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="#sdnode">Adding a new SelectionDAG node</a></li>
20 <li><a href="#type">Adding a new type</a>
22 <li><a href="#fund_type">Adding a new fundamental type</a></li>
23 <li><a href="#derived_type">Adding a new derived type</a></li>
27 <div class="doc_author">
28 <p>Written by <a href="http://misha.brukman.net">Misha Brukman</a>,
29 Brad Jones, Nate Begeman,
30 and <a href="http://nondot.org/sabre">Chris Lattner</a></p>
33 <!-- *********************************************************************** -->
34 <div class="doc_section">
35 <a name="introduction">Introduction and Warning</a>
37 <!-- *********************************************************************** -->
39 <div class="doc_text">
41 <p>During the course of using LLVM, you may wish to customize it for your
42 research project or for experimentation. At this point, you may realize that
43 you need to add something to LLVM, whether it be a new fundamental type, a new
44 intrinsic function, or a whole new instruction.</p>
46 <p>When you come to this realization, stop and think. Do you really need to
47 extend LLVM? Is it a new fundamental capability that LLVM does not support at
48 its current incarnation or can it be synthesized from already pre-existing LLVM
49 elements? If you are not sure, ask on the <a
50 href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM-dev</a> list. The
51 reason is that extending LLVM will get involved as you need to update all the
52 different passes that you intend to use with your extension, and there are
53 <em>many</em> LLVM analyses and transformations, so it may be quite a bit of
56 <p>Adding an <a href="#intrinsic">intrinsic function</a> is easier than adding
57 an instruction, and is transparent to optimization passes which treat it as an
58 unanalyzable function. If your added functionality can be expressed as a
59 function call, an intrinsic function is the method of choice for LLVM
62 <p>Before you invest a significant amount of effort into a non-trivial
63 extension, <span class="doc_warning">ask on the list</span> if what you are
64 looking to do can be done with already-existing infrastructure, or if maybe
65 someone else is already working on it. You will save yourself a lot of time and
66 effort by doing so.</p>
70 <!-- *********************************************************************** -->
71 <div class="doc_section">
72 <a name="intrinsic">Adding a new intrinsic function</a>
74 <!-- *********************************************************************** -->
76 <div class="doc_text">
78 <p>Adding a new intrinsic function to LLVM is much easier than adding a new
79 instruction. Almost all extensions to LLVM should start as an intrinsic
80 function and then be turned into an instruction if warranted.</p>
83 <li><tt>llvm/docs/LangRef.html</tt>:
84 Document the intrinsic. Decide whether it is code generator specific and
85 what the restrictions are. Talk to other people about it so that you are
86 sure it's a good idea.</li>
88 <li><tt>llvm/include/llvm/Intrinsics.h</tt>:
89 add an enum in the <tt>llvm::Intrinsic</tt> namespace</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/Analysis/ConstantFolding.cpp</tt>: If it is possible to
103 constant fold your intrinsic, add support to it in the
104 <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li>
106 <li><tt>llvm/lib/Transforms/Utils/Local.cpp</tt>: If your intrinsic has no side-
107 effects, add it to the list of intrinsics in the
108 <tt>isInstructionTriviallyDead</tt> function.</li>
110 <li><tt>llvm/test/Regression/*</tt>: Add test cases for your test cases to the
114 <p>Once the intrinsic has been added to the system, you must add code generator
115 support for it. Generally you must do the following steps:</p>
118 <dt>Add support to the C backend in <tt>lib/Target/CBackend/</tt></dt>
120 <dd>Depending on the intrinsic, there are a few ways to implement this. For
121 most intrinsics, it makes sense to add code to lower your intrinsic in
122 <tt>LowerIntrinsicCall</tt> in <tt>lib/CodeGen/IntrinsicLowering.cpp</tt>.
123 Second, if it makes sense to lower the intrinsic to an expanded sequence of C
124 code in all cases, just emit the expansion in <tt>visitCallInst</tt> in
125 <tt>Writer.cpp</tt>. If the intrinsic has some way to express it with GCC
126 (or any other compiler) extensions, it can be conditionally supported based on
127 the compiler compiling the CBE output (see <tt>llvm.prefetch</tt> for an
129 Third, if the intrinsic really has no way to be lowered, just have the code
130 generator emit code that prints an error message and calls abort if executed.
134 <dt>Add support to the SelectionDAG Instruction Selector in
135 <tt>lib/CodeGen/SelectionDAG/</tt></dt>
137 <dd>Since most targets in LLVM use the SelectionDAG framework for generating
138 code, you will likely need to add support for your intrinsic there as well.
