blob: 3c3bc1d60b1a24f7b115c04861992fe405e074d8 [file] [log] [blame]
Reid Spencere831f0b2004-07-29 00:13:04 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
Reid Spencer9acd4122004-05-22 02:28:36 +00002<html>
3<head>
Reid Spencer375d7842004-08-18 20:06:19 +00004 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
Reid Spencer9acd4122004-05-22 02:28:36 +00005 <title>LLVM Bytecode File Format</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
Reid Spencer4e0352d2004-07-05 08:18:07 +00007 <style type="text/css">
Reid Spencer435a1ab2004-08-03 20:21:05 +00008 TR, TD { border: 2px solid gray; padding-left: 4pt; padding-right: 4pt;
9 padding-top: 2pt; padding-bottom: 2pt; }
Reid Spencer4e0352d2004-07-05 08:18:07 +000010 TH { border: 2px solid gray; font-weight: bold; font-size: 105%; }
Reid Spencer9ce11e42004-07-05 19:04:27 +000011 TABLE { text-align: center; border: 2px solid black;
Reid Spencer435a1ab2004-08-03 20:21:05 +000012 border-collapse: collapse; margin-top: 1em; margin-left: 1em;
13 margin-right: 1em; margin-bottom: 1em; }
Reid Spencer9ce11e42004-07-05 19:04:27 +000014 .td_left { border: 2px solid gray; text-align: left; }
Reid Spencer9acd4122004-05-22 02:28:36 +000015 </style>
16</head>
17<body>
Reid Spencere831f0b2004-07-29 00:13:04 +000018<div class="doc_title"> LLVM Bytecode File Format </div>
Reid Spencer9acd4122004-05-22 02:28:36 +000019<ol>
20 <li><a href="#abstract">Abstract</a></li>
Reid Spencer4e0352d2004-07-05 08:18:07 +000021 <li><a href="#concepts">Concepts</a>
Reid Spencer9acd4122004-05-22 02:28:36 +000022 <ol>
23 <li><a href="#blocks">Blocks</a></li>
24 <li><a href="#lists">Lists</a></li>
25 <li><a href="#fields">Fields</a></li>
26 <li><a href="#align">Alignment</a></li>
Reid Spencer8e32af02004-07-07 13:34:26 +000027 <li><a href="#vbr">Variable Bit-Rate Encoding</a></li>
Reid Spencer4e0352d2004-07-05 08:18:07 +000028 <li><a href="#encoding">Encoding Primitives</a></li>
29 <li><a href="#slots">Slots</a></li>
30 </ol>
31 </li>
Reid Spencere2f97b32004-07-05 22:28:02 +000032 <li><a href="#general">General Structure</a> </li>
33 <li><a href="#blockdefs">Block Definitions</a>
Reid Spencer4e0352d2004-07-05 08:18:07 +000034 <ol>
Reid Spencereca340b2004-05-23 17:05:09 +000035 <li><a href="#signature">Signature Block</a></li>
36 <li><a href="#module">Module Block</a></li>
Reid Spencer4e0352d2004-07-05 08:18:07 +000037 <li><a href="#globaltypes">Global Type Pool</a></li>
38 <li><a href="#globalinfo">Module Info Block</a></li>
39 <li><a href="#constantpool">Global Constant Pool</a></li>
40 <li><a href="#functiondefs">Function Definition</a></li>
41 <li><a href="#compactiontable">Compaction Table</a></li>
Chris Lattnerdb137582005-11-05 22:20:06 +000042 <li><a href="#instructionlist">Instructions List</a></li>
43 <li><a href="#instructions">Instructions</a></li>
Reid Spencer4e0352d2004-07-05 08:18:07 +000044 <li><a href="#symtab">Symbol Table</a></li>
Reid Spencer9acd4122004-05-22 02:28:36 +000045 </ol>
46 </li>
Reid Spencerb14a0a62004-06-08 07:41:41 +000047 <li><a href="#versiondiffs">Version Differences</a>
48 <ol>
Chris Lattnercd196a92004-10-16 18:03:55 +000049 <li><a href="#vers13">Version 1.3 Differences From 1.4</a></li>
Reid Spencerb14a0a62004-06-08 07:41:41 +000050 <li><a href="#vers12">Version 1.2 Differences From 1.3</a></li>
51 <li><a href="#vers11">Version 1.1 Differences From 1.2</a></li>
52 <li><a href="#vers10">Version 1.0 Differences From 1.1</a></li>
53 </ol>
54 </li>
Reid Spencer9acd4122004-05-22 02:28:36 +000055</ol>
Chris Lattnere7d5ec22004-05-25 17:44:58 +000056<div class="doc_author">
57<p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
58</p>
Reid Spencer9acd4122004-05-22 02:28:36 +000059</div>
60<!-- *********************************************************************** -->
61<div class="doc_section"> <a name="abstract">Abstract </a></div>
62<!-- *********************************************************************** -->
63<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +000064<p>This document describes the LLVM bytecode file format. It specifies
65the binary encoding rules of the bytecode file format so that
66equivalent systems can encode bytecode files correctly. The LLVM
67bytecode representation is used to store the intermediate
68representation on disk in compacted form.</p>
69<p>The LLVM bytecode format may change in the future, but LLVM will
70always be backwards compatible with older formats. This document will
71only describe the most current version of the bytecode format. See <a
72 href="#versiondiffs">Version Differences</a> for the details on how
73the current version is different from previous versions.</p>
Reid Spencer9acd4122004-05-22 02:28:36 +000074</div>
75<!-- *********************************************************************** -->
Reid Spencer4e0352d2004-07-05 08:18:07 +000076<div class="doc_section"> <a name="concepts">Concepts</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +000077<!-- *********************************************************************** -->
78<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +000079<p>This section describes the general concepts of the bytecode file
80format without getting into specific layout details. It is recommended
81that you read this section thoroughly before interpreting the detailed
82descriptions.</p>
Reid Spencer9acd4122004-05-22 02:28:36 +000083</div>
84<!-- _______________________________________________________________________ -->
85<div class="doc_subsection"><a name="blocks">Blocks</a> </div>
86<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +000087<p>LLVM bytecode files consist simply of a sequence of blocks of bytes
88using a binary encoding Each block begins with an header of two
89unsigned integers. The first value identifies the type of block and the
90second value provides the size of the block in bytes. The block
91identifier is used because it is possible for entire blocks to be
92omitted from the file if they are empty. The block identifier helps the
93reader determine which kind of block is next in the file. Note that
94blocks can be nested within other blocks.</p>
95<p> All blocks are variable length, and the block header specifies the
96size of the block. All blocks begin on a byte index that is aligned to
97an even 32-bit boundary. That is, the first block is 32-bit aligned
98because it starts at offset 0. Each block is padded with zero fill
99bytes to ensure that the next block also starts on a 32-bit boundary.</p>
Reid Spencer9acd4122004-05-22 02:28:36 +0000100</div>
101<!-- _______________________________________________________________________ -->
102<div class="doc_subsection"><a name="lists">Lists</a> </div>
103<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000104<p>LLVM Bytecode blocks often contain lists of things of a similar
105type. For example, a function contains a list of instructions and a
106function type contains a list of argument types. There are two basic
107types of lists: length lists (<a href="#llist">llist</a>), and null
108terminated lists (<a href="#zlist">zlist</a>), as described below in
109the <a href="#encoding">Encoding Primitives</a>.</p>
Reid Spencer9acd4122004-05-22 02:28:36 +0000110</div>
111<!-- _______________________________________________________________________ -->
112<div class="doc_subsection"><a name="fields">Fields</a> </div>
113<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000114<p>Fields are units of information that LLVM knows how to write atomically. Most
115fields have a uniform length or some kind of length indication built into their
116encoding. For example, a constant string (array of bytes) is written simply as
117the length followed by the characters. Although this is similar to a list,
118constant strings are treated atomically and are thus fields.</p>
Reid Spencer9acd4122004-05-22 02:28:36 +0000119<p>Fields use a condensed bit format specific to the type of information
120they must contain. As few bits as possible are written for each field. The
Reid Spencere831f0b2004-07-29 00:13:04 +0000121sections that follow will provide the details on how these fields are
Reid Spencer9acd4122004-05-22 02:28:36 +0000122written and how the bits are to be interpreted.</p>
123</div>
124<!-- _______________________________________________________________________ -->
Reid Spencer4e0352d2004-07-05 08:18:07 +0000125<div class="doc_subsection"><a name="align">Alignment</a> </div>
Reid Spencer49ad21c2004-05-25 15:47:57 +0000126<div class="doc_text">
Reid Spencer40e93a42004-08-03 20:33:56 +0000127 <p>To support cross-platform differences, the bytecode file is aligned on
128 certain boundaries. This means that a small amount of padding (at most 3
129 bytes) will be added to ensure that the next entry is aligned to a 32-bit
130 boundary.</p>
Chris Lattnere7d5ec22004-05-25 17:44:58 +0000131</div>
Reid Spencer49ad21c2004-05-25 15:47:57 +0000132<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +0000133<div class="doc_subsection"><a name="vbr">Variable Bit-Rate Encoding</a>
Reid Spencer8e32af02004-07-07 13:34:26 +0000134</div>
Reid Spencere831f0b2004-07-29 00:13:04 +0000135<div class="doc_text">
136<p>Most of the values written to LLVM bytecode files are small integers. To
137minimize the number of bytes written for these quantities, an encoding scheme
138similar to UTF-8 is used to write integer data. The scheme is known as
139variable bit rate (vbr) encoding. In this encoding, the high bit of
140each byte is used to indicate if more bytes follow. If (byte &amp;
1410x80) is non-zero in any given byte, it means there is another byte
142immediately following that also contributes to the value. For the final
143byte (byte &amp; 0x80) is false (the high bit is not set). In each byte
144only the low seven bits contribute to the value. Consequently 32-bit
145quantities can take from one to <em>five</em> bytes to encode. In
146general, smaller quantities will encode in fewer bytes, as follows:</p>
147<table>
148 <tbody>
149 <tr>
150 <th>Byte #</th>
151 <th>Significant Bits</th>
152 <th>Maximum Value</th>
153 </tr>
154 <tr>
155 <td>1</td>
156 <td>0-6</td>
157 <td>127</td>
158 </tr>
159 <tr>
160 <td>2</td>
161 <td>7-13</td>
162 <td>16,383</td>
163 </tr>
164 <tr>
165 <td>3</td>
166 <td>14-20</td>
167 <td>2,097,151</td>
168 </tr>
169 <tr>
170 <td>4</td>
171 <td>21-27</td>
172 <td>268,435,455</td>
173 </tr>
174 <tr>
175 <td>5</td>
176 <td>28-34</td>
177 <td>34,359,738,367</td>
178 </tr>
179 <tr>
180 <td>6</td>
181 <td>35-41</td>
182 <td>4,398,046,511,103</td>
183 </tr>
184 <tr>
185 <td>7</td>
186 <td>42-48</td>
187 <td>562,949,953,421,311</td>
188 </tr>
189 <tr>
190 <td>8</td>
191 <td>49-55</td>
192 <td>72,057,594,037,927,935</td>
193 </tr>
194 <tr>
195 <td>9</td>
196 <td>56-62</td>
197 <td>9,223,372,036,854,775,807</td>
198 </tr>
199 <tr>
200 <td>10</td>
201 <td>63-69</td>
202 <td>1,180,591,620,717,411,303,423</td>
203 </tr>
204 </tbody>
205</table>
206<p>Note that in practice, the tenth byte could only encode bit 63 since
207the maximum quantity to use this encoding is a 64-bit integer.</p>
208<p><em>Signed</em> VBR values are encoded with the standard vbr
209encoding, but with the sign bit as the low order bit instead of the
210high order bit. This allows small negative quantities to be encoded
211efficiently. For example, -3
212is encoded as "((3 &lt;&lt; 1) | 1)" and 3 is encoded as "(3 &lt;&lt;
2131) | 0)", emitted with the standard vbr encoding above.</p>
214</div>
Reid Spencer8e32af02004-07-07 13:34:26 +0000215<!-- _______________________________________________________________________ -->
216<div class="doc_subsection"><a name="encoding">Encoding Primitives</a> </div>
217<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000218<p>Each field in the bytecode format is encoded into the file using a
219small set of primitive formats. The table below defines the encoding
220rules for the various primitives used and gives them each a type name.
