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
851 Module-Level Inline Assembly
852 ----------------------------
854 Modules may contain "module-level inline asm" blocks, which corresponds
855 to the GCC "file scope inline asm" blocks. These blocks are internally
856 concatenated by LLVM and treated as a single unit, but may be separated
857 in the ``.ll`` file if desired. The syntax is very simple:
861 module asm "inline asm code goes here"
862 module asm "more can go here"
864 The strings can contain any character by escaping non-printable
865 characters. The escape sequence used is simply "\\xx" where "xx" is the
866 two digit hex code for the number.
868 The inline asm code is simply printed to the machine code .s file when
869 assembly code is generated.
874 A module may specify a target specific data layout string that specifies
875 how data is to be laid out in memory. The syntax for the data layout is
880 target datalayout = "layout specification"
882 The *layout specification* consists of a list of specifications
883 separated by the minus sign character ('-'). Each specification starts
884 with a letter and may include other information after the letter to
885 define some aspect of the data layout. The specifications accepted are
889 Specifies that the target lays out data in big-endian form. That is,
890 the bits with the most significance have the lowest address
893 Specifies that the target lays out data in little-endian form. That
894 is, the bits with the least significance have the lowest address
897 Specifies the natural alignment of the stack in bits. Alignment
898 promotion of stack variables is limited to the natural stack
899 alignment to avoid dynamic stack realignment. The stack alignment
900 must be a multiple of 8-bits. If omitted, the natural stack
901 alignment defaults to "unspecified", which does not prevent any
902 alignment promotions.
903 ``p[n]:<size>:<abi>:<pref>``
904 This specifies the *size* of a pointer and its ``<abi>`` and
905 ``<pref>``\erred alignments for address space ``n``. All sizes are in
906 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
907 preceding ``:`` should be omitted too. The address space, ``n`` is
908 optional, and if not specified, denotes the default address space 0.
909 The value of ``n`` must be in the range [1,2^23).
910 ``i<size>:<abi>:<pref>``
911 This specifies the alignment for an integer type of a given bit
912 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
913 ``v<size>:<abi>:<pref>``
914 This specifies the alignment for a vector type of a given bit
916 ``f<size>:<abi>:<pref>``
917 This specifies the alignment for a floating point type of a given bit
918 ``<size>``. Only values of ``<size>`` that are supported by the target
919 will work. 32 (float) and 64 (double) are supported on all targets; 80
920 or 128 (different flavors of long double) are also supported on some
922 ``a<size>:<abi>:<pref>``
923 This specifies the alignment for an aggregate type of a given bit
925 ``s<size>:<abi>:<pref>``
926 This specifies the alignment for a stack object of a given bit
928 ``n<size1>:<size2>:<size3>...``
929 This specifies a set of native integer widths for the target CPU in
930 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
931 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
932 this set are considered to support most general arithmetic operations
935 When constructing the data layout for a given target, LLVM starts with a
936 default set of specifications which are then (possibly) overridden by
937 the specifications in the ``datalayout`` keyword. The default
938 specifications are given in this list:
941 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
942 - ``p1:32:32:32`` - 32-bit pointers with 32-bit alignment for address
944 - ``p2:16:32:32`` - 16-bit pointers with 32-bit alignment for address
946 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
947 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
948 - ``i16:16:16`` - i16 is 16-bit aligned
949 - ``i32:32:32`` - i32 is 32-bit aligned
950 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
952 - ``f32:32:32`` - float is 32-bit aligned
953 - ``f64:64:64`` - double is 64-bit aligned
954 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
955 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
956 - ``a0:0:1`` - aggregates are 8-bit aligned
957 - ``s0:64:64`` - stack objects are 64-bit aligned
959 When LLVM is determining the alignment for a given type, it uses the
962 #. If the type sought is an exact match for one of the specifications,
963 that specification is used.
964 #. If no match is found, and the type sought is an integer type, then
965 the smallest integer type that is larger than the bitwidth of the
966 sought type is used. If none of the specifications are larger than
967 the bitwidth then the largest integer type is used. For example,
968 given the default specifications above, the i7 type will use the
969 alignment of i8 (next largest) while both i65 and i256 will use the
970 alignment of i64 (largest specified).
971 #. If no match is found, and the type sought is a vector type, then the
972 largest vector type that is smaller than the sought vector type will
973 be used as a fall back. This happens because <128 x double> can be
974 implemented in terms of 64 <2 x double>, for example.
976 The function of the data layout string may not be what you expect.
977 Notably, this is not a specification from the frontend of what alignment
978 the code generator should use.
980 Instead, if specified, the target data layout is required to match what
981 the ultimate *code generator* expects. This string is used by the
982 mid-level optimizers to improve code, and this only works if it matches
983 what the ultimate code generator uses. If you would like to generate IR
984 that does not embed this target-specific detail into the IR, then you
985 don't have to specify the string. This will disable some optimizations
986 that require precise layout information, but this also prevents those
987 optimizations from introducing target specificity into the IR.
991 Pointer Aliasing Rules
992 ----------------------
994 Any memory access must be done through a pointer value associated with
995 an address range of the memory access, otherwise the behavior is
996 undefined. Pointer values are associated with address ranges according
997 to the following rules:
999 - A pointer value is associated with the addresses associated with any
1000 value it is *based* on.
1001 - An address of a global variable is associated with the address range
1002 of the variable's storage.
1003 - The result value of an allocation instruction is associated with the
1004 address range of the allocated storage.
1005 - A null pointer in the default address-space is associated with no
1007 - An integer constant other than zero or a pointer value returned from
1008 a function not defined within LLVM may be associated with address
1009 ranges allocated through mechanisms other than those provided by
1010 LLVM. Such ranges shall not overlap with any ranges of addresses
1011 allocated by mechanisms provided by LLVM.
1013 A pointer value is *based* on another pointer value according to the
1016 - A pointer value formed from a ``getelementptr`` operation is *based*
1017 on the first operand of the ``getelementptr``.
1018 - The result value of a ``bitcast`` is *based* on the operand of the
1020 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1021 values that contribute (directly or indirectly) to the computation of
1022 the pointer's value.
1023 - The "*based* on" relationship is transitive.
1025 Note that this definition of *"based"* is intentionally similar to the
1026 definition of *"based"* in C99, though it is slightly weaker.
1028 LLVM IR does not associate types with memory. The result type of a
1029 ``load`` merely indicates the size and alignment of the memory from
1030 which to load, as well as the interpretation of the value. The first
1031 operand type of a ``store`` similarly only indicates the size and
1032 alignment of the store.
1034 Consequently, type-based alias analysis, aka TBAA, aka
1035 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1036 :ref:`Metadata <metadata>` may be used to encode additional information
1037 which specialized optimization passes may use to implement type-based
1042 Volatile Memory Accesses
1043 ------------------------
1045 Certain memory accesses, such as :ref:`load <i_load>`'s,
1046 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1047 marked ``volatile``. The optimizers must not change the number of
1048 volatile operations or change their order of execution relative to other
1049 volatile operations. The optimizers *may* change the order of volatile
1050 operations relative to non-volatile operations. This is not Java's
1051 "volatile" and has no cross-thread synchronization behavior.
1055 Memory Model for Concurrent Operations
1056 --------------------------------------
1058 The LLVM IR does not define any way to start parallel threads of
1059 execution or to register signal handlers. Nonetheless, there are
1060 platform-specific ways to create them, and we define LLVM IR's behavior
1061 in their presence. This model is inspired by the C++0x memory model.
1063 For a more informal introduction to this model, see the :doc:`Atomics`.
1065 We define a *happens-before* partial order as the least partial order
1068 - Is a superset of single-thread program order, and
1069 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1070 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1071 techniques, like pthread locks, thread creation, thread joining,
1072 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1073 Constraints <ordering>`).
1075 Note that program order does not introduce *happens-before* edges
1076 between a thread and signals executing inside that thread.
1078 Every (defined) read operation (load instructions, memcpy, atomic
1079 loads/read-modify-writes, etc.) R reads a series of bytes written by
1080 (defined) write operations (store instructions, atomic
1081 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1082 section, initialized globals are considered to have a write of the
1083 initializer which is atomic and happens before any other read or write
1084 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1085 may see any write to the same byte, except:
1087 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1088 write\ :sub:`2` happens before R\ :sub:`byte`, then
1089 R\ :sub:`byte` does not see write\ :sub:`1`.
1090 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1091 R\ :sub:`byte` does not see write\ :sub:`3`.
1093 Given that definition, R\ :sub:`byte` is defined as follows:
1095 - If R is volatile, the result is target-dependent. (Volatile is
1096 supposed to give guarantees which can support ``sig_atomic_t`` in
1097 C/C++, and may be used for accesses to addresses which do not behave
1098 like normal memory. It does not generally provide cross-thread
1100 - Otherwise, if there is no write to the same byte that happens before
1101 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1102 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1103 R\ :sub:`byte` returns the value written by that write.
1104 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1105 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1106 Memory Ordering Constraints <ordering>` section for additional
1107 constraints on how the choice is made.
1108 - Otherwise R\ :sub:`byte` returns ``undef``.
1110 R returns the value composed of the series of bytes it read. This
1111 implies that some bytes within the value may be ``undef`` **without**
1112 the entire value being ``undef``. Note that this only defines the
1113 semantics of the operation; it doesn't mean that targets will emit more
1114 than one instruction to read the series of bytes.
1116 Note that in cases where none of the atomic intrinsics are used, this
1117 model places only one restriction on IR transformations on top of what
1118 is required for single-threaded execution: introducing a store to a byte
1119 which might not otherwise be stored is not allowed in general.
1120 (Specifically, in the case where another thread might write to and read
1121 from an address, introducing a store can change a load that may see
1122 exactly one write into a load that may see multiple writes.)
1126 Atomic Memory Ordering Constraints
1127 ----------------------------------
1129 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1130 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1131 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1132 an ordering parameter that determines which other atomic instructions on
1133 the same address they *synchronize with*. These semantics are borrowed
1134 from Java and C++0x, but are somewhat more colloquial. If these
1135 descriptions aren't precise enough, check those specs (see spec
1136 references in the :doc:`atomics guide <Atomics>`).
1137 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1138 differently since they don't take an address. See that instruction's
1139 documentation for details.
1141 For a simpler introduction to the ordering constraints, see the
1145 The set of values that can be read is governed by the happens-before
1146 partial order. A value cannot be read unless some operation wrote
1147 it. This is intended to provide a guarantee strong enough to model
1148 Java's non-volatile shared variables. This ordering cannot be
1149 specified for read-modify-write operations; it is not strong enough
1150 to make them atomic in any interesting way.
1152 In addition to the guarantees of ``unordered``, there is a single
1153 total order for modifications by ``monotonic`` operations on each
1154 address. All modification orders must be compatible with the
1155 happens-before order. There is no guarantee that the modification
1156 orders can be combined to a global total order for the whole program
1157 (and this often will not be possible). The read in an atomic
1158 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1159 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1160 order immediately before the value it writes. If one atomic read
1161 happens before another atomic read of the same address, the later
1162 read must see the same value or a later value in the address's
1163 modification order. This disallows reordering of ``monotonic`` (or
1164 stronger) operations on the same address. If an address is written
1165 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1166 read that address repeatedly, the other threads must eventually see
1167 the write. This corresponds to the C++0x/C1x
1168 ``memory_order_relaxed``.
1170 In addition to the guarantees of ``monotonic``, a
1171 *synchronizes-with* edge may be formed with a ``release`` operation.
1172 This is intended to model C++'s ``memory_order_acquire``.
1174 In addition to the guarantees of ``monotonic``, if this operation
1175 writes a value which is subsequently read by an ``acquire``
1176 operation, it *synchronizes-with* that operation. (This isn't a
1177 complete description; see the C++0x definition of a release
1178 sequence.) This corresponds to the C++0x/C1x
1179 ``memory_order_release``.
1180 ``acq_rel`` (acquire+release)
1181 Acts as both an ``acquire`` and ``release`` operation on its
1182 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1183 ``seq_cst`` (sequentially consistent)
1184 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1185 operation which only reads, ``release`` for an operation which only
1186 writes), there is a global total order on all
1187 sequentially-consistent operations on all addresses, which is
1188 consistent with the *happens-before* partial order and with the
1189 modification orders of all the affected addresses. Each
1190 sequentially-consistent read sees the last preceding write to the
1191 same address in this global order. This corresponds to the C++0x/C1x
1192 ``memory_order_seq_cst`` and Java volatile.
1196 If an atomic operation is marked ``singlethread``, it only *synchronizes
1197 with* or participates in modification and seq\_cst total orderings with
1198 other operations running in the same thread (for example, in signal
1206 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1207 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1208 :ref:`frem <i_frem>`) have the following flags that can set to enable
1209 otherwise unsafe floating point operations
1212 No NaNs - Allow optimizations to assume the arguments and result are not
1213 NaN. Such optimizations are required to retain defined behavior over
1214 NaNs, but the value of the result is undefined.
1217 No Infs - Allow optimizations to assume the arguments and result are not
1218 +/-Inf. Such optimizations are required to retain defined behavior over
1219 +/-Inf, but the value of the result is undefined.
1222 No Signed Zeros - Allow optimizations to treat the sign of a zero
1223 argument or result as insignificant.
1226 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1227 argument rather than perform division.
1230 Fast - Allow algebraically equivalent transformations that may
1231 dramatically change results in floating point (e.g. reassociate). This
1232 flag implies all the others.
1239 The LLVM type system is one of the most important features of the
1240 intermediate representation. Being typed enables a number of
1241 optimizations to be performed on the intermediate representation
1242 directly, without having to do extra analyses on the side before the
1243 transformation. A strong type system makes it easier to read the
1244 generated code and enables novel analyses and transformations that are
1245 not feasible to perform on normal three address code representations.
1247 Type Classifications
1248 --------------------
1250 The types fall into a few useful classifications:
1259 * - :ref:`integer <t_integer>`
1260 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1263 * - :ref:`floating point <t_floating>`
1264 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1272 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1273 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1274 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1275 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1277 * - :ref:`primitive <t_primitive>`
1278 - :ref:`label <t_label>`,
1279 :ref:`void <t_void>`,
1280 :ref:`integer <t_integer>`,
1281 :ref:`floating point <t_floating>`,
1282 :ref:`x86mmx <t_x86mmx>`,
1283 :ref:`metadata <t_metadata>`.
1285 * - :ref:`derived <t_derived>`
1286 - :ref:`array <t_array>`,
1287 :ref:`function <t_function>`,
1288 :ref:`pointer <t_pointer>`,
1289 :ref:`structure <t_struct>`,
1290 :ref:`vector <t_vector>`,
1291 :ref:`opaque <t_opaque>`.
1293 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1294 Values of these types are the only ones which can be produced by
1302 The primitive types are the fundamental building blocks of the LLVM
1313 The integer type is a very simple type that simply specifies an
1314 arbitrary bit width for the integer type desired. Any bit width from 1
1315 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1324 The number of bits the integer will occupy is specified by the ``N``
1330 +----------------+------------------------------------------------+
1331 | ``i1`` | a single-bit integer. |
1332 +----------------+------------------------------------------------+
1333 | ``i32`` | a 32-bit integer. |
1334 +----------------+------------------------------------------------+
1335 | ``i1942652`` | a really big integer of over 1 million bits. |
1336 +----------------+------------------------------------------------+
1340 Floating Point Types
1341 ^^^^^^^^^^^^^^^^^^^^
1350 - 16-bit floating point value
1353 - 32-bit floating point value
1356 - 64-bit floating point value
1359 - 128-bit floating point value (112-bit mantissa)
1362 - 80-bit floating point value (X87)
1365 - 128-bit floating point value (two 64-bits)
1375 The x86mmx type represents a value held in an MMX register on an x86
1376 machine. The operations allowed on it are quite limited: parameters and
1377 return values, load and store, and bitcast. User-specified MMX
1378 instructions are represented as intrinsic or asm calls with arguments
1379 and/or results of this type. There are no arrays, vectors or constants
1397 The void type does not represent any value and has no size.
1414 The label type represents code labels.
1431 The metadata type represents embedded metadata. No derived types may be
1432 created from metadata except for :ref:`function <t_function>` arguments.
1446 The real power in LLVM comes from the derived types in the system. This
1447 is what allows a programmer to represent arrays, functions, pointers,
1448 and other useful types. Each of these types contain one or more element
1449 types which may be a primitive type, or another derived type. For
1450 example, it is possible to have a two dimensional array, using an array
1451 as the element type of another array.
1458 Aggregate Types are a subset of derived types that can contain multiple
1459 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1460 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1471 The array type is a very simple derived type that arranges elements
1472 sequentially in memory. The array type requires a size (number of
1473 elements) and an underlying data type.
1480 [<# elements> x <elementtype>]
1482 The number of elements is a constant integer value; ``elementtype`` may
1483 be any type with a size.
1488 +------------------+--------------------------------------+
1489 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1490 +------------------+--------------------------------------+
1491 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1492 +------------------+--------------------------------------+
1493 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1494 +------------------+--------------------------------------+
1496 Here are some examples of multidimensional arrays:
1498 +-----------------------------+----------------------------------------------------------+
1499 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1500 +-----------------------------+----------------------------------------------------------+
1501 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1502 +-----------------------------+----------------------------------------------------------+
1503 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1504 +-----------------------------+----------------------------------------------------------+
1506 There is no restriction on indexing beyond the end of the array implied
1507 by a static type (though there are restrictions on indexing beyond the
1508 bounds of an allocated object in some cases). This means that
1509 single-dimension 'variable sized array' addressing can be implemented in
1510 LLVM with a zero length array type. An implementation of 'pascal style
1511 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1522 The function type can be thought of as a function signature. It consists
1523 of a return type and a list of formal parameter types. The return type
1524 of a function type is a first class type or a void type.
1531 <returntype> (<parameter list>)
1533 ...where '``<parameter list>``' is a comma-separated list of type
1534 specifiers. Optionally, the parameter list may include a type ``...``,
1535 which indicates that the function takes a variable number of arguments.
1536 Variable argument functions can access their arguments with the
1537 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1538 '``<returntype>``' is any type except :ref:`label <t_label>`.
1543 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1544 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1545 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1546 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1547 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1548 | ``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. |
1549 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1550 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1551 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1561 The structure type is used to represent a collection of data members
1562 together in memory. The elements of a structure may be any type that has
1565 Structures in memory are accessed using '``load``' and '``store``' by
1566 getting a pointer to a field with the '``getelementptr``' instruction.
1567 Structures in registers are accessed using the '``extractvalue``' and
1568 '``insertvalue``' instructions.
1570 Structures may optionally be "packed" structures, which indicate that
1571 the alignment of the struct is one byte, and that there is no padding
1572 between the elements. In non-packed structs, padding between field types
1573 is inserted as defined by the DataLayout string in the module, which is
1574 required to match what the underlying code generator expects.
1576 Structures can either be "literal" or "identified". A literal structure
1577 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1578 identified types are always defined at the top level with a name.
1579 Literal types are uniqued by their contents and can never be recursive
1580 or opaque since there is no way to write one. Identified types can be
1581 recursive, can be opaqued, and are never uniqued.
1588 %T1 = type { <type list> } ; Identified normal struct type
1589 %T2 = type <{ <type list> }> ; Identified packed struct type
1594 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1595 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1596 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1597 | ``{ 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``. |
1598 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1599 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1600 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1604 Opaque Structure Types
1605 ^^^^^^^^^^^^^^^^^^^^^^
1610 Opaque structure types are used to represent named structure types that
1611 do not have a body specified. This corresponds (for example) to the C
1612 notion of a forward declared structure.
1625 +--------------+-------------------+
1626 | ``opaque`` | An opaque type. |
1627 +--------------+-------------------+
1637 The pointer type is used to specify memory locations. Pointers are
1638 commonly used to reference objects in memory.
1640 Pointer types may have an optional address space attribute defining the
1641 numbered address space where the pointed-to object resides. The default
1642 address space is number zero. The semantics of non-zero address spaces
1643 are target-specific.
