1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
4 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
5 <title>LLVM Bytecode File Format</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7 <style type="text/css">
8 TR, TD { border: 2px solid gray; padding-left: 4pt; padding-right: 4pt;
9 padding-top: 2pt; padding-bottom: 2pt; }
10 TH { border: 2px solid gray; font-weight: bold; font-size: 105%; }
11 TABLE { text-align: center; border: 2px solid black;
12 border-collapse: collapse; margin-top: 1em; margin-left: 1em;
13 margin-right: 1em; margin-bottom: 1em; }
14 .td_left { border: 2px solid gray; text-align: left; }
18 <div class="doc_title"> LLVM Bytecode File Format </div>
20 <li><a href="#abstract">Abstract</a></li>
21 <li><a href="#concepts">Concepts</a>
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>
27 <li><a href="#vbr">Variable Bit-Rate Encoding</a></li>
28 <li><a href="#encoding">Encoding Primitives</a></li>
29 <li><a href="#slots">Slots</a></li>
32 <li><a href="#general">General Structure</a> </li>
33 <li><a href="#blockdefs">Block Definitions</a>
35 <li><a href="#signature">Signature Block</a></li>
36 <li><a href="#module">Module Block</a></li>
37 <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="#instructionlist">Instructions List</a></li>
42 <li><a href="#instructions">Instructions</a></li>
43 <li><a href="#symtab">Symbol Table</a></li>
46 <li><a href="#versiondiffs">Version Differences</a>
48 <li><a href="#vers13">Version 1.3 Differences From 1.4</a></li>
49 <li><a href="#vers12">Version 1.2 Differences From 1.3</a></li>
50 <li><a href="#vers11">Version 1.1 Differences From 1.2</a></li>
51 <li><a href="#vers10">Version 1.0 Differences From 1.1</a></li>
55 <div class="doc_author">
56 <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
59 <!-- *********************************************************************** -->
60 <div class="doc_section"> <a name="abstract">Abstract </a></div>
61 <!-- *********************************************************************** -->
62 <div class="doc_text">
63 <p>This document describes the LLVM bytecode file format. It specifies
64 the binary encoding rules of the bytecode file format so that
65 equivalent systems can encode bytecode files correctly. The LLVM
66 bytecode representation is used to store the intermediate
67 representation on disk in compacted form.</p>
68 <p>The LLVM bytecode format may change in the future, but LLVM will
69 always be backwards compatible with older formats. This document will
70 only describe the most current version of the bytecode format. See <a
71 href="#versiondiffs">Version Differences</a> for the details on how
72 the current version is different from previous versions.</p>
74 <!-- *********************************************************************** -->
75 <div class="doc_section"> <a name="concepts">Concepts</a> </div>
76 <!-- *********************************************************************** -->
77 <div class="doc_text">
78 <p>This section describes the general concepts of the bytecode file
79 format without getting into specific layout details. It is recommended
80 that you read this section thoroughly before interpreting the detailed
83 <!-- _______________________________________________________________________ -->
84 <div class="doc_subsection"><a name="blocks">Blocks</a> </div>
85 <div class="doc_text">
86 <p>LLVM bytecode files consist simply of a sequence of blocks of bytes
87 using a binary encoding Each block begins with an header of two
88 unsigned integers. The first value identifies the type of block and the
89 second value provides the size of the block in bytes. The block
90 identifier is used because it is possible for entire blocks to be
91 omitted from the file if they are empty. The block identifier helps the
92 reader determine which kind of block is next in the file. Note that
93 blocks can be nested within other blocks.</p>
94 <p> All blocks are variable length, and the block header specifies the
95 size of the block. All blocks begin on a byte index that is aligned to
96 an even 32-bit boundary. That is, the first block is 32-bit aligned
97 because it starts at offset 0. Each block is padded with zero fill
98 bytes to ensure that the next block also starts on a 32-bit boundary.</p>
100 <!-- _______________________________________________________________________ -->
101 <div class="doc_subsection"><a name="lists">Lists</a> </div>
102 <div class="doc_text">
103 <p>LLVM Bytecode blocks often contain lists of things of a similar
104 type. For example, a function contains a list of instructions and a
105 function type contains a list of argument types. There are two basic
106 types of lists: length lists (<a href="#llist">llist</a>), and null
107 terminated lists (<a href="#zlist">zlist</a>), as described below in
108 the <a href="#encoding">Encoding Primitives</a>.</p>
110 <!-- _______________________________________________________________________ -->
111 <div class="doc_subsection"><a name="fields">Fields</a> </div>
112 <div class="doc_text">
113 <p>Fields are units of information that LLVM knows how to write atomically. Most
114 fields have a uniform length or some kind of length indication built into their
115 encoding. For example, a constant string (array of bytes) is written simply as
116 the length followed by the characters. Although this is similar to a list,
117 constant strings are treated atomically and are thus fields.</p>
118 <p>Fields use a condensed bit format specific to the type of information
119 they must contain. As few bits as possible are written for each field. The
120 sections that follow will provide the details on how these fields are
121 written and how the bits are to be interpreted.</p>
123 <!-- _______________________________________________________________________ -->
124 <div class="doc_subsection"><a name="align">Alignment</a> </div>
125 <div class="doc_text">
126 <p>To support cross-platform differences, the bytecode file is aligned on
127 certain boundaries. This means that a small amount of padding (at most 3
128 bytes) will be added to ensure that the next entry is aligned to a 32-bit
131 <!-- _______________________________________________________________________ -->
132 <div class="doc_subsection"><a name="vbr">Variable Bit-Rate Encoding</a>
134 <div class="doc_text">
135 <p>Most of the values written to LLVM bytecode files are small integers. To
136 minimize the number of bytes written for these quantities, an encoding scheme
137 similar to UTF-8 is used to write integer data. The scheme is known as
138 variable bit rate (vbr) encoding. In this encoding, the high bit of
139 each byte is used to indicate if more bytes follow. If (byte &
140 0x80) is non-zero in any given byte, it means there is another byte
141 immediately following that also contributes to the value. For the final
142 byte (byte & 0x80) is false (the high bit is not set). In each byte
143 only the low seven bits contribute to the value. Consequently 32-bit
144 quantities can take from one to <em>five</em> bytes to encode. In
145 general, smaller quantities will encode in fewer bytes, as follows:</p>
150 <th>Significant Bits</th>
151 <th>Maximum Value</th>
176 <td>34,359,738,367</td>
181 <td>4,398,046,511,103</td>
186 <td>562,949,953,421,311</td>
191 <td>72,057,594,037,927,935</td>
196 <td>9,223,372,036,854,775,807</td>
201 <td>1,180,591,620,717,411,303,423</td>
205 <p>Note that in practice, the tenth byte could only encode bit 63 since
206 the maximum quantity to use this encoding is a 64-bit integer.</p>
207 <p><em>Signed</em> VBR values are encoded with the standard vbr
208 encoding, but with the sign bit as the low order bit instead of the
209 high order bit. This allows small negative quantities to be encoded
210 efficiently. For example, -3
211 is encoded as "((3 << 1) | 1)" and 3 is encoded as "(3 <<
212 1) | 0)", emitted with the standard vbr encoding above.</p>
214 <!-- _______________________________________________________________________ -->
215 <div class="doc_subsection"><a name="encoding">Encoding Primitives</a> </div>
216 <div class="doc_text">
217 <p>Each field in the bytecode format is encoded into the file using a
218 small set of primitive formats. The table below defines the encoding
219 rules for the various primitives used and gives them each a type name.
