[llvm] [ORC] Introduce IRPartitionLayer for common partition functionality. (PR #66812)

Sunho Kim via llvm-commits llvm-commits at lists.llvm.org
Tue Sep 19 13:12:50 PDT 2023


https://github.com/sunho created https://github.com/llvm/llvm-project/pull/66812

The IR module partitioning logic has been embeded in CompileOnDemandLayer but this functionality is also required for other purpose such as re-optimization.
This patch splits that IR partitioning logic out of CompileOnDemandLayer into a new IR layer IRPartitionLayer.

A motivation behind making it into new IR layer is for the possibility of "lazy re-optimizing JIT" where we'd like to have a composition like "CompileOnDemandLayer -> IRPartitionLayer -> ReOptLayer" as well as ordinary reoptimization like "IRPartitionLayer -> ReOptLayer"

>From adac015517504aa40eafb15fded4f7ab4b4b780d Mon Sep 17 00:00:00 2001
From: Sunho Kim <ksunhokim123 at gmail.com>
Date: Wed, 20 Sep 2023 05:11:15 +0900
Subject: [PATCH] [ORC] Introduce IRPartitionLayer for common partition
 functionality.

---
 .../BuildingAJIT/Chapter3/KaleidoscopeJIT.h   |   6 +-
 .../SpeculativeJIT/SpeculativeJIT.cpp         |   9 +-
 .../Orc/CompileOnDemandLayer.h                |  36 +--
 .../ExecutionEngine/Orc/IRPartitionLayer.h    |  83 +++++
 llvm/include/llvm/ExecutionEngine/Orc/LLJIT.h |   7 +-
 llvm/lib/ExecutionEngine/Orc/CMakeLists.txt   |   1 +
 .../Orc/CompileOnDemandLayer.cpp              | 282 +---------------
 .../ExecutionEngine/Orc/IRPartitionLayer.cpp  | 303 ++++++++++++++++++
 llvm/lib/ExecutionEngine/Orc/LLJIT.cpp        |   7 +-
 llvm/tools/lli/lli.cpp                        |   3 +-
 10 files changed, 415 insertions(+), 322 deletions(-)
 create mode 100644 llvm/include/llvm/ExecutionEngine/Orc/IRPartitionLayer.h
 create mode 100644 llvm/lib/ExecutionEngine/Orc/IRPartitionLayer.cpp

diff --git a/llvm/examples/Kaleidoscope/BuildingAJIT/Chapter3/KaleidoscopeJIT.h b/llvm/examples/Kaleidoscope/BuildingAJIT/Chapter3/KaleidoscopeJIT.h
index fd0e081ff2b4bd0..fee2d26e5d9254f 100644
--- a/llvm/examples/Kaleidoscope/BuildingAJIT/Chapter3/KaleidoscopeJIT.h
+++ b/llvm/examples/Kaleidoscope/BuildingAJIT/Chapter3/KaleidoscopeJIT.h
@@ -21,6 +21,7 @@
 #include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
 #include "llvm/ExecutionEngine/Orc/ExecutorProcessControl.h"
 #include "llvm/ExecutionEngine/Orc/IRCompileLayer.h"
+#include "llvm/ExecutionEngine/Orc/IRPartitionLayer.h"
 #include "llvm/ExecutionEngine/Orc/IRTransformLayer.h"
 #include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
 #include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
@@ -48,6 +49,7 @@ class KaleidoscopeJIT {
   RTDyldObjectLinkingLayer ObjectLayer;
   IRCompileLayer CompileLayer;
   IRTransformLayer OptimizeLayer;
+  IRPartitionLayer IPLayer;
   CompileOnDemandLayer CODLayer;
 
   JITDylib &MainJD;
@@ -68,8 +70,8 @@ class KaleidoscopeJIT {
         CompileLayer(*this->ES, ObjectLayer,
                      std::make_unique<ConcurrentIRCompiler>(std::move(JTMB))),
         OptimizeLayer(*this->ES, CompileLayer, optimizeModule),
-        CODLayer(*this->ES, OptimizeLayer,
-                 this->EPCIU->getLazyCallThroughManager(),
+        IPLayer(*this->ES, OptimizeLayer),
+        CODLayer(*this->ES, IPLayer, this->EPCIU->getLazyCallThroughManager(),
                  [this] { return this->EPCIU->createIndirectStubsManager(); }),
         MainJD(this->ES->createBareJITDylib("<main>")) {
     MainJD.addGenerator(
diff --git a/llvm/examples/SpeculativeJIT/SpeculativeJIT.cpp b/llvm/examples/SpeculativeJIT/SpeculativeJIT.cpp
index fdd376d82da5d89..a1722748c6ce069 100644
--- a/llvm/examples/SpeculativeJIT/SpeculativeJIT.cpp
+++ b/llvm/examples/SpeculativeJIT/SpeculativeJIT.cpp
@@ -3,6 +3,7 @@
 #include "llvm/ExecutionEngine/Orc/Core.h"
 #include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
 #include "llvm/ExecutionEngine/Orc/IRCompileLayer.h"
+#include "llvm/ExecutionEngine/Orc/IRPartitionLayer.h"
 #include "llvm/ExecutionEngine/Orc/IndirectionUtils.h"
 #include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
 #include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
@@ -107,13 +108,14 @@ class SpeculativeJIT {
       IndirectStubsManagerBuilderFunction ISMBuilder,
       std::unique_ptr<DynamicLibrarySearchGenerator> ProcessSymbolsGenerator)
       : ES(std::move(ES)), DL(std::move(DL)),
-        MainJD(this->ES->createBareJITDylib("<main>")), LCTMgr(std::move(LCTMgr)),
+        MainJD(this->ES->createBareJITDylib("<main>")),
+        LCTMgr(std::move(LCTMgr)),
         CompileLayer(*this->ES, ObjLayer,
                      std::make_unique<ConcurrentIRCompiler>(std::move(JTMB))),
         S(Imps, *this->ES),
         SpeculateLayer(*this->ES, CompileLayer, S, Mangle, BlockFreqQuery()),
-        CODLayer(*this->ES, SpeculateLayer, *this->LCTMgr,
-                 std::move(ISMBuilder)) {
+        IPLayer(*this->ES, SpeculateLayer),
+        CODLayer(*this->ES, IPLayer, *this->LCTMgr, std::move(ISMBuilder)) {
     MainJD.addGenerator(std::move(ProcessSymbolsGenerator));
     this->CODLayer.setImplMap(&Imps);
     this->ES->setDispatchTask(
@@ -147,6 +149,7 @@ class SpeculativeJIT {
   Speculator S;
   RTDyldObjectLinkingLayer ObjLayer{*ES, createMemMgr};
   IRSpeculationLayer SpeculateLayer;
+  IRPartitionLayer IPLayer;
   CompileOnDemandLayer CODLayer;
 };
 
