blob: f402361b0bcdba9df293c3c9461a8ea2c9a2b6de [file] [log] [blame]
Chris Lattnerb54c99c2004-02-06 05:42:53 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>TableGen Fundamentals</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7</head>
8<body>
9
10<div class="doc_title">TableGen Fundamentals</div>
11
12<ul>
13 <li><a href="#introduction">Introduction</a></li>
14 <ol>
15 <li><a href="#concepts">Basic concepts</a></li>
16 <li><a href="#example">An example record</a></li>
17 <li><a href="#running">Running TableGen</a></li>
18 </ol>
19 <li><a href="#syntax">TableGen syntax</a></li>
20 <ol>
21 <li><a href="#primitives">TableGen primitives</a></li>
22 <ol>
23 <li><a href="#comments">TableGen comments</a></li>
24 <li><a href="#types">The TableGen type system</a></li>
25 <li><a href="#values">TableGen values and expressions</a></li>
26 </ol>
27 <li><a href="#classesdefs">Classes and definitions</a></li>
28 <ol>
29 <li><a href="#valuedef">Value definitions</a></li>
30 <li><a href="#recordlet">'let' expressions</a></li>
31 <li><a href="#templateargs">Class template arguments</a></li>
32 </ol>
33 <li><a href="#filescope">File scope entities</a></li>
34 <ol>
35 <li><a href="#include">File inclusion</a></li>
36 <li><a href="#globallet">'let' expressions</a></li>
37 </ol>
38 </ol>
39 <li><a href="#backends">TableGen backends</a></li>
40 <ol>
41 <li><a href="#">x</a></li>
42 </ol>
43 <li><a href="#codegenerator">The LLVM code generator</a></li>
44 <ol>
45 <li><a href="#">x</a></li>
46 </ol>
47</ul>
48
49<!-- *********************************************************************** -->
50<div class="doc_section"><a name="introduction">Introduction</a></div>
51<!-- *********************************************************************** -->
52
53<div class="doc_text">
54
55<p>TableGen's purpose is to help a human develop and maintain records of
56domain-specific information. Because there may be a large number of these
57records, it is specifically designed to allow writing flexible descriptions and
58for common features of these records to be factored out. This reduces the
59amount of duplication in the description, reduces the chance of error, and
60makes it easier to structure domain specific information.</p>
61
62<p>The core part of TableGen <a href="#syntax">parses a file</a>, instantiates
63the declarations, and hands the result off to a domain-specific "<a
64href="#backends">TableGen backend</a>" for processing. The current major user
65of TableGen is the <a href="#codegenerator">LLVM code generator</a>.
66</p>
67
68</div>
69
70<!-- ======================================================================= -->
71<div class="doc_subsection">
72 <a name="running">Basic concepts</a>
73</div>
74
75<div class="doc_text">
76
77<p>
78TableGen files consist of two key parts: 'classes' and 'definitions', both of
79which are considered 'records'.
80</p>
81
82<p>
83<b>TableGen records</b> have a unique name, a list of values, and a list of
84superclasses. The list of values is main data that TableGen builds for each
85record, it is this that holds the domain specific information for the
86application. The interpretation of this data is left to a specific <a
87href="#backends">TableGen backend</a>, but the structure and format rules are
88taken care of and fixed by TableGen.
89</p>
90
91<p>
92<b>TableGen definitions</b> are the concrete form of 'records'. These generally
93do not have any undefined values, and are marked with the '<tt>def</tt>'
94keyword.
95</p>
96
97<p>
98<b>TableGen classes</b> are abstract records that are used to build and describe
99other records. These 'classes' allow the end-user to build abstractions for
100either the domain they are targetting (such as "Register", "RegisterClass", and
101"Instruction" in the LLVM code generator) or for the implementor to help factor
102out common properties of records (such as "FPInst", which is used to represent
103floating point instructions in the X86 backend). TableGen keeps track of all of
104the classes that are used to build up a definition, so the backend can find all
105definitions of a particular class, such as "Instruction".
