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 which 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.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
271 If none of the above identifiers are used, the global is externally
272 visible, meaning that it participates in linkage and can be used to
273 resolve external symbol references.
275 The next two types of linkage are targeted for Microsoft Windows
276 platform only. They are designed to support importing (exporting)
277 symbols from (to) DLLs (Dynamic Link Libraries).
280 "``dllimport``" linkage causes the compiler to reference a function
281 or variable via a global pointer to a pointer that is set up by the
282 DLL exporting the symbol. On Microsoft Windows targets, the pointer
283 name is formed by combining ``__imp_`` and the function or variable
286 "``dllexport``" linkage causes the compiler to provide a global
287 pointer to a pointer in a DLL, so that it can be referenced with the
288 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
289 name is formed by combining ``__imp_`` and the function or variable
292 For example, since the "``.LC0``" variable is defined to be internal, if
293 another module defined a "``.LC0``" variable and was linked with this
294 one, one of the two would be renamed, preventing a collision. Since
295 "``main``" and "``puts``" are external (i.e., lacking any linkage
296 declarations), they are accessible outside of the current module.
298 It is illegal for a function *declaration* to have any linkage type
299 other than ``external``, ``dllimport`` or ``extern_weak``.
306 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
307 :ref:`invokes <i_invoke>` can all have an optional calling convention
308 specified for the call. The calling convention of any pair of dynamic
309 caller/callee must match, or the behavior of the program is undefined.
310 The following calling conventions are supported by LLVM, and more may be
313 "``ccc``" - The C calling convention
314 This calling convention (the default if no other calling convention
315 is specified) matches the target C calling conventions. This calling
316 convention supports varargs function calls and tolerates some
317 mismatch in the declared prototype and implemented declaration of
318 the function (as does normal C).
319 "``fastcc``" - The fast calling convention
320 This calling convention attempts to make calls as fast as possible
321 (e.g. by passing things in registers). This calling convention
322 allows the target to use whatever tricks it wants to produce fast
323 code for the target, without having to conform to an externally
324 specified ABI (Application Binary Interface). `Tail calls can only
325 be optimized when this, the GHC or the HiPE convention is
326 used. <CodeGenerator.html#id80>`_ This calling convention does not
327 support varargs and requires the prototype of all callees to exactly
328 match the prototype of the function definition.
329 "``coldcc``" - The cold calling convention
330 This calling convention attempts to make code in the caller as
331 efficient as possible under the assumption that the call is not
332 commonly executed. As such, these calls often preserve all registers
333 so that the call does not break any live ranges in the caller side.
334 This calling convention does not support varargs and requires the
335 prototype of all callees to exactly match the prototype of the
337 "``cc 10``" - GHC convention
338 This calling convention has been implemented specifically for use by
339 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
340 It passes everything in registers, going to extremes to achieve this
341 by disabling callee save registers. This calling convention should
342 not be used lightly but only for specific situations such as an
343 alternative to the *register pinning* performance technique often
344 used when implementing functional programming languages. At the
345 moment only X86 supports this convention and it has the following
348 - On *X86-32* only supports up to 4 bit type parameters. No
349 floating point types are supported.
350 - On *X86-64* only supports up to 10 bit type parameters and 6
351 floating point parameters.
353 This calling convention supports `tail call
354 optimization <CodeGenerator.html#id80>`_ but requires both the
355 caller and callee are using it.
356 "``cc 11``" - The HiPE calling convention
357 This calling convention has been implemented specifically for use by
358 the `High-Performance Erlang
359 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
360 native code compiler of the `Ericsson's Open Source Erlang/OTP
361 system <http://www.erlang.org/download.shtml>`_. It uses more
362 registers for argument passing than the ordinary C calling
363 convention and defines no callee-saved registers. The calling
364 convention properly supports `tail call
365 optimization <CodeGenerator.html#id80>`_ but requires that both the
366 caller and the callee use it. It uses a *register pinning*
367 mechanism, similar to GHC's convention, for keeping frequently
368 accessed runtime components pinned to specific hardware registers.
369 At the moment only X86 supports this convention (both 32 and 64
371 "``cc <n>``" - Numbered convention
372 Any calling convention may be specified by number, allowing
373 target-specific calling conventions to be used. Target specific
374 calling conventions start at 64.
376 More calling conventions can be added/defined on an as-needed basis, to
377 support Pascal conventions or any other well-known target-independent
380 .. _visibilitystyles:
385 All Global Variables and Functions have one of the following visibility
388 "``default``" - Default style
389 On targets that use the ELF object file format, default visibility
390 means that the declaration is visible to other modules and, in
391 shared libraries, means that the declared entity may be overridden.
392 On Darwin, default visibility means that the declaration is visible
393 to other modules. Default visibility corresponds to "external
394 linkage" in the language.
395 "``hidden``" - Hidden style
396 Two declarations of an object with hidden visibility refer to the
397 same object if they are in the same shared object. Usually, hidden
398 visibility indicates that the symbol will not be placed into the
399 dynamic symbol table, so no other module (executable or shared
400 library) can reference it directly.
401 "``protected``" - Protected style
402 On ELF, protected visibility indicates that the symbol will be
403 placed in the dynamic symbol table, but that references within the
404 defining module will bind to the local symbol. That is, the symbol
405 cannot be overridden by another module.
412 LLVM IR allows you to specify name aliases for certain types. This can
413 make it easier to read the IR and make the IR more condensed
414 (particularly when recursive types are involved). An example of a name
419 %mytype = type { %mytype*, i32 }
421 You may give a name to any :ref:`type <typesystem>` except
422 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
423 expected with the syntax "%mytype".
425 Note that type names are aliases for the structural type that they
426 indicate, and that you can therefore specify multiple names for the same
427 type. This often leads to confusing behavior when dumping out a .ll
428 file. Since LLVM IR uses structural typing, the name is not part of the
429 type. When printing out LLVM IR, the printer will pick *one name* to
430 render all types of a particular shape. This means that if you have code
431 where two different source types end up having the same LLVM type, that
432 the dumper will sometimes print the "wrong" or unexpected type. This is
433 an important design point and isn't going to change.
440 Global variables define regions of memory allocated at compilation time
443 Global variables definitions must be initialized, may have an explicit section
444 to be placed in, and may have an optional explicit alignment specified.
446 Global variables in other translation units can also be declared, in which
447 case they don't have an initializer.
449 A variable may be defined as ``thread_local``, which means that it will
450 not be shared by threads (each thread will have a separated copy of the
451 variable). Not all targets support thread-local variables. Optionally, a
452 TLS model may be specified:
455 For variables that are only used within the current shared library.
457 For variables in modules that will not be loaded dynamically.
459 For variables defined in the executable and only used within it.
461 The models correspond to the ELF TLS models; see `ELF Handling For
462 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
463 more information on under which circumstances the different models may
464 be used. The target may choose a different TLS model if the specified
465 model is not supported, or if a better choice of model can be made.
467 A variable may be defined as a global ``constant``, which indicates that
468 the contents of the variable will **never** be modified (enabling better
469 optimization, allowing the global data to be placed in the read-only
470 section of an executable, etc). Note that variables that need runtime
471 initialization cannot be marked ``constant`` as there is a store to the
474 LLVM explicitly allows *declarations* of global variables to be marked
475 constant, even if the final definition of the global is not. This
476 capability can be used to enable slightly better optimization of the
477 program, but requires the language definition to guarantee that
478 optimizations based on the 'constantness' are valid for the translation
479 units that do not include the definition.
481 As SSA values, global variables define pointer values that are in scope
482 (i.e. they dominate) all basic blocks in the program. Global variables
483 always define a pointer to their "content" type because they describe a
484 region of memory, and all memory objects in LLVM are accessed through
487 Global variables can be marked with ``unnamed_addr`` which indicates
488 that the address is not significant, only the content. Constants marked
489 like this can be merged with other constants if they have the same
490 initializer. Note that a constant with significant address *can* be
491 merged with a ``unnamed_addr`` constant, the result being a constant
492 whose address is significant.
494 A global variable may be declared to reside in a target-specific
495 numbered address space. For targets that support them, address spaces
496 may affect how optimizations are performed and/or what target
497 instructions are used to access the variable. The default address space
498 is zero. The address space qualifier must precede any other attributes.
500 LLVM allows an explicit section to be specified for globals. If the
501 target supports it, it will emit globals to the section specified.
503 By default, global initializers are optimized by assuming that global
504 variables defined within the module are not modified from their
505 initial values before the start of the global initializer. This is
506 true even for variables potentially accessible from outside the
507 module, including those with external linkage or appearing in
508 ``@llvm.used``. This assumption may be suppressed by marking the
509 variable with ``externally_initialized``.
511 An explicit alignment may be specified for a global, which must be a
512 power of 2. If not present, or if the alignment is set to zero, the
513 alignment of the global is set by the target to whatever it feels
514 convenient. If an explicit alignment is specified, the global is forced
515 to have exactly that alignment. Targets and optimizers are not allowed
516 to over-align the global if the global has an assigned section. In this
517 case, the extra alignment could be observable: for example, code could
518 assume that the globals are densely packed in their section and try to
519 iterate over them as an array, alignment padding would break this
522 For example, the following defines a global in a numbered address space
523 with an initializer, section, and alignment:
527 @G = addrspace(5) constant float 1.0, section "foo", align 4
529 The following example just declares a global variable
533 @G = external global i32
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
556 curly brace, a list of basic blocks, and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, an optional
564 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
566 A function definition contains a list of basic blocks, forming the CFG (Control
567 Flow Graph) for the function. Each basic block may optionally start with a label
568 (giving the basic block a symbol table entry), contains a list of instructions,
569 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
570 function return). If an explicit label is not provided, a block is assigned an
571 implicit numbered label, using the next value from the same counter as used for
572 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
573 entry block does not have an explicit label, it will be assigned label "%0",
574 then the first unnamed temporary in that block will be "%1", etc.
576 The first basic block in a function is special in two ways: it is
577 immediately executed on entrance to the function, and it is not allowed
578 to have predecessor basic blocks (i.e. there can not be any branches to
579 the entry block of a function). Because the block can have no
580 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
582 LLVM allows an explicit section to be specified for functions. If the
583 target supports it, it will emit functions to the section specified.
585 An explicit alignment may be specified for a function. If not present,
586 or if the alignment is set to zero, the alignment of the function is set
587 by the target to whatever it feels convenient. If an explicit alignment
588 is specified, the function is forced to have at least that much
589 alignment. All alignments must be a power of 2.
591 If the ``unnamed_addr`` attribute is given, the address is know to not
592 be significant and two identical functions can be merged.
596 define [linkage] [visibility]
598 <ResultType> @<FunctionName> ([argument list])
599 [fn Attrs] [section "name"] [align N]
600 [gc] [prefix Constant] { ... }
607 Aliases act as "second name" for the aliasee value (which can be either
608 function, global variable, another alias or bitcast of global value).
609 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
610 :ref:`visibility style <visibility>`.
614 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
616 The linkage must be one of ``private``, ``linker_private``,
617 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
618 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
619 might not correctly handle dropping a weak symbol that is aliased by a non weak
622 .. _namedmetadatastructure:
627 Named metadata is a collection of metadata. :ref:`Metadata
628 nodes <metadata>` (but not metadata strings) are the only valid
629 operands for a named metadata.
633 ; Some unnamed metadata nodes, which are referenced by the named metadata.
634 !0 = metadata !{metadata !"zero"}
635 !1 = metadata !{metadata !"one"}
636 !2 = metadata !{metadata !"two"}
638 !name = !{!0, !1, !2}
645 The return type and each parameter of a function type may have a set of
646 *parameter attributes* associated with them. Parameter attributes are
647 used to communicate additional information about the result or
648 parameters of a function. Parameter attributes are considered to be part
649 of the function, not of the function type, so functions with different
650 parameter attributes can have the same function type.
652 Parameter attributes are simple keywords that follow the type specified.
653 If multiple parameter attributes are needed, they are space separated.
658 declare i32 @printf(i8* noalias nocapture, ...)
659 declare i32 @atoi(i8 zeroext)
660 declare signext i8 @returns_signed_char()
662 Note that any attributes for the function result (``nounwind``,
663 ``readonly``) come immediately after the argument list.
665 Currently, only the following parameter attributes are defined:
668 This indicates to the code generator that the parameter or return
669 value should be zero-extended to the extent required by the target's
670 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
671 the caller (for a parameter) or the callee (for a return value).
673 This indicates to the code generator that the parameter or return
674 value should be sign-extended to the extent required by the target's
675 ABI (which is usually 32-bits) by the caller (for a parameter) or
676 the callee (for a return value).
678 This indicates that this parameter or return value should be treated
679 in a special target-dependent fashion during while emitting code for
680 a function call or return (usually, by putting it in a register as
681 opposed to memory, though some targets use it to distinguish between
682 two different kinds of registers). Use of this attribute is
685 This indicates that the pointer parameter should really be passed by
686 value to the function. The attribute implies that a hidden copy of
687 the pointee is made between the caller and the callee, so the callee
688 is unable to modify the value in the caller. This attribute is only
689 valid on LLVM pointer arguments. It is generally used to pass
690 structs and arrays by value, but is also valid on pointers to
691 scalars. The copy is considered to belong to the caller not the
692 callee (for example, ``readonly`` functions should not write to
693 ``byval`` parameters). This is not a valid attribute for return
696 The byval attribute also supports specifying an alignment with the
697 align attribute. It indicates the alignment of the stack slot to
698 form and the known alignment of the pointer specified to the call
699 site. If the alignment is not specified, then the code generator
700 makes a target-specific assumption.
703 This indicates that the pointer parameter specifies the address of a
704 structure that is the return value of the function in the source
705 program. This pointer must be guaranteed by the caller to be valid:
706 loads and stores to the structure may be assumed by the callee
707 not to trap and to be properly aligned. This may only be applied to
708 the first parameter. This is not a valid attribute for return
711 This indicates that pointer values :ref:`based <pointeraliasing>` on
712 the argument or return value do not alias pointer values which are
713 not *based* on it, ignoring certain "irrelevant" dependencies. For a
714 call to the parent function, dependencies between memory references
715 from before or after the call and from those during the call are
716 "irrelevant" to the ``noalias`` keyword for the arguments and return
717 value used in that call. The caller shares the responsibility with
718 the callee for ensuring that these requirements are met. For further
719 details, please see the discussion of the NoAlias response in `alias
720 analysis <AliasAnalysis.html#MustMayNo>`_.
722 Note that this definition of ``noalias`` is intentionally similar
723 to the definition of ``restrict`` in C99 for function arguments,
724 though it is slightly weaker.
726 For function return values, C99's ``restrict`` is not meaningful,
727 while LLVM's ``noalias`` is.
729 This indicates that the callee does not make any copies of the
730 pointer that outlive the callee itself. This is not a valid
731 attribute for return values.
736 This indicates that the pointer parameter can be excised using the
737 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
738 attribute for return values and can only be applied to one parameter.
741 This indicates that the function always returns the argument as its return
742 value. This is an optimization hint to the code generator when generating
743 the caller, allowing tail call optimization and omission of register saves
744 and restores in some cases; it is not checked or enforced when generating
745 the callee. The parameter and the function return type must be valid
746 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
747 valid attribute for return values and can only be applied to one parameter.
751 Garbage Collector Names
752 -----------------------
754 Each function may specify a garbage collector name, which is simply a
759 define void @f() gc "name" { ... }
761 The compiler declares the supported values of *name*. Specifying a
762 collector which will cause the compiler to alter its output in order to
763 support the named garbage collection algorithm.
770 Prefix data is data associated with a function which the code generator
771 will emit immediately before the function body. The purpose of this feature
772 is to allow frontends to associate language-specific runtime metadata with
773 specific functions and make it available through the function pointer while
774 still allowing the function pointer to be called. To access the data for a
775 given function, a program may bitcast the function pointer to a pointer to
776 the constant's type. This implies that the IR symbol points to the start
779 To maintain the semantics of ordinary function calls, the prefix data must
780 have a particular format. Specifically, it must begin with a sequence of
781 bytes which decode to a sequence of machine instructions, valid for the
782 module's target, which transfer control to the point immediately succeeding
783 the prefix data, without performing any other visible action. This allows
784 the inliner and other passes to reason about the semantics of the function
785 definition without needing to reason about the prefix data. Obviously this
786 makes the format of the prefix data highly target dependent.
788 Prefix data is laid out as if it were an initializer for a global variable
789 of the prefix data's type. No padding is automatically placed between the
790 prefix data and the function body. If padding is required, it must be part
793 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
794 which encodes the ``nop`` instruction:
798 define void @f() prefix i8 144 { ... }
800 Generally prefix data can be formed by encoding a relative branch instruction
801 which skips the metadata, as in this example of valid prefix data for the
802 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
806 %0 = type <{ i8, i8, i8* }>
808 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
810 A function may have prefix data but no body. This has similar semantics
811 to the ``available_externally`` linkage in that the data may be used by the
812 optimizers but will not be emitted in the object file.
819 Attribute groups are groups of attributes that are referenced by objects within
820 the IR. They are important for keeping ``.ll`` files readable, because a lot of
821 functions will use the same set of attributes. In the degenerative case of a
822 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
823 group will capture the important command line flags used to build that file.
825 An attribute group is a module-level object. To use an attribute group, an
826 object references the attribute group's ID (e.g. ``#37``). An object may refer
827 to more than one attribute group. In that situation, the attributes from the
828 different groups are merged.
830 Here is an example of attribute groups for a function that should always be
831 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
835 ; Target-independent attributes:
836 attributes #0 = { alwaysinline alignstack=4 }
838 ; Target-dependent attributes:
839 attributes #1 = { "no-sse" }
841 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
842 define void @f() #0 #1 { ... }
849 Function attributes are set to communicate additional information about
850 a function. Function attributes are considered to be part of the
851 function, not of the function type, so functions with different function
852 attributes can have the same function type.
854 Function attributes are simple keywords that follow the type specified.
855 If multiple attributes are needed, they are space separated. For
860 define void @f() noinline { ... }
861 define void @f() alwaysinline { ... }
862 define void @f() alwaysinline optsize { ... }
863 define void @f() optsize { ... }
866 This attribute indicates that, when emitting the prologue and
867 epilogue, the backend should forcibly align the stack pointer.
868 Specify the desired alignment, which must be a power of two, in
871 This attribute indicates that the inliner should attempt to inline
872 this function into callers whenever possible, ignoring any active
873 inlining size threshold for this caller.
875 This indicates that the callee function at a call site should be
876 recognized as a built-in function, even though the function's declaration
877 uses the ``nobuiltin`` attribute. This is only valid at call sites for
878 direct calls to functions which are declared with the ``nobuiltin``
881 This attribute indicates that this function is rarely called. When
882 computing edge weights, basic blocks post-dominated by a cold
883 function call are also considered to be cold; and, thus, given low
886 This attribute indicates that the source code contained a hint that
887 inlining this function is desirable (such as the "inline" keyword in
888 C/C++). It is just a hint; it imposes no requirements on the
891 This attribute suggests that optimization passes and code generator
892 passes make choices that keep the code size of this function as small
893 as possible and perform optimizations that may sacrifice runtime
894 performance in order to minimize the size of the generated code.
896 This attribute disables prologue / epilogue emission for the
897 function. This can have very system-specific consequences.
899 This indicates that the callee function at a call site is not recognized as
900 a built-in function. LLVM will retain the original call and not replace it
901 with equivalent code based on the semantics of the built-in function, unless
902 the call site uses the ``builtin`` attribute. This is valid at call sites
903 and on function declarations and definitions.
905 This attribute indicates that calls to the function cannot be
906 duplicated. A call to a ``noduplicate`` function may be moved
907 within its parent function, but may not be duplicated within
910 A function containing a ``noduplicate`` call may still
911 be an inlining candidate, provided that the call is not
912 duplicated by inlining. That implies that the function has
913 internal linkage and only has one call site, so the original
914 call is dead after inlining.
916 This attributes disables implicit floating point instructions.
918 This attribute indicates that the inliner should never inline this
919 function in any situation. This attribute may not be used together
920 with the ``alwaysinline`` attribute.
922 This attribute suppresses lazy symbol binding for the function. This
923 may make calls to the function faster, at the cost of extra program
924 startup time if the function is not called during program startup.
926 This attribute indicates that the code generator should not use a
927 red zone, even if the target-specific ABI normally permits it.
929 This function attribute indicates that the function never returns
930 normally. This produces undefined behavior at runtime if the
931 function ever does dynamically return.
933 This function attribute indicates that the function never returns
934 with an unwind or exceptional control flow. If the function does
935 unwind, its runtime behavior is undefined.
937 This function attribute indicates that the function is not optimized
938 by any optimization or code generator passes with the
939 exception of interprocedural optimization passes.
940 This attribute cannot be used together with the ``alwaysinline``
941 attribute; this attribute is also incompatible
942 with the ``minsize`` attribute and the ``optsize`` attribute.
944 This attribute requires the ``noinline`` attribute to be specified on
945 the function as well, so the function is never inlined into any caller.
946 Only functions with the ``alwaysinline`` attribute are valid
947 candidates for inlining into the body of this function.
949 This attribute suggests that optimization passes and code generator
950 passes make choices that keep the code size of this function low,
951 and otherwise do optimizations specifically to reduce code size as
952 long as they do not significantly impact runtime performance.
954 On a function, this attribute indicates that the function computes its
955 result (or decides to unwind an exception) based strictly on its arguments,
956 without dereferencing any pointer arguments or otherwise accessing
957 any mutable state (e.g. memory, control registers, etc) visible to
958 caller functions. It does not write through any pointer arguments
959 (including ``byval`` arguments) and never changes any state visible
960 to callers. This means that it cannot unwind exceptions by calling
961 the ``C++`` exception throwing methods.
963 On an argument, this attribute indicates that the function does not
964 dereference that pointer argument, even though it may read or write the
965 memory that the pointer points to if accessed through other pointers.
967 On a function, this attribute indicates that the function does not write
968 through any pointer arguments (including ``byval`` arguments) or otherwise
969 modify any state (e.g. memory, control registers, etc) visible to
970 caller functions. It may dereference pointer arguments and read
971 state that may be set in the caller. A readonly function always
972 returns the same value (or unwinds an exception identically) when
973 called with the same set of arguments and global state. It cannot
974 unwind an exception by calling the ``C++`` exception throwing
977 On an argument, this attribute indicates that the function does not write
978 through this pointer argument, even though it may write to the memory that
979 the pointer points to.
981 This attribute indicates that this function can return twice. The C
982 ``setjmp`` is an example of such a function. The compiler disables
983 some optimizations (like tail calls) in the caller of these
986 This attribute indicates that AddressSanitizer checks
987 (dynamic address safety analysis) are enabled for this function.
989 This attribute indicates that MemorySanitizer checks (dynamic detection
990 of accesses to uninitialized memory) are enabled for this function.
992 This attribute indicates that ThreadSanitizer checks
993 (dynamic thread safety analysis) are enabled for this function.
995 This attribute indicates that the function should emit a stack
996 smashing protector. It is in the form of a "canary" --- a random value
997 placed on the stack before the local variables that's checked upon
998 return from the function to see if it has been overwritten. A
999 heuristic is used to determine if a function needs stack protectors
1000 or not. The heuristic used will enable protectors for functions with:
1002 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1003 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1004 - Calls to alloca() with variable sizes or constant sizes greater than
1005 ``ssp-buffer-size``.
1007 If a function that has an ``ssp`` attribute is inlined into a
1008 function that doesn't have an ``ssp`` attribute, then the resulting
1009 function will have an ``ssp`` attribute.
1011 This attribute indicates that the function should *always* emit a
1012 stack smashing protector. This overrides the ``ssp`` function
1015 If a function that has an ``sspreq`` attribute is inlined into a
1016 function that doesn't have an ``sspreq`` attribute or which has an
1017 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1018 an ``sspreq`` attribute.
1020 This attribute indicates that the function should emit a stack smashing
1021 protector. This attribute causes a strong heuristic to be used when
1022 determining if a function needs stack protectors. The strong heuristic
1023 will enable protectors for functions with:
1025 - Arrays of any size and type
1026 - Aggregates containing an array of any size and type.
