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>TableGen Fundamentals</title> 7 <link rel="stylesheet" href="llvm.css" type="text/css"> 8 </head> 9 <body> 10 11 <h1>TableGen Fundamentals</h1> 12 13 <div> 14 <ul> 15 <li><a href="#introduction">Introduction</a> 16 <ol> 17 <li><a href="#concepts">Basic concepts</a></li> 18 <li><a href="#example">An example record</a></li> 19 <li><a href="#running">Running TableGen</a></li> 20 </ol></li> 21 <li><a href="#syntax">TableGen syntax</a> 22 <ol> 23 <li><a href="#primitives">TableGen primitives</a> 24 <ol> 25 <li><a href="#comments">TableGen comments</a></li> 26 <li><a href="#types">The TableGen type system</a></li> 27 <li><a href="#values">TableGen values and expressions</a></li> 28 </ol></li> 29 <li><a href="#classesdefs">Classes and definitions</a> 30 <ol> 31 <li><a href="#valuedef">Value definitions</a></li> 32 <li><a href="#recordlet">'let' expressions</a></li> 33 <li><a href="#templateargs">Class template arguments</a></li> 34 <li><a href="#multiclass">Multiclass definitions and instances</a></li> 35 </ol></li> 36 <li><a href="#filescope">File scope entities</a> 37 <ol> 38 <li><a href="#include">File inclusion</a></li> 39 <li><a href="#globallet">'let' expressions</a></li> 40 <li><a href="#foreach">'foreach' blocks</a></li> 41 </ol></li> 42 </ol></li> 43 <li><a href="#backends">TableGen backends</a> 44 <ol> 45 <li><a href="#">todo</a></li> 46 </ol></li> 47 </ul> 48 </div> 49 50 <div class="doc_author"> 51 <p>Written by <a href="mailto:sabre (a] nondot.org">Chris Lattner</a></p> 52 </div> 53 54 <!-- *********************************************************************** --> 55 <h2><a name="introduction">Introduction</a></h2> 56 <!-- *********************************************************************** --> 57 58 <div> 59 60 <p>TableGen's purpose is to help a human develop and maintain records of 61 domain-specific information. Because there may be a large number of these 62 records, it is specifically designed to allow writing flexible descriptions and 63 for common features of these records to be factored out. This reduces the 64 amount of duplication in the description, reduces the chance of error, and 65 makes it easier to structure domain specific information.</p> 66 67 <p>The core part of TableGen <a href="#syntax">parses a file</a>, instantiates 68 the declarations, and hands the result off to a domain-specific "<a 69 href="#backends">TableGen backend</a>" for processing. The current major user 70 of TableGen is the <a href="CodeGenerator.html">LLVM code generator</a>.</p> 71 72 <p>Note that if you work on TableGen much, and use emacs or vim, that you can 73 find an emacs "TableGen mode" and a vim language file in the 74 <tt>llvm/utils/emacs</tt> and <tt>llvm/utils/vim</tt> directories of your LLVM 75 distribution, respectively.</p> 76 77 <!-- ======================================================================= --> 78 <h3><a name="concepts">Basic concepts</a></h3> 79 80 <div> 81 82 <p>TableGen files consist of two key parts: 'classes' and 'definitions', both 83 of which are considered 'records'.</p> 84 85 <p><b>TableGen records</b> have a unique name, a list of values, and a list of 86 superclasses. The list of values is the main data that TableGen builds for each 87 record; it is this that holds the domain specific information for the 88 application. The interpretation of this data is left to a specific <a 89 href="#backends">TableGen backend</a>, but the structure and format rules are 90 taken care of and are fixed by TableGen.</p> 91 92 <p><b>TableGen definitions</b> are the concrete form of 'records'. These 93 generally do not have any undefined values, and are marked with the 94 '<tt>def</tt>' keyword.</p> 95 96 <p><b>TableGen classes</b> are abstract records that are used to build and 97 describe other records. These 'classes' allow the end-user to build 98 abstractions for either the domain they are targeting (such as "Register", 99 "RegisterClass", and "Instruction" in the LLVM code generator) or for the 100 implementor to help factor out common properties of records (such as "FPInst", 101 which is used to represent floating point instructions in the X86 backend). 102 TableGen keeps track of all of the classes that are used to build up a 103 definition, so the backend can find all definitions of a particular class, such 104 as "Instruction".</p> 105 106 <p><b>TableGen multiclasses</b> are groups of abstract records that are 107 instantiated all at once. Each instantiation can result in multiple 108 TableGen definitions. If a multiclass inherits from another multiclass, 109 the definitions in the sub-multiclass become part of the current 110 multiclass, as if they were declared in the current multiclass.