1 .. role:: raw-html(raw) 2 :format: html 3 4 ======================== 5 LLVM Bitcode File Format 6 ======================== 7 8 .. contents:: 9 :local: 10 11 Abstract 12 ======== 13 14 This document describes the LLVM bitstream file format and the encoding of the 15 LLVM IR into it. 16 17 Overview 18 ======== 19 20 What is commonly known as the LLVM bitcode file format (also, sometimes 21 anachronistically known as bytecode) is actually two things: a `bitstream 22 container format`_ and an `encoding of LLVM IR`_ into the container format. 23 24 The bitstream format is an abstract encoding of structured data, very similar to 25 XML in some ways. Like XML, bitstream files contain tags, and nested 26 structures, and you can parse the file without having to understand the tags. 27 Unlike XML, the bitstream format is a binary encoding, and unlike XML it 28 provides a mechanism for the file to self-describe "abbreviations", which are 29 effectively size optimizations for the content. 30 31 LLVM IR files may be optionally embedded into a `wrapper`_ structure, or in a 32 `native object file`_. Both of these mechanisms make it easy to embed extra 33 data along with LLVM IR files. 34 35 This document first describes the LLVM bitstream format, describes the wrapper 36 format, then describes the record structure used by LLVM IR files. 37 38 .. _bitstream container format: 39 40 Bitstream Format 41 ================ 42 43 The bitstream format is literally a stream of bits, with a very simple 44 structure. This structure consists of the following concepts: 45 46 * A "`magic number`_" that identifies the contents of the stream. 47 48 * Encoding `primitives`_ like variable bit-rate integers. 49 50 * `Blocks`_, which define nested content. 51 52 * `Data Records`_, which describe entities within the file. 53 54 * Abbreviations, which specify compression optimizations for the file. 55 56 Note that the :doc:`llvm-bcanalyzer <CommandGuide/llvm-bcanalyzer>` tool can be 57 used to dump and inspect arbitrary bitstreams, which is very useful for 58 understanding the encoding. 59 60 .. _magic number: 61 62 Magic Numbers 63 ------------- 64 65 The first two bytes of a bitcode file are 'BC' (``0x42``, ``0x43``). The second 66 two bytes are an application-specific magic number. Generic bitcode tools can 67 look at only the first two bytes to verify the file is bitcode, while 68 application-specific programs will want to look at all four. 69 70 .. _primitives: 71 72 Primitives 73 ---------- 74 75 A bitstream literally consists of a stream of bits, which are read in order 76 starting with the least significant bit of each byte. The stream is made up of 77 a number of primitive values that encode a stream of unsigned integer values. 78 These integers are encoded in two ways: either as `Fixed Width Integers`_ or as 79 `Variable Width Integers`_. 80 81 .. _Fixed Width Integers: 82 .. _fixed-width value: 83 84 Fixed Width Integers 85 ^^^^^^^^^^^^^^^^^^^^ 86 87 Fixed-width integer values have their low bits emitted directly to the file. 88 For example, a 3-bit integer value encodes 1 as 001. Fixed width integers are 89 used when there are a well-known number of options for a field. For example, 90 boolean values are usually encoded with a 1-bit wide integer. 91 92 .. _Variable Width Integers: 93 .. _Variable Width Integer: 94 .. _variable-width value: 95 96 Variable Width Integers 97 ^^^^^^^^^^^^^^^^^^^^^^^ 98 99 Variable-width integer (VBR) values encode values of arbitrary size, optimizing 100 for the case where the values are small. Given a 4-bit VBR field, any 3-bit 101 value (0 through 7) is encoded directly, with the high bit set to zero. Values 102 larger than N-1 bits emit their bits in a series of N-1 bit chunks, where all 103 but the last set the high bit. 104 105 For example, the value 27 (0x1B) is encoded as 1011 0011 when emitted as a vbr4 106 value. The first set of four bits indicates the value 3 (011) with a 107 continuation piece (indicated by a high bit of 1). The next word indicates a 108 value of 24 (011 << 3) with no continuation. The sum (3+24) yields the value 109 27. 110 111 .. _char6-encoded value: 112 113 6-bit characters 114 ^^^^^^^^^^^^^^^^ 115 116 6-bit characters encode common characters into a fixed 6-bit field. They 117 represent the following characters with the following 6-bit values: 118 119 :: 120 121 'a' .. 'z' --- 0 .. 25 122 'A' .. 'Z' --- 26 .. 51 123 '0' .. '9' --- 52 .. 61 124 '.' --- 62 125 '_' --- 63 126 127 This encoding is only suitable for encoding characters and strings that consist 128 only of the above characters. It is completely incapable of encoding characters 129 not in the set. 130 131 Word Alignment 132 ^^^^^^^^^^^^^^ 133 134 Occasionally, it is useful to emit zero bits until the bitstream is a multiple 135 of 32 bits. This ensures that the bit position in the stream can be represented 136 as a multiple of 32-bit words. 137 138 Abbreviation IDs 139 ---------------- 140 141 A bitstream is a sequential series of `Blocks`_ and `Data Records`_. Both of 142 these start with an abbreviation ID encoded as a fixed-bitwidth field. The 143 width is specified by the current block, as described below. The value of the 144 abbreviation ID specifies either a builtin ID (which have special meanings, 145 defined below) or one of the abbreviation IDs defined for the current block by 146 the stream itself. 