Reland [NFC] Move DroppedVariableStats to its own file and redesign it to be extensible. #117042
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
Member
| @llvm/pr-subscribers-llvm-ir Author: Shubham Sandeep Rastogi (rastogishubham) ChangesMoved the IR unit test to the CodeGen folder to resolve linker errors: error: undefined reference to 'vtable for llvm::DroppedVariableStatsIR' Patch is 40.69 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/117042.diff 8 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/DroppedVariableStats.h b/llvm/include/llvm/CodeGen/DroppedVariableStats.h new file mode 100644 index 00000000000000..371d775b02e876 --- /dev/null +++ b/llvm/include/llvm/CodeGen/DroppedVariableStats.h @@ -0,0 +1,226 @@ +///===- DroppedVariableStats.h - Opt Diagnostics -*- C++ -*----------------===// +/// +/// 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 +/// +///===---------------------------------------------------------------------===// +/// \file +/// Dropped Variable Statistics for Debug Information. Reports any number +/// of #dbg_value that get dropped due to an optimization pass. +/// +///===---------------------------------------------------------------------===// + +#ifndef LLVM_CODEGEN_DROPPEDVARIABLESTATS_H +#define LLVM_CODEGEN_DROPPEDVARIABLESTATS_H + +#include "llvm/CodeGen/MachinePassManager.h" +#include "llvm/IR/DebugInfoMetadata.h" +#include "llvm/IR/DiagnosticInfo.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/PassInstrumentation.h" + +namespace llvm { + +/// A unique key that represents a debug variable. +/// First const DIScope *: Represents the scope of the debug variable. +/// Second const DIScope *: Represents the InlinedAt scope of the debug +/// variable. const DILocalVariable *: It is a pointer to the debug variable +/// itself. +using VarID = + std::tuple<const DIScope *, const DIScope *, const DILocalVariable *>; + +/// A base class to collect and print dropped debug information variable +/// statistics. +class DroppedVariableStats { +public: + DroppedVariableStats(bool DroppedVarStatsEnabled) + : DroppedVariableStatsEnabled(DroppedVarStatsEnabled) { + if (DroppedVarStatsEnabled) + llvm::outs() + << "Pass Level, Pass Name, Num of Dropped Variables, Func or " + "Module Name\n"; + }; + + virtual ~DroppedVariableStats() = default; + + // We intend this to be unique per-compilation, thus no copies. + DroppedVariableStats(const DroppedVariableStats &) = delete; + void operator=(const DroppedVariableStats &) = delete; + + bool getPassDroppedVariables() { return PassDroppedVariables; } + +protected: + void setup() { + DebugVariablesStack.push_back( + {DenseMap<const Function *, DebugVariables>()}); + InlinedAts.push_back( + {DenseMap<StringRef, DenseMap<VarID, DILocation *>>()}); + } + + void cleanup() { + assert(!DebugVariablesStack.empty() && + "DebugVariablesStack shouldn't be empty!"); + assert(!InlinedAts.empty() && "InlinedAts shouldn't be empty!"); + DebugVariablesStack.pop_back(); + InlinedAts.pop_back(); + } + + bool DroppedVariableStatsEnabled = false; + struct DebugVariables { + /// DenseSet of VarIDs before an optimization pass has run. + DenseSet<VarID> DebugVariablesBefore; + /// DenseSet of VarIDs after an optimization pass has run. + DenseSet<VarID> DebugVariablesAfter; + }; + +protected: + /// A stack of a DenseMap, that maps DebugVariables for every pass to an + /// llvm::Function. A stack is used because an optimization pass can call + /// other passes. + SmallVector<DenseMap<const Function *, DebugVariables>> DebugVariablesStack; + + /// A DenseSet tracking whether a scope was visited before. + DenseSet<const DIScope *> VisitedScope; + /// A stack of DenseMaps, which map the name of an llvm::Function to a + /// DenseMap of VarIDs and their inlinedAt locations before an optimization + /// pass has run. + SmallVector<DenseMap<StringRef, DenseMap<VarID, DILocation *>>> InlinedAts; + /// Calculate the number of dropped variables in an llvm::Function or + /// llvm::MachineFunction and print the relevant information to stdout. + void calculateDroppedStatsAndPrint(DebugVariables &DbgVariables, + StringRef FuncName, StringRef PassID, + StringRef