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, i8 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
132 It also shows a convention that we follow in this document. When
133 demonstrating instructions, we will follow an instruction with a comment
134 that defines the type and name of value produced.
142 LLVM programs are composed of ``Module``'s, each of which is a
143 translation unit of the input programs. Each module consists of
144 functions, global variables, and symbol table entries. Modules may be
145 combined together with the LLVM linker, which merges function (and
146 global variable) definitions, resolves forward declarations, and merges
147 symbol table entries. Here is an example of the "hello world" module:
151 ; Declare the string constant as a global constant.
152 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
154 ; External declaration of the puts function
155 declare i32 @puts(i8* nocapture) nounwind
157 ; Definition of main function
158 define i32 @main() { ; i32()*
159 ; Convert [13 x i8]* to i8 *...
160 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
162 ; Call puts function to write out the string to stdout.
163 call i32 @puts(i8* %cast210)
168 !1 = metadata !{i32 42}
171 This example is made up of a :ref:`global variable <globalvars>` named
172 "``.str``", an external declaration of the "``puts``" function, a
173 :ref:`function definition <functionstructure>` for "``main``" and
174 :ref:`named metadata <namedmetadatastructure>` "``foo``".
176 In general, a module is made up of a list of global values (where both
177 functions and global variables are global values). Global values are
178 represented by a pointer to a memory location (in this case, a pointer
179 to an array of char, and a pointer to a function), and have one of the
180 following :ref:`linkage types <linkage>`.
187 All Global Variables and Functions have one of the following types of
191 Global values with "``private``" linkage are only directly
192 accessible by objects in the current module. In particular, linking
193 code into a module with an private global value may cause the
194 private to be renamed as necessary to avoid collisions. Because the
195 symbol is private to the module, all references can be updated. This
196 doesn't show up in any symbol table in the object file.
198 Similar to ``private``, but the symbol is passed through the
199 assembler and evaluated by the linker. Unlike normal strong symbols,
200 they are removed by the linker from the final linked image
201 (executable or dynamic library).
202 ``linker_private_weak``
203 Similar to "``linker_private``", but the symbol is weak. Note that
204 ``linker_private_weak`` symbols are subject to coalescing by the
205 linker. The symbols are removed by the linker from the final linked
206 image (executable or dynamic library).
208 Similar to private, but the value shows as a local symbol
209 (``STB_LOCAL`` in the case of ELF) in the object file. This
210 corresponds to the notion of the '``static``' keyword in C.
211 ``available_externally``
212 Globals with "``available_externally``" linkage are never emitted
213 into the object file corresponding to the LLVM module. They exist to
214 allow inlining and other optimizations to take place given knowledge
215 of the definition of the global, which is known to be somewhere
216 outside the module. Globals with ``available_externally`` linkage
217 are allowed to be discarded at will, and are otherwise the same as
218 ``linkonce_odr``. This linkage type is only allowed on definitions,
221 Globals with "``linkonce``" linkage are merged with other globals of
222 the same name when linkage occurs. This can be used to implement
223 some forms of inline functions, templates, or other code which must
224 be generated in each translation unit that uses it, but where the
225 body may be overridden with a more definitive definition later.
226 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
227 that ``linkonce`` linkage does not actually allow the optimizer to
228 inline the body of this function into callers because it doesn't
229 know if this definition of the function is the definitive definition
230 within the program or whether it will be overridden by a stronger
231 definition. To enable inlining and other optimizations, use
232 "``linkonce_odr``" linkage.
234 "``weak``" linkage has the same merging semantics as ``linkonce``
235 linkage, except that unreferenced globals with ``weak`` linkage may
236 not be discarded. This is used for globals that are declared "weak"
239 "``common``" linkage is most similar to "``weak``" linkage, but they
240 are used for tentative definitions in C, such as "``int X;``" at
241 global scope. Symbols with "``common``" linkage are merged in the
242 same way as ``weak symbols``, and they may not be deleted if
243 unreferenced. ``common`` symbols may not have an explicit section,
244 must have a zero initializer, and may not be marked
245 ':ref:`constant <globalvars>`'. Functions and aliases may not have
248 .. _linkage_appending:
251 "``appending``" linkage may only be applied to global variables of
252 pointer to array type. When two global variables with appending
253 linkage are linked together, the two global arrays are appended
254 together. This is the LLVM, typesafe, equivalent of having the
255 system linker append together "sections" with identical names when
258 The semantics of this linkage follow the ELF object file model: the
259 symbol is weak until linked, if not linked, the symbol becomes null
260 instead of being an undefined reference.
261 ``linkonce_odr``, ``weak_odr``
262 Some languages allow differing globals to be merged, such as two
263 functions with different semantics. Other languages, such as
264 ``C++``, ensure that only equivalent globals are ever merged (the
265 "one definition rule" — "ODR"). Such languages can use the
266 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
267 global will only be merged with equivalent globals. These linkage
268 types are otherwise the same as their non-``odr`` versions.
269 ``linkonce_odr_auto_hide``
270 Similar to "``linkonce_odr``", but nothing in the translation unit
271 takes the address of this definition. For instance, functions that
272 had an inline definition, but the compiler decided not to inline it.
273 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
274 symbols are removed by the linker from the final linked image
275 (executable or dynamic library).
277 If none of the above identifiers are used, the global is externally
278 visible, meaning that it participates in linkage and can be used to
279 resolve external symbol references.
281 The next two types of linkage are targeted for Microsoft Windows
282 platform only. They are designed to support importing (exporting)
283 symbols from (to) DLLs (Dynamic Link Libraries).
286 "``dllimport``" linkage causes the compiler to reference a function
287 or variable via a global pointer to a pointer that is set up by the
288 DLL exporting the symbol. On Microsoft Windows targets, the pointer
289 name is formed by combining ``__imp_`` and the function or variable
292 "``dllexport``" linkage causes the compiler to provide a global
293 pointer to a pointer in a DLL, so that it can be referenced with the
294 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
295 name is formed by combining ``__imp_`` and the function or variable
298 For example, since the "``.LC0``" variable is defined to be internal, if
299 another module defined a "``.LC0``" variable and was linked with this
300 one, one of the two would be renamed, preventing a collision. Since
301 "``main``" and "``puts``" are external (i.e., lacking any linkage
302 declarations), they are accessible outside of the current module.
304 It is illegal for a function *declaration* to have any linkage type
305 other than ``external``, ``dllimport`` or ``extern_weak``.
307 Aliases can have only ``external``, ``internal``, ``weak`` or
308 ``weak_odr`` linkages.
315 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
316 :ref:`invokes <i_invoke>` can all have an optional calling convention
317 specified for the call. The calling convention of any pair of dynamic
318 caller/callee must match, or the behavior of the program is undefined.
319 The following calling conventions are supported by LLVM, and more may be
322 "``ccc``" - The C calling convention
323 This calling convention (the default if no other calling convention
324 is specified) matches the target C calling conventions. This calling
325 convention supports varargs function calls and tolerates some
326 mismatch in the declared prototype and implemented declaration of
327 the function (as does normal C).
328 "``fastcc``" - The fast calling convention
329 This calling convention attempts to make calls as fast as possible
330 (e.g. by passing things in registers). This calling convention
331 allows the target to use whatever tricks it wants to produce fast
332 code for the target, without having to conform to an externally
333 specified ABI (Application Binary Interface). `Tail calls can only
334 be optimized when this, the GHC or the HiPE convention is
335 used. <CodeGenerator.html#id80>`_ This calling convention does not
336 support varargs and requires the prototype of all callees to exactly
337 match the prototype of the function definition.
338 "``coldcc``" - The cold calling convention
339 This calling convention attempts to make code in the caller as
340 efficient as possible under the assumption that the call is not
341 commonly executed. As such, these calls often preserve all registers
342 so that the call does not break any live ranges in the caller side.
343 This calling convention does not support varargs and requires the
344 prototype of all callees to exactly match the prototype of the
346 "``cc 10``" - GHC convention
347 This calling convention has been implemented specifically for use by
348 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
349 It passes everything in registers, going to extremes to achieve this
350 by disabling callee save registers. This calling convention should
351 not be used lightly but only for specific situations such as an
352 alternative to the *register pinning* performance technique often
353 used when implementing functional programming languages. At the
354 moment only X86 supports this convention and it has the following
357 - On *X86-32* only supports up to 4 bit type parameters. No
358 floating point types are supported.
359 - On *X86-64* only supports up to 10 bit type parameters and 6
360 floating point parameters.
362 This calling convention supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires both the
364 caller and callee are using it.
365 "``cc 11``" - The HiPE calling convention
366 This calling convention has been implemented specifically for use by
367 the `High-Performance Erlang
368 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
369 native code compiler of the `Ericsson's Open Source Erlang/OTP
370 system <http://www.erlang.org/download.shtml>`_. It uses more
371 registers for argument passing than the ordinary C calling
372 convention and defines no callee-saved registers. The calling
373 convention properly supports `tail call
374 optimization <CodeGenerator.html#id80>`_ but requires that both the
375 caller and the callee use it. It uses a *register pinning*
376 mechanism, similar to GHC's convention, for keeping frequently
377 accessed runtime components pinned to specific hardware registers.
378 At the moment only X86 supports this convention (both 32 and 64
380 "``cc <n>``" - Numbered convention
381 Any calling convention may be specified by number, allowing
382 target-specific calling conventions to be used. Target specific
383 calling conventions start at 64.
385 More calling conventions can be added/defined on an as-needed basis, to
386 support Pascal conventions or any other well-known target-independent
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.
417 LLVM IR allows you to specify name aliases for certain types. This can
418 make it easier to read the IR and make the IR more condensed
419 (particularly when recursive types are involved). An example of a name
424 %mytype = type { %mytype*, i32 }
426 You may give a name to any :ref:`type <typesystem>` except
427 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
428 expected with the syntax "%mytype".
430 Note that type names are aliases for the structural type that they
431 indicate, and that you can therefore specify multiple names for the same
432 type. This often leads to confusing behavior when dumping out a .ll
433 file. Since LLVM IR uses structural typing, the name is not part of the
434 type. When printing out LLVM IR, the printer will pick *one name* to
435 render all types of a particular shape. This means that if you have code
436 where two different source types end up having the same LLVM type, that
437 the dumper will sometimes print the "wrong" or unexpected type. This is
438 an important design point and isn't going to change.
445 Global variables define regions of memory allocated at compilation time
446 instead of run-time. Global variables may optionally be initialized, may
447 have an explicit section to be placed in, and may have an optional
448 explicit alignment specified.
450 A variable may be defined as ``thread_local``, which means that it will
451 not be shared by threads (each thread will have a separated copy of the
452 variable). Not all targets support thread-local variables. Optionally, a
453 TLS model may be specified:
456 For variables that are only used within the current shared library.
458 For variables in modules that will not be loaded dynamically.
460 For variables defined in the executable and only used within it.
462 The models correspond to the ELF TLS models; see `ELF Handling For
463 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
464 more information on under which circumstances the different models may
465 be used. The target may choose a different TLS model if the specified
466 model is not supported, or if a better choice of model can be made.
468 A variable may be defined as a global "constant," which indicates that
469 the contents of the variable will **never** be modified (enabling better
470 optimization, allowing the global data to be placed in the read-only
471 section of an executable, etc). Note that variables that need runtime
472 initialization cannot be marked "constant" as there is a store to the
475 LLVM explicitly allows *declarations* of global variables to be marked
476 constant, even if the final definition of the global is not. This
477 capability can be used to enable slightly better optimization of the
478 program, but requires the language definition to guarantee that
479 optimizations based on the 'constantness' are valid for the translation
480 units that do not include the definition.
482 As SSA values, global variables define pointer values that are in scope
483 (i.e. they dominate) all basic blocks in the program. Global variables
484 always define a pointer to their "content" type because they describe a
485 region of memory, and all memory objects in LLVM are accessed through
488 Global variables can be marked with ``unnamed_addr`` which indicates
489 that the address is not significant, only the content. Constants marked
490 like this can be merged with other constants if they have the same
491 initializer. Note that a constant with significant address *can* be
492 merged with a ``unnamed_addr`` constant, the result being a constant
493 whose address is significant.
495 A global variable may be declared to reside in a target-specific
496 numbered address space. For targets that support them, address spaces
497 may affect how optimizations are performed and/or what target
498 instructions are used to access the variable. The default address space
499 is zero. The address space qualifier must precede any other attributes.
501 LLVM allows an explicit section to be specified for globals. If the
502 target supports it, it will emit globals to the section specified.
504 An explicit alignment may be specified for a global, which must be a
505 power of 2. If not present, or if the alignment is set to zero, the
506 alignment of the global is set by the target to whatever it feels
507 convenient. If an explicit alignment is specified, the global is forced
508 to have exactly that alignment. Targets and optimizers are not allowed
509 to over-align the global if the global has an assigned section. In this
510 case, the extra alignment could be observable: for example, code could
511 assume that the globals are densely packed in their section and try to
512 iterate over them as an array, alignment padding would break this
515 For example, the following defines a global in a numbered address space
516 with an initializer, section, and alignment:
520 @G = addrspace(5) constant float 1.0, section "foo", align 4
522 The following example defines a thread-local global with the
523 ``initialexec`` TLS model:
527 @G = thread_local(initialexec) global i32 0, align 4
529 .. _functionstructure:
534 LLVM function definitions consist of the "``define``" keyword, an
535 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
536 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
537 an optional ``unnamed_addr`` attribute, a return type, an optional
538 :ref:`parameter attribute <paramattrs>` for the return type, a function
539 name, a (possibly empty) argument list (each with optional :ref:`parameter
540 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
541 an optional section, an optional alignment, an optional :ref:`garbage
542 collector name <gc>`, an opening curly brace, a list of basic blocks,
543 and a closing curly brace.
545 LLVM function declarations consist of the "``declare``" keyword, an
546 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
547 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
548 an optional ``unnamed_addr`` attribute, a return type, an optional
549 :ref:`parameter attribute <paramattrs>` for the return type, a function
550 name, a possibly empty list of arguments, an optional alignment, and an
551 optional :ref:`garbage collector name <gc>`.
553 A function definition contains a list of basic blocks, forming the CFG
554 (Control Flow Graph) for the function. Each basic block may optionally
555 start with a label (giving the basic block a symbol table entry),
556 contains a list of instructions, and ends with a
557 :ref:`terminator <terminators>` instruction (such as a branch or function
560 The first basic block in a function is special in two ways: it is
561 immediately executed on entrance to the function, and it is not allowed
562 to have predecessor basic blocks (i.e. there can not be any branches to
563 the entry block of a function). Because the block can have no
564 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
566 LLVM allows an explicit section to be specified for functions. If the
567 target supports it, it will emit functions to the section specified.
569 An explicit alignment may be specified for a function. If not present,
570 or if the alignment is set to zero, the alignment of the function is set
571 by the target to whatever it feels convenient. If an explicit alignment
572 is specified, the function is forced to have at least that much
573 alignment. All alignments must be a power of 2.
575 If the ``unnamed_addr`` attribute is given, the address is know to not
576 be significant and two identical functions can be merged.
580 define [linkage] [visibility]
582 <ResultType> @<FunctionName> ([argument list])
583 [fn Attrs] [section "name"] [align N]
589 Aliases act as "second name" for the aliasee value (which can be either
590 function, global variable, another alias or bitcast of global value).
591 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
592 :ref:`visibility style <visibility>`.
596 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
598 .. _namedmetadatastructure:
603 Named metadata is a collection of metadata. :ref:`Metadata
604 nodes <metadata>` (but not metadata strings) are the only valid
605 operands for a named metadata.
609 ; Some unnamed metadata nodes, which are referenced by the named metadata.
610 !0 = metadata !{metadata !"zero"}
611 !1 = metadata !{metadata !"one"}
612 !2 = metadata !{metadata !"two"}
614 !name = !{!0, !1, !2}
621 The return type and each parameter of a function type may have a set of
622 *parameter attributes* associated with them. Parameter attributes are
623 used to communicate additional information about the result or
624 parameters of a function. Parameter attributes are considered to be part
625 of the function, not of the function type, so functions with different
626 parameter attributes can have the same function type.
628 Parameter attributes are simple keywords that follow the type specified.
629 If multiple parameter attributes are needed, they are space separated.
634 declare i32 @printf(i8* noalias nocapture, ...)
635 declare i32 @atoi(i8 zeroext)
636 declare signext i8 @returns_signed_char()
638 Note that any attributes for the function result (``nounwind``,
639 ``readonly``) come immediately after the argument list.
641 Currently, only the following parameter attributes are defined:
644 This indicates to the code generator that the parameter or return
645 value should be zero-extended to the extent required by the target's
646 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
647 the caller (for a parameter) or the callee (for a return value).
649 This indicates to the code generator that the parameter or return
650 value should be sign-extended to the extent required by the target's
651 ABI (which is usually 32-bits) by the caller (for a parameter) or
652 the callee (for a return value).
654 This indicates that this parameter or return value should be treated
655 in a special target-dependent fashion during while emitting code for
656 a function call or return (usually, by putting it in a register as
657 opposed to memory, though some targets use it to distinguish between
658 two different kinds of registers). Use of this attribute is
661 This indicates that the pointer parameter should really be passed by
662 value to the function. The attribute implies that a hidden copy of
663 the pointee is made between the caller and the callee, so the callee
664 is unable to modify the value in the caller. This attribute is only
665 valid on LLVM pointer arguments. It is generally used to pass
666 structs and arrays by value, but is also valid on pointers to
667 scalars. The copy is considered to belong to the caller not the
668 callee (for example, ``readonly`` functions should not write to
669 ``byval`` parameters). This is not a valid attribute for return
672 The byval attribute also supports specifying an alignment with the
673 align attribute. It indicates the alignment of the stack slot to
674 form and the known alignment of the pointer specified to the call
675 site. If the alignment is not specified, then the code generator
676 makes a target-specific assumption.
679 This indicates that the pointer parameter specifies the address of a
680 structure that is the return value of the function in the source
681 program. This pointer must be guaranteed by the caller to be valid:
682 loads and stores to the structure may be assumed by the callee to
683 not to trap and to be properly aligned. This may only be applied to
684 the first parameter. This is not a valid attribute for return
687 This indicates that pointer values `*based* <pointeraliasing>` on
688 the argument or return value do not alias pointer values which are
689 not *based* on it, ignoring certain "irrelevant" dependencies. For a
690 call to the parent function, dependencies between memory references
691 from before or after the call and from those during the call are
692 "irrelevant" to the ``noalias`` keyword for the arguments and return
693 value used in that call. The caller shares the responsibility with
694 the callee for ensuring that these requirements are met. For further
695 details, please see the discussion of the NoAlias response in `alias
696 analysis <AliasAnalysis.html#MustMayNo>`_.
698 Note that this definition of ``noalias`` is intentionally similar
699 to the definition of ``restrict`` in C99 for function arguments,
700 though it is slightly weaker.
702 For function return values, C99's ``restrict`` is not meaningful,
703 while LLVM's ``noalias`` is.
705 This indicates that the callee does not make any copies of the
706 pointer that outlive the callee itself. This is not a valid
707 attribute for return values.
712 This indicates that the pointer parameter can be excised using the
713 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
714 attribute for return values.
718 Garbage Collector Names
719 -----------------------
721 Each function may specify a garbage collector name, which is simply a
726 define void @f() gc "name" { ... }
728 The compiler declares the supported values of *name*. Specifying a
729 collector which will cause the compiler to alter its output in order to
730 support the named garbage collection algorithm.
737 Function attributes are set to communicate additional information about
738 a function. Function attributes are considered to be part of the
739 function, not of the function type, so functions with different function
740 attributes can have the same function type.
742 Function attributes are simple keywords that follow the type specified.
743 If multiple attributes are needed, they are space separated. For
748 define void @f() noinline { ... }
749 define void @f() alwaysinline { ... }
750 define void @f() alwaysinline optsize { ... }
751 define void @f() optsize { ... }
754 This attribute indicates that the address safety analysis is enabled
757 This attribute indicates that, when emitting the prologue and
758 epilogue, the backend should forcibly align the stack pointer.
759 Specify the desired alignment, which must be a power of two, in
762 This attribute indicates that the inliner should attempt to inline
763 this function into callers whenever possible, ignoring any active
764 inlining size threshold for this caller.
766 This attribute suppresses lazy symbol binding for the function. This
767 may make calls to the function faster, at the cost of extra program
768 startup time if the function is not called during program startup.
770 This attribute indicates that the source code contained a hint that
771 inlining this function is desirable (such as the "inline" keyword in
772 C/C++). It is just a hint; it imposes no requirements on the
775 This attribute disables prologue / epilogue emission for the
776 function. This can have very system-specific consequences.
778 This attributes disables implicit floating point instructions.
780 This attribute indicates that the inliner should never inline this
781 function in any situation. This attribute may not be used together
782 with the ``alwaysinline`` attribute.
784 This attribute indicates that the code generator should not use a
785 red zone, even if the target-specific ABI normally permits it.
787 This function attribute indicates that the function never returns
788 normally. This produces undefined behavior at runtime if the
789 function ever does dynamically return.
791 This function attribute indicates that the function never returns
792 with an unwind or exceptional control flow. If the function does
793 unwind, its runtime behavior is undefined.
795 This attribute suggests that optimization passes and code generator
796 passes make choices that keep the code size of this function low,
797 and otherwise do optimizations specifically to reduce code size.
799 This attribute indicates that the function computes its result (or
800 decides to unwind an exception) based strictly on its arguments,
801 without dereferencing any pointer arguments or otherwise accessing
802 any mutable state (e.g. memory, control registers, etc) visible to
803 caller functions. It does not write through any pointer arguments
804 (including ``byval`` arguments) and never changes any state visible
805 to callers. This means that it cannot unwind exceptions by calling
806 the ``C++`` exception throwing methods.
808 This attribute indicates that the function does not write through
809 any pointer arguments (including ``byval`` arguments) or otherwise
810 modify any state (e.g. memory, control registers, etc) visible to
811 caller functions. It may dereference pointer arguments and read
812 state that may be set in the caller. A readonly function always
813 returns the same value (or unwinds an exception identically) when
814 called with the same set of arguments and global state. It cannot
815 unwind an exception by calling the ``C++`` exception throwing
818 This attribute indicates that this function can return twice. The C
819 ``setjmp`` is an example of such a function. The compiler disables
820 some optimizations (like tail calls) in the caller of these
823 This attribute indicates that the function should emit a stack
824 smashing protector. It is in the form of a "canary"—a random value
825 placed on the stack before the local variables that's checked upon
826 return from the function to see if it has been overwritten. A
827 heuristic is used to determine if a function needs stack protectors
830 If a function that has an ``ssp`` attribute is inlined into a
831 function that doesn't have an ``ssp`` attribute, then the resulting
832 function will have an ``ssp`` attribute.