147 148 The set of builtin abbrev IDs is: 149 150 * 0 - `END_BLOCK`_ --- This abbrev ID marks the end of the current block. 151 152 * 1 - `ENTER_SUBBLOCK`_ --- This abbrev ID marks the beginning of a new 153 block. 154 155 * 2 - `DEFINE_ABBREV`_ --- This defines a new abbreviation. 156 157 * 3 - `UNABBREV_RECORD`_ --- This ID specifies the definition of an 158 unabbreviated record. 159 160 Abbreviation IDs 4 and above are defined by the stream itself, and specify an 161 `abbreviated record encoding`_. 162 163 .. _Blocks: 164 165 Blocks 166 ------ 167 168 Blocks in a bitstream denote nested regions of the stream, and are identified by 169 a content-specific id number (for example, LLVM IR uses an ID of 12 to represent 170 function bodies). Block IDs 0-7 are reserved for `standard blocks`_ whose 171 meaning is defined by Bitcode; block IDs 8 and greater are application 172 specific. Nested blocks capture the hierarchical structure of the data encoded 173 in it, and various properties are associated with blocks as the file is parsed. 174 Block definitions allow the reader to efficiently skip blocks in constant time 175 if the reader wants a summary of blocks, or if it wants to efficiently skip data 176 it does not understand. The LLVM IR reader uses this mechanism to skip function 177 bodies, lazily reading them on demand. 178 179 When reading and encoding the stream, several properties are maintained for the 180 block. In particular, each block maintains: 181 182 #. A current abbrev id width. This value starts at 2 at the beginning of the 183 stream, and is set every time a block record is entered. The block entry 184 specifies the abbrev id width for the body of the block. 185 186 #. A set of abbreviations. Abbreviations may be defined within a block, in 187 which case they are only defined in that block (neither subblocks nor 188 enclosing blocks see the abbreviation). Abbreviations can also be defined 189 inside a `BLOCKINFO`_ block, in which case they are defined in all blocks 190 that match the ID that the ``BLOCKINFO`` block is describing. 191 192 As sub blocks are entered, these properties are saved and the new sub-block has 193 its own set of abbreviations, and its own abbrev id width. When a sub-block is 194 popped, the saved values are restored. 195 196 .. _ENTER_SUBBLOCK: 197 198 ENTER_SUBBLOCK Encoding 199 ^^^^^^^^^^^^^^^^^^^^^^^ 200 201 :raw-html:`<tt>` 202 [ENTER_SUBBLOCK, blockid\ :sub:`vbr8`, newabbrevlen\ :sub:`vbr4`, <align32bits>, blocklen_32] 203 :raw-html:`</tt>` 204 205 The ``ENTER_SUBBLOCK`` abbreviation ID specifies the start of a new block 206 record. The ``blockid`` value is encoded as an 8-bit VBR identifier, and 207 indicates the type of block being entered, which can be a `standard block`_ or 208 an application-specific block. The ``newabbrevlen`` value is a 4-bit VBR, which 209 specifies the abbrev id width for the sub-block. The ``blocklen`` value is a 210 32-bit aligned value that specifies the size of the subblock in 32-bit 211 words. This value allows the reader to skip over the entire block in one jump. 212 213 .. _END_BLOCK: 214 215 END_BLOCK Encoding 216 ^^^^^^^^^^^^^^^^^^ 217 218 ``[END_BLOCK, <align32bits>]`` 219 220 The ``END_BLOCK`` abbreviation ID specifies the end of the current block record. 221 Its end is aligned to 32-bits to ensure that the size of the block is an even 222 multiple of 32-bits. 223 224 .. _Data Records: 225 226 Data Records 227 ------------ 228 229 Data records consist of a record code and a number of (up to) 64-bit integer 230 values. The interpretation of the code and values is application specific and 231 may vary between different block types. Records can be encoded either using an 232 unabbrev record, or with an abbreviation. In the LLVM IR format, for example, 233 there is a record which encodes the target triple of a module. The code is 234 ``MODULE_CODE_TRIPLE``, and the values of the record are the ASCII codes for the 235 characters in the string. 236 237 .. _UNABBREV_RECORD: 238 239 UNABBREV_RECORD Encoding 240 ^^^^^^^^^^^^^^^^^^^^^^^^ 241 242 :raw-html:`<tt>` 243 [UNABBREV_RECORD, code\ :sub:`vbr6`, numops\ :sub:`vbr6`, op0\ :sub:`vbr6`, op1\ :sub:`vbr6`, ...] 244 :raw-html:`</tt>` 245 246 An ``UNABBREV_RECORD`` provides a default fallback encoding, which is both 247 completely general and extremely inefficient. It can describe an arbitrary 248 record by emitting the code and operands as VBRs. 249 250 For example, emitting an LLVM IR target triple as an unabbreviated record 251 requires emitting the ``UNABBREV_RECORD`` abbrevid, a vbr6 for the 252 ``MODULE_CODE_TRIPLE`` code, a vbr6 for the length of the string, which is equal 253 to the number of operands, and a vbr6 for each character. Because there are no 254 letters with values less than 32, each letter would need to be emitted as at 255 least a two-part VBR, which means that each letter would require at least 12 256 bits. This is not an efficient encoding, but it is fully general. 257 258 .. _abbreviated record encoding: 259 260 Abbreviated Record Encoding 261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^ 262 263 ``[<abbrevid>, fields...]