1 =============================================
2 Building a JIT: Per-function Lazy Compilation
3 =============================================
8 **This tutorial is under active development. It is incomplete and details may
9 change frequently.** Nonetheless we invite you to try it out as it stands, and
10 we welcome any feedback.
12 Chapter 3 Introduction
13 ======================
15 **Warning: This text is currently out of date due to ORC API updates.**
17 **The example code has been updated and can be used. The text will be updated
18 once the API churn dies down.**
20 Welcome to Chapter 3 of the "Building an ORC-based JIT in LLVM" tutorial. This
21 chapter discusses lazy JITing and shows you how to enable it by adding an ORC
22 CompileOnDemand layer the JIT from `Chapter 2 <BuildingAJIT2.html>`_.
27 When we add a module to the KaleidoscopeJIT class from Chapter 2 it is
28 immediately optimized, compiled and linked for us by the IRTransformLayer,
29 IRCompileLayer and RTDyldObjectLinkingLayer respectively. This scheme, where all the
30 work to make a Module executable is done up front, is simple to understand and
31 its performance characteristics are easy to reason about. However, it will lead
32 to very high startup times if the amount of code to be compiled is large, and
33 may also do a lot of unnecessary compilation if only a few compiled functions
34 are ever called at runtime. A truly "just-in-time" compiler should allow us to
35 defer the compilation of any given function until the moment that function is
36 first called, improving launch times and eliminating redundant work. In fact,
37 the ORC APIs provide us with a layer to lazily compile LLVM IR:
38 *CompileOnDemandLayer*.
40 The CompileOnDemandLayer class conforms to the layer interface described in
41 Chapter 2, but its addModule method behaves quite differently from the layers
42 we have seen so far: rather than doing any work up front, it just scans the
43 Modules being added and arranges for each function in them to be compiled the
44 first time it is called. To do this, the CompileOnDemandLayer creates two small
45 utilities for each function that it scans: a *stub* and a *compile
46 callback*. The stub is a pair of a function pointer (which will be pointed at
47 the function's implementation once the function has been compiled) and an
48 indirect jump through the pointer. By fixing the address of the indirect jump
49 for the lifetime of the program we can give the function a permanent "effective
50 address", one that can be safely used for indirection and function pointer
51 comparison even if the function's implementation is never compiled, or if it is
52 compiled more than once (due to, for example, recompiling the function at a
53 higher optimization level) and changes address. The second utility, the compile
54 callback, represents a re-entry point from the program into the compiler that
55 will trigger compilation and then execution of a function. By initializing the
56 function's stub to point at the function's compile callback, we enable lazy
57 compilation: The first attempted call to the function will follow the function
58 pointer and trigger the compile callback instead. The compile callback will
59 compile the function, update the function pointer for the stub, then execute
60 the function. On all subsequent calls to the function, the function pointer
61 will point at the already-compiled function, so there is no further overhead
62 from the compiler. We will look at this process in more detail in the next
63 chapter of this tutorial, but for now we'll trust the CompileOnDemandLayer to
64 set all the stubs and callbacks up for us. All we need to do is to add the
65 CompileOnDemandLayer to the top of our stack and we'll get the benefits of
66 lazy compilation. We just need a few changes to the source:
71 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
72 #include "llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h"
73 #include "llvm/ExecutionEngine/Orc/CompileUtils.h"
77 class KaleidoscopeJIT {
79 std::unique_ptr<TargetMachine> TM;
81 RTDyldObjectLinkingLayer ObjectLayer;
82 IRCompileLayer<decltype(ObjectLayer), SimpleCompiler> CompileLayer;
84 using OptimizeFunction =
85 std::function<std::shared_ptr<Module>(std::shared_ptr<Module>)>;
87 IRTransformLayer<decltype(CompileLayer), OptimizeFunction> OptimizeLayer;
89 std::unique_ptr<JITCompileCallbackManager> CompileCallbackManager;
90 CompileOnDemandLayer<decltype(OptimizeLayer)> CODLayer;
93 using ModuleHandle = decltype(CODLayer)::ModuleHandleT;
95 First we need to include the CompileOnDemandLayer.h header, then add two new
96 members: a std::unique_ptr<JITCompileCallbackManager> and a CompileOnDemandLayer,
97 to our class. The CompileCallbackManager member is used by the CompileOnDemandLayer
98 to create the compile callback needed for each function.
