1.. _bitcode_format: 2 3.. role:: raw-html(raw) 4 :format: html 5 6======================== 7LLVM Bitcode File Format 8======================== 9 10.. contents:: 11 :local: 12 13Abstract 14======== 15 16This document describes the LLVM bitstream file format and the encoding of the 17LLVM IR into it. 18 19Overview 20======== 21 22What is commonly known as the LLVM bitcode file format (also, sometimes 23anachronistically known as bytecode) is actually two things: a `bitstream 24container format`_ and an `encoding of LLVM IR`_ into the container format. 25 26The bitstream format is an abstract encoding of structured data, very similar to 27XML in some ways. Like XML, bitstream files contain tags, and nested 28structures, and you can parse the file without having to understand the tags. 29Unlike XML, the bitstream format is a binary encoding, and unlike XML it 30provides a mechanism for the file to self-describe "abbreviations", which are 31effectively size optimizations for the content. 32 33LLVM IR files may be optionally embedded into a `wrapper`_ structure that makes 34it easy to embed extra data along with LLVM IR files. 35 36This document first describes the LLVM bitstream format, describes the wrapper 37format, then describes the record structure used by LLVM IR files. 38 39.. _bitstream container format: 40 41Bitstream Format 42================ 43 44The bitstream format is literally a stream of bits, with a very simple 45structure. This structure consists of the following concepts: 46 47* A "`magic number`_" that identifies the contents of the stream. 48 49* Encoding `primitives`_ like variable bit-rate integers. 50 51* `Blocks`_, which define nested content. 52 53* `Data Records`_, which describe entities within the file. 54 55* Abbreviations, which specify compression optimizations for the file. 56 57Note that the `llvm-bcanalyzer <CommandGuide/html/llvm-bcanalyzer.html>`_ tool 58can be used to dump and inspect arbitrary bitstreams, which is very useful for 59understanding the encoding. 60 61.. _magic number: 62 63Magic Numbers 64------------- 65 66The first two bytes of a bitcode file are 'BC' (``0x42``, ``0x43``). The second 67two bytes are an application-specific magic number. Generic bitcode tools can 68look at only the first two bytes to verify the file is bitcode, while 69application-specific programs will want to look at all four. 70 71.. _primitives: 72 73Primitives 74---------- 75 76A bitstream literally consists of a stream of bits, which are read in order 77starting with the least significant bit of each byte. The stream is made up of 78a number of primitive values that encode a stream of unsigned integer values. 79These integers are encoded in two ways: either as `Fixed Width Integers`_ or as 80`Variable Width Integers`_. 81 82.. _Fixed Width Integers: 83.. _fixed-width value: 84 85Fixed Width Integers 86^^^^^^^^^^^^^^^^^^^^ 87 88Fixed-width integer values have their low bits emitted directly to the file. 89For example, a 3-bit integer value encodes 1 as 001. Fixed width integers are 90used when there are a well-known number of options for a field. For example, 91boolean values are usually encoded with a 1-bit wide integer. 92 93.. _Variable Width Integers: 94.. _Variable Width Integer: 95.. _variable-width value: 96 97Variable Width Integers 98^^^^^^^^^^^^^^^^^^^^^^^ 99 100Variable-width integer (VBR) values encode values of arbitrary size, optimizing 101for the case where the values are small. Given a 4-bit VBR field, any 3-bit 102value (0 through 7) is encoded directly, with the high bit set to zero. Values 103larger than N-1 bits emit their bits in a series of N-1 bit chunks, where all 104but the last set the high bit. 105 106For example, the value 27 (0x1B) is encoded as 1011 0011 when emitted as a vbr4 107value. The first set of four bits indicates the value 3 (011) with a 108continuation piece (indicated by a high bit of 1). The next word indicates a 109value of 24 (011 << 3) with no continuation. The sum (3+24) yields the value 11027. 111 112.. _char6-encoded value: 113 1146-bit characters 115^^^^^^^^^^^^^^^^ 116 1176-bit characters encode common characters into a fixed 6-bit field. They 118represent the following characters with the following 6-bit values: 119 120:: 121 122 'a' .. 'z' --- 0 .. 25 123 'A' .. 'Z' --- 26 .. 51 124 '0' .. '9' --- 52 .. 61 125 '.' --- 62 126 '_' --- 63 127 128This encoding is only suitable for encoding characters and strings that consist 129only of the above characters. It is completely incapable of encoding characters 130not in the set. 131 132Word Alignment 133^^^^^^^^^^^^^^ 134 135Occasionally, it is useful to emit zero bits until the bitstream is a multiple 136of 32 bits. This ensures that the bit position in the stream can be represented 137as a multiple of 32-bit words. 