834 This attribute indicates that the function should *always* emit a
835 stack smashing protector. This overrides the ``ssp`` function
838 If a function that has an ``sspreq`` attribute is inlined into a
839 function that doesn't have an ``sspreq`` attribute or which has an
840 ``ssp`` attribute, then the resulting function will have an
841 ``sspreq`` attribute.
843 This attribute indicates that the ABI being targeted requires that
844 an unwind table entry be produce for this function even if we can
845 show that no exceptions passes by it. This is normally the case for
846 the ELF x86-64 abi, but it can be disabled for some compilation
849 This attribute indicates that calls to the function cannot be
850 duplicated. A call to a ``noduplicate`` function may be moved
851 within its parent function, but may not be duplicated within
854 A function containing a ``noduplicate`` call may still
855 be an inlining candidate, provided that the call is not
856 duplicated by inlining. That implies that the function has
857 internal linkage and only has one call site, so the original
858 call is dead after inlining.
862 Module-Level Inline Assembly
863 ----------------------------
865 Modules may contain "module-level inline asm" blocks, which corresponds
866 to the GCC "file scope inline asm" blocks. These blocks are internally
867 concatenated by LLVM and treated as a single unit, but may be separated
868 in the ``.ll`` file if desired. The syntax is very simple:
872 module asm "inline asm code goes here"
873 module asm "more can go here"
875 The strings can contain any character by escaping non-printable
876 characters. The escape sequence used is simply "\\xx" where "xx" is the
877 two digit hex code for the number.
879 The inline asm code is simply printed to the machine code .s file when
880 assembly code is generated.
885 A module may specify a target specific data layout string that specifies
886 how data is to be laid out in memory. The syntax for the data layout is
891 target datalayout = "layout specification"
893 The *layout specification* consists of a list of specifications
894 separated by the minus sign character ('-'). Each specification starts
895 with a letter and may include other information after the letter to
896 define some aspect of the data layout. The specifications accepted are
900 Specifies that the target lays out data in big-endian form. That is,
901 the bits with the most significance have the lowest address
904 Specifies that the target lays out data in little-endian form. That
905 is, the bits with the least significance have the lowest address
908 Specifies the natural alignment of the stack in bits. Alignment
909 promotion of stack variables is limited to the natural stack
910 alignment to avoid dynamic stack realignment. The stack alignment
911 must be a multiple of 8-bits. If omitted, the natural stack
912 alignment defaults to "unspecified", which does not prevent any
913 alignment promotions.
914 ``p[n]:<size>:<abi>:<pref>``
915 This specifies the *size* of a pointer and its ``<abi>`` and
916 ``<pref>``\erred alignments for address space ``n``. All sizes are in
917 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
918 preceding ``:`` should be omitted too. The address space, ``n`` is
919 optional, and if not specified, denotes the default address space 0.
920 The value of ``n`` must be in the range [1,2^23).
921 ``i<size>:<abi>:<pref>``
922 This specifies the alignment for an integer type of a given bit
923 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
924 ``v<size>:<abi>:<pref>``
925 This specifies the alignment for a vector type of a given bit
927 ``f<size>:<abi>:<pref>``
928 This specifies the alignment for a floating point type of a given bit
929 ``<size>``. Only values of ``<size>`` that are supported by the target
930 will work. 32 (float) and 64 (double) are supported on all targets; 80
931 or 128 (different flavors of long double) are also supported on some
933 ``a<size>:<abi>:<pref>``
934 This specifies the alignment for an aggregate type of a given bit
936 ``s<size>:<abi>:<pref>``
937 This specifies the alignment for a stack object of a given bit
939 ``n<size1>:<size2>:<size3>...``
940 This specifies a set of native integer widths for the target CPU in
941 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
942 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
943 this set are considered to support most general arithmetic operations
946 When constructing the data layout for a given target, LLVM starts with a
947 default set of specifications which are then (possibly) overridden by
948 the specifications in the ``datalayout`` keyword. The default
949 specifications are given in this list:
952 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
953 - ``p1:32:32:32`` - 32-bit pointers with 32-bit alignment for address
955 - ``p2:16:32:32`` - 16-bit pointers with 32-bit alignment for address
957 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
958 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
959 - ``i16:16:16`` - i16 is 16-bit aligned
960 - ``i32:32:32`` - i32 is 32-bit aligned
961 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
963 - ``f32:32:32`` - float is 32-bit aligned
964 - ``f64:64:64`` - double is 64-bit aligned
965 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
966 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
967 - ``a0:0:1`` - aggregates are 8-bit aligned
968 - ``s0:64:64`` - stack objects are 64-bit aligned
970 When LLVM is determining the alignment for a given type, it uses the
973 #. If the type sought is an exact match for one of the specifications,
974 that specification is used.
975 #. If no match is found, and the type sought is an integer type, then
976 the smallest integer type that is larger than the bitwidth of the
977 sought type is used. If none of the specifications are larger than
978 the bitwidth then the largest integer type is used. For example,
979 given the default specifications above, the i7 type will use the
980 alignment of i8 (next largest) while both i65 and i256 will use the
981 alignment of i64 (largest specified).
982 #. If no match is found, and the type sought is a vector type, then the
983 largest vector type that is smaller than the sought vector type will
984 be used as a fall back. This happens because <128 x double> can be
985 implemented in terms of 64 <2 x double>, for example.
987 The function of the data layout string may not be what you expect.
988 Notably, this is not a specification from the frontend of what alignment
989 the code generator should use.
991 Instead, if specified, the target data layout is required to match what
992 the ultimate *code generator* expects. This string is used by the
993 mid-level optimizers to improve code, and this only works if it matches
994 what the ultimate code generator uses. If you would like to generate IR
995 that does not embed this target-specific detail into the IR, then you
996 don't have to specify the string. This will disable some optimizations
997 that require precise layout information, but this also prevents those
998 optimizations from introducing target specificity into the IR.
1000 .. _pointeraliasing:
1002 Pointer Aliasing Rules
1003 ----------------------
1005 Any memory access must be done through a pointer value associated with
1006 an address range of the memory access, otherwise the behavior is
1007 undefined. Pointer values are associated with address ranges according
1008 to the following rules:
1010 - A pointer value is associated with the addresses associated with any
1011 value it is *based* on.
1012 - An address of a global variable is associated with the address range
1013 of the variable's storage.
1014 - The result value of an allocation instruction is associated with the
1015 address range of the allocated storage.
1016 - A null pointer in the default address-space is associated with no
1018 - An integer constant other than zero or a pointer value returned from
1019 a function not defined within LLVM may be associated with address
1020 ranges allocated through mechanisms other than those provided by
1021 LLVM. Such ranges shall not overlap with any ranges of addresses
1022 allocated by mechanisms provided by LLVM.
1024 A pointer value is *based* on another pointer value according to the
1027 - A pointer value formed from a ``getelementptr`` operation is *based*
1028 on the first operand of the ``getelementptr``.
1029 - The result value of a ``bitcast`` is *based* on the operand of the
1031 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1032 values that contribute (directly or indirectly) to the computation of
1033 the pointer's value.
1034 - The "*based* on" relationship is transitive.
1036 Note that this definition of *"based"* is intentionally similar to the
1037 definition of *"based"* in C99, though it is slightly weaker.
1039 LLVM IR does not associate types with memory. The result type of a
1040 ``load`` merely indicates the size and alignment of the memory from
1041 which to load, as well as the interpretation of the value. The first
1042 operand type of a ``store`` similarly only indicates the size and
1043 alignment of the store.
1045 Consequently, type-based alias analysis, aka TBAA, aka
1046 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1047 :ref:`Metadata <metadata>` may be used to encode additional information
1048 which specialized optimization passes may use to implement type-based
1053 Volatile Memory Accesses
1054 ------------------------
1056 Certain memory accesses, such as :ref:`load <i_load>`'s,
1057 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1058 marked ``volatile``. The optimizers must not change the number of
1059 volatile operations or change their order of execution relative to other
1060 volatile operations. The optimizers *may* change the order of volatile
1061 operations relative to non-volatile operations. This is not Java's
1062 "volatile" and has no cross-thread synchronization behavior.
1066 Memory Model for Concurrent Operations
1067 --------------------------------------
1069 The LLVM IR does not define any way to start parallel threads of
1070 execution or to register signal handlers. Nonetheless, there are
1071 platform-specific ways to create them, and we define LLVM IR's behavior
1072 in their presence. This model is inspired by the C++0x memory model.
1074 For a more informal introduction to this model, see the :doc:`Atomics`.
1076 We define a *happens-before* partial order as the least partial order
1079 - Is a superset of single-thread program order, and
1080 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1081 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1082 techniques, like pthread locks, thread creation, thread joining,
1083 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1084 Constraints <ordering>`).
1086 Note that program order does not introduce *happens-before* edges
1087 between a thread and signals executing inside that thread.
1089 Every (defined) read operation (load instructions, memcpy, atomic
1090 loads/read-modify-writes, etc.) R reads a series of bytes written by
1091 (defined) write operations (store instructions, atomic
1092 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1093 section, initialized globals are considered to have a write of the
1094 initializer which is atomic and happens before any other read or write
1095 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1096 may see any write to the same byte, except:
1098 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1099 write\ :sub:`2` happens before R\ :sub:`byte`, then
1100 R\ :sub:`byte` does not see write\ :sub:`1`.
1101 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1102 R\ :sub:`byte` does not see write\ :sub:`3`.
1104 Given that definition, R\ :sub:`byte` is defined as follows:
1106 - If R is volatile, the result is target-dependent. (Volatile is
1107 supposed to give guarantees which can support ``sig_atomic_t`` in
1108 C/C++, and may be used for accesses to addresses which do not behave
1109 like normal memory. It does not generally provide cross-thread
1111 - Otherwise, if there is no write to the same byte that happens before
1112 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1113 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1114 R\ :sub:`byte` returns the value written by that write.
1115 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1116 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1117 Memory Ordering Constraints <ordering>` section for additional
1118 constraints on how the choice is made.
1119 - Otherwise R\ :sub:`byte` returns ``undef``.
1121 R returns the value composed of the series of bytes it read. This
1122 implies that some bytes within the value may be ``undef`` **without**
1123 the entire value being ``undef``. Note that this only defines the
1124 semantics of the operation; it doesn't mean that targets will emit more
1125 than one instruction to read the series of bytes.
1127 Note that in cases where none of the atomic intrinsics are used, this
1128 model places only one restriction on IR transformations on top of what
1129 is required for single-threaded execution: introducing a store to a byte
1130 which might not otherwise be stored is not allowed in general.
1131 (Specifically, in the case where another thread might write to and read
1132 from an address, introducing a store can change a load that may see
1133 exactly one write into a load that may see multiple writes.)
1137 Atomic Memory Ordering Constraints
1138 ----------------------------------
1140 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1141 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1142 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1143 an ordering parameter that determines which other atomic instructions on
1144 the same address they *synchronize with*. These semantics are borrowed
1145 from Java and C++0x, but are somewhat more colloquial. If these
1146 descriptions aren't precise enough, check those specs (see spec
1147 references in the :doc:`atomics guide <Atomics>`).
1148 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1149 differently since they don't take an address. See that instruction's
1150 documentation for details.
1152 For a simpler introduction to the ordering constraints, see the
1156 The set of values that can be read is governed by the happens-before
1157 partial order. A value cannot be read unless some operation wrote
1158 it. This is intended to provide a guarantee strong enough to model
1159 Java's non-volatile shared variables. This ordering cannot be
1160 specified for read-modify-write operations; it is not strong enough
1161 to make them atomic in any interesting way.
1163 In addition to the guarantees of ``unordered``, there is a single
1164 total order for modifications by ``monotonic`` operations on each
1165 address. All modification orders must be compatible with the
1166 happens-before order. There is no guarantee that the modification
1167 orders can be combined to a global total order for the whole program
1168 (and this often will not be possible). The read in an atomic
1169 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1170 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1171 order immediately before the value it writes. If one atomic read
1172 happens before another atomic read of the same address, the later
1173 read must see the same value or a later value in the address's
1174 modification order. This disallows reordering of ``monotonic`` (or
1175 stronger) operations on the same address. If an address is written
1176 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1177 read that address repeatedly, the other threads must eventually see
1178 the write. This corresponds to the C++0x/C1x
1179 ``memory_order_relaxed``.
1181 In addition to the guarantees of ``monotonic``, a
1182 *synchronizes-with* edge may be formed with a ``release`` operation.
1183 This is intended to model C++'s ``memory_order_acquire``.
1185 In addition to the guarantees of ``monotonic``, if this operation
1186 writes a value which is subsequently read by an ``acquire``
1187 operation, it *synchronizes-with* that operation. (This isn't a
1188 complete description; see the C++0x definition of a release
1189 sequence.) This corresponds to the C++0x/C1x
1190 ``memory_order_release``.
1191 ``acq_rel`` (acquire+release)
1192 Acts as both an ``acquire`` and ``release`` operation on its
1193 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1194 ``seq_cst`` (sequentially consistent)
1195 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1196 operation which only reads, ``release`` for an operation which only
1197 writes), there is a global total order on all
1198 sequentially-consistent operations on all addresses, which is
1199 consistent with the *happens-before* partial order and with the
1200 modification orders of all the affected addresses. Each
1201 sequentially-consistent read sees the last preceding write to the
1202 same address in this global order. This corresponds to the C++0x/C1x
1203 ``memory_order_seq_cst`` and Java volatile.
1207 If an atomic operation is marked ``singlethread``, it only *synchronizes
1208 with* or participates in modification and seq\_cst total orderings with
1209 other operations running in the same thread (for example, in signal
1217 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1218 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1219 :ref:`frem <i_frem>`) have the following flags that can set to enable
1220 otherwise unsafe floating point operations
1223 No NaNs - Allow optimizations to assume the arguments and result are not
1224 NaN. Such optimizations are required to retain defined behavior over
1225 NaNs, but the value of the result is undefined.
1228 No Infs - Allow optimizations to assume the arguments and result are not
1229 +/-Inf. Such optimizations are required to retain defined behavior over
1230 +/-Inf, but the value of the result is undefined.
1233 No Signed Zeros - Allow optimizations to treat the sign of a zero
1234 argument or result as insignificant.
1237 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1238 argument rather than perform division.
1241 Fast - Allow algebraically equivalent transformations that may
1242 dramatically change results in floating point (e.g. reassociate). This
1243 flag implies all the others.
1250 The LLVM type system is one of the most important features of the
1251 intermediate representation. Being typed enables a number of
1252 optimizations to be performed on the intermediate representation
1253 directly, without having to do extra analyses on the side before the
1254 transformation. A strong type system makes it easier to read the
1255 generated code and enables novel analyses and transformations that are
1256 not feasible to perform on normal three address code representations.
1258 Type Classifications
1259 --------------------
1261 The types fall into a few useful classifications:
1270 * - :ref:`integer <t_integer>`
1271 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1274 * - :ref:`floating point <t_floating>`
1275 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1283 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1284 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1285 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1286 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1288 * - :ref:`primitive <t_primitive>`
1289 - :ref:`label <t_label>`,
1290 :ref:`void <t_void>`,
1291 :ref:`integer <t_integer>`,
1292 :ref:`floating point <t_floating>`,
1293 :ref:`x86mmx <t_x86mmx>`,
1294 :ref:`metadata <t_metadata>`.
1296 * - :ref:`derived <t_derived>`
1297 - :ref:`array <t_array>`,
1298 :ref:`function <t_function>`,
1299 :ref:`pointer <t_pointer>`,
1300 :ref:`structure <t_struct>`,
1301 :ref:`vector <t_vector>`,
1302 :ref:`opaque <t_opaque>`.
1304 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1305 Values of these types are the only ones which can be produced by
1313 The primitive types are the fundamental building blocks of the LLVM
1324 The integer type is a very simple type that simply specifies an
1325 arbitrary bit width for the integer type desired. Any bit width from 1
1326 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1335 The number of bits the integer will occupy is specified by the ``N``
1341 +----------------+------------------------------------------------+
1342 | ``i1`` | a single-bit integer. |
1343 +----------------+------------------------------------------------+
1344 | ``i32`` | a 32-bit integer. |
1345 +----------------+------------------------------------------------+
1346 | ``i1942652`` | a really big integer of over 1 million bits. |
1347 +----------------+------------------------------------------------+
1351 Floating Point Types
1352 ^^^^^^^^^^^^^^^^^^^^
1361 - 16-bit floating point value
1364 - 32-bit floating point value
1367 - 64-bit floating point value
1370 - 128-bit floating point value (112-bit mantissa)
1373 - 80-bit floating point value (X87)
1376 - 128-bit floating point value (two 64-bits)
1386 The x86mmx type represents a value held in an MMX register on an x86
1387 machine. The operations allowed on it are quite limited: parameters and
1388 return values, load and store, and bitcast. User-specified MMX
1389 instructions are represented as intrinsic or asm calls with arguments
1390 and/or results of this type. There are no arrays, vectors or constants
1408 The void type does not represent any value and has no size.
1425 The label type represents code labels.
1442 The metadata type represents embedded metadata. No derived types may be
1443 created from metadata except for :ref:`function <t_function>` arguments.
1457 The real power in LLVM comes from the derived types in the system. This
1458 is what allows a programmer to represent arrays, functions, pointers,
1459 and other useful types. Each of these types contain one or more element
1460 types which may be a primitive type, or another derived type. For
1461 example, it is possible to have a two dimensional array, using an array
1462 as the element type of another array.
1469 Aggregate Types are a subset of derived types that can contain multiple
1470 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1471 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1482 The array type is a very simple derived type that arranges elements
1483 sequentially in memory. The array type requires a size (number of
1484 elements) and an underlying data type.
1491 [<# elements> x <elementtype>]
1493 The number of elements is a constant integer value; ``elementtype`` may
1494 be any type with a size.
1499 +------------------+--------------------------------------+
1500 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1501 +------------------+--------------------------------------+
1502 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1503 +------------------+--------------------------------------+
1504 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1505 +------------------+--------------------------------------+
1507 Here are some examples of multidimensional arrays:
1509 +-----------------------------+----------------------------------------------------------+
1510 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1511 +-----------------------------+----------------------------------------------------------+
1512 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1513 +-----------------------------+----------------------------------------------------------+
1514 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1515 +-----------------------------+----------------------------------------------------------+
1517 There is no restriction on indexing beyond the end of the array implied
1518 by a static type (though there are restrictions on indexing beyond the
1519 bounds of an allocated object in some cases). This means that
1520 single-dimension 'variable sized array' addressing can be implemented in
1521 LLVM with a zero length array type. An implementation of 'pascal style
1522 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1533 The function type can be thought of as a function signature. It consists
1534 of a return type and a list of formal parameter types. The return type
1535 of a function type is a first class type or a void type.
1542 <returntype> (<parameter list>)
1544 ...where '``<parameter list>``' is a comma-separated list of type
1545 specifiers. Optionally, the parameter list may include a type ``...``,
1546 which indicates that the function takes a variable number of arguments.
1547 Variable argument functions can access their arguments with the
1548 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1549 '``<returntype>``' is any type except :ref:`label <t_label>`.
1554 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1555 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1556 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1557 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1558 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1559 | ``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. |
1560 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1561 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1562 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1572 The structure type is used to represent a collection of data members
1573 together in memory. The elements of a structure may be any type that has
1576 Structures in memory are accessed using '``load``' and '``store``' by
1577 getting a pointer to a field with the '``getelementptr``' instruction.
1578 Structures in registers are accessed using the '``extractvalue``' and
1579 '``insertvalue``' instructions.
1581 Structures may optionally be "packed" structures, which indicate that
1582 the alignment of the struct is one byte, and that there is no padding
1583 between the elements. In non-packed structs, padding between field types
1584 is inserted as defined by the DataLayout string in the module, which is
1585 required to match what the underlying code generator expects.
1587 Structures can either be "literal" or "identified". A literal structure
1588 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1589 identified types are always defined at the top level with a name.
1590 Literal types are uniqued by their contents and can never be recursive
1591 or opaque since there is no way to write one. Identified types can be
1592 recursive, can be opaqued, and are never uniqued.
1599 %T1 = type { <type list> } ; Identified normal struct type
1600 %T2 = type <{ <type list> }> ; Identified packed struct type
1605 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1606 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1607 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1608 | ``{ 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``. |
1609 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1610 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1611 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1615 Opaque Structure Types
1616 ^^^^^^^^^^^^^^^^^^^^^^
1621 Opaque structure types are used to represent named structure types that
1622 do not have a body specified. This corresponds (for example) to the C
1623 notion of a forward declared structure.
1636 +--------------+-------------------+
1637 | ``opaque`` | An opaque type. |
1638 +--------------+-------------------+
1648 The pointer type is used to specify memory locations. Pointers are
1649 commonly used to reference objects in memory.
1651 Pointer types may have an optional address space attribute defining the
1652 numbered address space where the pointed-to object resides. The default
1653 address space is number zero. The semantics of non-zero address spaces
1654 are target-specific.
1656 Note that LLVM does not permit pointers to void (``void*``) nor does it
1657 permit pointers to labels (``label*``). Use ``i8*`` instead.
1669 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1670 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1671 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1672 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1673 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1674 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1675 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1685 A vector type is a simple derived type that represents a vector of
1686 elements. Vector types are used when multiple primitive data are
1687 operated in parallel using a single instruction (SIMD). A vector type
1688 requires a size (number of elements) and an underlying primitive data
1689 type. Vector types are considered :ref:`first class <t_firstclass>`.
1696 < <# elements> x <elementtype> >
1698 The number of elements is a constant integer value larger than 0;
1699 elementtype may be any integer or floating point type, or a pointer to
1700 these types. Vectors of size zero are not allowed.
1705 +-------------------+--------------------------------------------------+
1706 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1707 +-------------------+--------------------------------------------------+
1708 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1709 +-------------------+--------------------------------------------------+
1710 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1711 +-------------------+--------------------------------------------------+
1712 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1713 +-------------------+--------------------------------------------------+
1718 LLVM has several different basic types of constants. This section
1719 describes them all and their syntax.
1724 **Boolean constants**
1725 The two strings '``true``' and '``false``' are both valid constants
1727 **Integer constants**
1728 Standard integers (such as '4') are constants of the
1729 :ref:`integer <t_integer>` type. Negative numbers may be used with
1731 **Floating point constants**
1732 Floating point constants use standard decimal notation (e.g.
1733 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1734 hexadecimal notation (see below). The assembler requires the exact
1735 decimal value of a floating-point constant. For example, the
1736 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1737 decimal in binary. Floating point constants must have a :ref:`floating
1738 point <t_floating>` type.
1739 **Null pointer constants**
1740 The identifier '``null``' is recognized as a null pointer constant
1741 and must be of :ref:`pointer type <t_pointer>`.
1743 The one non-intuitive notation for constants is the hexadecimal form of
1744 floating point constants. For example, the form
1745 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1746 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1747 constants are required (and the only time that they are generated by the
1748 disassembler) is when a floating point constant must be emitted but it
1749 cannot be represented as a decimal floating point number in a reasonable
1750 number of digits. For example, NaN's, infinities, and other special
1751 values are represented in their IEEE hexadecimal format so that assembly
1752 and disassembly do not cause any bits to change in the constants.
