1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3<html> 4<head> 5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> 6 <title>Extending LLVM: Adding instructions, intrinsics, types, etc.</title> 7 <link rel="stylesheet" href="_static/llvm.css" type="text/css"> 8</head> 9 10<body> 11 12<h1> 13 Extending LLVM: Adding instructions, intrinsics, types, etc. 14</h1> 15 16<ol> 17 <li><a href="#introduction">Introduction and Warning</a></li> 18 <li><a href="#intrinsic">Adding a new intrinsic function</a></li> 19 <li><a href="#instruction">Adding a new instruction</a></li> 20 <li><a href="#sdnode">Adding a new SelectionDAG node</a></li> 21 <li><a href="#type">Adding a new type</a> 22 <ol> 23 <li><a href="#fund_type">Adding a new fundamental type</a></li> 24 <li><a href="#derived_type">Adding a new derived type</a></li> 25 </ol></li> 26</ol> 27 28<div class="doc_author"> 29 <p>Written by <a href="http://misha.brukman.net">Misha Brukman</a>, 30 Brad Jones, Nate Begeman, 31 and <a href="http://nondot.org/sabre">Chris Lattner</a></p> 32</div> 33 34<!-- *********************************************************************** --> 35<h2> 36 <a name="introduction">Introduction and Warning</a> 37</h2> 38<!-- *********************************************************************** --> 39 40<div> 41 42<p>During the course of using LLVM, you may wish to customize it for your 43research project or for experimentation. At this point, you may realize that 44you need to add something to LLVM, whether it be a new fundamental type, a new 45intrinsic function, or a whole new instruction.</p> 46 47<p>When you come to this realization, stop and think. Do you really need to 48extend LLVM? Is it a new fundamental capability that LLVM does not support at 49its current incarnation or can it be synthesized from already pre-existing LLVM 50elements? If you are not sure, ask on the <a 51href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM-dev</a> list. The 52reason is that extending LLVM will get involved as you need to update all the 53different passes that you intend to use with your extension, and there are 54<em>many</em> LLVM analyses and transformations, so it may be quite a bit of 55work.</p> 56 57<p>Adding an <a href="#intrinsic">intrinsic function</a> is far easier than 58adding an instruction, and is transparent to optimization passes. If your added 59functionality can be expressed as a 60function call, an intrinsic function is the method of choice for LLVM 61extension.</p> 62 63<p>Before you invest a significant amount of effort into a non-trivial 64extension, <span class="doc_warning">ask on the list</span> if what you are 65looking to do can be done with already-existing infrastructure, or if maybe 66someone else is already working on it. You will save yourself a lot of time and 67effort by doing so.</p> 68 69</div> 70 71<!-- *********************************************************************** --> 72<h2> 73 <a name="intrinsic">Adding a new intrinsic function</a> 74</h2> 75<!-- *********************************************************************** --> 76 77<div> 78 79<p>Adding a new intrinsic function to LLVM is much easier than adding a new 80instruction. Almost all extensions to LLVM should start as an intrinsic 81function and then be turned into an instruction if warranted.</p> 82 83<ol> 84<li><tt>llvm/docs/LangRef.html</tt>: 85 Document the intrinsic. Decide whether it is code generator specific and 86 what the restrictions are. Talk to other people about it so that you are 87 sure it's a good idea.</li> 88 89<li><tt>llvm/include/llvm/Intrinsics*.td</tt>: 90 Add an entry for your intrinsic. Describe its memory access characteristics 91 for optimization (this controls whether it will be DCE'd, CSE'd, etc). Note 92 that any intrinsic using the <tt>llvm_int_ty</tt> type for an argument will 93 be deemed by <tt>tblgen</tt> as overloaded and the corresponding suffix 94 will be required on the intrinsic's name.</li> 95 96<li><tt>llvm/lib/Analysis/ConstantFolding.cpp</tt>: If it is possible to 97 constant fold your intrinsic, add support to it in the 98 <tt>canConstantFoldCallTo</tt> and <tt>ConstantFoldCall</tt> functions.</li> 99 100<li><tt>llvm/test/Regression/*</tt>: Add test cases for your test cases to the 101 test suite</li> 102</ol> 103 104<p>Once the intrinsic has been added to the system, you must add code generator 105support for it. Generally you must do the following steps:</p> 106 107<dl> 108 109<dt>Add support to the .td file for the target(s) of your choice in 110 <tt>lib/Target/*/*.td</tt>.</dt> 111 112<dd>This is usually a matter of adding a pattern to the .td file that matches 113 the intrinsic, though it may obviously require adding the instructions you 114 want to generate as well. There are lots of examples in the PowerPC and X86 115 backend to follow.</dd> 116</dl> 117 118</div> 119 120<!-- *********************************************************************** --> 121<h2> 122 <a name="sdnode">Adding a new SelectionDAG node</a> 123</h2> 124<!