`` 264 265 An abbreviated record is a abbreviation id followed by a set of fields that are 266 encoded according to the `abbreviation definition`_. This allows records to be 267 encoded significantly more densely than records encoded with the 268 `UNABBREV_RECORD`_ type, and allows the abbreviation types to be specified in 269 the stream itself, which allows the files to be completely self describing. The 270 actual encoding of abbreviations is defined below. 271 272 The record code, which is the first field of an abbreviated record, may be 273 encoded in the abbreviation definition (as a literal operand) or supplied in the 274 abbreviated record (as a Fixed or VBR operand value). 275 276 .. _abbreviation definition: 277 278 Abbreviations 279 ------------- 280 281 Abbreviations are an important form of compression for bitstreams. The idea is 282 to specify a dense encoding for a class of records once, then use that encoding 283 to emit many records. It takes space to emit the encoding into the file, but 284 the space is recouped (hopefully plus some) when the records that use it are 285 emitted. 286 287 Abbreviations can be determined dynamically per client, per file. Because the 288 abbreviations are stored in the bitstream itself, different streams of the same 289 format can contain different sets of abbreviations according to the needs of the 290 specific stream. As a concrete example, LLVM IR files usually emit an 291 abbreviation for binary operators. If a specific LLVM module contained no or 292 few binary operators, the abbreviation does not need to be emitted. 293 294 .. _DEFINE_ABBREV: 295 296 DEFINE_ABBREV Encoding 297 ^^^^^^^^^^^^^^^^^^^^^^ 298 299 :raw-html:`<tt>` 300 [DEFINE_ABBREV, numabbrevops\ :sub:`vbr5`, abbrevop0, abbrevop1, ...] 301 :raw-html:`</tt>` 302 303 A ``DEFINE_ABBREV`` record adds an abbreviation to the list of currently defined 304 abbreviations in the scope of this block. This definition only exists inside 305 this immediate block --- it is not visible in subblocks or enclosing blocks. 306 Abbreviations are implicitly assigned IDs sequentially starting from 4 (the 307 first application-defined abbreviation ID). Any abbreviations defined in a 308 ``BLOCKINFO`` record for the particular block type receive IDs first, in order, 309 followed by any abbreviations defined within the block itself. Abbreviated data 310 records reference this ID to indicate what abbreviation they are invoking. 311 312 An abbreviation definition consists of the ``DEFINE_ABBREV`` abbrevid followed 313 by a VBR that specifies the number of abbrev operands, then the abbrev operands 314 themselves. Abbreviation operands come in three forms. They all start with a 315 single bit that indicates whether the abbrev operand is a literal operand (when 316 the bit is 1) or an encoding operand (when the bit is 0). 317 318 #. Literal operands --- :raw-html:`<tt>` [1\ :sub:`1`, litvalue\ 319 :sub:`vbr8`] :raw-html:`</tt>` --- Literal operands specify that the value in 320 the result is always a single specific value. This specific value is emitted 321 as a vbr8 after the bit indicating that it is a literal operand. 322 323 #. Encoding info without data --- :raw-html:`<tt>` [0\ :sub:`1`, encoding\ 324 :sub:`3`] :raw-html:`</tt>` --- Operand encodings that do not have extra data 325 are just emitted as their code. 326 327 #. Encoding info with data --- :raw-html:`<tt>` [0\ :sub:`1`, encoding\ 328 :sub:`3`, value\ :sub:`vbr5`] :raw-html:`</tt>` --- Operand encodings that do 329 have extra data are emitted as their code, followed by the extra data. 330 331 The possible operand encodings are: 332 333 * Fixed (code 1): The field should be emitted as a `fixed-width value`_, whose 334 width is specified by the operand's extra data. 335 336 * VBR (code 2): The field should be emitted as a `variable-width value`_, whose 337 width is specified by the operand's extra data. 338 339 * Array (code 3): This field is an array of values. The array operand has no 340 extra data, but expects another operand to follow it, indicating the element 341 type of the array. When reading an array in an abbreviated record, the first 342 integer is a vbr6 that indicates the array length, followed by the encoded 343 elements of the array. An array may only occur as the last operand of an 344 abbreviation (except for the one final operand that gives the array's 345 type). 346 347 * Char6 (code 4): This field should be emitted as a `char6-encoded value`_. 348 This operand type takes no extra data. Char6 encoding is normally used as an 349 array element type. 350 351 * Blob (code 5): This field is emitted as a vbr6, followed by padding to a 352 32-bit boundary (for alignment) and an array of 8-bit objects. The array of 353 bytes is further followed by tail padding to ensure that its total length is a 354 multiple of 4 bytes. This makes it very efficient for the reader to decode 355 the data without having to make a copy of it: it can use a pointer to the data 356 in the mapped in file and poke directly at it. A blob may only occur as the 357 last operand of an abbreviation. 