1754 When using the hexadecimal form, constants of types half, float, and
1755 double are represented using the 16-digit form shown above (which
1756 matches the IEEE754 representation for double); half and float values
1757 must, however, be exactly representable as IEE754 half and single
1758 precision, respectively. Hexadecimal format is always used for long
1759 double, and there are three forms of long double. The 80-bit format used
1760 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1761 128-bit format used by PowerPC (two adjacent doubles) is represented by
1762 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1763 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1764 supported target uses this format. Long doubles will only work if they
1765 match the long double format on your target. The IEEE 16-bit format
1766 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1767 digits. All hexadecimal formats are big-endian (sign bit at the left).
1769 There are no constants of type x86mmx.
1774 Complex constants are a (potentially recursive) combination of simple
1775 constants and smaller complex constants.
1777 **Structure constants**
1778 Structure constants are represented with notation similar to
1779 structure type definitions (a comma separated list of elements,
1780 surrounded by braces (``{}``)). For example:
1781 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1782 "``@G = external global i32``". Structure constants must have
1783 :ref:`structure type <t_struct>`, and the number and types of elements
1784 must match those specified by the type.
1786 Array constants are represented with notation similar to array type
1787 definitions (a comma separated list of elements, surrounded by
1788 square brackets (``[]``)). For example:
1789 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1790 :ref:`array type <t_array>`, and the number and types of elements must
1791 match those specified by the type.
1792 **Vector constants**
1793 Vector constants are represented with notation similar to vector
1794 type definitions (a comma separated list of elements, surrounded by
1795 less-than/greater-than's (``<>``)). For example:
1796 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1797 must have :ref:`vector type <t_vector>`, and the number and types of
1798 elements must match those specified by the type.
1799 **Zero initialization**
1800 The string '``zeroinitializer``' can be used to zero initialize a
1801 value to zero of *any* type, including scalar and
1802 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1803 having to print large zero initializers (e.g. for large arrays) and
1804 is always exactly equivalent to using explicit zero initializers.
1806 A metadata node is a structure-like constant with :ref:`metadata
1807 type <t_metadata>`. For example:
1808 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1809 constants that are meant to be interpreted as part of the
1810 instruction stream, metadata is a place to attach additional
1811 information such as debug info.
1813 Global Variable and Function Addresses
1814 --------------------------------------
1816 The addresses of :ref:`global variables <globalvars>` and
1817 :ref:`functions <functionstructure>` are always implicitly valid
1818 (link-time) constants. These constants are explicitly referenced when
1819 the :ref:`identifier for the global <identifiers>` is used and always have
1820 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1823 .. code-block:: llvm
1827 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1834 The string '``undef``' can be used anywhere a constant is expected, and
1835 indicates that the user of the value may receive an unspecified
1836 bit-pattern. Undefined values may be of any type (other than '``label``'
1837 or '``void``') and be used anywhere a constant is permitted.
1839 Undefined values are useful because they indicate to the compiler that
1840 the program is well defined no matter what value is used. This gives the
1841 compiler more freedom to optimize. Here are some examples of
1842 (potentially surprising) transformations that are valid (in pseudo IR):
1844 .. code-block:: llvm
1854 This is safe because all of the output bits are affected by the undef
1855 bits. Any output bit can have a zero or one depending on the input bits.
1857 .. code-block:: llvm
1868 These logical operations have bits that are not always affected by the
1869 input. For example, if ``%X`` has a zero bit, then the output of the
1870 '``and``' operation will always be a zero for that bit, no matter what
1871 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1872 optimize or assume that the result of the '``and``' is '``undef``'.
1873 However, it is safe to assume that all bits of the '``undef``' could be
1874 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1875 all the bits of the '``undef``' operand to the '``or``' could be set,
1876 allowing the '``or``' to be folded to -1.
1878 .. code-block:: llvm
1880 %A = select undef, %X, %Y
1881 %B = select undef, 42, %Y
1882 %C = select %X, %Y, undef
1892 This set of examples shows that undefined '``select``' (and conditional
1893 branch) conditions can go *either way*, but they have to come from one
1894 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1895 both known to have a clear low bit, then ``%A`` would have to have a
1896 cleared low bit. However, in the ``%C`` example, the optimizer is
1897 allowed to assume that the '``undef``' operand could be the same as
1898 ``%Y``, allowing the whole '``select``' to be eliminated.
1900 .. code-block:: llvm
1902 %A = xor undef, undef
1919 This example points out that two '``undef``' operands are not
1920 necessarily the same. This can be surprising to people (and also matches
1921 C semantics) where they assume that "``X^X``" is always zero, even if
1922 ``X`` is undefined. This isn't true for a number of reasons, but the
1923 short answer is that an '``undef``' "variable" can arbitrarily change
1924 its value over its "live range". This is true because the variable
1925 doesn't actually *have a live range*. Instead, the value is logically
1926 read from arbitrary registers that happen to be around when needed, so
1927 the value is not necessarily consistent over time. In fact, ``%A`` and
1928 ``%C`` need to have the same semantics or the core LLVM "replace all
1929 uses with" concept would not hold.
1931 .. code-block:: llvm
1939 These examples show the crucial difference between an *undefined value*
1940 and *undefined behavior*. An undefined value (like '``undef``') is
1941 allowed to have an arbitrary bit-pattern. This means that the ``%A``
1942 operation can be constant folded to '``undef``', because the '``undef``'
1943 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
1944 However, in the second example, we can make a more aggressive
1945 assumption: because the ``undef`` is allowed to be an arbitrary value,
1946 we are allowed to assume that it could be zero. Since a divide by zero
1947 has *undefined behavior*, we are allowed to assume that the operation
1948 does not execute at all. This allows us to delete the divide and all
1949 code after it. Because the undefined operation "can't happen", the
1950 optimizer can assume that it occurs in dead code.
1952 .. code-block:: llvm
1954 a: store undef -> %X
1955 b: store %X -> undef
1960 These examples reiterate the ``fdiv`` example: a store *of* an undefined
1961 value can be assumed to not have any effect; we can assume that the
1962 value is overwritten with bits that happen to match what was already
1963 there. However, a store *to* an undefined location could clobber
1964 arbitrary memory, therefore, it has undefined behavior.
1971 Poison values are similar to :ref:`undef values <undefvalues>`, however
1972 they also represent the fact that an instruction or constant expression
1973 which cannot evoke side effects has nevertheless detected a condition
1974 which results in undefined behavior.
1976 There is currently no way of representing a poison value in the IR; they
1977 only exist when produced by operations such as :ref:`add <i_add>` with
1980 Poison value behavior is defined in terms of value *dependence*:
1982 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
1983 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
1984 their dynamic predecessor basic block.
1985 - Function arguments depend on the corresponding actual argument values
1986 in the dynamic callers of their functions.
1987 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
1988 instructions that dynamically transfer control back to them.
1989 - :ref:`Invoke <i_invoke>` instructions depend on the
1990 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
1991 call instructions that dynamically transfer control back to them.
1992 - Non-volatile loads and stores depend on the most recent stores to all
1993 of the referenced memory addresses, following the order in the IR
1994 (including loads and stores implied by intrinsics such as
1995 :ref:`@llvm.memcpy <int_memcpy>`.)
1996 - An instruction with externally visible side effects depends on the
1997 most recent preceding instruction with externally visible side
1998 effects, following the order in the IR. (This includes :ref:`volatile
1999 operations <volatile>`.)
2000 - An instruction *control-depends* on a :ref:`terminator
2001 instruction <terminators>` if the terminator instruction has
2002 multiple successors and the instruction is always executed when
2003 control transfers to one of the successors, and may not be executed
2004 when control is transferred to another.
2005 - Additionally, an instruction also *control-depends* on a terminator
2006 instruction if the set of instructions it otherwise depends on would
2007 be different if the terminator had transferred control to a different
2009 - Dependence is transitive.
2011 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2012 with the additional affect that any instruction which has a *dependence*
2013 on a poison value has undefined behavior.
2015 Here are some examples:
2017 .. code-block:: llvm
2020 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2021 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2022 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2023 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2025 store i32 %poison, i32* @g ; Poison value stored to memory.
2026 %poison2 = load i32* @g ; Poison value loaded back from memory.
2028 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2030 %narrowaddr = bitcast i32* @g to i16*
2031 %wideaddr = bitcast i32* @g to i64*
2032 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2033 %poison4 = load i64* %wideaddr ; Returns a poison value.
2035 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2036 br i1 %cmp, label %true, label %end ; Branch to either destination.
2039 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2040 ; it has undefined behavior.
2044 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2045 ; Both edges into this PHI are
2046 ; control-dependent on %cmp, so this
2047 ; always results in a poison value.
2049 store volatile i32 0, i32* @g ; This would depend on the store in %true
2050 ; if %cmp is true, or the store in %entry
2051 ; otherwise, so this is undefined behavior.
2053 br i1 %cmp, label %second_true, label %second_end
2054 ; The same branch again, but this time the
2055 ; true block doesn't have side effects.
2062 store volatile i32 0, i32* @g ; This time, the instruction always depends
2063 ; on the store in %end. Also, it is
2064 ; control-equivalent to %end, so this is
2065 ; well-defined (ignoring earlier undefined
2066 ; behavior in this example).
2070 Addresses of Basic Blocks
2071 -------------------------
2073 ``blockaddress(@function, %block)``
2075 The '``blockaddress``' constant computes the address of the specified
2076 basic block in the specified function, and always has an ``i8*`` type.
2077 Taking the address of the entry block is illegal.
2079 This value only has defined behavior when used as an operand to the
2080 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2081 against null. Pointer equality tests between labels addresses results in
2082 undefined behavior — though, again, comparison against null is ok, and
2083 no label is equal to the null pointer. This may be passed around as an
2084 opaque pointer sized value as long as the bits are not inspected. This
2085 allows ``ptrtoint`` and arithmetic to be performed on these values so
2086 long as the original value is reconstituted before the ``indirectbr``
2089 Finally, some targets may provide defined semantics when using the value
2090 as the operand to an inline assembly, but that is target specific.
2092 Constant Expressions
2093 --------------------
2095 Constant expressions are used to allow expressions involving other
2096 constants to be used as constants. Constant expressions may be of any
2097 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2098 that does not have side effects (e.g. load and call are not supported).
2099 The following is the syntax for constant expressions:
2101 ``trunc (CST to TYPE)``
2102 Truncate a constant to another type. The bit size of CST must be
2103 larger than the bit size of TYPE. Both types must be integers.
2104 ``zext (CST to TYPE)``
2105 Zero extend a constant to another type. The bit size of CST must be
2106 smaller than the bit size of TYPE. Both types must be integers.
2107 ``sext (CST to TYPE)``
2108 Sign extend a constant to another type. The bit size of CST must be
2109 smaller than the bit size of TYPE. Both types must be integers.
2110 ``fptrunc (CST to TYPE)``
2111 Truncate a floating point constant to another floating point type.
2112 The size of CST must be larger than the size of TYPE. Both types
2113 must be floating point.
2114 ``fpext (CST to TYPE)``
2115 Floating point extend a constant to another type. The size of CST
2116 must be smaller or equal to the size of TYPE. Both types must be
2118 ``fptoui (CST to TYPE)``
2119 Convert a floating point constant to the corresponding unsigned
2120 integer constant. TYPE must be a scalar or vector integer type. CST
2121 must be of scalar or vector floating point type. Both CST and TYPE
2122 must be scalars, or vectors of the same number of elements. If the
2123 value won't fit in the integer type, the results are undefined.
2124 ``fptosi (CST to TYPE)``
2125 Convert a floating point constant to the corresponding signed
2126 integer constant. TYPE must be a scalar or vector integer type. CST
2127 must be of scalar or vector floating point type. Both CST and TYPE
2128 must be scalars, or vectors of the same number of elements. If the
2129 value won't fit in the integer type, the results are undefined.
2130 ``uitofp (CST to TYPE)``
2131 Convert an unsigned integer constant to the corresponding floating
2132 point constant. TYPE must be a scalar or vector floating point type.
2133 CST must be of scalar or vector integer type. Both CST and TYPE must
2134 be scalars, or vectors of the same number of elements. If the value
2135 won't fit in the floating point type, the results are undefined.
2136 ``sitofp (CST to TYPE)``
2137 Convert a signed integer constant to the corresponding floating
2138 point constant. TYPE must be a scalar or vector floating point type.
2139 CST must be of scalar or vector integer type. Both CST and TYPE must
2140 be scalars, or vectors of the same number of elements. If the value
2141 won't fit in the floating point type, the results are undefined.
2142 ``ptrtoint (CST to TYPE)``
2143 Convert a pointer typed constant to the corresponding integer
2144 constant ``TYPE`` must be an integer type. ``CST`` must be of
2145 pointer type. The ``CST`` value is zero extended, truncated, or
2146 unchanged to make it fit in ``TYPE``.
2147 ``inttoptr (CST to TYPE)``
2148 Convert an integer constant to a pointer constant. TYPE must be a
2149 pointer type. CST must be of integer type. The CST value is zero
2150 extended, truncated, or unchanged to make it fit in a pointer size.
2151 This one is *really* dangerous!
2152 ``bitcast (CST to TYPE)``
2153 Convert a constant, CST, to another TYPE. The constraints of the
2154 operands are the same as those for the :ref:`bitcast
2155 instruction <i_bitcast>`.
2156 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2157 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2158 constants. As with the :ref:`getelementptr <i_getelementptr>`
2159 instruction, the index list may have zero or more indexes, which are
2160 required to make sense for the type of "CSTPTR".
2161 ``select (COND, VAL1, VAL2)``
2162 Perform the :ref:`select operation <i_select>` on constants.
2163 ``icmp COND (VAL1, VAL2)``
2164 Performs the :ref:`icmp operation <i_icmp>` on constants.
2165 ``fcmp COND (VAL1, VAL2)``
2166 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2167 ``extractelement (VAL, IDX)``
2168 Perform the :ref:`extractelement operation <i_extractelement>` on
2170 ``insertelement (VAL, ELT, IDX)``
2171 Perform the :ref:`insertelement operation <i_insertelement>` on
2173 ``shufflevector (VEC1, VEC2, IDXMASK)``
2174 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2176 ``extractvalue (VAL, IDX0, IDX1, ...)``
2177 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2178 constants. The index list is interpreted in a similar manner as
2179 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2180 least one index value must be specified.
2181 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2182 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2183 The index list is interpreted in a similar manner as indices in a
2184 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2185 value must be specified.
2186 ``OPCODE (LHS, RHS)``
2187 Perform the specified operation of the LHS and RHS constants. OPCODE
2188 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2189 binary <bitwiseops>` operations. The constraints on operands are
2190 the same as those for the corresponding instruction (e.g. no bitwise
2191 operations on floating point values are allowed).
2196 Inline Assembler Expressions
2197 ----------------------------
2199 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2200 Inline Assembly <moduleasm>`) through the use of a special value. This
2201 value represents the inline assembler as a string (containing the
2202 instructions to emit), a list of operand constraints (stored as a
2203 string), a flag that indicates whether or not the inline asm expression
2204 has side effects, and a flag indicating whether the function containing
2205 the asm needs to align its stack conservatively. An example inline
2206 assembler expression is:
2208 .. code-block:: llvm
2210 i32 (i32) asm "bswap $0", "=r,r"
2212 Inline assembler expressions may **only** be used as the callee operand
2213 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2214 Thus, typically we have:
2216 .. code-block:: llvm
2218 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2220 Inline asms with side effects not visible in the constraint list must be
2221 marked as having side effects. This is done through the use of the
2222 '``sideeffect``' keyword, like so:
2224 .. code-block:: llvm
2226 call void asm sideeffect "eieio", ""()
2228 In some cases inline asms will contain code that will not work unless
2229 the stack is aligned in some way, such as calls or SSE instructions on
2230 x86, yet will not contain code that does that alignment within the asm.
2231 The compiler should make conservative assumptions about what the asm
2232 might contain and should generate its usual stack alignment code in the
2233 prologue if the '``alignstack``' keyword is present:
2235 .. code-block:: llvm
2237 call void asm alignstack "eieio", ""()
2239 Inline asms also support using non-standard assembly dialects. The
2240 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2241 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2242 the only supported dialects. An example is:
2244 .. code-block:: llvm
2246 call void asm inteldialect "eieio", ""()
2248 If multiple keywords appear the '``sideeffect``' keyword must come
2249 first, the '``alignstack``' keyword second and the '``inteldialect``'
2255 The call instructions that wrap inline asm nodes may have a
2256 "``!srcloc``" MDNode attached to it that contains a list of constant
2257 integers. If present, the code generator will use the integer as the
2258 location cookie value when report errors through the ``LLVMContext``
2259 error reporting mechanisms. This allows a front-end to correlate backend
2260 errors that occur with inline asm back to the source code that produced
2263 .. code-block:: llvm
2265 call void asm sideeffect "something bad", ""(), !srcloc !42
2267 !42 = !{ i32 1234567 }
2269 It is up to the front-end to make sense of the magic numbers it places
2270 in the IR. If the MDNode contains multiple constants, the code generator
2271 will use the one that corresponds to the line of the asm that the error
2276 Metadata Nodes and Metadata Strings
2277 -----------------------------------
2279 LLVM IR allows metadata to be attached to instructions in the program
2280 that can convey extra information about the code to the optimizers and
2281 code generator. One example application of metadata is source-level
2282 debug information. There are two metadata primitives: strings and nodes.
2283 All metadata has the ``metadata`` type and is identified in syntax by a
2284 preceding exclamation point ('``!``').
2286 A metadata string is a string surrounded by double quotes. It can
2287 contain any character by escaping non-printable characters with
2288 "``\xx``" where "``xx``" is the two digit hex code. For example:
2291 Metadata nodes are represented with notation similar to structure
2292 constants (a comma separated list of elements, surrounded by braces and
2293 preceded by an exclamation point). Metadata nodes can have any values as
2294 their operand. For example:
2296 .. code-block:: llvm
2298 !{ metadata !"test\00", i32 10}
2300 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2301 metadata nodes, which can be looked up in the module symbol table. For
2304 .. code-block:: llvm
2306 !foo = metadata !{!4, !3}
2308 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2309 function is using two metadata arguments:
2311 .. code-block:: llvm
2313 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2315 Metadata can be attached with an instruction. Here metadata ``!21`` is
2316 attached to the ``add`` instruction using the ``!dbg`` identifier:
2318 .. code-block:: llvm
2320 %indvar.next = add i64 %indvar, 1, !dbg !21
2322 More information about specific metadata nodes recognized by the
2323 optimizers and code generator is found below.
2328 In LLVM IR, memory does not have types, so LLVM's own type system is not
2329 suitable for doing TBAA. Instead, metadata is added to the IR to
2330 describe a type system of a higher level language. This can be used to
2331 implement typical C/C++ TBAA, but it can also be used to implement
2332 custom alias analysis behavior for other languages.
2334 The current metadata format is very simple. TBAA metadata nodes have up
2335 to three fields, e.g.:
2337 .. code-block:: llvm
2339 !0 = metadata !{ metadata !"an example type tree" }
2340 !1 = metadata !{ metadata !"int", metadata !0 }
2341 !2 = metadata !{ metadata !"float", metadata !0 }
2342 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2344 The first field is an identity field. It can be any value, usually a
2345 metadata string, which uniquely identifies the type. The most important
2346 name in the tree is the name of the root node. Two trees with different
2347 root node names are entirely disjoint, even if they have leaves with
2350 The second field identifies the type's parent node in the tree, or is
2351 null or omitted for a root node. A type is considered to alias all of
2352 its descendants and all of its ancestors in the tree. Also, a type is
2353 considered to alias all types in other trees, so that bitcode produced
2354 from multiple front-ends is handled conservatively.
2356 If the third field is present, it's an integer which if equal to 1
2357 indicates that the type is "constant" (meaning
2358 ``pointsToConstantMemory`` should return true; see `other useful
2359 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2361 '``tbaa.struct``' Metadata
2362 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2364 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2365 aggregate assignment operations in C and similar languages, however it
2366 is defined to copy a contiguous region of memory, which is more than
2367 strictly necessary for aggregate types which contain holes due to
2368 padding. Also, it doesn't contain any TBAA information about the fields
2371 ``!tbaa.struct`` metadata can describe which memory subregions in a
2372 memcpy are padding and what the TBAA tags of the struct are.
2374 The current metadata format is very simple. ``!tbaa.struct`` metadata
2375 nodes are a list of operands which are in conceptual groups of three.
2376 For each group of three, the first operand gives the byte offset of a
2377 field in bytes, the second gives its size in bytes, and the third gives
2380 .. code-block:: llvm
2382 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2384 This describes a struct with two fields. The first is at offset 0 bytes
2385 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2386 and has size 4 bytes and has tbaa tag !2.
2388 Note that the fields need not be contiguous. In this example, there is a
2389 4 byte gap between the two fields. This gap represents padding which
2390 does not carry useful data and need not be preserved.
2392 '``fpmath``' Metadata
2393 ^^^^^^^^^^^^^^^^^^^^^
2395 ``fpmath`` metadata may be attached to any instruction of floating point
2396 type. It can be used to express the maximum acceptable error in the
2397 result of that instruction, in ULPs, thus potentially allowing the
2398 compiler to use a more efficient but less accurate method of computing
2399 it. ULP is defined as follows:
2401 If ``x`` is a real number that lies between two finite consecutive
2402 floating-point numbers ``a`` and ``b``, without being equal to one
2403 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2404 distance between the two non-equal finite floating-point numbers
2405 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2407 The metadata node shall consist of a single positive floating point
2408 number representing the maximum relative error, for example:
2410 .. code-block:: llvm
2412 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2414 '``range``' Metadata
2415 ^^^^^^^^^^^^^^^^^^^^
2417 ``range`` metadata may be attached only to loads of integer types. It
2418 expresses the possible ranges the loaded value is in. The ranges are
2419 represented with a flattened list of integers. The loaded value is known
2420 to be in the union of the ranges defined by each consecutive pair. Each
2421 pair has the following properties:
2423 - The type must match the type loaded by the instruction.
2424 - The pair ``a,b`` represents the range ``[a,b)``.
2425 - Both ``a`` and ``b`` are constants.
2426 - The range is allowed to wrap.
2427 - The range should not represent the full or empty set. That is,
2430 In addition, the pairs must be in signed order of the lower bound and
2431 they must be non-contiguous.
2435 .. code-block:: llvm
2437 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2438 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2439 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2440 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2442 !0 = metadata !{ i8 0, i8 2 }
2443 !1 = metadata !{ i8 255, i8 2 }
2444 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2445 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2447 Module Flags Metadata
2448 =====================
2450 Information about the module as a whole is difficult to convey to LLVM's
2451 subsystems. The LLVM IR isn't sufficient to transmit this information.