1027 - Calls to alloca().
1028 - Local variables that have had their address taken.
1030 This overrides the ``ssp`` function attribute.
1032 If a function that has an ``sspstrong`` attribute is inlined into a
1033 function that doesn't have an ``sspstrong`` attribute, then the
1034 resulting function will have an ``sspstrong`` attribute.
1036 This attribute indicates that the ABI being targeted requires that
1037 an unwind table entry be produce for this function even if we can
1038 show that no exceptions passes by it. This is normally the case for
1039 the ELF x86-64 abi, but it can be disabled for some compilation
1044 Module-Level Inline Assembly
1045 ----------------------------
1047 Modules may contain "module-level inline asm" blocks, which corresponds
1048 to the GCC "file scope inline asm" blocks. These blocks are internally
1049 concatenated by LLVM and treated as a single unit, but may be separated
1050 in the ``.ll`` file if desired. The syntax is very simple:
1052 .. code-block:: llvm
1054 module asm "inline asm code goes here"
1055 module asm "more can go here"
1057 The strings can contain any character by escaping non-printable
1058 characters. The escape sequence used is simply "\\xx" where "xx" is the
1059 two digit hex code for the number.
1061 The inline asm code is simply printed to the machine code .s file when
1062 assembly code is generated.
1064 .. _langref_datalayout:
1069 A module may specify a target specific data layout string that specifies
1070 how data is to be laid out in memory. The syntax for the data layout is
1073 .. code-block:: llvm
1075 target datalayout = "layout specification"
1077 The *layout specification* consists of a list of specifications
1078 separated by the minus sign character ('-'). Each specification starts
1079 with a letter and may include other information after the letter to
1080 define some aspect of the data layout. The specifications accepted are
1084 Specifies that the target lays out data in big-endian form. That is,
1085 the bits with the most significance have the lowest address
1088 Specifies that the target lays out data in little-endian form. That
1089 is, the bits with the least significance have the lowest address
1092 Specifies the natural alignment of the stack in bits. Alignment
1093 promotion of stack variables is limited to the natural stack
1094 alignment to avoid dynamic stack realignment. The stack alignment
1095 must be a multiple of 8-bits. If omitted, the natural stack
1096 alignment defaults to "unspecified", which does not prevent any
1097 alignment promotions.
1098 ``p[n]:<size>:<abi>:<pref>``
1099 This specifies the *size* of a pointer and its ``<abi>`` and
1100 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1101 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1102 preceding ``:`` should be omitted too. The address space, ``n`` is
1103 optional, and if not specified, denotes the default address space 0.
1104 The value of ``n`` must be in the range [1,2^23).
1105 ``i<size>:<abi>:<pref>``
1106 This specifies the alignment for an integer type of a given bit
1107 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1108 ``v<size>:<abi>:<pref>``
1109 This specifies the alignment for a vector type of a given bit
1111 ``f<size>:<abi>:<pref>``
1112 This specifies the alignment for a floating point type of a given bit
1113 ``<size>``. Only values of ``<size>`` that are supported by the target
1114 will work. 32 (float) and 64 (double) are supported on all targets; 80
1115 or 128 (different flavors of long double) are also supported on some
1117 ``a<size>:<abi>:<pref>``
1118 This specifies the alignment for an aggregate type of a given bit
1120 ``s<size>:<abi>:<pref>``
1121 This specifies the alignment for a stack object of a given bit
1123 ``n<size1>:<size2>:<size3>...``
1124 This specifies a set of native integer widths for the target CPU in
1125 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1126 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1127 this set are considered to support most general arithmetic operations
1130 When constructing the data layout for a given target, LLVM starts with a
1131 default set of specifications which are then (possibly) overridden by
1132 the specifications in the ``datalayout`` keyword. The default
1133 specifications are given in this list:
1135 - ``E`` - big endian
1136 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1137 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1138 same as the default address space.
1139 - ``S0`` - natural stack alignment is unspecified
1140 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1141 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1142 - ``i16:16:16`` - i16 is 16-bit aligned
1143 - ``i32:32:32`` - i32 is 32-bit aligned
1144 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1145 alignment of 64-bits
1146 - ``f16:16:16`` - half is 16-bit aligned
1147 - ``f32:32:32`` - float is 32-bit aligned
1148 - ``f64:64:64`` - double is 64-bit aligned
1149 - ``f128:128:128`` - quad is 128-bit aligned
1150 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1151 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1152 - ``a0:0:64`` - aggregates are 64-bit aligned
1154 When LLVM is determining the alignment for a given type, it uses the
1157 #. If the type sought is an exact match for one of the specifications,
1158 that specification is used.
1159 #. If no match is found, and the type sought is an integer type, then
1160 the smallest integer type that is larger than the bitwidth of the
1161 sought type is used. If none of the specifications are larger than
1162 the bitwidth then the largest integer type is used. For example,
1163 given the default specifications above, the i7 type will use the
1164 alignment of i8 (next largest) while both i65 and i256 will use the
1165 alignment of i64 (largest specified).
1166 #. If no match is found, and the type sought is a vector type, then the
1167 largest vector type that is smaller than the sought vector type will
1168 be used as a fall back. This happens because <128 x double> can be
1169 implemented in terms of 64 <2 x double>, for example.
1171 The function of the data layout string may not be what you expect.
1172 Notably, this is not a specification from the frontend of what alignment
1173 the code generator should use.
1175 Instead, if specified, the target data layout is required to match what
1176 the ultimate *code generator* expects. This string is used by the
1177 mid-level optimizers to improve code, and this only works if it matches
1178 what the ultimate code generator uses. If you would like to generate IR
1179 that does not embed this target-specific detail into the IR, then you
1180 don't have to specify the string. This will disable some optimizations
1181 that require precise layout information, but this also prevents those
1182 optimizations from introducing target specificity into the IR.
1189 A module may specify a target triple string that describes the target
1190 host. The syntax for the target triple is simply:
1192 .. code-block:: llvm
1194 target triple = "x86_64-apple-macosx10.7.0"
1196 The *target triple* string consists of a series of identifiers delimited
1197 by the minus sign character ('-'). The canonical forms are:
1201 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1202 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1204 This information is passed along to the backend so that it generates
1205 code for the proper architecture. It's possible to override this on the
1206 command line with the ``-mtriple`` command line option.
1208 .. _pointeraliasing:
1210 Pointer Aliasing Rules
1211 ----------------------
1213 Any memory access must be done through a pointer value associated with
1214 an address range of the memory access, otherwise the behavior is
1215 undefined. Pointer values are associated with address ranges according
1216 to the following rules:
1218 - A pointer value is associated with the addresses associated with any
1219 value it is *based* on.
1220 - An address of a global variable is associated with the address range
1221 of the variable's storage.
1222 - The result value of an allocation instruction is associated with the
1223 address range of the allocated storage.
1224 - A null pointer in the default address-space is associated with no
1226 - An integer constant other than zero or a pointer value returned from
1227 a function not defined within LLVM may be associated with address
1228 ranges allocated through mechanisms other than those provided by
1229 LLVM. Such ranges shall not overlap with any ranges of addresses
1230 allocated by mechanisms provided by LLVM.
1232 A pointer value is *based* on another pointer value according to the
1235 - A pointer value formed from a ``getelementptr`` operation is *based*
1236 on the first operand of the ``getelementptr``.
1237 - The result value of a ``bitcast`` is *based* on the operand of the
1239 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1240 values that contribute (directly or indirectly) to the computation of
1241 the pointer's value.
1242 - The "*based* on" relationship is transitive.
1244 Note that this definition of *"based"* is intentionally similar to the
1245 definition of *"based"* in C99, though it is slightly weaker.
1247 LLVM IR does not associate types with memory. The result type of a
1248 ``load`` merely indicates the size and alignment of the memory from
1249 which to load, as well as the interpretation of the value. The first
1250 operand type of a ``store`` similarly only indicates the size and
1251 alignment of the store.
1253 Consequently, type-based alias analysis, aka TBAA, aka
1254 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1255 :ref:`Metadata <metadata>` may be used to encode additional information
1256 which specialized optimization passes may use to implement type-based
1261 Volatile Memory Accesses
1262 ------------------------
1264 Certain memory accesses, such as :ref:`load <i_load>`'s,
1265 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1266 marked ``volatile``. The optimizers must not change the number of
1267 volatile operations or change their order of execution relative to other
1268 volatile operations. The optimizers *may* change the order of volatile
1269 operations relative to non-volatile operations. This is not Java's
1270 "volatile" and has no cross-thread synchronization behavior.
1272 IR-level volatile loads and stores cannot safely be optimized into
1273 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1274 flagged volatile. Likewise, the backend should never split or merge
1275 target-legal volatile load/store instructions.
1277 .. admonition:: Rationale
1279 Platforms may rely on volatile loads and stores of natively supported
1280 data width to be executed as single instruction. For example, in C
1281 this holds for an l-value of volatile primitive type with native
1282 hardware support, but not necessarily for aggregate types. The
1283 frontend upholds these expectations, which are intentionally
1284 unspecified in the IR. The rules above ensure that IR transformation
1285 do not violate the frontend's contract with the language.
1289 Memory Model for Concurrent Operations
1290 --------------------------------------
1292 The LLVM IR does not define any way to start parallel threads of
1293 execution or to register signal handlers. Nonetheless, there are
1294 platform-specific ways to create them, and we define LLVM IR's behavior
1295 in their presence. This model is inspired by the C++0x memory model.
1297 For a more informal introduction to this model, see the :doc:`Atomics`.
1299 We define a *happens-before* partial order as the least partial order
1302 - Is a superset of single-thread program order, and
1303 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1304 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1305 techniques, like pthread locks, thread creation, thread joining,
1306 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1307 Constraints <ordering>`).
1309 Note that program order does not introduce *happens-before* edges
1310 between a thread and signals executing inside that thread.
1312 Every (defined) read operation (load instructions, memcpy, atomic
1313 loads/read-modify-writes, etc.) R reads a series of bytes written by
1314 (defined) write operations (store instructions, atomic
1315 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1316 section, initialized globals are considered to have a write of the
1317 initializer which is atomic and happens before any other read or write
1318 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1319 may see any write to the same byte, except:
1321 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1322 write\ :sub:`2` happens before R\ :sub:`byte`, then
1323 R\ :sub:`byte` does not see write\ :sub:`1`.
1324 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1325 R\ :sub:`byte` does not see write\ :sub:`3`.
1327 Given that definition, R\ :sub:`byte` is defined as follows:
1329 - If R is volatile, the result is target-dependent. (Volatile is
1330 supposed to give guarantees which can support ``sig_atomic_t`` in
1331 C/C++, and may be used for accesses to addresses which do not behave
1332 like normal memory. It does not generally provide cross-thread
1334 - Otherwise, if there is no write to the same byte that happens before
1335 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1336 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1337 R\ :sub:`byte` returns the value written by that write.
1338 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1339 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1340 Memory Ordering Constraints <ordering>` section for additional
1341 constraints on how the choice is made.
1342 - Otherwise R\ :sub:`byte` returns ``undef``.
1344 R returns the value composed of the series of bytes it read. This
1345 implies that some bytes within the value may be ``undef`` **without**
1346 the entire value being ``undef``. Note that this only defines the
1347 semantics of the operation; it doesn't mean that targets will emit more
1348 than one instruction to read the series of bytes.
1350 Note that in cases where none of the atomic intrinsics are used, this
1351 model places only one restriction on IR transformations on top of what
1352 is required for single-threaded execution: introducing a store to a byte
1353 which might not otherwise be stored is not allowed in general.
1354 (Specifically, in the case where another thread might write to and read
1355 from an address, introducing a store can change a load that may see
1356 exactly one write into a load that may see multiple writes.)
1360 Atomic Memory Ordering Constraints
1361 ----------------------------------
1363 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1364 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1365 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1366 an ordering parameter that determines which other atomic instructions on
1367 the same address they *synchronize with*. These semantics are borrowed
1368 from Java and C++0x, but are somewhat more colloquial. If these
1369 descriptions aren't precise enough, check those specs (see spec
1370 references in the :doc:`atomics guide <Atomics>`).
1371 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1372 differently since they don't take an address. See that instruction's
1373 documentation for details.
1375 For a simpler introduction to the ordering constraints, see the
1379 The set of values that can be read is governed by the happens-before
1380 partial order. A value cannot be read unless some operation wrote
1381 it. This is intended to provide a guarantee strong enough to model
1382 Java's non-volatile shared variables. This ordering cannot be
1383 specified for read-modify-write operations; it is not strong enough
1384 to make them atomic in any interesting way.
1386 In addition to the guarantees of ``unordered``, there is a single
1387 total order for modifications by ``monotonic`` operations on each
1388 address. All modification orders must be compatible with the
1389 happens-before order. There is no guarantee that the modification
1390 orders can be combined to a global total order for the whole program
1391 (and this often will not be possible). The read in an atomic
1392 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1393 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1394 order immediately before the value it writes. If one atomic read
1395 happens before another atomic read of the same address, the later
1396 read must see the same value or a later value in the address's
1397 modification order. This disallows reordering of ``monotonic`` (or
1398 stronger) operations on the same address. If an address is written
1399 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1400 read that address repeatedly, the other threads must eventually see
1401 the write. This corresponds to the C++0x/C1x
1402 ``memory_order_relaxed``.
1404 In addition to the guarantees of ``monotonic``, a
1405 *synchronizes-with* edge may be formed with a ``release`` operation.
1406 This is intended to model C++'s ``memory_order_acquire``.
1408 In addition to the guarantees of ``monotonic``, if this operation
1409 writes a value which is subsequently read by an ``acquire``
1410 operation, it *synchronizes-with* that operation. (This isn't a
1411 complete description; see the C++0x definition of a release
1412 sequence.) This corresponds to the C++0x/C1x
1413 ``memory_order_release``.
1414 ``acq_rel`` (acquire+release)
1415 Acts as both an ``acquire`` and ``release`` operation on its
1416 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1417 ``seq_cst`` (sequentially consistent)
1418 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1419 operation which only reads, ``release`` for an operation which only
1420 writes), there is a global total order on all
1421 sequentially-consistent operations on all addresses, which is
1422 consistent with the *happens-before* partial order and with the
1423 modification orders of all the affected addresses. Each
1424 sequentially-consistent read sees the last preceding write to the
1425 same address in this global order. This corresponds to the C++0x/C1x
1426 ``memory_order_seq_cst`` and Java volatile.
1430 If an atomic operation is marked ``singlethread``, it only *synchronizes
1431 with* or participates in modification and seq\_cst total orderings with
1432 other operations running in the same thread (for example, in signal
1440 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1441 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1442 :ref:`frem <i_frem>`) have the following flags that can set to enable
1443 otherwise unsafe floating point operations
1446 No NaNs - Allow optimizations to assume the arguments and result are not
1447 NaN. Such optimizations are required to retain defined behavior over
1448 NaNs, but the value of the result is undefined.
1451 No Infs - Allow optimizations to assume the arguments and result are not
1452 +/-Inf. Such optimizations are required to retain defined behavior over
1453 +/-Inf, but the value of the result is undefined.
1456 No Signed Zeros - Allow optimizations to treat the sign of a zero
1457 argument or result as insignificant.
1460 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1461 argument rather than perform division.
1464 Fast - Allow algebraically equivalent transformations that may
1465 dramatically change results in floating point (e.g. reassociate). This
1466 flag implies all the others.
1473 The LLVM type system is one of the most important features of the
1474 intermediate representation. Being typed enables a number of
1475 optimizations to be performed on the intermediate representation
1476 directly, without having to do extra analyses on the side before the
1477 transformation. A strong type system makes it easier to read the
1478 generated code and enables novel analyses and transformations that are
1479 not feasible to perform on normal three address code representations.
1481 .. _typeclassifications:
1483 Type Classifications
1484 --------------------
1486 The types fall into a few useful classifications:
1495 * - :ref:`integer <t_integer>`
1496 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1499 * - :ref:`floating point <t_floating>`
1500 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1508 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1509 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1510 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1511 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1513 * - :ref:`primitive <t_primitive>`
1514 - :ref:`label <t_label>`,
1515 :ref:`void <t_void>`,
1516 :ref:`integer <t_integer>`,
1517 :ref:`floating point <t_floating>`,
1518 :ref:`x86mmx <t_x86mmx>`,
1519 :ref:`metadata <t_metadata>`.
1521 * - :ref:`derived <t_derived>`
1522 - :ref:`array <t_array>`,
1523 :ref:`function <t_function>`,
1524 :ref:`pointer <t_pointer>`,
1525 :ref:`structure <t_struct>`,
1526 :ref:`vector <t_vector>`,
1527 :ref:`opaque <t_opaque>`.
1529 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1530 Values of these types are the only ones which can be produced by
1538 The primitive types are the fundamental building blocks of the LLVM
1549 The integer type is a very simple type that simply specifies an
1550 arbitrary bit width for the integer type desired. Any bit width from 1
1551 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1560 The number of bits the integer will occupy is specified by the ``N``
1566 +----------------+------------------------------------------------+
1567 | ``i1`` | a single-bit integer. |
1568 +----------------+------------------------------------------------+
1569 | ``i32`` | a 32-bit integer. |
1570 +----------------+------------------------------------------------+
1571 | ``i1942652`` | a really big integer of over 1 million bits. |
1572 +----------------+------------------------------------------------+
1576 Floating Point Types
1577 ^^^^^^^^^^^^^^^^^^^^
1586 - 16-bit floating point value
1589 - 32-bit floating point value
1592 - 64-bit floating point value
1595 - 128-bit floating point value (112-bit mantissa)
1598 - 80-bit floating point value (X87)
1601 - 128-bit floating point value (two 64-bits)
1611 The x86mmx type represents a value held in an MMX register on an x86
1612 machine. The operations allowed on it are quite limited: parameters and
1613 return values, load and store, and bitcast. User-specified MMX
1614 instructions are represented as intrinsic or asm calls with arguments
1615 and/or results of this type. There are no arrays, vectors or constants
1633 The void type does not represent any value and has no size.
1650 The label type represents code labels.
1667 The metadata type represents embedded metadata. No derived types may be
1668 created from metadata except for :ref:`function <t_function>` arguments.
1682 The real power in LLVM comes from the derived types in the system. This
1683 is what allows a programmer to represent arrays, functions, pointers,
1684 and other useful types. Each of these types contain one or more element
1685 types which may be a primitive type, or another derived type. For
1686 example, it is possible to have a two dimensional array, using an array
1687 as the element type of another array.
1694 Aggregate Types are a subset of derived types that can contain multiple
1695 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1696 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1707 The array type is a very simple derived type that arranges elements
1708 sequentially in memory. The array type requires a size (number of
1709 elements) and an underlying data type.
1716 [<# elements> x <elementtype>]
1718 The number of elements is a constant integer value; ``elementtype`` may
1719 be any type with a size.
1724 +------------------+--------------------------------------+
1725 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1726 +------------------+--------------------------------------+
1727 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1728 +------------------+--------------------------------------+
1729 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1730 +------------------+--------------------------------------+
1732 Here are some examples of multidimensional arrays:
1734 +-----------------------------+----------------------------------------------------------+
1735 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1736 +-----------------------------+----------------------------------------------------------+
1737 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1738 +-----------------------------+----------------------------------------------------------+
1739 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1740 +-----------------------------+----------------------------------------------------------+
1742 There is no restriction on indexing beyond the end of the array implied
1743 by a static type (though there are restrictions on indexing beyond the
1744 bounds of an allocated object in some cases). This means that
1745 single-dimension 'variable sized array' addressing can be implemented in
1746 LLVM with a zero length array type. An implementation of 'pascal style
1747 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1758 The function type can be thought of as a function signature. It consists of a
1759 return type and a list of formal parameter types. The return type of a function
1760 type is a void type or first class type --- except for :ref:`label <t_label>`
1761 and :ref:`metadata <t_metadata>` types.
1768 <returntype> (<parameter list>)
1770 ...where '``<parameter list>``' is a comma-separated list of type
1771 specifiers. Optionally, the parameter list may include a type ``...``, which
1772 indicates that the function takes a variable number of arguments. Variable
1773 argument functions can access their arguments with the :ref:`variable argument
1774 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1775 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1780 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1781 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1782 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1783 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1784 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1785 | ``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. |
1786 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1787 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1788 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1798 The structure type is used to represent a collection of data members
1799 together in memory. The elements of a structure may be any type that has
1802 Structures in memory are accessed using '``load``' and '``store``' by
1803 getting a pointer to a field with the '``getelementptr``' instruction.
1804 Structures in registers are accessed using the '``extractvalue``' and
1805 '``insertvalue``' instructions.
1807 Structures may optionally be "packed" structures, which indicate that
1808 the alignment of the struct is one byte, and that there is no padding
1809 between the elements. In non-packed structs, padding between field types
1810 is inserted as defined by the DataLayout string in the module, which is
1811 required to match what the underlying code generator expects.
1813 Structures can either be "literal" or "identified". A literal structure
1814 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1815 identified types are always defined at the top level with a name.
1816 Literal types are uniqued by their contents and can never be recursive
1817 or opaque since there is no way to write one. Identified types can be
1818 recursive, can be opaqued, and are never uniqued.
1825 %T1 = type { <type list> } ; Identified normal struct type
1826 %T2 = type <{ <type list> }> ; Identified packed struct type
1831 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1832 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1833 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1834 | ``{ 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``. |
1835 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1836 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1837 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1841 Opaque Structure Types
1842 ^^^^^^^^^^^^^^^^^^^^^^
1847 Opaque structure types are used to represent named structure types that
1848 do not have a body specified. This corresponds (for example) to the C
1849 notion of a forward declared structure.
1862 +--------------+-------------------+
1863 | ``opaque`` | An opaque type. |
1864 +--------------+-------------------+
1874 The pointer type is used to specify memory locations. Pointers are
1875 commonly used to reference objects in memory.
1877 Pointer types may have an optional address space attribute defining the
1878 numbered address space where the pointed-to object resides. The default
1879 address space is number zero. The semantics of non-zero address spaces
1880 are target-specific.
1882 Note that LLVM does not permit pointers to void (``void*``) nor does it
1883 permit pointers to labels (``label*``). Use ``i8*`` instead.
1895 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1896 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1897 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1898 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1899 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1900 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1901 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1911 A vector type is a simple derived type that represents a vector of
1912 elements. Vector types are used when multiple primitive data are
1913 operated in parallel using a single instruction (SIMD). A vector type
1914 requires a size (number of elements) and an underlying primitive data
1915 type. Vector types are considered :ref:`first class <t_firstclass>`.
1922 < <# elements> x <elementtype> >
1924 The number of elements is a constant integer value larger than 0;
1925 elementtype may be any integer or floating point type, or a pointer to
1926 these types. Vectors of size zero are not allowed.
1931 +-------------------+--------------------------------------------------+
1932 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1933 +-------------------+--------------------------------------------------+
1934 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1935 +-------------------+--------------------------------------------------+
1936 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1937 +-------------------+--------------------------------------------------+
1938 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1939 +-------------------+--------------------------------------------------+
1944 LLVM has several different basic types of constants. This section
1945 describes them all and their syntax.
1950 **Boolean constants**
1951 The two strings '``true``' and '``false``' are both valid constants
1953 **Integer constants**
1954 Standard integers (such as '4') are constants of the
1955 :ref:`integer <t_integer>` type. Negative numbers may be used with
1957 **Floating point constants**
1958 Floating point constants use standard decimal notation (e.g.
1959 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1960 hexadecimal notation (see below). The assembler requires the exact
1961 decimal value of a floating-point constant. For example, the
1962 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1963 decimal in binary. Floating point constants must have a :ref:`floating
1964 point <t_floating>` type.
1965 **Null pointer constants**
1966 The identifier '``null``' is recognized as a null pointer constant
1967 and must be of :ref:`pointer type <t_pointer>`.
