1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>TableGen Fundamentals</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">TableGen Fundamentals</div>
12 <div class="doc_text">
14 <li><a href="#introduction">Introduction</a>
16 <li><a href="#concepts">Basic concepts</a></li>
17 <li><a href="#example">An example record</a></li>
18 <li><a href="#running">Running TableGen</a></li>
20 <li><a href="#syntax">TableGen syntax</a>
22 <li><a href="#primitives">TableGen primitives</a>
24 <li><a href="#comments">TableGen comments</a></li>
25 <li><a href="#types">The TableGen type system</a></li>
26 <li><a href="#values">TableGen values and expressions</a></li>
28 <li><a href="#classesdefs">Classes and definitions</a>
30 <li><a href="#valuedef">Value definitions</a></li>
31 <li><a href="#recordlet">'let' expressions</a></li>
32 <li><a href="#templateargs">Class template arguments</a></li>
34 <li><a href="#filescope">File scope entities</a>
36 <li><a href="#include">File inclusion</a></li>
37 <li><a href="#globallet">'let' expressions</a></li>
40 <li><a href="#backends">TableGen backends</a>
42 <li><a href="#">todo</a></li>
47 <div class="doc_author">
48 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
51 <!-- *********************************************************************** -->
52 <div class="doc_section"><a name="introduction">Introduction</a></div>
53 <!-- *********************************************************************** -->
55 <div class="doc_text">
57 <p>TableGen's purpose is to help a human develop and maintain records of
58 domain-specific information. Because there may be a large number of these
59 records, it is specifically designed to allow writing flexible descriptions and
60 for common features of these records to be factored out. This reduces the
61 amount of duplication in the description, reduces the chance of error, and
62 makes it easier to structure domain specific information.</p>
64 <p>The core part of TableGen <a href="#syntax">parses a file</a>, instantiates
65 the declarations, and hands the result off to a domain-specific "<a
66 href="#backends">TableGen backend</a>" for processing. The current major user
67 of TableGen is the <a href="CodeGenerator.html">LLVM code generator</a>.</p>
69 <p>Note that if you work on TableGen much, and use emacs or vim, that you can
70 find an emacs "TableGen mode" and a vim language file in
71 <tt>llvm/utils/emacs</tt> and <tt>llvm/utils/vim</tt> directory of your LLVM
72 distribution, respectively.</p>
76 <!-- ======================================================================= -->
77 <div class="doc_subsection"><a name="running">Basic concepts</a></div>
79 <div class="doc_text">
81 <p>TableGen files consist of two key parts: 'classes' and 'definitions', both
82 of which are considered 'records'.</p>
84 <p><b>TableGen records</b> have a unique name, a list of values, and a list of
85 superclasses. The list of values is main data that TableGen builds for each
86 record, it is this that holds the domain specific information for the
87 application. The interpretation of this data is left to a specific <a
88 href="#backends">TableGen backend</a>, but the structure and format rules are
89 taken care of and fixed by TableGen.</p>
91 <p><b>TableGen definitions</b> are the concrete form of 'records'. These
92 generally do not have any undefined values, and are marked with the
93 '<tt>def</tt>' keyword.</p>
95 <p><b>TableGen classes</b> are abstract records that are used to build and
96 describe other records. These 'classes' allow the end-user to build
97 abstractions for either the domain they are targetting (such as "Register",
98 "RegisterClass", and "Instruction" in the LLVM code generator) or for the
99 implementor to help factor out common properties of records (such as "FPInst",
100 which is used to represent floating point instructions in the X86 backend).