220 The type names used in the descriptions of blocks and fields in the <a
221 href="#details">Detailed Layout</a>next section. Any type name with
222 the suffix <em>_vbr</em> indicates a quantity that is encoded using
223 variable bit rate encoding as described above.</p>
224 <table class="doc_table">
228 <th class="td_left"><b>Rule</b></th>
231 <td><a name="unsigned"><b>unsigned</b></a></td>
232 <td class="td_left">A 32-bit unsigned integer that always occupies four
233 consecutive bytes. The unsigned integer is encoded using LSB first
234 ordering. That is bits 2<sup>0</sup> through 2<sup>7</sup> are in the
235 byte with the lowest file offset (little endian).</td>
238 <td style="vertical-align: top;"><a name="uint24_vbr">
239 <b>uint24_vbr</b></a></td>
240 <td style="vertical-align: top; text-align: left;">A 24-bit unsigned
241 integer that occupies from one to four bytes using variable bit rate
245 <td><a name="uint32_vbr"><b>uint32_vbr</b></a></td>
246 <td class="td_left">A 32-bit unsigned integer that occupies from one to
247 five bytes using variable bit rate encoding.</td>
250 <td><a name="uint64_vbr"><b>uint64_vbr</b></a></td>
251 <td class="td_left">A 64-bit unsigned integer that occupies from one to
252 ten bytes using variable bit rate encoding.</td>
255 <td><a name="int64_vbr"><b>int64_vbr</b></a></td>
256 <td class="td_left">A 64-bit signed integer that occupies from one to ten
257 bytes using the signed variable bit rate encoding.</td>
260 <td><a name="char"><b>char</b></a></td>
261 <td class="td_left">A single unsigned character encoded into one byte</td>
264 <td><a name="bit"><b>bit(n-m)</b></a></td>
265 <td class="td_left">A set of bit within some larger integer field. The
266 values of <code>n</code> and <code>m</code> specify the inclusive range
267 of bits that define the subfield. The value for <code>m</code> may be
268 omitted if its the same as <code>n</code>.</td>
271 <td style="vertical-align: top;"><b><a name="float"><b>float</b></a></b>
273 <td style="vertical-align: top; text-align: left;">A floating point
274 value encoded as a 32-bit IEEE value written in little-endian form.<br>
278 <td style="vertical-align: top;"><b><b><a name="double"><b>double</b></a>
280 <td style="vertical-align: top; text-align: left;">A floating point value
281 encoded as a64-bit IEEE value written in little-endian form</td>
284 <td><a name="string"><b>string</b></a></td>
285 <td class="td_left">A uint32_vbr indicating the type of the constant
286 string which also includes its length, immediately followed by the
287 characters of the string. There is no terminating null byte in the
291 <td><a name="data"><b>data</b></a></td>
292 <td class="td_left">An arbitrarily long segment of data to which no
293 interpretation is implied. This is used for constant initializers.<br>
297 <td><a name="llist"><b>llist(x)</b></a></td>
298 <td class="td_left">A length list of x. This means the list is encoded
299 as an <a href="#uint32_vbr">uint32_vbr</a> providing the length of the
300 list, followed by a sequence of that many "x" items. This implies that
301 the reader should iterate the number of times provided by the length.
305 <td><a name="zlist"><b>zlist(x)</b></a></td>
306 <td class="td_left">A zero-terminated list of x. This means the list is
307 encoded as a sequence of an indeterminate number of "x" items, followed
308 by an <a href="#uint32_vbr">uint32_vbr</a> terminating value. This
309 implies that none of the "x" items can have a zero value (or else the
310 list terminates).</td>
313 <td><a name="block"><b>block</b></a></td>
314 <td class="td_left">A block of data that is logically related. A block
315 is an unsigned 32-bit integer that encodes the type of the block in
316 the low 5 bits and the size of the block in the high 27 bits. The
317 length does not include the block header or any alignment bytes at the
318 end of the block. Blocks may compose other blocks. </td>
323 <!-- _______________________________________________________________________ -->
324 <div class="doc_subsection"><a name="notation">Field Notation</a> </div>
325 <div class="doc_text">
326 <p>In the detailed block and field descriptions that follow, a regex
327 like notation is used to describe optional and repeated fields. A very
328 limited subset of regex is used to describe these, as given in the
329 following table: </p>
330 <table class="doc_table">
333 <th><b>Character</b></th>
334 <th class="td_left"><b>Meaning</b></th>
337 <td><b><code>?</code></b></td>
338 <td class="td_left">The question mark indicates 0 or 1 occurrences of
339 the thing preceding it.</td>
342 <td><b><code>*</code></b></td>
343 <td class="td_left">The asterisk indicates 0 or more occurrences of the
344 thing preceding it.</td>
347 <td><b><code>+</code></b></td>
348 <td class="td_left">The plus sign indicates 1 or more occurrences of the
349 thing preceding it.</td>
352 <td><b><code>()</code></b></td>
353 <td class="td_left">Parentheses are used for grouping.</td>
356 <td><b><code>,</code></b></td>
357 <td class="td_left">The comma separates sequential fields.</td>
361 <p>So, for example, consider the following specifications:</p>
362 <div class="doc_code">
364 <li><code>string?</code></li>
365 <li><code>(uint32_vbr,uin32_vbr)+</code></li>
366 <li><code>(unsigned?,uint32_vbr)*</code></li>
367 <li><code>(llist(unsigned))?</code></li>
370 <p>with the following interpretations:</p>
372 <li>An optional string. Matches either nothing or a single string</li>
373 <li>One or more pairs of uint32_vbr.</li>
374 <li>Zero or more occurrences of either an unsigned followed by a uint32_vbr
375 or just a uint32_vbr.</li>
376 <li>An optional length list of unsigned values.</li>
379 <!-- _______________________________________________________________________ -->
380 <div class="doc_subsection"><a name="slots">Slots</a> </div>
381 <div class="doc_text">
382 <p>The bytecode format uses the notion of a "slot" to reference Types
383 and Values. Since the bytecode file is a <em>direct</em> representation of
384 LLVM's intermediate representation, there is a need to represent pointers in
385 the file. Slots are used for this purpose. For example, if one has the
388 <div class="doc_code"><code> %MyType = type { int, sbyte }<br>
389 %MyVar = external global %MyType
391 <p>there are two definitions. The definition of <tt>%MyVar</tt> uses
393 In the C++ IR this linkage between <tt>%MyVar</tt> and <tt>%MyType</tt>
394 is explicit through the use of C++ pointers. In bytecode, however, there's no
395 ability to store memory addresses. Instead, we compute and write out
396 slot numbers for every Type and Value written to the file.</p>
397 <p>A slot number is simply an unsigned 32-bit integer encoded in the variable
398 bit rate scheme (see <a href="#encoding">encoding</a>). This ensures that
399 low slot numbers are encoded in one byte. Through various bits of magic LLVM
400 attempts to always keep the slot numbers low. The first attempt is to associate
401 slot numbers with their "type plane". That is, Values of the same type
402 are written to the bytecode file in a list (sequentially). Their order in
403 that list determines their slot number. This means that slot #1 doesn't mean
404 anything unless you also specify for which type you want slot #1. Types are
405 always written to the file first (in the <a href="#globaltypes">Global Type
406 Pool</a>) and in such a way that both forward and backward references of the
407 types can often be resolved with a single pass through the type pool. </p>
408 <p>In summary then, a slot number can be thought of as just a vbr encoded index
409 into a list of Type* or Value*. To keep slot numbers low, Value* are indexed by
410 two slot numbers: the "type plane index" (type slot) and the "value index"
413 <!-- *********************************************************************** -->
414 <div class="doc_section"> <a name="general">General Structure</a> </div>
415 <!-- *********************************************************************** -->
416 <div class="doc_text">
417 <p>This section provides the general structure of the LLVM bytecode
418 file format. The bytecode file format requires blocks to be in a
419 certain order and nested in a particular way so that an LLVM module can
420 be constructed efficiently from the contents of the file. This ordering
421 defines a general structure for bytecode files as shown below. The
422 table below shows the order in which all block types may appear. Please
423 note that some of the blocks are optional and some may be repeated. The
424 structure is fairly loose because optional blocks, if empty, are
425 completely omitted from the file.</p>
443 <td class="td_left"><a href="#signature">Signature</a></td>
444 <td class="td_left">This contains the file signature (magic
445 number) that identifies the file as LLVM bytecode.</td>
453 <td class="td_left"><a href="#module">Module</a></td>
454 <td class="td_left">This is the top level block in a bytecode
455 file. It contains all the other blocks. </td>
463 <td class="td_left"> <a href="#globaltypes">Global Type Pool</a></td>
464 <td class="td_left">This block contains all the global (module)
473 <td class="td_left"> <a href="#globalinfo">Module Globals Info</a></td>
474 <td class="td_left">This block contains the type, constness, and
475 linkage for each of the global variables in the module. It also
476 contains the type of the functions and the constant initializers.</td>
484 <td class="td_left"> <a href="#constantpool">Module Constant Pool</a></td>
485 <td class="td_left">This block contains all the global constants
486 except function arguments, global values and constant strings.</td>
494 <td class="td_left"> <a href="#functiondefs">Function Definitions</a>*</td>
495 <td class="td_left">One function block is written for each
496 function in the module. The function block contains the instructions,
497 type constant pool, and symbol table for the function.</td>
505 <td class="td_left"> <a
506 href="#constantpool">Function Constant Pool</a></td>
507 <td class="td_left">Any constants (including types) used solely within
508 the function are emitted here in the function constant pool. </td>
516 <td class="td_left"> <a
517 href="#instructionlist">Instruction List</a></td>
518 <td class="td_left">This block contains all the instructions of the
519 function. The basic blocks are inferred by terminating instructions.
