS_Done)
return false;
bool Urgent =
!VirtReg.isSpillable() &&
(Intf->isSpillable() ||
RegClassInfo.getNumAllocatableRegs(MRI->getRegClass(VirtReg.reg())) <
// Only evict older cascades or live ranges without a cascade.
unsigned IntfCascade = RA.getExtraInfo().getCascade(Intf->reg());
if (Cascade <= IntfCascade) {
if (!Urgent)
return false;
LocalIntfs += (IsLocal && LIS->intervalIsInOneMBB(*Intf) &&
(!EnableLocalReassign || !canReassign(*Intf, PhysReg)));
// OK, so if we made it this far, this LR is an eviction candidate, load its
// features.
extractFeatures(InterferingIntervals, Largest, Pos, IsHint, LocalIntfs,
NrUrgent, LRPosInfo);
return true;
MCRegister MLEvictAdvisor::tryFindEvictionCandidate(
const LiveInterval &VirtReg, const AllocationOrder &Order,
uint8_t CostPerUseLimit, const SmallVirtRegSet &FixedRegisters) const {
auto MaybeOrderLimit = getOrderLimit(VirtReg, Order, CostPerUseLimit);
if (!MaybeOrderLimit)
return MCRegister::NoRegister;
unsigned OrderLimit = *MaybeOrderLimit;
// The heuristic sets initial costs such as, if CostPerUseLimit is
// max<uint8_t>, then any of the costs of the legally-evictable intervals
// would be lower. When that happens, one of those will be selected.
// Therefore, we allow the candidate be selected, unless the candidate is
// unspillable, in which case it would be incorrect to not find a register
// for it.
const bool MustFindEviction =
(!VirtReg.isSpillable() && CostPerUseLimit == static_cast<uint8_t>(~0u));
// Number of available candidates - if 0, no need to continue.
size_t Available = 0;
// Make sure we don't have leftover partial state from an attempt where we
// had no available candidates and bailed out early.
// Track the index->register mapping because AllocationOrder doesn't do that
// and we'd have to scan it.
// Also track their mask, to write asserts/debug.
CandidateRegList Regs;
Regs.fill({0, false});
// Track the largest value of features seen during this eviction session. We
// only normalize (some of) the float features, but it's just simpler to
// dimension 'Largest' to all the features, especially since we have the
// 'DoNotNormalize' list.
FeaturesListNormalizer Largest(FeatureIDs::FeatureCount, 0.0);
// Same overal idea as in the default eviction policy - we visit the values
// of AllocationOrder one at a time. If it's not legally available, we mask
// off the corresponding feature column (==do nothing because we already
// reset all the features to 0) Use Pos to capture the column we load
// features at - in AllocationOrder order.
size_t Pos = 0;
SmallVector<LRStartEndInfo, NumberOfInterferences> LRPosInfo;
for (auto I = Order.begin(), E = Order.getOrderLimitEnd(OrderLimit); I != E;
++I, ++Pos) {
MCRegister PhysReg = *I;
if (!canAllocatePhysReg(CostPerUseLimit, PhysReg)) {
if (loadInterferenceFeatures(VirtReg, PhysReg, I.isHint(), FixedRegisters,
Largest, Pos, LRPosInfo)) {
Regs[Pos] = std::make_pair(PhysReg, true);
if (Available == 0) {
// Nothing to decide, nothing to learn.
return MCRegister::NoRegister;
const size_t ValidPosLimit = Pos;
// If we must find eviction, the candidate should be masked out of the
// decision making process.
Regs[CandidateVirtRegPos].second = !MustFindEviction;
if (!MustFindEviction)
extractFeatures(SmallVector<const LiveInterval *, 1>(1, &VirtReg), Largest,
CandidateVirtRegPos, /*IsHint*/ 0,
/*LocalIntfsCount*/ 0,
/*NrUrgent*/ 0.0, LRPosInfo);
assert(InitialQSize > 0.0 && "We couldn't have gotten here if we had "
"nothing to allocate initially.");
if (EnableDevelopmentFeatures) {
LRPosInfo, Runner,
[this](SlotIndex InputIndex) -> int {
auto *CurrentMachineInstruction =
if (!CurrentMachineInstruction) {
return -1;
return CurrentMachineInstruction->getOpcode();
[this](SlotIndex InputIndex) -> float {
auto *CurrentMachineInstruction =
return MBFI.getBlockFreqRelativeToEntryBlock(
[this](SlotIndex InputIndex) -> MachineBasicBlock * {
auto *CurrentMachineInstruction =
return CurrentMachineInstruction->getParent();
FeatureIDs::instructions, FeatureIDs::instructions_mapping,
FeatureIDs::mbb_frequencies, FeatureIDs::mbb_mapping,
#endif // #ifdef LLVM_HAVE_TFLITE
// Normalize the features.