1645 Note that LLVM does not permit pointers to void (``void*``) nor does it
1646 permit pointers to labels (``label*``). Use ``i8*`` instead.
1658 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1659 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1660 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1661 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1662 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1663 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1664 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1674 A vector type is a simple derived type that represents a vector of
1675 elements. Vector types are used when multiple primitive data are
1676 operated in parallel using a single instruction (SIMD). A vector type
1677 requires a size (number of elements) and an underlying primitive data
1678 type. Vector types are considered :ref:`first class <t_firstclass>`.
1685 < <# elements> x <elementtype> >
1687 The number of elements is a constant integer value larger than 0;
1688 elementtype may be any integer or floating point type, or a pointer to
1689 these types. Vectors of size zero are not allowed.
1694 +-------------------+--------------------------------------------------+
1695 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1696 +-------------------+--------------------------------------------------+
1697 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1698 +-------------------+--------------------------------------------------+
1699 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1700 +-------------------+--------------------------------------------------+
1701 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1702 +-------------------+--------------------------------------------------+
1707 LLVM has several different basic types of constants. This section
1708 describes them all and their syntax.
1713 **Boolean constants**
1714 The two strings '``true``' and '``false``' are both valid constants
1716 **Integer constants**
1717 Standard integers (such as '4') are constants of the
1718 :ref:`integer <t_integer>` type. Negative numbers may be used with
1720 **Floating point constants**
1721 Floating point constants use standard decimal notation (e.g.
1722 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1723 hexadecimal notation (see below). The assembler requires the exact
1724 decimal value of a floating-point constant. For example, the
1725 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1726 decimal in binary. Floating point constants must have a :ref:`floating
1727 point <t_floating>` type.
1728 **Null pointer constants**
1729 The identifier '``null``' is recognized as a null pointer constant
1730 and must be of :ref:`pointer type <t_pointer>`.
1732 The one non-intuitive notation for constants is the hexadecimal form of
1733 floating point constants. For example, the form
1734 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1735 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1736 constants are required (and the only time that they are generated by the
1737 disassembler) is when a floating point constant must be emitted but it
1738 cannot be represented as a decimal floating point number in a reasonable
1739 number of digits. For example, NaN's, infinities, and other special
1740 values are represented in their IEEE hexadecimal format so that assembly
1741 and disassembly do not cause any bits to change in the constants.
1743 When using the hexadecimal form, constants of types half, float, and
1744 double are represented using the 16-digit form shown above (which
1745 matches the IEEE754 representation for double); half and float values
1746 must, however, be exactly representable as IEE754 half and single
1747 precision, respectively. Hexadecimal format is always used for long
1748 double, and there are three forms of long double. The 80-bit format used
1749 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1750 128-bit format used by PowerPC (two adjacent doubles) is represented by
1751 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1752 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1753 supported target uses this format. Long doubles will only work if they
1754 match the long double format on your target. The IEEE 16-bit format
1755 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1756 digits. All hexadecimal formats are big-endian (sign bit at the left).
1758 There are no constants of type x86mmx.
1763 Complex constants are a (potentially recursive) combination of simple
1764 constants and smaller complex constants.
1766 **Structure constants**
1767 Structure constants are represented with notation similar to
1768 structure type definitions (a comma separated list of elements,
1769 surrounded by braces (``{}``)). For example:
1770 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1771 "``@G = external global i32``". Structure constants must have
1772 :ref:`structure type <t_struct>`, and the number and types of elements
1773 must match those specified by the type.
1775 Array constants are represented with notation similar to array type
1776 definitions (a comma separated list of elements, surrounded by
1777 square brackets (``[]``)). For example:
1778 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1779 :ref:`array type <t_array>`, and the number and types of elements must
1780 match those specified by the type.
1781 **Vector constants**
1782 Vector constants are represented with notation similar to vector
1783 type definitions (a comma separated list of elements, surrounded by
1784 less-than/greater-than's (``<>``)). For example:
1785 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1786 must have :ref:`vector type <t_vector>`, and the number and types of
1787 elements must match those specified by the type.
1788 **Zero initialization**
1789 The string '``zeroinitializer``' can be used to zero initialize a
1790 value to zero of *any* type, including scalar and
1791 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1792 having to print large zero initializers (e.g. for large arrays) and
1793 is always exactly equivalent to using explicit zero initializers.
1795 A metadata node is a structure-like constant with :ref:`metadata
1796 type <t_metadata>`. For example:
1797 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1798 constants that are meant to be interpreted as part of the
1799 instruction stream, metadata is a place to attach additional
1800 information such as debug info.
1802 Global Variable and Function Addresses
1803 --------------------------------------
1805 The addresses of :ref:`global variables <globalvars>` and
1806 :ref:`functions <functionstructure>` are always implicitly valid
1807 (link-time) constants. These constants are explicitly referenced when
1808 the :ref:`identifier for the global <identifiers>` is used and always have
1809 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1812 .. code-block:: llvm
1816 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1823 The string '``undef``' can be used anywhere a constant is expected, and
1824 indicates that the user of the value may receive an unspecified
1825 bit-pattern. Undefined values may be of any type (other than '``label``'
1826 or '``void``') and be used anywhere a constant is permitted.
1828 Undefined values are useful because they indicate to the compiler that
1829 the program is well defined no matter what value is used. This gives the
1830 compiler more freedom to optimize. Here are some examples of
1831 (potentially surprising) transformations that are valid (in pseudo IR):
1833 .. code-block:: llvm
1843 This is safe because all of the output bits are affected by the undef
1844 bits. Any output bit can have a zero or one depending on the input bits.
1846 .. code-block:: llvm
1857 These logical operations have bits that are not always affected by the
1858 input. For example, if ``%X`` has a zero bit, then the output of the
1859 '``and``' operation will always be a zero for that bit, no matter what
1860 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1861 optimize or assume that the result of the '``and``' is '``undef``'.
1862 However, it is safe to assume that all bits of the '``undef``' could be
1863 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1864 all the bits of the '``undef``' operand to the '``or``' could be set,
1865 allowing the '``or``' to be folded to -1.
1867 .. code-block:: llvm
1869 %A = select undef, %X, %Y
1870 %B = select undef, 42, %Y
1871 %C = select %X, %Y, undef
1881 This set of examples shows that undefined '``select``' (and conditional
1882 branch) conditions can go *either way*, but they have to come from one
1883 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1884 both known to have a clear low bit, then ``%A`` would have to have a
1885 cleared low bit. However, in the ``%C`` example, the optimizer is
1886 allowed to assume that the '``undef``' operand could be the same as
1887 ``%Y``, allowing the whole '``select``' to be eliminated.
1889 .. code-block:: llvm
1891 %A = xor undef, undef
1908 This example points out that two '``undef``' operands are not
1909 necessarily the same. This can be surprising to people (and also matches
1910 C semantics) where they assume that "``X^X``" is always zero, even if
1911 ``X`` is undefined. This isn't true for a number of reasons, but the
1912 short answer is that an '``undef``' "variable" can arbitrarily change
1913 its value over its "live range". This is true because the variable
1914 doesn't actually *have a live range*. Instead, the value is logically
1915 read from arbitrary registers that happen to be around when needed, so
1916 the value is not necessarily consistent over time. In fact, ``%A`` and
1917 ``%C`` need to have the same semantics or the core LLVM "replace all
1918 uses with" concept would not hold.
1920 .. code-block:: llvm
1928 These examples show the crucial difference between an *undefined value*
1929 and *undefined behavior*. An undefined value (like '``undef``') is
1930 allowed to have an arbitrary bit-pattern. This means that the ``%A``
1931 operation can be constant folded to '``undef``', because the '``undef``'
1932 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
1933 However, in the second example, we can make a more aggressive
1934 assumption: because the ``undef`` is allowed to be an arbitrary value,
1935 we are allowed to assume that it could be zero. Since a divide by zero
1936 has *undefined behavior*, we are allowed to assume that the operation
1937 does not execute at all. This allows us to delete the divide and all
1938 code after it. Because the undefined operation "can't happen", the
1939 optimizer can assume that it occurs in dead code.
1941 .. code-block:: llvm
1943 a: store undef -> %X
1944 b: store %X -> undef
1949 These examples reiterate the ``fdiv`` example: a store *of* an undefined
1950 value can be assumed to not have any effect; we can assume that the
1951 value is overwritten with bits that happen to match what was already
1952 there. However, a store *to* an undefined location could clobber
1953 arbitrary memory, therefore, it has undefined behavior.
1960 Poison values are similar to :ref:`undef values <undefvalues>`, however
1961 they also represent the fact that an instruction or constant expression
1962 which cannot evoke side effects has nevertheless detected a condition
1963 which results in undefined behavior.
1965 There is currently no way of representing a poison value in the IR; they
1966 only exist when produced by operations such as :ref:`add <i_add>` with
1969 Poison value behavior is defined in terms of value *dependence*:
1971 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
1972 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
1973 their dynamic predecessor basic block.
1974 - Function arguments depend on the corresponding actual argument values
1975 in the dynamic callers of their functions.
1976 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
1977 instructions that dynamically transfer control back to them.
1978 - :ref:`Invoke <i_invoke>` instructions depend on the
1979 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
1980 call instructions that dynamically transfer control back to them.
1981 - Non-volatile loads and stores depend on the most recent stores to all
1982 of the referenced memory addresses, following the order in the IR
1983 (including loads and stores implied by intrinsics such as
1984 :ref:`@llvm.memcpy <int_memcpy>`.)
1985 - An instruction with externally visible side effects depends on the
1986 most recent preceding instruction with externally visible side
1987 effects, following the order in the IR. (This includes :ref:`volatile
1988 operations <volatile>`.)
1989 - An instruction *control-depends* on a :ref:`terminator
1990 instruction <terminators>` if the terminator instruction has
1991 multiple successors and the instruction is always executed when
1992 control transfers to one of the successors, and may not be executed
1993 when control is transferred to another.
1994 - Additionally, an instruction also *control-depends* on a terminator
1995 instruction if the set of instructions it otherwise depends on would
1996 be different if the terminator had transferred control to a different
1998 - Dependence is transitive.
2000 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2001 with the additional affect that any instruction which has a *dependence*
2002 on a poison value has undefined behavior.
2004 Here are some examples:
2006 .. code-block:: llvm
2009 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2010 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2011 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2012 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2014 store i32 %poison, i32* @g ; Poison value stored to memory.
2015 %poison2 = load i32* @g ; Poison value loaded back from memory.
2017 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2019 %narrowaddr = bitcast i32* @g to i16*
2020 %wideaddr = bitcast i32* @g to i64*
2021 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2022 %poison4 = load i64* %wideaddr ; Returns a poison value.
2024 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2025 br i1 %cmp, label %true, label %end ; Branch to either destination.
2028 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2029 ; it has undefined behavior.
2033 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2034 ; Both edges into this PHI are
2035 ; control-dependent on %cmp, so this
2036 ; always results in a poison value.
2038 store volatile i32 0, i32* @g ; This would depend on the store in %true
2039 ; if %cmp is true, or the store in %entry
2040 ; otherwise, so this is undefined behavior.
2042 br i1 %cmp, label %second_true, label %second_end
2043 ; The same branch again, but this time the
2044 ; true block doesn't have side effects.
2051 store volatile i32 0, i32* @g ; This time, the instruction always depends
2052 ; on the store in %end. Also, it is
2053 ; control-equivalent to %end, so this is
2054 ; well-defined (ignoring earlier undefined
2055 ; behavior in this example).
2059 Addresses of Basic Blocks
2060 -------------------------
2062 ``blockaddress(@function, %block)``
2064 The '``blockaddress``' constant computes the address of the specified
2065 basic block in the specified function, and always has an ``i8*`` type.
2066 Taking the address of the entry block is illegal.
2068 This value only has defined behavior when used as an operand to the
2069 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2070 against null. Pointer equality tests between labels addresses results in
2071 undefined behavior — though, again, comparison against null is ok, and
2072 no label is equal to the null pointer. This may be passed around as an
2073 opaque pointer sized value as long as the bits are not inspected. This
2074 allows ``ptrtoint`` and arithmetic to be performed on these values so
2075 long as the original value is reconstituted before the ``indirectbr``
2078 Finally, some targets may provide defined semantics when using the value
2079 as the operand to an inline assembly, but that is target specific.
2081 Constant Expressions
2082 --------------------
2084 Constant expressions are used to allow expressions involving other
2085 constants to be used as constants. Constant expressions may be of any
2086 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2087 that does not have side effects (e.g. load and call are not supported).
2088 The following is the syntax for constant expressions:
2090 ``trunc (CST to TYPE)``
2091 Truncate a constant to another type. The bit size of CST must be
2092 larger than the bit size of TYPE. Both types must be integers.
2093 ``zext (CST to TYPE)``
2094 Zero extend a constant to another type. The bit size of CST must be
2095 smaller than the bit size of TYPE. Both types must be integers.
2096 ``sext (CST to TYPE)``
2097 Sign extend a constant to another type. The bit size of CST must be
2098 smaller than the bit size of TYPE. Both types must be integers.
2099 ``fptrunc (CST to TYPE)``
2100 Truncate a floating point constant to another floating point type.
2101 The size of CST must be larger than the size of TYPE. Both types
2102 must be floating point.
2103 ``fpext (CST to TYPE)``
2104 Floating point extend a constant to another type. The size of CST
2105 must be smaller or equal to the size of TYPE. Both types must be
2107 ``fptoui (CST to TYPE)``
2108 Convert a floating point constant to the corresponding unsigned
2109 integer constant. TYPE must be a scalar or vector integer type. CST
2110 must be of scalar or vector floating point type. Both CST and TYPE
2111 must be scalars, or vectors of the same number of elements. If the
2112 value won't fit in the integer type, the results are undefined.
2113 ``fptosi (CST to TYPE)``
2114 Convert a floating point constant to the corresponding signed
2115 integer constant. TYPE must be a scalar or vector integer type. CST
2116 must be of scalar or vector floating point type. Both CST and TYPE
2117 must be scalars, or vectors of the same number of elements. If the
2118 value won't fit in the integer type, the results are undefined.
2119 ``uitofp (CST to TYPE)``
2120 Convert an unsigned integer constant to the corresponding floating
2121 point constant. TYPE must be a scalar or vector floating point type.
2122 CST must be of scalar or vector integer type. Both CST and TYPE must
2123 be scalars, or vectors of the same number of elements. If the value
2124 won't fit in the floating point type, the results are undefined.
2125 ``sitofp (CST to TYPE)``
2126 Convert a signed integer constant to the corresponding floating
2127 point constant. TYPE must be a scalar or vector floating point type.
2128 CST must be of scalar or vector integer type. Both CST and TYPE must
2129 be scalars, or vectors of the same number of elements. If the value
2130 won't fit in the floating point type, the results are undefined.
2131 ``ptrtoint (CST to TYPE)``
2132 Convert a pointer typed constant to the corresponding integer
2133 constant ``TYPE`` must be an integer type. ``CST`` must be of
2134 pointer type. The ``CST`` value is zero extended, truncated, or
2135 unchanged to make it fit in ``TYPE``.
2136 ``inttoptr (CST to TYPE)``
2137 Convert an integer constant to a pointer constant. TYPE must be a
2138 pointer type. CST must be of integer type. The CST value is zero
2139 extended, truncated, or unchanged to make it fit in a pointer size.
2140 This one is *really* dangerous!
2141 ``bitcast (CST to TYPE)``
2142 Convert a constant, CST, to another TYPE. The constraints of the
2143 operands are the same as those for the :ref:`bitcast
2144 instruction <i_bitcast>`.
2145 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2146 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2147 constants. As with the :ref:`getelementptr <i_getelementptr>`
2148 instruction, the index list may have zero or more indexes, which are
2149 required to make sense for the type of "CSTPTR".
2150 ``select (COND, VAL1, VAL2)``
2151 Perform the :ref:`select operation <i_select>` on constants.
2152 ``icmp COND (VAL1, VAL2)``
2153 Performs the :ref:`icmp operation <i_icmp>` on constants.
2154 ``fcmp COND (VAL1, VAL2)``
2155 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2156 ``extractelement (VAL, IDX)``
2157 Perform the :ref:`extractelement operation <i_extractelement>` on
2159 ``insertelement (VAL, ELT, IDX)``
2160 Perform the :ref:`insertelement operation <i_insertelement>` on
2162 ``shufflevector (VEC1, VEC2, IDXMASK)``
2163 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2165 ``extractvalue (VAL, IDX0, IDX1, ...)``
2166 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2167 constants. The index list is interpreted in a similar manner as
2168 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2169 least one index value must be specified.
2170 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2171 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2172 The index list is interpreted in a similar manner as indices in a
2173 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2174 value must be specified.
2175 ``OPCODE (LHS, RHS)``
2176 Perform the specified operation of the LHS and RHS constants. OPCODE
2177 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2178 binary <bitwiseops>` operations. The constraints on operands are
2179 the same as those for the corresponding instruction (e.g. no bitwise
2180 operations on floating point values are allowed).
2185 Inline Assembler Expressions
2186 ----------------------------
2188 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2189 Inline Assembly <moduleasm>`) through the use of a special value. This
2190 value represents the inline assembler as a string (containing the
2191 instructions to emit), a list of operand constraints (stored as a
2192 string), a flag that indicates whether or not the inline asm expression
2193 has side effects, and a flag indicating whether the function containing
2194 the asm needs to align its stack conservatively. An example inline
2195 assembler expression is:
2197 .. code-block:: llvm
2199 i32 (i32) asm "bswap $0", "=r,r"
2201 Inline assembler expressions may **only** be used as the callee operand
2202 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2203 Thus, typically we have:
2205 .. code-block:: llvm
2207 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2209 Inline asms with side effects not visible in the constraint list must be
2210 marked as having side effects. This is done through the use of the
2211 '``sideeffect``' keyword, like so:
2213 .. code-block:: llvm
2215 call void asm sideeffect "eieio", ""()
2217 In some cases inline asms will contain code that will not work unless
2218 the stack is aligned in some way, such as calls or SSE instructions on
2219 x86, yet will not contain code that does that alignment within the asm.
2220 The compiler should make conservative assumptions about what the asm
2221 might contain and should generate its usual stack alignment code in the
2222 prologue if the '``alignstack``' keyword is present:
2224 .. code-block:: llvm
2226 call void asm alignstack "eieio", ""()
2228 Inline asms also support using non-standard assembly dialects. The
2229 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2230 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2231 the only supported dialects. An example is:
2233 .. code-block:: llvm
2235 call void asm inteldialect "eieio", ""()
2237 If multiple keywords appear the '``sideeffect``' keyword must come
2238 first, the '``alignstack``' keyword second and the '``inteldialect``'
2244 The call instructions that wrap inline asm nodes may have a
2245 "``!srcloc``" MDNode attached to it that contains a list of constant
2246 integers. If present, the code generator will use the integer as the
2247 location cookie value when report errors through the ``LLVMContext``
2248 error reporting mechanisms. This allows a front-end to correlate backend
2249 errors that occur with inline asm back to the source code that produced
2252 .. code-block:: llvm
2254 call void asm sideeffect "something bad", ""(), !srcloc !42
2256 !42 = !{ i32 1234567 }
2258 It is up to the front-end to make sense of the magic numbers it places
2259 in the IR. If the MDNode contains multiple constants, the code generator
2260 will use the one that corresponds to the line of the asm that the error
2265 Metadata Nodes and Metadata Strings
2266 -----------------------------------
2268 LLVM IR allows metadata to be attached to instructions in the program
2269 that can convey extra information about the code to the optimizers and
2270 code generator. One example application of metadata is source-level
2271 debug information. There are two metadata primitives: strings and nodes.
2272 All metadata has the ``metadata`` type and is identified in syntax by a
2273 preceding exclamation point ('``!``').