528 <td class="td_left"> <a
529 href="#symtab">Function Symbol Table</a></td>
530 <td class="td_left">This symbol table provides the names for the function
531 specific values used (basic block labels mostly).</td>
539 <td class="td_left"> <a href="#symtab">Module Symbol Table</a></td>
540 <td class="td_left">This symbol table provides the names for the various
541 entries in the file that are not function specific (global vars, and
542 functions mostly).</td>
546 <p>Use the links in the table for details about the contents of each of
549 <!-- *********************************************************************** -->
550 <div class="doc_section"> <a name="blockdefs">Block Definitions</a> </div>
551 <!-- *********************************************************************** -->
552 <div class="doc_text">
553 <p>This section provides the detailed layout of the individual block
554 types in the LLVM bytecode file format. </p>
556 <!-- _______________________________________________________________________ -->
557 <div class="doc_subsection"><a name="signature">Signature Block</a> </div>
558 <div class="doc_text">
559 <p>The signature occurs in every LLVM bytecode file and is always first.
560 It simply provides a few bytes of data to identify the file as being an LLVM
561 bytecode file. This block is always four bytes in length and differs from the
562 other blocks because there is no identifier and no block length at the start
563 of the block. Essentially, this block is just the "magic number" for the file.
565 <p>There are two types of signatures for LLVM bytecode: uncompressed and
566 compressed as shown in the table below. </p>
571 <th class="td_left"><b>Uncompressed</b></th>
572 <th class="td_left"><b>Compressed</b></th>
575 <td><a href="#char">char</a></td>
576 <td class="td_left">Constant "l" (0x6C)</td>
577 <td class="td_left">Constant "l" (0x6C)</td>
580 <td><a href="#char">char</a></td>
581 <td class="td_left">Constant "l" (0x6C)</td>
582 <td class="td_left">Constant "l" (0x6C)</td>
585 <td><a href="#char">char</a></td>
586 <td class="td_left">Constant "v" (0x76)</td>
587 <td class="td_left">Constant "v" (0x76)</td>
590 <td><a href="#char">char</a></td>
591 <td class="td_left">Constant "m" (0x6D)</td>
592 <td class="td_left">Constant "c" (0x63)</td>
595 <td><a href="#char">char</a></td>
596 <td class="td_left">N/A</td>
597 <td class="td_left">'0'=null,'1'=gzip,'2'=bzip2</td>
601 <p>In other words, the uncompressed signature is just the characters 'llvm'
602 while the compressed signature is the characters 'llvc' followed by an ascii
603 digit ('0', '1', or '2') that indicates the kind of compression used. A value of
604 '0' indicates that null compression was used. This can happen when compression
605 was requested on a platform that wasn't configured for gzip or bzip2. A value of
606 '1' means that the rest of the file is compressed using the gzip algorithm and
607 should be uncompressed before interpretation. A value of '2' means that the rest
608 of the file is compressed using the bzip2 algorithm and should be uncompressed
609 before interpretation. In all cases, the data resulting from uncompression
610 should be interpreted as if it occurred immediately after the 'llvm'
611 signature (i.e. the uncompressed data begins with the
612 <a href="#module">Module Block</a></p>
613 <p><b>NOTE:</b> As of LLVM 1.4, all bytecode files produced by the LLVM tools
614 are compressed by default. To disable compression, pass the
615 <tt>--disable-compression</tt> option to the tool, if it supports it.
617 <!-- _______________________________________________________________________ -->
618 <div class="doc_subsection"><a name="module">Module Block</a> </div>
619 <div class="doc_text">
620 <p>The module block contains a small pre-amble and all the other blocks in
621 the file. The table below shows the structure of the module block. Note that it
622 only provides the module identifier, size of the module block, and the format
623 information. Everything else is contained in other blocks, described in other
629 <th class="td_left"><b>Field Description</b></th>
632 <td><a href="#unsigned">unsigned</a><br></td>
633 <td class="td_left"><a href="#mod_header">Module Block Identifier
637 <td><a href="#unsigned">unsigned</a></td>
638 <td class="td_left"><a href="#mod_header">Module Block Size</a></td>
641 <td><a href="#uint32_vbr">uint32_vbr</a></td>
642 <td class="td_left"><a href="#format">Format Information</a></td>
645 <td><a href="#block">block</a></td>
646 <td class="td_left"><a href="#globaltypes">Global Type Pool</a></td>
649 <td><a href="#block">block</a></td>
650 <td class="td_left"><a href="#globalinfo">Module Globals Info</a></td>
653 <td><a href="#block">block</a></td>
654 <td class="td_left"><a href="#constantpool">Module Constant Pool</a></td>
657 <td><a href="#block">block</a>*</td>
658 <td class="td_left"><a href="#functiondefs">Function Definitions</a></td>
661 <td><a href="#block">block</a></td>
662 <td class="td_left"><a href="#symtab">Module Symbol Table</a></td>
668 <!-- _______________________________________________________________________ -->
669 <div class="doc_subsubsection"><a name="mod_header">Module Block Header</a></div>
670 <div class="doc_text">
671 <p>The block header for the module block uses a longer format than the other
672 blocks in a bytecode file. Specifically, instead of encoding the type and size
673 of the block into a 32-bit integer with 5-bits for type and 27-bits for size,
674 the module block header uses two 32-bit unsigned values, one for type, and one
675 for size. While the 2<sup>27</sup> byte limit on block size is sufficient
676 for the blocks contained in the module, it isn't sufficient for the module
677 block itself because we want to ensure that bytecode files as large as
678 2<sup>32</sup> bytes are possible. For this reason, the module block (and
679 only the module block) uses a long format header.</p>
682 <!-- _______________________________________________________________________ -->
683 <div class="doc_subsubsection"><a name="format">Format Information</a></div>
684 <div class="doc_text">
685 <p>The format information field is encoded into a <a href="#uint32_vbr">uint32_vbr</a>.</p>
687 <p>Of particular note, the bytecode format number is simply a 32-bit
688 monotonically increasing integer that identifies the version of the bytecode
689 format (which is not directly related to the LLVM release number). The
690 bytecode versions defined so far are (note that this document only
691 describes the latest version, 2.0):</p>
693 <li>#0: LLVM 1.0 & 1.1</li>
694 <li>#1: LLVM 1.2</li>
695 <li>#2: LLVM 1.2.5 (not released)</li>
696 <li>#3: LLVM 1.3</li>
697 <li>#4: LLVM 1.3.x (not released)</li>
698 <li>#5: LLVM 1.4 through 1.8</li>
699 <li>#6: LLVM 1.9</li>
700 <li>#7: LLVM 2.0 and newer</li>
703 <!-- _______________________________________________________________________ -->
704 <div class="doc_subsection"><a name="globaltypes">Global Type Pool</a> </div>
705 <div class="doc_text">
706 <p>The global type pool consists of type definitions. Their order of appearance
707 in the file determines their type slot number (0 based). Slot numbers are
708 used to replace pointers in the intermediate representation. Each slot number
709 uniquely identifies one entry in a type plane (a collection of values of the
710 same type). Since all values have types and are associated with the order in
711 which the type pool is written, the global type pool <em>must</em> be written
712 as the first block of a module. If it is not, attempts to read the file will
713 fail because both forward and backward type resolution will not be possible.</p>
714 <p>The type pool is simply a list of type definitions, as shown in the
720 <th class="td_left"><b>Field Description</b></th>
723 <td><a href="#unsigned">block</a></td>
724 <td class="td_left">Type Pool Identifier (0x06) + Size<br>
728 <td><a href="#llist">llist</a>(<a href="#type">type</a>)</td>
729 <td class="td_left">A length list of type definitions.</td>
734 <!-- _______________________________________________________________________ -->
735 <div class="doc_subsubsection"><a name="type">Type Definitions</a></div>
736 <div class="doc_text">
737 <p>Types in the type pool are defined using a different format for each kind
