ORC Design and Implementation

Introduction

This document aims to provide a high-level overview of the design and implementation of the ORC JIT APIs. Except where otherwise stated all discussion refers to the modern ORCv2 APIs (available since LLVM 7). Clients wishing to transition from OrcV1 should see Section Transitioning from ORCv1 to ORCv2.

Use-cases

ORC provides a modular API for building JIT compilers. There are a number of use cases for such an API. For example:

1. The LLVM tutorials use a simple ORC-based JIT class to execute expressions compiled from a toy language: Kaleidoscope.

2. The LLVM debugger, LLDB, uses a cross-compiling JIT for expression evaluation. In this use case, cross compilation allows expressions compiled in the debugger process to be executed on the debug target process, which may be on a different device/architecture.

3. In high-performance JITs (e.g. JVMs, Julia) that want to make use of LLVM’s optimizations within an existing JIT infrastructure.

4. In interpreters and REPLs, e.g. Cling (C++) and the Swift interpreter.

By adopting a modular, library-based design we aim to make ORC useful in as many of these contexts as possible.

Features

ORC provides the following features:

JIT-linking

ORC provides APIs to link relocatable object files (COFF, ELF, MachO) [1] into a target process at runtime. The target process may be the same process that contains the JIT session object and jit-linker, or may be another process (even one running on a different machine or architecture) that communicates with the JIT via RPC.

LLVM IR compilation

ORC provides off the shelf components (IRCompileLayer, SimpleCompiler, ConcurrentIRCompiler) that make it easy to add LLVM IR to a JIT’d process.

Eager and lazy compilation

By default, ORC will compile symbols as soon as they are looked up in the JIT session object (ExecutionSession). Compiling eagerly by default makes it easy to use ORC as an in-memory compiler for an existing JIT (similar to how MCJIT is commonly used). However ORC also provides built-in support for lazy compilation via lazy-reexports (see Laziness).

Support for Custom Compilers and Program Representations

Clients can supply custom compilers for each symbol that they define in their JIT session. ORC will run the user-supplied compiler when the a definition of a symbol is needed. ORC is actually fully language agnostic: LLVM IR is not treated specially, and is supported via the same wrapper mechanism (the MaterializationUnit class) that is used for custom compilers.

Concurrent JIT’d code and Concurrent Compilation

JIT’d code may be executed in multiple threads, may spawn new threads, and may re-enter the ORC (e.g. to request lazy compilation) concurrently from multiple threads. Compilers launched my ORC can run concurrently (provided the client sets up an appropriate dispatcher). Built-in dependency tracking ensures that ORC does not release pointers to JIT’d code or data until all dependencies have also been JIT’d and they are safe to call or use.

Removable Code

Resources for JIT’d program representations

Orthogonality and Composability

Each of the features above can be used independently. It is possible to put ORC components together to make a non-lazy, in-process, single threaded JIT or a lazy, out-of-process, concurrent JIT, or anything in between.

LLJIT and LLLazyJIT

ORC provides two basic JIT classes off-the-shelf. These are useful both as examples of how to assemble ORC components to make a JIT, and as replacements for earlier LLVM JIT APIs (e.g. MCJIT).

The LLJIT class uses an IRCompileLayer and RTDyldObjectLinkingLayer to support compilation of LLVM IR and linking of relocatable object files. All operations are performed eagerly on symbol lookup (i.e. a symbol’s definition is compiled as soon as you attempt to look up its address). LLJIT is a suitable replacement for MCJIT in most cases (note: some more advanced features, e.g. JITEventListeners are not supported yet).

The LLLazyJIT extends LLJIT and adds a CompileOnDemandLayer to enable lazy compilation of LLVM IR. When an LLVM IR module is added via the addLazyIRModule method, function bodies in that module will not be compiled until they are first called. LLLazyJIT aims to provide a replacement of LLVM’s original (pre-MCJIT) JIT API.

