1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Bytecode File Format</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
8 table, tr, td { border: 2px solid gray }
9 th { border: 2px solid gray; font-weight: bold; }
10 table { border-collapse: collapse; margin-top: 1em margin-bottom: 1em }
14 <div class="doc_title"> LLVM Bytecode File Format </div>
16 <li><a href="#abstract">Abstract</a></li>
17 <li><a href="#general">General Concepts</a>
19 <li><a href="#blocks">Blocks</a></li>
20 <li><a href="#lists">Lists</a></li>
21 <li><a href="#fields">Fields</a></li>
22 <li><a href="#slots">Slots</a></li>
23 <li><a href="#encoding">Encoding Rules</a></li>
24 <li><a href="#align">Alignment</a></li>
27 <li><a href="#details">Detailed Layout</a>
29 <li><a href="#notation">Notation</a></li>
30 <li><a href="#blocktypes">Blocks Types</a></li>
31 <li><a href="#signature">Signature Block</a></li>
32 <li><a href="#module">Module Block</a></li>
33 <li><a href="#typeool">Global Type Pool</a></li>
34 <li><a href="#modinfo">Module Info Block</a></li>
35 <li><a href="#constants">Global Constant Pool</a></li>
36 <li><a href="#functions">Function Blocks</a></li>
37 <li><a href="#symtab">Module Symbol Table</a></li>
40 <li><a href="#versiondiffs">Version Differences</a>
42 <li><a href="#vers12">Version 1.2 Differences From 1.3</a></li>
43 <li><a href="#vers11">Version 1.1 Differences From 1.2</a></li>
44 <li><a href="#vers10">Version 1.0 Differences From 1.1</a></li>
48 <div class="doc_author">
49 <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
52 <div class="doc_warning">
53 <p>Warning: This is a work in progress.</p>
55 <!-- *********************************************************************** -->
56 <div class="doc_section"> <a name="abstract">Abstract </a></div>
57 <!-- *********************************************************************** -->
58 <div class="doc_text">
59 <p>This document describes the LLVM bytecode
60 file format. It specifies the binary encoding rules of the bytecode file format
61 so that equivalent systems can encode bytecode files correctly. The LLVM
62 bytecode representation is used to store the intermediate representation on
63 disk in compacted form.
66 <!-- *********************************************************************** -->
67 <div class="doc_section"> <a name="general">General Concepts</a> </div>
68 <!-- *********************************************************************** -->
69 <div class="doc_text">
70 <p>This section describes the general concepts of the bytecode file format
71 without getting into bit and byte level specifics. Note that the LLVM bytecode
72 format may change in the future, but will always be backwards compatible with
73 older formats. This document only describes the most current version of the
76 <!-- _______________________________________________________________________ -->
77 <div class="doc_subsection"><a name="blocks">Blocks</a> </div>
78 <div class="doc_text">
79 <p>LLVM bytecode files consist simply of a sequence of blocks of bytes.
80 Each block begins with an identification value that determines the type of
81 the next block. The possible types of blocks are described below in the section
82 <a href="#blocktypes">Block Types</a>. The block identifier is used because
83 it is possible for entire blocks to be omitted from the file if they are
84 empty. The block identifier helps the reader determine which kind of block is
86 <p>The following block identifiers are currently in use
87 (from llvm/Bytecode/Format.h):</p>
89 <li><b>Module (0x01)</b>.</li>
90 <li><b>Function (0x11)</b>.</li>
91 <li><b>ConstantPool (0x12)</b>.</li>
92 <li><b>SymbolTable (0x13)</b>.</li>
93 <li><b>ModuleGlobalInfo (0x14)</b>.</li>
94 <li><b>GlobalTypePlane (0x15)</b>.</li>
95 <li><b>BasicBlock (0x31)</b>.</li>
96 <li><b>InstructionList (0x32)</b>.</li>
97 <li><b>CompactionTable (0x33)</b>.</li>
99 <p> All blocks are variable length, and the block header specifies the size of
100 the block. All blocks are rounded aligned to even 32-bit boundaries, so they
101 always start and end of this boundary. Each block begins with an integer
102 identifier and the length of the block, which does not include the padding
103 bytes needed for alignment.</p>
105 <!-- _______________________________________________________________________ -->
106 <div class="doc_subsection"><a name="lists">Lists</a> </div>
107 <div class="doc_text">
108 <p>Most blocks are constructed of lists of information. Lists can be constructed
109 of other lists, etc. This decomposition of information follows the containment
110 hierarchy of the LLVM Intermediate Representation. For example, a function
111 contains a list of instructions (the terminator instructions implicitly define
112 the end of the basic blocks).</p>
113 <p>A list is encoded into the file simply by encoding the number of entries as
114 an integer followed by each of the entries. The reader knows when the list is
115 done because it will have filled the list with the required numbe of entries.
