38using namespace sampleprof;
42#define DEBUG_TYPE "fs-profile-loader"
46 cl::desc(
"Print setting flow sensitive branch probabilities"));
48 "fs-profile-debug-prob-diff-threshold",
cl::init(10),
49 cl::desc(
"Only show debug message if the branch probility is greater than "
50 "this value (in percentage)."));
53 "fs-profile-debug-bw-threshold",
cl::init(10000),
54 cl::desc(
"Only show debug message if the source branch weight is greater "
55 " than this value."));
59 cl::desc(
"View BFI before MIR loader"));
62 cl::desc(
"View BFI after MIR loader"));
68 "Load MIR Sample Profile",
99 if (
MI.isPseudoProbe()) {
101 Probe.
Id =
MI.getOperand(1).getImm();
102 Probe.
Type =
MI.getOperand(2).getImm();
103 Probe.
Attr =
MI.getOperand(3).getImm();
114namespace afdo_detail {
192 return std::error_code();
202 LLVM_DEBUG(
dbgs() <<
"\nPropagation complete. Setting branch probs\n");
211 Edge E = std::make_pair(BB, Succ);
215 if (BBWeight != SumEdgeWeight) {
216 LLVM_DEBUG(
dbgs() <<
"BBweight is not equal to SumEdgeWeight: BBWWeight="
217 << BBWeight <<
" SumEdgeWeight= " << SumEdgeWeight
219 BBWeight = SumEdgeWeight;
229 uint32_t MaxWeight = std::numeric_limits<uint32_t>::max();
231 if (BBWeight > MaxWeight) {
232 Factor = BBWeight / MaxWeight + 1;
241 Edge E = std::make_pair(BB, Succ);
243 EdgeWeight /= Factor;
245 assert(BBWeight >= EdgeWeight &&
246 "BBweight is larger than EdgeWeight -- should not happen.\n");
250 if (OldProb == NewProb)
258 if (OldProb > NewProb)
259 Diff = OldProb - NewProb;
261 Diff = NewProb - OldProb;
268 dbgs() <<
"Set branch fs prob: MBB (" << BB->
getNumber() <<
" -> "
271 dbgs() << DIL->getFilename() <<
":" << DIL->getLine() <<
":"
274 dbgs() <<
"-->" << SuccDIL->getFilename() <<
":" << SuccDIL->getLine()
275 <<
":" << SuccDIL->getColumn();
276 dbgs() <<
" W=" << BBWeightOrig <<
" " << OldProb <<
" --> " << NewProb
285 auto &Ctx = M.getContext();
289 if (std::error_code EC = ReaderOrErr.getError()) {
290 std::string Msg =
"Could not open profile: " + EC.message();
295 Reader = std::move(ReaderOrErr.get());
300 if (
Reader->profileIsProbeBased()) {
317 if (!
Reader->profileIsFS())
353 MIRSampleLoader = std::make_unique<MIRProfileLoader>(
354 FileName, RemappingFileName, std::move(VFS));
355 assert(LowBit < HighBit &&
"HighBit needs to be greater than Lowbit");
359 if (!MIRSampleLoader->isValid())
364 MBFI = &getAnalysis<MachineBlockFrequencyInfo>();
365 MIRSampleLoader->setInitVals(
366 &getAnalysis<MachineDominatorTree>(),
367 &getAnalysis<MachinePostDominatorTree>(), &getAnalysis<MachineLoopInfo>(),
368 MBFI, &getAnalysis<MachineOptimizationRemarkEmitterPass>().getORE());
374 MBFI->
view(
"MIR_Prof_loader_b." + MF.
getName(),
false);
377 bool Changed = MIRSampleLoader->runOnFunction(MF);
384 MBFI->
view(
"MIR_prof_loader_a." + MF.
getName(),
false);
390bool MIRProfileLoaderPass::doInitialization(
Module &M) {
391 LLVM_DEBUG(
dbgs() <<
"MIRProfileLoader pass working on Module " <<
M.getName()
394 MIRSampleLoader->setFSPass(
P);
395 return MIRSampleLoader->doInitialization(M);
398void MIRProfileLoaderPass::getAnalysisUsage(
AnalysisUsage &AU)
const {
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file defines the DenseMap class.