101 TableGen keeps track of all of the classes that are used to build up a
102 definition, so the backend can find all definitions of a particular class, such
103 as "Instruction".</p>
107 <!-- ======================================================================= -->
108 <div class="doc_subsection"><a name="example">An example record</a></div>
110 <div class="doc_text">
112 <p>With no other arguments, TableGen parses the specified file and prints out
113 all of the classes, then all of the definitions. This is a good way to see what
114 the various definitions expand to fully. Running this on the <tt>X86.td</tt>
115 file prints this (at the time of this writing):</p>
119 <b>def</b> ADDrr8 { <i>// Instruction X86Inst I2A8 Pattern</i>
120 <b>string</b> Name = "add";
121 <b>string</b> Namespace = "X86";
122 <b>list</b><Register> Uses = [];
123 <b>list</b><Register> Defs = [];
124 <b>bit</b> isReturn = 0;
125 <b>bit</b> isBranch = 0;
126 <b>bit</b> isCall = 0;
127 <b>bit</b> isTwoAddress = 1;
128 <b>bit</b> isTerminator = 0;
129 <b>dag</b> Pattern = (set R8, (plus R8, R8));
130 <b>bits</b><8> Opcode = { 0, 0, 0, 0, 0, 0, 0, 0 };
131 Format Form = MRMDestReg;
132 <b>bits</b><5> FormBits = { 0, 0, 0, 1, 1 };
134 <b>bits</b><3> TypeBits = { 0, 0, 1 };
135 <b>bit</b> hasOpSizePrefix = 0;
136 <b>bit</b> printImplicitUses = 0;
137 <b>bits</b><4> Prefix = { 0, 0, 0, 0 };
139 <b>bits</b><3> FPFormBits = { 0, 0, 0 };
144 <p>This definition corresponds to an 8-bit register-register add instruction in
145 the X86. The string after the '<tt>def</tt>' string indicates the name of the
146 record ("<tt>ADDrr8</tt>" in this case), and the comment at the end of the line
147 indicates the superclasses of the definition. The body of the record contains
148 all of the data that TableGen assembled for the record, indicating that the
149 instruction is part of the "X86" namespace, should be printed as "<tt>add</tt>"
150 in the assembly file, it is a two-address instruction, has a particular
151 encoding, etc. The contents and semantics of the information in the record is
152 specific to the needs of the X86 backend, and is only shown as an example.</p>
154 <p>As you can see, a lot of information is needed for every instruction
155 supported by the code generator, and specifying it all manually would be
156 unmaintainble, prone to bugs, and tiring to do in the first place. Because we
157 are using TableGen, all of the information was derived from the following
161 <b>def</b> ADDrr8 : I2A8<"add", 0x00, MRMDestReg>,
162 Pattern<(set R8, (plus R8, R8))>;
165 <p>This definition makes use of the custom I2A8 (two address instruction with
166 8-bit operand) class, which is defined in the X86-specific TableGen file to
167 factor out the common features that instructions of its class share. A key
168 feature of TableGen is that it allows the end-user to define the abstractions
169 they prefer to use when describing their information.</p>
173 <!-- ======================================================================= -->
174 <div class="doc_subsection"><a name="running">Running TableGen</a></div>
176 <div class="doc_text">
178 <p>TableGen runs just like any other LLVM tool. The first (optional) argument
179 specifies the file to read. If a filename is not specified, <tt>tblgen</tt>
180 reads from standard input.</p>
182 <p>To be useful, one of the <a href="#backends">TableGen backends</a> must be
183 used. These backends are selectable on the command line (type '<tt>tblgen
184 --help</tt>' for a list). For example, to get a list of all of the definitions
185 that subclass a particular type (which can be useful for building up an enum
186 list of these records), use the <tt>--print-enums</tt> option:</p>
189 $ tblgen X86.td -print-enums -class=Register
190 AH, AL, AX, BH, BL, BP, BX, CH, CL, CX, DH, DI, DL, DX,
191 EAX, EBP, EBX, ECX, EDI, EDX, ESI, ESP, FP0, FP1, FP2, FP3, FP4, FP5, FP6,
192 SI, SP, ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
194 $ tblgen X86.td -print-enums -class=Instruction
195 ADCrr32, ADDri16, ADDri16b, ADDri32, ADDri32b, ADDri8, ADDrr16, ADDrr32,
196 ADDrr8, ADJCALLSTACKDOWN, ADJCALLSTACKUP, ANDri16, ANDri16b, ANDri32, ANDri32b,
197 ANDri8, ANDrr16, ANDrr32, ANDrr8, BSWAPr32, CALLm32, CALLpcrel32, ...
200 <p>The default backend prints out all of the records, as described <a
201 href="#example">above</a>.</p>
203 <p>If you plan to use TableGen for some purpose, you will most likely have to
204 <a href="#backends">write a backend</a> that extracts the information specific
205 to what you need and formats it in the appropriate way.</p>
210 <!-- *********************************************************************** -->
211 <div class="doc_section"><a name="syntax">TableGen syntax</a></div>
212 <!-- *********************************************************************** -->
214 <div class="doc_text">
215 <p>TableGen doesn't care about the meaning of data (that is up to the backend
216 to define), but it does care about syntax, and it enforces a simple type system.
