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.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
313 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
314 :ref:`invokes <i_invoke>` can all have an optional calling convention
315 specified for the call. The calling convention of any pair of dynamic
316 caller/callee must match, or the behavior of the program is undefined.
317 The following calling conventions are supported by LLVM, and more may be
320 "``ccc``" - The C calling convention
321 This calling convention (the default if no other calling convention
322 is specified) matches the target C calling conventions. This calling
323 convention supports varargs function calls and tolerates some
324 mismatch in the declared prototype and implemented declaration of
325 the function (as does normal C).
326 "``fastcc``" - The fast calling convention
327 This calling convention attempts to make calls as fast as possible
328 (e.g. by passing things in registers). This calling convention
329 allows the target to use whatever tricks it wants to produce fast
330 code for the target, without having to conform to an externally
331 specified ABI (Application Binary Interface). `Tail calls can only
332 be optimized when this, the GHC or the HiPE convention is
333 used. <CodeGenerator.html#id80>`_ This calling convention does not
334 support varargs and requires the prototype of all callees to exactly
335 match the prototype of the function definition.
336 "``coldcc``" - The cold calling convention
337 This calling convention attempts to make code in the caller as
338 efficient as possible under the assumption that the call is not
339 commonly executed. As such, these calls often preserve all registers
340 so that the call does not break any live ranges in the caller side.
341 This calling convention does not support varargs and requires the
342 prototype of all callees to exactly match the prototype of the
344 "``cc 10``" - GHC convention
345 This calling convention has been implemented specifically for use by
346 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
347 It passes everything in registers, going to extremes to achieve this
348 by disabling callee save registers. This calling convention should
349 not be used lightly but only for specific situations such as an
350 alternative to the *register pinning* performance technique often
351 used when implementing functional programming languages. At the
352 moment only X86 supports this convention and it has the following
355 - On *X86-32* only supports up to 4 bit type parameters. No
356 floating point types are supported.
357 - On *X86-64* only supports up to 10 bit type parameters and 6
358 floating point parameters.
360 This calling convention supports `tail call
361 optimization <CodeGenerator.html#id80>`_ but requires both the
362 caller and callee are using it.
363 "``cc 11``" - The HiPE calling convention
364 This calling convention has been implemented specifically for use by
365 the `High-Performance Erlang
366 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
367 native code compiler of the `Ericsson's Open Source Erlang/OTP
368 system <http://www.erlang.org/download.shtml>`_. It uses more
369 registers for argument passing than the ordinary C calling
370 convention and defines no callee-saved registers. The calling
371 convention properly supports `tail call
372 optimization <CodeGenerator.html#id80>`_ but requires that both the
373 caller and the callee use it. It uses a *register pinning*
374 mechanism, similar to GHC's convention, for keeping frequently
375 accessed runtime components pinned to specific hardware registers.
376 At the moment only X86 supports this convention (both 32 and 64
378 "``cc <n>``" - Numbered convention
379 Any calling convention may be specified by number, allowing
380 target-specific calling conventions to be used. Target specific
381 calling conventions start at 64.
383 More calling conventions can be added/defined on an as-needed basis, to
384 support Pascal conventions or any other well-known target-independent
387 .. _visibilitystyles:
392 All Global Variables and Functions have one of the following visibility
395 "``default``" - Default style
396 On targets that use the ELF object file format, default visibility
397 means that the declaration is visible to other modules and, in
398 shared libraries, means that the declared entity may be overridden.
399 On Darwin, default visibility means that the declaration is visible
400 to other modules. Default visibility corresponds to "external
401 linkage" in the language.
402 "``hidden``" - Hidden style
403 Two declarations of an object with hidden visibility refer to the
404 same object if they are in the same shared object. Usually, hidden
405 visibility indicates that the symbol will not be placed into the
406 dynamic symbol table, so no other module (executable or shared
407 library) can reference it directly.
408 "``protected``" - Protected style
409 On ELF, protected visibility indicates that the symbol will be
410 placed in the dynamic symbol table, but that references within the
411 defining module will bind to the local symbol. That is, the symbol
412 cannot be overridden by another module.
419 LLVM IR allows you to specify name aliases for certain types. This can
420 make it easier to read the IR and make the IR more condensed
421 (particularly when recursive types are involved). An example of a name
426 %mytype = type { %mytype*, i32 }
428 You may give a name to any :ref:`type <typesystem>` except
429 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
430 expected with the syntax "%mytype".
432 Note that type names are aliases for the structural type that they
433 indicate, and that you can therefore specify multiple names for the same
434 type. This often leads to confusing behavior when dumping out a .ll
435 file. Since LLVM IR uses structural typing, the name is not part of the
436 type. When printing out LLVM IR, the printer will pick *one name* to
437 render all types of a particular shape. This means that if you have code
438 where two different source types end up having the same LLVM type, that
439 the dumper will sometimes print the "wrong" or unexpected type. This is
440 an important design point and isn't going to change.
447 Global variables define regions of memory allocated at compilation time
448 instead of run-time. Global variables may optionally be initialized, may
449 have an explicit section to be placed in, and may have an optional
450 explicit alignment specified.
452 A variable may be defined as ``thread_local``, which means that it will
453 not be shared by threads (each thread will have a separated copy of the
454 variable). Not all targets support thread-local variables. Optionally, a
455 TLS model may be specified:
458 For variables that are only used within the current shared library.
460 For variables in modules that will not be loaded dynamically.
462 For variables defined in the executable and only used within it.
464 The models correspond to the ELF TLS models; see `ELF Handling For
465 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
466 more information on under which circumstances the different models may
467 be used. The target may choose a different TLS model if the specified
468 model is not supported, or if a better choice of model can be made.
470 A variable may be defined as a global ``constant``, which indicates that
471 the contents of the variable will **never** be modified (enabling better
472 optimization, allowing the global data to be placed in the read-only
473 section of an executable, etc). Note that variables that need runtime
474 initialization cannot be marked ``constant`` as there is a store to the
477 LLVM explicitly allows *declarations* of global variables to be marked
478 constant, even if the final definition of the global is not. This
479 capability can be used to enable slightly better optimization of the
480 program, but requires the language definition to guarantee that
481 optimizations based on the 'constantness' are valid for the translation
482 units that do not include the definition.
484 As SSA values, global variables define pointer values that are in scope
485 (i.e. they dominate) all basic blocks in the program. Global variables
486 always define a pointer to their "content" type because they describe a
487 region of memory, and all memory objects in LLVM are accessed through
490 Global variables can be marked with ``unnamed_addr`` which indicates
491 that the address is not significant, only the content. Constants marked
492 like this can be merged with other constants if they have the same
493 initializer. Note that a constant with significant address *can* be
494 merged with a ``unnamed_addr`` constant, the result being a constant
495 whose address is significant.
497 A global variable may be declared to reside in a target-specific
498 numbered address space. For targets that support them, address spaces
499 may affect how optimizations are performed and/or what target
500 instructions are used to access the variable. The default address space
501 is zero. The address space qualifier must precede any other attributes.
503 LLVM allows an explicit section to be specified for globals. If the
504 target supports it, it will emit globals to the section specified.
506 By default, global initializers are optimized by assuming that global
507 variables defined within the module are not modified from their
508 initial values before the start of the global initializer. This is
509 true even for variables potentially accessible from outside the
510 module, including those with external linkage or appearing in
511 ``@llvm.used``. This assumption may be suppressed by marking the
512 variable with ``externally_initialized``.
514 An explicit alignment may be specified for a global, which must be a
515 power of 2. If not present, or if the alignment is set to zero, the
516 alignment of the global is set by the target to whatever it feels
517 convenient. If an explicit alignment is specified, the global is forced
518 to have exactly that alignment. Targets and optimizers are not allowed
519 to over-align the global if the global has an assigned section. In this
520 case, the extra alignment could be observable: for example, code could
521 assume that the globals are densely packed in their section and try to
522 iterate over them as an array, alignment padding would break this
525 For example, the following defines a global in a numbered address space
526 with an initializer, section, and alignment:
530 @G = addrspace(5) constant float 1.0, section "foo", align 4
532 The following example defines a thread-local global with the
533 ``initialexec`` TLS model:
537 @G = thread_local(initialexec) global i32 0, align 4
539 .. _functionstructure:
544 LLVM function definitions consist of the "``define``" keyword, an
545 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
546 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
547 an optional ``unnamed_addr`` attribute, a return type, an optional
548 :ref:`parameter attribute <paramattrs>` for the return type, a function
549 name, a (possibly empty) argument list (each with optional :ref:`parameter
550 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
551 an optional section, an optional alignment, an optional :ref:`garbage
552 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
553 curly brace, a list of basic blocks, and a closing curly brace.
555 LLVM function declarations consist of the "``declare``" keyword, an
556 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
557 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
558 an optional ``unnamed_addr`` attribute, a return type, an optional
559 :ref:`parameter attribute <paramattrs>` for the return type, a function
560 name, a possibly empty list of arguments, an optional alignment, an optional
561 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
563 A function definition contains a list of basic blocks, forming the CFG
564 (Control Flow Graph) for the function. Each basic block may optionally
565 start with a label (giving the basic block a symbol table entry),
566 contains a list of instructions, and ends with a
567 :ref:`terminator <terminators>` instruction (such as a branch or function
568 return). If explicit label is not provided, a block is assigned an
569 implicit numbered label, using a next value from the same counter as used
570 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
571 function entry block does not have explicit label, it will be assigned
572 label "%0", then first unnamed temporary in that block will be "%1", etc.
574 The first basic block in a function is special in two ways: it is
575 immediately executed on entrance to the function, and it is not allowed
576 to have predecessor basic blocks (i.e. there can not be any branches to
577 the entry block of a function). Because the block can have no
578 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
580 LLVM allows an explicit section to be specified for functions. If the
581 target supports it, it will emit functions to the section specified.
583 An explicit alignment may be specified for a function. If not present,
584 or if the alignment is set to zero, the alignment of the function is set
585 by the target to whatever it feels convenient. If an explicit alignment
586 is specified, the function is forced to have at least that much
587 alignment. All alignments must be a power of 2.
589 If the ``unnamed_addr`` attribute is given, the address is know to not
590 be significant and two identical functions can be merged.
594 define [linkage] [visibility]
596 <ResultType> @<FunctionName> ([argument list])
597 [fn Attrs] [section "name"] [align N]
598 [gc] [prefix Constant] { ... }
605 Aliases act as "second name" for the aliasee value (which can be either
606 function, global variable, another alias or bitcast of global value).
607 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
608 :ref:`visibility style <visibility>`.
612 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
614 The linkage must be one of ``private``, ``linker_private``,
615 ``linker_private_weak``, ``internal``, ``linkonce``, ``weak``,
616 ``linkonce_odr``, ``weak_odr``, ``linkonce_odr_auto_hide``, ``external``. Note
617 that some system linkers might not correctly handle dropping a weak symbol that
618 is aliased by a non weak alias.
620 .. _namedmetadatastructure:
625 Named metadata is a collection of metadata. :ref:`Metadata
626 nodes <metadata>` (but not metadata strings) are the only valid
627 operands for a named metadata.
631 ; Some unnamed metadata nodes, which are referenced by the named metadata.
632 !0 = metadata !{metadata !"zero"}
633 !1 = metadata !{metadata !"one"}
634 !2 = metadata !{metadata !"two"}
636 !name = !{!0, !1, !2}
643 The return type and each parameter of a function type may have a set of
644 *parameter attributes* associated with them. Parameter attributes are
645 used to communicate additional information about the result or
646 parameters of a function. Parameter attributes are considered to be part
647 of the function, not of the function type, so functions with different
648 parameter attributes can have the same function type.
650 Parameter attributes are simple keywords that follow the type specified.
651 If multiple parameter attributes are needed, they are space separated.
656 declare i32 @printf(i8* noalias nocapture, ...)
657 declare i32 @atoi(i8 zeroext)
658 declare signext i8 @returns_signed_char()
660 Note that any attributes for the function result (``nounwind``,
661 ``readonly``) come immediately after the argument list.
663 Currently, only the following parameter attributes are defined:
666 This indicates to the code generator that the parameter or return
667 value should be zero-extended to the extent required by the target's
668 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
669 the caller (for a parameter) or the callee (for a return value).
671 This indicates to the code generator that the parameter or return
672 value should be sign-extended to the extent required by the target's
673 ABI (which is usually 32-bits) by the caller (for a parameter) or
674 the callee (for a return value).
676 This indicates that this parameter or return value should be treated
677 in a special target-dependent fashion during while emitting code for
678 a function call or return (usually, by putting it in a register as
679 opposed to memory, though some targets use it to distinguish between
680 two different kinds of registers). Use of this attribute is
683 This indicates that the pointer parameter should really be passed by
684 value to the function. The attribute implies that a hidden copy of
685 the pointee is made between the caller and the callee, so the callee
686 is unable to modify the value in the caller. This attribute is only
687 valid on LLVM pointer arguments. It is generally used to pass
688 structs and arrays by value, but is also valid on pointers to
689 scalars. The copy is considered to belong to the caller not the
690 callee (for example, ``readonly`` functions should not write to
691 ``byval`` parameters). This is not a valid attribute for return
694 The byval attribute also supports specifying an alignment with the
695 align attribute. It indicates the alignment of the stack slot to
696 form and the known alignment of the pointer specified to the call
697 site. If the alignment is not specified, then the code generator
698 makes a target-specific assumption.
701 This indicates that the pointer parameter specifies the address of a
702 structure that is the return value of the function in the source
703 program. This pointer must be guaranteed by the caller to be valid:
704 loads and stores to the structure may be assumed by the callee
705 not to trap and to be properly aligned. This may only be applied to
706 the first parameter. This is not a valid attribute for return
709 This indicates that pointer values :ref:`based <pointeraliasing>` on
710 the argument or return value do not alias pointer values which are
711 not *based* on it, ignoring certain "irrelevant" dependencies. For a
712 call to the parent function, dependencies between memory references
713 from before or after the call and from those during the call are
714 "irrelevant" to the ``noalias`` keyword for the arguments and return
715 value used in that call. The caller shares the responsibility with
716 the callee for ensuring that these requirements are met. For further
717 details, please see the discussion of the NoAlias response in `alias
718 analysis <AliasAnalysis.html#MustMayNo>`_.
720 Note that this definition of ``noalias`` is intentionally similar
721 to the definition of ``restrict`` in C99 for function arguments,
722 though it is slightly weaker.
724 For function return values, C99's ``restrict`` is not meaningful,
725 while LLVM's ``noalias`` is.
727 This indicates that the callee does not make any copies of the
728 pointer that outlive the callee itself. This is not a valid
729 attribute for return values.
734 This indicates that the pointer parameter can be excised using the
735 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
736 attribute for return values and can only be applied to one parameter.
739 This indicates that the function always returns the argument as its return
740 value. This is an optimization hint to the code generator when generating
741 the caller, allowing tail call optimization and omission of register saves
742 and restores in some cases; it is not checked or enforced when generating
743 the callee. The parameter and the function return type must be valid
744 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
745 valid attribute for return values and can only be applied to one parameter.
749 Garbage Collector Names
750 -----------------------
752 Each function may specify a garbage collector name, which is simply a
757 define void @f() gc "name" { ... }
759 The compiler declares the supported values of *name*. Specifying a
760 collector which will cause the compiler to alter its output in order to
761 support the named garbage collection algorithm.
768 Prefix data is data associated with a function which the code generator
769 will emit immediately before the function body. The purpose of this feature
770 is to allow frontends to associate language-specific runtime metadata with
771 specific functions and make it available through the function pointer while
772 still allowing the function pointer to be called. To access the data for a
773 given function, a program may bitcast the function pointer to a pointer to
774 the constant's type. This implies that the IR symbol points to the start
777 To maintain the semantics of ordinary function calls, the prefix data must
778 have a particular format. Specifically, it must begin with a sequence of
779 bytes which decode to a sequence of machine instructions, valid for the
780 module's target, which transfer control to the point immediately succeeding
781 the prefix data, without performing any other visible action. This allows
782 the inliner and other passes to reason about the semantics of the function
783 definition without needing to reason about the prefix data. Obviously this
784 makes the format of the prefix data highly target dependent.
786 Prefix data is laid out as if it were an initializer for a global variable
787 of the prefix data's type. No padding is automatically placed between the
788 prefix data and the function body. If padding is required, it must be part
791 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
792 which encodes the ``nop`` instruction:
796 define void @f() prefix i8 144 { ... }
798 Generally prefix data can be formed by encoding a relative branch instruction
799 which skips the metadata, as in this example of valid prefix data for the
800 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
804 %0 = type <{ i8, i8, i8* }>
806 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
808 A function may have prefix data but no body. This has similar semantics
809 to the ``available_externally`` linkage in that the data may be used by the
810 optimizers but will not be emitted in the object file.
817 Attribute groups are groups of attributes that are referenced by objects within
818 the IR. They are important for keeping ``.ll`` files readable, because a lot of
819 functions will use the same set of attributes. In the degenerative case of a
820 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
821 group will capture the important command line flags used to build that file.
823 An attribute group is a module-level object. To use an attribute group, an
824 object references the attribute group's ID (e.g. ``#37``). An object may refer
825 to more than one attribute group. In that situation, the attributes from the
826 different groups are merged.
828 Here is an example of attribute groups for a function that should always be
829 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
833 ; Target-independent attributes:
834 attributes #0 = { alwaysinline alignstack=4 }
836 ; Target-dependent attributes:
837 attributes #1 = { "no-sse" }
839 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
840 define void @f() #0 #1 { ... }
847 Function attributes are set to communicate additional information about
848 a function. Function attributes are considered to be part of the
849 function, not of the function type, so functions with different function
850 attributes can have the same function type.
852 Function attributes are simple keywords that follow the type specified.
853 If multiple attributes are needed, they are space separated. For
858 define void @f() noinline { ... }
859 define void @f() alwaysinline { ... }
860 define void @f() alwaysinline optsize { ... }
861 define void @f() optsize { ... }
864 This attribute indicates that, when emitting the prologue and
865 epilogue, the backend should forcibly align the stack pointer.
866 Specify the desired alignment, which must be a power of two, in
869 This attribute indicates that the inliner should attempt to inline
870 this function into callers whenever possible, ignoring any active
871 inlining size threshold for this caller.
873 This indicates that the callee function at a call site should be
874 recognized as a built-in function, even though the function's declaration
875 uses the ``nobuiltin`` attribute. This is only valid at call sites for
876 direct calls to functions which are declared with the ``nobuiltin``
879 This attribute indicates that this function is rarely called. When
880 computing edge weights, basic blocks post-dominated by a cold
881 function call are also considered to be cold; and, thus, given low
884 This attribute indicates that the source code contained a hint that
885 inlining this function is desirable (such as the "inline" keyword in
886 C/C++). It is just a hint; it imposes no requirements on the
889 This attribute suggests that optimization passes and code generator
890 passes make choices that keep the code size of this function as small
891 as possible and perform optimizations that may sacrifice runtime
892 performance in order to minimize the size of the generated code.
894 This attribute disables prologue / epilogue emission for the
895 function. This can have very system-specific consequences.
897 This indicates that the callee function at a call site is not recognized as
898 a built-in function. LLVM will retain the original call and not replace it
899 with equivalent code based on the semantics of the built-in function, unless
900 the call site uses the ``builtin`` attribute. This is valid at call sites
901 and on function declarations and definitions.
903 This attribute indicates that calls to the function cannot be
904 duplicated. A call to a ``noduplicate`` function may be moved
905 within its parent function, but may not be duplicated within
908 A function containing a ``noduplicate`` call may still
909 be an inlining candidate, provided that the call is not
910 duplicated by inlining. That implies that the function has
911 internal linkage and only has one call site, so the original
912 call is dead after inlining.
914 This attributes disables implicit floating point instructions.
916 This attribute indicates that the inliner should never inline this
917 function in any situation. This attribute may not be used together
918 with the ``alwaysinline`` attribute.
920 This attribute suppresses lazy symbol binding for the function. This
921 may make calls to the function faster, at the cost of extra program
922 startup time if the function is not called during program startup.
924 This attribute indicates that the code generator should not use a
925 red zone, even if the target-specific ABI normally permits it.
927 This function attribute indicates that the function never returns
928 normally. This produces undefined behavior at runtime if the
929 function ever does dynamically return.
931 This function attribute indicates that the function never returns
932 with an unwind or exceptional control flow. If the function does
933 unwind, its runtime behavior is undefined.
935 This function attribute indicates that the function is not optimized
936 by any optimization or code generator passes with the
937 exception of interprocedural optimization passes.
938 This attribute cannot be used together with the ``alwaysinline``
939 attribute; this attribute is also incompatible
940 with the ``minsize`` attribute and the ``optsize`` attribute.
942 The inliner should never inline this function in any situation.
943 Only functions with the ``alwaysinline`` attribute are valid
944 candidates for inlining inside the body of this function.
946 This attribute suggests that optimization passes and code generator
947 passes make choices that keep the code size of this function low,
948 and otherwise do optimizations specifically to reduce code size as
949 long as they do not significantly impact runtime performance.
951 On a function, this attribute indicates that the function computes its
952 result (or decides to unwind an exception) based strictly on its arguments,
953 without dereferencing any pointer arguments or otherwise accessing
954 any mutable state (e.g. memory, control registers, etc) visible to
955 caller functions. It does not write through any pointer arguments
956 (including ``byval`` arguments) and never changes any state visible
957 to callers. This means that it cannot unwind exceptions by calling
958 the ``C++`` exception throwing methods.
960 On an argument, this attribute indicates that the function does not
961 dereference that pointer argument, even though it may read or write the
962 memory that the pointer points to if accessed through other pointers.
964 On a function, this attribute indicates that the function does not write
965 through any pointer arguments (including ``byval`` arguments) or otherwise
966 modify any state (e.g. memory, control registers, etc) visible to
967 caller functions. It may dereference pointer arguments and read
968 state that may be set in the caller. A readonly function always
969 returns the same value (or unwinds an exception identically) when
970 called with the same set of arguments and global state. It cannot
971 unwind an exception by calling the ``C++`` exception throwing
974 On an argument, this attribute indicates that the function does not write
975 through this pointer argument, even though it may write to the memory that
976 the pointer points to.
978 This attribute indicates that this function can return twice. The C
979 ``setjmp`` is an example of such a function. The compiler disables
980 some optimizations (like tail calls) in the caller of these
983 This attribute indicates that AddressSanitizer checks
984 (dynamic address safety analysis) are enabled for this function.
986 This attribute indicates that MemorySanitizer checks (dynamic detection
987 of accesses to uninitialized memory) are enabled for this function.
989 This attribute indicates that ThreadSanitizer checks
990 (dynamic thread safety analysis) are enabled for this function.
992 This attribute indicates that the function should emit a stack
993 smashing protector. It is in the form of a "canary" --- a random value
994 placed on the stack before the local variables that's checked upon
995 return from the function to see if it has been overwritten. A
996 heuristic is used to determine if a function needs stack protectors
997 or not. The heuristic used will enable protectors for functions with:
999 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1000 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1001 - Calls to alloca() with variable sizes or constant sizes greater than
1002 ``ssp-buffer-size``.
1004 If a function that has an ``ssp`` attribute is inlined into a
1005 function that doesn't have an ``ssp`` attribute, then the resulting
1006 function will have an ``ssp`` attribute.
1008 This attribute indicates that the function should *always* emit a
1009 stack smashing protector. This overrides the ``ssp`` function
1012 If a function that has an ``sspreq`` attribute is inlined into a
1013 function that doesn't have an ``sspreq`` attribute or which has an
1014 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1015 an ``sspreq`` attribute.
1017 This attribute indicates that the function should emit a stack smashing
1018 protector. This attribute causes a strong heuristic to be used when
1019 determining if a function needs stack protectors. The strong heuristic
1020 will enable protectors for functions with:
1022 - Arrays of any size and type
1023 - Aggregates containing an array of any size and type.
