llvm-project/llvm/lib/Analysis/BlockFrequencyInfo.cpp

338 lines
12 KiB
C++
Raw Normal View History

//===- BlockFrequencyInfo.cpp - Block Frequency Analysis ------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// Loops should be simplified before this analysis.
//
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/BlockFrequencyInfo.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/iterator.h"
#include "llvm/Analysis/BlockFrequencyInfoImpl.h"
#include "llvm/Analysis/BranchProbabilityInfo.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Pass.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/GraphWriter.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <string>
using namespace llvm;
#define DEBUG_TYPE "block-freq"
static cl::opt<GVDAGType> ViewBlockFreqPropagationDAG(
"view-block-freq-propagation-dags", cl::Hidden,
cl::desc("Pop up a window to show a dag displaying how block "
"frequencies propagation through the CFG."),
cl::values(clEnumValN(GVDT_None, "none", "do not display graphs."),
clEnumValN(GVDT_Fraction, "fraction",
"display a graph using the "
"fractional block frequency representation."),
clEnumValN(GVDT_Integer, "integer",
"display a graph using the raw "
"integer fractional block frequency representation."),
clEnumValN(GVDT_Count, "count", "display a graph using the real "
"profile count if available.")));
cl::opt<std::string>
ViewBlockFreqFuncName("view-bfi-func-name", cl::Hidden,
cl::desc("The option to specify "
"the name of the function "
"whose CFG will be displayed."));
cl::opt<unsigned>
ViewHotFreqPercent("view-hot-freq-percent", cl::init(10), cl::Hidden,
cl::desc("An integer in percent used to specify "
"the hot blocks/edges to be displayed "
"in red: a block or edge whose frequency "
"is no less than the max frequency of the "
"function multiplied by this percent."));
// Command line option to turn on CFG dot or text dump after profile annotation.
cl::opt<PGOViewCountsType> PGOViewCounts(
"pgo-view-counts", cl::Hidden,
cl::desc("A boolean option to show CFG dag or text with "
"block profile counts and branch probabilities "
"right after PGO profile annotation step. The "
"profile counts are computed using branch "
"probabilities from the runtime profile data and "
"block frequency propagation algorithm. To view "
"the raw counts from the profile, use option "
"-pgo-view-raw-counts instead. To limit graph "
"display to only one function, use filtering option "
"-view-bfi-func-name."),
cl::values(clEnumValN(PGOVCT_None, "none", "do not show."),
clEnumValN(PGOVCT_Graph, "graph", "show a graph."),
clEnumValN(PGOVCT_Text, "text", "show in text.")));
static cl::opt<bool> PrintBlockFreq(
"print-bfi", cl::init(false), cl::Hidden,
cl::desc("Print the block frequency info."));
cl::opt<std::string> PrintBlockFreqFuncName(
"print-bfi-func-name", cl::Hidden,
cl::desc("The option to specify the name of the function "
"whose block frequency info is printed."));
namespace llvm {
static GVDAGType getGVDT() {
if (PGOViewCounts == PGOVCT_Graph)
return GVDT_Count;
return ViewBlockFreqPropagationDAG;
}
template <>
struct GraphTraits<BlockFrequencyInfo *> {
using NodeRef = const BasicBlock *;
using ChildIteratorType = succ_const_iterator;
using nodes_iterator = pointer_iterator<Function::const_iterator>;
static NodeRef getEntryNode(const BlockFrequencyInfo *G) {