138 139Abbreviation IDs 140---------------- 141 142A bitstream is a sequential series of `Blocks`_ and `Data Records`_. Both of 143these start with an abbreviation ID encoded as a fixed-bitwidth field. The 144width is specified by the current block, as described below. The value of the 145abbreviation ID specifies either a builtin ID (which have special meanings, 146defined below) or one of the abbreviation IDs defined for the current block by 147the stream itself. 148 149The set of builtin abbrev IDs is: 150 151* 0 - `END_BLOCK`_ --- This abbrev ID marks the end of the current block. 152 153* 1 - `ENTER_SUBBLOCK`_ --- This abbrev ID marks the beginning of a new 154 block. 155 156* 2 - `DEFINE_ABBREV`_ --- This defines a new abbreviation. 157 158* 3 - `UNABBREV_RECORD`_ --- This ID specifies the definition of an 159 unabbreviated record. 160 161Abbreviation IDs 4 and above are defined by the stream itself, and specify an 162`abbreviated record encoding`_. 163 164.. _Blocks: 165 166Blocks 167------ 168 169Blocks in a bitstream denote nested regions of the stream, and are identified by 170a content-specific id number (for example, LLVM IR uses an ID of 12 to represent 171function bodies). Block IDs 0-7 are reserved for `standard blocks`_ whose 172meaning is defined by Bitcode; block IDs 8 and greater are application 173specific. Nested blocks capture the hierarchical structure of the data encoded 174in it, and various properties are associated with blocks as the file is parsed. 175Block definitions allow the reader to efficiently skip blocks in constant time 176if the reader wants a summary of blocks, or if it wants to efficiently skip data 177it does not understand. The LLVM IR reader uses this mechanism to skip function 178bodies, lazily reading them on demand. 179 180When reading and encoding the stream, several properties are maintained for the 181block. In particular, each block maintains: 182 183#. A current abbrev id width. This value starts at 2 at the beginning of the 184 stream, and is set every time a block record is entered. The block entry 185 specifies the abbrev id width for the body of the block. 186 187#. A set of abbreviations. Abbreviations may be defined within a block, in 188 which case they are only defined in that block (neither subblocks nor 189 enclosing blocks see the abbreviation). Abbreviations can also be defined 190 inside a `BLOCKINFO`_ block, in which case they are defined in all blocks 191 that match the ID that the ``BLOCKINFO`` block is describing. 192 193As sub blocks are entered, these properties are saved and the new sub-block has 194its own set of abbreviations, and its own abbrev id width. When a sub-block is 195popped, the saved values are restored. 196 197.. _ENTER_SUBBLOCK: 198 199ENTER_SUBBLOCK Encoding 200^^^^^^^^^^^^^^^^^^^^^^^ 201 202:raw-html:`<tt>` 203[ENTER_SUBBLOCK, blockid\ :sub:`vbr8`, newabbrevlen\ :sub:`vbr4`, <align32bits>, blocklen_32] 204:raw-html:`</tt>` 205 206The ``ENTER_SUBBLOCK`` abbreviation ID specifies the start of a new block 207record. The ``blockid`` value is encoded as an 8-bit VBR identifier, and 208indicates the type of block being entered, which can be a `standard block`_ or 209an application-specific block. The ``newabbrevlen`` value is a 4-bit VBR, which 210specifies the abbrev id width for the sub-block. The ``blocklen`` value is a 21132-bit aligned value that specifies the size of the subblock in 32-bit 212words. This value allows the reader to skip over the entire block in one jump. 213 214.. _END_BLOCK: 215 216END_BLOCK Encoding 217^^^^^^^^^^^^^^^^^^ 218 219``[END_BLOCK, <align32bits>]`` 220 221The ``END_BLOCK`` abbreviation ID specifies the end of the current block record. 222Its end is aligned to 32-bits to ensure that the size of the block is an even 223multiple of 32-bits. 224 225.. _Data Records: 226 227Data Records 228------------ 229 230Data records consist of a record code and a number of (up to) 64-bit integer 231values. The interpretation of the code and values is application specific and 232may vary between different block types. Records can be encoded either using an 233unabbrev record, or with an abbreviation. In the LLVM IR format, for example, 234there is a record which encodes the target triple of a module. The code is 235``MODULE_CODE_TRIPLE``, and the values of the record are the ASCII codes for the 236characters in the string. 237 238.. _UNABBREV_RECORD: 239 240UNABBREV_RECORD Encoding 241^^^^^^^^^^^^^^^^^^^^^^^^ 242 243:raw-html:`<tt>` 244[UNABBREV_RECORD, code\ :sub:`vbr6`, numops\ :sub:`vbr6`, op0\ :sub:`vbr6`, op1\ :sub:`vbr6`, ...] 245:raw-html:`</tt>` 246 247An ``UNABBREV_RECORD`` provides a default fallback encoding, which is both 248completely general and extremely inefficient. It can describe an arbitrary 249record by emitting the code and operands as VBRs. 250 251For example, emitting an LLVM IR target triple as an unabbreviated record 252requires emitting the ``UNABBREV_RECORD`` abbrevid, a vbr6 for the 253``MODULE_CODE_TRIPLE`` code, a vbr6 for the length of the string, which is equal 254to the number of operands, and a vbr6 for each character. Because there are no 255letters with values less than 32, each letter would need to be emitted as at 256least a two-part VBR, which means that each letter would require at least 12 257bits. This is not an efficient encoding, but it is fully general. 