2452 The ``llvm.module.flags`` named metadata exists in order to facilitate
2453 this. These flags are in the form of key / value pairs — much like a
2454 dictionary — making it easy for any subsystem who cares about a flag to
2457 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2458 Each triplet has the following form:
2460 - The first element is a *behavior* flag, which specifies the behavior
2461 when two (or more) modules are merged together, and it encounters two
2462 (or more) metadata with the same ID. The supported behaviors are
2464 - The second element is a metadata string that is a unique ID for the
2465 metadata. How each ID is interpreted is documented below.
2466 - The third element is the value of the flag.
2468 When two (or more) modules are merged together, the resulting
2469 ``llvm.module.flags`` metadata is the union of the modules'
2470 ``llvm.module.flags`` metadata. The only exception being a flag with the
2471 *Override* behavior, which may override another flag's value (see
2474 The following behaviors are supported:
2485 Emits an error if two values disagree. It is an error to have an
2486 ID with both an Error and a Warning behavior.
2490 Emits a warning if two values disagree.
2494 Emits an error when the specified value is not present or doesn't
2495 have the specified value. It is an error for two (or more)
2496 ``llvm.module.flags`` with the same ID to have the Require behavior
2497 but different values. There may be multiple Require flags per ID.
2501 Uses the specified value if the two values disagree. It is an
2502 error for two (or more) ``llvm.module.flags`` with the same ID
2503 to have the Override behavior but different values.
2505 An example of module flags:
2507 .. code-block:: llvm
2509 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2510 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2511 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2512 !3 = metadata !{ i32 3, metadata !"qux",
2514 metadata !"foo", i32 1
2517 !llvm.module.flags = !{ !0, !1, !2, !3 }
2519 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2520 if two or more ``!"foo"`` flags are seen is to emit an error if their
2521 values are not equal.
2523 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2524 behavior if two or more ``!"bar"`` flags are seen is to use the value
2525 '37' if their values are not equal.
2527 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2528 behavior if two or more ``!"qux"`` flags are seen is to emit a
2529 warning if their values are not equal.
2531 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2535 metadata !{ metadata !"foo", i32 1 }
2537 The behavior is to emit an error if the ``llvm.module.flags`` does
2538 not contain a flag with the ID ``!"foo"`` that has the value '1'. If
2539 two or more ``!"qux"`` flags exist, then they must have the same
2540 value or an error will be issued.
2542 Objective-C Garbage Collection Module Flags Metadata
2543 ----------------------------------------------------
2545 On the Mach-O platform, Objective-C stores metadata about garbage
2546 collection in a special section called "image info". The metadata
2547 consists of a version number and a bitmask specifying what types of
2548 garbage collection are supported (if any) by the file. If two or more
2549 modules are linked together their garbage collection metadata needs to
2550 be merged rather than appended together.
2552 The Objective-C garbage collection module flags metadata consists of the
2553 following key-value pairs:
2562 * - ``Objective-C Version``
2563 - **[Required]** — The Objective-C ABI version. Valid values are 1 and 2.
2565 * - ``Objective-C Image Info Version``
2566 - **[Required]** — The version of the image info section. Currently
2569 * - ``Objective-C Image Info Section``
2570 - **[Required]** — The section to place the metadata. Valid values are
2571 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2572 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2573 Objective-C ABI version 2.
2575 * - ``Objective-C Garbage Collection``
2576 - **[Required]** — Specifies whether garbage collection is supported or
2577 not. Valid values are 0, for no garbage collection, and 2, for garbage
2578 collection supported.
2580 * - ``Objective-C GC Only``
2581 - **[Optional]** — Specifies that only garbage collection is supported.
2582 If present, its value must be 6. This flag requires that the
2583 ``Objective-C Garbage Collection`` flag have the value 2.
2585 Some important flag interactions:
2587 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2588 merged with a module with ``Objective-C Garbage Collection`` set to
2589 2, then the resulting module has the
2590 ``Objective-C Garbage Collection`` flag set to 0.
2591 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2592 merged with a module with ``Objective-C GC Only`` set to 6.
2594 Intrinsic Global Variables
2595 ==========================
2597 LLVM has a number of "magic" global variables that contain data that
2598 affect code generation or other IR semantics. These are documented here.
2599 All globals of this sort should have a section specified as
2600 "``llvm.metadata``". This section and all globals that start with
2601 "``llvm.``" are reserved for use by LLVM.
2603 The '``llvm.used``' Global Variable
2604 -----------------------------------
2606 The ``@llvm.used`` global is an array with i8\* element type which has
2607 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2608 pointers to global variables and functions which may optionally have a
2609 pointer cast formed of bitcast or getelementptr. For example, a legal
2612 .. code-block:: llvm
2617 @llvm.used = appending global [2 x i8*] [
2619 i8* bitcast (i32* @Y to i8*)
2620 ], section "llvm.metadata"
2622 If a global variable appears in the ``@llvm.used`` list, then the
2623 compiler, assembler, and linker are required to treat the symbol as if
2624 there is a reference to the global that it cannot see. For example, if a
2625 variable has internal linkage and no references other than that from the
2626 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2627 represent references from inline asms and other things the compiler
2628 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2630 On some targets, the code generator must emit a directive to the
2631 assembler or object file to prevent the assembler and linker from
2632 molesting the symbol.
2634 The '``llvm.compiler.used``' Global Variable
2635 --------------------------------------------
2637 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2638 directive, except that it only prevents the compiler from touching the
2639 symbol. On targets that support it, this allows an intelligent linker to
2640 optimize references to the symbol without being impeded as it would be
2643 This is a rare construct that should only be used in rare circumstances,
2644 and should not be exposed to source languages.
2646 The '``llvm.global_ctors``' Global Variable
2647 -------------------------------------------
2649 .. code-block:: llvm
2651 %0 = type { i32, void ()* }
2652 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2654 The ``@llvm.global_ctors`` array contains a list of constructor
2655 functions and associated priorities. The functions referenced by this
2656 array will be called in ascending order of priority (i.e. lowest first)
2657 when the module is loaded. The order of functions with the same priority
2660 The '``llvm.global_dtors``' Global Variable
2661 -------------------------------------------
2663 .. code-block:: llvm
2665 %0 = type { i32, void ()* }
2666 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2668 The ``@llvm.global_dtors`` array contains a list of destructor functions
2669 and associated priorities. The functions referenced by this array will
2670 be called in descending order of priority (i.e. highest first) when the
2671 module is loaded. The order of functions with the same priority is not
2674 Instruction Reference
2675 =====================
2677 The LLVM instruction set consists of several different classifications
2678 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2679 instructions <binaryops>`, :ref:`bitwise binary
2680 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2681 :ref:`other instructions <otherops>`.
2685 Terminator Instructions
2686 -----------------------
2688 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2689 program ends with a "Terminator" instruction, which indicates which
2690 block should be executed after the current block is finished. These
2691 terminator instructions typically yield a '``void``' value: they produce
2692 control flow, not values (the one exception being the
2693 ':ref:`invoke <i_invoke>`' instruction).
2695 The terminator instructions are: ':ref:`ret <i_ret>`',
2696 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2697 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2698 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2702 '``ret``' Instruction
2703 ^^^^^^^^^^^^^^^^^^^^^
2710 ret <type> <value> ; Return a value from a non-void function
2711 ret void ; Return from void function
2716 The '``ret``' instruction is used to return control flow (and optionally
2717 a value) from a function back to the caller.
2719 There are two forms of the '``ret``' instruction: one that returns a
2720 value and then causes control flow, and one that just causes control
2726 The '``ret``' instruction optionally accepts a single argument, the
2727 return value. The type of the return value must be a ':ref:`first
2728 class <t_firstclass>`' type.
2730 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2731 return type and contains a '``ret``' instruction with no return value or
2732 a return value with a type that does not match its type, or if it has a
2733 void return type and contains a '``ret``' instruction with a return
2739 When the '``ret``' instruction is executed, control flow returns back to
2740 the calling function's context. If the caller is a
2741 ":ref:`call <i_call>`" instruction, execution continues at the
2742 instruction after the call. If the caller was an
2743 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2744 beginning of the "normal" destination block. If the instruction returns
2745 a value, that value shall set the call or invoke instruction's return
2751 .. code-block:: llvm
2753 ret i32 5 ; Return an integer value of 5
2754 ret void ; Return from a void function
2755 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2759 '``br``' Instruction
2760 ^^^^^^^^^^^^^^^^^^^^
2767 br i1 <cond>, label <iftrue>, label <iffalse>
2768 br label <dest> ; Unconditional branch
2773 The '``br``' instruction is used to cause control flow to transfer to a
2774 different basic block in the current function. There are two forms of
2775 this instruction, corresponding to a conditional branch and an
2776 unconditional branch.
2781 The conditional branch form of the '``br``' instruction takes a single
2782 '``i1``' value and two '``label``' values. The unconditional form of the
2783 '``br``' instruction takes a single '``label``' value as a target.
2788 Upon execution of a conditional '``br``' instruction, the '``i1``'
2789 argument is evaluated. If the value is ``true``, control flows to the
2790 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
2791 to the '``iffalse``' ``label`` argument.
2796 .. code-block:: llvm
2799 %cond = icmp eq i32 %a, %b
2800 br i1 %cond, label %IfEqual, label %IfUnequal
2808 '``switch``' Instruction
2809 ^^^^^^^^^^^^^^^^^^^^^^^^
2816 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2821 The '``switch``' instruction is used to transfer control flow to one of
2822 several different places. It is a generalization of the '``br``'
2823 instruction, allowing a branch to occur to one of many possible
2829 The '``switch``' instruction uses three parameters: an integer
2830 comparison value '``value``', a default '``label``' destination, and an
2831 array of pairs of comparison value constants and '``label``'s. The table
2832 is not allowed to contain duplicate constant entries.
2837 The ``switch`` instruction specifies a table of values and destinations.
2838 When the '``switch``' instruction is executed, this table is searched
2839 for the given value. If the value is found, control flow is transferred
2840 to the corresponding destination; otherwise, control flow is transferred
2841 to the default destination.
2846 Depending on properties of the target machine and the particular
2847 ``switch`` instruction, this instruction may be code generated in
2848 different ways. For example, it could be generated as a series of
2849 chained conditional branches or with a lookup table.
2854 .. code-block:: llvm
2856 ; Emulate a conditional br instruction
2857 %Val = zext i1 %value to i32
2858 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2860 ; Emulate an unconditional br instruction
2861 switch i32 0, label %dest [ ]
2863 ; Implement a jump table:
2864 switch i32 %val, label %otherwise [ i32 0, label %onzero
2866 i32 2, label %ontwo ]
2870 '``indirectbr``' Instruction
2871 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2878 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2883 The '``indirectbr``' instruction implements an indirect branch to a
2884 label within the current function, whose address is specified by
2885 "``address``". Address must be derived from a
2886 :ref:`blockaddress <blockaddress>` constant.
2891 The '``address``' argument is the address of the label to jump to. The
2892 rest of the arguments indicate the full set of possible destinations
2893 that the address may point to. Blocks are allowed to occur multiple
2894 times in the destination list, though this isn't particularly useful.
2896 This destination list is required so that dataflow analysis has an
2897 accurate understanding of the CFG.
2902 Control transfers to the block specified in the address argument. All
2903 possible destination blocks must be listed in the label list, otherwise
2904 this instruction has undefined behavior. This implies that jumps to
2905 labels defined in other functions have undefined behavior as well.
2910 This is typically implemented with a jump through a register.
2915 .. code-block:: llvm
2917 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2921 '``invoke``' Instruction
2922 ^^^^^^^^^^^^^^^^^^^^^^^^
2929 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
2930 to label <normal label> unwind label <exception label>
2935 The '``invoke``' instruction causes control to transfer to a specified
2936 function, with the possibility of control flow transfer to either the
2937 '``normal``' label or the '``exception``' label. If the callee function
2938 returns with the "``ret``" instruction, control flow will return to the
2939 "normal" label. If the callee (or any indirect callees) returns via the
2940 ":ref:`resume <i_resume>`" instruction or other exception handling
2941 mechanism, control is interrupted and continued at the dynamically
2942 nearest "exception" label.
2944 The '``exception``' label is a `landing
2945 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
2946 '``exception``' label is required to have the
2947 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
2948 information about the behavior of the program after unwinding happens,
2949 as its first non-PHI instruction. The restrictions on the
2950 "``landingpad``" instruction's tightly couples it to the "``invoke``"
2951 instruction, so that the important information contained within the
2952 "``landingpad``" instruction can't be lost through normal code motion.
2957 This instruction requires several arguments:
2959 #. The optional "cconv" marker indicates which :ref:`calling
2960 convention <callingconv>` the call should use. If none is
2961 specified, the call defaults to using C calling conventions.
2962 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
2963 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
2965 #. '``ptr to function ty``': shall be the signature of the pointer to
2966 function value being invoked. In most cases, this is a direct
2967 function invocation, but indirect ``invoke``'s are just as possible,
2968 branching off an arbitrary pointer to function value.
2969 #. '``function ptr val``': An LLVM value containing a pointer to a
2970 function to be invoked.
2971 #. '``function args``': argument list whose types match the function
2972 signature argument types and parameter attributes. All arguments must
2973 be of :ref:`first class <t_firstclass>` type. If the function signature
2974 indicates the function accepts a variable number of arguments, the
2975 extra arguments can be specified.
2976 #. '``normal label``': the label reached when the called function
2977 executes a '``ret``' instruction.
2978 #. '``exception label``': the label reached when a callee returns via
2979 the :ref:`resume <i_resume>` instruction or other exception handling
2981 #. The optional :ref:`function attributes <fnattrs>` list. Only
2982 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
2983 attributes are valid here.
2988 This instruction is designed to operate as a standard '``call``'
2989 instruction in most regards. The primary difference is that it
2990 establishes an association with a label, which is used by the runtime
2991 library to unwind the stack.
2993 This instruction is used in languages with destructors to ensure that
2994 proper cleanup is performed in the case of either a ``longjmp`` or a
2995 thrown exception. Additionally, this is important for implementation of
2996 '``catch``' clauses in high-level languages that support them.
2998 For the purposes of the SSA form, the definition of the value returned
2999 by the '``invoke``' instruction is deemed to occur on the edge from the
3000 current block to the "normal" label. If the callee unwinds then no
3001 return value is available.
3006 .. code-block:: llvm
3008 %retval = invoke i32 @Test(i32 15) to label %Continue
3009 unwind label %TestCleanup ; {i32}:retval set
3010 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3011 unwind label %TestCleanup ; {i32}:retval set
3015 '``resume``' Instruction
3016 ^^^^^^^^^^^^^^^^^^^^^^^^
3023 resume <type> <value>
3028 The '``resume``' instruction is a terminator instruction that has no
3034 The '``resume``' instruction requires one argument, which must have the
3035 same type as the result of any '``landingpad``' instruction in the same
3041 The '``resume``' instruction resumes propagation of an existing
3042 (in-flight) exception whose unwinding was interrupted with a
3043 :ref:`landingpad <i_landingpad>` instruction.
3048 .. code-block:: llvm
3050 resume { i8*, i32 } %exn
3054 '``unreachable``' Instruction
3055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3067 The '``unreachable``' instruction has no defined semantics. This
3068 instruction is used to inform the optimizer that a particular portion of
3069 the code is not reachable. This can be used to indicate that the code
3070 after a no-return function cannot be reached, and other facts.
3075 The '``unreachable``' instruction has no defined semantics.
3082 Binary operators are used to do most of the computation in a program.
3083 They require two operands of the same type, execute an operation on
3084 them, and produce a single value. The operands might represent multiple
3085 data, as is the case with the :ref:`vector <t_vector>` data type. The
3086 result value has the same type as its operands.
3088 There are several different binary operators:
3092 '``add``' Instruction
3093 ^^^^^^^^^^^^^^^^^^^^^
3100 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3101 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3102 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3103 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3108 The '``add``' instruction returns the sum of its two operands.
3113 The two arguments to the '``add``' instruction must be
3114 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3115 arguments must have identical types.
3120 The value produced is the integer sum of the two operands.
3122 If the sum has unsigned overflow, the result returned is the
3123 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3126 Because LLVM integers use a two's complement representation, this
3127 instruction is appropriate for both signed and unsigned integers.
3129 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3130 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3131 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3132 unsigned and/or signed overflow, respectively, occurs.
3137 .. code-block:: llvm
3139 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3143 '``fadd``' Instruction
3144 ^^^^^^^^^^^^^^^^^^^^^^
3151 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3156 The '``fadd``' instruction returns the sum of its two operands.
3161 The two arguments to the '``fadd``' instruction must be :ref:`floating
3162 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3163 Both arguments must have identical types.
3168 The value produced is the floating point sum of the two operands. This
3169 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3170 which are optimization hints to enable otherwise unsafe floating point
3176 .. code-block:: llvm
3178 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3180 '``sub``' Instruction
3181 ^^^^^^^^^^^^^^^^^^^^^
3188 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3189 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3190 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3191 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3196 The '``sub``' instruction returns the difference of its two operands.
3198 Note that the '``sub``' instruction is used to represent the '``neg``'
3199 instruction present in most other intermediate representations.
3204 The two arguments to the '``sub``' instruction must be
3205 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3206 arguments must have identical types.
3211 The value produced is the integer difference of the two operands.
3213 If the difference has unsigned overflow, the result returned is the
3214 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3217 Because LLVM integers use a two's complement representation, this
3218 instruction is appropriate for both signed and unsigned integers.
3220 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3221 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3222 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3223 unsigned and/or signed overflow, respectively, occurs.
3228 .. code-block:: llvm
3230 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3231 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3235 '``fsub``' Instruction
3236 ^^^^^^^^^^^^^^^^^^^^^^
3243 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3248 The '``fsub``' instruction returns the difference of its two operands.
3250 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3251 instruction present in most other intermediate representations.
3256 The two arguments to the '``fsub``' instruction must be :ref:`floating
3257 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3258 Both arguments must have identical types.
3263 The value produced is the floating point difference of the two operands.
3264 This instruction can also take any number of :ref:`fast-math
3265 flags <fastmath>`, which are optimization hints to enable otherwise
3266 unsafe floating point optimizations:
3271 .. code-block:: llvm
3273 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3274 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3276 '``mul``' Instruction
3277 ^^^^^^^^^^^^^^^^^^^^^
3284 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3285 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3286 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3287 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3292 The '``mul``' instruction returns the product of its two operands.
3297 The two arguments to the '``mul``' instruction must be
3298 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3299 arguments must have identical types.
3304 The value produced is the integer product of the two operands.
3306 If the result of the multiplication has unsigned overflow, the result
3307 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3308 bit width of the result.
3310 Because LLVM integers use a two's complement representation, and the
3311 result is the same width as the operands, this instruction returns the
3312 correct result for both signed and unsigned integers. If a full product
3313 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3314 sign-extended or zero-extended as appropriate to the width of the full
3317 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3318 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3319 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3320 unsigned and/or signed overflow, respectively, occurs.
3325 .. code-block:: llvm
3327 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3331 '``fmul``' Instruction
3332 ^^^^^^^^^^^^^^^^^^^^^^
3339 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3344 The '``fmul``' instruction returns the product of its two operands.
3349 The two arguments to the '``fmul``' instruction must be :ref:`floating
3350 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3351 Both arguments must have identical types.
3356 The value produced is the floating point product of the two operands.
3357 This instruction can also take any number of :ref:`fast-math
3358 flags <fastmath>`, which are optimization hints to enable otherwise
3359 unsafe floating point optimizations:
3364 .. code-block:: llvm
3366 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3368 '``udiv``' Instruction
3369 ^^^^^^^^^^^^^^^^^^^^^^
3376 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3377 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3382 The '``udiv``' instruction returns the quotient of its two operands.
3387 The two arguments to the '``udiv``' instruction must be
3388 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3389 arguments must have identical types.
3394 The value produced is the unsigned integer quotient of the two operands.
3396 Note that unsigned integer division and signed integer division are
3397 distinct operations; for signed integer division, use '``sdiv``'.
3399 Division by zero leads to undefined behavior.
3401 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3402 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3403 such, "((a udiv exact b) mul b) == a").
3408 .. code-block:: llvm
3410 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3412 '``sdiv``' Instruction
3413 ^^^^^^^^^^^^^^^^^^^^^^
3420 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3421 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3426 The '``sdiv``' instruction returns the quotient of its two operands.
3431 The two arguments to the '``sdiv``' instruction must be
3432 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3433 arguments must have identical types.
3438 The value produced is the signed integer quotient of the two operands
3439 rounded towards zero.
3441 Note that signed integer division and unsigned integer division are
3442 distinct operations; for unsigned integer division, use '``udiv``'.
3444 Division by zero leads to undefined behavior. Overflow also leads to
3445 undefined behavior; this is a rare case, but can occur, for example, by
3446 doing a 32-bit division of -2147483648 by -1.
3448 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3449 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3454 .. code-block:: llvm
3456 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3460 '``fdiv``' Instruction
3461 ^^^^^^^^^^^^^^^^^^^^^^
3468 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3473 The '``fdiv``' instruction returns the quotient of its two operands.
3478 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3479 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3480 Both arguments must have identical types.
3485 The value produced is the floating point quotient of the two operands.
3486 This instruction can also take any number of :ref:`fast-math
3487 flags <fastmath>`, which are optimization hints to enable otherwise
3488 unsafe floating point optimizations:
3493 .. code-block:: llvm
3495 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3497 '``urem``' Instruction
3498 ^^^^^^^^^^^^^^^^^^^^^^
3505 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3510 The '``urem``' instruction returns the remainder from the unsigned
3511 division of its two arguments.
3516 The two arguments to the '``urem``' instruction must be
3517 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3518 arguments must have identical types.
3523 This instruction returns the unsigned integer *remainder* of a division.
3524 This instruction always performs an unsigned division to get the
3527 Note that unsigned integer remainder and signed integer remainder are
3528 distinct operations; for signed integer remainder, use '``srem``'.
3530 Taking the remainder of a division by zero leads to undefined behavior.
3535 .. code-block:: llvm
3537 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3539 '``srem``' Instruction
3540 ^^^^^^^^^^^^^^^^^^^^^^
3547 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3552 The '``srem``' instruction returns the remainder from the signed
3553 division of its two operands. This instruction can also take
3554 :ref:`vector <t_vector>` versions of the values in which case the elements
3560 The two arguments to the '``srem``' instruction must be
3561 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3562 arguments must have identical types.
3567 This instruction returns the *remainder* of a division (where the result
3568 is either zero or has the same sign as the dividend, ``op1``), not the
3569 *modulo* operator (where the result is either zero or has the same sign
3570 as the divisor, ``op2``) of a value. For more information about the
3571 difference, see `The Math
3572 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3573 table of how this is implemented in various languages, please see
3575 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3577 Note that signed integer remainder and unsigned integer remainder are
3578 distinct operations; for unsigned integer remainder, use '``urem``'.