118 <!-- _______________________________________________________________________ -->
119 <div class="doc_subsection"><a name="fields">Fields</a> </div>
120 <div class="doc_text">
121 <p>Fields are units of information that LLVM knows how to write atomically.
122 Most fields have a uniform length or some kind of length indication built into
123 their encoding. For example, a constant string (array of bytes) is
124 written simply as the length followed by the characters. Although this is
125 similar to a list, constant strings are treated atomically and are thus
127 <p>Fields use a condensed bit format specific to the type of information
128 they must contain. As few bits as possible are written for each field. The
129 sections that follow will provide the details on how these fields are
130 written and how the bits are to be interpreted.</p>
132 <!-- _______________________________________________________________________ -->
133 <div class="doc_subsection"><a name="slots">Slots</a> </div>
134 <div class="doc_text">
135 <p>The bytecode format uses the notion of a "slot" to reference Types and
136 Values. Since the bytecode file is a <em>direct</em> representation of LLVM's
137 intermediate representation, there is a need to represent pointers in the file.
138 Slots are used for this purpose. For example, if one has the following assembly:
141 <div class="doc_code">
142 %MyType = type { int, sbyte }<br>
143 %MyVar = external global %MyType
146 <p>there are two definitions. The definition of <tt>%MyVar</tt> uses
147 <tt>%MyType</tt>. In the C++ IR this linkage between <tt>%MyVar</tt> and
149 explicit through the use of C++ pointers. In bytecode, however, there's no
150 ability to store memory addresses. Instead, we compute and write out slot
151 numbers for every type and Value written to the file.</p>
152 <p>A slot number is simply an unsigned 32-bit integer encoded in the variable
153 bit rate scheme (see <a href="#encoding">encoding</a> below). This ensures that
154 low slot numbers are encoded in one byte. Through various bits of magic LLVM
155 attempts to always keep the slot numbers low. The first attempt is to associate
156 slot numbers with their "type plane". That is, Values of the same type are
157 written to the bytecode file in a list (sequentially). Their order in that list
158 determines their slot number. This means that slot #1 doesn't mean anything
159 unless you also specify for which type you want slot #1. Types are handled
160 specially and are always written to the file first (in the Global Type Pool) and
161 in such a way that both forward and backward references of the types can often be
162 resolved with a single pass through the type pool. </p>
163 <p>Slot numbers are also kept small by rearranging their order. Because of the
164 structure of LLVM, certain values are much more likely to be used frequently
165 in the body of a function. For this reason, a compaction table is provided in
166 the body of a function if its use would make the function body smaller.
167 Suppose you have a function body that uses just the types "int*" and "{double}"
168 but uses them thousands of time. Its worthwhile to ensure that the slot number
169 for these types are low so they can be encoded in a single byte (via vbr).