LLJIT and LLLazyJIT instances can be created using their respective builder classes: LLJITBuilder and LLazyJITBuilder. For example, assuming you have a module M loaded on a ThreadSafeContext Ctx:

// Try to detect the host arch and construct an LLJIT instance.
auto JIT = LLJITBuilder().create();

// If we could not construct an instance, return an error.
if (!JIT)
  return JIT.takeError();

// Add the module.
if (auto Err = JIT->addIRModule(TheadSafeModule(std::move(M), Ctx)))
  return Err;

// Look up the JIT'd code entry point.
auto EntrySym = JIT->lookup("entry");
if (!EntrySym)
  return EntrySym.takeError();

// Cast the entry point address to a function pointer.
auto *Entry = (void(*)())EntrySym.getAddress();

// Call into JIT'd code.
Entry();

The builder classes provide a number of configuration options that can be specified before the JIT instance is constructed. For example:

// Build an LLLazyJIT instance that uses four worker threads for compilation,
// and jumps to a specific error handler (rather than null) on lazy compile
// failures.

void handleLazyCompileFailure() {
  // JIT'd code will jump here if lazy compilation fails, giving us an
  // opportunity to exit or throw an exception into JIT'd code.
  throw JITFailed();
}

auto JIT = LLLazyJITBuilder()
             .setNumCompileThreads(4)
             .setLazyCompileFailureAddr(
                 toJITTargetAddress(&handleLazyCompileFailure))
             .create();

// ...

For users wanting to get started with LLJIT a minimal example program can be found at llvm/examples/HowToUseLLJIT.

Design Overview

ORC’s JIT program model aims to emulate the linking and symbol resolution rules used by the static and dynamic linkers. This allows ORC to JIT arbitrary LLVM IR, including IR produced by an ordinary static compiler (e.g. clang) that uses constructs like symbol linkage and visibility, and weak [3] and common symbol definitions.

To see how this works, imagine a program foo which links against a pair of dynamic libraries: libA and libB. On the command line, building this program might look like:

$ clang++ -shared -o libA.dylib a1.cpp a2.cpp
$ clang++ -shared -o libB.dylib b1.cpp b2.cpp
$ clang++ -o myapp myapp.cpp -L. -lA -lB
$ ./myapp

In ORC, this would translate into API calls on a hypothetical CXXCompilingLayer (with error checking omitted for brevity) as:

ExecutionSession ES;
RTDyldObjectLinkingLayer ObjLinkingLayer(
    ES, []() { return std::make_unique<SectionMemoryManager>(); });
CXXCompileLayer CXXLayer(ES, ObjLinkingLayer);

// Create JITDylib "A" and add code to it using the CXX layer.
auto &LibA = ES.createJITDylib("A");
CXXLayer.add(LibA, MemoryBuffer::getFile("a1.cpp"));
CXXLayer.add(LibA, MemoryBuffer::getFile("a2.cpp"));

// Create JITDylib "B" and add code to it using the CXX layer.
auto &LibB = ES.createJITDylib("B");
CXXLayer.add(LibB, MemoryBuffer::getFile("b1.cpp"));
CXXLayer.add(LibB, MemoryBuffer::getFile("b2.cpp"));

// Create and specify the search order for the main JITDylib. This is
// equivalent to a "links against" relationship in a command-line link.
auto &MainJD = ES.createJITDylib("main");
MainJD.addToLinkOrder(&LibA);
MainJD.addToLinkOrder(&LibB);
CXXLayer.add(MainJD, MemoryBuffer::getFile("main.cpp"));

// Look up the JIT'd main, cast it to a function pointer, then call it.
auto MainSym = ExitOnErr(ES.lookup({&MainJD}, "main"));
auto *Main = (int(*)(int, char*[]))MainSym.getAddress();

int Result = Main(...);

This example tells us nothing about how or when compilation will happen. That will depend on the implementation of the hypothetical CXXCompilingLayer. The same linker-based symbol resolution rules will apply regardless of that implementation, however. For example, if a1.cpp and a2.cpp both define a function “foo” then ORCv2 will generate a duplicate definition error. On the other hand, if a1.cpp and b1.cpp both define “foo” there is no error (different dynamic libraries may define the same symbol). If main.cpp refers to “foo”, it should bind to the definition in LibA rather than the one in LibB, since main.cpp is part of the “main” dylib, and the main dylib links against LibA before LibB.