for (auto &V : Largest)
V = V ? V : 1.0;
for (size_t FeatureIndex = 0; FeatureIndex < FeatureIDs::FeatureCount;
++FeatureIndex) {
if (DoNotNormalize.test(FeatureIndex))
for (size_t Pos = 0; Pos < NumberOfInterferences; ++Pos) {
Runner->getTensor<float>(FeatureIndex)[Pos] /= Largest[FeatureIndex];
*Runner->getTensor<float>(FeatureIDs::progress) =
static_cast<float>(RA.getQueueSize()) / InitialQSize;
// Get a decision.
size_t CandidatePos = tryFindEvictionCandidatePosition(
VirtReg, Order, OrderLimit, CostPerUseLimit, FixedRegisters);
// The contract with the ML side is that CandidatePos is mask == 1 (i.e.
// Regs[CandidatePos].second)
if (CandidatePos == CandidateVirtRegPos) {
return MCRegister::NoRegister;
assert(CandidatePos < ValidPosLimit);
return Regs[CandidatePos].first;
const LIFeatureComponents &
MLEvictAdvisor::getLIFeatureComponents(const LiveInterval &LI) const {
RegID ID = LI.reg().id();
LIFeatureComponents Empty;
auto I = CachedFeatures.insert(std::make_pair(ID, Empty));
LIFeatureComponents &Ret = I.first->getSecond();
if (!I.second)
return Ret;
SmallPtrSet<MachineInstr *, 8> Visited;
const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
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++);
if (!Visited.insert(MI).second)
if (MI->isIdentityCopy() || MI->isImplicitDef())
bool Reads, Writes;
std::tie(Reads, Writes) = MI->readsWritesVirtualRegister(LI.reg());
float Freq = MBFI.getBlockFreqRelativeToEntryBlock(MI->getParent());
Ret.HottestBlockFreq = std::max(Freq, Ret.HottestBlockFreq);
Ret.R += (Reads && !Writes) * Freq;
Ret.W += (!Reads && Writes) * Freq;
Ret.RW += (Reads && Writes) * Freq;
auto *MBB = MI->getParent();
auto *Loop = Loops.getLoopFor(MBB);
bool IsExiting = Loop ? Loop->isLoopExiting(MBB) : false;
if (Writes && IsExiting && LIS->isLiveOutOfMBB(LI, MBB))
Ret.IndVarUpdates += Freq;
if (MI->isCopy() && VirtRegAuxInfo::copyHint(MI, LI.reg(), TRI, *MRI))
Ret.HintWeights += Freq;
Ret.IsRemat = VirtRegAuxInfo::isRematerializable(
LI, *LIS, *VRM, *MF.getSubtarget().getInstrInfo());
return Ret;
// Overall, this currently mimics what we do for weight calculation, but instead
// of accummulating the various features, we keep them separate.