221The type names used in the descriptions of blocks and fields in the <a
222 href="#details">Detailed Layout</a>next section. Any type name with
223the suffix <em>_vbr</em> indicates a quantity that is encoded using
224variable bit rate encoding as described above.</p>
225<table class="doc_table">
226 <tbody>
227 <tr>
228 <th><b>Type</b></th>
229 <th class="td_left"><b>Rule</b></th>
230 </tr>
231 <tr>
232 <td><a name="unsigned"><b>unsigned</b></a></td>
233 <td class="td_left">A 32-bit unsigned integer that always occupies four
Reid Spencereca340b2004-05-23 17:05:09 +0000234 consecutive bytes. The unsigned integer is encoded using LSB first
235 ordering. That is bits 2<sup>0</sup> through 2<sup>7</sup> are in the
236 byte with the lowest file offset (little endian).</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000237 </tr>
238 <tr>
Reid Spencer02d7f922004-08-03 20:57:56 +0000239 <td style="vertical-align: top;"><a name="uint24_vbr">
240 <b>uint24_vbr</b></a></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000241 <td style="vertical-align: top; text-align: left;">A 24-bit unsigned
242 integer that occupies from one to four bytes using variable bit rate
243 encoding.</td>
244 </tr>
245 <tr>
246 <td><a name="uint32_vbr"><b>uint32_vbr</b></a></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000247 <td class="td_left">A 32-bit unsigned integer that occupies from one to
248 five bytes using variable bit rate encoding.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000249 </tr>
250 <tr>
251 <td><a name="uint64_vbr"><b>uint64_vbr</b></a></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000252 <td class="td_left">A 64-bit unsigned integer that occupies from one to ten
253 bytes using variable bit rate encoding.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000254 </tr>
255 <tr>
256 <td><a name="int64_vbr"><b>int64_vbr</b></a></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000257 <td class="td_left">A 64-bit signed integer that occupies from one to ten
258 bytes using the signed variable bit rate encoding.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000259 </tr>
260 <tr>
261 <td><a name="char"><b>char</b></a></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000262 <td class="td_left">A single unsigned character encoded into one byte</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000263 </tr>
264 <tr>
265 <td><a name="bit"><b>bit(n-m)</b></a></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000266 <td class="td_left">A set of bit within some larger integer field. The values
267 of <code>n</code> and <code>m</code> specify the inclusive range of bits
268 that define the subfield. The value for <code>m</code> may be omitted if
269 its the same as <code>n</code>.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000270 </tr>
271 <tr>
272 <td style="vertical-align: top;"><b><a name="float"><b>float</b></a></b></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000273 <td style="vertical-align: top; text-align: left;">A floating point value encoded
274 as a 32-bit IEEE value written in little-endian form.<br>
Reid Spencere831f0b2004-07-29 00:13:04 +0000275 </td>
276 </tr>
277 <tr>
278 <td style="vertical-align: top;"><b><b><a name="double"><b>double</b></a></b></b></td>
Reid Spencera54f1102004-08-03 19:20:18 +0000279 <td style="vertical-align: top; text-align: left;">A floating point value encoded
280 as a64-bit IEEE value written in little-endian form</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000281 </tr>
282 <tr>
283 <td><a name="string"><b>string</b></a></td>
284 <td class="td_left">A uint32_vbr indicating the type of the
285constant string which also includes its length, immediately followed by
286the characters of the string. There is no terminating null byte in the
287string.</td>
288 </tr>
289 <tr>
290 <td><a name="data"><b>data</b></a></td>
291 <td class="td_left">An arbitrarily long segment of data to which
292no interpretation is implied. This is used for constant initializers.<br>
293 </td>
294 </tr>
295 <tr>
296 <td><a name="llist"><b>llist(x)</b></a></td>
297 <td class="td_left">A length list of x. This means the list is
298encoded as an <a href="#uint32_vbr">uint32_vbr</a> providing the
299length of the list, followed by a sequence of that many "x" items. This
300implies that the reader should iterate the number of times provided by
301the length.</td>
302 </tr>
303 <tr>
304 <td><a name="zlist"><b>zlist(x)</b></a></td>
305 <td class="td_left">A zero-terminated list of x. This means the
306list is encoded as a sequence of an indeterminate number of "x" items,
307followed by an <a href="#uint32_vbr">uint32_vbr</a> terminating value.
308This implies that none of the "x" items can have a zero value (or else
309the list terminates).</td>
310 </tr>
311 <tr>
312 <td><a name="block"><b>block</b></a></td>
313 <td class="td_left">A block of data that is logically related. A
314block is an unsigned 32-bit integer that encodes the type of the block
315in the low 5 bits and the size of the block in the high 27 bits. The
316length does not include the block header or any alignment bytes at the
317end of the block. Blocks may compose other blocks. </td>
318 </tr>
319 </tbody>
Reid Spencereca340b2004-05-23 17:05:09 +0000320</table>
321</div>
322<!-- _______________________________________________________________________ -->
Reid Spencer8e32af02004-07-07 13:34:26 +0000323<div class="doc_subsection"><a name="notation">Field Notation</a> </div>
324<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000325<p>In the detailed block and field descriptions that follow, a regex
326like notation is used to describe optional and repeated fields. A very
327limited subset of regex is used to describe these, as given in the
328following table: </p>
329<table class="doc_table">
330 <tbody>
Reid Spencer8e32af02004-07-07 13:34:26 +0000331 <tr>
332 <th><b>Character</b></th>
333 <th class="td_left"><b>Meaning</b></th>
Reid Spencere831f0b2004-07-29 00:13:04 +0000334 </tr>
335 <tr>
Reid Spencer8e32af02004-07-07 13:34:26 +0000336 <td><b><code>?</code></b></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000337 <td class="td_left">The question mark indicates 0 or 1
338occurrences of the thing preceding it.</td>
339 </tr>
340 <tr>
Reid Spencer8e32af02004-07-07 13:34:26 +0000341 <td><b><code>*</code></b></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000342 <td class="td_left">The asterisk indicates 0 or more occurrences
343of the thing preceding it.</td>
344 </tr>
345 <tr>
Reid Spencer8e32af02004-07-07 13:34:26 +0000346 <td><b><code>+</code></b></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000347 <td class="td_left">The plus sign indicates 1 or more occurrences
348of the thing preceding it.</td>
349 </tr>
350 <tr>
Reid Spencer8e32af02004-07-07 13:34:26 +0000351 <td><b><code>()</code></b></td>
352 <td class="td_left">Parentheses are used for grouping.</td>
Reid Spencer8e32af02004-07-07 13:34:26 +0000353 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +0000354 <tr>
355 <td><b><code>,</code></b></td>
356 <td class="td_left">The comma separates sequential fields.</td>
357 </tr>
358 </tbody>
359</table>
360<p>So, for example, consider the following specifications:</p>
361<div class="doc_code">
362<ol>
363 <li><code>string?</code></li>
364 <li><code>(uint32_vbr,uin32_vbr)+</code></li>
365 <li><code>(unsigned?,uint32_vbr)*</code></li>
366 <li><code>(llist(unsigned))?</code></li>
367</ol>
Reid Spencer8e32af02004-07-07 13:34:26 +0000368</div>
Reid Spencere831f0b2004-07-29 00:13:04 +0000369<p>with the following interpretations:</p>
370<ol>
371 <li>An optional string. Matches either nothing or a single string</li>
372 <li>One or more pairs of uint32_vbr.</li>
373 <li>Zero or more occurrences of either an unsigned followed by a
374uint32_vbr or just a uint32_vbr.</li>
375 <li>An optional length list of unsigned values.</li>
376</ol>
377</div>
Reid Spencer8e32af02004-07-07 13:34:26 +0000378<!-- _______________________________________________________________________ -->
Reid Spencer4e0352d2004-07-05 08:18:07 +0000379<div class="doc_subsection"><a name="slots">Slots</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000380<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000381<p>The bytecode format uses the notion of a "slot" to reference Types
382and Values. Since the bytecode file is a <em>direct</em> representation of
383LLVM's intermediate representation, there is a need to represent pointers in
384the file. Slots are used for this purpose. For example, if one has the following
385assembly:
Reid Spencer4e0352d2004-07-05 08:18:07 +0000386</p>
Reid Spencere831f0b2004-07-29 00:13:04 +0000387<div class="doc_code"><code> %MyType = type { int, sbyte }<br>
388%MyVar = external global %MyType
Reid Spencer8e32af02004-07-07 13:34:26 +0000389</code></div>
Reid Spencere831f0b2004-07-29 00:13:04 +0000390<p>there are two definitions. The definition of <tt>%MyVar</tt> uses <tt>%MyType</tt>.
391In the C++ IR this linkage between <tt>%MyVar</tt> and <tt>%MyType</tt>
392is explicit through the use of C++ pointers. In bytecode, however, there's no
393ability to store memory addresses. Instead, we compute and write out
394slot numbers for every Type and Value written to the file.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000395<p>A slot number is simply an unsigned 32-bit integer encoded in the variable
396bit rate scheme (see <a href="#encoding">encoding</a>). This ensures that
397low slot numbers are encoded in one byte. Through various bits of magic LLVM
398attempts to always keep the slot numbers low. The first attempt is to associate
Reid Spencere831f0b2004-07-29 00:13:04 +0000399slot numbers with their "type plane". That is, Values of the same type
400are written to the bytecode file in a list (sequentially). Their order in
401that list determines their slot number. This means that slot #1 doesn't mean
402anything unless you also specify for which type you want slot #1. Types are
Reid Spencer375d7842004-08-18 20:06:19 +0000403always written to the file first (in the <a href="#globaltypes">Global Type
404Pool</a>) and in such a way that both forward and backward references of the
405types can often be resolved with a single pass through the type pool. </p>
Reid Spencere831f0b2004-07-29 00:13:04 +0000406<p>Slot numbers are also kept small by rearranging their order. Because
407of the structure of LLVM, certain values are much more likely to be used
408frequently in the body of a function. For this reason, a compaction table is
409provided in the body of a function if its use would make the function body
410smaller. Suppose you have a function body that uses just the types "int*" and
411"{double}" but uses them thousands of time. Its worthwhile to ensure that the
412slot number for these types are low so they can be encoded in a single byte
413(via vbr). This is exactly what the compaction table does.</p>
Reid Spencer375d7842004-08-18 20:06:19 +0000414<p>In summary then, a slot number can be though of as just a vbr encoded index
415into a list of Type* or Value*. To keep slot numbers low, Value* are indexed by
416two slot numbers: the "type plane index" (type slot) and the "value index"
417(value slot).</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000418</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000419<!-- *********************************************************************** -->
Reid Spencere2f97b32004-07-05 22:28:02 +0000420<div class="doc_section"> <a name="general">General Structure</a> </div>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000421<!-- *********************************************************************** -->
422<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000423<p>This section provides the general structure of the LLVM bytecode
424file format. The bytecode file format requires blocks to be in a
425certain order and nested in a particular way so that an LLVM module can
426be constructed efficiently from the contents of the file. This ordering
427defines a general structure for bytecode files as shown below. The
428table below shows the order in which all block types may appear. Please
429note that some of the blocks are optional and some may be repeated. The
430structure is fairly loose because optional blocks, if empty, are
431completely omitted from the file.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000432<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000433 <tbody>
434 <tr>
435 <th>ID</th>
436 <th>Parent</th>
437 <th>Optional?</th>
438 <th>Repeated?</th>
439 <th>Level</th>
440 <th>Block Type</th>
441 <th>Description</th>
442 </tr>
443 <tr>
444 <td>N/A</td>
445 <td>File</td>
446 <td>No</td>
447 <td>No</td>
448 <td>0</td>
449 <td class="td_left"><a href="#signature">Signature</a></td>
450 <td class="td_left">This contains the file signature (magic
451number) that identifies the file as LLVM bytecode.</td>
452 </tr>
453 <tr>
454 <td>0x01</td>
455 <td>File</td>
456 <td>No</td>
457 <td>No</td>
458 <td>0</td>
459 <td class="td_left"><a href="#module">Module</a></td>
460 <td class="td_left">This is the top level block in a bytecode
461file. It contains all the other blocks. </td>
462 </tr>
463 <tr>
464 <td>0x06</td>
465 <td>Module</td>
466 <td>No</td>
467 <td>No</td>
468 <td>1</td>
469 <td class="td_left">&nbsp;&nbsp;&nbsp;<a href="#globaltypes">Global&nbsp;Type&nbsp;Pool</a></td>
470 <td class="td_left">This block contains all the global (module)
471level types.</td>
472 </tr>
473 <tr>
474 <td>0x05</td>
475 <td>Module</td>
476 <td>No</td>
477 <td>No</td>
478 <td>1</td>
479 <td class="td_left">&nbsp;&nbsp;&nbsp;<a href="#globalinfo">Module&nbsp;Globals&nbsp;Info</a></td>
480 <td class="td_left">This block contains the type, constness, and
481linkage for each of the global variables in the module. It also
482contains the type of the functions and the constant initializers.</td>
483 </tr>
484 <tr>
485 <td>0x03</td>
486 <td>Module</td>
487 <td>Yes</td>
488 <td>No</td>
489 <td>1</td>
490 <td class="td_left">&nbsp;&nbsp;&nbsp;<a href="#constantpool">Module&nbsp;Constant&nbsp;Pool</a></td>
491 <td class="td_left">This block contains all the global constants
492except function arguments, global values and constant strings.</td>
493 </tr>
494 <tr>
495 <td>0x02</td>
496 <td>Module</td>
497 <td>Yes</td>
498 <td>Yes</td>
499 <td>1</td>
500 <td class="td_left">&nbsp;&nbsp;&nbsp;<a href="#functiondefs">Function&nbsp;Definitions</a>*</td>
501 <td class="td_left">One function block is written for each
502function in the module. The function block contains the instructions,
503compaction table, type constant pool, and symbol table for the function.</td>
504 </tr>
505 <tr>
506 <td>0x03</td>
507 <td>Function</td>
508 <td>Yes</td>
509 <td>No</td>
510 <td>2</td>
511 <td class="td_left">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<a
512 href="#constantpool">Function&nbsp;Constant&nbsp;Pool</a></td>
513 <td class="td_left">Any constants (including types) used solely
514within the function are emitted here in the function constant pool. </td>
515 </tr>
516 <tr>
517 <td>0x08</td>
518 <td>Function</td>
519 <td>Yes</td>
520 <td>No</td>
521 <td>2</td>
522 <td class="td_left">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<a
523 href="#compactiontable">Compaction&nbsp;Table</a></td>
524 <td class="td_left">This table reduces bytecode size by providing
525a funtion-local mapping of type and value slot numbers to their global
526slot numbers</td>
527 </tr>
528 <tr>
529 <td>0x07</td>
530 <td>Function</td>
531 <td>No</td>
532 <td>No</td>
533 <td>2</td>
534 <td class="td_left">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<a
535 href="#instructionlist">Instruction&nbsp;List</a></td>
536 <td class="td_left">This block contains all the instructions of
537the function. The basic blocks are inferred by terminating
538instructions. </td>
539 </tr>
540 <tr>
541 <td>0x04</td>
542 <td>Function</td>
543 <td>Yes</td>
544 <td>No</td>
545 <td>2</td>
546 <td class="td_left">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<a
547 href="#symtab">Function&nbsp;Symbol&nbsp;Table</a></td>
548 <td class="td_left">This symbol table provides the names for the
549function specific values used (basic block labels mostly).</td>
550 </tr>
551 <tr>
552 <td>0x04</td>
553 <td>Module</td>
554 <td>Yes</td>
555 <td>No</td>
556 <td>1</td>
557 <td class="td_left">&nbsp;&nbsp;&nbsp;<a href="#symtab">Module&nbsp;Symbol&nbsp;Table</a></td>
558 <td class="td_left">This symbol table provides the names for the
559various entries in the file that are not function specific (global
560vars, and functions mostly).</td>
561 </tr>
562 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000563</table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000564<p>Use the links in the table for details about the contents of each of
565the block types.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000566</div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000567<!-- *********************************************************************** -->
Reid Spencere2f97b32004-07-05 22:28:02 +0000568<div class="doc_section"> <a name="blockdefs">Block Definitions</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000569<!-- *********************************************************************** -->
570<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000571<p>This section provides the detailed layout of the individual block
572types in the LLVM bytecode file format. </p>
Reid Spencer9acd4122004-05-22 02:28:36 +0000573</div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000574<!-- _______________________________________________________________________ -->
Reid Spencereca340b2004-05-23 17:05:09 +0000575<div class="doc_subsection"><a name="signature">Signature Block</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000576<div class="doc_text">
Chris Lattnera0390ec2004-05-24 05:35:17 +0000577<p>The signature occurs in every LLVM bytecode file and is always first.