106</p>
107
108</div>
109
110<!-- ======================================================================= -->
111<div class="doc_subsection">
112 <a name="example">An example record</a>
113</div>
114
115<div class="doc_text">
116
117<p>
118With no other arguments, TableGen parses the specified file and prints out all
119of the classes, then all of the definitions. This is a good way to see what the
120various definitions expand to fully. Running this on the <tt>X86.td</tt> file
121prints this (at the time of this writing):
122</p>
123
124<p>
125<pre>
126...
127def ADDrr8 { // Instruction X86Inst I2A8 Pattern
128 string Name = "add";
129 string Namespace = "X86";
130 list&lt;Register&gt; Uses = [];
131 list&lt;Register&gt; Defs = [];
132 bit isReturn = 0;
133 bit isBranch = 0;
134 bit isCall = 0;
135 bit isTwoAddress = 1;
136 bit isTerminator = 0;
137 dag Pattern = (set R8, (plus R8, R8));
138 bits&lt;8&gt; Opcode = { 0, 0, 0, 0, 0, 0, 0, 0 };
139 Format Form = MRMDestReg;
140 bits&lt;5&gt; FormBits = { 0, 0, 0, 1, 1 };
141 ArgType Type = Arg8;
142 bits&lt;3&gt; TypeBits = { 0, 0, 1 };
143 bit hasOpSizePrefix = 0;
144 bit printImplicitUses = 0;
145 bits&lt;4&gt; Prefix = { 0, 0, 0, 0 };
146 FPFormat FPForm = ?;
147 bits&lt;3&gt; FPFormBits = { 0, 0, 0 };
148}
149...
150</pre><p>
151
152<p>
153This definition corresponds to an 8-bit register-register add instruction in the
154X86. The string after the '<tt>def</tt>' string indicates the name of the
155record ("<tt>ADDrr8</tt>" in this case), and the comment at the end of the line
156indicates the superclasses of the definition. The body of the record contains
157all of the data that TableGen assembled for the record, indicating that the
158instruction is part of the "X86" namespace, should be printed as "<tt>add</tt>"
159in the assembly file, it is a two-address instruction, has a particular
160encoding, etc. The contents and semantics of the information in the record is
161specific to the needs of the X86 backend, and is only shown as an example.
162</p>
163
164<p>
165As you can see, a lot of information is needed for every instruction supported
166by the code generator, and specifying it all manually would be unmaintainble,
167prone to bugs, and tiring to do in the first place. Because we are using
168TableGen, all of the information was derived from the following definition:
169</p>
170
171<p><pre>
172def ADDrr8 : I2A8&lt;"add", 0x00, MRMDestReg&gt;,
173 Pattern&lt;(set R8, (plus R8, R8))&gt;;
174</pre></p>
175
176<p>
177This definition makes use of the custom I2A8 (two address instruction with 8-bit
178operand) class, which is defined in the X86-specific TableGen file to factor out
179the common features that instructions of its class share. A key feature of
180TableGen is that it allows the end-user to define the abstractions they prefer
181to use when describing their information.
182</p>
183
184</div>
185
186<!-- ======================================================================= -->
187<div class="doc_subsection">
188 <a name="running">Running TableGen</a>
189</div>
190
191<div class="doc_text">
192
193<p>
194TableGen runs just like any other LLVM tool. The first (optional) argument
195specifies the file to read. If a filename is not specified, <tt>tblgen</tt>
196reads from standard input.
197</p>
198
199<p>
200To be useful, one of the <a href="#backends">TableGen backends</a> must be used.