FuncOrModName, + StringRef PassLevel, const Function *Func); + + /// Check if a \p Var has been dropped or is a false positive. Also update the + /// \p DroppedCount if a debug variable is dropped. + bool updateDroppedCount(DILocation *DbgLoc, const DIScope *Scope, + const DIScope *DbgValScope, + DenseMap<VarID, DILocation *> &InlinedAtsMap, + VarID Var, unsigned &DroppedCount); + /// Run code to populate relevant data structures over an llvm::Function or + /// llvm::MachineFunction. + void run(DebugVariables &DbgVariables, StringRef FuncName, bool Before); + /// Populate the VarIDSet and InlinedAtMap with the relevant information + /// needed for before and after pass analysis to determine dropped variable + /// status. + void populateVarIDSetAndInlinedMap( + const DILocalVariable *DbgVar, DebugLoc DbgLoc, DenseSet<VarID> &VarIDSet, + DenseMap<StringRef, DenseMap<VarID, DILocation *>> &InlinedAtsMap, + StringRef FuncName, bool Before); + /// Visit every llvm::Instruction or llvm::MachineInstruction and check if the + /// debug variable denoted by its ID \p Var may have been dropped by an + /// optimization pass. + virtual void + visitEveryInstruction(unsigned &DroppedCount, + DenseMap<VarID, DILocation *> &InlinedAtsMap, + VarID Var) = 0; + /// Visit every debug record in an llvm::Function or llvm::MachineFunction + /// and call populateVarIDSetAndInlinedMap on it. + virtual void visitEveryDebugRecord( + DenseSet<VarID> &VarIDSet, + DenseMap<StringRef, DenseMap<VarID, DILocation *>> &InlinedAtsMap, + StringRef FuncName, bool Before) = 0; + +private: + /// Remove a dropped debug variable's VarID from all Sets in the + /// DroppedVariablesBefore stack. + void removeVarFromAllSets(VarID Var, const Function *F) { + // Do not remove Var from the last element, it will be popped from the + // stack. + for (auto &DebugVariablesMap : llvm::drop_end(DebugVariablesStack)) + DebugVariablesMap[F].DebugVariablesBefore.erase(Var); + } + /// Return true if \p Scope is the same as \p DbgValScope or a child scope of + /// \p DbgValScope, return false otherwise. + bool isScopeChildOfOrEqualTo(const DIScope *Scope, + const DIScope *DbgValScope); + /// Return true if \p InlinedAt is the same as \p DbgValInlinedAt or part of + /// the InlinedAt chain, return false otherwise. + bool isInlinedAtChildOfOrEqualTo(const DILocation *InlinedAt, + const DILocation *DbgValInlinedAt); + bool PassDroppedVariables = false; +}; + +/// A class to collect and print dropped debug information due to LLVM IR +/// optimization passes. After every LLVM IR pass is run, it will print how many +/// #dbg_values were dropped due to that pass. +class DroppedVariableStatsIR : public DroppedVariableStats { +public: + DroppedVariableStatsIR(bool DroppedVarStatsEnabled) + : llvm::DroppedVariableStats(DroppedVarStatsEnabled) {} + + virtual ~DroppedVariableStatsIR() = default; + + void runBeforePass(Any IR) { + setup(); + if (const auto *M = unwrapIR<Module>(IR)) + return this->runOnModule(M, true); + if (const auto *F = unwrapIR<Function>(IR)) + return this->runOnFunction(F, true); + } + + void runAfterPass(StringRef P, Any IR) { + if (const auto *M = unwrapIR<Module>(IR)) + runAfterPassModule(P, M); + else if (const auto *F = unwrapIR<Function>(IR)) + runAfterPassFunction(P, F); + cleanup(); + } + + void registerCallbacks(PassInstrumentationCallbacks &PIC); + +private: + const Function *Func; + + void runAfterPassFunction(StringRef PassID, const Function *F) { + runOnFunction(F, false); + calculateDroppedVarStatsOnFunction(F, PassID, F->getName().str(), + "Function"); + } + + void runAfterPassModule(StringRef PassID, const Module *M) { + runOnModule(M, false); + calculateDroppedVarStatsOnModule(M, PassID, M->getName().str(), "Module"); + } + /// Populate DebugVariablesBefore, DebugVariablesAfter, InlinedAts before or + /// after a pass has run to facilitate dropped variable calculation for an + /// llvm::Function. + void runOnFunction(const Function *F, bool Before); + /// Iterate over all Instructions in a Function and report any dropped debug + /// information. + void calculateDroppedVarStatsOnFunction(const Function *F, StringRef PassID, + StringRef