//===- InlineSizeEstimatorAnalysis.cpp - IR to native size from ML model --===//
//
// 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 implements feature and label extraction for offline supervised learning
// of a IR to native size model.
//
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/InlineSizeEstimatorAnalysis.h"
#ifdef LLVM_HAVE_TFLITE
#include "llvm/Analysis/Utils/TFUtils.h"
#endif
#include "llvm/IR/Function.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
AnalysisKey InlineSizeEstimatorAnalysis::Key;
#ifdef LLVM_HAVE_TFLITE
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Instructions.h"
#include "llvm/MC/MCAsmLayout.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include <algorithm>
#include <deque>
#include <optional>
cl::opt<std::string> TFIR2NativeModelPath(
"ml-inliner-ir2native-model", cl::Hidden,
cl::desc("Path to saved model evaluating native size from IR."));
#define DEBUG_TYPE "inline-size-estimator"
namespace {
unsigned getMaxInstructionID() {
#define LAST_OTHER_INST(NR) return NR;
#include "llvm/IR/Instruction.def"
}
class IRToNativeSizeLearning {
public:
enum class NamedFeatureIndex : size_t {
InitialSize,
Blocks,
Calls,
IsLocal,
IsLinkOnceODR,
IsLinkOnce,
Loops,
MaxLoopDepth,
MaxDomTreeLevel,
NumNamedFeatures
};
static const size_t NumNamedFeatures =
static_cast<size_t>(NamedFeatureIndex::NumNamedFeatures);
struct FunctionFeatures {
static const size_t FeatureCount;
std::array<int32_t, NumNamedFeatures> NamedFeatures = {0};
std::vector<int32_t> InstructionHistogram;
std::vector<int32_t> InstructionPairHistogram;
void fillTensor(int32_t *Ptr) const;
int32_t &operator[](NamedFeatureIndex Pos) {
return NamedFeatures[static_cast<size_t>(Pos)];
}
};
IRToNativeSizeLearning() = default;
static FunctionFeatures getFunctionFeatures(Function &F,
FunctionAnalysisManager &FAM);
};
// This is a point in time - we determined including these pairs of
// consecutive instructions (in the IR layout available at inline time) as
// features improves the model performance. We want to move away from manual
// feature selection.
// The array is given in opcode pairs rather than labels because 1) labels
// weren't readily available, and 2) the successions were hand - extracted.
//
// This array must be sorted.
static const std::array<std::pair<size_t, size_t>, 137>
ImportantInstructionSuccessions{
{{1, 1}, {1, 4}, {1, 5}, {1, 7}, {1, 8}, {1, 9}, {1, 11},
{1, 12}, {1, 13}, {1, 14}, {1, 18}, {1, 20}, {1, 22}, {1, 24},
{1, 25}, {1, 26}, {1, 27}, {1, 28}, {1, 29}, {1, 30}, {1, 31},
{1, 32}, {1, 33}, {1, 34}, {1, 39}, {1, 40}, {1, 42}, {1, 45},
{2, 1}, {2, 2}, {2, 13}, {2, 28}, {2, 29}, {2, 32}, {2, 33},
{2, 34}, {2, 38}, {2, 48}, {2, 49}, {2, 53}, {2, 55}, {2, 56},
{13, 2}, {13, 13}, {13, 26}, {13, 33}, {13, 34}, {13, 56}, {15, 27},
{28, 2}, {28, 48}, {28, 53}, {29, 2}, {29, 33}, {29, 56}, {31, 31},
{31, 33}, {31, 34}, {31, 49}, {32, 1}, {32, 2}, {32, 13}, {32, 15},
{32, 28}, {32, 29}, {32, 32}, {32, 33}, {32, 34}, {32, 39}, {32, 40},
{32, 48}, {32, 49}, {32, 53}, {32, 56}, {33, 1}, {33, 2}, {33, 32},
{33, 33}, {33, 34}, {33, 49}, {33, 53}, {33, 56}, {34, 1}, {34, 2},
{34, 32}, {34, 33}, {34, 34}, {34, 49}, {34, 53}, {34, 56}, {38, 34},
{39, 57}, {40, 34}, {47, 15}, {47, 49}, {48, 2}, {48, 34}, {48, 56},
{49, 1}, {49, 2}, {49, 28}, {49, 32}, {49, 33}, {49, 34}, {49, 39},
{49, 49}, {49, 56}, {53, 1}, {53, 2}, {53, 28}, {53, 34}, {53, 53},
{53, 57}, {55, 1}, {55, 28}, {55, 34}, {55, 53}, {55, 55}, {55, 56},
{56, 1}, {56, 2}, {56, 7}, {56, 13}, {56, 32}, {56, 33}, {56, 34},
{56, 49}, {56, 53}, {56, 56}, {56, 64}, {57, 34}, {57, 56}, {57, 57},
{64, 1}, {64, 64}, {65, 1}, {65, 65}}};
// We have: 9 calculated features (the features here); 1 feature for each
// instruction opcode; and 1 feature for each manually-identified sequence.