</p> 111 112 </div> 113 114 <!-- ======================================================================= --> 115 <h3><a name="example">An example record</a></h3> 116 117 <div> 118 119 <p>With no other arguments, TableGen parses the specified file and prints out 120 all of the classes, then all of the definitions. This is a good way to see what 121 the various definitions expand to fully. Running this on the <tt>X86.td</tt> 122 file prints this (at the time of this writing):</p> 123 124 <div class="doc_code"> 125 <pre> 126 ... 127 <b>def</b> ADD32rr { <i>// Instruction X86Inst I</i> 128 <b>string</b> Namespace = "X86"; 129 <b>dag</b> OutOperandList = (outs GR32:$dst); 130 <b>dag</b> InOperandList = (ins GR32:$src1, GR32:$src2); 131 <b>string</b> AsmString = "add{l}\t{$src2, $dst|$dst, $src2}"; 132 <b>list</b><dag> Pattern = [(set GR32:$dst, (add GR32:$src1, GR32:$src2))]; 133 <b>list</b><Register> Uses = []; 134 <b>list</b><Register> Defs = [EFLAGS]; 135 <b>list</b><Predicate> Predicates = []; 136 <b>int</b> CodeSize = 3; 137 <b>int</b> AddedComplexity = 0; 138 <b>bit</b> isReturn = 0; 139 <b>bit</b> isBranch = 0; 140 <b>bit</b> isIndirectBranch = 0; 141 <b>bit</b> isBarrier = 0; 142 <b>bit</b> isCall = 0; 143 <b>bit</b> canFoldAsLoad = 0; 144 <b>bit</b> mayLoad = 0; 145 <b>bit</b> mayStore = 0; 146 <b>bit</b> isImplicitDef = 0; 147 <b>bit</b> isConvertibleToThreeAddress = 1; 148 <b>bit</b> isCommutable = 1; 149 <b>bit</b> isTerminator = 0; 150 <b>bit</b> isReMaterializable = 0; 151 <b>bit</b> isPredicable = 0; 152 <b>bit</b> hasDelaySlot = 0; 153 <b>bit</b> usesCustomInserter = 0; 154 <b>bit</b> hasCtrlDep = 0; 155 <b>bit</b> isNotDuplicable = 0; 156 <b>bit</b> hasSideEffects = 0; 157 <b>bit</b> neverHasSideEffects = 0; 158 InstrItinClass Itinerary = NoItinerary; 159 <b>string</b> Constraints = ""; 160 <b>string</b> DisableEncoding = ""; 161 <b>bits</b><8> Opcode = { 0, 0, 0, 0, 0, 0, 0, 1 }; 162 Format Form = MRMDestReg; 163 <b>bits</b><6> FormBits = { 0, 0, 0, 0, 1, 1 }; 164 ImmType ImmT = NoImm; 165 <b>bits</b><3> ImmTypeBits = { 0, 0, 0 }; 166 <b>bit</b> hasOpSizePrefix = 0; 167 <b>bit</b> hasAdSizePrefix = 0; 168 <b>bits</b><4> Prefix = { 0, 0, 0, 0 }; 169 <b>bit</b> hasREX_WPrefix = 0; 170 FPFormat FPForm = ?; 171 <b>bits</b><3> FPFormBits = { 0, 0, 0 }; 172 } 173 ... 174 </pre> 175 </div> 176 177 <p>This definition corresponds to a 32-bit register-register add instruction in 178 the X86. The string after the '<tt>def</tt>' string indicates the name of the 179 record—"<tt>ADD32rr</tt>" in this case—and the comment at the end of 180 the line indicates the superclasses of the definition. The body of the record 181 contains all of the data that TableGen assembled for the record, indicating that 182 the instruction is part of the "X86" namespace, the pattern indicating how the 183 the instruction should be emitted into the assembly file, that it is a 184 two-address instruction, has a particular encoding, etc. The contents and 185 semantics of the information in the record is specific to the needs of the X86 186 backend, and is only shown as an example.</p> 187 188 <p>As you can see, a lot of information is needed for every instruction 189 supported by the code generator, and specifying it all manually would be 190 unmaintainable, prone to bugs, and tiring to do in the first place. Because we 191 are using TableGen, all of the information was derived from the following 192 definition:</p> 193 194 <div class="doc_code"> 195 <pre> 196 let Defs = [EFLAGS], 197 isCommutable = 1, <i>// X = ADD Y,Z --> X = ADD Z,Y</i> 198 isConvertibleToThreeAddress = 1 <b>in</b> <i>// Can transform into LEA.</i> 199 def ADD32rr : I<0x01, MRMDestReg, (outs GR32:$dst), 200 (ins GR32:$src1, GR32:$src2), 201 "add{l}\t{$src2, $dst|$dst, $src2}", 202 [(set GR32:$dst, (add GR32:$src1, GR32:$src2))]>; 203 </pre> 204 </div> 205 206 <p>This definition makes use of the custom class <tt>I</tt> (extended from the 207 custom class <tt>X86Inst</tt>), which is defined in the X86-specific TableGen 208 file, to factor out the common features that instructions of its class share. A 209 key feature of TableGen is that it allows the end-user to define the 210 abstractions they prefer to use when describing their information.</p> 211 212 <p>Each def record has a special entry called "NAME." This is the 213 name of the def ("ADD32rr" above). In the general case def names can 214 be formed from various kinds of string processing expressions and NAME 215 resolves to the final value obtained after resolving all of those 216 expressions. The user may refer to NAME anywhere she desires to use 217 the ultimate name of the def. NAME should not be defined anywhere 218 else in user code to avoid conflict problems.