void MLEvictAdvisor::extractFeatures(
const SmallVectorImpl<const LiveInterval *> &Intervals,
llvm::SmallVectorImpl<float> &Largest, size_t Pos, int64_t IsHint,
int64_t LocalIntfsCount, float NrUrgent,
SmallVectorImpl<LRStartEndInfo> &LRPosInfo) const {
int64_t NrDefsAndUses = 0;
int64_t NrBrokenHints = 0;
double R = 0.0;
double W = 0.0;
double RW = 0.0;
double IndVarUpdates = 0.0;
double HintWeights = 0.0;
float StartBBFreq = 0.0;
float EndBBFreq = 0.0;
float HottestBlockFreq = 0.0;
int32_t NrRematerializable = 0;
float TotalWeight = 0.0;
SlotIndex EndSI = LIS->getSlotIndexes()->getZeroIndex();
SlotIndex StartSI = LIS->getSlotIndexes()->getLastIndex();
int64_t MaxStage = 0;
int64_t MinStage =
Intervals.empty() ? 0 : std::numeric_limits<int64_t>::max();
for (const auto *L : Intervals) {
const LiveInterval &LI = *L;
MaxStage = std::max<int64_t>(
MaxStage, static_cast<int64_t>(RA.getExtraInfo().getStage(LI)));
MinStage = std::min<int64_t>(
MinStage, static_cast<int64_t>(RA.getExtraInfo().getStage(LI)));
TotalWeight = std::max(TotalWeight, LI.weight());
if (LI.beginIndex() < StartSI)
StartSI = LI.beginIndex();
if (LI.endIndex() > EndSI)
EndSI = LI.endIndex();
const LIFeatureComponents &LIFC = getLIFeatureComponents(LI);
NrBrokenHints += VRM->hasPreferredPhys(LI.reg());
NrDefsAndUses += LIFC.NrDefsAndUses;
HottestBlockFreq = std::max(HottestBlockFreq, LIFC.HottestBlockFreq);
R += LIFC.R;
W += LIFC.W;
IndVarUpdates += LIFC.IndVarUpdates;
HintWeights += LIFC.HintWeights;
NrRematerializable += LIFC.IsRemat;
if (EnableDevelopmentFeatures) {
for (auto CurrentSegment : LI) {
LRStartEndInfo{CurrentSegment.start, CurrentSegment.end, Pos});
size_t Size = 0;
if (!Intervals.empty()) {
StartBBFreq =
if (EndSI >= LIS->getSlotIndexes()->getLastIndex())
EndSI = LIS->getSlotIndexes()->getLastIndex().getPrevIndex();
EndBBFreq =
Size = StartSI.distance(EndSI);
// Set the features at the column 'Pos'.
#define SET(ID, TYPE, VAL) \
do { \
Runner->getTensor<TYPE>(FeatureIDs::ID)[Pos] = static_cast<TYPE>(VAL); \
if (!DoNotNormalize.test(FeatureIDs::ID)) \
Largest[FeatureIDs::ID] = \
std::max(Largest[FeatureIDs::ID], static_cast<float>(VAL)); \
} while (false)
SET(mask, int64_t, 1);
SET(is_free, int64_t, Intervals.empty());
SET(nr_urgent, float, NrUrgent);
SET(nr_broken_hints, float, NrBrokenHints);
SET(is_hint, int64_t, IsHint);
SET(is_local, int64_t, LocalIntfsCount);
SET(nr_rematerializable, float, NrRematerializable);
SET(nr_defs_and_uses, float, NrDefsAndUses);
SET(weighed_reads_by_max, float, R);
SET(weighed_writes_by_max, float, W);
SET(weighed_read_writes_by_max, float, RW);
SET(weighed_indvars_by_max, float, IndVarUpdates);
SET(hint_weights_by_max, float, HintWeights);
SET(start_bb_freq_by_max, float, StartBBFreq);
SET(end_bb_freq_by_max, float, EndBBFreq);
SET(hottest_bb_freq_by_max, float, HottestBlockFreq);
SET(liverange_size, float, Size);
SET(use_def_density, float, TotalWeight);
SET(max_stage, int64_t, MaxStage);
SET(min_stage, int64_t, MinStage);
#undef SET
void extractInstructionFeatures(
SmallVectorImpl<LRStartEndInfo> &LRPosInfo, MLModelRunner *RegallocRunner,
function_ref<int(SlotIndex)> GetOpcode,
function_ref<float(SlotIndex)> GetMBBFreq,
function_ref<MachineBasicBlock *(SlotIndex)> GetMBBReference,
const int InstructionsIndex, const int InstructionsMappingIndex,
const int MBBFreqIndex, const int MBBMappingIndex,
const SlotIndex LastIndex) {
// This function extracts instruction based features relevant to the eviction
// problem currently being solved. This function ends up extracting two
// tensors.
// 1 - A vector of size max instruction count. It contains the opcodes of the
// instructions spanned by all the intervals in the current instance of the
// eviction problem.
// 2 - A binary mapping matrix of size (LR count * max
// instruction count) which maps where the LRs are live to the actual opcodes
// for which they are live.