358 359 For example, target triples in LLVM modules are encoded as a record of the form 360 ``[TRIPLE, 'a', 'b', 'c', 'd']``. Consider if the bitstream emitted the 361 following abbrev entry: 362 363 :: 364 365 [0, Fixed, 4] 366 [0, Array] 367 [0, Char6] 368 369 When emitting a record with this abbreviation, the above entry would be emitted 370 as: 371 372 :raw-html:`<tt><blockquote>` 373 [4\ :sub:`abbrevwidth`, 2\ :sub:`4`, 4\ :sub:`vbr6`, 0\ :sub:`6`, 1\ :sub:`6`, 2\ :sub:`6`, 3\ :sub:`6`] 374 :raw-html:`</blockquote></tt>` 375 376 These values are: 377 378 #. The first value, 4, is the abbreviation ID for this abbreviation. 379 380 #. The second value, 2, is the record code for ``TRIPLE`` records within LLVM IR 381 file ``MODULE_BLOCK`` blocks. 382 383 #. The third value, 4, is the length of the array. 384 385 #. The rest of the values are the char6 encoded values for ``"abcd"``. 386 387 With this abbreviation, the triple is emitted with only 37 bits (assuming a 388 abbrev id width of 3). Without the abbreviation, significantly more space would 389 be required to emit the target triple. Also, because the ``TRIPLE`` value is 390 not emitted as a literal in the abbreviation, the abbreviation can also be used 391 for any other string value. 392 393 .. _standard blocks: 394 .. _standard block: 395 396 Standard Blocks 397 --------------- 398 399 In addition to the basic block structure and record encodings, the bitstream 400 also defines specific built-in block types. These block types specify how the 401 stream is to be decoded or other metadata. In the future, new standard blocks 402 may be added. Block IDs 0-7 are reserved for standard blocks. 403 404 .. _BLOCKINFO: 405 406 #0 - BLOCKINFO Block 407 ^^^^^^^^^^^^^^^^^^^^ 408 409 The ``BLOCKINFO`` block allows the description of metadata for other blocks. 410 The currently specified records are: 411 412 :: 413 414 [SETBID (#1), blockid] 415 [DEFINE_ABBREV, ...] 416 [BLOCKNAME, ...name...] 417 [SETRECORDNAME, RecordID, ...name...] 418 419 The ``SETBID`` record (code 1) indicates which block ID is being described. 420 ``SETBID`` records can occur multiple times throughout the block to change which 421 block ID is being described. There must be a ``SETBID`` record prior to any 422 other records. 423 424 Standard ``DEFINE_ABBREV`` records can occur inside ``BLOCKINFO`` blocks, but 425 unlike their occurrence in normal blocks, the abbreviation is defined for blocks 426 matching the block ID we are describing, *not* the ``BLOCKINFO`` block 427 itself. The abbreviations defined in ``BLOCKINFO`` blocks receive abbreviation 428 IDs as described in `DEFINE_ABBREV`_. 429 430 The ``BLOCKNAME`` record (code 2) can optionally occur in this block. The 431 elements of the record are the bytes of the string name of the block. 432 llvm-bcanalyzer can use this to dump out bitcode files symbolically. 433 434 The ``SETRECORDNAME`` record (code 3) can also optionally occur in this block. 435 The first operand value is a record ID number, and the rest of the elements of 436 the record are the bytes for the string name of the record. llvm-bcanalyzer can 437 use this to dump out bitcode files symbolically. 438 439 Note that although the data in ``BLOCKINFO`` blocks is described as "metadata," 440 the abbreviations they contain are essential for parsing records from the 441 corresponding blocks. It is not safe to skip them. 442 443 .. _wrapper: 444 445 Bitcode Wrapper Format 446 ====================== 447 448 Bitcode files for LLVM IR may optionally be wrapped in a simple wrapper 449 structure. This structure contains a simple header that indicates the offset 450 and size of the embedded BC file. This allows additional information to be 451 stored alongside the BC file. The structure of this file header is: 452 453 :raw-html:`<tt><blockquote>` 454 [Magic\ :sub:`32`, Version\ :sub:`32`, Offset\ :sub:`32`, Size\ :sub:`32`, CPUType\ :sub:`32`] 455 :raw-html:`</blockquote></tt>` 456 457 Each of the fields are 32-bit fields stored in little endian form (as with the 458 rest of the bitcode file fields). The Magic number is always ``0x0B17C0DE`` and 459 the version is currently always ``0``. The Offset field is the offset in bytes 460 to the start of the bitcode stream in the file, and the Size field is the size 461 in bytes of the stream. CPUType is a target-specific value that can be used to 462 encode the CPU of the target. 463 464 .. _native object file: 465 466 Native Object File Wrapper Format 467 ================================= 468 469 Bitcode files for LLVM IR may also be wrapped in a native object file 470 (i.e. ELF, COFF, Mach-O). The bitcode must be stored in a section of the 471 object file named ``.llvmbc``. This wrapper format is useful for accommodating 472 LTO in compilation pipelines where intermediate objects must be native object 473 files which contain metadata in other sections. 474 475 Not all tools support this format. 476 477 .. _encoding of LLVM IR: 478 479 LLVM IR Encoding 480 ================ 481 482 LLVM IR is encoded into a bitstream by defining blocks and records. It uses 483 blocks for things like constant pools, functions, symbol tables, etc. It uses 484 records for things like instructions, global variable descriptors, type 485 descriptions, etc. This document does not describe the set of abbreviations 486 that the writer uses, as these are fully self-described in the file, and the 487 reader is not allowed to build in any knowledge of this. 488 489 Basics 490 ------ 491 492 LLVM IR Magic Number 493 ^^^^^^^^^^^^^^^^^^^^ 494 495 The magic number for LLVM IR files is: 496 497 :raw-html:`<tt><blockquote>` 498 [0x0\ :sub:`4`, 0xC\ :sub:`4`, 0xE\ :sub:`4`, 0xD\ :sub:`4`] 499 :raw-html:`</blockquote></tt>` 500 501 When combined with the bitcode magic number and viewed as bytes, this is 502 ``"BC 0xC0DE"``. 