3580 Taking the remainder of a division by zero leads to undefined behavior.
3581 Overflow also leads to undefined behavior; this is a rare case, but can
3582 occur, for example, by taking the remainder of a 32-bit division of
3583 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3584 rule lets srem be implemented using instructions that return both the
3585 result of the division and the remainder.)
3590 .. code-block:: llvm
3592 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3596 '``frem``' Instruction
3597 ^^^^^^^^^^^^^^^^^^^^^^
3604 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3609 The '``frem``' instruction returns the remainder from the division of
3615 The two arguments to the '``frem``' instruction must be :ref:`floating
3616 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3617 Both arguments must have identical types.
3622 This instruction returns the *remainder* of a division. The remainder
3623 has the same sign as the dividend. This instruction can also take any
3624 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3625 to enable otherwise unsafe floating point optimizations:
3630 .. code-block:: llvm
3632 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3636 Bitwise Binary Operations
3637 -------------------------
3639 Bitwise binary operators are used to do various forms of bit-twiddling
3640 in a program. They are generally very efficient instructions and can
3641 commonly be strength reduced from other instructions. They require two
3642 operands of the same type, execute an operation on them, and produce a
3643 single value. The resulting value is the same type as its operands.
3645 '``shl``' Instruction
3646 ^^^^^^^^^^^^^^^^^^^^^
3653 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3654 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3655 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3656 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3661 The '``shl``' instruction returns the first operand shifted to the left
3662 a specified number of bits.
3667 Both arguments to the '``shl``' instruction must be the same
3668 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3669 '``op2``' is treated as an unsigned value.
3674 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3675 where ``n`` is the width of the result. If ``op2`` is (statically or
3676 dynamically) negative or equal to or larger than the number of bits in
3677 ``op1``, the result is undefined. If the arguments are vectors, each
3678 vector element of ``op1`` is shifted by the corresponding shift amount
3681 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3682 value <poisonvalues>` if it shifts out any non-zero bits. If the
3683 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3684 value <poisonvalues>` if it shifts out any bits that disagree with the
3685 resultant sign bit. As such, NUW/NSW have the same semantics as they
3686 would if the shift were expressed as a mul instruction with the same
3687 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3692 .. code-block:: llvm
3694 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3695 <result> = shl i32 4, 2 ; yields {i32}: 16
3696 <result> = shl i32 1, 10 ; yields {i32}: 1024
3697 <result> = shl i32 1, 32 ; undefined
3698 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3700 '``lshr``' Instruction
3701 ^^^^^^^^^^^^^^^^^^^^^^
3708 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3709 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3714 The '``lshr``' instruction (logical shift right) returns the first
3715 operand shifted to the right a specified number of bits with zero fill.
3720 Both arguments to the '``lshr``' instruction must be the same
3721 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3722 '``op2``' is treated as an unsigned value.
3727 This instruction always performs a logical shift right operation. The
3728 most significant bits of the result will be filled with zero bits after
3729 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3730 than the number of bits in ``op1``, the result is undefined. If the
3731 arguments are vectors, each vector element of ``op1`` is shifted by the
3732 corresponding shift amount in ``op2``.
3734 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3735 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3741 .. code-block:: llvm
3743 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3744 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3745 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3746 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3747 <result> = lshr i32 1, 32 ; undefined
3748 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3750 '``ashr``' Instruction
3751 ^^^^^^^^^^^^^^^^^^^^^^
3758 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
3759 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
3764 The '``ashr``' instruction (arithmetic shift right) returns the first
3765 operand shifted to the right a specified number of bits with sign
3771 Both arguments to the '``ashr``' instruction must be the same
3772 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3773 '``op2``' is treated as an unsigned value.
3778 This instruction always performs an arithmetic shift right operation,
3779 The most significant bits of the result will be filled with the sign bit
3780 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
3781 than the number of bits in ``op1``, the result is undefined. If the
3782 arguments are vectors, each vector element of ``op1`` is shifted by the
3783 corresponding shift amount in ``op2``.
3785 If the ``exact`` keyword is present, the result value of the ``ashr`` is
3786 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3792 .. code-block:: llvm
3794 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
3795 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
3796 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
3797 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
3798 <result> = ashr i32 1, 32 ; undefined
3799 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
3801 '``and``' Instruction
3802 ^^^^^^^^^^^^^^^^^^^^^
3809 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
3814 The '``and``' instruction returns the bitwise logical and of its two
3820 The two arguments to the '``and``' instruction must be
3821 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3822 arguments must have identical types.
3827 The truth table used for the '``and``' instruction is:
3844 .. code-block:: llvm
3846 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
3847 <result> = and i32 15, 40 ; yields {i32}:result = 8
3848 <result> = and i32 4, 8 ; yields {i32}:result = 0
3850 '``or``' Instruction
3851 ^^^^^^^^^^^^^^^^^^^^
3858 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
3863 The '``or``' instruction returns the bitwise logical inclusive or of its
3869 The two arguments to the '``or``' instruction must be
3870 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3871 arguments must have identical types.
3876 The truth table used for the '``or``' instruction is:
3895 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
3896 <result> = or i32 15, 40 ; yields {i32}:result = 47
3897 <result> = or i32 4, 8 ; yields {i32}:result = 12
3899 '``xor``' Instruction
3900 ^^^^^^^^^^^^^^^^^^^^^
3907 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
3912 The '``xor``' instruction returns the bitwise logical exclusive or of
3913 its two operands. The ``xor`` is used to implement the "one's
3914 complement" operation, which is the "~" operator in C.
3919 The two arguments to the '``xor``' instruction must be
3920 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3921 arguments must have identical types.
3926 The truth table used for the '``xor``' instruction is:
3943 .. code-block:: llvm
3945 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
3946 <result> = xor i32 15, 40 ; yields {i32}:result = 39
3947 <result> = xor i32 4, 8 ; yields {i32}:result = 12
3948 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
3953 LLVM supports several instructions to represent vector operations in a
3954 target-independent manner. These instructions cover the element-access
3955 and vector-specific operations needed to process vectors effectively.
3956 While LLVM does directly support these vector operations, many
3957 sophisticated algorithms will want to use target-specific intrinsics to
3958 take full advantage of a specific target.
3960 .. _i_extractelement:
3962 '``extractelement``' Instruction
3963 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3970 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
3975 The '``extractelement``' instruction extracts a single scalar element
3976 from a vector at a specified index.
3981 The first operand of an '``extractelement``' instruction is a value of
3982 :ref:`vector <t_vector>` type. The second operand is an index indicating
3983 the position from which to extract the element. The index may be a
3989 The result is a scalar of the same type as the element type of ``val``.
3990 Its value is the value at position ``idx`` of ``val``. If ``idx``
3991 exceeds the length of ``val``, the results are undefined.
3996 .. code-block:: llvm
3998 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4000 .. _i_insertelement:
4002 '``insertelement``' Instruction
4003 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4010 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4015 The '``insertelement``' instruction inserts a scalar element into a
4016 vector at a specified index.
4021 The first operand of an '``insertelement``' instruction is a value of
4022 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4023 type must equal the element type of the first operand. The third operand
4024 is an index indicating the position at which to insert the value. The
4025 index may be a variable.
4030 The result is a vector of the same type as ``val``. Its element values
4031 are those of ``val`` except at position ``idx``, where it gets the value
4032 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4038 .. code-block:: llvm
4040 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4042 .. _i_shufflevector:
4044 '``shufflevector``' Instruction
4045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4052 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4057 The '``shufflevector``' instruction constructs a permutation of elements
4058 from two input vectors, returning a vector with the same element type as
4059 the input and length that is the same as the shuffle mask.
4064 The first two operands of a '``shufflevector``' instruction are vectors
4065 with the same type. The third argument is a shuffle mask whose element
4066 type is always 'i32'. The result of the instruction is a vector whose
4067 length is the same as the shuffle mask and whose element type is the
4068 same as the element type of the first two operands.
4070 The shuffle mask operand is required to be a constant vector with either
4071 constant integer or undef values.
4076 The elements of the two input vectors are numbered from left to right
4077 across both of the vectors. The shuffle mask operand specifies, for each
4078 element of the result vector, which element of the two input vectors the
4079 result element gets. The element selector may be undef (meaning "don't
4080 care") and the second operand may be undef if performing a shuffle from
4086 .. code-block:: llvm
4088 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4089 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4090 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4091 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4092 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4093 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4094 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4095 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4097 Aggregate Operations
4098 --------------------
4100 LLVM supports several instructions for working with
4101 :ref:`aggregate <t_aggregate>` values.
4105 '``extractvalue``' Instruction
4106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4113 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4118 The '``extractvalue``' instruction extracts the value of a member field
4119 from an :ref:`aggregate <t_aggregate>` value.
4124 The first operand of an '``extractvalue``' instruction is a value of
4125 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4126 constant indices to specify which value to extract in a similar manner
4127 as indices in a '``getelementptr``' instruction.
4129 The major differences to ``getelementptr`` indexing are:
4131 - Since the value being indexed is not a pointer, the first index is
4132 omitted and assumed to be zero.
4133 - At least one index must be specified.
4134 - Not only struct indices but also array indices must be in bounds.
4139 The result is the value at the position in the aggregate specified by
4145 .. code-block:: llvm
4147 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4151 '``insertvalue``' Instruction
4152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4159 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4164 The '``insertvalue``' instruction inserts a value into a member field in
4165 an :ref:`aggregate <t_aggregate>` value.
4170 The first operand of an '``insertvalue``' instruction is a value of
4171 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4172 a first-class value to insert. The following operands are constant
4173 indices indicating the position at which to insert the value in a
4174 similar manner as indices in a '``extractvalue``' instruction. The value
4175 to insert must have the same type as the value identified by the
4181 The result is an aggregate of the same type as ``val``. Its value is
4182 that of ``val`` except that the value at the position specified by the
4183 indices is that of ``elt``.
4188 .. code-block:: llvm
4190 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4191 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4192 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4196 Memory Access and Addressing Operations
4197 ---------------------------------------
4199 A key design point of an SSA-based representation is how it represents
4200 memory. In LLVM, no memory locations are in SSA form, which makes things
4201 very simple. This section describes how to read, write, and allocate
4206 '``alloca``' Instruction
4207 ^^^^^^^^^^^^^^^^^^^^^^^^
4214 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4219 The '``alloca``' instruction allocates memory on the stack frame of the
4220 currently executing function, to be automatically released when this
4221 function returns to its caller. The object is always allocated in the
4222 generic address space (address space zero).
4227 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4228 bytes of memory on the runtime stack, returning a pointer of the
4229 appropriate type to the program. If "NumElements" is specified, it is
4230 the number of elements allocated, otherwise "NumElements" is defaulted
4231 to be one. If a constant alignment is specified, the value result of the
4232 allocation is guaranteed to be aligned to at least that boundary. If not
4233 specified, or if zero, the target can choose to align the allocation on
4234 any convenient boundary compatible with the type.
4236 '``type``' may be any sized type.
4241 Memory is allocated; a pointer is returned. The operation is undefined
4242 if there is insufficient stack space for the allocation. '``alloca``'d
4243 memory is automatically released when the function returns. The
4244 '``alloca``' instruction is commonly used to represent automatic
4245 variables that must have an address available. When the function returns
4246 (either with the ``ret`` or ``resume`` instructions), the memory is
4247 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4248 The order in which memory is allocated (ie., which way the stack grows)
4254 .. code-block:: llvm
4256 %ptr = alloca i32 ; yields {i32*}:ptr
4257 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4258 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4259 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4263 '``load``' Instruction
4264 ^^^^^^^^^^^^^^^^^^^^^^
4271 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4272 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4273 !<index> = !{ i32 1 }
4278 The '``load``' instruction is used to read from memory.
4283 The argument to the '``load``' instruction specifies the memory address
4284 from which to load. The pointer must point to a :ref:`first
4285 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4286 then the optimizer is not allowed to modify the number or order of
4287 execution of this ``load`` with other :ref:`volatile
4288 operations <volatile>`.
4290 If the ``load`` is marked as ``atomic``, it takes an extra
4291 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4292 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4293 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4294 when they may see multiple atomic stores. The type of the pointee must
4295 be an integer type whose bit width is a power of two greater than or
4296 equal to eight and less than or equal to a target-specific size limit.
4297 ``align`` must be explicitly specified on atomic loads, and the load has
4298 undefined behavior if the alignment is not set to a value which is at
4299 least the size in bytes of the pointee. ``!nontemporal`` does not have
4300 any defined semantics for atomic loads.
4302 The optional constant ``align`` argument specifies the alignment of the
4303 operation (that is, the alignment of the memory address). A value of 0
4304 or an omitted ``align`` argument means that the operation has the abi
4305 alignment for the target. It is the responsibility of the code emitter
4306 to ensure that the alignment information is correct. Overestimating the
4307 alignment results in undefined behavior. Underestimating the alignment
4308 may produce less efficient code. An alignment of 1 is always safe.
4310 The optional ``!nontemporal`` metadata must reference a single
4311 metatadata name <index> corresponding to a metadata node with one
4312 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4313 metatadata on the instruction tells the optimizer and code generator
4314 that this load is not expected to be reused in the cache. The code
4315 generator may select special instructions to save cache bandwidth, such
4316 as the ``MOVNT`` instruction on x86.
4318 The optional ``!invariant.load`` metadata must reference a single
4319 metatadata name <index> corresponding to a metadata node with no
4320 entries. The existence of the ``!invariant.load`` metatadata on the
4321 instruction tells the optimizer and code generator that this load
4322 address points to memory which does not change value during program
4323 execution. The optimizer may then move this load around, for example, by
4324 hoisting it out of loops using loop invariant code motion.
4329 The location of memory pointed to is loaded. If the value being loaded
4330 is of scalar type then the number of bytes read does not exceed the
4331 minimum number of bytes needed to hold all bits of the type. For
4332 example, loading an ``i24`` reads at most three bytes. When loading a
4333 value of a type like ``i20`` with a size that is not an integral number
4334 of bytes, the result is undefined if the value was not originally
4335 written using a store of the same type.
4340 .. code-block:: llvm
4342 %ptr = alloca i32 ; yields {i32*}:ptr
4343 store i32 3, i32* %ptr ; yields {void}
4344 %val = load i32* %ptr ; yields {i32}:val = i32 3
4348 '``store``' Instruction
4349 ^^^^^^^^^^^^^^^^^^^^^^^
4356 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4357 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4362 The '``store``' instruction is used to write to memory.
4367 There are two arguments to the '``store``' instruction: a value to store
4368 and an address at which to store it. The type of the '``<pointer>``'
4369 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4370 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4371 then the optimizer is not allowed to modify the number or order of
4372 execution of this ``store`` with other :ref:`volatile
4373 operations <volatile>`.
4375 If the ``store`` is marked as ``atomic``, it takes an extra
4376 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4377 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4378 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4379 when they may see multiple atomic stores. The type of the pointee must
4380 be an integer type whose bit width is a power of two greater than or
4381 equal to eight and less than or equal to a target-specific size limit.
4382 ``align`` must be explicitly specified on atomic stores, and the store
4383 has undefined behavior if the alignment is not set to a value which is
4384 at least the size in bytes of the pointee. ``!nontemporal`` does not
4385 have any defined semantics for atomic stores.
4387 The optional constant "align" argument specifies the alignment of the
4388 operation (that is, the alignment of the memory address). A value of 0
4389 or an omitted "align" argument means that the operation has the abi
4390 alignment for the target. It is the responsibility of the code emitter
4391 to ensure that the alignment information is correct. Overestimating the
4392 alignment results in an undefined behavior. Underestimating the
4393 alignment may produce less efficient code. An alignment of 1 is always
4396 The optional !nontemporal metadata must reference a single metatadata
4397 name <index> corresponding to a metadata node with one i32 entry of
4398 value 1. The existence of the !nontemporal metatadata on the instruction
4399 tells the optimizer and code generator that this load is not expected to
4400 be reused in the cache. The code generator may select special
4401 instructions to save cache bandwidth, such as the MOVNT instruction on
4407 The contents of memory are updated to contain '``<value>``' at the
4408 location specified by the '``<pointer>``' operand. If '``<value>``' is
4409 of scalar type then the number of bytes written does not exceed the
4410 minimum number of bytes needed to hold all bits of the type. For
4411 example, storing an ``i24`` writes at most three bytes. When writing a
4412 value of a type like ``i20`` with a size that is not an integral number
4413 of bytes, it is unspecified what happens to the extra bits that do not
4414 belong to the type, but they will typically be overwritten.
4419 .. code-block:: llvm
4421 %ptr = alloca i32 ; yields {i32*}:ptr
4422 store i32 3, i32* %ptr ; yields {void}
4423 %val = load i32* %ptr ; yields {i32}:val = i32 3
4427 '``fence``' Instruction
4428 ^^^^^^^^^^^^^^^^^^^^^^^
4435 fence [singlethread] <ordering> ; yields {void}
4440 The '``fence``' instruction is used to introduce happens-before edges
4446 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4447 defines what *synchronizes-with* edges they add. They can only be given
4448 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4453 A fence A which has (at least) ``release`` ordering semantics
4454 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4455 semantics if and only if there exist atomic operations X and Y, both
4456 operating on some atomic object M, such that A is sequenced before X, X
4457 modifies M (either directly or through some side effect of a sequence
4458 headed by X), Y is sequenced before B, and Y observes M. This provides a
4459 *happens-before* dependency between A and B. Rather than an explicit
4460 ``fence``, one (but not both) of the atomic operations X or Y might
4461 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4462 still *synchronize-with* the explicit ``fence`` and establish the
4463 *happens-before* edge.
4465 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4466 ``acquire`` and ``release`` semantics specified above, participates in
4467 the global program order of other ``seq_cst`` operations and/or fences.
4469 The optional ":ref:`singlethread <singlethread>`" argument specifies
4470 that the fence only synchronizes with other fences in the same thread.
4471 (This is useful for interacting with signal handlers.)
4476 .. code-block:: llvm
4478 fence acquire ; yields {void}
4479 fence singlethread seq_cst ; yields {void}
4483 '``cmpxchg``' Instruction
4484 ^^^^^^^^^^^^^^^^^^^^^^^^^
4491 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4496 The '``cmpxchg``' instruction is used to atomically modify memory. It
4497 loads a value in memory and compares it to a given value. If they are
4498 equal, it stores a new value into the memory.
4503 There are three arguments to the '``cmpxchg``' instruction: an address
4504 to operate on, a value to compare to the value currently be at that
4505 address, and a new value to place at that address if the compared values
4506 are equal. The type of '<cmp>' must be an integer type whose bit width
4507 is a power of two greater than or equal to eight and less than or equal
4508 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4509 type, and the type of '<pointer>' must be a pointer to that type. If the
4510 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4511 to modify the number or order of execution of this ``cmpxchg`` with
4512 other :ref:`volatile operations <volatile>`.
4514 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4515 synchronizes with other atomic operations.
4517 The optional "``singlethread``" argument declares that the ``cmpxchg``
4518 is only atomic with respect to code (usually signal handlers) running in
4519 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4520 respect to all other code in the system.
4522 The pointer passed into cmpxchg must have alignment greater than or
4523 equal to the size in memory of the operand.
4528 The contents of memory at the location specified by the '``<pointer>``'
4529 operand is read and compared to '``<cmp>``'; if the read value is the
4530 equal, '``<new>``' is written. The original value at the location is
4533 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4534 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4535 atomic load with an ordering parameter determined by dropping any
4536 ``release`` part of the ``cmpxchg``'s ordering.
4541 .. code-block:: llvm
4544 %orig = atomic load i32* %ptr unordered ; yields {i32}
4548 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4549 %squared = mul i32 %cmp, %cmp
4550 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4551 %success = icmp eq i32 %cmp, %old
4552 br i1 %success, label %done, label %loop
4559 '``atomicrmw``' Instruction
4560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4567 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4572 The '``atomicrmw``' instruction is used to atomically modify memory.
4577 There are three arguments to the '``atomicrmw``' instruction: an
4578 operation to apply, an address whose value to modify, an argument to the
4579 operation. The operation must be one of the following keywords:
4593 The type of '<value>' must be an integer type whose bit width is a power
4594 of two greater than or equal to eight and less than or equal to a
4595 target-specific size limit. The type of the '``<pointer>``' operand must
4596 be a pointer to that type. If the ``atomicrmw`` is marked as
4597 ``volatile``, then the optimizer is not allowed to modify the number or
4598 order of execution of this ``atomicrmw`` with other :ref:`volatile
4599 operations <volatile>`.
4604 The contents of memory at the location specified by the '``<pointer>``'
4605 operand are atomically read, modified, and written back. The original
4606 value at the location is returned. The modification is specified by the
4609 - xchg: ``*ptr = val``
4610 - add: ``*ptr = *ptr + val``
4611 - sub: ``*ptr = *ptr - val``
4612 - and: ``*ptr = *ptr & val``
4613 - nand: ``*ptr = ~(*ptr & val)``
4614 - or: ``*ptr = *ptr | val``
4615 - xor: ``*ptr = *ptr ^ val``
4616 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4617 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4618 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4620 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4626 .. code-block:: llvm
4628 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4630 .. _i_getelementptr:
4632 '``getelementptr``' Instruction
4633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4640 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4641 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4642 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4647 The '``getelementptr``' instruction is used to get the address of a
4648 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4649 address calculation only and does not access memory.
4654 The first argument is always a pointer or a vector of pointers, and
4655 forms the basis of the calculation. The remaining arguments are indices
4656 that indicate which of the elements of the aggregate object are indexed.
4657 The interpretation of each index is dependent on the type being indexed
4658 into. The first index always indexes the pointer value given as the
4659 first argument, the second index indexes a value of the type pointed to
4660 (not necessarily the value directly pointed to, since the first index
4661 can be non-zero), etc. The first type indexed into must be a pointer
4662 value, subsequent types can be arrays, vectors, and structs. Note that
4663 subsequent types being indexed into can never be pointers, since that
4664 would require loading the pointer before continuing calculation.
4666 The type of each index argument depends on the type it is indexing into.
4667 When indexing into a (optionally packed) structure, only ``i32`` integer
4668 **constants** are allowed (when using a vector of indices they must all
4669 be the **same** ``i32`` integer constant). When indexing into an array,
4670 pointer or vector, integers of any width are allowed, and they are not
4671 required to be constant. These integers are treated as signed values
4674 For example, let's consider a C code fragment and how it gets compiled
4690 int *foo(struct ST *s) {
4691 return &s[1].Z.B[5][13];
4694 The LLVM code generated by Clang is:
4696 .. code-block:: llvm
4698 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4699 %struct.ST = type { i32, double, %struct.RT }
4701 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4703 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4710 In the example above, the first index is indexing into the
4711 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4712 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4713 indexes into the third element of the structure, yielding a
4714 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4715 structure. The third index indexes into the second element of the
4716 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4717 dimensions of the array are subscripted into, yielding an '``i32``'
4718 type. The '``getelementptr``' instruction returns a pointer to this
4719 element, thus computing a value of '``i32*``' type.