Many JIT clients will have no need for this strict adherence to the usual ahead-of-time linking rules, and should be able to get by just fine by putting all of their code in a single JITDylib. However, clients who want to JIT code for languages/projects that traditionally rely on ahead-of-time linking (e.g. C++) will find that this feature makes life much easier.

Symbol lookup in ORC serves two other important functions, beyond providing addresses for symbols: (1) It triggers compilation of the symbol(s) searched for (if they have not been compiled already), and (2) it provides the synchronization mechanism for concurrent compilation. The pseudo-code for the lookup process is:

construct a query object from a query set and query handler
lock the session
lodge query against requested symbols, collect required materializers (if any)
unlock the session
dispatch materializers (if any)

In this context a materializer is something that provides a working definition of a symbol upon request. Usually materializers are just wrappers for compilers, but they may also wrap a jit-linker directly (if the program representation backing the definitions is an object file), or may even be a class that writes bits directly into memory (for example, if the definitions are stubs). Materialization is the blanket term for any actions (compiling, linking, splatting bits, registering with runtimes, etc.) that are required to generate a symbol definition that is safe to call or access.

As each materializer completes its work it notifies the JITDylib, which in turn notifies any query objects that are waiting on the newly materialized definitions. Each query object maintains a count of the number of symbols that it is still waiting on, and once this count reaches zero the query object calls the query handler with a SymbolMap (a map of symbol names to addresses) describing the result. If any symbol fails to materialize the query immediately calls the query handler with an error.

The collected materialization units are sent to the ExecutionSession to be dispatched, and the dispatch behavior can be set by the client. By default each materializer is run on the calling thread. Clients are free to create new threads to run materializers, or to send the work to a work queue for a thread pool (this is what LLJIT/LLLazyJIT do).

Top Level APIs

Many of ORC’s top-level APIs are visible in the example above:

• ExecutionSession represents the JIT’d program and provides context for the JIT: It contains the JITDylibs, error reporting mechanisms, and dispatches the materializers.

• JITDylibs provide the symbol tables.

• Layers (ObjLinkingLayer and CXXLayer) are wrappers around compilers and allow clients to add uncompiled program representations supported by those compilers to JITDylibs.

• ResourceTrackers allow you to remove code.

Several other important APIs are used explicitly. JIT clients need not be aware of them, but Layer authors will use them:

• MaterializationUnit - When XXXLayer::add is invoked it wraps the given program representation (in this example, C++ source) in a MaterializationUnit, which is then stored in the JITDylib. MaterializationUnits are responsible for describing the definitions they provide, and for unwrapping the program representation and passing it back to the layer when compilation is required (this ownership shuffle makes writing thread-safe layers easier, since the ownership of the program representation will be passed back on the stack, rather than having to be fished out of a Layer member, which would require synchronization).

• MaterializationResponsibility - When a MaterializationUnit hands a program representation back to the layer it comes with an associated MaterializationResponsibility object. This object tracks the definitions that must be materialized and provides a way to notify the JITDylib once they are either successfully materialized or a failure occurs.

Absolute Symbols, Aliases, and Reexports

ORC makes it easy to define symbols with absolute addresses, or symbols that are simply aliases of other symbols:

Absolute Symbols

Absolute symbols are symbols that map directly to addresses without requiring further materialization, for example: “foo” = 0x1234. One use case for absolute symbols is allowing resolution of process symbols. E.g.

With this mapping established code added to the JIT can refer to printf symbolically rather than requiring the address of printf to be “baked in”. This in turn allows cached versions of the JIT’d code (e.g. compiled objects) to be re-used across JIT sessions as the JIT’d code no longer changes, only the absolute symbol definition does.