Reid Spencereca340b2004-05-23 17:05:09 +0000578It simply provides a few bytes of data to identify the file as being an LLVM
579bytecode file. This block is always four bytes in length and differs from the
580other blocks because there is no identifier and no block length at the start
581of the block. Essentially, this block is just the "magic number" for the file.
Reid Spencere831f0b2004-07-29 00:13:04 +0000582</p>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000583<p>There are two types of signatures for LLVM bytecode: uncompressed and
584compressed as shown in the table below. </p>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000585<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000586 <tbody>
587 <tr>
588 <th><b>Type</b></th>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000589 <th class="td_left"><b>Uncompressed</b></th>
590 <th class="td_left"><b>Compressed</b></th>
Reid Spencere831f0b2004-07-29 00:13:04 +0000591 </tr>
592 <tr>
593 <td><a href="#char">char</a></td>
594 <td class="td_left">Constant "l" (0x6C)</td>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000595 <td class="td_left">Constant "l" (0x6C)</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000596 </tr>
597 <tr>
598 <td><a href="#char">char</a></td>
599 <td class="td_left">Constant "l" (0x6C)</td>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000600 <td class="td_left">Constant "l" (0x6C)</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000601 </tr>
602 <tr>
603 <td><a href="#char">char</a></td>
604 <td class="td_left">Constant "v" (0x76)</td>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000605 <td class="td_left">Constant "v" (0x76)</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000606 </tr>
607 <tr>
608 <td><a href="#char">char</a></td>
609 <td class="td_left">Constant "m" (0x6D)</td>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000610 <td class="td_left">Constant "c" (0x63)</td>
611 </tr>
612 <tr>
613 <td><a href="#char">char</a></td>
614 <td class="td_left">N/A</td>
615 <td class="td_left">'0'=null,'1'=gzip,'2'=bzip2</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000616 </tr>
617 </tbody>
Reid Spencereca340b2004-05-23 17:05:09 +0000618</table>
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000619<p>In other words, the uncompressed signature is just the characters 'llvm'
620while the compressed signature is the characters 'llvc' followed by an ascii
621digit ('0', '1', or '2') that indicates the kind of compression used. A value of
622'0' indicates that null compression was used. This can happen when compression
623was requested on a platform that wasn't configured for gzip or bzip2. A value of
624'1' means that the rest of the file is compressed using the gzip algorithm and
625should be uncompressed before interpretation. A value of '2' means that the rest
626of the file is compressed using the bzip2 algorithm and should be uncompressed
627before interpretation. In all cases, the data resulting from uncompression
628should be interpreted as if it occurred immediately after the 'llvm'
629signature (i.e. the uncompressed data begins with the
630<a href="#module">Module Block</a></p>
631<p><b>NOTE:</b> As of LLVM 1.4, all bytecode files produced by the LLVM tools
Reid Spencer1619c722004-11-08 09:10:50 +0000632are compressed by default. To disable compression, pass the
Reid Spencer812aa2ba2004-11-08 08:55:21 +0000633<tt>--disable-compression</tt> option to the tool, if it supports it.
Reid Spencereca340b2004-05-23 17:05:09 +0000634</div>
635<!-- _______________________________________________________________________ -->
636<div class="doc_subsection"><a name="module">Module Block</a> </div>
637<div class="doc_text">
638<p>The module block contains a small pre-amble and all the other blocks in
Reid Spencer4e0352d2004-07-05 08:18:07 +0000639the file. The table below shows the structure of the module block. Note that it
640only provides the module identifier, size of the module block, and the format
641information. Everything else is contained in other blocks, described in other
642sections.</p>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000643<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000644 <tbody>
645 <tr>
646 <th><b>Type</b></th>
647 <th class="td_left"><b>Field Description</b></th>
648 </tr>
649 <tr>
Reid Spencerf57b8b92004-08-16 19:24:36 +0000650 <td><a href="#unsigned">unsigned</a><br></td>
651 <td class="td_left"><a href="#mod_header">Module Block Identifier
652 (0x01)</a></td>
653 </tr>
654 <tr>
655 <td><a href="#unsigned">unsigned</a></td>
656 <td class="td_left"><a href="#mod_header">Module Block Size</a></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000657 </tr>
658 <tr>
659 <td><a href="#uint32_vbr">uint32_vbr</a></td>
660 <td class="td_left"><a href="#format">Format Information</a></td>
661 </tr>
662 <tr>
663 <td><a href="#block">block</a></td>
664 <td class="td_left"><a href="#globaltypes">Global Type Pool</a></td>
665 </tr>
666 <tr>
667 <td><a href="#block">block</a></td>
668 <td class="td_left"><a href="#globalinfo">Module Globals Info</a></td>
669 </tr>
670 <tr>
671 <td><a href="#block">block</a></td>
672 <td class="td_left"><a href="#constantpool">Module Constant Pool</a></td>
673 </tr>
674 <tr>
675 <td><a href="#block">block</a>*</td>
676 <td class="td_left"><a href="#functiondefs">Function Definitions</a></td>
677 </tr>
678 <tr>
679 <td><a href="#block">block</a></td>
Reid Spencer9a910812004-08-17 00:49:03 +0000680 <td class="td_left"><a href="#symtab">Module Symbol Table</a></td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000681 </tr>
682 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000683</table>
684</div>
Reid Spencerf57b8b92004-08-16 19:24:36 +0000685
686<!-- _______________________________________________________________________ -->
687<div class="doc_subsubsection"><a name="mod_header">Module Block Header</a></div>
688<div class="doc_text">
689 <p>The block header for the module block uses a longer format than the other
690 blocks in a bytecode file. Specifically, instead of encoding the type and size
691 of the block into a 32-bit integer with 5-bits for type and 27-bits for size,
692 the module block header uses two 32-bit unsigned values, one for type, and one
693 for size. While the 2<sup>27</sup> byte limit on block size is sufficient for the blocks
694 contained in the module, it isn't sufficient for the module block itself
695 because we want to ensure that bytecode files as large as 2<sup>32</sup> bytes
696 are possible. For this reason, the module block (and only the module block)
697 uses a long format header.</p>
698</div>
699
Reid Spencer4e0352d2004-07-05 08:18:07 +0000700<!-- _______________________________________________________________________ -->
701<div class="doc_subsubsection"><a name="format">Format Information</a></div>
702<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000703<p>The format information field is encoded into a <a href="#uint32_vbr">uint32_vbr</a>
704as shown in the following table.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000705<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000706 <tbody>
707 <tr>
708 <th><b>Type</b></th>
709 <th class="td_left"><b>Description</b></th>
710 </tr>
711 <tr>
712 <td><a href="#bit">bit(0)</a></td>
713 <td class="td_left">Target is big endian?</td>
714 </tr>
715 <tr>
716 <td><a href="#bit">bit(1)</a></td>
717 <td class="td_left">On target pointers are 64-bit?</td>
718 </tr>
719 <tr>
720 <td><a href="#bit">bit(2)</a></td>
721 <td class="td_left">Target has no endianess?</td>
722 </tr>
723 <tr>
724 <td><a href="#bit">bit(3)</a></td>
725 <td class="td_left">Target has no pointer size?</td>
726 </tr>
727 <tr>
728 <td><a href="#bit">bit(4-31)</a></td>
729 <td class="td_left">Bytecode format version</td>
730 </tr>
731 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000732</table>
733<p>
734Of particular note, the bytecode format number is simply a 28-bit
John Criswell2bbe7112005-10-24 16:20:10 +0000735monotonically increasing integer that identifies the version of the bytecode
Reid Spencere831f0b2004-07-29 00:13:04 +0000736format (which is not directly related to the LLVM release number). The
737bytecode versions defined so far are (note that this document only
738describes the latest version, 1.3):</p>
Chris Lattnera0390ec2004-05-24 05:35:17 +0000739<ul>
Reid Spencere831f0b2004-07-29 00:13:04 +0000740 <li>#0: LLVM 1.0 &amp; 1.1</li>
741 <li>#1: LLVM 1.2</li>
742 <li>#2: LLVM 1.2.5 (not released)</li>
John Criswellf23e9712005-10-24 17:10:57 +0000743 <li>#3: LLVM 1.3</li>
744 <li>#4: LLVM 1.3.x (not released)</li>
745 <li>#5: LLVM 1.4, 1.5, 1.6</li>
Reid Spencere831f0b2004-07-29 00:13:04 +0000746 </li>
Chris Lattnera0390ec2004-05-24 05:35:17 +0000747</ul>
Reid Spencere831f0b2004-07-29 00:13:04 +0000748<p>Note that we plan to eventually expand the target description
749capabilities
750of bytecode files to <a href="http://llvm.cs.uiuc.edu/PR263">target
751triples</a>.
