1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamcially
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as little
357 intrusive as possible. This calling convention behaves identical to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`, an opening
639 curly brace, a list of basic blocks, and a closing curly brace.
641 LLVM function declarations consist of the "``declare``" keyword, an
642 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
643 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
644 an optional :ref:`calling convention <callingconv>`,
645 an optional ``unnamed_addr`` attribute, a return type, an optional
646 :ref:`parameter attribute <paramattrs>` for the return type, a function
647 name, a possibly empty list of arguments, an optional alignment, an optional
648 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
649 and an optional :ref:`prologue <prologuedata>`.
651 A function definition contains a list of basic blocks, forming the CFG (Control
652 Flow Graph) for the function. Each basic block may optionally start with a label
653 (giving the basic block a symbol table entry), contains a list of instructions,
654 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
655 function return). If an explicit label is not provided, a block is assigned an
656 implicit numbered label, using the next value from the same counter as used for
657 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
658 entry block does not have an explicit label, it will be assigned label "%0",
659 then the first unnamed temporary in that block will be "%1", etc.
661 The first basic block in a function is special in two ways: it is
662 immediately executed on entrance to the function, and it is not allowed
663 to have predecessor basic blocks (i.e. there can not be any branches to
664 the entry block of a function). Because the block can have no
665 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
667 LLVM allows an explicit section to be specified for functions. If the
668 target supports it, it will emit functions to the section specified.
669 Additionally, the function can be placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is known to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
839 ; Some unnamed metadata nodes, which are referenced by the named metadata.
844 !name = !{!0, !1, !2}
851 The return type and each parameter of a function type may have a set of
852 *parameter attributes* associated with them. Parameter attributes are
853 used to communicate additional information about the result or
854 parameters of a function. Parameter attributes are considered to be part
855 of the function, not of the function type, so functions with different
856 parameter attributes can have the same function type.
858 Parameter attributes are simple keywords that follow the type specified.
859 If multiple parameter attributes are needed, they are space separated.
864 declare i32 @printf(i8* noalias nocapture, ...)
865 declare i32 @atoi(i8 zeroext)
866 declare signext i8 @returns_signed_char()
868 Note that any attributes for the function result (``nounwind``,
869 ``readonly``) come immediately after the argument list.
871 Currently, only the following parameter attributes are defined:
874 This indicates to the code generator that the parameter or return
875 value should be zero-extended to the extent required by the target's
876 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
877 the caller (for a parameter) or the callee (for a return value).
879 This indicates to the code generator that the parameter or return
880 value should be sign-extended to the extent required by the target's
881 ABI (which is usually 32-bits) by the caller (for a parameter) or
882 the callee (for a return value).
884 This indicates that this parameter or return value should be treated
885 in a special target-dependent fashion during while emitting code for
886 a function call or return (usually, by putting it in a register as
887 opposed to memory, though some targets use it to distinguish between
888 two different kinds of registers). Use of this attribute is
891 This indicates that the pointer parameter should really be passed by
892 value to the function. The attribute implies that a hidden copy of
893 the pointee is made between the caller and the callee, so the callee
894 is unable to modify the value in the caller. This attribute is only
895 valid on LLVM pointer arguments. It is generally used to pass
896 structs and arrays by value, but is also valid on pointers to
897 scalars. The copy is considered to belong to the caller not the
898 callee (for example, ``readonly`` functions should not write to
899 ``byval`` parameters). This is not a valid attribute for return
902 The byval attribute also supports specifying an alignment with the
903 align attribute. It indicates the alignment of the stack slot to
904 form and the known alignment of the pointer specified to the call
905 site. If the alignment is not specified, then the code generator
906 makes a target-specific assumption.
912 The ``inalloca`` argument attribute allows the caller to take the
913 address of outgoing stack arguments. An ``inalloca`` argument must
914 be a pointer to stack memory produced by an ``alloca`` instruction.
915 The alloca, or argument allocation, must also be tagged with the
916 inalloca keyword. Only the last argument may have the ``inalloca``
917 attribute, and that argument is guaranteed to be passed in memory.
919 An argument allocation may be used by a call at most once because
920 the call may deallocate it. The ``inalloca`` attribute cannot be
921 used in conjunction with other attributes that affect argument
922 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
923 ``inalloca`` attribute also disables LLVM's implicit lowering of
924 large aggregate return values, which means that frontend authors
925 must lower them with ``sret`` pointers.
927 When the call site is reached, the argument allocation must have
928 been the most recent stack allocation that is still live, or the
929 results are undefined. It is possible to allocate additional stack
930 space after an argument allocation and before its call site, but it
931 must be cleared off with :ref:`llvm.stackrestore
934 See :doc:`InAlloca` for more information on how to use this
938 This indicates that the pointer parameter specifies the address of a
939 structure that is the return value of the function in the source
940 program. This pointer must be guaranteed by the caller to be valid:
941 loads and stores to the structure may be assumed by the callee
942 not to trap and to be properly aligned. This may only be applied to
943 the first parameter. This is not a valid attribute for return
947 This indicates that the pointer value may be assumed by the optimizer to
948 have the specified alignment.
950 Note that this attribute has additional semantics when combined with the
956 This indicates that objects accessed via pointer values
957 :ref:`based <pointeraliasing>` on the argument or return value are not also
958 accessed, during the execution of the function, via pointer values not
959 *based* on the argument or return value. The attribute on a return value
960 also has additional semantics described below. The caller shares the
961 responsibility with the callee for ensuring that these requirements are met.
962 For further details, please see the discussion of the NoAlias response in
963 :ref:`alias analysis <Must, May, or No>`.
965 Note that this definition of ``noalias`` is intentionally similar
966 to the definition of ``restrict`` in C99 for function arguments.
968 For function return values, C99's ``restrict`` is not meaningful,
969 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
970 attribute on return values are stronger than the semantics of the attribute
971 when used on function arguments. On function return values, the ``noalias``
972 attribute indicates that the function acts like a system memory allocation
973 function, returning a pointer to allocated storage disjoint from the
974 storage for any other object accessible to the caller.
977 This indicates that the callee does not make any copies of the
978 pointer that outlive the callee itself. This is not a valid
979 attribute for return values.
984 This indicates that the pointer parameter can be excised using the
985 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
986 attribute for return values and can only be applied to one parameter.
989 This indicates that the function always returns the argument as its return
990 value. This is an optimization hint to the code generator when generating
991 the caller, allowing tail call optimization and omission of register saves
992 and restores in some cases; it is not checked or enforced when generating
993 the callee. The parameter and the function return type must be valid
994 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
995 valid attribute for return values and can only be applied to one parameter.
998 This indicates that the parameter or return pointer is not null. This
999 attribute may only be applied to pointer typed parameters. This is not
1000 checked or enforced by LLVM, the caller must ensure that the pointer
1001 passed in is non-null, or the callee must ensure that the returned pointer
1004 ``dereferenceable(<n>)``
1005 This indicates that the parameter or return pointer is dereferenceable. This
1006 attribute may only be applied to pointer typed parameters. A pointer that
1007 is dereferenceable can be loaded from speculatively without a risk of
1008 trapping. The number of bytes known to be dereferenceable must be provided
1009 in parentheses. It is legal for the number of bytes to be less than the
1010 size of the pointee type. The ``nonnull`` attribute does not imply
1011 dereferenceability (consider a pointer to one element past the end of an
1012 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1013 ``addrspace(0)`` (which is the default address space).
1017 Garbage Collector Strategy Names
1018 --------------------------------
1020 Each function may specify a garbage collector strategy name, which is simply a
1023 .. code-block:: llvm
1025 define void @f() gc "name" { ... }
1027 The supported values of *name* includes those :ref:`built in to LLVM
1028 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1029 strategy will cause the compiler to alter its output in order to support the
1030 named garbage collection algorithm. Note that LLVM itself does not contain a
1031 garbage collector, this functionality is restricted to generating machine code
1032 which can interoperate with a collector provided externally.
1039 Prefix data is data associated with a function which the code
1040 generator will emit immediately before the function's entrypoint.
1041 The purpose of this feature is to allow frontends to associate
1042 language-specific runtime metadata with specific functions and make it
1043 available through the function pointer while still allowing the
1044 function pointer to be called.
1046 To access the data for a given function, a program may bitcast the
1047 function pointer to a pointer to the constant's type and dereference
1048 index -1. This implies that the IR symbol points just past the end of
1049 the prefix data. For instance, take the example of a function annotated
1050 with a single ``i32``,
1052 .. code-block:: llvm
1054 define void @f() prefix i32 123 { ... }
1056 The prefix data can be referenced as,
1058 .. code-block:: llvm
1060 %0 = bitcast *void () @f to *i32
1061 %a = getelementptr inbounds *i32 %0, i32 -1
1064 Prefix data is laid out as if it were an initializer for a global variable
1065 of the prefix data's type. The function will be placed such that the
1066 beginning of the prefix data is aligned. This means that if the size
1067 of the prefix data is not a multiple of the alignment size, the
1068 function's entrypoint will not be aligned. If alignment of the
1069 function's entrypoint is desired, padding must be added to the prefix
1072 A function may have prefix data but no body. This has similar semantics
1073 to the ``available_externally`` linkage in that the data may be used by the
1074 optimizers but will not be emitted in the object file.
1081 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1082 be inserted prior to the function body. This can be used for enabling
1083 function hot-patching and instrumentation.
1085 To maintain the semantics of ordinary function calls, the prologue data must
1086 have a particular format. Specifically, it must begin with a sequence of
1087 bytes which decode to a sequence of machine instructions, valid for the
1088 module's target, which transfer control to the point immediately succeeding
1089 the prologue data, without performing any other visible action. This allows
1090 the inliner and other passes to reason about the semantics of the function
1091 definition without needing to reason about the prologue data. Obviously this
1092 makes the format of the prologue data highly target dependent.
1094 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1095 which encodes the ``nop`` instruction:
1097 .. code-block:: llvm
1099 define void @f() prologue i8 144 { ... }
1101 Generally prologue data can be formed by encoding a relative branch instruction
1102 which skips the metadata, as in this example of valid prologue data for the
1103 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1105 .. code-block:: llvm
1107 %0 = type <{ i8, i8, i8* }>
1109 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1111 A function may have prologue data but no body. This has similar semantics
1112 to the ``available_externally`` linkage in that the data may be used by the
1113 optimizers but will not be emitted in the object file.
1120 Attribute groups are groups of attributes that are referenced by objects within
1121 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1122 functions will use the same set of attributes. In the degenerative case of a
1123 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1124 group will capture the important command line flags used to build that file.
1126 An attribute group is a module-level object. To use an attribute group, an
1127 object references the attribute group's ID (e.g. ``#37``). An object may refer
1128 to more than one attribute group. In that situation, the attributes from the
1129 different groups are merged.
1131 Here is an example of attribute groups for a function that should always be
1132 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1134 .. code-block:: llvm
1136 ; Target-independent attributes:
1137 attributes #0 = { alwaysinline alignstack=4 }
1139 ; Target-dependent attributes:
1140 attributes #1 = { "no-sse" }
1142 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1143 define void @f() #0 #1 { ... }
1150 Function attributes are set to communicate additional information about
1151 a function. Function attributes are considered to be part of the
1152 function, not of the function type, so functions with different function
1153 attributes can have the same function type.
1155 Function attributes are simple keywords that follow the type specified.
1156 If multiple attributes are needed, they are space separated. For
1159 .. code-block:: llvm
1161 define void @f() noinline { ... }
1162 define void @f() alwaysinline { ... }
1163 define void @f() alwaysinline optsize { ... }
1164 define void @f() optsize { ... }
1167 This attribute indicates that, when emitting the prologue and
1168 epilogue, the backend should forcibly align the stack pointer.
1169 Specify the desired alignment, which must be a power of two, in
1172 This attribute indicates that the inliner should attempt to inline
1173 this function into callers whenever possible, ignoring any active
1174 inlining size threshold for this caller.
1176 This indicates that the callee function at a call site should be
1177 recognized as a built-in function, even though the function's declaration
1178 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1179 direct calls to functions that are declared with the ``nobuiltin``
1182 This attribute indicates that this function is rarely called. When
1183 computing edge weights, basic blocks post-dominated by a cold
1184 function call are also considered to be cold; and, thus, given low
1187 This attribute indicates that the source code contained a hint that
1188 inlining this function is desirable (such as the "inline" keyword in
1189 C/C++). It is just a hint; it imposes no requirements on the
1192 This attribute indicates that the function should be added to a
1193 jump-instruction table at code-generation time, and that all address-taken
1194 references to this function should be replaced with a reference to the
1195 appropriate jump-instruction-table function pointer. Note that this creates
1196 a new pointer for the original function, which means that code that depends
1197 on function-pointer identity can break. So, any function annotated with
1198 ``jumptable`` must also be ``unnamed_addr``.
1200 This attribute suggests that optimization passes and code generator
1201 passes make choices that keep the code size of this function as small
1202 as possible and perform optimizations that may sacrifice runtime
1203 performance in order to minimize the size of the generated code.
1205 This attribute disables prologue / epilogue emission for the
1206 function. This can have very system-specific consequences.
1208 This indicates that the callee function at a call site is not recognized as
1209 a built-in function. LLVM will retain the original call and not replace it
1210 with equivalent code based on the semantics of the built-in function, unless
1211 the call site uses the ``builtin`` attribute. This is valid at call sites
1212 and on function declarations and definitions.
1214 This attribute indicates that calls to the function cannot be
1215 duplicated. A call to a ``noduplicate`` function may be moved
1216 within its parent function, but may not be duplicated within
1217 its parent function.
1219 A function containing a ``noduplicate`` call may still
1220 be an inlining candidate, provided that the call is not
1221 duplicated by inlining. That implies that the function has
1222 internal linkage and only has one call site, so the original
1223 call is dead after inlining.
1225 This attributes disables implicit floating point instructions.
1227 This attribute indicates that the inliner should never inline this
1228 function in any situation. This attribute may not be used together
1229 with the ``alwaysinline`` attribute.
1231 This attribute suppresses lazy symbol binding for the function. This
1232 may make calls to the function faster, at the cost of extra program
1233 startup time if the function is not called during program startup.
1235 This attribute indicates that the code generator should not use a
1236 red zone, even if the target-specific ABI normally permits it.
1238 This function attribute indicates that the function never returns
1239 normally. This produces undefined behavior at runtime if the
1240 function ever does dynamically return.
1242 This function attribute indicates that the function never raises an
1243 exception. If the function does raise an exception, its runtime
1244 behavior is undefined. However, functions marked nounwind may still
1245 trap or generate asynchronous exceptions. Exception handling schemes
1246 that are recognized by LLVM to handle asynchronous exceptions, such
1247 as SEH, will still provide their implementation defined semantics.
1249 This function attribute indicates that the function is not optimized
1250 by any optimization or code generator passes with the
1251 exception of interprocedural optimization passes.
1252 This attribute cannot be used together with the ``alwaysinline``
1253 attribute; this attribute is also incompatible
1254 with the ``minsize`` attribute and the ``optsize`` attribute.
1256 This attribute requires the ``noinline`` attribute to be specified on
1257 the function as well, so the function is never inlined into any caller.
1258 Only functions with the ``alwaysinline`` attribute are valid
1259 candidates for inlining into the body of this function.
1261 This attribute suggests that optimization passes and code generator
1262 passes make choices that keep the code size of this function low,
1263 and otherwise do optimizations specifically to reduce code size as
1264 long as they do not significantly impact runtime performance.
1266 On a function, this attribute indicates that the function computes its
1267 result (or decides to unwind an exception) based strictly on its arguments,
1268 without dereferencing any pointer arguments or otherwise accessing
1269 any mutable state (e.g. memory, control registers, etc) visible to
1270 caller functions. It does not write through any pointer arguments
1271 (including ``byval`` arguments) and never changes any state visible
1272 to callers. This means that it cannot unwind exceptions by calling
1273 the ``C++`` exception throwing methods.
1275 On an argument, this attribute indicates that the function does not
1276 dereference that pointer argument, even though it may read or write the
1277 memory that the pointer points to if accessed through other pointers.
1279 On a function, this attribute indicates that the function does not write
1280 through any pointer arguments (including ``byval`` arguments) or otherwise
1281 modify any state (e.g. memory, control registers, etc) visible to
1282 caller functions. It may dereference pointer arguments and read
1283 state that may be set in the caller. A readonly function always
1284 returns the same value (or unwinds an exception identically) when
1285 called with the same set of arguments and global state. It cannot
1286 unwind an exception by calling the ``C++`` exception throwing
1289 On an argument, this attribute indicates that the function does not write
1290 through this pointer argument, even though it may write to the memory that
1291 the pointer points to.
1293 This attribute indicates that this function can return twice. The C
1294 ``setjmp`` is an example of such a function. The compiler disables
1295 some optimizations (like tail calls) in the caller of these
1297 ``sanitize_address``
1298 This attribute indicates that AddressSanitizer checks
1299 (dynamic address safety analysis) are enabled for this function.
1301 This attribute indicates that MemorySanitizer checks (dynamic detection
1302 of accesses to uninitialized memory) are enabled for this function.
1304 This attribute indicates that ThreadSanitizer checks
1305 (dynamic thread safety analysis) are enabled for this function.
1307 This attribute indicates that the function should emit a stack
1308 smashing protector. It is in the form of a "canary" --- a random value
1309 placed on the stack before the local variables that's checked upon
1310 return from the function to see if it has been overwritten. A
1311 heuristic is used to determine if a function needs stack protectors
1312 or not. The heuristic used will enable protectors for functions with:
1314 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1315 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1316 - Calls to alloca() with variable sizes or constant sizes greater than
1317 ``ssp-buffer-size``.
1319 Variables that are identified as requiring a protector will be arranged
1320 on the stack such that they are adjacent to the stack protector guard.
1322 If a function that has an ``ssp`` attribute is inlined into a
1323 function that doesn't have an ``ssp`` attribute, then the resulting
1324 function will have an ``ssp`` attribute.
1326 This attribute indicates that the function should *always* emit a
1327 stack smashing protector. This overrides the ``ssp`` function
1330 Variables that are identified as requiring a protector will be arranged
1331 on the stack such that they are adjacent to the stack protector guard.
1332 The specific layout rules are:
1334 #. Large arrays and structures containing large arrays
1335 (``>= ssp-buffer-size``) are closest to the stack protector.
1336 #. Small arrays and structures containing small arrays
1337 (``< ssp-buffer-size``) are 2nd closest to the protector.
1338 #. Variables that have had their address taken are 3rd closest to the
1341 If a function that has an ``sspreq`` attribute is inlined into a
1342 function that doesn't have an ``sspreq`` attribute or which has an
1343 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1344 an ``sspreq`` attribute.
1346 This attribute indicates that the function should emit a stack smashing
1347 protector. This attribute causes a strong heuristic to be used when
1348 determining if a function needs stack protectors. The strong heuristic
1349 will enable protectors for functions with:
1351 - Arrays of any size and type
1352 - Aggregates containing an array of any size and type.
1353 - Calls to alloca().
1354 - Local variables that have had their address taken.
1356 Variables that are identified as requiring a protector will be arranged
1357 on the stack such that they are adjacent to the stack protector guard.
1358 The specific layout rules are:
1360 #. Large arrays and structures containing large arrays
1361 (``>= ssp-buffer-size``) are closest to the stack protector.
1362 #. Small arrays and structures containing small arrays
1363 (``< ssp-buffer-size``) are 2nd closest to the protector.
1364 #. Variables that have had their address taken are 3rd closest to the
1367 This overrides the ``ssp`` function attribute.
1369 If a function that has an ``sspstrong`` attribute is inlined into a
1370 function that doesn't have an ``sspstrong`` attribute, then the
1371 resulting function will have an ``sspstrong`` attribute.
1373 This attribute indicates that the function will delegate to some other
1374 function with a tail call. The prototype of a thunk should not be used for
1375 optimization purposes. The caller is expected to cast the thunk prototype to
1376 match the thunk target prototype.
1378 This attribute indicates that the ABI being targeted requires that
1379 an unwind table entry be produce for this function even if we can
1380 show that no exceptions passes by it. This is normally the case for
1381 the ELF x86-64 abi, but it can be disabled for some compilation
1386 Module-Level Inline Assembly
1387 ----------------------------
1389 Modules may contain "module-level inline asm" blocks, which corresponds
1390 to the GCC "file scope inline asm" blocks. These blocks are internally
1391 concatenated by LLVM and treated as a single unit, but may be separated
1392 in the ``.ll`` file if desired. The syntax is very simple:
1394 .. code-block:: llvm
1396 module asm "inline asm code goes here"
1397 module asm "more can go here"
1399 The strings can contain any character by escaping non-printable
1400 characters. The escape sequence used is simply "\\xx" where "xx" is the
1401 two digit hex code for the number.
1403 The inline asm code is simply printed to the machine code .s file when
1404 assembly code is generated.
1406 .. _langref_datalayout:
1411 A module may specify a target specific data layout string that specifies
1412 how data is to be laid out in memory. The syntax for the data layout is
1415 .. code-block:: llvm
1417 target datalayout = "layout specification"
1419 The *layout specification* consists of a list of specifications
1420 separated by the minus sign character ('-'). Each specification starts
1421 with a letter and may include other information after the letter to
1422 define some aspect of the data layout. The specifications accepted are
1426 Specifies that the target lays out data in big-endian form. That is,
1427 the bits with the most significance have the lowest address
1430 Specifies that the target lays out data in little-endian form. That
1431 is, the bits with the least significance have the lowest address
1434 Specifies the natural alignment of the stack in bits. Alignment
1435 promotion of stack variables is limited to the natural stack
1436 alignment to avoid dynamic stack realignment. The stack alignment
1437 must be a multiple of 8-bits. If omitted, the natural stack
1438 alignment defaults to "unspecified", which does not prevent any
1439 alignment promotions.
1440 ``p[n]:<size>:<abi>:<pref>``
1441 This specifies the *size* of a pointer and its ``<abi>`` and
1442 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1443 bits. The address space, ``n`` is optional, and if not specified,
1444 denotes the default address space 0. The value of ``n`` must be
1445 in the range [1,2^23).
1446 ``i<size>:<abi>:<pref>``
1447 This specifies the alignment for an integer type of a given bit
1448 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1449 ``v<size>:<abi>:<pref>``
1450 This specifies the alignment for a vector type of a given bit
1452 ``f<size>:<abi>:<pref>``
1453 This specifies the alignment for a floating point type of a given bit
1454 ``<size>``. Only values of ``<size>`` that are supported by the target
1455 will work. 32 (float) and 64 (double) are supported on all targets; 80
1456 or 128 (different flavors of long double) are also supported on some
1459 This specifies the alignment for an object of aggregate type.
1461 If present, specifies that llvm names are mangled in the output. The
1464 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1465 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1466 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1467 symbols get a ``_`` prefix.
1468 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1469 functions also get a suffix based on the frame size.
1470 ``n<size1>:<size2>:<size3>...``
1471 This specifies a set of native integer widths for the target CPU in
1472 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1473 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1474 this set are considered to support most general arithmetic operations
1477 On every specification that takes a ``<abi>:<pref>``, specifying the
1478 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1479 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1481 When constructing the data layout for a given target, LLVM starts with a
1482 default set of specifications which are then (possibly) overridden by
1483 the specifications in the ``datalayout`` keyword. The default
1484 specifications are given in this list:
1486 - ``E`` - big endian
1487 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1488 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1489 same as the default address space.
1490 - ``S0`` - natural stack alignment is unspecified
1491 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1492 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1493 - ``i16:16:16`` - i16 is 16-bit aligned
1494 - ``i32:32:32`` - i32 is 32-bit aligned
1495 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1496 alignment of 64-bits
1497 - ``f16:16:16`` - half is 16-bit aligned
1498 - ``f32:32:32`` - float is 32-bit aligned
1499 - ``f64:64:64`` - double is 64-bit aligned
1500 - ``f128:128:128`` - quad is 128-bit aligned
1501 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1502 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1503 - ``a:0:64`` - aggregates are 64-bit aligned
1505 When LLVM is determining the alignment for a given type, it uses the
1508 #. If the type sought is an exact match for one of the specifications,
1509 that specification is used.
1510 #. If no match is found, and the type sought is an integer type, then
1511 the smallest integer type that is larger than the bitwidth of the
1512 sought type is used. If none of the specifications are larger than
1513 the bitwidth then the largest integer type is used. For example,
1514 given the default specifications above, the i7 type will use the
1515 alignment of i8 (next largest) while both i65 and i256 will use the
1516 alignment of i64 (largest specified).
1517 #. If no match is found, and the type sought is a vector type, then the
1518 largest vector type that is smaller than the sought vector type will
1519 be used as a fall back. This happens because <128 x double> can be
1520 implemented in terms of 64 <2 x double>, for example.
1522 The function of the data layout string may not be what you expect.
1523 Notably, this is not a specification from the frontend of what alignment
1524 the code generator should use.
1526 Instead, if specified, the target data layout is required to match what
1527 the ultimate *code generator* expects. This string is used by the
1528 mid-level optimizers to improve code, and this only works if it matches
1529 what the ultimate code generator uses. If you would like to generate IR
1530 that does not embed this target-specific detail into the IR, then you
1531 don't have to specify the string. This will disable some optimizations
1532 that require precise layout information, but this also prevents those
1533 optimizations from introducing target specificity into the IR.
1540 A module may specify a target triple string that describes the target
1541 host. The syntax for the target triple is simply:
1543 .. code-block:: llvm
1545 target triple = "x86_64-apple-macosx10.7.0"
1547 The *target triple* string consists of a series of identifiers delimited
1548 by the minus sign character ('-'). The canonical forms are:
1552 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1553 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1555 This information is passed along to the backend so that it generates
1556 code for the proper architecture. It's possible to override this on the
1557 command line with the ``-mtriple`` command line option.
1559 .. _pointeraliasing:
1561 Pointer Aliasing Rules
1562 ----------------------
1564 Any memory access must be done through a pointer value associated with
1565 an address range of the memory access, otherwise the behavior is
1566 undefined. Pointer values are associated with address ranges according
1567 to the following rules:
1569 - A pointer value is associated with the addresses associated with any
1570 value it is *based* on.
1571 - An address of a global variable is associated with the address range
1572 of the variable's storage.
1573 - The result value of an allocation instruction is associated with the
1574 address range of the allocated storage.
1575 - A null pointer in the default address-space is associated with no
1577 - An integer constant other than zero or a pointer value returned from
1578 a function not defined within LLVM may be associated with address
1579 ranges allocated through mechanisms other than those provided by
1580 LLVM. Such ranges shall not overlap with any ranges of addresses
1581 allocated by mechanisms provided by LLVM.
1583 A pointer value is *based* on another pointer value according to the
1586 - A pointer value formed from a ``getelementptr`` operation is *based*
1587 on the first operand of the ``getelementptr``.
