1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially
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
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
828 or not. The heuristic used will enable protectors for functions with:
830 - Character arrays larger than ``ssp-buffer-size`` (default 8).
831 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
832 - Calls to alloca() with variable sizes or constant sizes greater than
835 If a function that has an ``ssp`` attribute is inlined into a
836 function that doesn't have an ``ssp`` attribute, then the resulting
837 function will have an ``ssp`` attribute.
839 This attribute indicates that the function should *always* emit a
840 stack smashing protector. This overrides the ``ssp`` function
843 If a function that has an ``sspreq`` attribute is inlined into a
844 function that doesn't have an ``sspreq`` attribute or which has an
845 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
846 an ``sspreq`` attribute.
848 This attribute indicates that the function should emit a stack smashing
849 protector. This attribute causes a strong heuristic to be used when
850 determining if a function needs stack protectors. The strong heuristic
851 will enable protectors for functions with:
853 - Arrays of any size and type
854 - Aggregates containing an array of any size and type.
856 - Local variables that have had their address taken.
858 This overrides the ``ssp`` function attribute.
860 If a function that has an ``sspstrong`` attribute is inlined into a
861 function that doesn't have an ``sspstrong`` attribute, then the
862 resulting function will have an ``sspstrong`` attribute.
864 This attribute indicates that the ABI being targeted requires that
865 an unwind table entry be produce for this function even if we can
866 show that no exceptions passes by it. This is normally the case for
867 the ELF x86-64 abi, but it can be disabled for some compilation
870 This attribute indicates that calls to the function cannot be
871 duplicated. A call to a ``noduplicate`` function may be moved
872 within its parent function, but may not be duplicated within
875 A function containing a ``noduplicate`` call may still
876 be an inlining candidate, provided that the call is not
877 duplicated by inlining. That implies that the function has
878 internal linkage and only has one call site, so the original
879 call is dead after inlining.
883 Module-Level Inline Assembly
884 ----------------------------
886 Modules may contain "module-level inline asm" blocks, which corresponds
887 to the GCC "file scope inline asm" blocks. These blocks are internally
888 concatenated by LLVM and treated as a single unit, but may be separated
889 in the ``.ll`` file if desired. The syntax is very simple:
893 module asm "inline asm code goes here"
894 module asm "more can go here"
896 The strings can contain any character by escaping non-printable
897 characters. The escape sequence used is simply "\\xx" where "xx" is the
898 two digit hex code for the number.
900 The inline asm code is simply printed to the machine code .s file when
901 assembly code is generated.
906 A module may specify a target specific data layout string that specifies
907 how data is to be laid out in memory. The syntax for the data layout is
912 target datalayout = "layout specification"
914 The *layout specification* consists of a list of specifications
915 separated by the minus sign character ('-'). Each specification starts
916 with a letter and may include other information after the letter to
917 define some aspect of the data layout. The specifications accepted are
921 Specifies that the target lays out data in big-endian form. That is,
922 the bits with the most significance have the lowest address
925 Specifies that the target lays out data in little-endian form. That
926 is, the bits with the least significance have the lowest address
929 Specifies the natural alignment of the stack in bits. Alignment
930 promotion of stack variables is limited to the natural stack
931 alignment to avoid dynamic stack realignment. The stack alignment
932 must be a multiple of 8-bits. If omitted, the natural stack
933 alignment defaults to "unspecified", which does not prevent any
934 alignment promotions.
935 ``p[n]:<size>:<abi>:<pref>``
936 This specifies the *size* of a pointer and its ``<abi>`` and
937 ``<pref>``\erred alignments for address space ``n``. All sizes are in
938 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
939 preceding ``:`` should be omitted too. The address space, ``n`` is
940 optional, and if not specified, denotes the default address space 0.
941 The value of ``n`` must be in the range [1,2^23).
942 ``i<size>:<abi>:<pref>``
943 This specifies the alignment for an integer type of a given bit
944 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
945 ``v<size>:<abi>:<pref>``
946 This specifies the alignment for a vector type of a given bit
948 ``f<size>:<abi>:<pref>``
949 This specifies the alignment for a floating point type of a given bit
950 ``<size>``. Only values of ``<size>`` that are supported by the target
951 will work. 32 (float) and 64 (double) are supported on all targets; 80
952 or 128 (different flavors of long double) are also supported on some
954 ``a<size>:<abi>:<pref>``
955 This specifies the alignment for an aggregate type of a given bit
957 ``s<size>:<abi>:<pref>``
958 This specifies the alignment for a stack object of a given bit
960 ``n<size1>:<size2>:<size3>...``
961 This specifies a set of native integer widths for the target CPU in
962 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
963 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
964 this set are considered to support most general arithmetic operations
967 When constructing the data layout for a given target, LLVM starts with a
968 default set of specifications which are then (possibly) overridden by
969 the specifications in the ``datalayout`` keyword. The default
970 specifications are given in this list:
973 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
974 - ``S0`` - natural stack alignment is unspecified
975 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
976 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
977 - ``i16:16:16`` - i16 is 16-bit aligned
978 - ``i32:32:32`` - i32 is 32-bit aligned
979 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
981 - ``f16:16:16`` - half is 16-bit aligned
982 - ``f32:32:32`` - float is 32-bit aligned
983 - ``f64:64:64`` - double is 64-bit aligned
984 - ``f128:128:128`` - quad is 128-bit aligned
985 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
986 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
987 - ``a0:0:64`` - aggregates are 64-bit aligned
989 When LLVM is determining the alignment for a given type, it uses the
992 #. If the type sought is an exact match for one of the specifications,
993 that specification is used.
994 #. If no match is found, and the type sought is an integer type, then
995 the smallest integer type that is larger than the bitwidth of the
996 sought type is used. If none of the specifications are larger than
997 the bitwidth then the largest integer type is used. For example,
998 given the default specifications above, the i7 type will use the
999 alignment of i8 (next largest) while both i65 and i256 will use the
1000 alignment of i64 (largest specified).
1001 #. If no match is found, and the type sought is a vector type, then the
1002 largest vector type that is smaller than the sought vector type will
1003 be used as a fall back. This happens because <128 x double> can be
1004 implemented in terms of 64 <2 x double>, for example.
1006 The function of the data layout string may not be what you expect.
1007 Notably, this is not a specification from the frontend of what alignment
1008 the code generator should use.
1010 Instead, if specified, the target data layout is required to match what
1011 the ultimate *code generator* expects. This string is used by the
1012 mid-level optimizers to improve code, and this only works if it matches
1013 what the ultimate code generator uses. If you would like to generate IR
1014 that does not embed this target-specific detail into the IR, then you
1015 don't have to specify the string. This will disable some optimizations
1016 that require precise layout information, but this also prevents those
1017 optimizations from introducing target specificity into the IR.
1019 .. _pointeraliasing:
1021 Pointer Aliasing Rules
1022 ----------------------
1024 Any memory access must be done through a pointer value associated with
1025 an address range of the memory access, otherwise the behavior is
1026 undefined. Pointer values are associated with address ranges according
1027 to the following rules:
1029 - A pointer value is associated with the addresses associated with any
1030 value it is *based* on.
1031 - An address of a global variable is associated with the address range
1032 of the variable's storage.
1033 - The result value of an allocation instruction is associated with the
1034 address range of the allocated storage.
1035 - A null pointer in the default address-space is associated with no
1037 - An integer constant other than zero or a pointer value returned from
1038 a function not defined within LLVM may be associated with address
1039 ranges allocated through mechanisms other than those provided by
1040 LLVM. Such ranges shall not overlap with any ranges of addresses
1041 allocated by mechanisms provided by LLVM.
1043 A pointer value is *based* on another pointer value according to the
1046 - A pointer value formed from a ``getelementptr`` operation is *based*
1047 on the first operand of the ``getelementptr``.
1048 - The result value of a ``bitcast`` is *based* on the operand of the
1050 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1051 values that contribute (directly or indirectly) to the computation of
1052 the pointer's value.
1053 - The "*based* on" relationship is transitive.
1055 Note that this definition of *"based"* is intentionally similar to the
1056 definition of *"based"* in C99, though it is slightly weaker.
1058 LLVM IR does not associate types with memory. The result type of a
1059 ``load`` merely indicates the size and alignment of the memory from
1060 which to load, as well as the interpretation of the value. The first
1061 operand type of a ``store`` similarly only indicates the size and
1062 alignment of the store.
1064 Consequently, type-based alias analysis, aka TBAA, aka
1065 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1066 :ref:`Metadata <metadata>` may be used to encode additional information
1067 which specialized optimization passes may use to implement type-based
1072 Volatile Memory Accesses
1073 ------------------------
1075 Certain memory accesses, such as :ref:`load <i_load>`'s,
1076 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1077 marked ``volatile``. The optimizers must not change the number of
1078 volatile operations or change their order of execution relative to other
1079 volatile operations. The optimizers *may* change the order of volatile
1080 operations relative to non-volatile operations. This is not Java's
1081 "volatile" and has no cross-thread synchronization behavior.
1083 IR-level volatile loads and stores cannot safely be optimized into
1084 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1085 flagged volatile. Likewise, the backend should never split or merge
1086 target-legal volatile load/store instructions.
1090 Memory Model for Concurrent Operations
1091 --------------------------------------
1093 The LLVM IR does not define any way to start parallel threads of
1094 execution or to register signal handlers. Nonetheless, there are
1095 platform-specific ways to create them, and we define LLVM IR's behavior
1096 in their presence. This model is inspired by the C++0x memory model.
1098 For a more informal introduction to this model, see the :doc:`Atomics`.
1100 We define a *happens-before* partial order as the least partial order
1103 - Is a superset of single-thread program order, and
1104 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1105 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1106 techniques, like pthread locks, thread creation, thread joining,
1107 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1108 Constraints <ordering>`).
1110 Note that program order does not introduce *happens-before* edges
1111 between a thread and signals executing inside that thread.
1113 Every (defined) read operation (load instructions, memcpy, atomic
1114 loads/read-modify-writes, etc.) R reads a series of bytes written by
1115 (defined) write operations (store instructions, atomic
1116 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1117 section, initialized globals are considered to have a write of the
1118 initializer which is atomic and happens before any other read or write
1119 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1120 may see any write to the same byte, except:
1122 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1123 write\ :sub:`2` happens before R\ :sub:`byte`, then
1124 R\ :sub:`byte` does not see write\ :sub:`1`.
1125 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1126 R\ :sub:`byte` does not see write\ :sub:`3`.
1128 Given that definition, R\ :sub:`byte` is defined as follows:
1130 - If R is volatile, the result is target-dependent. (Volatile is
1131 supposed to give guarantees which can support ``sig_atomic_t`` in
1132 C/C++, and may be used for accesses to addresses which do not behave
1133 like normal memory. It does not generally provide cross-thread
1135 - Otherwise, if there is no write to the same byte that happens before
1136 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1137 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1138 R\ :sub:`byte` returns the value written by that write.