738 of type, as given in the following sections.</p>
739 <h3>Primitive Types</h3>
740 <p>The primitive types encompass the basic integer and floating point
741 types. They are encoded simply as their TypeID.</p>
746 <th class="td_left"><b>Description</b></th>
749 <td><a href="#uint24_vbr">uint24_vbr</a></td>
750 <td class="td_left">Type ID for the primitive types (values 1 to 11)
757 <li>The values for the Type IDs for the primitive types are provided by the
758 definition of the <code>llvm::Type::TypeID</code> enumeration in
759 <code>include/llvm/Type.h</code>. The enumeration gives the following mapping:
775 <h3>Function Types</h3>
780 <th class="td_left"><b>Description</b></th>
783 <td><a href="#uint24_vbr">uint24_vbr</a></td>
784 <td class="td_left">Type ID for function types (13)</td>
787 <td><a href="#uint24_vbr">uint24_vbr</a></td>
788 <td class="td_left">Type slot number of function's return type.</td>
791 <td><a href="#llist">llist</a>(<a href="#uint24_vbr">uint24_vbr</a>)</td>
792 <td class="td_left">Type slot number of each argument's type.</td>
795 <td><a href="#uint32_vbr">uint32_vbr</a>?</td>
796 <td class="td_left">Value 0 if this is a varargs function, missing
801 <h3>Structure Types</h3>
806 <th class="td_left"><b>Description</b></th>
809 <td><a href="#uint24_vbr">uint24_vbr</a></td>
810 <td class="td_left">Type ID for structure types (14)</td>
813 <td><a href="#zlist">zlist</a>(<a href="#uint24_vbr">uint24_vbr</a>)</td>
814 <td class="td_left">Slot number of each of the element's fields.</td>
823 <th class="td_left"><b>Description</b></th>
826 <td><a href="#uint24_vbr">uint24_vbr</a></td>
827 <td class="td_left">Type ID for Array Types (15)</td>
830 <td><a href="#uint24_vbr">uint24_vbr</a></td>
831 <td class="td_left">Type slot number of array's element type.</td>
834 <td><a href="#uint32_vbr">uint32_vbr</a></td>
835 <td class="td_left">The number of elements in the array.</td>
839 <h3>Pointer Types</h3>
844 <th class="td_left"><b>Description</b></th>
847 <td><a href="#uint24_vbr">uint24_vbr</a></td>
848 <td class="td_left">Type ID For Pointer Types (16)</td>
851 <td><a href="#uint24_vbr">uint24_vbr</a></td>
852 <td class="td_left">Type slot number of pointer's element type.</td>
856 <h3>Opaque Types</h3>
861 <th class="td_left"><b>Description</b></th>
864 <td><a href="#uint24_vbr">uint24_vbr</a></td>
865 <td class="td_left">Type ID For Opaque Types (17)</td>
869 <h3>Vector Types</h3>
874 <th class="td_left"><b>Description</b></th>
877 <td><a href="#uint24_vbr">uint24_vbr</a></td>
878 <td class="td_left">Type ID for Vector Types (18)</td>
881 <td><a href="#uint24_vbr">uint24_vbr</a></td>
882 <td class="td_left">Slot number of the vector's element type.</td>
885 <td><a href="#uint32_vbr">uint32_vbr</a></td>
886 <td class="td_left">The number of elements in the vector.</td>
890 <h3>Packed Structure Types</h3>
895 <th class="td_left"><b>Description</b></th>
898 <td><a href="#uint24_vbr">uint24_vbr</a></td>
899 <td class="td_left">Type ID for packed structure types (19)</td>
902 <td><a href="#zlist">zlist</a>(<a href="#uint24_vbr">uint24_vbr</a>)</td>
903 <td class="td_left">Slot number of each of the element's fields.</td>
908 <!-- _______________________________________________________________________ -->
909 <div class="doc_subsection"><a name="globalinfo">Module Global Info</a>
911 <div class="doc_text">
912 <p>The module global info block contains the definitions of all global
913 variables including their initializers and the <em>declaration</em> of
914 all functions. The format is shown in the table below:</p>
919 <th class="td_left"><b>Field Description</b></th>
922 <td><a href="#block">block</a></td>
923 <td class="td_left">Module global info identifier (0x05) + size</td>
926 <td><a href="#zlist">zlist</a>(<a href="#globalvar">globalvar</a>)</td>
927 <td class="td_left">A zero terminated list of global var definitions
928 occurring in the module.</td>
931 <td><a href="#zlist">zlist</a>(<a href="#funcfield">funcfield</a>)</td>
932 <td class="td_left">A zero terminated list of function definitions
933 occurring in the module.</td>
936 <td><a href="#llist">llist</a>(<a href="#string">string</a>)</td>
937 <td class="td_left">A length list of strings that specify the names of
938 the libraries that this module depends upon.</td>
941 <td><a href="#string">string</a></td>
942 <td class="td_left">The target triple for the module (blank means no
943 target triple specified, i.e. a platform-independent module).</td>
946 <td><a href="#string">string</a></td>
947 <td class="td_left">The data layout string describing the endianness,
948 pointer size, and type alignments for which the module was written
949 (blank means no data layout specified, i.e. a platform-independent
953 <td><a href="#llist">llist</a>(<a href="#string">string</a>)</td>
954 <td class="td_left">A length list of strings that defines a table of
955 section strings for globals. A global's SectionID is an index into
959 <td><a href="#string">string</a></td>
960 <td class="td_left">The inline asm block for this module.</td>
966 <!-- _______________________________________________________________________ -->
967 <div class="doc_subsubsection"><a name="globalvar">Global Variable Field</a>
970 <div class="doc_text">
972 <p>Global variables are written using an <a href="#uint32_vbr">uint32_vbr</a>
973 that encodes information about the global variable, an optional extension vbr,
974 and a an optional initializers for the global var.</p>
976 <p>The table below provides the bit layout of the first <a
977 href="#uint32_vbr">uint32_vbr</a> that describes the global variable.</p>
983 <th class="td_left"><b>Description</b></th>
986 <td><a href="#bit">bit(0)</a></td>
987 <td class="td_left">Is constant?</td>
990 <td><a href="#bit">bit(1)</a></td>
991 <td class="td_left">Has initializer? Note that this bit determines
992 whether the constant initializer field (described below) follows.</td>
995 <td><a href="#bit">bit(2-4)</a></td>
996 <td class="td_left">Linkage type: 0=External, 1=Weak,
997 2=Appending, 3=Internal, 4=LinkOnce, 5=DllImport,
998 6=DllExport, 7=ExternWeak</td>
1001 <td><a href="#bit">bit(5-31)</a></td>
1002 <td class="td_left">Type slot number of type for the global variable.</td>
1007 <p>When the Linkage type is set to 3 (internal) and the initializer field is set
1008 to 0 (an invalid combination), an extension word follows the first <a
1009 href="#uint32_vbr">uint32_vbr</a> which encodes the real linkage and init flag,
1010 and can includes more information:</p>
1015 <th><b>Type</b></th>
1016 <th class="td_left"><b>Description</b></th>
1019 <td><a href="#bit">bit(0)</a></td>
1020 <td class="td_left">Has initializer? Indicates the real value of the "Has
1021 initializer" field for the global. </td>
1024 <td><a href="#bit">bit(2-4)</a></td>
1025 <td class="td_left">Linkage type: Indicates the real value of the "linkage
1026 type" field for the global.</td>
1029 <td><a href="#bit">bit(4-8)</a></td>
1030 <td class="td_left">The log-base-2 of the alignment for the global.</td>
1033 <td><a href="#bit">bit(9)</a></td>
1034 <td class="td_left">If this bit is set, a SectionID follows this vbr.</td>
1037 <td><a href="#bit">bit(10-12)</a></td>
1038 <td class="td_left">Visibility style: 0=Default, 1=Hidden.</td>
1041 <td><a href="#bit">bit(13-31)</a></td>
1042 <td class="td_left">Currently unassigned.</td>
1047 <p>If the SectionID bit is set above, the following field is included:</p>
1052 <th><b>Type</b></th>
1053 <th class="td_left"><b>Description</b></th>
1056 <td><a href="#uint32_vbr">uint32_vbr</a>
1058 <td class="td_left">An optional section ID number, specifying the string
1059 to use for the section of the global. This an index (+1) of an entry
1060 into the SectionID llist in the
1061 <a href="#globalinfo">Module Global Info</a> block. If this value is
1062 0 or not present, the global has an empty section string.</td>
1067 <p>If the "Has initializer" field is set, the following field is included:</p>
1072 <th><b>Type</b></th>
1073 <th class="td_left"><b>Description</b></th>
1076 <td><a href="#uint32_vbr">uint32_vbr</a>
1078 <td class="td_left">An optional value slot number for the global
1079 variable's constant initializer.</td>
1085 <!-- _______________________________________________________________________ -->