1588 - The result value of a ``bitcast`` is *based* on the operand of the
1590 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1591 values that contribute (directly or indirectly) to the computation of
1592 the pointer's value.
1593 - The "*based* on" relationship is transitive.
1595 Note that this definition of *"based"* is intentionally similar to the
1596 definition of *"based"* in C99, though it is slightly weaker.
1598 LLVM IR does not associate types with memory. The result type of a
1599 ``load`` merely indicates the size and alignment of the memory from
1600 which to load, as well as the interpretation of the value. The first
1601 operand type of a ``store`` similarly only indicates the size and
1602 alignment of the store.
1604 Consequently, type-based alias analysis, aka TBAA, aka
1605 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1606 :ref:`Metadata <metadata>` may be used to encode additional information
1607 which specialized optimization passes may use to implement type-based
1612 Volatile Memory Accesses
1613 ------------------------
1615 Certain memory accesses, such as :ref:`load <i_load>`'s,
1616 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1617 marked ``volatile``. The optimizers must not change the number of
1618 volatile operations or change their order of execution relative to other
1619 volatile operations. The optimizers *may* change the order of volatile
1620 operations relative to non-volatile operations. This is not Java's
1621 "volatile" and has no cross-thread synchronization behavior.
1623 IR-level volatile loads and stores cannot safely be optimized into
1624 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1625 flagged volatile. Likewise, the backend should never split or merge
1626 target-legal volatile load/store instructions.
1628 .. admonition:: Rationale
1630 Platforms may rely on volatile loads and stores of natively supported
1631 data width to be executed as single instruction. For example, in C
1632 this holds for an l-value of volatile primitive type with native
1633 hardware support, but not necessarily for aggregate types. The
1634 frontend upholds these expectations, which are intentionally
1635 unspecified in the IR. The rules above ensure that IR transformation
1636 do not violate the frontend's contract with the language.
1640 Memory Model for Concurrent Operations
1641 --------------------------------------
1643 The LLVM IR does not define any way to start parallel threads of
1644 execution or to register signal handlers. Nonetheless, there are
1645 platform-specific ways to create them, and we define LLVM IR's behavior
1646 in their presence. This model is inspired by the C++0x memory model.
1648 For a more informal introduction to this model, see the :doc:`Atomics`.
1650 We define a *happens-before* partial order as the least partial order
1653 - Is a superset of single-thread program order, and
1654 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1655 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1656 techniques, like pthread locks, thread creation, thread joining,
1657 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1658 Constraints <ordering>`).
1660 Note that program order does not introduce *happens-before* edges
1661 between a thread and signals executing inside that thread.
1663 Every (defined) read operation (load instructions, memcpy, atomic
1664 loads/read-modify-writes, etc.) R reads a series of bytes written by
1665 (defined) write operations (store instructions, atomic
1666 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1667 section, initialized globals are considered to have a write of the
1668 initializer which is atomic and happens before any other read or write
1669 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1670 may see any write to the same byte, except:
1672 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1673 write\ :sub:`2` happens before R\ :sub:`byte`, then
1674 R\ :sub:`byte` does not see write\ :sub:`1`.
1675 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1676 R\ :sub:`byte` does not see write\ :sub:`3`.
1678 Given that definition, R\ :sub:`byte` is defined as follows:
1680 - If R is volatile, the result is target-dependent. (Volatile is
1681 supposed to give guarantees which can support ``sig_atomic_t`` in
1682 C/C++, and may be used for accesses to addresses that do not behave
1683 like normal memory. It does not generally provide cross-thread
1685 - Otherwise, if there is no write to the same byte that happens before
1686 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1687 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1688 R\ :sub:`byte` returns the value written by that write.
1689 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1690 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1691 Memory Ordering Constraints <ordering>` section for additional
1692 constraints on how the choice is made.
1693 - Otherwise R\ :sub:`byte` returns ``undef``.
1695 R returns the value composed of the series of bytes it read. This
1696 implies that some bytes within the value may be ``undef`` **without**
1697 the entire value being ``undef``. Note that this only defines the
1698 semantics of the operation; it doesn't mean that targets will emit more
1699 than one instruction to read the series of bytes.
1701 Note that in cases where none of the atomic intrinsics are used, this
1702 model places only one restriction on IR transformations on top of what
1703 is required for single-threaded execution: introducing a store to a byte
1704 which might not otherwise be stored is not allowed in general.
1705 (Specifically, in the case where another thread might write to and read
1706 from an address, introducing a store can change a load that may see
1707 exactly one write into a load that may see multiple writes.)
1711 Atomic Memory Ordering Constraints
1712 ----------------------------------
1714 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1715 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1716 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1717 ordering parameters that determine which other atomic instructions on
1718 the same address they *synchronize with*. These semantics are borrowed
1719 from Java and C++0x, but are somewhat more colloquial. If these
1720 descriptions aren't precise enough, check those specs (see spec
1721 references in the :doc:`atomics guide <Atomics>`).
1722 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1723 differently since they don't take an address. See that instruction's
1724 documentation for details.
1726 For a simpler introduction to the ordering constraints, see the
1730 The set of values that can be read is governed by the happens-before
1731 partial order. A value cannot be read unless some operation wrote
1732 it. This is intended to provide a guarantee strong enough to model
1733 Java's non-volatile shared variables. This ordering cannot be
1734 specified for read-modify-write operations; it is not strong enough
1735 to make them atomic in any interesting way.
1737 In addition to the guarantees of ``unordered``, there is a single
1738 total order for modifications by ``monotonic`` operations on each
1739 address. All modification orders must be compatible with the
1740 happens-before order. There is no guarantee that the modification
1741 orders can be combined to a global total order for the whole program
1742 (and this often will not be possible). The read in an atomic
1743 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1744 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1745 order immediately before the value it writes. If one atomic read
1746 happens before another atomic read of the same address, the later
1747 read must see the same value or a later value in the address's
1748 modification order. This disallows reordering of ``monotonic`` (or
1749 stronger) operations on the same address. If an address is written
1750 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1751 read that address repeatedly, the other threads must eventually see
1752 the write. This corresponds to the C++0x/C1x
1753 ``memory_order_relaxed``.
1755 In addition to the guarantees of ``monotonic``, a
1756 *synchronizes-with* edge may be formed with a ``release`` operation.
1757 This is intended to model C++'s ``memory_order_acquire``.
1759 In addition to the guarantees of ``monotonic``, if this operation
1760 writes a value which is subsequently read by an ``acquire``
1761 operation, it *synchronizes-with* that operation. (This isn't a
1762 complete description; see the C++0x definition of a release
1763 sequence.) This corresponds to the C++0x/C1x
1764 ``memory_order_release``.
1765 ``acq_rel`` (acquire+release)
1766 Acts as both an ``acquire`` and ``release`` operation on its
1767 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1768 ``seq_cst`` (sequentially consistent)
1769 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1770 operation that only reads, ``release`` for an operation that only
1771 writes), there is a global total order on all
1772 sequentially-consistent operations on all addresses, which is
1773 consistent with the *happens-before* partial order and with the
1774 modification orders of all the affected addresses. Each
1775 sequentially-consistent read sees the last preceding write to the
1776 same address in this global order. This corresponds to the C++0x/C1x
1777 ``memory_order_seq_cst`` and Java volatile.
1781 If an atomic operation is marked ``singlethread``, it only *synchronizes
1782 with* or participates in modification and seq\_cst total orderings with
1783 other operations running in the same thread (for example, in signal
1791 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1792 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1793 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1794 otherwise unsafe floating point operations
1797 No NaNs - Allow optimizations to assume the arguments and result are not
1798 NaN. Such optimizations are required to retain defined behavior over
1799 NaNs, but the value of the result is undefined.
1802 No Infs - Allow optimizations to assume the arguments and result are not
1803 +/-Inf. Such optimizations are required to retain defined behavior over
1804 +/-Inf, but the value of the result is undefined.
1807 No Signed Zeros - Allow optimizations to treat the sign of a zero
1808 argument or result as insignificant.
1811 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1812 argument rather than perform division.
1815 Fast - Allow algebraically equivalent transformations that may
1816 dramatically change results in floating point (e.g. reassociate). This
1817 flag implies all the others.
1821 Use-list Order Directives
1822 -------------------------
1824 Use-list directives encode the in-memory order of each use-list, allowing the
1825 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1826 indexes that are assigned to the referenced value's uses. The referenced
1827 value's use-list is immediately sorted by these indexes.
1829 Use-list directives may appear at function scope or global scope. They are not
1830 instructions, and have no effect on the semantics of the IR. When they're at
1831 function scope, they must appear after the terminator of the final basic block.
1833 If basic blocks have their address taken via ``blockaddress()`` expressions,
1834 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1841 uselistorder <ty> <value>, { <order-indexes> }
1842 uselistorder_bb @function, %block { <order-indexes> }
1848 define void @foo(i32 %arg1, i32 %arg2) {
1850 ; ... instructions ...
1852 ; ... instructions ...
1854 ; At function scope.
1855 uselistorder i32 %arg1, { 1, 0, 2 }
1856 uselistorder label %bb, { 1, 0 }
1860 uselistorder i32* @global, { 1, 2, 0 }
1861 uselistorder i32 7, { 1, 0 }
1862 uselistorder i32 (i32) @bar, { 1, 0 }
1863 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1870 The LLVM type system is one of the most important features of the
1871 intermediate representation. Being typed enables a number of
1872 optimizations to be performed on the intermediate representation
1873 directly, without having to do extra analyses on the side before the
1874 transformation. A strong type system makes it easier to read the
1875 generated code and enables novel analyses and transformations that are
1876 not feasible to perform on normal three address code representations.
1886 The void type does not represent any value and has no size.
1904 The function type can be thought of as a function signature. It consists of a
1905 return type and a list of formal parameter types. The return type of a function
1906 type is a void type or first class type --- except for :ref:`label <t_label>`
1907 and :ref:`metadata <t_metadata>` types.
1913 <returntype> (<parameter list>)
1915 ...where '``<parameter list>``' is a comma-separated list of type
1916 specifiers. Optionally, the parameter list may include a type ``...``, which
1917 indicates that the function takes a variable number of arguments. Variable
1918 argument functions can access their arguments with the :ref:`variable argument
1919 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1920 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1924 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1925 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1926 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1927 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1928 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1929 | ``i32 (i8*, ...)`` | A vararg function that takes at least one :ref:`pointer <t_pointer>` to ``i8`` (char in C), which returns an integer. This is the signature for ``printf`` in LLVM. |
1930 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1931 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1932 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1939 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1940 Values of these types are the only ones which can be produced by
1948 These are the types that are valid in registers from CodeGen's perspective.
1957 The integer type is a very simple type that simply specifies an
1958 arbitrary bit width for the integer type desired. Any bit width from 1
1959 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1967 The number of bits the integer will occupy is specified by the ``N``
1973 +----------------+------------------------------------------------+
1974 | ``i1`` | a single-bit integer. |
1975 +----------------+------------------------------------------------+
1976 | ``i32`` | a 32-bit integer. |
1977 +----------------+------------------------------------------------+
1978 | ``i1942652`` | a really big integer of over 1 million bits. |
1979 +----------------+------------------------------------------------+
1983 Floating Point Types
1984 """"""""""""""""""""
1993 - 16-bit floating point value
1996 - 32-bit floating point value
1999 - 64-bit floating point value
2002 - 128-bit floating point value (112-bit mantissa)
2005 - 80-bit floating point value (X87)
2008 - 128-bit floating point value (two 64-bits)
2015 The x86_mmx type represents a value held in an MMX register on an x86
2016 machine. The operations allowed on it are quite limited: parameters and
2017 return values, load and store, and bitcast. User-specified MMX
2018 instructions are represented as intrinsic or asm calls with arguments
2019 and/or results of this type. There are no arrays, vectors or constants
2036 The pointer type is used to specify memory locations. Pointers are
2037 commonly used to reference objects in memory.
2039 Pointer types may have an optional address space attribute defining the
2040 numbered address space where the pointed-to object resides. The default
2041 address space is number zero. The semantics of non-zero address spaces
2042 are target-specific.
2044 Note that LLVM does not permit pointers to void (``void*``) nor does it
2045 permit pointers to labels (``label*``). Use ``i8*`` instead.
2055 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2056 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2057 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2058 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2059 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2060 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2061 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2070 A vector type is a simple derived type that represents a vector of
2071 elements. Vector types are used when multiple primitive data are
2072 operated in parallel using a single instruction (SIMD). A vector type
2073 requires a size (number of elements) and an underlying primitive data
2074 type. Vector types are considered :ref:`first class <t_firstclass>`.
2080 < <# elements> x <elementtype> >
2082 The number of elements is a constant integer value larger than 0;
2083 elementtype may be any integer, floating point or pointer type. Vectors
2084 of size zero are not allowed.
2088 +-------------------+--------------------------------------------------+
2089 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2090 +-------------------+--------------------------------------------------+
2091 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2092 +-------------------+--------------------------------------------------+
2093 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2094 +-------------------+--------------------------------------------------+
2095 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2096 +-------------------+--------------------------------------------------+
2105 The label type represents code labels.
2120 The metadata type represents embedded metadata. No derived types may be
2121 created from metadata except for :ref:`function <t_function>` arguments.
2134 Aggregate Types are a subset of derived types that can contain multiple
2135 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2136 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2146 The array type is a very simple derived type that arranges elements
2147 sequentially in memory. The array type requires a size (number of
2148 elements) and an underlying data type.
2154 [<# elements> x <elementtype>]
2156 The number of elements is a constant integer value; ``elementtype`` may
2157 be any type with a size.
2161 +------------------+--------------------------------------+
2162 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2163 +------------------+--------------------------------------+
2164 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2165 +------------------+--------------------------------------+
2166 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2167 +------------------+--------------------------------------+
2169 Here are some examples of multidimensional arrays:
2171 +-----------------------------+----------------------------------------------------------+
2172 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2173 +-----------------------------+----------------------------------------------------------+
2174 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2175 +-----------------------------+----------------------------------------------------------+
2176 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2177 +-----------------------------+----------------------------------------------------------+
2179 There is no restriction on indexing beyond the end of the array implied
2180 by a static type (though there are restrictions on indexing beyond the
2181 bounds of an allocated object in some cases). This means that
2182 single-dimension 'variable sized array' addressing can be implemented in
2183 LLVM with a zero length array type. An implementation of 'pascal style
2184 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2194 The structure type is used to represent a collection of data members
2195 together in memory. The elements of a structure may be any type that has
2198 Structures in memory are accessed using '``load``' and '``store``' by
2199 getting a pointer to a field with the '``getelementptr``' instruction.
2200 Structures in registers are accessed using the '``extractvalue``' and
2201 '``insertvalue``' instructions.
2203 Structures may optionally be "packed" structures, which indicate that
2204 the alignment of the struct is one byte, and that there is no padding
2205 between the elements. In non-packed structs, padding between field types
2206 is inserted as defined by the DataLayout string in the module, which is
2207 required to match what the underlying code generator expects.
2209 Structures can either be "literal" or "identified". A literal structure
2210 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2211 identified types are always defined at the top level with a name.
2212 Literal types are uniqued by their contents and can never be recursive
2213 or opaque since there is no way to write one. Identified types can be
2214 recursive, can be opaqued, and are never uniqued.
2220 %T1 = type { <type list> } ; Identified normal struct type
2221 %T2 = type <{ <type list> }> ; Identified packed struct type
2225 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2226 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2227 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2228 | ``{ float, i32 (i32) * }`` | A pair, where the first element is a ``float`` and the second element is a :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32``, returning an ``i32``. |
2229 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2230 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2231 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2235 Opaque Structure Types
2236 """"""""""""""""""""""
2240 Opaque structure types are used to represent named structure types that
2241 do not have a body specified. This corresponds (for example) to the C
2242 notion of a forward declared structure.
2253 +--------------+-------------------+
2254 | ``opaque`` | An opaque type. |
2255 +--------------+-------------------+
2262 LLVM has several different basic types of constants. This section
2263 describes them all and their syntax.
2268 **Boolean constants**
2269 The two strings '``true``' and '``false``' are both valid constants
2271 **Integer constants**
2272 Standard integers (such as '4') are constants of the
2273 :ref:`integer <t_integer>` type. Negative numbers may be used with
2275 **Floating point constants**
2276 Floating point constants use standard decimal notation (e.g.
2277 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2278 hexadecimal notation (see below). The assembler requires the exact
2279 decimal value of a floating-point constant. For example, the
2280 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2281 decimal in binary. Floating point constants must have a :ref:`floating
2282 point <t_floating>` type.
2283 **Null pointer constants**
2284 The identifier '``null``' is recognized as a null pointer constant
2285 and must be of :ref:`pointer type <t_pointer>`.
2287 The one non-intuitive notation for constants is the hexadecimal form of
2288 floating point constants. For example, the form
2289 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2290 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2291 constants are required (and the only time that they are generated by the
2292 disassembler) is when a floating point constant must be emitted but it
2293 cannot be represented as a decimal floating point number in a reasonable
2294 number of digits. For example, NaN's, infinities, and other special
2295 values are represented in their IEEE hexadecimal format so that assembly
2296 and disassembly do not cause any bits to change in the constants.
2298 When using the hexadecimal form, constants of types half, float, and
2299 double are represented using the 16-digit form shown above (which
2300 matches the IEEE754 representation for double); half and float values
2301 must, however, be exactly representable as IEEE 754 half and single
2302 precision, respectively. Hexadecimal format is always used for long
2303 double, and there are three forms of long double. The 80-bit format used
2304 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2305 128-bit format used by PowerPC (two adjacent doubles) is represented by
2306 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2307 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2308 will only work if they match the long double format on your target.
2309 The IEEE 16-bit format (half precision) is represented by ``0xH``
2310 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2311 (sign bit at the left).
2313 There are no constants of type x86_mmx.
2315 .. _complexconstants:
2320 Complex constants are a (potentially recursive) combination of simple
2321 constants and smaller complex constants.
2323 **Structure constants**
2324 Structure constants are represented with notation similar to
2325 structure type definitions (a comma separated list of elements,
2326 surrounded by braces (``{}``)). For example:
2327 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2328 "``@G = external global i32``". Structure constants must have
2329 :ref:`structure type <t_struct>`, and the number and types of elements
2330 must match those specified by the type.
2332 Array constants are represented with notation similar to array type
2333 definitions (a comma separated list of elements, surrounded by
2334 square brackets (``[]``)). For example:
2335 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2336 :ref:`array type <t_array>`, and the number and types of elements must
2337 match those specified by the type. As a special case, character array
2338 constants may also be represented as a double-quoted string using the ``c``
2339 prefix. For example: "``c"Hello World\0A\00"``".
2340 **Vector constants**
2341 Vector constants are represented with notation similar to vector
2342 type definitions (a comma separated list of elements, surrounded by
2343 less-than/greater-than's (``<>``)). For example:
2344 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2345 must have :ref:`vector type <t_vector>`, and the number and types of
2346 elements must match those specified by the type.
2347 **Zero initialization**
2348 The string '``zeroinitializer``' can be used to zero initialize a
2349 value to zero of *any* type, including scalar and
2350 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2351 having to print large zero initializers (e.g. for large arrays) and
2352 is always exactly equivalent to using explicit zero initializers.
2354 A metadata node is a constant tuple without types. For example:
2355 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2356 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2357 Unlike other typed constants that are meant to be interpreted as part of
2358 the instruction stream, metadata is a place to attach additional
2359 information such as debug info.
2361 Global Variable and Function Addresses
2362 --------------------------------------
2364 The addresses of :ref:`global variables <globalvars>` and
2365 :ref:`functions <functionstructure>` are always implicitly valid
2366 (link-time) constants. These constants are explicitly referenced when
2367 the :ref:`identifier for the global <identifiers>` is used and always have
2368 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2371 .. code-block:: llvm
2375 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2382 The string '``undef``' can be used anywhere a constant is expected, and
2383 indicates that the user of the value may receive an unspecified
2384 bit-pattern. Undefined values may be of any type (other than '``label``'
2385 or '``void``') and be used anywhere a constant is permitted.
2387 Undefined values are useful because they indicate to the compiler that
2388 the program is well defined no matter what value is used. This gives the
2389 compiler more freedom to optimize. Here are some examples of
2390 (potentially surprising) transformations that are valid (in pseudo IR):
2392 .. code-block:: llvm
2402 This is safe because all of the output bits are affected by the undef
2403 bits. Any output bit can have a zero or one depending on the input bits.
2405 .. code-block:: llvm
2416 These logical operations have bits that are not always affected by the
2417 input. For example, if ``%X`` has a zero bit, then the output of the
2418 '``and``' operation will always be a zero for that bit, no matter what
2419 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2420 optimize or assume that the result of the '``and``' is '``undef``'.
2421 However, it is safe to assume that all bits of the '``undef``' could be
2422 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2423 all the bits of the '``undef``' operand to the '``or``' could be set,
2424 allowing the '``or``' to be folded to -1.
2426 .. code-block:: llvm
2428 %A = select undef, %X, %Y
2429 %B = select undef, 42, %Y
2430 %C = select %X, %Y, undef
2440 This set of examples shows that undefined '``select``' (and conditional
2441 branch) conditions can go *either way*, but they have to come from one
2442 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2443 both known to have a clear low bit, then ``%A`` would have to have a
2444 cleared low bit. However, in the ``%C`` example, the optimizer is
2445 allowed to assume that the '``undef``' operand could be the same as
2446 ``%Y``, allowing the whole '``select``' to be eliminated.
2448 .. code-block:: llvm
2450 %A = xor undef, undef
2467 This example points out that two '``undef``' operands are not
2468 necessarily the same. This can be surprising to people (and also matches
2469 C semantics) where they assume that "``X^X``" is always zero, even if
2470 ``X`` is undefined. This isn't true for a number of reasons, but the
2471 short answer is that an '``undef``' "variable" can arbitrarily change
2472 its value over its "live range". This is true because the variable
2473 doesn't actually *have a live range*. Instead, the value is logically
2474 read from arbitrary registers that happen to be around when needed, so
2475 the value is not necessarily consistent over time. In fact, ``%A`` and
2476 ``%C`` need to have the same semantics or the core LLVM "replace all
2477 uses with" concept would not hold.
2479 .. code-block:: llvm
2487 These examples show the crucial difference between an *undefined value*
2488 and *undefined behavior*. An undefined value (like '``undef``') is
2489 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2490 operation can be constant folded to '``undef``', because the '``undef``'
2491 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2492 However, in the second example, we can make a more aggressive
2493 assumption: because the ``undef`` is allowed to be an arbitrary value,
2494 we are allowed to assume that it could be zero. Since a divide by zero
2495 has *undefined behavior*, we are allowed to assume that the operation
2496 does not execute at all. This allows us to delete the divide and all
2497 code after it. Because the undefined operation "can't happen", the
2498 optimizer can assume that it occurs in dead code.
2500 .. code-block:: llvm
2502 a: store undef -> %X
2503 b: store %X -> undef
2508 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2509 value can be assumed to not have any effect; we can assume that the
2510 value is overwritten with bits that happen to match what was already
2511 there. However, a store *to* an undefined location could clobber
2512 arbitrary memory, therefore, it has undefined behavior.
2519 Poison values are similar to :ref:`undef values <undefvalues>`, however
2520 they also represent the fact that an instruction or constant expression
2521 that cannot evoke side effects has nevertheless detected a condition
2522 that results in undefined behavior.
2524 There is currently no way of representing a poison value in the IR; they
2525 only exist when produced by operations such as :ref:`add <i_add>` with
2528 Poison value behavior is defined in terms of value *dependence*:
2530 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2531 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2532 their dynamic predecessor basic block.
2533 - Function arguments depend on the corresponding actual argument values
2534 in the dynamic callers of their functions.
2535 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2536 instructions that dynamically transfer control back to them.
2537 - :ref:`Invoke <i_invoke>` instructions depend on the
2538 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2539 call instructions that dynamically transfer control back to them.
2540 - Non-volatile loads and stores depend on the most recent stores to all
2541 of the referenced memory addresses, following the order in the IR
2542 (including loads and stores implied by intrinsics such as
2543 :ref:`@llvm.memcpy <int_memcpy>`.)
2544 - An instruction with externally visible side effects depends on the
2545 most recent preceding instruction with externally visible side
2546 effects, following the order in the IR. (This includes :ref:`volatile
2547 operations <volatile>`.)
2548 - An instruction *control-depends* on a :ref:`terminator
2549 instruction <terminators>` if the terminator instruction has
2550 multiple successors and the instruction is always executed when
2551 control transfers to one of the successors, and may not be executed
2552 when control is transferred to another.
2553 - Additionally, an instruction also *control-depends* on a terminator
2554 instruction if the set of instructions it otherwise depends on would
2555 be different if the terminator had transferred control to a different
2557 - Dependence is transitive.
2559 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2560 with the additional effect that any instruction that has a *dependence*
2561 on a poison value has undefined behavior.
2563 Here are some examples:
2565 .. code-block:: llvm
2568 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2569 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2570 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2571 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2573 store i32 %poison, i32* @g ; Poison value stored to memory.
2574 %poison2 = load i32* @g ; Poison value loaded back from memory.
2576 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2578 %narrowaddr = bitcast i32* @g to i16*
2579 %wideaddr = bitcast i32* @g to i64*
2580 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2581 %poison4 = load i64* %wideaddr ; Returns a poison value.
2583 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2584 br i1 %cmp, label %true, label %end ; Branch to either destination.
2587 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2588 ; it has undefined behavior.
2592 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2593 ; Both edges into this PHI are
2594 ; control-dependent on %cmp, so this
2595 ; always results in a poison value.
2597 store volatile i32 0, i32* @g ; This would depend on the store in %true
2598 ; if %cmp is true, or the store in %entry
2599 ; otherwise, so this is undefined behavior.
2601 br i1 %cmp, label %second_true, label %second_end
2602 ; The same branch again, but this time the
2603 ; true block doesn't have side effects.
2610 store volatile i32 0, i32* @g ; This time, the instruction always depends
2611 ; on the store in %end. Also, it is
2612 ; control-equivalent to %end, so this is
2613 ; well-defined (ignoring earlier undefined
2614 ; behavior in this example).
2618 Addresses of Basic Blocks
2619 -------------------------
2621 ``blockaddress(@function, %block)``
2623 The '``blockaddress``' constant computes the address of the specified
2624 basic block in the specified function, and always has an ``i8*`` type.