-- *********************************************************************** --> 125 126<div> 127 128<p>As with intrinsics, adding a new SelectionDAG node to LLVM is much easier 129than adding a new instruction. New nodes are often added to help represent 130instructions common to many targets. These nodes often map to an LLVM 131instruction (add, sub) or intrinsic (byteswap, population count). In other 132cases, new nodes have been added to allow many targets to perform a common task 133(converting between floating point and integer representation) or capture more 134complicated behavior in a single node (rotate).</p> 135 136<ol> 137<li><tt>include/llvm/CodeGen/ISDOpcodes.h</tt>: 138 Add an enum value for the new SelectionDAG node.</li> 139<li><tt>lib/CodeGen/SelectionDAG/SelectionDAG.cpp</tt>: 140 Add code to print the node to <tt>getOperationName</tt>. If your new node 141 can be evaluated at compile time when given constant arguments (such as an 142 add of a constant with another constant), find the <tt>getNode</tt> method 143 that takes the appropriate number of arguments, and add a case for your node 144 to the switch statement that performs constant folding for nodes that take 145 the same number of arguments as your new node.</li> 146<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>: 147 Add code to <a href="CodeGenerator.html#selectiondag_legalize">legalize, 148 promote, and expand</a> the node as necessary. At a minimum, you will need 149 to add a case statement for your node in <tt>LegalizeOp</tt> which calls 150 LegalizeOp on the node's operands, and returns a new node if any of the 151 operands changed as a result of being legalized. It is likely that not all 152 targets supported by the SelectionDAG framework will natively support the 153 new node. In this case, you must also add code in your node's case 154 statement in <tt>LegalizeOp</tt> to Expand your node into simpler, legal 155 operations. The case for <tt>ISD::UREM</tt> for expanding a remainder into 156 a divide, multiply, and a subtract is a good example.</li> 157<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>: 158 If targets may support the new node being added only at certain sizes, you 159 will also need to add code to your node's case statement in 160 <tt>LegalizeOp</tt> to Promote your node's operands to a larger size, and 161 perform the correct operation. You will also need to add code to 162 <tt>PromoteOp</tt> to do this as well. For a good example, see 163 <tt>ISD::BSWAP</tt>, 164 which promotes its operand to a wider size, performs the byteswap, and then 165 shifts the correct bytes right to emulate the narrower byteswap in the 166 wider type.</li> 167<li><tt>lib/CodeGen/SelectionDAG/LegalizeDAG.cpp</tt>: 168 Add a case for your node in <tt>ExpandOp</tt> to teach the legalizer how to 169 perform the action represented by the new node on a value that has been 170 split into high and low halves. This case will be used to support your 171 node with a 64 bit operand on a 32 bit target.</li> 172<li><tt>lib/CodeGen/SelectionDAG/DAGCombiner.cpp</tt>: 173 If your node can be combined with itself, or other existing nodes in a 174 peephole-like fashion, add a visit function for it, and call that function 175 from <tt></tt>. There are several good examples for simple combines you 176 can do; <tt>visitFABS</tt> and <tt>visitSRL</tt> are good starting places. 177 </li> 178<li><tt>lib/Target/PowerPC/PPCISelLowering.cpp</tt>: 179 Each target has an implementation of the <tt>TargetLowering</tt> class, 180 usually in its own file (although some targets include it in the same 181 file as the DAGToDAGISel). The default behavior for a target is to 182 assume that your new node is legal for all types that are legal for 183 that target. If this target does not natively support your node, then 184 tell the target to either Promote it (if it is supported at a larger 185 type) or Expand it. This will cause the code you wrote in 186 <tt>LegalizeOp</tt> above to decompose your new node into other legal 187 nodes for this target.</li> 188<li><tt>lib/Target/TargetSelectionDAG.td</tt>: 189 Most current targets supported by LLVM generate code using the DAGToDAG 190 method, where SelectionDAG nodes are pattern matched to target-specific 191 nodes, which represent individual instructions. In order for the targets 192 to match an instruction to your new node, you must add a def for that node 193 to the list in this file, with the appropriate type constraints. Look at 194 <tt>add</tt>, <tt>bswap</tt>, and <tt>fadd</tt> for examples.</li> 195<li><tt>lib/Target/PowerPC/PPCInstrInfo.td</tt>: 196 Each target has a tablegen file that describes the target's instruction 197 set. For targets that use the DAGToDAG instruction selection framework, 198 add a pattern for your new node that uses one or more target nodes. 199 Documentation for this is a bit sparse right now, but there are several 200 decent examples. See the patterns for <tt>rotl</tt> in 201 <tt>PPCInstrInfo.td</tt>.</li> 202<li>TODO: document complex patterns.</li> 203<li><tt>llvm/test/Regression/CodeGen/*</tt>: Add test cases for your new node 204 to the test suite. <tt>llvm/test/Regression/CodeGen/X86/bswap.ll</tt> is 205 a good example.</li> 206</ol> 207 208</div> 209 210<!