1086 <div class="doc_subsubsection"><a name="funcfield">Function Field</a>
1088 <div class="doc_text">
1089 <p>Functions are written using an <a href="#uint32_vbr">uint32_vbr</a>
1090 that encodes information about the function and a set of flags. If needed,
1091 an extension word may follow this first field.</p>
1093 <p>The table below provides the bit layout of the <a
1094 href="#uint32_vbr">uint32_vbr</a> that describes the function.</p>
1099 <th><b>Type</b></th>
1100 <th class="td_left"><b>Description</b></th>
1103 <td><a href="#bit">bit(0-3)</a></td>
1104 <td class="td_left">
1105 Encodes the calling convention number of the function. The
1106 CC number of the function is the value of this field minus one.
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
1113 <a href="#functiondefs">Function Definiton Block</a> in the bytecode
1114 file for the function. If the function is external and has
1115 <tt>dllimport or extern_weak</tt> linkage additional field in the
1116 extension word is used to indicate the actual linkage type.</td>
1119 <td><a href="#bit">bit(5-30)</a></td>
1120 <td class="td_left">Type slot number of type for the function.</td>
1123 <td><a href="#bit">bit(31)</a></td>
1124 <td class="td_left">Indicates whether an extension word follows.</td>
1129 <p>If bit(31) is set, an additional <a href="#uint32_vbr">uint32_vbr</a> word
1130 follows with the following fields:</p>
1135 <th><b>Type</b></th>
1136 <th class="td_left"><b>Description</b></th>
1139 <td><a href="#bit">bit(0-4)</a></td>
1140 <td class="td_left">The log-base-2 of the alignment for the function.</td>
1143 <td><a href="#bit">bit(5-9)</a></td>
1144 <td class="td_left">The top nibble of the calling convention.</td>
1147 <td><a href="#bit">bit(10)</a></td>
1148 <td class="td_left">If this bit is set, a SectionID follows this vbr.</td>
1151 <td><a href="#bit">bit(11-12)</a></td>
1152 <td class="td_left">Linkage type for external functions. 0 - External
1153 linkage, 1 - DLLImport linkage, 2 - External weak linkage.</td>
1156 <td><a href="#bit">bit(13-31)</a></td>
1157 <td class="td_left">Currently unassigned.</td>
1162 <p>If the SectionID bit is set above, the following field is included:</p>
1167 <th><b>Type</b></th>
1168 <th class="td_left"><b>Description</b></th>
1171 <td><a href="#uint32_vbr">uint32_vbr</a>
1173 <td class="td_left">An optional section ID number, specifying the string
1174 to use for the section of the function. This an index (+1) of an entry
1175 into the SectionID llist in the
1176 <a href="#globalinfo">Module Global Info</a> block. If this value is
1177 0 or not present, the function has an empty section string.</td>
1184 <!-- _______________________________________________________________________ -->
1185 <div class="doc_subsection"><a name="constantpool">Constant Pool</a> </div>
1186 <div class="doc_text">
1187 <p>A constant pool defines as set of constant values. There are
1188 actually two types of constant pool blocks: one for modules and one for
1189 functions. For modules, the block begins with the constant strings
1190 encountered anywhere in the module. For functions, the block begins
1191 with types only encountered in the function. In both cases the header
1192 is identical. The tables that follow, show the header, module constant
1193 pool preamble, function constant pool preamble, and the part common to
1194 both function and module constant pools.</p>
1195 <p><b>Common Block Header</b></p>
1199 <th><b>Type</b></th>
1200 <th class="td_left"><b>Field Description</b></th>
1203 <td><a href="#block">block</a></td>
1204 <td class="td_left">Constant pool identifier (0x03) + size<br>
1209 <p><b>Module Constant Pool Preamble (constant strings)</b></p>
1213 <th><b>Type</b></th>
1214 <th class="td_left"><b>Field Description</b></th>
1217 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1218 <td class="td_left">The number of constant strings that follow.</td>
1221 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1222 <td class="td_left">Zero. This identifies the following "plane" as
1223 containing the constant strings. This is needed to identify it uniquely
1224 from other constant planes that follow. </td>
1227 <td><a href="#uint24_vbr">uint24_vbr</a>+</td>
1228 <td class="td_left">Type slot number of the constant string's type. Note
1229 that the constant string's type implicitly defines the length of the
1234 <p><b>Function Constant Pool Preamble (function types)</b></p>
1235 <p>The structure of the types for functions is identical to the <a
1236 href="#globaltypes">Global Type Pool</a>. Please refer to that section
1237 for the details. </p>
1238 <p><b>Common Part (other constants)</b></p>
1242 <th><b>Type</b></th>
1243 <th class="td_left"><b>Field Description</b></th>
1246 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1247 <td class="td_left">Number of entries in this type plane.</td>
1250 <td><a href="#uint24_vbr">uint24_vbr</a></td>
1251 <td class="td_left">Type slot number of this plane.</td>
1254 <td><a href="#constant">constant</a>+</td>
1255 <td class="td_left">The definition of a constant (see below).</td>
1261 <!-- _______________________________________________________________________ -->
1262 <div class="doc_subsubsection"><a name="constant">Simple Constant Pool
1265 <div class="doc_text">
1267 <p>Constant pool entries come in many shapes and flavors. The sections that
1268 follow define the format for each of them. All constants start with a <a
1269 href="#uint32_vbr">uint32_vbr</a> encoded integer that provides the
1270 number of operands for the constant. For primitive, structure, and
1271 array constants, this will always be zero to indicate that the form of the
1272 constant is solely determined by its type. In this case, we have the following
1273 field definitions, based on type:</p>
1276 <li><b>Bool</b>. This is written as an <a href="#uint32_vbr">uint32_vbr</a> of
1277 value 1U or 0U.</li>
1278 <li><b>Signed Integers (sbyte,short,int,long)</b>. These are written as an
1279 <a href="#int64_vbr">int64_vbr</a> with the corresponding value.</li>
1280 <li><b>Unsigned Integers (ubyte,ushort,uint,ulong)</b>. These are written as
1281 an <a href="#uint64_vbr">uint64_vbr</a> with the corresponding value. </li>
1282 <li><b>Floating Point</b>. Both the float and double types are written
1283 literally in binary format.</li>
1284 <li><b>Arrays</b>. Arrays are written simply as a list of
1285 <a href="#uint32_vbr">uint32_vbr</a> encoded value slot numbers to the
1286 constant element values.</li>
1287 <li><b>Structures</b>. Structures are written simply as a list of
1288 <a href="#uint32_vbr">uint32_vbr</a> encoded value slot numbers to the
1289 constant field values of the structure.</li>
1294 <!-- _______________________________________________________________________ -->
1295 <div class="doc_subsubsection">Undef Entries</div>
1297 <div class="doc_text">
1298 <p>When the number of operands to the constant is one, we have an 'undef' value
1299 of the specified type.</p>
1302 <!-- _______________________________________________________________________ -->
1303 <div class="doc_subsubsection">Inline Assembler Entries</div>
1305 <div class="doc_text">
1306 <p>Inline Assembler entries are stored in the constant pool, though they are not
1307 officially LLVM constants. These entries are marked with a value of
1308 "4294967295" (all ones) for the number of operands. They are encoded as
1314 <th><b>Type</b></th>
1315 <th class="td_left"><b>Field Description</b></th>
1318 <td><a href="#string">string</a></td>
1319 <td class="td_left">The asm string.</td>
1322 <td><a href="#string">string</a></td>
1323 <td class="td_left">The constraints string.</td>
1326 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1327 <td class="td_left">Flags</td>
1332 <p>Currently, the only defined flag, the low bit, indicates whether or not the
1333 inline assembler has side effects.</p>
1337 <!-- _______________________________________________________________________ -->
1338 <div class="doc_subsubsection">Constant Expression Entries</div>
1340 <div class="doc_text">
1342 <p>Otherwise, we have a constant expression. The format of the constant
1343 expression is specified in the table below, and the number is equal to the
1344 number of operands+1.</p>
1348 <th><b>Type</b></th>
1349 <th class="td_left"><b>Field Description</b></th>
1352 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1353 <td class="td_left">Op code of the instruction for the constant
1357 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1358 <td class="td_left">The value slot number of the constant value for an
1359 operand.<sup>1</sup></td>
1362 <td><a href="#uint24_vbr">uint24_vbr</a></td>
1363 <td class="td_left">The type slot number for the type of the constant
1364 value for an operand.<sup>1</sup></td>
1370 <li>Both these fields are repeatable but only in pairs.</li>
1373 <!-- _______________________________________________________________________ -->
1374 <div class="doc_subsection"><a name="functiondefs">Function Definition</a></div>
1375 <div class="doc_text">
1376 <p>Function definitions contain the linkage, constant pool, instruction list,
1377 and symbol table for a function. The following table shows the structure of
1378 a function definition.</p>
1382 <th><b>Type</b></th>
1383 <th class="td_left"><b>Field Description</b></th>
1386 <td><a href="#block">block</a><br>
1388 <td class="td_left">Function definition block identifier (0x02) +
1393 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1394 <td class="td_left">
1395 <a href="#funclinkage_and_visibility">The linkage and visibility</a>
1399 <td><a href="#block">block</a></td>
1400 <td class="td_left">The <a href="#constantpool">constant pool</a> block
1401 for this function.<sup>2</sup></td>
1404 <td><a href="#block">block</a></td>
1405 <td class="td_left">The <a href="#instructionlist">instruction list</a>
1406 for the function.</td>
1409 <td><a href="#block">block</a></td>
1410 <td class="td_left">The function's <a href="#symtab">symbol table</a>
1411 containing only those symbols pertinent to the function (mostly block
1418 <li>Note that if the linkage type is "External" then none of the
1419 other fields will be present as the function is defined elsewhere.</li>
1422 <!-- _______________________________________________________________________ -->
1423 <div class="doc_subsubsection"><a name="funclinkage_and_visibility">Linkage and
1426 <div class="doc_text">
1431 <th><b>Type</b></th>
1432 <th class="td_left"><b>Field Description</b></th>
1435 <td><a href="#bit">bit(0-15)</a></td>
1436 <td class="td_left">The linkage type of the function: 0=External, 1=Weak,
1437 2=Appending, 3=Internal, 4=LinkOnce, 5=DllImport,
1438 6=DllExport<sup>1</sup></td>
1441 <td><a href="#bit">bit(16-18)</a></td>
1442 <td class="td_left">Visibility style: 0=Default, 1=Hidden.</td>
1445 <td><a href="#bit">bit(19-31)</a></td>
1446 <td class="td_left">Currently unassigned.</td>
1452 <!-- _______________________________________________________________________ -->
1453 <div class="doc_subsection"><a name="instructionlist">Instruction List</a></div>
1454 <div class="doc_text">
1455 <p>The instructions in a function are written as a simple list. Basic
1456 blocks are inferred by the terminating instruction types. The format of
1457 the block is given in the following table.</p>
1461 <th><b>Type</b></th>
1462 <th class="td_left"><b>Field Description</b></th>
1465 <td><a href="#block">block</a><br>
1467 <td class="td_left">Instruction list identifier (0x07) + size<br>
1471 <td><a href="#instruction">instruction</a>+</td>
1472 <td class="td_left">An instruction. Instructions have a variety of
1473 formats. See <a href="#instruction">Instructions</a> for details.</td>
1479 <!-- _______________________________________________________________________ -->
1480 <div class="doc_subsection"><a name="instructions">Instructions</a></div>
1482 <div class="doc_text">
1483 <p>Instructions are written out one at a time as distinct units. Each
1485 record contains at least an <a href="#opcodes">opcode</a> and a type field,
1486 and may contain a <a href="#instoperands">list of operands</a> (whose
1487 interpretation depends on the opcode). Based on the number of operands, the
1488 <a href="#instencode">instruction is encoded</a> in a
1489 dense format that tries to encoded each instruction into 32-bits if
1493 <!-- _______________________________________________________________________ -->
1494 <div class="doc_subsubsection"><a name="opcodes">Instruction Opcodes</a></div>
1495 <div class="doc_text">
1496 <p>Instructions encode an opcode that identifies the kind of instruction.