1969 The one non-intuitive notation for constants is the hexadecimal form of
1970 floating point constants. For example, the form
1971 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1972 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1973 constants are required (and the only time that they are generated by the
1974 disassembler) is when a floating point constant must be emitted but it
1975 cannot be represented as a decimal floating point number in a reasonable
1976 number of digits. For example, NaN's, infinities, and other special
1977 values are represented in their IEEE hexadecimal format so that assembly
1978 and disassembly do not cause any bits to change in the constants.
1980 When using the hexadecimal form, constants of types half, float, and
1981 double are represented using the 16-digit form shown above (which
1982 matches the IEEE754 representation for double); half and float values
1983 must, however, be exactly representable as IEEE 754 half and single
1984 precision, respectively. Hexadecimal format is always used for long
1985 double, and there are three forms of long double. The 80-bit format used
1986 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1987 128-bit format used by PowerPC (two adjacent doubles) is represented by
1988 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1989 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1990 will only work if they match the long double format on your target.
1991 The IEEE 16-bit format (half precision) is represented by ``0xH``
1992 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1993 (sign bit at the left).
1995 There are no constants of type x86mmx.
1997 .. _complexconstants:
2002 Complex constants are a (potentially recursive) combination of simple
2003 constants and smaller complex constants.
2005 **Structure constants**
2006 Structure constants are represented with notation similar to
2007 structure type definitions (a comma separated list of elements,
2008 surrounded by braces (``{}``)). For example:
2009 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2010 "``@G = external global i32``". Structure constants must have
2011 :ref:`structure type <t_struct>`, and the number and types of elements
2012 must match those specified by the type.
2014 Array constants are represented with notation similar to array type
2015 definitions (a comma separated list of elements, surrounded by
2016 square brackets (``[]``)). For example:
2017 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2018 :ref:`array type <t_array>`, and the number and types of elements must
2019 match those specified by the type.
2020 **Vector constants**
2021 Vector constants are represented with notation similar to vector
2022 type definitions (a comma separated list of elements, surrounded by
2023 less-than/greater-than's (``<>``)). For example:
2024 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2025 must have :ref:`vector type <t_vector>`, and the number and types of
2026 elements must match those specified by the type.
2027 **Zero initialization**
2028 The string '``zeroinitializer``' can be used to zero initialize a
2029 value to zero of *any* type, including scalar and
2030 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2031 having to print large zero initializers (e.g. for large arrays) and
2032 is always exactly equivalent to using explicit zero initializers.
2034 A metadata node is a structure-like constant with :ref:`metadata
2035 type <t_metadata>`. For example:
2036 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2037 constants that are meant to be interpreted as part of the
2038 instruction stream, metadata is a place to attach additional
2039 information such as debug info.
2041 Global Variable and Function Addresses
2042 --------------------------------------
2044 The addresses of :ref:`global variables <globalvars>` and
2045 :ref:`functions <functionstructure>` are always implicitly valid
2046 (link-time) constants. These constants are explicitly referenced when
2047 the :ref:`identifier for the global <identifiers>` is used and always have
2048 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2051 .. code-block:: llvm
2055 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2062 The string '``undef``' can be used anywhere a constant is expected, and
2063 indicates that the user of the value may receive an unspecified
2064 bit-pattern. Undefined values may be of any type (other than '``label``'
2065 or '``void``') and be used anywhere a constant is permitted.
2067 Undefined values are useful because they indicate to the compiler that
2068 the program is well defined no matter what value is used. This gives the
2069 compiler more freedom to optimize. Here are some examples of
2070 (potentially surprising) transformations that are valid (in pseudo IR):
2072 .. code-block:: llvm
2082 This is safe because all of the output bits are affected by the undef
2083 bits. Any output bit can have a zero or one depending on the input bits.
2085 .. code-block:: llvm
2096 These logical operations have bits that are not always affected by the
2097 input. For example, if ``%X`` has a zero bit, then the output of the
2098 '``and``' operation will always be a zero for that bit, no matter what
2099 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2100 optimize or assume that the result of the '``and``' is '``undef``'.
2101 However, it is safe to assume that all bits of the '``undef``' could be
2102 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2103 all the bits of the '``undef``' operand to the '``or``' could be set,
2104 allowing the '``or``' to be folded to -1.
2106 .. code-block:: llvm
2108 %A = select undef, %X, %Y
2109 %B = select undef, 42, %Y
2110 %C = select %X, %Y, undef
2120 This set of examples shows that undefined '``select``' (and conditional
2121 branch) conditions can go *either way*, but they have to come from one
2122 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2123 both known to have a clear low bit, then ``%A`` would have to have a
2124 cleared low bit. However, in the ``%C`` example, the optimizer is
2125 allowed to assume that the '``undef``' operand could be the same as
2126 ``%Y``, allowing the whole '``select``' to be eliminated.
2128 .. code-block:: llvm
2130 %A = xor undef, undef
2147 This example points out that two '``undef``' operands are not
2148 necessarily the same. This can be surprising to people (and also matches
2149 C semantics) where they assume that "``X^X``" is always zero, even if
2150 ``X`` is undefined. This isn't true for a number of reasons, but the
2151 short answer is that an '``undef``' "variable" can arbitrarily change
2152 its value over its "live range". This is true because the variable
2153 doesn't actually *have a live range*. Instead, the value is logically
2154 read from arbitrary registers that happen to be around when needed, so
2155 the value is not necessarily consistent over time. In fact, ``%A`` and
2156 ``%C`` need to have the same semantics or the core LLVM "replace all
2157 uses with" concept would not hold.
2159 .. code-block:: llvm
2167 These examples show the crucial difference between an *undefined value*
2168 and *undefined behavior*. An undefined value (like '``undef``') is
2169 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2170 operation can be constant folded to '``undef``', because the '``undef``'
2171 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2172 However, in the second example, we can make a more aggressive
2173 assumption: because the ``undef`` is allowed to be an arbitrary value,
2174 we are allowed to assume that it could be zero. Since a divide by zero
2175 has *undefined behavior*, we are allowed to assume that the operation
2176 does not execute at all. This allows us to delete the divide and all
2177 code after it. Because the undefined operation "can't happen", the
2178 optimizer can assume that it occurs in dead code.
2180 .. code-block:: llvm
2182 a: store undef -> %X
2183 b: store %X -> undef
2188 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2189 value can be assumed to not have any effect; we can assume that the
2190 value is overwritten with bits that happen to match what was already
2191 there. However, a store *to* an undefined location could clobber
2192 arbitrary memory, therefore, it has undefined behavior.
2199 Poison values are similar to :ref:`undef values <undefvalues>`, however
2200 they also represent the fact that an instruction or constant expression
2201 which cannot evoke side effects has nevertheless detected a condition
2202 which results in undefined behavior.
2204 There is currently no way of representing a poison value in the IR; they
2205 only exist when produced by operations such as :ref:`add <i_add>` with
2208 Poison value behavior is defined in terms of value *dependence*:
2210 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2211 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2212 their dynamic predecessor basic block.
2213 - Function arguments depend on the corresponding actual argument values
2214 in the dynamic callers of their functions.
2215 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2216 instructions that dynamically transfer control back to them.
2217 - :ref:`Invoke <i_invoke>` instructions depend on the
2218 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2219 call instructions that dynamically transfer control back to them.
2220 - Non-volatile loads and stores depend on the most recent stores to all
2221 of the referenced memory addresses, following the order in the IR
2222 (including loads and stores implied by intrinsics such as
2223 :ref:`@llvm.memcpy <int_memcpy>`.)
2224 - An instruction with externally visible side effects depends on the
2225 most recent preceding instruction with externally visible side
2226 effects, following the order in the IR. (This includes :ref:`volatile
2227 operations <volatile>`.)
2228 - An instruction *control-depends* on a :ref:`terminator
2229 instruction <terminators>` if the terminator instruction has
2230 multiple successors and the instruction is always executed when
2231 control transfers to one of the successors, and may not be executed
2232 when control is transferred to another.
2233 - Additionally, an instruction also *control-depends* on a terminator
2234 instruction if the set of instructions it otherwise depends on would
2235 be different if the terminator had transferred control to a different
2237 - Dependence is transitive.
2239 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2240 with the additional affect that any instruction which has a *dependence*
2241 on a poison value has undefined behavior.
2243 Here are some examples:
2245 .. code-block:: llvm
2248 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2249 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2250 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2251 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2253 store i32 %poison, i32* @g ; Poison value stored to memory.
2254 %poison2 = load i32* @g ; Poison value loaded back from memory.
2256 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2258 %narrowaddr = bitcast i32* @g to i16*
2259 %wideaddr = bitcast i32* @g to i64*
2260 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2261 %poison4 = load i64* %wideaddr ; Returns a poison value.
2263 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2264 br i1 %cmp, label %true, label %end ; Branch to either destination.
2267 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2268 ; it has undefined behavior.
2272 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2273 ; Both edges into this PHI are
2274 ; control-dependent on %cmp, so this
2275 ; always results in a poison value.
2277 store volatile i32 0, i32* @g ; This would depend on the store in %true
2278 ; if %cmp is true, or the store in %entry
2279 ; otherwise, so this is undefined behavior.
2281 br i1 %cmp, label %second_true, label %second_end
2282 ; The same branch again, but this time the
2283 ; true block doesn't have side effects.
2290 store volatile i32 0, i32* @g ; This time, the instruction always depends
2291 ; on the store in %end. Also, it is
2292 ; control-equivalent to %end, so this is
2293 ; well-defined (ignoring earlier undefined
2294 ; behavior in this example).
2298 Addresses of Basic Blocks
2299 -------------------------
2301 ``blockaddress(@function, %block)``
2303 The '``blockaddress``' constant computes the address of the specified
2304 basic block in the specified function, and always has an ``i8*`` type.
2305 Taking the address of the entry block is illegal.
2307 This value only has defined behavior when used as an operand to the
2308 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2309 against null. Pointer equality tests between labels addresses results in
2310 undefined behavior --- though, again, comparison against null is ok, and
2311 no label is equal to the null pointer. This may be passed around as an
2312 opaque pointer sized value as long as the bits are not inspected. This
2313 allows ``ptrtoint`` and arithmetic to be performed on these values so
2314 long as the original value is reconstituted before the ``indirectbr``
2317 Finally, some targets may provide defined semantics when using the value
2318 as the operand to an inline assembly, but that is target specific.
2322 Constant Expressions
2323 --------------------
2325 Constant expressions are used to allow expressions involving other
2326 constants to be used as constants. Constant expressions may be of any
2327 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2328 that does not have side effects (e.g. load and call are not supported).
2329 The following is the syntax for constant expressions:
2331 ``trunc (CST to TYPE)``
2332 Truncate a constant to another type. The bit size of CST must be
2333 larger than the bit size of TYPE. Both types must be integers.
2334 ``zext (CST to TYPE)``
2335 Zero extend a constant to another type. The bit size of CST must be
2336 smaller than the bit size of TYPE. Both types must be integers.
2337 ``sext (CST to TYPE)``
2338 Sign extend a constant to another type. The bit size of CST must be
2339 smaller than the bit size of TYPE. Both types must be integers.
2340 ``fptrunc (CST to TYPE)``
2341 Truncate a floating point constant to another floating point type.
2342 The size of CST must be larger than the size of TYPE. Both types
2343 must be floating point.
2344 ``fpext (CST to TYPE)``
2345 Floating point extend a constant to another type. The size of CST
2346 must be smaller or equal to the size of TYPE. Both types must be
2348 ``fptoui (CST to TYPE)``
2349 Convert a floating point constant to the corresponding unsigned
2350 integer constant. TYPE must be a scalar or vector integer type. CST
2351 must be of scalar or vector floating point type. Both CST and TYPE
2352 must be scalars, or vectors of the same number of elements. If the
2353 value won't fit in the integer type, the results are undefined.
2354 ``fptosi (CST to TYPE)``
2355 Convert a floating point constant to the corresponding signed
2356 integer constant. TYPE must be a scalar or vector integer type. CST
2357 must be of scalar or vector floating point type. Both CST and TYPE
2358 must be scalars, or vectors of the same number of elements. If the
2359 value won't fit in the integer type, the results are undefined.
2360 ``uitofp (CST to TYPE)``
2361 Convert an unsigned integer constant to the corresponding floating
2362 point constant. TYPE must be a scalar or vector floating point type.
2363 CST must be of scalar or vector integer type. Both CST and TYPE must
2364 be scalars, or vectors of the same number of elements. If the value
2365 won't fit in the floating point type, the results are undefined.
2366 ``sitofp (CST to TYPE)``
2367 Convert a signed integer constant to the corresponding floating
2368 point constant. TYPE must be a scalar or vector floating point type.
2369 CST must be of scalar or vector integer type. Both CST and TYPE must
2370 be scalars, or vectors of the same number of elements. If the value
2371 won't fit in the floating point type, the results are undefined.
2372 ``ptrtoint (CST to TYPE)``
2373 Convert a pointer typed constant to the corresponding integer
2374 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2375 pointer type. The ``CST`` value is zero extended, truncated, or
2376 unchanged to make it fit in ``TYPE``.
2377 ``inttoptr (CST to TYPE)``
2378 Convert an integer constant to a pointer constant. TYPE must be a
2379 pointer type. CST must be of integer type. The CST value is zero
2380 extended, truncated, or unchanged to make it fit in a pointer size.
2381 This one is *really* dangerous!
2382 ``bitcast (CST to TYPE)``
2383 Convert a constant, CST, to another TYPE. The constraints of the
2384 operands are the same as those for the :ref:`bitcast
2385 instruction <i_bitcast>`.
2386 ``addrspacecast (CST to TYPE)``
2387 Convert a constant pointer or constant vector of pointer, CST, to another
2388 TYPE in a different address space. The constraints of the operands are the
2389 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2390 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2391 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2392 constants. As with the :ref:`getelementptr <i_getelementptr>`
2393 instruction, the index list may have zero or more indexes, which are
2394 required to make sense for the type of "CSTPTR".
2395 ``select (COND, VAL1, VAL2)``
2396 Perform the :ref:`select operation <i_select>` on constants.
2397 ``icmp COND (VAL1, VAL2)``
2398 Performs the :ref:`icmp operation <i_icmp>` on constants.
2399 ``fcmp COND (VAL1, VAL2)``
2400 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2401 ``extractelement (VAL, IDX)``
2402 Perform the :ref:`extractelement operation <i_extractelement>` on
2404 ``insertelement (VAL, ELT, IDX)``
2405 Perform the :ref:`insertelement operation <i_insertelement>` on
2407 ``shufflevector (VEC1, VEC2, IDXMASK)``
2408 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2410 ``extractvalue (VAL, IDX0, IDX1, ...)``
2411 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2412 constants. The index list is interpreted in a similar manner as
2413 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2414 least one index value must be specified.
2415 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2416 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2417 The index list is interpreted in a similar manner as indices in a
2418 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2419 value must be specified.
2420 ``OPCODE (LHS, RHS)``
2421 Perform the specified operation of the LHS and RHS constants. OPCODE
2422 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2423 binary <bitwiseops>` operations. The constraints on operands are
2424 the same as those for the corresponding instruction (e.g. no bitwise
2425 operations on floating point values are allowed).
2432 Inline Assembler Expressions
2433 ----------------------------
2435 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2436 Inline Assembly <moduleasm>`) through the use of a special value. This
2437 value represents the inline assembler as a string (containing the
2438 instructions to emit), a list of operand constraints (stored as a
2439 string), a flag that indicates whether or not the inline asm expression
2440 has side effects, and a flag indicating whether the function containing
2441 the asm needs to align its stack conservatively. An example inline
2442 assembler expression is:
2444 .. code-block:: llvm
2446 i32 (i32) asm "bswap $0", "=r,r"
2448 Inline assembler expressions may **only** be used as the callee operand
2449 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2450 Thus, typically we have:
2452 .. code-block:: llvm
2454 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2456 Inline asms with side effects not visible in the constraint list must be
2457 marked as having side effects. This is done through the use of the
2458 '``sideeffect``' keyword, like so:
2460 .. code-block:: llvm
2462 call void asm sideeffect "eieio", ""()
2464 In some cases inline asms will contain code that will not work unless
2465 the stack is aligned in some way, such as calls or SSE instructions on
2466 x86, yet will not contain code that does that alignment within the asm.
2467 The compiler should make conservative assumptions about what the asm
2468 might contain and should generate its usual stack alignment code in the
2469 prologue if the '``alignstack``' keyword is present:
2471 .. code-block:: llvm
2473 call void asm alignstack "eieio", ""()
2475 Inline asms also support using non-standard assembly dialects. The
2476 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2477 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2478 the only supported dialects. An example is:
2480 .. code-block:: llvm
2482 call void asm inteldialect "eieio", ""()
2484 If multiple keywords appear the '``sideeffect``' keyword must come
2485 first, the '``alignstack``' keyword second and the '``inteldialect``'
2491 The call instructions that wrap inline asm nodes may have a
2492 "``!srcloc``" MDNode attached to it that contains a list of constant
2493 integers. If present, the code generator will use the integer as the
2494 location cookie value when report errors through the ``LLVMContext``
2495 error reporting mechanisms. This allows a front-end to correlate backend
2496 errors that occur with inline asm back to the source code that produced
2499 .. code-block:: llvm
2501 call void asm sideeffect "something bad", ""(), !srcloc !42
2503 !42 = !{ i32 1234567 }
2505 It is up to the front-end to make sense of the magic numbers it places
2506 in the IR. If the MDNode contains multiple constants, the code generator
2507 will use the one that corresponds to the line of the asm that the error
2512 Metadata Nodes and Metadata Strings
2513 -----------------------------------
2515 LLVM IR allows metadata to be attached to instructions in the program
2516 that can convey extra information about the code to the optimizers and
2517 code generator. One example application of metadata is source-level
2518 debug information. There are two metadata primitives: strings and nodes.
2519 All metadata has the ``metadata`` type and is identified in syntax by a
2520 preceding exclamation point ('``!``').
2522 A metadata string is a string surrounded by double quotes. It can
2523 contain any character by escaping non-printable characters with
2524 "``\xx``" where "``xx``" is the two digit hex code. For example:
2527 Metadata nodes are represented with notation similar to structure
2528 constants (a comma separated list of elements, surrounded by braces and
2529 preceded by an exclamation point). Metadata nodes can have any values as
2530 their operand. For example:
2532 .. code-block:: llvm
2534 !{ metadata !"test\00", i32 10}
2536 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2537 metadata nodes, which can be looked up in the module symbol table. For
2540 .. code-block:: llvm
2542 !foo = metadata !{!4, !3}
2544 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2545 function is using two metadata arguments:
2547 .. code-block:: llvm
2549 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2551 Metadata can be attached with an instruction. Here metadata ``!21`` is
2552 attached to the ``add`` instruction using the ``!dbg`` identifier:
2554 .. code-block:: llvm
2556 %indvar.next = add i64 %indvar, 1, !dbg !21
2558 More information about specific metadata nodes recognized by the
2559 optimizers and code generator is found below.
2564 In LLVM IR, memory does not have types, so LLVM's own type system is not
2565 suitable for doing TBAA. Instead, metadata is added to the IR to
2566 describe a type system of a higher level language. This can be used to
2567 implement typical C/C++ TBAA, but it can also be used to implement
2568 custom alias analysis behavior for other languages.
2570 The current metadata format is very simple. TBAA metadata nodes have up
2571 to three fields, e.g.:
2573 .. code-block:: llvm
2575 !0 = metadata !{ metadata !"an example type tree" }
2576 !1 = metadata !{ metadata !"int", metadata !0 }
2577 !2 = metadata !{ metadata !"float", metadata !0 }
2578 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2580 The first field is an identity field. It can be any value, usually a
2581 metadata string, which uniquely identifies the type. The most important
2582 name in the tree is the name of the root node. Two trees with different
2583 root node names are entirely disjoint, even if they have leaves with
2586 The second field identifies the type's parent node in the tree, or is
2587 null or omitted for a root node. A type is considered to alias all of
2588 its descendants and all of its ancestors in the tree. Also, a type is
2589 considered to alias all types in other trees, so that bitcode produced
2590 from multiple front-ends is handled conservatively.
2592 If the third field is present, it's an integer which if equal to 1
2593 indicates that the type is "constant" (meaning
2594 ``pointsToConstantMemory`` should return true; see `other useful
2595 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2597 '``tbaa.struct``' Metadata
2598 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2600 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2601 aggregate assignment operations in C and similar languages, however it
2602 is defined to copy a contiguous region of memory, which is more than
2603 strictly necessary for aggregate types which contain holes due to
2604 padding. Also, it doesn't contain any TBAA information about the fields
2607 ``!tbaa.struct`` metadata can describe which memory subregions in a
2608 memcpy are padding and what the TBAA tags of the struct are.
2610 The current metadata format is very simple. ``!tbaa.struct`` metadata
2611 nodes are a list of operands which are in conceptual groups of three.
2612 For each group of three, the first operand gives the byte offset of a
2613 field in bytes, the second gives its size in bytes, and the third gives
2616 .. code-block:: llvm
2618 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2620 This describes a struct with two fields. The first is at offset 0 bytes
2621 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2622 and has size 4 bytes and has tbaa tag !2.
2624 Note that the fields need not be contiguous. In this example, there is a
2625 4 byte gap between the two fields. This gap represents padding which
2626 does not carry useful data and need not be preserved.
2628 '``fpmath``' Metadata
2629 ^^^^^^^^^^^^^^^^^^^^^
2631 ``fpmath`` metadata may be attached to any instruction of floating point
2632 type. It can be used to express the maximum acceptable error in the
2633 result of that instruction, in ULPs, thus potentially allowing the
2634 compiler to use a more efficient but less accurate method of computing
2635 it. ULP is defined as follows:
2637 If ``x`` is a real number that lies between two finite consecutive
2638 floating-point numbers ``a`` and ``b``, without being equal to one
2639 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2640 distance between the two non-equal finite floating-point numbers
2641 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2643 The metadata node shall consist of a single positive floating point
2644 number representing the maximum relative error, for example:
2646 .. code-block:: llvm
2648 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2650 '``range``' Metadata
2651 ^^^^^^^^^^^^^^^^^^^^
2653 ``range`` metadata may be attached only to loads of integer types. It
2654 expresses the possible ranges the loaded value is in. The ranges are
2655 represented with a flattened list of integers. The loaded value is known
2656 to be in the union of the ranges defined by each consecutive pair. Each
2657 pair has the following properties:
2659 - The type must match the type loaded by the instruction.
2660 - The pair ``a,b`` represents the range ``[a,b)``.
2661 - Both ``a`` and ``b`` are constants.
2662 - The range is allowed to wrap.
2663 - The range should not represent the full or empty set. That is,
2666 In addition, the pairs must be in signed order of the lower bound and
2667 they must be non-contiguous.
2671 .. code-block:: llvm
2673 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2674 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2675 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2676 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2678 !0 = metadata !{ i8 0, i8 2 }
2679 !1 = metadata !{ i8 255, i8 2 }
2680 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2681 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2686 It is sometimes useful to attach information to loop constructs. Currently,
2687 loop metadata is implemented as metadata attached to the branch instruction
2688 in the loop latch block. This type of metadata refer to a metadata node that is
2689 guaranteed to be separate for each loop. The loop identifier metadata is
2690 specified with the name ``llvm.loop``.
2692 The loop identifier metadata is implemented using a metadata that refers to
2693 itself to avoid merging it with any other identifier metadata, e.g.,
2694 during module linkage or function inlining. That is, each loop should refer
2695 to their own identification metadata even if they reside in separate functions.
2696 The following example contains loop identifier metadata for two separate loop
2699 .. code-block:: llvm
2701 !0 = metadata !{ metadata !0 }
2702 !1 = metadata !{ metadata !1 }
2704 The loop identifier metadata can be used to specify additional per-loop
2705 metadata. Any operands after the first operand can be treated as user-defined
2706 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2707 by the loop vectorizer to indicate how many times to unroll the loop:
2709 .. code-block:: llvm
2711 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2713 !0 = metadata !{ metadata !0, metadata !1 }
2714 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2719 Metadata types used to annotate memory accesses with information helpful
2720 for optimizations are prefixed with ``llvm.mem``.