2275 A metadata string is a string surrounded by double quotes. It can
2276 contain any character by escaping non-printable characters with
2277 "``\xx``" where "``xx``" is the two digit hex code. For example:
2280 Metadata nodes are represented with notation similar to structure
2281 constants (a comma separated list of elements, surrounded by braces and
2282 preceded by an exclamation point). Metadata nodes can have any values as
2283 their operand. For example:
2285 .. code-block:: llvm
2287 !{ metadata !"test\00", i32 10}
2289 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2290 metadata nodes, which can be looked up in the module symbol table. For
2293 .. code-block:: llvm
2295 !foo = metadata !{!4, !3}
2297 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2298 function is using two metadata arguments:
2300 .. code-block:: llvm
2302 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2304 Metadata can be attached with an instruction. Here metadata ``!21`` is
2305 attached to the ``add`` instruction using the ``!dbg`` identifier:
2307 .. code-block:: llvm
2309 %indvar.next = add i64 %indvar, 1, !dbg !21
2311 More information about specific metadata nodes recognized by the
2312 optimizers and code generator is found below.
2317 In LLVM IR, memory does not have types, so LLVM's own type system is not
2318 suitable for doing TBAA. Instead, metadata is added to the IR to
2319 describe a type system of a higher level language. This can be used to
2320 implement typical C/C++ TBAA, but it can also be used to implement
2321 custom alias analysis behavior for other languages.
2323 The current metadata format is very simple. TBAA metadata nodes have up
2324 to three fields, e.g.:
2326 .. code-block:: llvm
2328 !0 = metadata !{ metadata !"an example type tree" }
2329 !1 = metadata !{ metadata !"int", metadata !0 }
2330 !2 = metadata !{ metadata !"float", metadata !0 }
2331 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2333 The first field is an identity field. It can be any value, usually a
2334 metadata string, which uniquely identifies the type. The most important
2335 name in the tree is the name of the root node. Two trees with different
2336 root node names are entirely disjoint, even if they have leaves with
2339 The second field identifies the type's parent node in the tree, or is
2340 null or omitted for a root node. A type is considered to alias all of
2341 its descendants and all of its ancestors in the tree. Also, a type is
2342 considered to alias all types in other trees, so that bitcode produced
2343 from multiple front-ends is handled conservatively.
2345 If the third field is present, it's an integer which if equal to 1
2346 indicates that the type is "constant" (meaning
2347 ``pointsToConstantMemory`` should return true; see `other useful
2348 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2350 '``tbaa.struct``' Metadata
2351 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2353 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2354 aggregate assignment operations in C and similar languages, however it
2355 is defined to copy a contiguous region of memory, which is more than
2356 strictly necessary for aggregate types which contain holes due to
2357 padding. Also, it doesn't contain any TBAA information about the fields
2360 ``!tbaa.struct`` metadata can describe which memory subregions in a
2361 memcpy are padding and what the TBAA tags of the struct are.
2363 The current metadata format is very simple. ``!tbaa.struct`` metadata
2364 nodes are a list of operands which are in conceptual groups of three.
2365 For each group of three, the first operand gives the byte offset of a
2366 field in bytes, the second gives its size in bytes, and the third gives
2369 .. code-block:: llvm
2371 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2373 This describes a struct with two fields. The first is at offset 0 bytes
2374 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2375 and has size 4 bytes and has tbaa tag !2.
2377 Note that the fields need not be contiguous. In this example, there is a
2378 4 byte gap between the two fields. This gap represents padding which
2379 does not carry useful data and need not be preserved.
2381 '``fpmath``' Metadata
2382 ^^^^^^^^^^^^^^^^^^^^^
2384 ``fpmath`` metadata may be attached to any instruction of floating point
2385 type. It can be used to express the maximum acceptable error in the
2386 result of that instruction, in ULPs, thus potentially allowing the
2387 compiler to use a more efficient but less accurate method of computing
2388 it. ULP is defined as follows:
2390 If ``x`` is a real number that lies between two finite consecutive
2391 floating-point numbers ``a`` and ``b``, without being equal to one
2392 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2393 distance between the two non-equal finite floating-point numbers
2394 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2396 The metadata node shall consist of a single positive floating point
2397 number representing the maximum relative error, for example:
2399 .. code-block:: llvm
2401 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2403 '``range``' Metadata
2404 ^^^^^^^^^^^^^^^^^^^^
2406 ``range`` metadata may be attached only to loads of integer types. It
2407 expresses the possible ranges the loaded value is in. The ranges are
2408 represented with a flattened list of integers. The loaded value is known
2409 to be in the union of the ranges defined by each consecutive pair. Each
2410 pair has the following properties:
2412 - The type must match the type loaded by the instruction.
2413 - The pair ``a,b`` represents the range ``[a,b)``.
2414 - Both ``a`` and ``b`` are constants.
2415 - The range is allowed to wrap.
2416 - The range should not represent the full or empty set. That is,
2419 In addition, the pairs must be in signed order of the lower bound and
2420 they must be non-contiguous.
2424 .. code-block:: llvm
2426 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2427 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2428 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2429 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2431 !0 = metadata !{ i8 0, i8 2 }
2432 !1 = metadata !{ i8 255, i8 2 }
2433 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2434 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2436 Module Flags Metadata
2437 =====================
2439 Information about the module as a whole is difficult to convey to LLVM's
2440 subsystems. The LLVM IR isn't sufficient to transmit this information.
2441 The ``llvm.module.flags`` named metadata exists in order to facilitate
2442 this. These flags are in the form of key / value pairs — much like a
2443 dictionary — making it easy for any subsystem who cares about a flag to
2446 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2447 Each triplet has the following form:
2449 - The first element is a *behavior* flag, which specifies the behavior
2450 when two (or more) modules are merged together, and it encounters two
2451 (or more) metadata with the same ID. The supported behaviors are
2453 - The second element is a metadata string that is a unique ID for the
2454 metadata. How each ID is interpreted is documented below.
2455 - The third element is the value of the flag.
2457 When two (or more) modules are merged together, the resulting
2458 ``llvm.module.flags`` metadata is the union of the modules'
2459 ``llvm.module.flags`` metadata. The only exception being a flag with the
2460 *Override* behavior, which may override another flag's value (see
2463 The following behaviors are supported:
2474 Emits an error if two values disagree. It is an error to have an
2475 ID with both an Error and a Warning behavior.
2479 Emits a warning if two values disagree.
2483 Emits an error when the specified value is not present or doesn't
2484 have the specified value. It is an error for two (or more)
2485 ``llvm.module.flags`` with the same ID to have the Require behavior
2486 but different values. There may be multiple Require flags per ID.
2490 Uses the specified value if the two values disagree. It is an
2491 error for two (or more) ``llvm.module.flags`` with the same ID
2492 to have the Override behavior but different values.
2494 An example of module flags:
2496 .. code-block:: llvm
2498 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2499 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2500 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2501 !3 = metadata !{ i32 3, metadata !"qux",
2503 metadata !"foo", i32 1
2506 !llvm.module.flags = !{ !0, !1, !2, !3 }
2508 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2509 if two or more ``!"foo"`` flags are seen is to emit an error if their
2510 values are not equal.
2512 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2513 behavior if two or more ``!"bar"`` flags are seen is to use the value
2514 '37' if their values are not equal.
2516 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2517 behavior if two or more ``!"qux"`` flags are seen is to emit a
2518 warning if their values are not equal.
2520 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2524 metadata !{ metadata !"foo", i32 1 }
2526 The behavior is to emit an error if the ``llvm.module.flags`` does
2527 not contain a flag with the ID ``!"foo"`` that has the value '1'. If
2528 two or more ``!"qux"`` flags exist, then they must have the same
2529 value or an error will be issued.
2531 Objective-C Garbage Collection Module Flags Metadata
2532 ----------------------------------------------------
2534 On the Mach-O platform, Objective-C stores metadata about garbage
2535 collection in a special section called "image info". The metadata
2536 consists of a version number and a bitmask specifying what types of
2537 garbage collection are supported (if any) by the file. If two or more
2538 modules are linked together their garbage collection metadata needs to
2539 be merged rather than appended together.
2541 The Objective-C garbage collection module flags metadata consists of the
2542 following key-value pairs:
2551 * - ``Objective-C Version``
2552 - **[Required]** — The Objective-C ABI version. Valid values are 1 and 2.
2554 * - ``Objective-C Image Info Version``
2555 - **[Required]** — The version of the image info section. Currently
2558 * - ``Objective-C Image Info Section``
2559 - **[Required]** — The section to place the metadata. Valid values are
2560 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2561 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2562 Objective-C ABI version 2.
2564 * - ``Objective-C Garbage Collection``
2565 - **[Required]** — Specifies whether garbage collection is supported or
2566 not. Valid values are 0, for no garbage collection, and 2, for garbage
2567 collection supported.
2569 * - ``Objective-C GC Only``
2570 - **[Optional]** — Specifies that only garbage collection is supported.
2571 If present, its value must be 6. This flag requires that the
2572 ``Objective-C Garbage Collection`` flag have the value 2.
2574 Some important flag interactions:
2576 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2577 merged with a module with ``Objective-C Garbage Collection`` set to
2578 2, then the resulting module has the
2579 ``Objective-C Garbage Collection`` flag set to 0.
2580 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2581 merged with a module with ``Objective-C GC Only`` set to 6.
2583 Intrinsic Global Variables
2584 ==========================
2586 LLVM has a number of "magic" global variables that contain data that
2587 affect code generation or other IR semantics. These are documented here.
2588 All globals of this sort should have a section specified as
2589 "``llvm.metadata``". This section and all globals that start with
2590 "``llvm.``" are reserved for use by LLVM.
2592 The '``llvm.used``' Global Variable
2593 -----------------------------------
2595 The ``@llvm.used`` global is an array with i8\* element type which has
2596 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2597 pointers to global variables and functions which may optionally have a
2598 pointer cast formed of bitcast or getelementptr. For example, a legal
2601 .. code-block:: llvm
2606 @llvm.used = appending global [2 x i8*] [
2608 i8* bitcast (i32* @Y to i8*)
2609 ], section "llvm.metadata"
2611 If a global variable appears in the ``@llvm.used`` list, then the
2612 compiler, assembler, and linker are required to treat the symbol as if
2613 there is a reference to the global that it cannot see. For example, if a
2614 variable has internal linkage and no references other than that from the
2615 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2616 represent references from inline asms and other things the compiler
2617 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2619 On some targets, the code generator must emit a directive to the
2620 assembler or object file to prevent the assembler and linker from
2621 molesting the symbol.
2623 The '``llvm.compiler.used``' Global Variable
2624 --------------------------------------------
2626 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2627 directive, except that it only prevents the compiler from touching the
2628 symbol. On targets that support it, this allows an intelligent linker to
2629 optimize references to the symbol without being impeded as it would be
2632 This is a rare construct that should only be used in rare circumstances,
2633 and should not be exposed to source languages.
2635 The '``llvm.global_ctors``' Global Variable
2636 -------------------------------------------
2638 .. code-block:: llvm
2640 %0 = type { i32, void ()* }
2641 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2643 The ``@llvm.global_ctors`` array contains a list of constructor
2644 functions and associated priorities. The functions referenced by this
2645 array will be called in ascending order of priority (i.e. lowest first)
2646 when the module is loaded. The order of functions with the same priority
2649 The '``llvm.global_dtors``' Global Variable
2650 -------------------------------------------
2652 .. code-block:: llvm
2654 %0 = type { i32, void ()* }
2655 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2657 The ``@llvm.global_dtors`` array contains a list of destructor functions
2658 and associated priorities. The functions referenced by this array will
2659 be called in descending order of priority (i.e. highest first) when the
2660 module is loaded. The order of functions with the same priority is not
2663 Instruction Reference
2664 =====================
2666 The LLVM instruction set consists of several different classifications
2667 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2668 instructions <binaryops>`, :ref:`bitwise binary
2669 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2670 :ref:`other instructions <otherops>`.
2674 Terminator Instructions
2675 -----------------------
2677 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2678 program ends with a "Terminator" instruction, which indicates which
2679 block should be executed after the current block is finished. These
2680 terminator instructions typically yield a '``void``' value: they produce
2681 control flow, not values (the one exception being the
2682 ':ref:`invoke <i_invoke>`' instruction).
2684 The terminator instructions are: ':ref:`ret <i_ret>`',
2685 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2686 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2687 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2691 '``ret``' Instruction
2692 ^^^^^^^^^^^^^^^^^^^^^
2699 ret <type> <value> ; Return a value from a non-void function
2700 ret void ; Return from void function
2705 The '``ret``' instruction is used to return control flow (and optionally
2706 a value) from a function back to the caller.
2708 There are two forms of the '``ret``' instruction: one that returns a
2709 value and then causes control flow, and one that just causes control
2715 The '``ret``' instruction optionally accepts a single argument, the
2716 return value. The type of the return value must be a ':ref:`first
2717 class <t_firstclass>`' type.
2719 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2720 return type and contains a '``ret``' instruction with no return value or
2721 a return value with a type that does not match its type, or if it has a
2722 void return type and contains a '``ret``' instruction with a return
2728 When the '``ret``' instruction is executed, control flow returns back to
2729 the calling function's context. If the caller is a
2730 ":ref:`call <i_call>`" instruction, execution continues at the
2731 instruction after the call. If the caller was an
2732 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2733 beginning of the "normal" destination block. If the instruction returns
2734 a value, that value shall set the call or invoke instruction's return
2740 .. code-block:: llvm
2742 ret i32 5 ; Return an integer value of 5
2743 ret void ; Return from a void function
2744 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2748 '``br``' Instruction
2749 ^^^^^^^^^^^^^^^^^^^^
2756 br i1 <cond>, label <iftrue>, label <iffalse>
2757 br label <dest> ; Unconditional branch
2762 The '``br``' instruction is used to cause control flow to transfer to a
2763 different basic block in the current function. There are two forms of
2764 this instruction, corresponding to a conditional branch and an
2765 unconditional branch.
2770 The conditional branch form of the '``br``' instruction takes a single
2771 '``i1``' value and two '``label``' values. The unconditional form of the
2772 '``br``' instruction takes a single '``label``' value as a target.
2777 Upon execution of a conditional '``br``' instruction, the '``i1``'
2778 argument is evaluated. If the value is ``true``, control flows to the
2779 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
2780 to the '``iffalse``' ``label`` argument.
2785 .. code-block:: llvm
2788 %cond = icmp eq i32 %a, %b
2789 br i1 %cond, label %IfEqual, label %IfUnequal
2797 '``switch``' Instruction
2798 ^^^^^^^^^^^^^^^^^^^^^^^^
2805 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2810 The '``switch``' instruction is used to transfer control flow to one of
2811 several different places. It is a generalization of the '``br``'
2812 instruction, allowing a branch to occur to one of many possible
2818 The '``switch``' instruction uses three parameters: an integer
2819 comparison value '``value``', a default '``label``' destination, and an
2820 array of pairs of comparison value constants and '``label``'s. The table
2821 is not allowed to contain duplicate constant entries.
2826 The ``switch`` instruction specifies a table of values and destinations.
2827 When the '``switch``' instruction is executed, this table is searched
2828 for the given value. If the value is found, control flow is transferred
2829 to the corresponding destination; otherwise, control flow is transferred
2830 to the default destination.
2835 Depending on properties of the target machine and the particular
2836 ``switch`` instruction, this instruction may be code generated in
2837 different ways. For example, it could be generated as a series of
2838 chained conditional branches or with a lookup table.
2843 .. code-block:: llvm
2845 ; Emulate a conditional br instruction
2846 %Val = zext i1 %value to i32
2847 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2849 ; Emulate an unconditional br instruction
2850 switch i32 0, label %dest [ ]
2852 ; Implement a jump table:
2853 switch i32 %val, label %otherwise [ i32 0, label %onzero
2855 i32 2, label %ontwo ]
2859 '``indirectbr``' Instruction
2860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2867 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2872 The '``indirectbr``' instruction implements an indirect branch to a
2873 label within the current function, whose address is specified by
2874 "``address``". Address must be derived from a
2875 :ref:`blockaddress <blockaddress>` constant.
2880 The '``address``' argument is the address of the label to jump to. The
2881 rest of the arguments indicate the full set of possible destinations
2882 that the address may point to. Blocks are allowed to occur multiple
2883 times in the destination list, though this isn't particularly useful.
2885 This destination list is required so that dataflow analysis has an
2886 accurate understanding of the CFG.
2891 Control transfers to the block specified in the address argument. All
2892 possible destination blocks must be listed in the label list, otherwise
2893 this instruction has undefined behavior. This implies that jumps to
2894 labels defined in other functions have undefined behavior as well.
2899 This is typically implemented with a jump through a register.
2904 .. code-block:: llvm
2906 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2910 '``invoke``' Instruction
2911 ^^^^^^^^^^^^^^^^^^^^^^^^
2918 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
2919 to label <normal label> unwind label <exception label>
2924 The '``invoke``' instruction causes control to transfer to a specified
2925 function, with the possibility of control flow transfer to either the
2926 '``normal``' label or the '``exception``' label. If the callee function
2927 returns with the "``ret``" instruction, control flow will return to the
2928 "normal" label. If the callee (or any indirect callees) returns via the
2929 ":ref:`resume <i_resume>`" instruction or other exception handling
2930 mechanism, control is interrupted and continued at the dynamically
2931 nearest "exception" label.
2933 The '``exception``' label is a `landing
2934 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
2935 '``exception``' label is required to have the
2936 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
2937 information about the behavior of the program after unwinding happens,
2938 as its first non-PHI instruction. The restrictions on the
2939 "``landingpad``" instruction's tightly couples it to the "``invoke``"
2940 instruction, so that the important information contained within the
2941 "``landingpad``" instruction can't be lost through normal code motion.
2946 This instruction requires several arguments:
2948 #. The optional "cconv" marker indicates which :ref:`calling
2949 convention <callingconv>` the call should use. If none is
2950 specified, the call defaults to using C calling conventions.
2951 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
2952 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
2954 #. '``ptr to function ty``': shall be the signature of the pointer to
2955 function value being invoked. In most cases, this is a direct
2956 function invocation, but indirect ``invoke``'s are just as possible,
2957 branching off an arbitrary pointer to function value.
2958 #. '``function ptr val``': An LLVM value containing a pointer to a
2959 function to be invoked.
2960 #. '``function args``': argument list whose types match the function
2961 signature argument types and parameter attributes. All arguments must
2962 be of :ref:`first class <t_firstclass>` type. If the function signature
2963 indicates the function accepts a variable number of arguments, the
2964 extra arguments can be specified.
2965 #. '``normal label``': the label reached when the called function
2966 executes a '``ret``' instruction.
2967 #. '``exception label``': the label reached when a callee returns via
2968 the :ref:`resume <i_resume>` instruction or other exception handling
2970 #. The optional :ref:`function attributes <fnattrs>` list. Only
2971 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
2972 attributes are valid here.
2977 This instruction is designed to operate as a standard '``call``'
2978 instruction in most regards. The primary difference is that it
2979 establishes an association with a label, which is used by the runtime
2980 library to unwind the stack.
2982 This instruction is used in languages with destructors to ensure that
2983 proper cleanup is performed in the case of either a ``longjmp`` or a
2984 thrown exception. Additionally, this is important for implementation of
2985 '``catch``' clauses in high-level languages that support them.
2987 For the purposes of the SSA form, the definition of the value returned
2988 by the '``invoke``' instruction is deemed to occur on the edge from the
2989 current block to the "normal" label. If the callee unwinds then no
2990 return value is available.
2995 .. code-block:: llvm
2997 %retval = invoke i32 @Test(i32 15) to label %Continue
2998 unwind label %TestCleanup ; {i32}:retval set
2999 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3000 unwind label %TestCleanup ; {i32}:retval set
3004 '``resume``' Instruction
3005 ^^^^^^^^^^^^^^^^^^^^^^^^
3012 resume <type> <value>
3017 The '``resume``' instruction is a terminator instruction that has no
3023 The '``resume``' instruction requires one argument, which must have the
3024 same type as the result of any '``landingpad``' instruction in the same
3030 The '``resume``' instruction resumes propagation of an existing
3031 (in-flight) exception whose unwinding was interrupted with a
3032 :ref:`landingpad <i_landingpad>` instruction.