2625 Taking the address of the entry block is illegal.
2627 This value only has defined behavior when used as an operand to the
2628 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2629 against null. Pointer equality tests between labels addresses results in
2630 undefined behavior --- though, again, comparison against null is ok, and
2631 no label is equal to the null pointer. This may be passed around as an
2632 opaque pointer sized value as long as the bits are not inspected. This
2633 allows ``ptrtoint`` and arithmetic to be performed on these values so
2634 long as the original value is reconstituted before the ``indirectbr``
2637 Finally, some targets may provide defined semantics when using the value
2638 as the operand to an inline assembly, but that is target specific.
2642 Constant Expressions
2643 --------------------
2645 Constant expressions are used to allow expressions involving other
2646 constants to be used as constants. Constant expressions may be of any
2647 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2648 that does not have side effects (e.g. load and call are not supported).
2649 The following is the syntax for constant expressions:
2651 ``trunc (CST to TYPE)``
2652 Truncate a constant to another type. The bit size of CST must be
2653 larger than the bit size of TYPE. Both types must be integers.
2654 ``zext (CST to TYPE)``
2655 Zero extend a constant to another type. The bit size of CST must be
2656 smaller than the bit size of TYPE. Both types must be integers.
2657 ``sext (CST to TYPE)``
2658 Sign extend a constant to another type. The bit size of CST must be
2659 smaller than the bit size of TYPE. Both types must be integers.
2660 ``fptrunc (CST to TYPE)``
2661 Truncate a floating point constant to another floating point type.
2662 The size of CST must be larger than the size of TYPE. Both types
2663 must be floating point.
2664 ``fpext (CST to TYPE)``
2665 Floating point extend a constant to another type. The size of CST
2666 must be smaller or equal to the size of TYPE. Both types must be
2668 ``fptoui (CST to TYPE)``
2669 Convert a floating point constant to the corresponding unsigned
2670 integer constant. TYPE must be a scalar or vector integer type. CST
2671 must be of scalar or vector floating point type. Both CST and TYPE
2672 must be scalars, or vectors of the same number of elements. If the
2673 value won't fit in the integer type, the results are undefined.
2674 ``fptosi (CST to TYPE)``
2675 Convert a floating point constant to the corresponding signed
2676 integer constant. TYPE must be a scalar or vector integer type. CST
2677 must be of scalar or vector floating point type. Both CST and TYPE
2678 must be scalars, or vectors of the same number of elements. If the
2679 value won't fit in the integer type, the results are undefined.
2680 ``uitofp (CST to TYPE)``
2681 Convert an unsigned integer constant to the corresponding floating
2682 point constant. TYPE must be a scalar or vector floating point type.
2683 CST must be of scalar or vector integer type. Both CST and TYPE must
2684 be scalars, or vectors of the same number of elements. If the value
2685 won't fit in the floating point type, the results are undefined.
2686 ``sitofp (CST to TYPE)``
2687 Convert a signed integer constant to the corresponding floating
2688 point constant. TYPE must be a scalar or vector floating point type.
2689 CST must be of scalar or vector integer type. Both CST and TYPE must
2690 be scalars, or vectors of the same number of elements. If the value
2691 won't fit in the floating point type, the results are undefined.
2692 ``ptrtoint (CST to TYPE)``
2693 Convert a pointer typed constant to the corresponding integer
2694 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2695 pointer type. The ``CST`` value is zero extended, truncated, or
2696 unchanged to make it fit in ``TYPE``.
2697 ``inttoptr (CST to TYPE)``
2698 Convert an integer constant to a pointer constant. TYPE must be a
2699 pointer type. CST must be of integer type. The CST value is zero
2700 extended, truncated, or unchanged to make it fit in a pointer size.
2701 This one is *really* dangerous!
2702 ``bitcast (CST to TYPE)``
2703 Convert a constant, CST, to another TYPE. The constraints of the
2704 operands are the same as those for the :ref:`bitcast
2705 instruction <i_bitcast>`.
2706 ``addrspacecast (CST to TYPE)``
2707 Convert a constant pointer or constant vector of pointer, CST, to another
2708 TYPE in a different address space. The constraints of the operands are the
2709 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2710 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2711 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2712 constants. As with the :ref:`getelementptr <i_getelementptr>`
2713 instruction, the index list may have zero or more indexes, which are
2714 required to make sense for the type of "CSTPTR".
2715 ``select (COND, VAL1, VAL2)``
2716 Perform the :ref:`select operation <i_select>` on constants.
2717 ``icmp COND (VAL1, VAL2)``
2718 Performs the :ref:`icmp operation <i_icmp>` on constants.
2719 ``fcmp COND (VAL1, VAL2)``
2720 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2721 ``extractelement (VAL, IDX)``
2722 Perform the :ref:`extractelement operation <i_extractelement>` on
2724 ``insertelement (VAL, ELT, IDX)``
2725 Perform the :ref:`insertelement operation <i_insertelement>` on
2727 ``shufflevector (VEC1, VEC2, IDXMASK)``
2728 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2730 ``extractvalue (VAL, IDX0, IDX1, ...)``
2731 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2732 constants. The index list is interpreted in a similar manner as
2733 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2734 least one index value must be specified.
2735 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2736 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2737 The index list is interpreted in a similar manner as indices in a
2738 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2739 value must be specified.
2740 ``OPCODE (LHS, RHS)``
2741 Perform the specified operation of the LHS and RHS constants. OPCODE
2742 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2743 binary <bitwiseops>` operations. The constraints on operands are
2744 the same as those for the corresponding instruction (e.g. no bitwise
2745 operations on floating point values are allowed).
2752 Inline Assembler Expressions
2753 ----------------------------
2755 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2756 Inline Assembly <moduleasm>`) through the use of a special value. This
2757 value represents the inline assembler as a string (containing the
2758 instructions to emit), a list of operand constraints (stored as a
2759 string), a flag that indicates whether or not the inline asm expression
2760 has side effects, and a flag indicating whether the function containing
2761 the asm needs to align its stack conservatively. An example inline
2762 assembler expression is:
2764 .. code-block:: llvm
2766 i32 (i32) asm "bswap $0", "=r,r"
2768 Inline assembler expressions may **only** be used as the callee operand
2769 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2770 Thus, typically we have:
2772 .. code-block:: llvm
2774 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2776 Inline asms with side effects not visible in the constraint list must be
2777 marked as having side effects. This is done through the use of the
2778 '``sideeffect``' keyword, like so:
2780 .. code-block:: llvm
2782 call void asm sideeffect "eieio", ""()
2784 In some cases inline asms will contain code that will not work unless
2785 the stack is aligned in some way, such as calls or SSE instructions on
2786 x86, yet will not contain code that does that alignment within the asm.
2787 The compiler should make conservative assumptions about what the asm
2788 might contain and should generate its usual stack alignment code in the
2789 prologue if the '``alignstack``' keyword is present:
2791 .. code-block:: llvm
2793 call void asm alignstack "eieio", ""()
2795 Inline asms also support using non-standard assembly dialects. The
2796 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2797 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2798 the only supported dialects. An example is:
2800 .. code-block:: llvm
2802 call void asm inteldialect "eieio", ""()
2804 If multiple keywords appear the '``sideeffect``' keyword must come
2805 first, the '``alignstack``' keyword second and the '``inteldialect``'
2811 The call instructions that wrap inline asm nodes may have a
2812 "``!srcloc``" MDNode attached to it that contains a list of constant
2813 integers. If present, the code generator will use the integer as the
2814 location cookie value when report errors through the ``LLVMContext``
2815 error reporting mechanisms. This allows a front-end to correlate backend
2816 errors that occur with inline asm back to the source code that produced
2819 .. code-block:: llvm
2821 call void asm sideeffect "something bad", ""(), !srcloc !42
2823 !42 = !{ i32 1234567 }
2825 It is up to the front-end to make sense of the magic numbers it places
2826 in the IR. If the MDNode contains multiple constants, the code generator
2827 will use the one that corresponds to the line of the asm that the error
2835 LLVM IR allows metadata to be attached to instructions in the program
2836 that can convey extra information about the code to the optimizers and
2837 code generator. One example application of metadata is source-level
2838 debug information. There are two metadata primitives: strings and nodes.
2840 Metadata does not have a type, and is not a value. If referenced from a
2841 ``call`` instruction, it uses the ``metadata`` type.
2843 All metadata are identified in syntax by a exclamation point ('``!``').
2845 .. _metadata-string:
2847 Metadata Nodes and Metadata Strings
2848 -----------------------------------
2850 A metadata string is a string surrounded by double quotes. It can
2851 contain any character by escaping non-printable characters with
2852 "``\xx``" where "``xx``" is the two digit hex code. For example:
2855 Metadata nodes are represented with notation similar to structure
2856 constants (a comma separated list of elements, surrounded by braces and
2857 preceded by an exclamation point). Metadata nodes can have any values as
2858 their operand. For example:
2860 .. code-block:: llvm
2862 !{ !"test\00", i32 10}
2864 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2866 .. code-block:: llvm
2868 !0 = distinct !{!"test\00", i32 10}
2870 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2871 content. They can also occur when transformations cause uniquing collisions
2872 when metadata operands change.
2874 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2875 metadata nodes, which can be looked up in the module symbol table. For
2878 .. code-block:: llvm
2882 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2883 function is using two metadata arguments:
2885 .. code-block:: llvm
2887 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2889 Metadata can be attached with an instruction. Here metadata ``!21`` is
2890 attached to the ``add`` instruction using the ``!dbg`` identifier:
2892 .. code-block:: llvm
2894 %indvar.next = add i64 %indvar, 1, !dbg !21
2896 More information about specific metadata nodes recognized by the
2897 optimizers and code generator is found below.
2899 Specialized Metadata Nodes
2900 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2902 Specialized metadata nodes are custom data structures in metadata (as opposed
2903 to generic tuples). Their fields are labelled, and can be specified in any
2906 These aren't inherently debug info centric, but currently all the specialized
2907 metadata nodes are related to debug info.
2912 ``MDCompileUnit`` nodes represent a compile unit. The ``enums:``,
2913 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2914 tuples containing the debug info to be emitted along with the compile unit,
2915 regardless of code optimizations (some nodes are only emitted if there are
2916 references to them from instructions).
2918 .. code-block:: llvm
2920 !0 = !MDCompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2921 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2922 splitDebugFilename: "abc.debug", emissionKind: 1,
2923 enums: !2, retainedTypes: !3, subprograms: !4,
2924 globals: !5, imports: !6)
2929 ``MDFile`` nodes represent files. The ``filename:`` can include slashes.
2931 .. code-block:: llvm
2933 !0 = !MDFile(filename: "path/to/file", directory: "/path/to/dir")
2940 ``MDBasicType`` nodes represent primitive types. ``tag:`` defaults to
2941 ``DW_TAG_base_type``.
2943 .. code-block:: llvm
2945 !0 = !MDBasicType(name: "unsigned char", size: 8, align: 8,
2946 encoding: DW_ATE_unsigned_char)
2947 !1 = !MDBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
2949 .. _MDSubroutineType:
2954 ``MDSubroutineType`` nodes represent subroutine types. Their ``types:`` field
2955 refers to a tuple; the first operand is the return type, while the rest are the
2956 types of the formal arguments in order. If the first operand is ``null``, that
2957 represents a function with no return value (such as ``void foo() {}`` in C++).
2959 .. code-block:: llvm
2961 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
2962 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
2963 !2 = !MDSubroutineType(types: !{null, !0, !1}) ; void (int, char)
2968 ``MDDerivedType`` nodes represent types derived from other types, such as
2971 .. code-block:: llvm
2973 !0 = !MDBasicType(name: "unsigned char", size: 8, align: 8,
2974 encoding: DW_ATE_unsigned_char)
2975 !1 = !MDDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
2978 .. _MDCompositeType:
2983 ``MDCompositeType`` nodes represent types composed of other types, like
2984 structures and unions. ``elements:`` points to a tuple of the composed types.
2986 If the source language supports ODR, the ``identifier:`` field gives the unique
2987 identifier used for type merging between modules. When specified, other types
2988 can refer to composite types indirectly via a :ref:`metadata string
2989 <metadata-string>` that matches their identifier.
2991 .. code-block:: llvm
2993 !0 = !MDEnumerator(name: "SixKind", value: 7)
2994 !1 = !MDEnumerator(name: "SevenKind", value: 7)
2995 !2 = !MDEnumerator(name: "NegEightKind", value: -8)
2996 !3 = !MDCompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
2997 line: 2, size: 32, align: 32, identifier: "_M4Enum",
2998 elements: !{!0, !1, !2})
3003 ``MDSubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3004 :ref:`MDCompositeType`. ``count: -1`` indicates an empty array.
3006 .. code-block:: llvm
3008 !0 = !MDSubrange(count: 5, lowerBound: 0) ; array counting from 0
3009 !1 = !MDSubrange(count: 5, lowerBound: 1) ; array counting from 1
3010 !2 = !MDSubrange(count: -1) ; empty array.
3015 ``MDEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3016 variants of :ref:`MDCompositeType`.
3018 .. code-block:: llvm
3020 !0 = !MDEnumerator(name: "SixKind", value: 7)
3021 !1 = !MDEnumerator(name: "SevenKind", value: 7)
3022 !2 = !MDEnumerator(name: "NegEightKind", value: -8)
3024 MDTemplateTypeParameter
3025 """""""""""""""""""""""
3027 ``MDTemplateTypeParameter`` nodes represent type parameters to generic source
3028 language constructs. They are used (optionally) in :ref:`MDCompositeType` and
3029 :ref:`MDSubprogram` ``templateParams:`` fields.
3031 .. code-block:: llvm
3033 !0 = !MDTemplateTypeParameter(name: "Ty", type: !1)
3035 MDTemplateValueParameter
3036 """"""""""""""""""""""""
3038 ``MDTemplateValueParameter`` nodes represent value parameters to generic source
3039 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3040 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3041 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3042 :ref:`MDCompositeType` and :ref:`MDSubprogram` ``templateParams:`` fields.
3044 .. code-block:: llvm
3046 !0 = !MDTemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3051 ``MDNamespace`` nodes represent namespaces in the source language.
3053 .. code-block:: llvm
3055 !0 = !MDNamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3060 ``MDGlobalVariable`` nodes represent global variables in the source language.
3062 .. code-block:: llvm
3064 !0 = !MDGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3065 file: !2, line: 7, type: !3, isLocal: true,
3066 isDefinition: false, variable: i32* @foo,
3074 ``MDSubprogram`` nodes represent functions from the source language. The
3075 ``variables:`` field points at :ref:`variables <MDLocalVariable>` that must be
3076 retained, even if their IR counterparts are optimized out of the IR. The
3077 ``type:`` field must point at an :ref:`MDSubroutineType`.
3079 .. code-block:: llvm
3081 !0 = !MDSubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3082 file: !2, line: 7, type: !3, isLocal: true,
3083 isDefinition: false, scopeLine: 8, containingType: !4,
3084 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3085 flags: DIFlagPrototyped, isOptimized: true,
3086 function: void ()* @_Z3foov,
3087 templateParams: !5, declaration: !6, variables: !7)
3094 ``MDLexicalBlock`` nodes represent lexical blocks in the source language (a
3097 .. code-block:: llvm
3099 !0 = !MDLexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3101 .. _MDLexicalBlockFile:
3106 ``MDLexicalBlockFile`` nodes are used to discriminate between sections of a
3107 :ref:`lexical block <MDLexicalBlock>`. The ``file:`` field can be changed to
3108 indicate textual inclusion, or the ``discriminator:`` field can be used to
3109 discriminate between control flow within a single block in the source language.
3111 .. code-block:: llvm
3113 !0 = !MDLexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3114 !1 = !MDLexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3115 !2 = !MDLexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3120 ``MDLocation`` nodes represent source debug locations. The ``scope:`` field is
3121 mandatory, and points at an :ref:`MDLexicalBlockFile`, an
3122 :ref:`MDLexicalBlock`, or an :ref:`MDSubprogram`.
3124 .. code-block:: llvm
3126 !0 = !MDLocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3128 .. _MDLocalVariable:
3133 ``MDLocalVariable`` nodes represent local variables in the source language.
3134 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3135 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3136 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3137 specifies the argument position, and this variable will be included in the
3138 ``variables:`` field of its :ref:`MDSubprogram`.
3140 If set, the ``inlinedAt:`` field points at an :ref:`MDLocation`, and the
3141 variable represents an inlined version of a variable (with all other fields
3142 duplicated from the non-inlined version).
3144 .. code-block:: llvm
3146 !0 = !MDLocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3147 scope: !3, file: !2, line: 7, type: !3,
3148 flags: DIFlagArtificial, inlinedAt: !4)
3149 !1 = !MDLocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3150 scope: !4, file: !2, line: 7, type: !3,
3152 !1 = !MDLocalVariable(tag: DW_TAG_auto_variable, name: "y",
3153 scope: !5, file: !2, line: 7, type: !3,
3159 ``MDExpression`` nodes represent DWARF expression sequences. They are used in
3160 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3161 describe how the referenced LLVM variable relates to the source language
3164 The current supported vocabulary is limited:
3166 - ``DW_OP_deref`` dereferences the working expression.
3167 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3168 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3169 here, respectively) of the variable piece from the working expression.
3171 .. code-block:: llvm
3173 !0 = !MDExpression(DW_OP_deref)
3174 !1 = !MDExpression(DW_OP_plus, 3)
3175 !2 = !MDExpression(DW_OP_bit_piece, 3, 7)
3176 !3 = !MDExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3181 ``MDObjCProperty`` nodes represent Objective-C property nodes.
3183 .. code-block:: llvm
3185 !3 = !MDObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3186 getter: "getFoo", attributes: 7, type: !2)
3191 ``MDImportedEntity`` nodes represent entities (such as modules) imported into a
3194 .. code-block:: llvm
3196 !2 = !MDImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3197 entity: !1, line: 7)
3202 In LLVM IR, memory does not have types, so LLVM's own type system is not
3203 suitable for doing TBAA. Instead, metadata is added to the IR to
3204 describe a type system of a higher level language. This can be used to
3205 implement typical C/C++ TBAA, but it can also be used to implement
3206 custom alias analysis behavior for other languages.
3208 The current metadata format is very simple. TBAA metadata nodes have up
3209 to three fields, e.g.:
3211 .. code-block:: llvm
3213 !0 = !{ !"an example type tree" }
3214 !1 = !{ !"int", !0 }
3215 !2 = !{ !"float", !0 }
3216 !3 = !{ !"const float", !2, i64 1 }
3218 The first field is an identity field. It can be any value, usually a
3219 metadata string, which uniquely identifies the type. The most important
3220 name in the tree is the name of the root node. Two trees with different
3221 root node names are entirely disjoint, even if they have leaves with
3224 The second field identifies the type's parent node in the tree, or is
3225 null or omitted for a root node. A type is considered to alias all of
3226 its descendants and all of its ancestors in the tree. Also, a type is
3227 considered to alias all types in other trees, so that bitcode produced
3228 from multiple front-ends is handled conservatively.
3230 If the third field is present, it's an integer which if equal to 1
3231 indicates that the type is "constant" (meaning
3232 ``pointsToConstantMemory`` should return true; see `other useful
3233 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3235 '``tbaa.struct``' Metadata
3236 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3238 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3239 aggregate assignment operations in C and similar languages, however it
3240 is defined to copy a contiguous region of memory, which is more than
3241 strictly necessary for aggregate types which contain holes due to
3242 padding. Also, it doesn't contain any TBAA information about the fields
3245 ``!tbaa.struct`` metadata can describe which memory subregions in a
3246 memcpy are padding and what the TBAA tags of the struct are.
3248 The current metadata format is very simple. ``!tbaa.struct`` metadata
3249 nodes are a list of operands which are in conceptual groups of three.
3250 For each group of three, the first operand gives the byte offset of a
3251 field in bytes, the second gives its size in bytes, and the third gives
3254 .. code-block:: llvm
3256 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3258 This describes a struct with two fields. The first is at offset 0 bytes
3259 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3260 and has size 4 bytes and has tbaa tag !2.
3262 Note that the fields need not be contiguous. In this example, there is a
3263 4 byte gap between the two fields. This gap represents padding which
3264 does not carry useful data and need not be preserved.
3266 '``noalias``' and '``alias.scope``' Metadata
3267 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3269 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3270 noalias memory-access sets. This means that some collection of memory access
3271 instructions (loads, stores, memory-accessing calls, etc.) that carry
3272 ``noalias`` metadata can specifically be specified not to alias with some other
3273 collection of memory access instructions that carry ``alias.scope`` metadata.
3274 Each type of metadata specifies a list of scopes where each scope has an id and
3275 a domain. When evaluating an aliasing query, if for some some domain, the set
3276 of scopes with that domain in one instruction's ``alias.scope`` list is a
3277 subset of (or equal to) the set of scopes for that domain in another
3278 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3281 The metadata identifying each domain is itself a list containing one or two
3282 entries. The first entry is the name of the domain. Note that if the name is a
3283 string then it can be combined accross functions and translation units. A
3284 self-reference can be used to create globally unique domain names. A
3285 descriptive string may optionally be provided as a second list entry.
3287 The metadata identifying each scope is also itself a list containing two or
3288 three entries. The first entry is the name of the scope. Note that if the name
3289 is a string then it can be combined accross functions and translation units. A
3290 self-reference can be used to create globally unique scope names. A metadata
3291 reference to the scope's domain is the second entry. A descriptive string may
3292 optionally be provided as a third list entry.
3296 .. code-block:: llvm
3298 ; Two scope domains:
3302 ; Some scopes in these domains:
3308 !5 = !{!4} ; A list containing only scope !4
3312 ; These two instructions don't alias:
3313 %0 = load float* %c, align 4, !alias.scope !5
3314 store float %0, float* %arrayidx.i, align 4, !noalias !5
3316 ; These two instructions also don't alias (for domain !1, the set of scopes
3317 ; in the !alias.scope equals that in the !noalias list):
3318 %2 = load float* %c, align 4, !alias.scope !5
3319 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3321 ; These two instructions don't alias (for domain !0, the set of scopes in
3322 ; the !noalias list is not a superset of, or equal to, the scopes in the
3323 ; !alias.scope list):
3324 %2 = load float* %c, align 4, !alias.scope !6
3325 store float %0, float* %arrayidx.i, align 4, !noalias !7
3327 '``fpmath``' Metadata
3328 ^^^^^^^^^^^^^^^^^^^^^
3330 ``fpmath`` metadata may be attached to any instruction of floating point
3331 type. It can be used to express the maximum acceptable error in the
3332 result of that instruction, in ULPs, thus potentially allowing the
3333 compiler to use a more efficient but less accurate method of computing
3334 it. ULP is defined as follows:
3336 If ``x`` is a real number that lies between two finite consecutive
3337 floating-point numbers ``a`` and ``b``, without being equal to one
3338 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3339 distance between the two non-equal finite floating-point numbers
3340 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3342 The metadata node shall consist of a single positive floating point
3343 number representing the maximum relative error, for example:
3345 .. code-block:: llvm
3347 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3351 '``range``' Metadata
3352 ^^^^^^^^^^^^^^^^^^^^
3354 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3355 integer types. It expresses the possible ranges the loaded value or the value
3356 returned by the called function at this call site is in. The ranges are
3357 represented with a flattened list of integers. The loaded value or the value
3358 returned is known to be in the union of the ranges defined by each consecutive
3359 pair. Each pair has the following properties:
3361 - The type must match the type loaded by the instruction.
3362 - The pair ``a,b`` represents the range ``[a,b)``.
3363 - Both ``a`` and ``b`` are constants.
3364 - The range is allowed to wrap.
3365 - The range should not represent the full or empty set. That is,
3368 In addition, the pairs must be in signed order of the lower bound and
3369 they must be non-contiguous.
3373 .. code-block:: llvm
3375 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3376 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3377 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3378 %d = invoke i8 @bar() to label %cont
3379 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3381 !0 = !{ i8 0, i8 2 }
3382 !1 = !{ i8 255, i8 2 }
3383 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3384 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3389 It is sometimes useful to attach information to loop constructs. Currently,
3390 loop metadata is implemented as metadata attached to the branch instruction
3391 in the loop latch block. This type of metadata refer to a metadata node that is
3392 guaranteed to be separate for each loop. The loop identifier metadata is
3393 specified with the name ``llvm.loop``.
3395 The loop identifier metadata is implemented using a metadata that refers to
3396 itself to avoid merging it with any other identifier metadata, e.g.,
3397 during module linkage or function inlining. That is, each loop should refer
3398 to their own identification metadata even if they reside in separate functions.
3399 The following example contains loop identifier metadata for two separate loop
3402 .. code-block:: llvm
3407 The loop identifier metadata can be used to specify additional
3408 per-loop metadata. Any operands after the first operand can be treated
3409 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3410 suggests an unroll factor to the loop unroller:
3412 .. code-block:: llvm
3414 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3417 !1 = !{!"llvm.loop.unroll.count", i32 4}
3419 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3422 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3423 used to control per-loop vectorization and interleaving parameters such as
3424 vectorization width and interleave count. These metadata should be used in
3425 conjunction with ``llvm.loop`` loop identification metadata. The
3426 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3427 optimization hints and the optimizer will only interleave and vectorize loops if
3428 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3429 which contains information about loop-carried memory dependencies can be helpful
3430 in determining the safety of these transformations.
3432 '``llvm.loop.interleave.count``' Metadata
3433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3435 This metadata suggests an interleave count to the loop interleaver.
3436 The first operand is the string ``llvm.loop.interleave.count`` and the
3437 second operand is an integer specifying the interleave count. For
3440 .. code-block:: llvm
3442 !0 = !{!"llvm.loop.interleave.count", i32 4}
3444 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3445 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3446 then the interleave count will be determined automatically.
3448 '``llvm.loop.vectorize.enable``' Metadata
3449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3451 This metadata selectively enables or disables vectorization for the loop. The
3452 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3453 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3454 0 disables vectorization:
3456 .. code-block:: llvm
3458 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3459 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3461 '``llvm.loop.vectorize.width``' Metadata
3462 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3464 This metadata sets the target width of the vectorizer. The first
3465 operand is the string ``llvm.loop.vectorize.width`` and the second
3466 operand is an integer specifying the width. For example:
3468 .. code-block:: llvm
3470 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3472 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3473 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3474 0 or if the loop does not have this metadata the width will be
3475 determined automatically.
3477 '``llvm.loop.unroll``'
3478 ^^^^^^^^^^^^^^^^^^^^^^
3480 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3481 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3482 metadata should be used in conjunction with ``llvm.loop`` loop
3483 identification metadata. The ``llvm.loop.unroll`` metadata are only
3484 optimization hints and the unrolling will only be performed if the
3485 optimizer believes it is safe to do so.
3487 '``llvm.loop.unroll.count``' Metadata
3488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3490 This metadata suggests an unroll factor to the loop unroller. The
3491 first operand is the string ``llvm.loop.unroll.count`` and the second
3492 operand is a positive integer specifying the unroll factor. For
3495 .. code-block:: llvm
3497 !0 = !{!"llvm.loop.unroll.count", i32 4}
3499 If the trip count of the loop is less than the unroll count the loop
3500 will be partially unrolled.