258 259.. _abbreviated record encoding: 260 261Abbreviated Record Encoding 262^^^^^^^^^^^^^^^^^^^^^^^^^^^ 263 264``[<abbrevid>, fields...]`` 265 266An abbreviated record is a abbreviation id followed by a set of fields that are 267encoded according to the `abbreviation definition`_. This allows records to be 268encoded significantly more densely than records encoded with the 269`UNABBREV_RECORD`_ type, and allows the abbreviation types to be specified in 270the stream itself, which allows the files to be completely self describing. The 271actual encoding of abbreviations is defined below. 272 273The record code, which is the first field of an abbreviated record, may be 274encoded in the abbreviation definition (as a literal operand) or supplied in the 275abbreviated record (as a Fixed or VBR operand value). 276 277.. _abbreviation definition: 278 279Abbreviations 280------------- 281 282Abbreviations are an important form of compression for bitstreams. The idea is 283to specify a dense encoding for a class of records once, then use that encoding 284to emit many records. It takes space to emit the encoding into the file, but 285the space is recouped (hopefully plus some) when the records that use it are 286emitted. 287 288Abbreviations can be determined dynamically per client, per file. Because the 289abbreviations are stored in the bitstream itself, different streams of the same 290format can contain different sets of abbreviations according to the needs of the 291specific stream. As a concrete example, LLVM IR files usually emit an 292abbreviation for binary operators. If a specific LLVM module contained no or 293few binary operators, the abbreviation does not need to be emitted. 294 295.. _DEFINE_ABBREV: 296 297DEFINE_ABBREV Encoding 298^^^^^^^^^^^^^^^^^^^^^^ 299 300:raw-html:`<tt>` 301[DEFINE_ABBREV, numabbrevops\ :sub:`vbr5`, abbrevop0, abbrevop1, ...] 302:raw-html:`</tt>` 303 304A ``DEFINE_ABBREV`` record adds an abbreviation to the list of currently defined 305abbreviations in the scope of this block. This definition only exists inside 306this immediate block --- it is not visible in subblocks or enclosing blocks. 307Abbreviations are implicitly assigned IDs sequentially starting from 4 (the 308first application-defined abbreviation ID). Any abbreviations defined in a 309``BLOCKINFO`` record for the particular block type receive IDs first, in order, 310followed by any abbreviations defined within the block itself. Abbreviated data 311records reference this ID to indicate what abbreviation they are invoking. 312 313An abbreviation definition consists of the ``DEFINE_ABBREV`` abbrevid followed 314by a VBR that specifies the number of abbrev operands, then the abbrev operands 315themselves. Abbreviation operands come in three forms. They all start with a 316single bit that indicates whether the abbrev operand is a literal operand (when 317the bit is 1) or an encoding operand (when the bit is 0). 318 319#. Literal operands --- :raw-html:`<tt>` [1\ :sub:`1`, litvalue\ 320 :sub:`vbr8`] :raw-html:`</tt>` --- Literal operands specify that the value in 321 the result is always a single specific value. This specific value is emitted 322 as a vbr8 after the bit indicating that it is a literal operand. 323 324#. Encoding info without data --- :raw-html:`<tt>` [0\ :sub:`1`, encoding\ 325 :sub:`3`] :raw-html:`</tt>` --- Operand encodings that do not have extra data 326 are just emitted as their code. 327 328#. Encoding info with data --- :raw-html:`<tt>` [0\ :sub:`1`, encoding\ 329 :sub:`3`, value\ :sub:`vbr5`] :raw-html:`</tt>` --- Operand encodings that do 330 have extra data are emitted as their code, followed by the extra data. 331 332The possible operand encodings are: 333 334* Fixed (code 1): The field should be emitted as a `fixed-width value`_, whose 335 width is specified by the operand's extra data. 336 337* VBR (code 2): The field should be emitted as a `variable-width value`_, whose 338 width is specified by the operand's extra data. 339 340* Array (code 3): This field is an array of values. The array operand has no 341 extra data, but expects another operand to follow it, indicating the element 342 type of the array. When reading an array in an abbreviated record, the first 343 integer is a vbr6 that indicates the array length, followed by the encoded 344 elements of the array. An array may only occur as the last operand of an 345 abbreviation (except for the one final operand that gives the array's 346 type). 347 348* Char6 (code 4): This field should be emitted as a `char6-encoded value`_. 