</p> 219 220 </div> 221 222 <!-- ======================================================================= --> 223 <h3><a name="running">Running TableGen</a></h3> 224 225 <div> 226 227 <p>TableGen runs just like any other LLVM tool. The first (optional) argument 228 specifies the file to read. If a filename is not specified, <tt>tblgen</tt> 229 reads from standard input.</p> 230 231 <p>To be useful, one of the <a href="#backends">TableGen backends</a> must be 232 used. These backends are selectable on the command line (type '<tt>tblgen 233 -help</tt>' for a list). For example, to get a list of all of the definitions 234 that subclass a particular type (which can be useful for building up an enum 235 list of these records), use the <tt>-print-enums</tt> option:</p> 236 237 <div class="doc_code"> 238 <pre> 239 $ tblgen X86.td -print-enums -class=Register 240 AH, AL, AX, BH, BL, BP, BPL, BX, CH, CL, CX, DH, DI, DIL, DL, DX, EAX, EBP, EBX, 241 ECX, EDI, EDX, EFLAGS, EIP, ESI, ESP, FP0, FP1, FP2, FP3, FP4, FP5, FP6, IP, 242 MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7, R10, R10B, R10D, R10W, R11, R11B, R11D, 243 R11W, R12, R12B, R12D, R12W, R13, R13B, R13D, R13W, R14, R14B, R14D, R14W, R15, 244 R15B, R15D, R15W, R8, R8B, R8D, R8W, R9, R9B, R9D, R9W, RAX, RBP, RBX, RCX, RDI, 245 RDX, RIP, RSI, RSP, SI, SIL, SP, SPL, ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7, 246 XMM0, XMM1, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, XMM2, XMM3, XMM4, XMM5, 247 XMM6, XMM7, XMM8, XMM9, 248 249 $ tblgen X86.td -print-enums -class=Instruction 250 ABS_F, ABS_Fp32, ABS_Fp64, ABS_Fp80, ADC32mi, ADC32mi8, ADC32mr, ADC32ri, 251 ADC32ri8, ADC32rm, ADC32rr, ADC64mi32, ADC64mi8, ADC64mr, ADC64ri32, ADC64ri8, 252 ADC64rm, ADC64rr, ADD16mi, ADD16mi8, ADD16mr, ADD16ri, ADD16ri8, ADD16rm, 253 ADD16rr, ADD32mi, ADD32mi8, ADD32mr, ADD32ri, ADD32ri8, ADD32rm, ADD32rr, 254 ADD64mi32, ADD64mi8, ADD64mr, ADD64ri32, ... 255 </pre> 256 </div> 257 258 <p>The default backend prints out all of the records, as described <a 259 href="#example">above</a>.</p> 260 261 <p>If you plan to use TableGen, you will most likely have to <a 262 href="#backends">write a backend</a> that extracts the information specific to 263 what you need and formats it in the appropriate way.</p> 264 265 </div> 266 267 </div> 268 269 <!-- *********************************************************************** --> 270 <h2><a name="syntax">TableGen syntax</a></h2> 271 <!-- *********************************************************************** --> 272 273 <div> 274 275 <p>TableGen doesn't care about the meaning of data (that is up to the backend to 276 define), but it does care about syntax, and it enforces a simple type system. 277 This section describes the syntax and the constructs allowed in a TableGen file. 278 </p> 279 280 <!-- ======================================================================= --> 281 <h3><a name="primitives">TableGen primitives</a></h3> 282 283 <div> 284 285 <!-- --------------------------------------------------------------------------> 286 <h4><a name="comments">TableGen comments</a></h4> 287 288 <div> 289 290 <p>TableGen supports BCPL style "<tt>//</tt>" comments, which run to the end of 291 the line, and it also supports <b>nestable</b> "<tt>/* */</tt>" comments.</p> 292 293 </div> 294 295 <!-- --------------------------------------------------------------------------> 296 <h4> 297 <a name="types">The TableGen type system</a> 298 </h4> 299 300 <div> 301 302 <p>TableGen files are strongly typed, in a simple (but complete) type-system. 303 These types are used to perform automatic conversions, check for errors, and to 304 help interface designers constrain the input that they allow. Every <a 305 href="#valuedef">value definition</a> is required to have an associated type. 306 </p> 307 308 <p>TableGen supports a mixture of very low-level types (such as <tt>bit</tt>) 309 and very high-level types (such as <tt>dag</tt>). This flexibility is what 310 allows it to describe a wide range of information conveniently and compactly. 311 The TableGen types are:</p> 312 313 <dl> 314 <dt><tt><b>bit</b></tt></dt> 315 <dd>A 'bit' is a boolean value that can hold either 0 or 1.</dd> 316 317 <dt><tt><b>int</b></tt></dt> 318 <dd>The 'int' type represents a simple 32-bit integer value, such as 5.</dd> 319 320 <dt><tt><b>string</b></tt></dt> 321 <dd>The 'string' type represents an ordered sequence of characters of 322 arbitrary length.