139 This is usually accomplished by adding a new node, and then teaching the
140 SelectionDAG code how to handle that node. To do this, follow the steps in
141 the <a href="#sdnode">Adding a new SelectionDAG node</a> section.</dd>
144 <dt>Once you have added the new node, add code to
145 <tt>SelectionDAG/SelectionDAGISel.cpp</tt> to recognize the intrinsic. In most
146 cases, the intrinsic will just be turned into the node you just added. For an
147 example of this, see how <tt>visitIntrinsicCall</tt> handles
148 <tt>Intrinsic::ctpop_*</tt>.
153 <!-- *********************************************************************** -->
154 <div class="doc_section">
155 <a name="sdnode">Adding a new SelectionDAG node</a>
157 <!-- *********************************************************************** -->
159 <div class="doc_text">
161 <p>As with intrinsics, adding a new SelectionDAG node to LLVM is much easier
162 than adding a new instruction. New nodes are often added to help represent
163 instructions common to many targets. These nodes often map to an LLVM
164 instruction (add, sub) or intrinsic (byteswap, population count). In other
165 cases, new nodes have been added to allow many targets to perform a common task
166 (converting between floating point and integer representation) or capture more
167 complicated behavior in a single node (rotate).</p>
170 <li><tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>:
171 Add an enum value for the new SelectionDAG node.</li>
172 <li><tt>lib/CodeGen/SelectionDAG/SelectionDAG.cpp</tt>:
173 Add code to print the node to <tt>getOperationName</tt>. If your new node
174 can be evaluated at compile time when given constant arguments (such as an
175 add of a constant with another constant), find the <tt>getNode</tt> method
176 that takes the appropriate number of arguments, and add a case for your node
177 to the switch statement that performs constant folding for nodes that take
178 the same number of arguments as your new node.</li>
179 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
180 Add code to <a href="CodeGenerator.html#selectiondag_legalize">legalize,
181 promote, and expand</a> the node as necessary. At a minimum, you will need
182 to add a case statement for your node in <tt>LegalizeOp</tt> which calls
183 LegalizeOp on the node's operands, and returns a new node if any of the
184 operands changed as a result of being legalized. It is likely that not all
185 targets supported by the SelectionDAG framework will natively support the
186 new node. In this case, you must also add code in your node's case
187 statement in <tt>LegalizeOp</tt> to Expand your node into simpler, legal
188 operations. The case for <tt>ISD::UREM</tt> for expanding a remainder into
189 a divide, multiply, and a subtract is a good example.</li>
190 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
191 If targets may support the new node being added only at certain sizes, you
192 will also need to add code to your node's case statement in
193 <tt>LegalizeOp</tt> to Promote your node's operands to a larger size, and
194 perform the correct operation. You will also need to add code to
195 <tt>PromoteOp</tt> to do this as well. For a good example, see
197 which promotes its operand to a wider size, performs the byteswap, and then
198 shifts the correct bytes right to emulate the narrower byteswap in the
200 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
201 Add a case for your node in <tt>ExpandOp</tt> to teach the legalizer how to
202 perform the action represented by the new node on a value that has been
203 split into high and low halves. This case will be used to support your
204 node with a 64 bit operand on a 32 bit target.</li>
205 <li><tt>lib/CodeGen/SelectionDAG/DAGCombiner.cpp</tt>:
206 If your node can be combined with itself, or other existing nodes in a
207 peephole-like fashion, add a visit function for it, and call that function
208 from <tt></tt>. There are several good examples for simple combines you
209 can do; <tt>visitFABS</tt> and <tt>visitSRL</tt> are good starting places.
211 <li><tt>lib/Target/PowerPC/PPCISelLowering.cpp</tt>:
212 Each target has an implementation of the <tt>TargetLowering</tt> class,
213 usually in its own file (although some targets include it in the same
214 file as the DAGToDAGISel). The default behavior for a target is to
215 assume that your new node is legal for all types that are legal for
216 that target. If this target does not natively support your node, then
217 tell the target to either Promote it (if it is supported at a larger
218 type) or Expand it. This will cause the code you wrote in
219 <tt>LegalizeOp</tt> above to decompose your new node into other legal
220 nodes for this target.</li>
221 <li><tt>lib/Target/TargetSelectionDAG.td</tt>:
222 Most current targets supported by LLVM generate code using the DAGToDAG
223 method, where SelectionDAG nodes are pattern matched to target-specific
224 nodes, which represent individual instructions. In order for the targets
225 to match an instruction to your new node, you must add a def for that node
226 to the list in this file, with the appropriate type constraints. Look at
227 <tt>add</tt>, <tt>bswap</tt>, and <tt>fadd</tt> for examples.</li>
228 <li><tt>lib/Target/PowerPC/PPCInstrInfo.td</tt>:
229 Each target has a tablegen file that describes the target's instruction
230 set. For targets that use the DAGToDAG instruction selection framework,
231 add a pattern for your new node that uses one or more target nodes.