349 This operand type takes no extra data. Char6 encoding is normally used as an 350 array element type. 351 352* Blob (code 5): This field is emitted as a vbr6, followed by padding to a 353 32-bit boundary (for alignment) and an array of 8-bit objects. The array of 354 bytes is further followed by tail padding to ensure that its total length is a 355 multiple of 4 bytes. This makes it very efficient for the reader to decode 356 the data without having to make a copy of it: it can use a pointer to the data 357 in the mapped in file and poke directly at it. A blob may only occur as the 358 last operand of an abbreviation. 359 360For example, target triples in LLVM modules are encoded as a record of the form 361``[TRIPLE, 'a', 'b', 'c', 'd']``. Consider if the bitstream emitted the 362following abbrev entry: 363 364:: 365 366 [0, Fixed, 4] 367 [0, Array] 368 [0, Char6] 369 370When emitting a record with this abbreviation, the above entry would be emitted 371as: 372 373:raw-html:`<tt><blockquote>` 374[4\ :sub:`abbrevwidth`, 2\ :sub:`4`, 4\ :sub:`vbr6`, 0\ :sub:`6`, 1\ :sub:`6`, 2\ :sub:`6`, 3\ :sub:`6`] 375:raw-html:`</blockquote></tt>` 376 377These values are: 378 379#. The first value, 4, is the abbreviation ID for this abbreviation. 380 381#. The second value, 2, is the record code for ``TRIPLE`` records within LLVM IR 382 file ``MODULE_BLOCK`` blocks. 383 384#. The third value, 4, is the length of the array. 385 386#. The rest of the values are the char6 encoded values for ``"abcd"``. 387 388With this abbreviation, the triple is emitted with only 37 bits (assuming a 389abbrev id width of 3). Without the abbreviation, significantly more space would 390be required to emit the target triple. Also, because the ``TRIPLE`` value is 391not emitted as a literal in the abbreviation, the abbreviation can also be used 392for any other string value. 393 394.. _standard blocks: 395.. _standard block: 396 397Standard Blocks 398--------------- 399 400In addition to the basic block structure and record encodings, the bitstream 401also defines specific built-in block types. These block types specify how the 402stream is to be decoded or other metadata. In the future, new standard blocks 403may be added. Block IDs 0-7 are reserved for standard blocks. 404 405.. _BLOCKINFO: 406 407#0 - BLOCKINFO Block 408^^^^^^^^^^^^^^^^^^^^ 409 410The ``BLOCKINFO`` block allows the description of metadata for other blocks. 411The currently specified records are: 412 413:: 414 415 [SETBID (#1), blockid] 416 [DEFINE_ABBREV, ...] 417 [BLOCKNAME, ...name...] 418 [SETRECORDNAME, RecordID, ...name...] 419 420The ``SETBID`` record (code 1) indicates which block ID is being described. 421``SETBID`` records can occur multiple times throughout the block to change which 422block ID is being described. There must be a ``SETBID`` record prior to any 423other records. 424 425Standard ``DEFINE_ABBREV`` records can occur inside ``BLOCKINFO`` blocks, but 426unlike their occurrence in normal blocks, the abbreviation is defined for blocks 427matching the block ID we are describing, *not* the ``BLOCKINFO`` block 428itself. The abbreviations defined in ``BLOCKINFO`` blocks receive abbreviation 429IDs as described in `DEFINE_ABBREV`_. 430 431The ``BLOCKNAME`` record (code 2) can optionally occur in this block. The 432elements of the record are the bytes of the string name of the block. 433llvm-bcanalyzer can use this to dump out bitcode files symbolically. 434 435The ``SETRECORDNAME`` record (code 3) can also optionally occur in this block. 436The first operand value is a record ID number, and the rest of the elements of 437the record are the bytes for the string name of the record. llvm-bcanalyzer can 438use this to dump out bitcode files symbolically. 439 440Note that although the data in ``BLOCKINFO`` blocks is described as "metadata," 441the abbreviations they contain are essential for parsing records from the 442corresponding blocks. It is not safe to skip them. 443 444.. _wrapper: 445 446Bitcode Wrapper Format 447====================== 448 449Bitcode files for LLVM IR may optionally be wrapped in a simple wrapper 450structure. This structure contains a simple header that indicates the offset 451and size of the embedded BC file. This allows additional information to be 452stored alongside the BC file. The structure of this file header is: 453 454:raw-html:`<tt><blockquote>` 455[Magic\ :sub:`32`, Version\ :sub:`32`, Offset\ :sub:`32`, Size\ :sub:`32`, CPUType\ :sub:`32`] 456:raw-html:`</blockquote></tt>` 457 458Each of the fields are 32-bit fields stored in little endian form (as with the 459rest of the bitcode file fields). The Magic number is always ``0x0B17C0DE`` and 460the version is currently always ``0``. The Offset field is the offset in bytes 461to the start of the bitcode stream in the file, and the Size field is the size 462in bytes of the stream. CPUType is a target-specific value that can be used to 463encode the CPU of the target. 