Analysis: Remove implicit ilist iterator conversions Remove implicit ilist iterator conversions from LLVMAnalysis. I came across something really scary in `llvm::isKnownNotFullPoison()` which relied on `Instruction::getNextNode()` being completely broken (not surprising, but scary nevertheless). This function is documented (and coded to) return `nullptr` when it gets to the sentinel, but with an `ilist_half_node` as a sentinel, the sentinel check looks into some other memory and we don't recognize we've hit the end. Rooting out these scary cases is the reason I'm removing the implicit conversions before doing anything else with `ilist`; I'm not at all surprised that clients rely on badness. I found another scary case -- this time, not relying on badness, just bad (but I guess getting lucky so far) -- in `ObjectSizeOffsetEvaluator::compute_()`. Here, we save out the insertion point, do some things, and then restore it. Previously, we let the iterator auto-convert to `Instruction*`, and then set it back using the `Instruction*` version: Instruction *PrevInsertPoint = Builder.GetInsertPoint(); /* Logic that may change insert point */ if (PrevInsertPoint) Builder.SetInsertPoint(PrevInsertPoint); The check for `PrevInsertPoint` doesn't protect correctly against bad accesses. If the insertion point has been set to the end of a basic block (i.e., `SetInsertPoint(SomeBB)`), then `GetInsertPoint()` returns an iterator pointing at the list sentinel. The version of `SetInsertPoint()` that's getting called will then call `PrevInsertPoint->getParent()`, which explodes horribly. The only reason this hasn't blown up is that it's fairly unlikely the builder is adding to the end of the block; usually, we're adding instructions somewhere before the terminator. llvm-svn: 249925
2015-10-10 08:53:03 +08:00
return &G->getFunction()->front();
}
static ChildIteratorType child_begin(const NodeRef N) {
return succ_begin(N);
}
static ChildIteratorType child_end(const NodeRef N) { return succ_end(N); }
static nodes_iterator nodes_begin(const BlockFrequencyInfo *G) {
return nodes_iterator(G->getFunction()->begin());
}
static nodes_iterator nodes_end(const BlockFrequencyInfo *G) {
return nodes_iterator(G->getFunction()->end());
}
};
using BFIDOTGTraitsBase =
BFIDOTGraphTraitsBase<BlockFrequencyInfo, BranchProbabilityInfo>;
template <>
struct DOTGraphTraits<BlockFrequencyInfo *> : public BFIDOTGTraitsBase {
explicit DOTGraphTraits(bool isSimple = false)
: BFIDOTGTraitsBase(isSimple) {}
std::string getNodeLabel(const BasicBlock *Node,
const BlockFrequencyInfo *Graph) {
return BFIDOTGTraitsBase::getNodeLabel(Node, Graph, getGVDT());
}
std::string getNodeAttributes(const BasicBlock *Node,
const BlockFrequencyInfo *Graph) {
return BFIDOTGTraitsBase::getNodeAttributes(Node, Graph,
ViewHotFreqPercent);
}
std::string getEdgeAttributes(const BasicBlock *Node, EdgeIter EI,
const BlockFrequencyInfo *BFI) {
return BFIDOTGTraitsBase::getEdgeAttributes(Node, EI, BFI, BFI->getBPI(),
ViewHotFreqPercent);
}
};
} // end namespace llvm
BlockFrequencyInfo::BlockFrequencyInfo() = default;
BlockFrequencyInfo::BlockFrequencyInfo(const Function &F,
const BranchProbabilityInfo &BPI,
const LoopInfo &LI) {
calculate(F, BPI, LI);
}
BlockFrequencyInfo::BlockFrequencyInfo(BlockFrequencyInfo &&Arg)
: BFI(std::move(Arg.BFI)) {}
BlockFrequencyInfo &BlockFrequencyInfo::operator=(BlockFrequencyInfo &&RHS) {
releaseMemory();
BFI = std::move(RHS.BFI);
return *this;
}
// Explicitly define the default constructor otherwise it would be implicitly
// defined at the first ODR-use which is the BFI member in the
// LazyBlockFrequencyInfo header. The dtor needs the BlockFrequencyInfoImpl
// template instantiated which is not available in the header.