1139 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1140 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1141 Memory Ordering Constraints <ordering>` section for additional
1142 constraints on how the choice is made.
1143 - Otherwise R\ :sub:`byte` returns ``undef``.
1145 R returns the value composed of the series of bytes it read. This
1146 implies that some bytes within the value may be ``undef`` **without**
1147 the entire value being ``undef``. Note that this only defines the
1148 semantics of the operation; it doesn't mean that targets will emit more
1149 than one instruction to read the series of bytes.
1151 Note that in cases where none of the atomic intrinsics are used, this
1152 model places only one restriction on IR transformations on top of what
1153 is required for single-threaded execution: introducing a store to a byte
1154 which might not otherwise be stored is not allowed in general.
1155 (Specifically, in the case where another thread might write to and read
1156 from an address, introducing a store can change a load that may see
1157 exactly one write into a load that may see multiple writes.)
1161 Atomic Memory Ordering Constraints
1162 ----------------------------------
1164 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1165 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1166 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1167 an ordering parameter that determines which other atomic instructions on
1168 the same address they *synchronize with*. These semantics are borrowed
1169 from Java and C++0x, but are somewhat more colloquial. If these
1170 descriptions aren't precise enough, check those specs (see spec
1171 references in the :doc:`atomics guide <Atomics>`).
1172 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1173 differently since they don't take an address. See that instruction's
1174 documentation for details.
1176 For a simpler introduction to the ordering constraints, see the
1180 The set of values that can be read is governed by the happens-before
1181 partial order. A value cannot be read unless some operation wrote
1182 it. This is intended to provide a guarantee strong enough to model
1183 Java's non-volatile shared variables. This ordering cannot be
1184 specified for read-modify-write operations; it is not strong enough
1185 to make them atomic in any interesting way.
1187 In addition to the guarantees of ``unordered``, there is a single
1188 total order for modifications by ``monotonic`` operations on each
1189 address. All modification orders must be compatible with the
1190 happens-before order. There is no guarantee that the modification
1191 orders can be combined to a global total order for the whole program
1192 (and this often will not be possible). The read in an atomic
1193 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1194 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1195 order immediately before the value it writes. If one atomic read
1196 happens before another atomic read of the same address, the later
1197 read must see the same value or a later value in the address's
1198 modification order. This disallows reordering of ``monotonic`` (or
1199 stronger) operations on the same address. If an address is written
1200 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1201 read that address repeatedly, the other threads must eventually see
1202 the write. This corresponds to the C++0x/C1x
1203 ``memory_order_relaxed``.
1205 In addition to the guarantees of ``monotonic``, a
1206 *synchronizes-with* edge may be formed with a ``release`` operation.
1207 This is intended to model C++'s ``memory_order_acquire``.
1209 In addition to the guarantees of ``monotonic``, if this operation
1210 writes a value which is subsequently read by an ``acquire``
1211 operation, it *synchronizes-with* that operation. (This isn't a
1212 complete description; see the C++0x definition of a release
1213 sequence.) This corresponds to the C++0x/C1x
1214 ``memory_order_release``.
1215 ``acq_rel`` (acquire+release)
1216 Acts as both an ``acquire`` and ``release`` operation on its
1217 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1218 ``seq_cst`` (sequentially consistent)
1219 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1220 operation which only reads, ``release`` for an operation which only
1221 writes), there is a global total order on all
1222 sequentially-consistent operations on all addresses, which is
1223 consistent with the *happens-before* partial order and with the
1224 modification orders of all the affected addresses. Each
1225 sequentially-consistent read sees the last preceding write to the
1226 same address in this global order. This corresponds to the C++0x/C1x
1227 ``memory_order_seq_cst`` and Java volatile.
1231 If an atomic operation is marked ``singlethread``, it only *synchronizes
1232 with* or participates in modification and seq\_cst total orderings with
1233 other operations running in the same thread (for example, in signal
1241 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1242 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1243 :ref:`frem <i_frem>`) have the following flags that can set to enable
1244 otherwise unsafe floating point operations
1247 No NaNs - Allow optimizations to assume the arguments and result are not
1248 NaN. Such optimizations are required to retain defined behavior over
1249 NaNs, but the value of the result is undefined.
1252 No Infs - Allow optimizations to assume the arguments and result are not
1253 +/-Inf. Such optimizations are required to retain defined behavior over
1254 +/-Inf, but the value of the result is undefined.
1257 No Signed Zeros - Allow optimizations to treat the sign of a zero
1258 argument or result as insignificant.
1261 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1262 argument rather than perform division.
1265 Fast - Allow algebraically equivalent transformations that may
1266 dramatically change results in floating point (e.g. reassociate). This
1267 flag implies all the others.
1274 The LLVM type system is one of the most important features of the
1275 intermediate representation. Being typed enables a number of
1276 optimizations to be performed on the intermediate representation
1277 directly, without having to do extra analyses on the side before the
1278 transformation. A strong type system makes it easier to read the
1279 generated code and enables novel analyses and transformations that are
1280 not feasible to perform on normal three address code representations.
1282 Type Classifications
1283 --------------------
1285 The types fall into a few useful classifications:
1294 * - :ref:`integer <t_integer>`
1295 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1298 * - :ref:`floating point <t_floating>`
1299 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1307 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1308 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1309 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1310 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1312 * - :ref:`primitive <t_primitive>`
1313 - :ref:`label <t_label>`,
1314 :ref:`void <t_void>`,
1315 :ref:`integer <t_integer>`,
1316 :ref:`floating point <t_floating>`,
1317 :ref:`x86mmx <t_x86mmx>`,
1318 :ref:`metadata <t_metadata>`.
1320 * - :ref:`derived <t_derived>`
1321 - :ref:`array <t_array>`,
1322 :ref:`function <t_function>`,
1323 :ref:`pointer <t_pointer>`,
1324 :ref:`structure <t_struct>`,
1325 :ref:`vector <t_vector>`,
1326 :ref:`opaque <t_opaque>`.
1328 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1329 Values of these types are the only ones which can be produced by
1337 The primitive types are the fundamental building blocks of the LLVM
1348 The integer type is a very simple type that simply specifies an
1349 arbitrary bit width for the integer type desired. Any bit width from 1
1350 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1359 The number of bits the integer will occupy is specified by the ``N``
1365 +----------------+------------------------------------------------+
1366 | ``i1`` | a single-bit integer. |
1367 +----------------+------------------------------------------------+
1368 | ``i32`` | a 32-bit integer. |
1369 +----------------+------------------------------------------------+
1370 | ``i1942652`` | a really big integer of over 1 million bits. |
1371 +----------------+------------------------------------------------+
1375 Floating Point Types
1376 ^^^^^^^^^^^^^^^^^^^^
1385 - 16-bit floating point value
1388 - 32-bit floating point value
1391 - 64-bit floating point value
1394 - 128-bit floating point value (112-bit mantissa)
1397 - 80-bit floating point value (X87)
1400 - 128-bit floating point value (two 64-bits)
1410 The x86mmx type represents a value held in an MMX register on an x86
1411 machine. The operations allowed on it are quite limited: parameters and
1412 return values, load and store, and bitcast. User-specified MMX
1413 instructions are represented as intrinsic or asm calls with arguments
1414 and/or results of this type. There are no arrays, vectors or constants
1432 The void type does not represent any value and has no size.
1449 The label type represents code labels.
1466 The metadata type represents embedded metadata. No derived types may be
1467 created from metadata except for :ref:`function <t_function>` arguments.
1481 The real power in LLVM comes from the derived types in the system. This
1482 is what allows a programmer to represent arrays, functions, pointers,
1483 and other useful types. Each of these types contain one or more element
1484 types which may be a primitive type, or another derived type. For
1485 example, it is possible to have a two dimensional array, using an array
1486 as the element type of another array.
1493 Aggregate Types are a subset of derived types that can contain multiple
1494 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1495 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1506 The array type is a very simple derived type that arranges elements
1507 sequentially in memory. The array type requires a size (number of
1508 elements) and an underlying data type.
1515 [<# elements> x <elementtype>]
1517 The number of elements is a constant integer value; ``elementtype`` may
1518 be any type with a size.
1523 +------------------+--------------------------------------+
1524 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1525 +------------------+--------------------------------------+
1526 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1527 +------------------+--------------------------------------+
1528 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1529 +------------------+--------------------------------------+
1531 Here are some examples of multidimensional arrays:
1533 +-----------------------------+----------------------------------------------------------+
1534 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1535 +-----------------------------+----------------------------------------------------------+
1536 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1537 +-----------------------------+----------------------------------------------------------+
1538 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1539 +-----------------------------+----------------------------------------------------------+
1541 There is no restriction on indexing beyond the end of the array implied
1542 by a static type (though there are restrictions on indexing beyond the
1543 bounds of an allocated object in some cases). This means that
1544 single-dimension 'variable sized array' addressing can be implemented in
1545 LLVM with a zero length array type. An implementation of 'pascal style
1546 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1557 The function type can be thought of as a function signature. It consists
1558 of a return type and a list of formal parameter types. The return type
1559 of a function type is a first class type or a void type.
1566 <returntype> (<parameter list>)
1568 ...where '``<parameter list>``' is a comma-separated list of type
1569 specifiers. Optionally, the parameter list may include a type ``...``,
1570 which indicates that the function takes a variable number of arguments.
1571 Variable argument functions can access their arguments with the
1572 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1573 '``<returntype>``' is any type except :ref:`label <t_label>`.
1578 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1579 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1580 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1581 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1582 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1583 | ``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. |
1584 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1585 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1586 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1596 The structure type is used to represent a collection of data members
1597 together in memory. The elements of a structure may be any type that has
1600 Structures in memory are accessed using '``load``' and '``store``' by
1601 getting a pointer to a field with the '``getelementptr``' instruction.
1602 Structures in registers are accessed using the '``extractvalue``' and
1603 '``insertvalue``' instructions.
1605 Structures may optionally be "packed" structures, which indicate that
1606 the alignment of the struct is one byte, and that there is no padding
1607 between the elements. In non-packed structs, padding between field types
1608 is inserted as defined by the DataLayout string in the module, which is
1609 required to match what the underlying code generator expects.
1611 Structures can either be "literal" or "identified". A literal structure
1612 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1613 identified types are always defined at the top level with a name.
1614 Literal types are uniqued by their contents and can never be recursive
1615 or opaque since there is no way to write one. Identified types can be
1616 recursive, can be opaqued, and are never uniqued.
