Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 76bb6ae211b1d4d656c182a0b9e8a6e8877ee8d0 / . / docs / ExtendingLLVM.html
blob: b86d561794c727f307e1a1ffaf06a113e498fa5f [file] [log] [blame]
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>Extending LLVM: Adding instructions, intrinsics, types, etc.</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7</head>
8
9<body>
10
11<div class="doc_title">
12 Extending LLVM: Adding instructions, intrinsics, types, etc.
13</div>
14
15<ol>
16 <li><a href="#introduction">Introduction and Warning</a></li>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]17 <li><a href="#intrinsic">Adding a new intrinsic function</a></li>
Chris Lattner36365402004-04-09 19:24:20 +0000[diff] [blame]18 <li><a href="#instruction">Adding a new instruction</a></li>
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]19 <li><a href="#sdnode">Adding a new SelectionDAG node</a></li>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]20 <li><a href="#type">Adding a new type</a>
21 <ol>
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>
24 </ol></li>
25</ol>
26
Chris Lattner7911ce22004-05-23 21:07:27 +0000[diff] [blame]27<div class="doc_author">
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]28 <p>Written by <a href="http://misha.brukman.net">Misha Brukman</a>,
29 Brad Jones, and <a href="http://nondot.org/sabre">Chris Lattner</a></p>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]30</div>
31
32<!-- *********************************************************************** -->
33<div class="doc_section">
34 <a name="introduction">Introduction and Warning</a>
35</div>
36<!-- *********************************************************************** -->
37
38<div class="doc_text">
39
40<p>During the course of using LLVM, you may wish to customize it for your
41research project or for experimentation. At this point, you may realize that
42you need to add something to LLVM, whether it be a new fundamental type, a new
43intrinsic function, or a whole new instruction.</p>
44
45<p>When you come to this realization, stop and think. Do you really need to
46extend LLVM? Is it a new fundamental capability that LLVM does not support at
47its current incarnation or can it be synthesized from already pre-existing LLVM
48elements? If you are not sure, ask on the <a
49href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM-dev</a> list. The
50reason is that extending LLVM will get involved as you need to update all the
51different passes that you intend to use with your extension, and there are
52<em>many</em> LLVM analyses and transformations, so it may be quite a bit of
53work.</p>
54
Misha Brukmanb3b28272004-04-06 04:17:51 +0000[diff] [blame]55<p>Adding an <a href="#intrinsic">intrinsic function</a> is easier than adding
56an instruction, and is transparent to optimization passes which treat it as an
57unanalyzable function. If your added functionality can be expressed as a
58function call, an intrinsic function is the method of choice for LLVM
59extension.</p>
60
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]61<p>Before you invest a significant amount of effort into a non-trivial
62extension, <span class="doc_warning">ask on the list</span> if what you are
63looking to do can be done with already-existing infrastructure, or if maybe
64someone else is already working on it. You will save yourself a lot of time and
65effort by doing so.</p>
66
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]67</div>
68
69<!-- *********************************************************************** -->
70<div class="doc_section">
Chris Lattner36365402004-04-09 19:24:20 +0000[diff] [blame]71 <a name="intrinsic">Adding a new intrinsic function</a>
72</div>