4721 Note that it is perfectly legal to index partially through a structure,
4722 returning a pointer to an inner element. Because of this, the LLVM code
4723 for the given testcase is equivalent to:
4725 .. code-block:: llvm
4727 define i32* @foo(%struct.ST* %s) {
4728 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4729 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4730 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4731 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4732 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4736 If the ``inbounds`` keyword is present, the result value of the
4737 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4738 pointer is not an *in bounds* address of an allocated object, or if any
4739 of the addresses that would be formed by successive addition of the
4740 offsets implied by the indices to the base address with infinitely
4741 precise signed arithmetic are not an *in bounds* address of that
4742 allocated object. The *in bounds* addresses for an allocated object are
4743 all the addresses that point into the object, plus the address one byte
4744 past the end. In cases where the base is a vector of pointers the
4745 ``inbounds`` keyword applies to each of the computations element-wise.
4747 If the ``inbounds`` keyword is not present, the offsets are added to the
4748 base address with silently-wrapping two's complement arithmetic. If the
4749 offsets have a different width from the pointer, they are sign-extended
4750 or truncated to the width of the pointer. The result value of the
4751 ``getelementptr`` may be outside the object pointed to by the base
4752 pointer. The result value may not necessarily be used to access memory
4753 though, even if it happens to point into allocated storage. See the
4754 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4757 The getelementptr instruction is often confusing. For some more insight
4758 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
4763 .. code-block:: llvm
4765 ; yields [12 x i8]*:aptr
4766 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4768 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4770 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4772 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4774 In cases where the pointer argument is a vector of pointers, each index
4775 must be a vector with the same number of elements. For example:
4777 .. code-block:: llvm
4779 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
4781 Conversion Operations
4782 ---------------------
4784 The instructions in this category are the conversion instructions
4785 (casting) which all take a single operand and a type. They perform
4786 various bit conversions on the operand.
4788 '``trunc .. to``' Instruction
4789 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4796 <result> = trunc <ty> <value> to <ty2> ; yields ty2
4801 The '``trunc``' instruction truncates its operand to the type ``ty2``.
4806 The '``trunc``' instruction takes a value to trunc, and a type to trunc
4807 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
4808 of the same number of integers. The bit size of the ``value`` must be
4809 larger than the bit size of the destination type, ``ty2``. Equal sized
4810 types are not allowed.
4815 The '``trunc``' instruction truncates the high order bits in ``value``
4816 and converts the remaining bits to ``ty2``. Since the source size must
4817 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
4818 It will always truncate bits.
4823 .. code-block:: llvm
4825 %X = trunc i32 257 to i8 ; yields i8:1
4826 %Y = trunc i32 123 to i1 ; yields i1:true
4827 %Z = trunc i32 122 to i1 ; yields i1:false
4828 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
4830 '``zext .. to``' Instruction
4831 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4838 <result> = zext <ty> <value> to <ty2> ; yields ty2
4843 The '``zext``' instruction zero extends its operand to type ``ty2``.
4848 The '``zext``' instruction takes a value to cast, and a type to cast it
4849 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4850 the same number of integers. The bit size of the ``value`` must be
4851 smaller than the bit size of the destination type, ``ty2``.
4856 The ``zext`` fills the high order bits of the ``value`` with zero bits
4857 until it reaches the size of the destination type, ``ty2``.
4859 When zero extending from i1, the result will always be either 0 or 1.
4864 .. code-block:: llvm
4866 %X = zext i32 257 to i64 ; yields i64:257
4867 %Y = zext i1 true to i32 ; yields i32:1
4868 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4870 '``sext .. to``' Instruction
4871 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4878 <result> = sext <ty> <value> to <ty2> ; yields ty2
4883 The '``sext``' sign extends ``value`` to the type ``ty2``.
4888 The '``sext``' instruction takes a value to cast, and a type to cast it
4889 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4890 the same number of integers. The bit size of the ``value`` must be
4891 smaller than the bit size of the destination type, ``ty2``.
4896 The '``sext``' instruction performs a sign extension by copying the sign
4897 bit (highest order bit) of the ``value`` until it reaches the bit size
4898 of the type ``ty2``.
4900 When sign extending from i1, the extension always results in -1 or 0.
4905 .. code-block:: llvm
4907 %X = sext i8 -1 to i16 ; yields i16 :65535
4908 %Y = sext i1 true to i32 ; yields i32:-1
4909 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4911 '``fptrunc .. to``' Instruction
4912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4919 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
4924 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
4929 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
4930 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
4931 The size of ``value`` must be larger than the size of ``ty2``. This
4932 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
4937 The '``fptrunc``' instruction truncates a ``value`` from a larger
4938 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
4939 point <t_floating>` type. If the value cannot fit within the
4940 destination type, ``ty2``, then the results are undefined.
4945 .. code-block:: llvm
4947 %X = fptrunc double 123.0 to float ; yields float:123.0
4948 %Y = fptrunc double 1.0E+300 to float ; yields undefined
4950 '``fpext .. to``' Instruction
4951 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4958 <result> = fpext <ty> <value> to <ty2> ; yields ty2
4963 The '``fpext``' extends a floating point ``value`` to a larger floating
4969 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
4970 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
4971 to. The source type must be smaller than the destination type.
4976 The '``fpext``' instruction extends the ``value`` from a smaller
4977 :ref:`floating point <t_floating>` type to a larger :ref:`floating
4978 point <t_floating>` type. The ``fpext`` cannot be used to make a
4979 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
4980 *no-op cast* for a floating point cast.
4985 .. code-block:: llvm
4987 %X = fpext float 3.125 to double ; yields double:3.125000e+00
4988 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
4990 '``fptoui .. to``' Instruction
4991 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4998 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5003 The '``fptoui``' converts a floating point ``value`` to its unsigned
5004 integer equivalent of type ``ty2``.
5009 The '``fptoui``' instruction takes a value to cast, which must be a
5010 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5011 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5012 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5013 type with the same number of elements as ``ty``
5018 The '``fptoui``' instruction converts its :ref:`floating
5019 point <t_floating>` operand into the nearest (rounding towards zero)
5020 unsigned integer value. If the value cannot fit in ``ty2``, the results
5026 .. code-block:: llvm
5028 %X = fptoui double 123.0 to i32 ; yields i32:123
5029 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5030 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5032 '``fptosi .. to``' Instruction
5033 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5040 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5045 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5046 ``value`` to type ``ty2``.
5051 The '``fptosi``' instruction takes a value to cast, which must be a
5052 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5053 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5054 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5055 type with the same number of elements as ``ty``
5060 The '``fptosi``' instruction converts its :ref:`floating
5061 point <t_floating>` operand into the nearest (rounding towards zero)
5062 signed integer value. If the value cannot fit in ``ty2``, the results
5068 .. code-block:: llvm
5070 %X = fptosi double -123.0 to i32 ; yields i32:-123
5071 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5072 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5074 '``uitofp .. to``' Instruction
5075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5082 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5087 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5088 and converts that value to the ``ty2`` type.
5093 The '``uitofp``' instruction takes a value to cast, which must be a
5094 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5095 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5096 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5097 type with the same number of elements as ``ty``
5102 The '``uitofp``' instruction interprets its operand as an unsigned
5103 integer quantity and converts it to the corresponding floating point
5104 value. If the value cannot fit in the floating point value, the results
5110 .. code-block:: llvm
5112 %X = uitofp i32 257 to float ; yields float:257.0
5113 %Y = uitofp i8 -1 to double ; yields double:255.0
5115 '``sitofp .. to``' Instruction
5116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5123 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5128 The '``sitofp``' instruction regards ``value`` as a signed integer and
5129 converts that value to the ``ty2`` type.
5134 The '``sitofp``' instruction takes a value to cast, which must be a
5135 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5136 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5137 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5138 type with the same number of elements as ``ty``
5143 The '``sitofp``' instruction interprets its operand as a signed integer
5144 quantity and converts it to the corresponding floating point value. If
5145 the value cannot fit in the floating point value, the results are
5151 .. code-block:: llvm
5153 %X = sitofp i32 257 to float ; yields float:257.0
5154 %Y = sitofp i8 -1 to double ; yields double:-1.0
5158 '``ptrtoint .. to``' Instruction
5159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5166 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5171 The '``ptrtoint``' instruction converts the pointer or a vector of
5172 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5177 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5178 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5179 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5180 a vector of integers type.
5185 The '``ptrtoint``' instruction converts ``value`` to integer type
5186 ``ty2`` by interpreting the pointer value as an integer and either
5187 truncating or zero extending that value to the size of the integer type.
5188 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5189 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5190 the same size, then nothing is done (*no-op cast*) other than a type
5196 .. code-block:: llvm
5198 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5199 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5200 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5204 '``inttoptr .. to``' Instruction
5205 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5212 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5217 The '``inttoptr``' instruction converts an integer ``value`` to a
5218 pointer type, ``ty2``.
5223 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5224 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5230 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5231 applying either a zero extension or a truncation depending on the size
5232 of the integer ``value``. If ``value`` is larger than the size of a
5233 pointer then a truncation is done. If ``value`` is smaller than the size
5234 of a pointer then a zero extension is done. If they are the same size,
5235 nothing is done (*no-op cast*).
5240 .. code-block:: llvm
5242 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5243 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5244 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5245 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5249 '``bitcast .. to``' Instruction
5250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5257 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5262 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5268 The '``bitcast``' instruction takes a value to cast, which must be a
5269 non-aggregate first class value, and a type to cast it to, which must
5270 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5271 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5272 If the source type is a pointer, the destination type must also be a
5273 pointer. This instruction supports bitwise conversion of vectors to
5274 integers and to vectors of other types (as long as they have the same
5280 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5281 always a *no-op cast* because no bits change with this conversion. The
5282 conversion is done as if the ``value`` had been stored to memory and
5283 read back as type ``ty2``. Pointer (or vector of pointers) types may
5284 only be converted to other pointer (or vector of pointers) types with
5285 this instruction. To convert pointers to other types, use the
5286 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5292 .. code-block:: llvm
5294 %X = bitcast i8 255 to i8 ; yields i8 :-1
5295 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5296 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5297 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5304 The instructions in this category are the "miscellaneous" instructions,
5305 which defy better classification.
5309 '``icmp``' Instruction
5310 ^^^^^^^^^^^^^^^^^^^^^^
5317 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5322 The '``icmp``' instruction returns a boolean value or a vector of
5323 boolean values based on comparison of its two integer, integer vector,
5324 pointer, or pointer vector operands.
5329 The '``icmp``' instruction takes three operands. The first operand is
5330 the condition code indicating the kind of comparison to perform. It is
5331 not a value, just a keyword. The possible condition code are:
5334 #. ``ne``: not equal
5335 #. ``ugt``: unsigned greater than
5336 #. ``uge``: unsigned greater or equal
5337 #. ``ult``: unsigned less than
5338 #. ``ule``: unsigned less or equal
5339 #. ``sgt``: signed greater than
5340 #. ``sge``: signed greater or equal
5341 #. ``slt``: signed less than
5342 #. ``sle``: signed less or equal
5344 The remaining two arguments must be :ref:`integer <t_integer>` or
5345 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5346 must also be identical types.
5351 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5352 code given as ``cond``. The comparison performed always yields either an
5353 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5355 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5356 otherwise. No sign interpretation is necessary or performed.
5357 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5358 otherwise. No sign interpretation is necessary or performed.
5359 #. ``ugt``: interprets the operands as unsigned values and yields
5360 ``true`` if ``op1`` is greater than ``op2``.
5361 #. ``uge``: interprets the operands as unsigned values and yields
5362 ``true`` if ``op1`` is greater than or equal to ``op2``.
5363 #. ``ult``: interprets the operands as unsigned values and yields
5364 ``true`` if ``op1`` is less than ``op2``.
5365 #. ``ule``: interprets the operands as unsigned values and yields
5366 ``true`` if ``op1`` is less than or equal to ``op2``.
5367 #. ``sgt``: interprets the operands as signed values and yields ``true``
5368 if ``op1`` is greater than ``op2``.
5369 #. ``sge``: interprets the operands as signed values and yields ``true``
5370 if ``op1`` is greater than or equal to ``op2``.
5371 #. ``slt``: interprets the operands as signed values and yields ``true``
5372 if ``op1`` is less than ``op2``.
5373 #. ``sle``: interprets the operands as signed values and yields ``true``
5374 if ``op1`` is less than or equal to ``op2``.
5376 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5377 are compared as if they were integers.
5379 If the operands are integer vectors, then they are compared element by
5380 element. The result is an ``i1`` vector with the same number of elements
5381 as the values being compared. Otherwise, the result is an ``i1``.
5386 .. code-block:: llvm
5388 <result> = icmp eq i32 4, 5 ; yields: result=false
5389 <result> = icmp ne float* %X, %X ; yields: result=false
5390 <result> = icmp ult i16 4, 5 ; yields: result=true
5391 <result> = icmp sgt i16 4, 5 ; yields: result=false
5392 <result> = icmp ule i16 -4, 5 ; yields: result=false
5393 <result> = icmp sge i16 4, 5 ; yields: result=false
5395 Note that the code generator does not yet support vector types with the
5396 ``icmp`` instruction.
5400 '``fcmp``' Instruction
5401 ^^^^^^^^^^^^^^^^^^^^^^
5408 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5413 The '``fcmp``' instruction returns a boolean value or vector of boolean
5414 values based on comparison of its operands.
5416 If the operands are floating point scalars, then the result type is a
5417 boolean (:ref:`i1 <t_integer>`).
5419 If the operands are floating point vectors, then the result type is a
5420 vector of boolean with the same number of elements as the operands being
5426 The '``fcmp``' instruction takes three operands. The first operand is
5427 the condition code indicating the kind of comparison to perform. It is
5428 not a value, just a keyword. The possible condition code are:
5430 #. ``false``: no comparison, always returns false
5431 #. ``oeq``: ordered and equal
5432 #. ``ogt``: ordered and greater than
5433 #. ``oge``: ordered and greater than or equal
5434 #. ``olt``: ordered and less than
5435 #. ``ole``: ordered and less than or equal
5436 #. ``one``: ordered and not equal
5437 #. ``ord``: ordered (no nans)
5438 #. ``ueq``: unordered or equal
5439 #. ``ugt``: unordered or greater than
5440 #. ``uge``: unordered or greater than or equal
5441 #. ``ult``: unordered or less than
5442 #. ``ule``: unordered or less than or equal
5443 #. ``une``: unordered or not equal
5444 #. ``uno``: unordered (either nans)
5445 #. ``true``: no comparison, always returns true
5447 *Ordered* means that neither operand is a QNAN while *unordered* means
5448 that either operand may be a QNAN.
5450 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5451 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5452 type. They must have identical types.
5457 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5458 condition code given as ``cond``. If the operands are vectors, then the
5459 vectors are compared element by element. Each comparison performed
5460 always yields an :ref:`i1 <t_integer>` result, as follows:
5462 #. ``false``: always yields ``false``, regardless of operands.
5463 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5464 is equal to ``op2``.
5465 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5466 is greater than ``op2``.
5467 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5468 is greater than or equal to ``op2``.
5469 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5470 is less than ``op2``.
5471 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5472 is less than or equal to ``op2``.
5473 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5474 is not equal to ``op2``.
5475 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5476 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5478 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5479 greater than ``op2``.
5480 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5481 greater than or equal to ``op2``.
5482 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5484 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5485 less than or equal to ``op2``.
5486 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5487 not equal to ``op2``.
5488 #. ``uno``: yields ``true`` if either operand is a QNAN.
5489 #. ``true``: always yields ``true``, regardless of operands.
5494 .. code-block:: llvm
5496 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5497 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5498 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5499 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5501 Note that the code generator does not yet support vector types with the
5502 ``fcmp`` instruction.
5506 '``phi``' Instruction
5507 ^^^^^^^^^^^^^^^^^^^^^
5514 <result> = phi <ty> [ <val0>, <label0>], ...
5519 The '``phi``' instruction is used to implement the φ node in the SSA
5520 graph representing the function.
5525 The type of the incoming values is specified with the first type field.
5526 After this, the '``phi``' instruction takes a list of pairs as
5527 arguments, with one pair for each predecessor basic block of the current
5528 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5529 the value arguments to the PHI node. Only labels may be used as the
5532 There must be no non-phi instructions between the start of a basic block
5533 and the PHI instructions: i.e. PHI instructions must be first in a basic
5536 For the purposes of the SSA form, the use of each incoming value is
5537 deemed to occur on the edge from the corresponding predecessor block to
5538 the current block (but after any definition of an '``invoke``'
5539 instruction's return value on the same edge).
5544 At runtime, the '``phi``' instruction logically takes on the value
5545 specified by the pair corresponding to the predecessor basic block that
5546 executed just prior to the current block.
5551 .. code-block:: llvm
5553 Loop: ; Infinite loop that counts from 0 on up...
5554 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5555 %nextindvar = add i32 %indvar, 1
5560 '``select``' Instruction
5561 ^^^^^^^^^^^^^^^^^^^^^^^^
5568 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5570 selty is either i1 or {<N x i1>}
5575 The '``select``' instruction is used to choose one value based on a
5576 condition, without branching.
5581 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5582 values indicating the condition, and two values of the same :ref:`first
5583 class <t_firstclass>` type. If the val1/val2 are vectors and the
5584 condition is a scalar, then entire vectors are selected, not individual
5590 If the condition is an i1 and it evaluates to 1, the instruction returns
5591 the first value argument; otherwise, it returns the second value
5594 If the condition is a vector of i1, then the value arguments must be
5595 vectors of the same size, and the selection is done element by element.
5600 .. code-block:: llvm
5602 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5606 '``call``' Instruction
5607 ^^^^^^^^^^^^^^^^^^^^^^
5614 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5619 The '``call``' instruction represents a simple function call.
5624 This instruction requires several arguments:
5626 #. The optional "tail" marker indicates that the callee function does
5627 not access any allocas or varargs in the caller. Note that calls may
5628 be marked "tail" even if they do not occur before a
5629 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5630 function call is eligible for tail call optimization, but `might not
5631 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5632 The code generator may optimize calls marked "tail" with either 1)
5633 automatic `sibling call
5634 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5635 callee have matching signatures, or 2) forced tail call optimization
5636 when the following extra requirements are met:
5638 - Caller and callee both have the calling convention ``fastcc``.
5639 - The call is in tail position (ret immediately follows call and ret
5640 uses value of call or is void).
5641 - Option ``-tailcallopt`` is enabled, or
5642 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5643 - `Platform specific constraints are
5644 met. <CodeGenerator.html#tailcallopt>`_
5646 #. The optional "cconv" marker indicates which :ref:`calling
5647 convention <callingconv>` the call should use. If none is
5648 specified, the call defaults to using C calling conventions. The
5649 calling convention of the call must match the calling convention of
5650 the target function, or else the behavior is undefined.
5651 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5652 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5654 #. '``ty``': the type of the call instruction itself which is also the
5655 type of the return value. Functions that return no value are marked
5657 #. '``fnty``': shall be the signature of the pointer to function value
5658 being invoked. The argument types must match the types implied by
5659 this signature. This type can be omitted if the function is not
5660 varargs and if the function type does not return a pointer to a
5662 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5663 be invoked. In most cases, this is a direct function invocation, but
5664 indirect ``call``'s are just as possible, calling an arbitrary pointer
5666 #. '``function args``': argument list whose types match the function
5667 signature argument types and parameter attributes. All arguments must
5668 be of :ref:`first class <t_firstclass>` type. If the function signature
5669 indicates the function accepts a variable number of arguments, the
5670 extra arguments can be specified.
5671 #. The optional :ref:`function attributes <fnattrs>` list. Only
5672 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5673 attributes are valid here.
5678 The '``call``' instruction is used to cause control flow to transfer to
5679 a specified function, with its incoming arguments bound to the specified
5680 values. Upon a '``ret``' instruction in the called function, control
5681 flow continues with the instruction after the function call, and the
5682 return value of the function is bound to the result argument.
5687 .. code-block:: llvm
5689 %retval = call i32 @test(i32 %argc)
5690 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5691 %X = tail call i32 @foo() ; yields i32
5692 %Y = tail call fastcc i32 @foo() ; yields i32
5693 call void %foo(i8 97 signext)
5695 %struct.A = type { i32, i8 }
5696 %r = call %struct.A @foo() ; yields { 32, i8 }
5697 %gr = extractvalue %struct.A %r, 0 ; yields i32
5698 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5699 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5700 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5702 llvm treats calls to some functions with names and arguments that match
5703 the standard C99 library as being the C99 library functions, and may
5704 perform optimizations or generate code for them under that assumption.
5705 This is something we'd like to change in the future to provide better
5706 support for freestanding environments and non-C-based languages.
5710 '``va_arg``' Instruction
5711 ^^^^^^^^^^^^^^^^^^^^^^^^
5718 <resultval> = va_arg <va_list*> <arglist>, <argty>
5723 The '``va_arg``' instruction is used to access arguments passed through
5724 the "variable argument" area of a function call. It is used to implement
5725 the ``va_arg`` macro in C.
5730 This instruction takes a ``va_list*`` value and the type of the
5731 argument. It returns a value of the specified argument type and
5732 increments the ``va_list`` to point to the next argument. The actual
5733 type of ``va_list`` is target specific.
5738 The '``va_arg``' instruction loads an argument of the specified type
5739 from the specified ``va_list`` and causes the ``va_list`` to point to
5740 the next argument. For more information, see the variable argument
5741 handling :ref:`Intrinsic Functions <int_varargs>`.
5743 It is legal for this instruction to be called in a function which does
5744 not take a variable number of arguments, for example, the ``vfprintf``
5747 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5748 function <intrinsics>` because it takes a type as an argument.
5753 See the :ref:`variable argument processing <int_varargs>` section.
5755 Note that the code generator does not yet fully support va\_arg on many
5756 targets. Also, it does not currently support va\_arg with aggregate
5757 types on any target.
5761 '``landingpad``' Instruction
5762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5769 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
5770 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
5772 <clause> := catch <type> <value>
5773 <clause> := filter <array constant type> <array constant>
5778 The '``landingpad``' instruction is used by `LLVM's exception handling
5779 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5780 is a landing pad — one where the exception lands, and corresponds to the
5781 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
5782 defines values supplied by the personality function (``pers_fn``) upon
5783 re-entry to the function. The ``resultval`` has the type ``resultty``.
5788 This instruction takes a ``pers_fn`` value. This is the personality
5789 function associated with the unwinding mechanism. The optional
5790 ``cleanup`` flag indicates that the landing pad block is a cleanup.