1024 - Calls to alloca().
1025 - Local variables that have had their address taken.
1027 This overrides the ``ssp`` function attribute.
1029 If a function that has an ``sspstrong`` attribute is inlined into a
1030 function that doesn't have an ``sspstrong`` attribute, then the
1031 resulting function will have an ``sspstrong`` attribute.
1033 This attribute indicates that the ABI being targeted requires that
1034 an unwind table entry be produce for this function even if we can
1035 show that no exceptions passes by it. This is normally the case for
1036 the ELF x86-64 abi, but it can be disabled for some compilation
1041 Module-Level Inline Assembly
1042 ----------------------------
1044 Modules may contain "module-level inline asm" blocks, which corresponds
1045 to the GCC "file scope inline asm" blocks. These blocks are internally
1046 concatenated by LLVM and treated as a single unit, but may be separated
1047 in the ``.ll`` file if desired. The syntax is very simple:
1049 .. code-block:: llvm
1051 module asm "inline asm code goes here"
1052 module asm "more can go here"
1054 The strings can contain any character by escaping non-printable
1055 characters. The escape sequence used is simply "\\xx" where "xx" is the
1056 two digit hex code for the number.
1058 The inline asm code is simply printed to the machine code .s file when
1059 assembly code is generated.
1061 .. _langref_datalayout:
1066 A module may specify a target specific data layout string that specifies
1067 how data is to be laid out in memory. The syntax for the data layout is
1070 .. code-block:: llvm
1072 target datalayout = "layout specification"
1074 The *layout specification* consists of a list of specifications
1075 separated by the minus sign character ('-'). Each specification starts
1076 with a letter and may include other information after the letter to
1077 define some aspect of the data layout. The specifications accepted are
1081 Specifies that the target lays out data in big-endian form. That is,
1082 the bits with the most significance have the lowest address
1085 Specifies that the target lays out data in little-endian form. That
1086 is, the bits with the least significance have the lowest address
1089 Specifies the natural alignment of the stack in bits. Alignment
1090 promotion of stack variables is limited to the natural stack
1091 alignment to avoid dynamic stack realignment. The stack alignment
1092 must be a multiple of 8-bits. If omitted, the natural stack
1093 alignment defaults to "unspecified", which does not prevent any
1094 alignment promotions.
1095 ``p[n]:<size>:<abi>:<pref>``
1096 This specifies the *size* of a pointer and its ``<abi>`` and
1097 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1098 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1099 preceding ``:`` should be omitted too. The address space, ``n`` is
1100 optional, and if not specified, denotes the default address space 0.
1101 The value of ``n`` must be in the range [1,2^23).
1102 ``i<size>:<abi>:<pref>``
1103 This specifies the alignment for an integer type of a given bit
1104 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1105 ``v<size>:<abi>:<pref>``
1106 This specifies the alignment for a vector type of a given bit
1108 ``f<size>:<abi>:<pref>``
1109 This specifies the alignment for a floating point type of a given bit
1110 ``<size>``. Only values of ``<size>`` that are supported by the target
1111 will work. 32 (float) and 64 (double) are supported on all targets; 80
1112 or 128 (different flavors of long double) are also supported on some
1114 ``a<size>:<abi>:<pref>``
1115 This specifies the alignment for an aggregate type of a given bit
1117 ``s<size>:<abi>:<pref>``
1118 This specifies the alignment for a stack object of a given bit
1120 ``n<size1>:<size2>:<size3>...``
1121 This specifies a set of native integer widths for the target CPU in
1122 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1123 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1124 this set are considered to support most general arithmetic operations
1127 When constructing the data layout for a given target, LLVM starts with a
1128 default set of specifications which are then (possibly) overridden by
1129 the specifications in the ``datalayout`` keyword. The default
1130 specifications are given in this list:
1132 - ``E`` - big endian
1133 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1134 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1135 same as the default address space.
1136 - ``S0`` - natural stack alignment is unspecified
1137 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1138 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1139 - ``i16:16:16`` - i16 is 16-bit aligned
1140 - ``i32:32:32`` - i32 is 32-bit aligned
1141 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1142 alignment of 64-bits
1143 - ``f16:16:16`` - half is 16-bit aligned
1144 - ``f32:32:32`` - float is 32-bit aligned
1145 - ``f64:64:64`` - double is 64-bit aligned
1146 - ``f128:128:128`` - quad is 128-bit aligned
1147 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1148 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1149 - ``a0:0:64`` - aggregates are 64-bit aligned
1151 When LLVM is determining the alignment for a given type, it uses the
1154 #. If the type sought is an exact match for one of the specifications,
1155 that specification is used.
1156 #. If no match is found, and the type sought is an integer type, then
1157 the smallest integer type that is larger than the bitwidth of the
1158 sought type is used. If none of the specifications are larger than
1159 the bitwidth then the largest integer type is used. For example,
1160 given the default specifications above, the i7 type will use the
1161 alignment of i8 (next largest) while both i65 and i256 will use the
1162 alignment of i64 (largest specified).
1163 #. If no match is found, and the type sought is a vector type, then the
1164 largest vector type that is smaller than the sought vector type will
1165 be used as a fall back. This happens because <128 x double> can be
1166 implemented in terms of 64 <2 x double>, for example.
1168 The function of the data layout string may not be what you expect.
1169 Notably, this is not a specification from the frontend of what alignment
1170 the code generator should use.
1172 Instead, if specified, the target data layout is required to match what
1173 the ultimate *code generator* expects. This string is used by the
1174 mid-level optimizers to improve code, and this only works if it matches
1175 what the ultimate code generator uses. If you would like to generate IR
1176 that does not embed this target-specific detail into the IR, then you
1177 don't have to specify the string. This will disable some optimizations
1178 that require precise layout information, but this also prevents those
1179 optimizations from introducing target specificity into the IR.
1186 A module may specify a target triple string that describes the target
1187 host. The syntax for the target triple is simply:
1189 .. code-block:: llvm
1191 target triple = "x86_64-apple-macosx10.7.0"
1193 The *target triple* string consists of a series of identifiers delimited
1194 by the minus sign character ('-'). The canonical forms are:
1198 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1199 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1201 This information is passed along to the backend so that it generates
1202 code for the proper architecture. It's possible to override this on the
1203 command line with the ``-mtriple`` command line option.
1205 .. _pointeraliasing:
1207 Pointer Aliasing Rules
1208 ----------------------
1210 Any memory access must be done through a pointer value associated with
1211 an address range of the memory access, otherwise the behavior is
1212 undefined. Pointer values are associated with address ranges according
1213 to the following rules:
1215 - A pointer value is associated with the addresses associated with any
1216 value it is *based* on.
1217 - An address of a global variable is associated with the address range
1218 of the variable's storage.
1219 - The result value of an allocation instruction is associated with the
1220 address range of the allocated storage.
1221 - A null pointer in the default address-space is associated with no
1223 - An integer constant other than zero or a pointer value returned from
1224 a function not defined within LLVM may be associated with address
1225 ranges allocated through mechanisms other than those provided by
1226 LLVM. Such ranges shall not overlap with any ranges of addresses
1227 allocated by mechanisms provided by LLVM.
1229 A pointer value is *based* on another pointer value according to the
1232 - A pointer value formed from a ``getelementptr`` operation is *based*
1233 on the first operand of the ``getelementptr``.
1234 - The result value of a ``bitcast`` is *based* on the operand of the
1236 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1237 values that contribute (directly or indirectly) to the computation of
1238 the pointer's value.
1239 - The "*based* on" relationship is transitive.
1241 Note that this definition of *"based"* is intentionally similar to the
1242 definition of *"based"* in C99, though it is slightly weaker.
1244 LLVM IR does not associate types with memory. The result type of a
1245 ``load`` merely indicates the size and alignment of the memory from
1246 which to load, as well as the interpretation of the value. The first
1247 operand type of a ``store`` similarly only indicates the size and
1248 alignment of the store.
1250 Consequently, type-based alias analysis, aka TBAA, aka
1251 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1252 :ref:`Metadata <metadata>` may be used to encode additional information
1253 which specialized optimization passes may use to implement type-based
1258 Volatile Memory Accesses
1259 ------------------------
1261 Certain memory accesses, such as :ref:`load <i_load>`'s,
1262 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1263 marked ``volatile``. The optimizers must not change the number of
1264 volatile operations or change their order of execution relative to other
1265 volatile operations. The optimizers *may* change the order of volatile
1266 operations relative to non-volatile operations. This is not Java's
1267 "volatile" and has no cross-thread synchronization behavior.
1269 IR-level volatile loads and stores cannot safely be optimized into
1270 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1271 flagged volatile. Likewise, the backend should never split or merge
1272 target-legal volatile load/store instructions.
1274 .. admonition:: Rationale
1276 Platforms may rely on volatile loads and stores of natively supported
1277 data width to be executed as single instruction. For example, in C
1278 this holds for an l-value of volatile primitive type with native
1279 hardware support, but not necessarily for aggregate types. The
1280 frontend upholds these expectations, which are intentionally
1281 unspecified in the IR. The rules above ensure that IR transformation
1282 do not violate the frontend's contract with the language.
1286 Memory Model for Concurrent Operations
1287 --------------------------------------
1289 The LLVM IR does not define any way to start parallel threads of
1290 execution or to register signal handlers. Nonetheless, there are
1291 platform-specific ways to create them, and we define LLVM IR's behavior
1292 in their presence. This model is inspired by the C++0x memory model.
1294 For a more informal introduction to this model, see the :doc:`Atomics`.
1296 We define a *happens-before* partial order as the least partial order
1299 - Is a superset of single-thread program order, and
1300 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1301 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1302 techniques, like pthread locks, thread creation, thread joining,
1303 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1304 Constraints <ordering>`).
1306 Note that program order does not introduce *happens-before* edges
1307 between a thread and signals executing inside that thread.
1309 Every (defined) read operation (load instructions, memcpy, atomic
1310 loads/read-modify-writes, etc.) R reads a series of bytes written by
1311 (defined) write operations (store instructions, atomic
1312 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1313 section, initialized globals are considered to have a write of the
1314 initializer which is atomic and happens before any other read or write
1315 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1316 may see any write to the same byte, except:
1318 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1319 write\ :sub:`2` happens before R\ :sub:`byte`, then
1320 R\ :sub:`byte` does not see write\ :sub:`1`.
1321 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1322 R\ :sub:`byte` does not see write\ :sub:`3`.
1324 Given that definition, R\ :sub:`byte` is defined as follows:
1326 - If R is volatile, the result is target-dependent. (Volatile is
1327 supposed to give guarantees which can support ``sig_atomic_t`` in
1328 C/C++, and may be used for accesses to addresses which do not behave
1329 like normal memory. It does not generally provide cross-thread
1331 - Otherwise, if there is no write to the same byte that happens before
1332 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1333 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1334 R\ :sub:`byte` returns the value written by that write.
1335 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1336 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1337 Memory Ordering Constraints <ordering>` section for additional
1338 constraints on how the choice is made.
1339 - Otherwise R\ :sub:`byte` returns ``undef``.
1341 R returns the value composed of the series of bytes it read. This
1342 implies that some bytes within the value may be ``undef`` **without**
1343 the entire value being ``undef``. Note that this only defines the
1344 semantics of the operation; it doesn't mean that targets will emit more
1345 than one instruction to read the series of bytes.
1347 Note that in cases where none of the atomic intrinsics are used, this
1348 model places only one restriction on IR transformations on top of what
1349 is required for single-threaded execution: introducing a store to a byte
1350 which might not otherwise be stored is not allowed in general.
1351 (Specifically, in the case where another thread might write to and read
1352 from an address, introducing a store can change a load that may see
1353 exactly one write into a load that may see multiple writes.)
1357 Atomic Memory Ordering Constraints
1358 ----------------------------------
1360 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1361 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1362 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1363 an ordering parameter that determines which other atomic instructions on
1364 the same address they *synchronize with*. These semantics are borrowed
1365 from Java and C++0x, but are somewhat more colloquial. If these
1366 descriptions aren't precise enough, check those specs (see spec
1367 references in the :doc:`atomics guide <Atomics>`).
1368 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1369 differently since they don't take an address. See that instruction's
1370 documentation for details.
1372 For a simpler introduction to the ordering constraints, see the
1376 The set of values that can be read is governed by the happens-before
1377 partial order. A value cannot be read unless some operation wrote
1378 it. This is intended to provide a guarantee strong enough to model
1379 Java's non-volatile shared variables. This ordering cannot be
1380 specified for read-modify-write operations; it is not strong enough
1381 to make them atomic in any interesting way.
1383 In addition to the guarantees of ``unordered``, there is a single
1384 total order for modifications by ``monotonic`` operations on each
1385 address. All modification orders must be compatible with the
1386 happens-before order. There is no guarantee that the modification
1387 orders can be combined to a global total order for the whole program
1388 (and this often will not be possible). The read in an atomic
1389 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1390 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1391 order immediately before the value it writes. If one atomic read
1392 happens before another atomic read of the same address, the later
1393 read must see the same value or a later value in the address's
1394 modification order. This disallows reordering of ``monotonic`` (or
1395 stronger) operations on the same address. If an address is written
1396 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1397 read that address repeatedly, the other threads must eventually see
1398 the write. This corresponds to the C++0x/C1x
1399 ``memory_order_relaxed``.
1401 In addition to the guarantees of ``monotonic``, a
1402 *synchronizes-with* edge may be formed with a ``release`` operation.
1403 This is intended to model C++'s ``memory_order_acquire``.
1405 In addition to the guarantees of ``monotonic``, if this operation
1406 writes a value which is subsequently read by an ``acquire``
1407 operation, it *synchronizes-with* that operation. (This isn't a
1408 complete description; see the C++0x definition of a release
1409 sequence.) This corresponds to the C++0x/C1x
1410 ``memory_order_release``.
1411 ``acq_rel`` (acquire+release)
1412 Acts as both an ``acquire`` and ``release`` operation on its
1413 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1414 ``seq_cst`` (sequentially consistent)
1415 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1416 operation which only reads, ``release`` for an operation which only
1417 writes), there is a global total order on all
1418 sequentially-consistent operations on all addresses, which is
1419 consistent with the *happens-before* partial order and with the
1420 modification orders of all the affected addresses. Each
1421 sequentially-consistent read sees the last preceding write to the
1422 same address in this global order. This corresponds to the C++0x/C1x
1423 ``memory_order_seq_cst`` and Java volatile.
1427 If an atomic operation is marked ``singlethread``, it only *synchronizes
1428 with* or participates in modification and seq\_cst total orderings with
1429 other operations running in the same thread (for example, in signal
1437 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1438 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1439 :ref:`frem <i_frem>`) have the following flags that can set to enable
1440 otherwise unsafe floating point operations
1443 No NaNs - Allow optimizations to assume the arguments and result are not
1444 NaN. Such optimizations are required to retain defined behavior over
1445 NaNs, but the value of the result is undefined.
1448 No Infs - Allow optimizations to assume the arguments and result are not
1449 +/-Inf. Such optimizations are required to retain defined behavior over
1450 +/-Inf, but the value of the result is undefined.
1453 No Signed Zeros - Allow optimizations to treat the sign of a zero
1454 argument or result as insignificant.
1457 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1458 argument rather than perform division.
1461 Fast - Allow algebraically equivalent transformations that may
1462 dramatically change results in floating point (e.g. reassociate). This
1463 flag implies all the others.
1470 The LLVM type system is one of the most important features of the
1471 intermediate representation. Being typed enables a number of
1472 optimizations to be performed on the intermediate representation
1473 directly, without having to do extra analyses on the side before the
1474 transformation. A strong type system makes it easier to read the
1475 generated code and enables novel analyses and transformations that are
1476 not feasible to perform on normal three address code representations.
1478 .. _typeclassifications:
1480 Type Classifications
1481 --------------------
1483 The types fall into a few useful classifications:
1492 * - :ref:`integer <t_integer>`
1493 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1496 * - :ref:`floating point <t_floating>`
1497 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1505 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1506 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1507 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1508 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1510 * - :ref:`primitive <t_primitive>`
1511 - :ref:`label <t_label>`,
1512 :ref:`void <t_void>`,
1513 :ref:`integer <t_integer>`,
1514 :ref:`floating point <t_floating>`,
1515 :ref:`x86mmx <t_x86mmx>`,
1516 :ref:`metadata <t_metadata>`.
1518 * - :ref:`derived <t_derived>`
1519 - :ref:`array <t_array>`,
1520 :ref:`function <t_function>`,
1521 :ref:`pointer <t_pointer>`,
1522 :ref:`structure <t_struct>`,
1523 :ref:`vector <t_vector>`,
1524 :ref:`opaque <t_opaque>`.
1526 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1527 Values of these types are the only ones which can be produced by
1535 The primitive types are the fundamental building blocks of the LLVM
1546 The integer type is a very simple type that simply specifies an
1547 arbitrary bit width for the integer type desired. Any bit width from 1
1548 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1557 The number of bits the integer will occupy is specified by the ``N``
1563 +----------------+------------------------------------------------+
1564 | ``i1`` | a single-bit integer. |
1565 +----------------+------------------------------------------------+
1566 | ``i32`` | a 32-bit integer. |
1567 +----------------+------------------------------------------------+
1568 | ``i1942652`` | a really big integer of over 1 million bits. |
1569 +----------------+------------------------------------------------+
1573 Floating Point Types
1574 ^^^^^^^^^^^^^^^^^^^^
1583 - 16-bit floating point value
1586 - 32-bit floating point value
1589 - 64-bit floating point value
1592 - 128-bit floating point value (112-bit mantissa)
1595 - 80-bit floating point value (X87)
1598 - 128-bit floating point value (two 64-bits)
1608 The x86mmx type represents a value held in an MMX register on an x86
1609 machine. The operations allowed on it are quite limited: parameters and
1610 return values, load and store, and bitcast. User-specified MMX
1611 instructions are represented as intrinsic or asm calls with arguments
1612 and/or results of this type. There are no arrays, vectors or constants
1630 The void type does not represent any value and has no size.
1647 The label type represents code labels.
1664 The metadata type represents embedded metadata. No derived types may be
1665 created from metadata except for :ref:`function <t_function>` arguments.
1679 The real power in LLVM comes from the derived types in the system. This
1680 is what allows a programmer to represent arrays, functions, pointers,
1681 and other useful types. Each of these types contain one or more element
1682 types which may be a primitive type, or another derived type. For
1683 example, it is possible to have a two dimensional array, using an array
1684 as the element type of another array.
1691 Aggregate Types are a subset of derived types that can contain multiple
1692 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1693 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1704 The array type is a very simple derived type that arranges elements
1705 sequentially in memory. The array type requires a size (number of
1706 elements) and an underlying data type.
1713 [<# elements> x <elementtype>]
1715 The number of elements is a constant integer value; ``elementtype`` may
1716 be any type with a size.
1721 +------------------+--------------------------------------+
1722 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1723 +------------------+--------------------------------------+
1724 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1725 +------------------+--------------------------------------+
1726 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1727 +------------------+--------------------------------------+
1729 Here are some examples of multidimensional arrays:
1731 +-----------------------------+----------------------------------------------------------+
1732 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1733 +-----------------------------+----------------------------------------------------------+
1734 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1735 +-----------------------------+----------------------------------------------------------+
1736 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1737 +-----------------------------+----------------------------------------------------------+
1739 There is no restriction on indexing beyond the end of the array implied
1740 by a static type (though there are restrictions on indexing beyond the
1741 bounds of an allocated object in some cases). This means that
1742 single-dimension 'variable sized array' addressing can be implemented in
1743 LLVM with a zero length array type. An implementation of 'pascal style
1744 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1755 The function type can be thought of as a function signature. It consists
1756 of a return type and a list of formal parameter types. The return type
1757 of a function type is a first class type or a void type.
1764 <returntype> (<parameter list>)
1766 ...where '``<parameter list>``' is a comma-separated list of type
1767 specifiers. Optionally, the parameter list may include a type ``...``,
1768 which indicates that the function takes a variable number of arguments.
1769 Variable argument functions can access their arguments with the
1770 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1771 '``<returntype>``' is any type except :ref:`label <t_label>`.
1776 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1777 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1778 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1779 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1780 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1781 | ``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. |
1782 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1783 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1784 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1794 The structure type is used to represent a collection of data members
1795 together in memory. The elements of a structure may be any type that has
1798 Structures in memory are accessed using '``load``' and '``store``' by
1799 getting a pointer to a field with the '``getelementptr``' instruction.
1800 Structures in registers are accessed using the '``extractvalue``' and
1801 '``insertvalue``' instructions.
1803 Structures may optionally be "packed" structures, which indicate that
1804 the alignment of the struct is one byte, and that there is no padding
1805 between the elements. In non-packed structs, padding between field types
1806 is inserted as defined by the DataLayout string in the module, which is
1807 required to match what the underlying code generator expects.
1809 Structures can either be "literal" or "identified". A literal structure
1810 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1811 identified types are always defined at the top level with a name.
1812 Literal types are uniqued by their contents and can never be recursive
1813 or opaque since there is no way to write one. Identified types can be
1814 recursive, can be opaqued, and are never uniqued.
1821 %T1 = type { <type list> } ; Identified normal struct type
1822 %T2 = type <{ <type list> }> ; Identified packed struct type
1827 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1828 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1829 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1830 | ``{ 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``. |
1831 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1832 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1833 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1837 Opaque Structure Types
1838 ^^^^^^^^^^^^^^^^^^^^^^
1843 Opaque structure types are used to represent named structure types that
1844 do not have a body specified. This corresponds (for example) to the C
1845 notion of a forward declared structure.
1858 +--------------+-------------------+
1859 | ``opaque`` | An opaque type. |
1860 +--------------+-------------------+
1870 The pointer type is used to specify memory locations. Pointers are
1871 commonly used to reference objects in memory.
1873 Pointer types may have an optional address space attribute defining the
1874 numbered address space where the pointed-to object resides. The default
1875 address space is number zero. The semantics of non-zero address spaces
1876 are target-specific.
1878 Note that LLVM does not permit pointers to void (``void*``) nor does it
1879 permit pointers to labels (``label*``). Use ``i8*`` instead.
1891 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1892 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1893 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1894 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1895 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1896 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1897 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1907 A vector type is a simple derived type that represents a vector of
1908 elements. Vector types are used when multiple primitive data are
1909 operated in parallel using a single instruction (SIMD). A vector type
1910 requires a size (number of elements) and an underlying primitive data
1911 type. Vector types are considered :ref:`first class <t_firstclass>`.
1918 < <# elements> x <elementtype> >
1920 The number of elements is a constant integer value larger than 0;
1921 elementtype may be any integer or floating point type, or a pointer to
1922 these types. Vectors of size zero are not allowed.
1927 +-------------------+--------------------------------------------------+
1928 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1929 +-------------------+--------------------------------------------------+
1930 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1931 +-------------------+--------------------------------------------------+
1932 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1933 +-------------------+--------------------------------------------------+
1934 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1935 +-------------------+--------------------------------------------------+
1940 LLVM has several different basic types of constants. This section
1941 describes them all and their syntax.
1946 **Boolean constants**
1947 The two strings '``true``' and '``false``' are both valid constants
1949 **Integer constants**
1950 Standard integers (such as '4') are constants of the
1951 :ref:`integer <t_integer>` type. Negative numbers may be used with
1953 **Floating point constants**
1954 Floating point constants use standard decimal notation (e.g.
1955 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1956 hexadecimal notation (see below). The assembler requires the exact
1957 decimal value of a floating-point constant. For example, the
1958 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1959 decimal in binary. Floating point constants must have a :ref:`floating
1960 point <t_floating>` type.
1961 **Null pointer constants**
1962 The identifier '``null``' is recognized as a null pointer constant
1963 and must be of :ref:`pointer type <t_pointer>`.