</dd> 323 324 <dt><tt><b>bits</b><n></tt></dt> 325 <dd>A 'bits' type is an arbitrary, but fixed, size integer that is broken up 326 into individual bits. This type is useful because it can handle some bits 327 being defined while others are undefined.</dd> 328 329 <dt><tt><b>list</b><ty></tt></dt> 330 <dd>This type represents a list whose elements are some other type. The 331 contained type is arbitrary: it can even be another list type.</dd> 332 333 <dt>Class type</dt> 334 <dd>Specifying a class name in a type context means that the defined value 335 must be a subclass of the specified class. This is useful in conjunction with 336 the <b><tt>list</tt></b> type, for example, to constrain the elements of the 337 list to a common base class (e.g., a <tt><b>list</b><Register></tt> can 338 only contain definitions derived from the "<tt>Register</tt>" class).</dd> 339 340 <dt><tt><b>dag</b></tt></dt> 341 <dd>This type represents a nestable directed graph of elements.</dd> 342 343 <dt><tt><b>code</b></tt></dt> 344 <dd>This represents a big hunk of text. This is lexically distinct from 345 string values because it doesn't require escapeing double quotes and other 346 common characters that occur in code.</dd> 347 </dl> 348 349 <p>To date, these types have been sufficient for describing things that 350 TableGen has been used for, but it is straight-forward to extend this list if 351 needed.</p> 352 353 </div> 354 355 <!-- --------------------------------------------------------------------------> 356 <h4> 357 <a name="values">TableGen values and expressions</a> 358 </h4> 359 360 <div> 361 362 <p>TableGen allows for a pretty reasonable number of different expression forms 363 when building up values. These forms allow the TableGen file to be written in a 364 natural syntax and flavor for the application. The current expression forms 365 supported include:</p> 366 367 <dl> 368 <dt><tt>?</tt></dt> 369 <dd>uninitialized field</dd> 370 <dt><tt>0b1001011</tt></dt> 371 <dd>binary integer value</dd> 372 <dt><tt>07654321</tt></dt> 373 <dd>octal integer value (indicated by a leading 0)</dd> 374 <dt><tt>7</tt></dt> 375 <dd>decimal integer value</dd> 376 <dt><tt>0x7F</tt></dt> 377 <dd>hexadecimal integer value</dd> 378 <dt><tt>"foo"</tt></dt> 379 <dd>string value</dd> 380 <dt><tt>[{ ... }]</tt></dt> 381 <dd>code fragment</dd> 382 <dt><tt>[ X, Y, Z ]<type></tt></dt> 383 <dd>list value. <type> is the type of the list 384 element and is usually optional. In rare cases, 385 TableGen is unable to deduce the element type in 386 which case the user must specify it explicitly.</dd> 387 <dt><tt>{ a, b, c }</tt></dt> 388 <dd>initializer for a "bits<3>" value</dd> 389 <dt><tt>value</tt></dt> 390 <dd>value reference</dd> 391 <dt><tt>value{17}</tt></dt> 392 <dd>access to one bit of a value</dd> 393 <dt><tt>value{15-17}</tt></dt> 394 <dd>access to multiple bits of a value</dd> 395 <dt><tt>DEF</tt></dt> 396 <dd>reference to a record definition</dd> 397 <dt><tt>CLASS<val list></tt></dt> 398 <dd>reference to a new anonymous definition of CLASS with the specified 399 template arguments.</dd> 400 <dt><tt>X.Y</tt></dt> 401 <dd>reference to the subfield of a value</dd> 402 <dt><tt>list[4-7,17,2-3]</tt></dt> 403 <dd>A slice of the 'list' list, including elements 4,5,6,7,17,2, and 3 from 404 it. Elements may be included multiple times.</dd> 405 <dt><tt>foreach <var> = <list> in { <body> }</tt></dt> 406 <dt><tt>foreach <var> = <list> in <def></tt></dt> 407 <dd> Replicate <body> or <def>, replacing instances of 408 <var> with each value in <list>. <var> is scoped at the 409 level of the <tt>foreach</tt> loop and must not conflict with any other object 410 introduced in <body> or <def>. Currently only <tt>def</tt>s are 411 expanded within <body>. 412 </dd> 413 <dt><tt>(DEF a, b)</tt></dt> 414 <dd>a dag value. The first element is required to be a record definition, the 415 remaining elements in the list may be arbitrary other values, including nested 416 `<tt>dag</tt>' values.</dd> 417 <dt><tt>!strconcat(a, b)</tt></dt> 418 <dd>A string value that is the result of concatenating the 'a' and 'b' 419 strings.</dd> 420 <dt><tt>str1#str2</tt></dt> 421 <dd>"#" (paste) is a shorthand for !strconcat. It may concatenate 422 things that are not quoted strings, in which case an implicit 423 !cast<string> is done on the operand of the paste.</dd> 424 <dt><tt>!cast<type>(a)</tt></dt> 425 <dd>A symbol of type <em>type</em> obtained by looking up the string 'a' in 426 the symbol table. If the type of 'a' does not match <em>type</em>, TableGen 427 aborts with an error. !cast<string> is a special case in that the argument must 428 be an object defined by a 'def' construct.