5792 A ``clause`` begins with the clause type — ``catch`` or ``filter`` — and
5793 contains the global variable representing the "type" that may be caught
5794 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
5795 clause takes an array constant as its argument. Use
5796 "``[0 x i8**] undef``" for a filter which cannot throw. The
5797 '``landingpad``' instruction must contain *at least* one ``clause`` or
5798 the ``cleanup`` flag.
5803 The '``landingpad``' instruction defines the values which are set by the
5804 personality function (``pers_fn``) upon re-entry to the function, and
5805 therefore the "result type" of the ``landingpad`` instruction. As with
5806 calling conventions, how the personality function results are
5807 represented in LLVM IR is target specific.
5809 The clauses are applied in order from top to bottom. If two
5810 ``landingpad`` instructions are merged together through inlining, the
5811 clauses from the calling function are appended to the list of clauses.
5812 When the call stack is being unwound due to an exception being thrown,
5813 the exception is compared against each ``clause`` in turn. If it doesn't
5814 match any of the clauses, and the ``cleanup`` flag is not set, then
5815 unwinding continues further up the call stack.
5817 The ``landingpad`` instruction has several restrictions:
5819 - A landing pad block is a basic block which is the unwind destination
5820 of an '``invoke``' instruction.
5821 - A landing pad block must have a '``landingpad``' instruction as its
5822 first non-PHI instruction.
5823 - There can be only one '``landingpad``' instruction within the landing
5825 - A basic block that is not a landing pad block may not include a
5826 '``landingpad``' instruction.
5827 - All '``landingpad``' instructions in a function must have the same
5828 personality function.
5833 .. code-block:: llvm
5835 ;; A landing pad which can catch an integer.
5836 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5838 ;; A landing pad that is a cleanup.
5839 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5841 ;; A landing pad which can catch an integer and can only throw a double.
5842 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5844 filter [1 x i8**] [@_ZTId]
5851 LLVM supports the notion of an "intrinsic function". These functions
5852 have well known names and semantics and are required to follow certain
5853 restrictions. Overall, these intrinsics represent an extension mechanism
5854 for the LLVM language that does not require changing all of the
5855 transformations in LLVM when adding to the language (or the bitcode
5856 reader/writer, the parser, etc...).
5858 Intrinsic function names must all start with an "``llvm.``" prefix. This
5859 prefix is reserved in LLVM for intrinsic names; thus, function names may
5860 not begin with this prefix. Intrinsic functions must always be external
5861 functions: you cannot define the body of intrinsic functions. Intrinsic
5862 functions may only be used in call or invoke instructions: it is illegal
5863 to take the address of an intrinsic function. Additionally, because
5864 intrinsic functions are part of the LLVM language, it is required if any
5865 are added that they be documented here.
5867 Some intrinsic functions can be overloaded, i.e., the intrinsic
5868 represents a family of functions that perform the same operation but on
5869 different data types. Because LLVM can represent over 8 million
5870 different integer types, overloading is used commonly to allow an
5871 intrinsic function to operate on any integer type. One or more of the
5872 argument types or the result type can be overloaded to accept any
5873 integer type. Argument types may also be defined as exactly matching a
5874 previous argument's type or the result type. This allows an intrinsic
5875 function which accepts multiple arguments, but needs all of them to be
5876 of the same type, to only be overloaded with respect to a single
5877 argument or the result.
5879 Overloaded intrinsics will have the names of its overloaded argument
5880 types encoded into its function name, each preceded by a period. Only
5881 those types which are overloaded result in a name suffix. Arguments
5882 whose type is matched against another type do not. For example, the
5883 ``llvm.ctpop`` function can take an integer of any width and returns an
5884 integer of exactly the same integer width. This leads to a family of
5885 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
5886 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
5887 overloaded, and only one type suffix is required. Because the argument's
5888 type is matched against the return type, it does not require its own
5891 To learn how to add an intrinsic function, please see the `Extending
5892 LLVM Guide <ExtendingLLVM.html>`_.
5896 Variable Argument Handling Intrinsics
5897 -------------------------------------
5899 Variable argument support is defined in LLVM with the
5900 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
5901 functions. These functions are related to the similarly named macros
5902 defined in the ``<stdarg.h>`` header file.
5904 All of these functions operate on arguments that use a target-specific
5905 value type "``va_list``". The LLVM assembly language reference manual
5906 does not define what this type is, so all transformations should be
5907 prepared to handle these functions regardless of the type used.
5909 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
5910 variable argument handling intrinsic functions are used.
5912 .. code-block:: llvm
5914 define i32 @test(i32 %X, ...) {
5915 ; Initialize variable argument processing
5917 %ap2 = bitcast i8** %ap to i8*
5918 call void @llvm.va_start(i8* %ap2)
5920 ; Read a single integer argument
5921 %tmp = va_arg i8** %ap, i32
5923 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5925 %aq2 = bitcast i8** %aq to i8*
5926 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5927 call void @llvm.va_end(i8* %aq2)
5929 ; Stop processing of arguments.
5930 call void @llvm.va_end(i8* %ap2)
5934 declare void @llvm.va_start(i8*)
5935 declare void @llvm.va_copy(i8*, i8*)
5936 declare void @llvm.va_end(i8*)
5940 '``llvm.va_start``' Intrinsic
5941 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5948 declare void %llvm.va_start(i8* <arglist>)
5953 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
5954 subsequent use by ``va_arg``.
5959 The argument is a pointer to a ``va_list`` element to initialize.
5964 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
5965 available in C. In a target-dependent way, it initializes the
5966 ``va_list`` element to which the argument points, so that the next call
5967 to ``va_arg`` will produce the first variable argument passed to the
5968 function. Unlike the C ``va_start`` macro, this intrinsic does not need
5969 to know the last argument of the function as the compiler can figure
5972 '``llvm.va_end``' Intrinsic
5973 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5980 declare void @llvm.va_end(i8* <arglist>)
5985 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
5986 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
5991 The argument is a pointer to a ``va_list`` to destroy.
5996 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
5997 available in C. In a target-dependent way, it destroys the ``va_list``
5998 element to which the argument points. Calls to
5999 :ref:`llvm.va_start <int_va_start>` and
6000 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6005 '``llvm.va_copy``' Intrinsic
6006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6013 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6018 The '``llvm.va_copy``' intrinsic copies the current argument position
6019 from the source argument list to the destination argument list.
6024 The first argument is a pointer to a ``va_list`` element to initialize.
6025 The second argument is a pointer to a ``va_list`` element to copy from.
6030 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6031 available in C. In a target-dependent way, it copies the source
6032 ``va_list`` element into the destination ``va_list`` element. This
6033 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6034 arbitrarily complex and require, for example, memory allocation.
6036 Accurate Garbage Collection Intrinsics
6037 --------------------------------------
6039 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6040 (GC) requires the implementation and generation of these intrinsics.
6041 These intrinsics allow identification of :ref:`GC roots on the
6042 stack <int_gcroot>`, as well as garbage collector implementations that
6043 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6044 Front-ends for type-safe garbage collected languages should generate
6045 these intrinsics to make use of the LLVM garbage collectors. For more
6046 details, see `Accurate Garbage Collection with
6047 LLVM <GarbageCollection.html>`_.
6049 The garbage collection intrinsics only operate on objects in the generic
6050 address space (address space zero).
6054 '``llvm.gcroot``' Intrinsic
6055 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6062 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6067 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6068 the code generator, and allows some metadata to be associated with it.
6073 The first argument specifies the address of a stack object that contains
6074 the root pointer. The second pointer (which must be either a constant or
6075 a global value address) contains the meta-data to be associated with the
6081 At runtime, a call to this intrinsic stores a null pointer into the
6082 "ptrloc" location. At compile-time, the code generator generates
6083 information to allow the runtime to find the pointer at GC safe points.
6084 The '``llvm.gcroot``' intrinsic may only be used in a function which
6085 :ref:`specifies a GC algorithm <gc>`.
6089 '``llvm.gcread``' Intrinsic
6090 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6097 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6102 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6103 locations, allowing garbage collector implementations that require read
6109 The second argument is the address to read from, which should be an
6110 address allocated from the garbage collector. The first object is a
6111 pointer to the start of the referenced object, if needed by the language
6112 runtime (otherwise null).
6117 The '``llvm.gcread``' intrinsic has the same semantics as a load
6118 instruction, but may be replaced with substantially more complex code by
6119 the garbage collector runtime, as needed. The '``llvm.gcread``'
6120 intrinsic may only be used in a function which :ref:`specifies a GC
6125 '``llvm.gcwrite``' Intrinsic
6126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6133 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6138 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6139 locations, allowing garbage collector implementations that require write
6140 barriers (such as generational or reference counting collectors).
6145 The first argument is the reference to store, the second is the start of
6146 the object to store it to, and the third is the address of the field of
6147 Obj to store to. If the runtime does not require a pointer to the
6148 object, Obj may be null.
6153 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6154 instruction, but may be replaced with substantially more complex code by
6155 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6156 intrinsic may only be used in a function which :ref:`specifies a GC
6159 Code Generator Intrinsics
6160 -------------------------
6162 These intrinsics are provided by LLVM to expose special features that
6163 may only be implemented with code generator support.
6165 '``llvm.returnaddress``' Intrinsic
6166 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6173 declare i8 *@llvm.returnaddress(i32 <level>)
6178 The '``llvm.returnaddress``' intrinsic attempts to compute a
6179 target-specific value indicating the return address of the current
6180 function or one of its callers.
6185 The argument to this intrinsic indicates which function to return the
6186 address for. Zero indicates the calling function, one indicates its
6187 caller, etc. The argument is **required** to be a constant integer
6193 The '``llvm.returnaddress``' intrinsic either returns a pointer
6194 indicating the return address of the specified call frame, or zero if it
6195 cannot be identified. The value returned by this intrinsic is likely to
6196 be incorrect or 0 for arguments other than zero, so it should only be
6197 used for debugging purposes.
6199 Note that calling this intrinsic does not prevent function inlining or
6200 other aggressive transformations, so the value returned may not be that
6201 of the obvious source-language caller.
6203 '``llvm.frameaddress``' Intrinsic
6204 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6211 declare i8* @llvm.frameaddress(i32 <level>)
6216 The '``llvm.frameaddress``' intrinsic attempts to return the
6217 target-specific frame pointer value for the specified stack frame.
6222 The argument to this intrinsic indicates which function to return the
6223 frame pointer for. Zero indicates the calling function, one indicates
6224 its caller, etc. The argument is **required** to be a constant integer
6230 The '``llvm.frameaddress``' intrinsic either returns a pointer
6231 indicating the frame address of the specified call frame, or zero if it
6232 cannot be identified. The value returned by this intrinsic is likely to
6233 be incorrect or 0 for arguments other than zero, so it should only be
6234 used for debugging purposes.
6236 Note that calling this intrinsic does not prevent function inlining or
6237 other aggressive transformations, so the value returned may not be that
6238 of the obvious source-language caller.
6242 '``llvm.stacksave``' Intrinsic
6243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6250 declare i8* @llvm.stacksave()
6255 The '``llvm.stacksave``' intrinsic is used to remember the current state
6256 of the function stack, for use with
6257 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6258 implementing language features like scoped automatic variable sized
6264 This intrinsic returns a opaque pointer value that can be passed to
6265 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6266 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6267 ``llvm.stacksave``, it effectively restores the state of the stack to
6268 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6269 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6270 were allocated after the ``llvm.stacksave`` was executed.
6272 .. _int_stackrestore:
6274 '``llvm.stackrestore``' Intrinsic
6275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6282 declare void @llvm.stackrestore(i8* %ptr)
6287 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6288 the function stack to the state it was in when the corresponding
6289 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6290 useful for implementing language features like scoped automatic variable
6291 sized arrays in C99.
6296 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6298 '``llvm.prefetch``' Intrinsic
6299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6306 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6311 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6312 insert a prefetch instruction if supported; otherwise, it is a noop.
6313 Prefetches have no effect on the behavior of the program but can change
6314 its performance characteristics.
6319 ``address`` is the address to be prefetched, ``rw`` is the specifier
6320 determining if the fetch should be for a read (0) or write (1), and
6321 ``locality`` is a temporal locality specifier ranging from (0) - no
6322 locality, to (3) - extremely local keep in cache. The ``cache type``
6323 specifies whether the prefetch is performed on the data (1) or
6324 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6325 arguments must be constant integers.
6330 This intrinsic does not modify the behavior of the program. In
6331 particular, prefetches cannot trap and do not produce a value. On
6332 targets that support this intrinsic, the prefetch can provide hints to
6333 the processor cache for better performance.
6335 '``llvm.pcmarker``' Intrinsic
6336 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6343 declare void @llvm.pcmarker(i32 <id>)
6348 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6349 Counter (PC) in a region of code to simulators and other tools. The
6350 method is target specific, but it is expected that the marker will use
6351 exported symbols to transmit the PC of the marker. The marker makes no
6352 guarantees that it will remain with any specific instruction after
6353 optimizations. It is possible that the presence of a marker will inhibit
6354 optimizations. The intended use is to be inserted after optimizations to
6355 allow correlations of simulation runs.
6360 ``id`` is a numerical id identifying the marker.
6365 This intrinsic does not modify the behavior of the program. Backends
6366 that do not support this intrinsic may ignore it.
6368 '``llvm.readcyclecounter``' Intrinsic
6369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6376 declare i64 @llvm.readcyclecounter()
6381 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6382 counter register (or similar low latency, high accuracy clocks) on those
6383 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6384 should map to RPCC. As the backing counters overflow quickly (on the
6385 order of 9 seconds on alpha), this should only be used for small
6391 When directly supported, reading the cycle counter should not modify any
6392 memory. Implementations are allowed to either return a application
6393 specific value or a system wide value. On backends without support, this
6394 is lowered to a constant 0.
6396 Standard C Library Intrinsics
6397 -----------------------------
6399 LLVM provides intrinsics for a few important standard C library
6400 functions. These intrinsics allow source-language front-ends to pass
6401 information about the alignment of the pointer arguments to the code
6402 generator, providing opportunity for more efficient code generation.
6406 '``llvm.memcpy``' Intrinsic
6407 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6412 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6413 integer bit width and for different address spaces. Not all targets
6414 support all bit widths however.
6418 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6419 i32 <len>, i32 <align>, i1 <isvolatile>)
6420 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6421 i64 <len>, i32 <align>, i1 <isvolatile>)
6426 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6427 source location to the destination location.
6429 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6430 intrinsics do not return a value, takes extra alignment/isvolatile
6431 arguments and the pointers can be in specified address spaces.
6436 The first argument is a pointer to the destination, the second is a
6437 pointer to the source. The third argument is an integer argument
6438 specifying the number of bytes to copy, the fourth argument is the
6439 alignment of the source and destination locations, and the fifth is a
6440 boolean indicating a volatile access.
6442 If the call to this intrinsic has an alignment value that is not 0 or 1,
6443 then the caller guarantees that both the source and destination pointers
6444 are aligned to that boundary.
6446 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6447 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6448 very cleanly specified and it is unwise to depend on it.
6453 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6454 source location to the destination location, which are not allowed to
6455 overlap. It copies "len" bytes of memory over. If the argument is known
6456 to be aligned to some boundary, this can be specified as the fourth
6457 argument, otherwise it should be set to 0 or 1.
6459 '``llvm.memmove``' Intrinsic
6460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6465 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6466 bit width and for different address space. Not all targets support all
6471 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6472 i32 <len>, i32 <align>, i1 <isvolatile>)
6473 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6474 i64 <len>, i32 <align>, i1 <isvolatile>)
6479 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6480 source location to the destination location. It is similar to the
6481 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6484 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6485 intrinsics do not return a value, takes extra alignment/isvolatile
6486 arguments and the pointers can be in specified address spaces.
6491 The first argument is a pointer to the destination, the second is a
6492 pointer to the source. The third argument is an integer argument
6493 specifying the number of bytes to copy, the fourth argument is the
6494 alignment of the source and destination locations, and the fifth is a
6495 boolean indicating a volatile access.
6497 If the call to this intrinsic has an alignment value that is not 0 or 1,
6498 then the caller guarantees that the source and destination pointers are
6499 aligned to that boundary.
6501 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6502 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6503 not very cleanly specified and it is unwise to depend on it.
6508 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6509 source location to the destination location, which may overlap. It
6510 copies "len" bytes of memory over. If the argument is known to be
6511 aligned to some boundary, this can be specified as the fourth argument,
6512 otherwise it should be set to 0 or 1.
6514 '``llvm.memset.*``' Intrinsics
6515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6520 This is an overloaded intrinsic. You can use llvm.memset on any integer
6521 bit width and for different address spaces. However, not all targets
6522 support all bit widths.
6526 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6527 i32 <len>, i32 <align>, i1 <isvolatile>)
6528 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6529 i64 <len>, i32 <align>, i1 <isvolatile>)
6534 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6535 particular byte value.
6537 Note that, unlike the standard libc function, the ``llvm.memset``
6538 intrinsic does not return a value and takes extra alignment/volatile
6539 arguments. Also, the destination can be in an arbitrary address space.
6544 The first argument is a pointer to the destination to fill, the second
6545 is the byte value with which to fill it, the third argument is an
6546 integer argument specifying the number of bytes to fill, and the fourth
6547 argument is the known alignment of the destination location.
6549 If the call to this intrinsic has an alignment value that is not 0 or 1,
6550 then the caller guarantees that the destination pointer is aligned to
6553 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6554 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6555 very cleanly specified and it is unwise to depend on it.
6560 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6561 at the destination location. If the argument is known to be aligned to
6562 some boundary, this can be specified as the fourth argument, otherwise
6563 it should be set to 0 or 1.
6565 '``llvm.sqrt.*``' Intrinsic
6566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6571 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6572 floating point or vector of floating point type. Not all targets support
6577 declare float @llvm.sqrt.f32(float %Val)
6578 declare double @llvm.sqrt.f64(double %Val)
6579 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6580 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6581 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6586 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6587 returning the same value as the libm '``sqrt``' functions would. Unlike
6588 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6589 negative numbers other than -0.0 (which allows for better optimization,
6590 because there is no need to worry about errno being set).
6591 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6596 The argument and return value are floating point numbers of the same
6602 This function returns the sqrt of the specified operand if it is a
6603 nonnegative floating point number.
6605 '``llvm.powi.*``' Intrinsic
6606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6611 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6612 floating point or vector of floating point type. Not all targets support
6617 declare float @llvm.powi.f32(float %Val, i32 %power)
6618 declare double @llvm.powi.f64(double %Val, i32 %power)
6619 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6620 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6621 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6626 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6627 specified (positive or negative) power. The order of evaluation of
6628 multiplications is not defined. When a vector of floating point type is
6629 used, the second argument remains a scalar integer value.
6634 The second argument is an integer power, and the first is a value to
6635 raise to that power.
6640 This function returns the first value raised to the second power with an
6641 unspecified sequence of rounding operations.
6643 '``llvm.sin.*``' Intrinsic
6644 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6649 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6650 floating point or vector of floating point type. Not all targets support
6655 declare float @llvm.sin.f32(float %Val)
6656 declare double @llvm.sin.f64(double %Val)
6657 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6658 declare fp128 @llvm.sin.f128(fp128 %Val)
6659 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6664 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6669 The argument and return value are floating point numbers of the same
6675 This function returns the sine of the specified operand, returning the
6676 same values as the libm ``sin`` functions would, and handles error
6677 conditions in the same way.
6679 '``llvm.cos.*``' Intrinsic
6680 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6685 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6686 floating point or vector of floating point type. Not all targets support
6691 declare float @llvm.cos.f32(float %Val)
6692 declare double @llvm.cos.f64(double %Val)
6693 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6694 declare fp128 @llvm.cos.f128(fp128 %Val)
6695 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6700 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6705 The argument and return value are floating point numbers of the same
6711 This function returns the cosine of the specified operand, returning the
6712 same values as the libm ``cos`` functions would, and handles error
6713 conditions in the same way.
6715 '``llvm.pow.*``' Intrinsic
6716 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6721 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6722 floating point or vector of floating point type. Not all targets support
6727 declare float @llvm.pow.f32(float %Val, float %Power)
6728 declare double @llvm.pow.f64(double %Val, double %Power)
6729 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6730 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6731 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6736 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6737 specified (positive or negative) power.
6742 The second argument is a floating point power, and the first is a value
6743 to raise to that power.
6748 This function returns the first value raised to the second power,
6749 returning the same values as the libm ``pow`` functions would, and
6750 handles error conditions in the same way.
6752 '``llvm.exp.*``' Intrinsic
6753 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6758 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
6759 floating point or vector of floating point type. Not all targets support
6764 declare float @llvm.exp.f32(float %Val)
6765 declare double @llvm.exp.f64(double %Val)
6766 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6767 declare fp128 @llvm.exp.f128(fp128 %Val)
6768 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6773 The '``llvm.exp.*``' intrinsics perform the exp function.
6778 The argument and return value are floating point numbers of the same
6784 This function returns the same values as the libm ``exp`` functions
6785 would, and handles error conditions in the same way.
6787 '``llvm.exp2.*``' Intrinsic
6788 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6793 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
6794 floating point or vector of floating point type. Not all targets support
6799 declare float @llvm.exp2.f32(float %Val)
6800 declare double @llvm.exp2.f64(double %Val)
6801 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
6802 declare fp128 @llvm.exp2.f128(fp128 %Val)
6803 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
6808 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
6813 The argument and return value are floating point numbers of the same
6819 This function returns the same values as the libm ``exp2`` functions
6820 would, and handles error conditions in the same way.
6822 '``llvm.log.*``' Intrinsic
6823 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6828 This is an overloaded intrinsic. You can use ``llvm.log`` on any
6829 floating point or vector of floating point type. Not all targets support
6834 declare float @llvm.log.f32(float %Val)
6835 declare double @llvm.log.f64(double %Val)
6836 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6837 declare fp128 @llvm.log.f128(fp128 %Val)
6838 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6843 The '``llvm.log.*``' intrinsics perform the log function.
6848 The argument and return value are floating point numbers of the same
6854 This function returns the same values as the libm ``log`` functions
6855 would, and handles error conditions in the same way.
6857 '``llvm.log10.*``' Intrinsic
6858 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6863 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
6864 floating point or vector of floating point type. Not all targets support
6869 declare float @llvm.log10.f32(float %Val)
6870 declare double @llvm.log10.f64(double %Val)
6871 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
6872 declare fp128 @llvm.log10.f128(fp128 %Val)
6873 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
6878 The '``llvm.log10.*``' intrinsics perform the log10 function.
6883 The argument and return value are floating point numbers of the same
6889 This function returns the same values as the libm ``log10`` functions
6890 would, and handles error conditions in the same way.
6892 '``llvm.log2.*``' Intrinsic
6893 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6898 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
6899 floating point or vector of floating point type. Not all targets support
6904 declare float @llvm.log2.f32(float %Val)
6905 declare double @llvm.log2.f64(double %Val)
6906 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
6907 declare fp128 @llvm.log2.f128(fp128 %Val)
6908 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
6913 The '``llvm.log2.*``' intrinsics perform the log2 function.
6918 The argument and return value are floating point numbers of the same
6924 This function returns the same values as the libm ``log2`` functions
6925 would, and handles error conditions in the same way.