3037 .. code-block:: llvm
3039 resume { i8*, i32 } %exn
3043 '``unreachable``' Instruction
3044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3056 The '``unreachable``' instruction has no defined semantics. This
3057 instruction is used to inform the optimizer that a particular portion of
3058 the code is not reachable. This can be used to indicate that the code
3059 after a no-return function cannot be reached, and other facts.
3064 The '``unreachable``' instruction has no defined semantics.
3071 Binary operators are used to do most of the computation in a program.
3072 They require two operands of the same type, execute an operation on
3073 them, and produce a single value. The operands might represent multiple
3074 data, as is the case with the :ref:`vector <t_vector>` data type. The
3075 result value has the same type as its operands.
3077 There are several different binary operators:
3081 '``add``' Instruction
3082 ^^^^^^^^^^^^^^^^^^^^^
3089 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3090 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3091 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3092 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3097 The '``add``' instruction returns the sum of its two operands.
3102 The two arguments to the '``add``' instruction must be
3103 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3104 arguments must have identical types.
3109 The value produced is the integer sum of the two operands.
3111 If the sum has unsigned overflow, the result returned is the
3112 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3115 Because LLVM integers use a two's complement representation, this
3116 instruction is appropriate for both signed and unsigned integers.
3118 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3119 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3120 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3121 unsigned and/or signed overflow, respectively, occurs.
3126 .. code-block:: llvm
3128 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3132 '``fadd``' Instruction
3133 ^^^^^^^^^^^^^^^^^^^^^^
3140 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3145 The '``fadd``' instruction returns the sum of its two operands.
3150 The two arguments to the '``fadd``' instruction must be :ref:`floating
3151 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3152 Both arguments must have identical types.
3157 The value produced is the floating point sum of the two operands. This
3158 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3159 which are optimization hints to enable otherwise unsafe floating point
3165 .. code-block:: llvm
3167 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3169 '``sub``' Instruction
3170 ^^^^^^^^^^^^^^^^^^^^^
3177 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3178 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3179 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3180 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3185 The '``sub``' instruction returns the difference of its two operands.
3187 Note that the '``sub``' instruction is used to represent the '``neg``'
3188 instruction present in most other intermediate representations.
3193 The two arguments to the '``sub``' instruction must be
3194 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3195 arguments must have identical types.
3200 The value produced is the integer difference of the two operands.
3202 If the difference has unsigned overflow, the result returned is the
3203 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3206 Because LLVM integers use a two's complement representation, this
3207 instruction is appropriate for both signed and unsigned integers.
3209 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3210 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3211 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3212 unsigned and/or signed overflow, respectively, occurs.
3217 .. code-block:: llvm
3219 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3220 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3224 '``fsub``' Instruction
3225 ^^^^^^^^^^^^^^^^^^^^^^
3232 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3237 The '``fsub``' instruction returns the difference of its two operands.
3239 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3240 instruction present in most other intermediate representations.
3245 The two arguments to the '``fsub``' instruction must be :ref:`floating
3246 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3247 Both arguments must have identical types.
3252 The value produced is the floating point difference of the two operands.
3253 This instruction can also take any number of :ref:`fast-math
3254 flags <fastmath>`, which are optimization hints to enable otherwise
3255 unsafe floating point optimizations:
3260 .. code-block:: llvm
3262 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3263 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3265 '``mul``' Instruction
3266 ^^^^^^^^^^^^^^^^^^^^^
3273 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3274 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3275 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3276 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3281 The '``mul``' instruction returns the product of its two operands.
3286 The two arguments to the '``mul``' instruction must be
3287 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3288 arguments must have identical types.
3293 The value produced is the integer product of the two operands.
3295 If the result of the multiplication has unsigned overflow, the result
3296 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3297 bit width of the result.
3299 Because LLVM integers use a two's complement representation, and the
3300 result is the same width as the operands, this instruction returns the
3301 correct result for both signed and unsigned integers. If a full product
3302 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3303 sign-extended or zero-extended as appropriate to the width of the full
3306 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3307 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3308 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3309 unsigned and/or signed overflow, respectively, occurs.
3314 .. code-block:: llvm
3316 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3320 '``fmul``' Instruction
3321 ^^^^^^^^^^^^^^^^^^^^^^
3328 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3333 The '``fmul``' instruction returns the product of its two operands.
3338 The two arguments to the '``fmul``' instruction must be :ref:`floating
3339 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3340 Both arguments must have identical types.
3345 The value produced is the floating point product of the two operands.
3346 This instruction can also take any number of :ref:`fast-math
3347 flags <fastmath>`, which are optimization hints to enable otherwise
3348 unsafe floating point optimizations:
3353 .. code-block:: llvm
3355 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3357 '``udiv``' Instruction
3358 ^^^^^^^^^^^^^^^^^^^^^^
3365 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3366 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3371 The '``udiv``' instruction returns the quotient of its two operands.
3376 The two arguments to the '``udiv``' instruction must be
3377 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3378 arguments must have identical types.
3383 The value produced is the unsigned integer quotient of the two operands.
3385 Note that unsigned integer division and signed integer division are
3386 distinct operations; for signed integer division, use '``sdiv``'.
3388 Division by zero leads to undefined behavior.
3390 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3391 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3392 such, "((a udiv exact b) mul b) == a").
3397 .. code-block:: llvm
3399 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3401 '``sdiv``' Instruction
3402 ^^^^^^^^^^^^^^^^^^^^^^
3409 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3410 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3415 The '``sdiv``' instruction returns the quotient of its two operands.
3420 The two arguments to the '``sdiv``' instruction must be
3421 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3422 arguments must have identical types.
3427 The value produced is the signed integer quotient of the two operands
3428 rounded towards zero.
3430 Note that signed integer division and unsigned integer division are
3431 distinct operations; for unsigned integer division, use '``udiv``'.
3433 Division by zero leads to undefined behavior. Overflow also leads to
3434 undefined behavior; this is a rare case, but can occur, for example, by
3435 doing a 32-bit division of -2147483648 by -1.
3437 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3438 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3443 .. code-block:: llvm
3445 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3449 '``fdiv``' Instruction
3450 ^^^^^^^^^^^^^^^^^^^^^^
3457 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3462 The '``fdiv``' instruction returns the quotient of its two operands.
3467 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3468 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3469 Both arguments must have identical types.
3474 The value produced is the floating point quotient of the two operands.
3475 This instruction can also take any number of :ref:`fast-math
3476 flags <fastmath>`, which are optimization hints to enable otherwise
3477 unsafe floating point optimizations:
3482 .. code-block:: llvm
3484 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3486 '``urem``' Instruction
3487 ^^^^^^^^^^^^^^^^^^^^^^
3494 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3499 The '``urem``' instruction returns the remainder from the unsigned
3500 division of its two arguments.
3505 The two arguments to the '``urem``' instruction must be
3506 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3507 arguments must have identical types.
3512 This instruction returns the unsigned integer *remainder* of a division.
3513 This instruction always performs an unsigned division to get the
3516 Note that unsigned integer remainder and signed integer remainder are
3517 distinct operations; for signed integer remainder, use '``srem``'.
3519 Taking the remainder of a division by zero leads to undefined behavior.
3524 .. code-block:: llvm
3526 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3528 '``srem``' Instruction
3529 ^^^^^^^^^^^^^^^^^^^^^^
3536 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3541 The '``srem``' instruction returns the remainder from the signed
3542 division of its two operands. This instruction can also take
3543 :ref:`vector <t_vector>` versions of the values in which case the elements
3549 The two arguments to the '``srem``' instruction must be
3550 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3551 arguments must have identical types.
3556 This instruction returns the *remainder* of a division (where the result
3557 is either zero or has the same sign as the dividend, ``op1``), not the
3558 *modulo* operator (where the result is either zero or has the same sign
3559 as the divisor, ``op2``) of a value. For more information about the
3560 difference, see `The Math
3561 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3562 table of how this is implemented in various languages, please see
3564 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3566 Note that signed integer remainder and unsigned integer remainder are
3567 distinct operations; for unsigned integer remainder, use '``urem``'.
3569 Taking the remainder of a division by zero leads to undefined behavior.
3570 Overflow also leads to undefined behavior; this is a rare case, but can
3571 occur, for example, by taking the remainder of a 32-bit division of
3572 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3573 rule lets srem be implemented using instructions that return both the
3574 result of the division and the remainder.)
3579 .. code-block:: llvm
3581 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3585 '``frem``' Instruction
3586 ^^^^^^^^^^^^^^^^^^^^^^
3593 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3598 The '``frem``' instruction returns the remainder from the division of
3604 The two arguments to the '``frem``' instruction must be :ref:`floating
3605 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3606 Both arguments must have identical types.
3611 This instruction returns the *remainder* of a division. The remainder
3612 has the same sign as the dividend. This instruction can also take any
3613 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3614 to enable otherwise unsafe floating point optimizations:
3619 .. code-block:: llvm
3621 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3625 Bitwise Binary Operations
3626 -------------------------
3628 Bitwise binary operators are used to do various forms of bit-twiddling
3629 in a program. They are generally very efficient instructions and can
3630 commonly be strength reduced from other instructions. They require two
3631 operands of the same type, execute an operation on them, and produce a
3632 single value. The resulting value is the same type as its operands.
3634 '``shl``' Instruction
3635 ^^^^^^^^^^^^^^^^^^^^^
3642 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3643 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3644 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3645 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3650 The '``shl``' instruction returns the first operand shifted to the left
3651 a specified number of bits.
3656 Both arguments to the '``shl``' instruction must be the same
3657 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3658 '``op2``' is treated as an unsigned value.
3663 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3664 where ``n`` is the width of the result. If ``op2`` is (statically or
3665 dynamically) negative or equal to or larger than the number of bits in
3666 ``op1``, the result is undefined. If the arguments are vectors, each
3667 vector element of ``op1`` is shifted by the corresponding shift amount
3670 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3671 value <poisonvalues>` if it shifts out any non-zero bits. If the
3672 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3673 value <poisonvalues>` if it shifts out any bits that disagree with the
3674 resultant sign bit. As such, NUW/NSW have the same semantics as they
3675 would if the shift were expressed as a mul instruction with the same
3676 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3681 .. code-block:: llvm
3683 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3684 <result> = shl i32 4, 2 ; yields {i32}: 16
3685 <result> = shl i32 1, 10 ; yields {i32}: 1024
3686 <result> = shl i32 1, 32 ; undefined
3687 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3689 '``lshr``' Instruction
3690 ^^^^^^^^^^^^^^^^^^^^^^
3697 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3698 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3703 The '``lshr``' instruction (logical shift right) returns the first
3704 operand shifted to the right a specified number of bits with zero fill.
3709 Both arguments to the '``lshr``' instruction must be the same
3710 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3711 '``op2``' is treated as an unsigned value.
3716 This instruction always performs a logical shift right operation. The
3717 most significant bits of the result will be filled with zero bits after
3718 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3719 than the number of bits in ``op1``, the result is undefined. If the
3720 arguments are vectors, each vector element of ``op1`` is shifted by the
3721 corresponding shift amount in ``op2``.
3723 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3724 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3730 .. code-block:: llvm
3732 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3733 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3734 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3735 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3736 <result> = lshr i32 1, 32 ; undefined
3737 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3739 '``ashr``' Instruction
3740 ^^^^^^^^^^^^^^^^^^^^^^
3747 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
3748 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
3753 The '``ashr``' instruction (arithmetic shift right) returns the first
3754 operand shifted to the right a specified number of bits with sign
3760 Both arguments to the '``ashr``' instruction must be the same
3761 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3762 '``op2``' is treated as an unsigned value.
3767 This instruction always performs an arithmetic shift right operation,
3768 The most significant bits of the result will be filled with the sign bit
3769 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
3770 than the number of bits in ``op1``, the result is undefined. If the
3771 arguments are vectors, each vector element of ``op1`` is shifted by the
3772 corresponding shift amount in ``op2``.
3774 If the ``exact`` keyword is present, the result value of the ``ashr`` is
3775 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3781 .. code-block:: llvm
3783 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
3784 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
3785 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
3786 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
3787 <result> = ashr i32 1, 32 ; undefined
3788 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
3790 '``and``' Instruction
3791 ^^^^^^^^^^^^^^^^^^^^^
3798 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
3803 The '``and``' instruction returns the bitwise logical and of its two
3809 The two arguments to the '``and``' instruction must be
3810 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3811 arguments must have identical types.
3816 The truth table used for the '``and``' instruction is:
3833 .. code-block:: llvm
3835 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
3836 <result> = and i32 15, 40 ; yields {i32}:result = 8
3837 <result> = and i32 4, 8 ; yields {i32}:result = 0
3839 '``or``' Instruction
3840 ^^^^^^^^^^^^^^^^^^^^
3847 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
3852 The '``or``' instruction returns the bitwise logical inclusive or of its
3858 The two arguments to the '``or``' instruction must be
3859 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3860 arguments must have identical types.
3865 The truth table used for the '``or``' instruction is:
3884 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
3885 <result> = or i32 15, 40 ; yields {i32}:result = 47
3886 <result> = or i32 4, 8 ; yields {i32}:result = 12
3888 '``xor``' Instruction
3889 ^^^^^^^^^^^^^^^^^^^^^
3896 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
3901 The '``xor``' instruction returns the bitwise logical exclusive or of
3902 its two operands. The ``xor`` is used to implement the "one's
3903 complement" operation, which is the "~" operator in C.
3908 The two arguments to the '``xor``' instruction must be
3909 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3910 arguments must have identical types.
3915 The truth table used for the '``xor``' instruction is:
3932 .. code-block:: llvm
3934 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
3935 <result> = xor i32 15, 40 ; yields {i32}:result = 39
3936 <result> = xor i32 4, 8 ; yields {i32}:result = 12
3937 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
3942 LLVM supports several instructions to represent vector operations in a
3943 target-independent manner. These instructions cover the element-access
3944 and vector-specific operations needed to process vectors effectively.
3945 While LLVM does directly support these vector operations, many
3946 sophisticated algorithms will want to use target-specific intrinsics to
3947 take full advantage of a specific target.
3949 .. _i_extractelement:
3951 '``extractelement``' Instruction
3952 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3959 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
3964 The '``extractelement``' instruction extracts a single scalar element
3965 from a vector at a specified index.
3970 The first operand of an '``extractelement``' instruction is a value of
3971 :ref:`vector <t_vector>` type. The second operand is an index indicating
3972 the position from which to extract the element. The index may be a
3978 The result is a scalar of the same type as the element type of ``val``.
3979 Its value is the value at position ``idx`` of ``val``. If ``idx``
3980 exceeds the length of ``val``, the results are undefined.
3985 .. code-block:: llvm
3987 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
3989 .. _i_insertelement:
3991 '``insertelement``' Instruction
3992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3999 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4004 The '``insertelement``' instruction inserts a scalar element into a
4005 vector at a specified index.
4010 The first operand of an '``insertelement``' instruction is a value of
4011 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4012 type must equal the element type of the first operand. The third operand
4013 is an index indicating the position at which to insert the value. The
4014 index may be a variable.
4019 The result is a vector of the same type as ``val``. Its element values
4020 are those of ``val`` except at position ``idx``, where it gets the value
4021 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4027 .. code-block:: llvm
4029 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4031 .. _i_shufflevector:
4033 '``shufflevector``' Instruction
4034 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4041 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4046 The '``shufflevector``' instruction constructs a permutation of elements
4047 from two input vectors, returning a vector with the same element type as
4048 the input and length that is the same as the shuffle mask.
4053 The first two operands of a '``shufflevector``' instruction are vectors
4054 with the same type. The third argument is a shuffle mask whose element
4055 type is always 'i32'. The result of the instruction is a vector whose
4056 length is the same as the shuffle mask and whose element type is the
4057 same as the element type of the first two operands.
4059 The shuffle mask operand is required to be a constant vector with either
4060 constant integer or undef values.
4065 The elements of the two input vectors are numbered from left to right
4066 across both of the vectors. The shuffle mask operand specifies, for each
4067 element of the result vector, which element of the two input vectors the
4068 result element gets. The element selector may be undef (meaning "don't
4069 care") and the second operand may be undef if performing a shuffle from
4075 .. code-block:: llvm
4077 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4078 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4079 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4080 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4081 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4082 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4083 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4084 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4086 Aggregate Operations
4087 --------------------
4089 LLVM supports several instructions for working with
4090 :ref:`aggregate <t_aggregate>` values.
4094 '``extractvalue``' Instruction
4095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4102 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4107 The '``extractvalue``' instruction extracts the value of a member field
4108 from an :ref:`aggregate <t_aggregate>` value.
4113 The first operand of an '``extractvalue``' instruction is a value of
4114 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4115 constant indices to specify which value to extract in a similar manner
4116 as indices in a '``getelementptr``' instruction.
4118 The major differences to ``getelementptr`` indexing are:
4120 - Since the value being indexed is not a pointer, the first index is
4121 omitted and assumed to be zero.
4122 - At least one index must be specified.
4123 - Not only struct indices but also array indices must be in bounds.
4128 The result is the value at the position in the aggregate specified by
4134 .. code-block:: llvm
4136 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4140 '``insertvalue``' Instruction
4141 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4148 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4153 The '``insertvalue``' instruction inserts a value into a member field in
4154 an :ref:`aggregate <t_aggregate>` value.
4159 The first operand of an '``insertvalue``' instruction is a value of
4160 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4161 a first-class value to insert. The following operands are constant
4162 indices indicating the position at which to insert the value in a
4163 similar manner as indices in a '``extractvalue``' instruction. The value
4164 to insert must have the same type as the value identified by the
4170 The result is an aggregate of the same type as ``val``. Its value is
4171 that of ``val`` except that the value at the position specified by the
4172 indices is that of ``elt``.
4177 .. code-block:: llvm
4179 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4180 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4181 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4185 Memory Access and Addressing Operations
4186 ---------------------------------------
4188 A key design point of an SSA-based representation is how it represents
4189 memory. In LLVM, no memory locations are in SSA form, which makes things
4190 very simple. This section describes how to read, write, and allocate
4195 '``alloca``' Instruction
4196 ^^^^^^^^^^^^^^^^^^^^^^^^
4203 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4208 The '``alloca``' instruction allocates memory on the stack frame of the
4209 currently executing function, to be automatically released when this
4210 function returns to its caller. The object is always allocated in the
4211 generic address space (address space zero).
4216 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4217 bytes of memory on the runtime stack, returning a pointer of the
4218 appropriate type to the program. If "NumElements" is specified, it is
4219 the number of elements allocated, otherwise "NumElements" is defaulted
4220 to be one. If a constant alignment is specified, the value result of the
4221 allocation is guaranteed to be aligned to at least that boundary. If not
4222 specified, or if zero, the target can choose to align the allocation on
4223 any convenient boundary compatible with the type.
4225 '``type``' may be any sized type.
4230 Memory is allocated; a pointer is returned. The operation is undefined
4231 if there is insufficient stack space for the allocation. '``alloca``'d
4232 memory is automatically released when the function returns. The
4233 '``alloca``' instruction is commonly used to represent automatic
4234 variables that must have an address available. When the function returns
4235 (either with the ``ret`` or ``resume`` instructions), the memory is
4236 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4237 The order in which memory is allocated (ie., which way the stack grows)
4243 .. code-block:: llvm
4245 %ptr = alloca i32 ; yields {i32*}:ptr
4246 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4247 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4248 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4252 '``load``' Instruction
4253 ^^^^^^^^^^^^^^^^^^^^^^
4260 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4261 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4262 !<index> = !{ i32 1 }
4267 The '``load``' instruction is used to read from memory.
4272 The argument to the '``load``' instruction specifies the memory address
4273 from which to load. The pointer must point to a :ref:`first
4274 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4275 then the optimizer is not allowed to modify the number or order of
4276 execution of this ``load`` with other :ref:`volatile
4277 operations <volatile>`.