FuncOrModName, + StringRef PassLevel); + /// Populate DebugVariablesBefore, DebugVariablesAfter, InlinedAts before or + /// after a pass has run to facilitate dropped variable calculation for an + /// llvm::Module. Calls runOnFunction on every Function in the Module. + void runOnModule(const Module *M, bool Before); + /// Iterate over all Functions in a Module and report any dropped debug + /// information. Will call calculateDroppedVarStatsOnFunction on every + /// Function. + void calculateDroppedVarStatsOnModule(const Module *M, StringRef PassID, + StringRef FuncOrModName, + StringRef PassLevel); + /// Override base class method to run on an llvm::Function specifically. + virtual void + visitEveryInstruction(unsigned &DroppedCount, + DenseMap<VarID, DILocation *> &InlinedAtsMap, + VarID Var) override; + /// Override base class method to run on #dbg_values specifically. + virtual void visitEveryDebugRecord( + DenseSet<VarID> &VarIDSet, + DenseMap<StringRef, DenseMap<VarID, DILocation *>> &InlinedAtsMap, + StringRef FuncName, bool Before) override; + + template <typename IRUnitT> static const IRUnitT *unwrapIR(Any IR) { + const IRUnitT **IRPtr = llvm::any_cast<const IRUnitT *>(&IR); + return IRPtr ? *IRPtr : nullptr; + } +}; + +} // namespace llvm + +#endif diff --git a/llvm/include/llvm/Passes/StandardInstrumentations.h b/llvm/include/llvm/Passes/StandardInstrumentations.h index 9301a12c740eec..12a34c099eaffe 100644 --- a/llvm/include/llvm/Passes/StandardInstrumentations.h +++ b/llvm/include/llvm/Passes/StandardInstrumentations.h @@ -19,6 +19,7 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringRef.h" #include "llvm/ADT/StringSet.h" +#include "llvm/CodeGen/DroppedVariableStats.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/DebugInfoMetadata.h" @@ -579,83 +580,6 @@ class PrintCrashIRInstrumentation { static void SignalHandler(void *); }; -/// A class to collect and print dropped debug information variable statistics. -/// After every LLVM IR pass is run, it will print how many #dbg_values were -/// dropped due to that pass. -class DroppedVariableStats { -public: - DroppedVariableStats(bool DroppedVarStatsEnabled) { - if (DroppedVarStatsEnabled) - llvm::outs() - << "Pass Level, Pass Name, Num of Dropped Variables, Func or " - "Module Name\n"; - }; - // We intend this to be unique per-compilation, thus no copies. - DroppedVariableStats(const DroppedVariableStats &) = delete; - void operator=(const DroppedVariableStats &) = delete; - - void registerCallbacks(PassInstrumentationCallbacks &PIC); - void runBeforePass(StringRef PassID, Any IR); - void runAfterPass(StringRef PassID, Any IR, const PreservedAnalyses &PA); - void runAfterPassInvalidated(StringRef PassID, const PreservedAnalyses &PA); - bool getPassDroppedVariables() { return PassDroppedVariables; } - -private: - bool PassDroppedVariables = false; - /// A unique key that represents a #dbg_value. - using VarID = - std::tuple<const DIScope *, const DIScope *, const DILocalVariable *>; - - struct DebugVariables { - /// DenseSet of VarIDs before an optimization pass has run. - DenseSet<VarID> DebugVariablesBefore; - /// DenseSet of VarIDs after an optimization pass has run. - DenseSet<VarID> DebugVariablesAfter; - }; - - /// A stack of a DenseMap, that maps DebugVariables for every pass to an - /// llvm::Function. A stack is used because an optimization pass can call - /// other passes. - SmallVector<DenseMap<const Function *, DebugVariables>> DebugVariablesStack; - - /// A DenseSet tracking whether a scope was visited before. - DenseSet<const DIScope *> VisitedScope; - /// A stack of DenseMaps, which map the name of an llvm::Function to a - /// DenseMap of VarIDs and their inlinedAt locations before an optimization - /// pass has run. - SmallVector<DenseMap<StringRef, DenseMap<VarID, DILocation *>>> InlinedAts; - - /// Iterate over all Functions in a Module and report any dropped debug - /// information. Will call calculateDroppedVarStatsOnFunction on every - /// Function. - void calculateDroppedVarStatsOnModule(const Module *M, StringRef PassID, - std::string FuncOrModName, - std::string PassLevel); - /// Iterate over all Instructions