464 465.. _encoding of LLVM IR: 466 467LLVM IR Encoding 468================ 469 470LLVM IR is encoded into a bitstream by defining blocks and records. It uses 471blocks for things like constant pools, functions, symbol tables, etc. It uses 472records for things like instructions, global variable descriptors, type 473descriptions, etc. This document does not describe the set of abbreviations 474that the writer uses, as these are fully self-described in the file, and the 475reader is not allowed to build in any knowledge of this. 476 477Basics 478------ 479 480LLVM IR Magic Number 481^^^^^^^^^^^^^^^^^^^^ 482 483The magic number for LLVM IR files is: 484 485:raw-html:`<tt><blockquote>` 486[0x0\ :sub:`4`, 0xC\ :sub:`4`, 0xE\ :sub:`4`, 0xD\ :sub:`4`] 487:raw-html:`</blockquote></tt>` 488 489When combined with the bitcode magic number and viewed as bytes, this is 490``"BC 0xC0DE"``. 491 492.. _Signed VBRs: 493 494Signed VBRs 495^^^^^^^^^^^ 496 497`Variable Width Integer`_ encoding is an efficient way to encode arbitrary sized 498unsigned values, but is an extremely inefficient for encoding signed values, as 499signed values are otherwise treated as maximally large unsigned values. 500 501As such, signed VBR values of a specific width are emitted as follows: 502 503* Positive values are emitted as VBRs of the specified width, but with their 504 value shifted left by one. 505 506* Negative values are emitted as VBRs of the specified width, but the negated 507 value is shifted left by one, and the low bit is set. 508 509With this encoding, small positive and small negative values can both be emitted 510efficiently. Signed VBR encoding is used in ``CST_CODE_INTEGER`` and 511``CST_CODE_WIDE_INTEGER`` records within ``CONSTANTS_BLOCK`` blocks. 512It is also used for phi instruction operands in `MODULE_CODE_VERSION`_ 1. 513 514LLVM IR Blocks 515^^^^^^^^^^^^^^ 516 517LLVM IR is defined with the following blocks: 518 519* 8 --- `MODULE_BLOCK`_ --- This is the top-level block that contains the entire 520 module, and describes a variety of per-module information. 521 522* 9 --- `PARAMATTR_BLOCK`_ --- This enumerates the parameter attributes. 523 524* 10 --- `TYPE_BLOCK`_ --- This describes all of the types in the module. 525 526* 11 --- `CONSTANTS_BLOCK`_ --- This describes constants for a module or 527 function. 528 529* 12 --- `FUNCTION_BLOCK`_ --- This describes a function body. 530 531* 13 --- `TYPE_SYMTAB_BLOCK`_ --- This describes the type symbol table. 532 533* 14 --- `VALUE_SYMTAB_BLOCK`_ --- This describes a value symbol table. 534 535* 15 --- `METADATA_BLOCK`_ --- This describes metadata items. 536 537* 16 --- `METADATA_ATTACHMENT`_ --- This contains records associating metadata 538 with function instruction values. 539 540.. _MODULE_BLOCK: 541 542MODULE_BLOCK Contents 543--------------------- 544 545The ``MODULE_BLOCK`` block (id 8) is the top-level block for LLVM bitcode files, 546and each bitcode file must contain exactly one. In addition to records 547(described below) containing information about the module, a ``MODULE_BLOCK`` 548block may contain the following sub-blocks: 549 550* `BLOCKINFO`_ 551* `PARAMATTR_BLOCK`_ 552* `TYPE_BLOCK`_ 553* `TYPE_SYMTAB_BLOCK`_ 554* `VALUE_SYMTAB_BLOCK`_ 555* `CONSTANTS_BLOCK`_ 556* `FUNCTION_BLOCK`_ 557* `METADATA_BLOCK`_ 558 559.. _MODULE_CODE_VERSION: 560 561MODULE_CODE_VERSION Record 562^^^^^^^^^^^^^^^^^^^^^^^^^^ 563 564``[VERSION, version#]`` 565 566The ``VERSION`` record (code 1) contains a single value indicating the format 567version. Versions 0 and 1 are supported at this time. The difference between 568version 0 and 1 is in the encoding of instruction operands in 569each `FUNCTION_BLOCK`_. 570 571In version 0, each value defined by an instruction is assigned an ID 572unique to the function. Function-level value IDs are assigned starting from 573``NumModuleValues`` since they share the same namespace as module-level 574values. The value enumerator resets after each function. When a value is 575an operand of an instruction, the value ID is used to represent the operand. 576For large functions or large modules, these operand values can be large. 577 578The encoding in version 1 attempts to avoid large operand values 579in common cases. Instead of using the value ID directly, operands are 580encoded as relative to the current instruction. Thus, if an operand 581is the value defined by the previous instruction, the operand 582will be encoded as 1. 583 584For example, instead of 585 586.. code-block:: llvm 587 588 #n = load #n-1 589 #n+1 = icmp eq #n, #const0 590 br #n+1, label #(bb1), label #(bb2) 591 592version 1 will encode the instructions as 593 594.. code-block:: llvm 595 596 #n = load #1 597 #n+1 = icmp eq #1, (#n+1)-#const0 598 br #1, label #(bb1), label #(bb2) 599 600Note in the example that operands which are constants also use 601the relative encoding, while operands like basic block labels 602do not use the relative encoding. 603 604Forward references will result in a negative value. 605This can be inefficient, as operands are normally encoded 606as unsigned VBRs. However, forward references are rare, except in the 607case of phi instructions. For phi instructions, operands are encoded as 608`Signed VBRs`_ to deal with forward references. 609 610 611MODULE_CODE_TRIPLE Record 612^^^^^^^^^^^^^^^^^^^^^^^^^ 613 614``[TRIPLE, ...string...]