Reid Spencer4e0352d2004-07-05 08:18:07 +0000752</p>
Reid Spencer9acd4122004-05-22 02:28:36 +0000753</div>
754<!-- _______________________________________________________________________ -->
Reid Spencer4e0352d2004-07-05 08:18:07 +0000755<div class="doc_subsection"><a name="globaltypes">Global Type Pool</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000756<div class="doc_text">
Chris Lattnera0390ec2004-05-24 05:35:17 +0000757<p>The global type pool consists of type definitions. Their order of appearance
Reid Spencer375d7842004-08-18 20:06:19 +0000758in the file determines their type slot number (0 based). Slot numbers are
Reid Spencere831f0b2004-07-29 00:13:04 +0000759used to replace pointers in the intermediate representation. Each slot number
760uniquely identifies one entry in a type plane (a collection of values of the
761same type). Since all values have types and are associated with the order in
762which the type pool is written, the global type pool <em>must</em> be written
763as the first block of a module. If it is not, attempts to read the file will
764fail because both forward and backward type resolution will not be possible.</p>
765<p>The type pool is simply a list of type definitions, as shown in the
766table below.</p>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000767<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000768 <tbody>
769 <tr>
770 <th><b>Type</b></th>
771 <th class="td_left"><b>Field Description</b></th>
772 </tr>
773 <tr>
774 <td><a href="#unsigned">block</a></td>
775 <td class="td_left">Type Pool Identifier (0x06) + Size<br>
776 </td>
777 </tr>
778 <tr>
779 <td><a href="#llist">llist</a>(<a href="#type">type</a>)</td>
780 <td class="td_left">A length list of type definitions.</td>
781 </tr>
782 </tbody>
Reid Spencereca340b2004-05-23 17:05:09 +0000783</table>
Reid Spencer9acd4122004-05-22 02:28:36 +0000784</div>
785<!-- _______________________________________________________________________ -->
Reid Spencer4e0352d2004-07-05 08:18:07 +0000786<div class="doc_subsubsection"><a name="type">Type Definitions</a></div>
787<div class="doc_text">
Reid Spencer8e32af02004-07-07 13:34:26 +0000788<p>Types in the type pool are defined using a different format for each kind
789of type, as given in the following sections.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000790<h3>Primitive Types</h3>
Reid Spencere831f0b2004-07-29 00:13:04 +0000791<p>The primitive types encompass the basic integer and floating point
Reid Spencer375d7842004-08-18 20:06:19 +0000792types. They are encoded simply as their TypeID.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000793<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000794 <tbody>
795 <tr>
796 <th><b>Type</b></th>
797 <th class="td_left"><b>Description</b></th>
798 </tr>
799 <tr>
800 <td><a href="#uint24_vbr">uint24_vbr</a></td>
801 <td class="td_left">Type ID for the primitive types (values 1 to
80211) <sup>1</sup></td>
803 </tr>
804 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000805</table>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000806Notes:
807<ol>
Reid Spencere831f0b2004-07-29 00:13:04 +0000808 <li>The values for the Type IDs for the primitive types are provided
809by the definition of the <code>llvm::Type::TypeID</code> enumeration
810in <code>include/llvm/Type.h</code>. The enumeration gives the
811following mapping:
812 <ol>
813 <li>bool</li>
814 <li>ubyte</li>
815 <li>sbyte</li>
816 <li>ushort</li>
817 <li>short</li>
818 <li>uint</li>
819 <li>int</li>
820 <li>ulong</li>
821 <li>long</li>
822 <li>float</li>
823 <li>double</li>
824 </ol>
825 </li>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000826</ol>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000827<h3>Function Types</h3>
828<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000829 <tbody>
830 <tr>
831 <th><b>Type</b></th>
832 <th class="td_left"><b>Description</b></th>
833 </tr>
834 <tr>
835 <td><a href="#uint24_vbr">uint24_vbr</a></td>
836 <td class="td_left">Type ID for function types (13)</td>
837 </tr>
838 <tr>
839 <td><a href="#uint24_vbr">uint24_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +0000840 <td class="td_left">Type slot number of function's return type.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000841 </tr>
842 <tr>
843 <td><a href="#llist">llist</a>(<a href="#uint24_vbr">uint24_vbr</a>)</td>
Reid Spencer375d7842004-08-18 20:06:19 +0000844 <td class="td_left">Type slot number of each argument's type.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000845 </tr>
846 <tr>
847 <td><a href="#uint32_vbr">uint32_vbr</a>?</td>
848 <td class="td_left">Value 0 if this is a varargs function,
849missing otherwise.</td>
850 </tr>
851 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000852</table>
853<h3>Structure Types</h3>
854<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000855 <tbody>
856 <tr>
857 <th><b>Type</b></th>
858 <th class="td_left"><b>Description</b></th>
859 </tr>
860 <tr>
861 <td><a href="#uint24_vbr">uint24_vbr</a></td>
862 <td class="td_left">Type ID for structure types (14)</td>
863 </tr>
864 <tr>
865 <td><a href="#zlist">zlist</a>(<a href="#uint24_vbr">uint24_vbr</a>)</td>
866 <td class="td_left">Slot number of each of the element's fields.</td>
867 </tr>
868 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000869</table>
870<h3>Array Types</h3>
871<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000872 <tbody>
873 <tr>
874 <th><b>Type</b></th>
875 <th class="td_left"><b>Description</b></th>
876 </tr>
877 <tr>
878 <td><a href="#uint24_vbr">uint24_vbr</a></td>
879 <td class="td_left">Type ID for Array Types (15)</td>
880 </tr>
881 <tr>
882 <td><a href="#uint24_vbr">uint24_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +0000883 <td class="td_left">Type slot number of array's element type.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000884 </tr>
885 <tr>
886 <td><a href="#uint32_vbr">uint32_vbr</a></td>
887 <td class="td_left">The number of elements in the array.</td>
888 </tr>
889 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000890</table>
891<h3>Pointer Types</h3>
892<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000893 <tbody>
894 <tr>
895 <th><b>Type</b></th>
896 <th class="td_left"><b>Description</b></th>
897 </tr>
898 <tr>
899 <td><a href="#uint24_vbr">uint24_vbr</a></td>
900 <td class="td_left">Type ID For Pointer Types (16)</td>
901 </tr>
902 <tr>
903 <td><a href="#uint24_vbr">uint24_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +0000904 <td class="td_left">Type slot number of pointer's element type.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000905 </tr>
906 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000907</table>
908<h3>Opaque Types</h3>
909<table>
Reid Spencere831f0b2004-07-29 00:13:04 +0000910 <tbody>
911 <tr>
912 <th><b>Type</b></th>
913 <th class="td_left"><b>Description</b></th>
914 </tr>
915 <tr>
916 <td><a href="#uint24_vbr">uint24_vbr</a></td>
917 <td class="td_left">Type ID For Opaque Types (17)</td>
918 </tr>
919 </tbody>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000920</table>
Brian Gaeke02209042004-08-20 06:00:58 +0000921<h3>Packed Types</h3>
922<table>
923 <tbody>
924 <tr>
925 <th><b>Type</b></th>
926 <th class="td_left"><b>Description</b></th>
927 </tr>
928 <tr>
929 <td><a href="#uint24_vbr">uint24_vbr</a></td>
930 <td class="td_left">Type ID for Packed Types (18)</td>
931 </tr>
932 <tr>
933 <td><a href="#uint24_vbr">uint24_vbr</a></td>
934 <td class="td_left">Slot number of packed vector's element type.</td>
935 </tr>
936 <tr>
937 <td><a href="#uint32_vbr">uint32_vbr</a></td>
938 <td class="td_left">The number of elements in the packed vector.</td>
939 </tr>
940 </tbody>
941</table>
Reid Spencer4e0352d2004-07-05 08:18:07 +0000942</div>
943<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +0000944<div class="doc_subsection"><a name="globalinfo">Module Global Info</a>
945</div>
Reid Spencer9acd4122004-05-22 02:28:36 +0000946<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +0000947<p>The module global info block contains the definitions of all global
948variables including their initializers and the <em>declaration</em> of
949all functions. The format is shown in the table below:</p>
950<table>
951 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000952 <tr>
953 <th><b>Type</b></th>
954 <th class="td_left"><b>Field Description</b></th>
Reid Spencer9ce11e42004-07-05 19:04:27 +0000955 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +0000956 <tr>
957 <td><a href="#block">block</a></td>
958 <td class="td_left">Module global info identifier (0x05) + size<br>
959 </td>
960 </tr>
961 <tr>
962 <td><a href="#zlist">zlist</a>(<a href="#globalvar">globalvar</a>)</td>
963 <td class="td_left">A zero terminated list of global var
Reid Spencer375d7842004-08-18 20:06:19 +0000964definitions occurring in the module.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000965 </tr>
966 <tr>
Chris Lattnercd196a92004-10-16 18:03:55 +0000967 <td><a href="#zlist">zlist</a>(<a href="#funcfield">funcfield</a>)</td>
968 <td class="td_left">A zero terminated list of function definitions
Reid Spencer375d7842004-08-18 20:06:19 +0000969occurring in the module.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +0000970 </tr>
971 <tr>
972 <td style="vertical-align: top;"><a href="#llist">llist</a>(<a
973 href="#string">string</a>)<br>
974 </td>
975 <td style="vertical-align: top; text-align: left;">A length list
976of strings that specify the names of the libraries that this module
977depends upon.<br>
978 </td>
979 </tr>
980 <tr>
981 <td style="vertical-align: top;"><a href="#string">string</a><br>
982 </td>
983 <td style="vertical-align: top; text-align: left;">The target
984triple for the module (blank means no target triple specified, i.e. a
985platform independent module).<br>
986 </td>
987 </tr>
988 </tbody>
989</table>
Reid Spencer9acd4122004-05-22 02:28:36 +0000990</div>
Chris Lattnercd196a92004-10-16 18:03:55 +0000991
Reid Spencer9ce11e42004-07-05 19:04:27 +0000992<!-- _______________________________________________________________________ -->
993<div class="doc_subsubsection"><a name="globalvar">Global Variable Field</a>
994</div>
Chris Lattner84acbcc2005-11-06 07:20:25 +0000995
Reid Spencer9ce11e42004-07-05 19:04:27 +0000996<div class="doc_text">
Chris Lattner84acbcc2005-11-06 07:20:25 +0000997
Reid Spencere831f0b2004-07-29 00:13:04 +0000998<p>Global variables are written using an <a href="#uint32_vbr">uint32_vbr</a>
Chris Lattner84acbcc2005-11-06 07:20:25 +0000999that encodes information about the global variable, an optional extension vbr,
1000and a an optional initializers for the global var.</p>
1001
Reid Spencere831f0b2004-07-29 00:13:04 +00001002<p>The table below provides the bit layout of the first <a
1003 href="#uint32_vbr">uint32_vbr</a> that describes the global variable.</p>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001004
Reid Spencere831f0b2004-07-29 00:13:04 +00001005<table>
1006 <tbody>
Reid Spencer8e32af02004-07-07 13:34:26 +00001007 <tr>
1008 <th><b>Type</b></th>
1009 <th class="td_left"><b>Description</b></th>
Reid Spencer8e32af02004-07-07 13:34:26 +00001010 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001011 <tr>
1012 <td><a href="#bit">bit(0)</a></td>
1013 <td class="td_left">Is constant?</td>
1014 </tr>
1015 <tr>
1016 <td><a href="#bit">bit(1)</a></td>
1017 <td class="td_left">Has initializer? Note that this bit
1018determines whether the constant initializer field (described below)
1019follows. </td>
1020 </tr>
1021 <tr>
1022 <td><a href="#bit">bit(2-4)</a></td>
1023 <td class="td_left">Linkage type: 0=External, 1=Weak,
10242=Appending, 3=Internal, 4=LinkOnce</td>
1025 </tr>
1026 <tr>
1027 <td><a href="#bit">bit(5-31)</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +00001028 <td class="td_left">Type slot number of type for the global variable.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001029 </tr>
1030 </tbody>
1031</table>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001032
1033<p>When the Linkage type is set to 3 (internal) and the initializer field is set
1034to 0 (an invalid combination), an extension word follows the first <a
1035href="#uint32_vbr">uint32_vbr</a> which encodes the real linkage and init flag,
1036and can includes more information:</p>
1037
1038<table>
1039 <tbody>
1040 <tr>
1041 <th><b>Type</b></th>
1042 <th class="td_left"><b>Description</b></th>
1043 </tr>
1044 <tr>
1045 <td><a href="#bit">bit(0)</a></td>
1046 <td class="td_left">Has initializer? Indicates the real value of the "Has
1047 initializer" field for the global. </td>
1048 </tr>
1049 <tr>
1050 <td><a href="#bit">bit(2-4)</a></td>
1051 <td class="td_left">Linkage type: Indicates the real value of the "linkage
1052 type" field for the global.</td>
1053 </tr>
1054 <tr>
1055 <td><a href="#bit">bit(4-8)</a></td>
1056 <td class="td_left">The log-base-2 of the alignment for the global.</td>
1057 </tr>
1058 <tr>
1059 <td><a href="#bit">bit(9-31)</a></td>
1060 <td class="td_left">Currently unassigned.</td>
1061 </tr>
1062 </tbody>
1063</table>
1064
Reid Spencere831f0b2004-07-29 00:13:04 +00001065<p>The table below provides the format of the constant initializers for
Chris Lattner84acbcc2005-11-06 07:20:25 +00001066the global variable field, if it has one (indicated by the "Has initializer"
1067field).</p>
1068
Reid Spencere831f0b2004-07-29 00:13:04 +00001069<table>
1070 <tbody>
1071 <tr>
1072 <th><b>Type</b></th>
1073 <th class="td_left"><b>Description</b></th>
1074 </tr>
1075 <tr>
1076 <td>(<a href="#zlist">zlist</a>(<a href="#uint32_vbr">uint32_vbr</a>))?