BlockFrequencyInfo::~BlockFrequencyInfo() = default;
bool BlockFrequencyInfo::invalidate(Function &F, const PreservedAnalyses &PA,
FunctionAnalysisManager::Invalidator &) {
// Check whether the analysis, all analyses on functions, or the function's
// CFG have been preserved.
auto PAC = PA.getChecker<BlockFrequencyAnalysis>();
return !(PAC.preserved() || PAC.preservedSet<AllAnalysesOn<Function>>() ||
PAC.preservedSet<CFGAnalyses>());
}
void BlockFrequencyInfo::calculate(const Function &F,
const BranchProbabilityInfo &BPI,
const LoopInfo &LI) {
if (!BFI)
BFI.reset(new ImplType);
BFI->calculate(F, BPI, LI);
if (ViewBlockFreqPropagationDAG != GVDT_None &&
(ViewBlockFreqFuncName.empty() ||
F.getName().equals(ViewBlockFreqFuncName))) {
view();
}
if (PrintBlockFreq &&
(PrintBlockFreqFuncName.empty() ||
F.getName().equals(PrintBlockFreqFuncName))) {
print(dbgs());
}
}
BlockFrequency BlockFrequencyInfo::getBlockFreq(const BasicBlock *BB) const {
return BFI ? BFI->getBlockFreq(BB) : 0;
}
Optional<uint64_t>
BlockFrequencyInfo::getBlockProfileCount(const BasicBlock *BB) const {
if (!BFI)
return None;
return BFI->getBlockProfileCount(*getFunction(), BB);
}
Optional<uint64_t>
BlockFrequencyInfo::getProfileCountFromFreq(uint64_t Freq) const {
if (!BFI)
return None;
return BFI->getProfileCountFromFreq(*getFunction(), Freq);
}
void BlockFrequencyInfo::setBlockFreq(const BasicBlock *BB, uint64_t Freq) {
assert(BFI && "Expected analysis to be available");
BFI->setBlockFreq(BB, Freq);
}
void BlockFrequencyInfo::setBlockFreqAndScale(
const BasicBlock *ReferenceBB, uint64_t Freq,
SmallPtrSetImpl<BasicBlock *> &BlocksToScale) {
assert(BFI && "Expected analysis to be available");
// Use 128 bits APInt to avoid overflow.
APInt NewFreq(128, Freq);
APInt OldFreq(128, BFI->getBlockFreq(ReferenceBB).getFrequency());
APInt BBFreq(128, 0);
for (auto *BB : BlocksToScale) {
BBFreq = BFI->getBlockFreq(BB).getFrequency();
// Multiply first by NewFreq and then divide by OldFreq
// to minimize loss of precision.
BBFreq *= NewFreq;
// udiv is an expensive operation in the general case. If this ends up being
// a hot spot, one of the options proposed in
// https://reviews.llvm.org/D28535#650071 could be used to avoid this.
BBFreq = BBFreq.udiv(OldFreq);
BFI->setBlockFreq(BB, BBFreq.getLimitedValue());
}
BFI->setBlockFreq(ReferenceBB, Freq);
}
/// Pop up a ghostview window with the current block frequency propagation
/// rendered using dot.
void BlockFrequencyInfo::view() const {
ViewGraph(const_cast<BlockFrequencyInfo *>(this), "BlockFrequencyDAGs");
}
const Function *BlockFrequencyInfo::getFunction() const {
return BFI ? BFI->getFunction() : nullptr;
}
const BranchProbabilityInfo *BlockFrequencyInfo::getBPI() const {
return BFI ? &BFI->getBPI() : nullptr;
}
raw_ostream &BlockFrequencyInfo::
printBlockFreq(raw_ostream &OS, const BlockFrequency Freq) const {
return BFI ? BFI->printBlockFreq(OS, Freq) : OS;
}
raw_ostream &
BlockFrequencyInfo::printBlockFreq(raw_ostream &OS,
const BasicBlock *BB) const {
return BFI ? BFI->printBlockFreq(OS, BB) : OS;
}
uint64_t BlockFrequencyInfo::getEntryFreq() const {
return BFI ? BFI->getEntryFreq() : 0;
}
void BlockFrequencyInfo::releaseMemory() { BFI.reset(); }
void BlockFrequencyInfo::print(raw_ostream &OS) const {
if (BFI)
BFI->print(OS);
}
INITIALIZE_PASS_BEGIN(BlockFrequencyInfoWrapperPass, "block-freq",
"Block Frequency Analysis", true, true)
INITIALIZE_PASS_DEPENDENCY(BranchProbabilityInfoWrapperPass)
INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
INITIALIZE_PASS_END(BlockFrequencyInfoWrapperPass, "block-freq",
"Block Frequency Analysis", true, true)
char BlockFrequencyInfoWrapperPass::ID = 0;