201These backends are selectable on the command line (type '<tt>tblgen --help</tt>'
202for a list). For example, to get a list of all of the definitions that subclass
203a particular type (which can be useful for building up an enum list of these
204records), use the <tt>--print-enums</tt> option:
205</p>
206
207<p><pre>
208$ tblgen X86.td -print-enums -class=Register
209AH, AL, AX, BH, BL, BP, BX, CH, CL, CX, DH, DI, DL, DX,
210EAX, EBP, EBX, ECX, EDI, EDX, ESI, ESP, FP0, FP1, FP2, FP3, FP4, FP5, FP6,
211SI, SP, ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
212
213$ tblgen X86.td -print-enums -class=Instruction
214ADCrr32, ADDri16, ADDri16b, ADDri32, ADDri32b, ADDri8, ADDrr16, ADDrr32,
215ADDrr8, ADJCALLSTACKDOWN, ADJCALLSTACKUP, ANDri16, ANDri16b, ANDri32, ANDri32b,
216ANDri8, ANDrr16, ANDrr32, ANDrr8, BSWAPr32, CALLm32, CALLpcrel32, ...
217</pre></p>
218
219<p>
220The default backend prints out all of the records, as described <a
221href="#example">above</a>.
222</p>
223
224<p>
225If you plan to use TableGen for some purpose, you will most likely have to <a
226href="#backends">write a backend</a> that extracts the information specific to
227what you need and formats it in the appropriate way.
228</p>
229
230</div>
231
232
233<!-- *********************************************************************** -->
234<div class="doc_section"><a name="syntax">TableGen syntax</a></div>
235<!-- *********************************************************************** -->
236
237<div class="doc_text">
238
239<p>
240TableGen doesn't care about the meaning of data (that is up to the backend to
241define), but it does care about syntax, and it enforces a simple type system.
242This section describes the syntax and the constructs allowed in a TableGen file.
243</p>
244
245</div>
246
247<!-- ======================================================================= -->
248<div class="doc_subsection">
249 <a name="primitives">TableGen primitives</tt></a>
250</div>
251
252<!----------------------------------------------------------------------------->
253<div class="doc_subsubsection">
254 <a name="comments">TableGen comments</tt></a>
255</div>
256
257<div class="doc_text">
258
259<p>TableGen supports BCPL style "<tt>//</tt>" comments, which run to the end of
260the line, and it also supports <b>nestable</b> "<tt>/* */</tt>" comments.</p>
261
262</div>
263
264
265<!----------------------------------------------------------------------------->
266<div class="doc_subsubsection">
267 <a name="types">The TableGen type system</tt></a>
268</div>
269
270<div class="doc_text">
271<p>
272TableGen files are strongly typed, in a simple (but complete) type-system.
273These types are used to perform automatic conversions, check for errors, and to
274help interface designers constrain the input that they allow. Every <a
275href="#valuedef">value definition</a> is required to have an associated type.
276</p>
277
278<p>
279TableGen supports a mixture of very low-level types (such as <tt>bit</tt>) and
280very high-level types (such as <tt>dag</tt>). This flexibility is what allows
281it to describe a wide range of information conveniently and compactly. The
282TableGen types are:
283</p>
284
285<p>
286<ul>
287<li>"<tt>bit</tt>" - A 'bit' is a boolean value that can hold either 0 or
2881.</li>
289
290<li>"<tt>int</tt>" - The 'int' type represents a simple 32-bit integer value, such as 5.</li>
291
292<li>"<tt>string</tt>" - The 'string' type represents an ordered sequence of
293characters of arbitrary length.</li>
294
295<li>"<tt>bits&lt;n&gt;</tt>" - A 'bits' type is a arbitrary, but fixed, size
296integer that is broken up into individual bits. This type is useful because it
297can handle some bits being defined while others are undefined.</li>
298
299<li>"<tt>list&lt;ty&gt;</tt>" - This type represents a list whose elements are
300some other type. The contained type is arbitrary: it can even be another list
301type.</li>
302
303<li>Class type - Specifying a class name in a type context means that the
304defined value must be a subclass of the specified class. This is useful in
305conjunction with the "list" type, for example, to constrain the elements of the
306list to a common base class (e.g., a <tt>list&lt;Register&gt;</tt> can only
307contain definitions derived from the "<tt>Register</tt>" class).</li>
308
309<li>"<tt>code</tt>" - This represents a big hunk of text. NOTE: I don't
310remember why this is distinct from string!</li>
311
312<li>"<tt>dag</tt>" - This type represents a nestable directed graph of
313elements.</li>
314</ul>
315</p>
316
317<p>
318To date, these types have been sufficient for describing things that TableGen
319has been used for, but it is straight-forward to extend this list if needed.