// 3 - A vector of size max supported MBB count storing MBB frequencies,
// encompassing all of the MBBs covered by the eviction problem.
// 4 - A vector of size max instruction count of indices to members of the MBB
// frequency vector, mapping each instruction to its associated MBB.
// Start off by sorting the segments based on the beginning slot index.
LRPosInfo.begin(), LRPosInfo.end(),
[](LRStartEndInfo A, LRStartEndInfo B) { return A.Begin < B.Begin; });
size_t InstructionIndex = 0;
size_t CurrentSegmentIndex = 0;
SlotIndex CurrentIndex = LRPosInfo[0].Begin;
std::map<MachineBasicBlock *, size_t> VisitedMBBs;
size_t CurrentMBBIndex = 0;
// This loop processes all the segments sequentially by starting at the
// beginning slot index of the first segment, iterating through all the slot
// indices before the end slot index of that segment (while checking for
// overlaps with segments that start at greater slot indices). After hitting
// that end index, the current segment being processed gets bumped until they
// are all processed or the max instruction count is hit, where everything is
// just truncated.
while (true) {
// If the index that we are currently at is within the current segment and
// we haven't hit the max instruction count, continue processing the current
// segment.
while (CurrentIndex <= LRPosInfo[CurrentSegmentIndex].End &&
InstructionIndex < ModelMaxSupportedInstructionCount) {
int CurrentOpcode = GetOpcode(CurrentIndex);
// If the current machine instruction is null, skip it
if (CurrentOpcode == -1) {
// If we're currently at the last index in the SlotIndex analysis,
// we can't go any further, so return from the function
if (CurrentIndex >= LastIndex) {
CurrentIndex = CurrentIndex.getNextIndex();
MachineBasicBlock *CurrentMBBReference = GetMBBReference(CurrentIndex);
if (VisitedMBBs.count(CurrentMBBReference) == 0) {
VisitedMBBs[CurrentMBBReference] = CurrentMBBIndex;
extractMBBFrequency(CurrentIndex, InstructionIndex, VisitedMBBs,
GetMBBFreq, CurrentMBBReference, RegallocRunner,
MBBFreqIndex, MBBMappingIndex);
// Current code assumes we're not going to get any disjointed segments
assert(LRPosInfo[CurrentSegmentIndex].Begin <= CurrentIndex);
RegallocRunner->getTensor<int64_t>(InstructionsIndex)[InstructionIndex] =
CurrentOpcode < OpcodeValueCutoff ? CurrentOpcode : 0;
// set value in the binary mapping matrix for the current instruction
auto CurrentSegmentPosition = LRPosInfo[CurrentSegmentIndex].Pos;
InstructionsMappingIndex)[CurrentSegmentPosition *
ModelMaxSupportedInstructionCount +
InstructionIndex] = 1;
// All of the segments are sorted based on the beginning slot index, but
// this doesn't mean that the beginning slot index of the next segment is
// after the end segment of the one being currently processed. This while
// loop checks for overlapping segments and modifies the portion of the
// column in the mapping matrix for the currently processed instruction
// for the LR it is checking. Also make sure that the beginning of the
// current segment we're checking for overlap in is less than the current
// index, otherwise we're done checking overlaps.
size_t OverlapCheckCurrentSegment = CurrentSegmentIndex + 1;
while (OverlapCheckCurrentSegment < LRPosInfo.size() &&
LRPosInfo[OverlapCheckCurrentSegment].Begin <= CurrentIndex) {
auto OverlapCurrentSegmentPosition =
if (LRPosInfo[OverlapCheckCurrentSegment].End >= CurrentIndex) {
InstructionsMappingIndex)[OverlapCurrentSegmentPosition *
ModelMaxSupportedInstructionCount +
InstructionIndex] = 1;
if (CurrentIndex >= LastIndex) {
CurrentIndex = CurrentIndex.getNextIndex();
// if we've just finished processing through the last segment or if we've
// hit the maximum number of instructions, break out of the loop.
if (CurrentSegmentIndex == LRPosInfo.size() - 1 ||
InstructionIndex >= ModelMaxSupportedInstructionCount) {
// If the segments are not overlapping, we need to move to the beginning
// index of the next segment to avoid having instructions not attached to
// any register.