`` 615 616The ``TRIPLE`` record (code 2) contains a variable number of values representing 617the bytes of the ``target triple`` specification string. 618 619MODULE_CODE_DATALAYOUT Record 620^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 621 622``[DATALAYOUT, ...string...]`` 623 624The ``DATALAYOUT`` record (code 3) contains a variable number of values 625representing the bytes of the ``target datalayout`` specification string. 626 627MODULE_CODE_ASM Record 628^^^^^^^^^^^^^^^^^^^^^^ 629 630``[ASM, ...string...]`` 631 632The ``ASM`` record (code 4) contains a variable number of values representing 633the bytes of ``module asm`` strings, with individual assembly blocks separated 634by newline (ASCII 10) characters. 635 636.. _MODULE_CODE_SECTIONNAME: 637 638MODULE_CODE_SECTIONNAME Record 639^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 640 641``[SECTIONNAME, ...string...]`` 642 643The ``SECTIONNAME`` record (code 5) contains a variable number of values 644representing the bytes of a single section name string. There should be one 645``SECTIONNAME`` record for each section name referenced (e.g., in global 646variable or function ``section`` attributes) within the module. These records 647can be referenced by the 1-based index in the *section* fields of ``GLOBALVAR`` 648or ``FUNCTION`` records. 649 650MODULE_CODE_DEPLIB Record 651^^^^^^^^^^^^^^^^^^^^^^^^^ 652 653``[DEPLIB, ...string...]`` 654 655The ``DEPLIB`` record (code 6) contains a variable number of values representing 656the bytes of a single dependent library name string, one of the libraries 657mentioned in a ``deplibs`` declaration. There should be one ``DEPLIB`` record 658for each library name referenced. 659 660MODULE_CODE_GLOBALVAR Record 661^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 662 663``[GLOBALVAR, pointer type, isconst, initid, linkage, alignment, section, visibility, threadlocal, unnamed_addr]`` 664 665The ``GLOBALVAR`` record (code 7) marks the declaration or definition of a 666global variable. The operand fields are: 667 668* *pointer type*: The type index of the pointer type used to point to this 669 global variable 670 671* *isconst*: Non-zero if the variable is treated as constant within the module, 672 or zero if it is not 673 674* *initid*: If non-zero, the value index of the initializer for this variable, 675 plus 1. 676 677.. _linkage type: 678 679* *linkage*: An encoding of the linkage type for this variable: 680 * ``external``: code 0 681 * ``weak``: code 1 682 * ``appending``: code 2 683 * ``internal``: code 3 684 * ``linkonce``: code 4 685 * ``dllimport``: code 5 686 * ``dllexport``: code 6 687 * ``extern_weak``: code 7 688 * ``common``: code 8 689 * ``private``: code 9 690 * ``weak_odr``: code 10 691 * ``linkonce_odr``: code 11 692 * ``available_externally``: code 12 693 * ``linker_private``: code 13 694 695* alignment*: The logarithm base 2 of the variable's requested alignment, plus 1 696 697* *section*: If non-zero, the 1-based section index in the table of 698 `MODULE_CODE_SECTIONNAME`_ entries. 699 700.. _visibility: 701 702* *visibility*: If present, an encoding of the visibility of this variable: 703 * ``default``: code 0 704 * ``hidden``: code 1 705 * ``protected``: code 2 706 707* *threadlocal*: If present, an encoding of the thread local storage mode of the 708 variable: 709 * ``not thread local``: code 0 710 * ``thread local; default TLS model``: code 1 711 * ``localdynamic``: code 2 712 * ``initialexec``: code 3 713 * ``localexec``: code 4 714 715* *unnamed_addr*: If present and non-zero, indicates that the variable has 716 ``unnamed_addr`` 717 718.. _FUNCTION: 719 720MODULE_CODE_FUNCTION Record 721^^^^^^^^^^^^^^^^^^^^^^^^^^^ 722 723``[FUNCTION, type, callingconv, isproto, linkage, paramattr, alignment, section, visibility, gc]`` 724 725The ``FUNCTION`` record (code 8) marks the declaration or definition of a 726function. The operand fields are: 727 728* *type*: The type index of the function type describing this function 729 730* *callingconv*: The calling convention number: 731 * ``ccc``: code 0 732 * ``fastcc``: code 8 733 * ``coldcc``: code 9 734 * ``x86_stdcallcc``: code 64 735 * ``x86_fastcallcc``: code 65 736 * ``arm_apcscc``: code 66 737 * ``arm_aapcscc``: code 67 738 * ``arm_aapcs_vfpcc``: code 68 739 740* isproto*: Non-zero if this entry represents a declaration rather than a 741 definition 742 743* *linkage*: An encoding of the `linkage type`_ for this function 744 745* *paramattr*: If nonzero, the 1-based parameter attribute index into the table 746 of `PARAMATTR_CODE_ENTRY`_ entries. 747 748* *alignment*: The logarithm base 2 of the function's requested alignment, plus 749 1 750 751* *section*: If non-zero, the 1-based section index in the table of 752 `MODULE_CODE_SECTIONNAME`_ entries. 753 754* *visibility*: An encoding of the `visibility`_ of this function 755 756* *gc*: If present and nonzero, the 1-based garbage collector index in the table 757 of `MODULE_CODE_GCNAME`_ entries. 