1077 </td>
Reid Spencer375d7842004-08-18 20:06:19 +00001078 <td class="td_left">An optional zero-terminated list of value slot
Reid Spencere831f0b2004-07-29 00:13:04 +00001079numbers of the global variable's constant initializer.</td>
1080 </tr>
1081 </tbody>
1082</table>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001083</div>
Chris Lattnercd196a92004-10-16 18:03:55 +00001084
1085<!-- _______________________________________________________________________ -->
1086<div class="doc_subsubsection"><a name="funcfield">Function Field</a>
1087</div>
1088<div class="doc_text">
1089<p>Functions are written using an <a href="#uint32_vbr">uint32_vbr</a>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001090that encodes information about the function and a set of flags. If needed,
1091an extension word may follow this first field.</p>
Chris Lattnercd196a92004-10-16 18:03:55 +00001092
1093<p>The table below provides the bit layout of the <a
1094href="#uint32_vbr">uint32_vbr</a> that describes the function.</p>
1095
1096<table>
1097 <tbody>
1098 <tr>
1099 <th><b>Type</b></th>
1100 <th class="td_left"><b>Description</b></th>
1101 </tr>
1102 <tr>
Chris Lattner97f5c0e2004-11-15 22:54:50 +00001103 <td><a href="#bit">bit(0-3)</a></td>
Chris Lattner0132aff2005-05-06 22:57:40 +00001104 <td class="td_left">
Chris Lattner9e0781f2005-11-06 07:48:11 +00001105 Encodes the calling convention number of the function. The
Chris Lattner0132aff2005-05-06 22:57:40 +00001106 CC number of the function is the value of this field minus one.
1107 </td>
Chris Lattner97f5c0e2004-11-15 22:54:50 +00001108 </tr>
1109 <tr>
1110 <td><a href="#bit">bit(4)</a></td>
1111 <td class="td_left">If this bit is set to 1, the indicated function is
1112 external, and there is no <a href="#functiondefs">Function Definiton
1113 Block</a> in the bytecode file for the function.</td>
Chris Lattnercd196a92004-10-16 18:03:55 +00001114 </tr>
1115 <tr>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001116 <td><a href="#bit">bit(5-30)</a></td>
Chris Lattnercd196a92004-10-16 18:03:55 +00001117 <td class="td_left">Type slot number of type for the function.</td>
1118 </tr>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001119 <tr>
1120 <td><a href="#bit">bit(31)</a></td>
1121 <td class="td_left">Indicates whether an extension word follows.</td>
1122 </tr>
1123 </tbody>
1124</table>
1125
1126<p>If bit(31) is set, an additional <a href="#uint32_vbr">uint32_vbr</a> word
1127follows with the following fields:</p>
1128
1129<table>
1130 <tbody>
1131 <tr>
1132 <th><b>Type</b></th>
1133 <th class="td_left"><b>Description</b></th>
1134 </tr>
1135 <tr>
1136 <td><a href="#bit">bit(0-4)</a></td>
1137 <td class="td_left">The log-base-2 of the alignment for the function.</td>
1138 </tr>
1139 <tr>
Chris Lattner9e0781f2005-11-06 07:48:11 +00001140 <td><a href="#bit">bit(5-9)</a></td>
1141 <td class="td_left">The top nibble of the calling convention.</td>
1142 </tr>
1143 <tr>
1144 <td><a href="#bit">bit(10-31)</a></td>
Chris Lattner84acbcc2005-11-06 07:20:25 +00001145 <td class="td_left">Currently unassigned.</td>
1146 </tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001147 </tbody>
1148</table>
1149
1150</div>
1151
Reid Spencer9acd4122004-05-22 02:28:36 +00001152<!-- _______________________________________________________________________ -->
Reid Spencer4e0352d2004-07-05 08:18:07 +00001153<div class="doc_subsection"><a name="constantpool">Constant Pool</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +00001154<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001155<p>A constant pool defines as set of constant values. There are
1156actually two types of constant pool blocks: one for modules and one for
1157functions. For modules, the block begins with the constant strings
1158encountered anywhere in the module. For functions, the block begins
1159with types only encountered in the function. In both cases the header
1160is identical. The tables that follow, show the header, module constant
1161pool preamble, function constant pool preamble, and the part common to
1162both function and module constant pools.</p>
1163<p><b>Common Block Header</b></p>
1164<table>
1165 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001166 <tr>
1167 <th><b>Type</b></th>
1168 <th class="td_left"><b>Field Description</b></th>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001169 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001170 <tr>
1171 <td><a href="#block">block</a></td>
1172 <td class="td_left">Constant pool identifier (0x03) + size<br>
1173 </td>
1174 </tr>
1175 </tbody>
1176</table>
1177<p><b>Module Constant Pool Preamble (constant strings)</b></p>
1178<table>
1179 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001180 <tr>
1181 <th><b>Type</b></th>
1182 <th class="td_left"><b>Field Description</b></th>
Reid Spencere831f0b2004-07-29 00:13:04 +00001183 </tr>
1184 <tr>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001185 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1186 <td class="td_left">The number of constant strings that follow.</td>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001187 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001188 <tr>
1189 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1190 <td class="td_left">Zero. This identifies the following "plane"
1191as containing the constant strings. This is needed to identify it
1192uniquely from other constant planes that follow. </td>
1193 </tr>
1194 <tr>
1195 <td><a href="#uint24_vbr">uint24_vbr</a>+</td>
Reid Spencer375d7842004-08-18 20:06:19 +00001196 <td class="td_left">Type slot number of the constant string's type.
Reid Spencere831f0b2004-07-29 00:13:04 +00001197Note that the constant string's type implicitly defines the length of
1198the string. </td>
1199 </tr>
1200 </tbody>
1201</table>
1202<p><b>Function Constant Pool Preamble (function types)</b></p>
1203<p>The structure of the types for functions is identical to the <a
1204 href="#globaltypes">Global Type Pool</a>. Please refer to that section
1205for the details. </p>
1206<p><b>Common Part (other constants)</b></p>
1207<table>
1208 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001209 <tr>
1210 <th><b>Type</b></th>
1211 <th class="td_left"><b>Field Description</b></th>
Reid Spencere831f0b2004-07-29 00:13:04 +00001212 </tr>
1213 <tr>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001214 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1215 <td class="td_left">Number of entries in this type plane.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001216 </tr>
1217 <tr>
1218 <td><a href="#uint24_vbr">uint24_vbr</a></td>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001219 <td class="td_left">Type slot number of this plane.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001220 </tr>
1221 <tr>
Reid Spencer8e32af02004-07-07 13:34:26 +00001222 <td><a href="#constant">constant</a>+</td>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001223 <td class="td_left">The definition of a constant (see below).</td>
1224 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001225 </tbody>
1226</table>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001227</div>
1228<!-- _______________________________________________________________________ -->
1229<div class="doc_subsubsection"><a name="constant">Constant Field</a></div>
1230<div class="doc_text">
Reid Spencer375d7842004-08-18 20:06:19 +00001231<p>Constants come in many shapes and flavors. The sections that follow
Reid Spencere831f0b2004-07-29 00:13:04 +00001232define the format for each of them. All constants start with a <a
1233 href="#uint32_vbr">uint32_vbr</a> encoded integer that provides the
1234number of operands for the constant. For primitive, structure, and
1235array constants, this will always be zero since those types of
1236constants have no operands. In this case, we have the following field
1237definitions:</p>
1238<ul>
1239 <li><b>Bool</b>. This is written as an <a href="#uint32_vbr">uint32_vbr</a>
1240of value 1U or 0U.</li>
1241 <li><b>Signed Integers (sbyte,short,int,long)</b>. These are written
1242as an <a href="#int64_vbr">int64_vbr</a> with the corresponding value.</li>
1243 <li><b>Unsigned Integers (ubyte,ushort,uint,ulong)</b>. These are
1244written as an <a href="#uint64_vbr">uint64_vbr</a> with the
1245corresponding value. </li>
1246 <li><b>Floating Point</b>. Both the float and double types are
1247written literally in binary format.</li>
1248 <li><b>Arrays</b>. Arrays are written simply as a list of <a
Reid Spencer375d7842004-08-18 20:06:19 +00001249 href="#uint32_vbr">uint32_vbr</a> encoded value slot numbers to the constant
Reid Spencere831f0b2004-07-29 00:13:04 +00001250element values.</li>
1251 <li><b>Structures</b>. Structures are written simply as a list of <a
Reid Spencer375d7842004-08-18 20:06:19 +00001252 href="#uint32_vbr">uint32_vbr</a> encoded value slot numbers to the constant
Reid Spencere831f0b2004-07-29 00:13:04 +00001253field values of the structure.</li>
1254</ul>
Chris Lattnercd196a92004-10-16 18:03:55 +00001255
1256<p>When the number of operands to the constant is one, we have an 'undef' value
1257of the specified type.</p>
1258
1259<p>When the number of operands to the constant is greater than one, we have a
1260constant expression and its field format is provided in the table below, and the
1261number is equal to the number of operands+1.</p>
Reid Spencere831f0b2004-07-29 00:13:04 +00001262<table>
1263 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001264 <tr>
1265 <th><b>Type</b></th>
1266 <th class="td_left"><b>Field Description</b></th>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001267 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001268 <tr>
1269 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1270 <td class="td_left">Op code of the instruction for the constant
1271expression.</td>
1272 </tr>
1273 <tr>
1274 <td><a href="#uint32_vbr">uint32_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +00001275 <td class="td_left">The value slot number of the constant value for an
Reid Spencere831f0b2004-07-29 00:13:04 +00001276operand.<sup>1</sup></td>
1277 </tr>
1278 <tr>
1279 <td><a href="#uint24_vbr">uint24_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +00001280 <td class="td_left">The type slot number for the type of the constant
Reid Spencere831f0b2004-07-29 00:13:04 +00001281value for an operand.<sup>1</sup></td>
1282 </tr>
1283 </tbody>
1284</table>
1285Notes:
1286<ol>
1287 <li>Both these fields are repeatable but only in pairs.</li>
1288</ol>
Reid Spencer9acd4122004-05-22 02:28:36 +00001289</div>
1290<!-- _______________________________________________________________________ -->
Reid Spencere2f97b32004-07-05 22:28:02 +00001291<div class="doc_subsection"><a name="functiondefs">Function Definition</a></div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001292<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001293<p>Function definitions contain the linkage, constant pool or
1294compaction table, instruction list, and symbol table for a function.
