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.td</tt>:
89 Add an entry for your intrinsic.</li>
91 <li><tt>llvm/lib/Analysis/ConstantFolding.cpp</tt>: If it is possible to
92 constant fold your intrinsic, add support to it in the
93 <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li>
95 <li><tt>llvm/lib/Transforms/Utils/Local.cpp</tt>: If your intrinsic has no side-
96 effects, add it to the list of intrinsics in the
97 <tt>isInstructionTriviallyDead</tt> function.</li>
99 <li><tt>llvm/test/Regression/*</tt>: Add test cases for your test cases to the
103 <p>Once the intrinsic has been added to the system, you must add code generator
104 support for it. Generally you must do the following steps:</p>
107 <dt>Add support to the C backend in <tt>lib/Target/CBackend/</tt></dt>
109 <dd>Depending on the intrinsic, there are a few ways to implement this. For
110 most intrinsics, it makes sense to add code to lower your intrinsic in
111 <tt>LowerIntrinsicCall</tt> in <tt>lib/CodeGen/IntrinsicLowering.cpp</tt>.
112 Second, if it makes sense to lower the intrinsic to an expanded sequence of C
113 code in all cases, just emit the expansion in <tt>visitCallInst</tt> in
114 <tt>Writer.cpp</tt>. If the intrinsic has some way to express it with GCC
115 (or any other compiler) extensions, it can be conditionally supported based on
116 the compiler compiling the CBE output (see <tt>llvm.prefetch</tt> for an
118 Third, if the intrinsic really has no way to be lowered, just have the code
119 generator emit code that prints an error message and calls abort if executed.
123 <dt>Add support to the SelectionDAG Instruction Selector in
124 <tt>lib/CodeGen/SelectionDAG/</tt></dt>
126 <dd>Since most targets in LLVM use the SelectionDAG framework for generating
127 code, you will likely need to add support for your intrinsic there as well.
128 This is usually accomplished by adding a new node, and then teaching the
129 SelectionDAG code how to handle that node. To do this, follow the steps in
130 the <a href="#sdnode">Adding a new SelectionDAG node</a> section.</dd>
133 <dt>Once you have added the new node, add code to
134 <tt>SelectionDAG/SelectionDAGISel.cpp</tt> to recognize the intrinsic. In most
135 cases, the intrinsic will just be turned into the node you just added. For an
136 example of this, see how <tt>visitIntrinsicCall</tt> handles
137 <tt>Intrinsic::ctpop_*</tt>.
142 <!-- *********************************************************************** -->
143 <div class="doc_section">
144 <a name="sdnode">Adding a new SelectionDAG node</a>
146 <!-- *********************************************************************** -->
148 <div class="doc_text">
150 <p>As with intrinsics, adding a new SelectionDAG node to LLVM is much easier
151 than adding a new instruction. New nodes are often added to help represent
152 instructions common to many targets. These nodes often map to an LLVM
153 instruction (add, sub) or intrinsic (byteswap, population count). In other
154 cases, new nodes have been added to allow many targets to perform a common task
155 (converting between floating point and integer representation) or capture more
156 complicated behavior in a single node (rotate).</p>
159 <li><tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>:
160 Add an enum value for the new SelectionDAG node.</li>
161 <li><tt>lib/CodeGen/SelectionDAG/SelectionDAG.cpp</tt>:
162 Add code to print the node to <tt>getOperationName</tt>. If your new node
163 can be evaluated at compile time when given constant arguments (such as an
164 add of a constant with another constant), find the <tt>getNode</tt> method
165 that takes the appropriate number of arguments, and add a case for your node
