1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Extending LLVM: Adding instructions, intrinsics, types, etc.</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title">
12 Extending LLVM: Adding instructions, intrinsics, types, etc.
16 <li><a href="#introduction">Introduction and Warning</a></li>
17 <li><a href="#intrinsic">Adding a new intrinsic function</a></li>
18 <li><a href="#instruction">Adding a new instruction</a></li>
19 <li><a href="#type">Adding a new type</a>
21 <li><a href="#fund_type">Adding a new fundamental type</a></li>
22 <li><a href="#derived_type">Adding a new derived type</a></li>
26 <div class="doc_author">
27 <p>Written by <a href="http://misha.brukman.net">Misha Brukman</a></p>
30 <!-- *********************************************************************** -->
31 <div class="doc_section">
32 <a name="introduction">Introduction and Warning</a>
34 <!-- *********************************************************************** -->
36 <div class="doc_text">
38 <p>During the course of using LLVM, you may wish to customize it for your
39 research project or for experimentation. At this point, you may realize that
40 you need to add something to LLVM, whether it be a new fundamental type, a new
41 intrinsic function, or a whole new instruction.</p>
43 <p>When you come to this realization, stop and think. Do you really need to
44 extend LLVM? Is it a new fundamental capability that LLVM does not support at
45 its current incarnation or can it be synthesized from already pre-existing LLVM
46 elements? If you are not sure, ask on the <a
47 href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM-dev</a> list. The
48 reason is that extending LLVM will get involved as you need to update all the
49 different passes that you intend to use with your extension, and there are
50 <em>many</em> LLVM analyses and transformations, so it may be quite a bit of
53 <p>Adding an <a href="#intrinsic">intrinsic function</a> is easier than adding
54 an instruction, and is transparent to optimization passes which treat it as an
55 unanalyzable function. If your added functionality can be expressed as a
56 function call, an intrinsic function is the method of choice for LLVM
59 <p>Before you invest a significant amount of effort into a non-trivial
60 extension, <span class="doc_warning">ask on the list</span> if what you are
61 looking to do can be done with already-existing infrastructure, or if maybe
62 someone else is already working on it. You will save yourself a lot of time and
63 effort by doing so.</p>
67 <!-- *********************************************************************** -->
68 <div class="doc_section">
69 <a name="intrinsic">Adding a new intrinsic function</a>
71 <!-- *********************************************************************** -->
73 <div class="doc_text">
75 <p>Adding a new intrinsic function to LLVM is much easier than adding a new
76 instruction. Almost all extensions to LLVM should start as an intrinsic
77 function and then be turned into an instruction if warranted.</p>
80 <li><tt>llvm/docs/LangRef.html</tt>:
81 Document the intrinsic. Decide whether it is code generator specific and
82 what the restrictions are. Talk to other people about it so that you are
83 sure it's a good idea.</li>
85 <li><tt>llvm/include/llvm/Intrinsics.h</tt>:
86 add an enum in the <tt>llvm::Intrinsic</tt> namespace</li>
88 <li><tt>llvm/lib/CodeGen/IntrinsicLowering.cpp</tt>:
89 implement the lowering for this intrinsic</li>
91 <li><tt>llvm/lib/VMCore/Verifier.cpp</tt>:
92 Add code to check the invariants of the intrinsic are respected.</li>
94 <li><tt>llvm/lib/VMCore/Function.cpp (<tt>Function::getIntrinsicID()</tt>)</tt>:
95 Identify the new intrinsic function, returning the enum for the intrinsic
98 <li><tt>llvm/lib/Analysis/BasicAliasAnalysis.cpp</tt>: If the new intrinsic does
99 not access memory or does not write to memory, add it to the relevant list
102 <li><tt>llvm/lib/Transforms/Utils/Local.cpp</tt>: If it is possible to constant
103 propagate your intrinsic, add support to it in the
104 <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li>
106 <li>Test your intrinsic</li>
107 <li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li>
110 <p>If this intrinsic requires code generator support (ie, it cannot be lowered).
