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