3502 '``llvm.loop.unroll.disable``' Metadata
3503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3505 This metadata either disables loop unrolling. The metadata has a single operand
3506 which is the string ``llvm.loop.unroll.disable``. For example:
3508 .. code-block:: llvm
3510 !0 = !{!"llvm.loop.unroll.disable"}
3512 '``llvm.loop.unroll.full``' Metadata
3513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3515 This metadata either suggests that the loop should be unrolled fully. The
3516 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3519 .. code-block:: llvm
3521 !0 = !{!"llvm.loop.unroll.full"}
3526 Metadata types used to annotate memory accesses with information helpful
3527 for optimizations are prefixed with ``llvm.mem``.
3529 '``llvm.mem.parallel_loop_access``' Metadata
3530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3532 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3533 or metadata containing a list of loop identifiers for nested loops.
3534 The metadata is attached to memory accessing instructions and denotes that
3535 no loop carried memory dependence exist between it and other instructions denoted
3536 with the same loop identifier.
3538 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3539 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3540 set of loops associated with that metadata, respectively, then there is no loop
3541 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3544 As a special case, if all memory accessing instructions in a loop have
3545 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3546 loop has no loop carried memory dependences and is considered to be a parallel
3549 Note that if not all memory access instructions have such metadata referring to
3550 the loop, then the loop is considered not being trivially parallel. Additional
3551 memory dependence analysis is required to make that determination. As a fail
3552 safe mechanism, this causes loops that were originally parallel to be considered
3553 sequential (if optimization passes that are unaware of the parallel semantics
3554 insert new memory instructions into the loop body).
3556 Example of a loop that is considered parallel due to its correct use of
3557 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3558 metadata types that refer to the same loop identifier metadata.
3560 .. code-block:: llvm
3564 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3566 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3568 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3574 It is also possible to have nested parallel loops. In that case the
3575 memory accesses refer to a list of loop identifier metadata nodes instead of
3576 the loop identifier metadata node directly:
3578 .. code-block:: llvm
3582 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3584 br label %inner.for.body
3588 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3590 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3592 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3596 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3598 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3600 outer.for.end: ; preds = %for.body
3602 !0 = !{!1, !2} ; a list of loop identifiers
3603 !1 = !{!1} ; an identifier for the inner loop
3604 !2 = !{!2} ; an identifier for the outer loop
3609 The ``llvm.bitsets`` global metadata is used to implement
3610 :doc:`bitsets <BitSets>`.
3612 Module Flags Metadata
3613 =====================
3615 Information about the module as a whole is difficult to convey to LLVM's
3616 subsystems. The LLVM IR isn't sufficient to transmit this information.
3617 The ``llvm.module.flags`` named metadata exists in order to facilitate
3618 this. These flags are in the form of key / value pairs --- much like a
3619 dictionary --- making it easy for any subsystem who cares about a flag to
3622 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3623 Each triplet has the following form:
3625 - The first element is a *behavior* flag, which specifies the behavior
3626 when two (or more) modules are merged together, and it encounters two
3627 (or more) metadata with the same ID. The supported behaviors are
3629 - The second element is a metadata string that is a unique ID for the
3630 metadata. Each module may only have one flag entry for each unique ID (not
3631 including entries with the **Require** behavior).
3632 - The third element is the value of the flag.
3634 When two (or more) modules are merged together, the resulting
3635 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3636 each unique metadata ID string, there will be exactly one entry in the merged
3637 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3638 be determined by the merge behavior flag, as described below. The only exception
3639 is that entries with the *Require* behavior are always preserved.
3641 The following behaviors are supported:
3652 Emits an error if two values disagree, otherwise the resulting value
3653 is that of the operands.
3657 Emits a warning if two values disagree. The result value will be the
3658 operand for the flag from the first module being linked.
3662 Adds a requirement that another module flag be present and have a
3663 specified value after linking is performed. The value must be a
3664 metadata pair, where the first element of the pair is the ID of the
3665 module flag to be restricted, and the second element of the pair is
3666 the value the module flag should be restricted to. This behavior can
3667 be used to restrict the allowable results (via triggering of an
3668 error) of linking IDs with the **Override** behavior.
3672 Uses the specified value, regardless of the behavior or value of the
3673 other module. If both modules specify **Override**, but the values
3674 differ, an error will be emitted.
3678 Appends the two values, which are required to be metadata nodes.
3682 Appends the two values, which are required to be metadata
3683 nodes. However, duplicate entries in the second list are dropped
3684 during the append operation.
3686 It is an error for a particular unique flag ID to have multiple behaviors,
3687 except in the case of **Require** (which adds restrictions on another metadata
3688 value) or **Override**.
3690 An example of module flags:
3692 .. code-block:: llvm
3694 !0 = !{ i32 1, !"foo", i32 1 }
3695 !1 = !{ i32 4, !"bar", i32 37 }
3696 !2 = !{ i32 2, !"qux", i32 42 }
3697 !3 = !{ i32 3, !"qux",
3702 !llvm.module.flags = !{ !0, !1, !2, !3 }
3704 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3705 if two or more ``!"foo"`` flags are seen is to emit an error if their
3706 values are not equal.
3708 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3709 behavior if two or more ``!"bar"`` flags are seen is to use the value
3712 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3713 behavior if two or more ``!"qux"`` flags are seen is to emit a
3714 warning if their values are not equal.
3716 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3722 The behavior is to emit an error if the ``llvm.module.flags`` does not
3723 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3726 Objective-C Garbage Collection Module Flags Metadata
3727 ----------------------------------------------------
3729 On the Mach-O platform, Objective-C stores metadata about garbage
3730 collection in a special section called "image info". The metadata
3731 consists of a version number and a bitmask specifying what types of
3732 garbage collection are supported (if any) by the file. If two or more
3733 modules are linked together their garbage collection metadata needs to
3734 be merged rather than appended together.
3736 The Objective-C garbage collection module flags metadata consists of the
3737 following key-value pairs:
3746 * - ``Objective-C Version``
3747 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3749 * - ``Objective-C Image Info Version``
3750 - **[Required]** --- The version of the image info section. Currently
3753 * - ``Objective-C Image Info Section``
3754 - **[Required]** --- The section to place the metadata. Valid values are
3755 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3756 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3757 Objective-C ABI version 2.
3759 * - ``Objective-C Garbage Collection``
3760 - **[Required]** --- Specifies whether garbage collection is supported or
3761 not. Valid values are 0, for no garbage collection, and 2, for garbage
3762 collection supported.
3764 * - ``Objective-C GC Only``
3765 - **[Optional]** --- Specifies that only garbage collection is supported.
3766 If present, its value must be 6. This flag requires that the
3767 ``Objective-C Garbage Collection`` flag have the value 2.
3769 Some important flag interactions:
3771 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3772 merged with a module with ``Objective-C Garbage Collection`` set to
3773 2, then the resulting module has the
3774 ``Objective-C Garbage Collection`` flag set to 0.
3775 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3776 merged with a module with ``Objective-C GC Only`` set to 6.
3778 Automatic Linker Flags Module Flags Metadata
3779 --------------------------------------------
3781 Some targets support embedding flags to the linker inside individual object
3782 files. Typically this is used in conjunction with language extensions which
3783 allow source files to explicitly declare the libraries they depend on, and have
3784 these automatically be transmitted to the linker via object files.
3786 These flags are encoded in the IR using metadata in the module flags section,
3787 using the ``Linker Options`` key. The merge behavior for this flag is required
3788 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3789 node which should be a list of other metadata nodes, each of which should be a
3790 list of metadata strings defining linker options.
3792 For example, the following metadata section specifies two separate sets of
3793 linker options, presumably to link against ``libz`` and the ``Cocoa``
3796 !0 = !{ i32 6, !"Linker Options",
3799 !{ !"-framework", !"Cocoa" } } }
3800 !llvm.module.flags = !{ !0 }
3802 The metadata encoding as lists of lists of options, as opposed to a collapsed
3803 list of options, is chosen so that the IR encoding can use multiple option
3804 strings to specify e.g., a single library, while still having that specifier be
3805 preserved as an atomic element that can be recognized by a target specific
3806 assembly writer or object file emitter.
3808 Each individual option is required to be either a valid option for the target's
3809 linker, or an option that is reserved by the target specific assembly writer or
3810 object file emitter. No other aspect of these options is defined by the IR.
3812 C type width Module Flags Metadata
3813 ----------------------------------
3815 The ARM backend emits a section into each generated object file describing the
3816 options that it was compiled with (in a compiler-independent way) to prevent
3817 linking incompatible objects, and to allow automatic library selection. Some
3818 of these options are not visible at the IR level, namely wchar_t width and enum
3821 To pass this information to the backend, these options are encoded in module
3822 flags metadata, using the following key-value pairs:
3832 - * 0 --- sizeof(wchar_t) == 4
3833 * 1 --- sizeof(wchar_t) == 2
3836 - * 0 --- Enums are at least as large as an ``int``.
3837 * 1 --- Enums are stored in the smallest integer type which can
3838 represent all of its values.
3840 For example, the following metadata section specifies that the module was
3841 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3842 enum is the smallest type which can represent all of its values::
3844 !llvm.module.flags = !{!0, !1}
3845 !0 = !{i32 1, !"short_wchar", i32 1}
3846 !1 = !{i32 1, !"short_enum", i32 0}
3848 .. _intrinsicglobalvariables:
3850 Intrinsic Global Variables
3851 ==========================
3853 LLVM has a number of "magic" global variables that contain data that
3854 affect code generation or other IR semantics. These are documented here.
3855 All globals of this sort should have a section specified as
3856 "``llvm.metadata``". This section and all globals that start with
3857 "``llvm.``" are reserved for use by LLVM.
3861 The '``llvm.used``' Global Variable
3862 -----------------------------------
3864 The ``@llvm.used`` global is an array which has
3865 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3866 pointers to named global variables, functions and aliases which may optionally
3867 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3870 .. code-block:: llvm
3875 @llvm.used = appending global [2 x i8*] [
3877 i8* bitcast (i32* @Y to i8*)
3878 ], section "llvm.metadata"
3880 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3881 and linker are required to treat the symbol as if there is a reference to the
3882 symbol that it cannot see (which is why they have to be named). For example, if
3883 a variable has internal linkage and no references other than that from the
3884 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3885 references from inline asms and other things the compiler cannot "see", and
3886 corresponds to "``attribute((used))``" in GNU C.
3888 On some targets, the code generator must emit a directive to the
3889 assembler or object file to prevent the assembler and linker from
3890 molesting the symbol.
3892 .. _gv_llvmcompilerused:
3894 The '``llvm.compiler.used``' Global Variable
3895 --------------------------------------------
3897 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3898 directive, except that it only prevents the compiler from touching the
3899 symbol. On targets that support it, this allows an intelligent linker to
3900 optimize references to the symbol without being impeded as it would be
3903 This is a rare construct that should only be used in rare circumstances,
3904 and should not be exposed to source languages.
3906 .. _gv_llvmglobalctors:
3908 The '``llvm.global_ctors``' Global Variable
3909 -------------------------------------------
3911 .. code-block:: llvm
3913 %0 = type { i32, void ()*, i8* }
3914 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3916 The ``@llvm.global_ctors`` array contains a list of constructor
3917 functions, priorities, and an optional associated global or function.
3918 The functions referenced by this array will be called in ascending order
3919 of priority (i.e. lowest first) when the module is loaded. The order of
3920 functions with the same priority is not defined.
3922 If the third field is present, non-null, and points to a global variable
3923 or function, the initializer function will only run if the associated
3924 data from the current module is not discarded.
3926 .. _llvmglobaldtors:
3928 The '``llvm.global_dtors``' Global Variable
3929 -------------------------------------------
3931 .. code-block:: llvm
3933 %0 = type { i32, void ()*, i8* }
3934 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3936 The ``@llvm.global_dtors`` array contains a list of destructor
3937 functions, priorities, and an optional associated global or function.
3938 The functions referenced by this array will be called in descending
3939 order of priority (i.e. highest first) when the module is unloaded. The
3940 order of functions with the same priority is not defined.
3942 If the third field is present, non-null, and points to a global variable
3943 or function, the destructor function will only run if the associated
3944 data from the current module is not discarded.
3946 Instruction Reference
3947 =====================
3949 The LLVM instruction set consists of several different classifications
3950 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3951 instructions <binaryops>`, :ref:`bitwise binary
3952 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3953 :ref:`other instructions <otherops>`.
3957 Terminator Instructions
3958 -----------------------
3960 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3961 program ends with a "Terminator" instruction, which indicates which
3962 block should be executed after the current block is finished. These
3963 terminator instructions typically yield a '``void``' value: they produce
3964 control flow, not values (the one exception being the
3965 ':ref:`invoke <i_invoke>`' instruction).
3967 The terminator instructions are: ':ref:`ret <i_ret>`',
3968 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3969 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3970 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3974 '``ret``' Instruction
3975 ^^^^^^^^^^^^^^^^^^^^^
3982 ret <type> <value> ; Return a value from a non-void function
3983 ret void ; Return from void function
3988 The '``ret``' instruction is used to return control flow (and optionally
3989 a value) from a function back to the caller.
3991 There are two forms of the '``ret``' instruction: one that returns a
3992 value and then causes control flow, and one that just causes control
3998 The '``ret``' instruction optionally accepts a single argument, the
3999 return value. The type of the return value must be a ':ref:`first
4000 class <t_firstclass>`' type.
4002 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4003 return type and contains a '``ret``' instruction with no return value or
4004 a return value with a type that does not match its type, or if it has a
4005 void return type and contains a '``ret``' instruction with a return
4011 When the '``ret``' instruction is executed, control flow returns back to
4012 the calling function's context. If the caller is a
4013 ":ref:`call <i_call>`" instruction, execution continues at the
4014 instruction after the call. If the caller was an
4015 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4016 beginning of the "normal" destination block. If the instruction returns
4017 a value, that value shall set the call or invoke instruction's return
4023 .. code-block:: llvm
4025 ret i32 5 ; Return an integer value of 5
4026 ret void ; Return from a void function
4027 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4031 '``br``' Instruction
4032 ^^^^^^^^^^^^^^^^^^^^
4039 br i1 <cond>, label <iftrue>, label <iffalse>
4040 br label <dest> ; Unconditional branch
4045 The '``br``' instruction is used to cause control flow to transfer to a
4046 different basic block in the current function. There are two forms of
4047 this instruction, corresponding to a conditional branch and an
4048 unconditional branch.
4053 The conditional branch form of the '``br``' instruction takes a single
4054 '``i1``' value and two '``label``' values. The unconditional form of the
4055 '``br``' instruction takes a single '``label``' value as a target.
4060 Upon execution of a conditional '``br``' instruction, the '``i1``'
4061 argument is evaluated. If the value is ``true``, control flows to the
4062 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4063 to the '``iffalse``' ``label`` argument.
4068 .. code-block:: llvm
4071 %cond = icmp eq i32 %a, %b
4072 br i1 %cond, label %IfEqual, label %IfUnequal
4080 '``switch``' Instruction
4081 ^^^^^^^^^^^^^^^^^^^^^^^^
4088 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4093 The '``switch``' instruction is used to transfer control flow to one of
4094 several different places. It is a generalization of the '``br``'
4095 instruction, allowing a branch to occur to one of many possible
4101 The '``switch``' instruction uses three parameters: an integer
4102 comparison value '``value``', a default '``label``' destination, and an
4103 array of pairs of comparison value constants and '``label``'s. The table
4104 is not allowed to contain duplicate constant entries.
4109 The ``switch`` instruction specifies a table of values and destinations.
4110 When the '``switch``' instruction is executed, this table is searched
4111 for the given value. If the value is found, control flow is transferred
4112 to the corresponding destination; otherwise, control flow is transferred
4113 to the default destination.
4118 Depending on properties of the target machine and the particular
4119 ``switch`` instruction, this instruction may be code generated in
4120 different ways. For example, it could be generated as a series of
4121 chained conditional branches or with a lookup table.
4126 .. code-block:: llvm
4128 ; Emulate a conditional br instruction
4129 %Val = zext i1 %value to i32
4130 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4132 ; Emulate an unconditional br instruction
4133 switch i32 0, label %dest [ ]
4135 ; Implement a jump table:
4136 switch i32 %val, label %otherwise [ i32 0, label %onzero
4138 i32 2, label %ontwo ]
4142 '``indirectbr``' Instruction
4143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4150 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4155 The '``indirectbr``' instruction implements an indirect branch to a
4156 label within the current function, whose address is specified by
4157 "``address``". Address must be derived from a
4158 :ref:`blockaddress <blockaddress>` constant.
4163 The '``address``' argument is the address of the label to jump to. The
4164 rest of the arguments indicate the full set of possible destinations
4165 that the address may point to. Blocks are allowed to occur multiple
4166 times in the destination list, though this isn't particularly useful.
4168 This destination list is required so that dataflow analysis has an
4169 accurate understanding of the CFG.
4174 Control transfers to the block specified in the address argument. All
4175 possible destination blocks must be listed in the label list, otherwise
4176 this instruction has undefined behavior. This implies that jumps to
4177 labels defined in other functions have undefined behavior as well.
4182 This is typically implemented with a jump through a register.
4187 .. code-block:: llvm
4189 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4193 '``invoke``' Instruction
4194 ^^^^^^^^^^^^^^^^^^^^^^^^
4201 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4202 to label <normal label> unwind label <exception label>
4207 The '``invoke``' instruction causes control to transfer to a specified
4208 function, with the possibility of control flow transfer to either the
4209 '``normal``' label or the '``exception``' label. If the callee function
4210 returns with the "``ret``" instruction, control flow will return to the
4211 "normal" label. If the callee (or any indirect callees) returns via the
4212 ":ref:`resume <i_resume>`" instruction or other exception handling
4213 mechanism, control is interrupted and continued at the dynamically
4214 nearest "exception" label.
4216 The '``exception``' label is a `landing
4217 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4218 '``exception``' label is required to have the
4219 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4220 information about the behavior of the program after unwinding happens,
4221 as its first non-PHI instruction. The restrictions on the
4222 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4223 instruction, so that the important information contained within the
4224 "``landingpad``" instruction can't be lost through normal code motion.
4229 This instruction requires several arguments:
4231 #. The optional "cconv" marker indicates which :ref:`calling
4232 convention <callingconv>` the call should use. If none is
4233 specified, the call defaults to using C calling conventions.
4234 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4235 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4237 #. '``ptr to function ty``': shall be the signature of the pointer to
4238 function value being invoked. In most cases, this is a direct
4239 function invocation, but indirect ``invoke``'s are just as possible,
4240 branching off an arbitrary pointer to function value.
4241 #. '``function ptr val``': An LLVM value containing a pointer to a
4242 function to be invoked.
4243 #. '``function args``': argument list whose types match the function
4244 signature argument types and parameter attributes. All arguments must
4245 be of :ref:`first class <t_firstclass>` type. If the function signature
4246 indicates the function accepts a variable number of arguments, the
4247 extra arguments can be specified.
4248 #. '``normal label``': the label reached when the called function
4249 executes a '``ret``' instruction.
4250 #. '``exception label``': the label reached when a callee returns via
4251 the :ref:`resume <i_resume>` instruction or other exception handling
4253 #. The optional :ref:`function attributes <fnattrs>` list. Only
4254 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4255 attributes are valid here.
4260 This instruction is designed to operate as a standard '``call``'
4261 instruction in most regards. The primary difference is that it
4262 establishes an association with a label, which is used by the runtime
4263 library to unwind the stack.
4265 This instruction is used in languages with destructors to ensure that
4266 proper cleanup is performed in the case of either a ``longjmp`` or a
4267 thrown exception. Additionally, this is important for implementation of
4268 '``catch``' clauses in high-level languages that support them.
4270 For the purposes of the SSA form, the definition of the value returned
4271 by the '``invoke``' instruction is deemed to occur on the edge from the
4272 current block to the "normal" label. If the callee unwinds then no
4273 return value is available.
4278 .. code-block:: llvm
4280 %retval = invoke i32 @Test(i32 15) to label %Continue
4281 unwind label %TestCleanup ; i32:retval set
4282 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4283 unwind label %TestCleanup ; i32:retval set
4287 '``resume``' Instruction
4288 ^^^^^^^^^^^^^^^^^^^^^^^^
4295 resume <type> <value>
4300 The '``resume``' instruction is a terminator instruction that has no
4306 The '``resume``' instruction requires one argument, which must have the
4307 same type as the result of any '``landingpad``' instruction in the same
4313 The '``resume``' instruction resumes propagation of an existing
4314 (in-flight) exception whose unwinding was interrupted with a
4315 :ref:`landingpad <i_landingpad>` instruction.
4320 .. code-block:: llvm
4322 resume { i8*, i32 } %exn
4326 '``unreachable``' Instruction
4327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4339 The '``unreachable``' instruction has no defined semantics. This
4340 instruction is used to inform the optimizer that a particular portion of
4341 the code is not reachable. This can be used to indicate that the code
4342 after a no-return function cannot be reached, and other facts.
4347 The '``unreachable``' instruction has no defined semantics.
4354 Binary operators are used to do most of the computation in a program.
4355 They require two operands of the same type, execute an operation on
4356 them, and produce a single value. The operands might represent multiple
4357 data, as is the case with the :ref:`vector <t_vector>` data type. The
4358 result value has the same type as its operands.
4360 There are several different binary operators:
4364 '``add``' Instruction
4365 ^^^^^^^^^^^^^^^^^^^^^
4372 <result> = add <ty> <op1>, <op2> ; yields ty:result
4373 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4374 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4375 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4380 The '``add``' instruction returns the sum of its two operands.
4385 The two arguments to the '``add``' instruction must be
4386 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4387 arguments must have identical types.
4392 The value produced is the integer sum of the two operands.
4394 If the sum has unsigned overflow, the result returned is the
4395 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4398 Because LLVM integers use a two's complement representation, this
4399 instruction is appropriate for both signed and unsigned integers.
4401 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4402 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4403 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4404 unsigned and/or signed overflow, respectively, occurs.
4409 .. code-block:: llvm
4411 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4415 '``fadd``' Instruction
4416 ^^^^^^^^^^^^^^^^^^^^^^
4423 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4428 The '``fadd``' instruction returns the sum of its two operands.
4433 The two arguments to the '``fadd``' instruction must be :ref:`floating
4434 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4435 Both arguments must have identical types.
4440 The value produced is the floating point sum of the two operands. This
4441 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4442 which are optimization hints to enable otherwise unsafe floating point
4448 .. code-block:: llvm
4450 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4452 '``sub``' Instruction
4453 ^^^^^^^^^^^^^^^^^^^^^
4460 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4461 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4462 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4463 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4468 The '``sub``' instruction returns the difference of its two operands.
4470 Note that the '``sub``' instruction is used to represent the '``neg``'
4471 instruction present in most other intermediate representations.
4476 The two arguments to the '``sub``' instruction must be
4477 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4478 arguments must have identical types.
4483 The value produced is the integer difference of the two operands.
4485 If the difference has unsigned overflow, the result returned is the
4486 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4489 Because LLVM integers use a two's complement representation, this
4490 instruction is appropriate for both signed and unsigned integers.
4492 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4493 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4494 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4495 unsigned and/or signed overflow, respectively, occurs.
4500 .. code-block:: llvm
4502 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4503 <result> = sub i32 0, %val ; yields i32:result = -%var
4507 '``fsub``' Instruction
4508 ^^^^^^^^^^^^^^^^^^^^^^
4515 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4520 The '``fsub``' instruction returns the difference of its two operands.
4522 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4523 instruction present in most other intermediate representations.
4528 The two arguments to the '``fsub``' instruction must be :ref:`floating
4529 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4530 Both arguments must have identical types.
4535 The value produced is the floating point difference of the two operands.
4536 This instruction can also take any number of :ref:`fast-math
4537 flags <fastmath>`, which are optimization hints to enable otherwise
4538 unsafe floating point optimizations:
4543 .. code-block:: llvm
4545 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4546 <result> = fsub float -0.0, %val ; yields float:result = -%var
4548 '``mul``' Instruction
4549 ^^^^^^^^^^^^^^^^^^^^^
4556 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4557 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4558 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4559 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4564 The '``mul``' instruction returns the product of its two operands.
4569 The two arguments to the '``mul``' instruction must be
4570 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4571 arguments must have identical types.
4576 The value produced is the integer product of the two operands.
4578 If the result of the multiplication has unsigned overflow, the result
4579 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4580 bit width of the result.
4582 Because LLVM integers use a two's complement representation, and the
4583 result is the same width as the operands, this instruction returns the
4584 correct result for both signed and unsigned integers. If a full product
4585 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4586 sign-extended or zero-extended as appropriate to the width of the full
4589 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4590 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4591 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4592 unsigned and/or signed overflow, respectively, occurs.
4597 .. code-block:: llvm
4599 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4603 '``fmul``' Instruction
4604 ^^^^^^^^^^^^^^^^^^^^^^
4611 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4616 The '``fmul``' instruction returns the product of its two operands.
4621 The two arguments to the '``fmul``' instruction must be :ref:`floating
4622 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4623 Both arguments must have identical types.
4628 The value produced is the floating point product of the two operands.
4629 This instruction can also take any number of :ref:`fast-math
4630 flags <fastmath>`, which are optimization hints to enable otherwise
4631 unsafe floating point optimizations:
4636 .. code-block:: llvm
4638 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4640 '``udiv``' Instruction
4641 ^^^^^^^^^^^^^^^^^^^^^^
4648 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4649 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4654 The '``udiv``' instruction returns the quotient of its two operands.
4659 The two arguments to the '``udiv``' instruction must be
4660 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4661 arguments must have identical types.
4666 The value produced is the unsigned integer quotient of the two operands.
4668 Note that unsigned integer division and signed integer division are
4669 distinct operations; for signed integer division, use '``sdiv``'.
4671 Division by zero leads to undefined behavior.
4673 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4674 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4675 such, "((a udiv exact b) mul b) == a").
4680 .. code-block:: llvm
4682 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4684 '``sdiv``' Instruction
4685 ^^^^^^^^^^^^^^^^^^^^^^
4692 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4693 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4698 The '``sdiv``' instruction returns the quotient of its two operands.
4703 The two arguments to the '``sdiv``' instruction must be
4704 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4705 arguments must have identical types.
4710 The value produced is the signed integer quotient of the two operands
4711 rounded towards zero.
4713 Note that signed integer division and unsigned integer division are
4714 distinct operations; for unsigned integer division, use '``udiv``'.
4716 Division by zero leads to undefined behavior. Overflow also leads to
4717 undefined behavior; this is a rare case, but can occur, for example, by
4718 doing a 32-bit division of -2147483648 by -1.