232 Documentation for this is a bit sparse right now, but there are several
233 decent examples. See the patterns for <tt>rotl</tt> in
234 <tt>PPCInstrInfo.td</tt>.</li>
235 <li>TODO: document complex patterns.</li>
236 <li><tt>llvm/test/Regression/CodeGen/*</tt>: Add test cases for your new node
237 to the test suite. <tt>llvm/test/Regression/CodeGen/X86/bswap.ll</tt> is
243 <!-- *********************************************************************** -->
244 <div class="doc_section">
245 <a name="instruction">Adding a new instruction</a>
247 <!-- *********************************************************************** -->
249 <div class="doc_text">
251 <p><span class="doc_warning">WARNING: adding instructions changes the bytecode
252 format, and it will take some effort to maintain compatibility with
253 the previous version.</span> Only add an instruction if it is absolutely
258 <li><tt>llvm/include/llvm/Instruction.def</tt>:
259 add a number for your instruction and an enum name</li>
261 <li><tt>llvm/include/llvm/Instructions.h</tt>:
262 add a definition for the class that will represent your instruction</li>
264 <li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>:
265 add a prototype for a visitor to your new instruction type</li>
267 <li><tt>llvm/lib/AsmParser/Lexer.l</tt>:
268 add a new token to parse your instruction from assembly text file</li>
270 <li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>:
271 add the grammar on how your instruction can be read and what it will
272 construct as a result</li>
274 <li><tt>llvm/lib/Bytecode/Reader/Reader.cpp</tt>:
275 add a case for your instruction and how it will be parsed from bytecode</li>
277 <li><tt>llvm/lib/VMCore/Instruction.cpp</tt>:
278 add a case for how your instruction will be printed out to assembly</li>
280 <li><tt>llvm/lib/VMCore/Instructions.cpp</tt>:
281 implement the class you defined in
282 <tt>llvm/include/llvm/Instructions.h</tt></li>
284 <li>Test your instruction</li>
286 <li><tt>llvm/lib/Target/*</tt>:
287 Add support for your instruction to code generators, or add a lowering
290 <li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li>
294 <p>Also, you need to implement (or modify) any analyses or passes that you want
295 to understand this new instruction.</p>
300 <!-- *********************************************************************** -->
301 <div class="doc_section">
302 <a name="type">Adding a new type</a>
304 <!-- *********************************************************************** -->
306 <div class="doc_text">
308 <p><span class="doc_warning">WARNING: adding new types changes the bytecode
309 format, and will break compatibility with currently-existing LLVM
310 installations.</span> Only add new types if it is absolutely necessary.</p>
314 <!-- ======================================================================= -->
315 <div class="doc_subsection">
316 <a name="fund_type">Adding a fundamental type</a>
319 <div class="doc_text">
323 <li><tt>llvm/include/llvm/Type.h</tt>:
324 add enum for the new type; add static <tt>Type*</tt> for this type</li>
326 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
327 add mapping from <tt>TypeID</tt> => <tt>Type*</tt>;
328 initialize the static <tt>Type*</tt></li>
330 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
331 add ability to parse in the type from text assembly</li>
333 <li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>:
334 add a token for that type</li>
340 <!-- ======================================================================= -->
341 <div class="doc_subsection">
342 <a name="derived_type">Adding a derived type</a>
345 <div class="doc_text">
348 <li><tt>llvm/include/llvm/Type.h</tt>:
349 add enum for the new type; add a forward declaration of the type
352 <li><tt>llvm/include/llvm/DerivedTypes.h</tt>:
353 add new class to represent new class in the hierarchy; add forward
354 declaration to the TypeMap value type</li>
356 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
357 add support for derived type to:
358 <div class="doc_code">
360 std::string getTypeDescription(const Type &Ty,
361 std::vector<const Type*> &TypeStack)
362 bool TypesEqual(const Type *Ty, const Type *Ty2,
363 std::map<const Type*, const Type*> & EqTypes)
366 add necessary member functions for type, and factory methods</li>
368 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
369 add ability to parse in the type from text assembly</li>
371 <li><tt>llvm/lib/ByteCode/Writer/Writer.cpp</tt>:
372 modify <tt>void BytecodeWriter::outputType(const Type *T)</tt> to serialize
375 <li><tt>llvm/lib/ByteCode/Reader/Reader.cpp</tt>:
376 modify <tt>const Type *BytecodeReader::ParseType()</tt> to read your data
379 <li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>:
381 <div class="doc_code">
383 void calcTypeName(const Type *Ty,
384 std::vector<const Type*> &TypeStack,
385 std::map<const Type*,std::string> &TypeNames,
386 std::string & Result)
389 to output the new derived type
397 <!-- *********************************************************************** -->
401 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
402 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
403 <a href="http://validator.w3.org/check/referer"><img
404 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
406 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
408 Last modified: $Date$