320</p>
321
322</div>
323
324<!----------------------------------------------------------------------------->
325<div class="doc_subsubsection">
326 <a name="values">TableGen values and expressions</tt></a>
327</div>
328
329<div>
330<p>
331TableGen allows for a pretty reasonable number of different expression forms
332when building up values. These forms allow the TableGen file to be written in a
333natural syntax and flavor for the application. The current expression forms
334supported include:
335</p>
336
337<p><ul>
338<li>? - Uninitialized field.</li>
339<li>0b1001011 - Binary integer value.</li>
340<li>07654321 - Octal integer value (indicated by a leading 0).</li>
341<li>7 - Decimal integer value.</li>
342<li>0x7F - Hexadecimal integer value.</li>
343<li>"foo" - String value.</li>
344<li>[{ .... }] - Code fragment.</li>
345<li>[ X, Y, Z ] - List value.</li>
346<li>{ a, b, c } - Initializer for a "bits&lt;3&gt;" value.</li>
347<li>value - Value reference.</li>
348<li>value{17} - Access to one or more bits of a value.</li>
349<li>DEF - Reference to a record definition.</li>
350<li>X.Y - Reference to the subfield of a value.</li>
351
352<li>(DEF a, b) - A dag value. The first element is required to be a record
353definition, the remaining elements in the list may be arbitrary other values,
354including nested 'dag' values.</li>
355
356</ul></p>
357
358<p>
359Note that all of the values have rules specifying how they convert to to values
360for different types. These rules allow you to assign a value like "7" to a
361"bits&lt;4&gt;" value, for example.
362</p>
363
364
365
366</div>
367
368
369<!-- ======================================================================= -->
370<div class="doc_subsection">
371 <a name="classesdefs">Classes and definitions</tt></a>
372</div>
373
374<div>
375<p>
376As mentioned in the <a href="#concepts">intro</a>, classes and definitions
377(collectively known as 'records') in TableGen are the main high-level unit of
378information that TableGen collects. Records are defined with a <tt>def</tt> or
379<tt>class</tt> keyword, the record name, and an optional list of "<a
380href="templateargs">template arguments</a>". If the record has superclasses,
381they are specified as a comma seperated list that starts with a colon character
382(":"). If <a href="#valuedef">value definitions</a> or <a href="#recordlet">let
383expressions</a> are needed for the class they are enclosed in curly braces
384("{}"), otherwise the record ends with a semicolon. Here is a simple TableGen
385file:
386</p>
387
388<p><pre>
389class C { bit V = 1; }
390def X : C;
391def Y : C {
392 string Greeting = "hello";
393}
394</pre></p>
395
396<p>
397This example defines two definitions, <tt>X</tt> and <tt>Y</tt>, both of which
398derive from the <tt>C</tt> class. Because of this, they both get the <tt>V</tt>
399bit value. The <tt>Y</tt> definition also gets the Greeting member as well.
400</p>
401
402</div>
403
404<!----------------------------------------------------------------------------->
405<div class="doc_subsubsection">
406 <a name="valuedef">Value definitions</tt></a>
407</div>
408
409<div class="doc_text">
410<p>
411Value definitions define named entries in records. A value must be defined
412before it can be referred to as the operand for another value definition, or
413before the value is reset with a <a href="#recordlet">let expression</a>. A
414value is defined by specifying a <a href="#types">TableGen type</a> and a name.
415If an initial value is available, it may be specified after the type with an
416equal sign. Value definitions require terminating semicolons.