503 504 .. _Signed VBRs: 505 506 Signed VBRs 507 ^^^^^^^^^^^ 508 509 `Variable Width Integer`_ encoding is an efficient way to encode arbitrary sized 510 unsigned values, but is an extremely inefficient for encoding signed values, as 511 signed values are otherwise treated as maximally large unsigned values. 512 513 As such, signed VBR values of a specific width are emitted as follows: 514 515 * Positive values are emitted as VBRs of the specified width, but with their 516 value shifted left by one. 517 518 * Negative values are emitted as VBRs of the specified width, but the negated 519 value is shifted left by one, and the low bit is set. 520 521 With this encoding, small positive and small negative values can both be emitted 522 efficiently. Signed VBR encoding is used in ``CST_CODE_INTEGER`` and 523 ``CST_CODE_WIDE_INTEGER`` records within ``CONSTANTS_BLOCK`` blocks. 524 It is also used for phi instruction operands in `MODULE_CODE_VERSION`_ 1. 525 526 LLVM IR Blocks 527 ^^^^^^^^^^^^^^ 528 529 LLVM IR is defined with the following blocks: 530 531 * 8 --- `MODULE_BLOCK`_ --- This is the top-level block that contains the entire 532 module, and describes a variety of per-module information. 533 534 * 9 --- `PARAMATTR_BLOCK`_ --- This enumerates the parameter attributes. 535 536 * 10 --- `TYPE_BLOCK`_ --- This describes all of the types in the module. 537 538 * 11 --- `CONSTANTS_BLOCK`_ --- This describes constants for a module or 539 function. 540 541 * 12 --- `FUNCTION_BLOCK`_ --- This describes a function body. 542 543 * 13 --- `TYPE_SYMTAB_BLOCK`_ --- This describes the type symbol table. 544 545 * 14 --- `VALUE_SYMTAB_BLOCK`_ --- This describes a value symbol table. 546 547 * 15 --- `METADATA_BLOCK`_ --- This describes metadata items. 548 549 * 16 --- `METADATA_ATTACHMENT`_ --- This contains records associating metadata 550 with function instruction values. 551 552 .. _MODULE_BLOCK: 553 554 MODULE_BLOCK Contents 555 --------------------- 556 557 The ``MODULE_BLOCK`` block (id 8) is the top-level block for LLVM bitcode files, 558 and each bitcode file must contain exactly one. In addition to records 559 (described below) containing information about the module, a ``MODULE_BLOCK`` 560 block may contain the following sub-blocks: 561 562 * `BLOCKINFO`_ 563 * `PARAMATTR_BLOCK`_ 564 * `TYPE_BLOCK`_ 565 * `TYPE_SYMTAB_BLOCK`_ 566 * `VALUE_SYMTAB_BLOCK`_ 567 * `CONSTANTS_BLOCK`_ 568 * `FUNCTION_BLOCK`_ 569 * `METADATA_BLOCK`_ 570 571 .. _MODULE_CODE_VERSION: 572 573 MODULE_CODE_VERSION Record 574 ^^^^^^^^^^^^^^^^^^^^^^^^^^ 575 576 ``[VERSION, version#]`` 577 578 The ``VERSION`` record (code 1) contains a single value indicating the format 579 version. Versions 0 and 1 are supported at this time. The difference between 580 version 0 and 1 is in the encoding of instruction operands in 581 each `FUNCTION_BLOCK`_. 582 583 In version 0, each value defined by an instruction is assigned an ID 584 unique to the function. Function-level value IDs are assigned starting from 585 ``NumModuleValues`` since they share the same namespace as module-level 586 values. The value enumerator resets after each function. When a value is 587 an operand of an instruction, the value ID is used to represent the operand. 588 For large functions or large modules, these operand values can be large. 589 590 The encoding in version 1 attempts to avoid large operand values 591 in common cases. Instead of using the value ID directly, operands are 592 encoded as relative to the current instruction. Thus, if an operand 593 is the value defined by the previous instruction, the operand 594 will be encoded as 1. 595 596 For example, instead of 597 598 .. code-block:: llvm 599 600 #n = load #n-1 601 #n+1 = icmp eq #n, #const0 602 br #n+1, label #(bb1), label #(bb2) 603 604 version 1 will encode the instructions as 605 606 .. code-block:: llvm 607 608 #n = load #1 609 #n+1 = icmp eq #1, (#n+1)-#const0 610 br #1, label #(bb1), label #(bb2) 611 612 Note in the example that operands which are constants also use 613 the relative encoding, while operands like basic block labels 614 do not use the relative encoding. 615 616 Forward references will result in a negative value. 617 This can be inefficient, as operands are normally encoded 618 as unsigned VBRs. However, forward references are rare, except in the 619 case of phi instructions. For phi instructions, operands are encoded as 620 `Signed VBRs`_ to deal with forward references. 621 622 623 MODULE_CODE_TRIPLE Record 624 ^^^^^^^^^^^^^^^^^^^^^^^^^ 625 626 ``[TRIPLE, ...string...]`` 627 628 The ``TRIPLE`` record (code 2) contains a variable number of values representing 629 the bytes of the ``target triple`` specification string. 630 631 MODULE_CODE_DATALAYOUT Record 632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 633 634 ``[DATALAYOUT, ...string...]`` 635 636 The ``DATALAYOUT`` record (code 3) contains a variable number of values 637 representing the bytes of the ``target datalayout`` specification string. 638 639 MODULE_CODE_ASM Record 640 ^^^^^^^^^^^^^^^^^^^^^^ 641 642 ``[ASM, ...string...]