diff --git a/llvm/include/llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h b/llvm/include/llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h
index 23a64b5836db85f..ee556d11b708352 100644
--- a/llvm/include/llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h
+++ b/llvm/include/llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h
@@ -54,37 +54,15 @@ namespace llvm {
 namespace orc {
 
 class CompileOnDemandLayer : public IRLayer {
-  friend class PartitioningIRMaterializationUnit;
-
 public:
   /// Builder for IndirectStubsManagers.
   using IndirectStubsManagerBuilder =
       std::function<std::unique_ptr<IndirectStubsManager>()>;
 
-  using GlobalValueSet = std::set<const GlobalValue *>;
-
-  /// Partitioning function.
-  using PartitionFunction =
-      std::function<std::optional<GlobalValueSet>(GlobalValueSet Requested)>;
-
-  /// Off-the-shelf partitioning which compiles all requested symbols (usually
-  /// a single function at a time).
-  static std::optional<GlobalValueSet>
-  compileRequested(GlobalValueSet Requested);
-
-  /// Off-the-shelf partitioning which compiles whole modules whenever any
-  /// symbol in them is requested.
-  static std::optional<GlobalValueSet>
-  compileWholeModule(GlobalValueSet Requested);
-
   /// Construct a CompileOnDemandLayer.
   CompileOnDemandLayer(ExecutionSession &ES, IRLayer &BaseLayer,
-                        LazyCallThroughManager &LCTMgr,
-                        IndirectStubsManagerBuilder BuildIndirectStubsManager);
-
-  /// Sets the partition function.
-  void setPartitionFunction(PartitionFunction Partition);
-
+                       LazyCallThroughManager &LCTMgr,
+                       IndirectStubsManagerBuilder BuildIndirectStubsManager);
   /// Sets the ImplSymbolMap
   void setImplMap(ImplSymbolMap *Imp);
 
@@ -111,22 +89,12 @@ class CompileOnDemandLayer : public IRLayer {
 
   PerDylibResources &getPerDylibResources(JITDylib &TargetD);
 
-  void cleanUpModule(Module &M);
-
-  void expandPartition(GlobalValueSet &Partition);
-
-  void emitPartition(std::unique_ptr<MaterializationResponsibility> R,
-                     ThreadSafeModule TSM,
-                     IRMaterializationUnit::SymbolNameToDefinitionMap Defs);
-
   mutable std::mutex CODLayerMutex;
 
   IRLayer &BaseLayer;
   LazyCallThroughManager &LCTMgr;
   IndirectStubsManagerBuilder BuildIndirectStubsManager;
   PerDylibResourcesMap DylibResources;
-  PartitionFunction Partition = compileRequested;
-  SymbolLinkagePromoter PromoteSymbols;
   ImplSymbolMap *AliaseeImpls = nullptr;
 };
 