417</div>
418
419<!----------------------------------------------------------------------------->
420<div class="doc_subsubsection">
421 <a name="recordlet">'let' expressions</tt></a>
422</div>
423
424<div class="doc_text">
425<p>
426A record-level let expression is used to change the value of a value definition
427in a record. This is primarily useful when a superclass defines a value that a
428derived class or definitions wants to override. Let expressions consist of the
429'<tt>let</tt>' keyword, followed by a value name, an equal sign ("="), and a new
430value for example, a new class could be added to the example above, redefining
431the <tt>V</tt> field for all of its subclasses:</p>
432
433<p><pre>
434class D : C { let V = 0; }
435def Z : D;
436</pre></p>
437
438<p>
439In this case, the <tt>Z</tt> definition will have a zero value for its "V"
440value, despite the fact that it derives (indirectly) from the <tt>C</tt> class,
441because the <tt>D</tt> class overrode its value.
442</p>
443
444</div>
445
446<!----------------------------------------------------------------------------->
447<div class="doc_subsubsection">
448 <a name="templateargs">Class template arguments</tt></a>
449</div>
450
451<div class="doc_text">
452and default values...
453</div>
454
455
456
457<!-- ======================================================================= -->
458<div class="doc_subsection">
459 <a name="filescope">File scope entities</tt></a>
460</div>
461
462<!----------------------------------------------------------------------------->
463<div class="doc_subsubsection">
464 <a name="include">File inclusion</tt></a>
465</div>
466
467<div class="doc_text">
468<p>
469TableGen supports the '<tt>include</tt>' token, which textually substitutes the
470specified file in place of the include directive. The filename should be
471specified as a double quoted string immediately after the '<tt>include</tt>'
472keyword. Example:
473
474<p><pre>
475 include "foo.td"
476</pre></p>
477
478</div>
479
480<!----------------------------------------------------------------------------->
481<div class="doc_subsubsection">
482 <a name="globallet">'let' expressions</tt></a>
483</div>
484
485<div class="doc_text">
486<p>
487"let" expressions at file scope are similar to <a href="#recordlet">"let"
488expressions within a record</a>, except they can specify a value binding for
489multiple records at a time, and may be useful in certain other cases.
490File-scope let expressions are really just another way that TableGen allows the
491end-user to factor out commonality from the records.
492</p>
493
494<p>
495File-scope "let" expressions take a comma-seperated list of bindings to apply,
496and one of more records to bind the values in. Here are some examples:
497</p>
498
499<p><pre>
500let isTerminator = 1, isReturn = 1 in
501 def RET : X86Inst&lt;"ret", 0xC3, RawFrm, NoArg&gt;;
502
503let isCall = 1 in
504 // All calls clobber the non-callee saved registers...
505 let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6] in {
506 def CALLpcrel32 : X86Inst&lt;"call", 0xE8, RawFrm, NoArg&gt;;
507 def CALLr32 : X86Inst&lt;"call", 0xFF, MRMS2r, Arg32&gt;;
508 def CALLm32 : X86Inst&lt;"call", 0xFF, MRMS2m, Arg32&gt;;
509 }
510</pre></p>
511
512<p>
513File-scope "let" expressions are often useful when a couple of definitions need
514to be added to several records, and the records do not otherwise need to be
515opened, as in the case with the CALL* instructions above.
516</p>
517</div>
518
519
520<!-- *********************************************************************** -->
521<div class="doc_section"><a name="backends">TableGen backends</a></div>
522<!-- *********************************************************************** -->
523
524<div class="doc_text">
525
526<p>
527How they work, how to write one. This section should not contain details about
528any particular backend, except maybe -print-enums as an example. This should
529highlight the APIs in TableGen/Record.h.
530</p>
531
532</div>
533
534
535<!-- *********************************************************************** -->
536<div class="doc_section"><a name="codegenerator">The LLVM code generator</a></div>
537<!-- *********************************************************************** -->
538
539<div class="doc_text">
540
541<p>
542This is just a temporary, convenient, place to put stuff about the code
543generator before it gets its own document. This should describe all of the
544tablegen backends used by the code generator and the classes/definitions they
545expect.
546</p>
547
548</div>
549
550
551
552<!-- *********************************************************************** -->
553<hr>
554<div class="doc_footer">
555 <address><a href="mailto:sabre@nondot.org">Chris Lattner</a></address>
556 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
557 <br>
558 Last modified: $Date$
559</div>
560
561</body>
562</html>