</dd> 429 <dt><tt>!subst(a, b, c)</tt></dt> 430 <dd>If 'a' and 'b' are of string type or are symbol references, substitute 431 'b' for 'a' in 'c.' This operation is analogous to $(subst) in GNU make.</dd> 432 <dt><tt>!foreach(a, b, c)</tt></dt> 433 <dd>For each member 'b' of dag or list 'a' apply operator 'c.' 'b' is a 434 dummy variable that should be declared as a member variable of an instantiated 435 class. This operation is analogous to $(foreach) in GNU make.</dd> 436 <dt><tt>!head(a)</tt></dt> 437 <dd>The first element of list 'a.'</dd> 438 <dt><tt>!tail(a)</tt></dt> 439 <dd>The 2nd-N elements of list 'a.'</dd> 440 <dt><tt>!empty(a)</tt></dt> 441 <dd>An integer {0,1} indicating whether list 'a' is empty.</dd> 442 <dt><tt>!if(a,b,c)</tt></dt> 443 <dd>'b' if the result of 'int' or 'bit' operator 'a' is nonzero, 444 'c' otherwise.</dd> 445 <dt><tt>!eq(a,b)</tt></dt> 446 <dd>'bit 1' if string a is equal to string b, 0 otherwise. This 447 only operates on string, int and bit objects. Use !cast<string> to 448 compare other types of objects.</dd> 449 </dl> 450 451 <p>Note that all of the values have rules specifying how they convert to values 452 for different types. These rules allow you to assign a value like "<tt>7</tt>" 453 to a "<tt>bits<4></tt>" value, for example.</p> 454 455 </div> 456 457 </div> 458 459 <!-- ======================================================================= --> 460 <h3> 461 <a name="classesdefs">Classes and definitions</a> 462 </h3> 463 464 <div> 465 466 <p>As mentioned in the <a href="#concepts">intro</a>, classes and definitions 467 (collectively known as 'records') in TableGen are the main high-level unit of 468 information that TableGen collects. Records are defined with a <tt>def</tt> or 469 <tt>class</tt> keyword, the record name, and an optional list of "<a 470 href="#templateargs">template arguments</a>". If the record has superclasses, 471 they are specified as a comma separated list that starts with a colon character 472 ("<tt>:</tt>"). If <a href="#valuedef">value definitions</a> or <a 473 href="#recordlet">let expressions</a> are needed for the class, they are 474 enclosed in curly braces ("<tt>{}</tt>"); otherwise, the record ends with a 475 semicolon.</p> 476 477 <p>Here is a simple TableGen file:</p> 478 479 <div class="doc_code"> 480 <pre> 481 <b>class</b> C { <b>bit</b> V = 1; } 482 <b>def</b> X : C; 483 <b>def</b> Y : C { 484 <b>string</b> Greeting = "hello"; 485 } 486 </pre> 487 </div> 488 489 <p>This example defines two definitions, <tt>X</tt> and <tt>Y</tt>, both of 490 which derive from the <tt>C</tt> class. Because of this, they both get the 491 <tt>V</tt> bit value. The <tt>Y</tt> definition also gets the Greeting member 492 as well.</p> 493 494 <p>In general, classes are useful for collecting together the commonality 495 between a group of records and isolating it in a single place. Also, classes 496 permit the specification of default values for their subclasses, allowing the 497 subclasses to override them as they wish.</p> 498 499 <!----------------------------------------------------------------------------> 500 <h4> 501 <a name="valuedef">Value definitions</a> 502 </h4> 503 504 <div> 505 506 <p>Value definitions define named entries in records. A value must be defined 507 before it can be referred to as the operand for another value definition or 508 before the value is reset with a <a href="#recordlet">let expression</a>. A 509 value is defined by specifying a <a href="#types">TableGen type</a> and a name. 510 If an initial value is available, it may be specified after the type with an 511 equal sign. Value definitions require terminating semicolons.</p> 512 513 </div> 514 515 <!-- --------------------------------------------------------------------------> 516 <h4> 517 <a name="recordlet">'let' expressions</a> 518 </h4> 519 520 <div> 521 522 <p>A record-level let expression is used to change the value of a value 523 definition in a record. This is primarily useful when a superclass defines a 524 value that a derived class or definition wants to override. Let expressions 525 consist of the '<tt>let</tt>' keyword followed by a value name, an equal sign 526 ("<tt>=</tt>"), and a new value. For example, a new class could be added to the 527 example above, redefining the <tt>V</tt> field for all of its subclasses:</p> 528 529 <div class="doc_code"> 530 <pre> 531 <b>class</b> D : C { let V = 0; } 532 <b>def</b> Z : D; 533 </pre> 534 </div> 535 536 <p>In this case, the <tt>Z</tt> definition will have a zero value for its "V" 537 value, despite the fact that it derives (indirectly) from the <tt>C</tt> class, 538 because the <tt>D</tt> class overrode its value.