docs/ExtendingLLVM.html

   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>
  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>
  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
  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>
  31 </div>
  32
  33 <!-- *********************************************************************** -->
  34 <div class="doc_section">
  35   <a name="introduction">Introduction and Warning</a>
  36 </div>
  37 <!-- *********************************************************************** -->
  38
  39 <div class="doc_text">
  40
  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>
  45
  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
  54 work.</p>
  55
  56 <p>Adding an <a href="#intrinsic">intrinsic function</a> is far easier than
  57 adding an instruction, and is transparent to optimization passes.  If your added
  58 functionality can be expressed as a
  59 function call, an intrinsic function is the method of choice for LLVM
  60 extension.</p>
  61
  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>
  67
  68 </div>
  69
  70 <!-- *********************************************************************** -->
  71 <div class="doc_section">
  72   <a name="intrinsic">Adding a new intrinsic function</a>
  73 </div>
  74 <!-- *********************************************************************** -->
  75
  76 <div class="doc_text">
  77
  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>
  81
  82 <ol>
  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>
  87
  88 <li><tt>llvm/include/llvm/Intrinsics*.td</tt>:
  89     Add an entry for your intrinsic.  Describe its memory access characteristics
  90     for optimization (this controls whether it will be DCE'd, CSE'd, etc). Note
  91     that any intrinsic using the <tt>llvm_int_ty</tt> type for an argument will
  92     be deemed by <tt>tblgen</tt> as overloaded and the corresponding suffix
  93     will be required on the intrinsic's name.</li>
  94
  95 <li><tt>llvm/lib/Analysis/ConstantFolding.cpp</tt>: If it is possible to
  96     constant fold your intrinsic, add support to it in the
  97     <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li>
  98
  99 <li><tt>llvm/test/Regression/*</tt>: Add test cases for your test cases to the
 100     test suite</li>
 101 </ol>
 102
 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>
 105
 106 <dl>
 107 <dt>Add support to the C backend in <tt>lib/Target/CBackend/</tt></dt>
 108
 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
 113     C 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
 116     on the compiler compiling the CBE output (see <tt>llvm.prefetch</tt> for an
 117     example).  Third, if the intrinsic really has no way to be lowered, just
 118     have the code generator emit code that prints an error message and calls
 119     abort if executed.</dd>
 120
 121 <dt>Add support to the .td file for the target(s) of your choice in
 122    <tt>lib/Target/*/*.td</tt>.</dt>
 123
 124 <dd>This is usually a matter of adding a pattern to the .td file that matches
 125     the intrinsic, though it may obviously require adding the instructions you
 126     want to generate as well.  There are lots of examples in the PowerPC and X86
 127     backend to follow.</dd>
 128 </dl>
 129
 130 </div>
 131
 132 <!-- *********************************************************************** -->
 133 <div class="doc_section">
 134   <a name="sdnode">Adding a new SelectionDAG node</a>
 135 </div>
 136 <!-- *********************************************************************** -->
 137
 138 <div class="doc_text">
 139
 140 <p>As with intrinsics, adding a new SelectionDAG node to LLVM is much easier
 141 than adding a new instruction.  New nodes are often added to help represent
 142 instructions common to many targets.  These nodes often map to an LLVM
 143 instruction (add, sub) or intrinsic (byteswap, population count).  In other
 144 cases, new nodes have been added to allow many targets to perform a common task
 145 (converting between floating point and integer representation) or capture more
 146 complicated behavior in a single node (rotate).</p>
 147
 148 <ol>
 149 <li><tt>include/llvm/CodeGen/SelectionDAGNodes.h</tt>:
 150     Add an enum value for the new SelectionDAG node.</li>
 151 <li><tt>lib/CodeGen/SelectionDAG/SelectionDAG.cpp</tt>:
 152     Add code to print the node to <tt>getOperationName</tt>.  If your new node
 153     can be evaluated at compile time when given constant arguments (such as an
 154     add of a constant with another constant), find the <tt>getNode</tt> method
 155     that takes the appropriate number of arguments, and add a case for your node
 156     to the switch statement that performs constant folding for nodes that take
 157     the same number of arguments as your new node.</li>
 158 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
 159     Add code to <a href="CodeGenerator.html#selectiondag_legalize">legalize,
 160     promote, and expand</a> the node as necessary.  At a minimum, you will need