166 to the switch statement that performs constant folding for nodes that take
167 the same number of arguments as your new node.</li>
168 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
169 Add code to <a href="CodeGenerator.html#selectiondag_legalize">legalize,
170 promote, and expand</a> the node as necessary. At a minimum, you will need
171 to add a case statement for your node in <tt>LegalizeOp</tt> which calls
172 LegalizeOp on the node's operands, and returns a new node if any of the
173 operands changed as a result of being legalized. It is likely that not all
174 targets supported by the SelectionDAG framework will natively support the
175 new node. In this case, you must also add code in your node's case
176 statement in <tt>LegalizeOp</tt> to Expand your node into simpler, legal
177 operations. The case for <tt>ISD::UREM</tt> for expanding a remainder into
178 a divide, multiply, and a subtract is a good example.</li>
179 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
180 If targets may support the new node being added only at certain sizes, you
181 will also need to add code to your node's case statement in
182 <tt>LegalizeOp</tt> to Promote your node's operands to a larger size, and
183 perform the correct operation. You will also need to add code to
184 <tt>PromoteOp</tt> to do this as well. For a good example, see
186 which promotes its operand to a wider size, performs the byteswap, and then
187 shifts the correct bytes right to emulate the narrower byteswap in the
189 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
190 Add a case for your node in <tt>ExpandOp</tt> to teach the legalizer how to
191 perform the action represented by the new node on a value that has been
192 split into high and low halves. This case will be used to support your
193 node with a 64 bit operand on a 32 bit target.</li>
194 <li><tt>lib/CodeGen/SelectionDAG/DAGCombiner.cpp</tt>:
195 If your node can be combined with itself, or other existing nodes in a
196 peephole-like fashion, add a visit function for it, and call that function
197 from <tt></tt>. There are several good examples for simple combines you
198 can do; <tt>visitFABS</tt> and <tt>visitSRL</tt> are good starting places.
200 <li><tt>lib/Target/PowerPC/PPCISelLowering.cpp</tt>:
201 Each target has an implementation of the <tt>TargetLowering</tt> class,
202 usually in its own file (although some targets include it in the same
203 file as the DAGToDAGISel). The default behavior for a target is to
204 assume that your new node is legal for all types that are legal for
205 that target. If this target does not natively support your node, then
206 tell the target to either Promote it (if it is supported at a larger
207 type) or Expand it. This will cause the code you wrote in
208 <tt>LegalizeOp</tt> above to decompose your new node into other legal
209 nodes for this target.</li>
210 <li><tt>lib/Target/TargetSelectionDAG.td</tt>:
211 Most current targets supported by LLVM generate code using the DAGToDAG
212 method, where SelectionDAG nodes are pattern matched to target-specific
213 nodes, which represent individual instructions. In order for the targets
214 to match an instruction to your new node, you must add a def for that node
215 to the list in this file, with the appropriate type constraints. Look at
216 <tt>add</tt>, <tt>bswap</tt>, and <tt>fadd</tt> for examples.</li>
217 <li><tt>lib/Target/PowerPC/PPCInstrInfo.td</tt>:
218 Each target has a tablegen file that describes the target's instruction
219 set. For targets that use the DAGToDAG instruction selection framework,
220 add a pattern for your new node that uses one or more target nodes.