2722 '``llvm.mem.parallel_loop_access``' Metadata
2723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2725 For a loop to be parallel, in addition to using
2726 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2727 also all of the memory accessing instructions in the loop body need to be
2728 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2729 is at least one memory accessing instruction not marked with the metadata,
2730 the loop must be considered a sequential loop. This causes parallel loops to be
2731 converted to sequential loops due to optimization passes that are unaware of
2732 the parallel semantics and that insert new memory instructions to the loop
2735 Example of a loop that is considered parallel due to its correct use of
2736 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2737 metadata types that refer to the same loop identifier metadata.
2739 .. code-block:: llvm
2743 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2745 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2747 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2751 !0 = metadata !{ metadata !0 }
2753 It is also possible to have nested parallel loops. In that case the
2754 memory accesses refer to a list of loop identifier metadata nodes instead of
2755 the loop identifier metadata node directly:
2757 .. code-block:: llvm
2764 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2766 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2768 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2772 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2774 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2776 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2778 outer.for.end: ; preds = %for.body
2780 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2781 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2782 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2784 '``llvm.vectorizer``'
2785 ^^^^^^^^^^^^^^^^^^^^^
2787 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2788 vectorization parameters such as vectorization factor and unroll factor.
2790 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2791 loop identification metadata.
2793 '``llvm.vectorizer.unroll``' Metadata
2794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2796 This metadata instructs the loop vectorizer to unroll the specified
2797 loop exactly ``N`` times.
2799 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2800 operand is an integer specifying the unroll factor. For example:
2802 .. code-block:: llvm
2804 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2806 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2809 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2810 determined automatically.
2812 '``llvm.vectorizer.width``' Metadata
2813 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2815 This metadata sets the target width of the vectorizer to ``N``. Without
2816 this metadata, the vectorizer will choose a width automatically.
2817 Regardless of this metadata, the vectorizer will only vectorize loops if
2818 it believes it is valid to do so.
2820 The first operand is the string ``llvm.vectorizer.width`` and the second
2821 operand is an integer specifying the width. For example:
2823 .. code-block:: llvm
2825 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2827 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2830 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2833 Module Flags Metadata
2834 =====================
2836 Information about the module as a whole is difficult to convey to LLVM's
2837 subsystems. The LLVM IR isn't sufficient to transmit this information.
2838 The ``llvm.module.flags`` named metadata exists in order to facilitate
2839 this. These flags are in the form of key / value pairs --- much like a
2840 dictionary --- making it easy for any subsystem who cares about a flag to
2843 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2844 Each triplet has the following form:
2846 - The first element is a *behavior* flag, which specifies the behavior
2847 when two (or more) modules are merged together, and it encounters two
2848 (or more) metadata with the same ID. The supported behaviors are
2850 - The second element is a metadata string that is a unique ID for the
2851 metadata. Each module may only have one flag entry for each unique ID (not
2852 including entries with the **Require** behavior).
2853 - The third element is the value of the flag.
2855 When two (or more) modules are merged together, the resulting
2856 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2857 each unique metadata ID string, there will be exactly one entry in the merged
2858 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2859 be determined by the merge behavior flag, as described below. The only exception
2860 is that entries with the *Require* behavior are always preserved.
2862 The following behaviors are supported:
2873 Emits an error if two values disagree, otherwise the resulting value
2874 is that of the operands.
2878 Emits a warning if two values disagree. The result value will be the
2879 operand for the flag from the first module being linked.
2883 Adds a requirement that another module flag be present and have a
2884 specified value after linking is performed. The value must be a
2885 metadata pair, where the first element of the pair is the ID of the
2886 module flag to be restricted, and the second element of the pair is
2887 the value the module flag should be restricted to. This behavior can
2888 be used to restrict the allowable results (via triggering of an
2889 error) of linking IDs with the **Override** behavior.
2893 Uses the specified value, regardless of the behavior or value of the
2894 other module. If both modules specify **Override**, but the values
2895 differ, an error will be emitted.
2899 Appends the two values, which are required to be metadata nodes.
2903 Appends the two values, which are required to be metadata
2904 nodes. However, duplicate entries in the second list are dropped
2905 during the append operation.
2907 It is an error for a particular unique flag ID to have multiple behaviors,
2908 except in the case of **Require** (which adds restrictions on another metadata
2909 value) or **Override**.
2911 An example of module flags:
2913 .. code-block:: llvm
2915 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2916 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2917 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2918 !3 = metadata !{ i32 3, metadata !"qux",
2920 metadata !"foo", i32 1
2923 !llvm.module.flags = !{ !0, !1, !2, !3 }
2925 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2926 if two or more ``!"foo"`` flags are seen is to emit an error if their
2927 values are not equal.
2929 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2930 behavior if two or more ``!"bar"`` flags are seen is to use the value
2933 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2934 behavior if two or more ``!"qux"`` flags are seen is to emit a
2935 warning if their values are not equal.
2937 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2941 metadata !{ metadata !"foo", i32 1 }
2943 The behavior is to emit an error if the ``llvm.module.flags`` does not
2944 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2947 Objective-C Garbage Collection Module Flags Metadata
2948 ----------------------------------------------------
2950 On the Mach-O platform, Objective-C stores metadata about garbage
2951 collection in a special section called "image info". The metadata
2952 consists of a version number and a bitmask specifying what types of
2953 garbage collection are supported (if any) by the file. If two or more
2954 modules are linked together their garbage collection metadata needs to
2955 be merged rather than appended together.
2957 The Objective-C garbage collection module flags metadata consists of the
2958 following key-value pairs:
2967 * - ``Objective-C Version``
2968 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2970 * - ``Objective-C Image Info Version``
2971 - **[Required]** --- The version of the image info section. Currently
2974 * - ``Objective-C Image Info Section``
2975 - **[Required]** --- The section to place the metadata. Valid values are
2976 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2977 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2978 Objective-C ABI version 2.
2980 * - ``Objective-C Garbage Collection``
2981 - **[Required]** --- Specifies whether garbage collection is supported or
2982 not. Valid values are 0, for no garbage collection, and 2, for garbage
2983 collection supported.
2985 * - ``Objective-C GC Only``
2986 - **[Optional]** --- Specifies that only garbage collection is supported.
2987 If present, its value must be 6. This flag requires that the
2988 ``Objective-C Garbage Collection`` flag have the value 2.
2990 Some important flag interactions:
2992 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2993 merged with a module with ``Objective-C Garbage Collection`` set to
2994 2, then the resulting module has the
2995 ``Objective-C Garbage Collection`` flag set to 0.
2996 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2997 merged with a module with ``Objective-C GC Only`` set to 6.
2999 Automatic Linker Flags Module Flags Metadata
3000 --------------------------------------------
3002 Some targets support embedding flags to the linker inside individual object
3003 files. Typically this is used in conjunction with language extensions which
3004 allow source files to explicitly declare the libraries they depend on, and have
3005 these automatically be transmitted to the linker via object files.
3007 These flags are encoded in the IR using metadata in the module flags section,
3008 using the ``Linker Options`` key. The merge behavior for this flag is required
3009 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3010 node which should be a list of other metadata nodes, each of which should be a
3011 list of metadata strings defining linker options.
3013 For example, the following metadata section specifies two separate sets of
3014 linker options, presumably to link against ``libz`` and the ``Cocoa``
3017 !0 = metadata !{ i32 6, metadata !"Linker Options",
3019 metadata !{ metadata !"-lz" },
3020 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3021 !llvm.module.flags = !{ !0 }
3023 The metadata encoding as lists of lists of options, as opposed to a collapsed
3024 list of options, is chosen so that the IR encoding can use multiple option
3025 strings to specify e.g., a single library, while still having that specifier be
3026 preserved as an atomic element that can be recognized by a target specific
3027 assembly writer or object file emitter.
3029 Each individual option is required to be either a valid option for the target's
3030 linker, or an option that is reserved by the target specific assembly writer or
3031 object file emitter. No other aspect of these options is defined by the IR.
3033 .. _intrinsicglobalvariables:
3035 Intrinsic Global Variables
3036 ==========================
3038 LLVM has a number of "magic" global variables that contain data that
3039 affect code generation or other IR semantics. These are documented here.
3040 All globals of this sort should have a section specified as
3041 "``llvm.metadata``". This section and all globals that start with
3042 "``llvm.``" are reserved for use by LLVM.
3046 The '``llvm.used``' Global Variable
3047 -----------------------------------
3049 The ``@llvm.used`` global is an array which has
3050 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3051 pointers to named global variables, functions and aliases which may optionally
3052 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3055 .. code-block:: llvm
3060 @llvm.used = appending global [2 x i8*] [
3062 i8* bitcast (i32* @Y to i8*)
3063 ], section "llvm.metadata"
3065 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3066 and linker are required to treat the symbol as if there is a reference to the
3067 symbol that it cannot see (which is why they have to be named). For example, if
3068 a variable has internal linkage and no references other than that from the
3069 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3070 references from inline asms and other things the compiler cannot "see", and
3071 corresponds to "``attribute((used))``" in GNU C.
3073 On some targets, the code generator must emit a directive to the
3074 assembler or object file to prevent the assembler and linker from
3075 molesting the symbol.
3077 .. _gv_llvmcompilerused:
3079 The '``llvm.compiler.used``' Global Variable
3080 --------------------------------------------
3082 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3083 directive, except that it only prevents the compiler from touching the
3084 symbol. On targets that support it, this allows an intelligent linker to
3085 optimize references to the symbol without being impeded as it would be
3088 This is a rare construct that should only be used in rare circumstances,
3089 and should not be exposed to source languages.
3091 .. _gv_llvmglobalctors:
3093 The '``llvm.global_ctors``' Global Variable
3094 -------------------------------------------
3096 .. code-block:: llvm
3098 %0 = type { i32, void ()* }
3099 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3101 The ``@llvm.global_ctors`` array contains a list of constructor
3102 functions and associated priorities. The functions referenced by this
3103 array will be called in ascending order of priority (i.e. lowest first)
3104 when the module is loaded. The order of functions with the same priority
3107 .. _llvmglobaldtors:
3109 The '``llvm.global_dtors``' Global Variable
3110 -------------------------------------------
3112 .. code-block:: llvm
3114 %0 = type { i32, void ()* }
3115 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3117 The ``@llvm.global_dtors`` array contains a list of destructor functions
3118 and associated priorities. The functions referenced by this array will
3119 be called in descending order of priority (i.e. highest first) when the
3120 module is loaded. The order of functions with the same priority is not
3123 Instruction Reference
3124 =====================
3126 The LLVM instruction set consists of several different classifications
3127 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3128 instructions <binaryops>`, :ref:`bitwise binary
3129 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3130 :ref:`other instructions <otherops>`.
3134 Terminator Instructions
3135 -----------------------
3137 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3138 program ends with a "Terminator" instruction, which indicates which
3139 block should be executed after the current block is finished. These
3140 terminator instructions typically yield a '``void``' value: they produce
3141 control flow, not values (the one exception being the
3142 ':ref:`invoke <i_invoke>`' instruction).
3144 The terminator instructions are: ':ref:`ret <i_ret>`',
3145 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3146 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3147 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3151 '``ret``' Instruction
3152 ^^^^^^^^^^^^^^^^^^^^^
3159 ret <type> <value> ; Return a value from a non-void function
3160 ret void ; Return from void function
3165 The '``ret``' instruction is used to return control flow (and optionally
3166 a value) from a function back to the caller.
3168 There are two forms of the '``ret``' instruction: one that returns a
3169 value and then causes control flow, and one that just causes control
3175 The '``ret``' instruction optionally accepts a single argument, the
3176 return value. The type of the return value must be a ':ref:`first
3177 class <t_firstclass>`' type.
3179 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3180 return type and contains a '``ret``' instruction with no return value or
3181 a return value with a type that does not match its type, or if it has a
3182 void return type and contains a '``ret``' instruction with a return
3188 When the '``ret``' instruction is executed, control flow returns back to
3189 the calling function's context. If the caller is a
3190 ":ref:`call <i_call>`" instruction, execution continues at the
3191 instruction after the call. If the caller was an
3192 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3193 beginning of the "normal" destination block. If the instruction returns
3194 a value, that value shall set the call or invoke instruction's return
3200 .. code-block:: llvm
3202 ret i32 5 ; Return an integer value of 5
3203 ret void ; Return from a void function
3204 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3208 '``br``' Instruction
3209 ^^^^^^^^^^^^^^^^^^^^
3216 br i1 <cond>, label <iftrue>, label <iffalse>
3217 br label <dest> ; Unconditional branch
3222 The '``br``' instruction is used to cause control flow to transfer to a
3223 different basic block in the current function. There are two forms of
3224 this instruction, corresponding to a conditional branch and an
3225 unconditional branch.
3230 The conditional branch form of the '``br``' instruction takes a single
3231 '``i1``' value and two '``label``' values. The unconditional form of the
3232 '``br``' instruction takes a single '``label``' value as a target.
3237 Upon execution of a conditional '``br``' instruction, the '``i1``'
3238 argument is evaluated. If the value is ``true``, control flows to the
3239 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3240 to the '``iffalse``' ``label`` argument.
3245 .. code-block:: llvm
3248 %cond = icmp eq i32 %a, %b
3249 br i1 %cond, label %IfEqual, label %IfUnequal
3257 '``switch``' Instruction
3258 ^^^^^^^^^^^^^^^^^^^^^^^^
3265 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3270 The '``switch``' instruction is used to transfer control flow to one of
3271 several different places. It is a generalization of the '``br``'
3272 instruction, allowing a branch to occur to one of many possible
3278 The '``switch``' instruction uses three parameters: an integer
3279 comparison value '``value``', a default '``label``' destination, and an
3280 array of pairs of comparison value constants and '``label``'s. The table
3281 is not allowed to contain duplicate constant entries.
3286 The ``switch`` instruction specifies a table of values and destinations.
3287 When the '``switch``' instruction is executed, this table is searched
3288 for the given value. If the value is found, control flow is transferred
3289 to the corresponding destination; otherwise, control flow is transferred
3290 to the default destination.
3295 Depending on properties of the target machine and the particular
3296 ``switch`` instruction, this instruction may be code generated in
3297 different ways. For example, it could be generated as a series of
3298 chained conditional branches or with a lookup table.
3303 .. code-block:: llvm
3305 ; Emulate a conditional br instruction
3306 %Val = zext i1 %value to i32
3307 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3309 ; Emulate an unconditional br instruction
3310 switch i32 0, label %dest [ ]
3312 ; Implement a jump table:
3313 switch i32 %val, label %otherwise [ i32 0, label %onzero
3315 i32 2, label %ontwo ]
3319 '``indirectbr``' Instruction
3320 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3327 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3332 The '``indirectbr``' instruction implements an indirect branch to a
3333 label within the current function, whose address is specified by
3334 "``address``". Address must be derived from a
3335 :ref:`blockaddress <blockaddress>` constant.
3340 The '``address``' argument is the address of the label to jump to. The
3341 rest of the arguments indicate the full set of possible destinations
3342 that the address may point to. Blocks are allowed to occur multiple
3343 times in the destination list, though this isn't particularly useful.
3345 This destination list is required so that dataflow analysis has an
3346 accurate understanding of the CFG.
3351 Control transfers to the block specified in the address argument. All
3352 possible destination blocks must be listed in the label list, otherwise
3353 this instruction has undefined behavior. This implies that jumps to
3354 labels defined in other functions have undefined behavior as well.
3359 This is typically implemented with a jump through a register.
3364 .. code-block:: llvm
3366 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3370 '``invoke``' Instruction
3371 ^^^^^^^^^^^^^^^^^^^^^^^^
3378 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3379 to label <normal label> unwind label <exception label>
3384 The '``invoke``' instruction causes control to transfer to a specified
3385 function, with the possibility of control flow transfer to either the
3386 '``normal``' label or the '``exception``' label. If the callee function
3387 returns with the "``ret``" instruction, control flow will return to the
3388 "normal" label. If the callee (or any indirect callees) returns via the
3389 ":ref:`resume <i_resume>`" instruction or other exception handling
3390 mechanism, control is interrupted and continued at the dynamically
3391 nearest "exception" label.
3393 The '``exception``' label is a `landing
3394 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3395 '``exception``' label is required to have the
3396 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3397 information about the behavior of the program after unwinding happens,
3398 as its first non-PHI instruction. The restrictions on the
3399 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3400 instruction, so that the important information contained within the
3401 "``landingpad``" instruction can't be lost through normal code motion.
3406 This instruction requires several arguments:
3408 #. The optional "cconv" marker indicates which :ref:`calling
3409 convention <callingconv>` the call should use. If none is
3410 specified, the call defaults to using C calling conventions.
3411 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3412 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3414 #. '``ptr to function ty``': shall be the signature of the pointer to
3415 function value being invoked. In most cases, this is a direct
3416 function invocation, but indirect ``invoke``'s are just as possible,
3417 branching off an arbitrary pointer to function value.
3418 #. '``function ptr val``': An LLVM value containing a pointer to a
3419 function to be invoked.
3420 #. '``function args``': argument list whose types match the function
3421 signature argument types and parameter attributes. All arguments must
3422 be of :ref:`first class <t_firstclass>` type. If the function signature
3423 indicates the function accepts a variable number of arguments, the
3424 extra arguments can be specified.
3425 #. '``normal label``': the label reached when the called function
3426 executes a '``ret``' instruction.
3427 #. '``exception label``': the label reached when a callee returns via
3428 the :ref:`resume <i_resume>` instruction or other exception handling
3430 #. The optional :ref:`function attributes <fnattrs>` list. Only
3431 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3432 attributes are valid here.
3437 This instruction is designed to operate as a standard '``call``'
3438 instruction in most regards. The primary difference is that it
3439 establishes an association with a label, which is used by the runtime
3440 library to unwind the stack.
3442 This instruction is used in languages with destructors to ensure that
3443 proper cleanup is performed in the case of either a ``longjmp`` or a
3444 thrown exception. Additionally, this is important for implementation of
3445 '``catch``' clauses in high-level languages that support them.
3447 For the purposes of the SSA form, the definition of the value returned
3448 by the '``invoke``' instruction is deemed to occur on the edge from the
3449 current block to the "normal" label. If the callee unwinds then no
3450 return value is available.
3455 .. code-block:: llvm
3457 %retval = invoke i32 @Test(i32 15) to label %Continue
3458 unwind label %TestCleanup ; {i32}:retval set
3459 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3460 unwind label %TestCleanup ; {i32}:retval set
3464 '``resume``' Instruction
3465 ^^^^^^^^^^^^^^^^^^^^^^^^
3472 resume <type> <value>
3477 The '``resume``' instruction is a terminator instruction that has no
3483 The '``resume``' instruction requires one argument, which must have the
3484 same type as the result of any '``landingpad``' instruction in the same
3490 The '``resume``' instruction resumes propagation of an existing
3491 (in-flight) exception whose unwinding was interrupted with a
3492 :ref:`landingpad <i_landingpad>` instruction.
3497 .. code-block:: llvm
3499 resume { i8*, i32 } %exn
3503 '``unreachable``' Instruction
3504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3516 The '``unreachable``' instruction has no defined semantics. This
3517 instruction is used to inform the optimizer that a particular portion of
3518 the code is not reachable. This can be used to indicate that the code
3519 after a no-return function cannot be reached, and other facts.
3524 The '``unreachable``' instruction has no defined semantics.
3531 Binary operators are used to do most of the computation in a program.
3532 They require two operands of the same type, execute an operation on
3533 them, and produce a single value. The operands might represent multiple
3534 data, as is the case with the :ref:`vector <t_vector>` data type. The
3535 result value has the same type as its operands.
3537 There are several different binary operators:
3541 '``add``' Instruction
3542 ^^^^^^^^^^^^^^^^^^^^^
3549 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3550 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3551 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3552 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3557 The '``add``' instruction returns the sum of its two operands.
3562 The two arguments to the '``add``' instruction must be
3563 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3564 arguments must have identical types.
3569 The value produced is the integer sum of the two operands.
3571 If the sum has unsigned overflow, the result returned is the
3572 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3575 Because LLVM integers use a two's complement representation, this
3576 instruction is appropriate for both signed and unsigned integers.
3578 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3579 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3580 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3581 unsigned and/or signed overflow, respectively, occurs.
3586 .. code-block:: llvm
3588 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3592 '``fadd``' Instruction
3593 ^^^^^^^^^^^^^^^^^^^^^^
3600 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3605 The '``fadd``' instruction returns the sum of its two operands.
3610 The two arguments to the '``fadd``' instruction must be :ref:`floating
3611 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3612 Both arguments must have identical types.
3617 The value produced is the floating point sum of the two operands. This
3618 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3619 which are optimization hints to enable otherwise unsafe floating point
3625 .. code-block:: llvm
3627 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3629 '``sub``' Instruction
3630 ^^^^^^^^^^^^^^^^^^^^^
3637 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3638 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3639 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3640 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3645 The '``sub``' instruction returns the difference of its two operands.
3647 Note that the '``sub``' instruction is used to represent the '``neg``'
3648 instruction present in most other intermediate representations.
3653 The two arguments to the '``sub``' instruction must be
3654 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3655 arguments must have identical types.
3660 The value produced is the integer difference of the two operands.
3662 If the difference has unsigned overflow, the result returned is the
3663 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3666 Because LLVM integers use a two's complement representation, this
3667 instruction is appropriate for both signed and unsigned integers.
3669 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3670 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3671 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3672 unsigned and/or signed overflow, respectively, occurs.
3677 .. code-block:: llvm
3679 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3680 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3684 '``fsub``' Instruction
3685 ^^^^^^^^^^^^^^^^^^^^^^
3692 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3697 The '``fsub``' instruction returns the difference of its two operands.
3699 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3700 instruction present in most other intermediate representations.
3705 The two arguments to the '``fsub``' instruction must be :ref:`floating
3706 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3707 Both arguments must have identical types.
3712 The value produced is the floating point difference of the two operands.
3713 This instruction can also take any number of :ref:`fast-math
3714 flags <fastmath>`, which are optimization hints to enable otherwise
3715 unsafe floating point optimizations:
3720 .. code-block:: llvm
3722 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3723 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3725 '``mul``' Instruction
3726 ^^^^^^^^^^^^^^^^^^^^^
3733 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3734 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3735 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3736 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3741 The '``mul``' instruction returns the product of its two operands.
3746 The two arguments to the '``mul``' instruction must be
3747 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3748 arguments must have identical types.
3753 The value produced is the integer product of the two operands.
3755 If the result of the multiplication has unsigned overflow, the result
3756 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3757 bit width of the result.
3759 Because LLVM integers use a two's complement representation, and the
3760 result is the same width as the operands, this instruction returns the
3761 correct result for both signed and unsigned integers. If a full product
3762 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3763 sign-extended or zero-extended as appropriate to the width of the full
3766 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3767 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3768 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3769 unsigned and/or signed overflow, respectively, occurs.
3774 .. code-block:: llvm
3776 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3780 '``fmul``' Instruction
3781 ^^^^^^^^^^^^^^^^^^^^^^
3788 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3793 The '``fmul``' instruction returns the product of its two operands.
3798 The two arguments to the '``fmul``' instruction must be :ref:`floating
3799 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3800 Both arguments must have identical types.
3805 The value produced is the floating point product of the two operands.
3806 This instruction can also take any number of :ref:`fast-math
3807 flags <fastmath>`, which are optimization hints to enable otherwise
3808 unsafe floating point optimizations:
3813 .. code-block:: llvm
3815 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3817 '``udiv``' Instruction
3818 ^^^^^^^^^^^^^^^^^^^^^^
3825 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3826 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3831 The '``udiv``' instruction returns the quotient of its two operands.
3836 The two arguments to the '``udiv``' instruction must be
3837 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3838 arguments must have identical types.
3843 The value produced is the unsigned integer quotient of the two operands.
3845 Note that unsigned integer division and signed integer division are
3846 distinct operations; for signed integer division, use '``sdiv``'.