1965 The one non-intuitive notation for constants is the hexadecimal form of
1966 floating point constants. For example, the form
1967 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1968 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1969 constants are required (and the only time that they are generated by the
1970 disassembler) is when a floating point constant must be emitted but it
1971 cannot be represented as a decimal floating point number in a reasonable
1972 number of digits. For example, NaN's, infinities, and other special
1973 values are represented in their IEEE hexadecimal format so that assembly
1974 and disassembly do not cause any bits to change in the constants.
1976 When using the hexadecimal form, constants of types half, float, and
1977 double are represented using the 16-digit form shown above (which
1978 matches the IEEE754 representation for double); half and float values
1979 must, however, be exactly representable as IEEE 754 half and single
1980 precision, respectively. Hexadecimal format is always used for long
1981 double, and there are three forms of long double. The 80-bit format used
1982 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1983 128-bit format used by PowerPC (two adjacent doubles) is represented by
1984 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1985 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1986 will only work if they match the long double format on your target.
1987 The IEEE 16-bit format (half precision) is represented by ``0xH``
1988 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1989 (sign bit at the left).
1991 There are no constants of type x86mmx.
1993 .. _complexconstants:
1998 Complex constants are a (potentially recursive) combination of simple
1999 constants and smaller complex constants.
2001 **Structure constants**
2002 Structure constants are represented with notation similar to
2003 structure type definitions (a comma separated list of elements,
2004 surrounded by braces (``{}``)). For example:
2005 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2006 "``@G = external global i32``". Structure constants must have
2007 :ref:`structure type <t_struct>`, and the number and types of elements
2008 must match those specified by the type.
2010 Array constants are represented with notation similar to array type
2011 definitions (a comma separated list of elements, surrounded by
2012 square brackets (``[]``)). For example:
2013 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2014 :ref:`array type <t_array>`, and the number and types of elements must
2015 match those specified by the type.
2016 **Vector constants**
2017 Vector constants are represented with notation similar to vector
2018 type definitions (a comma separated list of elements, surrounded by
2019 less-than/greater-than's (``<>``)). For example:
2020 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2021 must have :ref:`vector type <t_vector>`, and the number and types of
2022 elements must match those specified by the type.
2023 **Zero initialization**
2024 The string '``zeroinitializer``' can be used to zero initialize a
2025 value to zero of *any* type, including scalar and
2026 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2027 having to print large zero initializers (e.g. for large arrays) and
2028 is always exactly equivalent to using explicit zero initializers.
2030 A metadata node is a structure-like constant with :ref:`metadata
2031 type <t_metadata>`. For example:
2032 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2033 constants that are meant to be interpreted as part of the
2034 instruction stream, metadata is a place to attach additional
2035 information such as debug info.
2037 Global Variable and Function Addresses
2038 --------------------------------------
2040 The addresses of :ref:`global variables <globalvars>` and
2041 :ref:`functions <functionstructure>` are always implicitly valid
2042 (link-time) constants. These constants are explicitly referenced when
2043 the :ref:`identifier for the global <identifiers>` is used and always have
2044 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2047 .. code-block:: llvm
2051 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2058 The string '``undef``' can be used anywhere a constant is expected, and
2059 indicates that the user of the value may receive an unspecified
2060 bit-pattern. Undefined values may be of any type (other than '``label``'
2061 or '``void``') and be used anywhere a constant is permitted.
2063 Undefined values are useful because they indicate to the compiler that
2064 the program is well defined no matter what value is used. This gives the
2065 compiler more freedom to optimize. Here are some examples of
2066 (potentially surprising) transformations that are valid (in pseudo IR):
2068 .. code-block:: llvm
2078 This is safe because all of the output bits are affected by the undef
2079 bits. Any output bit can have a zero or one depending on the input bits.
2081 .. code-block:: llvm
2092 These logical operations have bits that are not always affected by the
2093 input. For example, if ``%X`` has a zero bit, then the output of the
2094 '``and``' operation will always be a zero for that bit, no matter what
2095 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2096 optimize or assume that the result of the '``and``' is '``undef``'.
2097 However, it is safe to assume that all bits of the '``undef``' could be
2098 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2099 all the bits of the '``undef``' operand to the '``or``' could be set,
2100 allowing the '``or``' to be folded to -1.
2102 .. code-block:: llvm
2104 %A = select undef, %X, %Y
2105 %B = select undef, 42, %Y
2106 %C = select %X, %Y, undef
2116 This set of examples shows that undefined '``select``' (and conditional
2117 branch) conditions can go *either way*, but they have to come from one
2118 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2119 both known to have a clear low bit, then ``%A`` would have to have a
2120 cleared low bit. However, in the ``%C`` example, the optimizer is
2121 allowed to assume that the '``undef``' operand could be the same as
2122 ``%Y``, allowing the whole '``select``' to be eliminated.
2124 .. code-block:: llvm
2126 %A = xor undef, undef
2143 This example points out that two '``undef``' operands are not
2144 necessarily the same. This can be surprising to people (and also matches
2145 C semantics) where they assume that "``X^X``" is always zero, even if
2146 ``X`` is undefined. This isn't true for a number of reasons, but the
2147 short answer is that an '``undef``' "variable" can arbitrarily change
2148 its value over its "live range". This is true because the variable
2149 doesn't actually *have a live range*. Instead, the value is logically
2150 read from arbitrary registers that happen to be around when needed, so
2151 the value is not necessarily consistent over time. In fact, ``%A`` and
2152 ``%C`` need to have the same semantics or the core LLVM "replace all
2153 uses with" concept would not hold.
2155 .. code-block:: llvm
2163 These examples show the crucial difference between an *undefined value*
2164 and *undefined behavior*. An undefined value (like '``undef``') is
2165 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2166 operation can be constant folded to '``undef``', because the '``undef``'
2167 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2168 However, in the second example, we can make a more aggressive
2169 assumption: because the ``undef`` is allowed to be an arbitrary value,
2170 we are allowed to assume that it could be zero. Since a divide by zero
2171 has *undefined behavior*, we are allowed to assume that the operation
2172 does not execute at all. This allows us to delete the divide and all
2173 code after it. Because the undefined operation "can't happen", the
2174 optimizer can assume that it occurs in dead code.
2176 .. code-block:: llvm
2178 a: store undef -> %X
2179 b: store %X -> undef
2184 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2185 value can be assumed to not have any effect; we can assume that the
2186 value is overwritten with bits that happen to match what was already
2187 there. However, a store *to* an undefined location could clobber
2188 arbitrary memory, therefore, it has undefined behavior.
2195 Poison values are similar to :ref:`undef values <undefvalues>`, however
2196 they also represent the fact that an instruction or constant expression
2197 which cannot evoke side effects has nevertheless detected a condition
2198 which results in undefined behavior.
2200 There is currently no way of representing a poison value in the IR; they
2201 only exist when produced by operations such as :ref:`add <i_add>` with
2204 Poison value behavior is defined in terms of value *dependence*:
2206 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2207 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2208 their dynamic predecessor basic block.
2209 - Function arguments depend on the corresponding actual argument values
2210 in the dynamic callers of their functions.
2211 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2212 instructions that dynamically transfer control back to them.
2213 - :ref:`Invoke <i_invoke>` instructions depend on the
2214 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2215 call instructions that dynamically transfer control back to them.
2216 - Non-volatile loads and stores depend on the most recent stores to all
2217 of the referenced memory addresses, following the order in the IR
2218 (including loads and stores implied by intrinsics such as
2219 :ref:`@llvm.memcpy <int_memcpy>`.)
2220 - An instruction with externally visible side effects depends on the
2221 most recent preceding instruction with externally visible side
2222 effects, following the order in the IR. (This includes :ref:`volatile
2223 operations <volatile>`.)
2224 - An instruction *control-depends* on a :ref:`terminator
2225 instruction <terminators>` if the terminator instruction has
2226 multiple successors and the instruction is always executed when
2227 control transfers to one of the successors, and may not be executed
2228 when control is transferred to another.
2229 - Additionally, an instruction also *control-depends* on a terminator
2230 instruction if the set of instructions it otherwise depends on would
2231 be different if the terminator had transferred control to a different
2233 - Dependence is transitive.
2235 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2236 with the additional affect that any instruction which has a *dependence*
2237 on a poison value has undefined behavior.
2239 Here are some examples:
2241 .. code-block:: llvm
2244 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2245 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2246 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2247 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2249 store i32 %poison, i32* @g ; Poison value stored to memory.
2250 %poison2 = load i32* @g ; Poison value loaded back from memory.
2252 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2254 %narrowaddr = bitcast i32* @g to i16*
2255 %wideaddr = bitcast i32* @g to i64*
2256 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2257 %poison4 = load i64* %wideaddr ; Returns a poison value.
2259 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2260 br i1 %cmp, label %true, label %end ; Branch to either destination.
2263 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2264 ; it has undefined behavior.
2268 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2269 ; Both edges into this PHI are
2270 ; control-dependent on %cmp, so this
2271 ; always results in a poison value.
2273 store volatile i32 0, i32* @g ; This would depend on the store in %true
2274 ; if %cmp is true, or the store in %entry
2275 ; otherwise, so this is undefined behavior.
2277 br i1 %cmp, label %second_true, label %second_end
2278 ; The same branch again, but this time the
2279 ; true block doesn't have side effects.
2286 store volatile i32 0, i32* @g ; This time, the instruction always depends
2287 ; on the store in %end. Also, it is
2288 ; control-equivalent to %end, so this is
2289 ; well-defined (ignoring earlier undefined
2290 ; behavior in this example).
2294 Addresses of Basic Blocks
2295 -------------------------
2297 ``blockaddress(@function, %block)``
2299 The '``blockaddress``' constant computes the address of the specified
2300 basic block in the specified function, and always has an ``i8*`` type.
2301 Taking the address of the entry block is illegal.
2303 This value only has defined behavior when used as an operand to the
2304 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2305 against null. Pointer equality tests between labels addresses results in
2306 undefined behavior --- though, again, comparison against null is ok, and
2307 no label is equal to the null pointer. This may be passed around as an
2308 opaque pointer sized value as long as the bits are not inspected. This
2309 allows ``ptrtoint`` and arithmetic to be performed on these values so
2310 long as the original value is reconstituted before the ``indirectbr``
2313 Finally, some targets may provide defined semantics when using the value
2314 as the operand to an inline assembly, but that is target specific.
2318 Constant Expressions
2319 --------------------
2321 Constant expressions are used to allow expressions involving other
2322 constants to be used as constants. Constant expressions may be of any
2323 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2324 that does not have side effects (e.g. load and call are not supported).
2325 The following is the syntax for constant expressions:
2327 ``trunc (CST to TYPE)``
2328 Truncate a constant to another type. The bit size of CST must be
2329 larger than the bit size of TYPE. Both types must be integers.
2330 ``zext (CST to TYPE)``
2331 Zero extend a constant to another type. The bit size of CST must be
2332 smaller than the bit size of TYPE. Both types must be integers.
2333 ``sext (CST to TYPE)``
2334 Sign extend a constant to another type. The bit size of CST must be
2335 smaller than the bit size of TYPE. Both types must be integers.
2336 ``fptrunc (CST to TYPE)``
2337 Truncate a floating point constant to another floating point type.
2338 The size of CST must be larger than the size of TYPE. Both types
2339 must be floating point.
2340 ``fpext (CST to TYPE)``
2341 Floating point extend a constant to another type. The size of CST
2342 must be smaller or equal to the size of TYPE. Both types must be
2344 ``fptoui (CST to TYPE)``
2345 Convert a floating point constant to the corresponding unsigned
2346 integer constant. TYPE must be a scalar or vector integer type. CST
2347 must be of scalar or vector floating point type. Both CST and TYPE
2348 must be scalars, or vectors of the same number of elements. If the
2349 value won't fit in the integer type, the results are undefined.
2350 ``fptosi (CST to TYPE)``
2351 Convert a floating point constant to the corresponding signed
2352 integer constant. TYPE must be a scalar or vector integer type. CST
2353 must be of scalar or vector floating point type. Both CST and TYPE
2354 must be scalars, or vectors of the same number of elements. If the
2355 value won't fit in the integer type, the results are undefined.
2356 ``uitofp (CST to TYPE)``
2357 Convert an unsigned integer constant to the corresponding floating
2358 point constant. TYPE must be a scalar or vector floating point type.
2359 CST must be of scalar or vector integer type. Both CST and TYPE must
2360 be scalars, or vectors of the same number of elements. If the value
2361 won't fit in the floating point type, the results are undefined.
2362 ``sitofp (CST to TYPE)``
2363 Convert a signed integer constant to the corresponding floating
2364 point constant. TYPE must be a scalar or vector floating point type.
2365 CST must be of scalar or vector integer type. Both CST and TYPE must
2366 be scalars, or vectors of the same number of elements. If the value
2367 won't fit in the floating point type, the results are undefined.
2368 ``ptrtoint (CST to TYPE)``
2369 Convert a pointer typed constant to the corresponding integer
2370 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2371 pointer type. The ``CST`` value is zero extended, truncated, or
2372 unchanged to make it fit in ``TYPE``.
2373 ``inttoptr (CST to TYPE)``
2374 Convert an integer constant to a pointer constant. TYPE must be a
2375 pointer type. CST must be of integer type. The CST value is zero
2376 extended, truncated, or unchanged to make it fit in a pointer size.
2377 This one is *really* dangerous!
2378 ``bitcast (CST to TYPE)``
2379 Convert a constant, CST, to another TYPE. The constraints of the
2380 operands are the same as those for the :ref:`bitcast
2381 instruction <i_bitcast>`.
2382 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2383 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2384 constants. As with the :ref:`getelementptr <i_getelementptr>`
2385 instruction, the index list may have zero or more indexes, which are
2386 required to make sense for the type of "CSTPTR".
2387 ``select (COND, VAL1, VAL2)``
2388 Perform the :ref:`select operation <i_select>` on constants.
2389 ``icmp COND (VAL1, VAL2)``
2390 Performs the :ref:`icmp operation <i_icmp>` on constants.
2391 ``fcmp COND (VAL1, VAL2)``
2392 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2393 ``extractelement (VAL, IDX)``
2394 Perform the :ref:`extractelement operation <i_extractelement>` on
2396 ``insertelement (VAL, ELT, IDX)``
2397 Perform the :ref:`insertelement operation <i_insertelement>` on
2399 ``shufflevector (VEC1, VEC2, IDXMASK)``
2400 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2402 ``extractvalue (VAL, IDX0, IDX1, ...)``
2403 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2404 constants. The index list is interpreted in a similar manner as
2405 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2406 least one index value must be specified.
2407 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2408 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2409 The index list is interpreted in a similar manner as indices in a
2410 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2411 value must be specified.
2412 ``OPCODE (LHS, RHS)``
2413 Perform the specified operation of the LHS and RHS constants. OPCODE
2414 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2415 binary <bitwiseops>` operations. The constraints on operands are
2416 the same as those for the corresponding instruction (e.g. no bitwise
2417 operations on floating point values are allowed).
2424 Inline Assembler Expressions
2425 ----------------------------
2427 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2428 Inline Assembly <moduleasm>`) through the use of a special value. This
2429 value represents the inline assembler as a string (containing the
2430 instructions to emit), a list of operand constraints (stored as a
2431 string), a flag that indicates whether or not the inline asm expression
2432 has side effects, and a flag indicating whether the function containing
2433 the asm needs to align its stack conservatively. An example inline
2434 assembler expression is:
2436 .. code-block:: llvm
2438 i32 (i32) asm "bswap $0", "=r,r"
2440 Inline assembler expressions may **only** be used as the callee operand
2441 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2442 Thus, typically we have:
2444 .. code-block:: llvm
2446 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2448 Inline asms with side effects not visible in the constraint list must be
2449 marked as having side effects. This is done through the use of the
2450 '``sideeffect``' keyword, like so:
2452 .. code-block:: llvm
2454 call void asm sideeffect "eieio", ""()
2456 In some cases inline asms will contain code that will not work unless
2457 the stack is aligned in some way, such as calls or SSE instructions on
2458 x86, yet will not contain code that does that alignment within the asm.
2459 The compiler should make conservative assumptions about what the asm
2460 might contain and should generate its usual stack alignment code in the
2461 prologue if the '``alignstack``' keyword is present:
2463 .. code-block:: llvm
2465 call void asm alignstack "eieio", ""()
2467 Inline asms also support using non-standard assembly dialects. The
2468 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2469 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2470 the only supported dialects. An example is:
2472 .. code-block:: llvm
2474 call void asm inteldialect "eieio", ""()
2476 If multiple keywords appear the '``sideeffect``' keyword must come
2477 first, the '``alignstack``' keyword second and the '``inteldialect``'
2483 The call instructions that wrap inline asm nodes may have a
2484 "``!srcloc``" MDNode attached to it that contains a list of constant
2485 integers. If present, the code generator will use the integer as the
2486 location cookie value when report errors through the ``LLVMContext``
2487 error reporting mechanisms. This allows a front-end to correlate backend
2488 errors that occur with inline asm back to the source code that produced
2491 .. code-block:: llvm
2493 call void asm sideeffect "something bad", ""(), !srcloc !42
2495 !42 = !{ i32 1234567 }
2497 It is up to the front-end to make sense of the magic numbers it places
2498 in the IR. If the MDNode contains multiple constants, the code generator
2499 will use the one that corresponds to the line of the asm that the error
2504 Metadata Nodes and Metadata Strings
2505 -----------------------------------
2507 LLVM IR allows metadata to be attached to instructions in the program
2508 that can convey extra information about the code to the optimizers and
2509 code generator. One example application of metadata is source-level
2510 debug information. There are two metadata primitives: strings and nodes.
2511 All metadata has the ``metadata`` type and is identified in syntax by a
2512 preceding exclamation point ('``!``').
2514 A metadata string is a string surrounded by double quotes. It can
2515 contain any character by escaping non-printable characters with
2516 "``\xx``" where "``xx``" is the two digit hex code. For example:
2519 Metadata nodes are represented with notation similar to structure
2520 constants (a comma separated list of elements, surrounded by braces and
2521 preceded by an exclamation point). Metadata nodes can have any values as
2522 their operand. For example:
2524 .. code-block:: llvm
2526 !{ metadata !"test\00", i32 10}
2528 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2529 metadata nodes, which can be looked up in the module symbol table. For
2532 .. code-block:: llvm
2534 !foo = metadata !{!4, !3}
2536 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2537 function is using two metadata arguments:
2539 .. code-block:: llvm
2541 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2543 Metadata can be attached with an instruction. Here metadata ``!21`` is
2544 attached to the ``add`` instruction using the ``!dbg`` identifier:
2546 .. code-block:: llvm
2548 %indvar.next = add i64 %indvar, 1, !dbg !21
2550 More information about specific metadata nodes recognized by the
2551 optimizers and code generator is found below.
2556 In LLVM IR, memory does not have types, so LLVM's own type system is not
2557 suitable for doing TBAA. Instead, metadata is added to the IR to
2558 describe a type system of a higher level language. This can be used to
2559 implement typical C/C++ TBAA, but it can also be used to implement
2560 custom alias analysis behavior for other languages.
2562 The current metadata format is very simple. TBAA metadata nodes have up
2563 to three fields, e.g.:
2565 .. code-block:: llvm
2567 !0 = metadata !{ metadata !"an example type tree" }
2568 !1 = metadata !{ metadata !"int", metadata !0 }
2569 !2 = metadata !{ metadata !"float", metadata !0 }
2570 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2572 The first field is an identity field. It can be any value, usually a
2573 metadata string, which uniquely identifies the type. The most important
2574 name in the tree is the name of the root node. Two trees with different
2575 root node names are entirely disjoint, even if they have leaves with
2578 The second field identifies the type's parent node in the tree, or is
2579 null or omitted for a root node. A type is considered to alias all of
2580 its descendants and all of its ancestors in the tree. Also, a type is
2581 considered to alias all types in other trees, so that bitcode produced
2582 from multiple front-ends is handled conservatively.
2584 If the third field is present, it's an integer which if equal to 1
2585 indicates that the type is "constant" (meaning
2586 ``pointsToConstantMemory`` should return true; see `other useful
2587 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2589 '``tbaa.struct``' Metadata
2590 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2592 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2593 aggregate assignment operations in C and similar languages, however it
2594 is defined to copy a contiguous region of memory, which is more than
2595 strictly necessary for aggregate types which contain holes due to
2596 padding. Also, it doesn't contain any TBAA information about the fields
2599 ``!tbaa.struct`` metadata can describe which memory subregions in a
2600 memcpy are padding and what the TBAA tags of the struct are.
2602 The current metadata format is very simple. ``!tbaa.struct`` metadata
2603 nodes are a list of operands which are in conceptual groups of three.
2604 For each group of three, the first operand gives the byte offset of a
2605 field in bytes, the second gives its size in bytes, and the third gives
2608 .. code-block:: llvm
2610 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2612 This describes a struct with two fields. The first is at offset 0 bytes
2613 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2614 and has size 4 bytes and has tbaa tag !2.
2616 Note that the fields need not be contiguous. In this example, there is a
2617 4 byte gap between the two fields. This gap represents padding which
2618 does not carry useful data and need not be preserved.
2620 '``fpmath``' Metadata
2621 ^^^^^^^^^^^^^^^^^^^^^
2623 ``fpmath`` metadata may be attached to any instruction of floating point
2624 type. It can be used to express the maximum acceptable error in the
2625 result of that instruction, in ULPs, thus potentially allowing the
2626 compiler to use a more efficient but less accurate method of computing
2627 it. ULP is defined as follows:
2629 If ``x`` is a real number that lies between two finite consecutive
2630 floating-point numbers ``a`` and ``b``, without being equal to one
2631 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2632 distance between the two non-equal finite floating-point numbers
2633 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2635 The metadata node shall consist of a single positive floating point
2636 number representing the maximum relative error, for example:
2638 .. code-block:: llvm
2640 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2642 '``range``' Metadata
2643 ^^^^^^^^^^^^^^^^^^^^
2645 ``range`` metadata may be attached only to loads of integer types. It
2646 expresses the possible ranges the loaded value is in. The ranges are
2647 represented with a flattened list of integers. The loaded value is known
2648 to be in the union of the ranges defined by each consecutive pair. Each
2649 pair has the following properties:
2651 - The type must match the type loaded by the instruction.
2652 - The pair ``a,b`` represents the range ``[a,b)``.
2653 - Both ``a`` and ``b`` are constants.
2654 - The range is allowed to wrap.
2655 - The range should not represent the full or empty set. That is,
2658 In addition, the pairs must be in signed order of the lower bound and
2659 they must be non-contiguous.
2663 .. code-block:: llvm
2665 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2666 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2667 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2668 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2670 !0 = metadata !{ i8 0, i8 2 }
2671 !1 = metadata !{ i8 255, i8 2 }
2672 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2673 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2678 It is sometimes useful to attach information to loop constructs. Currently,
2679 loop metadata is implemented as metadata attached to the branch instruction
2680 in the loop latch block. This type of metadata refer to a metadata node that is
2681 guaranteed to be separate for each loop. The loop identifier metadata is
2682 specified with the name ``llvm.loop``.
2684 The loop identifier metadata is implemented using a metadata that refers to
2685 itself to avoid merging it with any other identifier metadata, e.g.,
2686 during module linkage or function inlining. That is, each loop should refer
2687 to their own identification metadata even if they reside in separate functions.
2688 The following example contains loop identifier metadata for two separate loop
2691 .. code-block:: llvm
2693 !0 = metadata !{ metadata !0 }
2694 !1 = metadata !{ metadata !1 }
2696 The loop identifier metadata can be used to specify additional per-loop
2697 metadata. Any operands after the first operand can be treated as user-defined
2698 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2699 by the loop vectorizer to indicate how many times to unroll the loop:
2701 .. code-block:: llvm
2703 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2705 !0 = metadata !{ metadata !0, metadata !1 }
2706 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2711 Metadata types used to annotate memory accesses with information helpful
2712 for optimizations are prefixed with ``llvm.mem``.