1623 %T1 = type { <type list> } ; Identified normal struct type
1624 %T2 = type <{ <type list> }> ; Identified packed struct type
1629 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1630 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1631 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1632 | ``{ 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``. |
1633 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1634 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1635 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1639 Opaque Structure Types
1640 ^^^^^^^^^^^^^^^^^^^^^^
1645 Opaque structure types are used to represent named structure types that
1646 do not have a body specified. This corresponds (for example) to the C
1647 notion of a forward declared structure.
1660 +--------------+-------------------+
1661 | ``opaque`` | An opaque type. |
1662 +--------------+-------------------+
1672 The pointer type is used to specify memory locations. Pointers are
1673 commonly used to reference objects in memory.
1675 Pointer types may have an optional address space attribute defining the
1676 numbered address space where the pointed-to object resides. The default
1677 address space is number zero. The semantics of non-zero address spaces
1678 are target-specific.
1680 Note that LLVM does not permit pointers to void (``void*``) nor does it
1681 permit pointers to labels (``label*``). Use ``i8*`` instead.
1693 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1694 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1695 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1696 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1697 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1698 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1699 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1709 A vector type is a simple derived type that represents a vector of
1710 elements. Vector types are used when multiple primitive data are
1711 operated in parallel using a single instruction (SIMD). A vector type
1712 requires a size (number of elements) and an underlying primitive data
1713 type. Vector types are considered :ref:`first class <t_firstclass>`.
1720 < <# elements> x <elementtype> >
1722 The number of elements is a constant integer value larger than 0;
1723 elementtype may be any integer or floating point type, or a pointer to
1724 these types. Vectors of size zero are not allowed.
1729 +-------------------+--------------------------------------------------+
1730 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1731 +-------------------+--------------------------------------------------+
1732 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1733 +-------------------+--------------------------------------------------+
1734 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1735 +-------------------+--------------------------------------------------+
1736 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1737 +-------------------+--------------------------------------------------+
1742 LLVM has several different basic types of constants. This section
1743 describes them all and their syntax.
1748 **Boolean constants**
1749 The two strings '``true``' and '``false``' are both valid constants
1751 **Integer constants**
1752 Standard integers (such as '4') are constants of the
1753 :ref:`integer <t_integer>` type. Negative numbers may be used with
1755 **Floating point constants**
1756 Floating point constants use standard decimal notation (e.g.
1757 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1758 hexadecimal notation (see below). The assembler requires the exact
1759 decimal value of a floating-point constant. For example, the
1760 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1761 decimal in binary. Floating point constants must have a :ref:`floating
1762 point <t_floating>` type.
1763 **Null pointer constants**
1764 The identifier '``null``' is recognized as a null pointer constant
1765 and must be of :ref:`pointer type <t_pointer>`.
1767 The one non-intuitive notation for constants is the hexadecimal form of
1768 floating point constants. For example, the form
1769 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1770 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1771 constants are required (and the only time that they are generated by the
1772 disassembler) is when a floating point constant must be emitted but it
1773 cannot be represented as a decimal floating point number in a reasonable
1774 number of digits. For example, NaN's, infinities, and other special
1775 values are represented in their IEEE hexadecimal format so that assembly
1776 and disassembly do not cause any bits to change in the constants.
1778 When using the hexadecimal form, constants of types half, float, and
1779 double are represented using the 16-digit form shown above (which
1780 matches the IEEE754 representation for double); half and float values
1781 must, however, be exactly representable as IEEE 754 half and single
1782 precision, respectively. Hexadecimal format is always used for long
1783 double, and there are three forms of long double. The 80-bit format used
1784 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1785 128-bit format used by PowerPC (two adjacent doubles) is represented by
1786 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1787 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1788 supported target uses this format. Long doubles will only work if they
1789 match the long double format on your target. The IEEE 16-bit format
1790 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1791 digits. All hexadecimal formats are big-endian (sign bit at the left).
1793 There are no constants of type x86mmx.
1798 Complex constants are a (potentially recursive) combination of simple
1799 constants and smaller complex constants.
1801 **Structure constants**
1802 Structure constants are represented with notation similar to
1803 structure type definitions (a comma separated list of elements,
1804 surrounded by braces (``{}``)). For example:
1805 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1806 "``@G = external global i32``". Structure constants must have
1807 :ref:`structure type <t_struct>`, and the number and types of elements
1808 must match those specified by the type.
1810 Array constants are represented with notation similar to array type
1811 definitions (a comma separated list of elements, surrounded by
1812 square brackets (``[]``)). For example:
1813 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1814 :ref:`array type <t_array>`, and the number and types of elements must
1815 match those specified by the type.
1816 **Vector constants**
1817 Vector constants are represented with notation similar to vector
1818 type definitions (a comma separated list of elements, surrounded by
1819 less-than/greater-than's (``<>``)). For example:
1820 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1821 must have :ref:`vector type <t_vector>`, and the number and types of
1822 elements must match those specified by the type.
1823 **Zero initialization**
1824 The string '``zeroinitializer``' can be used to zero initialize a
1825 value to zero of *any* type, including scalar and
1826 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1827 having to print large zero initializers (e.g. for large arrays) and
1828 is always exactly equivalent to using explicit zero initializers.
1830 A metadata node is a structure-like constant with :ref:`metadata
1831 type <t_metadata>`. For example:
1832 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1833 constants that are meant to be interpreted as part of the
1834 instruction stream, metadata is a place to attach additional
1835 information such as debug info.
1837 Global Variable and Function Addresses
1838 --------------------------------------
1840 The addresses of :ref:`global variables <globalvars>` and
1841 :ref:`functions <functionstructure>` are always implicitly valid
1842 (link-time) constants. These constants are explicitly referenced when
1843 the :ref:`identifier for the global <identifiers>` is used and always have
1844 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1847 .. code-block:: llvm
1851 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1858 The string '``undef``' can be used anywhere a constant is expected, and
1859 indicates that the user of the value may receive an unspecified
1860 bit-pattern. Undefined values may be of any type (other than '``label``'
1861 or '``void``') and be used anywhere a constant is permitted.
1863 Undefined values are useful because they indicate to the compiler that
1864 the program is well defined no matter what value is used. This gives the
1865 compiler more freedom to optimize. Here are some examples of
1866 (potentially surprising) transformations that are valid (in pseudo IR):
1868 .. code-block:: llvm
1878 This is safe because all of the output bits are affected by the undef
1879 bits. Any output bit can have a zero or one depending on the input bits.
1881 .. code-block:: llvm
1892 These logical operations have bits that are not always affected by the
1893 input. For example, if ``%X`` has a zero bit, then the output of the
1894 '``and``' operation will always be a zero for that bit, no matter what
1895 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1896 optimize or assume that the result of the '``and``' is '``undef``'.
1897 However, it is safe to assume that all bits of the '``undef``' could be
1898 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1899 all the bits of the '``undef``' operand to the '``or``' could be set,
1900 allowing the '``or``' to be folded to -1.
1902 .. code-block:: llvm
1904 %A = select undef, %X, %Y
1905 %B = select undef, 42, %Y
1906 %C = select %X, %Y, undef
1916 This set of examples shows that undefined '``select``' (and conditional
1917 branch) conditions can go *either way*, but they have to come from one
1918 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1919 both known to have a clear low bit, then ``%A`` would have to have a
1920 cleared low bit. However, in the ``%C`` example, the optimizer is
1921 allowed to assume that the '``undef``' operand could be the same as
1922 ``%Y``, allowing the whole '``select``' to be eliminated.
1924 .. code-block:: llvm
1926 %A = xor undef, undef
1943 This example points out that two '``undef``' operands are not
1944 necessarily the same. This can be surprising to people (and also matches
1945 C semantics) where they assume that "``X^X``" is always zero, even if
1946 ``X`` is undefined. This isn't true for a number of reasons, but the
1947 short answer is that an '``undef``' "variable" can arbitrarily change
1948 its value over its "live range". This is true because the variable
1949 doesn't actually *have a live range*. Instead, the value is logically
1950 read from arbitrary registers that happen to be around when needed, so
1951 the value is not necessarily consistent over time. In fact, ``%A`` and
1952 ``%C`` need to have the same semantics or the core LLVM "replace all
1953 uses with" concept would not hold.
1955 .. code-block:: llvm
1963 These examples show the crucial difference between an *undefined value*
1964 and *undefined behavior*. An undefined value (like '``undef``') is
1965 allowed to have an arbitrary bit-pattern. This means that the ``%A``
1966 operation can be constant folded to '``undef``', because the '``undef``'
1967 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
1968 However, in the second example, we can make a more aggressive
1969 assumption: because the ``undef`` is allowed to be an arbitrary value,
1970 we are allowed to assume that it could be zero. Since a divide by zero
1971 has *undefined behavior*, we are allowed to assume that the operation
1972 does not execute at all. This allows us to delete the divide and all
1973 code after it. Because the undefined operation "can't happen", the
1974 optimizer can assume that it occurs in dead code.
1976 .. code-block:: llvm
1978 a: store undef -> %X
1979 b: store %X -> undef
1984 These examples reiterate the ``fdiv`` example: a store *of* an undefined
1985 value can be assumed to not have any effect; we can assume that the
1986 value is overwritten with bits that happen to match what was already
1987 there. However, a store *to* an undefined location could clobber
1988 arbitrary memory, therefore, it has undefined behavior.
1995 Poison values are similar to :ref:`undef values <undefvalues>`, however
1996 they also represent the fact that an instruction or constant expression
1997 which cannot evoke side effects has nevertheless detected a condition
1998 which results in undefined behavior.
2000 There is currently no way of representing a poison value in the IR; they
2001 only exist when produced by operations such as :ref:`add <i_add>` with
2004 Poison value behavior is defined in terms of value *dependence*:
2006 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2007 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2008 their dynamic predecessor basic block.
2009 - Function arguments depend on the corresponding actual argument values
2010 in the dynamic callers of their functions.
2011 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2012 instructions that dynamically transfer control back to them.
2013 - :ref:`Invoke <i_invoke>` instructions depend on the
2014 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2015 call instructions that dynamically transfer control back to them.
2016 - Non-volatile loads and stores depend on the most recent stores to all
2017 of the referenced memory addresses, following the order in the IR
2018 (including loads and stores implied by intrinsics such as
2019 :ref:`@llvm.memcpy <int_memcpy>`.)
2020 - An instruction with externally visible side effects depends on the
2021 most recent preceding instruction with externally visible side
2022 effects, following the order in the IR. (This includes :ref:`volatile
2023 operations <volatile>`.)
2024 - An instruction *control-depends* on a :ref:`terminator
2025 instruction <terminators>` if the terminator instruction has
2026 multiple successors and the instruction is always executed when
2027 control transfers to one of the successors, and may not be executed
2028 when control is transferred to another.
2029 - Additionally, an instruction also *control-depends* on a terminator
2030 instruction if the set of instructions it otherwise depends on would
2031 be different if the terminator had transferred control to a different
2033 - Dependence is transitive.