1295The following table shows the structure of a function definition.</p>
1296<table>
1297 <tbody>
Reid Spencere2f97b32004-07-05 22:28:02 +00001298 <tr>
1299 <th><b>Type</b></th>
1300 <th class="td_left"><b>Field Description</b></th>
Reid Spencere2f97b32004-07-05 22:28:02 +00001301 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001302 <tr>
1303 <td><a href="#block">block</a><br>
1304 </td>
1305 <td class="td_left">Function definition block identifier (0x02) +
1306size<br>
1307 </td>
1308 </tr>
1309 <tr>
1310 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1311 <td class="td_left">The linkage type of the function: 0=External,
13121=Weak, 2=Appending, 3=Internal, 4=LinkOnce<sup>1</sup></td>
1313 </tr>
1314 <tr>
1315 <td><a href="#block">block</a></td>
1316 <td class="td_left">The <a href="#constantpool">constant pool</a>
1317block for this function.<sup>2</sup></td>
1318 </tr>
1319 <tr>
1320 <td><a href="#block">block</a></td>
1321 <td class="td_left">The <a href="#compactiontable">compaction
1322table</a> block for the function.<sup>2</sup></td>
1323 </tr>
1324 <tr>
1325 <td><a href="#block">block</a></td>
1326 <td class="td_left">The <a href="#instructionlist">instruction
1327list</a> for the function.</td>
1328 </tr>
1329 <tr>
1330 <td><a href="#block">block</a></td>
Reid Spencer9a910812004-08-17 00:49:03 +00001331 <td class="td_left">The function's <a href="#symtab">symbol
Reid Spencere831f0b2004-07-29 00:13:04 +00001332table</a> containing only those symbols pertinent to the function
1333(mostly block labels).</td>
1334 </tr>
1335 </tbody>
1336</table>
1337Notes:
1338<ol>
1339 <li>Note that if the linkage type is "External" then none of the
1340other fields will be present as the function is defined elsewhere.</li>
1341 <li>Note that only one of the constant pool or compaction table will
1342be written. Compaction tables are only written if they will actually
1343save bytecode space. If not, then a regular constant pool is written.</li>
1344</ol>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001345</div>
1346<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001347<div class="doc_subsection"><a name="compactiontable">Compaction Table</a>
1348</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001349<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001350<p>Compaction tables are part of a function definition. They are merely
1351a device for reducing the size of bytecode files. The size of a
Reid Spencer375d7842004-08-18 20:06:19 +00001352bytecode file is dependent on the <em>values</em> of the slot numbers
Reid Spencere831f0b2004-07-29 00:13:04 +00001353used because larger values use more bytes in the variable bit rate
1354encoding scheme. Furthermore, the compressed instruction format
1355reserves only six bits for the type of the instruction. In large
1356modules, declaring hundreds or thousands of types, the values of the
1357slot numbers can be quite large. However, functions may use only a
1358small fraction of the global types. In such cases a compaction table is
1359created that maps the global type and value slot numbers to smaller
1360values used by a function. Functions will contain either a
1361function-specific constant pool <em>or</em> a compaction table but not
1362both. Compaction tables have the format shown in the table below.</p>
1363<table>
1364 <tbody>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001365 <tr>
1366 <th><b>Type</b></th>
1367 <th class="td_left"><b>Field Description</b></th>
Reid Spencere831f0b2004-07-29 00:13:04 +00001368 </tr>
1369 <tr>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001370 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1371 <td class="td_left">The number of types that follow</td>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001372 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001373 <tr>
1374 <td><a href="#uint24_vbr">uint24_vbr</a>+</td>
Reid Spencer375d7842004-08-18 20:06:19 +00001375 <td class="td_left">The type slot number in the global types of
Reid Spencere831f0b2004-07-29 00:13:04 +00001376the type that will be referenced in the function with the index of this
1377entry in the compaction table.</td>
1378 </tr>
1379 <tr>
1380 <td><a href="#type_len">type_len</a></td>
1381 <td class="td_left">An encoding of the type and number of values
1382that follow. This field's encoding varies depending on the size of the
1383type plane. See <a href="#type_len">Type and Length</a> for further
1384details.</td>
1385 </tr>
1386 <tr>
1387 <td><a href="#uint32_vbr">uint32_vbr</a>+</td>
Reid Spencer375d7842004-08-18 20:06:19 +00001388 <td class="td_left">The value slot number in the global values
Reid Spencere831f0b2004-07-29 00:13:04 +00001389that will be referenced in the function with the index of this entry in
Reid Spencer375d7842004-08-18 20:06:19 +00001390the compaction table.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001391 </tr>
1392 </tbody>
1393</table>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001394</div>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001395<!-- _______________________________________________________________________ -->
1396<div class="doc_subsubsection"><a name="type_len">Type and Length</a></div>
1397<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001398<p>The type and length of a compaction table type plane is encoded
1399differently depending on the length of the plane. For planes of length
14001 or 2, the length is encoded into bits 0 and 1 of a <a
1401 href="#uint32_vbr">uint32_vbr</a> and the type is encoded into bits
14022-31. Because type numbers are often small, this often saves an extra
1403byte per plane. If the length of the plane is greater than 2 then the
1404encoding uses a <a href="#uint32_vbr">uint32_vbr</a> for each of the
1405length and type, in that order.</p>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001406</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001407<!-- _______________________________________________________________________ -->
Reid Spencer375d7842004-08-18 20:06:19 +00001408<div class="doc_subsection"><a name="instructionlist">Instruction List</a></div>
Reid Spencer9acd4122004-05-22 02:28:36 +00001409<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001410<p>The instructions in a function are written as a simple list. Basic
1411blocks are inferred by the terminating instruction types. The format of
1412the block is given in the following table.</p>
1413<table>
1414 <tbody>
Reid Spencere2f97b32004-07-05 22:28:02 +00001415 <tr>
1416 <th><b>Type</b></th>
1417 <th class="td_left"><b>Field Description</b></th>
Reid Spencere2f97b32004-07-05 22:28:02 +00001418 </tr>
Reid Spencere831f0b2004-07-29 00:13:04 +00001419 <tr>
1420 <td><a href="#block">block</a><br>
1421 </td>
1422 <td class="td_left">Instruction list identifier (0x07) + size<br>
1423 </td>
1424 </tr>
1425 <tr>
1426 <td><a href="#instruction">instruction</a>+</td>
1427 <td class="td_left">An instruction. Instructions have a variety
1428of formats. See <a href="#instruction">Instructions</a> for details.</td>
1429 </tr>
1430 </tbody>
1431</table>
Reid Spencer9acd4122004-05-22 02:28:36 +00001432</div>
Chris Lattnerdb137582005-11-05 22:20:06 +00001433
Reid Spencere2f97b32004-07-05 22:28:02 +00001434<!-- _______________________________________________________________________ -->
Chris Lattnerdb137582005-11-05 22:20:06 +00001435<div class="doc_subsection"><a name="instructions">Instructions</a></div>
1436
Reid Spencere2f97b32004-07-05 22:28:02 +00001437<div class="doc_text">
Chris Lattnerdb137582005-11-05 22:20:06 +00001438<p>Instructions are written out one at a time as distinct units. Each
1439instruction
1440record contains at least an <a href="#opcodes">opcode</a> and a type field,
Chris Lattner16025ee2005-11-05 22:32:06 +00001441and may contain a <a href="#instoperands">list of operands</a> (whose
1442interpretation depends on the opcode). Based on the number of operands, the
Chris Lattnerdb137582005-11-05 22:20:06 +00001443<a href="#instencode">instruction is encoded</a> in a
1444dense format that tries to encoded each instruction into 32-bits if
1445possible. </p>
Reid Spencere2f97b32004-07-05 22:28:02 +00001446</div>
Reid Spencer375d7842004-08-18 20:06:19 +00001447
1448<!-- _______________________________________________________________________ -->
Chris Lattnerdb137582005-11-05 22:20:06 +00001449<div class="doc_subsubsection"><a name="opcodes">Instruction Opcodes</a></div>
Reid Spencer375d7842004-08-18 20:06:19 +00001450<div class="doc_text">
1451 <p>Instructions encode an opcode that identifies the kind of instruction.
1452 Opcodes are an enumerated integer value. The specific values used depend on
1453 the version of LLVM you're using. The opcode values are defined in the
Misha Brukmanaeee58f2004-11-07 00:59:58 +00001454 <a href="http://llvm.cs.uiuc.edu/cvsweb/cvsweb.cgi/llvm/include/llvm/Instruction.def">
Reid Spencer375d7842004-08-18 20:06:19 +00001455 <tt>include/llvm/Instruction.def</tt></a> file. You should check there for the
1456 most recent definitions. The table below provides the opcodes defined as of
Nate Begeman0c1e6802004-08-27 07:59:37 +00001457 the writing of this document. The table associates each opcode mnemonic with
Reid Spencer375d7842004-08-18 20:06:19 +00001458 its enumeration value and the bytecode and LLVM version numbers in which the
1459 opcode was introduced.</p>
1460 <table>
1461 <tbody>
1462 <tr>
1463 <th>Opcode</th>
1464 <th>Number</th>
1465 <th>Bytecode Version</th>
1466 <th>LLVM Version</th>
1467 </tr>
1468 <tr><td colspan="4"><b>Terminator Instructions</b></td></tr>
1469 <tr><td>Ret</td><td>1</td><td>1</td><td>1.0</td></tr>
1470 <tr><td>Br</td><td>2</td><td>1</td><td>1.0</td></tr>
1471 <tr><td>Switch</td><td>3</td><td>1</td><td>1.0</td></tr>
1472 <tr><td>Invoke</td><td>4</td><td>1</td><td>1.0</td></tr>
1473 <tr><td>Unwind</td><td>5</td><td>1</td><td>1.0</td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001474 <tr><td>Unreachable</td><td>6</td><td>1</td><td>1.4</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001475 <tr><td colspan="4"><b>Binary Operators</b></td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001476 <tr><td>Add</td><td>7</td><td>1</td><td>1.0</td></tr>
1477 <tr><td>Sub</td><td>8</td><td>1</td><td>1.0</td></tr>
1478 <tr><td>Mul</td><td>9</td><td>1</td><td>1.0</td></tr>
1479 <tr><td>Div</td><td>10</td><td>1</td><td>1.0</td></tr>
1480 <tr><td>Rem</td><td>11</td><td>1</td><td>1.0</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001481 <tr><td colspan="4"><b>Logical Operators</b></td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001482 <tr><td>And</td><td>12</td><td>1</td><td>1.0</td></tr>
1483 <tr><td>Or</td><td>13</td><td>1</td><td>1.0</td></tr>
1484 <tr><td>Xor</td><td>14</td><td>1</td><td>1.0</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001485 <tr><td colspan="4"><b>Binary Comparison Operators</b></td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001486 <tr><td>SetEQ</td><td>15</td><td>1</td><td>1.0</td></tr>
1487 <tr><td>SetNE</td><td>16</td><td>1</td><td>1.0</td></tr>
1488 <tr><td>SetLE</td><td>17</td><td>1</td><td>1.0</td></tr>
1489 <tr><td>SetGE</td><td>18</td><td>1</td><td>1.0</td></tr>
1490 <tr><td>SetLT</td><td>19</td><td>1</td><td>1.0</td></tr>
1491 <tr><td>SetGT</td><td>20</td><td>1</td><td>1.0</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001492 <tr><td colspan="4"><b>Memory Operators</b></td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001493 <tr><td>Malloc</td><td>21</td><td>1</td><td>1.0</td></tr>
1494 <tr><td>Free</td><td>22</td><td>1</td><td>1.0</td></tr>
1495 <tr><td>Alloca</td><td>23</td><td>1</td><td>1.0</td></tr>
1496 <tr><td>Load</td><td>24</td><td>1</td><td>1.0</td></tr>
1497 <tr><td>Store</td><td>25</td><td>1</td><td>1.0</td></tr>
1498 <tr><td>GetElementPtr</td><td>26</td><td>1</td><td>1.0</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001499 <tr><td colspan="4"><b>Other Operators</b></td></tr>
Chris Lattnercd196a92004-10-16 18:03:55 +00001500 <tr><td>PHI</td><td>27</td><td>1</td><td>1.0</td></tr>
1501 <tr><td>Cast</td><td>28</td><td>1</td><td>1.0</td></tr>
1502 <tr><td>Call</td><td>29</td><td>1</td><td>1.0</td></tr>
1503 <tr><td>Shl</td><td>30</td><td>1</td><td>1.0</td></tr>
1504 <tr><td>Shr</td><td>31</td><td>1</td><td>1.0</td></tr>
1505 <tr><td>VANext</td><td>32</td><td>1</td><td>1.0</td></tr>
1506 <tr><td>VAArg</td><td>33</td><td>1</td><td>1.0</td></tr>
1507 <tr><td>Select</td><td>34</td><td>2</td><td>1.2</td></tr>
Reid Spencer621fd112005-05-14 00:06:06 +00001508 <tr><td colspan="4">
1509 <b>Pseudo Instructions<a href="#pi_note">*</a></b>
1510 </td></tr>
1511 <tr><td>Invoke+CC </td><td>56</td><td>5</td><td>1.5</td></tr>
1512 <tr><td>Invoke+FastCC</td><td>57</td><td>5</td><td>1.5</td></tr>
1513 <tr><td>Call+CC</td><td>58</td><td>5</td><td>1.5</td></tr>
1514 <tr><td>Call+FastCC+TailCall</td><td>59</td><td>5</td><td>1.5</td></tr>
1515 <tr><td>Call+FastCC</td><td>60</td><td>5</td><td>1.5</td></tr>
1516 <tr><td>Call+CCC+TailCall</td><td>61</td><td>5</td><td>1.5</td></tr>
1517 <tr><td>Load+Volatile</td><td>62</td><td>3</td><td>1.3</td></tr>
1518 <tr><td>Store+Volatile</td><td>63</td><td>3</td><td>1.3</td></tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001519 </tbody>
1520 </table>
Reid Spencer375d7842004-08-18 20:06:19 +00001521
Reid Spencer621fd112005-05-14 00:06:06 +00001522<p><b><a name="pi_note">* Note: </a></b>
Chris Lattnerdb137582005-11-05 22:20:06 +00001523These aren't really opcodes from an LLVM language perspective. They encode
Reid Spencer621fd112005-05-14 00:06:06 +00001524information into other opcodes without reserving space for that information.
1525For example, opcode=63 is a Volatile Store. The opcode for this
1526instruction is 25 (Store) but we encode it as 63 to indicate that is a Volatile
1527Store. The same is done for the calling conventions and tail calls.
1528In each of these entries in range 56-63, the opcode is documented as the base
1529opcode (Invoke, Call, Store) plus some set of modifiers, as follows:</p>
1530<dl>
1531 <dt>CC</dt>
Chris Lattnerf47e4c12005-05-14 01:30:15 +00001532 <dd>This means an arbitrary calling convention is specified
1533 in a VBR that follows the opcode. This is used when the instruction cannot
1534 be encoded with one of the more compact forms.