4720 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4721 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4726 .. code-block:: llvm
4728 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4732 '``fdiv``' Instruction
4733 ^^^^^^^^^^^^^^^^^^^^^^
4740 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4745 The '``fdiv``' instruction returns the quotient of its two operands.
4750 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4751 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4752 Both arguments must have identical types.
4757 The value produced is the floating point quotient of the two operands.
4758 This instruction can also take any number of :ref:`fast-math
4759 flags <fastmath>`, which are optimization hints to enable otherwise
4760 unsafe floating point optimizations:
4765 .. code-block:: llvm
4767 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4769 '``urem``' Instruction
4770 ^^^^^^^^^^^^^^^^^^^^^^
4777 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4782 The '``urem``' instruction returns the remainder from the unsigned
4783 division of its two arguments.
4788 The two arguments to the '``urem``' instruction must be
4789 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4790 arguments must have identical types.
4795 This instruction returns the unsigned integer *remainder* of a division.
4796 This instruction always performs an unsigned division to get the
4799 Note that unsigned integer remainder and signed integer remainder are
4800 distinct operations; for signed integer remainder, use '``srem``'.
4802 Taking the remainder of a division by zero leads to undefined behavior.
4807 .. code-block:: llvm
4809 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4811 '``srem``' Instruction
4812 ^^^^^^^^^^^^^^^^^^^^^^
4819 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4824 The '``srem``' instruction returns the remainder from the signed
4825 division of its two operands. This instruction can also take
4826 :ref:`vector <t_vector>` versions of the values in which case the elements
4832 The two arguments to the '``srem``' instruction must be
4833 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4834 arguments must have identical types.
4839 This instruction returns the *remainder* of a division (where the result
4840 is either zero or has the same sign as the dividend, ``op1``), not the
4841 *modulo* operator (where the result is either zero or has the same sign
4842 as the divisor, ``op2``) of a value. For more information about the
4843 difference, see `The Math
4844 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4845 table of how this is implemented in various languages, please see
4847 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4849 Note that signed integer remainder and unsigned integer remainder are
4850 distinct operations; for unsigned integer remainder, use '``urem``'.
4852 Taking the remainder of a division by zero leads to undefined behavior.
4853 Overflow also leads to undefined behavior; this is a rare case, but can
4854 occur, for example, by taking the remainder of a 32-bit division of
4855 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4856 rule lets srem be implemented using instructions that return both the
4857 result of the division and the remainder.)
4862 .. code-block:: llvm
4864 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4868 '``frem``' Instruction
4869 ^^^^^^^^^^^^^^^^^^^^^^
4876 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4881 The '``frem``' instruction returns the remainder from the division of
4887 The two arguments to the '``frem``' instruction must be :ref:`floating
4888 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4889 Both arguments must have identical types.
4894 This instruction returns the *remainder* of a division. The remainder
4895 has the same sign as the dividend. This instruction can also take any
4896 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4897 to enable otherwise unsafe floating point optimizations:
4902 .. code-block:: llvm
4904 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4908 Bitwise Binary Operations
4909 -------------------------
4911 Bitwise binary operators are used to do various forms of bit-twiddling
4912 in a program. They are generally very efficient instructions and can
4913 commonly be strength reduced from other instructions. They require two
4914 operands of the same type, execute an operation on them, and produce a
4915 single value. The resulting value is the same type as its operands.
4917 '``shl``' Instruction
4918 ^^^^^^^^^^^^^^^^^^^^^
4925 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4926 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4927 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4928 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4933 The '``shl``' instruction returns the first operand shifted to the left
4934 a specified number of bits.
4939 Both arguments to the '``shl``' instruction must be the same
4940 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4941 '``op2``' is treated as an unsigned value.
4946 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4947 where ``n`` is the width of the result. If ``op2`` is (statically or
4948 dynamically) negative or equal to or larger than the number of bits in
4949 ``op1``, the result is undefined. If the arguments are vectors, each
4950 vector element of ``op1`` is shifted by the corresponding shift amount
4953 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4954 value <poisonvalues>` if it shifts out any non-zero bits. If the
4955 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4956 value <poisonvalues>` if it shifts out any bits that disagree with the
4957 resultant sign bit. As such, NUW/NSW have the same semantics as they
4958 would if the shift were expressed as a mul instruction with the same
4959 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4964 .. code-block:: llvm
4966 <result> = shl i32 4, %var ; yields i32: 4 << %var
4967 <result> = shl i32 4, 2 ; yields i32: 16
4968 <result> = shl i32 1, 10 ; yields i32: 1024
4969 <result> = shl i32 1, 32 ; undefined
4970 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4972 '``lshr``' Instruction
4973 ^^^^^^^^^^^^^^^^^^^^^^
4980 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4981 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4986 The '``lshr``' instruction (logical shift right) returns the first
4987 operand shifted to the right a specified number of bits with zero fill.
4992 Both arguments to the '``lshr``' instruction must be the same
4993 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4994 '``op2``' is treated as an unsigned value.
4999 This instruction always performs a logical shift right operation. The
5000 most significant bits of the result will be filled with zero bits after
5001 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5002 than the number of bits in ``op1``, the result is undefined. If the
5003 arguments are vectors, each vector element of ``op1`` is shifted by the
5004 corresponding shift amount in ``op2``.
5006 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5007 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5013 .. code-block:: llvm
5015 <result> = lshr i32 4, 1 ; yields i32:result = 2
5016 <result> = lshr i32 4, 2 ; yields i32:result = 1
5017 <result> = lshr i8 4, 3 ; yields i8:result = 0
5018 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5019 <result> = lshr i32 1, 32 ; undefined
5020 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5022 '``ashr``' Instruction
5023 ^^^^^^^^^^^^^^^^^^^^^^
5030 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5031 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5036 The '``ashr``' instruction (arithmetic shift right) returns the first
5037 operand shifted to the right a specified number of bits with sign
5043 Both arguments to the '``ashr``' instruction must be the same
5044 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5045 '``op2``' is treated as an unsigned value.
5050 This instruction always performs an arithmetic shift right operation,
5051 The most significant bits of the result will be filled with the sign bit
5052 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5053 than the number of bits in ``op1``, the result is undefined. If the
5054 arguments are vectors, each vector element of ``op1`` is shifted by the
5055 corresponding shift amount in ``op2``.
5057 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5058 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5064 .. code-block:: llvm
5066 <result> = ashr i32 4, 1 ; yields i32:result = 2
5067 <result> = ashr i32 4, 2 ; yields i32:result = 1
5068 <result> = ashr i8 4, 3 ; yields i8:result = 0
5069 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5070 <result> = ashr i32 1, 32 ; undefined
5071 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5073 '``and``' Instruction
5074 ^^^^^^^^^^^^^^^^^^^^^
5081 <result> = and <ty> <op1>, <op2> ; yields ty:result
5086 The '``and``' instruction returns the bitwise logical and of its two
5092 The two arguments to the '``and``' instruction must be
5093 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5094 arguments must have identical types.
5099 The truth table used for the '``and``' instruction is:
5116 .. code-block:: llvm
5118 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5119 <result> = and i32 15, 40 ; yields i32:result = 8
5120 <result> = and i32 4, 8 ; yields i32:result = 0
5122 '``or``' Instruction
5123 ^^^^^^^^^^^^^^^^^^^^
5130 <result> = or <ty> <op1>, <op2> ; yields ty:result
5135 The '``or``' instruction returns the bitwise logical inclusive or of its
5141 The two arguments to the '``or``' instruction must be
5142 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5143 arguments must have identical types.
5148 The truth table used for the '``or``' instruction is:
5167 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5168 <result> = or i32 15, 40 ; yields i32:result = 47
5169 <result> = or i32 4, 8 ; yields i32:result = 12
5171 '``xor``' Instruction
5172 ^^^^^^^^^^^^^^^^^^^^^
5179 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5184 The '``xor``' instruction returns the bitwise logical exclusive or of
5185 its two operands. The ``xor`` is used to implement the "one's
5186 complement" operation, which is the "~" operator in C.
5191 The two arguments to the '``xor``' instruction must be
5192 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5193 arguments must have identical types.
5198 The truth table used for the '``xor``' instruction is:
5215 .. code-block:: llvm
5217 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5218 <result> = xor i32 15, 40 ; yields i32:result = 39
5219 <result> = xor i32 4, 8 ; yields i32:result = 12
5220 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5225 LLVM supports several instructions to represent vector operations in a
5226 target-independent manner. These instructions cover the element-access
5227 and vector-specific operations needed to process vectors effectively.
5228 While LLVM does directly support these vector operations, many
5229 sophisticated algorithms will want to use target-specific intrinsics to
5230 take full advantage of a specific target.
5232 .. _i_extractelement:
5234 '``extractelement``' Instruction
5235 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5242 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5247 The '``extractelement``' instruction extracts a single scalar element
5248 from a vector at a specified index.
5253 The first operand of an '``extractelement``' instruction is a value of
5254 :ref:`vector <t_vector>` type. The second operand is an index indicating
5255 the position from which to extract the element. The index may be a
5256 variable of any integer type.
5261 The result is a scalar of the same type as the element type of ``val``.
5262 Its value is the value at position ``idx`` of ``val``. If ``idx``
5263 exceeds the length of ``val``, the results are undefined.
5268 .. code-block:: llvm
5270 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5272 .. _i_insertelement:
5274 '``insertelement``' Instruction
5275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5282 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5287 The '``insertelement``' instruction inserts a scalar element into a
5288 vector at a specified index.
5293 The first operand of an '``insertelement``' instruction is a value of
5294 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5295 type must equal the element type of the first operand. The third operand
5296 is an index indicating the position at which to insert the value. The
5297 index may be a variable of any integer type.
5302 The result is a vector of the same type as ``val``. Its element values
5303 are those of ``val`` except at position ``idx``, where it gets the value
5304 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5310 .. code-block:: llvm
5312 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5314 .. _i_shufflevector:
5316 '``shufflevector``' Instruction
5317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5324 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5329 The '``shufflevector``' instruction constructs a permutation of elements
5330 from two input vectors, returning a vector with the same element type as
5331 the input and length that is the same as the shuffle mask.
5336 The first two operands of a '``shufflevector``' instruction are vectors
5337 with the same type. The third argument is a shuffle mask whose element
5338 type is always 'i32'. The result of the instruction is a vector whose
5339 length is the same as the shuffle mask and whose element type is the
5340 same as the element type of the first two operands.
5342 The shuffle mask operand is required to be a constant vector with either
5343 constant integer or undef values.
5348 The elements of the two input vectors are numbered from left to right
5349 across both of the vectors. The shuffle mask operand specifies, for each
5350 element of the result vector, which element of the two input vectors the
5351 result element gets. The element selector may be undef (meaning "don't
5352 care") and the second operand may be undef if performing a shuffle from
5358 .. code-block:: llvm
5360 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5361 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5362 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5363 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5364 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5365 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5366 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5367 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5369 Aggregate Operations
5370 --------------------
5372 LLVM supports several instructions for working with
5373 :ref:`aggregate <t_aggregate>` values.
5377 '``extractvalue``' Instruction
5378 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5385 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5390 The '``extractvalue``' instruction extracts the value of a member field
5391 from an :ref:`aggregate <t_aggregate>` value.
5396 The first operand of an '``extractvalue``' instruction is a value of
5397 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5398 constant indices to specify which value to extract in a similar manner
5399 as indices in a '``getelementptr``' instruction.
5401 The major differences to ``getelementptr`` indexing are:
5403 - Since the value being indexed is not a pointer, the first index is
5404 omitted and assumed to be zero.
5405 - At least one index must be specified.
5406 - Not only struct indices but also array indices must be in bounds.
5411 The result is the value at the position in the aggregate specified by
5417 .. code-block:: llvm
5419 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5423 '``insertvalue``' Instruction
5424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5431 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5436 The '``insertvalue``' instruction inserts a value into a member field in
5437 an :ref:`aggregate <t_aggregate>` value.
5442 The first operand of an '``insertvalue``' instruction is a value of
5443 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5444 a first-class value to insert. The following operands are constant
5445 indices indicating the position at which to insert the value in a
5446 similar manner as indices in a '``extractvalue``' instruction. The value
5447 to insert must have the same type as the value identified by the
5453 The result is an aggregate of the same type as ``val``. Its value is
5454 that of ``val`` except that the value at the position specified by the
5455 indices is that of ``elt``.
5460 .. code-block:: llvm
5462 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5463 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5464 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5468 Memory Access and Addressing Operations
5469 ---------------------------------------
5471 A key design point of an SSA-based representation is how it represents
5472 memory. In LLVM, no memory locations are in SSA form, which makes things
5473 very simple. This section describes how to read, write, and allocate
5478 '``alloca``' Instruction
5479 ^^^^^^^^^^^^^^^^^^^^^^^^
5486 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5491 The '``alloca``' instruction allocates memory on the stack frame of the
5492 currently executing function, to be automatically released when this
5493 function returns to its caller. The object is always allocated in the
5494 generic address space (address space zero).
5499 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5500 bytes of memory on the runtime stack, returning a pointer of the
5501 appropriate type to the program. If "NumElements" is specified, it is
5502 the number of elements allocated, otherwise "NumElements" is defaulted
5503 to be one. If a constant alignment is specified, the value result of the
5504 allocation is guaranteed to be aligned to at least that boundary. The
5505 alignment may not be greater than ``1 << 29``. If not specified, or if
5506 zero, the target can choose to align the allocation on any convenient
5507 boundary compatible with the type.
5509 '``type``' may be any sized type.
5514 Memory is allocated; a pointer is returned. The operation is undefined
5515 if there is insufficient stack space for the allocation. '``alloca``'d
5516 memory is automatically released when the function returns. The
5517 '``alloca``' instruction is commonly used to represent automatic
5518 variables that must have an address available. When the function returns
5519 (either with the ``ret`` or ``resume`` instructions), the memory is
5520 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5521 The order in which memory is allocated (ie., which way the stack grows)
5527 .. code-block:: llvm
5529 %ptr = alloca i32 ; yields i32*:ptr
5530 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5531 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5532 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5536 '``load``' Instruction
5537 ^^^^^^^^^^^^^^^^^^^^^^
5544 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>]
5545 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5546 !<index> = !{ i32 1 }
5551 The '``load``' instruction is used to read from memory.
5556 The argument to the ``load`` instruction specifies the memory address
5557 from which to load. The pointer must point to a :ref:`first
5558 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5559 then the optimizer is not allowed to modify the number or order of
5560 execution of this ``load`` with other :ref:`volatile
5561 operations <volatile>`.
5563 If the ``load`` is marked as ``atomic``, it takes an extra
5564 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5565 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5566 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5567 when they may see multiple atomic stores. The type of the pointee must
5568 be an integer type whose bit width is a power of two greater than or
5569 equal to eight and less than or equal to a target-specific size limit.
5570 ``align`` must be explicitly specified on atomic loads, and the load has
5571 undefined behavior if the alignment is not set to a value which is at
5572 least the size in bytes of the pointee. ``!nontemporal`` does not have
5573 any defined semantics for atomic loads.
5575 The optional constant ``align`` argument specifies the alignment of the
5576 operation (that is, the alignment of the memory address). A value of 0
5577 or an omitted ``align`` argument means that the operation has the ABI
5578 alignment for the target. It is the responsibility of the code emitter
5579 to ensure that the alignment information is correct. Overestimating the
5580 alignment results in undefined behavior. Underestimating the alignment
5581 may produce less efficient code. An alignment of 1 is always safe. The
5582 maximum possible alignment is ``1 << 29``.
5584 The optional ``!nontemporal`` metadata must reference a single
5585 metadata name ``<index>`` corresponding to a metadata node with one
5586 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5587 metadata on the instruction tells the optimizer and code generator
5588 that this load is not expected to be reused in the cache. The code
5589 generator may select special instructions to save cache bandwidth, such
5590 as the ``MOVNT`` instruction on x86.
5592 The optional ``!invariant.load`` metadata must reference a single
5593 metadata name ``<index>`` corresponding to a metadata node with no
5594 entries. The existence of the ``!invariant.load`` metadata on the
5595 instruction tells the optimizer and code generator that the address
5596 operand to this load points to memory which can be assumed unchanged.
5597 Being invariant does not imply that a location is dereferenceable,
5598 but it does imply that once the location is known dereferenceable
5599 its value is henceforth unchanging.
5601 The optional ``!nonnull`` metadata must reference a single
5602 metadata name ``<index>`` corresponding to a metadata node with no
5603 entries. The existence of the ``!nonnull`` metadata on the
5604 instruction tells the optimizer that the value loaded is known to
5605 never be null. This is analogous to the ''nonnull'' attribute
5606 on parameters and return values. This metadata can only be applied
5607 to loads of a pointer type.
5612 The location of memory pointed to is loaded. If the value being loaded
5613 is of scalar type then the number of bytes read does not exceed the
5614 minimum number of bytes needed to hold all bits of the type. For
5615 example, loading an ``i24`` reads at most three bytes. When loading a
5616 value of a type like ``i20`` with a size that is not an integral number
5617 of bytes, the result is undefined if the value was not originally
5618 written using a store of the same type.
5623 .. code-block:: llvm
5625 %ptr = alloca i32 ; yields i32*:ptr
5626 store i32 3, i32* %ptr ; yields void
5627 %val = load i32* %ptr ; yields i32:val = i32 3
5631 '``store``' Instruction
5632 ^^^^^^^^^^^^^^^^^^^^^^^
5639 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5640 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5645 The '``store``' instruction is used to write to memory.
5650 There are two arguments to the ``store`` instruction: a value to store
5651 and an address at which to store it. The type of the ``<pointer>``
5652 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5653 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5654 then the optimizer is not allowed to modify the number or order of
5655 execution of this ``store`` with other :ref:`volatile
5656 operations <volatile>`.
5658 If the ``store`` is marked as ``atomic``, it takes an extra
5659 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5660 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5661 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5662 when they may see multiple atomic stores. The type of the pointee must
5663 be an integer type whose bit width is a power of two greater than or
5664 equal to eight and less than or equal to a target-specific size limit.
5665 ``align`` must be explicitly specified on atomic stores, and the store
5666 has undefined behavior if the alignment is not set to a value which is
5667 at least the size in bytes of the pointee. ``!nontemporal`` does not
5668 have any defined semantics for atomic stores.
5670 The optional constant ``align`` argument specifies the alignment of the
5671 operation (that is, the alignment of the memory address). A value of 0
5672 or an omitted ``align`` argument means that the operation has the ABI
5673 alignment for the target. It is the responsibility of the code emitter
5674 to ensure that the alignment information is correct. Overestimating the
5675 alignment results in undefined behavior. Underestimating the
5676 alignment may produce less efficient code. An alignment of 1 is always
5677 safe. The maximum possible alignment is ``1 << 29``.
5679 The optional ``!nontemporal`` metadata must reference a single metadata
5680 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5681 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5682 tells the optimizer and code generator that this load is not expected to
5683 be reused in the cache. The code generator may select special
5684 instructions to save cache bandwidth, such as the MOVNT instruction on
5690 The contents of memory are updated to contain ``<value>`` at the
5691 location specified by the ``<pointer>`` operand. If ``<value>`` is
5692 of scalar type then the number of bytes written does not exceed the
5693 minimum number of bytes needed to hold all bits of the type. For
5694 example, storing an ``i24`` writes at most three bytes. When writing a
5695 value of a type like ``i20`` with a size that is not an integral number
5696 of bytes, it is unspecified what happens to the extra bits that do not
5697 belong to the type, but they will typically be overwritten.
5702 .. code-block:: llvm
5704 %ptr = alloca i32 ; yields i32*:ptr
5705 store i32 3, i32* %ptr ; yields void
5706 %val = load i32* %ptr ; yields i32:val = i32 3
5710 '``fence``' Instruction
5711 ^^^^^^^^^^^^^^^^^^^^^^^
5718 fence [singlethread] <ordering> ; yields void
5723 The '``fence``' instruction is used to introduce happens-before edges
5729 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5730 defines what *synchronizes-with* edges they add. They can only be given
5731 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5736 A fence A which has (at least) ``release`` ordering semantics
5737 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5738 semantics if and only if there exist atomic operations X and Y, both
5739 operating on some atomic object M, such that A is sequenced before X, X
5740 modifies M (either directly or through some side effect of a sequence
5741 headed by X), Y is sequenced before B, and Y observes M. This provides a
5742 *happens-before* dependency between A and B. Rather than an explicit
5743 ``fence``, one (but not both) of the atomic operations X or Y might
5744 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5745 still *synchronize-with* the explicit ``fence`` and establish the
5746 *happens-before* edge.
5748 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5749 ``acquire`` and ``release`` semantics specified above, participates in
5750 the global program order of other ``seq_cst`` operations and/or fences.
5752 The optional ":ref:`singlethread <singlethread>`" argument specifies
5753 that the fence only synchronizes with other fences in the same thread.
5754 (This is useful for interacting with signal handlers.)
5759 .. code-block:: llvm
5761 fence acquire ; yields void
5762 fence singlethread seq_cst ; yields void
5766 '``cmpxchg``' Instruction
5767 ^^^^^^^^^^^^^^^^^^^^^^^^^
5774 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5779 The '``cmpxchg``' instruction is used to atomically modify memory. It
5780 loads a value in memory and compares it to a given value. If they are
5781 equal, it tries to store a new value into the memory.
5786 There are three arguments to the '``cmpxchg``' instruction: an address
5787 to operate on, a value to compare to the value currently be at that
5788 address, and a new value to place at that address if the compared values
5789 are equal. The type of '<cmp>' must be an integer type whose bit width
5790 is a power of two greater than or equal to eight and less than or equal
5791 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5792 type, and the type of '<pointer>' must be a pointer to that type. If the
5793 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5794 to modify the number or order of execution of this ``cmpxchg`` with
5795 other :ref:`volatile operations <volatile>`.
5797 The success and failure :ref:`ordering <ordering>` arguments specify how this
5798 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5799 must be at least ``monotonic``, the ordering constraint on failure must be no
5800 stronger than that on success, and the failure ordering cannot be either
5801 ``release`` or ``acq_rel``.
5803 The optional "``singlethread``" argument declares that the ``cmpxchg``
5804 is only atomic with respect to code (usually signal handlers) running in
5805 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5806 respect to all other code in the system.
5808 The pointer passed into cmpxchg must have alignment greater than or
5809 equal to the size in memory of the operand.
5814 The contents of memory at the location specified by the '``<pointer>``' operand
5815 is read and compared to '``<cmp>``'; if the read value is the equal, the
5816 '``<new>``' is written. The original value at the location is returned, together
5817 with a flag indicating success (true) or failure (false).
5819 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5820 permitted: the operation may not write ``<new>`` even if the comparison
5823 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5824 if the value loaded equals ``cmp``.
5826 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5827 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5828 load with an ordering parameter determined the second ordering parameter.
5833 .. code-block:: llvm
5836 %orig = atomic load i32* %ptr unordered ; yields i32
5840 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5841 %squared = mul i32 %cmp, %cmp
5842 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5843 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5844 %success = extractvalue { i32, i1 } %val_success, 1
5845 br i1 %success, label %done, label %loop
5852 '``atomicrmw``' Instruction
5853 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5860 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5865 The '``atomicrmw``' instruction is used to atomically modify memory.
5870 There are three arguments to the '``atomicrmw``' instruction: an
5871 operation to apply, an address whose value to modify, an argument to the
5872 operation. The operation must be one of the following keywords:
5886 The type of '<value>' must be an integer type whose bit width is a power
5887 of two greater than or equal to eight and less than or equal to a
5888 target-specific size limit. The type of the '``<pointer>``' operand must
5889 be a pointer to that type. If the ``atomicrmw`` is marked as
5890 ``volatile``, then the optimizer is not allowed to modify the number or
5891 order of execution of this ``atomicrmw`` with other :ref:`volatile
5892 operations <volatile>`.
5897 The contents of memory at the location specified by the '``<pointer>``'
5898 operand are atomically read, modified, and written back. The original
5899 value at the location is returned. The modification is specified by the
5902 - xchg: ``*ptr = val``
5903 - add: ``*ptr = *ptr + val``
5904 - sub: ``*ptr = *ptr - val``
5905 - and: ``*ptr = *ptr & val``
5906 - nand: ``*ptr = ~(*ptr & val)``
5907 - or: ``*ptr = *ptr | val``
5908 - xor: ``*ptr = *ptr ^ val``
5909 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5910 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5911 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5913 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5919 .. code-block:: llvm
5921 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5923 .. _i_getelementptr:
5925 '``getelementptr``' Instruction
5926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5933 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5934 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5935 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5940 The '``getelementptr``' instruction is used to get the address of a
5941 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5942 address calculation only and does not access memory.
5947 The first argument is always a pointer or a vector of pointers, and
5948 forms the basis of the calculation. The remaining arguments are indices
5949 that indicate which of the elements of the aggregate object are indexed.
5950 The interpretation of each index is dependent on the type being indexed
5951 into. The first index always indexes the pointer value given as the
5952 first argument, the second index indexes a value of the type pointed to
5953 (not necessarily the value directly pointed to, since the first index
5954 can be non-zero), etc. The first type indexed into must be a pointer
5955 value, subsequent types can be arrays, vectors, and structs. Note that
5956 subsequent types being indexed into can never be pointers, since that
5957 would require loading the pointer before continuing calculation.
5959 The type of each index argument depends on the type it is indexing into.
5960 When indexing into a (optionally packed) structure, only ``i32`` integer
5961 **constants** are allowed (when using a vector of indices they must all
5962 be the **same** ``i32`` integer constant). When indexing into an array,
5963 pointer or vector, integers of any width are allowed, and they are not
5964 required to be constant. These integers are treated as signed values
5967 For example, let's consider a C code fragment and how it gets compiled
5983 int *foo(struct ST *s) {
5984 return &s[1].Z.B[5][13];
5987 The LLVM code generated by Clang is:
5989 .. code-block:: llvm
5991 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5992 %struct.ST = type { i32, double, %struct.RT }
5994 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5996 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6003 In the example above, the first index is indexing into the
6004 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6005 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6006 indexes into the third element of the structure, yielding a
6007 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6008 structure. The third index indexes into the second element of the
6009 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6010 dimensions of the array are subscripted into, yielding an '``i32``'
6011 type. The '``getelementptr``' instruction returns a pointer to this
6012 element, thus computing a value of '``i32*``' type.