170 This is exactly what the compaction table does.</p>
172 <!-- _______________________________________________________________________ -->
173 <div class="doc_subsection"><a name="encoding">Encoding Primitives</a> </div>
174 <div class="doc_text">
175 <p>Each field that can be put out is encoded into the file using a small set
176 of primitives. The rules for these primitives are described below.</p>
177 <h3>Variable Bit Rate Encoding</h3>
178 <p>Most of the values written to LLVM bytecode files are small integers. To
179 minimize the number of bytes written for these quantities, an encoding
180 scheme similar to UTF-8 is used to write integer data. The scheme is known as
181 variable bit rate (vbr) encoding. In this encoding, the high bit of each
182 byte is used to indicate if more bytes follow. If (byte & 0x80) is non-zero
183 in any given byte, it means there is another byte immediately following that
184 also contributes to the value. For the final byte (byte & 0x80) is false
185 (the high bit is not set). In each byte only the low seven bits contribute to
186 the value. Consequently 32-bit quantities can take from one to <em>five</em>
187 bytes to encode. In general, smaller quantities will encode in fewer bytes,
189 <table class="doc_table_nw">
192 <th>Significant Bits</th>
193 <th>Maximum Value</th>
195 <tr><td>1</td><td>0-6</td><td>127</td></tr>
196 <tr><td>2</td><td>7-13</td><td>16,383</td></tr>
197 <tr><td>3</td><td>14-20</td><td>2,097,151</td></tr>
198 <tr><td>4</td><td>21-27</td><td>268,435,455</td></tr>
199 <tr><td>5</td><td>28-34</td><td>34,359,738,367</td></tr>
200 <tr><td>6</td><td>35-41</td><td>4,398,046,511,103</td></tr>
201 <tr><td>7</td><td>42-48</td><td>562,949,953,421,311</td></tr>
202 <tr><td>8</td><td>49-55</td><td>72,057,594,037,927,935</td></tr>
203 <tr><td>9</td><td>56-62</td><td>9,223,372,036,854,775,807</td></tr>
204 <tr><td>10</td><td>63-69</td><td>1,180,591,620,717,411,303,423</td></tr>
206 <p>Note that in practice, the tenth byte could only encode bit 63
207 since the maximum quantity to use this encoding is a 64-bit integer.</p>
209 <p><em>Signed</em> VBR values are encoded with the standard vbr encoding, but
210 with the sign bit as the low order bit instead of the high order bit. This
211 allows small negative quantities to be encoded efficiently. For example, -3
212 is encoded as "((3 << 1) | 1)" and 3 is encoded as "(3 << 1) |
213 0)", emitted with the standard vbr encoding above.</p>
215 <p>The table below defines the encoding rules for type names used in the
216 descriptions of blocks and fields in the next section. Any type name with
217 the suffix <em>_vbr</em> indicate a quantity that is encoded using
218 variable bit rate encoding as described above.</p>
219 <table class="doc_table" >
222 <th align="left"><b>Rule</b></th>
226 <td align="left">A 32-bit unsigned integer that always occupies four
227 consecutive bytes. The unsigned integer is encoded using LSB first
228 ordering. That is bits 2<sup>0</sup> through 2<sup>7</sup> are in the
229 byte with the lowest file offset (little endian).</td>
232 <td align="left">A 32-bit unsigned integer that occupies from one to five
233 bytes using variable bit rate encoding.</td>
236 <td align="left">A 64-bit unsigned integer that occupies from one to ten
237 bytes using variable bit rate encoding.</td>
240 <td align="left">A 64-bit signed integer that occupies from one to ten
241 bytes using the signed variable bit rate encoding.</td>
244 <td align="left">A single unsigned character encoded into one byte</td>
247 <td align="left">A single bit within a byte.</td>
250 <td align="left">A uint_vbr indicating the length of the character string
251 immediately followed by the characters of the string. There is no
252 terminating null byte in the string.</td>
255 <td align="left">An arbitrarily long segment of data to which no
256 interpretation is implied. This is used for float, double, and constant
261 <!-- _______________________________________________________________________ -->
262 <div class="doc_subsection"><a name="align">Alignment</a> </div>
263 <div class="doc_text">
264 <p>To support cross-platform differences, the bytecode file is aligned on
265 certain boundaries. This means that a small amount of padding (at most 3 bytes)
266 will be added to ensure that the next entry is aligned to a 32-bit boundary.