2035 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2036 with the additional affect that any instruction which has a *dependence*
2037 on a poison value has undefined behavior.
2039 Here are some examples:
2041 .. code-block:: llvm
2044 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2045 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2046 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2047 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2049 store i32 %poison, i32* @g ; Poison value stored to memory.
2050 %poison2 = load i32* @g ; Poison value loaded back from memory.
2052 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2054 %narrowaddr = bitcast i32* @g to i16*
2055 %wideaddr = bitcast i32* @g to i64*
2056 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2057 %poison4 = load i64* %wideaddr ; Returns a poison value.
2059 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2060 br i1 %cmp, label %true, label %end ; Branch to either destination.
2063 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2064 ; it has undefined behavior.
2068 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2069 ; Both edges into this PHI are
2070 ; control-dependent on %cmp, so this
2071 ; always results in a poison value.
2073 store volatile i32 0, i32* @g ; This would depend on the store in %true
2074 ; if %cmp is true, or the store in %entry
2075 ; otherwise, so this is undefined behavior.
2077 br i1 %cmp, label %second_true, label %second_end
2078 ; The same branch again, but this time the
2079 ; true block doesn't have side effects.
2086 store volatile i32 0, i32* @g ; This time, the instruction always depends
2087 ; on the store in %end. Also, it is
2088 ; control-equivalent to %end, so this is
2089 ; well-defined (ignoring earlier undefined
2090 ; behavior in this example).
2094 Addresses of Basic Blocks
2095 -------------------------
2097 ``blockaddress(@function, %block)``
2099 The '``blockaddress``' constant computes the address of the specified
2100 basic block in the specified function, and always has an ``i8*`` type.
2101 Taking the address of the entry block is illegal.
2103 This value only has defined behavior when used as an operand to the
2104 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2105 against null. Pointer equality tests between labels addresses results in
2106 undefined behavior --- though, again, comparison against null is ok, and
2107 no label is equal to the null pointer. This may be passed around as an
2108 opaque pointer sized value as long as the bits are not inspected. This
2109 allows ``ptrtoint`` and arithmetic to be performed on these values so
2110 long as the original value is reconstituted before the ``indirectbr``
2113 Finally, some targets may provide defined semantics when using the value
2114 as the operand to an inline assembly, but that is target specific.
2116 Constant Expressions
2117 --------------------
2119 Constant expressions are used to allow expressions involving other
2120 constants to be used as constants. Constant expressions may be of any
2121 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2122 that does not have side effects (e.g. load and call are not supported).
2123 The following is the syntax for constant expressions:
2125 ``trunc (CST to TYPE)``
2126 Truncate a constant to another type. The bit size of CST must be
2127 larger than the bit size of TYPE. Both types must be integers.
2128 ``zext (CST to TYPE)``
2129 Zero extend a constant to another type. The bit size of CST must be
2130 smaller than the bit size of TYPE. Both types must be integers.
2131 ``sext (CST to TYPE)``
2132 Sign extend a constant to another type. The bit size of CST must be
2133 smaller than the bit size of TYPE. Both types must be integers.
2134 ``fptrunc (CST to TYPE)``
2135 Truncate a floating point constant to another floating point type.
2136 The size of CST must be larger than the size of TYPE. Both types
2137 must be floating point.
2138 ``fpext (CST to TYPE)``
2139 Floating point extend a constant to another type. The size of CST
2140 must be smaller or equal to the size of TYPE. Both types must be
2142 ``fptoui (CST to TYPE)``
2143 Convert a floating point constant to the corresponding unsigned
2144 integer constant. TYPE must be a scalar or vector integer type. CST
2145 must be of scalar or vector floating point type. Both CST and TYPE
2146 must be scalars, or vectors of the same number of elements. If the
2147 value won't fit in the integer type, the results are undefined.
2148 ``fptosi (CST to TYPE)``
2149 Convert a floating point constant to the corresponding signed
2150 integer constant. TYPE must be a scalar or vector integer type. CST
2151 must be of scalar or vector floating point type. Both CST and TYPE
2152 must be scalars, or vectors of the same number of elements. If the
2153 value won't fit in the integer type, the results are undefined.
2154 ``uitofp (CST to TYPE)``
2155 Convert an unsigned integer constant to the corresponding floating
2156 point constant. TYPE must be a scalar or vector floating point type.
2157 CST must be of scalar or vector integer type. Both CST and TYPE must
2158 be scalars, or vectors of the same number of elements. If the value
2159 won't fit in the floating point type, the results are undefined.
2160 ``sitofp (CST to TYPE)``
2161 Convert a signed integer constant to the corresponding floating
2162 point constant. TYPE must be a scalar or vector floating point type.
2163 CST must be of scalar or vector integer type. Both CST and TYPE must
2164 be scalars, or vectors of the same number of elements. If the value
2165 won't fit in the floating point type, the results are undefined.
2166 ``ptrtoint (CST to TYPE)``
2167 Convert a pointer typed constant to the corresponding integer
2168 constant ``TYPE`` must be an integer type. ``CST`` must be of
2169 pointer type. The ``CST`` value is zero extended, truncated, or
2170 unchanged to make it fit in ``TYPE``.
2171 ``inttoptr (CST to TYPE)``
2172 Convert an integer constant to a pointer constant. TYPE must be a
2173 pointer type. CST must be of integer type. The CST value is zero
2174 extended, truncated, or unchanged to make it fit in a pointer size.
2175 This one is *really* dangerous!
2176 ``bitcast (CST to TYPE)``
2177 Convert a constant, CST, to another TYPE. The constraints of the
2178 operands are the same as those for the :ref:`bitcast
2179 instruction <i_bitcast>`.
2180 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2181 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2182 constants. As with the :ref:`getelementptr <i_getelementptr>`
2183 instruction, the index list may have zero or more indexes, which are
2184 required to make sense for the type of "CSTPTR".
2185 ``select (COND, VAL1, VAL2)``
2186 Perform the :ref:`select operation <i_select>` on constants.
2187 ``icmp COND (VAL1, VAL2)``
2188 Performs the :ref:`icmp operation <i_icmp>` on constants.
2189 ``fcmp COND (VAL1, VAL2)``
2190 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2191 ``extractelement (VAL, IDX)``
2192 Perform the :ref:`extractelement operation <i_extractelement>` on
2194 ``insertelement (VAL, ELT, IDX)``
2195 Perform the :ref:`insertelement operation <i_insertelement>` on
2197 ``shufflevector (VEC1, VEC2, IDXMASK)``
2198 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2200 ``extractvalue (VAL, IDX0, IDX1, ...)``
2201 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2202 constants. The index list is interpreted in a similar manner as
2203 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2204 least one index value must be specified.
2205 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2206 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2207 The index list is interpreted in a similar manner as indices in a
2208 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2209 value must be specified.
2210 ``OPCODE (LHS, RHS)``
2211 Perform the specified operation of the LHS and RHS constants. OPCODE
2212 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2213 binary <bitwiseops>` operations. The constraints on operands are
2214 the same as those for the corresponding instruction (e.g. no bitwise
2215 operations on floating point values are allowed).
2220 Inline Assembler Expressions
2221 ----------------------------
2223 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2224 Inline Assembly <moduleasm>`) through the use of a special value. This
2225 value represents the inline assembler as a string (containing the
2226 instructions to emit), a list of operand constraints (stored as a
2227 string), a flag that indicates whether or not the inline asm expression
2228 has side effects, and a flag indicating whether the function containing
2229 the asm needs to align its stack conservatively. An example inline
2230 assembler expression is:
2232 .. code-block:: llvm
2234 i32 (i32) asm "bswap $0", "=r,r"
2236 Inline assembler expressions may **only** be used as the callee operand
2237 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2238 Thus, typically we have:
2240 .. code-block:: llvm
2242 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2244 Inline asms with side effects not visible in the constraint list must be
2245 marked as having side effects. This is done through the use of the
2246 '``sideeffect``' keyword, like so:
2248 .. code-block:: llvm
2250 call void asm sideeffect "eieio", ""()
2252 In some cases inline asms will contain code that will not work unless
2253 the stack is aligned in some way, such as calls or SSE instructions on
2254 x86, yet will not contain code that does that alignment within the asm.
2255 The compiler should make conservative assumptions about what the asm
2256 might contain and should generate its usual stack alignment code in the
2257 prologue if the '``alignstack``' keyword is present:
2259 .. code-block:: llvm
2261 call void asm alignstack "eieio", ""()
2263 Inline asms also support using non-standard assembly dialects. The
2264 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2265 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2266 the only supported dialects. An example is:
2268 .. code-block:: llvm
2270 call void asm inteldialect "eieio", ""()
2272 If multiple keywords appear the '``sideeffect``' keyword must come
2273 first, the '``alignstack``' keyword second and the '``inteldialect``'
2279 The call instructions that wrap inline asm nodes may have a
2280 "``!srcloc``" MDNode attached to it that contains a list of constant
2281 integers. If present, the code generator will use the integer as the
2282 location cookie value when report errors through the ``LLVMContext``
2283 error reporting mechanisms. This allows a front-end to correlate backend
2284 errors that occur with inline asm back to the source code that produced
2287 .. code-block:: llvm
2289 call void asm sideeffect "something bad", ""(), !srcloc !42
2291 !42 = !{ i32 1234567 }
2293 It is up to the front-end to make sense of the magic numbers it places
2294 in the IR. If the MDNode contains multiple constants, the code generator
2295 will use the one that corresponds to the line of the asm that the error
2300 Metadata Nodes and Metadata Strings
2301 -----------------------------------
2303 LLVM IR allows metadata to be attached to instructions in the program
2304 that can convey extra information about the code to the optimizers and
2305 code generator. One example application of metadata is source-level
2306 debug information. There are two metadata primitives: strings and nodes.
2307 All metadata has the ``metadata`` type and is identified in syntax by a
2308 preceding exclamation point ('``!``').
2310 A metadata string is a string surrounded by double quotes. It can
2311 contain any character by escaping non-printable characters with
2312 "``\xx``" where "``xx``" is the two digit hex code. For example:
2315 Metadata nodes are represented with notation similar to structure
2316 constants (a comma separated list of elements, surrounded by braces and
2317 preceded by an exclamation point). Metadata nodes can have any values as
2318 their operand. For example:
2320 .. code-block:: llvm
2322 !{ metadata !"test\00", i32 10}
2324 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2325 metadata nodes, which can be looked up in the module symbol table. For
2328 .. code-block:: llvm
2330 !foo = metadata !{!4, !3}
2332 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2333 function is using two metadata arguments:
2335 .. code-block:: llvm
2337 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2339 Metadata can be attached with an instruction. Here metadata ``!21`` is
2340 attached to the ``add`` instruction using the ``!dbg`` identifier:
2342 .. code-block:: llvm
2344 %indvar.next = add i64 %indvar, 1, !dbg !21
2346 More information about specific metadata nodes recognized by the
2347 optimizers and code generator is found below.