</p> 539 540 </div> 541 542 <!-- --------------------------------------------------------------------------> 543 <h4> 544 <a name="templateargs">Class template arguments</a> 545 </h4> 546 547 <div> 548 549 <p>TableGen permits the definition of parameterized classes as well as normal 550 concrete classes. Parameterized TableGen classes specify a list of variable 551 bindings (which may optionally have defaults) that are bound when used. Here is 552 a simple example:</p> 553 554 <div class="doc_code"> 555 <pre> 556 <b>class</b> FPFormat<<b>bits</b><3> val> { 557 <b>bits</b><3> Value = val; 558 } 559 <b>def</b> NotFP : FPFormat<0>; 560 <b>def</b> ZeroArgFP : FPFormat<1>; 561 <b>def</b> OneArgFP : FPFormat<2>; 562 <b>def</b> OneArgFPRW : FPFormat<3>; 563 <b>def</b> TwoArgFP : FPFormat<4>; 564 <b>def</b> CompareFP : FPFormat<5>; 565 <b>def</b> CondMovFP : FPFormat<6>; 566 <b>def</b> SpecialFP : FPFormat<7>; 567 </pre> 568 </div> 569 570 <p>In this case, template arguments are used as a space efficient way to specify 571 a list of "enumeration values", each with a "<tt>Value</tt>" field set to the 572 specified integer.</p> 573 574 <p>The more esoteric forms of <a href="#values">TableGen expressions</a> are 575 useful in conjunction with template arguments. As an example:</p> 576 577 <div class="doc_code"> 578 <pre> 579 <b>class</b> ModRefVal<<b>bits</b><2> val> { 580 <b>bits</b><2> Value = val; 581 } 582 583 <b>def</b> None : ModRefVal<0>; 584 <b>def</b> Mod : ModRefVal<1>; 585 <b>def</b> Ref : ModRefVal<2>; 586 <b>def</b> ModRef : ModRefVal<3>; 587 588 <b>class</b> Value<ModRefVal MR> { 589 <i>// Decode some information into a more convenient format, while providing 590 // a nice interface to the user of the "Value" class.</i> 591 <b>bit</b> isMod = MR.Value{0}; 592 <b>bit</b> isRef = MR.Value{1}; 593 594 <i>// other stuff...</i> 595 } 596 597 <i>// Example uses</i> 598 <b>def</b> bork : Value<Mod>; 599 <b>def</b> zork : Value<Ref>; 600 <b>def</b> hork : Value<ModRef>; 601 </pre> 602 </div> 603 604 <p>This is obviously a contrived example, but it shows how template arguments 605 can be used to decouple the interface provided to the user of the class from the 606 actual internal data representation expected by the class. In this case, 607 running <tt>tblgen</tt> on the example prints the following definitions:</p> 608 609 <div class="doc_code"> 610 <pre> 611 <b>def</b> bork { <i>// Value</i> 612 <b>bit</b> isMod = 1; 613 <b>bit</b> isRef = 0; 614 } 615 <b>def</b> hork { <i>// Value</i> 616 <b>bit</b> isMod = 1; 617 <b>bit</b> isRef = 1; 618 } 619 <b>def</b> zork { <i>// Value</i> 620 <b>bit</b> isMod = 0; 621 <b>bit</b> isRef = 1; 622 } 623 </pre> 624 </div> 625 626 <p> This shows that TableGen was able to dig into the argument and extract a 627 piece of information that was requested by the designer of the "Value" class. 628 For more realistic examples, please see existing users of TableGen, such as the 629 X86 backend.</p> 630 631 </div> 632 633 <!-- --------------------------------------------------------------------------> 634 <h4> 635 <a name="multiclass">Multiclass definitions and instances</a> 636 </h4> 637 638 <div> 639 640 <p> 641 While classes with template arguments are a good way to factor commonality 642 between two instances of a definition, multiclasses allow a convenient notation 643 for defining multiple definitions at once (instances of implicitly constructed 644 classes). For example, consider an 3-address instruction set whose instructions 645 come in two forms: "<tt>reg = reg op reg</tt>" and "<tt>reg = reg op imm</tt>" 646 (e.g. SPARC). In this case, you'd like to specify in one place that this 647 commonality exists, then in a separate place indicate what all the ops are. 648 </p> 649 650 <p> 651 Here is an example TableGen fragment that shows this idea: 652 </p> 653 654 <div class="doc_code"> 655 <pre> 656 <b>def</b> ops; 657 <b>def</b> GPR; 658 <b>def</b> Imm; 659 <b>class</b> inst<<b>int</b> opc, <b>string</b> asmstr, <b>dag</b> operandlist>; 660 661 <b>multiclass</b> ri_inst<<b>int</b> opc, <b>string</b> asmstr> { 662 def _rr : inst<opc, !strconcat(asmstr, " $dst, $src1, $src2"), 663 (ops GPR:$dst, GPR:$src1, GPR:$src2)>; 664 def _ri : inst<opc, !strconcat(asmstr, " $dst, $src1, $src2"), 665 (ops GPR:$dst, GPR:$src1, Imm:$src2)>; 666 } 667 668 <i>// Instantiations of the ri_inst multiclass.</i> 669 <b>defm</b> ADD : ri_inst<0b111, "add">; 670 <b>defm</b> SUB : ri_inst<0b101, "sub">; 671 <b>defm</b> MUL : ri_inst<0b100, "mul">; 672 ... 