2714 '``llvm.mem.parallel_loop_access``' Metadata
2715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2717 For a loop to be parallel, in addition to using
2718 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2719 also all of the memory accessing instructions in the loop body need to be
2720 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2721 is at least one memory accessing instruction not marked with the metadata,
2722 the loop must be considered a sequential loop. This causes parallel loops to be
2723 converted to sequential loops due to optimization passes that are unaware of
2724 the parallel semantics and that insert new memory instructions to the loop
2727 Example of a loop that is considered parallel due to its correct use of
2728 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2729 metadata types that refer to the same loop identifier metadata.
2731 .. code-block:: llvm
2735 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2737 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2739 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2743 !0 = metadata !{ metadata !0 }
2745 It is also possible to have nested parallel loops. In that case the
2746 memory accesses refer to a list of loop identifier metadata nodes instead of
2747 the loop identifier metadata node directly:
2749 .. code-block:: llvm
2756 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2758 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2760 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
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 %outer.for.end, label %outer.for.body, !llvm.loop !2
2770 outer.for.end: ; preds = %for.body
2772 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2773 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2774 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2776 '``llvm.vectorizer``'
2777 ^^^^^^^^^^^^^^^^^^^^^
2779 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2780 vectorization parameters such as vectorization factor and unroll factor.
2782 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2783 loop identification metadata.
2785 '``llvm.vectorizer.unroll``' Metadata
2786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2788 This metadata instructs the loop vectorizer to unroll the specified
2789 loop exactly ``N`` times.
2791 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2792 operand is an integer specifying the unroll factor. For example:
2794 .. code-block:: llvm
2796 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2798 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2801 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2802 determined automatically.
2804 '``llvm.vectorizer.width``' Metadata
2805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2807 This metadata sets the target width of the vectorizer to ``N``. Without
2808 this metadata, the vectorizer will choose a width automatically.
2809 Regardless of this metadata, the vectorizer will only vectorize loops if
2810 it believes it is valid to do so.
2812 The first operand is the string ``llvm.vectorizer.width`` and the second
2813 operand is an integer specifying the width. For example:
2815 .. code-block:: llvm
2817 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2819 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2822 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2825 Module Flags Metadata
2826 =====================
2828 Information about the module as a whole is difficult to convey to LLVM's
2829 subsystems. The LLVM IR isn't sufficient to transmit this information.
2830 The ``llvm.module.flags`` named metadata exists in order to facilitate
2831 this. These flags are in the form of key / value pairs --- much like a
2832 dictionary --- making it easy for any subsystem who cares about a flag to
2835 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2836 Each triplet has the following form:
2838 - The first element is a *behavior* flag, which specifies the behavior
2839 when two (or more) modules are merged together, and it encounters two
2840 (or more) metadata with the same ID. The supported behaviors are
2842 - The second element is a metadata string that is a unique ID for the
2843 metadata. Each module may only have one flag entry for each unique ID (not
2844 including entries with the **Require** behavior).
2845 - The third element is the value of the flag.
2847 When two (or more) modules are merged together, the resulting
2848 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2849 each unique metadata ID string, there will be exactly one entry in the merged
2850 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2851 be determined by the merge behavior flag, as described below. The only exception
2852 is that entries with the *Require* behavior are always preserved.
2854 The following behaviors are supported:
2865 Emits an error if two values disagree, otherwise the resulting value
2866 is that of the operands.
2870 Emits a warning if two values disagree. The result value will be the
2871 operand for the flag from the first module being linked.
2875 Adds a requirement that another module flag be present and have a
2876 specified value after linking is performed. The value must be a
2877 metadata pair, where the first element of the pair is the ID of the
2878 module flag to be restricted, and the second element of the pair is
2879 the value the module flag should be restricted to. This behavior can
2880 be used to restrict the allowable results (via triggering of an
2881 error) of linking IDs with the **Override** behavior.
2885 Uses the specified value, regardless of the behavior or value of the
2886 other module. If both modules specify **Override**, but the values
2887 differ, an error will be emitted.
2891 Appends the two values, which are required to be metadata nodes.
2895 Appends the two values, which are required to be metadata
2896 nodes. However, duplicate entries in the second list are dropped
2897 during the append operation.
2899 It is an error for a particular unique flag ID to have multiple behaviors,
2900 except in the case of **Require** (which adds restrictions on another metadata
2901 value) or **Override**.
2903 An example of module flags:
2905 .. code-block:: llvm
2907 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2908 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2909 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2910 !3 = metadata !{ i32 3, metadata !"qux",
2912 metadata !"foo", i32 1
2915 !llvm.module.flags = !{ !0, !1, !2, !3 }
2917 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2918 if two or more ``!"foo"`` flags are seen is to emit an error if their
2919 values are not equal.
2921 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2922 behavior if two or more ``!"bar"`` flags are seen is to use the value
2925 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2926 behavior if two or more ``!"qux"`` flags are seen is to emit a
2927 warning if their values are not equal.
2929 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2933 metadata !{ metadata !"foo", i32 1 }
2935 The behavior is to emit an error if the ``llvm.module.flags`` does not
2936 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2939 Objective-C Garbage Collection Module Flags Metadata
2940 ----------------------------------------------------
2942 On the Mach-O platform, Objective-C stores metadata about garbage
2943 collection in a special section called "image info". The metadata
2944 consists of a version number and a bitmask specifying what types of
2945 garbage collection are supported (if any) by the file. If two or more
2946 modules are linked together their garbage collection metadata needs to
2947 be merged rather than appended together.
2949 The Objective-C garbage collection module flags metadata consists of the
2950 following key-value pairs:
2959 * - ``Objective-C Version``
2960 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2962 * - ``Objective-C Image Info Version``
2963 - **[Required]** --- The version of the image info section. Currently
2966 * - ``Objective-C Image Info Section``
2967 - **[Required]** --- The section to place the metadata. Valid values are
2968 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2969 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2970 Objective-C ABI version 2.
2972 * - ``Objective-C Garbage Collection``
2973 - **[Required]** --- Specifies whether garbage collection is supported or
2974 not. Valid values are 0, for no garbage collection, and 2, for garbage
2975 collection supported.
2977 * - ``Objective-C GC Only``
2978 - **[Optional]** --- Specifies that only garbage collection is supported.
2979 If present, its value must be 6. This flag requires that the
2980 ``Objective-C Garbage Collection`` flag have the value 2.
2982 Some important flag interactions:
2984 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2985 merged with a module with ``Objective-C Garbage Collection`` set to
2986 2, then the resulting module has the
2987 ``Objective-C Garbage Collection`` flag set to 0.
2988 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2989 merged with a module with ``Objective-C GC Only`` set to 6.
2991 Automatic Linker Flags Module Flags Metadata
2992 --------------------------------------------
2994 Some targets support embedding flags to the linker inside individual object
2995 files. Typically this is used in conjunction with language extensions which
2996 allow source files to explicitly declare the libraries they depend on, and have
2997 these automatically be transmitted to the linker via object files.
2999 These flags are encoded in the IR using metadata in the module flags section,
3000 using the ``Linker Options`` key. The merge behavior for this flag is required
3001 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3002 node which should be a list of other metadata nodes, each of which should be a
3003 list of metadata strings defining linker options.
3005 For example, the following metadata section specifies two separate sets of
3006 linker options, presumably to link against ``libz`` and the ``Cocoa``
3009 !0 = metadata !{ i32 6, metadata !"Linker Options",
3011 metadata !{ metadata !"-lz" },
3012 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3013 !llvm.module.flags = !{ !0 }
3015 The metadata encoding as lists of lists of options, as opposed to a collapsed
3016 list of options, is chosen so that the IR encoding can use multiple option
3017 strings to specify e.g., a single library, while still having that specifier be
3018 preserved as an atomic element that can be recognized by a target specific
3019 assembly writer or object file emitter.
3021 Each individual option is required to be either a valid option for the target's
3022 linker, or an option that is reserved by the target specific assembly writer or
3023 object file emitter. No other aspect of these options is defined by the IR.
3025 .. _intrinsicglobalvariables:
3027 Intrinsic Global Variables
3028 ==========================
3030 LLVM has a number of "magic" global variables that contain data that
3031 affect code generation or other IR semantics. These are documented here.
3032 All globals of this sort should have a section specified as
3033 "``llvm.metadata``". This section and all globals that start with
3034 "``llvm.``" are reserved for use by LLVM.
3038 The '``llvm.used``' Global Variable
3039 -----------------------------------
3041 The ``@llvm.used`` global is an array which has
3042 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3043 pointers to named global variables, functions and aliases which may optionally
3044 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3047 .. code-block:: llvm
3052 @llvm.used = appending global [2 x i8*] [
3054 i8* bitcast (i32* @Y to i8*)
3055 ], section "llvm.metadata"
3057 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3058 and linker are required to treat the symbol as if there is a reference to the
3059 symbol that it cannot see (which is why they have to be named). For example, if
3060 a variable has internal linkage and no references other than that from the
3061 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3062 references from inline asms and other things the compiler cannot "see", and
3063 corresponds to "``attribute((used))``" in GNU C.
3065 On some targets, the code generator must emit a directive to the
3066 assembler or object file to prevent the assembler and linker from
3067 molesting the symbol.
3069 .. _gv_llvmcompilerused:
3071 The '``llvm.compiler.used``' Global Variable
3072 --------------------------------------------
3074 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3075 directive, except that it only prevents the compiler from touching the
3076 symbol. On targets that support it, this allows an intelligent linker to
3077 optimize references to the symbol without being impeded as it would be
3080 This is a rare construct that should only be used in rare circumstances,
3081 and should not be exposed to source languages.
3083 .. _gv_llvmglobalctors:
3085 The '``llvm.global_ctors``' Global Variable
3086 -------------------------------------------
3088 .. code-block:: llvm
3090 %0 = type { i32, void ()* }
3091 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3093 The ``@llvm.global_ctors`` array contains a list of constructor
3094 functions and associated priorities. The functions referenced by this
3095 array will be called in ascending order of priority (i.e. lowest first)
3096 when the module is loaded. The order of functions with the same priority
3099 .. _llvmglobaldtors:
3101 The '``llvm.global_dtors``' Global Variable
3102 -------------------------------------------
3104 .. code-block:: llvm
3106 %0 = type { i32, void ()* }
3107 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3109 The ``@llvm.global_dtors`` array contains a list of destructor functions
3110 and associated priorities. The functions referenced by this array will
3111 be called in descending order of priority (i.e. highest first) when the
3112 module is loaded. The order of functions with the same priority is not
3115 Instruction Reference
3116 =====================
3118 The LLVM instruction set consists of several different classifications
3119 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3120 instructions <binaryops>`, :ref:`bitwise binary
3121 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3122 :ref:`other instructions <otherops>`.
3126 Terminator Instructions
3127 -----------------------
3129 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3130 program ends with a "Terminator" instruction, which indicates which
3131 block should be executed after the current block is finished. These
3132 terminator instructions typically yield a '``void``' value: they produce
3133 control flow, not values (the one exception being the
3134 ':ref:`invoke <i_invoke>`' instruction).
3136 The terminator instructions are: ':ref:`ret <i_ret>`',
3137 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3138 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3139 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3143 '``ret``' Instruction
3144 ^^^^^^^^^^^^^^^^^^^^^
3151 ret <type> <value> ; Return a value from a non-void function
3152 ret void ; Return from void function
3157 The '``ret``' instruction is used to return control flow (and optionally
3158 a value) from a function back to the caller.
3160 There are two forms of the '``ret``' instruction: one that returns a
3161 value and then causes control flow, and one that just causes control
3167 The '``ret``' instruction optionally accepts a single argument, the
3168 return value. The type of the return value must be a ':ref:`first
3169 class <t_firstclass>`' type.
3171 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3172 return type and contains a '``ret``' instruction with no return value or
3173 a return value with a type that does not match its type, or if it has a
3174 void return type and contains a '``ret``' instruction with a return
3180 When the '``ret``' instruction is executed, control flow returns back to
3181 the calling function's context. If the caller is a
3182 ":ref:`call <i_call>`" instruction, execution continues at the
3183 instruction after the call. If the caller was an
3184 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3185 beginning of the "normal" destination block. If the instruction returns
3186 a value, that value shall set the call or invoke instruction's return
3192 .. code-block:: llvm
3194 ret i32 5 ; Return an integer value of 5
3195 ret void ; Return from a void function
3196 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3200 '``br``' Instruction
3201 ^^^^^^^^^^^^^^^^^^^^
3208 br i1 <cond>, label <iftrue>, label <iffalse>
3209 br label <dest> ; Unconditional branch
3214 The '``br``' instruction is used to cause control flow to transfer to a
3215 different basic block in the current function. There are two forms of
3216 this instruction, corresponding to a conditional branch and an
3217 unconditional branch.
3222 The conditional branch form of the '``br``' instruction takes a single
3223 '``i1``' value and two '``label``' values. The unconditional form of the
3224 '``br``' instruction takes a single '``label``' value as a target.
3229 Upon execution of a conditional '``br``' instruction, the '``i1``'
3230 argument is evaluated. If the value is ``true``, control flows to the
3231 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3232 to the '``iffalse``' ``label`` argument.
3237 .. code-block:: llvm
3240 %cond = icmp eq i32 %a, %b
3241 br i1 %cond, label %IfEqual, label %IfUnequal
3249 '``switch``' Instruction
3250 ^^^^^^^^^^^^^^^^^^^^^^^^
3257 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3262 The '``switch``' instruction is used to transfer control flow to one of
3263 several different places. It is a generalization of the '``br``'
3264 instruction, allowing a branch to occur to one of many possible
3270 The '``switch``' instruction uses three parameters: an integer
3271 comparison value '``value``', a default '``label``' destination, and an
3272 array of pairs of comparison value constants and '``label``'s. The table
3273 is not allowed to contain duplicate constant entries.
3278 The ``switch`` instruction specifies a table of values and destinations.
3279 When the '``switch``' instruction is executed, this table is searched
3280 for the given value. If the value is found, control flow is transferred
3281 to the corresponding destination; otherwise, control flow is transferred
3282 to the default destination.
3287 Depending on properties of the target machine and the particular
3288 ``switch`` instruction, this instruction may be code generated in
3289 different ways. For example, it could be generated as a series of
3290 chained conditional branches or with a lookup table.
3295 .. code-block:: llvm
3297 ; Emulate a conditional br instruction
3298 %Val = zext i1 %value to i32
3299 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3301 ; Emulate an unconditional br instruction
3302 switch i32 0, label %dest [ ]
3304 ; Implement a jump table:
3305 switch i32 %val, label %otherwise [ i32 0, label %onzero
3307 i32 2, label %ontwo ]
3311 '``indirectbr``' Instruction
3312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3319 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3324 The '``indirectbr``' instruction implements an indirect branch to a
3325 label within the current function, whose address is specified by
3326 "``address``". Address must be derived from a
3327 :ref:`blockaddress <blockaddress>` constant.
3332 The '``address``' argument is the address of the label to jump to. The
3333 rest of the arguments indicate the full set of possible destinations
3334 that the address may point to. Blocks are allowed to occur multiple
3335 times in the destination list, though this isn't particularly useful.
3337 This destination list is required so that dataflow analysis has an
3338 accurate understanding of the CFG.
3343 Control transfers to the block specified in the address argument. All
3344 possible destination blocks must be listed in the label list, otherwise
3345 this instruction has undefined behavior. This implies that jumps to
3346 labels defined in other functions have undefined behavior as well.
3351 This is typically implemented with a jump through a register.
3356 .. code-block:: llvm
3358 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3362 '``invoke``' Instruction
3363 ^^^^^^^^^^^^^^^^^^^^^^^^
3370 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3371 to label <normal label> unwind label <exception label>
3376 The '``invoke``' instruction causes control to transfer to a specified
3377 function, with the possibility of control flow transfer to either the
3378 '``normal``' label or the '``exception``' label. If the callee function
3379 returns with the "``ret``" instruction, control flow will return to the
3380 "normal" label. If the callee (or any indirect callees) returns via the
3381 ":ref:`resume <i_resume>`" instruction or other exception handling
3382 mechanism, control is interrupted and continued at the dynamically
3383 nearest "exception" label.
3385 The '``exception``' label is a `landing
3386 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3387 '``exception``' label is required to have the
3388 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3389 information about the behavior of the program after unwinding happens,
3390 as its first non-PHI instruction. The restrictions on the
3391 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3392 instruction, so that the important information contained within the
3393 "``landingpad``" instruction can't be lost through normal code motion.
3398 This instruction requires several arguments:
3400 #. The optional "cconv" marker indicates which :ref:`calling
3401 convention <callingconv>` the call should use. If none is
3402 specified, the call defaults to using C calling conventions.
3403 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3404 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3406 #. '``ptr to function ty``': shall be the signature of the pointer to
3407 function value being invoked. In most cases, this is a direct
3408 function invocation, but indirect ``invoke``'s are just as possible,
3409 branching off an arbitrary pointer to function value.
3410 #. '``function ptr val``': An LLVM value containing a pointer to a
3411 function to be invoked.
3412 #. '``function args``': argument list whose types match the function
3413 signature argument types and parameter attributes. All arguments must
3414 be of :ref:`first class <t_firstclass>` type. If the function signature
3415 indicates the function accepts a variable number of arguments, the
3416 extra arguments can be specified.
3417 #. '``normal label``': the label reached when the called function
3418 executes a '``ret``' instruction.
3419 #. '``exception label``': the label reached when a callee returns via
3420 the :ref:`resume <i_resume>` instruction or other exception handling
3422 #. The optional :ref:`function attributes <fnattrs>` list. Only
3423 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3424 attributes are valid here.
3429 This instruction is designed to operate as a standard '``call``'
3430 instruction in most regards. The primary difference is that it
3431 establishes an association with a label, which is used by the runtime
3432 library to unwind the stack.
3434 This instruction is used in languages with destructors to ensure that
3435 proper cleanup is performed in the case of either a ``longjmp`` or a
3436 thrown exception. Additionally, this is important for implementation of
3437 '``catch``' clauses in high-level languages that support them.
3439 For the purposes of the SSA form, the definition of the value returned
3440 by the '``invoke``' instruction is deemed to occur on the edge from the
3441 current block to the "normal" label. If the callee unwinds then no
3442 return value is available.
3447 .. code-block:: llvm
3449 %retval = invoke i32 @Test(i32 15) to label %Continue
3450 unwind label %TestCleanup ; {i32}:retval set
3451 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3452 unwind label %TestCleanup ; {i32}:retval set
3456 '``resume``' Instruction
3457 ^^^^^^^^^^^^^^^^^^^^^^^^
3464 resume <type> <value>
3469 The '``resume``' instruction is a terminator instruction that has no
3475 The '``resume``' instruction requires one argument, which must have the
3476 same type as the result of any '``landingpad``' instruction in the same
3482 The '``resume``' instruction resumes propagation of an existing
3483 (in-flight) exception whose unwinding was interrupted with a
3484 :ref:`landingpad <i_landingpad>` instruction.
3489 .. code-block:: llvm
3491 resume { i8*, i32 } %exn
3495 '``unreachable``' Instruction
3496 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3508 The '``unreachable``' instruction has no defined semantics. This
3509 instruction is used to inform the optimizer that a particular portion of
3510 the code is not reachable. This can be used to indicate that the code
3511 after a no-return function cannot be reached, and other facts.
3516 The '``unreachable``' instruction has no defined semantics.
3523 Binary operators are used to do most of the computation in a program.
3524 They require two operands of the same type, execute an operation on
3525 them, and produce a single value. The operands might represent multiple
3526 data, as is the case with the :ref:`vector <t_vector>` data type. The
3527 result value has the same type as its operands.
3529 There are several different binary operators:
3533 '``add``' Instruction
3534 ^^^^^^^^^^^^^^^^^^^^^
3541 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3542 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3543 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3544 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3549 The '``add``' instruction returns the sum of its two operands.
3554 The two arguments to the '``add``' instruction must be
3555 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3556 arguments must have identical types.
3561 The value produced is the integer sum of the two operands.
3563 If the sum has unsigned overflow, the result returned is the
3564 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3567 Because LLVM integers use a two's complement representation, this
3568 instruction is appropriate for both signed and unsigned integers.
3570 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3571 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3572 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3573 unsigned and/or signed overflow, respectively, occurs.
3578 .. code-block:: llvm
3580 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3584 '``fadd``' Instruction
3585 ^^^^^^^^^^^^^^^^^^^^^^
3592 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3597 The '``fadd``' instruction returns the sum of its two operands.
3602 The two arguments to the '``fadd``' instruction must be :ref:`floating
3603 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3604 Both arguments must have identical types.
3609 The value produced is the floating point sum of the two operands. This
3610 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3611 which are optimization hints to enable otherwise unsafe floating point
3617 .. code-block:: llvm
3619 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3621 '``sub``' Instruction
3622 ^^^^^^^^^^^^^^^^^^^^^
3629 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3630 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3631 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3632 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3637 The '``sub``' instruction returns the difference of its two operands.
3639 Note that the '``sub``' instruction is used to represent the '``neg``'
3640 instruction present in most other intermediate representations.
3645 The two arguments to the '``sub``' instruction must be
3646 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3647 arguments must have identical types.
3652 The value produced is the integer difference of the two operands.
3654 If the difference has unsigned overflow, the result returned is the
3655 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3658 Because LLVM integers use a two's complement representation, this
3659 instruction is appropriate for both signed and unsigned integers.
3661 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3662 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3663 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3664 unsigned and/or signed overflow, respectively, occurs.
3669 .. code-block:: llvm
3671 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3672 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3676 '``fsub``' Instruction
3677 ^^^^^^^^^^^^^^^^^^^^^^
3684 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3689 The '``fsub``' instruction returns the difference of its two operands.
3691 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3692 instruction present in most other intermediate representations.
3697 The two arguments to the '``fsub``' instruction must be :ref:`floating
3698 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3699 Both arguments must have identical types.
3704 The value produced is the floating point difference of the two operands.
3705 This instruction can also take any number of :ref:`fast-math
3706 flags <fastmath>`, which are optimization hints to enable otherwise
3707 unsafe floating point optimizations:
3712 .. code-block:: llvm
3714 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3715 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3717 '``mul``' Instruction
3718 ^^^^^^^^^^^^^^^^^^^^^
3725 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3726 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3727 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3728 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3733 The '``mul``' instruction returns the product of its two operands.
3738 The two arguments to the '``mul``' instruction must be
3739 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3740 arguments must have identical types.
3745 The value produced is the integer product of the two operands.
3747 If the result of the multiplication has unsigned overflow, the result
3748 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3749 bit width of the result.
3751 Because LLVM integers use a two's complement representation, and the
3752 result is the same width as the operands, this instruction returns the
3753 correct result for both signed and unsigned integers. If a full product
3754 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3755 sign-extended or zero-extended as appropriate to the width of the full
3758 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3759 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3760 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3761 unsigned and/or signed overflow, respectively, occurs.
3766 .. code-block:: llvm
3768 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3772 '``fmul``' Instruction
3773 ^^^^^^^^^^^^^^^^^^^^^^
3780 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3785 The '``fmul``' instruction returns the product of its two operands.
3790 The two arguments to the '``fmul``' instruction must be :ref:`floating
3791 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3792 Both arguments must have identical types.
3797 The value produced is the floating point product of the two operands.
3798 This instruction can also take any number of :ref:`fast-math
3799 flags <fastmath>`, which are optimization hints to enable otherwise
3800 unsafe floating point optimizations:
3805 .. code-block:: llvm
3807 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3809 '``udiv``' Instruction
3810 ^^^^^^^^^^^^^^^^^^^^^^
3817 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3818 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3823 The '``udiv``' instruction returns the quotient of its two operands.
3828 The two arguments to the '``udiv``' instruction must be
3829 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3830 arguments must have identical types.
3835 The value produced is the unsigned integer quotient of the two operands.
3837 Note that unsigned integer division and signed integer division are
3838 distinct operations; for signed integer division, use '``sdiv``'.