2352 In LLVM IR, memory does not have types, so LLVM's own type system is not
2353 suitable for doing TBAA. Instead, metadata is added to the IR to
2354 describe a type system of a higher level language. This can be used to
2355 implement typical C/C++ TBAA, but it can also be used to implement
2356 custom alias analysis behavior for other languages.
2358 The current metadata format is very simple. TBAA metadata nodes have up
2359 to three fields, e.g.:
2361 .. code-block:: llvm
2363 !0 = metadata !{ metadata !"an example type tree" }
2364 !1 = metadata !{ metadata !"int", metadata !0 }
2365 !2 = metadata !{ metadata !"float", metadata !0 }
2366 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2368 The first field is an identity field. It can be any value, usually a
2369 metadata string, which uniquely identifies the type. The most important
2370 name in the tree is the name of the root node. Two trees with different
2371 root node names are entirely disjoint, even if they have leaves with
2374 The second field identifies the type's parent node in the tree, or is
2375 null or omitted for a root node. A type is considered to alias all of
2376 its descendants and all of its ancestors in the tree. Also, a type is
2377 considered to alias all types in other trees, so that bitcode produced
2378 from multiple front-ends is handled conservatively.
2380 If the third field is present, it's an integer which if equal to 1
2381 indicates that the type is "constant" (meaning
2382 ``pointsToConstantMemory`` should return true; see `other useful
2383 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2385 '``tbaa.struct``' Metadata
2386 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2388 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2389 aggregate assignment operations in C and similar languages, however it
2390 is defined to copy a contiguous region of memory, which is more than
2391 strictly necessary for aggregate types which contain holes due to
2392 padding. Also, it doesn't contain any TBAA information about the fields
2395 ``!tbaa.struct`` metadata can describe which memory subregions in a
2396 memcpy are padding and what the TBAA tags of the struct are.
2398 The current metadata format is very simple. ``!tbaa.struct`` metadata
2399 nodes are a list of operands which are in conceptual groups of three.
2400 For each group of three, the first operand gives the byte offset of a
2401 field in bytes, the second gives its size in bytes, and the third gives
2404 .. code-block:: llvm
2406 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2408 This describes a struct with two fields. The first is at offset 0 bytes
2409 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2410 and has size 4 bytes and has tbaa tag !2.
2412 Note that the fields need not be contiguous. In this example, there is a
2413 4 byte gap between the two fields. This gap represents padding which
2414 does not carry useful data and need not be preserved.
2416 '``fpmath``' Metadata
2417 ^^^^^^^^^^^^^^^^^^^^^
2419 ``fpmath`` metadata may be attached to any instruction of floating point
2420 type. It can be used to express the maximum acceptable error in the
2421 result of that instruction, in ULPs, thus potentially allowing the
2422 compiler to use a more efficient but less accurate method of computing
2423 it. ULP is defined as follows:
2425 If ``x`` is a real number that lies between two finite consecutive
2426 floating-point numbers ``a`` and ``b``, without being equal to one
2427 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2428 distance between the two non-equal finite floating-point numbers
2429 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2431 The metadata node shall consist of a single positive floating point
2432 number representing the maximum relative error, for example:
2434 .. code-block:: llvm
2436 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2438 '``range``' Metadata
2439 ^^^^^^^^^^^^^^^^^^^^
2441 ``range`` metadata may be attached only to loads of integer types. It
2442 expresses the possible ranges the loaded value is in. The ranges are
2443 represented with a flattened list of integers. The loaded value is known
2444 to be in the union of the ranges defined by each consecutive pair. Each
2445 pair has the following properties:
2447 - The type must match the type loaded by the instruction.
2448 - The pair ``a,b`` represents the range ``[a,b)``.
2449 - Both ``a`` and ``b`` are constants.
2450 - The range is allowed to wrap.
2451 - The range should not represent the full or empty set. That is,
2454 In addition, the pairs must be in signed order of the lower bound and
2455 they must be non-contiguous.
2459 .. code-block:: llvm
2461 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2462 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2463 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2464 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2466 !0 = metadata !{ i8 0, i8 2 }
2467 !1 = metadata !{ i8 255, i8 2 }
2468 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2469 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2471 Module Flags Metadata
2472 =====================
2474 Information about the module as a whole is difficult to convey to LLVM's
2475 subsystems. The LLVM IR isn't sufficient to transmit this information.
2476 The ``llvm.module.flags`` named metadata exists in order to facilitate
2477 this. These flags are in the form of key / value pairs --- much like a
2478 dictionary --- making it easy for any subsystem who cares about a flag to
2481 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2482 Each triplet has the following form:
2484 - The first element is a *behavior* flag, which specifies the behavior
2485 when two (or more) modules are merged together, and it encounters two
2486 (or more) metadata with the same ID. The supported behaviors are
2488 - The second element is a metadata string that is a unique ID for the
2489 metadata. Each module may only have one flag entry for each unique ID (not
2490 including entries with the **Require** behavior).
2491 - The third element is the value of the flag.
2493 When two (or more) modules are merged together, the resulting
2494 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2495 each unique metadata ID string, there will be exactly one entry in the merged
2496 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2497 be determined by the merge behavior flag, as described below. The only exception
2498 is that entries with the *Require* behavior are always preserved.
2500 The following behaviors are supported:
2511 Emits an error if two values disagree, otherwise the resulting value
2512 is that of the operands.
2516 Emits a warning if two values disagree. The result value will be the
2517 operand for the flag from the first module being linked.
2521 Adds a requirement that another module flag be present and have a
2522 specified value after linking is performed. The value must be a
2523 metadata pair, where the first element of the pair is the ID of the
2524 module flag to be restricted, and the second element of the pair is
2525 the value the module flag should be restricted to. This behavior can
2526 be used to restrict the allowable results (via triggering of an
2527 error) of linking IDs with the **Override** behavior.
2531 Uses the specified value, regardless of the behavior or value of the
2532 other module. If both modules specify **Override**, but the values
2533 differ, an error will be emitted.
2537 Appends the two values, which are required to be metadata nodes.
2541 Appends the two values, which are required to be metadata
2542 nodes. However, duplicate entries in the second list are dropped
2543 during the append operation.
2545 It is an error for a particular unique flag ID to have multiple behaviors,
2546 except in the case of **Require** (which adds restrictions on another metadata
2547 value) or **Override**.
2549 An example of module flags:
2551 .. code-block:: llvm
2553 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2554 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2555 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2556 !3 = metadata !{ i32 3, metadata !"qux",
2558 metadata !"foo", i32 1
2561 !llvm.module.flags = !{ !0, !1, !2, !3 }
2563 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2564 if two or more ``!"foo"`` flags are seen is to emit an error if their
2565 values are not equal.
2567 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2568 behavior if two or more ``!"bar"`` flags are seen is to use the value
2571 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2572 behavior if two or more ``!"qux"`` flags are seen is to emit a
2573 warning if their values are not equal.
2575 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2579 metadata !{ metadata !"foo", i32 1 }
2581 The behavior is to emit an error if the ``llvm.module.flags`` does not
2582 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2585 Objective-C Garbage Collection Module Flags Metadata
2586 ----------------------------------------------------
2588 On the Mach-O platform, Objective-C stores metadata about garbage
2589 collection in a special section called "image info". The metadata
2590 consists of a version number and a bitmask specifying what types of
2591 garbage collection are supported (if any) by the file. If two or more
2592 modules are linked together their garbage collection metadata needs to
2593 be merged rather than appended together.
2595 The Objective-C garbage collection module flags metadata consists of the
2596 following key-value pairs:
2605 * - ``Objective-C Version``
2606 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2608 * - ``Objective-C Image Info Version``
2609 - **[Required]** --- The version of the image info section. Currently
2612 * - ``Objective-C Image Info Section``
2613 - **[Required]** --- The section to place the metadata. Valid values are
2614 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2615 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2616 Objective-C ABI version 2.
2618 * - ``Objective-C Garbage Collection``
2619 - **[Required]** --- Specifies whether garbage collection is supported or
2620 not. Valid values are 0, for no garbage collection, and 2, for garbage
2621 collection supported.
2623 * - ``Objective-C GC Only``
2624 - **[Optional]** --- Specifies that only garbage collection is supported.
2625 If present, its value must be 6. This flag requires that the
2626 ``Objective-C Garbage Collection`` flag have the value 2.
2628 Some important flag interactions:
2630 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2631 merged with a module with ``Objective-C Garbage Collection`` set to
2632 2, then the resulting module has the
2633 ``Objective-C Garbage Collection`` flag set to 0.
2634 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2635 merged with a module with ``Objective-C GC Only`` set to 6.
2637 Automatic Linker Flags Module Flags Metadata
2638 --------------------------------------------
2640 Some targets support embedding flags to the linker inside individual object
2641 files. Typically this is used in conjunction with language extensions which
2642 allow source files to explicitly declare the libraries they depend on, and have
2643 these automatically be transmitted to the linker via object files.
2645 These flags are encoded in the IR using metadata in the module flags section,
2646 using the ``Linker Options`` key. The merge behavior for this flag is required
2647 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2648 node which should be a list of other metadata nodes, each of which should be a
2649 list of metadata strings defining linker options.
2651 For example, the following metadata section specifies two separate sets of
2652 linker options, presumably to link against ``libz`` and the ``Cocoa``
2655 !0 = metadata !{ i32 6, metadata !"Linker Options",
2657 metadata !{ metadata !"-lz" },
2658 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2659 !llvm.module.flags = !{ !0 }
2661 The metadata encoding as lists of lists of options, as opposed to a collapsed
2662 list of options, is chosen so that the IR encoding can use multiple option
2663 strings to specify e.g., a single library, while still having that specifier be
2664 preserved as an atomic element that can be recognized by a target specific
2665 assembly writer or object file emitter.
2667 Each individual option is required to be either a valid option for the target's
2668 linker, or an option that is reserved by the target specific assembly writer or
2669 object file emitter. No other aspect of these options is defined by the IR.
2671 Intrinsic Global Variables
2672 ==========================
2674 LLVM has a number of "magic" global variables that contain data that
2675 affect code generation or other IR semantics. These are documented here.
2676 All globals of this sort should have a section specified as
2677 "``llvm.metadata``". This section and all globals that start with
2678 "``llvm.``" are reserved for use by LLVM.