6014 Note that it is perfectly legal to index partially through a structure,
6015 returning a pointer to an inner element. Because of this, the LLVM code
6016 for the given testcase is equivalent to:
6018 .. code-block:: llvm
6020 define i32* @foo(%struct.ST* %s) {
6021 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6022 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6023 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6024 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6025 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6029 If the ``inbounds`` keyword is present, the result value of the
6030 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6031 pointer is not an *in bounds* address of an allocated object, or if any
6032 of the addresses that would be formed by successive addition of the
6033 offsets implied by the indices to the base address with infinitely
6034 precise signed arithmetic are not an *in bounds* address of that
6035 allocated object. The *in bounds* addresses for an allocated object are
6036 all the addresses that point into the object, plus the address one byte
6037 past the end. In cases where the base is a vector of pointers the
6038 ``inbounds`` keyword applies to each of the computations element-wise.
6040 If the ``inbounds`` keyword is not present, the offsets are added to the
6041 base address with silently-wrapping two's complement arithmetic. If the
6042 offsets have a different width from the pointer, they are sign-extended
6043 or truncated to the width of the pointer. The result value of the
6044 ``getelementptr`` may be outside the object pointed to by the base
6045 pointer. The result value may not necessarily be used to access memory
6046 though, even if it happens to point into allocated storage. See the
6047 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6050 The getelementptr instruction is often confusing. For some more insight
6051 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6056 .. code-block:: llvm
6058 ; yields [12 x i8]*:aptr
6059 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
6061 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6063 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
6065 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
6067 In cases where the pointer argument is a vector of pointers, each index
6068 must be a vector with the same number of elements. For example:
6070 .. code-block:: llvm
6072 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
6074 Conversion Operations
6075 ---------------------
6077 The instructions in this category are the conversion instructions
6078 (casting) which all take a single operand and a type. They perform
6079 various bit conversions on the operand.
6081 '``trunc .. to``' Instruction
6082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6089 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6094 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6099 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6100 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6101 of the same number of integers. The bit size of the ``value`` must be
6102 larger than the bit size of the destination type, ``ty2``. Equal sized
6103 types are not allowed.
6108 The '``trunc``' instruction truncates the high order bits in ``value``
6109 and converts the remaining bits to ``ty2``. Since the source size must
6110 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6111 It will always truncate bits.
6116 .. code-block:: llvm
6118 %X = trunc i32 257 to i8 ; yields i8:1
6119 %Y = trunc i32 123 to i1 ; yields i1:true
6120 %Z = trunc i32 122 to i1 ; yields i1:false
6121 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6123 '``zext .. to``' Instruction
6124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6131 <result> = zext <ty> <value> to <ty2> ; yields ty2
6136 The '``zext``' instruction zero extends its operand to type ``ty2``.
6141 The '``zext``' instruction takes a value to cast, and a type to cast it
6142 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6143 the same number of integers. The bit size of the ``value`` must be
6144 smaller than the bit size of the destination type, ``ty2``.
6149 The ``zext`` fills the high order bits of the ``value`` with zero bits
6150 until it reaches the size of the destination type, ``ty2``.
6152 When zero extending from i1, the result will always be either 0 or 1.
6157 .. code-block:: llvm
6159 %X = zext i32 257 to i64 ; yields i64:257
6160 %Y = zext i1 true to i32 ; yields i32:1
6161 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6163 '``sext .. to``' Instruction
6164 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6171 <result> = sext <ty> <value> to <ty2> ; yields ty2
6176 The '``sext``' sign extends ``value`` to the type ``ty2``.
6181 The '``sext``' instruction takes a value to cast, and a type to cast it
6182 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6183 the same number of integers. The bit size of the ``value`` must be
6184 smaller than the bit size of the destination type, ``ty2``.
6189 The '``sext``' instruction performs a sign extension by copying the sign
6190 bit (highest order bit) of the ``value`` until it reaches the bit size
6191 of the type ``ty2``.
6193 When sign extending from i1, the extension always results in -1 or 0.
6198 .. code-block:: llvm
6200 %X = sext i8 -1 to i16 ; yields i16 :65535
6201 %Y = sext i1 true to i32 ; yields i32:-1
6202 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6204 '``fptrunc .. to``' Instruction
6205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6212 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6217 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6222 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6223 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6224 The size of ``value`` must be larger than the size of ``ty2``. This
6225 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6230 The '``fptrunc``' instruction truncates a ``value`` from a larger
6231 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6232 point <t_floating>` type. If the value cannot fit within the
6233 destination type, ``ty2``, then the results are undefined.
6238 .. code-block:: llvm
6240 %X = fptrunc double 123.0 to float ; yields float:123.0
6241 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6243 '``fpext .. to``' Instruction
6244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6251 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6256 The '``fpext``' extends a floating point ``value`` to a larger floating
6262 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6263 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6264 to. The source type must be smaller than the destination type.
6269 The '``fpext``' instruction extends the ``value`` from a smaller
6270 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6271 point <t_floating>` type. The ``fpext`` cannot be used to make a
6272 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6273 *no-op cast* for a floating point cast.
6278 .. code-block:: llvm
6280 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6281 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6283 '``fptoui .. to``' Instruction
6284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6291 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6296 The '``fptoui``' converts a floating point ``value`` to its unsigned
6297 integer equivalent of type ``ty2``.
6302 The '``fptoui``' instruction takes a value to cast, which must be a
6303 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6304 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6305 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6306 type with the same number of elements as ``ty``
6311 The '``fptoui``' instruction converts its :ref:`floating
6312 point <t_floating>` operand into the nearest (rounding towards zero)
6313 unsigned integer value. If the value cannot fit in ``ty2``, the results
6319 .. code-block:: llvm
6321 %X = fptoui double 123.0 to i32 ; yields i32:123
6322 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6323 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6325 '``fptosi .. to``' Instruction
6326 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6333 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6338 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6339 ``value`` to type ``ty2``.
6344 The '``fptosi``' instruction takes a value to cast, which must be a
6345 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6346 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6347 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6348 type with the same number of elements as ``ty``
6353 The '``fptosi``' instruction converts its :ref:`floating
6354 point <t_floating>` operand into the nearest (rounding towards zero)
6355 signed integer value. If the value cannot fit in ``ty2``, the results
6361 .. code-block:: llvm
6363 %X = fptosi double -123.0 to i32 ; yields i32:-123
6364 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6365 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6367 '``uitofp .. to``' Instruction
6368 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6375 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6380 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6381 and converts that value to the ``ty2`` type.
6386 The '``uitofp``' instruction takes a value to cast, which must be a
6387 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6388 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6389 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6390 type with the same number of elements as ``ty``
6395 The '``uitofp``' instruction interprets its operand as an unsigned
6396 integer quantity and converts it to the corresponding floating point
6397 value. If the value cannot fit in the floating point value, the results
6403 .. code-block:: llvm
6405 %X = uitofp i32 257 to float ; yields float:257.0
6406 %Y = uitofp i8 -1 to double ; yields double:255.0
6408 '``sitofp .. to``' Instruction
6409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6416 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6421 The '``sitofp``' instruction regards ``value`` as a signed integer and
6422 converts that value to the ``ty2`` type.
6427 The '``sitofp``' instruction takes a value to cast, which must be a
6428 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6429 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6430 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6431 type with the same number of elements as ``ty``
6436 The '``sitofp``' instruction interprets its operand as a signed integer
6437 quantity and converts it to the corresponding floating point value. If
6438 the value cannot fit in the floating point value, the results are
6444 .. code-block:: llvm
6446 %X = sitofp i32 257 to float ; yields float:257.0
6447 %Y = sitofp i8 -1 to double ; yields double:-1.0
6451 '``ptrtoint .. to``' Instruction
6452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6459 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6464 The '``ptrtoint``' instruction converts the pointer or a vector of
6465 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6470 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6471 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6472 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6473 a vector of integers type.
6478 The '``ptrtoint``' instruction converts ``value`` to integer type
6479 ``ty2`` by interpreting the pointer value as an integer and either
6480 truncating or zero extending that value to the size of the integer type.
6481 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6482 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6483 the same size, then nothing is done (*no-op cast*) other than a type
6489 .. code-block:: llvm
6491 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6492 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6493 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6497 '``inttoptr .. to``' Instruction
6498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6505 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6510 The '``inttoptr``' instruction converts an integer ``value`` to a
6511 pointer type, ``ty2``.
6516 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6517 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6523 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6524 applying either a zero extension or a truncation depending on the size
6525 of the integer ``value``. If ``value`` is larger than the size of a
6526 pointer then a truncation is done. If ``value`` is smaller than the size
6527 of a pointer then a zero extension is done. If they are the same size,
6528 nothing is done (*no-op cast*).
6533 .. code-block:: llvm
6535 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6536 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6537 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6538 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6542 '``bitcast .. to``' Instruction
6543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6550 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6555 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6561 The '``bitcast``' instruction takes a value to cast, which must be a
6562 non-aggregate first class value, and a type to cast it to, which must
6563 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6564 bit sizes of ``value`` and the destination type, ``ty2``, must be
6565 identical. If the source type is a pointer, the destination type must
6566 also be a pointer of the same size. This instruction supports bitwise
6567 conversion of vectors to integers and to vectors of other types (as
6568 long as they have the same size).
6573 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6574 is always a *no-op cast* because no bits change with this
6575 conversion. The conversion is done as if the ``value`` had been stored
6576 to memory and read back as type ``ty2``. Pointer (or vector of
6577 pointers) types may only be converted to other pointer (or vector of
6578 pointers) types with the same address space through this instruction.
6579 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6580 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6585 .. code-block:: llvm
6587 %X = bitcast i8 255 to i8 ; yields i8 :-1
6588 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6589 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6590 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6592 .. _i_addrspacecast:
6594 '``addrspacecast .. to``' Instruction
6595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6602 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6607 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6608 address space ``n`` to type ``pty2`` in address space ``m``.
6613 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6614 to cast and a pointer type to cast it to, which must have a different
6620 The '``addrspacecast``' instruction converts the pointer value
6621 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6622 value modification, depending on the target and the address space
6623 pair. Pointer conversions within the same address space must be
6624 performed with the ``bitcast`` instruction. Note that if the address space
6625 conversion is legal then both result and operand refer to the same memory
6631 .. code-block:: llvm
6633 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6634 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6635 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6642 The instructions in this category are the "miscellaneous" instructions,
6643 which defy better classification.
6647 '``icmp``' Instruction
6648 ^^^^^^^^^^^^^^^^^^^^^^
6655 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6660 The '``icmp``' instruction returns a boolean value or a vector of
6661 boolean values based on comparison of its two integer, integer vector,
6662 pointer, or pointer vector operands.
6667 The '``icmp``' instruction takes three operands. The first operand is
6668 the condition code indicating the kind of comparison to perform. It is
6669 not a value, just a keyword. The possible condition code are:
6672 #. ``ne``: not equal
6673 #. ``ugt``: unsigned greater than
6674 #. ``uge``: unsigned greater or equal
6675 #. ``ult``: unsigned less than
6676 #. ``ule``: unsigned less or equal
6677 #. ``sgt``: signed greater than
6678 #. ``sge``: signed greater or equal
6679 #. ``slt``: signed less than
6680 #. ``sle``: signed less or equal
6682 The remaining two arguments must be :ref:`integer <t_integer>` or
6683 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6684 must also be identical types.
6689 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6690 code given as ``cond``. The comparison performed always yields either an
6691 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6693 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6694 otherwise. No sign interpretation is necessary or performed.
6695 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6696 otherwise. No sign interpretation is necessary or performed.
6697 #. ``ugt``: interprets the operands as unsigned values and yields
6698 ``true`` if ``op1`` is greater than ``op2``.
6699 #. ``uge``: interprets the operands as unsigned values and yields
6700 ``true`` if ``op1`` is greater than or equal to ``op2``.
6701 #. ``ult``: interprets the operands as unsigned values and yields
6702 ``true`` if ``op1`` is less than ``op2``.
6703 #. ``ule``: interprets the operands as unsigned values and yields
6704 ``true`` if ``op1`` is less than or equal to ``op2``.
6705 #. ``sgt``: interprets the operands as signed values and yields ``true``
6706 if ``op1`` is greater than ``op2``.
6707 #. ``sge``: interprets the operands as signed values and yields ``true``
6708 if ``op1`` is greater than or equal to ``op2``.
6709 #. ``slt``: interprets the operands as signed values and yields ``true``
6710 if ``op1`` is less than ``op2``.
6711 #. ``sle``: interprets the operands as signed values and yields ``true``
6712 if ``op1`` is less than or equal to ``op2``.
6714 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6715 are compared as if they were integers.
6717 If the operands are integer vectors, then they are compared element by
6718 element. The result is an ``i1`` vector with the same number of elements
6719 as the values being compared. Otherwise, the result is an ``i1``.
6724 .. code-block:: llvm
6726 <result> = icmp eq i32 4, 5 ; yields: result=false
6727 <result> = icmp ne float* %X, %X ; yields: result=false
6728 <result> = icmp ult i16 4, 5 ; yields: result=true
6729 <result> = icmp sgt i16 4, 5 ; yields: result=false
6730 <result> = icmp ule i16 -4, 5 ; yields: result=false
6731 <result> = icmp sge i16 4, 5 ; yields: result=false
6733 Note that the code generator does not yet support vector types with the
6734 ``icmp`` instruction.
6738 '``fcmp``' Instruction
6739 ^^^^^^^^^^^^^^^^^^^^^^
6746 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6751 The '``fcmp``' instruction returns a boolean value or vector of boolean
6752 values based on comparison of its operands.
6754 If the operands are floating point scalars, then the result type is a
6755 boolean (:ref:`i1 <t_integer>`).
6757 If the operands are floating point vectors, then the result type is a
6758 vector of boolean with the same number of elements as the operands being
6764 The '``fcmp``' instruction takes three operands. The first operand is
6765 the condition code indicating the kind of comparison to perform. It is
6766 not a value, just a keyword. The possible condition code are:
6768 #. ``false``: no comparison, always returns false
6769 #. ``oeq``: ordered and equal
6770 #. ``ogt``: ordered and greater than
6771 #. ``oge``: ordered and greater than or equal
6772 #. ``olt``: ordered and less than
6773 #. ``ole``: ordered and less than or equal
6774 #. ``one``: ordered and not equal
6775 #. ``ord``: ordered (no nans)
6776 #. ``ueq``: unordered or equal
6777 #. ``ugt``: unordered or greater than
6778 #. ``uge``: unordered or greater than or equal
6779 #. ``ult``: unordered or less than
6780 #. ``ule``: unordered or less than or equal
6781 #. ``une``: unordered or not equal
6782 #. ``uno``: unordered (either nans)
6783 #. ``true``: no comparison, always returns true
6785 *Ordered* means that neither operand is a QNAN while *unordered* means
6786 that either operand may be a QNAN.
6788 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6789 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6790 type. They must have identical types.
6795 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6796 condition code given as ``cond``. If the operands are vectors, then the
6797 vectors are compared element by element. Each comparison performed
6798 always yields an :ref:`i1 <t_integer>` result, as follows:
6800 #. ``false``: always yields ``false``, regardless of operands.
6801 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6802 is equal to ``op2``.
6803 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6804 is greater than ``op2``.
6805 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6806 is greater than or equal to ``op2``.
6807 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6808 is less than ``op2``.
6809 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6810 is less than or equal to ``op2``.
6811 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6812 is not equal to ``op2``.
6813 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6814 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6816 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6817 greater than ``op2``.
6818 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6819 greater than or equal to ``op2``.
6820 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6822 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6823 less than or equal to ``op2``.
6824 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6825 not equal to ``op2``.
6826 #. ``uno``: yields ``true`` if either operand is a QNAN.
6827 #. ``true``: always yields ``true``, regardless of operands.
6832 .. code-block:: llvm
6834 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6835 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6836 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6837 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6839 Note that the code generator does not yet support vector types with the
6840 ``fcmp`` instruction.
6844 '``phi``' Instruction
6845 ^^^^^^^^^^^^^^^^^^^^^
6852 <result> = phi <ty> [ <val0>, <label0>], ...
6857 The '``phi``' instruction is used to implement the φ node in the SSA
6858 graph representing the function.
6863 The type of the incoming values is specified with the first type field.
6864 After this, the '``phi``' instruction takes a list of pairs as
6865 arguments, with one pair for each predecessor basic block of the current
6866 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6867 the value arguments to the PHI node. Only labels may be used as the
6870 There must be no non-phi instructions between the start of a basic block
6871 and the PHI instructions: i.e. PHI instructions must be first in a basic
6874 For the purposes of the SSA form, the use of each incoming value is
6875 deemed to occur on the edge from the corresponding predecessor block to
6876 the current block (but after any definition of an '``invoke``'
6877 instruction's return value on the same edge).
6882 At runtime, the '``phi``' instruction logically takes on the value
6883 specified by the pair corresponding to the predecessor basic block that
6884 executed just prior to the current block.
6889 .. code-block:: llvm
6891 Loop: ; Infinite loop that counts from 0 on up...
6892 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6893 %nextindvar = add i32 %indvar, 1
6898 '``select``' Instruction
6899 ^^^^^^^^^^^^^^^^^^^^^^^^
6906 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6908 selty is either i1 or {<N x i1>}
6913 The '``select``' instruction is used to choose one value based on a
6914 condition, without IR-level branching.
6919 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6920 values indicating the condition, and two values of the same :ref:`first
6921 class <t_firstclass>` type.
6926 If the condition is an i1 and it evaluates to 1, the instruction returns
6927 the first value argument; otherwise, it returns the second value
6930 If the condition is a vector of i1, then the value arguments must be
6931 vectors of the same size, and the selection is done element by element.
6933 If the condition is an i1 and the value arguments are vectors of the
6934 same size, then an entire vector is selected.
6939 .. code-block:: llvm
6941 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6945 '``call``' Instruction
6946 ^^^^^^^^^^^^^^^^^^^^^^
6953 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6958 The '``call``' instruction represents a simple function call.
6963 This instruction requires several arguments:
6965 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6966 should perform tail call optimization. The ``tail`` marker is a hint that
6967 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6968 means that the call must be tail call optimized in order for the program to
6969 be correct. The ``musttail`` marker provides these guarantees:
6971 #. The call will not cause unbounded stack growth if it is part of a
6972 recursive cycle in the call graph.
6973 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6976 Both markers imply that the callee does not access allocas or varargs from
6977 the caller. Calls marked ``musttail`` must obey the following additional
6980 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6981 or a pointer bitcast followed by a ret instruction.
6982 - The ret instruction must return the (possibly bitcasted) value
6983 produced by the call or void.
6984 - The caller and callee prototypes must match. Pointer types of
6985 parameters or return types may differ in pointee type, but not
6987 - The calling conventions of the caller and callee must match.
6988 - All ABI-impacting function attributes, such as sret, byval, inreg,
6989 returned, and inalloca, must match.
6990 - The callee must be varargs iff the caller is varargs. Bitcasting a
6991 non-varargs function to the appropriate varargs type is legal so
6992 long as the non-varargs prefixes obey the other rules.
6994 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6995 the following conditions are met:
6997 - Caller and callee both have the calling convention ``fastcc``.
6998 - The call is in tail position (ret immediately follows call and ret
6999 uses value of call or is void).
7000 - Option ``-tailcallopt`` is enabled, or
7001 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7002 - `Platform-specific constraints are
7003 met. <CodeGenerator.html#tailcallopt>`_
7005 #. The optional "cconv" marker indicates which :ref:`calling
7006 convention <callingconv>` the call should use. If none is
7007 specified, the call defaults to using C calling conventions. The
7008 calling convention of the call must match the calling convention of
7009 the target function, or else the behavior is undefined.
7010 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7011 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7013 #. '``ty``': the type of the call instruction itself which is also the
7014 type of the return value. Functions that return no value are marked
7016 #. '``fnty``': shall be the signature of the pointer to function value
7017 being invoked. The argument types must match the types implied by
7018 this signature. This type can be omitted if the function is not
7019 varargs and if the function type does not return a pointer to a
7021 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7022 be invoked. In most cases, this is a direct function invocation, but
7023 indirect ``call``'s are just as possible, calling an arbitrary pointer
7025 #. '``function args``': argument list whose types match the function
7026 signature argument types and parameter attributes. All arguments must
7027 be of :ref:`first class <t_firstclass>` type. If the function signature
7028 indicates the function accepts a variable number of arguments, the
7029 extra arguments can be specified.
7030 #. The optional :ref:`function attributes <fnattrs>` list. Only
7031 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7032 attributes are valid here.
7037 The '``call``' instruction is used to cause control flow to transfer to
7038 a specified function, with its incoming arguments bound to the specified
7039 values. Upon a '``ret``' instruction in the called function, control
7040 flow continues with the instruction after the function call, and the
7041 return value of the function is bound to the result argument.
7046 .. code-block:: llvm
7048 %retval = call i32 @test(i32 %argc)
7049 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7050 %X = tail call i32 @foo() ; yields i32
7051 %Y = tail call fastcc i32 @foo() ; yields i32
7052 call void %foo(i8 97 signext)
7054 %struct.A = type { i32, i8 }
7055 %r = call %struct.A @foo() ; yields { i32, i8 }
7056 %gr = extractvalue %struct.A %r, 0 ; yields i32
7057 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7058 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7059 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7061 llvm treats calls to some functions with names and arguments that match
7062 the standard C99 library as being the C99 library functions, and may
7063 perform optimizations or generate code for them under that assumption.
7064 This is something we'd like to change in the future to provide better
7065 support for freestanding environments and non-C-based languages.
7069 '``va_arg``' Instruction
7070 ^^^^^^^^^^^^^^^^^^^^^^^^
7077 <resultval> = va_arg <va_list*> <arglist>, <argty>
7082 The '``va_arg``' instruction is used to access arguments passed through
7083 the "variable argument" area of a function call. It is used to implement
7084 the ``va_arg`` macro in C.
7089 This instruction takes a ``va_list*`` value and the type of the
7090 argument. It returns a value of the specified argument type and
7091 increments the ``va_list`` to point to the next argument. The actual
7092 type of ``va_list`` is target specific.
7097 The '``va_arg``' instruction loads an argument of the specified type
7098 from the specified ``va_list`` and causes the ``va_list`` to point to
7099 the next argument. For more information, see the variable argument
7100 handling :ref:`Intrinsic Functions <int_varargs>`.
7102 It is legal for this instruction to be called in a function which does
7103 not take a variable number of arguments, for example, the ``vfprintf``
7106 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7107 function <intrinsics>` because it takes a type as an argument.
7112 See the :ref:`variable argument processing <int_varargs>` section.
7114 Note that the code generator does not yet fully support va\_arg on many
7115 targets. Also, it does not currently support va\_arg with aggregate
7116 types on any target.
7120 '``landingpad``' Instruction
7121 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7128 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
7129 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
7131 <clause> := catch <type> <value>
7132 <clause> := filter <array constant type> <array constant>
7137 The '``landingpad``' instruction is used by `LLVM's exception handling
7138 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7139 is a landing pad --- one where the exception lands, and corresponds to the
7140 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7141 defines values supplied by the personality function (``pers_fn``) upon
7142 re-entry to the function. The ``resultval`` has the type ``resultty``.
7147 This instruction takes a ``pers_fn`` value. This is the personality
7148 function associated with the unwinding mechanism. The optional
7149 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7151 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7152 contains the global variable representing the "type" that may be caught
7153 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7154 clause takes an array constant as its argument. Use
7155 "``[0 x i8**] undef``" for a filter which cannot throw. The
7156 '``landingpad``' instruction must contain *at least* one ``clause`` or
7157 the ``cleanup`` flag.
7162 The '``landingpad``' instruction defines the values which are set by the
7163 personality function (``pers_fn``) upon re-entry to the function, and
7164 therefore the "result type" of the ``landingpad`` instruction. As with
7165 calling conventions, how the personality function results are
7166 represented in LLVM IR is target specific.
7168 The clauses are applied in order from top to bottom. If two
7169 ``landingpad`` instructions are merged together through inlining, the
7170 clauses from the calling function are appended to the list of clauses.
7171 When the call stack is being unwound due to an exception being thrown,
7172 the exception is compared against each ``clause`` in turn. If it doesn't
7173 match any of the clauses, and the ``cleanup`` flag is not set, then
7174 unwinding continues further up the call stack.
7176 The ``landingpad`` instruction has several restrictions:
7178 - A landing pad block is a basic block which is the unwind destination
7179 of an '``invoke``' instruction.
7180 - A landing pad block must have a '``landingpad``' instruction as its
7181 first non-PHI instruction.
7182 - There can be only one '``landingpad``' instruction within the landing
7184 - A basic block that is not a landing pad block may not include a
7185 '``landingpad``' instruction.
7186 - All '``landingpad``' instructions in a function must have the same
7187 personality function.
7192 .. code-block:: llvm
7194 ;; A landing pad which can catch an integer.
7195 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7197 ;; A landing pad that is a cleanup.
7198 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7200 ;; A landing pad which can catch an integer and can only throw a double.
7201 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7203 filter [1 x i8**] [@_ZTId]
7210 LLVM supports the notion of an "intrinsic function". These functions
7211 have well known names and semantics and are required to follow certain
7212 restrictions. Overall, these intrinsics represent an extension mechanism
7213 for the LLVM language that does not require changing all of the
7214 transformations in LLVM when adding to the language (or the bitcode
7215 reader/writer, the parser, etc...).
7217 Intrinsic function names must all start with an "``llvm.``" prefix. This
7218 prefix is reserved in LLVM for intrinsic names; thus, function names may
7219 not begin with this prefix. Intrinsic functions must always be external
7220 functions: you cannot define the body of intrinsic functions. Intrinsic
7221 functions may only be used in call or invoke instructions: it is illegal
7222 to take the address of an intrinsic function. Additionally, because
7223 intrinsic functions are part of the LLVM language, it is required if any
7224 are added that they be documented here.
7226 Some intrinsic functions can be overloaded, i.e., the intrinsic
7227 represents a family of functions that perform the same operation but on
7228 different data types. Because LLVM can represent over 8 million
7229 different integer types, overloading is used commonly to allow an
7230 intrinsic function to operate on any integer type. One or more of the
7231 argument types or the result type can be overloaded to accept any
7232 integer type. Argument types may also be defined as exactly matching a
7233 previous argument's type or the result type. This allows an intrinsic
7234 function which accepts multiple arguments, but needs all of them to be
7235 of the same type, to only be overloaded with respect to a single
7236 argument or the result.
7238 Overloaded intrinsics will have the names of its overloaded argument
7239 types encoded into its function name, each preceded by a period. Only
7240 those types which are overloaded result in a name suffix. Arguments
7241 whose type is matched against another type do not. For example, the
7242 ``llvm.ctpop`` function can take an integer of any width and returns an
7243 integer of exactly the same integer width. This leads to a family of
7244 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7245 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7246 overloaded, and only one type suffix is required. Because the argument's
7247 type is matched against the return type, it does not require its own
7250 To learn how to add an intrinsic function, please see the `Extending
7251 LLVM Guide <ExtendingLLVM.html>`_.
7255 Variable Argument Handling Intrinsics
7256 -------------------------------------
7258 Variable argument support is defined in LLVM with the
7259 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7260 functions. These functions are related to the similarly named macros
7261 defined in the ``<stdarg.h>`` header file.