73<!-- *********************************************************************** -->
74
75<div class="doc_text">
76
77<p>Adding a new intrinsic function to LLVM is much easier than adding a new
78instruction. Almost all extensions to LLVM should start as an intrinsic
79function and then be turned into an instruction if warranted.</p>
80
81<ol>
82<li><tt>llvm/docs/LangRef.html</tt>:
83 Document the intrinsic. Decide whether it is code generator specific and
84 what the restrictions are. Talk to other people about it so that you are
85 sure it's a good idea.</li>
86
87<li><tt>llvm/include/llvm/Intrinsics.h</tt>:
88 add an enum in the <tt>llvm::Intrinsic</tt> namespace</li>
89
Chris Lattner36365402004-04-09 19:24:20 +0000[diff] [blame]90<li><tt>llvm/lib/VMCore/Verifier.cpp</tt>:
91 Add code to check the invariants of the intrinsic are respected.</li>
92
93<li><tt>llvm/lib/VMCore/Function.cpp (<tt>Function::getIntrinsicID()</tt>)</tt>:
94 Identify the new intrinsic function, returning the enum for the intrinsic
95 that you added.</li>
Chris Lattner0190fdb2004-04-10 06:56:53 +0000[diff] [blame]96
97<li><tt>llvm/lib/Analysis/BasicAliasAnalysis.cpp</tt>: If the new intrinsic does
Chris Lattner81519d92004-06-20 07:53:22 +0000[diff] [blame]98 not access memory or does not write to memory, add it to the relevant list
Chris Lattner0190fdb2004-04-10 06:56:53 +0000[diff] [blame]99 of functions.</li>
100
Nate Begeman2f86c222006-01-14 01:27:10 +0000[diff] [blame]101<li><tt>llvm/lib/Analysis/ConstantFolding.cpp</tt>: If it is possible to
102 constant fold your intrinsic, add support to it in the
103 <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li>
104
105<li><tt>llvm/lib/Transforms/Utils/Local.cpp</tt>: If your intrinsic has no side-
106 effects, add it to the list of intrinsics in the
107 <tt>isInstructionTriviallyDead</tt> function.</li>
Chris Lattnerd828bc62004-04-13 19:48:55 +0000[diff] [blame]108
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]109<li><tt>llvm/test/Regression/*</tt>: Add test cases for your test cases to the
110 test suite</li>
Chris Lattner36365402004-04-09 19:24:20 +0000[diff] [blame]111</ol>
112
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]113<p>Once the intrinsic has been added to the system, you must add code generator
114support for it. Generally you must do the following steps:</p>
115
116<dl>
117<dt>Add support to the C backend in <tt>lib/Target/CBackend/</tt></dt>
118
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]119<dd>Depending on the intrinsic, there are a few ways to implement this. For
120most intrinsics, it makes sense to add code to lower your intrinsic in
121<tt>LowerIntrinsicCall</tt> in <tt>lib/CodeGen/IntrinsicLowering.cpp</tt>.
122Second, if it makes sense to lower the intrinsic to an expanded sequence of C
123code in all cases, just emit the expansion in <tt>visitCallInst</tt> in
124<tt>Writer.cpp</tt>. If the intrinsic has some way to express it with GCC
125(or any other compiler) extensions, it can be conditionally supported based on
126the compiler compiling the CBE output (see llvm.prefetch for an example).
127Third, if the intrinsic really has no way to be lowered, just have the code
128generator emit code that prints an error message and calls abort if executed.
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]129</dd>
130
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]131<dl>
132<dt>Add support to the SelectionDAG Instruction Selector in
133<tt>lib/CodeGen/SelectionDAG/</tt></dt>
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]134
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]135<dd>Since most targets in LLVM use the SelectionDAG framework for generating
136code, you will likely need to add support for your intrinsic there as well.