3848 Division by zero leads to undefined behavior.
3850 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3851 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3852 such, "((a udiv exact b) mul b) == a").
3857 .. code-block:: llvm
3859 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3861 '``sdiv``' Instruction
3862 ^^^^^^^^^^^^^^^^^^^^^^
3869 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3870 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3875 The '``sdiv``' instruction returns the quotient of its two operands.
3880 The two arguments to the '``sdiv``' instruction must be
3881 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3882 arguments must have identical types.
3887 The value produced is the signed integer quotient of the two operands
3888 rounded towards zero.
3890 Note that signed integer division and unsigned integer division are
3891 distinct operations; for unsigned integer division, use '``udiv``'.
3893 Division by zero leads to undefined behavior. Overflow also leads to
3894 undefined behavior; this is a rare case, but can occur, for example, by
3895 doing a 32-bit division of -2147483648 by -1.
3897 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3898 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3903 .. code-block:: llvm
3905 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3909 '``fdiv``' Instruction
3910 ^^^^^^^^^^^^^^^^^^^^^^
3917 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3922 The '``fdiv``' instruction returns the quotient of its two operands.
3927 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3928 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3929 Both arguments must have identical types.
3934 The value produced is the floating point quotient of the two operands.
3935 This instruction can also take any number of :ref:`fast-math
3936 flags <fastmath>`, which are optimization hints to enable otherwise
3937 unsafe floating point optimizations:
3942 .. code-block:: llvm
3944 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3946 '``urem``' Instruction
3947 ^^^^^^^^^^^^^^^^^^^^^^
3954 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3959 The '``urem``' instruction returns the remainder from the unsigned
3960 division of its two arguments.
3965 The two arguments to the '``urem``' instruction must be
3966 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3967 arguments must have identical types.
3972 This instruction returns the unsigned integer *remainder* of a division.
3973 This instruction always performs an unsigned division to get the
3976 Note that unsigned integer remainder and signed integer remainder are
3977 distinct operations; for signed integer remainder, use '``srem``'.
3979 Taking the remainder of a division by zero leads to undefined behavior.
3984 .. code-block:: llvm
3986 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3988 '``srem``' Instruction
3989 ^^^^^^^^^^^^^^^^^^^^^^
3996 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4001 The '``srem``' instruction returns the remainder from the signed
4002 division of its two operands. This instruction can also take
4003 :ref:`vector <t_vector>` versions of the values in which case the elements
4009 The two arguments to the '``srem``' instruction must be
4010 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4011 arguments must have identical types.
4016 This instruction returns the *remainder* of a division (where the result
4017 is either zero or has the same sign as the dividend, ``op1``), not the
4018 *modulo* operator (where the result is either zero or has the same sign
4019 as the divisor, ``op2``) of a value. For more information about the
4020 difference, see `The Math
4021 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4022 table of how this is implemented in various languages, please see
4024 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4026 Note that signed integer remainder and unsigned integer remainder are
4027 distinct operations; for unsigned integer remainder, use '``urem``'.
4029 Taking the remainder of a division by zero leads to undefined behavior.
4030 Overflow also leads to undefined behavior; this is a rare case, but can
4031 occur, for example, by taking the remainder of a 32-bit division of
4032 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4033 rule lets srem be implemented using instructions that return both the
4034 result of the division and the remainder.)
4039 .. code-block:: llvm
4041 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4045 '``frem``' Instruction
4046 ^^^^^^^^^^^^^^^^^^^^^^
4053 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4058 The '``frem``' instruction returns the remainder from the division of
4064 The two arguments to the '``frem``' instruction must be :ref:`floating
4065 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4066 Both arguments must have identical types.
4071 This instruction returns the *remainder* of a division. The remainder
4072 has the same sign as the dividend. This instruction can also take any
4073 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4074 to enable otherwise unsafe floating point optimizations:
4079 .. code-block:: llvm
4081 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4085 Bitwise Binary Operations
4086 -------------------------
4088 Bitwise binary operators are used to do various forms of bit-twiddling
4089 in a program. They are generally very efficient instructions and can
4090 commonly be strength reduced from other instructions. They require two
4091 operands of the same type, execute an operation on them, and produce a
4092 single value. The resulting value is the same type as its operands.
4094 '``shl``' Instruction
4095 ^^^^^^^^^^^^^^^^^^^^^
4102 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4103 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4104 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4105 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4110 The '``shl``' instruction returns the first operand shifted to the left
4111 a specified number of bits.
4116 Both arguments to the '``shl``' instruction must be the same
4117 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4118 '``op2``' is treated as an unsigned value.
4123 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4124 where ``n`` is the width of the result. If ``op2`` is (statically or
4125 dynamically) negative or equal to or larger than the number of bits in
4126 ``op1``, the result is undefined. If the arguments are vectors, each
4127 vector element of ``op1`` is shifted by the corresponding shift amount
4130 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4131 value <poisonvalues>` if it shifts out any non-zero bits. If the
4132 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4133 value <poisonvalues>` if it shifts out any bits that disagree with the
4134 resultant sign bit. As such, NUW/NSW have the same semantics as they
4135 would if the shift were expressed as a mul instruction with the same
4136 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4141 .. code-block:: llvm
4143 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4144 <result> = shl i32 4, 2 ; yields {i32}: 16
4145 <result> = shl i32 1, 10 ; yields {i32}: 1024
4146 <result> = shl i32 1, 32 ; undefined
4147 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4149 '``lshr``' Instruction
4150 ^^^^^^^^^^^^^^^^^^^^^^
4157 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4158 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4163 The '``lshr``' instruction (logical shift right) returns the first
4164 operand shifted to the right a specified number of bits with zero fill.
4169 Both arguments to the '``lshr``' instruction must be the same
4170 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4171 '``op2``' is treated as an unsigned value.
4176 This instruction always performs a logical shift right operation. The
4177 most significant bits of the result will be filled with zero bits after
4178 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4179 than the number of bits in ``op1``, the result is undefined. If the
4180 arguments are vectors, each vector element of ``op1`` is shifted by the
4181 corresponding shift amount in ``op2``.
4183 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4184 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4190 .. code-block:: llvm
4192 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4193 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4194 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4195 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4196 <result> = lshr i32 1, 32 ; undefined
4197 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4199 '``ashr``' Instruction
4200 ^^^^^^^^^^^^^^^^^^^^^^
4207 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4208 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4213 The '``ashr``' instruction (arithmetic shift right) returns the first
4214 operand shifted to the right a specified number of bits with sign
4220 Both arguments to the '``ashr``' instruction must be the same
4221 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4222 '``op2``' is treated as an unsigned value.
4227 This instruction always performs an arithmetic shift right operation,
4228 The most significant bits of the result will be filled with the sign bit
4229 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4230 than the number of bits in ``op1``, the result is undefined. If the
4231 arguments are vectors, each vector element of ``op1`` is shifted by the
4232 corresponding shift amount in ``op2``.
4234 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4235 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4241 .. code-block:: llvm
4243 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4244 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4245 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4246 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4247 <result> = ashr i32 1, 32 ; undefined
4248 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4250 '``and``' Instruction
4251 ^^^^^^^^^^^^^^^^^^^^^
4258 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4263 The '``and``' instruction returns the bitwise logical and of its two
4269 The two arguments to the '``and``' instruction must be
4270 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4271 arguments must have identical types.
4276 The truth table used for the '``and``' instruction is:
4293 .. code-block:: llvm
4295 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4296 <result> = and i32 15, 40 ; yields {i32}:result = 8
4297 <result> = and i32 4, 8 ; yields {i32}:result = 0
4299 '``or``' Instruction
4300 ^^^^^^^^^^^^^^^^^^^^
4307 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4312 The '``or``' instruction returns the bitwise logical inclusive or of its
4318 The two arguments to the '``or``' instruction must be
4319 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4320 arguments must have identical types.
4325 The truth table used for the '``or``' instruction is:
4344 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4345 <result> = or i32 15, 40 ; yields {i32}:result = 47
4346 <result> = or i32 4, 8 ; yields {i32}:result = 12
4348 '``xor``' Instruction
4349 ^^^^^^^^^^^^^^^^^^^^^
4356 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4361 The '``xor``' instruction returns the bitwise logical exclusive or of
4362 its two operands. The ``xor`` is used to implement the "one's
4363 complement" operation, which is the "~" operator in C.
4368 The two arguments to the '``xor``' instruction must be
4369 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4370 arguments must have identical types.
4375 The truth table used for the '``xor``' instruction is:
4392 .. code-block:: llvm
4394 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4395 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4396 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4397 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4402 LLVM supports several instructions to represent vector operations in a
4403 target-independent manner. These instructions cover the element-access
4404 and vector-specific operations needed to process vectors effectively.
4405 While LLVM does directly support these vector operations, many
4406 sophisticated algorithms will want to use target-specific intrinsics to
4407 take full advantage of a specific target.
4409 .. _i_extractelement:
4411 '``extractelement``' Instruction
4412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4419 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4424 The '``extractelement``' instruction extracts a single scalar element
4425 from a vector at a specified index.
4430 The first operand of an '``extractelement``' instruction is a value of
4431 :ref:`vector <t_vector>` type. The second operand is an index indicating
4432 the position from which to extract the element. The index may be a
4438 The result is a scalar of the same type as the element type of ``val``.
4439 Its value is the value at position ``idx`` of ``val``. If ``idx``
4440 exceeds the length of ``val``, the results are undefined.
4445 .. code-block:: llvm
4447 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4449 .. _i_insertelement:
4451 '``insertelement``' Instruction
4452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4459 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4464 The '``insertelement``' instruction inserts a scalar element into a
4465 vector at a specified index.
4470 The first operand of an '``insertelement``' instruction is a value of
4471 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4472 type must equal the element type of the first operand. The third operand
4473 is an index indicating the position at which to insert the value. The
4474 index may be a variable.
4479 The result is a vector of the same type as ``val``. Its element values
4480 are those of ``val`` except at position ``idx``, where it gets the value
4481 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4487 .. code-block:: llvm
4489 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4491 .. _i_shufflevector:
4493 '``shufflevector``' Instruction
4494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4501 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4506 The '``shufflevector``' instruction constructs a permutation of elements
4507 from two input vectors, returning a vector with the same element type as
4508 the input and length that is the same as the shuffle mask.
4513 The first two operands of a '``shufflevector``' instruction are vectors
4514 with the same type. The third argument is a shuffle mask whose element
4515 type is always 'i32'. The result of the instruction is a vector whose
4516 length is the same as the shuffle mask and whose element type is the
4517 same as the element type of the first two operands.
4519 The shuffle mask operand is required to be a constant vector with either
4520 constant integer or undef values.
4525 The elements of the two input vectors are numbered from left to right
4526 across both of the vectors. The shuffle mask operand specifies, for each
4527 element of the result vector, which element of the two input vectors the
4528 result element gets. The element selector may be undef (meaning "don't
4529 care") and the second operand may be undef if performing a shuffle from
4535 .. code-block:: llvm
4537 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4538 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4539 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4540 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4541 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4542 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4543 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4544 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4546 Aggregate Operations
4547 --------------------
4549 LLVM supports several instructions for working with
4550 :ref:`aggregate <t_aggregate>` values.
4554 '``extractvalue``' Instruction
4555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4562 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4567 The '``extractvalue``' instruction extracts the value of a member field
4568 from an :ref:`aggregate <t_aggregate>` value.
4573 The first operand of an '``extractvalue``' instruction is a value of
4574 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4575 constant indices to specify which value to extract in a similar manner
4576 as indices in a '``getelementptr``' instruction.
4578 The major differences to ``getelementptr`` indexing are:
4580 - Since the value being indexed is not a pointer, the first index is
4581 omitted and assumed to be zero.
4582 - At least one index must be specified.
4583 - Not only struct indices but also array indices must be in bounds.
4588 The result is the value at the position in the aggregate specified by
4594 .. code-block:: llvm
4596 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4600 '``insertvalue``' Instruction
4601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4608 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4613 The '``insertvalue``' instruction inserts a value into a member field in
4614 an :ref:`aggregate <t_aggregate>` value.
4619 The first operand of an '``insertvalue``' instruction is a value of
4620 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4621 a first-class value to insert. The following operands are constant
4622 indices indicating the position at which to insert the value in a
4623 similar manner as indices in a '``extractvalue``' instruction. The value
4624 to insert must have the same type as the value identified by the
4630 The result is an aggregate of the same type as ``val``. Its value is
4631 that of ``val`` except that the value at the position specified by the
4632 indices is that of ``elt``.
4637 .. code-block:: llvm
4639 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4640 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4641 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4645 Memory Access and Addressing Operations
4646 ---------------------------------------
4648 A key design point of an SSA-based representation is how it represents
4649 memory. In LLVM, no memory locations are in SSA form, which makes things
4650 very simple. This section describes how to read, write, and allocate
4655 '``alloca``' Instruction
4656 ^^^^^^^^^^^^^^^^^^^^^^^^
4663 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4668 The '``alloca``' instruction allocates memory on the stack frame of the
4669 currently executing function, to be automatically released when this
4670 function returns to its caller. The object is always allocated in the
4671 generic address space (address space zero).
4676 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4677 bytes of memory on the runtime stack, returning a pointer of the
4678 appropriate type to the program. If "NumElements" is specified, it is
4679 the number of elements allocated, otherwise "NumElements" is defaulted
4680 to be one. If a constant alignment is specified, the value result of the
4681 allocation is guaranteed to be aligned to at least that boundary. If not
4682 specified, or if zero, the target can choose to align the allocation on
4683 any convenient boundary compatible with the type.
4685 '``type``' may be any sized type.
4690 Memory is allocated; a pointer is returned. The operation is undefined
4691 if there is insufficient stack space for the allocation. '``alloca``'d
4692 memory is automatically released when the function returns. The
4693 '``alloca``' instruction is commonly used to represent automatic
4694 variables that must have an address available. When the function returns
4695 (either with the ``ret`` or ``resume`` instructions), the memory is
4696 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4697 The order in which memory is allocated (ie., which way the stack grows)
4703 .. code-block:: llvm
4705 %ptr = alloca i32 ; yields {i32*}:ptr
4706 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4707 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4708 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4712 '``load``' Instruction
4713 ^^^^^^^^^^^^^^^^^^^^^^
4720 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4721 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4722 !<index> = !{ i32 1 }
4727 The '``load``' instruction is used to read from memory.
4732 The argument to the ``load`` instruction specifies the memory address
4733 from which to load. The pointer must point to a :ref:`first
4734 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4735 then the optimizer is not allowed to modify the number or order of
4736 execution of this ``load`` with other :ref:`volatile
4737 operations <volatile>`.
4739 If the ``load`` is marked as ``atomic``, it takes an extra
4740 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4741 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4742 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4743 when they may see multiple atomic stores. The type of the pointee must
4744 be an integer type whose bit width is a power of two greater than or
4745 equal to eight and less than or equal to a target-specific size limit.
4746 ``align`` must be explicitly specified on atomic loads, and the load has
4747 undefined behavior if the alignment is not set to a value which is at
4748 least the size in bytes of the pointee. ``!nontemporal`` does not have
4749 any defined semantics for atomic loads.
4751 The optional constant ``align`` argument specifies the alignment of the
4752 operation (that is, the alignment of the memory address). A value of 0
4753 or an omitted ``align`` argument means that the operation has the ABI
4754 alignment for the target. It is the responsibility of the code emitter
4755 to ensure that the alignment information is correct. Overestimating the
4756 alignment results in undefined behavior. Underestimating the alignment
4757 may produce less efficient code. An alignment of 1 is always safe.
4759 The optional ``!nontemporal`` metadata must reference a single
4760 metadata name ``<index>`` corresponding to a metadata node with one
4761 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4762 metadata on the instruction tells the optimizer and code generator
4763 that this load is not expected to be reused in the cache. The code
4764 generator may select special instructions to save cache bandwidth, such
4765 as the ``MOVNT`` instruction on x86.
4767 The optional ``!invariant.load`` metadata must reference a single
4768 metadata name ``<index>`` corresponding to a metadata node with no
4769 entries. The existence of the ``!invariant.load`` metadata on the
4770 instruction tells the optimizer and code generator that this load
4771 address points to memory which does not change value during program
4772 execution. The optimizer may then move this load around, for example, by
4773 hoisting it out of loops using loop invariant code motion.
4778 The location of memory pointed to is loaded. If the value being loaded
4779 is of scalar type then the number of bytes read does not exceed the
4780 minimum number of bytes needed to hold all bits of the type. For
4781 example, loading an ``i24`` reads at most three bytes. When loading a
4782 value of a type like ``i20`` with a size that is not an integral number
4783 of bytes, the result is undefined if the value was not originally
4784 written using a store of the same type.
4789 .. code-block:: llvm
4791 %ptr = alloca i32 ; yields {i32*}:ptr
4792 store i32 3, i32* %ptr ; yields {void}
4793 %val = load i32* %ptr ; yields {i32}:val = i32 3
4797 '``store``' Instruction
4798 ^^^^^^^^^^^^^^^^^^^^^^^
4805 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4806 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4811 The '``store``' instruction is used to write to memory.
4816 There are two arguments to the ``store`` instruction: a value to store
4817 and an address at which to store it. The type of the ``<pointer>``
4818 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4819 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4820 then the optimizer is not allowed to modify the number or order of
4821 execution of this ``store`` with other :ref:`volatile
4822 operations <volatile>`.
4824 If the ``store`` is marked as ``atomic``, it takes an extra
4825 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4826 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4827 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4828 when they may see multiple atomic stores. The type of the pointee must
4829 be an integer type whose bit width is a power of two greater than or
4830 equal to eight and less than or equal to a target-specific size limit.
4831 ``align`` must be explicitly specified on atomic stores, and the store
4832 has undefined behavior if the alignment is not set to a value which is
4833 at least the size in bytes of the pointee. ``!nontemporal`` does not
4834 have any defined semantics for atomic stores.
4836 The optional constant ``align`` argument specifies the alignment of the
4837 operation (that is, the alignment of the memory address). A value of 0
4838 or an omitted ``align`` argument means that the operation has the ABI
4839 alignment for the target. It is the responsibility of the code emitter
4840 to ensure that the alignment information is correct. Overestimating the
4841 alignment results in undefined behavior. Underestimating the
4842 alignment may produce less efficient code. An alignment of 1 is always
4845 The optional ``!nontemporal`` metadata must reference a single metadata
4846 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4847 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4848 tells the optimizer and code generator that this load is not expected to
4849 be reused in the cache. The code generator may select special
4850 instructions to save cache bandwidth, such as the MOVNT instruction on
4856 The contents of memory are updated to contain ``<value>`` at the
4857 location specified by the ``<pointer>`` operand. If ``<value>`` is
4858 of scalar type then the number of bytes written does not exceed the
4859 minimum number of bytes needed to hold all bits of the type. For
4860 example, storing an ``i24`` writes at most three bytes. When writing a
4861 value of a type like ``i20`` with a size that is not an integral number
4862 of bytes, it is unspecified what happens to the extra bits that do not
4863 belong to the type, but they will typically be overwritten.
4868 .. code-block:: llvm
4870 %ptr = alloca i32 ; yields {i32*}:ptr
4871 store i32 3, i32* %ptr ; yields {void}
4872 %val = load i32* %ptr ; yields {i32}:val = i32 3
4876 '``fence``' Instruction
4877 ^^^^^^^^^^^^^^^^^^^^^^^
4884 fence [singlethread] <ordering> ; yields {void}
4889 The '``fence``' instruction is used to introduce happens-before edges
4895 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4896 defines what *synchronizes-with* edges they add. They can only be given
4897 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4902 A fence A which has (at least) ``release`` ordering semantics
4903 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4904 semantics if and only if there exist atomic operations X and Y, both
4905 operating on some atomic object M, such that A is sequenced before X, X
4906 modifies M (either directly or through some side effect of a sequence
4907 headed by X), Y is sequenced before B, and Y observes M. This provides a
4908 *happens-before* dependency between A and B. Rather than an explicit
4909 ``fence``, one (but not both) of the atomic operations X or Y might
4910 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4911 still *synchronize-with* the explicit ``fence`` and establish the
4912 *happens-before* edge.
4914 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4915 ``acquire`` and ``release`` semantics specified above, participates in
4916 the global program order of other ``seq_cst`` operations and/or fences.
4918 The optional ":ref:`singlethread <singlethread>`" argument specifies
4919 that the fence only synchronizes with other fences in the same thread.
4920 (This is useful for interacting with signal handlers.)
4925 .. code-block:: llvm
4927 fence acquire ; yields {void}
4928 fence singlethread seq_cst ; yields {void}
4932 '``cmpxchg``' Instruction
4933 ^^^^^^^^^^^^^^^^^^^^^^^^^
4940 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4945 The '``cmpxchg``' instruction is used to atomically modify memory. It
4946 loads a value in memory and compares it to a given value. If they are
4947 equal, it stores a new value into the memory.
4952 There are three arguments to the '``cmpxchg``' instruction: an address
4953 to operate on, a value to compare to the value currently be at that
4954 address, and a new value to place at that address if the compared values
4955 are equal. The type of '<cmp>' must be an integer type whose bit width
4956 is a power of two greater than or equal to eight and less than or equal
4957 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4958 type, and the type of '<pointer>' must be a pointer to that type. If the
4959 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4960 to modify the number or order of execution of this ``cmpxchg`` with
4961 other :ref:`volatile operations <volatile>`.
4963 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4964 synchronizes with other atomic operations.
4966 The optional "``singlethread``" argument declares that the ``cmpxchg``
4967 is only atomic with respect to code (usually signal handlers) running in
4968 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4969 respect to all other code in the system.
4971 The pointer passed into cmpxchg must have alignment greater than or
4972 equal to the size in memory of the operand.
4977 The contents of memory at the location specified by the '``<pointer>``'
4978 operand is read and compared to '``<cmp>``'; if the read value is the
4979 equal, '``<new>``' is written. The original value at the location is
4982 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4983 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4984 atomic load with an ordering parameter determined by dropping any
4985 ``release`` part of the ``cmpxchg``'s ordering.
4990 .. code-block:: llvm
4993 %orig = atomic load i32* %ptr unordered ; yields {i32}
4997 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4998 %squared = mul i32 %cmp, %cmp
4999 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
5000 %success = icmp eq i32 %cmp, %old
5001 br i1 %success, label %done, label %loop
5008 '``atomicrmw``' Instruction
5009 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5016 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5021 The '``atomicrmw``' instruction is used to atomically modify memory.
5026 There are three arguments to the '``atomicrmw``' instruction: an
5027 operation to apply, an address whose value to modify, an argument to the
5028 operation. The operation must be one of the following keywords:
5042 The type of '<value>' must be an integer type whose bit width is a power
5043 of two greater than or equal to eight and less than or equal to a
5044 target-specific size limit. The type of the '``<pointer>``' operand must
5045 be a pointer to that type. If the ``atomicrmw`` is marked as
5046 ``volatile``, then the optimizer is not allowed to modify the number or
5047 order of execution of this ``atomicrmw`` with other :ref:`volatile
5048 operations <volatile>`.
5053 The contents of memory at the location specified by the '``<pointer>``'
5054 operand are atomically read, modified, and written back. The original
5055 value at the location is returned. The modification is specified by the
5058 - xchg: ``*ptr = val``
5059 - add: ``*ptr = *ptr + val``
5060 - sub: ``*ptr = *ptr - val``
5061 - and: ``*ptr = *ptr & val``
5062 - nand: ``*ptr = ~(*ptr & val)``
5063 - or: ``*ptr = *ptr | val``
5064 - xor: ``*ptr = *ptr ^ val``
5065 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5066 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5067 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5069 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5075 .. code-block:: llvm
5077 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5079 .. _i_getelementptr:
5081 '``getelementptr``' Instruction
5082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5089 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5090 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5091 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5096 The '``getelementptr``' instruction is used to get the address of a
5097 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5098 address calculation only and does not access memory.
5103 The first argument is always a pointer or a vector of pointers, and
5104 forms the basis of the calculation. The remaining arguments are indices
5105 that indicate which of the elements of the aggregate object are indexed.