diff --git a/llvm/include/llvm/ExecutionEngine/Orc/IRPartitionLayer.h b/llvm/include/llvm/ExecutionEngine/Orc/IRPartitionLayer.h
new file mode 100644
index 000000000000000..0535ce473837474
--- /dev/null
+++ b/llvm/include/llvm/ExecutionEngine/Orc/IRPartitionLayer.h
@@ -0,0 +1,83 @@
+//===- IRPartitionLayer.h - Partition IR module on lookup -------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// JIT layer for breaking up modules into smaller submodules that only contains
+// looked up symbols.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_EXECUTIONENGINE_ORC_IRPARTITIONLAYER_H
+#define LLVM_EXECUTIONENGINE_ORC_IRPARTITIONLAYER_H
+
+#include "llvm/ExecutionEngine/Orc/IndirectionUtils.h"
+#include "llvm/ExecutionEngine/Orc/Layer.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/Constant.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/Mangler.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
+
+namespace llvm {
+namespace orc {
+
+class IRPartitionLayer : public IRLayer {
+  friend class PartitioningIRMaterializationUnit;
+
+public:
+  using GlobalValueSet = std::set<const GlobalValue *>;
+
+  /// Partitioning function.
+  using PartitionFunction =
+      std::function<std::optional<GlobalValueSet>(GlobalValueSet Requested)>;
+
+  /// Construct a IRPartitionLayer.
+  IRPartitionLayer(ExecutionSession &ES, IRLayer &BaseLayer);
+
+  /// Off-the-shelf partitioning which compiles all requested symbols (usually
+  /// a single function at a time).
+  static std::optional<GlobalValueSet>
+  compileRequested(GlobalValueSet Requested);
+
+  /// Off-the-shelf partitioning which compiles whole modules whenever any
+  /// symbol in them is requested.
+  static std::optional<GlobalValueSet>
+  compileWholeModule(GlobalValueSet Requested);
+
+  /// Sets the partition function.
+  void setPartitionFunction(PartitionFunction Partition);
+
+  /// Emits the given module. This should not be called by clients: it will be
+  /// called by the JIT when a definition added via the add method is requested.
+  void emit(std::unique_ptr<MaterializationResponsibility> R,
+            ThreadSafeModule TSM) override;
+
+private:
+  void cleanUpModule(Module &M);
+
+  void expandPartition(GlobalValueSet &Partition);
+
+  void emitPartition(std::unique_ptr<MaterializationResponsibility> R,
+                     ThreadSafeModule TSM,
+                     IRMaterializationUnit::SymbolNameToDefinitionMap Defs);
+
+  IRLayer &BaseLayer;
+  PartitionFunction Partition = compileRequested;
+  SymbolLinkagePromoter PromoteSymbols;
+};
+
+} // namespace orc
+} // namespace llvm
+
+#endif
diff --git a/llvm/include/llvm/ExecutionEngine/Orc/LLJIT.h b/llvm/include/llvm/ExecutionEngine/Orc/LLJIT.h
index ab54e0c2c288205..896de521f13af80 100644
--- a/llvm/include/llvm/ExecutionEngine/Orc/LLJIT.h
+++ b/llvm/include/llvm/ExecutionEngine/Orc/LLJIT.h
@@ -17,6 +17,7 @@
 #include "llvm/ExecutionEngine/Orc/CompileUtils.h"
 #include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
 #include "llvm/ExecutionEngine/Orc/IRCompileLayer.h"
+#include "llvm/ExecutionEngine/Orc/IRPartitionLayer.h"
 #include "llvm/ExecutionEngine/Orc/IRTransformLayer.h"
 #include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
 #include "llvm/ExecutionEngine/Orc/ThreadSafeModule.h"
@@ -271,9 +272,8 @@ class LLLazyJIT : public LLJIT {
 public:
 
   /// Sets the partition function.
-  void
-  setPartitionFunction(CompileOnDemandLayer::PartitionFunction Partition) {
-    CODLayer->setPartitionFunction(std::move(Partition));
+  void setPartitionFunction(IRPartitionLayer::PartitionFunction Partition) {
+    IPLayer->setPartitionFunction(std::move(Partition));
   }
 
   /// Returns a reference to the on-demand layer.
@@ -293,6 +293,7 @@ class LLLazyJIT : public LLJIT {
   LLLazyJIT(LLLazyJITBuilderState &S, Error &Err);
 
   std::unique_ptr<LazyCallThroughManager> LCTMgr;
+  std::unique_ptr<IRPartitionLayer> IPLayer;
   std::unique_ptr<CompileOnDemandLayer> CODLayer;
 };
 
diff --git a/llvm/lib/ExecutionEngine/Orc/CMakeLists.txt b/llvm/lib/ExecutionEngine/Orc/CMakeLists.txt
index 3256ed8b7362c66..51d1ff1a86f2dbc 100644
--- a/llvm/lib/ExecutionEngine/Orc/CMakeLists.txt
+++ b/llvm/lib/ExecutionEngine/Orc/CMakeLists.txt
@@ -27,6 +27,7 @@ add_llvm_component_library(LLVMOrcJIT
   IndirectionUtils.cpp
   IRCompileLayer.cpp
   IRTransformLayer.cpp
+  IRPartitionLayer.cpp
   JITTargetMachineBuilder.cpp
   LazyReexports.cpp
   Layer.cpp
diff --git a/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp b/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
index 6448adaa0ceb36f..9296bc2b389ab24 100644
--- a/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
@@ -9,6 +9,7 @@
 #include "llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
+#include "llvm/ExecutionEngine/Orc/Layer.h"
 #include "llvm/IR/Mangler.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/FormatVariadic.h"
@@ -17,101 +18,6 @@
 using namespace llvm;
 using namespace llvm::orc;
 