269 <!-- *********************************************************************** -->
270 <div class="doc_section"> <a name="details">Detailed Layout</a> </div>
271 <!-- *********************************************************************** -->
272 <div class="doc_text">
273 <p>This section provides the detailed layout of the LLVM bytecode file format.
274 bit and byte level specifics.</p>
276 <!-- _______________________________________________________________________ -->
277 <div class="doc_subsection"><a name="notation">Notation</a></div>
278 <div class="doc_text">
279 <p>The descriptions of the bytecode format that follow describe the bit
280 fields in detail. These descriptions are provided in tabular form. Each table
281 has four columns that specify:</p>
283 <li><b>Byte(s)</b>: The offset in bytes of the field from the start of
284 its container (block, list, other field).</li>
285 <li><b>Bit(s)</b>: The offset in bits of the field from the start of
286 the byte field. Bits are always little endian. That is, bit addresses with
287 smaller values have smaller address (i.e. 2<sup>0</sup> is at bit 0,
288 2<sup>1</sup> at 1, etc.)
290 <li><b>Align?</b>: Indicates if this field is aligned to 32 bits or not.
291 This indicates where the <em>next</em> field starts, always on a 32 bit
293 <li><b>Type</b>: The basic type of information contained in the field.</li>
294 <li><b>Description</b>: Describes the contents of the field.</li>
297 <!-- _______________________________________________________________________ -->
298 <div class="doc_subsection"><a name="blocktypes">Block Types</a></div>
299 <div class="doc_text">
300 <p>The bytecode format encodes the intermediate representation into groups
301 of bytes known as blocks. The blocks are written sequentially to the file in
302 the following order:</p>
304 <li><a href="#signature">Signature</a>: This contains the file signature
305 (magic number) that identifies the file as LLVM bytecode and the bytecode
307 <li><a href="#module">Module Block</a>: This is the top level block in a
308 bytecode file. It contains all the other blocks.</li>
309 <li><a href="#gtypepool">Global Type Pool</a>: This block contains all the
310 global (module) level types.</li>
311 <li><a href="#modinfo">Module Info</a>: This block contains the types of the
312 global variables and functions in the module as well as the constant
313 initializers for the global variables</li>
314 <li><a href="#constants">Constants</a>: This block contains all the global
315 constants except function arguments, global values and constant strings.</li>
316 <li><a href="#functions">Functions</a>: One function block is written for
317 each function in the module. </li>
318 <li><a href="#symtab">Symbol Table</a>: The module level symbol table that
319 provides names for the various other entries in the file is the final block
323 <!-- _______________________________________________________________________ -->
324 <div class="doc_subsection"><a name="signature">Signature Block</a> </div>
325 <div class="doc_text">
326 <p>The signature occurs in every LLVM bytecode file and is always first.
327 It simply provides a few bytes of data to identify the file as being an LLVM
328 bytecode file. This block is always four bytes in length and differs from the
329 other blocks because there is no identifier and no block length at the start
330 of the block. Essentially, this block is just the "magic number" for the file.