`` 643 644 The ``ASM`` record (code 4) contains a variable number of values representing 645 the bytes of ``module asm`` strings, with individual assembly blocks separated 646 by newline (ASCII 10) characters. 647 648 .. _MODULE_CODE_SECTIONNAME: 649 650 MODULE_CODE_SECTIONNAME Record 651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 652 653 ``[SECTIONNAME, ...string...]`` 654 655 The ``SECTIONNAME`` record (code 5) contains a variable number of values 656 representing the bytes of a single section name string. There should be one 657 ``SECTIONNAME`` record for each section name referenced (e.g., in global 658 variable or function ``section`` attributes) within the module. These records 659 can be referenced by the 1-based index in the *section* fields of ``GLOBALVAR`` 660 or ``FUNCTION`` records. 661 662 MODULE_CODE_DEPLIB Record 663 ^^^^^^^^^^^^^^^^^^^^^^^^^ 664 665 ``[DEPLIB, ...string...]`` 666 667 The ``DEPLIB`` record (code 6) contains a variable number of values representing 668 the bytes of a single dependent library name string, one of the libraries 669 mentioned in a ``deplibs`` declaration. There should be one ``DEPLIB`` record 670 for each library name referenced. 671 672 MODULE_CODE_GLOBALVAR Record 673 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 674 675 ``[GLOBALVAR, pointer type, isconst, initid, linkage, alignment, section, visibility, threadlocal, unnamed_addr, externally_initialized, dllstorageclass, comdat]`` 676 677 The ``GLOBALVAR`` record (code 7) marks the declaration or definition of a 678 global variable. The operand fields are: 679 680 * *pointer type*: The type index of the pointer type used to point to this 681 global variable 682 683 * *isconst*: Non-zero if the variable is treated as constant within the module, 684 or zero if it is not 685 686 * *initid*: If non-zero, the value index of the initializer for this variable, 687 plus 1. 688 689 .. _linkage type: 690 691 * *linkage*: An encoding of the linkage type for this variable: 692 * ``external``: code 0 693 * ``weak``: code 1 694 * ``appending``: code 2 695 * ``internal``: code 3 696 * ``linkonce``: code 4 697 * ``dllimport``: code 5 698 * ``dllexport``: code 6 699 * ``extern_weak``: code 7 700 * ``common``: code 8 701 * ``private``: code 9 702 * ``weak_odr``: code 10 703 * ``linkonce_odr``: code 11 704 * ``available_externally``: code 12 705 * deprecated : code 13 706 * deprecated : code 14 707 708 * alignment*: The logarithm base 2 of the variable's requested alignment, plus 1 709 710 * *section*: If non-zero, the 1-based section index in the table of 711 `MODULE_CODE_SECTIONNAME`_ entries. 712 713 .. _visibility: 714 715 * *visibility*: If present, an encoding of the visibility of this variable: 716 * ``default``: code 0 717 * ``hidden``: code 1 718 * ``protected``: code 2 719 720 * *threadlocal*: If present, an encoding of the thread local storage mode of the 721 variable: 722 * ``not thread local``: code 0 723 * ``thread local; default TLS model``: code 1 724 * ``localdynamic``: code 2 725 * ``initialexec``: code 3 726 * ``localexec``: code 4 727 728 * *unnamed_addr*: If present and non-zero, indicates that the variable has 729 ``unnamed_addr`` 730 731 .. _bcdllstorageclass: 732 733 * *dllstorageclass*: If present, an encoding of the DLL storage class of this variable: 734 735 * ``default``: code 0 736 * ``dllimport``: code 1 737 * ``dllexport``: code 2 738 739 .. _FUNCTION: 740 741 MODULE_CODE_FUNCTION Record 742 ^^^^^^^^^^^^^^^^^^^^^^^^^^^ 743 744 ``[FUNCTION, type, callingconv, isproto, linkage, paramattr, alignment, section, visibility, gc, prologuedata, dllstorageclass, comdat, prefixdata, personalityfn]`` 745 746 The ``FUNCTION`` record (code 8) marks the declaration or definition of a 747 function. The operand fields are: 748 749 * *type*: The type index of the function type describing this function 750 751 * *callingconv*: The calling convention number: 752 * ``ccc``: code 0 753 * ``fastcc``: code 8 754 * ``coldcc``: code 9 755 * ``webkit_jscc``: code 12 756 * ``anyregcc``: code 13 757 * ``preserve_mostcc``: code 14 758 * ``preserve_allcc``: code 15 759 * ``cxx_fast_tlscc``: code 17 760 * ``x86_stdcallcc``: code 64 761 * ``x86_fastcallcc``: code 65 762 * ``arm_apcscc``: code 66 763 * ``arm_aapcscc``: code 67 764 * ``arm_aapcs_vfpcc``: code 68 765 766 * isproto*: Non-zero if this entry represents a declaration rather than a 767 definition 768 769 * *linkage*: An encoding of the `linkage type`_ for this function 770 771 * *paramattr*: If nonzero, the 1-based parameter attribute index into the table 772 of `PARAMATTR_CODE_ENTRY`_ entries. 773 774 * *alignment*: The logarithm base 2 of the function's requested alignment, plus 775 1 776 777 * *section*: If non-zero, the 1-based section index in the table of 778 `MODULE_CODE_SECTIONNAME`_ entries. 779 780 * *visibility*: An encoding of the `visibility`_ of this function 781 782 * *gc*: If present and nonzero, the 1-based garbage collector index in the table 783 of `MODULE_CODE_GCNAME`_ entries. 784 785 * *unnamed_addr*: If present and non-zero, indicates that the function has 786 ``unnamed_addr`` 787 788 * *prologuedata*: If non-zero, the value index of the prologue data for this function, 789 plus 1. 