-- *********************************************************************** --> 211<h2> 212 <a name="instruction">Adding a new instruction</a> 213</h2> 214<!-- *********************************************************************** --> 215 216<div> 217 218<p><span class="doc_warning">WARNING: adding instructions changes the bitcode 219format, and it will take some effort to maintain compatibility with 220the previous version.</span> Only add an instruction if it is absolutely 221necessary.</p> 222 223<ol> 224 225<li><tt>llvm/include/llvm/Instruction.def</tt>: 226 add a number for your instruction and an enum name</li> 227 228<li><tt>llvm/include/llvm/Instructions.h</tt>: 229 add a definition for the class that will represent your instruction</li> 230 231<li><tt>llvm/include/llvm/Support/InstVisitor.h</tt>: 232 add a prototype for a visitor to your new instruction type</li> 233 234<li><tt>llvm/lib/AsmParser/Lexer.l</tt>: 235 add a new token to parse your instruction from assembly text file</li> 236 237<li><tt>llvm/lib/AsmParser/llvmAsmParser.y</tt>: 238 add the grammar on how your instruction can be read and what it will 239 construct as a result</li> 240 241<li><tt>llvm/lib/Bitcode/Reader/Reader.cpp</tt>: 242 add a case for your instruction and how it will be parsed from bitcode</li> 243 244<li><tt>llvm/lib/VMCore/Instruction.cpp</tt>: 245 add a case for how your instruction will be printed out to assembly</li> 246 247<li><tt>llvm/lib/VMCore/Instructions.cpp</tt>: 248 implement the class you defined in 249 <tt>llvm/include/llvm/Instructions.h</tt></li> 250 251<li>Test your instruction</li> 252 253<li><tt>llvm/lib/Target/*</tt>: 254 Add support for your instruction to code generators, or add a lowering 255 pass.</li> 256 257<li><tt>llvm/test/Regression/*</tt>: add your test cases to the test suite.</li> 258 259</ol> 260 261<p>Also, you need to implement (or modify) any analyses or passes that you want 262to understand this new instruction.</p> 263 264</div> 265 266 267<!-- *********************************************************************** --> 268<h2> 269 <a name="type">Adding a new type</a> 270</h2> 271<!-- *********************************************************************** --> 272 273<div> 274 275<p><span class="doc_warning">WARNING: adding new types changes the bitcode 276format, and will break compatibility with currently-existing LLVM 277installations.</span> Only add new types if it is absolutely necessary.</p> 278 279<!-- ======================================================================= --> 280<h3> 281 <a name="fund_type">Adding a fundamental type</a> 282</h3> 283 284<div> 285 286<ol> 287 288<li><tt>llvm/include/llvm/Type.h</tt>: 289 add enum for the new type; add static <tt>Type*</tt> for this type</li> 290 291<li><tt>llvm/lib/VMCore/Type.cpp</tt>: 292 add mapping from <tt>TypeID</tt> => <tt>Type*</tt>; 293 initialize the static <tt>Type*</tt></li> 294 295<li><tt>llvm/lib/AsmReader/Lexer.l</tt>: 296 add ability to parse in the type from text assembly</li> 297 298<li><tt>llvm/lib/AsmReader/llvmAsmParser.y</tt>: 299 add a token for that type</li> 300 301</ol> 302 303</div> 304 305<!-- ======================================================================= --> 306<h3> 307 <a name="derived_type">Adding a derived type</a> 308</h3> 309 310<div> 311 312<ol> 313<li><tt>llvm/include/llvm/Type.h</tt>: 314 add enum for the new type; add a forward declaration of the type 315 also</li> 316 317<li><tt>llvm/include/llvm/DerivedTypes.h</tt>: 318 add new class to represent new class in the hierarchy; add forward 319 declaration to the TypeMap value type</li> 320 321<li><tt>llvm/lib/VMCore/Type.cpp</tt>: 322 add support for derived type to: 323<div class="doc_code"> 324<pre> 325std::string getTypeDescription(const Type &Ty, 326 std::vector<const Type*> &TypeStack) 327bool TypesEqual(const Type *Ty, const Type *Ty2, 328 std::map<const Type*, const Type*> & EqTypes) 329</pre> 330</div> 331 add necessary member functions for type, and factory methods</li> 332 333<li><tt>llvm/lib/AsmReader/Lexer.l</tt>: 334 add ability to parse in the type from text assembly</li> 335 336<li><tt>llvm/lib/BitCode/Writer/Writer.cpp</tt>: 337 modify <tt>void BitcodeWriter::outputType(const Type *T)</tt> to serialize 338 your type</li> 339 340<li><tt>llvm/lib/BitCode/Reader/Reader.cpp</tt>: 341 modify <tt>const Type *BitcodeReader::ParseType()</tt> to read your data 342 type</li> 343 344<li><tt>llvm/lib/VMCore/AsmWriter.cpp</tt>: 345 modify 346<div class="doc_code"> 347<pre> 348void calcTypeName(const Type *Ty, 349 std::vector<const Type*> &TypeStack, 350 std::map<const Type*,std::string> &TypeNames, 351 std::string & Result) 352</pre> 353</div> 354 to output the new derived type 355</li> 356 357 358</ol> 359 360</div> 361 362</div> 363 364<!-- *********************************************************************** --> 365 366<hr> 367<address> 368 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 369 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 370 <a href="http://validator.w3.org/check/referer"><img 371 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 372 373 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a> 374 <br> 375 Last modified: $Date$ 376</address> 377 378</body> 379</html> 380