331 <table class="doc_table_nw" >
333 <th><b>Byte(s)</b></th>
334 <th><b>Bit(s)</b></th>
335 <th><b>Align?</b></th>
337 <th align="left"><b>Field Description</b></th>
339 <td>00</td><td>-</td><td>No</td><td>char</td>
340 <td align="left">Constant "l" (0x6C)</td>
342 <td>01</td><td>-</td><td>No</td><td>char</td>
343 <td align="left">Constant "l" (0x6C)</td>
345 <td>02</td><td>-</td><td>No</td><td>char</td>
346 <td align="left">Constant "v" (0x76)</td>
348 <td>03</td><td>-</td><td>No</td><td>char</td>
349 <td align="left">Constant "m" (0x6D)</td>
353 <!-- _______________________________________________________________________ -->
354 <div class="doc_subsection"><a name="module">Module Block</a> </div>
355 <div class="doc_text">
356 <p>The module block contains a small pre-amble and all the other blocks in
357 the file. Of particular note, the bytecode format number is simply a 28-bit
358 monotonically increase integer that identifiers the version of the bytecode
359 format (which is not directly related to the LLVM release number). The
360 bytecode versions defined so far are (note that this document only describes
361 the latest version): </p>
364 <li>#0: LLVM 1.0 & 1.1</li>
365 <li>#1: LLVM 1.2</li>
366 <li>#2: LLVM 1.3</li>
369 <p>The table below shows the format of the module block header. It is defined
370 by blocks described in other sections.</p>
371 <table class="doc_table_nw" >
373 <th><b>Byte(s)</b></th>
374 <th><b>Bit(s)</b></th>
375 <th><b>Align?</b></th>
377 <th align="left"><b>Field Description</b></th>
379 <td>04-07</td><td>-</td><td>No</td><td>unsigned</td>
380 <td align="left">Module Identifier (0x01)</td>
382 <td>08-11</td><td>-</td><td>No</td><td>unsigned</td>
383 <td align="left">Size of the module block in bytes</td>
385 <td>12-15</td><td>00</td><td>Yes</td><td>uint32_vbr</td>
386 <td align="left">Format Information</td>
388 <td>''</td><td>0</td><td>-</td><td>bit</td>
389 <td align="left">Big Endian?</td>
391 <td>''</td><td>1</td><td>-</td><td>bit</td>
392 <td align="left">Pointers Are 64-bit?</td>
394 <td>''</td><td>2</td><td>-</td><td>bit</td>
395 <td align="left">Has No Endianess?</td>
397 <td>''</td><td>3</td><td>-</td><td>bit</td>
398 <td align="left">Has No Pointer Size?</td>
400 <td>''</td><td>4-31</td><td>-</td><td>bit</td>
401 <td align="left">Bytecode Format Version</td>
403 <td>16-end</td><td>-</td><td>-</td><td>blocks</td>
404 <td align="left">The remaining bytes in the block consist
405 solely of other block types in sequence.</td>
409 <p>Note that we plan to eventually expand the target description capabilities
410 of bytecode files to <a href="http://llvm.cs.uiuc.edu/PR263">target
415 <!-- _______________________________________________________________________ -->
416 <div class="doc_subsection"><a name="gtypepool">Global Type Pool</a> </div>
417 <div class="doc_text">
418 <p>The global type pool consists of type definitions. Their order of appearance
419 in the file determines their slot number (0 based). Slot numbers are used to
420 replace pointers in the intermediate representation. Each slot number uniquely
421 identifies one entry in a type plane (a collection of values of the same type).
422 Since all values have types and are associated with the order in which the type
423 pool is written, the global type pool <em>must</em> be written as the first
424 block of a module. If it is not, attempts to read the file will fail because
425 both forward and backward type resolution will not be possible.</p>
426 <p>The type pool is simply a list of types definitions, as shown in the table
428 <table class="doc_table_nw" >
430 <th><b>Byte(s)</b></th>
431 <th><b>Bit(s)</b></th>
432 <th><b>Align?</b></th>
434 <th align="left"><b>Field Description</b></th>
436 <td>00-03</td><td>-</td><td>No</td><td>unsigned</td>
437 <td align="left">Type Pool Identifier (0x13)</td>
439 <td>04-07</td><td>-</td><td>No</td><td>unsigned</td>
440 <td align="left">Size in bytes of the symbol table block.</td>
442 <td>08-11<sup>1</sup></td><td>-</td><td>No</td><td>uint32_vbr</td>
443 <td align="left">Number of entries in type plane</td>
445 <td>12-15<sup>1</sup></td><td>-</td><td>No</td><td>uint32_vbr</td>
446 <td align="left">Type plane index for following entries</td>
448 <td>16-end<sup>1,2</sup></td><td>-</td><td>No</td><td>type</td>
449 <td align="left">Each of the type definitions.</td>
451 <td align="left" colspan="5"><sup>1</sup>Maximum length shown,
452 may be smaller<br><sup>2</sup>Repeated field.