// For the latter 2, we build a histogram: we count the number of
// occurrences of each instruction opcode or succession of instructions,
// respectively.
// Note that instruction opcodes start from 1. For convenience, we also have an
// always 0 feature for the '0' opcode, hence the extra 1.
const size_t IRToNativeSizeLearning::FunctionFeatures::FeatureCount =
ImportantInstructionSuccessions.size() + getMaxInstructionID() + 1 +
IRToNativeSizeLearning::NumNamedFeatures;
size_t getSize(Function &F, TargetTransformInfo &TTI) {
size_t Ret = 0;
for (const auto &BB : F)
for (const auto &I : BB)
Ret += *(TTI.getInstructionCost(
&I, TargetTransformInfo::TargetCostKind::TCK_CodeSize).getValue());
return Ret;
}
size_t getSize(Function &F, FunctionAnalysisManager &FAM) {
auto &TTI = FAM.getResult<TargetIRAnalysis>(F);
return getSize(F, TTI);
}
unsigned getMaxDominatorTreeDepth(const Function &F,
const DominatorTree &Tree) {
unsigned Ret = 0;
for (const auto &BB : F)
if (const auto *TN = Tree.getNode(&BB))
Ret = std::max(Ret, TN->getLevel());
return Ret;
}
} // namespace
IRToNativeSizeLearning::FunctionFeatures
IRToNativeSizeLearning::getFunctionFeatures(Function &F,
FunctionAnalysisManager &FAM) {
assert(llvm::is_sorted(ImportantInstructionSuccessions) &&
"expected function features are sorted");
auto &DomTree = FAM.getResult<DominatorTreeAnalysis>(F);
FunctionFeatures FF;
size_t InstrCount = getMaxInstructionID() + 1;
FF.InstructionHistogram.resize(InstrCount);
FF.InstructionPairHistogram.resize(ImportantInstructionSuccessions.size());
int StartID = 0;
int LastID = StartID;
auto getPairIndex = [](size_t a, size_t b) {
auto I = llvm::find(ImportantInstructionSuccessions, std::make_pair(a, b));
if (I == ImportantInstructionSuccessions.end())
return -1;
return static_cast<int>(
std::distance(ImportantInstructionSuccessions.begin(), I));
};
// We don't want debug calls, because they'd just add noise.