3840 Division by zero leads to undefined behavior.
3842 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3843 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3844 such, "((a udiv exact b) mul b) == a").
3849 .. code-block:: llvm
3851 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3853 '``sdiv``' Instruction
3854 ^^^^^^^^^^^^^^^^^^^^^^
3861 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3862 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3867 The '``sdiv``' instruction returns the quotient of its two operands.
3872 The two arguments to the '``sdiv``' instruction must be
3873 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3874 arguments must have identical types.
3879 The value produced is the signed integer quotient of the two operands
3880 rounded towards zero.
3882 Note that signed integer division and unsigned integer division are
3883 distinct operations; for unsigned integer division, use '``udiv``'.
3885 Division by zero leads to undefined behavior. Overflow also leads to
3886 undefined behavior; this is a rare case, but can occur, for example, by
3887 doing a 32-bit division of -2147483648 by -1.
3889 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3890 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3895 .. code-block:: llvm
3897 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3901 '``fdiv``' Instruction
3902 ^^^^^^^^^^^^^^^^^^^^^^
3909 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3914 The '``fdiv``' instruction returns the quotient of its two operands.
3919 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3920 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3921 Both arguments must have identical types.
3926 The value produced is the floating point quotient of the two operands.
3927 This instruction can also take any number of :ref:`fast-math
3928 flags <fastmath>`, which are optimization hints to enable otherwise
3929 unsafe floating point optimizations:
3934 .. code-block:: llvm
3936 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3938 '``urem``' Instruction
3939 ^^^^^^^^^^^^^^^^^^^^^^
3946 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3951 The '``urem``' instruction returns the remainder from the unsigned
3952 division of its two arguments.
3957 The two arguments to the '``urem``' instruction must be
3958 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3959 arguments must have identical types.
3964 This instruction returns the unsigned integer *remainder* of a division.
3965 This instruction always performs an unsigned division to get the
3968 Note that unsigned integer remainder and signed integer remainder are
3969 distinct operations; for signed integer remainder, use '``srem``'.
3971 Taking the remainder of a division by zero leads to undefined behavior.
3976 .. code-block:: llvm
3978 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3980 '``srem``' Instruction
3981 ^^^^^^^^^^^^^^^^^^^^^^
3988 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3993 The '``srem``' instruction returns the remainder from the signed
3994 division of its two operands. This instruction can also take
3995 :ref:`vector <t_vector>` versions of the values in which case the elements
4001 The two arguments to the '``srem``' instruction must be
4002 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4003 arguments must have identical types.
4008 This instruction returns the *remainder* of a division (where the result
4009 is either zero or has the same sign as the dividend, ``op1``), not the
4010 *modulo* operator (where the result is either zero or has the same sign
4011 as the divisor, ``op2``) of a value. For more information about the
4012 difference, see `The Math
4013 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4014 table of how this is implemented in various languages, please see
4016 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4018 Note that signed integer remainder and unsigned integer remainder are
4019 distinct operations; for unsigned integer remainder, use '``urem``'.
4021 Taking the remainder of a division by zero leads to undefined behavior.
4022 Overflow also leads to undefined behavior; this is a rare case, but can
4023 occur, for example, by taking the remainder of a 32-bit division of
4024 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4025 rule lets srem be implemented using instructions that return both the
4026 result of the division and the remainder.)
4031 .. code-block:: llvm
4033 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4037 '``frem``' Instruction
4038 ^^^^^^^^^^^^^^^^^^^^^^
4045 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4050 The '``frem``' instruction returns the remainder from the division of
4056 The two arguments to the '``frem``' instruction must be :ref:`floating
4057 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4058 Both arguments must have identical types.
4063 This instruction returns the *remainder* of a division. The remainder
4064 has the same sign as the dividend. This instruction can also take any
4065 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4066 to enable otherwise unsafe floating point optimizations:
4071 .. code-block:: llvm
4073 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4077 Bitwise Binary Operations
4078 -------------------------
4080 Bitwise binary operators are used to do various forms of bit-twiddling
4081 in a program. They are generally very efficient instructions and can
4082 commonly be strength reduced from other instructions. They require two
4083 operands of the same type, execute an operation on them, and produce a
4084 single value. The resulting value is the same type as its operands.
4086 '``shl``' Instruction
4087 ^^^^^^^^^^^^^^^^^^^^^
4094 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4095 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4096 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4097 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4102 The '``shl``' instruction returns the first operand shifted to the left
4103 a specified number of bits.
4108 Both arguments to the '``shl``' instruction must be the same
4109 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4110 '``op2``' is treated as an unsigned value.
4115 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4116 where ``n`` is the width of the result. If ``op2`` is (statically or
4117 dynamically) negative or equal to or larger than the number of bits in
4118 ``op1``, the result is undefined. If the arguments are vectors, each
4119 vector element of ``op1`` is shifted by the corresponding shift amount
4122 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4123 value <poisonvalues>` if it shifts out any non-zero bits. If the
4124 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4125 value <poisonvalues>` if it shifts out any bits that disagree with the
4126 resultant sign bit. As such, NUW/NSW have the same semantics as they
4127 would if the shift were expressed as a mul instruction with the same
4128 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4133 .. code-block:: llvm
4135 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4136 <result> = shl i32 4, 2 ; yields {i32}: 16
4137 <result> = shl i32 1, 10 ; yields {i32}: 1024
4138 <result> = shl i32 1, 32 ; undefined
4139 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4141 '``lshr``' Instruction
4142 ^^^^^^^^^^^^^^^^^^^^^^
4149 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4150 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4155 The '``lshr``' instruction (logical shift right) returns the first
4156 operand shifted to the right a specified number of bits with zero fill.
4161 Both arguments to the '``lshr``' instruction must be the same
4162 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4163 '``op2``' is treated as an unsigned value.
4168 This instruction always performs a logical shift right operation. The
4169 most significant bits of the result will be filled with zero bits after
4170 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4171 than the number of bits in ``op1``, the result is undefined. If the
4172 arguments are vectors, each vector element of ``op1`` is shifted by the
4173 corresponding shift amount in ``op2``.
4175 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4176 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4182 .. code-block:: llvm
4184 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4185 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4186 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4187 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4188 <result> = lshr i32 1, 32 ; undefined
4189 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4191 '``ashr``' Instruction
4192 ^^^^^^^^^^^^^^^^^^^^^^
4199 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4200 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4205 The '``ashr``' instruction (arithmetic shift right) returns the first
4206 operand shifted to the right a specified number of bits with sign
4212 Both arguments to the '``ashr``' instruction must be the same
4213 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4214 '``op2``' is treated as an unsigned value.
4219 This instruction always performs an arithmetic shift right operation,
4220 The most significant bits of the result will be filled with the sign bit
4221 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4222 than the number of bits in ``op1``, the result is undefined. If the
4223 arguments are vectors, each vector element of ``op1`` is shifted by the
4224 corresponding shift amount in ``op2``.
4226 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4227 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4233 .. code-block:: llvm
4235 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4236 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4237 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4238 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4239 <result> = ashr i32 1, 32 ; undefined
4240 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4242 '``and``' Instruction
4243 ^^^^^^^^^^^^^^^^^^^^^
4250 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4255 The '``and``' instruction returns the bitwise logical and of its two
4261 The two arguments to the '``and``' instruction must be
4262 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4263 arguments must have identical types.
4268 The truth table used for the '``and``' instruction is:
4285 .. code-block:: llvm
4287 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4288 <result> = and i32 15, 40 ; yields {i32}:result = 8
4289 <result> = and i32 4, 8 ; yields {i32}:result = 0
4291 '``or``' Instruction
4292 ^^^^^^^^^^^^^^^^^^^^
4299 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4304 The '``or``' instruction returns the bitwise logical inclusive or of its
4310 The two arguments to the '``or``' instruction must be
4311 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4312 arguments must have identical types.
4317 The truth table used for the '``or``' instruction is:
4336 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4337 <result> = or i32 15, 40 ; yields {i32}:result = 47
4338 <result> = or i32 4, 8 ; yields {i32}:result = 12
4340 '``xor``' Instruction
4341 ^^^^^^^^^^^^^^^^^^^^^
4348 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4353 The '``xor``' instruction returns the bitwise logical exclusive or of
4354 its two operands. The ``xor`` is used to implement the "one's
4355 complement" operation, which is the "~" operator in C.
4360 The two arguments to the '``xor``' instruction must be
4361 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4362 arguments must have identical types.
4367 The truth table used for the '``xor``' instruction is:
4384 .. code-block:: llvm
4386 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4387 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4388 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4389 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4394 LLVM supports several instructions to represent vector operations in a
4395 target-independent manner. These instructions cover the element-access
4396 and vector-specific operations needed to process vectors effectively.
4397 While LLVM does directly support these vector operations, many
4398 sophisticated algorithms will want to use target-specific intrinsics to
4399 take full advantage of a specific target.
4401 .. _i_extractelement:
4403 '``extractelement``' Instruction
4404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4411 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4416 The '``extractelement``' instruction extracts a single scalar element
4417 from a vector at a specified index.
4422 The first operand of an '``extractelement``' instruction is a value of
4423 :ref:`vector <t_vector>` type. The second operand is an index indicating
4424 the position from which to extract the element. The index may be a
4430 The result is a scalar of the same type as the element type of ``val``.
4431 Its value is the value at position ``idx`` of ``val``. If ``idx``
4432 exceeds the length of ``val``, the results are undefined.
4437 .. code-block:: llvm
4439 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4441 .. _i_insertelement:
4443 '``insertelement``' Instruction
4444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4451 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4456 The '``insertelement``' instruction inserts a scalar element into a
4457 vector at a specified index.
4462 The first operand of an '``insertelement``' instruction is a value of
4463 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4464 type must equal the element type of the first operand. The third operand
4465 is an index indicating the position at which to insert the value. The
4466 index may be a variable.
4471 The result is a vector of the same type as ``val``. Its element values
4472 are those of ``val`` except at position ``idx``, where it gets the value
4473 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4479 .. code-block:: llvm
4481 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4483 .. _i_shufflevector:
4485 '``shufflevector``' Instruction
4486 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4493 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4498 The '``shufflevector``' instruction constructs a permutation of elements
4499 from two input vectors, returning a vector with the same element type as
4500 the input and length that is the same as the shuffle mask.
4505 The first two operands of a '``shufflevector``' instruction are vectors
4506 with the same type. The third argument is a shuffle mask whose element
4507 type is always 'i32'. The result of the instruction is a vector whose
4508 length is the same as the shuffle mask and whose element type is the
4509 same as the element type of the first two operands.
4511 The shuffle mask operand is required to be a constant vector with either
4512 constant integer or undef values.
4517 The elements of the two input vectors are numbered from left to right
4518 across both of the vectors. The shuffle mask operand specifies, for each
4519 element of the result vector, which element of the two input vectors the
4520 result element gets. The element selector may be undef (meaning "don't
4521 care") and the second operand may be undef if performing a shuffle from
4527 .. code-block:: llvm
4529 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4530 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4531 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4532 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4533 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4534 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4535 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4536 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4538 Aggregate Operations
4539 --------------------
4541 LLVM supports several instructions for working with
4542 :ref:`aggregate <t_aggregate>` values.
4546 '``extractvalue``' Instruction
4547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4554 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4559 The '``extractvalue``' instruction extracts the value of a member field
4560 from an :ref:`aggregate <t_aggregate>` value.
4565 The first operand of an '``extractvalue``' instruction is a value of
4566 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4567 constant indices to specify which value to extract in a similar manner
4568 as indices in a '``getelementptr``' instruction.
4570 The major differences to ``getelementptr`` indexing are:
4572 - Since the value being indexed is not a pointer, the first index is
4573 omitted and assumed to be zero.
4574 - At least one index must be specified.
4575 - Not only struct indices but also array indices must be in bounds.
4580 The result is the value at the position in the aggregate specified by
4586 .. code-block:: llvm
4588 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4592 '``insertvalue``' Instruction
4593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4600 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4605 The '``insertvalue``' instruction inserts a value into a member field in
4606 an :ref:`aggregate <t_aggregate>` value.
4611 The first operand of an '``insertvalue``' instruction is a value of
4612 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4613 a first-class value to insert. The following operands are constant
4614 indices indicating the position at which to insert the value in a
4615 similar manner as indices in a '``extractvalue``' instruction. The value
4616 to insert must have the same type as the value identified by the
4622 The result is an aggregate of the same type as ``val``. Its value is
4623 that of ``val`` except that the value at the position specified by the
4624 indices is that of ``elt``.
4629 .. code-block:: llvm
4631 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4632 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4633 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4637 Memory Access and Addressing Operations
4638 ---------------------------------------
4640 A key design point of an SSA-based representation is how it represents
4641 memory. In LLVM, no memory locations are in SSA form, which makes things
4642 very simple. This section describes how to read, write, and allocate
4647 '``alloca``' Instruction
4648 ^^^^^^^^^^^^^^^^^^^^^^^^
4655 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4660 The '``alloca``' instruction allocates memory on the stack frame of the
4661 currently executing function, to be automatically released when this
4662 function returns to its caller. The object is always allocated in the
4663 generic address space (address space zero).
4668 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4669 bytes of memory on the runtime stack, returning a pointer of the
4670 appropriate type to the program. If "NumElements" is specified, it is
4671 the number of elements allocated, otherwise "NumElements" is defaulted
4672 to be one. If a constant alignment is specified, the value result of the
4673 allocation is guaranteed to be aligned to at least that boundary. If not
4674 specified, or if zero, the target can choose to align the allocation on
4675 any convenient boundary compatible with the type.
4677 '``type``' may be any sized type.
4682 Memory is allocated; a pointer is returned. The operation is undefined
4683 if there is insufficient stack space for the allocation. '``alloca``'d
4684 memory is automatically released when the function returns. The
4685 '``alloca``' instruction is commonly used to represent automatic
4686 variables that must have an address available. When the function returns
4687 (either with the ``ret`` or ``resume`` instructions), the memory is
4688 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4689 The order in which memory is allocated (ie., which way the stack grows)
4695 .. code-block:: llvm
4697 %ptr = alloca i32 ; yields {i32*}:ptr
4698 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4699 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4700 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4704 '``load``' Instruction
4705 ^^^^^^^^^^^^^^^^^^^^^^
4712 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4713 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4714 !<index> = !{ i32 1 }
4719 The '``load``' instruction is used to read from memory.
4724 The argument to the ``load`` instruction specifies the memory address
4725 from which to load. The pointer must point to a :ref:`first
4726 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4727 then the optimizer is not allowed to modify the number or order of
4728 execution of this ``load`` with other :ref:`volatile
4729 operations <volatile>`.
4731 If the ``load`` is marked as ``atomic``, it takes an extra
4732 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4733 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4734 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4735 when they may see multiple atomic stores. The type of the pointee must
4736 be an integer type whose bit width is a power of two greater than or
4737 equal to eight and less than or equal to a target-specific size limit.
4738 ``align`` must be explicitly specified on atomic loads, and the load has
4739 undefined behavior if the alignment is not set to a value which is at
4740 least the size in bytes of the pointee. ``!nontemporal`` does not have
4741 any defined semantics for atomic loads.
4743 The optional constant ``align`` argument specifies the alignment of the
4744 operation (that is, the alignment of the memory address). A value of 0
4745 or an omitted ``align`` argument means that the operation has the ABI
4746 alignment for the target. It is the responsibility of the code emitter
4747 to ensure that the alignment information is correct. Overestimating the
4748 alignment results in undefined behavior. Underestimating the alignment
4749 may produce less efficient code. An alignment of 1 is always safe.
4751 The optional ``!nontemporal`` metadata must reference a single
4752 metadata name ``<index>`` corresponding to a metadata node with one
4753 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4754 metadata on the instruction tells the optimizer and code generator
4755 that this load is not expected to be reused in the cache. The code
4756 generator may select special instructions to save cache bandwidth, such
4757 as the ``MOVNT`` instruction on x86.
4759 The optional ``!invariant.load`` metadata must reference a single
4760 metadata name ``<index>`` corresponding to a metadata node with no
4761 entries. The existence of the ``!invariant.load`` metadata on the
4762 instruction tells the optimizer and code generator that this load
4763 address points to memory which does not change value during program
4764 execution. The optimizer may then move this load around, for example, by
4765 hoisting it out of loops using loop invariant code motion.
4770 The location of memory pointed to is loaded. If the value being loaded
4771 is of scalar type then the number of bytes read does not exceed the
4772 minimum number of bytes needed to hold all bits of the type. For
4773 example, loading an ``i24`` reads at most three bytes. When loading a
4774 value of a type like ``i20`` with a size that is not an integral number
4775 of bytes, the result is undefined if the value was not originally
4776 written using a store of the same type.
4781 .. code-block:: llvm
4783 %ptr = alloca i32 ; yields {i32*}:ptr
4784 store i32 3, i32* %ptr ; yields {void}
4785 %val = load i32* %ptr ; yields {i32}:val = i32 3
4789 '``store``' Instruction
4790 ^^^^^^^^^^^^^^^^^^^^^^^
4797 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4798 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4803 The '``store``' instruction is used to write to memory.
4808 There are two arguments to the ``store`` instruction: a value to store
4809 and an address at which to store it. The type of the ``<pointer>``
4810 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4811 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4812 then the optimizer is not allowed to modify the number or order of
4813 execution of this ``store`` with other :ref:`volatile
4814 operations <volatile>`.
4816 If the ``store`` is marked as ``atomic``, it takes an extra
4817 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4818 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4819 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4820 when they may see multiple atomic stores. The type of the pointee must
4821 be an integer type whose bit width is a power of two greater than or
4822 equal to eight and less than or equal to a target-specific size limit.
4823 ``align`` must be explicitly specified on atomic stores, and the store
4824 has undefined behavior if the alignment is not set to a value which is
4825 at least the size in bytes of the pointee. ``!nontemporal`` does not
4826 have any defined semantics for atomic stores.
4828 The optional constant ``align`` argument specifies the alignment of the
4829 operation (that is, the alignment of the memory address). A value of 0
4830 or an omitted ``align`` argument means that the operation has the ABI
4831 alignment for the target. It is the responsibility of the code emitter
4832 to ensure that the alignment information is correct. Overestimating the
4833 alignment results in undefined behavior. Underestimating the
4834 alignment may produce less efficient code. An alignment of 1 is always
4837 The optional ``!nontemporal`` metadata must reference a single metadata
4838 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4839 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4840 tells the optimizer and code generator that this load is not expected to
4841 be reused in the cache. The code generator may select special
4842 instructions to save cache bandwidth, such as the MOVNT instruction on
4848 The contents of memory are updated to contain ``<value>`` at the
4849 location specified by the ``<pointer>`` operand. If ``<value>`` is
4850 of scalar type then the number of bytes written does not exceed the
4851 minimum number of bytes needed to hold all bits of the type. For
4852 example, storing an ``i24`` writes at most three bytes. When writing a
4853 value of a type like ``i20`` with a size that is not an integral number
4854 of bytes, it is unspecified what happens to the extra bits that do not
4855 belong to the type, but they will typically be overwritten.
4860 .. code-block:: llvm
4862 %ptr = alloca i32 ; yields {i32*}:ptr
4863 store i32 3, i32* %ptr ; yields {void}
4864 %val = load i32* %ptr ; yields {i32}:val = i32 3
4868 '``fence``' Instruction
4869 ^^^^^^^^^^^^^^^^^^^^^^^
4876 fence [singlethread] <ordering> ; yields {void}
4881 The '``fence``' instruction is used to introduce happens-before edges
4887 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4888 defines what *synchronizes-with* edges they add. They can only be given
4889 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4894 A fence A which has (at least) ``release`` ordering semantics
4895 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4896 semantics if and only if there exist atomic operations X and Y, both
4897 operating on some atomic object M, such that A is sequenced before X, X
4898 modifies M (either directly or through some side effect of a sequence
4899 headed by X), Y is sequenced before B, and Y observes M. This provides a
4900 *happens-before* dependency between A and B. Rather than an explicit
4901 ``fence``, one (but not both) of the atomic operations X or Y might
4902 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4903 still *synchronize-with* the explicit ``fence`` and establish the
4904 *happens-before* edge.
4906 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4907 ``acquire`` and ``release`` semantics specified above, participates in
4908 the global program order of other ``seq_cst`` operations and/or fences.
4910 The optional ":ref:`singlethread <singlethread>`" argument specifies
4911 that the fence only synchronizes with other fences in the same thread.
4912 (This is useful for interacting with signal handlers.)
4917 .. code-block:: llvm
4919 fence acquire ; yields {void}
4920 fence singlethread seq_cst ; yields {void}
4924 '``cmpxchg``' Instruction
4925 ^^^^^^^^^^^^^^^^^^^^^^^^^
4932 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4937 The '``cmpxchg``' instruction is used to atomically modify memory. It
4938 loads a value in memory and compares it to a given value. If they are
4939 equal, it stores a new value into the memory.
4944 There are three arguments to the '``cmpxchg``' instruction: an address
4945 to operate on, a value to compare to the value currently be at that
4946 address, and a new value to place at that address if the compared values
4947 are equal. The type of '<cmp>' must be an integer type whose bit width
4948 is a power of two greater than or equal to eight and less than or equal
4949 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4950 type, and the type of '<pointer>' must be a pointer to that type. If the
4951 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4952 to modify the number or order of execution of this ``cmpxchg`` with
4953 other :ref:`volatile operations <volatile>`.
4955 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4956 synchronizes with other atomic operations.
4958 The optional "``singlethread``" argument declares that the ``cmpxchg``
4959 is only atomic with respect to code (usually signal handlers) running in
4960 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4961 respect to all other code in the system.
4963 The pointer passed into cmpxchg must have alignment greater than or
4964 equal to the size in memory of the operand.
4969 The contents of memory at the location specified by the '``<pointer>``'
4970 operand is read and compared to '``<cmp>``'; if the read value is the
4971 equal, '``<new>``' is written. The original value at the location is
4974 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4975 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4976 atomic load with an ordering parameter determined by dropping any
4977 ``release`` part of the ``cmpxchg``'s ordering.
4982 .. code-block:: llvm
4985 %orig = atomic load i32* %ptr unordered ; yields {i32}
4989 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4990 %squared = mul i32 %cmp, %cmp
4991 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4992 %success = icmp eq i32 %cmp, %old
4993 br i1 %success, label %done, label %loop
5000 '``atomicrmw``' Instruction
5001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5008 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5013 The '``atomicrmw``' instruction is used to atomically modify memory.
5018 There are three arguments to the '``atomicrmw``' instruction: an
5019 operation to apply, an address whose value to modify, an argument to the
5020 operation. The operation must be one of the following keywords:
5034 The type of '<value>' must be an integer type whose bit width is a power
5035 of two greater than or equal to eight and less than or equal to a
5036 target-specific size limit. The type of the '``<pointer>``' operand must
5037 be a pointer to that type. If the ``atomicrmw`` is marked as
5038 ``volatile``, then the optimizer is not allowed to modify the number or
5039 order of execution of this ``atomicrmw`` with other :ref:`volatile
5040 operations <volatile>`.
5045 The contents of memory at the location specified by the '``<pointer>``'
5046 operand are atomically read, modified, and written back. The original
5047 value at the location is returned. The modification is specified by the
5050 - xchg: ``*ptr = val``
5051 - add: ``*ptr = *ptr + val``
5052 - sub: ``*ptr = *ptr - val``
5053 - and: ``*ptr = *ptr & val``
5054 - nand: ``*ptr = ~(*ptr & val)``
5055 - or: ``*ptr = *ptr | val``
5056 - xor: ``*ptr = *ptr ^ val``
5057 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5058 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5059 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5061 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5067 .. code-block:: llvm
5069 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5071 .. _i_getelementptr:
5073 '``getelementptr``' Instruction
5074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5081 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5082 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5083 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5088 The '``getelementptr``' instruction is used to get the address of a
5089 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5090 address calculation only and does not access memory.