For process and library symbols the DynamicLibrarySearchGenerator utility (See How to Add Process and Library Symbols to JITDylibs) can be used to automatically build absolute symbol mappings for you. However the absoluteSymbols function is still useful for making non-global objects in your JIT visible to JIT’d code. For example, imagine that your JIT standard library needs access to your JIT object to make some calls. We could bake the address of your object into the library, but then it would need to be recompiled for each session:

We can turn this into a symbolic reference in the JIT standard library:

And then make our JIT object visible to the JIT standard library with an absolute symbol definition when the JIT is started:

Aliases and Reexports

Aliases and reexports allow you to define new symbols that map to existing symbols. This can be useful for changing linkage relationships between symbols across sessions without having to recompile code. For example, imagine that JIT’d code has access to a log function, void log(const char*) for which there are two implementations in the JIT standard library: log_fast and log_detailed. Your JIT can choose which one of these definitions will be used when the log symbol is referenced by setting up an alias at JIT startup time:

The symbolAliases function allows you to define aliases within a single JITDylib. The reexports function provides the same functionality, but operates across JITDylib boundaries. E.g.

The reexports utility can be handy for composing a single JITDylib interface by re-exporting symbols from several other JITDylibs.

Laziness

Laziness in ORC is provided by a utility called “lazy reexports”. A lazy reexport is similar to a regular reexport or alias: It provides a new name for an existing symbol. Unlike regular reexports however, lookups of lazy reexports do not trigger immediate materialization of the reexported symbol. Instead, they only trigger materialization of a function stub. This function stub is initialized to point at a lazy call-through, which provides reentry into the JIT. If the stub is called at runtime then the lazy call-through will look up the reexported symbol (triggering materialization for it if necessary), update the stub (to call directly to the reexported symbol on subsequent calls), and then return via the reexported symbol. By re-using the existing symbol lookup mechanism, lazy reexports inherit the same concurrency guarantees: calls to lazy reexports can be made from multiple threads concurrently, and the reexported symbol can be any state of compilation (uncompiled, already in the process of being compiled, or already compiled) and the call will succeed. This allows laziness to be safely mixed with features like remote compilation, concurrent compilation, concurrent JIT’d code, and speculative compilation.

There is one other key difference between regular reexports and lazy reexports that some clients must be aware of: The address of a lazy reexport will be different from the address of the reexported symbol (whereas a regular reexport is guaranteed to have the same address as the reexported symbol). Clients who care about pointer equality will generally want to use the address of the reexport as the canonical address of the reexported symbol. This will allow the address to be taken without forcing materialization of the reexport.

Usage example:

If JITDylib JD contains definitions for symbols foo_body and bar_body, we can create lazy entry points Foo and Bar in JITDylib JD2 by calling:

auto ReexportFlags = JITSymbolFlags::Exported | JITSymbolFlags::Callable;
JD2.define(
  lazyReexports(CallThroughMgr, StubsMgr, JD,
                SymbolAliasMap({
                  { Mangle("foo"), { Mangle("foo_body"), ReexportedFlags } },
                  { Mangle("bar"), { Mangle("bar_body"), ReexportedFlags } }
                }));

A full example of how to use lazyReexports with the LLJIT class can be found at llvm/examples/OrcV2Examples/LLJITWithLazyReexports.

Supporting Custom Compilers

TBD.

Transitioning from ORCv1 to ORCv2

Since LLVM 7.0, new ORC development work has focused on adding support for concurrent JIT compilation. The new APIs (including new layer interfaces and implementations, and new utilities) that support concurrency are collectively referred to as ORCv2, and the original, non-concurrent layers and utilities are now referred to as ORCv1.

The majority of the ORCv1 layers and utilities were renamed with a ‘Legacy’ prefix in LLVM 8.0, and have deprecation warnings attached in LLVM 9.0. In LLVM 12.0 ORCv1 will be removed entirely.

Transitioning from ORCv1 to ORCv2 should be easy for most clients. Most of the ORCv1 layers and utilities have ORCv2 counterparts [2] that can be directly substituted. However there are some design differences between ORCv1 and ORCv2 to be aware of:

1. ORCv2 fully adopts the JIT-as-linker model that began with MCJIT. Modules (and other program representations, e.g. Object Files) are no longer added directly to JIT classes or layers. Instead, they are added to JITDylib instances by layers. The JITDylib determines where the definitions reside, the layers determine how the definitions will be compiled. Linkage relationships between JITDylibs determine how inter-module references are resolved, and symbol resolvers are no longer used. See the section Design Overview for more details.