221 Documentation for this is a bit sparse right now, but there are several
222 decent examples. See the patterns for <tt>rotl</tt> in
223 <tt>PPCInstrInfo.td</tt>.</li>
224 <li>TODO: document complex patterns.</li>
225 <li><tt>llvm/test/Regression/CodeGen/*</tt>: Add test cases for your new node
226 to the test suite. <tt>llvm/test/Regression/CodeGen/X86/bswap.ll</tt> is
232 <!-- *********************************************************************** -->
233 <div class="doc_section">
234 <a name="instruction">Adding a new instruction</a>
236 <!-- *********************************************************************** -->
238 <div class="doc_text">
240 <p><span class="doc_warning">WARNING: adding instructions changes the bytecode
241 format, and it will take some effort to maintain compatibility with
242 the previous version.</span> Only add an instruction if it is absolutely
247 <li><tt>llvm/include/llvm/Instruction.def</tt>:
248 add a number for your instruction and an enum name</li>
250 <li><tt>llvm/include/llvm/Instructions.h</tt>:
251 add a definition for the class that will represent your instruction</li>
253 <li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>:
254 add a prototype for a visitor to your new instruction type</li>
256 <li><tt>llvm/lib/AsmParser/Lexer.l</tt>:
257 add a new token to parse your instruction from assembly text file</li>
259 <li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>:
260 add the grammar on how your instruction can be read and what it will
261 construct as a result</li>
263 <li><tt>llvm/lib/Bytecode/Reader/Reader.cpp</tt>:
264 add a case for your instruction and how it will be parsed from bytecode</li>
266 <li><tt>llvm/lib/VMCore/Instruction.cpp</tt>:
267 add a case for how your instruction will be printed out to assembly</li>
269 <li><tt>llvm/lib/VMCore/Instructions.cpp</tt>:
270 implement the class you defined in
271 <tt>llvm/include/llvm/Instructions.h</tt></li>
273 <li>Test your instruction</li>
275 <li><tt>llvm/lib/Target/*</tt>:
276 Add support for your instruction to code generators, or add a lowering
279 <li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li>
283 <p>Also, you need to implement (or modify) any analyses or passes that you want
284 to understand this new instruction.</p>
289 <!-- *********************************************************************** -->
290 <div class="doc_section">
291 <a name="type">Adding a new type</a>
293 <!-- *********************************************************************** -->
295 <div class="doc_text">
297 <p><span class="doc_warning">WARNING: adding new types changes the bytecode
298 format, and will break compatibility with currently-existing LLVM
299 installations.</span> Only add new types if it is absolutely necessary.</p>
303 <!-- ======================================================================= -->
304 <div class="doc_subsection">
305 <a name="fund_type">Adding a fundamental type</a>
308 <div class="doc_text">
312 <li><tt>llvm/include/llvm/Type.h</tt>:
313 add enum for the new type; add static <tt>Type*</tt> for this type</li>
315 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
316 add mapping from <tt>TypeID</tt> => <tt>Type*</tt>;
317 initialize the static <tt>Type*</tt></li>
319 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
320 add ability to parse in the type from text assembly</li>
322 <li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>:
323 add a token for that type</li>
329 <!-- ======================================================================= -->
330 <div class="doc_subsection">
331 <a name="derived_type">Adding a derived type</a>
334 <div class="doc_text">
337 <li><tt>llvm/include/llvm/Type.h</tt>:
338 add enum for the new type; add a forward declaration of the type
341 <li><tt>llvm/include/llvm/DerivedTypes.h</tt>:
342 add new class to represent new class in the hierarchy; add forward
343 declaration to the TypeMap value type</li>
345 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
346 add support for derived type to:
347 <div class="doc_code">
349 std::string getTypeDescription(const Type &Ty,
350 std::vector<const Type*> &TypeStack)
351 bool TypesEqual(const Type *Ty, const Type *Ty2,
352 std::map<const Type*, const Type*> & EqTypes)
355 add necessary member functions for type, and factory methods</li>
357 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
358 add ability to parse in the type from text assembly</li>
360 <li><tt>llvm/lib/ByteCode/Writer/Writer.cpp</tt>:
361 modify <tt>void BytecodeWriter::outputType(const Type *T)</tt> to serialize
364 <li><tt>llvm/lib/ByteCode/Reader/Reader.cpp</tt>:
365 modify <tt>const Type *BytecodeReader::ParseType()</tt> to read your data
368 <li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>:
370 <div class="doc_code">
372 void calcTypeName(const Type *Ty,
373 std::vector<const Type*> &TypeStack,
374 std::map<const Type*,std::string> &TypeNames,
375 std::string & Result)
378 to output the new derived type
386 <!-- *********************************************************************** -->
390 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
391 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
392 <a href="http://validator.w3.org/check/referer"><img
393 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
395 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
397 Last modified: $Date$