-static ThreadSafeModule extractSubModule(ThreadSafeModule &TSM,
-                                         StringRef Suffix,
-                                         GVPredicate ShouldExtract) {
-
-  auto DeleteExtractedDefs = [](GlobalValue &GV) {
-    // Bump the linkage: this global will be provided by the external module.
-    GV.setLinkage(GlobalValue::ExternalLinkage);
-
-    // Delete the definition in the source module.
-    if (isa<Function>(GV)) {
-      auto &F = cast<Function>(GV);
-      F.deleteBody();
-      F.setPersonalityFn(nullptr);
-    } else if (isa<GlobalVariable>(GV)) {
-      cast<GlobalVariable>(GV).setInitializer(nullptr);
-    } else if (isa<GlobalAlias>(GV)) {
-      // We need to turn deleted aliases into function or variable decls based
-      // on the type of their aliasee.
-      auto &A = cast<GlobalAlias>(GV);
-      Constant *Aliasee = A.getAliasee();
-      assert(A.hasName() && "Anonymous alias?");
-      assert(Aliasee->hasName() && "Anonymous aliasee");
-      std::string AliasName = std::string(A.getName());
-
-      if (isa<Function>(Aliasee)) {
-        auto *F = cloneFunctionDecl(*A.getParent(), *cast<Function>(Aliasee));
-        A.replaceAllUsesWith(F);
-        A.eraseFromParent();
-        F->setName(AliasName);
-      } else if (isa<GlobalVariable>(Aliasee)) {
-        auto *G = cloneGlobalVariableDecl(*A.getParent(),
-                                          *cast<GlobalVariable>(Aliasee));
-        A.replaceAllUsesWith(G);
-        A.eraseFromParent();
-        G->setName(AliasName);
-      } else
-        llvm_unreachable("Alias to unsupported type");
-    } else
-      llvm_unreachable("Unsupported global type");
-  };
-
-  auto NewTSM = cloneToNewContext(TSM, ShouldExtract, DeleteExtractedDefs);
-  NewTSM.withModuleDo([&](Module &M) {
-    M.setModuleIdentifier((M.getModuleIdentifier() + Suffix).str());
-  });
-
-  return NewTSM;
-}
-
-namespace llvm {
-namespace orc {
-
-class PartitioningIRMaterializationUnit : public IRMaterializationUnit {
-public:
-  PartitioningIRMaterializationUnit(ExecutionSession &ES,
-                                    const IRSymbolMapper::ManglingOptions &MO,
-                                    ThreadSafeModule TSM,
-                                    CompileOnDemandLayer &Parent)
-      : IRMaterializationUnit(ES, MO, std::move(TSM)), Parent(Parent) {}
-
-  PartitioningIRMaterializationUnit(
-      ThreadSafeModule TSM, Interface I,
-      SymbolNameToDefinitionMap SymbolToDefinition,
-      CompileOnDemandLayer &Parent)
-      : IRMaterializationUnit(std::move(TSM), std::move(I),
-                              std::move(SymbolToDefinition)),
-        Parent(Parent) {}
-
-private:
-  void materialize(std::unique_ptr<MaterializationResponsibility> R) override {
-    Parent.emitPartition(std::move(R), std::move(TSM),
-                         std::move(SymbolToDefinition));
-  }
-
-  void discard(const JITDylib &V, const SymbolStringPtr &Name) override {
-    // All original symbols were materialized by the CODLayer and should be
-    // final. The function bodies provided by M should never be overridden.
-    llvm_unreachable("Discard should never be called on an "
-                     "ExtractingIRMaterializationUnit");
-  }
-
-  mutable std::mutex SourceModuleMutex;
-  CompileOnDemandLayer &Parent;
-};
-
-std::optional<CompileOnDemandLayer::GlobalValueSet>
-CompileOnDemandLayer::compileRequested(GlobalValueSet Requested) {
-  return std::move(Requested);
-}
-
-std::optional<CompileOnDemandLayer::GlobalValueSet>
-CompileOnDemandLayer::compileWholeModule(GlobalValueSet Requested) {
-  return std::nullopt;
-}
-
 CompileOnDemandLayer::CompileOnDemandLayer(
     ExecutionSession &ES, IRLayer &BaseLayer, LazyCallThroughManager &LCTMgr,
     IndirectStubsManagerBuilder BuildIndirectStubsManager)
@@ -119,13 +25,10 @@ CompileOnDemandLayer::CompileOnDemandLayer(
       LCTMgr(LCTMgr),
       BuildIndirectStubsManager(std::move(BuildIndirectStubsManager)) {}
 
-void CompileOnDemandLayer::setPartitionFunction(PartitionFunction Partition) {
-  this->Partition = std::move(Partition);
-}
-
 void CompileOnDemandLayer::setImplMap(ImplSymbolMap *Imp) {
   this->AliaseeImpls = Imp;
 }
+
 void CompileOnDemandLayer::emit(
     std::unique_ptr<MaterializationResponsibility> R, ThreadSafeModule TSM) {
   assert(TSM && "Null module");
@@ -138,10 +41,6 @@ void CompileOnDemandLayer::emit(
 
   SymbolAliasMap NonCallables;
   SymbolAliasMap Callables;
-  TSM.withModuleDo([&](Module &M) {
-    // First, do some cleanup on the module:
-    cleanUpModule(M);
-  });
 
   for (auto &KV : R->getSymbols()) {
     auto &Name = KV.first;
@@ -152,11 +51,10 @@ void CompileOnDemandLayer::emit(
       NonCallables[Name] = SymbolAliasMapEntry(Name, Flags);
   }
 