Reid Spencer621fd112005-05-14 00:06:06 +00001535 </dd>
1536 <dt>FastCC</dt>
1537 <dd>This indicates that the Call or Invoke is using the FastCC calling
Chris Lattnerf47e4c12005-05-14 01:30:15 +00001538 convention.</dd>
Reid Spencer621fd112005-05-14 00:06:06 +00001539 <dt>CCC</dt>
Chris Lattnerf47e4c12005-05-14 01:30:15 +00001540 <dd>This indicates that the Call or Invoke is using the native "C" calling
1541 convention.</dd>
Reid Spencer621fd112005-05-14 00:06:06 +00001542 <dt>TailCall</dt>
Chris Lattnerf47e4c12005-05-14 01:30:15 +00001543 <dd>This indicates that the Call has the 'tail' modifier.</dd>
Reid Spencer621fd112005-05-14 00:06:06 +00001544</dl>
Chris Lattnerdb137582005-11-05 22:20:06 +00001545</div>
1546
Chris Lattner16025ee2005-11-05 22:32:06 +00001547<!-- _______________________________________________________________________ -->
1548<div class="doc_subsubsection"><a name="instoperands">Instruction
1549Operands</a></div>
1550
1551<div class="doc_text">
1552<p>
1553Based on the instruction opcode and type, the bytecode format implicitly (to
1554save space) specifies the interpretation of the operand list. For most
1555instructions, the type of each operand is implicit from the type of the
1556instruction itself (e.g. the type of operands of a binary operator must match
1557the type of the instruction). As such, the bytecode format generally only
1558encodes the value number of the operand, not the type.</p>
1559
1560<p>In some cases, however, this is not sufficient. This section enumerates
1561those cases:</p>
1562
1563<ul>
1564<li>getelementptr: the slot numbers for sequential type indexes are shifted up
1565two bits. This allows the low order bits will encode the type of index used,
1566as follows: 0=uint, 1=int, 2=ulong, 3=long.</li>
1567<li>cast: the result type number is encoded as the second operand.</li>
1568<li>alloca/malloc: If the allocation has an explicit alignment, the log2 of the
1569 alignment is encoded as the second operand.</li>
1570<li>call: If the tail marker and calling convention cannot be <a
1571 href="#pi_note">encoded into the opcode</a> of the call, it is passed as an
1572 additional operand. The low bit of the operand is a flag indicating whether
1573 the call is a tail call. The rest of the bits contain the calling
1574 convention number (shifted left by one bit).</li>
1575</ul>
1576</div>
Chris Lattnerdb137582005-11-05 22:20:06 +00001577
1578<!-- _______________________________________________________________________ -->
1579<div class="doc_subsubsection"><a name="instencode">Instruction
1580Encoding</a></div>
1581
1582<div class="doc_text">
1583<p>For brevity, instructions are written in one of four formats,
1584depending on the number of operands to the instruction. Each
1585instruction begins with a <a href="#uint32_vbr">uint32_vbr</a> that
1586encodes the type of the instruction as well as other things. The tables
1587that follow describe the format of this first part of each instruction.</p>
1588<p><b>Instruction Format 0</b></p>
1589<p>This format is used for a few instructions that can't easily be
1590shortened because they have large numbers of operands (e.g. PHI Node or
1591getelementptr). Each of the opcode, type, and operand fields is found in
1592successive fields.</p>
1593<table>
1594 <tbody>
1595 <tr>
1596 <th><b>Type</b></th>
1597 <th class="td_left"><b>Field Description</b></th>
1598 </tr>
1599 <tr>
1600 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1601 <td class="td_left">Specifies the opcode of the instruction. Note
1602that for compatibility with the other instruction formats, the opcode
1603is shifted left by 2 bits. Bits 0 and 1 must have value zero for this
1604format.</td>
1605 </tr>
1606 <tr>
1607 <td><a href="#uint24_vbr">uint24_vbr</a></td>
1608 <td class="td_left">Provides the type slot number of the result type of
1609 the instruction.</td>
1610 </tr>
1611 <tr>
1612 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1613 <td class="td_left">The number of operands that follow.</td>
1614 </tr>
1615 <tr>
1616 <td><a href="#uint32_vbr">uint32_vbr</a>+</td>
1617 <td class="td_left">The slot number of the value(s) for the operand(s).
Chris Lattner16025ee2005-11-05 22:32:06 +00001618 </td>
Chris Lattnerdb137582005-11-05 22:20:06 +00001619 </tr>
1620 </tbody>
1621</table>
Chris Lattner16025ee2005-11-05 22:32:06 +00001622
Chris Lattnerdb137582005-11-05 22:20:06 +00001623<p><b>Instruction Format 1</b></p>
1624<p>This format encodes the opcode, type and a single operand into a
1625single <a href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1626<table>
1627 <tbody>
1628 <tr>
1629 <th><b>Bits</b></th>
1630 <th><b>Type</b></th>
1631 <th class="td_left"><b>Field Description</b></th>
1632 </tr>
1633 <tr>
1634 <td>0-1</td>
1635 <td>constant "1"</td>
1636 <td class="td_left">These two bits must be the value 1 which identifies
1637 this as an instruction of format 1.</td>
1638 </tr>
1639 <tr>
1640 <td>2-7</td>
1641 <td><a href="#instructions">opcode</a></td>
1642 <td class="td_left">Specifies the opcode of the instruction. Note that
1643 the maximum opcode value is 63.</td>
1644 </tr>
1645 <tr>
1646 <td>8-19</td>
1647 <td><a href="#unsigned">unsigned</a></td>
1648 <td class="td_left">Specifies the slot number of the type for this
1649 instruction. Maximum slot number is 2<sup>12</sup>-1=4095.</td>
1650 </tr>
1651 <tr>
1652 <td>20-31</td>
1653 <td><a href="#unsigned">unsigned</a></td>
1654 <td class="td_left">Specifies the slot number of the value for the
1655 first operand. Maximum slot number is 2<sup>12</sup>-1=4095. Note that
1656 the value 2<sup>12</sup>-1 denotes zero operands.</td>
1657 </tr>
1658 </tbody>
1659</table>
1660<p><b>Instruction Format 2</b></p>
1661<p>This format encodes the opcode, type and two operands into a single <a
1662 href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1663<table>
1664 <tbody>
1665 <tr>
1666 <th><b>Bits</b></th>
1667 <th><b>Type</b></th>
1668 <th class="td_left"><b>Field Description</b></th>
1669 </tr>
1670 <tr>
1671 <td>0-1</td>
1672 <td>constant "2"</td>
1673 <td class="td_left">These two bits must be the value 2 which identifies
1674 this as an instruction of format 2.</td>
1675 </tr>
1676 <tr>
1677 <td>2-7</td>
1678 <td><a href="#instructions">opcode</a></td>
1679 <td class="td_left">Specifies the opcode of the instruction. Note that
1680 the maximum opcode value is 63.</td>
1681 </tr>
1682 <tr>
1683 <td>8-15</td>
1684 <td><a href="#unsigned">unsigned</a></td>
1685 <td class="td_left">Specifies the slot number of the type for this
1686 instruction. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1687 </tr>
1688 <tr>
1689 <td>16-23</td>
1690 <td><a href="#unsigned">unsigned</a></td>
1691 <td class="td_left">Specifies the slot number of the value for the first
1692 operand. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1693 </tr>
1694 <tr>
1695 <td>24-31</td>
1696 <td><a href="#unsigned">unsigned</a></td>
1697 <td class="td_left">Specifies the slot number of the value for the second
1698 operand. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1699 </tr>
1700 </tbody>
1701</table>
1702<p><b>Instruction Format 3</b></p>
1703<p>This format encodes the opcode, type and three operands into a
1704single <a href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1705<table>
1706 <tbody>
1707 <tr>
1708 <th><b>Bits</b></th>
1709 <th><b>Type</b></th>
1710 <th class="td_left"><b>Field Description</b></th>
1711 </tr>
1712 <tr>
1713 <td>0-1</td>
1714 <td>constant "3"</td>
1715 <td class="td_left">These two bits must be the value 3 which identifies
1716 this as an instruction of format 3.</td>
1717 </tr>
1718 <tr>
1719 <td>2-7</td>
1720 <td><a href="#instructions">opcode</a></td>
1721 <td class="td_left">Specifies the opcode of the instruction. Note that
1722 the maximum opcode value is 63.</td>
1723 </tr>
1724 <tr>
1725 <td>8-13</td>
1726 <td><a href="#unsigned">unsigned</a></td>
1727 <td class="td_left">Specifies the slot number of the type for this
1728 instruction. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1729 </tr>
1730 <tr>
1731 <td>14-19</td>
1732 <td><a href="#unsigned">unsigned</a></td>
1733 <td class="td_left">Specifies the slot number of the value for the first
1734 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1735 </tr>
1736 <tr>
1737 <td>20-25</td>
1738 <td><a href="#unsigned">unsigned</a></td>
1739 <td class="td_left">Specifies the slot number of the value for the second
1740 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1741 </tr>
1742 <tr>
1743 <td>26-31</td>
1744 <td><a href="#unsigned">unsigned</a></td>
1745 <td class="td_left">Specifies the slot number of the value for the third
1746 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1747 </tr>
1748 </tbody>
1749</table>
1750</div>
Reid Spencer621fd112005-05-14 00:06:06 +00001751
Reid Spencer9acd4122004-05-22 02:28:36 +00001752<!-- _______________________________________________________________________ -->
Reid Spencereca340b2004-05-23 17:05:09 +00001753<div class="doc_subsection"><a name="symtab">Symbol Table</a> </div>
Reid Spencer9acd4122004-05-22 02:28:36 +00001754<div class="doc_text">
Reid Spencer375d7842004-08-18 20:06:19 +00001755<p>A symbol table can be put out in conjunction with a module or a function. A
1756symbol table has a list of name/type associations followed by a list of
1757name/value associations. The name/value associations are organized into "type
1758planes" so that all values of a common type are listed together. Each type
1759plane starts with the number of entries in the plane and the type slot number
1760for all the values in that plane (so the type can be looked up in the global
1761type pool). For each entry in a type plane, the slot number of the value and
1762the name associated with that value are written. The format is given in the
1763table below. </p>
Reid Spencer9ce11e42004-07-05 19:04:27 +00001764<table>
Reid Spencere831f0b2004-07-29 00:13:04 +00001765 <tbody>
1766 <tr>
1767 <th><b>Type</b></th>
1768 <th class="td_left"><b>Field Description</b></th>
1769 </tr>
1770 <tr>
1771 <td><a href="#block">block</a><br>
1772 </td>
1773 <td class="td_left">Symbol Table Identifier (0x04)</td>
1774 </tr>
1775 <tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001776 <td><a href="#llist">llist</a>(<a href="#symtab_entry">type_entry</a>)</td>
Reid Spencer9a910812004-08-17 00:49:03 +00001777 <td class="td_left">A length list of symbol table entries for
1778 <tt>Type</tt>s
1779 </td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001780 </tr>
1781 <tr>
Reid Spencer9a910812004-08-17 00:49:03 +00001782 <td><a href="#zlist">llist</a>(<a href="#symtab_plane">symtab_plane</a>)</td>
Reid Spencer375d7842004-08-18 20:06:19 +00001783 <td class="td_left">A length list of "type planes" of symbol table
Reid Spencer9a910812004-08-17 00:49:03 +00001784 entries for <tt>Value</tt>s</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001785 </tr>
1786 </tbody>
Reid Spencereca340b2004-05-23 17:05:09 +00001787</table>
Reid Spencer9acd4122004-05-22 02:28:36 +00001788</div>
Reid Spencer375d7842004-08-18 20:06:19 +00001789
1790<!-- _______________________________________________________________________ -->
1791<div class="doc_subsubsection"> <a name="type_entry">Symbol Table Type
1792Entry</a>
1793</div>
1794<div class="doc_text">
1795<p>A symbol table type entry associates a name with a type. The name is provided
1796simply as an array of chars. The type is provided as a type slot number (index)
1797into the global type pool. The format is given in the following table:</p>
1798<table>
1799 <tbody>
1800 <tr>
1801 <th><b>Type</b></th>
1802 <th class="td_left"><b>Field Description</b></th>
1803 </tr>
1804 <tr>
1805 <td><a href="#uint32_vbr">uint24_vbr</a></td>
1806 <td class="td_left">Type slot number of the type being given a
1807 name relative to the global type pool.