1497 Opcodes are an enumerated integer value. The specific values used depend on
1498 the version of LLVM you're using. The opcode values are defined in the
1499 <a href="http://llvm.org/cvsweb/cvsweb.cgi/llvm/include/llvm/Instruction.def">
1500 <tt>include/llvm/Instruction.def</tt></a> file. You should check there for the
1501 most recent definitions. The table below provides the opcodes defined as of
1502 the writing of this document. The table associates each opcode mnemonic with
1503 its enumeration value and the bytecode and LLVM version numbers in which the
1504 opcode was introduced.</p>
1510 <th>Bytecode Version</th>
1511 <th>LLVM Version</th>
1513 <tr><td colspan="4"><b>Terminator Instructions</b></td></tr>
1514 <tr><td>Ret</td><td>1</td><td>1</td><td>1.0</td></tr>
1515 <tr><td>Br</td><td>2</td><td>1</td><td>1.0</td></tr>
1516 <tr><td>Switch</td><td>3</td><td>1</td><td>1.0</td></tr>
1517 <tr><td>Invoke</td><td>4</td><td>1</td><td>1.0</td></tr>
1518 <tr><td>Unwind</td><td>5</td><td>1</td><td>1.0</td></tr>
1519 <tr><td>Unreachable</td><td>6</td><td>1</td><td>1.4</td></tr>
1520 <tr><td colspan="4"><b>Binary Operators</b></td></tr>
1521 <tr><td>Add</td><td>7</td><td>1</td><td>1.0</td></tr>
1522 <tr><td>Sub</td><td>8</td><td>1</td><td>1.0</td></tr>
1523 <tr><td>Mul</td><td>9</td><td>1</td><td>1.0</td></tr>
1524 <tr><td>UDiv</td><td>10</td><td>6</td><td>1.9</td></tr>
1525 <tr><td>SDiv</td><td>11</td><td>6</td><td>1.9</td></tr>
1526 <tr><td>FDiv</td><td>12</td><td>6</td><td>1.9</td></tr>
1527 <tr><td>URem</td><td>13</td><td>6</td><td>1.9</td></tr>
1528 <tr><td>SRem</td><td>14</td><td>6</td><td>1.9</td></tr>
1529 <tr><td>FRem</td><td>15</td><td>6</td><td>1.9</td></tr>
1530 <tr><td colspan="4"><b>Logical Operators</b></td></tr>
1531 <tr><td>Shl</td><td>16</td><td>1</td><td>1.0</td></tr>
1532 <tr><td>LShr</td><td>17</td><td>6</td><td>1.9</td></tr>
1533 <tr><td>AShr</td><td>18</td><td>6</td><td>1.9</td></tr>
1534 <tr><td>And</td><td>19</td><td>1</td><td>1.0</td></tr>
1535 <tr><td>Or</td><td>20</td><td>1</td><td>1.0</td></tr>
1536 <tr><td>Xor</td><td>21</td><td>1</td><td>1.0</td></tr>
1537 <tr><td colspan="4"><b>Memory Operators</b></td></tr>
1538 <tr><td>Malloc</td><td>22</td><td>1</td><td>1.0</td></tr>
1539 <tr><td>Free</td><td>23</td><td>1</td><td>1.0</td></tr>
1540 <tr><td>Alloca</td><td>24</td><td>1</td><td>1.0</td></tr>
1541 <tr><td>Load</td><td>25</td><td>1</td><td>1.0</td></tr>
1542 <tr><td>Store</td><td>26</td><td>1</td><td>1.0</td></tr>
1543 <tr><td>GetElementPtr</td><td>27</td><td>1</td><td>1.0</td></tr>
1544 <tr><td colspan="4"><b>Cast Operators</b></td></tr>
1545 <tr><td>Trunc</td><td>28</td><td>7</td><td>2.0</td></tr>
1546 <tr><td>ZExt</td><td>29</td><td>7</td><td>2.0</td></tr>
1547 <tr><td>SExt</td><td>30</td><td>7</td><td>2.0</td></tr>
1548 <tr><td>FPToUI</td><td>31</td><td>7</td><td>2.0</td></tr>
1549 <tr><td>FPToSI</td><td>32</td><td>7</td><td>2.0</td></tr>
1550 <tr><td>UIToFP</td><td>33</td><td>7</td><td>2.0</td></tr>
1551 <tr><td>SIToFP</td><td>34</td><td>7</td><td>2.0</td></tr>
1552 <tr><td>FPTrunc</td><td>35</td><td>7</td><td>2.0</td></tr>
1553 <tr><td>FPExt</td><td>36</td><td>7</td><td>2.0</td></tr>
1554 <tr><td>PtrToInt</td><td>37</td><td>7</td><td>2.0</td></tr>
1555 <tr><td>IntToPtr</td><td>38</td><td>7</td><td>2.0</td></tr>
1556 <tr><td>BitCast</td><td>39</td><td>7</td><td>2.0</td></tr>
1557 <tr><td colspan="4"><b>Other Operators</b></td></tr>
1558 <tr><td>ICmp</td><td>40</td><td>7</td><td>2.0</td></tr>
1559 <tr><td>FCmp</td><td>41</td><td>7</td><td>2.0</td></tr>
1560 <tr><td>PHI</td><td>42</td><td>1</td><td>1.0</td></tr>
1561 <tr><td>Call</td><td>43</td><td>1</td><td>1.0</td></tr>
1562 <tr><td>Select</td><td>44</td><td>2</td><td>1.2</td></tr>
1563 <tr><td>UserOp1</td><td>45</td><td>1</td><td>1.0</td></tr>
1564 <tr><td>UserOp2</td><td>46</td><td>1</td><td>1.0</td></tr>
1565 <tr><td>VAArg</td><td>47</td><td>5</td><td>1.5</td></tr>
1566 <tr><td>ExtractElement</td><td>48</td><td>5</td><td>1.5</td></tr>
1567 <tr><td>InsertElement</td><td>49</td><td>5</td><td>1.5</td></tr>
1568 <tr><td>ShuffleElement</td><td>50</td><td>5</td><td>1.5</td></tr>
1569 <tr><td colspan="4">
1570 <b>Pseudo Instructions<a href="#pi_note">*</a></b>
1572 <tr><td>Invoke+CC </td><td>56</td><td>5</td><td>1.5</td></tr>
1573 <tr><td>Invoke+FastCC</td><td>57</td><td>5</td><td>1.5</td></tr>
1574 <tr><td>Call+CC</td><td>58</td><td>5</td><td>1.5</td></tr>
1575 <tr><td>Call+FastCC+TailCall</td><td>59</td><td>5</td><td>1.5</td></tr>
1576 <tr><td>Call+FastCC</td><td>60</td><td>5</td><td>1.5</td></tr>
1577 <tr><td>Call+CCC+TailCall</td><td>61</td><td>5</td><td>1.5</td></tr>
1578 <tr><td>Load+Volatile</td><td>62</td><td>3</td><td>1.3</td></tr>
1579 <tr><td>Store+Volatile</td><td>63</td><td>3</td><td>1.3</td></tr>
1583 <p><b><a name="pi_note">* Note: </a></b>
1584 These aren't really opcodes from an LLVM language perspective. They encode
1585 information into other opcodes without reserving space for that information.