2680 The '``llvm.used``' Global Variable
2681 -----------------------------------
2683 The ``@llvm.used`` global is an array with i8\* element type which has
2684 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2685 pointers to global variables and functions which may optionally have a
2686 pointer cast formed of bitcast or getelementptr. For example, a legal
2689 .. code-block:: llvm
2694 @llvm.used = appending global [2 x i8*] [
2696 i8* bitcast (i32* @Y to i8*)
2697 ], section "llvm.metadata"
2699 If a global variable appears in the ``@llvm.used`` list, then the
2700 compiler, assembler, and linker are required to treat the symbol as if
2701 there is a reference to the global that it cannot see. For example, if a
2702 variable has internal linkage and no references other than that from the
2703 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2704 represent references from inline asms and other things the compiler
2705 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2707 On some targets, the code generator must emit a directive to the
2708 assembler or object file to prevent the assembler and linker from
2709 molesting the symbol.
2711 The '``llvm.compiler.used``' Global Variable
2712 --------------------------------------------
2714 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2715 directive, except that it only prevents the compiler from touching the
2716 symbol. On targets that support it, this allows an intelligent linker to
2717 optimize references to the symbol without being impeded as it would be
2720 This is a rare construct that should only be used in rare circumstances,
2721 and should not be exposed to source languages.
2723 The '``llvm.global_ctors``' Global Variable
2724 -------------------------------------------
2726 .. code-block:: llvm
2728 %0 = type { i32, void ()* }
2729 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2731 The ``@llvm.global_ctors`` array contains a list of constructor
2732 functions and associated priorities. The functions referenced by this
2733 array will be called in ascending order of priority (i.e. lowest first)
2734 when the module is loaded. The order of functions with the same priority
2737 The '``llvm.global_dtors``' Global Variable
2738 -------------------------------------------
2740 .. code-block:: llvm
2742 %0 = type { i32, void ()* }
2743 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2745 The ``@llvm.global_dtors`` array contains a list of destructor functions
2746 and associated priorities. The functions referenced by this array will
2747 be called in descending order of priority (i.e. highest first) when the
2748 module is loaded. The order of functions with the same priority is not
2751 Instruction Reference
2752 =====================
2754 The LLVM instruction set consists of several different classifications
2755 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2756 instructions <binaryops>`, :ref:`bitwise binary
2757 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2758 :ref:`other instructions <otherops>`.
2762 Terminator Instructions
2763 -----------------------
2765 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2766 program ends with a "Terminator" instruction, which indicates which
2767 block should be executed after the current block is finished. These
2768 terminator instructions typically yield a '``void``' value: they produce
2769 control flow, not values (the one exception being the
2770 ':ref:`invoke <i_invoke>`' instruction).
2772 The terminator instructions are: ':ref:`ret <i_ret>`',
2773 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2774 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2775 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2779 '``ret``' Instruction
2780 ^^^^^^^^^^^^^^^^^^^^^
2787 ret <type> <value> ; Return a value from a non-void function
2788 ret void ; Return from void function
2793 The '``ret``' instruction is used to return control flow (and optionally
2794 a value) from a function back to the caller.
2796 There are two forms of the '``ret``' instruction: one that returns a
2797 value and then causes control flow, and one that just causes control
2803 The '``ret``' instruction optionally accepts a single argument, the
2804 return value. The type of the return value must be a ':ref:`first
2805 class <t_firstclass>`' type.
2807 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2808 return type and contains a '``ret``' instruction with no return value or
2809 a return value with a type that does not match its type, or if it has a
2810 void return type and contains a '``ret``' instruction with a return
2816 When the '``ret``' instruction is executed, control flow returns back to
2817 the calling function's context. If the caller is a
2818 ":ref:`call <i_call>`" instruction, execution continues at the
2819 instruction after the call. If the caller was an
2820 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2821 beginning of the "normal" destination block. If the instruction returns
2822 a value, that value shall set the call or invoke instruction's return
2828 .. code-block:: llvm
2830 ret i32 5 ; Return an integer value of 5
2831 ret void ; Return from a void function
2832 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2836 '``br``' Instruction
2837 ^^^^^^^^^^^^^^^^^^^^
2844 br i1 <cond>, label <iftrue>, label <iffalse>
2845 br label <dest> ; Unconditional branch
2850 The '``br``' instruction is used to cause control flow to transfer to a
2851 different basic block in the current function. There are two forms of
2852 this instruction, corresponding to a conditional branch and an
2853 unconditional branch.
2858 The conditional branch form of the '``br``' instruction takes a single
2859 '``i1``' value and two '``label``' values. The unconditional form of the
2860 '``br``' instruction takes a single '``label``' value as a target.
2865 Upon execution of a conditional '``br``' instruction, the '``i1``'
2866 argument is evaluated. If the value is ``true``, control flows to the
2867 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
2868 to the '``iffalse``' ``label`` argument.
2873 .. code-block:: llvm
2876 %cond = icmp eq i32 %a, %b
2877 br i1 %cond, label %IfEqual, label %IfUnequal
2885 '``switch``' Instruction
2886 ^^^^^^^^^^^^^^^^^^^^^^^^
2893 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2898 The '``switch``' instruction is used to transfer control flow to one of
2899 several different places. It is a generalization of the '``br``'
2900 instruction, allowing a branch to occur to one of many possible
2906 The '``switch``' instruction uses three parameters: an integer
2907 comparison value '``value``', a default '``label``' destination, and an
2908 array of pairs of comparison value constants and '``label``'s. The table
2909 is not allowed to contain duplicate constant entries.
2914 The ``switch`` instruction specifies a table of values and destinations.
2915 When the '``switch``' instruction is executed, this table is searched
2916 for the given value. If the value is found, control flow is transferred
2917 to the corresponding destination; otherwise, control flow is transferred
2918 to the default destination.
2923 Depending on properties of the target machine and the particular
2924 ``switch`` instruction, this instruction may be code generated in
2925 different ways. For example, it could be generated as a series of
2926 chained conditional branches or with a lookup table.
2931 .. code-block:: llvm
2933 ; Emulate a conditional br instruction
2934 %Val = zext i1 %value to i32
2935 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2937 ; Emulate an unconditional br instruction
2938 switch i32 0, label %dest [ ]
2940 ; Implement a jump table:
2941 switch i32 %val, label %otherwise [ i32 0, label %onzero
2943 i32 2, label %ontwo ]
2947 '``indirectbr``' Instruction
2948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2955 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2960 The '``indirectbr``' instruction implements an indirect branch to a
2961 label within the current function, whose address is specified by
2962 "``address``". Address must be derived from a
2963 :ref:`blockaddress <blockaddress>` constant.
2968 The '``address``' argument is the address of the label to jump to. The
2969 rest of the arguments indicate the full set of possible destinations
2970 that the address may point to. Blocks are allowed to occur multiple
2971 times in the destination list, though this isn't particularly useful.
2973 This destination list is required so that dataflow analysis has an
2974 accurate understanding of the CFG.
2979 Control transfers to the block specified in the address argument. All
2980 possible destination blocks must be listed in the label list, otherwise
2981 this instruction has undefined behavior. This implies that jumps to
2982 labels defined in other functions have undefined behavior as well.
2987 This is typically implemented with a jump through a register.
2992 .. code-block:: llvm
2994 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2998 '``invoke``' Instruction
2999 ^^^^^^^^^^^^^^^^^^^^^^^^
3006 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3007 to label <normal label> unwind label <exception label>
3012 The '``invoke``' instruction causes control to transfer to a specified
3013 function, with the possibility of control flow transfer to either the
3014 '``normal``' label or the '``exception``' label. If the callee function
3015 returns with the "``ret``" instruction, control flow will return to the
3016 "normal" label. If the callee (or any indirect callees) returns via the
3017 ":ref:`resume <i_resume>`" instruction or other exception handling
3018 mechanism, control is interrupted and continued at the dynamically
3019 nearest "exception" label.
3021 The '``exception``' label is a `landing
3022 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3023 '``exception``' label is required to have the
3024 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3025 information about the behavior of the program after unwinding happens,
3026 as its first non-PHI instruction. The restrictions on the
3027 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3028 instruction, so that the important information contained within the
3029 "``landingpad``" instruction can't be lost through normal code motion.
3034 This instruction requires several arguments:
3036 #. The optional "cconv" marker indicates which :ref:`calling
3037 convention <callingconv>` the call should use. If none is
3038 specified, the call defaults to using C calling conventions.
3039 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3040 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3042 #. '``ptr to function ty``': shall be the signature of the pointer to
3043 function value being invoked. In most cases, this is a direct
3044 function invocation, but indirect ``invoke``'s are just as possible,
3045 branching off an arbitrary pointer to function value.
3046 #. '``function ptr val``': An LLVM value containing a pointer to a
3047 function to be invoked.
3048 #. '``function args``': argument list whose types match the function
3049 signature argument types and parameter attributes. All arguments must
3050 be of :ref:`first class <t_firstclass>` type. If the function signature
3051 indicates the function accepts a variable number of arguments, the
3052 extra arguments can be specified.
3053 #. '``normal label``': the label reached when the called function
3054 executes a '``ret``' instruction.
3055 #. '``exception label``': the label reached when a callee returns via
3056 the :ref:`resume <i_resume>` instruction or other exception handling
3058 #. The optional :ref:`function attributes <fnattrs>` list. Only
3059 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3060 attributes are valid here.
3065 This instruction is designed to operate as a standard '``call``'
3066 instruction in most regards. The primary difference is that it
3067 establishes an association with a label, which is used by the runtime
3068 library to unwind the stack.
3070 This instruction is used in languages with destructors to ensure that
3071 proper cleanup is performed in the case of either a ``longjmp`` or a
3072 thrown exception. Additionally, this is important for implementation of
3073 '``catch``' clauses in high-level languages that support them.
3075 For the purposes of the SSA form, the definition of the value returned
3076 by the '``invoke``' instruction is deemed to occur on the edge from the
3077 current block to the "normal" label. If the callee unwinds then no
3078 return value is available.
3083 .. code-block:: llvm
3085 %retval = invoke i32 @Test(i32 15) to label %Continue
3086 unwind label %TestCleanup ; {i32}:retval set
3087 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3088 unwind label %TestCleanup ; {i32}:retval set
3092 '``resume``' Instruction
3093 ^^^^^^^^^^^^^^^^^^^^^^^^
3100 resume <type> <value>
3105 The '``resume``' instruction is a terminator instruction that has no
3111 The '``resume``' instruction requires one argument, which must have the
3112 same type as the result of any '``landingpad``' instruction in the same
3118 The '``resume``' instruction resumes propagation of an existing
3119 (in-flight) exception whose unwinding was interrupted with a
3120 :ref:`landingpad <i_landingpad>` instruction.
3125 .. code-block:: llvm
3127 resume { i8*, i32 } %exn
3131 '``unreachable``' Instruction
3132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3144 The '``unreachable``' instruction has no defined semantics. This
3145 instruction is used to inform the optimizer that a particular portion of
3146 the code is not reachable. This can be used to indicate that the code
3147 after a no-return function cannot be reached, and other facts.