5106 The interpretation of each index is dependent on the type being indexed
5107 into. The first index always indexes the pointer value given as the
5108 first argument, the second index indexes a value of the type pointed to
5109 (not necessarily the value directly pointed to, since the first index
5110 can be non-zero), etc. The first type indexed into must be a pointer
5111 value, subsequent types can be arrays, vectors, and structs. Note that
5112 subsequent types being indexed into can never be pointers, since that
5113 would require loading the pointer before continuing calculation.
5115 The type of each index argument depends on the type it is indexing into.
5116 When indexing into a (optionally packed) structure, only ``i32`` integer
5117 **constants** are allowed (when using a vector of indices they must all
5118 be the **same** ``i32`` integer constant). When indexing into an array,
5119 pointer or vector, integers of any width are allowed, and they are not
5120 required to be constant. These integers are treated as signed values
5123 For example, let's consider a C code fragment and how it gets compiled
5139 int *foo(struct ST *s) {
5140 return &s[1].Z.B[5][13];
5143 The LLVM code generated by Clang is:
5145 .. code-block:: llvm
5147 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5148 %struct.ST = type { i32, double, %struct.RT }
5150 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5152 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5159 In the example above, the first index is indexing into the
5160 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5161 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5162 indexes into the third element of the structure, yielding a
5163 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5164 structure. The third index indexes into the second element of the
5165 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5166 dimensions of the array are subscripted into, yielding an '``i32``'
5167 type. The '``getelementptr``' instruction returns a pointer to this
5168 element, thus computing a value of '``i32*``' type.
5170 Note that it is perfectly legal to index partially through a structure,
5171 returning a pointer to an inner element. Because of this, the LLVM code
5172 for the given testcase is equivalent to:
5174 .. code-block:: llvm
5176 define i32* @foo(%struct.ST* %s) {
5177 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5178 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5179 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5180 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5181 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5185 If the ``inbounds`` keyword is present, the result value of the
5186 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5187 pointer is not an *in bounds* address of an allocated object, or if any
5188 of the addresses that would be formed by successive addition of the
5189 offsets implied by the indices to the base address with infinitely
5190 precise signed arithmetic are not an *in bounds* address of that
5191 allocated object. The *in bounds* addresses for an allocated object are
5192 all the addresses that point into the object, plus the address one byte
5193 past the end. In cases where the base is a vector of pointers the
5194 ``inbounds`` keyword applies to each of the computations element-wise.
5196 If the ``inbounds`` keyword is not present, the offsets are added to the
5197 base address with silently-wrapping two's complement arithmetic. If the
5198 offsets have a different width from the pointer, they are sign-extended
5199 or truncated to the width of the pointer. The result value of the
5200 ``getelementptr`` may be outside the object pointed to by the base
5201 pointer. The result value may not necessarily be used to access memory
5202 though, even if it happens to point into allocated storage. See the
5203 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5206 The getelementptr instruction is often confusing. For some more insight
5207 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5212 .. code-block:: llvm
5214 ; yields [12 x i8]*:aptr
5215 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5217 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5219 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5221 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5223 In cases where the pointer argument is a vector of pointers, each index
5224 must be a vector with the same number of elements. For example:
5226 .. code-block:: llvm
5228 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5230 Conversion Operations
5231 ---------------------
5233 The instructions in this category are the conversion instructions
5234 (casting) which all take a single operand and a type. They perform
5235 various bit conversions on the operand.
5237 '``trunc .. to``' Instruction
5238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5245 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5250 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5255 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5256 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5257 of the same number of integers. The bit size of the ``value`` must be
5258 larger than the bit size of the destination type, ``ty2``. Equal sized
5259 types are not allowed.
5264 The '``trunc``' instruction truncates the high order bits in ``value``
5265 and converts the remaining bits to ``ty2``. Since the source size must
5266 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5267 It will always truncate bits.
5272 .. code-block:: llvm
5274 %X = trunc i32 257 to i8 ; yields i8:1
5275 %Y = trunc i32 123 to i1 ; yields i1:true
5276 %Z = trunc i32 122 to i1 ; yields i1:false
5277 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5279 '``zext .. to``' Instruction
5280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5287 <result> = zext <ty> <value> to <ty2> ; yields ty2
5292 The '``zext``' instruction zero extends its operand to type ``ty2``.
5297 The '``zext``' instruction takes a value to cast, and a type to cast it
5298 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5299 the same number of integers. The bit size of the ``value`` must be
5300 smaller than the bit size of the destination type, ``ty2``.
5305 The ``zext`` fills the high order bits of the ``value`` with zero bits
5306 until it reaches the size of the destination type, ``ty2``.
5308 When zero extending from i1, the result will always be either 0 or 1.
5313 .. code-block:: llvm
5315 %X = zext i32 257 to i64 ; yields i64:257
5316 %Y = zext i1 true to i32 ; yields i32:1
5317 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5319 '``sext .. to``' Instruction
5320 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5327 <result> = sext <ty> <value> to <ty2> ; yields ty2
5332 The '``sext``' sign extends ``value`` to the type ``ty2``.
5337 The '``sext``' instruction takes a value to cast, and a type to cast it
5338 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5339 the same number of integers. The bit size of the ``value`` must be
5340 smaller than the bit size of the destination type, ``ty2``.
5345 The '``sext``' instruction performs a sign extension by copying the sign
5346 bit (highest order bit) of the ``value`` until it reaches the bit size
5347 of the type ``ty2``.
5349 When sign extending from i1, the extension always results in -1 or 0.
5354 .. code-block:: llvm
5356 %X = sext i8 -1 to i16 ; yields i16 :65535
5357 %Y = sext i1 true to i32 ; yields i32:-1
5358 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5360 '``fptrunc .. to``' Instruction
5361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5368 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5373 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5378 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5379 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5380 The size of ``value`` must be larger than the size of ``ty2``. This
5381 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5386 The '``fptrunc``' instruction truncates a ``value`` from a larger
5387 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5388 point <t_floating>` type. If the value cannot fit within the
5389 destination type, ``ty2``, then the results are undefined.
5394 .. code-block:: llvm
5396 %X = fptrunc double 123.0 to float ; yields float:123.0
5397 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5399 '``fpext .. to``' Instruction
5400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5407 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5412 The '``fpext``' extends a floating point ``value`` to a larger floating
5418 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5419 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5420 to. The source type must be smaller than the destination type.
5425 The '``fpext``' instruction extends the ``value`` from a smaller
5426 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5427 point <t_floating>` type. The ``fpext`` cannot be used to make a
5428 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5429 *no-op cast* for a floating point cast.
5434 .. code-block:: llvm
5436 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5437 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5439 '``fptoui .. to``' Instruction
5440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5447 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5452 The '``fptoui``' converts a floating point ``value`` to its unsigned
5453 integer equivalent of type ``ty2``.
5458 The '``fptoui``' instruction takes a value to cast, which must be a
5459 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5460 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5461 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5462 type with the same number of elements as ``ty``
5467 The '``fptoui``' instruction converts its :ref:`floating
5468 point <t_floating>` operand into the nearest (rounding towards zero)
5469 unsigned integer value. If the value cannot fit in ``ty2``, the results
5475 .. code-block:: llvm
5477 %X = fptoui double 123.0 to i32 ; yields i32:123
5478 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5479 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5481 '``fptosi .. to``' Instruction
5482 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5489 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5494 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5495 ``value`` to type ``ty2``.
5500 The '``fptosi``' instruction takes a value to cast, which must be a
5501 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5502 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5503 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5504 type with the same number of elements as ``ty``
5509 The '``fptosi``' instruction converts its :ref:`floating
5510 point <t_floating>` operand into the nearest (rounding towards zero)
5511 signed integer value. If the value cannot fit in ``ty2``, the results
5517 .. code-block:: llvm
5519 %X = fptosi double -123.0 to i32 ; yields i32:-123
5520 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5521 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5523 '``uitofp .. to``' Instruction
5524 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5531 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5536 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5537 and converts that value to the ``ty2`` type.
5542 The '``uitofp``' instruction takes a value to cast, which must be a
5543 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5544 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5545 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5546 type with the same number of elements as ``ty``
5551 The '``uitofp``' instruction interprets its operand as an unsigned
5552 integer quantity and converts it to the corresponding floating point
5553 value. If the value cannot fit in the floating point value, the results
5559 .. code-block:: llvm
5561 %X = uitofp i32 257 to float ; yields float:257.0
5562 %Y = uitofp i8 -1 to double ; yields double:255.0
5564 '``sitofp .. to``' Instruction
5565 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5572 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5577 The '``sitofp``' instruction regards ``value`` as a signed integer and
5578 converts that value to the ``ty2`` type.
5583 The '``sitofp``' instruction takes a value to cast, which must be a
5584 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5585 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5586 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5587 type with the same number of elements as ``ty``
5592 The '``sitofp``' instruction interprets its operand as a signed integer
5593 quantity and converts it to the corresponding floating point value. If
5594 the value cannot fit in the floating point value, the results are
5600 .. code-block:: llvm
5602 %X = sitofp i32 257 to float ; yields float:257.0
5603 %Y = sitofp i8 -1 to double ; yields double:-1.0
5607 '``ptrtoint .. to``' Instruction
5608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5615 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5620 The '``ptrtoint``' instruction converts the pointer or a vector of
5621 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5626 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5627 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5628 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5629 a vector of integers type.
5634 The '``ptrtoint``' instruction converts ``value`` to integer type
5635 ``ty2`` by interpreting the pointer value as an integer and either
5636 truncating or zero extending that value to the size of the integer type.
5637 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5638 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5639 the same size, then nothing is done (*no-op cast*) other than a type
5645 .. code-block:: llvm
5647 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5648 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5649 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5653 '``inttoptr .. to``' Instruction
5654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5661 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5666 The '``inttoptr``' instruction converts an integer ``value`` to a
5667 pointer type, ``ty2``.
5672 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5673 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5679 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5680 applying either a zero extension or a truncation depending on the size
5681 of the integer ``value``. If ``value`` is larger than the size of a
5682 pointer then a truncation is done. If ``value`` is smaller than the size
5683 of a pointer then a zero extension is done. If they are the same size,
5684 nothing is done (*no-op cast*).
5689 .. code-block:: llvm
5691 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5692 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5693 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5694 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5698 '``bitcast .. to``' Instruction
5699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5706 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5711 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5717 The '``bitcast``' instruction takes a value to cast, which must be a
5718 non-aggregate first class value, and a type to cast it to, which must
5719 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5720 bit sizes of ``value`` and the destination type, ``ty2``, must be
5721 identical. If the source type is a pointer, the destination type must
5722 also be a pointer of the same size. This instruction supports bitwise
5723 conversion of vectors to integers and to vectors of other types (as
5724 long as they have the same size).
5729 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5730 is always a *no-op cast* because no bits change with this
5731 conversion. The conversion is done as if the ``value`` had been stored
5732 to memory and read back as type ``ty2``. Pointer (or vector of
5733 pointers) types may only be converted to other pointer (or vector of
5734 pointers) types with the same address space through this instruction.
5735 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5736 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5741 .. code-block:: llvm
5743 %X = bitcast i8 255 to i8 ; yields i8 :-1
5744 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5745 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5746 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5748 .. _i_addrspacecast:
5750 '``addrspacecast .. to``' Instruction
5751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5758 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5763 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5764 address space ``n`` to type ``pty2`` in address space ``m``.
5769 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5770 to cast and a pointer type to cast it to, which must have a different
5776 The '``addrspacecast``' instruction converts the pointer value
5777 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5778 value modification, depending on the target and the address space
5779 pair. Pointer conversions within the same address space must be
5780 performed with the ``bitcast`` instruction. Note that if the address space
5781 conversion is legal then both result and operand refer to the same memory
5787 .. code-block:: llvm
5789 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5790 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5791 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5798 The instructions in this category are the "miscellaneous" instructions,
5799 which defy better classification.
5803 '``icmp``' Instruction
5804 ^^^^^^^^^^^^^^^^^^^^^^
5811 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5816 The '``icmp``' instruction returns a boolean value or a vector of
5817 boolean values based on comparison of its two integer, integer vector,
5818 pointer, or pointer vector operands.
5823 The '``icmp``' instruction takes three operands. The first operand is
5824 the condition code indicating the kind of comparison to perform. It is
5825 not a value, just a keyword. The possible condition code are:
5828 #. ``ne``: not equal
5829 #. ``ugt``: unsigned greater than
5830 #. ``uge``: unsigned greater or equal
5831 #. ``ult``: unsigned less than
5832 #. ``ule``: unsigned less or equal
5833 #. ``sgt``: signed greater than
5834 #. ``sge``: signed greater or equal
5835 #. ``slt``: signed less than
5836 #. ``sle``: signed less or equal
5838 The remaining two arguments must be :ref:`integer <t_integer>` or
5839 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5840 must also be identical types.
5845 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5846 code given as ``cond``. The comparison performed always yields either an
5847 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5849 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5850 otherwise. No sign interpretation is necessary or performed.
5851 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5852 otherwise. No sign interpretation is necessary or performed.
5853 #. ``ugt``: interprets the operands as unsigned values and yields
5854 ``true`` if ``op1`` is greater than ``op2``.
5855 #. ``uge``: interprets the operands as unsigned values and yields
5856 ``true`` if ``op1`` is greater than or equal to ``op2``.
5857 #. ``ult``: interprets the operands as unsigned values and yields
5858 ``true`` if ``op1`` is less than ``op2``.
5859 #. ``ule``: interprets the operands as unsigned values and yields
5860 ``true`` if ``op1`` is less than or equal to ``op2``.
5861 #. ``sgt``: interprets the operands as signed values and yields ``true``
5862 if ``op1`` is greater than ``op2``.
5863 #. ``sge``: interprets the operands as signed values and yields ``true``
5864 if ``op1`` is greater than or equal to ``op2``.
5865 #. ``slt``: interprets the operands as signed values and yields ``true``
5866 if ``op1`` is less than ``op2``.
5867 #. ``sle``: interprets the operands as signed values and yields ``true``
5868 if ``op1`` is less than or equal to ``op2``.
5870 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5871 are compared as if they were integers.
5873 If the operands are integer vectors, then they are compared element by
5874 element. The result is an ``i1`` vector with the same number of elements
5875 as the values being compared. Otherwise, the result is an ``i1``.
5880 .. code-block:: llvm
5882 <result> = icmp eq i32 4, 5 ; yields: result=false
5883 <result> = icmp ne float* %X, %X ; yields: result=false
5884 <result> = icmp ult i16 4, 5 ; yields: result=true
5885 <result> = icmp sgt i16 4, 5 ; yields: result=false
5886 <result> = icmp ule i16 -4, 5 ; yields: result=false
5887 <result> = icmp sge i16 4, 5 ; yields: result=false
5889 Note that the code generator does not yet support vector types with the
5890 ``icmp`` instruction.
5894 '``fcmp``' Instruction
5895 ^^^^^^^^^^^^^^^^^^^^^^
5902 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5907 The '``fcmp``' instruction returns a boolean value or vector of boolean
5908 values based on comparison of its operands.
5910 If the operands are floating point scalars, then the result type is a
5911 boolean (:ref:`i1 <t_integer>`).
5913 If the operands are floating point vectors, then the result type is a
5914 vector of boolean with the same number of elements as the operands being
5920 The '``fcmp``' instruction takes three operands. The first operand is
5921 the condition code indicating the kind of comparison to perform. It is
5922 not a value, just a keyword. The possible condition code are:
5924 #. ``false``: no comparison, always returns false
5925 #. ``oeq``: ordered and equal
5926 #. ``ogt``: ordered and greater than
5927 #. ``oge``: ordered and greater than or equal
5928 #. ``olt``: ordered and less than
5929 #. ``ole``: ordered and less than or equal
5930 #. ``one``: ordered and not equal
5931 #. ``ord``: ordered (no nans)
5932 #. ``ueq``: unordered or equal
5933 #. ``ugt``: unordered or greater than
5934 #. ``uge``: unordered or greater than or equal
5935 #. ``ult``: unordered or less than
5936 #. ``ule``: unordered or less than or equal
5937 #. ``une``: unordered or not equal
5938 #. ``uno``: unordered (either nans)
5939 #. ``true``: no comparison, always returns true
5941 *Ordered* means that neither operand is a QNAN while *unordered* means
5942 that either operand may be a QNAN.
5944 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5945 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5946 type. They must have identical types.
5951 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5952 condition code given as ``cond``. If the operands are vectors, then the
5953 vectors are compared element by element. Each comparison performed
5954 always yields an :ref:`i1 <t_integer>` result, as follows:
5956 #. ``false``: always yields ``false``, regardless of operands.
5957 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5958 is equal to ``op2``.
5959 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5960 is greater than ``op2``.
5961 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5962 is greater than or equal to ``op2``.
5963 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5964 is less than ``op2``.
5965 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5966 is less than or equal to ``op2``.
5967 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5968 is not equal to ``op2``.
5969 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5970 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5972 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5973 greater than ``op2``.
5974 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5975 greater than or equal to ``op2``.
5976 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5978 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5979 less than or equal to ``op2``.
5980 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5981 not equal to ``op2``.
5982 #. ``uno``: yields ``true`` if either operand is a QNAN.
5983 #. ``true``: always yields ``true``, regardless of operands.
5988 .. code-block:: llvm
5990 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5991 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5992 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5993 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5995 Note that the code generator does not yet support vector types with the
5996 ``fcmp`` instruction.
6000 '``phi``' Instruction
6001 ^^^^^^^^^^^^^^^^^^^^^
6008 <result> = phi <ty> [ <val0>, <label0>], ...
6013 The '``phi``' instruction is used to implement the φ node in the SSA
6014 graph representing the function.
6019 The type of the incoming values is specified with the first type field.
6020 After this, the '``phi``' instruction takes a list of pairs as
6021 arguments, with one pair for each predecessor basic block of the current
6022 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6023 the value arguments to the PHI node. Only labels may be used as the
6026 There must be no non-phi instructions between the start of a basic block
6027 and the PHI instructions: i.e. PHI instructions must be first in a basic
6030 For the purposes of the SSA form, the use of each incoming value is
6031 deemed to occur on the edge from the corresponding predecessor block to
6032 the current block (but after any definition of an '``invoke``'
6033 instruction's return value on the same edge).
6038 At runtime, the '``phi``' instruction logically takes on the value
6039 specified by the pair corresponding to the predecessor basic block that
6040 executed just prior to the current block.
6045 .. code-block:: llvm
6047 Loop: ; Infinite loop that counts from 0 on up...
6048 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6049 %nextindvar = add i32 %indvar, 1
6054 '``select``' Instruction
6055 ^^^^^^^^^^^^^^^^^^^^^^^^
6062 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6064 selty is either i1 or {<N x i1>}
6069 The '``select``' instruction is used to choose one value based on a
6070 condition, without branching.
6075 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6076 values indicating the condition, and two values of the same :ref:`first
6077 class <t_firstclass>` type. If the val1/val2 are vectors and the
6078 condition is a scalar, then entire vectors are selected, not individual
6084 If the condition is an i1 and it evaluates to 1, the instruction returns
6085 the first value argument; otherwise, it returns the second value
6088 If the condition is a vector of i1, then the value arguments must be
6089 vectors of the same size, and the selection is done element by element.
6094 .. code-block:: llvm
6096 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6100 '``call``' Instruction
6101 ^^^^^^^^^^^^^^^^^^^^^^
6108 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6113 The '``call``' instruction represents a simple function call.
6118 This instruction requires several arguments:
6120 #. The optional "tail" marker indicates that the callee function does
6121 not access any allocas or varargs in the caller. Note that calls may
6122 be marked "tail" even if they do not occur before a
6123 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6124 function call is eligible for tail call optimization, but `might not
6125 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6126 The code generator may optimize calls marked "tail" with either 1)
6127 automatic `sibling call
6128 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6129 callee have matching signatures, or 2) forced tail call optimization
6130 when the following extra requirements are met:
6132 - Caller and callee both have the calling convention ``fastcc``.
6133 - The call is in tail position (ret immediately follows call and ret
6134 uses value of call or is void).
6135 - Option ``-tailcallopt`` is enabled, or
6136 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6137 - `Platform specific constraints are
6138 met. <CodeGenerator.html#tailcallopt>`_
6140 #. The optional "cconv" marker indicates which :ref:`calling
6141 convention <callingconv>` the call should use. If none is
6142 specified, the call defaults to using C calling conventions. The
6143 calling convention of the call must match the calling convention of
6144 the target function, or else the behavior is undefined.
6145 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6146 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6148 #. '``ty``': the type of the call instruction itself which is also the
6149 type of the return value. Functions that return no value are marked
6151 #. '``fnty``': shall be the signature of the pointer to function value
6152 being invoked. The argument types must match the types implied by
6153 this signature. This type can be omitted if the function is not
6154 varargs and if the function type does not return a pointer to a
6156 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6157 be invoked. In most cases, this is a direct function invocation, but
6158 indirect ``call``'s are just as possible, calling an arbitrary pointer
6160 #. '``function args``': argument list whose types match the function
6161 signature argument types and parameter attributes. All arguments must
6162 be of :ref:`first class <t_firstclass>` type. If the function signature
6163 indicates the function accepts a variable number of arguments, the
6164 extra arguments can be specified.
6165 #. The optional :ref:`function attributes <fnattrs>` list. Only
6166 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6167 attributes are valid here.
6172 The '``call``' instruction is used to cause control flow to transfer to
6173 a specified function, with its incoming arguments bound to the specified
6174 values. Upon a '``ret``' instruction in the called function, control
6175 flow continues with the instruction after the function call, and the
6176 return value of the function is bound to the result argument.
6181 .. code-block:: llvm
6183 %retval = call i32 @test(i32 %argc)
6184 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6185 %X = tail call i32 @foo() ; yields i32
6186 %Y = tail call fastcc i32 @foo() ; yields i32
6187 call void %foo(i8 97 signext)
6189 %struct.A = type { i32, i8 }
6190 %r = call %struct.A @foo() ; yields { 32, i8 }
6191 %gr = extractvalue %struct.A %r, 0 ; yields i32
6192 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6193 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6194 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6196 llvm treats calls to some functions with names and arguments that match
6197 the standard C99 library as being the C99 library functions, and may
6198 perform optimizations or generate code for them under that assumption.
6199 This is something we'd like to change in the future to provide better
6200 support for freestanding environments and non-C-based languages.
6204 '``va_arg``' Instruction
6205 ^^^^^^^^^^^^^^^^^^^^^^^^
6212 <resultval> = va_arg <va_list*> <arglist>, <argty>
6217 The '``va_arg``' instruction is used to access arguments passed through
6218 the "variable argument" area of a function call. It is used to implement
6219 the ``va_arg`` macro in C.
6224 This instruction takes a ``va_list*`` value and the type of the
6225 argument. It returns a value of the specified argument type and
6226 increments the ``va_list`` to point to the next argument. The actual
6227 type of ``va_list`` is target specific.
6232 The '``va_arg``' instruction loads an argument of the specified type
6233 from the specified ``va_list`` and causes the ``va_list`` to point to
6234 the next argument. For more information, see the variable argument
6235 handling :ref:`Intrinsic Functions <int_varargs>`.
6237 It is legal for this instruction to be called in a function which does
6238 not take a variable number of arguments, for example, the ``vfprintf``
6241 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6242 function <intrinsics>` because it takes a type as an argument.
6247 See the :ref:`variable argument processing <int_varargs>` section.
6249 Note that the code generator does not yet fully support va\_arg on many
6250 targets. Also, it does not currently support va\_arg with aggregate
6251 types on any target.
6255 '``landingpad``' Instruction
6256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6263 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6264 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6266 <clause> := catch <type> <value>
6267 <clause> := filter <array constant type> <array constant>
6272 The '``landingpad``' instruction is used by `LLVM's exception handling
6273 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6274 is a landing pad --- one where the exception lands, and corresponds to the
6275 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6276 defines values supplied by the personality function (``pers_fn``) upon
6277 re-entry to the function. The ``resultval`` has the type ``resultty``.