   Unless multiple JITDylibs are needed to model linkage relationships, ORCv1 clients should place all code in a single JITDylib. MCJIT clients should use LLJIT (see LLJIT and LLLazyJIT), and can place code in LLJIT’s default created main JITDylib (See LLJIT::getMainJITDylib()).

2. All JIT stacks now need an ExecutionSession instance. ExecutionSession manages the string pool, error reporting, synchronization, and symbol lookup.

3. ORCv2 uses uniqued strings (SymbolStringPtr instances) rather than string values in order to reduce memory overhead and improve lookup performance. See the subsection How to manage symbol strings.

4. IR layers require ThreadSafeModule instances, rather than std::unique_ptr<Module>s. ThreadSafeModule is a wrapper that ensures that Modules that use the same LLVMContext are not accessed concurrently. See How to use ThreadSafeModule and ThreadSafeContext.

5. Symbol lookup is no longer handled by layers. Instead, there is a lookup method on JITDylib that takes a list of JITDylibs to scan.

   ExecutionSession ES;
   JITDylib &JD1 = ...;
   JITDylib &JD2 = ...;
   
   auto Sym = ES.lookup({&JD1, &JD2}, ES.intern("_main"));
   

6. The removeModule/removeObject methods are replaced by ResourceTracker::remove. See the subsection How to remove code.

For code examples and suggestions of how to use the ORCv2 APIs, please see the section How-tos.

How-tos

How to manage symbol strings

Symbol strings in ORC are uniqued to improve lookup performance, reduce memory overhead, and allow symbol names to function as efficient keys. To get the unique SymbolStringPtr for a string value, call the ExecutionSession::intern method:

ExecutionSession ES;
/// ...
auto MainSymbolName = ES.intern("main");

If you wish to perform lookup using the C/IR name of a symbol you will also need to apply the platform linker-mangling before interning the string. On Linux this mangling is a no-op, but on other platforms it usually involves adding a prefix to the string (e.g. ‘_’ on Darwin). The mangling scheme is based on the DataLayout for the target. Given a DataLayout and an ExecutionSession, you can create a MangleAndInterner function object that will perform both jobs for you:

ExecutionSession ES;
const DataLayout &DL = ...;
MangleAndInterner Mangle(ES, DL);

// ...

// Portable IR-symbol-name lookup:
auto Sym = ES.lookup({&MainJD}, Mangle("main"));

How to create JITDylibs and set up linkage relationships

In ORC, all symbol definitions reside in JITDylibs. JITDylibs are created by calling the ExecutionSession::createJITDylib method with a unique name:

ExecutionSession ES;
auto &JD = ES.createJITDylib("libFoo.dylib");

The JITDylib is owned by the ExecutionEngine instance and will be freed when it is destroyed.

How to remove code

To remove an individual module from a JITDylib it must first be added using an explicit ResourceTracker. The module can then be removed by calling ResourceTracker::remove:

auto &JD = ... ;
auto M = ... ;

auto RT = JD.createResourceTracker();
Layer.add(RT, std::move(M)); // Add M to JD, tracking resources with RT

RT.remove(); // Remove M from JD.

Modules added directly to a JITDylib will be tracked by that JITDylib’s default resource tracker.

All code can be removed from a JITDylib by calling JITDylib::clear. This leaves the cleared JITDylib in an empty but usable state.

JITDylibs can be removed by calling ExecutionSession::removeJITDylib. This clears the JITDylib and then puts it into a defunct state. No further operations can be performed on the JITDylib, and it will be destroyed as soon as the last handle to it is released.

An example of how to use the resource management APIs can be found at llvm/examples/OrcV2Examples/LLJITRemovableCode.

How to add the support for custom program representation

In order to add the support for a custom program representation, a custom MaterializationUnit for the program representation, and a custom Layer are needed. The Layer will have two operations: add and emit. The add operation takes an instance of your program representation, builds one of your custom MaterializationUnits to hold it, then adds it to a JITDylib. The emit operation takes a MaterializationResponsibility object and an instance of your program representation and materializes it, usually by compiling it and handing the resulting object off to an ObjectLinkingLayer.