758 759* *unnamed_addr*: If present and non-zero, indicates that the function has 760 ``unnamed_addr`` 761 762MODULE_CODE_ALIAS Record 763^^^^^^^^^^^^^^^^^^^^^^^^ 764 765``[ALIAS, alias type, aliasee val#, linkage, visibility]`` 766 767The ``ALIAS`` record (code 9) marks the definition of an alias. The operand 768fields are 769 770* *alias type*: The type index of the alias 771 772* *aliasee val#*: The value index of the aliased value 773 774* *linkage*: An encoding of the `linkage type`_ for this alias 775 776* *visibility*: If present, an encoding of the `visibility`_ of the alias 777 778MODULE_CODE_PURGEVALS Record 779^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 780 781``[PURGEVALS, numvals]`` 782 783The ``PURGEVALS`` record (code 10) resets the module-level value list to the 784size given by the single operand value. Module-level value list items are added 785by ``GLOBALVAR``, ``FUNCTION``, and ``ALIAS`` records. After a ``PURGEVALS`` 786record is seen, new value indices will start from the given *numvals* value. 787 788.. _MODULE_CODE_GCNAME: 789 790MODULE_CODE_GCNAME Record 791^^^^^^^^^^^^^^^^^^^^^^^^^ 792 793``[GCNAME, ...string...]`` 794 795The ``GCNAME`` record (code 11) contains a variable number of values 796representing the bytes of a single garbage collector name string. There should 797be one ``GCNAME`` record for each garbage collector name referenced in function 798``gc`` attributes within the module. These records can be referenced by 1-based 799index in the *gc* fields of ``FUNCTION`` records. 800 801.. _PARAMATTR_BLOCK: 802 803PARAMATTR_BLOCK Contents 804------------------------ 805 806The ``PARAMATTR_BLOCK`` block (id 9) contains a table of entries describing the 807attributes of function parameters. These entries are referenced by 1-based index 808in the *paramattr* field of module block `FUNCTION`_ records, or within the 809*attr* field of function block ``INST_INVOKE`` and ``INST_CALL`` records. 810 811Entries within ``PARAMATTR_BLOCK`` are constructed to ensure that each is unique 812(i.e., no two indicies represent equivalent attribute lists). 813 814.. _PARAMATTR_CODE_ENTRY: 815 816PARAMATTR_CODE_ENTRY Record 817^^^^^^^^^^^^^^^^^^^^^^^^^^^ 818 819``[ENTRY, paramidx0, attr0, paramidx1, attr1...]`` 820 821The ``ENTRY`` record (code 1) contains an even number of values describing a 822unique set of function parameter attributes. Each *paramidx* value indicates 823which set of attributes is represented, with 0 representing the return value 824attributes, 0xFFFFFFFF representing function attributes, and other values 825representing 1-based function parameters. Each *attr* value is a bitmap with the 826following interpretation: 827 828* bit 0: ``zeroext`` 829* bit 1: ``signext`` 830* bit 2: ``noreturn`` 831* bit 3: ``inreg`` 832* bit 4: ``sret`` 833* bit 5: ``nounwind`` 834* bit 6: ``noalias`` 835* bit 7: ``byval`` 836* bit 8: ``nest`` 837* bit 9: ``readnone`` 838* bit 10: ``readonly`` 839* bit 11: ``noinline`` 840* bit 12: ``alwaysinline`` 841* bit 13: ``optsize`` 842* bit 14: ``ssp`` 843* bit 15: ``sspreq`` 844* bits 16-31: ``align n`` 845* bit 32: ``nocapture`` 846* bit 33: ``noredzone`` 847* bit 34: ``noimplicitfloat`` 848* bit 35: ``naked`` 849* bit 36: ``inlinehint`` 850* bits 37-39: ``alignstack n``, represented as the logarithm 851 base 2 of the requested alignment, plus 1 852 853.. _TYPE_BLOCK: 854 855TYPE_BLOCK Contents 856------------------- 857 858The ``TYPE_BLOCK`` block (id 10) contains records which constitute a table of 859type operator entries used to represent types referenced within an LLVM 860module. Each record (with the exception of `NUMENTRY`_) generates a single type 861table entry, which may be referenced by 0-based index from instructions, 862constants, metadata, type symbol table entries, or other type operator records. 863 864Entries within ``TYPE_BLOCK`` are constructed to ensure that each entry is 865unique (i.e., no two indicies represent structurally equivalent types). 866 867.. _TYPE_CODE_NUMENTRY: 868.. _NUMENTRY: 869 870TYPE_CODE_NUMENTRY Record 871^^^^^^^^^^^^^^^^^^^^^^^^^ 872 873``[NUMENTRY, numentries]`` 874 875The ``NUMENTRY`` record (code 1) contains a single value which indicates the 876total number of type code entries in the type table of the module. If present, 877``NUMENTRY`` should be the first record in the block. 878 879TYPE_CODE_VOID Record 880^^^^^^^^^^^^^^^^^^^^^ 881 882``[VOID]`` 883 884The ``VOID`` record (code 2) adds a ``void`` type to the type table. 885 886TYPE_CODE_HALF Record 887^^^^^^^^^^^^^^^^^^^^^ 888 889``[HALF]`` 890 891The ``HALF`` record (code 10) adds a ``half`` (16-bit floating point) type to 892the type table. 893 894TYPE_CODE_FLOAT Record 895^^^^^^^^^^^^^^^^^^^^^^ 896 897``[FLOAT]`` 898 899The ``FLOAT`` record (code 3) adds a ``float`` (32-bit floating point) type to 900the type table. 901 902TYPE_CODE_DOUBLE Record 903^^^^^^^^^^^^^^^^^^^^^^^ 904 905``[DOUBLE]`` 906 907The ``DOUBLE`` record (code 4) adds a ``double`` (64-bit floating point) type to 908the type table. 