217 This section describes the syntax and the constructs allowed in a TableGen file.
221 <!-- ======================================================================= -->
222 <div class="doc_subsection"><a name="primitives">TableGen primitives</a></div>
224 <!-- -------------------------------------------------------------------------->
225 <div class="doc_subsubsection"><a name="comments">TableGen comments</a></div>
227 <div class="doc_text">
228 <p>TableGen supports BCPL style "<tt>//</tt>" comments, which run to the end of
229 the line, and it also supports <b>nestable</b> "<tt>/* */</tt>" comments.</p>
232 <!-- -------------------------------------------------------------------------->
233 <div class="doc_subsubsection">
234 <a name="types">The TableGen type system</a>
237 <div class="doc_text">
238 <p>TableGen files are strongly typed, in a simple (but complete) type-system.
239 These types are used to perform automatic conversions, check for errors, and to
240 help interface designers constrain the input that they allow. Every <a
241 href="#valuedef">value definition</a> is required to have an associated type.
244 <p>TableGen supports a mixture of very low-level types (such as <tt>bit</tt>)
245 and very high-level types (such as <tt>dag</tt>). This flexibility is what
246 allows it to describe a wide range of information conveniently and compactly.
247 The TableGen types are:</p>
250 <li>"<tt><b>bit</b></tt>" - A 'bit' is a boolean value that can hold either 0 or
253 <li>"<tt><b>int</b></tt>" - The 'int' type represents a simple 32-bit integer
254 value, such as 5.</li>
256 <li>"<tt><b>string</b></tt>" - The 'string' type represents an ordered sequence
257 of characters of arbitrary length.</li>
259 <li>"<tt><b>bits</b><n></tt>" - A 'bits' type is an arbitrary, but fixed,
260 size integer that is broken up into individual bits. This type is useful
261 because it can handle some bits being defined while others are undefined.</li>
263 <li>"<tt><b>list</b><ty></tt>" - This type represents a list whose
264 elements are some other type. The contained type is arbitrary: it can even be
265 another list type.</li>
267 <li>Class type - Specifying a class name in a type context means that the
268 defined value must be a subclass of the specified class. This is useful in
269 conjunction with the "list" type, for example, to constrain the elements of the
270 list to a common base class (e.g., a <tt><b>list</b><Register></tt> can
271 only contain definitions derived from the "<tt>Register</tt>" class).</li>
273 <li>"<tt><b>code</b></tt>" - This represents a big hunk of text. NOTE: I don't
274 remember why this is distinct from string!</li>
276 <li>"<tt><b>dag</b></tt>" - This type represents a nestable directed graph of
280 <p>To date, these types have been sufficient for describing things that
281 TableGen has been used for, but it is straight-forward to extend this list if
286 <!-- -------------------------------------------------------------------------->
287 <div class="doc_subsubsection">
288 <a name="values">TableGen values and expressions</a>
291 <div class="doc_text">
293 <p>TableGen allows for a pretty reasonable number of different expression forms
294 when building up values. These forms allow the TableGen file to be written in a
295 natural syntax and flavor for the application. The current expression forms
296 supported include:</p>
299 <li><tt>?</tt> - uninitialized field</li>
300 <li><tt>0b1001011</tt> - binary integer value</li>
301 <li><tt>07654321</tt> - octal integer value (indicated by a leading 0)</li>
302 <li><tt>7</tt> - decimal integer value</li>
303 <li><tt>0x7F</tt> - hexadecimal integer value</li>
304 <li><tt>"foo"</tt> - string value</li>
305 <li><tt>[{ ... }]</tt> - code fragment</li>
306 <li><tt>[ X, Y, Z ]</tt> - list value.</li>
307 <li><tt>{ a, b, c }</tt> - initializer for a "bits<3>" value</li>
308 <li><tt>value</tt> - value reference</li>
309 <li><tt>value{17}</tt> - access to one bit of a value</li>
310 <li><tt>value{15-17}</tt> - access to multiple bits of a value</li>
311 <li><tt>DEF</tt> - reference to a record definition</li>
312 <li><tt>CLASS<val list></tt> - reference to a new anonymous definition of
313 CLASS with the specified template arguments.</li>
314 <li><tt>X.Y</tt> - reference to the subfield of a value</li>
315 <li><tt>list[4-7,17,2-3]</tt> - A slice of the 'list' list, including elements
316 4,5,6,7,17,2, and 3 from it. Elements may be included multiple times.</li>