if (LRPosInfo[CurrentSegmentIndex + 1].Begin >
LRPosInfo[CurrentSegmentIndex].End) {
CurrentIndex = LRPosInfo[CurrentSegmentIndex + 1].Begin;
void extractMBBFrequency(const SlotIndex CurrentIndex,
const size_t CurrentInstructionIndex,
std::map<MachineBasicBlock *, size_t> &VisitedMBBs,
function_ref<float(SlotIndex)> GetMBBFreq,
MachineBasicBlock *CurrentMBBReference,
MLModelRunner *RegallocRunner, const int MBBFreqIndex,
const int MBBMappingIndex) {
size_t CurrentMBBIndex = VisitedMBBs[CurrentMBBReference];
float CurrentMBBFreq = GetMBBFreq(CurrentIndex);
if (CurrentMBBIndex < ModelMaxSupportedMBBCount) {
RegallocRunner->getTensor<float>(MBBFreqIndex)[CurrentMBBIndex] =
MBBMappingIndex)[CurrentInstructionIndex] = CurrentMBBIndex;
// Development mode-specific implementations
RegAllocEvictionAdvisorAnalysis *llvm::createDevelopmentModeAdvisor() {
return new DevelopmentModeEvictionAdvisorAnalysis();
int64_t DevelopmentModeEvictAdvisor::tryFindEvictionCandidatePosition(
const LiveInterval &VirtReg, const AllocationOrder &Order,
unsigned OrderLimit, uint8_t CostPerUseLimit,
const SmallVirtRegSet &FixedRegisters) const {
int64_t Ret = 0;
if (isa<ModelUnderTrainingRunner>(getRunner())) {
Ret = MLEvictAdvisor::tryFindEvictionCandidatePosition(
VirtReg, Order, OrderLimit, CostPerUseLimit, FixedRegisters);
} else {
MCRegister PhysReg = getDefaultAdvisor().tryFindEvictionCandidate(
VirtReg, Order, CostPerUseLimit, FixedRegisters);
// Find the index of the selected PhysReg. We need it for logging,
// otherwise this is wasted cycles (but so would starting development mode
// without a model nor logging)
if (!PhysReg)
Ret = CandidateVirtRegPos;
for (auto I = Order.begin(), E = Order.getOrderLimitEnd(OrderLimit);
I != E; ++I, ++Ret)
if (*I == PhysReg)
if (TrainingLog.empty())
return Ret;
// TODO(mtrofin): when we support optional rewards, this can go away. In the
// meantime, we log the "pretend" reward (0) for the previous observation
// before starting a new one.
if (Log->hasObservationInProgress())
size_t CurrentFeature = 0;
size_t FeatureCount = EnableDevelopmentFeatures
? FeatureIDs::FeaturesWithDevelopmentCount
: FeatureIDs::FeatureCount;
for (; CurrentFeature < FeatureCount; ++CurrentFeature) {
reinterpret_cast<const char *>(
if (auto *MUTR = dyn_cast<ModelUnderTrainingRunner>(&getRunner()))
for (size_t I = 0; I < MUTR->extraOutputsForLoggingSpecs().size();
++I, ++CurrentFeature)
reinterpret_cast<const char *>(MUTR->getUntypedExtraOutputValue(I)));
// The output is right after the features and the extra outputs
Log->logTensorValue(CurrentFeature, reinterpret_cast<const char *>(&Ret));
return Ret;
bool RegAllocScoring::runOnMachineFunction(MachineFunction &MF) {
std::optional<float> CachedReward;
auto GetReward = [&]() {
if (!CachedReward)
CachedReward = static_cast<float>(
calculateRegAllocScore(MF, getAnalysis<MachineBlockFrequencyInfo>())
return *CachedReward;
return false;
#endif // #ifdef LLVM_HAVE_TFLITE
RegAllocEvictionAdvisorAnalysis *llvm::createReleaseModeAdvisor() {
return llvm::isEmbeddedModelEvaluatorValid<CompiledModelType>() ||
? new ReleaseModeEvictionAdvisorAnalysis()
: nullptr;
// In all cases except development mode, we don't need scoring.
#if !defined(LLVM_HAVE_TFLITE)
bool RegAllocScoring::runOnMachineFunction(MachineFunction &) { return false; }