in a Function and report any dropped debug - /// information. - void calculateDroppedVarStatsOnFunction(const Function *F, StringRef PassID, - std::string FuncOrModName, - std::string PassLevel); - /// Populate DebugVariablesBefore, DebugVariablesAfter, InlinedAts before or - /// after a pass has run to facilitate dropped variable calculation for an - /// llvm::Function. - void runOnFunction(const Function *F, bool Before); - /// Populate DebugVariablesBefore, DebugVariablesAfter, InlinedAts before or - /// after a pass has run to facilitate dropped variable calculation for an - /// llvm::Module. Calls runOnFunction on every Function in the Module. - void runOnModule(const Module *M, bool Before); - /// Remove a dropped #dbg_value VarID from all Sets in the - /// DroppedVariablesBefore stack. - void removeVarFromAllSets(VarID Var, const Function *F); - /// Return true if \p Scope is the same as \p DbgValScope or a child scope of - /// \p DbgValScope, return false otherwise. - bool isScopeChildOfOrEqualTo(DIScope *Scope, const DIScope *DbgValScope); - /// Return true if \p InlinedAt is the same as \p DbgValInlinedAt or part of - /// the InlinedAt chain, return false otherwise. - bool isInlinedAtChildOfOrEqualTo(const DILocation *InlinedAt, - const DILocation *DbgValInlinedAt); -}; - /// This class provides an interface to register all the standard pass /// instrumentations and manages their state (if any). class StandardInstrumentations { @@ -673,7 +597,7 @@ class StandardInstrumentations { PrintCrashIRInstrumentation PrintCrashIR; IRChangedTester ChangeTester; VerifyInstrumentation Verify; - DroppedVariableStats DroppedStats; + DroppedVariableStatsIR DroppedStatsIR; bool VerifyEach; diff --git a/llvm/lib/CodeGen/CMakeLists.txt b/llvm/lib/CodeGen/CMakeLists.txt index 7b47c0e6f75dbe..263d4a9ee94d2e 100644 --- a/llvm/lib/CodeGen/CMakeLists.txt +++ b/llvm/lib/CodeGen/CMakeLists.txt @@ -50,6 +50,7 @@ add_llvm_component_library(LLVMCodeGen DeadMachineInstructionElim.cpp DetectDeadLanes.cpp DFAPacketizer.cpp + DroppedVariableStats.cpp DwarfEHPrepare.cpp EarlyIfConversion.cpp EdgeBundles.cpp diff --git a/llvm/lib/CodeGen/DroppedVariableStats.cpp b/llvm/lib/CodeGen/DroppedVariableStats.cpp new file mode 100644 index 00000000000000..122fcad1293f1e --- /dev/null +++ b/llvm/lib/CodeGen/DroppedVariableStats.cpp @@ -0,0 +1,194 @@ +///===- DroppedVariableStats.cpp ------------------------------------------===// +/// +/// 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 +/// +///===---------------------------------------------------------------------===// +/// \file +/// Dropped Variable Statistics for Debug Information. Reports any number +/// of #dbg_value that get dropped due to an optimization pass. +/// +///===---------------------------------------------------------------------===// + +#include "llvm/CodeGen/DroppedVariableStats.h" +#include "llvm/IR/DebugInfoMetadata.h" +#include "llvm/IR/InstIterator.h" +#include "llvm/IR/Module.h" + +using namespace llvm; + +bool DroppedVariableStats::isScopeChildOfOrEqualTo(const DIScope *Scope, + const DIScope *DbgValScope) { + while (Scope != nullptr) { + if (VisitedScope.find(Scope) == VisitedScope.end()) { + VisitedScope.insert(Scope); + if (Scope == DbgValScope) { + VisitedScope.clear(); + return true; + } + Scope = Scope->getScope(); + } else { + VisitedScope.clear(); + return false; + } + } + return false; +} + +bool DroppedVariableStats::isInlinedAtChildOfOrEqualTo( + const DILocation *InlinedAt, const DILocation *DbgValInlinedAt) { + if (DbgValInlinedAt == InlinedAt) + return true; + if (!DbgValInlinedAt) + return false; + auto *IA = InlinedAt; + while (IA) { + if (IA == DbgValInlinedAt) + return true; + IA = IA->getInlinedAt(); + } + return false; +} + +void DroppedVariableStats::calculateDroppedStatsAndPrint( + DebugVariables &DbgVariables, StringRef FuncName, StringRef PassID, + StringRef FuncOrModName, StringRef PassLevel, const Function *Func) { + unsigned DroppedCount = 0; + DenseSet<VarID> &DebugVariablesBeforeSet = DbgVariables.DebugVariablesBefore; + DenseSet<VarID> &DebugVariablesAfterSet = DbgVariables.DebugVariablesAfter; + DenseMap<VarID, DILocation *> &InlinedAtsMap = InlinedAts.back()[FuncName]; + // Find an Instruction that shares the same scope as the dropped #dbg_value or + // has a scope that is the child of the scope of the #dbg_value, and has an + // inlinedAt equal to the inlinedAt of the #dbg_value or it's inlinedAt chain + // contains the inlinedAt of the #dbg_value, if such an Instruction is found, + // debug information is dropped. + for (VarID Var : DebugVariablesBeforeSet) { + if (DebugVariablesAfterSet.contains(Var)) + continue; + visitEveryInstruction(DroppedCount, InlinedAtsMap, Var); + removeVarFromAllSets(Var, Func); + } + if (DroppedCount > 0) { + llvm::outs() << PassLevel << ", " << PassID << ", " << DroppedCount << ", " + << FuncOrModName << "\n"; + PassDroppedVariables = true; + } else + PassDroppedVariables = false; +} + +bool DroppedVariableStats::updateDroppedCount( + DILocation *DbgLoc, const DIScope *Scope, const DIScope *DbgValScope, + DenseMap<VarID, DILocation *> &InlinedAtsMap, VarID Var, + unsigned &DroppedCount) { + + // If the Scope is a child of, or equal to the DbgValScope and is inlined at + // the Var's InlinedAt location, return true to signify that the Var has been + // dropped. + if (isScopeChildOfOrEqualTo(Scope, DbgValScope)) + if (isInlinedAtChildOfOrEqualTo(DbgLoc->getInlinedAt(), + InlinedAtsMap[Var])) { + // Found another instruction in the variable's scope, so there exists a + // break point at which the variable could be observed. Count it as + // dropped. + DroppedCount++; + return true; + } + return false; +} + +void DroppedVariableStats::run(DebugVariables &DbgVariables, StringRef FuncName, + bool Before) { + auto &VarIDSet = (Before ? DbgVariables.DebugVariablesBefore + : DbgVariables.DebugVariablesAfter); + auto &InlinedAtsMap = InlinedAts.back(); + if (Before) + InlinedAtsMap.try_emplace(FuncName, DenseMap<VarID, DILocation *>()); + VarIDSet = DenseSet<VarID>(); + visitEveryDebugRecord(VarIDSet, InlinedAtsMap, FuncName, Before); +} + +void DroppedVariableStats::populateVarIDSetAndInlinedMap( + const DILocalVariable *DbgVar, DebugLoc DbgLoc, DenseSet<VarID> &VarIDSet, + DenseMap<StringRef, DenseMap<VarID, DILocation *>> &InlinedAtsMap, + StringRef FuncName, bool Before) { + VarID Key{DbgVar->getScope(), DbgLoc->getInlinedAtScope(), DbgVar}; + VarIDSet.insert(Key); + if (Before) + InlinedAt... [truncated] |
…t to be extensible. Moved the IR unit test to the CodeGen folder to resolve linker errors: error: undefined reference to 'vtable for llvm::DroppedVariableStatsIR'
rastogishubham force-pushed the DroppedRefactorFix branch from 1f4f368 to cecf619 Compare 
Collaborator
| LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/133/builds/7786 Here is the relevant piece of the build log for the reference |
rastogishubham added a commit that referenced this pull request Dec 3, 2024
Collaborator
| LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/76/builds/5002 Here is the relevant piece of the build log for the reference |
rastogishubham added a commit that referenced this pull request Dec 3, 2024
…design it to be extensible. (#117042)" This reverts commit acf3b1a. Broke https://lab.llvm.org/buildbot/#/builders/76/builds/5002 tools/clang/lib/CodeGen/CMakeFiles/obj.clangCodeGen.dir/BackendUtil.cpp.o:(.toc+0x258): undefined reference to `vtable for llvm::DroppedVariableStatsIR'
rastogishubham added a commit that referenced this pull request Dec 4, 2024
…edesign it to be extensible. (#117042)" (#118546)" This reverts commit 0c8928d. Broke Bot: https://lab.llvm.org/buildbot/#/builders/76/builds/5008 error: undefined reference to `vtable for llvm::DroppedVariableStatsIR'
Collaborator
| LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/146/builds/1751 Here is the relevant piece of the build log for the reference |
Collaborator
| LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/64/builds/1612 Here is the relevant piece of the build log for the reference |
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
Moved the IR unit test to the CodeGen folder to resolve linker errors:
error: undefined reference to 'vtable for llvm::DroppedVariableStatsIR'This patch is trying to reland #115563