-  // Create a partitioning materialization unit and lodge it with the
-  // implementation dylib.
+  // Lodge symbols with the implementation dylib.
   if (auto Err = PDR.getImplDylib().define(
-          std::make_unique<PartitioningIRMaterializationUnit>(
-              ES, *getManglingOptions(), std::move(TSM), *this))) {
+          std::make_unique<BasicIRLayerMaterializationUnit>(
+              BaseLayer, *getManglingOptions(), std::move(TSM)))) {
     ES.reportError(std::move(Err));
     R->failMaterialization();
     return;
@@ -210,173 +108,3 @@ CompileOnDemandLayer::getPerDylibResources(JITDylib &TargetD) {
 
   return I->second;
 }
-
-void CompileOnDemandLayer::cleanUpModule(Module &M) {
-  for (auto &F : M.functions()) {
-    if (F.isDeclaration())
-      continue;
-
-    if (F.hasAvailableExternallyLinkage()) {
-      F.deleteBody();
-      F.setPersonalityFn(nullptr);
-      continue;
-    }
-  }
-}
-
-void CompileOnDemandLayer::expandPartition(GlobalValueSet &Partition) {
-  // Expands the partition to ensure the following rules hold:
-  // (1) If any alias is in the partition, its aliasee is also in the partition.
-  // (2) If any aliasee is in the partition, its aliases are also in the
-  //     partiton.
-  // (3) If any global variable is in the partition then all global variables
-  //     are in the partition.
-  assert(!Partition.empty() && "Unexpected empty partition");
-
-  const Module &M = *(*Partition.begin())->getParent();
-  bool ContainsGlobalVariables = false;
-  std::vector<const GlobalValue *> GVsToAdd;
-
-  for (const auto *GV : Partition)
-    if (isa<GlobalAlias>(GV))
-      GVsToAdd.push_back(
-          cast<GlobalValue>(cast<GlobalAlias>(GV)->getAliasee()));
-    else if (isa<GlobalVariable>(GV))
-      ContainsGlobalVariables = true;
-
-  for (auto &A : M.aliases())
-    if (Partition.count(cast<GlobalValue>(A.getAliasee())))
-      GVsToAdd.push_back(&A);
-
-  if (ContainsGlobalVariables)
-    for (auto &G : M.globals())
-      GVsToAdd.push_back(&G);
-
-  for (const auto *GV : GVsToAdd)
-    Partition.insert(GV);
-}
-
-void CompileOnDemandLayer::emitPartition(
-    std::unique_ptr<MaterializationResponsibility> R, ThreadSafeModule TSM,
-    IRMaterializationUnit::SymbolNameToDefinitionMap Defs) {
-
-  // FIXME: Need a 'notify lazy-extracting/emitting' callback to tie the
-  //        extracted module key, extracted module, and source module key
-  //        together. This could be used, for example, to provide a specific
-  //        memory manager instance to the linking layer.
-
-  auto &ES = getExecutionSession();
-  GlobalValueSet RequestedGVs;
-  for (auto &Name : R->getRequestedSymbols()) {
-    if (Name == R->getInitializerSymbol())
-      TSM.withModuleDo([&](Module &M) {
-        for (auto &GV : getStaticInitGVs(M))
-          RequestedGVs.insert(&GV);
-      });
-    else {
-      assert(Defs.count(Name) && "No definition for symbol");
-      RequestedGVs.insert(Defs[Name]);
-    }
-  }
-
-  /// Perform partitioning with the context lock held, since the partition
-  /// function is allowed to access the globals to compute the partition.
-  auto GVsToExtract =
-      TSM.withModuleDo([&](Module &M) { return Partition(RequestedGVs); });
-
-  // Take a 'None' partition to mean the whole module (as opposed to an empty
-  // partition, which means "materialize nothing"). Emit the whole module
-  // unmodified to the base layer.
-  if (GVsToExtract == std::nullopt) {
-    Defs.clear();
-    BaseLayer.emit(std::move(R), std::move(TSM));
-    return;
-  }
-
-  // If the partition is empty, return the whole module to the symbol table.
-  if (GVsToExtract->empty()) {
-    if (auto Err =
-            R->replace(std::make_unique<PartitioningIRMaterializationUnit>(
-                std::move(TSM),
-                MaterializationUnit::Interface(R->getSymbols(),
-                                               R->getInitializerSymbol()),
-                std::move(Defs), *this))) {
-      getExecutionSession().reportError(std::move(Err));
-      R->failMaterialization();
-      return;
-    }
-    return;
-  }
-
-  // Ok -- we actually need to partition the symbols. Promote the symbol
-  // linkages/names, expand the partition to include any required symbols
-  // (i.e. symbols that can't be separated from our partition), and
-  // then extract the partition.
-  //
-  // FIXME: We apply this promotion once per partitioning. It's safe, but
-  // overkill.
-  auto ExtractedTSM =
-      TSM.withModuleDo([&](Module &M) -> Expected<ThreadSafeModule> {
-        auto PromotedGlobals = PromoteSymbols(M);
-        if (!PromotedGlobals.empty()) {
-
-          MangleAndInterner Mangle(ES, M.getDataLayout());
-          SymbolFlagsMap SymbolFlags;
-          IRSymbolMapper::add(ES, *getManglingOptions(),
-                              PromotedGlobals, SymbolFlags);
-
-          if (auto Err = R->defineMaterializing(SymbolFlags))
-            return std::move(Err);
-        }
-
-        expandPartition(*GVsToExtract);
-
-        // Submodule name is given by hashing the names of the globals.
-        std::string SubModuleName;
-        {
-          std::vector<const GlobalValue*> HashGVs;
-          HashGVs.reserve(GVsToExtract->size());
-          for (const auto *GV : *GVsToExtract)
-            HashGVs.push_back(GV);
-          llvm::sort(HashGVs, [](const GlobalValue *LHS, const GlobalValue *RHS) {
-              return LHS->getName() < RHS->getName();
-            });
-          hash_code HC(0);
-          for (const auto *GV : HashGVs) {
-            assert(GV->hasName() && "All GVs to extract should be named by now");
-            auto GVName = GV->getName();
-            HC = hash_combine(HC, hash_combine_range(GVName.begin(), GVName.end()));
-          }
-          raw_string_ostream(SubModuleName)
-            << ".submodule."
-            << formatv(sizeof(size_t) == 8 ? "{0:x16}" : "{0:x8}",
-                       static_cast<size_t>(HC))
-            << ".ll";
-        }
-
-        // Extract the requested partiton (plus any necessary aliases) and