3152 The '``unreachable``' instruction has no defined semantics.
3159 Binary operators are used to do most of the computation in a program.
3160 They require two operands of the same type, execute an operation on
3161 them, and produce a single value. The operands might represent multiple
3162 data, as is the case with the :ref:`vector <t_vector>` data type. The
3163 result value has the same type as its operands.
3165 There are several different binary operators:
3169 '``add``' Instruction
3170 ^^^^^^^^^^^^^^^^^^^^^
3177 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3178 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3179 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3180 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3185 The '``add``' instruction returns the sum of its two operands.
3190 The two arguments to the '``add``' instruction must be
3191 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3192 arguments must have identical types.
3197 The value produced is the integer sum of the two operands.
3199 If the sum has unsigned overflow, the result returned is the
3200 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3203 Because LLVM integers use a two's complement representation, this
3204 instruction is appropriate for both signed and unsigned integers.
3206 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3207 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3208 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3209 unsigned and/or signed overflow, respectively, occurs.
3214 .. code-block:: llvm
3216 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3220 '``fadd``' Instruction
3221 ^^^^^^^^^^^^^^^^^^^^^^
3228 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3233 The '``fadd``' instruction returns the sum of its two operands.
3238 The two arguments to the '``fadd``' instruction must be :ref:`floating
3239 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3240 Both arguments must have identical types.
3245 The value produced is the floating point sum of the two operands. This
3246 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3247 which are optimization hints to enable otherwise unsafe floating point
3253 .. code-block:: llvm
3255 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3257 '``sub``' Instruction
3258 ^^^^^^^^^^^^^^^^^^^^^
3265 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3266 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3267 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3268 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3273 The '``sub``' instruction returns the difference of its two operands.
3275 Note that the '``sub``' instruction is used to represent the '``neg``'
3276 instruction present in most other intermediate representations.
3281 The two arguments to the '``sub``' instruction must be
3282 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3283 arguments must have identical types.
3288 The value produced is the integer difference of the two operands.
3290 If the difference has unsigned overflow, the result returned is the
3291 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3294 Because LLVM integers use a two's complement representation, this
3295 instruction is appropriate for both signed and unsigned integers.
3297 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3298 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3299 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3300 unsigned and/or signed overflow, respectively, occurs.
3305 .. code-block:: llvm
3307 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3308 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3312 '``fsub``' Instruction
3313 ^^^^^^^^^^^^^^^^^^^^^^
3320 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3325 The '``fsub``' instruction returns the difference of its two operands.
3327 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3328 instruction present in most other intermediate representations.
3333 The two arguments to the '``fsub``' instruction must be :ref:`floating
3334 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3335 Both arguments must have identical types.
3340 The value produced is the floating point difference of the two operands.
3341 This instruction can also take any number of :ref:`fast-math
3342 flags <fastmath>`, which are optimization hints to enable otherwise
3343 unsafe floating point optimizations:
3348 .. code-block:: llvm
3350 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3351 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3353 '``mul``' Instruction
3354 ^^^^^^^^^^^^^^^^^^^^^
3361 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3362 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3363 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3364 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3369 The '``mul``' instruction returns the product of its two operands.
3374 The two arguments to the '``mul``' instruction must be
3375 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3376 arguments must have identical types.
3381 The value produced is the integer product of the two operands.
3383 If the result of the multiplication has unsigned overflow, the result
3384 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3385 bit width of the result.
3387 Because LLVM integers use a two's complement representation, and the
3388 result is the same width as the operands, this instruction returns the
3389 correct result for both signed and unsigned integers. If a full product
3390 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3391 sign-extended or zero-extended as appropriate to the width of the full
3394 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3395 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3396 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3397 unsigned and/or signed overflow, respectively, occurs.
3402 .. code-block:: llvm
3404 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3408 '``fmul``' Instruction
3409 ^^^^^^^^^^^^^^^^^^^^^^
3416 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3421 The '``fmul``' instruction returns the product of its two operands.
3426 The two arguments to the '``fmul``' instruction must be :ref:`floating
3427 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3428 Both arguments must have identical types.
3433 The value produced is the floating point product of the two operands.
3434 This instruction can also take any number of :ref:`fast-math
3435 flags <fastmath>`, which are optimization hints to enable otherwise
3436 unsafe floating point optimizations:
3441 .. code-block:: llvm
3443 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3445 '``udiv``' Instruction
3446 ^^^^^^^^^^^^^^^^^^^^^^
3453 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3454 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3459 The '``udiv``' instruction returns the quotient of its two operands.
3464 The two arguments to the '``udiv``' instruction must be
3465 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3466 arguments must have identical types.
3471 The value produced is the unsigned integer quotient of the two operands.
3473 Note that unsigned integer division and signed integer division are
3474 distinct operations; for signed integer division, use '``sdiv``'.
3476 Division by zero leads to undefined behavior.
3478 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3479 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3480 such, "((a udiv exact b) mul b) == a").
3485 .. code-block:: llvm
3487 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3489 '``sdiv``' Instruction
3490 ^^^^^^^^^^^^^^^^^^^^^^
3497 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3498 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3503 The '``sdiv``' instruction returns the quotient of its two operands.
3508 The two arguments to the '``sdiv``' instruction must be
3509 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3510 arguments must have identical types.
3515 The value produced is the signed integer quotient of the two operands
3516 rounded towards zero.
3518 Note that signed integer division and unsigned integer division are
3519 distinct operations; for unsigned integer division, use '``udiv``'.
3521 Division by zero leads to undefined behavior. Overflow also leads to
3522 undefined behavior; this is a rare case, but can occur, for example, by
3523 doing a 32-bit division of -2147483648 by -1.
3525 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3526 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3531 .. code-block:: llvm
3533 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3537 '``fdiv``' Instruction
3538 ^^^^^^^^^^^^^^^^^^^^^^
3545 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3550 The '``fdiv``' instruction returns the quotient of its two operands.
3555 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3556 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3557 Both arguments must have identical types.
3562 The value produced is the floating point quotient of the two operands.
3563 This instruction can also take any number of :ref:`fast-math
3564 flags <fastmath>`, which are optimization hints to enable otherwise
3565 unsafe floating point optimizations:
3570 .. code-block:: llvm
3572 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3574 '``urem``' Instruction
3575 ^^^^^^^^^^^^^^^^^^^^^^
3582 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3587 The '``urem``' instruction returns the remainder from the unsigned
3588 division of its two arguments.
3593 The two arguments to the '``urem``' instruction must be
3594 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3595 arguments must have identical types.
3600 This instruction returns the unsigned integer *remainder* of a division.
3601 This instruction always performs an unsigned division to get the
3604 Note that unsigned integer remainder and signed integer remainder are
3605 distinct operations; for signed integer remainder, use '``srem``'.
3607 Taking the remainder of a division by zero leads to undefined behavior.
3612 .. code-block:: llvm
3614 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3616 '``srem``' Instruction
3617 ^^^^^^^^^^^^^^^^^^^^^^
3624 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3629 The '``srem``' instruction returns the remainder from the signed
3630 division of its two operands. This instruction can also take
3631 :ref:`vector <t_vector>` versions of the values in which case the elements
3637 The two arguments to the '``srem``' instruction must be
3638 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3639 arguments must have identical types.
3644 This instruction returns the *remainder* of a division (where the result
3645 is either zero or has the same sign as the dividend, ``op1``), not the
3646 *modulo* operator (where the result is either zero or has the same sign
3647 as the divisor, ``op2``) of a value. For more information about the
3648 difference, see `The Math
3649 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3650 table of how this is implemented in various languages, please see
3652 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3654 Note that signed integer remainder and unsigned integer remainder are
3655 distinct operations; for unsigned integer remainder, use '``urem``'.
3657 Taking the remainder of a division by zero leads to undefined behavior.
3658 Overflow also leads to undefined behavior; this is a rare case, but can
3659 occur, for example, by taking the remainder of a 32-bit division of
3660 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3661 rule lets srem be implemented using instructions that return both the
3662 result of the division and the remainder.)
3667 .. code-block:: llvm
3669 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3673 '``frem``' Instruction
3674 ^^^^^^^^^^^^^^^^^^^^^^
3681 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3686 The '``frem``' instruction returns the remainder from the division of
3692 The two arguments to the '``frem``' instruction must be :ref:`floating
3693 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3694 Both arguments must have identical types.
3699 This instruction returns the *remainder* of a division. The remainder
3700 has the same sign as the dividend. This instruction can also take any
3701 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3702 to enable otherwise unsafe floating point optimizations:
3707 .. code-block:: llvm
3709 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3713 Bitwise Binary Operations
3714 -------------------------
3716 Bitwise binary operators are used to do various forms of bit-twiddling
3717 in a program. They are generally very efficient instructions and can
3718 commonly be strength reduced from other instructions. They require two
3719 operands of the same type, execute an operation on them, and produce a
3720 single value. The resulting value is the same type as its operands.
3722 '``shl``' Instruction
3723 ^^^^^^^^^^^^^^^^^^^^^
3730 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3731 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3732 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3733 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3738 The '``shl``' instruction returns the first operand shifted to the left
3739 a specified number of bits.
3744 Both arguments to the '``shl``' instruction must be the same
3745 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3746 '``op2``' is treated as an unsigned value.
3751 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3752 where ``n`` is the width of the result. If ``op2`` is (statically or
3753 dynamically) negative or equal to or larger than the number of bits in
3754 ``op1``, the result is undefined. If the arguments are vectors, each
3755 vector element of ``op1`` is shifted by the corresponding shift amount
3758 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3759 value <poisonvalues>` if it shifts out any non-zero bits. If the
3760 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3761 value <poisonvalues>` if it shifts out any bits that disagree with the
3762 resultant sign bit. As such, NUW/NSW have the same semantics as they
3763 would if the shift were expressed as a mul instruction with the same
3764 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3769 .. code-block:: llvm
3771 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3772 <result> = shl i32 4, 2 ; yields {i32}: 16
3773 <result> = shl i32 1, 10 ; yields {i32}: 1024
3774 <result> = shl i32 1, 32 ; undefined
3775 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3777 '``lshr``' Instruction
3778 ^^^^^^^^^^^^^^^^^^^^^^
3785 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3786 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3791 The '``lshr``' instruction (logical shift right) returns the first
3792 operand shifted to the right a specified number of bits with zero fill.
3797 Both arguments to the '``lshr``' instruction must be the same
3798 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3799 '``op2``' is treated as an unsigned value.
3804 This instruction always performs a logical shift right operation. The
3805 most significant bits of the result will be filled with zero bits after
3806 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3807 than the number of bits in ``op1``, the result is undefined. If the
3808 arguments are vectors, each vector element of ``op1`` is shifted by the
3809 corresponding shift amount in ``op2``.