7263 All of these functions operate on arguments that use a target-specific
7264 value type "``va_list``". The LLVM assembly language reference manual
7265 does not define what this type is, so all transformations should be
7266 prepared to handle these functions regardless of the type used.
7268 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7269 variable argument handling intrinsic functions are used.
7271 .. code-block:: llvm
7273 ; This struct is different for every platform. For most platforms,
7274 ; it is merely an i8*.
7275 %struct.va_list = type { i8* }
7277 ; For Unix x86_64 platforms, va_list is the following struct:
7278 ; %struct.va_list = type { i32, i32, i8*, i8* }
7280 define i32 @test(i32 %X, ...) {
7281 ; Initialize variable argument processing
7282 %ap = alloca %struct.va_list
7283 %ap2 = bitcast %struct.va_list* %ap to i8*
7284 call void @llvm.va_start(i8* %ap2)
7286 ; Read a single integer argument
7287 %tmp = va_arg i8* %ap2, i32
7289 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7291 %aq2 = bitcast i8** %aq to i8*
7292 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7293 call void @llvm.va_end(i8* %aq2)
7295 ; Stop processing of arguments.
7296 call void @llvm.va_end(i8* %ap2)
7300 declare void @llvm.va_start(i8*)
7301 declare void @llvm.va_copy(i8*, i8*)
7302 declare void @llvm.va_end(i8*)
7306 '``llvm.va_start``' Intrinsic
7307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7314 declare void @llvm.va_start(i8* <arglist>)
7319 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7320 subsequent use by ``va_arg``.
7325 The argument is a pointer to a ``va_list`` element to initialize.
7330 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7331 available in C. In a target-dependent way, it initializes the
7332 ``va_list`` element to which the argument points, so that the next call
7333 to ``va_arg`` will produce the first variable argument passed to the
7334 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7335 to know the last argument of the function as the compiler can figure
7338 '``llvm.va_end``' Intrinsic
7339 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7346 declare void @llvm.va_end(i8* <arglist>)
7351 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7352 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7357 The argument is a pointer to a ``va_list`` to destroy.
7362 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7363 available in C. In a target-dependent way, it destroys the ``va_list``
7364 element to which the argument points. Calls to
7365 :ref:`llvm.va_start <int_va_start>` and
7366 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7371 '``llvm.va_copy``' Intrinsic
7372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7379 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7384 The '``llvm.va_copy``' intrinsic copies the current argument position
7385 from the source argument list to the destination argument list.
7390 The first argument is a pointer to a ``va_list`` element to initialize.
7391 The second argument is a pointer to a ``va_list`` element to copy from.
7396 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7397 available in C. In a target-dependent way, it copies the source
7398 ``va_list`` element into the destination ``va_list`` element. This
7399 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7400 arbitrarily complex and require, for example, memory allocation.
7402 Accurate Garbage Collection Intrinsics
7403 --------------------------------------
7405 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7406 (GC) requires the frontend to generate code containing appropriate intrinsic
7407 calls and select an appropriate GC strategy which knows how to lower these
7408 intrinsics in a manner which is appropriate for the target collector.
7410 These intrinsics allow identification of :ref:`GC roots on the
7411 stack <int_gcroot>`, as well as garbage collector implementations that
7412 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7413 Frontends for type-safe garbage collected languages should generate
7414 these intrinsics to make use of the LLVM garbage collectors. For more
7415 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7417 Experimental Statepoint Intrinsics
7418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7420 LLVM provides an second experimental set of intrinsics for describing garbage
7421 collection safepoints in compiled code. These intrinsics are an alternative
7422 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7423 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7424 differences in approach are covered in the `Garbage Collection with LLVM
7425 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7426 described in :doc:`Statepoints`.
7430 '``llvm.gcroot``' Intrinsic
7431 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7438 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7443 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7444 the code generator, and allows some metadata to be associated with it.
7449 The first argument specifies the address of a stack object that contains
7450 the root pointer. The second pointer (which must be either a constant or
7451 a global value address) contains the meta-data to be associated with the
7457 At runtime, a call to this intrinsic stores a null pointer into the
7458 "ptrloc" location. At compile-time, the code generator generates
7459 information to allow the runtime to find the pointer at GC safe points.
7460 The '``llvm.gcroot``' intrinsic may only be used in a function which
7461 :ref:`specifies a GC algorithm <gc>`.
7465 '``llvm.gcread``' Intrinsic
7466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7473 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7478 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7479 locations, allowing garbage collector implementations that require read
7485 The second argument is the address to read from, which should be an
7486 address allocated from the garbage collector. The first object is a
7487 pointer to the start of the referenced object, if needed by the language
7488 runtime (otherwise null).
7493 The '``llvm.gcread``' intrinsic has the same semantics as a load
7494 instruction, but may be replaced with substantially more complex code by
7495 the garbage collector runtime, as needed. The '``llvm.gcread``'
7496 intrinsic may only be used in a function which :ref:`specifies a GC
7501 '``llvm.gcwrite``' Intrinsic
7502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7509 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7514 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7515 locations, allowing garbage collector implementations that require write
7516 barriers (such as generational or reference counting collectors).
7521 The first argument is the reference to store, the second is the start of
7522 the object to store it to, and the third is the address of the field of
7523 Obj to store to. If the runtime does not require a pointer to the
7524 object, Obj may be null.
7529 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7530 instruction, but may be replaced with substantially more complex code by
7531 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7532 intrinsic may only be used in a function which :ref:`specifies a GC
7535 Code Generator Intrinsics
7536 -------------------------
7538 These intrinsics are provided by LLVM to expose special features that
7539 may only be implemented with code generator support.
7541 '``llvm.returnaddress``' Intrinsic
7542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7549 declare i8 *@llvm.returnaddress(i32 <level>)
7554 The '``llvm.returnaddress``' intrinsic attempts to compute a
7555 target-specific value indicating the return address of the current
7556 function or one of its callers.
7561 The argument to this intrinsic indicates which function to return the
7562 address for. Zero indicates the calling function, one indicates its
7563 caller, etc. The argument is **required** to be a constant integer
7569 The '``llvm.returnaddress``' intrinsic either returns a pointer
7570 indicating the return address of the specified call frame, or zero if it
7571 cannot be identified. The value returned by this intrinsic is likely to
7572 be incorrect or 0 for arguments other than zero, so it should only be
7573 used for debugging purposes.
7575 Note that calling this intrinsic does not prevent function inlining or
7576 other aggressive transformations, so the value returned may not be that
7577 of the obvious source-language caller.
7579 '``llvm.frameaddress``' Intrinsic
7580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7587 declare i8* @llvm.frameaddress(i32 <level>)
7592 The '``llvm.frameaddress``' intrinsic attempts to return the
7593 target-specific frame pointer value for the specified stack frame.
7598 The argument to this intrinsic indicates which function to return the
7599 frame pointer for. Zero indicates the calling function, one indicates
7600 its caller, etc. The argument is **required** to be a constant integer
7606 The '``llvm.frameaddress``' intrinsic either returns a pointer
7607 indicating the frame address of the specified call frame, or zero if it
7608 cannot be identified. The value returned by this intrinsic is likely to
7609 be incorrect or 0 for arguments other than zero, so it should only be
7610 used for debugging purposes.
7612 Note that calling this intrinsic does not prevent function inlining or
7613 other aggressive transformations, so the value returned may not be that
7614 of the obvious source-language caller.
7616 '``llvm.frameallocate``' and '``llvm.framerecover``' Intrinsics
7617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7624 declare i8* @llvm.frameallocate(i32 %size)
7625 declare i8* @llvm.framerecover(i8* %func, i8* %fp)
7630 The '``llvm.frameallocate``' intrinsic allocates stack memory at some fixed
7631 offset from the frame pointer, and the '``llvm.framerecover``'
7632 intrinsic applies that offset to a live frame pointer to recover the address of
7633 the allocation. The offset is computed during frame layout of the caller of
7634 ``llvm.frameallocate``.
7639 The ``size`` argument to '``llvm.frameallocate``' must be a constant integer
7640 indicating the amount of stack memory to allocate. As with allocas, allocating
7641 zero bytes is legal, but the result is undefined.
7643 The ``func`` argument to '``llvm.framerecover``' must be a constant
7644 bitcasted pointer to a function defined in the current module. The code
7645 generator cannot determine the frame allocation offset of functions defined in
7648 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7649 pointer of a call frame that is currently live. The return value of
7650 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7651 also expose the frame pointer through stack unwinding mechanisms.
7656 These intrinsics allow a group of functions to access one stack memory
7657 allocation in an ancestor stack frame. The memory returned from
7658 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7659 memory is only aligned to the ABI-required stack alignment. Each function may
7660 only call '``llvm.frameallocate``' one or zero times from the function entry
7661 block. The frame allocation intrinsic inhibits inlining, as any frame
7662 allocations in the inlined function frame are likely to be at a different
7663 offset from the one used by '``llvm.framerecover``' called with the
7666 .. _int_read_register:
7667 .. _int_write_register:
7669 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7677 declare i32 @llvm.read_register.i32(metadata)
7678 declare i64 @llvm.read_register.i64(metadata)
7679 declare void @llvm.write_register.i32(metadata, i32 @value)
7680 declare void @llvm.write_register.i64(metadata, i64 @value)
7686 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7687 provides access to the named register. The register must be valid on
7688 the architecture being compiled to. The type needs to be compatible
7689 with the register being read.
7694 The '``llvm.read_register``' intrinsic returns the current value of the
7695 register, where possible. The '``llvm.write_register``' intrinsic sets
7696 the current value of the register, where possible.
7698 This is useful to implement named register global variables that need
7699 to always be mapped to a specific register, as is common practice on
7700 bare-metal programs including OS kernels.
7702 The compiler doesn't check for register availability or use of the used
7703 register in surrounding code, including inline assembly. Because of that,
7704 allocatable registers are not supported.
7706 Warning: So far it only works with the stack pointer on selected
7707 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7708 work is needed to support other registers and even more so, allocatable
7713 '``llvm.stacksave``' Intrinsic
7714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7721 declare i8* @llvm.stacksave()
7726 The '``llvm.stacksave``' intrinsic is used to remember the current state
7727 of the function stack, for use with
7728 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7729 implementing language features like scoped automatic variable sized
7735 This intrinsic returns a opaque pointer value that can be passed to
7736 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7737 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7738 ``llvm.stacksave``, it effectively restores the state of the stack to
7739 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7740 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7741 were allocated after the ``llvm.stacksave`` was executed.
7743 .. _int_stackrestore:
7745 '``llvm.stackrestore``' Intrinsic
7746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7753 declare void @llvm.stackrestore(i8* %ptr)
7758 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7759 the function stack to the state it was in when the corresponding
7760 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7761 useful for implementing language features like scoped automatic variable
7762 sized arrays in C99.
7767 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7769 '``llvm.prefetch``' Intrinsic
7770 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7777 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7782 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7783 insert a prefetch instruction if supported; otherwise, it is a noop.
7784 Prefetches have no effect on the behavior of the program but can change
7785 its performance characteristics.
7790 ``address`` is the address to be prefetched, ``rw`` is the specifier
7791 determining if the fetch should be for a read (0) or write (1), and
7792 ``locality`` is a temporal locality specifier ranging from (0) - no
7793 locality, to (3) - extremely local keep in cache. The ``cache type``
7794 specifies whether the prefetch is performed on the data (1) or
7795 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7796 arguments must be constant integers.
7801 This intrinsic does not modify the behavior of the program. In
7802 particular, prefetches cannot trap and do not produce a value. On
7803 targets that support this intrinsic, the prefetch can provide hints to
7804 the processor cache for better performance.
7806 '``llvm.pcmarker``' Intrinsic
7807 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7814 declare void @llvm.pcmarker(i32 <id>)
7819 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7820 Counter (PC) in a region of code to simulators and other tools. The
7821 method is target specific, but it is expected that the marker will use
7822 exported symbols to transmit the PC of the marker. The marker makes no
7823 guarantees that it will remain with any specific instruction after
7824 optimizations. It is possible that the presence of a marker will inhibit
7825 optimizations. The intended use is to be inserted after optimizations to
7826 allow correlations of simulation runs.
7831 ``id`` is a numerical id identifying the marker.
7836 This intrinsic does not modify the behavior of the program. Backends
7837 that do not support this intrinsic may ignore it.
7839 '``llvm.readcyclecounter``' Intrinsic
7840 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7847 declare i64 @llvm.readcyclecounter()
7852 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7853 counter register (or similar low latency, high accuracy clocks) on those
7854 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7855 should map to RPCC. As the backing counters overflow quickly (on the
7856 order of 9 seconds on alpha), this should only be used for small
7862 When directly supported, reading the cycle counter should not modify any
7863 memory. Implementations are allowed to either return a application
7864 specific value or a system wide value. On backends without support, this
7865 is lowered to a constant 0.
7867 Note that runtime support may be conditional on the privilege-level code is
7868 running at and the host platform.
7870 '``llvm.clear_cache``' Intrinsic
7871 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7878 declare void @llvm.clear_cache(i8*, i8*)
7883 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7884 in the specified range to the execution unit of the processor. On
7885 targets with non-unified instruction and data cache, the implementation
7886 flushes the instruction cache.
7891 On platforms with coherent instruction and data caches (e.g. x86), this
7892 intrinsic is a nop. On platforms with non-coherent instruction and data
7893 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7894 instructions or a system call, if cache flushing requires special
7897 The default behavior is to emit a call to ``__clear_cache`` from the run
7900 This instrinsic does *not* empty the instruction pipeline. Modifications
7901 of the current function are outside the scope of the intrinsic.
7903 '``llvm.instrprof_increment``' Intrinsic
7904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7911 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
7912 i32 <num-counters>, i32 <index>)
7917 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
7918 frontend for use with instrumentation based profiling. These will be
7919 lowered by the ``-instrprof`` pass to generate execution counts of a
7925 The first argument is a pointer to a global variable containing the
7926 name of the entity being instrumented. This should generally be the
7927 (mangled) function name for a set of counters.
7929 The second argument is a hash value that can be used by the consumer
7930 of the profile data to detect changes to the instrumented source, and
7931 the third is the number of counters associated with ``name``. It is an
7932 error if ``hash`` or ``num-counters`` differ between two instances of
7933 ``instrprof_increment`` that refer to the same name.
7935 The last argument refers to which of the counters for ``name`` should
7936 be incremented. It should be a value between 0 and ``num-counters``.
7941 This intrinsic represents an increment of a profiling counter. It will
7942 cause the ``-instrprof`` pass to generate the appropriate data
7943 structures and the code to increment the appropriate value, in a
7944 format that can be written out by a compiler runtime and consumed via
7945 the ``llvm-profdata`` tool.
7947 Standard C Library Intrinsics
7948 -----------------------------
7950 LLVM provides intrinsics for a few important standard C library
7951 functions. These intrinsics allow source-language front-ends to pass
7952 information about the alignment of the pointer arguments to the code
7953 generator, providing opportunity for more efficient code generation.
7957 '``llvm.memcpy``' Intrinsic
7958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7963 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7964 integer bit width and for different address spaces. Not all targets
7965 support all bit widths however.
7969 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7970 i32 <len>, i32 <align>, i1 <isvolatile>)
7971 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7972 i64 <len>, i32 <align>, i1 <isvolatile>)
7977 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7978 source location to the destination location.
7980 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7981 intrinsics do not return a value, takes extra alignment/isvolatile
7982 arguments and the pointers can be in specified address spaces.
7987 The first argument is a pointer to the destination, the second is a
7988 pointer to the source. The third argument is an integer argument
7989 specifying the number of bytes to copy, the fourth argument is the
7990 alignment of the source and destination locations, and the fifth is a
7991 boolean indicating a volatile access.
7993 If the call to this intrinsic has an alignment value that is not 0 or 1,
7994 then the caller guarantees that both the source and destination pointers
7995 are aligned to that boundary.
7997 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7998 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7999 very cleanly specified and it is unwise to depend on it.
8004 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8005 source location to the destination location, which are not allowed to
8006 overlap. It copies "len" bytes of memory over. If the argument is known
8007 to be aligned to some boundary, this can be specified as the fourth
8008 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8010 '``llvm.memmove``' Intrinsic
8011 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8016 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8017 bit width and for different address space. Not all targets support all
8022 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8023 i32 <len>, i32 <align>, i1 <isvolatile>)
8024 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8025 i64 <len>, i32 <align>, i1 <isvolatile>)
8030 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8031 source location to the destination location. It is similar to the
8032 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8035 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8036 intrinsics do not return a value, takes extra alignment/isvolatile
8037 arguments and the pointers can be in specified address spaces.
8042 The first argument is a pointer to the destination, the second is a
8043 pointer to the source. The third argument is an integer argument
8044 specifying the number of bytes to copy, the fourth argument is the
8045 alignment of the source and destination locations, and the fifth is a
8046 boolean indicating a volatile access.
8048 If the call to this intrinsic has an alignment value that is not 0 or 1,
8049 then the caller guarantees that the source and destination pointers are
8050 aligned to that boundary.
8052 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8053 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8054 not very cleanly specified and it is unwise to depend on it.
8059 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8060 source location to the destination location, which may overlap. It
8061 copies "len" bytes of memory over. If the argument is known to be
8062 aligned to some boundary, this can be specified as the fourth argument,
8063 otherwise it should be set to 0 or 1 (both meaning no alignment).
8065 '``llvm.memset.*``' Intrinsics
8066 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8071 This is an overloaded intrinsic. You can use llvm.memset on any integer
8072 bit width and for different address spaces. However, not all targets
8073 support all bit widths.
8077 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8078 i32 <len>, i32 <align>, i1 <isvolatile>)
8079 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8080 i64 <len>, i32 <align>, i1 <isvolatile>)
8085 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8086 particular byte value.
8088 Note that, unlike the standard libc function, the ``llvm.memset``
8089 intrinsic does not return a value and takes extra alignment/volatile
8090 arguments. Also, the destination can be in an arbitrary address space.
8095 The first argument is a pointer to the destination to fill, the second
8096 is the byte value with which to fill it, the third argument is an
8097 integer argument specifying the number of bytes to fill, and the fourth
8098 argument is the known alignment of the destination location.
8100 If the call to this intrinsic has an alignment value that is not 0 or 1,
8101 then the caller guarantees that the destination pointer is aligned to
8104 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8105 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8106 very cleanly specified and it is unwise to depend on it.
8111 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8112 at the destination location. If the argument is known to be aligned to
8113 some boundary, this can be specified as the fourth argument, otherwise
8114 it should be set to 0 or 1 (both meaning no alignment).
8116 '``llvm.sqrt.*``' Intrinsic
8117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8122 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8123 floating point or vector of floating point type. Not all targets support
8128 declare float @llvm.sqrt.f32(float %Val)
8129 declare double @llvm.sqrt.f64(double %Val)
8130 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8131 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8132 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8137 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8138 returning the same value as the libm '``sqrt``' functions would. Unlike
8139 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8140 negative numbers other than -0.0 (which allows for better optimization,
8141 because there is no need to worry about errno being set).
8142 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8147 The argument and return value are floating point numbers of the same
8153 This function returns the sqrt of the specified operand if it is a
8154 nonnegative floating point number.
8156 '``llvm.powi.*``' Intrinsic
8157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8162 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8163 floating point or vector of floating point type. Not all targets support
8168 declare float @llvm.powi.f32(float %Val, i32 %power)
8169 declare double @llvm.powi.f64(double %Val, i32 %power)
8170 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8171 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8172 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8177 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8178 specified (positive or negative) power. The order of evaluation of
8179 multiplications is not defined. When a vector of floating point type is
8180 used, the second argument remains a scalar integer value.
8185 The second argument is an integer power, and the first is a value to
8186 raise to that power.
8191 This function returns the first value raised to the second power with an
8192 unspecified sequence of rounding operations.
8194 '``llvm.sin.*``' Intrinsic
8195 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8200 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8201 floating point or vector of floating point type. Not all targets support
8206 declare float @llvm.sin.f32(float %Val)
8207 declare double @llvm.sin.f64(double %Val)
8208 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8209 declare fp128 @llvm.sin.f128(fp128 %Val)
8210 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8215 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8220 The argument and return value are floating point numbers of the same
8226 This function returns the sine of the specified operand, returning the
8227 same values as the libm ``sin`` functions would, and handles error
8228 conditions in the same way.
8230 '``llvm.cos.*``' Intrinsic
8231 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8236 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8237 floating point or vector of floating point type. Not all targets support
8242 declare float @llvm.cos.f32(float %Val)
8243 declare double @llvm.cos.f64(double %Val)
8244 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8245 declare fp128 @llvm.cos.f128(fp128 %Val)
8246 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8251 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8256 The argument and return value are floating point numbers of the same
8262 This function returns the cosine of the specified operand, returning the
8263 same values as the libm ``cos`` functions would, and handles error
8264 conditions in the same way.
8266 '``llvm.pow.*``' Intrinsic
8267 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8272 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8273 floating point or vector of floating point type. Not all targets support
8278 declare float @llvm.pow.f32(float %Val, float %Power)
8279 declare double @llvm.pow.f64(double %Val, double %Power)
8280 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8281 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8282 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8287 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8288 specified (positive or negative) power.
8293 The second argument is a floating point power, and the first is a value
8294 to raise to that power.
8299 This function returns the first value raised to the second power,
8300 returning the same values as the libm ``pow`` functions would, and
8301 handles error conditions in the same way.
8303 '``llvm.exp.*``' Intrinsic
8304 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8309 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8310 floating point or vector of floating point type. Not all targets support
8315 declare float @llvm.exp.f32(float %Val)
8316 declare double @llvm.exp.f64(double %Val)
8317 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8318 declare fp128 @llvm.exp.f128(fp128 %Val)
8319 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8324 The '``llvm.exp.*``' intrinsics perform the exp function.
8329 The argument and return value are floating point numbers of the same
8335 This function returns the same values as the libm ``exp`` functions
8336 would, and handles error conditions in the same way.
8338 '``llvm.exp2.*``' Intrinsic
8339 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8344 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8345 floating point or vector of floating point type. Not all targets support
8350 declare float @llvm.exp2.f32(float %Val)
8351 declare double @llvm.exp2.f64(double %Val)
8352 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8353 declare fp128 @llvm.exp2.f128(fp128 %Val)
8354 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8359 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8364 The argument and return value are floating point numbers of the same
8370 This function returns the same values as the libm ``exp2`` functions
8371 would, and handles error conditions in the same way.
8373 '``llvm.log.*``' Intrinsic
8374 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8379 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8380 floating point or vector of floating point type. Not all targets support
8385 declare float @llvm.log.f32(float %Val)
8386 declare double @llvm.log.f64(double %Val)
8387 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8388 declare fp128 @llvm.log.f128(fp128 %Val)
8389 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8394 The '``llvm.log.*``' intrinsics perform the log function.
8399 The argument and return value are floating point numbers of the same
8405 This function returns the same values as the libm ``log`` functions
8406 would, and handles error conditions in the same way.
8408 '``llvm.log10.*``' Intrinsic
8409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8414 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8415 floating point or vector of floating point type. Not all targets support
8420 declare float @llvm.log10.f32(float %Val)
8421 declare double @llvm.log10.f64(double %Val)
8422 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8423 declare fp128 @llvm.log10.f128(fp128 %Val)
8424 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8429 The '``llvm.log10.*``' intrinsics perform the log10 function.
8434 The argument and return value are floating point numbers of the same
8440 This function returns the same values as the libm ``log10`` functions
8441 would, and handles error conditions in the same way.
8443 '``llvm.log2.*``' Intrinsic
8444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8449 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8450 floating point or vector of floating point type. Not all targets support
8455 declare float @llvm.log2.f32(float %Val)
8456 declare double @llvm.log2.f64(double %Val)
8457 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8458 declare fp128 @llvm.log2.f128(fp128 %Val)
8459 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8464 The '``llvm.log2.*``' intrinsics perform the log2 function.
8469 The argument and return value are floating point numbers of the same
8475 This function returns the same values as the libm ``log2`` functions
8476 would, and handles error conditions in the same way.
8478 '``llvm.fma.*``' Intrinsic
8479 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8484 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8485 floating point or vector of floating point type. Not all targets support
8490 declare float @llvm.fma.f32(float %a, float %b, float %c)
8491 declare double @llvm.fma.f64(double %a, double %b, double %c)
8492 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8493 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8494 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8499 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8505 The argument and return value are floating point numbers of the same
8511 This function returns the same values as the libm ``fma`` functions
8512 would, and does not set errno.
8514 '``llvm.fabs.*``' Intrinsic
8515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8520 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8521 floating point or vector of floating point type. Not all targets support
8526 declare float @llvm.fabs.f32(float %Val)
8527 declare double @llvm.fabs.f64(double %Val)
8528 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8529 declare fp128 @llvm.fabs.f128(fp128 %Val)
8530 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8535 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8541 The argument and return value are floating point numbers of the same
8547 This function returns the same values as the libm ``fabs`` functions
8548 would, and handles error conditions in the same way.
8550 '``llvm.minnum.*``' Intrinsic
8551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8556 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8557 floating point or vector of floating point type. Not all targets support
8562 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8563 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8564 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8565 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8566 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8571 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8578 The arguments and return value are floating point numbers of the same
8584 Follows the IEEE-754 semantics for minNum, which also match for libm's
8587 If either operand is a NaN, returns the other non-NaN operand. Returns
8588 NaN only if both operands are NaN. If the operands compare equal,
8589 returns a value that compares equal to both operands. This means that
8590 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8592 '``llvm.maxnum.*``' Intrinsic
8593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8598 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8599 floating point or vector of floating point type. Not all targets support
8604 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8605 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8606 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8607 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8608 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8613 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8620 The arguments and return value are floating point numbers of the same
8625 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8628 If either operand is a NaN, returns the other non-NaN operand. Returns
8629 NaN only if both operands are NaN. If the operands compare equal,
8630 returns a value that compares equal to both operands. This means that
8631 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8633 '``llvm.copysign.*``' Intrinsic
8634 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8639 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8640 floating point or vector of floating point type. Not all targets support
8645 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8646 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8647 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8648 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8649 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8654 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8655 first operand and the sign of the second operand.
8660 The arguments and return value are floating point numbers of the same
8666 This function returns the same values as the libm ``copysign``
8667 functions would, and handles error conditions in the same way.