103 : TM(EngineBuilder().selectTarget()), DL(TM->createDataLayout()),
104 ObjectLayer([]() { return std::make_shared<SectionMemoryManager>(); }),
105 CompileLayer(ObjectLayer, SimpleCompiler(*TM)),
106 OptimizeLayer(CompileLayer,
107 [this](std::shared_ptr<Module> M) {
108 return optimizeModule(std::move(M));
110 CompileCallbackManager(
111 orc::createLocalCompileCallbackManager(TM->getTargetTriple(), 0)),
112 CODLayer(OptimizeLayer,
113 [this](Function &F) { return std::set<Function*>({&F}); },
114 *CompileCallbackManager,
115 orc::createLocalIndirectStubsManagerBuilder(
116 TM->getTargetTriple())) {
117 llvm::sys::DynamicLibrary::LoadLibraryPermanently(nullptr);
120 Next we have to update our constructor to initialize the new members. To create
121 an appropriate compile callback manager we use the
122 createLocalCompileCallbackManager function, which takes a TargetMachine and a
123 JITTargetAddress to call if it receives a request to compile an unknown
124 function. In our simple JIT this situation is unlikely to come up, so we'll
125 cheat and just pass '0' here. In a production quality JIT you could give the
126 address of a function that throws an exception in order to unwind the JIT'd
129 Now we can construct our CompileOnDemandLayer. Following the pattern from
130 previous layers we start by passing a reference to the next layer down in our
131 stack -- the OptimizeLayer. Next we need to supply a 'partitioning function':
132 when a not-yet-compiled function is called, the CompileOnDemandLayer will call
133 this function to ask us what we would like to compile. At a minimum we need to
134 compile the function being called (given by the argument to the partitioning
135 function), but we could also request that the CompileOnDemandLayer compile other
136 functions that are unconditionally called (or highly likely to be called) from
137 the function being called. For KaleidoscopeJIT we'll keep it simple and just
138 request compilation of the function that was called. Next we pass a reference to
139 our CompileCallbackManager. Finally, we need to supply an "indirect stubs
140 manager builder": a utility function that constructs IndirectStubManagers, which
141 are in turn used to build the stubs for the functions in each module. The
142 CompileOnDemandLayer will call the indirect stub manager builder once for each
143 call to addModule, and use the resulting indirect stubs manager to create
144 stubs for all functions in all modules in the set. If/when the module set is
145 removed from the JIT the indirect stubs manager will be deleted, freeing any
146 memory allocated to the stubs. We supply this function by using the
147 createLocalIndirectStubsManagerBuilder utility.
152 if (auto Sym = CODLayer.findSymbol(Name, false))
154 return cantFail(CODLayer.addModule(std::move(Ms),
155 std::move(Resolver)));
159 return CODLayer.findSymbol(MangledNameStream.str(), true);
163 CODLayer.removeModule(H);
166 Finally, we need to replace the references to OptimizeLayer in our addModule,
167 findSymbol, and removeModule methods. With that, we're up and running.
171 ** Chapter conclusion.**
176 Here is the complete code listing for our running example with a CompileOnDemand
177 layer added to enable lazy function-at-a-time compilation. To build this example, use:
182 clang++ -g toy.cpp `llvm-config --cxxflags --ldflags --system-libs --libs core orcjit native` -O3 -o toy
188 .. literalinclude:: ../../examples/Kaleidoscope/BuildingAJIT/Chapter3/KaleidoscopeJIT.h
191 `Next: Extreme Laziness -- Using Compile Callbacks to JIT directly from ASTs <BuildingAJIT4.html>`_