Your custom MaterializationUnit will have two operations: materialize and discard. The materialize function will be called for you when any symbol provided by the unit is looked up, and it should just call the emit function on your layer, passing in the given MaterializationResponsibility and the wrapped program representation. The discard function will be called if some weak symbol provided by your unit is not needed (because the JIT found an overriding definition). You can use this to drop your definition early, or just ignore it and let the linker drops the definition later.

Here is an example of an ASTLayer:

// ... In you JIT class
AstLayer astLayer;
// ...


class AstMaterializationUnit : public orc::MaterializationUnit {
public:
  AstMaterializationUnit(AstLayer &l, Ast &ast)
  : llvm::orc::MaterializationUnit(l.getInterface(ast)), astLayer(l),
  ast(ast) {};

  llvm::StringRef getName() const override {
    return "AstMaterializationUnit";
  }

  void materialize(std::unique_ptr<orc::MaterializationResponsibility> r) override {
    astLayer.emit(std::move(r), ast);
  };

private:
  void discard(const llvm::orc::JITDylib &jd, const llvm::orc::SymbolStringPtr &sym) override {
    llvm_unreachable("functions are not overridable");
  }


  AstLayer &astLayer;
  Ast &ast;
};

class AstLayer {
  llvhm::orc::IRLayer &baseLayer;
  llvhm::orc::MangleAndInterner &mangler;

public:
  AstLayer(llvm::orc::IRLayer &baseLayer, llvm::orc::MangleAndInterner &mangler)
  : baseLayer(baseLayer), mangler(mangler){};

  llvm::Error add(llvm::orc::ResourceTrackerSP &rt, Ast &ast) {
    return rt->getJITDylib().define(std::make_unique<AstMaterializationUnit>(*this, ast), rt);
  }

  void emit(std::unique_ptr<orc::MaterializationResponsibility> mr, Ast &ast) {
    // compileAst is just function that compiles the given AST and returns
    // a `llvm::orc::ThreadSafeModule`
    baseLayer.emit(std::move(mr), compileAst(ast));
  }

  llvm::orc::MaterializationUnit::Interface getInterface(Ast &ast) {
      SymbolFlagsMap Symbols;
      // Find all the symbols in the AST and for each of them
      // add it to the Symbols map.
      Symbols[mangler(someNameFromAST)] =
        JITSymbolFlags(JITSymbolFlags::Exported | JITSymbolFlags::Callable);
      return MaterializationUnit::Interface(std::move(Symbols), nullptr);
  }
};

Take look at the source code of Building A JIT’s Chapter 4 for a complete example.

How to use ThreadSafeModule and ThreadSafeContext

ThreadSafeModule and ThreadSafeContext are wrappers around Modules and LLVMContexts respectively. A ThreadSafeModule is a pair of a std::unique_ptr<Module> and a (possibly shared) ThreadSafeContext value. A ThreadSafeContext is a pair of a std::unique_ptr<LLVMContext> and a lock. This design serves two purposes: providing a locking scheme and lifetime management for LLVMContexts. The ThreadSafeContext may be locked to prevent accidental concurrent access by two Modules that use the same LLVMContext. The underlying LLVMContext is freed once all ThreadSafeContext values pointing to it are destroyed, allowing the context memory to be reclaimed as soon as the Modules referring to it are destroyed.

ThreadSafeContexts can be explicitly constructed from a std::unique_ptr<LLVMContext>:

ThreadSafeContext TSCtx(std::make_unique<LLVMContext>());

ThreadSafeModules can be constructed from a pair of a std::unique_ptr<Module> and a ThreadSafeContext value. ThreadSafeContext values may be shared between multiple ThreadSafeModules:

ThreadSafeModule TSM1(
  std::make_unique<Module>("M1", *TSCtx.getContext()), TSCtx);

ThreadSafeModule TSM2(
  std::make_unique<Module>("M2", *TSCtx.getContext()), TSCtx);

Before using a ThreadSafeContext, clients should ensure that either the context is only accessible on the current thread, or that the context is locked. In the example above (where the context is never locked) we rely on the fact that both TSM1 and TSM2, and TSCtx are all created on one thread. If a context is going to be shared between threads then it must be locked before any accessing or creating any Modules attached to it. E.g.