8669 '``llvm.floor.*``' Intrinsic
8670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8675 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8676 floating point or vector of floating point type. Not all targets support
8681 declare float @llvm.floor.f32(float %Val)
8682 declare double @llvm.floor.f64(double %Val)
8683 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8684 declare fp128 @llvm.floor.f128(fp128 %Val)
8685 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8690 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8695 The argument and return value are floating point numbers of the same
8701 This function returns the same values as the libm ``floor`` functions
8702 would, and handles error conditions in the same way.
8704 '``llvm.ceil.*``' Intrinsic
8705 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8710 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8711 floating point or vector of floating point type. Not all targets support
8716 declare float @llvm.ceil.f32(float %Val)
8717 declare double @llvm.ceil.f64(double %Val)
8718 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8719 declare fp128 @llvm.ceil.f128(fp128 %Val)
8720 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8725 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8730 The argument and return value are floating point numbers of the same
8736 This function returns the same values as the libm ``ceil`` functions
8737 would, and handles error conditions in the same way.
8739 '``llvm.trunc.*``' Intrinsic
8740 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8745 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8746 floating point or vector of floating point type. Not all targets support
8751 declare float @llvm.trunc.f32(float %Val)
8752 declare double @llvm.trunc.f64(double %Val)
8753 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8754 declare fp128 @llvm.trunc.f128(fp128 %Val)
8755 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8760 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8761 nearest integer not larger in magnitude than the operand.
8766 The argument and return value are floating point numbers of the same
8772 This function returns the same values as the libm ``trunc`` functions
8773 would, and handles error conditions in the same way.
8775 '``llvm.rint.*``' Intrinsic
8776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8781 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8782 floating point or vector of floating point type. Not all targets support
8787 declare float @llvm.rint.f32(float %Val)
8788 declare double @llvm.rint.f64(double %Val)
8789 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8790 declare fp128 @llvm.rint.f128(fp128 %Val)
8791 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8796 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8797 nearest integer. It may raise an inexact floating-point exception if the
8798 operand isn't an integer.
8803 The argument and return value are floating point numbers of the same
8809 This function returns the same values as the libm ``rint`` functions
8810 would, and handles error conditions in the same way.
8812 '``llvm.nearbyint.*``' Intrinsic
8813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8818 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8819 floating point or vector of floating point type. Not all targets support
8824 declare float @llvm.nearbyint.f32(float %Val)
8825 declare double @llvm.nearbyint.f64(double %Val)
8826 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8827 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8828 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8833 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8839 The argument and return value are floating point numbers of the same
8845 This function returns the same values as the libm ``nearbyint``
8846 functions would, and handles error conditions in the same way.
8848 '``llvm.round.*``' Intrinsic
8849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8854 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8855 floating point or vector of floating point type. Not all targets support
8860 declare float @llvm.round.f32(float %Val)
8861 declare double @llvm.round.f64(double %Val)
8862 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8863 declare fp128 @llvm.round.f128(fp128 %Val)
8864 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8869 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8875 The argument and return value are floating point numbers of the same
8881 This function returns the same values as the libm ``round``
8882 functions would, and handles error conditions in the same way.
8884 Bit Manipulation Intrinsics
8885 ---------------------------
8887 LLVM provides intrinsics for a few important bit manipulation
8888 operations. These allow efficient code generation for some algorithms.
8890 '``llvm.bswap.*``' Intrinsics
8891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8896 This is an overloaded intrinsic function. You can use bswap on any
8897 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8901 declare i16 @llvm.bswap.i16(i16 <id>)
8902 declare i32 @llvm.bswap.i32(i32 <id>)
8903 declare i64 @llvm.bswap.i64(i64 <id>)
8908 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8909 values with an even number of bytes (positive multiple of 16 bits).
8910 These are useful for performing operations on data that is not in the
8911 target's native byte order.
8916 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8917 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8918 intrinsic returns an i32 value that has the four bytes of the input i32
8919 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8920 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8921 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8922 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8925 '``llvm.ctpop.*``' Intrinsic
8926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8931 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8932 bit width, or on any vector with integer elements. Not all targets
8933 support all bit widths or vector types, however.
8937 declare i8 @llvm.ctpop.i8(i8 <src>)
8938 declare i16 @llvm.ctpop.i16(i16 <src>)
8939 declare i32 @llvm.ctpop.i32(i32 <src>)
8940 declare i64 @llvm.ctpop.i64(i64 <src>)
8941 declare i256 @llvm.ctpop.i256(i256 <src>)
8942 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8947 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8953 The only argument is the value to be counted. The argument may be of any
8954 integer type, or a vector with integer elements. The return type must
8955 match the argument type.
8960 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8961 each element of a vector.
8963 '``llvm.ctlz.*``' Intrinsic
8964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8969 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8970 integer bit width, or any vector whose elements are integers. Not all
8971 targets support all bit widths or vector types, however.
8975 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8976 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8977 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8978 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8979 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8980 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8985 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8986 leading zeros in a variable.
8991 The first argument is the value to be counted. This argument may be of
8992 any integer type, or a vector with integer element type. The return
8993 type must match the first argument type.
8995 The second argument must be a constant and is a flag to indicate whether
8996 the intrinsic should ensure that a zero as the first argument produces a
8997 defined result. Historically some architectures did not provide a
8998 defined result for zero values as efficiently, and many algorithms are
8999 now predicated on avoiding zero-value inputs.
9004 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9005 zeros in a variable, or within each element of the vector. If
9006 ``src == 0`` then the result is the size in bits of the type of ``src``
9007 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9008 ``llvm.ctlz(i32 2) = 30``.
9010 '``llvm.cttz.*``' Intrinsic
9011 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9016 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9017 integer bit width, or any vector of integer elements. Not all targets
9018 support all bit widths or vector types, however.
9022 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9023 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9024 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9025 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9026 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9027 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9032 The '``llvm.cttz``' family of intrinsic functions counts the number of
9038 The first argument is the value to be counted. This argument may be of
9039 any integer type, or a vector with integer element type. The return
9040 type must match the first argument type.
9042 The second argument must be a constant and is a flag to indicate whether
9043 the intrinsic should ensure that a zero as the first argument produces a
9044 defined result. Historically some architectures did not provide a
9045 defined result for zero values as efficiently, and many algorithms are
9046 now predicated on avoiding zero-value inputs.
9051 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9052 zeros in a variable, or within each element of a vector. If ``src == 0``
9053 then the result is the size in bits of the type of ``src`` if
9054 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9055 ``llvm.cttz(2) = 1``.
9057 Arithmetic with Overflow Intrinsics
9058 -----------------------------------
9060 LLVM provides intrinsics for some arithmetic with overflow operations.
9062 '``llvm.sadd.with.overflow.*``' Intrinsics
9063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9068 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9069 on any integer bit width.
9073 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9074 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9075 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9080 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9081 a signed addition of the two arguments, and indicate whether an overflow
9082 occurred during the signed summation.
9087 The arguments (%a and %b) and the first element of the result structure
9088 may be of integer types of any bit width, but they must have the same
9089 bit width. The second element of the result structure must be of type
9090 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9096 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9097 a signed addition of the two variables. They return a structure --- the
9098 first element of which is the signed summation, and the second element
9099 of which is a bit specifying if the signed summation resulted in an
9105 .. code-block:: llvm
9107 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9108 %sum = extractvalue {i32, i1} %res, 0
9109 %obit = extractvalue {i32, i1} %res, 1
9110 br i1 %obit, label %overflow, label %normal
9112 '``llvm.uadd.with.overflow.*``' Intrinsics
9113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9118 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9119 on any integer bit width.
9123 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9124 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9125 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9130 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9131 an unsigned addition of the two arguments, and indicate whether a carry
9132 occurred during the unsigned summation.
9137 The arguments (%a and %b) and the first element of the result structure
9138 may be of integer types of any bit width, but they must have the same
9139 bit width. The second element of the result structure must be of type
9140 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9146 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9147 an unsigned addition of the two arguments. They return a structure --- the
9148 first element of which is the sum, and the second element of which is a
9149 bit specifying if the unsigned summation resulted in a carry.
9154 .. code-block:: llvm
9156 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9157 %sum = extractvalue {i32, i1} %res, 0
9158 %obit = extractvalue {i32, i1} %res, 1
9159 br i1 %obit, label %carry, label %normal
9161 '``llvm.ssub.with.overflow.*``' Intrinsics
9162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9167 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9168 on any integer bit width.
9172 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9173 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9174 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9179 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9180 a signed subtraction of the two arguments, and indicate whether an
9181 overflow occurred during the signed subtraction.
9186 The arguments (%a and %b) and the first element of the result structure
9187 may be of integer types of any bit width, but they must have the same
9188 bit width. The second element of the result structure must be of type
9189 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9195 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9196 a signed subtraction of the two arguments. They return a structure --- the
9197 first element of which is the subtraction, and the second element of
9198 which is a bit specifying if the signed subtraction resulted in an
9204 .. code-block:: llvm
9206 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9207 %sum = extractvalue {i32, i1} %res, 0
9208 %obit = extractvalue {i32, i1} %res, 1
9209 br i1 %obit, label %overflow, label %normal
9211 '``llvm.usub.with.overflow.*``' Intrinsics
9212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9217 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9218 on any integer bit width.
9222 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9223 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9224 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9229 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9230 an unsigned subtraction of the two arguments, and indicate whether an
9231 overflow occurred during the unsigned subtraction.
9236 The arguments (%a and %b) and the first element of the result structure
9237 may be of integer types of any bit width, but they must have the same
9238 bit width. The second element of the result structure must be of type
9239 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9245 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9246 an unsigned subtraction of the two arguments. They return a structure ---
9247 the first element of which is the subtraction, and the second element of
9248 which is a bit specifying if the unsigned subtraction resulted in an
9254 .. code-block:: llvm
9256 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9257 %sum = extractvalue {i32, i1} %res, 0
9258 %obit = extractvalue {i32, i1} %res, 1
9259 br i1 %obit, label %overflow, label %normal
9261 '``llvm.smul.with.overflow.*``' Intrinsics
9262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9267 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9268 on any integer bit width.
9272 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9273 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9274 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9279 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9280 a signed multiplication of the two arguments, and indicate whether an
9281 overflow occurred during the signed multiplication.
9286 The arguments (%a and %b) and the first element of the result structure
9287 may be of integer types of any bit width, but they must have the same
9288 bit width. The second element of the result structure must be of type
9289 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9295 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9296 a signed multiplication of the two arguments. They return a structure ---
9297 the first element of which is the multiplication, and the second element
9298 of which is a bit specifying if the signed multiplication resulted in an
9304 .. code-block:: llvm
9306 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9307 %sum = extractvalue {i32, i1} %res, 0
9308 %obit = extractvalue {i32, i1} %res, 1
9309 br i1 %obit, label %overflow, label %normal
9311 '``llvm.umul.with.overflow.*``' Intrinsics
9312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9317 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9318 on any integer bit width.
9322 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9323 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9324 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9329 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9330 a unsigned multiplication of the two arguments, and indicate whether an
9331 overflow occurred during the unsigned multiplication.
9336 The arguments (%a and %b) and the first element of the result structure
9337 may be of integer types of any bit width, but they must have the same
9338 bit width. The second element of the result structure must be of type
9339 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9345 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9346 an unsigned multiplication of the two arguments. They return a structure ---
9347 the first element of which is the multiplication, and the second
9348 element of which is a bit specifying if the unsigned multiplication
9349 resulted in an overflow.
9354 .. code-block:: llvm
9356 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9357 %sum = extractvalue {i32, i1} %res, 0
9358 %obit = extractvalue {i32, i1} %res, 1
9359 br i1 %obit, label %overflow, label %normal
9361 Specialised Arithmetic Intrinsics
9362 ---------------------------------
9364 '``llvm.fmuladd.*``' Intrinsic
9365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9372 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9373 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9378 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9379 expressions that can be fused if the code generator determines that (a) the
9380 target instruction set has support for a fused operation, and (b) that the
9381 fused operation is more efficient than the equivalent, separate pair of mul
9382 and add instructions.
9387 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9388 multiplicands, a and b, and an addend c.
9397 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9399 is equivalent to the expression a \* b + c, except that rounding will
9400 not be performed between the multiplication and addition steps if the
9401 code generator fuses the operations. Fusion is not guaranteed, even if
9402 the target platform supports it. If a fused multiply-add is required the
9403 corresponding llvm.fma.\* intrinsic function should be used
9404 instead. This never sets errno, just as '``llvm.fma.*``'.
9409 .. code-block:: llvm
9411 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9413 Half Precision Floating Point Intrinsics
9414 ----------------------------------------
9416 For most target platforms, half precision floating point is a
9417 storage-only format. This means that it is a dense encoding (in memory)
9418 but does not support computation in the format.
9420 This means that code must first load the half-precision floating point
9421 value as an i16, then convert it to float with
9422 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9423 then be performed on the float value (including extending to double
9424 etc). To store the value back to memory, it is first converted to float
9425 if needed, then converted to i16 with
9426 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9429 .. _int_convert_to_fp16:
9431 '``llvm.convert.to.fp16``' Intrinsic
9432 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9439 declare i16 @llvm.convert.to.fp16.f32(float %a)
9440 declare i16 @llvm.convert.to.fp16.f64(double %a)
9445 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9446 conventional floating point type to half precision floating point format.
9451 The intrinsic function contains single argument - the value to be
9457 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9458 conventional floating point format to half precision floating point format. The
9459 return value is an ``i16`` which contains the converted number.
9464 .. code-block:: llvm
9466 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9467 store i16 %res, i16* @x, align 2
9469 .. _int_convert_from_fp16:
9471 '``llvm.convert.from.fp16``' Intrinsic
9472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9479 declare float @llvm.convert.from.fp16.f32(i16 %a)
9480 declare double @llvm.convert.from.fp16.f64(i16 %a)
9485 The '``llvm.convert.from.fp16``' intrinsic function performs a
9486 conversion from half precision floating point format to single precision
9487 floating point format.
9492 The intrinsic function contains single argument - the value to be
9498 The '``llvm.convert.from.fp16``' intrinsic function performs a
9499 conversion from half single precision floating point format to single
9500 precision floating point format. The input half-float value is
9501 represented by an ``i16`` value.
9506 .. code-block:: llvm
9508 %a = load i16* @x, align 2
9509 %res = call float @llvm.convert.from.fp16(i16 %a)
9516 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9517 prefix), are described in the `LLVM Source Level
9518 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9521 Exception Handling Intrinsics
9522 -----------------------------
9524 The LLVM exception handling intrinsics (which all start with
9525 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9526 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9530 Trampoline Intrinsics
9531 ---------------------
9533 These intrinsics make it possible to excise one parameter, marked with
9534 the :ref:`nest <nest>` attribute, from a function. The result is a
9535 callable function pointer lacking the nest parameter - the caller does
9536 not need to provide a value for it. Instead, the value to use is stored
9537 in advance in a "trampoline", a block of memory usually allocated on the
9538 stack, which also contains code to splice the nest value into the
9539 argument list. This is used to implement the GCC nested function address
9542 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9543 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9544 It can be created as follows:
9546 .. code-block:: llvm
9548 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9549 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
9550 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9551 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9552 %fp = bitcast i8* %p to i32 (i32, i32)*
9554 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9555 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9559 '``llvm.init.trampoline``' Intrinsic
9560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9567 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9572 This fills the memory pointed to by ``tramp`` with executable code,
9573 turning it into a trampoline.
9578 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9579 pointers. The ``tramp`` argument must point to a sufficiently large and
9580 sufficiently aligned block of memory; this memory is written to by the
9581 intrinsic. Note that the size and the alignment are target-specific -
9582 LLVM currently provides no portable way of determining them, so a
9583 front-end that generates this intrinsic needs to have some
9584 target-specific knowledge. The ``func`` argument must hold a function
9585 bitcast to an ``i8*``.
9590 The block of memory pointed to by ``tramp`` is filled with target
9591 dependent code, turning it into a function. Then ``tramp`` needs to be
9592 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9593 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9594 function's signature is the same as that of ``func`` with any arguments
9595 marked with the ``nest`` attribute removed. At most one such ``nest``
9596 argument is allowed, and it must be of pointer type. Calling the new
9597 function is equivalent to calling ``func`` with the same argument list,
9598 but with ``nval`` used for the missing ``nest`` argument. If, after
9599 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9600 modified, then the effect of any later call to the returned function
9601 pointer is undefined.
9605 '``llvm.adjust.trampoline``' Intrinsic
9606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9613 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9618 This performs any required machine-specific adjustment to the address of
9619 a trampoline (passed as ``tramp``).
9624 ``tramp`` must point to a block of memory which already has trampoline
9625 code filled in by a previous call to
9626 :ref:`llvm.init.trampoline <int_it>`.
9631 On some architectures the address of the code to be executed needs to be
9632 different than the address where the trampoline is actually stored. This
9633 intrinsic returns the executable address corresponding to ``tramp``
9634 after performing the required machine specific adjustments. The pointer
9635 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9637 Masked Vector Load and Store Intrinsics
9638 ---------------------------------------
9640 LLVM provides intrinsics for predicated vector load and store operations. The predicate is specified by a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits of the mask are on, the intrinsic is identical to a regular vector load or store. When all bits are off, no memory is accessed.
9644 '``llvm.masked.load.*``' Intrinsics
9645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9649 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9653 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9654 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9659 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes in the passthru operand.
9665 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean 'i1' values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of passthru operand are the same vector types.
9671 The '``llvm.masked.load``' intrinsic is designed for conditional reading of selected vector elements in a single IR operation. It is useful for targets that support vector masked loads and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar load operations.
9672 The result of this operation is equivalent to a regular vector load instruction followed by a 'select' between the loaded and the passthru values, predicated on the same mask. However, using this intrinsic prevents exceptions on memory access to masked-off lanes.
9677 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9679 ;; The result of the two following instructions is identical aside from potential memory access exception
9680 %loadlal = load <16 x float>* %ptr, align 4
9681 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9685 '``llvm.masked.store.*``' Intrinsics
9686 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9690 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9694 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9695 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9700 Writes a vector to memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9705 The first operand is the vector value to be written to memory. The second operand is the base pointer for the store, it has the same underlying type as the value operand. The third operand is the alignment of the destination location. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9711 The '``llvm.masked.store``' intrinsics is designed for conditional writing of selected vector elements in a single IR operation. It is useful for targets that support vector masked store and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9712 The result of this operation is equivalent to a load-modify-store sequence. However, using this intrinsic prevents exceptions and data races on memory access to masked-off lanes.
9716 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9718 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9719 %oldval = load <16 x float>* %ptr, align 4
9720 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9721 store <16 x float> %res, <16 x float>* %ptr, align 4
9727 This class of intrinsics provides information about the lifetime of
9728 memory objects and ranges where variables are immutable.
9732 '``llvm.lifetime.start``' Intrinsic
9733 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9740 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9745 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9751 The first argument is a constant integer representing the size of the
9752 object, or -1 if it is variable sized. The second argument is a pointer
9758 This intrinsic indicates that before this point in the code, the value
9759 of the memory pointed to by ``ptr`` is dead. This means that it is known
9760 to never be used and has an undefined value. A load from the pointer
9761 that precedes this intrinsic can be replaced with ``'undef'``.
9765 '``llvm.lifetime.end``' Intrinsic
9766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9773 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9778 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9784 The first argument is a constant integer representing the size of the
9785 object, or -1 if it is variable sized. The second argument is a pointer
9791 This intrinsic indicates that after this point in the code, the value of
9792 the memory pointed to by ``ptr`` is dead. This means that it is known to
9793 never be used and has an undefined value. Any stores into the memory
9794 object following this intrinsic may be removed as dead.
9796 '``llvm.invariant.start``' Intrinsic
9797 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9804 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9809 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9810 a memory object will not change.
9815 The first argument is a constant integer representing the size of the
9816 object, or -1 if it is variable sized. The second argument is a pointer
9822 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9823 the return value, the referenced memory location is constant and
9826 '``llvm.invariant.end``' Intrinsic
9827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9834 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9839 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9840 memory object are mutable.
9845 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9846 The second argument is a constant integer representing the size of the
9847 object, or -1 if it is variable sized and the third argument is a
9848 pointer to the object.
9853 This intrinsic indicates that the memory is mutable again.
9858 This class of intrinsics is designed to be generic and has no specific
9861 '``llvm.var.annotation``' Intrinsic
9862 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9869 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9874 The '``llvm.var.annotation``' intrinsic.
9879 The first argument is a pointer to a value, the second is a pointer to a
9880 global string, the third is a pointer to a global string which is the
9881 source file name, and the last argument is the line number.
9886 This intrinsic allows annotation of local variables with arbitrary
9887 strings. This can be useful for special purpose optimizations that want
9888 to look for these annotations. These have no other defined use; they are
9889 ignored by code generation and optimization.
9891 '``llvm.ptr.annotation.*``' Intrinsic
9892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9897 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9898 pointer to an integer of any width. *NOTE* you must specify an address space for
9899 the pointer. The identifier for the default address space is the integer
9904 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9905 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9906 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9907 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9908 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9913 The '``llvm.ptr.annotation``' intrinsic.
9918 The first argument is a pointer to an integer value of arbitrary bitwidth
9919 (result of some expression), the second is a pointer to a global string, the
9920 third is a pointer to a global string which is the source file name, and the
9921 last argument is the line number. It returns the value of the first argument.
9926 This intrinsic allows annotation of a pointer to an integer with arbitrary
9927 strings. This can be useful for special purpose optimizations that want to look
9928 for these annotations. These have no other defined use; they are ignored by code
9929 generation and optimization.
9931 '``llvm.annotation.*``' Intrinsic
9932 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9937 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9938 any integer bit width.
9942 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9943 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9944 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9945 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9946 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9951 The '``llvm.annotation``' intrinsic.
9956 The first argument is an integer value (result of some expression), the
9957 second is a pointer to a global string, the third is a pointer to a
9958 global string which is the source file name, and the last argument is
9959 the line number. It returns the value of the first argument.
9964 This intrinsic allows annotations to be put on arbitrary expressions
9965 with arbitrary strings. This can be useful for special purpose
9966 optimizations that want to look for these annotations. These have no
9967 other defined use; they are ignored by code generation and optimization.
9969 '``llvm.trap``' Intrinsic
9970 ^^^^^^^^^^^^^^^^^^^^^^^^^
9977 declare void @llvm.trap() noreturn nounwind
9982 The '``llvm.trap``' intrinsic.
9992 This intrinsic is lowered to the target dependent trap instruction. If
9993 the target does not have a trap instruction, this intrinsic will be
9994 lowered to a call of the ``abort()`` function.
9996 '``llvm.debugtrap``' Intrinsic
9997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10004 declare void @llvm.debugtrap() nounwind
10009 The '``llvm.debugtrap``' intrinsic.
10019 This intrinsic is lowered to code which is intended to cause an
10020 execution trap with the intention of requesting the attention of a
10023 '``llvm.stackprotector``' Intrinsic
10024 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10031 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10036 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10037 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10038 is placed on the stack before local variables.
10043 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10044 The first argument is the value loaded from the stack guard
10045 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10046 enough space to hold the value of the guard.
10051 This intrinsic causes the prologue/epilogue inserter to force the position of
10052 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10053 to ensure that if a local variable on the stack is overwritten, it will destroy
10054 the value of the guard. When the function exits, the guard on the stack is
10055 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10056 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10057 calling the ``__stack_chk_fail()`` function.
10059 '``llvm.stackprotectorcheck``' Intrinsic
10060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10067 declare void @llvm.stackprotectorcheck(i8** <guard>)
10072 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10073 created stack protector and if they are not equal calls the
10074 ``__stack_chk_fail()`` function.
10079 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10080 the variable ``@__stack_chk_guard``.
10085 This intrinsic is provided to perform the stack protector check by comparing
10086 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10087 values do not match call the ``__stack_chk_fail()`` function.
10089 The reason to provide this as an IR level intrinsic instead of implementing it
10090 via other IR operations is that in order to perform this operation at the IR
10091 level without an intrinsic, one would need to create additional basic blocks to
10092 handle the success/failure cases. This makes it difficult to stop the stack
10093 protector check from disrupting sibling tail calls in Codegen. With this
10094 intrinsic, we are able to generate the stack protector basic blocks late in
10095 codegen after the tail call decision has occurred.
10097 '``llvm.objectsize``' Intrinsic
10098 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10105 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10106 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10111 The ``llvm.objectsize`` intrinsic is designed to provide information to
10112 the optimizers to determine at compile time whether a) an operation
10113 (like memcpy) will overflow a buffer that corresponds to an object, or
10114 b) that a runtime check for overflow isn't necessary. An object in this
10115 context means an allocation of a specific class, structure, array, or
10121 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10122 argument is a pointer to or into the ``object``. The second argument is
10123 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10124 or -1 (if false) when the object size is unknown. The second argument
10125 only accepts constants.
10130 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10131 the size of the object concerned. If the size cannot be determined at
10132 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10133 on the ``min`` argument).
10135 '``llvm.expect``' Intrinsic
10136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10141 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10146 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10147 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10148 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10153 The ``llvm.expect`` intrinsic provides information about expected (the
10154 most probable) value of ``val``, which can be used by optimizers.
10159 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10160 a value. The second argument is an expected value, this needs to be a
10161 constant value, variables are not allowed.
10166 This intrinsic is lowered to the ``val``.
10168 '``llvm.assume``' Intrinsic
10169 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10176 declare void @llvm.assume(i1 %cond)
10181 The ``llvm.assume`` allows the optimizer to assume that the provided
10182 condition is true. This information can then be used in simplifying other parts
10188 The condition which the optimizer may assume is always true.
10193 The intrinsic allows the optimizer to assume that the provided condition is
10194 always true whenever the control flow reaches the intrinsic call. No code is
10195 generated for this intrinsic, and instructions that contribute only to the
10196 provided condition are not used for code generation. If the condition is
10197 violated during execution, the behavior is undefined.
10199 Note that the optimizer might limit the transformations performed on values
10200 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10201 only used to form the intrinsic's input argument. This might prove undesirable
10202 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10203 sufficient overall improvement in code quality. For this reason,
10204 ``llvm.assume`` should not be used to document basic mathematical invariants
10205 that the optimizer can otherwise deduce or facts that are of little use to the
10210 '``llvm.bitset.test``' Intrinsic
10211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10218 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10224 The first argument is a pointer to be tested. The second argument is a
10225 metadata string containing the name of a :doc:`bitset <BitSets>`.
10230 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10231 member of the given bitset.
10233 '``llvm.donothing``' Intrinsic
10234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10241 declare void @llvm.donothing() nounwind readnone
10246 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10247 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10248 with an invoke instruction.
10258 This intrinsic does nothing, and it's removed by optimizers and ignored
10261 Stack Map Intrinsics
10262 --------------------
10264 LLVM provides experimental intrinsics to support runtime patching
10265 mechanisms commonly desired in dynamic language JITs. These intrinsics
10266 are described in :doc:`StackMaps`.