6282 This instruction takes a ``pers_fn`` value. This is the personality
6283 function associated with the unwinding mechanism. The optional
6284 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6286 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6287 contains the global variable representing the "type" that may be caught
6288 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6289 clause takes an array constant as its argument. Use
6290 "``[0 x i8**] undef``" for a filter which cannot throw. The
6291 '``landingpad``' instruction must contain *at least* one ``clause`` or
6292 the ``cleanup`` flag.
6297 The '``landingpad``' instruction defines the values which are set by the
6298 personality function (``pers_fn``) upon re-entry to the function, and
6299 therefore the "result type" of the ``landingpad`` instruction. As with
6300 calling conventions, how the personality function results are
6301 represented in LLVM IR is target specific.
6303 The clauses are applied in order from top to bottom. If two
6304 ``landingpad`` instructions are merged together through inlining, the
6305 clauses from the calling function are appended to the list of clauses.
6306 When the call stack is being unwound due to an exception being thrown,
6307 the exception is compared against each ``clause`` in turn. If it doesn't
6308 match any of the clauses, and the ``cleanup`` flag is not set, then
6309 unwinding continues further up the call stack.
6311 The ``landingpad`` instruction has several restrictions:
6313 - A landing pad block is a basic block which is the unwind destination
6314 of an '``invoke``' instruction.
6315 - A landing pad block must have a '``landingpad``' instruction as its
6316 first non-PHI instruction.
6317 - There can be only one '``landingpad``' instruction within the landing
6319 - A basic block that is not a landing pad block may not include a
6320 '``landingpad``' instruction.
6321 - All '``landingpad``' instructions in a function must have the same
6322 personality function.
6327 .. code-block:: llvm
6329 ;; A landing pad which can catch an integer.
6330 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6332 ;; A landing pad that is a cleanup.
6333 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6335 ;; A landing pad which can catch an integer and can only throw a double.
6336 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6338 filter [1 x i8**] [@_ZTId]
6345 LLVM supports the notion of an "intrinsic function". These functions
6346 have well known names and semantics and are required to follow certain
6347 restrictions. Overall, these intrinsics represent an extension mechanism
6348 for the LLVM language that does not require changing all of the
6349 transformations in LLVM when adding to the language (or the bitcode
6350 reader/writer, the parser, etc...).
6352 Intrinsic function names must all start with an "``llvm.``" prefix. This
6353 prefix is reserved in LLVM for intrinsic names; thus, function names may
6354 not begin with this prefix. Intrinsic functions must always be external
6355 functions: you cannot define the body of intrinsic functions. Intrinsic
6356 functions may only be used in call or invoke instructions: it is illegal
6357 to take the address of an intrinsic function. Additionally, because
6358 intrinsic functions are part of the LLVM language, it is required if any
6359 are added that they be documented here.
6361 Some intrinsic functions can be overloaded, i.e., the intrinsic
6362 represents a family of functions that perform the same operation but on
6363 different data types. Because LLVM can represent over 8 million
6364 different integer types, overloading is used commonly to allow an
6365 intrinsic function to operate on any integer type. One or more of the
6366 argument types or the result type can be overloaded to accept any
6367 integer type. Argument types may also be defined as exactly matching a
6368 previous argument's type or the result type. This allows an intrinsic
6369 function which accepts multiple arguments, but needs all of them to be
6370 of the same type, to only be overloaded with respect to a single
6371 argument or the result.
6373 Overloaded intrinsics will have the names of its overloaded argument
6374 types encoded into its function name, each preceded by a period. Only
6375 those types which are overloaded result in a name suffix. Arguments
6376 whose type is matched against another type do not. For example, the
6377 ``llvm.ctpop`` function can take an integer of any width and returns an
6378 integer of exactly the same integer width. This leads to a family of
6379 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6380 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6381 overloaded, and only one type suffix is required. Because the argument's
6382 type is matched against the return type, it does not require its own
6385 To learn how to add an intrinsic function, please see the `Extending
6386 LLVM Guide <ExtendingLLVM.html>`_.
6390 Variable Argument Handling Intrinsics
6391 -------------------------------------
6393 Variable argument support is defined in LLVM with the
6394 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6395 functions. These functions are related to the similarly named macros
6396 defined in the ``<stdarg.h>`` header file.
6398 All of these functions operate on arguments that use a target-specific
6399 value type "``va_list``". The LLVM assembly language reference manual
6400 does not define what this type is, so all transformations should be
6401 prepared to handle these functions regardless of the type used.
6403 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6404 variable argument handling intrinsic functions are used.
6406 .. code-block:: llvm
6408 define i32 @test(i32 %X, ...) {
6409 ; Initialize variable argument processing
6411 %ap2 = bitcast i8** %ap to i8*
6412 call void @llvm.va_start(i8* %ap2)
6414 ; Read a single integer argument
6415 %tmp = va_arg i8** %ap, i32
6417 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6419 %aq2 = bitcast i8** %aq to i8*
6420 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6421 call void @llvm.va_end(i8* %aq2)
6423 ; Stop processing of arguments.
6424 call void @llvm.va_end(i8* %ap2)
6428 declare void @llvm.va_start(i8*)
6429 declare void @llvm.va_copy(i8*, i8*)
6430 declare void @llvm.va_end(i8*)
6434 '``llvm.va_start``' Intrinsic
6435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6442 declare void @llvm.va_start(i8* <arglist>)
6447 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6448 subsequent use by ``va_arg``.
6453 The argument is a pointer to a ``va_list`` element to initialize.
6458 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6459 available in C. In a target-dependent way, it initializes the
6460 ``va_list`` element to which the argument points, so that the next call
6461 to ``va_arg`` will produce the first variable argument passed to the
6462 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6463 to know the last argument of the function as the compiler can figure
6466 '``llvm.va_end``' Intrinsic
6467 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6474 declare void @llvm.va_end(i8* <arglist>)
6479 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6480 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6485 The argument is a pointer to a ``va_list`` to destroy.
6490 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6491 available in C. In a target-dependent way, it destroys the ``va_list``
6492 element to which the argument points. Calls to
6493 :ref:`llvm.va_start <int_va_start>` and
6494 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6499 '``llvm.va_copy``' Intrinsic
6500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6507 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6512 The '``llvm.va_copy``' intrinsic copies the current argument position
6513 from the source argument list to the destination argument list.
6518 The first argument is a pointer to a ``va_list`` element to initialize.
6519 The second argument is a pointer to a ``va_list`` element to copy from.
6524 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6525 available in C. In a target-dependent way, it copies the source
6526 ``va_list`` element into the destination ``va_list`` element. This
6527 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6528 arbitrarily complex and require, for example, memory allocation.
6530 Accurate Garbage Collection Intrinsics
6531 --------------------------------------
6533 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6534 (GC) requires the implementation and generation of these intrinsics.
6535 These intrinsics allow identification of :ref:`GC roots on the
6536 stack <int_gcroot>`, as well as garbage collector implementations that
6537 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6538 Front-ends for type-safe garbage collected languages should generate
6539 these intrinsics to make use of the LLVM garbage collectors. For more
6540 details, see `Accurate Garbage Collection with
6541 LLVM <GarbageCollection.html>`_.
6543 The garbage collection intrinsics only operate on objects in the generic
6544 address space (address space zero).
6548 '``llvm.gcroot``' Intrinsic
6549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6556 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6561 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6562 the code generator, and allows some metadata to be associated with it.
6567 The first argument specifies the address of a stack object that contains
6568 the root pointer. The second pointer (which must be either a constant or
6569 a global value address) contains the meta-data to be associated with the
6575 At runtime, a call to this intrinsic stores a null pointer into the
6576 "ptrloc" location. At compile-time, the code generator generates
6577 information to allow the runtime to find the pointer at GC safe points.
6578 The '``llvm.gcroot``' intrinsic may only be used in a function which
6579 :ref:`specifies a GC algorithm <gc>`.
6583 '``llvm.gcread``' Intrinsic
6584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6591 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6596 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6597 locations, allowing garbage collector implementations that require read
6603 The second argument is the address to read from, which should be an
6604 address allocated from the garbage collector. The first object is a
6605 pointer to the start of the referenced object, if needed by the language
6606 runtime (otherwise null).
6611 The '``llvm.gcread``' intrinsic has the same semantics as a load
6612 instruction, but may be replaced with substantially more complex code by
6613 the garbage collector runtime, as needed. The '``llvm.gcread``'
6614 intrinsic may only be used in a function which :ref:`specifies a GC
6619 '``llvm.gcwrite``' Intrinsic
6620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6627 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6632 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6633 locations, allowing garbage collector implementations that require write
6634 barriers (such as generational or reference counting collectors).
6639 The first argument is the reference to store, the second is the start of
6640 the object to store it to, and the third is the address of the field of
6641 Obj to store to. If the runtime does not require a pointer to the
6642 object, Obj may be null.
6647 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6648 instruction, but may be replaced with substantially more complex code by
6649 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6650 intrinsic may only be used in a function which :ref:`specifies a GC
6653 Code Generator Intrinsics
6654 -------------------------
6656 These intrinsics are provided by LLVM to expose special features that
6657 may only be implemented with code generator support.
6659 '``llvm.returnaddress``' Intrinsic
6660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6667 declare i8 *@llvm.returnaddress(i32 <level>)
6672 The '``llvm.returnaddress``' intrinsic attempts to compute a
6673 target-specific value indicating the return address of the current
6674 function or one of its callers.
6679 The argument to this intrinsic indicates which function to return the
6680 address for. Zero indicates the calling function, one indicates its
6681 caller, etc. The argument is **required** to be a constant integer
6687 The '``llvm.returnaddress``' intrinsic either returns a pointer
6688 indicating the return address of the specified call frame, or zero if it
6689 cannot be identified. The value returned by this intrinsic is likely to
6690 be incorrect or 0 for arguments other than zero, so it should only be
6691 used for debugging purposes.
6693 Note that calling this intrinsic does not prevent function inlining or
6694 other aggressive transformations, so the value returned may not be that
6695 of the obvious source-language caller.
6697 '``llvm.frameaddress``' Intrinsic
6698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6705 declare i8* @llvm.frameaddress(i32 <level>)
6710 The '``llvm.frameaddress``' intrinsic attempts to return the
6711 target-specific frame pointer value for the specified stack frame.
6716 The argument to this intrinsic indicates which function to return the
6717 frame pointer for. Zero indicates the calling function, one indicates
6718 its caller, etc. The argument is **required** to be a constant integer
6724 The '``llvm.frameaddress``' intrinsic either returns a pointer
6725 indicating the frame address of the specified call frame, or zero if it
6726 cannot be identified. The value returned by this intrinsic is likely to
6727 be incorrect or 0 for arguments other than zero, so it should only be
6728 used for debugging purposes.
6730 Note that calling this intrinsic does not prevent function inlining or
6731 other aggressive transformations, so the value returned may not be that
6732 of the obvious source-language caller.
6736 '``llvm.stacksave``' Intrinsic
6737 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6744 declare i8* @llvm.stacksave()
6749 The '``llvm.stacksave``' intrinsic is used to remember the current state
6750 of the function stack, for use with
6751 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6752 implementing language features like scoped automatic variable sized
6758 This intrinsic returns a opaque pointer value that can be passed to
6759 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6760 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6761 ``llvm.stacksave``, it effectively restores the state of the stack to
6762 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6763 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6764 were allocated after the ``llvm.stacksave`` was executed.
6766 .. _int_stackrestore:
6768 '``llvm.stackrestore``' Intrinsic
6769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6776 declare void @llvm.stackrestore(i8* %ptr)
6781 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6782 the function stack to the state it was in when the corresponding
6783 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6784 useful for implementing language features like scoped automatic variable
6785 sized arrays in C99.
6790 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6792 '``llvm.prefetch``' Intrinsic
6793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6800 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6805 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6806 insert a prefetch instruction if supported; otherwise, it is a noop.
6807 Prefetches have no effect on the behavior of the program but can change
6808 its performance characteristics.
6813 ``address`` is the address to be prefetched, ``rw`` is the specifier
6814 determining if the fetch should be for a read (0) or write (1), and
6815 ``locality`` is a temporal locality specifier ranging from (0) - no
6816 locality, to (3) - extremely local keep in cache. The ``cache type``
6817 specifies whether the prefetch is performed on the data (1) or
6818 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6819 arguments must be constant integers.
6824 This intrinsic does not modify the behavior of the program. In
6825 particular, prefetches cannot trap and do not produce a value. On
6826 targets that support this intrinsic, the prefetch can provide hints to
6827 the processor cache for better performance.
6829 '``llvm.pcmarker``' Intrinsic
6830 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6837 declare void @llvm.pcmarker(i32 <id>)
6842 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6843 Counter (PC) in a region of code to simulators and other tools. The
6844 method is target specific, but it is expected that the marker will use
6845 exported symbols to transmit the PC of the marker. The marker makes no
6846 guarantees that it will remain with any specific instruction after
6847 optimizations. It is possible that the presence of a marker will inhibit
6848 optimizations. The intended use is to be inserted after optimizations to
6849 allow correlations of simulation runs.
6854 ``id`` is a numerical id identifying the marker.
6859 This intrinsic does not modify the behavior of the program. Backends
6860 that do not support this intrinsic may ignore it.
6862 '``llvm.readcyclecounter``' Intrinsic
6863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6870 declare i64 @llvm.readcyclecounter()
6875 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6876 counter register (or similar low latency, high accuracy clocks) on those
6877 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6878 should map to RPCC. As the backing counters overflow quickly (on the
6879 order of 9 seconds on alpha), this should only be used for small
6885 When directly supported, reading the cycle counter should not modify any
6886 memory. Implementations are allowed to either return a application
6887 specific value or a system wide value. On backends without support, this
6888 is lowered to a constant 0.
6890 Note that runtime support may be conditional on the privilege-level code is
6891 running at and the host platform.
6893 Standard C Library Intrinsics
6894 -----------------------------
6896 LLVM provides intrinsics for a few important standard C library
6897 functions. These intrinsics allow source-language front-ends to pass
6898 information about the alignment of the pointer arguments to the code
6899 generator, providing opportunity for more efficient code generation.
6903 '``llvm.memcpy``' Intrinsic
6904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6909 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6910 integer bit width and for different address spaces. Not all targets
6911 support all bit widths however.
6915 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6916 i32 <len>, i32 <align>, i1 <isvolatile>)
6917 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6918 i64 <len>, i32 <align>, i1 <isvolatile>)
6923 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6924 source location to the destination location.
6926 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6927 intrinsics do not return a value, takes extra alignment/isvolatile
6928 arguments and the pointers can be in specified address spaces.
6933 The first argument is a pointer to the destination, the second is a
6934 pointer to the source. The third argument is an integer argument
6935 specifying the number of bytes to copy, the fourth argument is the
6936 alignment of the source and destination locations, and the fifth is a
6937 boolean indicating a volatile access.
6939 If the call to this intrinsic has an alignment value that is not 0 or 1,
6940 then the caller guarantees that both the source and destination pointers
6941 are aligned to that boundary.
6943 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6944 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6945 very cleanly specified and it is unwise to depend on it.
6950 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6951 source location to the destination location, which are not allowed to
6952 overlap. It copies "len" bytes of memory over. If the argument is known
6953 to be aligned to some boundary, this can be specified as the fourth
6954 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6956 '``llvm.memmove``' Intrinsic
6957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6962 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6963 bit width and for different address space. Not all targets support all
6968 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6969 i32 <len>, i32 <align>, i1 <isvolatile>)
6970 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6971 i64 <len>, i32 <align>, i1 <isvolatile>)
6976 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6977 source location to the destination location. It is similar to the
6978 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6981 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6982 intrinsics do not return a value, takes extra alignment/isvolatile
6983 arguments and the pointers can be in specified address spaces.
6988 The first argument is a pointer to the destination, the second is a
6989 pointer to the source. The third argument is an integer argument
6990 specifying the number of bytes to copy, the fourth argument is the
6991 alignment of the source and destination locations, and the fifth is a
6992 boolean indicating a volatile access.
6994 If the call to this intrinsic has an alignment value that is not 0 or 1,
6995 then the caller guarantees that the source and destination pointers are
6996 aligned to that boundary.
6998 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6999 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7000 not very cleanly specified and it is unwise to depend on it.
7005 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7006 source location to the destination location, which may overlap. It
7007 copies "len" bytes of memory over. If the argument is known to be
7008 aligned to some boundary, this can be specified as the fourth argument,
7009 otherwise it should be set to 0 or 1 (both meaning no alignment).
7011 '``llvm.memset.*``' Intrinsics
7012 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7017 This is an overloaded intrinsic. You can use llvm.memset on any integer
7018 bit width and for different address spaces. However, not all targets
7019 support all bit widths.
7023 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7024 i32 <len>, i32 <align>, i1 <isvolatile>)
7025 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7026 i64 <len>, i32 <align>, i1 <isvolatile>)
7031 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7032 particular byte value.
7034 Note that, unlike the standard libc function, the ``llvm.memset``
7035 intrinsic does not return a value and takes extra alignment/volatile
7036 arguments. Also, the destination can be in an arbitrary address space.
7041 The first argument is a pointer to the destination to fill, the second
7042 is the byte value with which to fill it, the third argument is an
7043 integer argument specifying the number of bytes to fill, and the fourth
7044 argument is the known alignment of the destination location.
7046 If the call to this intrinsic has an alignment value that is not 0 or 1,
7047 then the caller guarantees that the destination pointer is aligned to
7050 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7051 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7052 very cleanly specified and it is unwise to depend on it.
7057 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7058 at the destination location. If the argument is known to be aligned to
7059 some boundary, this can be specified as the fourth argument, otherwise
7060 it should be set to 0 or 1 (both meaning no alignment).
7062 '``llvm.sqrt.*``' Intrinsic
7063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7068 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7069 floating point or vector of floating point type. Not all targets support
7074 declare float @llvm.sqrt.f32(float %Val)
7075 declare double @llvm.sqrt.f64(double %Val)
7076 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7077 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7078 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7083 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7084 returning the same value as the libm '``sqrt``' functions would. Unlike
7085 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7086 negative numbers other than -0.0 (which allows for better optimization,
7087 because there is no need to worry about errno being set).
7088 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7093 The argument and return value are floating point numbers of the same
7099 This function returns the sqrt of the specified operand if it is a
7100 nonnegative floating point number.
7102 '``llvm.powi.*``' Intrinsic
7103 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7108 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7109 floating point or vector of floating point type. Not all targets support
7114 declare float @llvm.powi.f32(float %Val, i32 %power)
7115 declare double @llvm.powi.f64(double %Val, i32 %power)
7116 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7117 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7118 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7123 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7124 specified (positive or negative) power. The order of evaluation of
7125 multiplications is not defined. When a vector of floating point type is
7126 used, the second argument remains a scalar integer value.
7131 The second argument is an integer power, and the first is a value to
7132 raise to that power.
7137 This function returns the first value raised to the second power with an
7138 unspecified sequence of rounding operations.
7140 '``llvm.sin.*``' Intrinsic
7141 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7146 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7147 floating point or vector of floating point type. Not all targets support
7152 declare float @llvm.sin.f32(float %Val)
7153 declare double @llvm.sin.f64(double %Val)
7154 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7155 declare fp128 @llvm.sin.f128(fp128 %Val)
7156 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7161 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7166 The argument and return value are floating point numbers of the same
7172 This function returns the sine of the specified operand, returning the
7173 same values as the libm ``sin`` functions would, and handles error
7174 conditions in the same way.
7176 '``llvm.cos.*``' Intrinsic
7177 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7182 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7183 floating point or vector of floating point type. Not all targets support
7188 declare float @llvm.cos.f32(float %Val)
7189 declare double @llvm.cos.f64(double %Val)
7190 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7191 declare fp128 @llvm.cos.f128(fp128 %Val)
7192 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7197 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7202 The argument and return value are floating point numbers of the same
7208 This function returns the cosine of the specified operand, returning the
7209 same values as the libm ``cos`` functions would, and handles error
7210 conditions in the same way.
7212 '``llvm.pow.*``' Intrinsic
7213 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7218 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7219 floating point or vector of floating point type. Not all targets support
7224 declare float @llvm.pow.f32(float %Val, float %Power)
7225 declare double @llvm.pow.f64(double %Val, double %Power)
7226 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7227 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7228 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7233 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7234 specified (positive or negative) power.
7239 The second argument is a floating point power, and the first is a value
7240 to raise to that power.
7245 This function returns the first value raised to the second power,
7246 returning the same values as the libm ``pow`` functions would, and
7247 handles error conditions in the same way.
7249 '``llvm.exp.*``' Intrinsic
7250 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7255 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7256 floating point or vector of floating point type. Not all targets support
7261 declare float @llvm.exp.f32(float %Val)
7262 declare double @llvm.exp.f64(double %Val)
7263 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7264 declare fp128 @llvm.exp.f128(fp128 %Val)
7265 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7270 The '``llvm.exp.*``' intrinsics perform the exp function.
7275 The argument and return value are floating point numbers of the same
7281 This function returns the same values as the libm ``exp`` functions
7282 would, and handles error conditions in the same way.
7284 '``llvm.exp2.*``' Intrinsic
7285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7290 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7291 floating point or vector of floating point type. Not all targets support
7296 declare float @llvm.exp2.f32(float %Val)
7297 declare double @llvm.exp2.f64(double %Val)
7298 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7299 declare fp128 @llvm.exp2.f128(fp128 %Val)
7300 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7305 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7310 The argument and return value are floating point numbers of the same
7316 This function returns the same values as the libm ``exp2`` functions
7317 would, and handles error conditions in the same way.
7319 '``llvm.log.*``' Intrinsic
7320 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7325 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7326 floating point or vector of floating point type. Not all targets support
7331 declare float @llvm.log.f32(float %Val)
7332 declare double @llvm.log.f64(double %Val)
7333 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7334 declare fp128 @llvm.log.f128(fp128 %Val)
7335 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7340 The '``llvm.log.*``' intrinsics perform the log function.
7345 The argument and return value are floating point numbers of the same
7351 This function returns the same values as the libm ``log`` functions
7352 would, and handles error conditions in the same way.
7354 '``llvm.log10.*``' Intrinsic
7355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7360 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7361 floating point or vector of floating point type. Not all targets support
7366 declare float @llvm.log10.f32(float %Val)
7367 declare double @llvm.log10.f64(double %Val)
7368 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7369 declare fp128 @llvm.log10.f128(fp128 %Val)
7370 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7375 The '``llvm.log10.*``' intrinsics perform the log10 function.
7380 The argument and return value are floating point numbers of the same
7386 This function returns the same values as the libm ``log10`` functions
7387 would, and handles error conditions in the same way.
7389 '``llvm.log2.*``' Intrinsic
7390 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7395 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7396 floating point or vector of floating point type. Not all targets support
7401 declare float @llvm.log2.f32(float %Val)
7402 declare double @llvm.log2.f64(double %Val)
7403 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7404 declare fp128 @llvm.log2.f128(fp128 %Val)
7405 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7410 The '``llvm.log2.*``' intrinsics perform the log2 function.
7415 The argument and return value are floating point numbers of the same
7421 This function returns the same values as the libm ``log2`` functions
7422 would, and handles error conditions in the same way.
7424 '``llvm.fma.*``' Intrinsic
7425 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7430 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7431 floating point or vector of floating point type. Not all targets support
7436 declare float @llvm.fma.f32(float %a, float %b, float %c)
7437 declare double @llvm.fma.f64(double %a, double %b, double %c)
7438 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7439 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7440 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7445 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7451 The argument and return value are floating point numbers of the same
7457 This function returns the same values as the libm ``fma`` functions
7460 '``llvm.fabs.*``' Intrinsic
7461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7466 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7467 floating point or vector of floating point type. Not all targets support
7472 declare float @llvm.fabs.f32(float %Val)
7473 declare double @llvm.fabs.f64(double %Val)
7474 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7475 declare fp128 @llvm.fabs.f128(fp128 %Val)
7476 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7481 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7487 The argument and return value are floating point numbers of the same
7493 This function returns the same values as the libm ``fabs`` functions
7494 would, and handles error conditions in the same way.