4279 If the ``load`` is marked as ``atomic``, it takes an extra
4280 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4281 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4282 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4283 when they may see multiple atomic stores. The type of the pointee must
4284 be an integer type whose bit width is a power of two greater than or
4285 equal to eight and less than or equal to a target-specific size limit.
4286 ``align`` must be explicitly specified on atomic loads, and the load has
4287 undefined behavior if the alignment is not set to a value which is at
4288 least the size in bytes of the pointee. ``!nontemporal`` does not have
4289 any defined semantics for atomic loads.
4291 The optional constant ``align`` argument specifies the alignment of the
4292 operation (that is, the alignment of the memory address). A value of 0
4293 or an omitted ``align`` argument means that the operation has the abi
4294 alignment for the target. It is the responsibility of the code emitter
4295 to ensure that the alignment information is correct. Overestimating the
4296 alignment results in undefined behavior. Underestimating the alignment
4297 may produce less efficient code. An alignment of 1 is always safe.
4299 The optional ``!nontemporal`` metadata must reference a single
4300 metatadata name <index> corresponding to a metadata node with one
4301 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4302 metatadata on the instruction tells the optimizer and code generator
4303 that this load is not expected to be reused in the cache. The code
4304 generator may select special instructions to save cache bandwidth, such
4305 as the ``MOVNT`` instruction on x86.
4307 The optional ``!invariant.load`` metadata must reference a single
4308 metatadata name <index> corresponding to a metadata node with no
4309 entries. The existence of the ``!invariant.load`` metatadata on the
4310 instruction tells the optimizer and code generator that this load
4311 address points to memory which does not change value during program
4312 execution. The optimizer may then move this load around, for example, by
4313 hoisting it out of loops using loop invariant code motion.
4318 The location of memory pointed to is loaded. If the value being loaded
4319 is of scalar type then the number of bytes read does not exceed the
4320 minimum number of bytes needed to hold all bits of the type. For
4321 example, loading an ``i24`` reads at most three bytes. When loading a
4322 value of a type like ``i20`` with a size that is not an integral number
4323 of bytes, the result is undefined if the value was not originally
4324 written using a store of the same type.
4329 .. code-block:: llvm
4331 %ptr = alloca i32 ; yields {i32*}:ptr
4332 store i32 3, i32* %ptr ; yields {void}
4333 %val = load i32* %ptr ; yields {i32}:val = i32 3
4337 '``store``' Instruction
4338 ^^^^^^^^^^^^^^^^^^^^^^^
4345 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4346 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4351 The '``store``' instruction is used to write to memory.
4356 There are two arguments to the '``store``' instruction: a value to store
4357 and an address at which to store it. The type of the '``<pointer>``'
4358 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4359 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4360 then the optimizer is not allowed to modify the number or order of
4361 execution of this ``store`` with other :ref:`volatile
4362 operations <volatile>`.
4364 If the ``store`` is marked as ``atomic``, it takes an extra
4365 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4366 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4367 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4368 when they may see multiple atomic stores. The type of the pointee must
4369 be an integer type whose bit width is a power of two greater than or
4370 equal to eight and less than or equal to a target-specific size limit.
4371 ``align`` must be explicitly specified on atomic stores, and the store
4372 has undefined behavior if the alignment is not set to a value which is
4373 at least the size in bytes of the pointee. ``!nontemporal`` does not
4374 have any defined semantics for atomic stores.
4376 The optional constant "align" argument specifies the alignment of the
4377 operation (that is, the alignment of the memory address). A value of 0
4378 or an omitted "align" argument means that the operation has the abi
4379 alignment for the target. It is the responsibility of the code emitter
4380 to ensure that the alignment information is correct. Overestimating the
4381 alignment results in an undefined behavior. Underestimating the
4382 alignment may produce less efficient code. An alignment of 1 is always
4385 The optional !nontemporal metadata must reference a single metatadata
4386 name <index> corresponding to a metadata node with one i32 entry of
4387 value 1. The existence of the !nontemporal metatadata on the instruction
4388 tells the optimizer and code generator that this load is not expected to
4389 be reused in the cache. The code generator may select special
4390 instructions to save cache bandwidth, such as the MOVNT instruction on
4396 The contents of memory are updated to contain '``<value>``' at the
4397 location specified by the '``<pointer>``' operand. If '``<value>``' is
4398 of scalar type then the number of bytes written does not exceed the
4399 minimum number of bytes needed to hold all bits of the type. For
4400 example, storing an ``i24`` writes at most three bytes. When writing a
4401 value of a type like ``i20`` with a size that is not an integral number
4402 of bytes, it is unspecified what happens to the extra bits that do not
4403 belong to the type, but they will typically be overwritten.
4408 .. code-block:: llvm
4410 %ptr = alloca i32 ; yields {i32*}:ptr
4411 store i32 3, i32* %ptr ; yields {void}
4412 %val = load i32* %ptr ; yields {i32}:val = i32 3
4416 '``fence``' Instruction
4417 ^^^^^^^^^^^^^^^^^^^^^^^
4424 fence [singlethread] <ordering> ; yields {void}
4429 The '``fence``' instruction is used to introduce happens-before edges
4435 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4436 defines what *synchronizes-with* edges they add. They can only be given
4437 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4442 A fence A which has (at least) ``release`` ordering semantics
4443 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4444 semantics if and only if there exist atomic operations X and Y, both
4445 operating on some atomic object M, such that A is sequenced before X, X
4446 modifies M (either directly or through some side effect of a sequence
4447 headed by X), Y is sequenced before B, and Y observes M. This provides a
4448 *happens-before* dependency between A and B. Rather than an explicit
4449 ``fence``, one (but not both) of the atomic operations X or Y might
4450 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4451 still *synchronize-with* the explicit ``fence`` and establish the
4452 *happens-before* edge.
4454 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4455 ``acquire`` and ``release`` semantics specified above, participates in
4456 the global program order of other ``seq_cst`` operations and/or fences.
4458 The optional ":ref:`singlethread <singlethread>`" argument specifies
4459 that the fence only synchronizes with other fences in the same thread.
4460 (This is useful for interacting with signal handlers.)
4465 .. code-block:: llvm
4467 fence acquire ; yields {void}
4468 fence singlethread seq_cst ; yields {void}
4472 '``cmpxchg``' Instruction
4473 ^^^^^^^^^^^^^^^^^^^^^^^^^
4480 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4485 The '``cmpxchg``' instruction is used to atomically modify memory. It
4486 loads a value in memory and compares it to a given value. If they are
4487 equal, it stores a new value into the memory.
4492 There are three arguments to the '``cmpxchg``' instruction: an address
4493 to operate on, a value to compare to the value currently be at that
4494 address, and a new value to place at that address if the compared values
4495 are equal. The type of '<cmp>' must be an integer type whose bit width
4496 is a power of two greater than or equal to eight and less than or equal
4497 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4498 type, and the type of '<pointer>' must be a pointer to that type. If the
4499 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4500 to modify the number or order of execution of this ``cmpxchg`` with
4501 other :ref:`volatile operations <volatile>`.
4503 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4504 synchronizes with other atomic operations.
4506 The optional "``singlethread``" argument declares that the ``cmpxchg``
4507 is only atomic with respect to code (usually signal handlers) running in
4508 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4509 respect to all other code in the system.
4511 The pointer passed into cmpxchg must have alignment greater than or
4512 equal to the size in memory of the operand.
4517 The contents of memory at the location specified by the '``<pointer>``'
4518 operand is read and compared to '``<cmp>``'; if the read value is the
4519 equal, '``<new>``' is written. The original value at the location is
4522 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4523 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4524 atomic load with an ordering parameter determined by dropping any
4525 ``release`` part of the ``cmpxchg``'s ordering.
4530 .. code-block:: llvm
4533 %orig = atomic load i32* %ptr unordered ; yields {i32}
4537 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4538 %squared = mul i32 %cmp, %cmp
4539 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4540 %success = icmp eq i32 %cmp, %old
4541 br i1 %success, label %done, label %loop
4548 '``atomicrmw``' Instruction
4549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4556 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4561 The '``atomicrmw``' instruction is used to atomically modify memory.
4566 There are three arguments to the '``atomicrmw``' instruction: an
4567 operation to apply, an address whose value to modify, an argument to the
4568 operation. The operation must be one of the following keywords:
4582 The type of '<value>' must be an integer type whose bit width is a power
4583 of two greater than or equal to eight and less than or equal to a
4584 target-specific size limit. The type of the '``<pointer>``' operand must
4585 be a pointer to that type. If the ``atomicrmw`` is marked as
4586 ``volatile``, then the optimizer is not allowed to modify the number or
4587 order of execution of this ``atomicrmw`` with other :ref:`volatile
4588 operations <volatile>`.
4593 The contents of memory at the location specified by the '``<pointer>``'
4594 operand are atomically read, modified, and written back. The original
4595 value at the location is returned. The modification is specified by the
4598 - xchg: ``*ptr = val``
4599 - add: ``*ptr = *ptr + val``
4600 - sub: ``*ptr = *ptr - val``
4601 - and: ``*ptr = *ptr & val``
4602 - nand: ``*ptr = ~(*ptr & val)``
4603 - or: ``*ptr = *ptr | val``
4604 - xor: ``*ptr = *ptr ^ val``
4605 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4606 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4607 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4609 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4615 .. code-block:: llvm
4617 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4619 .. _i_getelementptr:
4621 '``getelementptr``' Instruction
4622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4629 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4630 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4631 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4636 The '``getelementptr``' instruction is used to get the address of a
4637 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4638 address calculation only and does not access memory.
4643 The first argument is always a pointer or a vector of pointers, and
4644 forms the basis of the calculation. The remaining arguments are indices
4645 that indicate which of the elements of the aggregate object are indexed.
4646 The interpretation of each index is dependent on the type being indexed
4647 into. The first index always indexes the pointer value given as the
4648 first argument, the second index indexes a value of the type pointed to
4649 (not necessarily the value directly pointed to, since the first index
4650 can be non-zero), etc. The first type indexed into must be a pointer
4651 value, subsequent types can be arrays, vectors, and structs. Note that
4652 subsequent types being indexed into can never be pointers, since that
4653 would require loading the pointer before continuing calculation.
4655 The type of each index argument depends on the type it is indexing into.
4656 When indexing into a (optionally packed) structure, only ``i32`` integer
4657 **constants** are allowed (when using a vector of indices they must all
4658 be the **same** ``i32`` integer constant). When indexing into an array,
4659 pointer or vector, integers of any width are allowed, and they are not
4660 required to be constant. These integers are treated as signed values
4663 For example, let's consider a C code fragment and how it gets compiled
4679 int *foo(struct ST *s) {
4680 return &s[1].Z.B[5][13];
4683 The LLVM code generated by Clang is:
4685 .. code-block:: llvm
4687 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4688 %struct.ST = type { i32, double, %struct.RT }
4690 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4692 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4699 In the example above, the first index is indexing into the
4700 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4701 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4702 indexes into the third element of the structure, yielding a
4703 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4704 structure. The third index indexes into the second element of the
4705 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4706 dimensions of the array are subscripted into, yielding an '``i32``'
4707 type. The '``getelementptr``' instruction returns a pointer to this
4708 element, thus computing a value of '``i32*``' type.
4710 Note that it is perfectly legal to index partially through a structure,
4711 returning a pointer to an inner element. Because of this, the LLVM code
4712 for the given testcase is equivalent to:
4714 .. code-block:: llvm
4716 define i32* @foo(%struct.ST* %s) {
4717 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4718 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4719 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4720 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4721 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4725 If the ``inbounds`` keyword is present, the result value of the
4726 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4727 pointer is not an *in bounds* address of an allocated object, or if any
4728 of the addresses that would be formed by successive addition of the
4729 offsets implied by the indices to the base address with infinitely
4730 precise signed arithmetic are not an *in bounds* address of that
4731 allocated object. The *in bounds* addresses for an allocated object are
4732 all the addresses that point into the object, plus the address one byte
4733 past the end. In cases where the base is a vector of pointers the
4734 ``inbounds`` keyword applies to each of the computations element-wise.
4736 If the ``inbounds`` keyword is not present, the offsets are added to the
4737 base address with silently-wrapping two's complement arithmetic. If the
4738 offsets have a different width from the pointer, they are sign-extended
4739 or truncated to the width of the pointer. The result value of the
4740 ``getelementptr`` may be outside the object pointed to by the base
4741 pointer. The result value may not necessarily be used to access memory
4742 though, even if it happens to point into allocated storage. See the
4743 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4746 The getelementptr instruction is often confusing. For some more insight
4747 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
4752 .. code-block:: llvm
4754 ; yields [12 x i8]*:aptr
4755 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4757 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4759 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4761 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4763 In cases where the pointer argument is a vector of pointers, each index
4764 must be a vector with the same number of elements. For example:
4766 .. code-block:: llvm
4768 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
4770 Conversion Operations
4771 ---------------------
4773 The instructions in this category are the conversion instructions
4774 (casting) which all take a single operand and a type. They perform
4775 various bit conversions on the operand.
4777 '``trunc .. to``' Instruction
4778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4785 <result> = trunc <ty> <value> to <ty2> ; yields ty2
4790 The '``trunc``' instruction truncates its operand to the type ``ty2``.
4795 The '``trunc``' instruction takes a value to trunc, and a type to trunc
4796 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
4797 of the same number of integers. The bit size of the ``value`` must be
4798 larger than the bit size of the destination type, ``ty2``. Equal sized
4799 types are not allowed.
4804 The '``trunc``' instruction truncates the high order bits in ``value``
4805 and converts the remaining bits to ``ty2``. Since the source size must
4806 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
4807 It will always truncate bits.
4812 .. code-block:: llvm
4814 %X = trunc i32 257 to i8 ; yields i8:1
4815 %Y = trunc i32 123 to i1 ; yields i1:true
4816 %Z = trunc i32 122 to i1 ; yields i1:false
4817 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
4819 '``zext .. to``' Instruction
4820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4827 <result> = zext <ty> <value> to <ty2> ; yields ty2
4832 The '``zext``' instruction zero extends its operand to type ``ty2``.
4837 The '``zext``' instruction takes a value to cast, and a type to cast it
4838 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4839 the same number of integers. The bit size of the ``value`` must be
4840 smaller than the bit size of the destination type, ``ty2``.
4845 The ``zext`` fills the high order bits of the ``value`` with zero bits
4846 until it reaches the size of the destination type, ``ty2``.
4848 When zero extending from i1, the result will always be either 0 or 1.
4853 .. code-block:: llvm
4855 %X = zext i32 257 to i64 ; yields i64:257
4856 %Y = zext i1 true to i32 ; yields i32:1
4857 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4859 '``sext .. to``' Instruction
4860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4867 <result> = sext <ty> <value> to <ty2> ; yields ty2
4872 The '``sext``' sign extends ``value`` to the type ``ty2``.
4877 The '``sext``' instruction takes a value to cast, and a type to cast it
4878 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4879 the same number of integers. The bit size of the ``value`` must be
4880 smaller than the bit size of the destination type, ``ty2``.
4885 The '``sext``' instruction performs a sign extension by copying the sign
4886 bit (highest order bit) of the ``value`` until it reaches the bit size
4887 of the type ``ty2``.
4889 When sign extending from i1, the extension always results in -1 or 0.
4894 .. code-block:: llvm
4896 %X = sext i8 -1 to i16 ; yields i16 :65535
4897 %Y = sext i1 true to i32 ; yields i32:-1
4898 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4900 '``fptrunc .. to``' Instruction
4901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4908 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
4913 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
4918 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
4919 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
4920 The size of ``value`` must be larger than the size of ``ty2``. This
4921 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
4926 The '``fptrunc``' instruction truncates a ``value`` from a larger
4927 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
4928 point <t_floating>` type. If the value cannot fit within the
4929 destination type, ``ty2``, then the results are undefined.
4934 .. code-block:: llvm
4936 %X = fptrunc double 123.0 to float ; yields float:123.0
4937 %Y = fptrunc double 1.0E+300 to float ; yields undefined
4939 '``fpext .. to``' Instruction
4940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4947 <result> = fpext <ty> <value> to <ty2> ; yields ty2
4952 The '``fpext``' extends a floating point ``value`` to a larger floating
4958 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
4959 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
4960 to. The source type must be smaller than the destination type.
4965 The '``fpext``' instruction extends the ``value`` from a smaller
4966 :ref:`floating point <t_floating>` type to a larger :ref:`floating
4967 point <t_floating>` type. The ``fpext`` cannot be used to make a
4968 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
4969 *no-op cast* for a floating point cast.
4974 .. code-block:: llvm
4976 %X = fpext float 3.125 to double ; yields double:3.125000e+00
4977 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
4979 '``fptoui .. to``' Instruction
4980 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4987 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
4992 The '``fptoui``' converts a floating point ``value`` to its unsigned
4993 integer equivalent of type ``ty2``.
4998 The '``fptoui``' instruction takes a value to cast, which must be a
4999 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5000 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5001 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5002 type with the same number of elements as ``ty``
5007 The '``fptoui``' instruction converts its :ref:`floating
5008 point <t_floating>` operand into the nearest (rounding towards zero)
5009 unsigned integer value. If the value cannot fit in ``ty2``, the results
5015 .. code-block:: llvm
5017 %X = fptoui double 123.0 to i32 ; yields i32:123
5018 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5019 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5021 '``fptosi .. to``' Instruction
5022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5029 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5034 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5035 ``value`` to type ``ty2``.
5040 The '``fptosi``' instruction takes a value to cast, which must be a
5041 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5042 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5043 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5044 type with the same number of elements as ``ty``
5049 The '``fptosi``' instruction converts its :ref:`floating
5050 point <t_floating>` operand into the nearest (rounding towards zero)
5051 signed integer value. If the value cannot fit in ``ty2``, the results
5057 .. code-block:: llvm
5059 %X = fptosi double -123.0 to i32 ; yields i32:-123
5060 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5061 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5063 '``uitofp .. to``' Instruction
5064 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5071 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5076 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5077 and converts that value to the ``ty2`` type.
5082 The '``uitofp``' instruction takes a value to cast, which must be a
5083 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5084 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5085 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5086 type with the same number of elements as ``ty``
5091 The '``uitofp``' instruction interprets its operand as an unsigned
5092 integer quantity and converts it to the corresponding floating point
5093 value. If the value cannot fit in the floating point value, the results
5099 .. code-block:: llvm
5101 %X = uitofp i32 257 to float ; yields float:257.0
5102 %Y = uitofp i8 -1 to double ; yields double:255.0
5104 '``sitofp .. to``' Instruction
5105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5112 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5117 The '``sitofp``' instruction regards ``value`` as a signed integer and
5118 converts that value to the ``ty2`` type.
5123 The '``sitofp``' instruction takes a value to cast, which must be a
5124 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5125 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5126 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5127 type with the same number of elements as ``ty``
5132 The '``sitofp``' instruction interprets its operand as a signed integer
5133 quantity and converts it to the corresponding floating point value. If
5134 the value cannot fit in the floating point value, the results are
5140 .. code-block:: llvm
5142 %X = sitofp i32 257 to float ; yields float:257.0
5143 %Y = sitofp i8 -1 to double ; yields double:-1.0
5147 '``ptrtoint .. to``' Instruction
5148 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5155 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5160 The '``ptrtoint``' instruction converts the pointer or a vector of
5161 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5166 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5167 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5168 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5169 a vector of integers type.
5174 The '``ptrtoint``' instruction converts ``value`` to integer type
5175 ``ty2`` by interpreting the pointer value as an integer and either
5176 truncating or zero extending that value to the size of the integer type.
5177 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5178 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5179 the same size, then nothing is done (*no-op cast*) other than a type
5185 .. code-block:: llvm
5187 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5188 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5189 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5193 '``inttoptr .. to``' Instruction
5194 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5201 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5206 The '``inttoptr``' instruction converts an integer ``value`` to a
5207 pointer type, ``ty2``.