1586 For example, opcode=63 is a Volatile Store. The opcode for this
1587 instruction is 25 (Store) but we encode it as 63 to indicate that is a Volatile
1588 Store. The same is done for the calling conventions and tail calls.
1589 In each of these entries in range 56-63, the opcode is documented as the base
1590 opcode (Invoke, Call, Store) plus some set of modifiers, as follows:</p>
1593 <dd>This means an arbitrary calling convention is specified
1594 in a VBR that follows the opcode. This is used when the instruction cannot
1595 be encoded with one of the more compact forms.
1598 <dd>This indicates that the Call or Invoke is using the FastCC calling
1601 <dd>This indicates that the Call or Invoke is using the native "C" calling
1604 <dd>This indicates that the Call has the 'tail' modifier.</dd>
1608 <!-- _______________________________________________________________________ -->
1609 <div class="doc_subsubsection"><a name="instoperands">Instruction
1612 <div class="doc_text">
1614 Based on the instruction opcode and type, the bytecode format implicitly (to
1615 save space) specifies the interpretation of the operand list. For most
1616 instructions, the type of each operand is implicit from the type of the
1617 instruction itself (e.g. the type of operands of a binary operator must match
1618 the type of the instruction). As such, the bytecode format generally only
1619 encodes the value number of the operand, not the type.</p>
1621 <p>In some cases, however, this is not sufficient. This section enumerates
1625 <li>getelementptr: the slot numbers for sequential type indexes are shifted
1626 up two bits. This allows the low order bits will encode the type of index
1627 used, as follows: 0=uint, 1=int, 2=ulong, 3=long.</li>
1628 <li>cast: the result type number is encoded as the second operand.</li>
1629 <li>alloca/malloc: If the allocation has an explicit alignment, the log2 of
1630 the alignment is encoded as the second operand.</li>
1631 <li>call: If the tail marker and calling convention cannot be
1632 <a href="#pi_note">encoded into the opcode</a> of the call, it is passed as
1633 an additional operand. The low bit of the operand is a flag indicating
1634 whether the call is a tail call. The rest of the bits contain the calling
1635 convention number (shifted left by one bit).</li>
1639 <!-- _______________________________________________________________________ -->
1640 <div class="doc_subsubsection"><a name="instencode">Instruction
1643 <div class="doc_text">
1644 <p>For brevity, instructions are written in one of four formats,
1645 depending on the number of operands to the instruction. Each
1646 instruction begins with a <a href="#uint32_vbr">uint32_vbr</a> that
1647 encodes the type of the instruction as well as other things. The tables
1648 that follow describe the format of this first part of each instruction.</p>
1649 <p><b>Instruction Format 0</b></p>
1650 <p>This format is used for a few instructions that can't easily be
1651 shortened because they have large numbers of operands (e.g. PHI Node or
1652 getelementptr). Each of the opcode, type, and operand fields is found in
1653 successive fields.</p>
1657 <th><b>Type</b></th>
1658 <th class="td_left"><b>Field Description</b></th>
1661 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1662 <td class="td_left">Specifies the opcode of the instruction. Note that
1663 for compatibility with the other instruction formats, the opcode is
1664 shifted left by 2 bits. Bits 0 and 1 must have value zero for this
1668 <td><a href="#uint24_vbr">uint24_vbr</a></td>
1669 <td class="td_left">Provides the type slot number of the result type of
1670 the instruction.</td>
1673 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1674 <td class="td_left">The number of operands that follow.</td>
1677 <td><a href="#uint32_vbr">uint32_vbr</a>+</td>
1678 <td class="td_left">The slot number of the value(s) for the operand(s).
1684 <p><b>Instruction Format 1</b></p>
1685 <p>This format encodes the opcode, type and a single operand into a
1686 single <a href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1690 <th><b>Bits</b></th>
1691 <th><b>Type</b></th>
1692 <th class="td_left"><b>Field Description</b></th>
1696 <td>constant "1"</td>
1697 <td class="td_left">These two bits must be the value 1 which identifies
1698 this as an instruction of format 1.</td>
1702 <td><a href="#instructions">opcode</a></td>
1703 <td class="td_left">Specifies the opcode of the instruction. Note that
1704 the maximum opcode value is 63.</td>
1708 <td><a href="#unsigned">unsigned</a></td>
1709 <td class="td_left">Specifies the slot number of the type for this
1710 instruction. Maximum slot number is 2<sup>12</sup>-1=4095.</td>
1714 <td><a href="#unsigned">unsigned</a></td>
1715 <td class="td_left">Specifies the slot number of the value for the
1716 first operand. Maximum slot number is 2<sup>12</sup>-1=4095. Note that
1717 the value 2<sup>12</sup>-1 denotes zero operands.</td>
1721 <p><b>Instruction Format 2</b></p>
1722 <p>This format encodes the opcode, type and two operands into a single <a
1723 href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1727 <th><b>Bits</b></th>
1728 <th><b>Type</b></th>
1729 <th class="td_left"><b>Field Description</b></th>
1733 <td>constant "2"</td>
1734 <td class="td_left">These two bits must be the value 2 which identifies
1735 this as an instruction of format 2.</td>
1739 <td><a href="#instructions">opcode</a></td>
1740 <td class="td_left">Specifies the opcode of the instruction. Note that
1741 the maximum opcode value is 63.</td>
1745 <td><a href="#unsigned">unsigned</a></td>
1746 <td class="td_left">Specifies the slot number of the type for this
1747 instruction. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1751 <td><a href="#unsigned">unsigned</a></td>
1752 <td class="td_left">Specifies the slot number of the value for the first
1753 operand. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1757 <td><a href="#unsigned">unsigned</a></td>
1758 <td class="td_left">Specifies the slot number of the value for the second
1759 operand. Maximum slot number is 2<sup>8</sup>-1=255.</td>
1763 <p><b>Instruction Format 3</b></p>
1764 <p>This format encodes the opcode, type and three operands into a
1765 single <a href="#uint32_vbr">uint32_vbr</a> as follows:</p>
1769 <th><b>Bits</b></th>
1770 <th><b>Type</b></th>
1771 <th class="td_left"><b>Field Description</b></th>
1775 <td>constant "3"</td>
1776 <td class="td_left">These two bits must be the value 3 which identifies
1777 this as an instruction of format 3.</td>
1781 <td><a href="#instructions">opcode</a></td>
1782 <td class="td_left">Specifies the opcode of the instruction. Note that
1783 the maximum opcode value is 63.</td>
1787 <td><a href="#unsigned">unsigned</a></td>
1788 <td class="td_left">Specifies the slot number of the type for this
1789 instruction. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1793 <td><a href="#unsigned">unsigned</a></td>
1794 <td class="td_left">Specifies the slot number of the value for the first
1795 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1799 <td><a href="#unsigned">unsigned</a></td>
1800 <td class="td_left">Specifies the slot number of the value for the second
1801 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1805 <td><a href="#unsigned">unsigned</a></td>
1806 <td class="td_left">Specifies the slot number of the value for the third
1807 operand. Maximum slot number is 2<sup>6</sup>-1=63.</td>
1813 <!-- _______________________________________________________________________ -->
1814 <div class="doc_subsection"><a name="symtab">Symbol Table</a> </div>
1815 <div class="doc_text">
1816 <p>A symbol table can be put out in conjunction with a module or a function. A
1817 symbol table has a list of name/type associations followed by a list of
1818 name/value associations. The name/value associations are organized into "type
1819 planes" so that all values of a common type are listed together. Each type
1820 plane starts with the number of entries in the plane and the type slot number
1821 for all the values in that plane (so the type can be looked up in the global
1822 type pool). For each entry in a type plane, the slot number of the value and
1823 the name associated with that value are written. The format is given in the
1828 <th><b>Type</b></th>
1829 <th class="td_left"><b>Field Description</b></th>
1832 <td><a href="#block">block</a><br>
1834 <td class="td_left">Symbol Table Identifier (0x04)</td>
1837 <td><a href="#llist">llist</a>(<a href="#symtab_entry">type_entry</a>)
1839 <td class="td_left">A length list of symbol table entries for
1844 <td><a href="#zlist">llist</a>(<a href="#symtab_plane">symtab_plane</a>)
1846 <td class="td_left">A length list of "type planes" of symbol table
1847 entries for <tt>Value</tt>s</td>
1853 <!-- _______________________________________________________________________ -->
1854 <div class="doc_subsubsection"> <a name="type_entry">Symbol Table Type
1857 <div class="doc_text">
1858 <p>A symbol table type entry associates a name with a type. The name is provided
1859 simply as an array of chars. The type is provided as a type slot number (index)
1860 into the global type pool. The format is given in the following table:</p>
1864 <th><b>Type</b></th>
1865 <th class="td_left"><b>Field Description</b></th>
1868 <td><a href="#uint32_vbr">uint24_vbr</a></td>
1869 <td class="td_left">Type slot number of the type being given a
1870 name relative to the global type pool.