7496 '``llvm.copysign.*``' Intrinsic
7497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7502 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7503 floating point or vector of floating point type. Not all targets support
7508 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7509 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7510 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7511 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7512 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7517 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7518 first operand and the sign of the second operand.
7523 The arguments and return value are floating point numbers of the same
7529 This function returns the same values as the libm ``copysign``
7530 functions would, and handles error conditions in the same way.
7532 '``llvm.floor.*``' Intrinsic
7533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7538 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7539 floating point or vector of floating point type. Not all targets support
7544 declare float @llvm.floor.f32(float %Val)
7545 declare double @llvm.floor.f64(double %Val)
7546 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7547 declare fp128 @llvm.floor.f128(fp128 %Val)
7548 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7553 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7558 The argument and return value are floating point numbers of the same
7564 This function returns the same values as the libm ``floor`` functions
7565 would, and handles error conditions in the same way.
7567 '``llvm.ceil.*``' Intrinsic
7568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7573 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7574 floating point or vector of floating point type. Not all targets support
7579 declare float @llvm.ceil.f32(float %Val)
7580 declare double @llvm.ceil.f64(double %Val)
7581 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7582 declare fp128 @llvm.ceil.f128(fp128 %Val)
7583 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7588 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7593 The argument and return value are floating point numbers of the same
7599 This function returns the same values as the libm ``ceil`` functions
7600 would, and handles error conditions in the same way.
7602 '``llvm.trunc.*``' Intrinsic
7603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7608 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7609 floating point or vector of floating point type. Not all targets support
7614 declare float @llvm.trunc.f32(float %Val)
7615 declare double @llvm.trunc.f64(double %Val)
7616 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7617 declare fp128 @llvm.trunc.f128(fp128 %Val)
7618 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7623 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7624 nearest integer not larger in magnitude than the operand.
7629 The argument and return value are floating point numbers of the same
7635 This function returns the same values as the libm ``trunc`` functions
7636 would, and handles error conditions in the same way.
7638 '``llvm.rint.*``' Intrinsic
7639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7644 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7645 floating point or vector of floating point type. Not all targets support
7650 declare float @llvm.rint.f32(float %Val)
7651 declare double @llvm.rint.f64(double %Val)
7652 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7653 declare fp128 @llvm.rint.f128(fp128 %Val)
7654 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7659 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7660 nearest integer. It may raise an inexact floating-point exception if the
7661 operand isn't an integer.
7666 The argument and return value are floating point numbers of the same
7672 This function returns the same values as the libm ``rint`` functions
7673 would, and handles error conditions in the same way.
7675 '``llvm.nearbyint.*``' Intrinsic
7676 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7681 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7682 floating point or vector of floating point type. Not all targets support
7687 declare float @llvm.nearbyint.f32(float %Val)
7688 declare double @llvm.nearbyint.f64(double %Val)
7689 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7690 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7691 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7696 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7702 The argument and return value are floating point numbers of the same
7708 This function returns the same values as the libm ``nearbyint``
7709 functions would, and handles error conditions in the same way.
7711 '``llvm.round.*``' Intrinsic
7712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7717 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7718 floating point or vector of floating point type. Not all targets support
7723 declare float @llvm.round.f32(float %Val)
7724 declare double @llvm.round.f64(double %Val)
7725 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7726 declare fp128 @llvm.round.f128(fp128 %Val)
7727 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7732 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7738 The argument and return value are floating point numbers of the same
7744 This function returns the same values as the libm ``round``
7745 functions would, and handles error conditions in the same way.
7747 Bit Manipulation Intrinsics
7748 ---------------------------
7750 LLVM provides intrinsics for a few important bit manipulation
7751 operations. These allow efficient code generation for some algorithms.
7753 '``llvm.bswap.*``' Intrinsics
7754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7759 This is an overloaded intrinsic function. You can use bswap on any
7760 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7764 declare i16 @llvm.bswap.i16(i16 <id>)
7765 declare i32 @llvm.bswap.i32(i32 <id>)
7766 declare i64 @llvm.bswap.i64(i64 <id>)
7771 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7772 values with an even number of bytes (positive multiple of 16 bits).
7773 These are useful for performing operations on data that is not in the
7774 target's native byte order.
7779 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7780 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7781 intrinsic returns an i32 value that has the four bytes of the input i32
7782 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7783 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7784 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7785 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7788 '``llvm.ctpop.*``' Intrinsic
7789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7794 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7795 bit width, or on any vector with integer elements. Not all targets
7796 support all bit widths or vector types, however.
7800 declare i8 @llvm.ctpop.i8(i8 <src>)
7801 declare i16 @llvm.ctpop.i16(i16 <src>)
7802 declare i32 @llvm.ctpop.i32(i32 <src>)
7803 declare i64 @llvm.ctpop.i64(i64 <src>)
7804 declare i256 @llvm.ctpop.i256(i256 <src>)
7805 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7810 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7816 The only argument is the value to be counted. The argument may be of any
7817 integer type, or a vector with integer elements. The return type must
7818 match the argument type.
7823 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7824 each element of a vector.
7826 '``llvm.ctlz.*``' Intrinsic
7827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7832 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7833 integer bit width, or any vector whose elements are integers. Not all
7834 targets support all bit widths or vector types, however.
7838 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7839 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7840 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7841 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7842 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7843 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7848 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7849 leading zeros in a variable.
7854 The first argument is the value to be counted. This argument may be of
7855 any integer type, or a vectory with integer element type. The return
7856 type must match the first argument type.
7858 The second argument must be a constant and is a flag to indicate whether
7859 the intrinsic should ensure that a zero as the first argument produces a
7860 defined result. Historically some architectures did not provide a
7861 defined result for zero values as efficiently, and many algorithms are
7862 now predicated on avoiding zero-value inputs.
7867 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7868 zeros in a variable, or within each element of the vector. If
7869 ``src == 0`` then the result is the size in bits of the type of ``src``
7870 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7871 ``llvm.ctlz(i32 2) = 30``.
7873 '``llvm.cttz.*``' Intrinsic
7874 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7879 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7880 integer bit width, or any vector of integer elements. Not all targets
7881 support all bit widths or vector types, however.
7885 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7886 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7887 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7888 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7889 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7890 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7895 The '``llvm.cttz``' family of intrinsic functions counts the number of
7901 The first argument is the value to be counted. This argument may be of
7902 any integer type, or a vectory with integer element type. The return
7903 type must match the first argument type.
7905 The second argument must be a constant and is a flag to indicate whether
7906 the intrinsic should ensure that a zero as the first argument produces a
7907 defined result. Historically some architectures did not provide a
7908 defined result for zero values as efficiently, and many algorithms are
7909 now predicated on avoiding zero-value inputs.
7914 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7915 zeros in a variable, or within each element of a vector. If ``src == 0``
7916 then the result is the size in bits of the type of ``src`` if
7917 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7918 ``llvm.cttz(2) = 1``.
7920 Arithmetic with Overflow Intrinsics
7921 -----------------------------------
7923 LLVM provides intrinsics for some arithmetic with overflow operations.
7925 '``llvm.sadd.with.overflow.*``' Intrinsics
7926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7931 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7932 on any integer bit width.
7936 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7937 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7938 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7943 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7944 a signed addition of the two arguments, and indicate whether an overflow
7945 occurred during the signed summation.
7950 The arguments (%a and %b) and the first element of the result structure
7951 may be of integer types of any bit width, but they must have the same
7952 bit width. The second element of the result structure must be of type
7953 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7959 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7960 a signed addition of the two variables. They return a structure --- the
7961 first element of which is the signed summation, and the second element
7962 of which is a bit specifying if the signed summation resulted in an
7968 .. code-block:: llvm
7970 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7971 %sum = extractvalue {i32, i1} %res, 0
7972 %obit = extractvalue {i32, i1} %res, 1
7973 br i1 %obit, label %overflow, label %normal
7975 '``llvm.uadd.with.overflow.*``' Intrinsics
7976 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7981 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7982 on any integer bit width.
7986 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7987 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7988 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7993 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7994 an unsigned addition of the two arguments, and indicate whether a carry
7995 occurred during the unsigned summation.
8000 The arguments (%a and %b) and the first element of the result structure
8001 may be of integer types of any bit width, but they must have the same
8002 bit width. The second element of the result structure must be of type
8003 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8009 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8010 an unsigned addition of the two arguments. They return a structure --- the
8011 first element of which is the sum, and the second element of which is a
8012 bit specifying if the unsigned summation resulted in a carry.
8017 .. code-block:: llvm
8019 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8020 %sum = extractvalue {i32, i1} %res, 0
8021 %obit = extractvalue {i32, i1} %res, 1
8022 br i1 %obit, label %carry, label %normal
8024 '``llvm.ssub.with.overflow.*``' Intrinsics
8025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8030 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8031 on any integer bit width.
8035 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8036 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8037 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8042 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8043 a signed subtraction of the two arguments, and indicate whether an
8044 overflow occurred during the signed subtraction.
8049 The arguments (%a and %b) and the first element of the result structure
8050 may be of integer types of any bit width, but they must have the same
8051 bit width. The second element of the result structure must be of type
8052 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8058 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8059 a signed subtraction of the two arguments. They return a structure --- the
8060 first element of which is the subtraction, and the second element of
8061 which is a bit specifying if the signed subtraction resulted in an
8067 .. code-block:: llvm
8069 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8070 %sum = extractvalue {i32, i1} %res, 0
8071 %obit = extractvalue {i32, i1} %res, 1
8072 br i1 %obit, label %overflow, label %normal
8074 '``llvm.usub.with.overflow.*``' Intrinsics
8075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8080 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8081 on any integer bit width.
8085 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8086 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8087 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8092 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8093 an unsigned subtraction of the two arguments, and indicate whether an
8094 overflow occurred during the unsigned subtraction.
8099 The arguments (%a and %b) and the first element of the result structure
8100 may be of integer types of any bit width, but they must have the same
8101 bit width. The second element of the result structure must be of type
8102 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8108 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8109 an unsigned subtraction of the two arguments. They return a structure ---
8110 the first element of which is the subtraction, and the second element of
8111 which is a bit specifying if the unsigned subtraction resulted in an
8117 .. code-block:: llvm
8119 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8120 %sum = extractvalue {i32, i1} %res, 0
8121 %obit = extractvalue {i32, i1} %res, 1
8122 br i1 %obit, label %overflow, label %normal
8124 '``llvm.smul.with.overflow.*``' Intrinsics
8125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8130 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8131 on any integer bit width.
8135 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8136 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8137 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8142 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8143 a signed multiplication of the two arguments, and indicate whether an
8144 overflow occurred during the signed multiplication.
8149 The arguments (%a and %b) and the first element of the result structure
8150 may be of integer types of any bit width, but they must have the same
8151 bit width. The second element of the result structure must be of type
8152 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8158 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8159 a signed multiplication of the two arguments. They return a structure ---
8160 the first element of which is the multiplication, and the second element
8161 of which is a bit specifying if the signed multiplication resulted in an
8167 .. code-block:: llvm
8169 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8170 %sum = extractvalue {i32, i1} %res, 0
8171 %obit = extractvalue {i32, i1} %res, 1
8172 br i1 %obit, label %overflow, label %normal
8174 '``llvm.umul.with.overflow.*``' Intrinsics
8175 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8180 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8181 on any integer bit width.
8185 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8186 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8187 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8192 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8193 a unsigned multiplication of the two arguments, and indicate whether an
8194 overflow occurred during the unsigned multiplication.
8199 The arguments (%a and %b) and the first element of the result structure
8200 may be of integer types of any bit width, but they must have the same
8201 bit width. The second element of the result structure must be of type
8202 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8208 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8209 an unsigned multiplication of the two arguments. They return a structure ---
8210 the first element of which is the multiplication, and the second
8211 element of which is a bit specifying if the unsigned multiplication
8212 resulted in an overflow.
8217 .. code-block:: llvm
8219 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8220 %sum = extractvalue {i32, i1} %res, 0
8221 %obit = extractvalue {i32, i1} %res, 1
8222 br i1 %obit, label %overflow, label %normal
8224 Specialised Arithmetic Intrinsics
8225 ---------------------------------
8227 '``llvm.fmuladd.*``' Intrinsic
8228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8235 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8236 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8241 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8242 expressions that can be fused if the code generator determines that (a) the
8243 target instruction set has support for a fused operation, and (b) that the
8244 fused operation is more efficient than the equivalent, separate pair of mul
8245 and add instructions.
8250 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8251 multiplicands, a and b, and an addend c.
8260 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8262 is equivalent to the expression a \* b + c, except that rounding will
8263 not be performed between the multiplication and addition steps if the
8264 code generator fuses the operations. Fusion is not guaranteed, even if
8265 the target platform supports it. If a fused multiply-add is required the
8266 corresponding llvm.fma.\* intrinsic function should be used instead.
8271 .. code-block:: llvm
8273 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8275 Half Precision Floating Point Intrinsics
8276 ----------------------------------------
8278 For most target platforms, half precision floating point is a
8279 storage-only format. This means that it is a dense encoding (in memory)
8280 but does not support computation in the format.
8282 This means that code must first load the half-precision floating point
8283 value as an i16, then convert it to float with
8284 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8285 then be performed on the float value (including extending to double
8286 etc). To store the value back to memory, it is first converted to float
8287 if needed, then converted to i16 with
8288 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8291 .. _int_convert_to_fp16:
8293 '``llvm.convert.to.fp16``' Intrinsic
8294 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8301 declare i16 @llvm.convert.to.fp16(f32 %a)
8306 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8307 from single precision floating point format to half precision floating
8313 The intrinsic function contains single argument - the value to be
8319 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8320 from single precision floating point format to half precision floating
8321 point format. The return value is an ``i16`` which contains the
8327 .. code-block:: llvm
8329 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8330 store i16 %res, i16* @x, align 2
8332 .. _int_convert_from_fp16:
8334 '``llvm.convert.from.fp16``' Intrinsic
8335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8342 declare f32 @llvm.convert.from.fp16(i16 %a)
8347 The '``llvm.convert.from.fp16``' intrinsic function performs a
8348 conversion from half precision floating point format to single precision
8349 floating point format.
8354 The intrinsic function contains single argument - the value to be
8360 The '``llvm.convert.from.fp16``' intrinsic function performs a
8361 conversion from half single precision floating point format to single
8362 precision floating point format. The input half-float value is
8363 represented by an ``i16`` value.
8368 .. code-block:: llvm
8370 %a = load i16* @x, align 2
8371 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8376 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8377 prefix), are described in the `LLVM Source Level
8378 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8381 Exception Handling Intrinsics
8382 -----------------------------
8384 The LLVM exception handling intrinsics (which all start with
8385 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8386 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8390 Trampoline Intrinsics
8391 ---------------------
8393 These intrinsics make it possible to excise one parameter, marked with
8394 the :ref:`nest <nest>` attribute, from a function. The result is a
8395 callable function pointer lacking the nest parameter - the caller does
8396 not need to provide a value for it. Instead, the value to use is stored
8397 in advance in a "trampoline", a block of memory usually allocated on the
8398 stack, which also contains code to splice the nest value into the
8399 argument list. This is used to implement the GCC nested function address
8402 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8403 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8404 It can be created as follows:
8406 .. code-block:: llvm
8408 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8409 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8410 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8411 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8412 %fp = bitcast i8* %p to i32 (i32, i32)*
8414 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8415 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8419 '``llvm.init.trampoline``' Intrinsic
8420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8427 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8432 This fills the memory pointed to by ``tramp`` with executable code,
8433 turning it into a trampoline.
8438 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8439 pointers. The ``tramp`` argument must point to a sufficiently large and
8440 sufficiently aligned block of memory; this memory is written to by the
8441 intrinsic. Note that the size and the alignment are target-specific -
8442 LLVM currently provides no portable way of determining them, so a
8443 front-end that generates this intrinsic needs to have some
8444 target-specific knowledge. The ``func`` argument must hold a function
8445 bitcast to an ``i8*``.
8450 The block of memory pointed to by ``tramp`` is filled with target
8451 dependent code, turning it into a function. Then ``tramp`` needs to be
8452 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8453 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8454 function's signature is the same as that of ``func`` with any arguments
8455 marked with the ``nest`` attribute removed. At most one such ``nest``
8456 argument is allowed, and it must be of pointer type. Calling the new
8457 function is equivalent to calling ``func`` with the same argument list,
8458 but with ``nval`` used for the missing ``nest`` argument. If, after
8459 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8460 modified, then the effect of any later call to the returned function
8461 pointer is undefined.
8465 '``llvm.adjust.trampoline``' Intrinsic
8466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8473 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8478 This performs any required machine-specific adjustment to the address of
8479 a trampoline (passed as ``tramp``).
8484 ``tramp`` must point to a block of memory which already has trampoline
8485 code filled in by a previous call to
8486 :ref:`llvm.init.trampoline <int_it>`.
8491 On some architectures the address of the code to be executed needs to be
8492 different to the address where the trampoline is actually stored. This
8493 intrinsic returns the executable address corresponding to ``tramp``
8494 after performing the required machine specific adjustments. The pointer
8495 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8500 This class of intrinsics exists to information about the lifetime of
8501 memory objects and ranges where variables are immutable.
8503 '``llvm.lifetime.start``' Intrinsic
8504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8511 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8516 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8522 The first argument is a constant integer representing the size of the
8523 object, or -1 if it is variable sized. The second argument is a pointer
8529 This intrinsic indicates that before this point in the code, the value
8530 of the memory pointed to by ``ptr`` is dead. This means that it is known
8531 to never be used and has an undefined value. A load from the pointer
8532 that precedes this intrinsic can be replaced with ``'undef'``.
8534 '``llvm.lifetime.end``' Intrinsic
8535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8542 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8547 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8553 The first argument is a constant integer representing the size of the
8554 object, or -1 if it is variable sized. The second argument is a pointer
8560 This intrinsic indicates that after this point in the code, the value of
8561 the memory pointed to by ``ptr`` is dead. This means that it is known to
8562 never be used and has an undefined value. Any stores into the memory
8563 object following this intrinsic may be removed as dead.
8565 '``llvm.invariant.start``' Intrinsic
8566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8573 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8578 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8579 a memory object will not change.
8584 The first argument is a constant integer representing the size of the
8585 object, or -1 if it is variable sized. The second argument is a pointer
8591 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8592 the return value, the referenced memory location is constant and
8595 '``llvm.invariant.end``' Intrinsic
8596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8603 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8608 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8609 memory object are mutable.
8614 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8615 The second argument is a constant integer representing the size of the
8616 object, or -1 if it is variable sized and the third argument is a
8617 pointer to the object.
8622 This intrinsic indicates that the memory is mutable again.
8627 This class of intrinsics is designed to be generic and has no specific
8630 '``llvm.var.annotation``' Intrinsic
8631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8638 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8643 The '``llvm.var.annotation``' intrinsic.
8648 The first argument is a pointer to a value, the second is a pointer to a
8649 global string, the third is a pointer to a global string which is the
8650 source file name, and the last argument is the line number.
8655 This intrinsic allows annotation of local variables with arbitrary
8656 strings. This can be useful for special purpose optimizations that want
8657 to look for these annotations. These have no other defined use; they are
8658 ignored by code generation and optimization.
8660 '``llvm.ptr.annotation.*``' Intrinsic
8661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8666 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8667 pointer to an integer of any width. *NOTE* you must specify an address space for
8668 the pointer. The identifier for the default address space is the integer
8673 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8674 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8675 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8676 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8677 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8682 The '``llvm.ptr.annotation``' intrinsic.
8687 The first argument is a pointer to an integer value of arbitrary bitwidth
8688 (result of some expression), the second is a pointer to a global string, the
8689 third is a pointer to a global string which is the source file name, and the
8690 last argument is the line number. It returns the value of the first argument.
8695 This intrinsic allows annotation of a pointer to an integer with arbitrary
8696 strings. This can be useful for special purpose optimizations that want to look
8697 for these annotations. These have no other defined use; they are ignored by code
8698 generation and optimization.
8700 '``llvm.annotation.*``' Intrinsic
8701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8706 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8707 any integer bit width.
8711 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8712 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8713 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8714 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8715 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8720 The '``llvm.annotation``' intrinsic.
8725 The first argument is an integer value (result of some expression), the
8726 second is a pointer to a global string, the third is a pointer to a
8727 global string which is the source file name, and the last argument is
8728 the line number. It returns the value of the first argument.
8733 This intrinsic allows annotations to be put on arbitrary expressions
8734 with arbitrary strings. This can be useful for special purpose
8735 optimizations that want to look for these annotations. These have no
8736 other defined use; they are ignored by code generation and optimization.
8738 '``llvm.trap``' Intrinsic
8739 ^^^^^^^^^^^^^^^^^^^^^^^^^
8746 declare void @llvm.trap() noreturn nounwind
8751 The '``llvm.trap``' intrinsic.
8761 This intrinsic is lowered to the target dependent trap instruction. If
8762 the target does not have a trap instruction, this intrinsic will be
8763 lowered to a call of the ``abort()`` function.
8765 '``llvm.debugtrap``' Intrinsic
8766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8773 declare void @llvm.debugtrap() nounwind
8778 The '``llvm.debugtrap``' intrinsic.
8788 This intrinsic is lowered to code which is intended to cause an
8789 execution trap with the intention of requesting the attention of a
8792 '``llvm.stackprotector``' Intrinsic
8793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8800 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8805 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8806 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8807 is placed on the stack before local variables.
8812 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8813 The first argument is the value loaded from the stack guard
8814 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8815 enough space to hold the value of the guard.
8820 This intrinsic causes the prologue/epilogue inserter to force the position of
8821 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8822 to ensure that if a local variable on the stack is overwritten, it will destroy
8823 the value of the guard. When the function exits, the guard on the stack is
8824 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8825 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8826 calling the ``__stack_chk_fail()`` function.
8828 '``llvm.stackprotectorcheck``' Intrinsic
8829 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8836 declare void @llvm.stackprotectorcheck(i8** <guard>)
8841 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8842 created stack protector and if they are not equal calls the
8843 ``__stack_chk_fail()`` function.
8848 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8849 the variable ``@__stack_chk_guard``.
8854 This intrinsic is provided to perform the stack protector check by comparing
8855 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8856 values do not match call the ``__stack_chk_fail()`` function.
8858 The reason to provide this as an IR level intrinsic instead of implementing it
8859 via other IR operations is that in order to perform this operation at the IR
8860 level without an intrinsic, one would need to create additional basic blocks to
8861 handle the success/failure cases. This makes it difficult to stop the stack
8862 protector check from disrupting sibling tail calls in Codegen. With this
8863 intrinsic, we are able to generate the stack protector basic blocks late in
8864 codegen after the tail call decision has occurred.
8866 '``llvm.objectsize``' Intrinsic
8867 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8874 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8875 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8880 The ``llvm.objectsize`` intrinsic is designed to provide information to
8881 the optimizers to determine at compile time whether a) an operation
8882 (like memcpy) will overflow a buffer that corresponds to an object, or
8883 b) that a runtime check for overflow isn't necessary. An object in this
8884 context means an allocation of a specific class, structure, array, or
8890 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8891 argument is a pointer to or into the ``object``. The second argument is
8892 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8893 or -1 (if false) when the object size is unknown. The second argument
8894 only accepts constants.
8899 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8900 the size of the object concerned. If the size cannot be determined at
8901 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8902 on the ``min`` argument).
8904 '``llvm.expect``' Intrinsic
8905 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8912 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8913 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8918 The ``llvm.expect`` intrinsic provides information about expected (the
8919 most probable) value of ``val``, which can be used by optimizers.
8924 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8925 a value. The second argument is an expected value, this needs to be a
8926 constant value, variables are not allowed.
8931 This intrinsic is lowered to the ``val``.
8933 '``llvm.donothing``' Intrinsic
8934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8941 declare void @llvm.donothing() nounwind readnone
8946 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8947 only intrinsic that can be called with an invoke instruction.
8957 This intrinsic does nothing, and it's removed by optimizers and ignored