673 </pre> 674 </div> 675 676 <p>The name of the resultant definitions has the multidef fragment names 677 appended to them, so this defines <tt>ADD_rr</tt>, <tt>ADD_ri</tt>, 678 <tt>SUB_rr</tt>, etc. A defm may inherit from multiple multiclasses, 679 instantiating definitions from each multiclass. Using a multiclass 680 this way is exactly equivalent to instantiating the classes multiple 681 times yourself, e.g. by writing:</p> 682 683 <div class="doc_code"> 684 <pre> 685 <b>def</b> ops; 686 <b>def</b> GPR; 687 <b>def</b> Imm; 688 <b>class</b> inst<<b>int</b> opc, <b>string</b> asmstr, <b>dag</b> operandlist>; 689 690 <b>class</b> rrinst<<b>int</b> opc, <b>string</b> asmstr> 691 : inst<opc, !strconcat(asmstr, " $dst, $src1, $src2"), 692 (ops GPR:$dst, GPR:$src1, GPR:$src2)>; 693 694 <b>class</b> riinst<<b>int</b> opc, <b>string</b> asmstr> 695 : inst<opc, !strconcat(asmstr, " $dst, $src1, $src2"), 696 (ops GPR:$dst, GPR:$src1, Imm:$src2)>; 697 698 <i>// Instantiations of the ri_inst multiclass.</i> 699 <b>def</b> ADD_rr : rrinst<0b111, "add">; 700 <b>def</b> ADD_ri : riinst<0b111, "add">; 701 <b>def</b> SUB_rr : rrinst<0b101, "sub">; 702 <b>def</b> SUB_ri : riinst<0b101, "sub">; 703 <b>def</b> MUL_rr : rrinst<0b100, "mul">; 704 <b>def</b> MUL_ri : riinst<0b100, "mul">; 705 ... 706 </pre> 707 </div> 708 709 <p> 710 A defm can also be used inside a multiclass providing several levels of 711 multiclass instanciations. 712 </p> 713 714 <div class="doc_code"> 715 <pre> 716 <b>class</b> Instruction<bits<4> opc, string Name> { 717 bits<4> opcode = opc; 718 string name = Name; 719 } 720 721 <b>multiclass</b> basic_r<bits<4> opc> { 722 <b>def</b> rr : Instruction<opc, "rr">; 723 <b>def</b> rm : Instruction<opc, "rm">; 724 } 725 726 <b>multiclass</b> basic_s<bits<4> opc> { 727 <b>defm</b> SS : basic_r<opc>; 728 <b>defm</b> SD : basic_r<opc>; 729 <b>def</b> X : Instruction<opc, "x">; 730 } 731 732 <b>multiclass</b> basic_p<bits<4> opc> { 733 <b>defm</b> PS : basic_r<opc>; 734 <b>defm</b> PD : basic_r<opc>; 735 <b>def</b> Y : Instruction<opc, "y">; 736 } 737 738 <b>defm</b> ADD : basic_s<0xf>, basic_p<0xf>; 739 ... 740 741 <i>// Results</i> 742 <b>def</b> ADDPDrm { ... 743 <b>def</b> ADDPDrr { ... 744 <b>def</b> ADDPSrm { ... 745 <b>def</b> ADDPSrr { ... 746 <b>def</b> ADDSDrm { ... 747 <b>def</b> ADDSDrr { ... 748 <b>def</b> ADDY { ... 749 <b>def</b> ADDX { ... 750 </pre> 751 </div> 752 753 <p> 754 defm declarations can inherit from classes too, the 755 rule to follow is that the class list must start after the 756 last multiclass, and there must be at least one multiclass 757 before them. 758 </p> 759 760 <div class="doc_code"> 761 <pre> 762 <b>class</b> XD { bits<4> Prefix = 11; } 763 <b>class</b> XS { bits<4> Prefix = 12; } 764 765 <b>class</b> I<bits<4> op> { 766 bits<4> opcode = op; 767 } 768 769 <b>multiclass</b> R { 770 <b>def</b> rr : I<4>; 771 <b>def</b> rm : I<2>; 772 } 773 774 <b>multiclass</b> Y { 775 <b>defm</b> SS : R, XD; 776 <b>defm</b> SD : R, XS; 777 } 778 779 <b>defm</b> Instr : Y; 780 781 <i>// Results</i> 782 <b>def</b> InstrSDrm { 783 bits<4> opcode = { 0, 0, 1, 0 }; 784 bits<4> Prefix = { 1, 1, 0, 0 }; 785 } 786 ... 787 <b>def</b> InstrSSrr { 788 bits<4> opcode = { 0, 1, 0, 0 }; 789 bits<4> Prefix = { 1, 0, 1, 1 }; 790 } 791 </pre> 792 </div> 793 794 </div> 795 796 </div> 797 798 <!-- ======================================================================= --> 799 <h3> 800 <a name="filescope">File scope entities</a> 801 </h3> 802 803 <div> 804 805 <!-- --------------------------------------------------------------------------> 806 <h4> 807 <a name="include">File inclusion</a> 808 </h4> 809 810 <div> 811 <p>TableGen supports the '<tt>include</tt>' token, which textually substitutes 812 the specified file in place of the include directive. The filename should be 813 specified as a double quoted string immediately after the '<tt>include</tt>' 814 keyword. Example:</p> 815 816 <div class="doc_code"> 817 <pre> 818 <b>include</b> "foo.td" 819 </pre> 820 </div> 821 822 </div> 823 824 <!-- --------------------------------------------------------------------------> 825 <h4> 826 <a name="globallet">'let' expressions</a> 827 </h4> 828 829 <div> 830 831 <p>"Let" expressions at file scope are similar to <a href="#recordlet">"let" 832 expressions within a record</a>, except they can specify a value binding for 833 multiple records at a time, and may be useful in certain other cases. 834 File-scope let expressions are really just another way that TableGen allows the 835 end-user to factor out commonality from the records.