5212 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5213 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5219 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5220 applying either a zero extension or a truncation depending on the size
5221 of the integer ``value``. If ``value`` is larger than the size of a
5222 pointer then a truncation is done. If ``value`` is smaller than the size
5223 of a pointer then a zero extension is done. If they are the same size,
5224 nothing is done (*no-op cast*).
5229 .. code-block:: llvm
5231 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5232 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5233 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5234 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5238 '``bitcast .. to``' Instruction
5239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5246 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5251 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5257 The '``bitcast``' instruction takes a value to cast, which must be a
5258 non-aggregate first class value, and a type to cast it to, which must
5259 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5260 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5261 If the source type is a pointer, the destination type must also be a
5262 pointer. This instruction supports bitwise conversion of vectors to
5263 integers and to vectors of other types (as long as they have the same
5269 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5270 always a *no-op cast* because no bits change with this conversion. The
5271 conversion is done as if the ``value`` had been stored to memory and
5272 read back as type ``ty2``. Pointer (or vector of pointers) types may
5273 only be converted to other pointer (or vector of pointers) types with
5274 this instruction. To convert pointers to other types, use the
5275 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5281 .. code-block:: llvm
5283 %X = bitcast i8 255 to i8 ; yields i8 :-1
5284 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5285 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5286 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5293 The instructions in this category are the "miscellaneous" instructions,
5294 which defy better classification.
5298 '``icmp``' Instruction
5299 ^^^^^^^^^^^^^^^^^^^^^^
5306 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5311 The '``icmp``' instruction returns a boolean value or a vector of
5312 boolean values based on comparison of its two integer, integer vector,
5313 pointer, or pointer vector operands.
5318 The '``icmp``' instruction takes three operands. The first operand is
5319 the condition code indicating the kind of comparison to perform. It is
5320 not a value, just a keyword. The possible condition code are:
5323 #. ``ne``: not equal
5324 #. ``ugt``: unsigned greater than
5325 #. ``uge``: unsigned greater or equal
5326 #. ``ult``: unsigned less than
5327 #. ``ule``: unsigned less or equal
5328 #. ``sgt``: signed greater than
5329 #. ``sge``: signed greater or equal
5330 #. ``slt``: signed less than
5331 #. ``sle``: signed less or equal
5333 The remaining two arguments must be :ref:`integer <t_integer>` or
5334 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5335 must also be identical types.
5340 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5341 code given as ``cond``. The comparison performed always yields either an
5342 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5344 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5345 otherwise. No sign interpretation is necessary or performed.
5346 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5347 otherwise. No sign interpretation is necessary or performed.
5348 #. ``ugt``: interprets the operands as unsigned values and yields
5349 ``true`` if ``op1`` is greater than ``op2``.
5350 #. ``uge``: interprets the operands as unsigned values and yields
5351 ``true`` if ``op1`` is greater than or equal to ``op2``.
5352 #. ``ult``: interprets the operands as unsigned values and yields
5353 ``true`` if ``op1`` is less than ``op2``.
5354 #. ``ule``: interprets the operands as unsigned values and yields
5355 ``true`` if ``op1`` is less than or equal to ``op2``.
5356 #. ``sgt``: interprets the operands as signed values and yields ``true``
5357 if ``op1`` is greater than ``op2``.
5358 #. ``sge``: interprets the operands as signed values and yields ``true``
5359 if ``op1`` is greater than or equal to ``op2``.
5360 #. ``slt``: interprets the operands as signed values and yields ``true``
5361 if ``op1`` is less than ``op2``.
5362 #. ``sle``: interprets the operands as signed values and yields ``true``
5363 if ``op1`` is less than or equal to ``op2``.
5365 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5366 are compared as if they were integers.
5368 If the operands are integer vectors, then they are compared element by
5369 element. The result is an ``i1`` vector with the same number of elements
5370 as the values being compared. Otherwise, the result is an ``i1``.
5375 .. code-block:: llvm
5377 <result> = icmp eq i32 4, 5 ; yields: result=false
5378 <result> = icmp ne float* %X, %X ; yields: result=false
5379 <result> = icmp ult i16 4, 5 ; yields: result=true
5380 <result> = icmp sgt i16 4, 5 ; yields: result=false
5381 <result> = icmp ule i16 -4, 5 ; yields: result=false
5382 <result> = icmp sge i16 4, 5 ; yields: result=false
5384 Note that the code generator does not yet support vector types with the
5385 ``icmp`` instruction.
5389 '``fcmp``' Instruction
5390 ^^^^^^^^^^^^^^^^^^^^^^
5397 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5402 The '``fcmp``' instruction returns a boolean value or vector of boolean
5403 values based on comparison of its operands.
5405 If the operands are floating point scalars, then the result type is a
5406 boolean (:ref:`i1 <t_integer>`).
5408 If the operands are floating point vectors, then the result type is a
5409 vector of boolean with the same number of elements as the operands being
5415 The '``fcmp``' instruction takes three operands. The first operand is
5416 the condition code indicating the kind of comparison to perform. It is
5417 not a value, just a keyword. The possible condition code are:
5419 #. ``false``: no comparison, always returns false
5420 #. ``oeq``: ordered and equal
5421 #. ``ogt``: ordered and greater than
5422 #. ``oge``: ordered and greater than or equal
5423 #. ``olt``: ordered and less than
5424 #. ``ole``: ordered and less than or equal
5425 #. ``one``: ordered and not equal
5426 #. ``ord``: ordered (no nans)
5427 #. ``ueq``: unordered or equal
5428 #. ``ugt``: unordered or greater than
5429 #. ``uge``: unordered or greater than or equal
5430 #. ``ult``: unordered or less than
5431 #. ``ule``: unordered or less than or equal
5432 #. ``une``: unordered or not equal
5433 #. ``uno``: unordered (either nans)
5434 #. ``true``: no comparison, always returns true
5436 *Ordered* means that neither operand is a QNAN while *unordered* means
5437 that either operand may be a QNAN.
5439 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5440 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5441 type. They must have identical types.
5446 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5447 condition code given as ``cond``. If the operands are vectors, then the
5448 vectors are compared element by element. Each comparison performed
5449 always yields an :ref:`i1 <t_integer>` result, as follows:
5451 #. ``false``: always yields ``false``, regardless of operands.
5452 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5453 is equal to ``op2``.
5454 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5455 is greater than ``op2``.
5456 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5457 is greater than or equal to ``op2``.
5458 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5459 is less than ``op2``.
5460 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5461 is less than or equal to ``op2``.
5462 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5463 is not equal to ``op2``.
5464 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5465 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5467 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5468 greater than ``op2``.
5469 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5470 greater than or equal to ``op2``.
5471 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5473 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5474 less than or equal to ``op2``.
5475 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5476 not equal to ``op2``.
5477 #. ``uno``: yields ``true`` if either operand is a QNAN.
5478 #. ``true``: always yields ``true``, regardless of operands.
5483 .. code-block:: llvm
5485 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5486 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5487 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5488 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5490 Note that the code generator does not yet support vector types with the
5491 ``fcmp`` instruction.
5495 '``phi``' Instruction
5496 ^^^^^^^^^^^^^^^^^^^^^
5503 <result> = phi <ty> [ <val0>, <label0>], ...
5508 The '``phi``' instruction is used to implement the φ node in the SSA
5509 graph representing the function.
5514 The type of the incoming values is specified with the first type field.
5515 After this, the '``phi``' instruction takes a list of pairs as
5516 arguments, with one pair for each predecessor basic block of the current
5517 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5518 the value arguments to the PHI node. Only labels may be used as the
5521 There must be no non-phi instructions between the start of a basic block
5522 and the PHI instructions: i.e. PHI instructions must be first in a basic
5525 For the purposes of the SSA form, the use of each incoming value is
5526 deemed to occur on the edge from the corresponding predecessor block to
5527 the current block (but after any definition of an '``invoke``'
5528 instruction's return value on the same edge).
5533 At runtime, the '``phi``' instruction logically takes on the value
5534 specified by the pair corresponding to the predecessor basic block that
5535 executed just prior to the current block.
5540 .. code-block:: llvm
5542 Loop: ; Infinite loop that counts from 0 on up...
5543 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5544 %nextindvar = add i32 %indvar, 1
5549 '``select``' Instruction
5550 ^^^^^^^^^^^^^^^^^^^^^^^^
5557 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5559 selty is either i1 or {<N x i1>}
5564 The '``select``' instruction is used to choose one value based on a
5565 condition, without branching.
5570 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5571 values indicating the condition, and two values of the same :ref:`first
5572 class <t_firstclass>` type. If the val1/val2 are vectors and the
5573 condition is a scalar, then entire vectors are selected, not individual
5579 If the condition is an i1 and it evaluates to 1, the instruction returns
5580 the first value argument; otherwise, it returns the second value
5583 If the condition is a vector of i1, then the value arguments must be
5584 vectors of the same size, and the selection is done element by element.
5589 .. code-block:: llvm
5591 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5595 '``call``' Instruction
5596 ^^^^^^^^^^^^^^^^^^^^^^
5603 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5608 The '``call``' instruction represents a simple function call.
5613 This instruction requires several arguments:
5615 #. The optional "tail" marker indicates that the callee function does
5616 not access any allocas or varargs in the caller. Note that calls may
5617 be marked "tail" even if they do not occur before a
5618 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5619 function call is eligible for tail call optimization, but `might not
5620 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5621 The code generator may optimize calls marked "tail" with either 1)
5622 automatic `sibling call
5623 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5624 callee have matching signatures, or 2) forced tail call optimization
5625 when the following extra requirements are met:
5627 - Caller and callee both have the calling convention ``fastcc``.
5628 - The call is in tail position (ret immediately follows call and ret
5629 uses value of call or is void).
5630 - Option ``-tailcallopt`` is enabled, or
5631 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5632 - `Platform specific constraints are
5633 met. <CodeGenerator.html#tailcallopt>`_
5635 #. The optional "cconv" marker indicates which :ref:`calling
5636 convention <callingconv>` the call should use. If none is
5637 specified, the call defaults to using C calling conventions. The
5638 calling convention of the call must match the calling convention of
5639 the target function, or else the behavior is undefined.
5640 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5641 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5643 #. '``ty``': the type of the call instruction itself which is also the
5644 type of the return value. Functions that return no value are marked
5646 #. '``fnty``': shall be the signature of the pointer to function value
5647 being invoked. The argument types must match the types implied by
5648 this signature. This type can be omitted if the function is not
5649 varargs and if the function type does not return a pointer to a
5651 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5652 be invoked. In most cases, this is a direct function invocation, but
5653 indirect ``call``'s are just as possible, calling an arbitrary pointer
5655 #. '``function args``': argument list whose types match the function
5656 signature argument types and parameter attributes. All arguments must
5657 be of :ref:`first class <t_firstclass>` type. If the function signature
5658 indicates the function accepts a variable number of arguments, the
5659 extra arguments can be specified.
5660 #. The optional :ref:`function attributes <fnattrs>` list. Only
5661 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5662 attributes are valid here.
5667 The '``call``' instruction is used to cause control flow to transfer to
5668 a specified function, with its incoming arguments bound to the specified
5669 values. Upon a '``ret``' instruction in the called function, control
5670 flow continues with the instruction after the function call, and the
5671 return value of the function is bound to the result argument.
5676 .. code-block:: llvm
5678 %retval = call i32 @test(i32 %argc)
5679 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5680 %X = tail call i32 @foo() ; yields i32
5681 %Y = tail call fastcc i32 @foo() ; yields i32
5682 call void %foo(i8 97 signext)
5684 %struct.A = type { i32, i8 }
5685 %r = call %struct.A @foo() ; yields { 32, i8 }
5686 %gr = extractvalue %struct.A %r, 0 ; yields i32
5687 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5688 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5689 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5691 llvm treats calls to some functions with names and arguments that match
5692 the standard C99 library as being the C99 library functions, and may
5693 perform optimizations or generate code for them under that assumption.
5694 This is something we'd like to change in the future to provide better
5695 support for freestanding environments and non-C-based languages.
5699 '``va_arg``' Instruction
5700 ^^^^^^^^^^^^^^^^^^^^^^^^
5707 <resultval> = va_arg <va_list*> <arglist>, <argty>
5712 The '``va_arg``' instruction is used to access arguments passed through
5713 the "variable argument" area of a function call. It is used to implement
5714 the ``va_arg`` macro in C.
5719 This instruction takes a ``va_list*`` value and the type of the
5720 argument. It returns a value of the specified argument type and
5721 increments the ``va_list`` to point to the next argument. The actual
5722 type of ``va_list`` is target specific.
5727 The '``va_arg``' instruction loads an argument of the specified type
5728 from the specified ``va_list`` and causes the ``va_list`` to point to
5729 the next argument. For more information, see the variable argument
5730 handling :ref:`Intrinsic Functions <int_varargs>`.
5732 It is legal for this instruction to be called in a function which does
5733 not take a variable number of arguments, for example, the ``vfprintf``
5736 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5737 function <intrinsics>` because it takes a type as an argument.
5742 See the :ref:`variable argument processing <int_varargs>` section.
5744 Note that the code generator does not yet fully support va\_arg on many
5745 targets. Also, it does not currently support va\_arg with aggregate
5746 types on any target.
5750 '``landingpad``' Instruction
5751 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5758 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
5759 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
5761 <clause> := catch <type> <value>
5762 <clause> := filter <array constant type> <array constant>
5767 The '``landingpad``' instruction is used by `LLVM's exception handling
5768 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5769 is a landing pad — one where the exception lands, and corresponds to the
5770 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
5771 defines values supplied by the personality function (``pers_fn``) upon
5772 re-entry to the function. The ``resultval`` has the type ``resultty``.
5777 This instruction takes a ``pers_fn`` value. This is the personality
5778 function associated with the unwinding mechanism. The optional
5779 ``cleanup`` flag indicates that the landing pad block is a cleanup.
5781 A ``clause`` begins with the clause type — ``catch`` or ``filter`` — and
5782 contains the global variable representing the "type" that may be caught
5783 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
5784 clause takes an array constant as its argument. Use
5785 "``[0 x i8**] undef``" for a filter which cannot throw. The
5786 '``landingpad``' instruction must contain *at least* one ``clause`` or
5787 the ``cleanup`` flag.
5792 The '``landingpad``' instruction defines the values which are set by the
5793 personality function (``pers_fn``) upon re-entry to the function, and
5794 therefore the "result type" of the ``landingpad`` instruction. As with
5795 calling conventions, how the personality function results are
5796 represented in LLVM IR is target specific.
5798 The clauses are applied in order from top to bottom. If two
5799 ``landingpad`` instructions are merged together through inlining, the
5800 clauses from the calling function are appended to the list of clauses.
5801 When the call stack is being unwound due to an exception being thrown,
5802 the exception is compared against each ``clause`` in turn. If it doesn't
5803 match any of the clauses, and the ``cleanup`` flag is not set, then
5804 unwinding continues further up the call stack.
5806 The ``landingpad`` instruction has several restrictions:
5808 - A landing pad block is a basic block which is the unwind destination
5809 of an '``invoke``' instruction.
5810 - A landing pad block must have a '``landingpad``' instruction as its
5811 first non-PHI instruction.
5812 - There can be only one '``landingpad``' instruction within the landing
5814 - A basic block that is not a landing pad block may not include a
5815 '``landingpad``' instruction.
5816 - All '``landingpad``' instructions in a function must have the same
5817 personality function.
5822 .. code-block:: llvm
5824 ;; A landing pad which can catch an integer.
5825 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5827 ;; A landing pad that is a cleanup.
5828 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5830 ;; A landing pad which can catch an integer and can only throw a double.
5831 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5833 filter [1 x i8**] [@_ZTId]
5840 LLVM supports the notion of an "intrinsic function". These functions
5841 have well known names and semantics and are required to follow certain
5842 restrictions. Overall, these intrinsics represent an extension mechanism
5843 for the LLVM language that does not require changing all of the
5844 transformations in LLVM when adding to the language (or the bitcode
5845 reader/writer, the parser, etc...).
5847 Intrinsic function names must all start with an "``llvm.``" prefix. This
5848 prefix is reserved in LLVM for intrinsic names; thus, function names may
5849 not begin with this prefix. Intrinsic functions must always be external
5850 functions: you cannot define the body of intrinsic functions. Intrinsic
5851 functions may only be used in call or invoke instructions: it is illegal
5852 to take the address of an intrinsic function. Additionally, because
5853 intrinsic functions are part of the LLVM language, it is required if any
5854 are added that they be documented here.
5856 Some intrinsic functions can be overloaded, i.e., the intrinsic
5857 represents a family of functions that perform the same operation but on
5858 different data types. Because LLVM can represent over 8 million
5859 different integer types, overloading is used commonly to allow an
5860 intrinsic function to operate on any integer type. One or more of the
5861 argument types or the result type can be overloaded to accept any
5862 integer type. Argument types may also be defined as exactly matching a
5863 previous argument's type or the result type. This allows an intrinsic
5864 function which accepts multiple arguments, but needs all of them to be
5865 of the same type, to only be overloaded with respect to a single
5866 argument or the result.
5868 Overloaded intrinsics will have the names of its overloaded argument
5869 types encoded into its function name, each preceded by a period. Only
5870 those types which are overloaded result in a name suffix. Arguments
5871 whose type is matched against another type do not. For example, the
5872 ``llvm.ctpop`` function can take an integer of any width and returns an
5873 integer of exactly the same integer width. This leads to a family of
5874 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
5875 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
5876 overloaded, and only one type suffix is required. Because the argument's
5877 type is matched against the return type, it does not require its own
5880 To learn how to add an intrinsic function, please see the `Extending
5881 LLVM Guide <ExtendingLLVM.html>`_.
5885 Variable Argument Handling Intrinsics
5886 -------------------------------------
5888 Variable argument support is defined in LLVM with the
5889 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
5890 functions. These functions are related to the similarly named macros
5891 defined in the ``<stdarg.h>`` header file.
5893 All of these functions operate on arguments that use a target-specific
5894 value type "``va_list``". The LLVM assembly language reference manual
5895 does not define what this type is, so all transformations should be
5896 prepared to handle these functions regardless of the type used.
5898 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
5899 variable argument handling intrinsic functions are used.
5901 .. code-block:: llvm
5903 define i32 @test(i32 %X, ...) {
5904 ; Initialize variable argument processing
5906 %ap2 = bitcast i8** %ap to i8*
5907 call void @llvm.va_start(i8* %ap2)
5909 ; Read a single integer argument
5910 %tmp = va_arg i8** %ap, i32
5912 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5914 %aq2 = bitcast i8** %aq to i8*
5915 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5916 call void @llvm.va_end(i8* %aq2)
5918 ; Stop processing of arguments.
5919 call void @llvm.va_end(i8* %ap2)
5923 declare void @llvm.va_start(i8*)
5924 declare void @llvm.va_copy(i8*, i8*)
5925 declare void @llvm.va_end(i8*)
5929 '``llvm.va_start``' Intrinsic
5930 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5937 declare void %llvm.va_start(i8* <arglist>)
5942 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
5943 subsequent use by ``va_arg``.
5948 The argument is a pointer to a ``va_list`` element to initialize.
5953 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
5954 available in C. In a target-dependent way, it initializes the
5955 ``va_list`` element to which the argument points, so that the next call
5956 to ``va_arg`` will produce the first variable argument passed to the
5957 function. Unlike the C ``va_start`` macro, this intrinsic does not need
5958 to know the last argument of the function as the compiler can figure
5961 '``llvm.va_end``' Intrinsic
5962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5969 declare void @llvm.va_end(i8* <arglist>)
5974 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
5975 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
5980 The argument is a pointer to a ``va_list`` to destroy.
5985 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
5986 available in C. In a target-dependent way, it destroys the ``va_list``
5987 element to which the argument points. Calls to
5988 :ref:`llvm.va_start <int_va_start>` and
5989 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
5994 '``llvm.va_copy``' Intrinsic
5995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6002 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6007 The '``llvm.va_copy``' intrinsic copies the current argument position
6008 from the source argument list to the destination argument list.