5095 The first argument is always a pointer or a vector of pointers, and
5096 forms the basis of the calculation. The remaining arguments are indices
5097 that indicate which of the elements of the aggregate object are indexed.
5098 The interpretation of each index is dependent on the type being indexed
5099 into. The first index always indexes the pointer value given as the
5100 first argument, the second index indexes a value of the type pointed to
5101 (not necessarily the value directly pointed to, since the first index
5102 can be non-zero), etc. The first type indexed into must be a pointer
5103 value, subsequent types can be arrays, vectors, and structs. Note that
5104 subsequent types being indexed into can never be pointers, since that
5105 would require loading the pointer before continuing calculation.
5107 The type of each index argument depends on the type it is indexing into.
5108 When indexing into a (optionally packed) structure, only ``i32`` integer
5109 **constants** are allowed (when using a vector of indices they must all
5110 be the **same** ``i32`` integer constant). When indexing into an array,
5111 pointer or vector, integers of any width are allowed, and they are not
5112 required to be constant. These integers are treated as signed values
5115 For example, let's consider a C code fragment and how it gets compiled
5131 int *foo(struct ST *s) {
5132 return &s[1].Z.B[5][13];
5135 The LLVM code generated by Clang is:
5137 .. code-block:: llvm
5139 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5140 %struct.ST = type { i32, double, %struct.RT }
5142 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5144 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5151 In the example above, the first index is indexing into the
5152 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5153 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5154 indexes into the third element of the structure, yielding a
5155 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5156 structure. The third index indexes into the second element of the
5157 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5158 dimensions of the array are subscripted into, yielding an '``i32``'
5159 type. The '``getelementptr``' instruction returns a pointer to this
5160 element, thus computing a value of '``i32*``' type.
5162 Note that it is perfectly legal to index partially through a structure,
5163 returning a pointer to an inner element. Because of this, the LLVM code
5164 for the given testcase is equivalent to:
5166 .. code-block:: llvm
5168 define i32* @foo(%struct.ST* %s) {
5169 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5170 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5171 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5172 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5173 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5177 If the ``inbounds`` keyword is present, the result value of the
5178 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5179 pointer is not an *in bounds* address of an allocated object, or if any
5180 of the addresses that would be formed by successive addition of the
5181 offsets implied by the indices to the base address with infinitely
5182 precise signed arithmetic are not an *in bounds* address of that
5183 allocated object. The *in bounds* addresses for an allocated object are
5184 all the addresses that point into the object, plus the address one byte
5185 past the end. In cases where the base is a vector of pointers the
5186 ``inbounds`` keyword applies to each of the computations element-wise.
5188 If the ``inbounds`` keyword is not present, the offsets are added to the
5189 base address with silently-wrapping two's complement arithmetic. If the
5190 offsets have a different width from the pointer, they are sign-extended
5191 or truncated to the width of the pointer. The result value of the
5192 ``getelementptr`` may be outside the object pointed to by the base
5193 pointer. The result value may not necessarily be used to access memory
5194 though, even if it happens to point into allocated storage. See the
5195 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5198 The getelementptr instruction is often confusing. For some more insight
5199 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5204 .. code-block:: llvm
5206 ; yields [12 x i8]*:aptr
5207 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5209 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5211 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5213 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5215 In cases where the pointer argument is a vector of pointers, each index
5216 must be a vector with the same number of elements. For example:
5218 .. code-block:: llvm
5220 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5222 Conversion Operations
5223 ---------------------
5225 The instructions in this category are the conversion instructions
5226 (casting) which all take a single operand and a type. They perform
5227 various bit conversions on the operand.
5229 '``trunc .. to``' Instruction
5230 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5237 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5242 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5247 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5248 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5249 of the same number of integers. The bit size of the ``value`` must be
5250 larger than the bit size of the destination type, ``ty2``. Equal sized
5251 types are not allowed.
5256 The '``trunc``' instruction truncates the high order bits in ``value``
5257 and converts the remaining bits to ``ty2``. Since the source size must
5258 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5259 It will always truncate bits.
5264 .. code-block:: llvm
5266 %X = trunc i32 257 to i8 ; yields i8:1
5267 %Y = trunc i32 123 to i1 ; yields i1:true
5268 %Z = trunc i32 122 to i1 ; yields i1:false
5269 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5271 '``zext .. to``' Instruction
5272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5279 <result> = zext <ty> <value> to <ty2> ; yields ty2
5284 The '``zext``' instruction zero extends its operand to type ``ty2``.
5289 The '``zext``' instruction takes a value to cast, and a type to cast it
5290 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5291 the same number of integers. The bit size of the ``value`` must be
5292 smaller than the bit size of the destination type, ``ty2``.
5297 The ``zext`` fills the high order bits of the ``value`` with zero bits
5298 until it reaches the size of the destination type, ``ty2``.
5300 When zero extending from i1, the result will always be either 0 or 1.
5305 .. code-block:: llvm
5307 %X = zext i32 257 to i64 ; yields i64:257
5308 %Y = zext i1 true to i32 ; yields i32:1
5309 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5311 '``sext .. to``' Instruction
5312 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5319 <result> = sext <ty> <value> to <ty2> ; yields ty2
5324 The '``sext``' sign extends ``value`` to the type ``ty2``.
5329 The '``sext``' instruction takes a value to cast, and a type to cast it
5330 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5331 the same number of integers. The bit size of the ``value`` must be
5332 smaller than the bit size of the destination type, ``ty2``.
5337 The '``sext``' instruction performs a sign extension by copying the sign
5338 bit (highest order bit) of the ``value`` until it reaches the bit size
5339 of the type ``ty2``.
5341 When sign extending from i1, the extension always results in -1 or 0.
5346 .. code-block:: llvm
5348 %X = sext i8 -1 to i16 ; yields i16 :65535
5349 %Y = sext i1 true to i32 ; yields i32:-1
5350 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5352 '``fptrunc .. to``' Instruction
5353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5360 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5365 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5370 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5371 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5372 The size of ``value`` must be larger than the size of ``ty2``. This
5373 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5378 The '``fptrunc``' instruction truncates a ``value`` from a larger
5379 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5380 point <t_floating>` type. If the value cannot fit within the
5381 destination type, ``ty2``, then the results are undefined.
5386 .. code-block:: llvm
5388 %X = fptrunc double 123.0 to float ; yields float:123.0
5389 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5391 '``fpext .. to``' Instruction
5392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5399 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5404 The '``fpext``' extends a floating point ``value`` to a larger floating
5410 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5411 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5412 to. The source type must be smaller than the destination type.
5417 The '``fpext``' instruction extends the ``value`` from a smaller
5418 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5419 point <t_floating>` type. The ``fpext`` cannot be used to make a
5420 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5421 *no-op cast* for a floating point cast.
5426 .. code-block:: llvm
5428 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5429 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5431 '``fptoui .. to``' Instruction
5432 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5439 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5444 The '``fptoui``' converts a floating point ``value`` to its unsigned
5445 integer equivalent of type ``ty2``.
5450 The '``fptoui``' instruction takes a value to cast, which must be a
5451 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5452 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5453 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5454 type with the same number of elements as ``ty``
5459 The '``fptoui``' instruction converts its :ref:`floating
5460 point <t_floating>` operand into the nearest (rounding towards zero)
5461 unsigned integer value. If the value cannot fit in ``ty2``, the results
5467 .. code-block:: llvm
5469 %X = fptoui double 123.0 to i32 ; yields i32:123
5470 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5471 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5473 '``fptosi .. to``' Instruction
5474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5481 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5486 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5487 ``value`` to type ``ty2``.
5492 The '``fptosi``' instruction takes a value to cast, which must be a
5493 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5494 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5495 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5496 type with the same number of elements as ``ty``
5501 The '``fptosi``' instruction converts its :ref:`floating
5502 point <t_floating>` operand into the nearest (rounding towards zero)
5503 signed integer value. If the value cannot fit in ``ty2``, the results
5509 .. code-block:: llvm
5511 %X = fptosi double -123.0 to i32 ; yields i32:-123
5512 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5513 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5515 '``uitofp .. to``' Instruction
5516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5523 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5528 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5529 and converts that value to the ``ty2`` type.
5534 The '``uitofp``' instruction takes a value to cast, which must be a
5535 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5536 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5537 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5538 type with the same number of elements as ``ty``
5543 The '``uitofp``' instruction interprets its operand as an unsigned
5544 integer quantity and converts it to the corresponding floating point
5545 value. If the value cannot fit in the floating point value, the results
5551 .. code-block:: llvm
5553 %X = uitofp i32 257 to float ; yields float:257.0
5554 %Y = uitofp i8 -1 to double ; yields double:255.0
5556 '``sitofp .. to``' Instruction
5557 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5564 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5569 The '``sitofp``' instruction regards ``value`` as a signed integer and
5570 converts that value to the ``ty2`` type.
5575 The '``sitofp``' instruction takes a value to cast, which must be a
5576 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5577 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5578 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5579 type with the same number of elements as ``ty``
5584 The '``sitofp``' instruction interprets its operand as a signed integer
5585 quantity and converts it to the corresponding floating point value. If
5586 the value cannot fit in the floating point value, the results are
5592 .. code-block:: llvm
5594 %X = sitofp i32 257 to float ; yields float:257.0
5595 %Y = sitofp i8 -1 to double ; yields double:-1.0
5599 '``ptrtoint .. to``' Instruction
5600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5607 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5612 The '``ptrtoint``' instruction converts the pointer or a vector of
5613 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5618 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5619 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5620 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5621 a vector of integers type.
5626 The '``ptrtoint``' instruction converts ``value`` to integer type
5627 ``ty2`` by interpreting the pointer value as an integer and either
5628 truncating or zero extending that value to the size of the integer type.
5629 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5630 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5631 the same size, then nothing is done (*no-op cast*) other than a type
5637 .. code-block:: llvm
5639 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5640 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5641 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5645 '``inttoptr .. to``' Instruction
5646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5653 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5658 The '``inttoptr``' instruction converts an integer ``value`` to a
5659 pointer type, ``ty2``.
5664 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5665 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5671 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5672 applying either a zero extension or a truncation depending on the size
5673 of the integer ``value``. If ``value`` is larger than the size of a
5674 pointer then a truncation is done. If ``value`` is smaller than the size
5675 of a pointer then a zero extension is done. If they are the same size,
5676 nothing is done (*no-op cast*).
5681 .. code-block:: llvm
5683 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5684 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5685 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5686 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5690 '``bitcast .. to``' Instruction
5691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5698 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5703 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5709 The '``bitcast``' instruction takes a value to cast, which must be a
5710 non-aggregate first class value, and a type to cast it to, which must
5711 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5712 bit sizes of ``value`` and the destination type, ``ty2``, must be
5713 identical. If the source type is a pointer, the destination type must
5714 also be a pointer of the same size. This instruction supports bitwise
5715 conversion of vectors to integers and to vectors of other types (as
5716 long as they have the same size).
5721 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5722 is always a *no-op cast* because no bits change with this
5723 conversion. The conversion is done as if the ``value`` had been stored
5724 to memory and read back as type ``ty2``. Pointer (or vector of
5725 pointers) types may only be converted to other pointer (or vector of
5726 pointers) types with this instruction if the pointer sizes are
5727 equal. To convert pointers to other types, use the :ref:`inttoptr
5728 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5733 .. code-block:: llvm
5735 %X = bitcast i8 255 to i8 ; yields i8 :-1
5736 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5737 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5738 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5745 The instructions in this category are the "miscellaneous" instructions,
5746 which defy better classification.
5750 '``icmp``' Instruction
5751 ^^^^^^^^^^^^^^^^^^^^^^
5758 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5763 The '``icmp``' instruction returns a boolean value or a vector of
5764 boolean values based on comparison of its two integer, integer vector,
5765 pointer, or pointer vector operands.
5770 The '``icmp``' instruction takes three operands. The first operand is
5771 the condition code indicating the kind of comparison to perform. It is
5772 not a value, just a keyword. The possible condition code are:
5775 #. ``ne``: not equal
5776 #. ``ugt``: unsigned greater than
5777 #. ``uge``: unsigned greater or equal
5778 #. ``ult``: unsigned less than
5779 #. ``ule``: unsigned less or equal
5780 #. ``sgt``: signed greater than
5781 #. ``sge``: signed greater or equal
5782 #. ``slt``: signed less than
5783 #. ``sle``: signed less or equal
5785 The remaining two arguments must be :ref:`integer <t_integer>` or
5786 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5787 must also be identical types.
5792 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5793 code given as ``cond``. The comparison performed always yields either an
5794 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5796 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5797 otherwise. No sign interpretation is necessary or performed.
5798 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5799 otherwise. No sign interpretation is necessary or performed.
5800 #. ``ugt``: interprets the operands as unsigned values and yields
5801 ``true`` if ``op1`` is greater than ``op2``.
5802 #. ``uge``: interprets the operands as unsigned values and yields
5803 ``true`` if ``op1`` is greater than or equal to ``op2``.
5804 #. ``ult``: interprets the operands as unsigned values and yields
5805 ``true`` if ``op1`` is less than ``op2``.
5806 #. ``ule``: interprets the operands as unsigned values and yields
5807 ``true`` if ``op1`` is less than or equal to ``op2``.
5808 #. ``sgt``: interprets the operands as signed values and yields ``true``
5809 if ``op1`` is greater than ``op2``.
5810 #. ``sge``: interprets the operands as signed values and yields ``true``
5811 if ``op1`` is greater than or equal to ``op2``.
5812 #. ``slt``: interprets the operands as signed values and yields ``true``
5813 if ``op1`` is less than ``op2``.
5814 #. ``sle``: interprets the operands as signed values and yields ``true``
5815 if ``op1`` is less than or equal to ``op2``.
5817 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5818 are compared as if they were integers.
5820 If the operands are integer vectors, then they are compared element by
5821 element. The result is an ``i1`` vector with the same number of elements
5822 as the values being compared. Otherwise, the result is an ``i1``.
5827 .. code-block:: llvm
5829 <result> = icmp eq i32 4, 5 ; yields: result=false
5830 <result> = icmp ne float* %X, %X ; yields: result=false
5831 <result> = icmp ult i16 4, 5 ; yields: result=true
5832 <result> = icmp sgt i16 4, 5 ; yields: result=false
5833 <result> = icmp ule i16 -4, 5 ; yields: result=false
5834 <result> = icmp sge i16 4, 5 ; yields: result=false
5836 Note that the code generator does not yet support vector types with the
5837 ``icmp`` instruction.
5841 '``fcmp``' Instruction
5842 ^^^^^^^^^^^^^^^^^^^^^^
5849 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5854 The '``fcmp``' instruction returns a boolean value or vector of boolean
5855 values based on comparison of its operands.
5857 If the operands are floating point scalars, then the result type is a
5858 boolean (:ref:`i1 <t_integer>`).
5860 If the operands are floating point vectors, then the result type is a
5861 vector of boolean with the same number of elements as the operands being
5867 The '``fcmp``' instruction takes three operands. The first operand is
5868 the condition code indicating the kind of comparison to perform. It is
5869 not a value, just a keyword. The possible condition code are:
5871 #. ``false``: no comparison, always returns false
5872 #. ``oeq``: ordered and equal
5873 #. ``ogt``: ordered and greater than
5874 #. ``oge``: ordered and greater than or equal
5875 #. ``olt``: ordered and less than
5876 #. ``ole``: ordered and less than or equal
5877 #. ``one``: ordered and not equal
5878 #. ``ord``: ordered (no nans)
5879 #. ``ueq``: unordered or equal
5880 #. ``ugt``: unordered or greater than
5881 #. ``uge``: unordered or greater than or equal
5882 #. ``ult``: unordered or less than
5883 #. ``ule``: unordered or less than or equal
5884 #. ``une``: unordered or not equal
5885 #. ``uno``: unordered (either nans)
5886 #. ``true``: no comparison, always returns true
5888 *Ordered* means that neither operand is a QNAN while *unordered* means
5889 that either operand may be a QNAN.
5891 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5892 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5893 type. They must have identical types.
5898 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5899 condition code given as ``cond``. If the operands are vectors, then the
5900 vectors are compared element by element. Each comparison performed
5901 always yields an :ref:`i1 <t_integer>` result, as follows:
5903 #. ``false``: always yields ``false``, regardless of operands.
5904 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5905 is equal to ``op2``.
5906 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5907 is greater than ``op2``.
5908 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5909 is greater than or equal to ``op2``.
5910 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5911 is less than ``op2``.
5912 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5913 is less than or equal to ``op2``.
5914 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5915 is not equal to ``op2``.
5916 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5917 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5919 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5920 greater than ``op2``.
5921 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5922 greater than or equal to ``op2``.
5923 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5925 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5926 less than or equal to ``op2``.
5927 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5928 not equal to ``op2``.
5929 #. ``uno``: yields ``true`` if either operand is a QNAN.
5930 #. ``true``: always yields ``true``, regardless of operands.
5935 .. code-block:: llvm
5937 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5938 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5939 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5940 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5942 Note that the code generator does not yet support vector types with the
5943 ``fcmp`` instruction.
5947 '``phi``' Instruction
5948 ^^^^^^^^^^^^^^^^^^^^^
5955 <result> = phi <ty> [ <val0>, <label0>], ...
5960 The '``phi``' instruction is used to implement the φ node in the SSA
5961 graph representing the function.
5966 The type of the incoming values is specified with the first type field.
5967 After this, the '``phi``' instruction takes a list of pairs as
5968 arguments, with one pair for each predecessor basic block of the current
5969 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5970 the value arguments to the PHI node. Only labels may be used as the
5973 There must be no non-phi instructions between the start of a basic block
5974 and the PHI instructions: i.e. PHI instructions must be first in a basic
5977 For the purposes of the SSA form, the use of each incoming value is
5978 deemed to occur on the edge from the corresponding predecessor block to
5979 the current block (but after any definition of an '``invoke``'
5980 instruction's return value on the same edge).
5985 At runtime, the '``phi``' instruction logically takes on the value
5986 specified by the pair corresponding to the predecessor basic block that
5987 executed just prior to the current block.
5992 .. code-block:: llvm
5994 Loop: ; Infinite loop that counts from 0 on up...
5995 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5996 %nextindvar = add i32 %indvar, 1
6001 '``select``' Instruction
6002 ^^^^^^^^^^^^^^^^^^^^^^^^
6009 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6011 selty is either i1 or {<N x i1>}
6016 The '``select``' instruction is used to choose one value based on a
6017 condition, without branching.
6022 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6023 values indicating the condition, and two values of the same :ref:`first
6024 class <t_firstclass>` type. If the val1/val2 are vectors and the
6025 condition is a scalar, then entire vectors are selected, not individual
6031 If the condition is an i1 and it evaluates to 1, the instruction returns
6032 the first value argument; otherwise, it returns the second value
6035 If the condition is a vector of i1, then the value arguments must be
6036 vectors of the same size, and the selection is done element by element.
6041 .. code-block:: llvm
6043 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6047 '``call``' Instruction
6048 ^^^^^^^^^^^^^^^^^^^^^^
6055 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6060 The '``call``' instruction represents a simple function call.
6065 This instruction requires several arguments:
6067 #. The optional "tail" marker indicates that the callee function does
6068 not access any allocas or varargs in the caller. Note that calls may
6069 be marked "tail" even if they do not occur before a
6070 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6071 function call is eligible for tail call optimization, but `might not
6072 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6073 The code generator may optimize calls marked "tail" with either 1)
6074 automatic `sibling call
6075 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6076 callee have matching signatures, or 2) forced tail call optimization
6077 when the following extra requirements are met:
6079 - Caller and callee both have the calling convention ``fastcc``.
6080 - The call is in tail position (ret immediately follows call and ret
6081 uses value of call or is void).
6082 - Option ``-tailcallopt`` is enabled, or
6083 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6084 - `Platform specific constraints are
6085 met. <CodeGenerator.html#tailcallopt>`_
6087 #. The optional "cconv" marker indicates which :ref:`calling
6088 convention <callingconv>` the call should use. If none is
6089 specified, the call defaults to using C calling conventions. The
6090 calling convention of the call must match the calling convention of
6091 the target function, or else the behavior is undefined.
6092 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6093 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6095 #. '``ty``': the type of the call instruction itself which is also the
6096 type of the return value. Functions that return no value are marked
6098 #. '``fnty``': shall be the signature of the pointer to function value
6099 being invoked. The argument types must match the types implied by
6100 this signature. This type can be omitted if the function is not
6101 varargs and if the function type does not return a pointer to a
6103 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6104 be invoked. In most cases, this is a direct function invocation, but
6105 indirect ``call``'s are just as possible, calling an arbitrary pointer
6107 #. '``function args``': argument list whose types match the function
6108 signature argument types and parameter attributes. All arguments must
6109 be of :ref:`first class <t_firstclass>` type. If the function signature
6110 indicates the function accepts a variable number of arguments, the
6111 extra arguments can be specified.
6112 #. The optional :ref:`function attributes <fnattrs>` list. Only
6113 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6114 attributes are valid here.
6119 The '``call``' instruction is used to cause control flow to transfer to
6120 a specified function, with its incoming arguments bound to the specified
6121 values. Upon a '``ret``' instruction in the called function, control
6122 flow continues with the instruction after the function call, and the
6123 return value of the function is bound to the result argument.
6128 .. code-block:: llvm
6130 %retval = call i32 @test(i32 %argc)
6131 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6132 %X = tail call i32 @foo() ; yields i32
6133 %Y = tail call fastcc i32 @foo() ; yields i32
6134 call void %foo(i8 97 signext)
6136 %struct.A = type { i32, i8 }
6137 %r = call %struct.A @foo() ; yields { 32, i8 }
6138 %gr = extractvalue %struct.A %r, 0 ; yields i32
6139 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6140 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6141 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6143 llvm treats calls to some functions with names and arguments that match
6144 the standard C99 library as being the C99 library functions, and may
6145 perform optimizations or generate code for them under that assumption.
6146 This is something we'd like to change in the future to provide better
6147 support for freestanding environments and non-C-based languages.
6151 '``va_arg``' Instruction
6152 ^^^^^^^^^^^^^^^^^^^^^^^^
6159 <resultval> = va_arg <va_list*> <arglist>, <argty>
6164 The '``va_arg``' instruction is used to access arguments passed through
6165 the "variable argument" area of a function call. It is used to implement
6166 the ``va_arg`` macro in C.
6171 This instruction takes a ``va_list*`` value and the type of the
6172 argument. It returns a value of the specified argument type and
6173 increments the ``va_list`` to point to the next argument. The actual
6174 type of ``va_list`` is target specific.
6179 The '``va_arg``' instruction loads an argument of the specified type
6180 from the specified ``va_list`` and causes the ``va_list`` to point to
6181 the next argument. For more information, see the variable argument
6182 handling :ref:`Intrinsic Functions <int_varargs>`.
6184 It is legal for this instruction to be called in a function which does
6185 not take a variable number of arguments, for example, the ``vfprintf``
6188 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6189 function <intrinsics>` because it takes a type as an argument.
6194 See the :ref:`variable argument processing <int_varargs>` section.
6196 Note that the code generator does not yet fully support va\_arg on many
6197 targets. Also, it does not currently support va\_arg with aggregate
6198 types on any target.
6202 '``landingpad``' Instruction
6203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6210 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6211 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6213 <clause> := catch <type> <value>
6214 <clause> := filter <array constant type> <array constant>
6219 The '``landingpad``' instruction is used by `LLVM's exception handling
6220 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6221 is a landing pad --- one where the exception lands, and corresponds to the
6222 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6223 defines values supplied by the personality function (``pers_fn``) upon
6224 re-entry to the function. The ``resultval`` has the type ``resultty``.
6229 This instruction takes a ``pers_fn`` value. This is the personality
6230 function associated with the unwinding mechanism. The optional
6231 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6233 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6234 contains the global variable representing the "type" that may be caught
6235 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6236 clause takes an array constant as its argument. Use
6237 "``[0 x i8**] undef``" for a filter which cannot throw. The
6238 '``landingpad``' instruction must contain *at least* one ``clause`` or
6239 the ``cleanup`` flag.