-        // put the rest back into the impl dylib.
-        auto ShouldExtract = [&](const GlobalValue &GV) -> bool {
-          return GVsToExtract->count(&GV);
-        };
-
-        return extractSubModule(TSM, SubModuleName , ShouldExtract);
-      });
-
-  if (!ExtractedTSM) {
-    ES.reportError(ExtractedTSM.takeError());
-    R->failMaterialization();
-    return;
-  }
-
-  if (auto Err = R->replace(std::make_unique<PartitioningIRMaterializationUnit>(
-          ES, *getManglingOptions(), std::move(TSM), *this))) {
-    ES.reportError(std::move(Err));
-    R->failMaterialization();
-    return;
-  }
-  BaseLayer.emit(std::move(R), std::move(*ExtractedTSM));
-}
-
-} // end namespace orc
-} // end namespace llvm
diff --git a/llvm/lib/ExecutionEngine/Orc/IRPartitionLayer.cpp b/llvm/lib/ExecutionEngine/Orc/IRPartitionLayer.cpp
new file mode 100644
index 000000000000000..9ad171beac7fe2e
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/Orc/IRPartitionLayer.cpp
@@ -0,0 +1,303 @@
+//===----- IRPartitionLayer.cpp - Partition IR module into submodules -----===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ExecutionEngine/Orc/IRPartitionLayer.h"
+#include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
+#include "llvm/ExecutionEngine/Orc/IndirectionUtils.h"
+
+using namespace llvm;
+using namespace llvm::orc;
+
+static ThreadSafeModule extractSubModule(ThreadSafeModule &TSM,
+                                         StringRef Suffix,
+                                         GVPredicate ShouldExtract) {
+
+  auto DeleteExtractedDefs = [](GlobalValue &GV) {
+    // Bump the linkage: this global will be provided by the external module.
+    GV.setLinkage(GlobalValue::ExternalLinkage);
+
+    // Delete the definition in the source module.
+    if (isa<Function>(GV)) {
+      auto &F = cast<Function>(GV);
+      F.deleteBody();
+      F.setPersonalityFn(nullptr);
+    } else if (isa<GlobalVariable>(GV)) {
+      cast<GlobalVariable>(GV).setInitializer(nullptr);
+    } else if (isa<GlobalAlias>(GV)) {
+      // We need to turn deleted aliases into function or variable decls based
+      // on the type of their aliasee.
+      auto &A = cast<GlobalAlias>(GV);
+      Constant *Aliasee = A.getAliasee();
+      assert(A.hasName() && "Anonymous alias?");
+      assert(Aliasee->hasName() && "Anonymous aliasee");
+      std::string AliasName = std::string(A.getName());
+
+      if (isa<Function>(Aliasee)) {
+        auto *F = cloneFunctionDecl(*A.getParent(), *cast<Function>(Aliasee));
+        A.replaceAllUsesWith(F);
+        A.eraseFromParent();
+        F->setName(AliasName);
+      } else if (isa<GlobalVariable>(Aliasee)) {
+        auto *G = cloneGlobalVariableDecl(*A.getParent(),
+                                          *cast<GlobalVariable>(Aliasee));
+        A.replaceAllUsesWith(G);
+        A.eraseFromParent();
+        G->setName(AliasName);
+      } else
+        llvm_unreachable("Alias to unsupported type");
+    } else
+      llvm_unreachable("Unsupported global type");
+  };
+
+  auto NewTSM = cloneToNewContext(TSM, ShouldExtract, DeleteExtractedDefs);
+  NewTSM.withModuleDo([&](Module &M) {
+    M.setModuleIdentifier((M.getModuleIdentifier() + Suffix).str());
+  });
+
+  return NewTSM;
+}
+
+namespace llvm {
+namespace orc {
+
+class PartitioningIRMaterializationUnit : public IRMaterializationUnit {
+public:
+  PartitioningIRMaterializationUnit(ExecutionSession &ES,
+                                    const IRSymbolMapper::ManglingOptions &MO,
+                                    ThreadSafeModule TSM,
+                                    IRPartitionLayer &Parent)
+      : IRMaterializationUnit(ES, MO, std::move(TSM)), Parent(Parent) {}
+
+  PartitioningIRMaterializationUnit(
+      ThreadSafeModule TSM, Interface I,
+      SymbolNameToDefinitionMap SymbolToDefinition, IRPartitionLayer &Parent)
+      : IRMaterializationUnit(std::move(TSM), std::move(I),
+                              std::move(SymbolToDefinition)),
+        Parent(Parent) {}
+
+private:
+  void materialize(std::unique_ptr<MaterializationResponsibility> R) override {
+    Parent.emitPartition(std::move(R), std::move(TSM),
+                         std::move(SymbolToDefinition));
+  }
+
+  void discard(const JITDylib &V, const SymbolStringPtr &Name) override {
+    // All original symbols were materialized by the CODLayer and should be
+    // final. The function bodies provided by M should never be overridden.
+    llvm_unreachable("Discard should never be called on an "
+                     "ExtractingIRMaterializationUnit");
+  }
+
+  IRPartitionLayer &Parent;
+};
+
+} // namespace orc
+} // namespace llvm
+
+IRPartitionLayer::IRPartitionLayer(ExecutionSession &ES, IRLayer &BaseLayer)
+    : IRLayer(ES, BaseLayer.getManglingOptions()), BaseLayer(BaseLayer) {}
+
+void IRPartitionLayer::setPartitionFunction(PartitionFunction Partition) {
+  this->Partition = Partition;
+}
+
+std::optional<IRPartitionLayer::GlobalValueSet>
+IRPartitionLayer::compileRequested(GlobalValueSet Requested) {
+  return std::move(Requested);
+}
+
+std::optional<IRPartitionLayer::GlobalValueSet>
+IRPartitionLayer::compileWholeModule(GlobalValueSet Requested) {
+  return std::nullopt;
+}
+
+void IRPartitionLayer::emit(std::unique_ptr<MaterializationResponsibility> R,
+                            ThreadSafeModule TSM) {
+  assert(TSM && "Null module");
+
+  auto &ES = getExecutionSession();
+  TSM.withModuleDo([&](Module &M) {
+    // First, do some cleanup on the module:
+    cleanUpModule(M);
+  });
+
+  // Create a partitioning materialization unit and pass the responsibility.
+  if (auto Err = R->replace(std::make_unique<PartitioningIRMaterializationUnit>(
+          ES, *getManglingOptions(), std::move(TSM), *this))) {
+    ES.reportError(std::move(Err));
+    R->failMaterialization();
+    return;
+  }
+}
+
+void IRPartitionLayer::cleanUpModule(Module &M) {
+  for (auto &F : M.functions()) {
+    if (F.isDeclaration())
+      continue;
+
+    if (F.hasAvailableExternallyLinkage()) {
+      F.deleteBody();
+      F.setPersonalityFn(nullptr);