1808 </td>
1809 </tr>
1810 <tr>
1811 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1812 <td class="td_left">Length of the character array that follows.</td>
1813 </tr>
1814 <tr>
1815 <td><a href="#char">char</a>+</td>
1816 <td class="td_left">The characters of the name.</td>
1817 </tr>
1818 </tbody>
1819</table>
1820</div>
Reid Spencere2f97b32004-07-05 22:28:02 +00001821<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001822<div class="doc_subsubsection"> <a name="symtab_plane">Symbol Table
1823Plane</a>
Reid Spencere2f97b32004-07-05 22:28:02 +00001824</div>
1825<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001826<p>A symbol table plane provides the symbol table entries for all
1827values of a common type. The encoding is given in the following table:</p>
Reid Spencere2f97b32004-07-05 22:28:02 +00001828<table>
Reid Spencere831f0b2004-07-29 00:13:04 +00001829 <tbody>
1830 <tr>
1831 <th><b>Type</b></th>
1832 <th class="td_left"><b>Field Description</b></th>
1833 </tr>
1834 <tr>
1835 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1836 <td class="td_left">Number of entries in this plane.</td>
1837 </tr>
1838 <tr>
1839 <td><a href="#uint32_vbr">uint32_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +00001840 <td class="td_left">Type slot number of type for all values in this plane..</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001841 </tr>
1842 <tr>
Reid Spencer375d7842004-08-18 20:06:19 +00001843 <td><a href="#value_entry">value_entry</a>+</td>
1844 <td class="td_left">The symbol table entries for to associate values with
1845 names.</td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001846 </tr>
1847 </tbody>
Reid Spencere2f97b32004-07-05 22:28:02 +00001848</table>
1849</div>
Reid Spencere2f97b32004-07-05 22:28:02 +00001850<!-- _______________________________________________________________________ -->
Reid Spencer375d7842004-08-18 20:06:19 +00001851<div class="doc_subsubsection"><a name="value_entry">Symbol Table Value
Reid Spencere831f0b2004-07-29 00:13:04 +00001852Entry</a>
Reid Spencere2f97b32004-07-05 22:28:02 +00001853</div>
1854<div class="doc_text">
Reid Spencer375d7842004-08-18 20:06:19 +00001855<p>A symbol table value entry provides the assocation between a value and the
1856name given to the value. The value is referenced by its slot number. The
Reid Spencere831f0b2004-07-29 00:13:04 +00001857format is given in the following table:</p>
Reid Spencere2f97b32004-07-05 22:28:02 +00001858<table>
Reid Spencere831f0b2004-07-29 00:13:04 +00001859 <tbody>
1860 <tr>
1861 <th><b>Type</b></th>
1862 <th class="td_left"><b>Field Description</b></th>
1863 </tr>
1864 <tr>
1865 <td><a href="#uint32_vbr">uint24_vbr</a></td>
Reid Spencer375d7842004-08-18 20:06:19 +00001866 <td class="td_left">Value slot number of the value being given a name.
Reid Spencer9a910812004-08-17 00:49:03 +00001867 </td>
Reid Spencere831f0b2004-07-29 00:13:04 +00001868 </tr>
1869 <tr>
1870 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1871 <td class="td_left">Length of the character array that follows.</td>
1872 </tr>
1873 <tr>
1874 <td><a href="#char">char</a>+</td>
1875 <td class="td_left">The characters of the name.</td>
1876 </tr>
1877 </tbody>
Reid Spencere2f97b32004-07-05 22:28:02 +00001878</table>
1879</div>
Reid Spencer375d7842004-08-18 20:06:19 +00001880
Reid Spencerb14a0a62004-06-08 07:41:41 +00001881<!-- *********************************************************************** -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001882<div class="doc_section"> <a name="versiondiffs">Version Differences</a>
1883</div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00001884<!-- *********************************************************************** -->
1885<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001886<p>This section describes the differences in the Bytecode Format across
1887LLVM
1888versions. The versions are listed in reverse order because it assumes
1889the current version is as documented in the previous sections. Each
1890section here
Chris Lattnere5589ea2004-07-05 18:05:48 +00001891describes the differences between that version and the one that <i>follows</i>.
Reid Spencerb14a0a62004-06-08 07:41:41 +00001892</p>
1893</div>
Chris Lattnercd196a92004-10-16 18:03:55 +00001894
Reid Spencerb14a0a62004-06-08 07:41:41 +00001895<!-- _______________________________________________________________________ -->
Reid Spencer3cff8a042004-08-18 20:17:05 +00001896<div class="doc_subsection"><a name="vers13">Version 1.3 Differences From
Chris Lattnercd196a92004-10-16 18:03:55 +00001897 1.4</a></div>
Reid Spencer0b85d032004-08-17 07:43:43 +00001898<!-- _______________________________________________________________________ -->
Chris Lattnercd196a92004-10-16 18:03:55 +00001899
1900<div class="doc_subsubsection">Unreachable Instruction</div>
1901<div class="doc_text">
1902 <p>The LLVM <a href="LangRef.html#i_unreachable">Unreachable</a> instruction
1903 was added in version 1.4 of LLVM. This caused all instruction numbers after
1904 it to shift down by one.</p>
1905</div>
1906
1907<div class="doc_subsubsection">Function Flags</div>
1908<div class="doc_text">
1909 <p>LLVM bytecode versions prior to 1.4 did not include the 5 bit offset
1910 in <a href="#funcfield">the function list</a> in the <a
1911 href="#globalinfo">Module Global Info</a> block.</p>
1912</div>
1913
1914<div class="doc_subsubsection">Function Flags</div>
1915<div class="doc_text">
1916 <p>LLVM bytecode versions prior to 1.4 did not include the 'undef' constant
1917 value, which affects the encoding of <a href="#constant">Constant
1918 Fields</a>.</p>
1919</div>
1920
Chris Lattnerdb6a9fd2004-10-16 00:29:30 +00001921<!--
Reid Spencer0b85d032004-08-17 07:43:43 +00001922<div class="doc_subsubsection">Aligned Data</div>
1923<div class="doc_text">
1924 <p>In version 1.3, certain data items were aligned to 32-bit boundaries. In
1925 version 1.4, alignment of data was done away with completely. The need for
1926 alignment has gone away and the only thing it adds is bytecode file size
1927 overhead. In most cases this overhead was small. However, in functions with
1928 large numbers of format 0 instructions (GEPs and PHIs with lots of parameters)
1929 or regular instructions with large valued operands (e.g. because there's just
1930 a lot of instructions in the function) the overhead can be extreme. In one
1931 test case, the overhead was 44,000 bytes (34% of the total file size).
1932 Consequently in release 1.4, the decision was made to eliminate alignment
1933 altogether.</p>
1934 <p>In version 1.3 format, the following bytecode constructs were aligned (i.e.
1935 they were followed by one to three bytes of padding):</p>
1936 <ul>
1937 <li>All blocks.</li>
1938 <li>Instructions using the long format (format 0).</li>
1939 <li>All call instructions that called a var args function.</li>
1940 <li>The target triple (a string field at the end of the module block).</li>
1941 <li>The version field (immediately following the signature).</li>
1942 </ul>
1943 <p>None of these constructs are aligned in version 1.4</p>
1944</div>
Chris Lattnerdb6a9fd2004-10-16 00:29:30 +00001945-->
Chris Lattnercd196a92004-10-16 18:03:55 +00001946
Reid Spencer0b85d032004-08-17 07:43:43 +00001947<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001948<div class="doc_subsection"><a name="vers12">Version 1.2 Differences
1949From 1.3</a></div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001950<!-- _______________________________________________________________________ -->
Chris Lattnercd196a92004-10-16 18:03:55 +00001951
Reid Spencer4e0352d2004-07-05 08:18:07 +00001952<div class="doc_subsubsection">Type Derives From Value</div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00001953<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001954<p>In version 1.2, the Type class in the LLVM IR derives from the Value
1955class. This is not the case in version 1.3. Consequently, in version
19561.2 the notion of a "Type Type" was used to write out values that were
1957Types. The types always occuped plane 12 (corresponding to the
1958TypeTyID) of any type planed set of values. In 1.3 this representation
1959is not convenient because the TypeTyID (12) is not present and its
1960value is now used for LabelTyID. Consequently, the data structures
1961written that involve types do so by writing all the types first and
1962then each of the value planes according to those types. In version 1.2,
1963the types would have been written intermingled with the values.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001964</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001965<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001966<div class="doc_subsubsection">Restricted getelementptr Types</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00001967<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00001968<p>In version 1.2, the getelementptr instruction required a ubyte type
1969index for accessing a structure field and a long type index for
1970accessing an array element. Consequently, it was only possible to
1971access structures of 255 or fewer elements. Starting in version 1.3,
1972this restriction was lifted. Structures must now be indexed with uint
1973constants. Arrays may now be indexed with int, uint, long, or ulong
1974typed values. The consequence of this was that the bytecode format had
1975to change in order to accommodate the larger range of structure indices.</p>
Reid Spencerb14a0a62004-06-08 07:41:41 +00001976</div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00001977<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00001978<div class="doc_subsubsection">Short Block Headers</div>
1979<div class="doc_text">
1980<p>In version 1.2, block headers were always 8 bytes being comprised of
1981both an unsigned integer type and an unsigned integer size. For very
1982small modules, these block headers turn out to be a large fraction of
1983the total bytecode file size. In an attempt to make these small files
1984smaller, the type and size information was encoded into a single
1985unsigned integer (4 bytes) comprised of 5 bits for the block type
1986(maximum 31 block types) and 27 bits for the block size (max
1987~134MBytes). These limits seemed sufficient for any blocks or sizes
1988forseen in the future. Note that the module block, which encloses all
1989the other blocks is still written as 8 bytes since bytecode files
1990larger than 134MBytes might be possible.</p>
1991</div>
1992<!-- _______________________________________________________________________ -->
1993<div class="doc_subsubsection">Dependent Libraries and Target Triples</div>
1994<div class="doc_text">
1995<p>In version 1.2, the bytecode format does not store module's target
1996triple or dependent. These fields have been added to the end of the <a
1997 href="#globalinfo">module global info block</a>. The purpose of these
1998fields is to allow a front end compiler to specifiy that the generated
1999module is specific to a particular target triple (operating
2000system/manufacturer/processor) which makes it non-portable; and to
2001allow front end compilers to specify the list of libraries that the
2002module depends on for successful linking.</p>
2003</div>
2004<!-- _______________________________________________________________________ -->
2005<div class="doc_subsubsection">Types Restricted to 24-bits</div>
2006<div class="doc_text">
2007<p>In version 1.2, type slot identifiers were written as 32-bit VBR
2008quantities. In 1.3 this has been reduced to 24-bits in order to ensure
2009that it is not possible to overflow the type field of a global variable
2010definition. 24-bits for type slot numbers is deemed sufficient for any
2011practical use of LLVM.</p>
2012</div>
2013<!-- _______________________________________________________________________ -->
2014<!-- _______________________________________________________________________ -->
2015<div class="doc_subsection"><a name="vers11">Version 1.1 Differences
2016From 1.2 </a></div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00002017<!-- _______________________________________________________________________ -->
2018<div class="doc_subsubsection">Explicit Primitive Zeros</div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00002019<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00002020<p>In version 1.1, the zero value for primitives was explicitly encoded
2021into the bytecode format. Since these zero values are constant values
2022in the LLVM IR and never change, there is no reason to explicitly
2023encode them. This explicit encoding was removed in version 1.2.</p>
Reid Spencer4e0352d2004-07-05 08:18:07 +00002024</div>
Reid Spencer4e0352d2004-07-05 08:18:07 +00002025<!-- _______________________________________________________________________ -->
2026<div class="doc_subsubsection">Inconsistent Module Global Info</div>
2027<div class="doc_text">
Reid Spencere831f0b2004-07-29 00:13:04 +00002028<p>In version 1.1, the Module Global Info block was not aligned causing
2029the next block to be read in on an unaligned boundary. This problem was
2030corrected in version 1.2.<br>
2031<br>
2032</p>
Reid Spencerb14a0a62004-06-08 07:41:41 +00002033</div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00002034<!-- _______________________________________________________________________ -->
Reid Spencere831f0b2004-07-29 00:13:04 +00002035<div class="doc_subsection"><a name="vers10">Version 1.0 Differences
2036From 1.1</a></div>
Reid Spencerb14a0a62004-06-08 07:41:41 +00002037<div class="doc_text">
Reid Spencer4e0352d2004-07-05 08:18:07 +00002038<p>None. Version 1.0 and 1.1 bytecode formats are identical.</p>
Reid Spencerb14a0a62004-06-08 07:41:41 +00002039</div>
Reid Spencer9acd4122004-05-22 02:28:36 +00002040<!-- *********************************************************************** -->
2041<hr>
Reid Spencere831f0b2004-07-29 00:13:04 +00002042<address> <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2043 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
2044<a href="http://validator.w3.org/check/referer"><img
2045 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
2046<a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and <a
2047 href="mailto:sabre@nondot.org">Chris Lattner</a><br>
2048<a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
2049Last modified: $Date$
Reid Spencer9acd4122004-05-22 02:28:36 +00002050</address>
Reid Spencere831f0b2004-07-29 00:13:04 +00002051</body>
2052</html>