137This is usually accomplished by adding a new node, and then teaching the
138SelectionDAG code how to handle that node. To do this, follow the steps in
139the next section, Adding a new SelectionDAG node.</dd>
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]140
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]141<dl>
142<dt>Once you have added the new node, add code to
143<tt>SelectionDAG/SelectionDAGISel.cpp</tt> to recognize the intrinsic. In most
144cases, the intrinsic will just be turned into the node you just added. For an
145example of this, see how <tt>visitIntrinsicCall</tt> handles Intrinsic::ctpop
146</dt>
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]147
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]148</div>
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]149
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]150<!-- *********************************************************************** -->
151<div class="doc_section">
152 <a name="sdnode">Adding a new SelectionDAG node</a>
153</div>
154<!-- *********************************************************************** -->
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]155
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]156<div class="doc_text">
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]157
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]158<p>As with intrinsics, adding a new SelectionDAG node to LLVM is much easier
159than adding a new instruction. New nodes are often added to help represent
160instructions common to many targets. These nodes often map to an LLVM
161instruction (add, sub) or intrinsic (byteswap, population count). In other
162cases, new nodes have been added to allow many targets to perform a common task
163(converting between floating point and integer representation) or capture more
164complicated behavior in a single node (rotate).</p>
Chris Lattner5eb9f0d2005-05-11 03:53:53 +0000[diff] [blame]165
Nate Begeman099d76c2006-01-16 07:54:23 +0000[diff] [blame]166<ol>
167<li><tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>:
168 Add an enum value for the new SelectionDAG node.</li>
169<li><tt>lib/CodeGen/SelectionDAG/SelectionDAG.cpp</tt>:
170 Add code to print the node to <tt>getOperationName</tt>. If your new node
171 can be evaluated at compile time when given constant arguments (such as an
172 add of a constant with another constant), find the <tt>getNode</tt> method
173 that takes the appropriate number of arguments, and add a case for your node
174 to the switch statement that performs constant folding for nodes that take
175 the same number of arguments as your new node.</li>
176<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
177 Add code to <a href="CodeGenerator.html#selectiondag_legalize">legalize,
178 promote, and expand</a> the node as necessary. At a minimum, you will need
179 to add a case statement for your node in <tt>LegalizeOp</tt> which calls
180 LegalizeOp on the node's operands, and returns a new node if any of the
181 operands changed as a result of being legalized. It is likely that not all
182 targets supported by the SelectionDAG framework will natively support the
183 new node. In this case, you must also add code in your node's case
184 statement in <tt>LegalizeOp</tt> to Expand your node into simpler, legal
185 operations. The case for ISD::UREM for expanding a remainder into a
186 multiply and a subtract is a good example.</li>
187<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
188 If targets may support the new node being added only at certain sizes, you
189 will also need to add code to your node's case statement in
190 <tt>LegalizeOp</tt> to Promote your node's operands to a larger size, and
191 perform the correct operation. You will also need to add code to
192 <tt>PromoteOp</tt> to do this as well. For a good example, see ISD::BSWAP,
193 which promotes its operand to a wider size, performs the byteswap, and then
194 shifts the correct bytes right to emulate the narrower byteswap in the
195 wider type.</li>
196<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
197 Add a case for your node in <tt>ExpandOp</tt> to teach the legalizer how to
198 perform the action represented by the new node on a value that has been
199 split into high and low halves. This case will be used to support your
200 node with a 64 bit operand on a 32 bit target.</li>
201<li><tt>lib/CodeGen/SelectionDAG/DAGCombiner.cpp</tt>:
202 If your node can be combined with itself, or other existing nodes in a
203 peephole-like fashion, add a visit function for it, and call that function
204 from <tt></tt>. There are several good examples for simple combines you
205 can do; <tt>visitFABS</tt> and <tt>visitSRL</tt> are good starting places.
206 </li>
207<li><tt>lib/Target/PowerPC/PPCISelLowering.cpp</tt>:
208 Each target has an implementation of the <tt>TargetLowering</tt> class,
209 usually in its own file (although some targets include it in the same
210 file as the DAGToDAGISel). The default behavior for a target is to
211 assume that your new node is legal for all types that are legal for
212 that target. If this target does not natively support your node, then
213 tell the target to either Promote it (if it is supported at a larger
214 type) or Expand it. This will cause the code you wrote in
215 <tt>LegalizeOp</tt> above to decompose your new node into other legal
216 nodes for this target.</li>
217<li><tt>lib/Target/TargetSelectionDAG.td</tt>:
218 Most current targets supported by LLVM generate code using the DAGToDAG
219 method, where SelectionDAG nodes are pattern matched to target-specific
220 nodes, which represent individual instructions. In order for the targets
221 to match an instruction to your new node, you must add a def for that node
222 to the list in this file, with the appropriate type constraints. Look at
223 <tt>add</tt>, <tt>bswap</tt>, and <tt>fadd</tt> for examples.</li>
224<li><tt>lib/Target/PowerPC/PPCInstrInfo.td</tt>:
225 Each target has a tablegen file that describes the target's instruction
226 set. For targets that use the DAGToDAG instruction selection framework,
227 add a pattern for your new node that uses one or more target nodes.