</p> 836 837 <p>File-scope "let" expressions take a comma-separated list of bindings to 838 apply, and one or more records to bind the values in. Here are some 839 examples:</p> 840 841 <div class="doc_code"> 842 <pre> 843 <b>let</b> isTerminator = 1, isReturn = 1, isBarrier = 1, hasCtrlDep = 1 <b>in</b> 844 <b>def</b> RET : I<0xC3, RawFrm, (outs), (ins), "ret", [(X86retflag 0)]>; 845 846 <b>let</b> isCall = 1 <b>in</b> 847 <i>// All calls clobber the non-callee saved registers...</i> 848 <b>let</b> Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0, 849 MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7, 850 XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7, EFLAGS] <b>in</b> { 851 <b>def</b> CALLpcrel32 : Ii32<0xE8, RawFrm, (outs), (ins i32imm:$dst,variable_ops), 852 "call\t${dst:call}", []>; 853 <b>def</b> CALL32r : I<0xFF, MRM2r, (outs), (ins GR32:$dst, variable_ops), 854 "call\t{*}$dst", [(X86call GR32:$dst)]>; 855 <b>def</b> CALL32m : I<0xFF, MRM2m, (outs), (ins i32mem:$dst, variable_ops), 856 "call\t{*}$dst", []>; 857 } 858 </pre> 859 </div> 860 861 <p>File-scope "let" expressions are often useful when a couple of definitions 862 need to be added to several records, and the records do not otherwise need to be 863 opened, as in the case with the <tt>CALL*</tt> instructions above.</p> 864 865 <p>It's also possible to use "let" expressions inside multiclasses, providing 866 more ways to factor out commonality from the records, specially if using 867 several levels of multiclass instanciations. This also avoids the need of using 868 "let" expressions within subsequent records inside a multiclass.</p> 869 870 <pre class="doc_code"> 871 <b>multiclass </b>basic_r<bits<4> opc> { 872 <b>let </b>Predicates = [HasSSE2] in { 873 <b>def </b>rr : Instruction<opc, "rr">; 874 <b>def </b>rm : Instruction<opc, "rm">; 875 } 876 <b>let </b>Predicates = [HasSSE3] in 877 <b>def </b>rx : Instruction<opc, "rx">; 878 } 879 880 <b>multiclass </b>basic_ss<bits<4> opc> { 881 <b>let </b>IsDouble = 0 in 882 <b>defm </b>SS : basic_r<opc>; 883 884 <b>let </b>IsDouble = 1 in 885 <b>defm </b>SD : basic_r<opc>; 886 } 887 888 <b>defm </b>ADD : basic_ss<0xf>; 889 </pre> 890 </div> 891 892 <!-- --------------------------------------------------------------------------> 893 <h4> 894 <a name="foreach">Looping</a> 895 </h4> 896 897 <div> 898 <p>TableGen supports the '<tt>foreach</tt>' block, which textually replicates 899 the loop body, substituting iterator values for iterator references in the 900 body. Example:</p> 901 902 <div class="doc_code"> 903 <pre> 904 <b>foreach</b> i = [0, 1, 2, 3] in { 905 <b>def</b> R#i : Register<...>; 906 <b>def</b> F#i : Register<...>; 907 } 908 </pre> 909 </div> 910 911 <p>This will create objects <tt>R0</tt>, <tt>R1</tt>, <tt>R2</tt> and 912 <tt>R3</tt>. <tt>foreach</tt> blocks may be nested. If there is only 913 one item in the body the braces may be elided:</p> 914 915 <div class="doc_code"> 916 <pre> 917 <b>foreach</b> i = [0, 1, 2, 3] in 918 <b>def</b> R#i : Register<...>; 919 920 </pre> 921 </div> 922 923 </div> 924 925 </div> 926 927 </div> 928 929 <!-- *********************************************************************** --> 930 <h2><a name="codegen">Code Generator backend info</a></h2> 931 <!-- *********************************************************************** --> 932 933 <div> 934 935 <p>Expressions used by code generator to describe instructions and isel 936 patterns:</p> 937 938 <dl> 939 <dt><tt>(implicit a)</tt></dt> 940 <dd>an implicitly defined physical register. This tells the dag instruction 941 selection emitter the input pattern's extra definitions matches implicit 942 physical register definitions.</dd> 943 </dl> 944 </div> 945 946 <!-- *********************************************************************** --> 947 <h2><a name="backends">TableGen backends</a></h2> 948 <!-- *********************************************************************** --> 949 950 <div> 951 952 <p>TODO: How they work, how to write one. This section should not contain 953 details about any particular backend, except maybe -print-enums as an example. 954 This should highlight the APIs in <tt>TableGen/Record.h</tt>.</p> 955 956 </div> 957 958 <!-- *********************************************************************** --> 959 960 <hr> 961 <address> 962 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 963 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 964 <a href="http://validator.w3.org/check/referer"><img 965 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 966 967 <a href="mailto:sabre (a] nondot.org">Chris Lattner</a><br> 968 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br> 969 Last modified: $Date$ 970 </address> 971 972 </body> 973 </html> 974