6013 The first argument is a pointer to a ``va_list`` element to initialize.
6014 The second argument is a pointer to a ``va_list`` element to copy from.
6019 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6020 available in C. In a target-dependent way, it copies the source
6021 ``va_list`` element into the destination ``va_list`` element. This
6022 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6023 arbitrarily complex and require, for example, memory allocation.
6025 Accurate Garbage Collection Intrinsics
6026 --------------------------------------
6028 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6029 (GC) requires the implementation and generation of these intrinsics.
6030 These intrinsics allow identification of :ref:`GC roots on the
6031 stack <int_gcroot>`, as well as garbage collector implementations that
6032 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6033 Front-ends for type-safe garbage collected languages should generate
6034 these intrinsics to make use of the LLVM garbage collectors. For more
6035 details, see `Accurate Garbage Collection with
6036 LLVM <GarbageCollection.html>`_.
6038 The garbage collection intrinsics only operate on objects in the generic
6039 address space (address space zero).
6043 '``llvm.gcroot``' Intrinsic
6044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6051 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6056 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6057 the code generator, and allows some metadata to be associated with it.
6062 The first argument specifies the address of a stack object that contains
6063 the root pointer. The second pointer (which must be either a constant or
6064 a global value address) contains the meta-data to be associated with the
6070 At runtime, a call to this intrinsic stores a null pointer into the
6071 "ptrloc" location. At compile-time, the code generator generates
6072 information to allow the runtime to find the pointer at GC safe points.
6073 The '``llvm.gcroot``' intrinsic may only be used in a function which
6074 :ref:`specifies a GC algorithm <gc>`.
6078 '``llvm.gcread``' Intrinsic
6079 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6086 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6091 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6092 locations, allowing garbage collector implementations that require read
6098 The second argument is the address to read from, which should be an
6099 address allocated from the garbage collector. The first object is a
6100 pointer to the start of the referenced object, if needed by the language
6101 runtime (otherwise null).
6106 The '``llvm.gcread``' intrinsic has the same semantics as a load
6107 instruction, but may be replaced with substantially more complex code by
6108 the garbage collector runtime, as needed. The '``llvm.gcread``'
6109 intrinsic may only be used in a function which :ref:`specifies a GC
6114 '``llvm.gcwrite``' Intrinsic
6115 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6122 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6127 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6128 locations, allowing garbage collector implementations that require write
6129 barriers (such as generational or reference counting collectors).
6134 The first argument is the reference to store, the second is the start of
6135 the object to store it to, and the third is the address of the field of
6136 Obj to store to. If the runtime does not require a pointer to the
6137 object, Obj may be null.
6142 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6143 instruction, but may be replaced with substantially more complex code by
6144 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6145 intrinsic may only be used in a function which :ref:`specifies a GC
6148 Code Generator Intrinsics
6149 -------------------------
6151 These intrinsics are provided by LLVM to expose special features that
6152 may only be implemented with code generator support.
6154 '``llvm.returnaddress``' Intrinsic
6155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6162 declare i8 *@llvm.returnaddress(i32 <level>)
6167 The '``llvm.returnaddress``' intrinsic attempts to compute a
6168 target-specific value indicating the return address of the current
6169 function or one of its callers.
6174 The argument to this intrinsic indicates which function to return the
6175 address for. Zero indicates the calling function, one indicates its
6176 caller, etc. The argument is **required** to be a constant integer
6182 The '``llvm.returnaddress``' intrinsic either returns a pointer
6183 indicating the return address of the specified call frame, or zero if it
6184 cannot be identified. The value returned by this intrinsic is likely to
6185 be incorrect or 0 for arguments other than zero, so it should only be
6186 used for debugging purposes.
6188 Note that calling this intrinsic does not prevent function inlining or
6189 other aggressive transformations, so the value returned may not be that
6190 of the obvious source-language caller.
6192 '``llvm.frameaddress``' Intrinsic
6193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6200 declare i8* @llvm.frameaddress(i32 <level>)
6205 The '``llvm.frameaddress``' intrinsic attempts to return the
6206 target-specific frame pointer value for the specified stack frame.
6211 The argument to this intrinsic indicates which function to return the
6212 frame pointer for. Zero indicates the calling function, one indicates
6213 its caller, etc. The argument is **required** to be a constant integer
6219 The '``llvm.frameaddress``' intrinsic either returns a pointer
6220 indicating the frame address of the specified call frame, or zero if it
6221 cannot be identified. The value returned by this intrinsic is likely to
6222 be incorrect or 0 for arguments other than zero, so it should only be
6223 used for debugging purposes.
6225 Note that calling this intrinsic does not prevent function inlining or
6226 other aggressive transformations, so the value returned may not be that
6227 of the obvious source-language caller.
6231 '``llvm.stacksave``' Intrinsic
6232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6239 declare i8* @llvm.stacksave()
6244 The '``llvm.stacksave``' intrinsic is used to remember the current state
6245 of the function stack, for use with
6246 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6247 implementing language features like scoped automatic variable sized
6253 This intrinsic returns a opaque pointer value that can be passed to
6254 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6255 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6256 ``llvm.stacksave``, it effectively restores the state of the stack to
6257 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6258 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6259 were allocated after the ``llvm.stacksave`` was executed.
6261 .. _int_stackrestore:
6263 '``llvm.stackrestore``' Intrinsic
6264 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6271 declare void @llvm.stackrestore(i8* %ptr)
6276 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6277 the function stack to the state it was in when the corresponding
6278 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6279 useful for implementing language features like scoped automatic variable
6280 sized arrays in C99.
6285 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6287 '``llvm.prefetch``' Intrinsic
6288 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6295 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6300 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6301 insert a prefetch instruction if supported; otherwise, it is a noop.
6302 Prefetches have no effect on the behavior of the program but can change
6303 its performance characteristics.
6308 ``address`` is the address to be prefetched, ``rw`` is the specifier
6309 determining if the fetch should be for a read (0) or write (1), and
6310 ``locality`` is a temporal locality specifier ranging from (0) - no
6311 locality, to (3) - extremely local keep in cache. The ``cache type``
6312 specifies whether the prefetch is performed on the data (1) or
6313 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6314 arguments must be constant integers.
6319 This intrinsic does not modify the behavior of the program. In
6320 particular, prefetches cannot trap and do not produce a value. On
6321 targets that support this intrinsic, the prefetch can provide hints to
6322 the processor cache for better performance.
6324 '``llvm.pcmarker``' Intrinsic
6325 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6332 declare void @llvm.pcmarker(i32 <id>)
6337 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6338 Counter (PC) in a region of code to simulators and other tools. The
6339 method is target specific, but it is expected that the marker will use
6340 exported symbols to transmit the PC of the marker. The marker makes no
6341 guarantees that it will remain with any specific instruction after
6342 optimizations. It is possible that the presence of a marker will inhibit
6343 optimizations. The intended use is to be inserted after optimizations to
6344 allow correlations of simulation runs.
6349 ``id`` is a numerical id identifying the marker.
6354 This intrinsic does not modify the behavior of the program. Backends
6355 that do not support this intrinsic may ignore it.
6357 '``llvm.readcyclecounter``' Intrinsic
6358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6365 declare i64 @llvm.readcyclecounter()
6370 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6371 counter register (or similar low latency, high accuracy clocks) on those
6372 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6373 should map to RPCC. As the backing counters overflow quickly (on the
6374 order of 9 seconds on alpha), this should only be used for small
6380 When directly supported, reading the cycle counter should not modify any
6381 memory. Implementations are allowed to either return a application
6382 specific value or a system wide value. On backends without support, this
6383 is lowered to a constant 0.
6385 Standard C Library Intrinsics
6386 -----------------------------
6388 LLVM provides intrinsics for a few important standard C library
6389 functions. These intrinsics allow source-language front-ends to pass
6390 information about the alignment of the pointer arguments to the code
6391 generator, providing opportunity for more efficient code generation.
6395 '``llvm.memcpy``' Intrinsic
6396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6401 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6402 integer bit width and for different address spaces. Not all targets
6403 support all bit widths however.
6407 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6408 i32 <len>, i32 <align>, i1 <isvolatile>)
6409 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6410 i64 <len>, i32 <align>, i1 <isvolatile>)
6415 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6416 source location to the destination location.
6418 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6419 intrinsics do not return a value, takes extra alignment/isvolatile
6420 arguments and the pointers can be in specified address spaces.
6425 The first argument is a pointer to the destination, the second is a
6426 pointer to the source. The third argument is an integer argument
6427 specifying the number of bytes to copy, the fourth argument is the
6428 alignment of the source and destination locations, and the fifth is a
6429 boolean indicating a volatile access.
6431 If the call to this intrinsic has an alignment value that is not 0 or 1,
6432 then the caller guarantees that both the source and destination pointers
6433 are aligned to that boundary.
6435 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6436 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6437 very cleanly specified and it is unwise to depend on it.
6442 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6443 source location to the destination location, which are not allowed to
6444 overlap. It copies "len" bytes of memory over. If the argument is known
6445 to be aligned to some boundary, this can be specified as the fourth
6446 argument, otherwise it should be set to 0 or 1.
6448 '``llvm.memmove``' Intrinsic
6449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6454 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6455 bit width and for different address space. Not all targets support all
6460 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6461 i32 <len>, i32 <align>, i1 <isvolatile>)
6462 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6463 i64 <len>, i32 <align>, i1 <isvolatile>)
6468 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6469 source location to the destination location. It is similar to the
6470 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6473 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6474 intrinsics do not return a value, takes extra alignment/isvolatile
6475 arguments and the pointers can be in specified address spaces.
6480 The first argument is a pointer to the destination, the second is a
6481 pointer to the source. The third argument is an integer argument
6482 specifying the number of bytes to copy, the fourth argument is the
6483 alignment of the source and destination locations, and the fifth is a
6484 boolean indicating a volatile access.
6486 If the call to this intrinsic has an alignment value that is not 0 or 1,
6487 then the caller guarantees that the source and destination pointers are
6488 aligned to that boundary.
6490 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6491 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6492 not very cleanly specified and it is unwise to depend on it.
6497 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6498 source location to the destination location, which may overlap. It
6499 copies "len" bytes of memory over. If the argument is known to be
6500 aligned to some boundary, this can be specified as the fourth argument,
6501 otherwise it should be set to 0 or 1.
6503 '``llvm.memset.*``' Intrinsics
6504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6509 This is an overloaded intrinsic. You can use llvm.memset on any integer
6510 bit width and for different address spaces. However, not all targets
6511 support all bit widths.
6515 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6516 i32 <len>, i32 <align>, i1 <isvolatile>)
6517 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6518 i64 <len>, i32 <align>, i1 <isvolatile>)
6523 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6524 particular byte value.
6526 Note that, unlike the standard libc function, the ``llvm.memset``
6527 intrinsic does not return a value and takes extra alignment/volatile
6528 arguments. Also, the destination can be in an arbitrary address space.
6533 The first argument is a pointer to the destination to fill, the second
6534 is the byte value with which to fill it, the third argument is an
6535 integer argument specifying the number of bytes to fill, and the fourth
6536 argument is the known alignment of the destination location.
6538 If the call to this intrinsic has an alignment value that is not 0 or 1,
6539 then the caller guarantees that the destination pointer is aligned to
6542 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6543 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6544 very cleanly specified and it is unwise to depend on it.
6549 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6550 at the destination location. If the argument is known to be aligned to
6551 some boundary, this can be specified as the fourth argument, otherwise
6552 it should be set to 0 or 1.
6554 '``llvm.sqrt.*``' Intrinsic
6555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6560 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6561 floating point or vector of floating point type. Not all targets support
6566 declare float @llvm.sqrt.f32(float %Val)
6567 declare double @llvm.sqrt.f64(double %Val)
6568 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6569 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6570 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6575 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6576 returning the same value as the libm '``sqrt``' functions would. Unlike
6577 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6578 negative numbers other than -0.0 (which allows for better optimization,
6579 because there is no need to worry about errno being set).
6580 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6585 The argument and return value are floating point numbers of the same
6591 This function returns the sqrt of the specified operand if it is a
6592 nonnegative floating point number.
6594 '``llvm.powi.*``' Intrinsic
6595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6600 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6601 floating point or vector of floating point type. Not all targets support
6606 declare float @llvm.powi.f32(float %Val, i32 %power)
6607 declare double @llvm.powi.f64(double %Val, i32 %power)
6608 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6609 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6610 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6615 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6616 specified (positive or negative) power. The order of evaluation of
6617 multiplications is not defined. When a vector of floating point type is
6618 used, the second argument remains a scalar integer value.
6623 The second argument is an integer power, and the first is a value to
6624 raise to that power.
6629 This function returns the first value raised to the second power with an
6630 unspecified sequence of rounding operations.
6632 '``llvm.sin.*``' Intrinsic
6633 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6638 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6639 floating point or vector of floating point type. Not all targets support
6644 declare float @llvm.sin.f32(float %Val)
6645 declare double @llvm.sin.f64(double %Val)
6646 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6647 declare fp128 @llvm.sin.f128(fp128 %Val)
6648 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6653 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6658 The argument and return value are floating point numbers of the same
6664 This function returns the sine of the specified operand, returning the
6665 same values as the libm ``sin`` functions would, and handles error
6666 conditions in the same way.
6668 '``llvm.cos.*``' Intrinsic
6669 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6674 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6675 floating point or vector of floating point type. Not all targets support
6680 declare float @llvm.cos.f32(float %Val)
6681 declare double @llvm.cos.f64(double %Val)
6682 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6683 declare fp128 @llvm.cos.f128(fp128 %Val)
6684 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6689 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6694 The argument and return value are floating point numbers of the same
6700 This function returns the cosine of the specified operand, returning the
6701 same values as the libm ``cos`` functions would, and handles error
6702 conditions in the same way.
6704 '``llvm.pow.*``' Intrinsic
6705 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6710 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6711 floating point or vector of floating point type. Not all targets support
6716 declare float @llvm.pow.f32(float %Val, float %Power)
6717 declare double @llvm.pow.f64(double %Val, double %Power)
6718 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6719 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6720 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6725 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6726 specified (positive or negative) power.
6731 The second argument is a floating point power, and the first is a value
6732 to raise to that power.
6737 This function returns the first value raised to the second power,
6738 returning the same values as the libm ``pow`` functions would, and
6739 handles error conditions in the same way.
6741 '``llvm.exp.*``' Intrinsic
6742 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6747 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
6748 floating point or vector of floating point type. Not all targets support
6753 declare float @llvm.exp.f32(float %Val)
6754 declare double @llvm.exp.f64(double %Val)
6755 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6756 declare fp128 @llvm.exp.f128(fp128 %Val)
6757 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6762 The '``llvm.exp.*``' intrinsics perform the exp function.
6767 The argument and return value are floating point numbers of the same
6773 This function returns the same values as the libm ``exp`` functions
6774 would, and handles error conditions in the same way.
6776 '``llvm.exp2.*``' Intrinsic
6777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6782 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
6783 floating point or vector of floating point type. Not all targets support
6788 declare float @llvm.exp2.f32(float %Val)
6789 declare double @llvm.exp2.f64(double %Val)
6790 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
6791 declare fp128 @llvm.exp2.f128(fp128 %Val)
6792 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
6797 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
6802 The argument and return value are floating point numbers of the same
6808 This function returns the same values as the libm ``exp2`` functions
6809 would, and handles error conditions in the same way.
6811 '``llvm.log.*``' Intrinsic
6812 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6817 This is an overloaded intrinsic. You can use ``llvm.log`` on any
6818 floating point or vector of floating point type. Not all targets support
6823 declare float @llvm.log.f32(float %Val)
6824 declare double @llvm.log.f64(double %Val)
6825 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6826 declare fp128 @llvm.log.f128(fp128 %Val)
6827 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6832 The '``llvm.log.*``' intrinsics perform the log function.
6837 The argument and return value are floating point numbers of the same
6843 This function returns the same values as the libm ``log`` functions
6844 would, and handles error conditions in the same way.
6846 '``llvm.log10.*``' Intrinsic
6847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6852 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
6853 floating point or vector of floating point type. Not all targets support
6858 declare float @llvm.log10.f32(float %Val)
6859 declare double @llvm.log10.f64(double %Val)
6860 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
6861 declare fp128 @llvm.log10.f128(fp128 %Val)
6862 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
6867 The '``llvm.log10.*``' intrinsics perform the log10 function.
6872 The argument and return value are floating point numbers of the same
6878 This function returns the same values as the libm ``log10`` functions
6879 would, and handles error conditions in the same way.
6881 '``llvm.log2.*``' Intrinsic
6882 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6887 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
6888 floating point or vector of floating point type. Not all targets support
6893 declare float @llvm.log2.f32(float %Val)
6894 declare double @llvm.log2.f64(double %Val)
6895 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
6896 declare fp128 @llvm.log2.f128(fp128 %Val)
6897 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
6902 The '``llvm.log2.*``' intrinsics perform the log2 function.
6907 The argument and return value are floating point numbers of the same
6913 This function returns the same values as the libm ``log2`` functions
6914 would, and handles error conditions in the same way.
6916 '``llvm.fma.*``' Intrinsic
6917 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6922 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
6923 floating point or vector of floating point type. Not all targets support
6928 declare float @llvm.fma.f32(float %a, float %b, float %c)
6929 declare double @llvm.fma.f64(double %a, double %b, double %c)
6930 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
6931 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
6932 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
6937 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
6943 The argument and return value are floating point numbers of the same
6949 This function returns the same values as the libm ``fma`` functions
6952 '``llvm.fabs.*``' Intrinsic
6953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6958 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
6959 floating point or vector of floating point type. Not all targets support
6964 declare float @llvm.fabs.f32(float %Val)
6965 declare double @llvm.fabs.f64(double %Val)
6966 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
6967 declare fp128 @llvm.fabs.f128(fp128 %Val)
6968 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
6973 The '``llvm.fabs.*``' intrinsics return the absolute value of the
6979 The argument and return value are floating point numbers of the same
6985 This function returns the same values as the libm ``fabs`` functions
6986 would, and handles error conditions in the same way.
6988 '``llvm.floor.*``' Intrinsic
6989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6994 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
6995 floating point or vector of floating point type. Not all targets support
7000 declare float @llvm.floor.f32(float %Val)
7001 declare double @llvm.floor.f64(double %Val)
7002 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7003 declare fp128 @llvm.floor.f128(fp128 %Val)
7004 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7009 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7014 The argument and return value are floating point numbers of the same
7020 This function returns the same values as the libm ``floor`` functions
7021 would, and handles error conditions in the same way.
7023 '``llvm.ceil.*``' Intrinsic
7024 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7029 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7030 floating point or vector of floating point type. Not all targets support
7035 declare float @llvm.ceil.f32(float %Val)
7036 declare double @llvm.ceil.f64(double %Val)
7037 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7038 declare fp128 @llvm.ceil.f128(fp128 %Val)
7039 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7044 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7049 The argument and return value are floating point numbers of the same
7055 This function returns the same values as the libm ``ceil`` functions
7056 would, and handles error conditions in the same way.
7058 '``llvm.trunc.*``' Intrinsic
7059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7064 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7065 floating point or vector of floating point type. Not all targets support
7070 declare float @llvm.trunc.f32(float %Val)
7071 declare double @llvm.trunc.f64(double %Val)
7072 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7073 declare fp128 @llvm.trunc.f128(fp128 %Val)
7074 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7079 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7080 nearest integer not larger in magnitude than the operand.
7085 The argument and return value are floating point numbers of the same
7091 This function returns the same values as the libm ``trunc`` functions
7092 would, and handles error conditions in the same way.
7094 '``llvm.rint.*``' Intrinsic
7095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7100 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7101 floating point or vector of floating point type. Not all targets support
7106 declare float @llvm.rint.f32(float %Val)
7107 declare double @llvm.rint.f64(double %Val)
7108 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7109 declare fp128 @llvm.rint.f128(fp128 %Val)
7110 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7115 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7116 nearest integer. It may raise an inexact floating-point exception if the
7117 operand isn't an integer.