ThreadSafeContext TSCtx(std::make_unique<LLVMContext>());

ThreadPool TP(NumThreads);
JITStack J;

for (auto &ModulePath : ModulePaths) {
  TP.async(
    [&]() {
      auto Lock = TSCtx.getLock();
      auto M = loadModuleOnContext(ModulePath, TSCtx.getContext());
      J.addModule(ThreadSafeModule(std::move(M), TSCtx));
    });
}

TP.wait();

To make exclusive access to Modules easier to manage the ThreadSafeModule class provides a convenience function, withModuleDo, that implicitly (1) locks the associated context, (2) runs a given function object, (3) unlocks the context, and (3) returns the result generated by the function object. E.g.

ThreadSafeModule TSM = getModule(...);

// Dump the module:
size_t NumFunctionsInModule =
  TSM.withModuleDo(
    [](Module &M) { // <- Context locked before entering lambda.
      return M.size();
    } // <- Context unlocked after leaving.
  );

Clients wishing to maximize possibilities for concurrent compilation will want to create every new ThreadSafeModule on a new ThreadSafeContext. For this reason a convenience constructor for ThreadSafeModule is provided that implicitly constructs a new ThreadSafeContext value from a std::unique_ptr<LLVMContext>:

// Maximize concurrency opportunities by loading every module on a
// separate context.
for (const auto &IRPath : IRPaths) {
  auto Ctx = std::make_unique<LLVMContext>();
  auto M = std::make_unique<LLVMContext>("M", *Ctx);
  CompileLayer.add(MainJD, ThreadSafeModule(std::move(M), std::move(Ctx)));
}

Clients who plan to run single-threaded may choose to save memory by loading all modules on the same context:

// Save memory by using one context for all Modules:
ThreadSafeContext TSCtx(std::make_unique<LLVMContext>());
for (const auto &IRPath : IRPaths) {
  ThreadSafeModule TSM(parsePath(IRPath, *TSCtx.getContext()), TSCtx);
  CompileLayer.add(MainJD, ThreadSafeModule(std::move(TSM));
}

How to Add Process and Library Symbols to JITDylibs

JIT’d code may need to access symbols in the host program or in supporting libraries. The best way to enable this is to reflect these symbols into your JITDylibs so that they appear the same as any other symbol defined within the execution session (i.e. they are findable via ExecutionSession::lookup, and so visible to the JIT linker during linking).

One way to reflect external symbols is to add them manually using the absoluteSymbols function:

const DataLayout &DL = getDataLayout();
MangleAndInterner Mangle(ES, DL);

auto &JD = ES.createJITDylib("main");

JD.define(
  absoluteSymbols({
    { Mangle("puts"), pointerToJITTargetAddress(&puts)},
    { Mangle("gets"), pointerToJITTargetAddress(&getS)}
  }));

Using absoluteSymbols is reasonable if the set of symbols to be reflected is small and fixed. On the other hand, if the set of symbols is large or variable it may make more sense to have the definitions added for you on demand by a definition generator.A definition generator is an object that can be attached to a JITDylib, receiving a callback whenever a lookup within that JITDylib fails to find one or more symbols. The definition generator is given a chance to produce a definition of the missing symbol(s) before the lookup proceeds.

ORC provides the DynamicLibrarySearchGenerator utility for reflecting symbols from the process (or a specific dynamic library) for you. For example, to reflect the whole interface of a runtime library:

const DataLayout &DL = getDataLayout();
auto &JD = ES.createJITDylib("main");

if (auto DLSGOrErr =
    DynamicLibrarySearchGenerator::Load("/path/to/lib"
                                        DL.getGlobalPrefix()))
  JD.addGenerator(std::move(*DLSGOrErr);
else
  return DLSGOrErr.takeError();

// IR added to JD can now link against all symbols exported by the library
// at '/path/to/lib'.
CompileLayer.add(JD, loadModule(...));