790 791 * *dllstorageclass*: An encoding of the 792 :ref:`dllstorageclass<bcdllstorageclass>` of this function 793 794 * *comdat*: An encoding of the COMDAT of this function 795 796 * *prefixdata*: If non-zero, the value index of the prefix data for this function, 797 plus 1. 798 799 * *personalityfn*: If non-zero, the value index of the personality function for this function, 800 plus 1. 801 802 MODULE_CODE_ALIAS Record 803 ^^^^^^^^^^^^^^^^^^^^^^^^ 804 805 ``[ALIAS, alias type, aliasee val#, linkage, visibility, dllstorageclass]`` 806 807 The ``ALIAS`` record (code 9) marks the definition of an alias. The operand 808 fields are 809 810 * *alias type*: The type index of the alias 811 812 * *aliasee val#*: The value index of the aliased value 813 814 * *linkage*: An encoding of the `linkage type`_ for this alias 815 816 * *visibility*: If present, an encoding of the `visibility`_ of the alias 817 818 * *dllstorageclass*: If present, an encoding of the 819 :ref:`dllstorageclass<bcdllstorageclass>` of the alias 820 821 MODULE_CODE_PURGEVALS Record 822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 823 824 ``[PURGEVALS, numvals]`` 825 826 The ``PURGEVALS`` record (code 10) resets the module-level value list to the 827 size given by the single operand value. Module-level value list items are added 828 by ``GLOBALVAR``, ``FUNCTION``, and ``ALIAS`` records. After a ``PURGEVALS`` 829 record is seen, new value indices will start from the given *numvals* value. 830 831 .. _MODULE_CODE_GCNAME: 832 833 MODULE_CODE_GCNAME Record 834 ^^^^^^^^^^^^^^^^^^^^^^^^^ 835 836 ``[GCNAME, ...string...]`` 837 838 The ``GCNAME`` record (code 11) contains a variable number of values 839 representing the bytes of a single garbage collector name string. There should 840 be one ``GCNAME`` record for each garbage collector name referenced in function 841 ``gc`` attributes within the module. These records can be referenced by 1-based 842 index in the *gc* fields of ``FUNCTION`` records. 843 844 .. _PARAMATTR_BLOCK: 845 846 PARAMATTR_BLOCK Contents 847 ------------------------ 848 849 The ``PARAMATTR_BLOCK`` block (id 9) contains a table of entries describing the 850 attributes of function parameters. These entries are referenced by 1-based index 851 in the *paramattr* field of module block `FUNCTION`_ records, or within the 852 *attr* field of function block ``INST_INVOKE`` and ``INST_CALL`` records. 853 854 Entries within ``PARAMATTR_BLOCK`` are constructed to ensure that each is unique 855 (i.e., no two indices represent equivalent attribute lists). 856 857 .. _PARAMATTR_CODE_ENTRY: 858 859 PARAMATTR_CODE_ENTRY Record 860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^ 861 862 ``[ENTRY, paramidx0, attr0, paramidx1, attr1...]`` 863 864 The ``ENTRY`` record (code 1) contains an even number of values describing a 865 unique set of function parameter attributes. Each *paramidx* value indicates 866 which set of attributes is represented, with 0 representing the return value 867 attributes, 0xFFFFFFFF representing function attributes, and other values 868 representing 1-based function parameters. Each *attr* value is a bitmap with the 869 following interpretation: 870 871 * bit 0: ``zeroext`` 872 * bit 1: ``signext`` 873 * bit 2: ``noreturn`` 874 * bit 3: ``inreg`` 875 * bit 4: ``sret`` 876 * bit 5: ``nounwind`` 877 * bit 6: ``noalias`` 878 * bit 7: ``byval`` 879 * bit 8: ``nest`` 880 * bit 9: ``readnone`` 881 * bit 10: ``readonly`` 882 * bit 11: ``noinline`` 883 * bit 12: ``alwaysinline`` 884 * bit 13: ``optsize`` 885 * bit 14: ``ssp`` 886 * bit 15: ``sspreq`` 887 * bits 16-31: ``align n`` 888 * bit 32: ``nocapture`` 889 * bit 33: ``noredzone`` 890 * bit 34: ``noimplicitfloat`` 891 * bit 35: ``naked`` 892 * bit 36: ``inlinehint`` 893 * bits 37-39: ``alignstack n``, represented as the logarithm 894 base 2 of the requested alignment, plus 1 895 896 .. _TYPE_BLOCK: 897 898 TYPE_BLOCK Contents 899 ------------------- 900 901 The ``TYPE_BLOCK`` block (id 10) contains records which constitute a table of 902 type operator entries used to represent types referenced within an LLVM 903 module. Each record (with the exception of `NUMENTRY`_) generates a single type 904 table entry, which may be referenced by 0-based index from instructions, 905 constants, metadata, type symbol table entries, or other type operator records. 906 907 Entries within ``TYPE_BLOCK`` are constructed to ensure that each entry is 908 unique (i.e., no two indices represent structurally equivalent types). 909 910 .. _TYPE_CODE_NUMENTRY: 911 .. _NUMENTRY: 912 913 TYPE_CODE_NUMENTRY Record 914 ^^^^^^^^^^^^^^^^^^^^^^^^^ 915 916 ``[NUMENTRY, numentries]`` 917 918 The ``NUMENTRY`` record (code 1) contains a single value which indicates the 919 total number of type code entries in the type table of the module. If present, 920 ``NUMENTRY`` should be the first record in the block. 921 922 TYPE_CODE_VOID Record 923 ^^^^^^^^^^^^^^^^^^^^^ 924 925 ``[VOID]`` 926 927 The ``VOID`` record (code 2) adds a ``void`` type to the type table. 928 929 TYPE_CODE_HALF Record 930 ^^^^^^^^^^^^^^^^^^^^^ 931 932 ``[HALF]`` 933 934 The ``HALF`` record (code 10) adds a ``half`` (16-bit floating point) type to 935 the type table. 936 937 TYPE_CODE_FLOAT Record 938 ^^^^^^^^^^^^^^^^^^^^^^ 939 940 ``[FLOAT]`` 941 942 The ``FLOAT`` record (code 3) adds a ``float`` (32-bit floating point) type to 943 the type table. 