+      continue;
+    }
+  }
+}
+
+void IRPartitionLayer::expandPartition(GlobalValueSet &Partition) {
+  // Expands the partition to ensure the following rules hold:
+  // (1) If any alias is in the partition, its aliasee is also in the partition.
+  // (2) If any aliasee is in the partition, its aliases are also in the
+  //     partiton.
+  // (3) If any global variable is in the partition then all global variables
+  //     are in the partition.
+  assert(!Partition.empty() && "Unexpected empty partition");
+
+  const Module &M = *(*Partition.begin())->getParent();
+  bool ContainsGlobalVariables = false;
+  std::vector<const GlobalValue *> GVsToAdd;
+
+  for (const auto *GV : Partition)
+    if (isa<GlobalAlias>(GV))
+      GVsToAdd.push_back(
+          cast<GlobalValue>(cast<GlobalAlias>(GV)->getAliasee()));
+    else if (isa<GlobalVariable>(GV))
+      ContainsGlobalVariables = true;
+
+  for (auto &A : M.aliases())
+    if (Partition.count(cast<GlobalValue>(A.getAliasee())))
+      GVsToAdd.push_back(&A);
+
+  if (ContainsGlobalVariables)
+    for (auto &G : M.globals())
+      GVsToAdd.push_back(&G);
+
+  for (const auto *GV : GVsToAdd)
+    Partition.insert(GV);
+}
+
+void IRPartitionLayer::emitPartition(
+    std::unique_ptr<MaterializationResponsibility> R, ThreadSafeModule TSM,
+    IRMaterializationUnit::SymbolNameToDefinitionMap Defs) {
+
+  // FIXME: Need a 'notify lazy-extracting/emitting' callback to tie the
+  //        extracted module key, extracted module, and source module key
+  //        together. This could be used, for example, to provide a specific
+  //        memory manager instance to the linking layer.
+
+  auto &ES = getExecutionSession();
+  GlobalValueSet RequestedGVs;
+  for (auto &Name : R->getRequestedSymbols()) {
+    if (Name == R->getInitializerSymbol())
+      TSM.withModuleDo([&](Module &M) {
+        for (auto &GV : getStaticInitGVs(M))
+          RequestedGVs.insert(&GV);
+      });
+    else {
+      assert(Defs.count(Name) && "No definition for symbol");
+      RequestedGVs.insert(Defs[Name]);
+    }
+  }
+
+  /// Perform partitioning with the context lock held, since the partition
+  /// function is allowed to access the globals to compute the partition.
+  auto GVsToExtract =
+      TSM.withModuleDo([&](Module &M) { return Partition(RequestedGVs); });
+
+  // Take a 'None' partition to mean the whole module (as opposed to an empty
+  // partition, which means "materialize nothing"). Emit the whole module
+  // unmodified to the base layer.
+  if (GVsToExtract == std::nullopt) {
+    Defs.clear();
+    BaseLayer.emit(std::move(R), std::move(TSM));
+    return;
+  }
+
+  // If the partition is empty, return the whole module to the symbol table.
+  if (GVsToExtract->empty()) {
+    if (auto Err =
+            R->replace(std::make_unique<PartitioningIRMaterializationUnit>(
+                std::move(TSM),
+                MaterializationUnit::Interface(R->getSymbols(),
+                                               R->getInitializerSymbol()),
+                std::move(Defs), *this))) {
+      getExecutionSession().reportError(std::move(Err));
+      R->failMaterialization();
+      return;
+    }
+    return;
+  }
+
+  // Ok -- we actually need to partition the symbols. Promote the symbol
+  // linkages/names, expand the partition to include any required symbols
+  // (i.e. symbols that can't be separated from our partition), and
+  // then extract the partition.
+  //
+  // FIXME: We apply this promotion once per partitioning. It's safe, but
+  // overkill.
+  auto ExtractedTSM = TSM.withModuleDo([&](Module &M)
+                                           -> Expected<ThreadSafeModule> {
+    auto PromotedGlobals = PromoteSymbols(M);
+    if (!PromotedGlobals.empty()) {
+
+      MangleAndInterner Mangle(ES, M.getDataLayout());
+      SymbolFlagsMap SymbolFlags;
+      IRSymbolMapper::add(ES, *getManglingOptions(), PromotedGlobals,
+                          SymbolFlags);
+
+      if (auto Err = R->defineMaterializing(SymbolFlags))
+        return std::move(Err);
+    }
+
+    expandPartition(*GVsToExtract);
+
+    // Submodule name is given by hashing the names of the globals.
+    std::string SubModuleName;
+    {
+      std::vector<const GlobalValue *> HashGVs;
+      HashGVs.reserve(GVsToExtract->size());
+      for (const auto *GV : *GVsToExtract)
+        HashGVs.push_back(GV);
+      llvm::sort(HashGVs, [](const GlobalValue *LHS, const GlobalValue *RHS) {
+        return LHS->getName() < RHS->getName();
+      });
+      hash_code HC(0);
+      for (const auto *GV : HashGVs) {
+        assert(GV->hasName() && "All GVs to extract should be named by now");
+        auto GVName = GV->getName();
+        HC = hash_combine(HC, hash_combine_range(GVName.begin(), GVName.end()));
+      }
+      raw_string_ostream(SubModuleName)
+          << ".submodule."
+          << formatv(sizeof(size_t) == 8 ? "{0:x16}" : "{0:x8}",
+                     static_cast<size_t>(HC))
+          << ".ll";
+    }
+
+    // Extract the requested partiton (plus any necessary aliases) and
+    // put the rest back into the impl dylib.
+    auto ShouldExtract = [&](const GlobalValue &GV) -> bool {
+      return GVsToExtract->count(&GV);
+    };
+
+    return extractSubModule(TSM, SubModuleName, ShouldExtract);
+  });
+
+  if (!ExtractedTSM) {
+    ES.reportError(ExtractedTSM.takeError());
+    R->failMaterialization();
+    return;
+  }
+
+  if (auto Err = R->replace(std::make_unique<PartitioningIRMaterializationUnit>(
+          ES, *getManglingOptions(), std::move(TSM), *this))) {
+    ES.reportError(std::move(Err));
+    R->failMaterialization();
+    return;
+  }
+  BaseLayer.emit(std::move(R), std::move(*ExtractedTSM));
+}
diff --git a/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp b/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
index 18ab9e6b940fefe..15fc50bad94aabf 100644
--- a/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
@@ -1288,9 +1288,12 @@ LLLazyJIT::LLLazyJIT(LLLazyJITBuilderState &S, Error &Err) : LLJIT(S, Err) {
     return;
   }
 