317 <li><tt>(DEF a, b)</tt> - a dag value. The first element is required to be a
318 record definition, the remaining elements in the list may be arbitrary other
319 values, including nested `<tt>dag</tt>' values.</li>
322 <p>Note that all of the values have rules specifying how they convert to values
323 for different types. These rules allow you to assign a value like "7" to a
324 "bits<4>" value, for example.</p>
328 <!-- ======================================================================= -->
329 <div class="doc_subsection">
330 <a name="classesdefs">Classes and definitions</a>
333 <div class="doc_text">
335 <p>As mentioned in the <a href="#concepts">intro</a>, classes and definitions
336 (collectively known as 'records') in TableGen are the main high-level unit of
337 information that TableGen collects. Records are defined with a <tt>def</tt> or
338 <tt>class</tt> keyword, the record name, and an optional list of "<a
339 href="#templateargs">template arguments</a>". If the record has superclasses,
340 they are specified as a comma separated list that starts with a colon character
341 (":"). If <a href="#valuedef">value definitions</a> or <a href="#recordlet">let
342 expressions</a> are needed for the class, they are enclosed in curly braces
343 ("{}"); otherwise, the record ends with a semicolon. Here is a simple TableGen
347 <b>class</b> C { <b>bit</b> V = 1; }
350 <b>string</b> Greeting = "hello";
354 <p>This example defines two definitions, <tt>X</tt> and <tt>Y</tt>, both of
355 which derive from the <tt>C</tt> class. Because of this, they both get the
356 <tt>V</tt> bit value. The <tt>Y</tt> definition also gets the Greeting member
359 <p>In general, classes are useful for collecting together the commonality
360 between a group of records and isolating it in a single place. Also, classes
361 permit the specification of default values for their subclasses, allowing the
362 subclasses to override them as they wish.</p>
366 <!---------------------------------------------------------------------------->
367 <div class="doc_subsubsection">
368 <a name="valuedef">Value definitions</a>
371 <div class="doc_text">
372 <p>Value definitions define named entries in records. A value must be defined
373 before it can be referred to as the operand for another value definition or
374 before the value is reset with a <a href="#recordlet">let expression</a>. A
375 value is defined by specifying a <a href="#types">TableGen type</a> and a name.
376 If an initial value is available, it may be specified after the type with an
377 equal sign. Value definitions require terminating semicolons.</p>
380 <!-- -------------------------------------------------------------------------->
381 <div class="doc_subsubsection">
382 <a name="recordlet">'let' expressions</a>
385 <div class="doc_text">
386 <p>A record-level let expression is used to change the value of a value
387 definition in a record. This is primarily useful when a superclass defines a
388 value that a derived class or definition wants to override. Let expressions
389 consist of the '<tt>let</tt>' keyword followed by a value name, an equal sign
390 ("="), and a new value. For example, a new class could be added to the example
391 above, redefining the <tt>V</tt> field for all of its subclasses:</p>
394 <b>class</b> D : C { let V = 0; }
398 <p>In this case, the <tt>Z</tt> definition will have a zero value for its "V"
399 value, despite the fact that it derives (indirectly) from the <tt>C</tt> class,
400 because the <tt>D</tt> class overrode its value.</p>
404 <!-- -------------------------------------------------------------------------->
405 <div class="doc_subsubsection">
406 <a name="templateargs">Class template arguments</a>
409 <div class="doc_text">
410 <p>TableGen permits the definition of parameterized classes as well as normal
411 concrete classes. Parameterized TableGen classes specify a list of variable
412 bindings (which may optionally have defaults) that are bound when used. Here is
413 a simple example:</p>
416 <b>class</b> FPFormat<<b>bits</b><3> val> {
417 <b>bits</b><3> Value = val;
419 <b>def</b> NotFP : FPFormat<0>;
420 <b>def</b> ZeroArgFP : FPFormat<1>;
421 <b>def</b> OneArgFP : FPFormat<2>;
422 <b>def</b> OneArgFPRW : FPFormat<3>;
423 <b>def</b> TwoArgFP : FPFormat<4>;
424 <b>def</b> SpecialFP : FPFormat<5>;
427 <p>In this case, template arguments are used as a space efficient way to specify
428 a list of "enumeration values", each with a "Value" field set to the specified