6244 The '``landingpad``' instruction defines the values which are set by the
6245 personality function (``pers_fn``) upon re-entry to the function, and
6246 therefore the "result type" of the ``landingpad`` instruction. As with
6247 calling conventions, how the personality function results are
6248 represented in LLVM IR is target specific.
6250 The clauses are applied in order from top to bottom. If two
6251 ``landingpad`` instructions are merged together through inlining, the
6252 clauses from the calling function are appended to the list of clauses.
6253 When the call stack is being unwound due to an exception being thrown,
6254 the exception is compared against each ``clause`` in turn. If it doesn't
6255 match any of the clauses, and the ``cleanup`` flag is not set, then
6256 unwinding continues further up the call stack.
6258 The ``landingpad`` instruction has several restrictions:
6260 - A landing pad block is a basic block which is the unwind destination
6261 of an '``invoke``' instruction.
6262 - A landing pad block must have a '``landingpad``' instruction as its
6263 first non-PHI instruction.
6264 - There can be only one '``landingpad``' instruction within the landing
6266 - A basic block that is not a landing pad block may not include a
6267 '``landingpad``' instruction.
6268 - All '``landingpad``' instructions in a function must have the same
6269 personality function.
6274 .. code-block:: llvm
6276 ;; A landing pad which can catch an integer.
6277 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6279 ;; A landing pad that is a cleanup.
6280 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6282 ;; A landing pad which can catch an integer and can only throw a double.
6283 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6285 filter [1 x i8**] [@_ZTId]
6292 LLVM supports the notion of an "intrinsic function". These functions
6293 have well known names and semantics and are required to follow certain
6294 restrictions. Overall, these intrinsics represent an extension mechanism
6295 for the LLVM language that does not require changing all of the
6296 transformations in LLVM when adding to the language (or the bitcode
6297 reader/writer, the parser, etc...).
6299 Intrinsic function names must all start with an "``llvm.``" prefix. This
6300 prefix is reserved in LLVM for intrinsic names; thus, function names may
6301 not begin with this prefix. Intrinsic functions must always be external
6302 functions: you cannot define the body of intrinsic functions. Intrinsic
6303 functions may only be used in call or invoke instructions: it is illegal
6304 to take the address of an intrinsic function. Additionally, because
6305 intrinsic functions are part of the LLVM language, it is required if any
6306 are added that they be documented here.
6308 Some intrinsic functions can be overloaded, i.e., the intrinsic
6309 represents a family of functions that perform the same operation but on
6310 different data types. Because LLVM can represent over 8 million
6311 different integer types, overloading is used commonly to allow an
6312 intrinsic function to operate on any integer type. One or more of the
6313 argument types or the result type can be overloaded to accept any
6314 integer type. Argument types may also be defined as exactly matching a
6315 previous argument's type or the result type. This allows an intrinsic
6316 function which accepts multiple arguments, but needs all of them to be
6317 of the same type, to only be overloaded with respect to a single
6318 argument or the result.
6320 Overloaded intrinsics will have the names of its overloaded argument
6321 types encoded into its function name, each preceded by a period. Only
6322 those types which are overloaded result in a name suffix. Arguments
6323 whose type is matched against another type do not. For example, the
6324 ``llvm.ctpop`` function can take an integer of any width and returns an
6325 integer of exactly the same integer width. This leads to a family of
6326 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6327 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6328 overloaded, and only one type suffix is required. Because the argument's
6329 type is matched against the return type, it does not require its own
6332 To learn how to add an intrinsic function, please see the `Extending
6333 LLVM Guide <ExtendingLLVM.html>`_.
6337 Variable Argument Handling Intrinsics
6338 -------------------------------------
6340 Variable argument support is defined in LLVM with the
6341 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6342 functions. These functions are related to the similarly named macros
6343 defined in the ``<stdarg.h>`` header file.
6345 All of these functions operate on arguments that use a target-specific
6346 value type "``va_list``". The LLVM assembly language reference manual
6347 does not define what this type is, so all transformations should be
6348 prepared to handle these functions regardless of the type used.
6350 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6351 variable argument handling intrinsic functions are used.
6353 .. code-block:: llvm
6355 define i32 @test(i32 %X, ...) {
6356 ; Initialize variable argument processing
6358 %ap2 = bitcast i8** %ap to i8*
6359 call void @llvm.va_start(i8* %ap2)
6361 ; Read a single integer argument
6362 %tmp = va_arg i8** %ap, i32
6364 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6366 %aq2 = bitcast i8** %aq to i8*
6367 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6368 call void @llvm.va_end(i8* %aq2)
6370 ; Stop processing of arguments.
6371 call void @llvm.va_end(i8* %ap2)
6375 declare void @llvm.va_start(i8*)
6376 declare void @llvm.va_copy(i8*, i8*)
6377 declare void @llvm.va_end(i8*)
6381 '``llvm.va_start``' Intrinsic
6382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6389 declare void @llvm.va_start(i8* <arglist>)
6394 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6395 subsequent use by ``va_arg``.
6400 The argument is a pointer to a ``va_list`` element to initialize.
6405 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6406 available in C. In a target-dependent way, it initializes the
6407 ``va_list`` element to which the argument points, so that the next call
6408 to ``va_arg`` will produce the first variable argument passed to the
6409 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6410 to know the last argument of the function as the compiler can figure
6413 '``llvm.va_end``' Intrinsic
6414 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6421 declare void @llvm.va_end(i8* <arglist>)
6426 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6427 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6432 The argument is a pointer to a ``va_list`` to destroy.
6437 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6438 available in C. In a target-dependent way, it destroys the ``va_list``
6439 element to which the argument points. Calls to
6440 :ref:`llvm.va_start <int_va_start>` and
6441 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6446 '``llvm.va_copy``' Intrinsic
6447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6454 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6459 The '``llvm.va_copy``' intrinsic copies the current argument position
6460 from the source argument list to the destination argument list.
6465 The first argument is a pointer to a ``va_list`` element to initialize.
6466 The second argument is a pointer to a ``va_list`` element to copy from.
6471 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6472 available in C. In a target-dependent way, it copies the source
6473 ``va_list`` element into the destination ``va_list`` element. This
6474 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6475 arbitrarily complex and require, for example, memory allocation.
6477 Accurate Garbage Collection Intrinsics
6478 --------------------------------------
6480 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6481 (GC) requires the implementation and generation of these intrinsics.
6482 These intrinsics allow identification of :ref:`GC roots on the
6483 stack <int_gcroot>`, as well as garbage collector implementations that
6484 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6485 Front-ends for type-safe garbage collected languages should generate
6486 these intrinsics to make use of the LLVM garbage collectors. For more
6487 details, see `Accurate Garbage Collection with
6488 LLVM <GarbageCollection.html>`_.
6490 The garbage collection intrinsics only operate on objects in the generic
6491 address space (address space zero).
6495 '``llvm.gcroot``' Intrinsic
6496 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6503 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6508 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6509 the code generator, and allows some metadata to be associated with it.
6514 The first argument specifies the address of a stack object that contains
6515 the root pointer. The second pointer (which must be either a constant or
6516 a global value address) contains the meta-data to be associated with the
6522 At runtime, a call to this intrinsic stores a null pointer into the
6523 "ptrloc" location. At compile-time, the code generator generates
6524 information to allow the runtime to find the pointer at GC safe points.
6525 The '``llvm.gcroot``' intrinsic may only be used in a function which
6526 :ref:`specifies a GC algorithm <gc>`.
6530 '``llvm.gcread``' Intrinsic
6531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6538 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6543 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6544 locations, allowing garbage collector implementations that require read
6550 The second argument is the address to read from, which should be an
6551 address allocated from the garbage collector. The first object is a
6552 pointer to the start of the referenced object, if needed by the language
6553 runtime (otherwise null).
6558 The '``llvm.gcread``' intrinsic has the same semantics as a load
6559 instruction, but may be replaced with substantially more complex code by
6560 the garbage collector runtime, as needed. The '``llvm.gcread``'
6561 intrinsic may only be used in a function which :ref:`specifies a GC
6566 '``llvm.gcwrite``' Intrinsic
6567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6574 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6579 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6580 locations, allowing garbage collector implementations that require write
6581 barriers (such as generational or reference counting collectors).
6586 The first argument is the reference to store, the second is the start of
6587 the object to store it to, and the third is the address of the field of
6588 Obj to store to. If the runtime does not require a pointer to the
6589 object, Obj may be null.
6594 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6595 instruction, but may be replaced with substantially more complex code by
6596 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6597 intrinsic may only be used in a function which :ref:`specifies a GC
6600 Code Generator Intrinsics
6601 -------------------------
6603 These intrinsics are provided by LLVM to expose special features that
6604 may only be implemented with code generator support.
6606 '``llvm.returnaddress``' Intrinsic
6607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6614 declare i8 *@llvm.returnaddress(i32 <level>)
6619 The '``llvm.returnaddress``' intrinsic attempts to compute a
6620 target-specific value indicating the return address of the current
6621 function or one of its callers.
6626 The argument to this intrinsic indicates which function to return the
6627 address for. Zero indicates the calling function, one indicates its
6628 caller, etc. The argument is **required** to be a constant integer
6634 The '``llvm.returnaddress``' intrinsic either returns a pointer
6635 indicating the return address of the specified call frame, or zero if it
6636 cannot be identified. The value returned by this intrinsic is likely to
6637 be incorrect or 0 for arguments other than zero, so it should only be
6638 used for debugging purposes.
6640 Note that calling this intrinsic does not prevent function inlining or
6641 other aggressive transformations, so the value returned may not be that
6642 of the obvious source-language caller.
6644 '``llvm.frameaddress``' Intrinsic
6645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6652 declare i8* @llvm.frameaddress(i32 <level>)
6657 The '``llvm.frameaddress``' intrinsic attempts to return the
6658 target-specific frame pointer value for the specified stack frame.
6663 The argument to this intrinsic indicates which function to return the
6664 frame pointer for. Zero indicates the calling function, one indicates
6665 its caller, etc. The argument is **required** to be a constant integer
6671 The '``llvm.frameaddress``' intrinsic either returns a pointer
6672 indicating the frame address of the specified call frame, or zero if it
6673 cannot be identified. The value returned by this intrinsic is likely to
6674 be incorrect or 0 for arguments other than zero, so it should only be
6675 used for debugging purposes.
6677 Note that calling this intrinsic does not prevent function inlining or
6678 other aggressive transformations, so the value returned may not be that
6679 of the obvious source-language caller.
6683 '``llvm.stacksave``' Intrinsic
6684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6691 declare i8* @llvm.stacksave()
6696 The '``llvm.stacksave``' intrinsic is used to remember the current state
6697 of the function stack, for use with
6698 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6699 implementing language features like scoped automatic variable sized
6705 This intrinsic returns a opaque pointer value that can be passed to
6706 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6707 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6708 ``llvm.stacksave``, it effectively restores the state of the stack to
6709 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6710 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6711 were allocated after the ``llvm.stacksave`` was executed.
6713 .. _int_stackrestore:
6715 '``llvm.stackrestore``' Intrinsic
6716 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6723 declare void @llvm.stackrestore(i8* %ptr)
6728 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6729 the function stack to the state it was in when the corresponding
6730 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6731 useful for implementing language features like scoped automatic variable
6732 sized arrays in C99.
6737 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6739 '``llvm.prefetch``' Intrinsic
6740 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6747 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6752 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6753 insert a prefetch instruction if supported; otherwise, it is a noop.
6754 Prefetches have no effect on the behavior of the program but can change
6755 its performance characteristics.
6760 ``address`` is the address to be prefetched, ``rw`` is the specifier
6761 determining if the fetch should be for a read (0) or write (1), and
6762 ``locality`` is a temporal locality specifier ranging from (0) - no
6763 locality, to (3) - extremely local keep in cache. The ``cache type``
6764 specifies whether the prefetch is performed on the data (1) or
6765 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6766 arguments must be constant integers.
6771 This intrinsic does not modify the behavior of the program. In
6772 particular, prefetches cannot trap and do not produce a value. On
6773 targets that support this intrinsic, the prefetch can provide hints to
6774 the processor cache for better performance.
6776 '``llvm.pcmarker``' Intrinsic
6777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6784 declare void @llvm.pcmarker(i32 <id>)
6789 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6790 Counter (PC) in a region of code to simulators and other tools. The
6791 method is target specific, but it is expected that the marker will use
6792 exported symbols to transmit the PC of the marker. The marker makes no
6793 guarantees that it will remain with any specific instruction after
6794 optimizations. It is possible that the presence of a marker will inhibit
6795 optimizations. The intended use is to be inserted after optimizations to
6796 allow correlations of simulation runs.
6801 ``id`` is a numerical id identifying the marker.
6806 This intrinsic does not modify the behavior of the program. Backends
6807 that do not support this intrinsic may ignore it.
6809 '``llvm.readcyclecounter``' Intrinsic
6810 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6817 declare i64 @llvm.readcyclecounter()
6822 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6823 counter register (or similar low latency, high accuracy clocks) on those
6824 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6825 should map to RPCC. As the backing counters overflow quickly (on the
6826 order of 9 seconds on alpha), this should only be used for small
6832 When directly supported, reading the cycle counter should not modify any
6833 memory. Implementations are allowed to either return a application
6834 specific value or a system wide value. On backends without support, this
6835 is lowered to a constant 0.
6837 Note that runtime support may be conditional on the privilege-level code is
6838 running at and the host platform.
6840 Standard C Library Intrinsics
6841 -----------------------------
6843 LLVM provides intrinsics for a few important standard C library
6844 functions. These intrinsics allow source-language front-ends to pass
6845 information about the alignment of the pointer arguments to the code
6846 generator, providing opportunity for more efficient code generation.
6850 '``llvm.memcpy``' Intrinsic
6851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6856 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6857 integer bit width and for different address spaces. Not all targets
6858 support all bit widths however.
6862 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6863 i32 <len>, i32 <align>, i1 <isvolatile>)
6864 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6865 i64 <len>, i32 <align>, i1 <isvolatile>)
6870 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6871 source location to the destination location.
6873 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6874 intrinsics do not return a value, takes extra alignment/isvolatile
6875 arguments and the pointers can be in specified address spaces.
6880 The first argument is a pointer to the destination, the second is a
6881 pointer to the source. The third argument is an integer argument
6882 specifying the number of bytes to copy, the fourth argument is the
6883 alignment of the source and destination locations, and the fifth is a
6884 boolean indicating a volatile access.
6886 If the call to this intrinsic has an alignment value that is not 0 or 1,
6887 then the caller guarantees that both the source and destination pointers
6888 are aligned to that boundary.
6890 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6891 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6892 very cleanly specified and it is unwise to depend on it.
6897 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6898 source location to the destination location, which are not allowed to
6899 overlap. It copies "len" bytes of memory over. If the argument is known
6900 to be aligned to some boundary, this can be specified as the fourth
6901 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
6903 '``llvm.memmove``' Intrinsic
6904 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6909 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6910 bit width and for different address space. Not all targets support all
6915 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6916 i32 <len>, i32 <align>, i1 <isvolatile>)
6917 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6918 i64 <len>, i32 <align>, i1 <isvolatile>)
6923 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6924 source location to the destination location. It is similar to the
6925 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6928 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6929 intrinsics do not return a value, takes extra alignment/isvolatile
6930 arguments and the pointers can be in specified address spaces.
6935 The first argument is a pointer to the destination, the second is a
6936 pointer to the source. The third argument is an integer argument
6937 specifying the number of bytes to copy, the fourth argument is the
6938 alignment of the source and destination locations, and the fifth is a
6939 boolean indicating a volatile access.
6941 If the call to this intrinsic has an alignment value that is not 0 or 1,
6942 then the caller guarantees that the source and destination pointers are
6943 aligned to that boundary.
6945 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6946 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6947 not very cleanly specified and it is unwise to depend on it.
6952 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6953 source location to the destination location, which may overlap. It
6954 copies "len" bytes of memory over. If the argument is known to be
6955 aligned to some boundary, this can be specified as the fourth argument,
6956 otherwise it should be set to 0 or 1 (both meaning no alignment).
6958 '``llvm.memset.*``' Intrinsics
6959 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6964 This is an overloaded intrinsic. You can use llvm.memset on any integer
6965 bit width and for different address spaces. However, not all targets
6966 support all bit widths.
6970 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6971 i32 <len>, i32 <align>, i1 <isvolatile>)
6972 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6973 i64 <len>, i32 <align>, i1 <isvolatile>)
6978 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6979 particular byte value.
6981 Note that, unlike the standard libc function, the ``llvm.memset``
6982 intrinsic does not return a value and takes extra alignment/volatile
6983 arguments. Also, the destination can be in an arbitrary address space.
6988 The first argument is a pointer to the destination to fill, the second
6989 is the byte value with which to fill it, the third argument is an
6990 integer argument specifying the number of bytes to fill, and the fourth
6991 argument is the known alignment of the destination location.
6993 If the call to this intrinsic has an alignment value that is not 0 or 1,
6994 then the caller guarantees that the destination pointer is aligned to
6997 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6998 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6999 very cleanly specified and it is unwise to depend on it.
7004 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7005 at the destination location. If the argument is known to be aligned to
7006 some boundary, this can be specified as the fourth argument, otherwise
7007 it should be set to 0 or 1 (both meaning no alignment).
7009 '``llvm.sqrt.*``' Intrinsic
7010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7015 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7016 floating point or vector of floating point type. Not all targets support
7021 declare float @llvm.sqrt.f32(float %Val)
7022 declare double @llvm.sqrt.f64(double %Val)
7023 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7024 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7025 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7030 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7031 returning the same value as the libm '``sqrt``' functions would. Unlike
7032 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7033 negative numbers other than -0.0 (which allows for better optimization,
7034 because there is no need to worry about errno being set).
7035 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7040 The argument and return value are floating point numbers of the same
7046 This function returns the sqrt of the specified operand if it is a
7047 nonnegative floating point number.
7049 '``llvm.powi.*``' Intrinsic
7050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7055 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7056 floating point or vector of floating point type. Not all targets support
7061 declare float @llvm.powi.f32(float %Val, i32 %power)
7062 declare double @llvm.powi.f64(double %Val, i32 %power)
7063 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7064 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7065 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7070 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7071 specified (positive or negative) power. The order of evaluation of
7072 multiplications is not defined. When a vector of floating point type is
7073 used, the second argument remains a scalar integer value.
7078 The second argument is an integer power, and the first is a value to
7079 raise to that power.
7084 This function returns the first value raised to the second power with an
7085 unspecified sequence of rounding operations.
7087 '``llvm.sin.*``' Intrinsic
7088 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7093 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7094 floating point or vector of floating point type. Not all targets support
7099 declare float @llvm.sin.f32(float %Val)
7100 declare double @llvm.sin.f64(double %Val)
7101 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7102 declare fp128 @llvm.sin.f128(fp128 %Val)
7103 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7108 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7113 The argument and return value are floating point numbers of the same
7119 This function returns the sine of the specified operand, returning the
7120 same values as the libm ``sin`` functions would, and handles error
7121 conditions in the same way.
7123 '``llvm.cos.*``' Intrinsic
7124 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7129 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7130 floating point or vector of floating point type. Not all targets support
7135 declare float @llvm.cos.f32(float %Val)
7136 declare double @llvm.cos.f64(double %Val)
7137 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7138 declare fp128 @llvm.cos.f128(fp128 %Val)
7139 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7144 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7149 The argument and return value are floating point numbers of the same
7155 This function returns the cosine of the specified operand, returning the
7156 same values as the libm ``cos`` functions would, and handles error
7157 conditions in the same way.
7159 '``llvm.pow.*``' Intrinsic
7160 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7165 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7166 floating point or vector of floating point type. Not all targets support
7171 declare float @llvm.pow.f32(float %Val, float %Power)
7172 declare double @llvm.pow.f64(double %Val, double %Power)
7173 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7174 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7175 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7180 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7181 specified (positive or negative) power.
7186 The second argument is a floating point power, and the first is a value
7187 to raise to that power.
7192 This function returns the first value raised to the second power,
7193 returning the same values as the libm ``pow`` functions would, and
7194 handles error conditions in the same way.
7196 '``llvm.exp.*``' Intrinsic
7197 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7202 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7203 floating point or vector of floating point type. Not all targets support
7208 declare float @llvm.exp.f32(float %Val)
7209 declare double @llvm.exp.f64(double %Val)
7210 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7211 declare fp128 @llvm.exp.f128(fp128 %Val)
7212 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7217 The '``llvm.exp.*``' intrinsics perform the exp function.
7222 The argument and return value are floating point numbers of the same
7228 This function returns the same values as the libm ``exp`` functions
7229 would, and handles error conditions in the same way.
7231 '``llvm.exp2.*``' Intrinsic
7232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7237 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7238 floating point or vector of floating point type. Not all targets support
7243 declare float @llvm.exp2.f32(float %Val)
7244 declare double @llvm.exp2.f64(double %Val)
7245 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7246 declare fp128 @llvm.exp2.f128(fp128 %Val)
7247 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7252 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7257 The argument and return value are floating point numbers of the same
7263 This function returns the same values as the libm ``exp2`` functions
7264 would, and handles error conditions in the same way.
7266 '``llvm.log.*``' Intrinsic
7267 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7272 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7273 floating point or vector of floating point type. Not all targets support
7278 declare float @llvm.log.f32(float %Val)
7279 declare double @llvm.log.f64(double %Val)
7280 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7281 declare fp128 @llvm.log.f128(fp128 %Val)
7282 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7287 The '``llvm.log.*``' intrinsics perform the log function.
7292 The argument and return value are floating point numbers of the same
7298 This function returns the same values as the libm ``log`` functions
7299 would, and handles error conditions in the same way.
7301 '``llvm.log10.*``' Intrinsic
7302 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7307 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7308 floating point or vector of floating point type. Not all targets support
7313 declare float @llvm.log10.f32(float %Val)
7314 declare double @llvm.log10.f64(double %Val)
7315 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7316 declare fp128 @llvm.log10.f128(fp128 %Val)
7317 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7322 The '``llvm.log10.*``' intrinsics perform the log10 function.
7327 The argument and return value are floating point numbers of the same
7333 This function returns the same values as the libm ``log10`` functions
7334 would, and handles error conditions in the same way.
7336 '``llvm.log2.*``' Intrinsic
7337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7342 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7343 floating point or vector of floating point type. Not all targets support
7348 declare float @llvm.log2.f32(float %Val)
7349 declare double @llvm.log2.f64(double %Val)
7350 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7351 declare fp128 @llvm.log2.f128(fp128 %Val)
7352 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7357 The '``llvm.log2.*``' intrinsics perform the log2 function.
7362 The argument and return value are floating point numbers of the same
7368 This function returns the same values as the libm ``log2`` functions
7369 would, and handles error conditions in the same way.
7371 '``llvm.fma.*``' Intrinsic
7372 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7377 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7378 floating point or vector of floating point type. Not all targets support
7383 declare float @llvm.fma.f32(float %a, float %b, float %c)
7384 declare double @llvm.fma.f64(double %a, double %b, double %c)
7385 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7386 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7387 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7392 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7398 The argument and return value are floating point numbers of the same
7404 This function returns the same values as the libm ``fma`` functions
7407 '``llvm.fabs.*``' Intrinsic
7408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7413 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7414 floating point or vector of floating point type. Not all targets support
7419 declare float @llvm.fabs.f32(float %Val)
7420 declare double @llvm.fabs.f64(double %Val)
7421 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7422 declare fp128 @llvm.fabs.f128(fp128 %Val)
7423 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7428 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7434 The argument and return value are floating point numbers of the same
7440 This function returns the same values as the libm ``fabs`` functions
7441 would, and handles error conditions in the same way.