This file defines the DenseSet and SmallDenseSet classes.
cl::opt< bool > ImprovedFSDiscriminator("improved-fs-discriminator", cl::Hidden, cl::init(false), cl::desc("New FS discriminators encoding (incompatible with the original " "encoding)"))
static cl::opt< bool > ShowFSBranchProb("show-fs-branchprob", cl::Hidden, cl::init(false), cl::desc("Print setting flow sensitive branch probabilities"))
static cl::opt< bool > ViewBFIAfter("fs-viewbfi-after", cl::Hidden, cl::init(false), cl::desc("View BFI after MIR loader"))
static cl::opt< unsigned > FSProfileDebugBWThreshold("fs-profile-debug-bw-threshold", cl::init(10000), cl::desc("Only show debug message if the source branch weight is greater " " than this value."))
cl::opt< bool > ImprovedFSDiscriminator
static cl::opt< bool > ViewBFIBefore("fs-viewbfi-before", cl::Hidden, cl::init(false), cl::desc("View BFI before MIR loader"))
static cl::opt< unsigned > FSProfileDebugProbDiffThreshold("fs-profile-debug-prob-diff-threshold", cl::init(10), cl::desc("Only show debug message if the branch probility is greater than " "this value (in percentage)."))
#define INITIALIZE_PASS_DEPENDENCY(depName)
#define INITIALIZE_PASS_END(passName, arg, name, cfg, analysis)
#define INITIALIZE_PASS_BEGIN(passName, arg, name, cfg, analysis)
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
This file provides the interface for the sampled PGO profile loader base implementation.
This file provides the utility functions for the sampled PGO loader base implementation.
Defines the virtual file system interface vfs::FileSystem.
Represent the analysis usage information of a pass.
AnalysisUsage & addRequired()
void setPreservesAll()
Set by analyses that do not transform their input at all.
AnalysisUsage & addRequiredTransitive()
Implements a dense probed hash-table based set.
Diagnostic information for the sample profiler.
Represents either an error or a value T.
FunctionPass class - This class is used to implement most global optimizations.
Class to represent profile counts.
A smart pointer to a reference-counted object that inherits from RefCountedBase or ThreadSafeRefCount...
MIRProfileLoaderPass(std::string FileName="", std::string RemappingFileName="", FSDiscriminatorPass P=FSDiscriminatorPass::Pass1, IntrusiveRefCntPtr< vfs::FileSystem > FS=nullptr)
FS bits will only use the '1' bits in the Mask.
MIRProfileLoader(StringRef Name, StringRef RemapName, IntrusiveRefCntPtr< vfs::FileSystem > FS)
void setBranchProbs(MachineFunction &F)
ErrorOr< uint64_t > getInstWeight(const MachineInstr &MI) override
bool runOnFunction(MachineFunction &F)
MachineBlockFrequencyInfo * BFI
Hold the information of the basic block frequency.
FSDiscriminatorPass P
PassNum is the sequence number this pass is called, start from 1.
bool doInitialization(Module &M)
void setInitVals(MachineDominatorTree *MDT, MachinePostDominatorTree *MPDT, MachineLoopInfo *MLI, MachineBlockFrequencyInfo *MBFI, MachineOptimizationRemarkEmitter *MORE)
void setFSPass(FSDiscriminatorPass Pass)
int getNumber() const
MachineBasicBlocks are uniquely numbered at the function level, unless they're not in a MachineFuncti...
void setSuccProbability(succ_iterator I, BranchProbability Prob)
Set successor probability of a given iterator.
succ_iterator succ_begin()
unsigned succ_size() const
std::vector< MachineBasicBlock * >::iterator succ_iterator
DebugLoc findBranchDebugLoc()
Find and return the merged DebugLoc of the branch instructions of the block.
iterator_range< succ_iterator > successors()
iterator_range< pred_iterator > predecessors()
MachineBlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate machine basic b...
void view(const Twine &Name, bool isSimple=true) const
Pop up a ghostview window with the current block frequency propagation rendered using dot.
const MachineBranchProbabilityInfo * getMBPI() const
void calculate(const MachineFunction &F, const MachineBranchProbabilityInfo &MBPI, const MachineLoopInfo &MLI)
calculate - compute block frequency info for the given function.
BranchProbability getEdgeProbability(const MachineBasicBlock *Src, const MachineBasicBlock *Dst) const
DominatorTree Class - Concrete subclass of DominatorTreeBase that is used to compute a normal dominat...