+  // Create the IP Layer.
+  IPLayer = std::make_unique<IRPartitionLayer>(*ES, *InitHelperTransformLayer);
+
   // Create the COD layer.
-  CODLayer = std::make_unique<CompileOnDemandLayer>(
-      *ES, *InitHelperTransformLayer, *LCTMgr, std::move(ISMBuilder));
+  CODLayer = std::make_unique<CompileOnDemandLayer>(*ES, *IPLayer, *LCTMgr,
+                                                    std::move(ISMBuilder));
 
   if (S.NumCompileThreads > 0)
     CODLayer->setCloneToNewContextOnEmit(true);
diff --git a/llvm/tools/lli/lli.cpp b/llvm/tools/lli/lli.cpp
index 360ffa628a47b3f..1378622cb71511d 100644
--- a/llvm/tools/lli/lli.cpp
+++ b/llvm/tools/lli/lli.cpp
@@ -30,6 +30,7 @@
 #include "llvm/ExecutionEngine/Orc/EPCEHFrameRegistrar.h"
 #include "llvm/ExecutionEngine/Orc/EPCGenericRTDyldMemoryManager.h"
 #include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
+#include "llvm/ExecutionEngine/Orc/IRPartitionLayer.h"
 #include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
 #include "llvm/ExecutionEngine/Orc/LLJIT.h"
 #include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
@@ -984,7 +985,7 @@ int runOrcJIT(const char *ProgName) {
   }
 
   if (PerModuleLazy)
-    J->setPartitionFunction(orc::CompileOnDemandLayer::compileWholeModule);
+    J->setPartitionFunction(orc::IRPartitionLayer::compileWholeModule);
 
   auto Dump = createDebugDumper();
 



More information about the llvm-commits mailing list