228 Documentation for this is a bit sparse right now, but there are several
229 decent examples. See the patterns for <tt>rotl</tt> in
230 <tt>PPCInstrInfo.td</tt>.</li>
231<li>TODO: document complex patterns.</li>
232<li><tt>llvm/test/Regression/CodeGen/*</tt>: Add test cases for your new node
233 to the test suite. <tt>llvm/test/Regression/CodeGen/X86/bswap.ll</tt> is
234 a good example.</li>
235</ol>
Chris Lattner36365402004-04-09 19:24:20 +0000[diff] [blame]236
237</div>
238
239<!-- *********************************************************************** -->
240<div class="doc_section">
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]241 <a name="instruction">Adding a new instruction</a>
242</div>
243<!-- *********************************************************************** -->
244
245<div class="doc_text">
246
247<p><span class="doc_warning">WARNING: adding instructions changes the bytecode
Misha Brukmanb3b28272004-04-06 04:17:51 +0000[diff] [blame]248format, and it will take some effort to maintain compatibility with
249the previous version.</span> Only add an instruction if it is absolutely
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]250necessary.</p>
251
252<ol>
Misha Brukmanb3b28272004-04-06 04:17:51 +0000[diff] [blame]253
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]254<li><tt>llvm/include/llvm/Instruction.def</tt>:
255 add a number for your instruction and an enum name</li>
256
Misha Brukman47b14a42004-07-29 17:30:56 +0000[diff] [blame]257<li><tt>llvm/include/llvm/Instructions.h</tt>:
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]258 add a definition for the class that will represent your instruction</li>
259
260<li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>:
261 add a prototype for a visitor to your new instruction type</li>
262
263<li><tt>llvm/lib/AsmParser/Lexer.l</tt>:
264 add a new token to parse your instruction from assembly text file</li>
265
266<li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>:
267 add the grammar on how your instruction can be read and what it will
268 construct as a result</li>
269
Misha Brukmane39cd632004-09-28 16:58:12 +0000[diff] [blame]270<li><tt>llvm/lib/Bytecode/Reader/Reader.cpp</tt>:
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]271 add a case for your instruction and how it will be parsed from bytecode</li>
272
273<li><tt>llvm/lib/VMCore/Instruction.cpp</tt>:
274 add a case for how your instruction will be printed out to assembly</li>
275
Chris Lattner8f363212004-07-29 17:31:57 +0000[diff] [blame]276<li><tt>llvm/lib/VMCore/Instructions.cpp</tt>:
Misha Brukmane39cd632004-09-28 16:58:12 +0000[diff] [blame]277 implement the class you defined in
278 <tt>llvm/include/llvm/Instructions.h</tt></li>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]279
Misha Brukmana4242282004-12-01 20:58:54 +0000[diff] [blame]280<li>Test your instruction</li>
281
282<li><tt>llvm/lib/Target/*</tt>:
283 Add support for your instruction to code generators, or add a lowering
284 pass.</li>
285
286<li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li>
287
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]288</ol>
289
290<p>Also, you need to implement (or modify) any analyses or passes that you want
291to understand this new instruction.</p>
292
293</div>
294
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]295
296<!-- *********************************************************************** -->
297<div class="doc_section">
298 <a name="type">Adding a new type</a>
299</div>
300<!-- *********************************************************************** -->
301
302<div class="doc_text">
303
304<p><span class="doc_warning">WARNING: adding new types changes the bytecode
305format, and will break compatibility with currently-existing LLVM
306installations.</span> Only add new types if it is absolutely necessary.</p>
307
308</div>
309
310<!-- ======================================================================= -->
311<div class="doc_subsection">