431 <p>The more esoteric forms of <a href="#values">TableGen expressions</a> are
432 useful in conjunction with template arguments. As an example:</p>
435 <b>class</b> ModRefVal<<b>bits</b><2> val> {
436 <b>bits</b><2> Value = val;
439 <b>def</b> None : ModRefVal<0>;
440 <b>def</b> Mod : ModRefVal<1>;
441 <b>def</b> Ref : ModRefVal<2>;
442 <b>def</b> ModRef : ModRefVal<3>;
444 <b>class</b> Value<ModRefVal MR> {
445 <i>// decode some information into a more convenient format, while providing
446 // a nice interface to the user of the "Value" class.</i>
447 <b>bit</b> isMod = MR.Value{0};
448 <b>bit</b> isRef = MR.Value{1};
450 <i>// other stuff...</i>
453 <i>// Example uses</i>
454 <b>def</b> bork : Value<Mod>;
455 <b>def</b> zork : Value<Ref>;
456 <b>def</b> hork : Value<ModRef>;
459 <p>This is obviously a contrived example, but it shows how template arguments
460 can be used to decouple the interface provided to the user of the class from the
461 actual internal data representation expected by the class. In this case,
462 running <tt>tblgen</tt> on the example prints the following definitions:</p>
465 <b>def</b> bork { <i>// Value</i>
469 <b>def</b> hork { <i>// Value</i>
473 <b>def</b> zork { <i>// Value</i>
479 <p> This shows that TableGen was able to dig into the argument and extract a
480 piece of information that was requested by the designer of the "Value" class.
481 For more realistic examples, please see existing users of TableGen, such as the
486 <!-- ======================================================================= -->
487 <div class="doc_subsection">
488 <a name="filescope">File scope entities</a>
491 <!-- -------------------------------------------------------------------------->
492 <div class="doc_subsubsection">
493 <a name="include">File inclusion</a>
496 <div class="doc_text">
497 <p>TableGen supports the '<tt>include</tt>' token, which textually substitutes
498 the specified file in place of the include directive. The filename should be
499 specified as a double quoted string immediately after the '<tt>include</tt>'
500 keyword. Example:</p>
503 <b>include</b> "foo.td"
508 <!-- -------------------------------------------------------------------------->
509 <div class="doc_subsubsection">
510 <a name="globallet">'let' expressions</a>
513 <div class="doc_text">
514 <p> "let" expressions at file scope are similar to <a href="#recordlet">"let"
515 expressions within a record</a>, except they can specify a value binding for
516 multiple records at a time, and may be useful in certain other cases.
517 File-scope let expressions are really just another way that TableGen allows the
518 end-user to factor out commonality from the records.</p>
520 <p>File-scope "let" expressions take a comma-separated list of bindings to
521 apply, and one of more records to bind the values in. Here are some
525 <b>let</b> isTerminator = 1, isReturn = 1 <b>in</b>
526 <b>def</b> RET : X86Inst<"ret", 0xC3, RawFrm, NoArg>;
528 <b>let</b> isCall = 1 <b>in</b>
529 <i>// All calls clobber the non-callee saved registers...</i>
530 <b>let</b> Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6] in {
531 <b>def</b> CALLpcrel32 : X86Inst<"call", 0xE8, RawFrm, NoArg>;
532 <b>def</b> CALLr32 : X86Inst<"call", 0xFF, MRMS2r, Arg32>;
533 <b>def</b> CALLm32 : X86Inst<"call", 0xFF, MRMS2m, Arg32>;
537 <p>File-scope "let" expressions are often useful when a couple of definitions
538 need to be added to several records, and the records do not otherwise need to be
539 opened, as in the case with the CALL* instructions above.</p>
542 <!-- *********************************************************************** -->
543 <div class="doc_section"><a name="backends">TableGen backends</a></div>
544 <!-- *********************************************************************** -->
546 <div class="doc_text">
547 <p>How they work, how to write one. This section should not contain details
548 about any particular backend, except maybe -print-enums as an example. This
549 should highlight the APIs in <tt>TableGen/Record.h</tt>.</p>
552 <!-- *********************************************************************** -->
556 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
557 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
558 <a href="http://validator.w3.org/check/referer"><img
559 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
561 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
562 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
563 Last modified: $Date$