111 You should also add support to the code generator in question.</p>
115 <!-- *********************************************************************** -->
116 <div class="doc_section">
117 <a name="instruction">Adding a new instruction</a>
119 <!-- *********************************************************************** -->
121 <div class="doc_text">
123 <p><span class="doc_warning">WARNING: adding instructions changes the bytecode
124 format, and it will take some effort to maintain compatibility with
125 the previous version.</span> Only add an instruction if it is absolutely
130 <li><tt>llvm/include/llvm/Instruction.def</tt>:
131 add a number for your instruction and an enum name</li>
133 <li><tt>llvm/include/llvm/Instructions.h</tt>:
134 add a definition for the class that will represent your instruction</li>
136 <li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>:
137 add a prototype for a visitor to your new instruction type</li>
139 <li><tt>llvm/lib/AsmParser/Lexer.l</tt>:
140 add a new token to parse your instruction from assembly text file</li>
142 <li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>:
143 add the grammar on how your instruction can be read and what it will
144 construct as a result</li>
146 <li><tt>llvm/lib/Bytecode/Reader/InstructionReader.cpp</tt>:
147 add a case for your instruction and how it will be parsed from bytecode</li>
149 <li><tt>llvm/lib/VMCore/Instruction.cpp</tt>:
150 add a case for how your instruction will be printed out to assembly</li>
152 <li><tt>llvm/lib/VMCore/Instructions.cpp</tt>:
153 implement the class you defined in <tt>llvm/include/llvm/Instructions.h</tt></li>
157 <p>Also, you need to implement (or modify) any analyses or passes that you want
158 to understand this new instruction.</p>
163 <!-- *********************************************************************** -->
164 <div class="doc_section">
165 <a name="type">Adding a new type</a>
167 <!-- *********************************************************************** -->
169 <div class="doc_text">
171 <p><span class="doc_warning">WARNING: adding new types changes the bytecode
172 format, and will break compatibility with currently-existing LLVM
173 installations.</span> Only add new types if it is absolutely necessary.</p>
177 <!-- ======================================================================= -->
178 <div class="doc_subsection">
179 <a name="fund_type">Adding a fundamental type</a>
182 <div class="doc_text">
186 <li><tt>llvm/include/llvm/Type.def</tt>:
187 add enum for the type</li>
189 <li><tt>llvm/include/llvm/Type.h</tt>:
190 add ID number for the new type; add static <tt>Type*</tt> for this type</li>
192 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
193 add mapping from <tt>TypeID</tt> => <tt>Type*</tt>;
194 initialize the static <tt>Type*</tt></li>
196 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
197 add ability to parse in the type from text assembly</li>
199 <li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>:
200 add a token for that type</li>
206 <!-- ======================================================================= -->
207 <div class="doc_subsection">
208 <a name="derived_type">Adding a derived type</a>
211 <div class="doc_text">
214 <li><tt>llvm/include/llvm/Type.def</tt>:
215 add enum for the type</li>
217 <li><tt>llvm/include/llvm/Type.h</tt>:
218 add ID number for the new type; add a forward declaration of the type
221 <li><tt>llvm/include/llvm/DerivedType.h</tt>:
222 add new class to represent new class in the hierarchy; add forward
223 declaration to the TypeMap value type</li>
225 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
226 add support for derived type to:
227 <div class="doc_code">
229 std::string getTypeDescription(const Type &Ty,
230 std::vector<const Type*> &TypeStack)
231 bool TypesEqual(const Type *Ty, const Type *Ty2,
232 std::map<const Type*, const Type*> & EqTypes)
235 add necessary member functions for type, and factory methods</li>
237 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
238 add ability to parse in the type from text assembly</li>
240 <li><tt>llvm/lib/ByteCode/Writer/Writer.cpp</tt>:
241 modify <tt>void BytecodeWriter::outputType(const Type *T)</tt> to serialize
244 <li><tt>llvm/lib/ByteCode/Reader/Reader.cpp</tt>:
245 modify <tt>const Type *BytecodeReader::ParseType()</tt> to read your data
248 <li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>:
250 <div class="doc_code">
252 void calcTypeName(const Type *Ty,
253 std::vector<const Type*> &TypeStack,
254 std::map<const Type*,std::string> &TypeNames,
255 std::string & Result)
258 to output the new derived type
266 <!-- *********************************************************************** -->
270 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
271 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
272 <a href="http://validator.w3.org/check/referer"><img
273 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
275 <a href="http://misha.brukman.net">Misha Brukman</a><br>
276 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
278 Last modified: $Date$