7122 The argument and return value are floating point numbers of the same
7128 This function returns the same values as the libm ``rint`` functions
7129 would, and handles error conditions in the same way.
7131 '``llvm.nearbyint.*``' Intrinsic
7132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7137 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7138 floating point or vector of floating point type. Not all targets support
7143 declare float @llvm.nearbyint.f32(float %Val)
7144 declare double @llvm.nearbyint.f64(double %Val)
7145 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7146 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7147 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7152 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7158 The argument and return value are floating point numbers of the same
7164 This function returns the same values as the libm ``nearbyint``
7165 functions would, and handles error conditions in the same way.
7167 Bit Manipulation Intrinsics
7168 ---------------------------
7170 LLVM provides intrinsics for a few important bit manipulation
7171 operations. These allow efficient code generation for some algorithms.
7173 '``llvm.bswap.*``' Intrinsics
7174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7179 This is an overloaded intrinsic function. You can use bswap on any
7180 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7184 declare i16 @llvm.bswap.i16(i16 <id>)
7185 declare i32 @llvm.bswap.i32(i32 <id>)
7186 declare i64 @llvm.bswap.i64(i64 <id>)
7191 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7192 values with an even number of bytes (positive multiple of 16 bits).
7193 These are useful for performing operations on data that is not in the
7194 target's native byte order.
7199 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7200 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7201 intrinsic returns an i32 value that has the four bytes of the input i32
7202 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7203 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7204 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7205 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7208 '``llvm.ctpop.*``' Intrinsic
7209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7214 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7215 bit width, or on any vector with integer elements. Not all targets
7216 support all bit widths or vector types, however.
7220 declare i8 @llvm.ctpop.i8(i8 <src>)
7221 declare i16 @llvm.ctpop.i16(i16 <src>)
7222 declare i32 @llvm.ctpop.i32(i32 <src>)
7223 declare i64 @llvm.ctpop.i64(i64 <src>)
7224 declare i256 @llvm.ctpop.i256(i256 <src>)
7225 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7230 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7236 The only argument is the value to be counted. The argument may be of any
7237 integer type, or a vector with integer elements. The return type must
7238 match the argument type.
7243 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7244 each element of a vector.
7246 '``llvm.ctlz.*``' Intrinsic
7247 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7252 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7253 integer bit width, or any vector whose elements are integers. Not all
7254 targets support all bit widths or vector types, however.
7258 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7259 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7260 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7261 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7262 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7263 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7268 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7269 leading zeros in a variable.
7274 The first argument is the value to be counted. This argument may be of
7275 any integer type, or a vectory with integer element type. The return
7276 type must match the first argument type.
7278 The second argument must be a constant and is a flag to indicate whether
7279 the intrinsic should ensure that a zero as the first argument produces a
7280 defined result. Historically some architectures did not provide a
7281 defined result for zero values as efficiently, and many algorithms are
7282 now predicated on avoiding zero-value inputs.
7287 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7288 zeros in a variable, or within each element of the vector. If
7289 ``src == 0`` then the result is the size in bits of the type of ``src``
7290 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7291 ``llvm.ctlz(i32 2) = 30``.
7293 '``llvm.cttz.*``' Intrinsic
7294 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7299 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7300 integer bit width, or any vector of integer elements. Not all targets
7301 support all bit widths or vector types, however.
7305 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7306 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7307 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7308 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7309 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7310 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7315 The '``llvm.cttz``' family of intrinsic functions counts the number of
7321 The first argument is the value to be counted. This argument may be of
7322 any integer type, or a vectory with integer element type. The return
7323 type must match the first argument type.
7325 The second argument must be a constant and is a flag to indicate whether
7326 the intrinsic should ensure that a zero as the first argument produces a
7327 defined result. Historically some architectures did not provide a
7328 defined result for zero values as efficiently, and many algorithms are
7329 now predicated on avoiding zero-value inputs.
7334 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7335 zeros in a variable, or within each element of a vector. If ``src == 0``
7336 then the result is the size in bits of the type of ``src`` if
7337 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7338 ``llvm.cttz(2) = 1``.
7340 Arithmetic with Overflow Intrinsics
7341 -----------------------------------
7343 LLVM provides intrinsics for some arithmetic with overflow operations.
7345 '``llvm.sadd.with.overflow.*``' Intrinsics
7346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7351 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7352 on any integer bit width.
7356 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7357 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7358 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7363 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7364 a signed addition of the two arguments, and indicate whether an overflow
7365 occurred during the signed summation.
7370 The arguments (%a and %b) and the first element of the result structure
7371 may be of integer types of any bit width, but they must have the same
7372 bit width. The second element of the result structure must be of type
7373 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7379 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7380 a signed addition of the two variables. They return a structure — the
7381 first element of which is the signed summation, and the second element
7382 of which is a bit specifying if the signed summation resulted in an
7388 .. code-block:: llvm
7390 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7391 %sum = extractvalue {i32, i1} %res, 0
7392 %obit = extractvalue {i32, i1} %res, 1
7393 br i1 %obit, label %overflow, label %normal
7395 '``llvm.uadd.with.overflow.*``' Intrinsics
7396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7401 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7402 on any integer bit width.
7406 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7407 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7408 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7413 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7414 an unsigned addition of the two arguments, and indicate whether a carry
7415 occurred during the unsigned summation.
7420 The arguments (%a and %b) and the first element of the result structure
7421 may be of integer types of any bit width, but they must have the same
7422 bit width. The second element of the result structure must be of type
7423 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7429 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7430 an unsigned addition of the two arguments. They return a structure — the
7431 first element of which is the sum, and the second element of which is a
7432 bit specifying if the unsigned summation resulted in a carry.
7437 .. code-block:: llvm
7439 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7440 %sum = extractvalue {i32, i1} %res, 0
7441 %obit = extractvalue {i32, i1} %res, 1
7442 br i1 %obit, label %carry, label %normal
7444 '``llvm.ssub.with.overflow.*``' Intrinsics
7445 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7450 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7451 on any integer bit width.
7455 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7456 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7457 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7462 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7463 a signed subtraction of the two arguments, and indicate whether an
7464 overflow occurred during the signed subtraction.
7469 The arguments (%a and %b) and the first element of the result structure
7470 may be of integer types of any bit width, but they must have the same
7471 bit width. The second element of the result structure must be of type
7472 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7478 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7479 a signed subtraction of the two arguments. They return a structure — the
7480 first element of which is the subtraction, and the second element of
7481 which is a bit specifying if the signed subtraction resulted in an
7487 .. code-block:: llvm
7489 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7490 %sum = extractvalue {i32, i1} %res, 0
7491 %obit = extractvalue {i32, i1} %res, 1
7492 br i1 %obit, label %overflow, label %normal
7494 '``llvm.usub.with.overflow.*``' Intrinsics
7495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7500 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7501 on any integer bit width.
7505 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7506 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7507 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7512 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7513 an unsigned subtraction of the two arguments, and indicate whether an
7514 overflow occurred during the unsigned subtraction.
7519 The arguments (%a and %b) and the first element of the result structure
7520 may be of integer types of any bit width, but they must have the same
7521 bit width. The second element of the result structure must be of type
7522 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7528 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7529 an unsigned subtraction of the two arguments. They return a structure —
7530 the first element of which is the subtraction, and the second element of
7531 which is a bit specifying if the unsigned subtraction resulted in an
7537 .. code-block:: llvm
7539 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7540 %sum = extractvalue {i32, i1} %res, 0
7541 %obit = extractvalue {i32, i1} %res, 1
7542 br i1 %obit, label %overflow, label %normal
7544 '``llvm.smul.with.overflow.*``' Intrinsics
7545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7550 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7551 on any integer bit width.
7555 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7556 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7557 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7562 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7563 a signed multiplication of the two arguments, and indicate whether an
7564 overflow occurred during the signed multiplication.
7569 The arguments (%a and %b) and the first element of the result structure
7570 may be of integer types of any bit width, but they must have the same
7571 bit width. The second element of the result structure must be of type
7572 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7578 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7579 a signed multiplication of the two arguments. They return a structure —
7580 the first element of which is the multiplication, and the second element
7581 of which is a bit specifying if the signed multiplication resulted in an
7587 .. code-block:: llvm
7589 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7590 %sum = extractvalue {i32, i1} %res, 0
7591 %obit = extractvalue {i32, i1} %res, 1
7592 br i1 %obit, label %overflow, label %normal
7594 '``llvm.umul.with.overflow.*``' Intrinsics
7595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7600 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7601 on any integer bit width.
7605 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7606 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7607 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7612 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7613 a unsigned multiplication of the two arguments, and indicate whether an
7614 overflow occurred during the unsigned multiplication.
7619 The arguments (%a and %b) and the first element of the result structure
7620 may be of integer types of any bit width, but they must have the same
7621 bit width. The second element of the result structure must be of type
7622 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7628 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7629 an unsigned multiplication of the two arguments. They return a structure
7630 — the first element of which is the multiplication, and the second
7631 element of which is a bit specifying if the unsigned multiplication
7632 resulted in an overflow.
7637 .. code-block:: llvm
7639 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7640 %sum = extractvalue {i32, i1} %res, 0
7641 %obit = extractvalue {i32, i1} %res, 1
7642 br i1 %obit, label %overflow, label %normal
7644 Specialised Arithmetic Intrinsics
7645 ---------------------------------
7647 '``llvm.fmuladd.*``' Intrinsic
7648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7655 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7656 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7661 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7662 expressions that can be fused if the code generator determines that the
7663 fused expression would be legal and efficient.
7668 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7669 multiplicands, a and b, and an addend c.
7678 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7680 is equivalent to the expression a \* b + c, except that rounding will
7681 not be performed between the multiplication and addition steps if the
7682 code generator fuses the operations. Fusion is not guaranteed, even if
7683 the target platform supports it. If a fused multiply-add is required the
7684 corresponding llvm.fma.\* intrinsic function should be used instead.
7689 .. code-block:: llvm
7691 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7693 Half Precision Floating Point Intrinsics
7694 ----------------------------------------
7696 For most target platforms, half precision floating point is a
7697 storage-only format. This means that it is a dense encoding (in memory)
7698 but does not support computation in the format.
7700 This means that code must first load the half-precision floating point
7701 value as an i16, then convert it to float with
7702 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7703 then be performed on the float value (including extending to double
7704 etc). To store the value back to memory, it is first converted to float
7705 if needed, then converted to i16 with
7706 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7709 .. _int_convert_to_fp16:
7711 '``llvm.convert.to.fp16``' Intrinsic
7712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7719 declare i16 @llvm.convert.to.fp16(f32 %a)
7724 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7725 from single precision floating point format to half precision floating
7731 The intrinsic function contains single argument - the value to be
7737 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7738 from single precision floating point format to half precision floating
7739 point format. The return value is an ``i16`` which contains the
7745 .. code-block:: llvm
7747 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7748 store i16 %res, i16* @x, align 2
7750 .. _int_convert_from_fp16:
7752 '``llvm.convert.from.fp16``' Intrinsic
7753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7760 declare f32 @llvm.convert.from.fp16(i16 %a)
7765 The '``llvm.convert.from.fp16``' intrinsic function performs a
7766 conversion from half precision floating point format to single precision
7767 floating point format.
7772 The intrinsic function contains single argument - the value to be
7778 The '``llvm.convert.from.fp16``' intrinsic function performs a
7779 conversion from half single precision floating point format to single
7780 precision floating point format. The input half-float value is
7781 represented by an ``i16`` value.
7786 .. code-block:: llvm
7788 %a = load i16* @x, align 2
7789 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7794 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
7795 prefix), are described in the `LLVM Source Level
7796 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
7799 Exception Handling Intrinsics
7800 -----------------------------
7802 The LLVM exception handling intrinsics (which all start with
7803 ``llvm.eh.`` prefix), are described in the `LLVM Exception
7804 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
7808 Trampoline Intrinsics
7809 ---------------------
7811 These intrinsics make it possible to excise one parameter, marked with
7812 the :ref:`nest <nest>` attribute, from a function. The result is a
7813 callable function pointer lacking the nest parameter - the caller does
7814 not need to provide a value for it. Instead, the value to use is stored
7815 in advance in a "trampoline", a block of memory usually allocated on the
7816 stack, which also contains code to splice the nest value into the
7817 argument list. This is used to implement the GCC nested function address
7820 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
7821 then the resulting function pointer has signature ``i32 (i32, i32)*``.
7822 It can be created as follows:
7824 .. code-block:: llvm
7826 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7827 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7828 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7829 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7830 %fp = bitcast i8* %p to i32 (i32, i32)*
7832 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
7833 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
7837 '``llvm.init.trampoline``' Intrinsic
7838 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7845 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7850 This fills the memory pointed to by ``tramp`` with executable code,
7851 turning it into a trampoline.
7856 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
7857 pointers. The ``tramp`` argument must point to a sufficiently large and
7858 sufficiently aligned block of memory; this memory is written to by the
7859 intrinsic. Note that the size and the alignment are target-specific -
7860 LLVM currently provides no portable way of determining them, so a
7861 front-end that generates this intrinsic needs to have some
7862 target-specific knowledge. The ``func`` argument must hold a function
7863 bitcast to an ``i8*``.
7868 The block of memory pointed to by ``tramp`` is filled with target
7869 dependent code, turning it into a function. Then ``tramp`` needs to be
7870 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
7871 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
7872 function's signature is the same as that of ``func`` with any arguments
7873 marked with the ``nest`` attribute removed. At most one such ``nest``
7874 argument is allowed, and it must be of pointer type. Calling the new
7875 function is equivalent to calling ``func`` with the same argument list,
7876 but with ``nval`` used for the missing ``nest`` argument. If, after
7877 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
7878 modified, then the effect of any later call to the returned function
7879 pointer is undefined.
7883 '``llvm.adjust.trampoline``' Intrinsic
7884 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7891 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7896 This performs any required machine-specific adjustment to the address of
7897 a trampoline (passed as ``tramp``).
7902 ``tramp`` must point to a block of memory which already has trampoline
7903 code filled in by a previous call to
7904 :ref:`llvm.init.trampoline <int_it>`.
7909 On some architectures the address of the code to be executed needs to be
7910 different to the address where the trampoline is actually stored. This
7911 intrinsic returns the executable address corresponding to ``tramp``
7912 after performing the required machine specific adjustments. The pointer
7913 returned can then be :ref:`bitcast and executed <int_trampoline>`.
7918 This class of intrinsics exists to information about the lifetime of
7919 memory objects and ranges where variables are immutable.
7921 '``llvm.lifetime.start``' Intrinsic
7922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7929 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7934 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
7940 The first argument is a constant integer representing the size of the
7941 object, or -1 if it is variable sized. The second argument is a pointer
7947 This intrinsic indicates that before this point in the code, the value
7948 of the memory pointed to by ``ptr`` is dead. This means that it is known
7949 to never be used and has an undefined value. A load from the pointer
7950 that precedes this intrinsic can be replaced with ``'undef'``.
7952 '``llvm.lifetime.end``' Intrinsic
7953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7960 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7965 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
7971 The first argument is a constant integer representing the size of the
7972 object, or -1 if it is variable sized. The second argument is a pointer
7978 This intrinsic indicates that after this point in the code, the value of
7979 the memory pointed to by ``ptr`` is dead. This means that it is known to
7980 never be used and has an undefined value. Any stores into the memory
7981 object following this intrinsic may be removed as dead.
7983 '``llvm.invariant.start``' Intrinsic
7984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7991 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
7996 The '``llvm.invariant.start``' intrinsic specifies that the contents of
7997 a memory object will not change.
8002 The first argument is a constant integer representing the size of the
8003 object, or -1 if it is variable sized. The second argument is a pointer
8009 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8010 the return value, the referenced memory location is constant and
8013 '``llvm.invariant.end``' Intrinsic
8014 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8021 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8026 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8027 memory object are mutable.
8032 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8033 The second argument is a constant integer representing the size of the
8034 object, or -1 if it is variable sized and the third argument is a
8035 pointer to the object.
8040 This intrinsic indicates that the memory is mutable again.
8045 This class of intrinsics is designed to be generic and has no specific
8048 '``llvm.var.annotation``' Intrinsic
8049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8056 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8061 The '``llvm.var.annotation``' intrinsic.
8066 The first argument is a pointer to a value, the second is a pointer to a
8067 global string, the third is a pointer to a global string which is the
8068 source file name, and the last argument is the line number.
8073 This intrinsic allows annotation of local variables with arbitrary
8074 strings. This can be useful for special purpose optimizations that want
8075 to look for these annotations. These have no other defined use; they are
8076 ignored by code generation and optimization.
8078 '``llvm.annotation.*``' Intrinsic
8079 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8084 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8085 any integer bit width.
8089 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8090 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8091 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8092 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8093 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8098 The '``llvm.annotation``' intrinsic.
8103 The first argument is an integer value (result of some expression), the
8104 second is a pointer to a global string, the third is a pointer to a
8105 global string which is the source file name, and the last argument is
8106 the line number. It returns the value of the first argument.
8111 This intrinsic allows annotations to be put on arbitrary expressions
8112 with arbitrary strings. This can be useful for special purpose
8113 optimizations that want to look for these annotations. These have no
8114 other defined use; they are ignored by code generation and optimization.
8116 '``llvm.trap``' Intrinsic
8117 ^^^^^^^^^^^^^^^^^^^^^^^^^
8124 declare void @llvm.trap() noreturn nounwind
8129 The '``llvm.trap``' intrinsic.
8139 This intrinsic is lowered to the target dependent trap instruction. If
8140 the target does not have a trap instruction, this intrinsic will be
8141 lowered to a call of the ``abort()`` function.
8143 '``llvm.debugtrap``' Intrinsic
8144 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8151 declare void @llvm.debugtrap() nounwind
8156 The '``llvm.debugtrap``' intrinsic.
8166 This intrinsic is lowered to code which is intended to cause an
8167 execution trap with the intention of requesting the attention of a
8170 '``llvm.stackprotector``' Intrinsic
8171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8178 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8183 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8184 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8185 is placed on the stack before local variables.
8190 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8191 The first argument is the value loaded from the stack guard
8192 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8193 enough space to hold the value of the guard.
8198 This intrinsic causes the prologue/epilogue inserter to force the
8199 position of the ``AllocaInst`` stack slot to be before local variables
8200 on the stack. This is to ensure that if a local variable on the stack is
8201 overwritten, it will destroy the value of the guard. When the function
8202 exits, the guard on the stack is checked against the original guard. If
8203 they are different, then the program aborts by calling the
8204 ``__stack_chk_fail()`` function.
8206 '``llvm.objectsize``' Intrinsic
8207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8214 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8215 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8220 The ``llvm.objectsize`` intrinsic is designed to provide information to
8221 the optimizers to determine at compile time whether a) an operation
8222 (like memcpy) will overflow a buffer that corresponds to an object, or
8223 b) that a runtime check for overflow isn't necessary. An object in this
8224 context means an allocation of a specific class, structure, array, or
8230 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8231 argument is a pointer to or into the ``object``. The second argument is
8232 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8233 or -1 (if false) when the object size is unknown. The second argument
8234 only accepts constants.
8239 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8240 the size of the object concerned. If the size cannot be determined at
8241 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8242 on the ``min`` argument).
8244 '``llvm.expect``' Intrinsic
8245 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8252 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8253 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8258 The ``llvm.expect`` intrinsic provides information about expected (the
8259 most probable) value of ``val``, which can be used by optimizers.
8264 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8265 a value. The second argument is an expected value, this needs to be a
8266 constant value, variables are not allowed.
8271 This intrinsic is lowered to the ``val``.
8273 '``llvm.donothing``' Intrinsic
8274 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8281 declare void @llvm.donothing() nounwind readnone
8286 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8287 only intrinsic that can be called with an invoke instruction.
8297 This intrinsic does nothing, and it's removed by optimizers and ignored