7443 '``llvm.copysign.*``' Intrinsic
7444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7449 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7450 floating point or vector of floating point type. Not all targets support
7455 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7456 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7457 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7458 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7459 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7464 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7465 first operand and the sign of the second operand.
7470 The arguments and return value are floating point numbers of the same
7476 This function returns the same values as the libm ``copysign``
7477 functions would, and handles error conditions in the same way.
7479 '``llvm.floor.*``' Intrinsic
7480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7485 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7486 floating point or vector of floating point type. Not all targets support
7491 declare float @llvm.floor.f32(float %Val)
7492 declare double @llvm.floor.f64(double %Val)
7493 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7494 declare fp128 @llvm.floor.f128(fp128 %Val)
7495 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7500 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7505 The argument and return value are floating point numbers of the same
7511 This function returns the same values as the libm ``floor`` functions
7512 would, and handles error conditions in the same way.
7514 '``llvm.ceil.*``' Intrinsic
7515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7520 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7521 floating point or vector of floating point type. Not all targets support
7526 declare float @llvm.ceil.f32(float %Val)
7527 declare double @llvm.ceil.f64(double %Val)
7528 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7529 declare fp128 @llvm.ceil.f128(fp128 %Val)
7530 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7535 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7540 The argument and return value are floating point numbers of the same
7546 This function returns the same values as the libm ``ceil`` functions
7547 would, and handles error conditions in the same way.
7549 '``llvm.trunc.*``' Intrinsic
7550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7555 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7556 floating point or vector of floating point type. Not all targets support
7561 declare float @llvm.trunc.f32(float %Val)
7562 declare double @llvm.trunc.f64(double %Val)
7563 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7564 declare fp128 @llvm.trunc.f128(fp128 %Val)
7565 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7570 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7571 nearest integer not larger in magnitude than the operand.
7576 The argument and return value are floating point numbers of the same
7582 This function returns the same values as the libm ``trunc`` functions
7583 would, and handles error conditions in the same way.
7585 '``llvm.rint.*``' Intrinsic
7586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7591 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7592 floating point or vector of floating point type. Not all targets support
7597 declare float @llvm.rint.f32(float %Val)
7598 declare double @llvm.rint.f64(double %Val)
7599 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7600 declare fp128 @llvm.rint.f128(fp128 %Val)
7601 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7606 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7607 nearest integer. It may raise an inexact floating-point exception if the
7608 operand isn't an integer.
7613 The argument and return value are floating point numbers of the same
7619 This function returns the same values as the libm ``rint`` functions
7620 would, and handles error conditions in the same way.
7622 '``llvm.nearbyint.*``' Intrinsic
7623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7628 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7629 floating point or vector of floating point type. Not all targets support
7634 declare float @llvm.nearbyint.f32(float %Val)
7635 declare double @llvm.nearbyint.f64(double %Val)
7636 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7637 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7638 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7643 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7649 The argument and return value are floating point numbers of the same
7655 This function returns the same values as the libm ``nearbyint``
7656 functions would, and handles error conditions in the same way.
7658 '``llvm.round.*``' Intrinsic
7659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7664 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7665 floating point or vector of floating point type. Not all targets support
7670 declare float @llvm.round.f32(float %Val)
7671 declare double @llvm.round.f64(double %Val)
7672 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7673 declare fp128 @llvm.round.f128(fp128 %Val)
7674 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7679 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7685 The argument and return value are floating point numbers of the same
7691 This function returns the same values as the libm ``round``
7692 functions would, and handles error conditions in the same way.
7694 Bit Manipulation Intrinsics
7695 ---------------------------
7697 LLVM provides intrinsics for a few important bit manipulation
7698 operations. These allow efficient code generation for some algorithms.
7700 '``llvm.bswap.*``' Intrinsics
7701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7706 This is an overloaded intrinsic function. You can use bswap on any
7707 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7711 declare i16 @llvm.bswap.i16(i16 <id>)
7712 declare i32 @llvm.bswap.i32(i32 <id>)
7713 declare i64 @llvm.bswap.i64(i64 <id>)
7718 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7719 values with an even number of bytes (positive multiple of 16 bits).
7720 These are useful for performing operations on data that is not in the
7721 target's native byte order.
7726 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7727 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7728 intrinsic returns an i32 value that has the four bytes of the input i32
7729 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7730 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7731 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7732 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7735 '``llvm.ctpop.*``' Intrinsic
7736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7741 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7742 bit width, or on any vector with integer elements. Not all targets
7743 support all bit widths or vector types, however.
7747 declare i8 @llvm.ctpop.i8(i8 <src>)
7748 declare i16 @llvm.ctpop.i16(i16 <src>)
7749 declare i32 @llvm.ctpop.i32(i32 <src>)
7750 declare i64 @llvm.ctpop.i64(i64 <src>)
7751 declare i256 @llvm.ctpop.i256(i256 <src>)
7752 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7757 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7763 The only argument is the value to be counted. The argument may be of any
7764 integer type, or a vector with integer elements. The return type must
7765 match the argument type.
7770 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7771 each element of a vector.
7773 '``llvm.ctlz.*``' Intrinsic
7774 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7779 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7780 integer bit width, or any vector whose elements are integers. Not all
7781 targets support all bit widths or vector types, however.
7785 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7786 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7787 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7788 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7789 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7790 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7795 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7796 leading zeros in a variable.
7801 The first argument is the value to be counted. This argument may be of
7802 any integer type, or a vectory with integer element type. The return
7803 type must match the first argument type.
7805 The second argument must be a constant and is a flag to indicate whether
7806 the intrinsic should ensure that a zero as the first argument produces a
7807 defined result. Historically some architectures did not provide a
7808 defined result for zero values as efficiently, and many algorithms are
7809 now predicated on avoiding zero-value inputs.
7814 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7815 zeros in a variable, or within each element of the vector. If
7816 ``src == 0`` then the result is the size in bits of the type of ``src``
7817 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7818 ``llvm.ctlz(i32 2) = 30``.
7820 '``llvm.cttz.*``' Intrinsic
7821 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7826 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7827 integer bit width, or any vector of integer elements. Not all targets
7828 support all bit widths or vector types, however.
7832 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7833 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7834 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7835 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7836 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7837 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7842 The '``llvm.cttz``' family of intrinsic functions counts the number of
7848 The first argument is the value to be counted. This argument may be of
7849 any integer type, or a vectory with integer element type. The return
7850 type must match the first argument type.
7852 The second argument must be a constant and is a flag to indicate whether
7853 the intrinsic should ensure that a zero as the first argument produces a
7854 defined result. Historically some architectures did not provide a
7855 defined result for zero values as efficiently, and many algorithms are
7856 now predicated on avoiding zero-value inputs.
7861 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7862 zeros in a variable, or within each element of a vector. If ``src == 0``
7863 then the result is the size in bits of the type of ``src`` if
7864 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7865 ``llvm.cttz(2) = 1``.
7867 Arithmetic with Overflow Intrinsics
7868 -----------------------------------
7870 LLVM provides intrinsics for some arithmetic with overflow operations.
7872 '``llvm.sadd.with.overflow.*``' Intrinsics
7873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7878 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7879 on any integer bit width.
7883 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7884 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7885 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7890 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7891 a signed addition of the two arguments, and indicate whether an overflow
7892 occurred during the signed summation.
7897 The arguments (%a and %b) and the first element of the result structure
7898 may be of integer types of any bit width, but they must have the same
7899 bit width. The second element of the result structure must be of type
7900 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7906 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7907 a signed addition of the two variables. They return a structure --- the
7908 first element of which is the signed summation, and the second element
7909 of which is a bit specifying if the signed summation resulted in an
7915 .. code-block:: llvm
7917 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7918 %sum = extractvalue {i32, i1} %res, 0
7919 %obit = extractvalue {i32, i1} %res, 1
7920 br i1 %obit, label %overflow, label %normal
7922 '``llvm.uadd.with.overflow.*``' Intrinsics
7923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7928 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7929 on any integer bit width.
7933 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7934 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7935 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7940 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7941 an unsigned addition of the two arguments, and indicate whether a carry
7942 occurred during the unsigned summation.
7947 The arguments (%a and %b) and the first element of the result structure
7948 may be of integer types of any bit width, but they must have the same
7949 bit width. The second element of the result structure must be of type
7950 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7956 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7957 an unsigned addition of the two arguments. They return a structure --- the
7958 first element of which is the sum, and the second element of which is a
7959 bit specifying if the unsigned summation resulted in a carry.
7964 .. code-block:: llvm
7966 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7967 %sum = extractvalue {i32, i1} %res, 0
7968 %obit = extractvalue {i32, i1} %res, 1
7969 br i1 %obit, label %carry, label %normal
7971 '``llvm.ssub.with.overflow.*``' Intrinsics
7972 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7977 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7978 on any integer bit width.
7982 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7983 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7984 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7989 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7990 a signed subtraction of the two arguments, and indicate whether an
7991 overflow occurred during the signed subtraction.
7996 The arguments (%a and %b) and the first element of the result structure
7997 may be of integer types of any bit width, but they must have the same
7998 bit width. The second element of the result structure must be of type
7999 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8005 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8006 a signed subtraction of the two arguments. They return a structure --- the
8007 first element of which is the subtraction, and the second element of
8008 which is a bit specifying if the signed subtraction resulted in an
8014 .. code-block:: llvm
8016 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8017 %sum = extractvalue {i32, i1} %res, 0
8018 %obit = extractvalue {i32, i1} %res, 1
8019 br i1 %obit, label %overflow, label %normal
8021 '``llvm.usub.with.overflow.*``' Intrinsics
8022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8027 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8028 on any integer bit width.
8032 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8033 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8034 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8039 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8040 an unsigned subtraction of the two arguments, and indicate whether an
8041 overflow occurred during the unsigned subtraction.
8046 The arguments (%a and %b) and the first element of the result structure
8047 may be of integer types of any bit width, but they must have the same
8048 bit width. The second element of the result structure must be of type
8049 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8055 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8056 an unsigned subtraction of the two arguments. They return a structure ---
8057 the first element of which is the subtraction, and the second element of
8058 which is a bit specifying if the unsigned subtraction resulted in an
8064 .. code-block:: llvm
8066 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8067 %sum = extractvalue {i32, i1} %res, 0
8068 %obit = extractvalue {i32, i1} %res, 1
8069 br i1 %obit, label %overflow, label %normal
8071 '``llvm.smul.with.overflow.*``' Intrinsics
8072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8077 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8078 on any integer bit width.
8082 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8083 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8084 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8089 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8090 a signed multiplication of the two arguments, and indicate whether an
8091 overflow occurred during the signed multiplication.
8096 The arguments (%a and %b) and the first element of the result structure
8097 may be of integer types of any bit width, but they must have the same
8098 bit width. The second element of the result structure must be of type
8099 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8105 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8106 a signed multiplication of the two arguments. They return a structure ---
8107 the first element of which is the multiplication, and the second element
8108 of which is a bit specifying if the signed multiplication resulted in an
8114 .. code-block:: llvm
8116 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8117 %sum = extractvalue {i32, i1} %res, 0
8118 %obit = extractvalue {i32, i1} %res, 1
8119 br i1 %obit, label %overflow, label %normal
8121 '``llvm.umul.with.overflow.*``' Intrinsics
8122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8127 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8128 on any integer bit width.
8132 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8133 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8134 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8139 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8140 a unsigned multiplication of the two arguments, and indicate whether an
8141 overflow occurred during the unsigned multiplication.
8146 The arguments (%a and %b) and the first element of the result structure
8147 may be of integer types of any bit width, but they must have the same
8148 bit width. The second element of the result structure must be of type
8149 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8155 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8156 an unsigned multiplication of the two arguments. They return a structure ---
8157 the first element of which is the multiplication, and the second
8158 element of which is a bit specifying if the unsigned multiplication
8159 resulted in an overflow.
8164 .. code-block:: llvm
8166 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8167 %sum = extractvalue {i32, i1} %res, 0
8168 %obit = extractvalue {i32, i1} %res, 1
8169 br i1 %obit, label %overflow, label %normal
8171 Specialised Arithmetic Intrinsics
8172 ---------------------------------
8174 '``llvm.fmuladd.*``' Intrinsic
8175 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8182 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8183 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8188 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8189 expressions that can be fused if the code generator determines that (a) the
8190 target instruction set has support for a fused operation, and (b) that the
8191 fused operation is more efficient than the equivalent, separate pair of mul
8192 and add instructions.
8197 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8198 multiplicands, a and b, and an addend c.
8207 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8209 is equivalent to the expression a \* b + c, except that rounding will
8210 not be performed between the multiplication and addition steps if the
8211 code generator fuses the operations. Fusion is not guaranteed, even if
8212 the target platform supports it. If a fused multiply-add is required the
8213 corresponding llvm.fma.\* intrinsic function should be used instead.
8218 .. code-block:: llvm
8220 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8222 Half Precision Floating Point Intrinsics
8223 ----------------------------------------
8225 For most target platforms, half precision floating point is a
8226 storage-only format. This means that it is a dense encoding (in memory)
8227 but does not support computation in the format.
8229 This means that code must first load the half-precision floating point
8230 value as an i16, then convert it to float with
8231 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8232 then be performed on the float value (including extending to double
8233 etc). To store the value back to memory, it is first converted to float
8234 if needed, then converted to i16 with
8235 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8238 .. _int_convert_to_fp16:
8240 '``llvm.convert.to.fp16``' Intrinsic
8241 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8248 declare i16 @llvm.convert.to.fp16(f32 %a)
8253 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8254 from single precision floating point format to half precision floating
8260 The intrinsic function contains single argument - the value to be
8266 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8267 from single precision floating point format to half precision floating
8268 point format. The return value is an ``i16`` which contains the
8274 .. code-block:: llvm
8276 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8277 store i16 %res, i16* @x, align 2
8279 .. _int_convert_from_fp16:
8281 '``llvm.convert.from.fp16``' Intrinsic
8282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8289 declare f32 @llvm.convert.from.fp16(i16 %a)
8294 The '``llvm.convert.from.fp16``' intrinsic function performs a
8295 conversion from half precision floating point format to single precision
8296 floating point format.
8301 The intrinsic function contains single argument - the value to be
8307 The '``llvm.convert.from.fp16``' intrinsic function performs a
8308 conversion from half single precision floating point format to single
8309 precision floating point format. The input half-float value is
8310 represented by an ``i16`` value.
8315 .. code-block:: llvm
8317 %a = load i16* @x, align 2
8318 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8323 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8324 prefix), are described in the `LLVM Source Level
8325 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8328 Exception Handling Intrinsics
8329 -----------------------------
8331 The LLVM exception handling intrinsics (which all start with
8332 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8333 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8337 Trampoline Intrinsics
8338 ---------------------
8340 These intrinsics make it possible to excise one parameter, marked with
8341 the :ref:`nest <nest>` attribute, from a function. The result is a
8342 callable function pointer lacking the nest parameter - the caller does
8343 not need to provide a value for it. Instead, the value to use is stored
8344 in advance in a "trampoline", a block of memory usually allocated on the
8345 stack, which also contains code to splice the nest value into the
8346 argument list. This is used to implement the GCC nested function address
8349 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8350 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8351 It can be created as follows:
8353 .. code-block:: llvm
8355 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8356 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8357 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8358 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8359 %fp = bitcast i8* %p to i32 (i32, i32)*
8361 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8362 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8366 '``llvm.init.trampoline``' Intrinsic
8367 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8374 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8379 This fills the memory pointed to by ``tramp`` with executable code,
8380 turning it into a trampoline.
8385 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8386 pointers. The ``tramp`` argument must point to a sufficiently large and
8387 sufficiently aligned block of memory; this memory is written to by the
8388 intrinsic. Note that the size and the alignment are target-specific -
8389 LLVM currently provides no portable way of determining them, so a
8390 front-end that generates this intrinsic needs to have some
8391 target-specific knowledge. The ``func`` argument must hold a function
8392 bitcast to an ``i8*``.
8397 The block of memory pointed to by ``tramp`` is filled with target
8398 dependent code, turning it into a function. Then ``tramp`` needs to be
8399 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8400 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8401 function's signature is the same as that of ``func`` with any arguments
8402 marked with the ``nest`` attribute removed. At most one such ``nest``
8403 argument is allowed, and it must be of pointer type. Calling the new
8404 function is equivalent to calling ``func`` with the same argument list,
8405 but with ``nval`` used for the missing ``nest`` argument. If, after
8406 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8407 modified, then the effect of any later call to the returned function
8408 pointer is undefined.
8412 '``llvm.adjust.trampoline``' Intrinsic
8413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8420 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8425 This performs any required machine-specific adjustment to the address of
8426 a trampoline (passed as ``tramp``).
8431 ``tramp`` must point to a block of memory which already has trampoline
8432 code filled in by a previous call to
8433 :ref:`llvm.init.trampoline <int_it>`.
8438 On some architectures the address of the code to be executed needs to be
8439 different to the address where the trampoline is actually stored. This
8440 intrinsic returns the executable address corresponding to ``tramp``
8441 after performing the required machine specific adjustments. The pointer
8442 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8447 This class of intrinsics exists to information about the lifetime of
8448 memory objects and ranges where variables are immutable.
8450 '``llvm.lifetime.start``' Intrinsic
8451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8458 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8463 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8469 The first argument is a constant integer representing the size of the
8470 object, or -1 if it is variable sized. The second argument is a pointer
8476 This intrinsic indicates that before this point in the code, the value
8477 of the memory pointed to by ``ptr`` is dead. This means that it is known
8478 to never be used and has an undefined value. A load from the pointer
8479 that precedes this intrinsic can be replaced with ``'undef'``.
8481 '``llvm.lifetime.end``' Intrinsic
8482 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8489 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8494 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8500 The first argument is a constant integer representing the size of the
8501 object, or -1 if it is variable sized. The second argument is a pointer
8507 This intrinsic indicates that after this point in the code, the value of
8508 the memory pointed to by ``ptr`` is dead. This means that it is known to
8509 never be used and has an undefined value. Any stores into the memory
8510 object following this intrinsic may be removed as dead.
8512 '``llvm.invariant.start``' Intrinsic
8513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8520 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8525 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8526 a memory object will not change.
8531 The first argument is a constant integer representing the size of the
8532 object, or -1 if it is variable sized. The second argument is a pointer
8538 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8539 the return value, the referenced memory location is constant and
8542 '``llvm.invariant.end``' Intrinsic
8543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8550 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8555 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8556 memory object are mutable.
8561 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8562 The second argument is a constant integer representing the size of the
8563 object, or -1 if it is variable sized and the third argument is a
8564 pointer to the object.
8569 This intrinsic indicates that the memory is mutable again.
8574 This class of intrinsics is designed to be generic and has no specific
8577 '``llvm.var.annotation``' Intrinsic
8578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8585 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8590 The '``llvm.var.annotation``' intrinsic.
8595 The first argument is a pointer to a value, the second is a pointer to a
8596 global string, the third is a pointer to a global string which is the
8597 source file name, and the last argument is the line number.
8602 This intrinsic allows annotation of local variables with arbitrary
8603 strings. This can be useful for special purpose optimizations that want
8604 to look for these annotations. These have no other defined use; they are
8605 ignored by code generation and optimization.
8607 '``llvm.ptr.annotation.*``' Intrinsic
8608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8613 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8614 pointer to an integer of any width. *NOTE* you must specify an address space for
8615 the pointer. The identifier for the default address space is the integer
8620 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8621 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8622 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8623 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8624 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8629 The '``llvm.ptr.annotation``' intrinsic.
8634 The first argument is a pointer to an integer value of arbitrary bitwidth
8635 (result of some expression), the second is a pointer to a global string, the
8636 third is a pointer to a global string which is the source file name, and the
8637 last argument is the line number. It returns the value of the first argument.
8642 This intrinsic allows annotation of a pointer to an integer with arbitrary
8643 strings. This can be useful for special purpose optimizations that want to look
8644 for these annotations. These have no other defined use; they are ignored by code
8645 generation and optimization.
8647 '``llvm.annotation.*``' Intrinsic
8648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8653 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8654 any integer bit width.
8658 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8659 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8660 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8661 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8662 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8667 The '``llvm.annotation``' intrinsic.
8672 The first argument is an integer value (result of some expression), the
8673 second is a pointer to a global string, the third is a pointer to a
8674 global string which is the source file name, and the last argument is
8675 the line number. It returns the value of the first argument.
8680 This intrinsic allows annotations to be put on arbitrary expressions
8681 with arbitrary strings. This can be useful for special purpose
8682 optimizations that want to look for these annotations. These have no
8683 other defined use; they are ignored by code generation and optimization.
8685 '``llvm.trap``' Intrinsic
8686 ^^^^^^^^^^^^^^^^^^^^^^^^^
8693 declare void @llvm.trap() noreturn nounwind
8698 The '``llvm.trap``' intrinsic.
8708 This intrinsic is lowered to the target dependent trap instruction. If
8709 the target does not have a trap instruction, this intrinsic will be
8710 lowered to a call of the ``abort()`` function.
8712 '``llvm.debugtrap``' Intrinsic
8713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8720 declare void @llvm.debugtrap() nounwind
8725 The '``llvm.debugtrap``' intrinsic.
8735 This intrinsic is lowered to code which is intended to cause an
8736 execution trap with the intention of requesting the attention of a
8739 '``llvm.stackprotector``' Intrinsic
8740 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8747 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8752 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8753 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8754 is placed on the stack before local variables.
8759 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8760 The first argument is the value loaded from the stack guard
8761 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8762 enough space to hold the value of the guard.
8767 This intrinsic causes the prologue/epilogue inserter to force the position of
8768 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8769 to ensure that if a local variable on the stack is overwritten, it will destroy
8770 the value of the guard. When the function exits, the guard on the stack is
8771 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8772 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8773 calling the ``__stack_chk_fail()`` function.
8775 '``llvm.stackprotectorcheck``' Intrinsic
8776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8783 declare void @llvm.stackprotectorcheck(i8** <guard>)
8788 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8789 created stack protector and if they are not equal calls the
8790 ``__stack_chk_fail()`` function.
8795 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8796 the variable ``@__stack_chk_guard``.
8801 This intrinsic is provided to perform the stack protector check by comparing
8802 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8803 values do not match call the ``__stack_chk_fail()`` function.
8805 The reason to provide this as an IR level intrinsic instead of implementing it
8806 via other IR operations is that in order to perform this operation at the IR
8807 level without an intrinsic, one would need to create additional basic blocks to
8808 handle the success/failure cases. This makes it difficult to stop the stack
8809 protector check from disrupting sibling tail calls in Codegen. With this
8810 intrinsic, we are able to generate the stack protector basic blocks late in
8811 codegen after the tail call decision has occured.
8813 '``llvm.objectsize``' Intrinsic
8814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8821 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8822 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8827 The ``llvm.objectsize`` intrinsic is designed to provide information to
8828 the optimizers to determine at compile time whether a) an operation
8829 (like memcpy) will overflow a buffer that corresponds to an object, or
8830 b) that a runtime check for overflow isn't necessary. An object in this
8831 context means an allocation of a specific class, structure, array, or
8837 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8838 argument is a pointer to or into the ``object``. The second argument is
8839 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8840 or -1 (if false) when the object size is unknown. The second argument
8841 only accepts constants.
8846 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8847 the size of the object concerned. If the size cannot be determined at
8848 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8849 on the ``min`` argument).
8851 '``llvm.expect``' Intrinsic
8852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8859 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8860 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8865 The ``llvm.expect`` intrinsic provides information about expected (the
8866 most probable) value of ``val``, which can be used by optimizers.
8871 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8872 a value. The second argument is an expected value, this needs to be a
8873 constant value, variables are not allowed.
8878 This intrinsic is lowered to the ``val``.
8880 '``llvm.donothing``' Intrinsic
8881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8888 declare void @llvm.donothing() nounwind readnone
8893 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8894 only intrinsic that can be called with an invoke instruction.
8904 This intrinsic does nothing, and it's removed by optimizers and ignored