6927 '``llvm.fma.*``' Intrinsic
6928 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6933 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
6934 floating point or vector of floating point type. Not all targets support
6939 declare float @llvm.fma.f32(float %a, float %b, float %c)
6940 declare double @llvm.fma.f64(double %a, double %b, double %c)
6941 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
6942 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
6943 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
6948 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
6954 The argument and return value are floating point numbers of the same
6960 This function returns the same values as the libm ``fma`` functions
6963 '``llvm.fabs.*``' Intrinsic
6964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6969 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
6970 floating point or vector of floating point type. Not all targets support
6975 declare float @llvm.fabs.f32(float %Val)
6976 declare double @llvm.fabs.f64(double %Val)
6977 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
6978 declare fp128 @llvm.fabs.f128(fp128 %Val)
6979 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
6984 The '``llvm.fabs.*``' intrinsics return the absolute value of the
6990 The argument and return value are floating point numbers of the same
6996 This function returns the same values as the libm ``fabs`` functions
6997 would, and handles error conditions in the same way.
6999 '``llvm.floor.*``' Intrinsic
7000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7005 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7006 floating point or vector of floating point type. Not all targets support
7011 declare float @llvm.floor.f32(float %Val)
7012 declare double @llvm.floor.f64(double %Val)
7013 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7014 declare fp128 @llvm.floor.f128(fp128 %Val)
7015 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7020 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7025 The argument and return value are floating point numbers of the same
7031 This function returns the same values as the libm ``floor`` functions
7032 would, and handles error conditions in the same way.
7034 '``llvm.ceil.*``' Intrinsic
7035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7040 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7041 floating point or vector of floating point type. Not all targets support
7046 declare float @llvm.ceil.f32(float %Val)
7047 declare double @llvm.ceil.f64(double %Val)
7048 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7049 declare fp128 @llvm.ceil.f128(fp128 %Val)
7050 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7055 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7060 The argument and return value are floating point numbers of the same
7066 This function returns the same values as the libm ``ceil`` functions
7067 would, and handles error conditions in the same way.
7069 '``llvm.trunc.*``' Intrinsic
7070 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7075 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7076 floating point or vector of floating point type. Not all targets support
7081 declare float @llvm.trunc.f32(float %Val)
7082 declare double @llvm.trunc.f64(double %Val)
7083 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7084 declare fp128 @llvm.trunc.f128(fp128 %Val)
7085 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7090 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7091 nearest integer not larger in magnitude than the operand.
7096 The argument and return value are floating point numbers of the same
7102 This function returns the same values as the libm ``trunc`` functions
7103 would, and handles error conditions in the same way.
7105 '``llvm.rint.*``' Intrinsic
7106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7111 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7112 floating point or vector of floating point type. Not all targets support
7117 declare float @llvm.rint.f32(float %Val)
7118 declare double @llvm.rint.f64(double %Val)
7119 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7120 declare fp128 @llvm.rint.f128(fp128 %Val)
7121 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7126 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7127 nearest integer. It may raise an inexact floating-point exception if the
7128 operand isn't an integer.
7133 The argument and return value are floating point numbers of the same
7139 This function returns the same values as the libm ``rint`` functions
7140 would, and handles error conditions in the same way.
7142 '``llvm.nearbyint.*``' Intrinsic
7143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7148 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7149 floating point or vector of floating point type. Not all targets support
7154 declare float @llvm.nearbyint.f32(float %Val)
7155 declare double @llvm.nearbyint.f64(double %Val)
7156 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7157 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7158 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7163 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7169 The argument and return value are floating point numbers of the same
7175 This function returns the same values as the libm ``nearbyint``
7176 functions would, and handles error conditions in the same way.
7178 Bit Manipulation Intrinsics
7179 ---------------------------
7181 LLVM provides intrinsics for a few important bit manipulation
7182 operations. These allow efficient code generation for some algorithms.
7184 '``llvm.bswap.*``' Intrinsics
7185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7190 This is an overloaded intrinsic function. You can use bswap on any
7191 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7195 declare i16 @llvm.bswap.i16(i16 <id>)
7196 declare i32 @llvm.bswap.i32(i32 <id>)
7197 declare i64 @llvm.bswap.i64(i64 <id>)
7202 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7203 values with an even number of bytes (positive multiple of 16 bits).
7204 These are useful for performing operations on data that is not in the
7205 target's native byte order.
7210 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7211 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7212 intrinsic returns an i32 value that has the four bytes of the input i32
7213 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7214 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7215 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7216 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7219 '``llvm.ctpop.*``' Intrinsic
7220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7225 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7226 bit width, or on any vector with integer elements. Not all targets
7227 support all bit widths or vector types, however.
7231 declare i8 @llvm.ctpop.i8(i8 <src>)
7232 declare i16 @llvm.ctpop.i16(i16 <src>)
7233 declare i32 @llvm.ctpop.i32(i32 <src>)
7234 declare i64 @llvm.ctpop.i64(i64 <src>)
7235 declare i256 @llvm.ctpop.i256(i256 <src>)
7236 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7241 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7247 The only argument is the value to be counted. The argument may be of any
7248 integer type, or a vector with integer elements. The return type must
7249 match the argument type.
7254 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7255 each element of a vector.
7257 '``llvm.ctlz.*``' Intrinsic
7258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7263 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7264 integer bit width, or any vector whose elements are integers. Not all
7265 targets support all bit widths or vector types, however.
7269 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7270 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7271 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7272 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7273 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7274 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7279 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7280 leading zeros in a variable.
7285 The first argument is the value to be counted. This argument may be of
7286 any integer type, or a vectory with integer element type. The return
7287 type must match the first argument type.
7289 The second argument must be a constant and is a flag to indicate whether
7290 the intrinsic should ensure that a zero as the first argument produces a
7291 defined result. Historically some architectures did not provide a
7292 defined result for zero values as efficiently, and many algorithms are
7293 now predicated on avoiding zero-value inputs.
7298 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7299 zeros in a variable, or within each element of the vector. If
7300 ``src == 0`` then the result is the size in bits of the type of ``src``
7301 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7302 ``llvm.ctlz(i32 2) = 30``.
7304 '``llvm.cttz.*``' Intrinsic
7305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7310 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7311 integer bit width, or any vector of integer elements. Not all targets
7312 support all bit widths or vector types, however.
7316 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7317 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7318 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7319 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7320 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7321 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7326 The '``llvm.cttz``' family of intrinsic functions counts the number of
7332 The first argument is the value to be counted. This argument may be of
7333 any integer type, or a vectory with integer element type. The return
7334 type must match the first argument type.
7336 The second argument must be a constant and is a flag to indicate whether
7337 the intrinsic should ensure that a zero as the first argument produces a
7338 defined result. Historically some architectures did not provide a
7339 defined result for zero values as efficiently, and many algorithms are
7340 now predicated on avoiding zero-value inputs.
7345 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7346 zeros in a variable, or within each element of a vector. If ``src == 0``
7347 then the result is the size in bits of the type of ``src`` if
7348 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7349 ``llvm.cttz(2) = 1``.
7351 Arithmetic with Overflow Intrinsics
7352 -----------------------------------
7354 LLVM provides intrinsics for some arithmetic with overflow operations.
7356 '``llvm.sadd.with.overflow.*``' Intrinsics
7357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7362 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7363 on any integer bit width.
7367 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7368 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7369 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7374 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7375 a signed addition of the two arguments, and indicate whether an overflow
7376 occurred during the signed summation.
7381 The arguments (%a and %b) and the first element of the result structure
7382 may be of integer types of any bit width, but they must have the same
7383 bit width. The second element of the result structure must be of type
7384 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7390 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7391 a signed addition of the two variables. They return a structure — the
7392 first element of which is the signed summation, and the second element
7393 of which is a bit specifying if the signed summation resulted in an
7399 .. code-block:: llvm
7401 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7402 %sum = extractvalue {i32, i1} %res, 0
7403 %obit = extractvalue {i32, i1} %res, 1
7404 br i1 %obit, label %overflow, label %normal
7406 '``llvm.uadd.with.overflow.*``' Intrinsics
7407 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7412 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7413 on any integer bit width.
7417 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7418 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7419 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7424 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7425 an unsigned addition of the two arguments, and indicate whether a carry
7426 occurred during the unsigned summation.
7431 The arguments (%a and %b) and the first element of the result structure
7432 may be of integer types of any bit width, but they must have the same
7433 bit width. The second element of the result structure must be of type
7434 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7440 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7441 an unsigned addition of the two arguments. They return a structure — the
7442 first element of which is the sum, and the second element of which is a
7443 bit specifying if the unsigned summation resulted in a carry.
7448 .. code-block:: llvm
7450 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7451 %sum = extractvalue {i32, i1} %res, 0
7452 %obit = extractvalue {i32, i1} %res, 1
7453 br i1 %obit, label %carry, label %normal
7455 '``llvm.ssub.with.overflow.*``' Intrinsics
7456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7461 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7462 on any integer bit width.
7466 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7467 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7468 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7473 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7474 a signed subtraction of the two arguments, and indicate whether an
7475 overflow occurred during the signed subtraction.
7480 The arguments (%a and %b) and the first element of the result structure
7481 may be of integer types of any bit width, but they must have the same
7482 bit width. The second element of the result structure must be of type
7483 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7489 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7490 a signed subtraction of the two arguments. They return a structure — the
7491 first element of which is the subtraction, and the second element of
7492 which is a bit specifying if the signed subtraction resulted in an
7498 .. code-block:: llvm
7500 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7501 %sum = extractvalue {i32, i1} %res, 0
7502 %obit = extractvalue {i32, i1} %res, 1
7503 br i1 %obit, label %overflow, label %normal
7505 '``llvm.usub.with.overflow.*``' Intrinsics
7506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7511 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7512 on any integer bit width.
7516 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7517 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7518 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7523 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7524 an unsigned subtraction of the two arguments, and indicate whether an
7525 overflow occurred during the unsigned subtraction.
7530 The arguments (%a and %b) and the first element of the result structure
7531 may be of integer types of any bit width, but they must have the same
7532 bit width. The second element of the result structure must be of type
7533 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7539 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7540 an unsigned subtraction of the two arguments. They return a structure —
7541 the first element of which is the subtraction, and the second element of
7542 which is a bit specifying if the unsigned subtraction resulted in an
7548 .. code-block:: llvm
7550 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7551 %sum = extractvalue {i32, i1} %res, 0
7552 %obit = extractvalue {i32, i1} %res, 1
7553 br i1 %obit, label %overflow, label %normal
7555 '``llvm.smul.with.overflow.*``' Intrinsics
7556 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7561 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7562 on any integer bit width.
7566 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7567 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7568 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7573 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7574 a signed multiplication of the two arguments, and indicate whether an
7575 overflow occurred during the signed multiplication.
7580 The arguments (%a and %b) and the first element of the result structure
7581 may be of integer types of any bit width, but they must have the same
7582 bit width. The second element of the result structure must be of type
7583 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7589 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7590 a signed multiplication of the two arguments. They return a structure —
7591 the first element of which is the multiplication, and the second element
7592 of which is a bit specifying if the signed multiplication resulted in an
7598 .. code-block:: llvm
7600 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7601 %sum = extractvalue {i32, i1} %res, 0
7602 %obit = extractvalue {i32, i1} %res, 1
7603 br i1 %obit, label %overflow, label %normal
7605 '``llvm.umul.with.overflow.*``' Intrinsics
7606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7611 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7612 on any integer bit width.
7616 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7617 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7618 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7623 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7624 a unsigned multiplication of the two arguments, and indicate whether an
7625 overflow occurred during the unsigned multiplication.
7630 The arguments (%a and %b) and the first element of the result structure
7631 may be of integer types of any bit width, but they must have the same
7632 bit width. The second element of the result structure must be of type
7633 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7639 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7640 an unsigned multiplication of the two arguments. They return a structure
7641 — the first element of which is the multiplication, and the second
7642 element of which is a bit specifying if the unsigned multiplication
7643 resulted in an overflow.
7648 .. code-block:: llvm
7650 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7651 %sum = extractvalue {i32, i1} %res, 0
7652 %obit = extractvalue {i32, i1} %res, 1
7653 br i1 %obit, label %overflow, label %normal
7655 Specialised Arithmetic Intrinsics
7656 ---------------------------------
7658 '``llvm.fmuladd.*``' Intrinsic
7659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7666 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7667 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7672 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7673 expressions that can be fused if the code generator determines that the
7674 fused expression would be legal and efficient.
7679 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7680 multiplicands, a and b, and an addend c.
7689 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7691 is equivalent to the expression a \* b + c, except that rounding will
7692 not be performed between the multiplication and addition steps if the
7693 code generator fuses the operations. Fusion is not guaranteed, even if
7694 the target platform supports it. If a fused multiply-add is required the
7695 corresponding llvm.fma.\* intrinsic function should be used instead.
7700 .. code-block:: llvm
7702 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7704 Half Precision Floating Point Intrinsics
7705 ----------------------------------------
7707 For most target platforms, half precision floating point is a
7708 storage-only format. This means that it is a dense encoding (in memory)
7709 but does not support computation in the format.
7711 This means that code must first load the half-precision floating point
7712 value as an i16, then convert it to float with
7713 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7714 then be performed on the float value (including extending to double
7715 etc). To store the value back to memory, it is first converted to float
7716 if needed, then converted to i16 with
7717 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7720 .. _int_convert_to_fp16:
7722 '``llvm.convert.to.fp16``' Intrinsic
7723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7730 declare i16 @llvm.convert.to.fp16(f32 %a)
7735 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7736 from single precision floating point format to half precision floating
7742 The intrinsic function contains single argument - the value to be
7748 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7749 from single precision floating point format to half precision floating
7750 point format. The return value is an ``i16`` which contains the
7756 .. code-block:: llvm
7758 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7759 store i16 %res, i16* @x, align 2
7761 .. _int_convert_from_fp16:
7763 '``llvm.convert.from.fp16``' Intrinsic
7764 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7771 declare f32 @llvm.convert.from.fp16(i16 %a)
7776 The '``llvm.convert.from.fp16``' intrinsic function performs a
7777 conversion from half precision floating point format to single precision
7778 floating point format.
7783 The intrinsic function contains single argument - the value to be
7789 The '``llvm.convert.from.fp16``' intrinsic function performs a
7790 conversion from half single precision floating point format to single
7791 precision floating point format. The input half-float value is
7792 represented by an ``i16`` value.
7797 .. code-block:: llvm
7799 %a = load i16* @x, align 2
7800 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7805 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
7806 prefix), are described in the `LLVM Source Level
7807 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
7810 Exception Handling Intrinsics
7811 -----------------------------
7813 The LLVM exception handling intrinsics (which all start with
7814 ``llvm.eh.`` prefix), are described in the `LLVM Exception
7815 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
7819 Trampoline Intrinsics
7820 ---------------------
7822 These intrinsics make it possible to excise one parameter, marked with
7823 the :ref:`nest <nest>` attribute, from a function. The result is a
7824 callable function pointer lacking the nest parameter - the caller does
7825 not need to provide a value for it. Instead, the value to use is stored
7826 in advance in a "trampoline", a block of memory usually allocated on the
7827 stack, which also contains code to splice the nest value into the
7828 argument list. This is used to implement the GCC nested function address
7831 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
7832 then the resulting function pointer has signature ``i32 (i32, i32)*``.
7833 It can be created as follows:
7835 .. code-block:: llvm
7837 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7838 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7839 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7840 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7841 %fp = bitcast i8* %p to i32 (i32, i32)*
7843 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
7844 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
7848 '``llvm.init.trampoline``' Intrinsic
7849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7856 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7861 This fills the memory pointed to by ``tramp`` with executable code,
7862 turning it into a trampoline.
7867 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
7868 pointers. The ``tramp`` argument must point to a sufficiently large and
7869 sufficiently aligned block of memory; this memory is written to by the
7870 intrinsic. Note that the size and the alignment are target-specific -
7871 LLVM currently provides no portable way of determining them, so a
7872 front-end that generates this intrinsic needs to have some
7873 target-specific knowledge. The ``func`` argument must hold a function
7874 bitcast to an ``i8*``.
7879 The block of memory pointed to by ``tramp`` is filled with target
7880 dependent code, turning it into a function. Then ``tramp`` needs to be
7881 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
7882 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
7883 function's signature is the same as that of ``func`` with any arguments
7884 marked with the ``nest`` attribute removed. At most one such ``nest``
7885 argument is allowed, and it must be of pointer type. Calling the new
7886 function is equivalent to calling ``func`` with the same argument list,
7887 but with ``nval`` used for the missing ``nest`` argument. If, after
7888 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
7889 modified, then the effect of any later call to the returned function
7890 pointer is undefined.
7894 '``llvm.adjust.trampoline``' Intrinsic
7895 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7902 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7907 This performs any required machine-specific adjustment to the address of
7908 a trampoline (passed as ``tramp``).
7913 ``tramp`` must point to a block of memory which already has trampoline
7914 code filled in by a previous call to
7915 :ref:`llvm.init.trampoline <int_it>`.
7920 On some architectures the address of the code to be executed needs to be
7921 different to the address where the trampoline is actually stored. This
7922 intrinsic returns the executable address corresponding to ``tramp``
7923 after performing the required machine specific adjustments. The pointer
7924 returned can then be :ref:`bitcast and executed <int_trampoline>`.
7929 This class of intrinsics exists to information about the lifetime of
7930 memory objects and ranges where variables are immutable.
7932 '``llvm.lifetime.start``' Intrinsic
7933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7940 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7945 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
7951 The first argument is a constant integer representing the size of the
7952 object, or -1 if it is variable sized. The second argument is a pointer
7958 This intrinsic indicates that before this point in the code, the value
7959 of the memory pointed to by ``ptr`` is dead. This means that it is known
7960 to never be used and has an undefined value. A load from the pointer
7961 that precedes this intrinsic can be replaced with ``'undef'``.
7963 '``llvm.lifetime.end``' Intrinsic
7964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7971 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7976 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
7982 The first argument is a constant integer representing the size of the
7983 object, or -1 if it is variable sized. The second argument is a pointer
7989 This intrinsic indicates that after this point in the code, the value of
7990 the memory pointed to by ``ptr`` is dead. This means that it is known to
7991 never be used and has an undefined value. Any stores into the memory
7992 object following this intrinsic may be removed as dead.
7994 '``llvm.invariant.start``' Intrinsic
7995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8002 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8007 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8008 a memory object will not change.
8013 The first argument is a constant integer representing the size of the
8014 object, or -1 if it is variable sized. The second argument is a pointer
8020 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8021 the return value, the referenced memory location is constant and
8024 '``llvm.invariant.end``' Intrinsic
8025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8032 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8037 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8038 memory object are mutable.
8043 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8044 The second argument is a constant integer representing the size of the
8045 object, or -1 if it is variable sized and the third argument is a
8046 pointer to the object.
8051 This intrinsic indicates that the memory is mutable again.
8056 This class of intrinsics is designed to be generic and has no specific
8059 '``llvm.var.annotation``' Intrinsic
8060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8067 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8072 The '``llvm.var.annotation``' intrinsic.
8077 The first argument is a pointer to a value, the second is a pointer to a
8078 global string, the third is a pointer to a global string which is the
8079 source file name, and the last argument is the line number.
8084 This intrinsic allows annotation of local variables with arbitrary
8085 strings. This can be useful for special purpose optimizations that want
8086 to look for these annotations. These have no other defined use; they are
8087 ignored by code generation and optimization.
8089 '``llvm.annotation.*``' Intrinsic
8090 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8095 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8096 any integer bit width.
8100 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8101 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8102 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8103 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8104 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8109 The '``llvm.annotation``' intrinsic.
8114 The first argument is an integer value (result of some expression), the
8115 second is a pointer to a global string, the third is a pointer to a
8116 global string which is the source file name, and the last argument is
8117 the line number. It returns the value of the first argument.
8122 This intrinsic allows annotations to be put on arbitrary expressions
8123 with arbitrary strings. This can be useful for special purpose
8124 optimizations that want to look for these annotations. These have no
8125 other defined use; they are ignored by code generation and optimization.
8127 '``llvm.trap``' Intrinsic
8128 ^^^^^^^^^^^^^^^^^^^^^^^^^
8135 declare void @llvm.trap() noreturn nounwind
8140 The '``llvm.trap``' intrinsic.
8150 This intrinsic is lowered to the target dependent trap instruction. If
8151 the target does not have a trap instruction, this intrinsic will be
8152 lowered to a call of the ``abort()`` function.
8154 '``llvm.debugtrap``' Intrinsic
8155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8162 declare void @llvm.debugtrap() nounwind
8167 The '``llvm.debugtrap``' intrinsic.
8177 This intrinsic is lowered to code which is intended to cause an
8178 execution trap with the intention of requesting the attention of a
8181 '``llvm.stackprotector``' Intrinsic
8182 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8189 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8194 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8195 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8196 is placed on the stack before local variables.
8201 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8202 The first argument is the value loaded from the stack guard
8203 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8204 enough space to hold the value of the guard.
8209 This intrinsic causes the prologue/epilogue inserter to force the
8210 position of the ``AllocaInst`` stack slot to be before local variables
8211 on the stack. This is to ensure that if a local variable on the stack is
8212 overwritten, it will destroy the value of the guard. When the function
8213 exits, the guard on the stack is checked against the original guard. If
8214 they are different, then the program aborts by calling the
8215 ``__stack_chk_fail()`` function.
8217 '``llvm.objectsize``' Intrinsic
8218 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8225 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8226 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8231 The ``llvm.objectsize`` intrinsic is designed to provide information to
8232 the optimizers to determine at compile time whether a) an operation
8233 (like memcpy) will overflow a buffer that corresponds to an object, or
8234 b) that a runtime check for overflow isn't necessary. An object in this
8235 context means an allocation of a specific class, structure, array, or
8241 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8242 argument is a pointer to or into the ``object``. The second argument is
8243 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8244 or -1 (if false) when the object size is unknown. The second argument
8245 only accepts constants.
8250 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8251 the size of the object concerned. If the size cannot be determined at
8252 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8253 on the ``min`` argument).
8255 '``llvm.expect``' Intrinsic
8256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8263 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8264 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8269 The ``llvm.expect`` intrinsic provides information about expected (the
8270 most probable) value of ``val``, which can be used by optimizers.
8275 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8276 a value. The second argument is an expected value, this needs to be a
8277 constant value, variables are not allowed.
8282 This intrinsic is lowered to the ``val``.
8284 '``llvm.donothing``' Intrinsic
8285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8292 declare void @llvm.donothing() nounwind readnone
8297 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8298 only intrinsic that can be called with an invoke instruction.
8308 This intrinsic does nothing, and it's removed by optimizers and ignored