 161     to add a case statement for your node in <tt>LegalizeOp</tt> which calls
 162     LegalizeOp on the node's operands, and returns a new node if any of the
 163     operands changed as a result of being legalized.  It is likely that not all
 164     targets supported by the SelectionDAG framework will natively support the
 165     new node.  In this case, you must also add code in your node's case
 166     statement in <tt>LegalizeOp</tt> to Expand your node into simpler, legal
 167     operations.  The case for <tt>ISD::UREM</tt> for expanding a remainder into
 168     a divide, multiply, and a subtract is a good example.</li>
 169 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
 170     If targets may support the new node being added only at certain sizes, you
 171     will also need to add code to your node's case statement in
 172     <tt>LegalizeOp</tt> to Promote your node's operands to a larger size, and
 173     perform the correct operation.  You will also need to add code to
 174     <tt>PromoteOp</tt> to do this as well.  For a good example, see
 175     <tt>ISD::BSWAP</tt>,
 176     which promotes its operand to a wider size, performs the byteswap, and then
 177     shifts the correct bytes right to emulate the narrower byteswap in the
 178     wider type.</li>
 179 <li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>:
 180     Add a case for your node in <tt>ExpandOp</tt> to teach the legalizer how to
 181     perform the action represented by the new node on a value that has been
 182     split into high and low halves.  This case will be used to support your
 183     node with a 64 bit operand on a 32 bit target.</li>
 184 <li><tt>lib/CodeGen/SelectionDAG/DAGCombiner.cpp</tt>:
 185     If your node can be combined with itself, or other existing nodes in a
 186     peephole-like fashion, add a visit function for it, and call that function
 187     from <tt></tt>.  There are several good examples for simple combines you
 188     can do; <tt>visitFABS</tt> and <tt>visitSRL</tt> are good starting places.
 189     </li>
 190 <li><tt>lib/Target/PowerPC/PPCISelLowering.cpp</tt>:
 191     Each target has an implementation of the <tt>TargetLowering</tt> class,
 192     usually in its own file (although some targets include it in the same
 193     file as the DAGToDAGISel).  The default behavior for a target is to
 194     assume that your new node is legal for all types that are legal for
 195     that target.  If this target does not natively support your node, then
 196     tell the target to either Promote it (if it is supported at a larger
 197     type) or Expand it.  This will cause the code you wrote in
 198     <tt>LegalizeOp</tt> above to decompose your new node into other legal
 199     nodes for this target.</li>
 200 <li><tt>lib/Target/TargetSelectionDAG.td</tt>:
 201     Most current targets supported by LLVM generate code using the DAGToDAG
 202     method, where SelectionDAG nodes are pattern matched to target-specific
 203     nodes, which represent individual instructions.  In order for the targets
 204     to match an instruction to your new node, you must add a def for that node
 205     to the list in this file, with the appropriate type constraints. Look at
 206     <tt>add</tt>, <tt>bswap</tt>, and <tt>fadd</tt> for examples.</li>
 207 <li><tt>lib/Target/PowerPC/PPCInstrInfo.td</tt>:
 208     Each target has a tablegen file that describes the target's instruction
 209     set.  For targets that use the DAGToDAG instruction selection framework,
 210     add a pattern for your new node that uses one or more target nodes.
 211     Documentation for this is a bit sparse right now, but there are several
 212     decent examples.  See the patterns for <tt>rotl</tt> in
 213     <tt>PPCInstrInfo.td</tt>.</li>
 214 <li>TODO: document complex patterns.</li>
 215 <li><tt>llvm/test/Regression/CodeGen/*</tt>: Add test cases for your new node
 216     to the test suite.  <tt>llvm/test/Regression/CodeGen/X86/bswap.ll</tt> is
 217     a good example.</li>
 218 </ol>
 219
 220 </div>
 221
 222 <!-- *********************************************************************** -->
 223 <div class="doc_section">
 224   <a name="instruction">Adding a new instruction</a>
 225 </div>
 226 <!-- *********************************************************************** -->
 227
 228 <div class="doc_text">
 229
 230 <p><span class="doc_warning">WARNING: adding instructions changes the bitcode
 231 format, and it will take some effort to maintain compatibility with
 232 the previous version.</span> Only add an instruction if it is absolutely
 233 necessary.</p>
 234
 235 <ol>
 236
 237 <li><tt>llvm/include/llvm/Instruction.def</tt>:
 238     add a number for your instruction and an enum name</li>
 239
 240 <li><tt>llvm/include/llvm/Instructions.h</tt>:
 241     add a definition for the class that will represent your instruction</li>
 242
 243 <li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>:
 244     add a prototype for a visitor to your new instruction type</li>
 245
 246 <li><tt>llvm/lib/AsmParser/Lexer.l</tt>:
 247     add a new token to parse your instruction from assembly text file</li>
 248
 249 <li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>:
 250     add the grammar on how your instruction can be read and what it will
 251     construct as a result</li>
 252
 253 <li><tt>llvm/lib/Bitcode/Reader/Reader.cpp</tt>:
 254     add a case for your instruction and how it will be parsed from bitcode</li>
 255
 256 <li><tt>llvm/lib/VMCore/Instruction.cpp</tt>:
 257     add a case for how your instruction will be printed out to assembly</li>
 258
 259 <li><tt>llvm/lib/VMCore/Instructions.cpp</tt>:
 260     implement the class you defined in
 261     <tt>llvm/include/llvm/Instructions.h</tt></li>
 262
 263 <li>Test your instruction</li>
 264
 265 <li><tt>llvm/lib/Target/*</tt>:
 266     Add support for your instruction to code generators, or add a lowering
 267     pass.</li>
 268
 269 <li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li>
 270
 271 </ol>
 272
 273 <p>Also, you need to implement (or modify) any analyses or passes that you want
 274 to understand this new instruction.</p>
 275
 276 </div>
 277
 278
 279 <!-- *********************************************************************** -->
 280 <div class="doc_section">
 281   <a name="type">Adding a new type</a>
 282 </div>
 283 <!-- *********************************************************************** -->
 284
 285 <div class="doc_text">
 286
 287 <p><span class="doc_warning">WARNING: adding new types changes the bitcode
 288 format, and will break compatibility with currently-existing LLVM
 289 installations.</span> Only add new types if it is absolutely necessary.</p>
 290
 291 </div>
 292
 293 <!-- ======================================================================= -->
 294 <div class="doc_subsection">
 295   <a name="fund_type">Adding a fundamental type</a>
 296 </div>
 297
 298 <div class="doc_text">
 299
 300 <ol>
 301
 302 <li><tt>llvm/include/llvm/Type.h</tt>:
 303     add enum for the new type; add static <tt>Type*</tt> for this type</li>
 304
 305 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
 306     add mapping from <tt>TypeID</tt> =&gt; <tt>Type*</tt>;
 307     initialize the static <tt>Type*</tt></li>
 308
 309 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
 310     add ability to parse in the type from text assembly</li>
 311
 312 <li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>:
 313     add a token for that type</li>
 314
 315 </ol>
 316
 317 </div>
 318
 319 <!-- ======================================================================= -->
 320 <div class="doc_subsection">
 321   <a name="derived_type">Adding a derived type</a>
 322 </div>
 323
 324 <div class="doc_text">
 325
 326 <ol>
 327 <li><tt>llvm/include/llvm/Type.h</tt>:
 328     add enum for the new type; add a forward declaration of the type
 329     also</li>
 330
 331 <li><tt>llvm/include/llvm/DerivedTypes.h</tt>:
 332     add new class to represent new class in the hierarchy; add forward
 333     declaration to the TypeMap value type</li>
 334
 335 <li><tt>llvm/lib/VMCore/Type.cpp</tt>:
 336     add support for derived type to:
 337 <div class="doc_code">
 338 <pre>
 339 std::string getTypeDescription(const Type &amp;Ty,
 340   std::vector&lt;const Type*&gt; &amp;TypeStack)
 341 bool TypesEqual(const Type *Ty, const Type *Ty2,
 342   std::map&lt;const Type*, const Type*&gt; &amp; EqTypes)
 343 </pre>
 344 </div>
 345     add necessary member functions for type, and factory methods</li>
 346
 347 <li><tt>llvm/lib/AsmReader/Lexer.l</tt>:
 348     add ability to parse in the type from text assembly</li>
 349
 350 <li><tt>llvm/lib/BitCode/Writer/Writer.cpp</tt>:
 351     modify <tt>void BitcodeWriter::outputType(const Type *T)</tt> to serialize
 352     your type</li>
 353
 354 <li><tt>llvm/lib/BitCode/Reader/Reader.cpp</tt>:
 355     modify <tt>const Type *BitcodeReader::ParseType()</tt> to read your data
 356     type</li>
 357
 358 <li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>:
 359     modify
 360 <div class="doc_code">
 361 <pre>
 362 void calcTypeName(const Type *Ty,
 363                   std::vector&lt;const Type*&gt; &amp;TypeStack,
 364                   std::map&lt;const Type*,std::string&gt; &amp;TypeNames,
 365                   std::string &amp; Result)
 366 </pre>
 367 </div>
 368     to output the new derived type
 369 </li>
 370
 371
 372 </ol>
 373
 374 </div>
 375
 376 <!-- *********************************************************************** -->
 377
 378 <hr>
 379 <address>
 380   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
 381   src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
 382   <a href="http://validator.w3.org/check/referer"><img
 383   src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
 384
 385   <a href="http://llvm.org">The LLVM Compiler Infrastructure</a>
 386   <br>
 387   Last modified: $Date$
 388 </address>
 389
 390 </body>
 391 </html>