944 945 TYPE_CODE_DOUBLE Record 946 ^^^^^^^^^^^^^^^^^^^^^^^ 947 948 ``[DOUBLE]`` 949 950 The ``DOUBLE`` record (code 4) adds a ``double`` (64-bit floating point) type to 951 the type table. 952 953 TYPE_CODE_LABEL Record 954 ^^^^^^^^^^^^^^^^^^^^^^ 955 956 ``[LABEL]`` 957 958 The ``LABEL`` record (code 5) adds a ``label`` type to the type table. 959 960 TYPE_CODE_OPAQUE Record 961 ^^^^^^^^^^^^^^^^^^^^^^^ 962 963 ``[OPAQUE]`` 964 965 The ``OPAQUE`` record (code 6) adds an ``opaque`` type to the type table. Note 966 that distinct ``opaque`` types are not unified. 967 968 TYPE_CODE_INTEGER Record 969 ^^^^^^^^^^^^^^^^^^^^^^^^ 970 971 ``[INTEGER, width]`` 972 973 The ``INTEGER`` record (code 7) adds an integer type to the type table. The 974 single *width* field indicates the width of the integer type. 975 976 TYPE_CODE_POINTER Record 977 ^^^^^^^^^^^^^^^^^^^^^^^^ 978 979 ``[POINTER, pointee type, address space]`` 980 981 The ``POINTER`` record (code 8) adds a pointer type to the type table. The 982 operand fields are 983 984 * *pointee type*: The type index of the pointed-to type 985 986 * *address space*: If supplied, the target-specific numbered address space where 987 the pointed-to object resides. Otherwise, the default address space is zero. 988 989 TYPE_CODE_FUNCTION Record 990 ^^^^^^^^^^^^^^^^^^^^^^^^^ 991 992 ``[FUNCTION, vararg, ignored, retty, ...paramty... ]`` 993 994 The ``FUNCTION`` record (code 9) adds a function type to the type table. The 995 operand fields are 996 997 * *vararg*: Non-zero if the type represents a varargs function 998 999 * *ignored*: This value field is present for backward compatibility only, and is 1000 ignored 1001 1002 * *retty*: The type index of the function's return type 1003 1004 * *paramty*: Zero or more type indices representing the parameter types of the 1005 function 1006 1007 TYPE_CODE_STRUCT Record 1008 ^^^^^^^^^^^^^^^^^^^^^^^ 1009 1010 ``[STRUCT, ispacked, ...eltty...]`` 1011 1012 The ``STRUCT`` record (code 10) adds a struct type to the type table. The 1013 operand fields are 1014 1015 * *ispacked*: Non-zero if the type represents a packed structure 1016 1017 * *eltty*: Zero or more type indices representing the element types of the 1018 structure 1019 1020 TYPE_CODE_ARRAY Record 1021 ^^^^^^^^^^^^^^^^^^^^^^ 1022 1023 ``[ARRAY, numelts, eltty]`` 1024 1025 The ``ARRAY`` record (code 11) adds an array type to the type table. The 1026 operand fields are 1027 1028 * *numelts*: The number of elements in arrays of this type 1029 1030 * *eltty*: The type index of the array element type 1031 1032 TYPE_CODE_VECTOR Record 1033 ^^^^^^^^^^^^^^^^^^^^^^^ 1034 1035 ``[VECTOR, numelts, eltty]`` 1036 1037 The ``VECTOR`` record (code 12) adds a vector type to the type table. The 1038 operand fields are 1039 1040 * *numelts*: The number of elements in vectors of this type 1041 1042 * *eltty*: The type index of the vector element type 1043 1044 TYPE_CODE_X86_FP80 Record 1045 ^^^^^^^^^^^^^^^^^^^^^^^^^ 1046 1047 ``[X86_FP80]`` 1048 1049 The ``X86_FP80`` record (code 13) adds an ``x86_fp80`` (80-bit floating point) 1050 type to the type table. 1051 1052 TYPE_CODE_FP128 Record 1053 ^^^^^^^^^^^^^^^^^^^^^^ 1054 1055 ``[FP128]`` 1056 1057 The ``FP128`` record (code 14) adds an ``fp128`` (128-bit floating point) type 1058 to the type table. 1059 1060 TYPE_CODE_PPC_FP128 Record 1061 ^^^^^^^^^^^^^^^^^^^^^^^^^^ 1062 1063 ``[PPC_FP128]`` 1064 1065 The ``PPC_FP128`` record (code 15) adds a ``ppc_fp128`` (128-bit floating point) 1066 type to the type table. 1067 1068 TYPE_CODE_METADATA Record 1069 ^^^^^^^^^^^^^^^^^^^^^^^^^ 1070 1071 ``[METADATA]`` 1072 1073 The ``METADATA`` record (code 16) adds a ``metadata`` type to the type table. 1074 1075 .. _CONSTANTS_BLOCK: 1076 1077 CONSTANTS_BLOCK Contents 1078 ------------------------ 1079 1080 The ``CONSTANTS_BLOCK`` block (id 11) ... 1081 1082 .. _FUNCTION_BLOCK: 1083 1084 FUNCTION_BLOCK Contents 1085 ----------------------- 1086 1087 The ``FUNCTION_BLOCK`` block (id 12) ... 1088 1089 In addition to the record types described below, a ``FUNCTION_BLOCK`` block may 1090 contain the following sub-blocks: 1091 1092 * `CONSTANTS_BLOCK`_ 1093 * `VALUE_SYMTAB_BLOCK`_ 1094 * `METADATA_ATTACHMENT`_ 1095 1096 .. _TYPE_SYMTAB_BLOCK: 1097 1098 TYPE_SYMTAB_BLOCK Contents 1099 -------------------------- 1100 1101 The ``TYPE_SYMTAB_BLOCK`` block (id 13) contains entries which map between 1102 module-level named types and their corresponding type indices. 1103 1104 .. _TST_CODE_ENTRY: 1105 1106 TST_CODE_ENTRY Record 1107 ^^^^^^^^^^^^^^^^^^^^^ 1108 1109 ``[ENTRY, typeid, ...string...]`` 1110 1111 The ``ENTRY`` record (code 1) contains a variable number of values, with the 1112 first giving the type index of the designated type, and the remaining values 1113 giving the character codes of the type name. Each entry corresponds to a single 1114 named type. 1115 1116 .. _VALUE_SYMTAB_BLOCK: 1117 1118 VALUE_SYMTAB_BLOCK Contents 1119 --------------------------- 1120 1121 The ``VALUE_SYMTAB_BLOCK`` block (id 14) ... 1122 1123 .. _METADATA_BLOCK: 1124 1125 METADATA_BLOCK Contents 1126 ----------------------- 1127 1128 The ``METADATA_BLOCK`` block (id 15) ... 1129 1130 .. _METADATA_ATTACHMENT: 1131 1132 METADATA_ATTACHMENT Contents 1133 ---------------------------- 1134 1135 The ``METADATA_ATTACHMENT`` block (id 16) ... 1136