BlockFrequencyInfoWrapperPass::BlockFrequencyInfoWrapperPass()
: FunctionPass(ID) {
initializeBlockFrequencyInfoWrapperPassPass(*PassRegistry::getPassRegistry());
}
BlockFrequencyInfoWrapperPass::~BlockFrequencyInfoWrapperPass() = default;
void BlockFrequencyInfoWrapperPass::print(raw_ostream &OS,
const Module *) const {
BFI.print(OS);
}
void BlockFrequencyInfoWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequired<BranchProbabilityInfoWrapperPass>();
AU.addRequired<LoopInfoWrapperPass>();
AU.setPreservesAll();
}
void BlockFrequencyInfoWrapperPass::releaseMemory() { BFI.releaseMemory(); }
bool BlockFrequencyInfoWrapperPass::runOnFunction(Function &F) {
BranchProbabilityInfo &BPI =
getAnalysis<BranchProbabilityInfoWrapperPass>().getBPI();
LoopInfo &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
BFI.calculate(F, BPI, LI);
return false;
}
[PM] Change the static object whose address is used to uniquely identify analyses to have a common type which is enforced rather than using a char object and a `void *` type when used as an identifier. This has a number of advantages. First, it at least helps some of the confusion raised in Justin Lebar's code review of why `void *` was being used everywhere by having a stronger type that connects to documentation about this. However, perhaps more importantly, it addresses a serious issue where the alignment of these pointer-like identifiers was unknown. This made it hard to use them in pointer-like data structures. We were already dodging this in dangerous ways to create the "all analyses" entry. In a subsequent patch I attempted to use these with TinyPtrVector and things fell apart in a very bad way. And it isn't just a compile time or type system issue. Worse than that, the actual alignment of these pointer-like opaque identifiers wasn't guaranteed to be a useful alignment as they were just characters. This change introduces a type to use as the "key" object whose address forms the opaque identifier. This both forces the objects to have proper alignment, and provides type checking that we get it right everywhere. It also makes the types somewhat less mysterious than `void *`. We could go one step further and introduce a truly opaque pointer-like type to return from the `ID()` static function rather than returning `AnalysisKey *`, but that didn't seem to be a clear win so this is just the initial change to get to a reliably typed and aligned object serving is a key for all the analyses. Thanks to Richard Smith and Justin Lebar for helping pick plausible names and avoid making this refactoring many times. =] And thanks to Sean for the super fast review! While here, I've tried to move away from the "PassID" nomenclature entirely as it wasn't really helping and is overloaded with old pass manager constructs. Now we have IDs for analyses, and key objects whose address can be used as IDs. Where possible and clear I've shortened this to just "ID". In a few places I kept "AnalysisID" to make it clear what was being identified. Differential Revision: https://reviews.llvm.org/D27031 llvm-svn: 287783
2016-11-24 01:53:26 +08:00
AnalysisKey BlockFrequencyAnalysis::Key;
BlockFrequencyInfo BlockFrequencyAnalysis::run(Function &F,
FunctionAnalysisManager &AM) {
BlockFrequencyInfo BFI;
BFI.calculate(F, AM.getResult<BranchProbabilityAnalysis>(F),
AM.getResult<LoopAnalysis>(F));
return BFI;
}
PreservedAnalyses
BlockFrequencyPrinterPass::run(Function &F, FunctionAnalysisManager &AM) {
OS << "Printing analysis results of BFI for function "
<< "'" << F.getName() << "':"
<< "\n";
AM.getResult<BlockFrequencyAnalysis>(F).print(OS);
return PreservedAnalyses::all();
}