312 <a name="fund_type">Adding a fundamental type</a>
313</div>
314
315<div class="doc_text">
316
317<ol>
318
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]319<li><tt>llvm/include/llvm/Type.h</tt>:
Chris Lattner55f95012005-04-23 21:59:11 +0000[diff] [blame]320 add enum for the new type; add static <tt>Type*</tt> for this type</li>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]321
322<li><tt>llvm/lib/VMCore/Type.cpp</tt>:
323 add mapping from <tt>TypeID</tt> =&gt; <tt>Type*</tt>;
324 initialize the static <tt>Type*</tt></li>
325
326<li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
327 add ability to parse in the type from text assembly</li>
328
329<li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>:
330 add a token for that type</li>
331
332</ol>
333
334</div>
335
336<!-- ======================================================================= -->
337<div class="doc_subsection">
338 <a name="derived_type">Adding a derived type</a>
339</div>
340
341<div class="doc_text">
342
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]343<ol>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]344<li><tt>llvm/include/llvm/Type.h</tt>:
Chris Lattner55f95012005-04-23 21:59:11 +0000[diff] [blame]345 add enum for the new type; add a forward declaration of the type
Misha Brukman7cc8a892004-08-12 19:58:43 +0000[diff] [blame]346 also</li>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]347
Chris Lattner47746aa2005-11-13 02:09:55 +0000[diff] [blame]348<li><tt>llvm/include/llvm/DerivedTypes.h</tt>:
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]349 add new class to represent new class in the hierarchy; add forward
350 declaration to the TypeMap value type</li>
351
352<li><tt>llvm/lib/VMCore/Type.cpp</tt>:
353 add support for derived type to:
Misha Brukman7cc8a892004-08-12 19:58:43 +0000[diff] [blame]354<div class="doc_code">
355<pre>
356std::string getTypeDescription(const Type &amp;Ty,
357 std::vector&lt;const Type*&gt; &amp;TypeStack)
358bool TypesEqual(const Type *Ty, const Type *Ty2,
359 std::map&lt;const Type*, const Type*&gt; &amp; EqTypes)
360</pre>
361</div>
362 add necessary member functions for type, and factory methods</li>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]363
364<li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
365 add ability to parse in the type from text assembly</li>
366
367<li><tt>llvm/lib/ByteCode/Writer/Writer.cpp</tt>:
Misha Brukman7cc8a892004-08-12 19:58:43 +0000[diff] [blame]368 modify <tt>void BytecodeWriter::outputType(const Type *T)</tt> to serialize
369 your type</li>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]370
371<li><tt>llvm/lib/ByteCode/Reader/Reader.cpp</tt>:
Misha Brukman7cc8a892004-08-12 19:58:43 +0000[diff] [blame]372 modify <tt>const Type *BytecodeReader::ParseType()</tt> to read your data
373 type</li>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]374
375<li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>:
Misha Brukman7cc8a892004-08-12 19:58:43 +0000[diff] [blame]376 modify
377<div class="doc_code">
378<pre>
379void calcTypeName(const Type *Ty,
380 std::vector&lt;const Type*&gt; &amp;TypeStack,
381 std::map&lt;const Type*,std::string&gt; &amp;TypeNames,
382 std::string &amp; Result)
383</pre>
384</div>
Chris Lattner8dad40c2004-08-12 19:06:24 +0000[diff] [blame]385 to output the new derived type
386</li>
387
388
389</ol>
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]390
391</div>
392
393<!-- *********************************************************************** -->
394
395<hr>
396<address>
397 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
398 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
399 <a href="http://validator.w3.org/check/referer"><img
400 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
401
Misha Brukmana3ce4292004-04-06 03:53:49 +0000[diff] [blame]402 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
403 <br>
404 Last modified: $Date$
405</address>
406
407</body>
408</html>