1874 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1875 <td class="td_left">Length of the character array that follows.</td>
1878 <td><a href="#char">char</a>+</td>
1879 <td class="td_left">The characters of the name.</td>
1884 <!-- _______________________________________________________________________ -->
1885 <div class="doc_subsubsection"> <a name="symtab_plane">Symbol Table
1888 <div class="doc_text">
1889 <p>A symbol table plane provides the symbol table entries for all
1890 values of a common type. The encoding is given in the following table:</p>
1894 <th><b>Type</b></th>
1895 <th class="td_left"><b>Field Description</b></th>
1898 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1899 <td class="td_left">Number of entries in this plane.</td>
1902 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1903 <td class="td_left">Type slot number of type for all values in this plane.
1907 <td><a href="#value_entry">value_entry</a>+</td>
1908 <td class="td_left">The symbol table entries for to associate values with
1914 <!-- _______________________________________________________________________ -->
1915 <div class="doc_subsubsection"><a name="value_entry">Symbol Table Value
1918 <div class="doc_text">
1919 <p>A symbol table value entry provides the assocation between a value and the
1920 name given to the value. The value is referenced by its slot number. The
1921 format is given in the following table:</p>
1925 <th><b>Type</b></th>
1926 <th class="td_left"><b>Field Description</b></th>
1929 <td><a href="#uint32_vbr">uint24_vbr</a></td>
1930 <td class="td_left">Value slot number of the value being given a name.
1934 <td><a href="#uint32_vbr">uint32_vbr</a></td>
1935 <td class="td_left">Length of the character array that follows.</td>
1938 <td><a href="#char">char</a>+</td>
1939 <td class="td_left">The characters of the name.</td>
1945 <!-- *********************************************************************** -->
1946 <div class="doc_section"> <a name="versiondiffs">Version Differences</a>
1948 <!-- *********************************************************************** -->
1949 <div class="doc_text">
1950 <p>This section describes the differences in the Bytecode Format across
1952 versions. The versions are listed in reverse order because it assumes
1953 the current version is as documented in the previous sections. Each
1955 describes the differences between that version and the one that <i>follows</i>.
1959 <!-- _______________________________________________________________________ -->
1960 <div class="doc_subsection"><a name="vers13">Version 1.3 Differences From
1962 <!-- _______________________________________________________________________ -->
1964 <div class="doc_subsubsection">Unreachable Instruction</div>
1965 <div class="doc_text">
1966 <p>The LLVM <a href="LangRef.html#i_unreachable">Unreachable</a> instruction
1967 was added in version 1.4 of LLVM. This caused all instruction numbers after
1968 it to shift down by one.</p>
1971 <div class="doc_subsubsection">Function Flags</div>
1972 <div class="doc_text">
1973 <p>LLVM bytecode versions prior to 1.4 did not include the 5 bit offset
1974 in <a href="#funcfield">the function list</a> in the <a
1975 href="#globalinfo">Module Global Info</a> block.</p>
1978 <div class="doc_subsubsection">Function Flags</div>
1979 <div class="doc_text">
1980 <p>LLVM bytecode versions prior to 1.4 did not include the 'undef' constant
1981 value, which affects the encoding of <a href="#constant">Constant Fields</a>.
1986 <!-- _______________________________________________________________________ -->
1987 <div class="doc_subsection"><a name="vers12">Version 1.2 Differences
1989 <!-- _______________________________________________________________________ -->
1991 <div class="doc_subsubsection">Type Derives From Value</div>
1992 <div class="doc_text">
1993 <p>In version 1.2, the Type class in the LLVM IR derives from the Value
1994 class. This is not the case in version 1.3. Consequently, in version
1995 1.2 the notion of a "Type Type" was used to write out values that were
1996 Types. The types always occuped plane 12 (corresponding to the
1997 TypeTyID) of any type planed set of values. In 1.3 this representation
1998 is not convenient because the TypeTyID (12) is not present and its
1999 value is now used for LabelTyID. Consequently, the data structures
2000 written that involve types do so by writing all the types first and
2001 then each of the value planes according to those types. In version 1.2,
2002 the types would have been written intermingled with the values.</p>
2004 <!-- _______________________________________________________________________ -->
2005 <div class="doc_subsubsection">Restricted getelementptr Types</div>
2006 <div class="doc_text">
2007 <p>In version 1.2, the getelementptr instruction required a ubyte type
2008 index for accessing a structure field and a long type index for
2009 accessing an array element. Consequently, it was only possible to
2010 access structures of 255 or fewer elements. Starting in version 1.3,
2011 this restriction was lifted. Structures must now be indexed with uint
2012 constants. Arrays may now be indexed with int, uint, long, or ulong
2013 typed values. The consequence of this was that the bytecode format had
2014 to change in order to accommodate the larger range of structure indices.</p>
2016 <!-- _______________________________________________________________________ -->
2017 <div class="doc_subsubsection">Short Block Headers</div>
2018 <div class="doc_text">
2019 <p>In version 1.2, block headers were always 8 bytes being comprised of
2020 both an unsigned integer type and an unsigned integer size. For very
2021 small modules, these block headers turn out to be a large fraction of
2022 the total bytecode file size. In an attempt to make these small files
2023 smaller, the type and size information was encoded into a single
2024 unsigned integer (4 bytes) comprised of 5 bits for the block type
2025 (maximum 31 block types) and 27 bits for the block size (max
2026 ~134MBytes). These limits seemed sufficient for any blocks or sizes
2027 forseen in the future. Note that the module block, which encloses all
2028 the other blocks is still written as 8 bytes since bytecode files
2029 larger than 134MBytes might be possible.</p>
2031 <!-- _______________________________________________________________________ -->
2032 <div class="doc_subsubsection">Dependent Libraries and Target Triples</div>
2033 <div class="doc_text">
2034 <p>In version 1.2, the bytecode format does not store module's target
2035 triple or dependent. These fields have been added to the end of the <a
2036 href="#globalinfo">module global info block</a>. The purpose of these
2037 fields is to allow a front end compiler to specifiy that the generated
2038 module is specific to a particular target triple (operating
2039 system/manufacturer/processor) which makes it non-portable; and to
2040 allow front end compilers to specify the list of libraries that the
2041 module depends on for successful linking.</p>
2043 <!-- _______________________________________________________________________ -->
2044 <div class="doc_subsubsection">Types Restricted to 24-bits</div>
2045 <div class="doc_text">
2046 <p>In version 1.2, type slot identifiers were written as 32-bit VBR
2047 quantities. In 1.3 this has been reduced to 24-bits in order to ensure
2048 that it is not possible to overflow the type field of a global variable
2049 definition. 24-bits for type slot numbers is deemed sufficient for any
2050 practical use of LLVM.</p>
2052 <!-- _______________________________________________________________________ -->
2053 <!-- _______________________________________________________________________ -->
2054 <div class="doc_subsection"><a name="vers11">Version 1.1 Differences
2056 <!-- _______________________________________________________________________ -->
2057 <div class="doc_subsubsection">Explicit Primitive Zeros</div>
2058 <div class="doc_text">
2059 <p>In version 1.1, the zero value for primitives was explicitly encoded
2060 into the bytecode format. Since these zero values are constant values
2061 in the LLVM IR and never change, there is no reason to explicitly
2062 encode them. This explicit encoding was removed in version 1.2.</p>
2064 <!-- _______________________________________________________________________ -->
2065 <div class="doc_subsubsection">Inconsistent Module Global Info</div>
2066 <div class="doc_text">
2067 <p>In version 1.1, the Module Global Info block was not aligned causing
2068 the next block to be read in on an unaligned boundary. This problem was
2069 corrected in version 1.2.<br>
2073 <!-- _______________________________________________________________________ -->
2074 <div class="doc_subsection"><a name="vers10">Version 1.0 Differences
2076 <div class="doc_text">
2077 <p>None. Version 1.0 and 1.1 bytecode formats are identical.</p>
2079 <!-- *********************************************************************** -->
2081 <address> <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2082 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
2083 <a href="http://validator.w3.org/check/referer"><img
2084 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
2085 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and <a
2086 href="mailto:sabre@nondot.org">Chris Lattner</a><br>
2087 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
2088 Last modified: $Date$