909 910TYPE_CODE_LABEL Record 911^^^^^^^^^^^^^^^^^^^^^^ 912 913``[LABEL]`` 914 915The ``LABEL`` record (code 5) adds a ``label`` type to the type table. 916 917TYPE_CODE_OPAQUE Record 918^^^^^^^^^^^^^^^^^^^^^^^ 919 920``[OPAQUE]`` 921 922The ``OPAQUE`` record (code 6) adds an ``opaque`` type to the type table. Note 923that distinct ``opaque`` types are not unified. 924 925TYPE_CODE_INTEGER Record 926^^^^^^^^^^^^^^^^^^^^^^^^ 927 928``[INTEGER, width]`` 929 930The ``INTEGER`` record (code 7) adds an integer type to the type table. The 931single *width* field indicates the width of the integer type. 932 933TYPE_CODE_POINTER Record 934^^^^^^^^^^^^^^^^^^^^^^^^ 935 936``[POINTER, pointee type, address space]`` 937 938The ``POINTER`` record (code 8) adds a pointer type to the type table. The 939operand fields are 940 941* *pointee type*: The type index of the pointed-to type 942 943* *address space*: If supplied, the target-specific numbered address space where 944 the pointed-to object resides. Otherwise, the default address space is zero. 945 946TYPE_CODE_FUNCTION Record 947^^^^^^^^^^^^^^^^^^^^^^^^^ 948 949``[FUNCTION, vararg, ignored, retty, ...paramty... ]`` 950 951The ``FUNCTION`` record (code 9) adds a function type to the type table. The 952operand fields are 953 954* *vararg*: Non-zero if the type represents a varargs function 955 956* *ignored*: This value field is present for backward compatibility only, and is 957 ignored 958 959* *retty*: The type index of the function's return type 960 961* *paramty*: Zero or more type indices representing the parameter types of the 962 function 963 964TYPE_CODE_STRUCT Record 965^^^^^^^^^^^^^^^^^^^^^^^ 966 967``[STRUCT, ispacked, ...eltty...]`` 968 969The ``STRUCT`` record (code 10) adds a struct type to the type table. The 970operand fields are 971 972* *ispacked*: Non-zero if the type represents a packed structure 973 974* *eltty*: Zero or more type indices representing the element types of the 975 structure 976 977TYPE_CODE_ARRAY Record 978^^^^^^^^^^^^^^^^^^^^^^ 979 980``[ARRAY, numelts, eltty]`` 981 982The ``ARRAY`` record (code 11) adds an array type to the type table. The 983operand fields are 984 985* *numelts*: The number of elements in arrays of this type 986 987* *eltty*: The type index of the array element type 988 989TYPE_CODE_VECTOR Record 990^^^^^^^^^^^^^^^^^^^^^^^ 991 992``[VECTOR, numelts, eltty]`` 993 994The ``VECTOR`` record (code 12) adds a vector type to the type table. The 995operand fields are 996 997* *numelts*: The number of elements in vectors of this type 998 999* *eltty*: The type index of the vector element type 1000 1001TYPE_CODE_X86_FP80 Record 1002^^^^^^^^^^^^^^^^^^^^^^^^^ 1003 1004``[X86_FP80]`` 1005 1006The ``X86_FP80`` record (code 13) adds an ``x86_fp80`` (80-bit floating point) 1007type to the type table. 1008 1009TYPE_CODE_FP128 Record 1010^^^^^^^^^^^^^^^^^^^^^^ 1011 1012``[FP128]`` 1013 1014The ``FP128`` record (code 14) adds an ``fp128`` (128-bit floating point) type 1015to the type table. 1016 1017TYPE_CODE_PPC_FP128 Record 1018^^^^^^^^^^^^^^^^^^^^^^^^^^ 1019 1020``[PPC_FP128]`` 1021 1022The ``PPC_FP128`` record (code 15) adds a ``ppc_fp128`` (128-bit floating point) 1023type to the type table. 1024 1025TYPE_CODE_METADATA Record 1026^^^^^^^^^^^^^^^^^^^^^^^^^ 1027 1028``[METADATA]`` 1029 1030The ``METADATA`` record (code 16) adds a ``metadata`` type to the type table. 1031 1032.. _CONSTANTS_BLOCK: 1033 1034CONSTANTS_BLOCK Contents 1035------------------------ 1036 1037The ``CONSTANTS_BLOCK`` block (id 11) ... 1038 1039.. _FUNCTION_BLOCK: 1040 1041FUNCTION_BLOCK Contents 1042----------------------- 1043 1044The ``FUNCTION_BLOCK`` block (id 12) ... 1045 1046In addition to the record types described below, a ``FUNCTION_BLOCK`` block may 1047contain the following sub-blocks: 1048 1049* `CONSTANTS_BLOCK`_ 1050* `VALUE_SYMTAB_BLOCK`_ 1051* `METADATA_ATTACHMENT`_ 1052 1053.. _TYPE_SYMTAB_BLOCK: 1054 1055TYPE_SYMTAB_BLOCK Contents 1056-------------------------- 1057 1058The ``TYPE_SYMTAB_BLOCK`` block (id 13) contains entries which map between 1059module-level named types and their corresponding type indices. 1060 1061.. _TST_CODE_ENTRY: 1062 1063TST_CODE_ENTRY Record 1064^^^^^^^^^^^^^^^^^^^^^ 1065 1066``[ENTRY, typeid, ...string...]`` 1067 1068The ``ENTRY`` record (code 1) contains a variable number of values, with the 1069first giving the type index of the designated type, and the remaining values 1070giving the character codes of the type name. Each entry corresponds to a single 1071named type. 1072 1073.. _VALUE_SYMTAB_BLOCK: 1074 1075VALUE_SYMTAB_BLOCK Contents 1076--------------------------- 1077 1078The ``VALUE_SYMTAB_BLOCK`` block (id 14) ... 1079 1080.. _METADATA_BLOCK: 1081 1082METADATA_BLOCK Contents 1083----------------------- 1084 1085The ``METADATA_BLOCK`` block (id 15) ... 1086 1087.. _METADATA_ATTACHMENT: 1088 1089METADATA_ATTACHMENT Contents 1090---------------------------- 1091 1092The ``METADATA_ATTACHMENT`` block (id 16) ... 1093