456 <!-- _______________________________________________________________________ -->
457 <div class="doc_subsection"><a name="modinfo">Module Info</a> </div>
458 <div class="doc_text">
459 <p>To be determined.</p>
461 <!-- _______________________________________________________________________ -->
462 <div class="doc_subsection"><a name="constants">Constants</a> </div>
463 <div class="doc_text">
464 <p>To be determined.</p>
466 <!-- _______________________________________________________________________ -->
467 <div class="doc_subsection"><a name="functions">Functions</a> </div>
468 <div class="doc_text">
469 <p>To be determined.</p>
471 <!-- _______________________________________________________________________ -->
472 <div class="doc_subsection"><a name="symtab">Symbol Table</a> </div>
473 <div class="doc_text">
474 <p>A symbol table can be put out in conjunction with a module or a function.
475 A symbol table is a list of type planes. Each type plane starts with the number
476 of entries in the plane and the type plane's slot number (so the type can be
477 looked up in the global type pool). For each entry in a type plane, the slot
478 number of the value and the name associated with that value are written. The
479 format is given in the table below. </p>
480 <table class="doc_table_nw" >
482 <th><b>Byte(s)</b></th>
483 <th><b>Bit(s)</b></th>
484 <th><b>Align?</b></th>
486 <th align="left"><b>Field Description</b></th>
488 <td>00-03</td><td>-</td><td>No</td><td>unsigned</td>
489 <td align="left">Symbol Table Identifier (0x13)</td>
491 <td>04-07</td><td>-</td><td>No</td><td>unsigned</td>
492 <td align="left">Size in bytes of the symbol table block.</td>
494 <td>08-11<sup>1</sup></td><td>-</td><td>No</td><td>uint32_vbr</td>
495 <td align="left">Number of entries in type plane</td>
497 <td>12-15<sup>1</sup></td><td>-</td><td>No</td><td>uint32_vbr</td>
498 <td align="left">Type plane index for following entries</td>
500 <td>16-19<sup>1,2</sup></td><td>-</td><td>No</td><td>uint32_vbr</td>
501 <td align="left">Slot number of a value.</td>
503 <td>variable<sup>1,2</sup></td><td>-</td><td>No</td><td>string</td>
504 <td align="left">Name of the value in the symbol table.</td>
507 <td align="left" colspan="5"><sup>1</sup>Maximum length shown,
508 may be smaller<br><sup>2</sup>Repeated field.
512 <!-- *********************************************************************** -->
513 <div class="doc_section"> <a name="versiondiffs">Version Differences</a> </div>
514 <!-- *********************************************************************** -->
515 <div class="doc_text">
516 <p>This section describes the differences in the Bytecode Format across LLVM
517 versions. The versions are listed in reverse order because it assumes the
518 current version is as documented in the previous sections. Each section here
519 describes the differences between that version and the one that <i>follows</i>
522 <!-- _______________________________________________________________________ -->
523 <div class="doc_subsection">
524 <a name="vers12">Version 1.2 Differences From 1.3</a></div>
525 <div class="doc_text">
526 <p>TBD: How version 1.2 differs from version 1.3</p>
529 <!-- _______________________________________________________________________ -->
530 <div class="doc_subsection">
531 <a name="vers11">Version 1.1 Differences From 1.2 </a></div>
532 <div class="doc_text">
533 <p>TBD: How version 1.1 differs from version 1.2</p>
536 <!-- _______________________________________________________________________ -->
537 <div class="doc_subsection">
538 <a name="vers11">Version 1.0 Differences From 1.1</a></div>
539 <div class="doc_text">
540 <p>TBD: How version 1.0 differs from version 1.1</p>
543 <!-- *********************************************************************** -->
546 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
547 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
548 <a href="http://validator.w3.org/check/referer"><img
549 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
551 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
552 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
553 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
554 Last modified: $Date$