for (const auto &BB : F) {
for (const auto &I : BB.instructionsWithoutDebug()) {
auto ID = I.getOpcode();
++FF.InstructionHistogram[ID];
int PairIndex = getPairIndex(LastID, ID);
if (PairIndex >= 0)
++FF.InstructionPairHistogram[PairIndex];
LastID = ID;
if (isa<CallBase>(I))
++FF[NamedFeatureIndex::Calls];
}
}
FF[NamedFeatureIndex::InitialSize] = getSize(F, FAM);
FF[NamedFeatureIndex::IsLocal] = F.hasLocalLinkage();
FF[NamedFeatureIndex::IsLinkOnceODR] = F.hasLinkOnceODRLinkage();
FF[NamedFeatureIndex::IsLinkOnce] = F.hasLinkOnceLinkage();
FF[NamedFeatureIndex::Blocks] = F.size();
auto &LI = FAM.getResult<LoopAnalysis>(F);
FF[NamedFeatureIndex::Loops] = std::distance(LI.begin(), LI.end());
for (auto &L : LI)
FF[NamedFeatureIndex::MaxLoopDepth] =
std::max(FF[NamedFeatureIndex::MaxLoopDepth],
static_cast<int32_t>(L->getLoopDepth()));
FF[NamedFeatureIndex::MaxDomTreeLevel] = getMaxDominatorTreeDepth(F, DomTree);
return FF;
}
void IRToNativeSizeLearning::FunctionFeatures::fillTensor(int32_t *Ptr) const {
std::copy(NamedFeatures.begin(), NamedFeatures.end(), Ptr);
Ptr += NamedFeatures.size();
std::copy(InstructionHistogram.begin(), InstructionHistogram.end(), Ptr);
Ptr += InstructionHistogram.size();
std::copy(InstructionPairHistogram.begin(), InstructionPairHistogram.end(),
Ptr);
}
bool InlineSizeEstimatorAnalysis::isEvaluatorRequested() {
return !TFIR2NativeModelPath.empty();
}
InlineSizeEstimatorAnalysis::InlineSizeEstimatorAnalysis() {
if (!isEvaluatorRequested()) {
return;
}
std::vector<TensorSpec> InputSpecs{TensorSpec::createSpec<int32_t>(
"serving_default_input_1",
{1, static_cast<int64_t>(
IRToNativeSizeLearning::FunctionFeatures::FeatureCount)})};
std::vector<TensorSpec> OutputSpecs{
TensorSpec::createSpec<float>("StatefulPartitionedCall", {1})};
Evaluator = std::make_unique<TFModelEvaluator>(
TFIR2NativeModelPath.getValue().c_str(), InputSpecs, OutputSpecs);
if (!Evaluator || !Evaluator->isValid()) {
Evaluator.reset();
return;
}
}
InlineSizeEstimatorAnalysis::Result
InlineSizeEstimatorAnalysis::run(const Function &F,
FunctionAnalysisManager &FAM) {
if (!Evaluator)
return std::nullopt;
auto Features = IRToNativeSizeLearning::getFunctionFeatures(
const_cast<Function &>(F), FAM);
int32_t *V = Evaluator->getInput<int32_t>(0);
Features.fillTensor(V);
auto ER = Evaluator->evaluate();
if (!ER)
return std::nullopt;
float Ret = *ER->getTensorValue<float>(0);
if (Ret < 0.0)
Ret = 0.0;
return static_cast<size_t>(Ret);
}
InlineSizeEstimatorAnalysis::~InlineSizeEstimatorAnalysis() {}
InlineSizeEstimatorAnalysis::InlineSizeEstimatorAnalysis(
InlineSizeEstimatorAnalysis &&Other)
: Evaluator(std::move(Other.Evaluator)) {}
#else
namespace llvm {
class TFModelEvaluator {};
} // namespace llvm
InlineSizeEstimatorAnalysis::InlineSizeEstimatorAnalysis() = default;
InlineSizeEstimatorAnalysis ::InlineSizeEstimatorAnalysis(
InlineSizeEstimatorAnalysis &&) {}
InlineSizeEstimatorAnalysis::~InlineSizeEstimatorAnalysis() = default;
InlineSizeEstimatorAnalysis::Result
InlineSizeEstimatorAnalysis::run(const Function &F,
FunctionAnalysisManager &FAM) {
return std::nullopt;
}
bool InlineSizeEstimatorAnalysis::isEvaluatorRequested() { return false; }
#endif
PreservedAnalyses
InlineSizeEstimatorAnalysisPrinterPass::run(Function &F,
FunctionAnalysisManager &AM) {
OS << "[InlineSizeEstimatorAnalysis] size estimate for " << F.getName()
<< ": " << AM.getResult<InlineSizeEstimatorAnalysis>(F) << "\n";
return PreservedAnalyses::all();
}