The DynamicLibrarySearchGenerator utility can also be constructed with a filter function to restrict the set of symbols that may be reflected. For example, to expose an allowed set of symbols from the main process:

const DataLayout &DL = getDataLayout();
MangleAndInterner Mangle(ES, DL);

auto &JD = ES.createJITDylib("main");

DenseSet<SymbolStringPtr> AllowList({
    Mangle("puts"),
    Mangle("gets")
  });

// Use GetForCurrentProcess with a predicate function that checks the
// allowed list.
JD.addGenerator(cantFail(DynamicLibrarySearchGenerator::GetForCurrentProcess(
      DL.getGlobalPrefix(),
      [&](const SymbolStringPtr &S) { return AllowList.count(S); })));

// IR added to JD can now link against any symbols exported by the process
// and contained in the list.
CompileLayer.add(JD, loadModule(...));

References to process or library symbols could also be hardcoded into your IR or object files using the symbols’ raw addresses, however symbolic resolution using the JIT symbol tables should be preferred: it keeps the IR and objects readable and reusable in subsequent JIT sessions. Hardcoded addresses are difficult to read, and usually only good for one session.

Roadmap

ORC is still undergoing active development. Some current and future works are listed below.

Current Work

1. TargetProcessControl: Improvements to in-tree support for out-of-process execution

   The TargetProcessControl API provides various operations on the JIT target process (the one which will execute the JIT’d code), including memory allocation, memory writes, function execution, and process queries (e.g. for the target triple). By targeting this API new components can be developed which will work equally well for in-process and out-of-process JITing.

2. ORC RPC based TargetProcessControl implementation

   An ORC RPC based implementation of the TargetProcessControl API is currently under development to enable easy out-of-process JITing via file descriptors / sockets.

3. Core State Machine Cleanup

   The core ORC state machine is currently implemented between JITDylib and ExecutionSession. Methods are slowly being moved to ExecutionSession. This will tidy up the code base, and also allow us to support asynchronous removal of JITDylibs (in practice deleting an associated state object in ExecutionSession and leaving the JITDylib instance in a defunct state until all references to it have been released).

Near Future Work

1. ORC JIT Runtime Libraries

   We need a runtime library for JIT’d code. This would include things like TLS registration, reentry functions, registration code for language runtimes (e.g. Objective C and Swift) and other JIT specific runtime code. This should be built in a similar manner to compiler-rt (possibly even as part of it).

2. Remote jit_dlopen / jit_dlclose

   To more fully mimic the environment that static programs operate in we would like JIT’d code to be able to “dlopen” and “dlclose” JITDylibs, running all of their initializers/deinitializers on the current thread. This would require support from the runtime library described above.

3. Debugging support

   ORC currently supports the GDBRegistrationListener API when using RuntimeDyld as the underlying JIT linker. We will need a new solution for JITLink based platforms.

Further Future Work

1. Speculative Compilation

   ORC’s support for concurrent compilation allows us to easily enable speculative JIT compilation: compilation of code that is not needed yet, but which we have reason to believe will be needed in the future. This can be used to hide compile latency and improve JIT throughput. A proof-of-concept example of speculative compilation with ORC has already been developed (see llvm/examples/SpeculativeJIT). Future work on this is likely to focus on re-using and improving existing profiling support (currently used by PGO) to feed speculation decisions, as well as built-in tools to simplify use of speculative compilation.

[1]

Formats/architectures vary in terms of supported features. MachO and ELF tend to have better support than COFF. Patches very welcome!

[2]

The LazyEmittingLayer, RemoteObjectClientLayer and RemoteObjectServerLayer do not have counterparts in the new system. In the case of LazyEmittingLayer it was simply no longer needed: in ORCv2, deferring compilation until symbols are looked up is the default. The removal of RemoteObjectClientLayer and RemoteObjectServerLayer means that JIT stacks can no longer be split across processes, however this functionality appears not to have been used.

[3]

Weak definitions are currently handled correctly within dylibs, but if multiple dylibs provide a weak definition of a symbol then each will end up with its own definition (similar to how weak definitions are handled in Windows DLLs). This will be fixed in the future.