MachineFunctionPass - This class adapts the FunctionPass interface to allow convenient creation of pa...
void getAnalysisUsage(AnalysisUsage &AU) const override
getAnalysisUsage - Subclasses that override getAnalysisUsage must call this.
StringRef getName() const
getName - Return the name of the corresponding LLVM function.
Function & getFunction()
Return the LLVM function that this machine code represents.
void RenumberBlocks(MachineBasicBlock *MBBFrom=nullptr)
RenumberBlocks - This discards all of the MachineBasicBlock numbers and recomputes them.
Representation of each machine instruction.
MachinePostDominatorTree - an analysis pass wrapper for DominatorTree used to compute the post-domina...
A Module instance is used to store all the information related to an LLVM module.
Pass interface - Implemented by all 'passes'.
bool computeAndPropagateWeights(FunctionT &F, const DenseSet< GlobalValue::GUID > &InlinedGUIDs)
Generate branch weight metadata for all branches in F.
void computeDominanceAndLoopInfo(FunctionT &F)
IntrusiveRefCntPtr< vfs::FileSystem > FS
VirtualFileSystem to load profile files from.
EdgeWeightMap EdgeWeights
Map edges to their computed weights.
OptRemarkEmitterT * ORE
Optimization Remark Emitter used to emit diagnostic remarks.
unsigned getFunctionLoc(FunctionT &Func)
Get the line number for the function header.
ErrorOr< uint64_t > getInstWeightImpl(const InstructionT &Inst)
EquivalenceClassMap EquivalenceClass
Equivalence classes for block weights.
std::unique_ptr< SampleProfileReader > Reader
Profile reader object.
DominatorTreePtrT DT
Dominance, post-dominance and loop information.
std::unique_ptr< PseudoProbeManager > ProbeManager
std::string Filename
Name of the profile file to load.
PostDominatorTreePtrT PDT
virtual ErrorOr< uint64_t > getProbeWeight(const InstructionT &Inst)
std::string RemappingFilename
Name of the profile remapping file to load.
FunctionSamples * Samples
Samples collected for the body of this function.
std::pair< const BasicBlockT *, const BasicBlockT * > Edge
void clearFunctionData(bool ResetDT=true)
Clear all the per-function data used to load samples and propagate weights.
BlockWeightMap BlockWeights
Map basic blocks to their computed weights.
StringRef - Represent a constant reference to a string, i.e.
bool equals(StringRef RHS) const
equals - Check for string equality, this is more efficient than compare() when the relative ordering ...
StringRef getName() const
Return a constant reference to the value's name.
A range adaptor for a pair of iterators.
static bool ProfileIsProbeBased
static ErrorOr< std::unique_ptr< SampleProfileReader > > create(const std::string Filename, LLVMContext &C, vfs::FileSystem &FS, FSDiscriminatorPass P=FSDiscriminatorPass::Base, const std::string RemapFilename="")
Create a sample profile reader appropriate to the file format.
initializer< Ty > init(const Ty &Val)
IntrusiveRefCntPtr< FileSystem > getRealFileSystem()
Gets an vfs::FileSystem for the 'real' file system, as seen by the operating system.
This is an optimization pass for GlobalISel generic memory operations.
static unsigned getFSPassBitBegin(sampleprof::FSDiscriminatorPass P)
char & MIRProfileLoaderPassID
This pass reads flow sensitive profile.
static unsigned getFSPassBitEnd(sampleprof::FSDiscriminatorPass P)
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."))
std::optional< PseudoProbe > extractProbe(const Instruction &Inst)
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
cl::opt< GVDAGType > ViewBlockLayoutWithBFI("view-block-layout-with-bfi", cl::Hidden, cl::desc("Pop up a window to show a dag displaying MBP layout and associated " "block frequencies of 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.")))
FunctionPass * createMIRProfileLoaderPass(std::string File, std::string RemappingFile, sampleprof::FSDiscriminatorPass P, IntrusiveRefCntPtr< vfs::FileSystem > FS)
Read Flow Sensitive Profile.
OutputIt move(R &&Range, OutputIt Out)
Provide wrappers to std::move which take ranges instead of having to pass begin/end explicitly.
Implement std::hash so that hash_code can be used in STL containers.
static PredRangeT getPredecessors(MachineBasicBlock *BB)
static SuccRangeT getSuccessors(MachineBasicBlock *BB)
static const MachineBasicBlock * getEntryBB(const MachineFunction *F)
static Function & getFunction(MachineFunction &F)