3811 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3812 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3818 .. code-block:: llvm
3820 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3821 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3822 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3823 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3824 <result> = lshr i32 1, 32 ; undefined
3825 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3827 '``ashr``' Instruction
3828 ^^^^^^^^^^^^^^^^^^^^^^
3835 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
3836 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
3841 The '``ashr``' instruction (arithmetic shift right) returns the first
3842 operand shifted to the right a specified number of bits with sign
3848 Both arguments to the '``ashr``' instruction must be the same
3849 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3850 '``op2``' is treated as an unsigned value.
3855 This instruction always performs an arithmetic shift right operation,
3856 The most significant bits of the result will be filled with the sign bit
3857 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
3858 than the number of bits in ``op1``, the result is undefined. If the
3859 arguments are vectors, each vector element of ``op1`` is shifted by the
3860 corresponding shift amount in ``op2``.
3862 If the ``exact`` keyword is present, the result value of the ``ashr`` is
3863 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3869 .. code-block:: llvm
3871 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
3872 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
3873 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
3874 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
3875 <result> = ashr i32 1, 32 ; undefined
3876 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
3878 '``and``' Instruction
3879 ^^^^^^^^^^^^^^^^^^^^^
3886 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
3891 The '``and``' instruction returns the bitwise logical and of its two
3897 The two arguments to the '``and``' instruction must be
3898 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3899 arguments must have identical types.
3904 The truth table used for the '``and``' instruction is:
3921 .. code-block:: llvm
3923 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
3924 <result> = and i32 15, 40 ; yields {i32}:result = 8
3925 <result> = and i32 4, 8 ; yields {i32}:result = 0
3927 '``or``' Instruction
3928 ^^^^^^^^^^^^^^^^^^^^
3935 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
3940 The '``or``' instruction returns the bitwise logical inclusive or of its
3946 The two arguments to the '``or``' instruction must be
3947 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3948 arguments must have identical types.
3953 The truth table used for the '``or``' instruction is:
3972 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
3973 <result> = or i32 15, 40 ; yields {i32}:result = 47
3974 <result> = or i32 4, 8 ; yields {i32}:result = 12
3976 '``xor``' Instruction
3977 ^^^^^^^^^^^^^^^^^^^^^
3984 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
3989 The '``xor``' instruction returns the bitwise logical exclusive or of
3990 its two operands. The ``xor`` is used to implement the "one's
3991 complement" operation, which is the "~" operator in C.
3996 The two arguments to the '``xor``' instruction must be
3997 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3998 arguments must have identical types.
4003 The truth table used for the '``xor``' instruction is:
4020 .. code-block:: llvm
4022 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4023 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4024 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4025 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4030 LLVM supports several instructions to represent vector operations in a
4031 target-independent manner. These instructions cover the element-access
4032 and vector-specific operations needed to process vectors effectively.
4033 While LLVM does directly support these vector operations, many
4034 sophisticated algorithms will want to use target-specific intrinsics to
4035 take full advantage of a specific target.
4037 .. _i_extractelement:
4039 '``extractelement``' Instruction
4040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4047 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4052 The '``extractelement``' instruction extracts a single scalar element
4053 from a vector at a specified index.
4058 The first operand of an '``extractelement``' instruction is a value of
4059 :ref:`vector <t_vector>` type. The second operand is an index indicating
4060 the position from which to extract the element. The index may be a
4066 The result is a scalar of the same type as the element type of ``val``.
4067 Its value is the value at position ``idx`` of ``val``. If ``idx``
4068 exceeds the length of ``val``, the results are undefined.
4073 .. code-block:: llvm
4075 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4077 .. _i_insertelement:
4079 '``insertelement``' Instruction
4080 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4087 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4092 The '``insertelement``' instruction inserts a scalar element into a
4093 vector at a specified index.
4098 The first operand of an '``insertelement``' instruction is a value of
4099 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4100 type must equal the element type of the first operand. The third operand
4101 is an index indicating the position at which to insert the value. The
4102 index may be a variable.
4107 The result is a vector of the same type as ``val``. Its element values
4108 are those of ``val`` except at position ``idx``, where it gets the value
4109 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4115 .. code-block:: llvm
4117 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4119 .. _i_shufflevector:
4121 '``shufflevector``' Instruction
4122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4129 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4134 The '``shufflevector``' instruction constructs a permutation of elements
4135 from two input vectors, returning a vector with the same element type as
4136 the input and length that is the same as the shuffle mask.
4141 The first two operands of a '``shufflevector``' instruction are vectors
4142 with the same type. The third argument is a shuffle mask whose element
4143 type is always 'i32'. The result of the instruction is a vector whose
4144 length is the same as the shuffle mask and whose element type is the
4145 same as the element type of the first two operands.
4147 The shuffle mask operand is required to be a constant vector with either
4148 constant integer or undef values.
4153 The elements of the two input vectors are numbered from left to right
4154 across both of the vectors. The shuffle mask operand specifies, for each
4155 element of the result vector, which element of the two input vectors the
4156 result element gets. The element selector may be undef (meaning "don't
4157 care") and the second operand may be undef if performing a shuffle from
4163 .. code-block:: llvm
4165 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4166 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4167 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4168 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4169 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4170 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4171 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4172 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4174 Aggregate Operations
4175 --------------------
4177 LLVM supports several instructions for working with
4178 :ref:`aggregate <t_aggregate>` values.
4182 '``extractvalue``' Instruction
4183 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4190 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4195 The '``extractvalue``' instruction extracts the value of a member field
4196 from an :ref:`aggregate <t_aggregate>` value.
4201 The first operand of an '``extractvalue``' instruction is a value of
4202 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4203 constant indices to specify which value to extract in a similar manner
4204 as indices in a '``getelementptr``' instruction.
4206 The major differences to ``getelementptr`` indexing are:
4208 - Since the value being indexed is not a pointer, the first index is
4209 omitted and assumed to be zero.
4210 - At least one index must be specified.
4211 - Not only struct indices but also array indices must be in bounds.
4216 The result is the value at the position in the aggregate specified by
4222 .. code-block:: llvm
4224 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4228 '``insertvalue``' Instruction
4229 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4236 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4241 The '``insertvalue``' instruction inserts a value into a member field in
4242 an :ref:`aggregate <t_aggregate>` value.
4247 The first operand of an '``insertvalue``' instruction is a value of
4248 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4249 a first-class value to insert. The following operands are constant
4250 indices indicating the position at which to insert the value in a
4251 similar manner as indices in a '``extractvalue``' instruction. The value
4252 to insert must have the same type as the value identified by the
4258 The result is an aggregate of the same type as ``val``. Its value is
4259 that of ``val`` except that the value at the position specified by the
4260 indices is that of ``elt``.
4265 .. code-block:: llvm
4267 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4268 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4269 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4273 Memory Access and Addressing Operations
4274 ---------------------------------------
4276 A key design point of an SSA-based representation is how it represents
4277 memory. In LLVM, no memory locations are in SSA form, which makes things
4278 very simple. This section describes how to read, write, and allocate
4283 '``alloca``' Instruction
4284 ^^^^^^^^^^^^^^^^^^^^^^^^
4291 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4296 The '``alloca``' instruction allocates memory on the stack frame of the
4297 currently executing function, to be automatically released when this
4298 function returns to its caller. The object is always allocated in the
4299 generic address space (address space zero).
4304 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4305 bytes of memory on the runtime stack, returning a pointer of the
4306 appropriate type to the program. If "NumElements" is specified, it is
4307 the number of elements allocated, otherwise "NumElements" is defaulted
4308 to be one. If a constant alignment is specified, the value result of the
4309 allocation is guaranteed to be aligned to at least that boundary. If not
4310 specified, or if zero, the target can choose to align the allocation on
4311 any convenient boundary compatible with the type.
4313 '``type``' may be any sized type.
4318 Memory is allocated; a pointer is returned. The operation is undefined
4319 if there is insufficient stack space for the allocation. '``alloca``'d
4320 memory is automatically released when the function returns. The
4321 '``alloca``' instruction is commonly used to represent automatic
4322 variables that must have an address available. When the function returns
4323 (either with the ``ret`` or ``resume`` instructions), the memory is
4324 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4325 The order in which memory is allocated (ie., which way the stack grows)
4331 .. code-block:: llvm
4333 %ptr = alloca i32 ; yields {i32*}:ptr
4334 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4335 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4336 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4340 '``load``' Instruction
4341 ^^^^^^^^^^^^^^^^^^^^^^
4348 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4349 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4350 !<index> = !{ i32 1 }
4355 The '``load``' instruction is used to read from memory.
4360 The argument to the '``load``' instruction specifies the memory address
4361 from which to load. The pointer must point to a :ref:`first
4362 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4363 then the optimizer is not allowed to modify the number or order of
4364 execution of this ``load`` with other :ref:`volatile
4365 operations <volatile>`.
4367 If the ``load`` is marked as ``atomic``, it takes an extra
4368 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4369 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4370 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4371 when they may see multiple atomic stores. The type of the pointee must
4372 be an integer type whose bit width is a power of two greater than or
4373 equal to eight and less than or equal to a target-specific size limit.
4374 ``align`` must be explicitly specified on atomic loads, and the load has
4375 undefined behavior if the alignment is not set to a value which is at
4376 least the size in bytes of the pointee. ``!nontemporal`` does not have
4377 any defined semantics for atomic loads.
4379 The optional constant ``align`` argument specifies the alignment of the
4380 operation (that is, the alignment of the memory address). A value of 0
4381 or an omitted ``align`` argument means that the operation has the abi
4382 alignment for the target. It is the responsibility of the code emitter
4383 to ensure that the alignment information is correct. Overestimating the
4384 alignment results in undefined behavior. Underestimating the alignment
4385 may produce less efficient code. An alignment of 1 is always safe.
4387 The optional ``!nontemporal`` metadata must reference a single
4388 metatadata name <index> corresponding to a metadata node with one
4389 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4390 metatadata on the instruction tells the optimizer and code generator
4391 that this load is not expected to be reused in the cache. The code
4392 generator may select special instructions to save cache bandwidth, such
4393 as the ``MOVNT`` instruction on x86.
4395 The optional ``!invariant.load`` metadata must reference a single
4396 metatadata name <index> corresponding to a metadata node with no
4397 entries. The existence of the ``!invariant.load`` metatadata on the
4398 instruction tells the optimizer and code generator that this load
4399 address points to memory which does not change value during program
4400 execution. The optimizer may then move this load around, for example, by
4401 hoisting it out of loops using loop invariant code motion.
4406 The location of memory pointed to is loaded. If the value being loaded
4407 is of scalar type then the number of bytes read does not exceed the
4408 minimum number of bytes needed to hold all bits of the type. For
4409 example, loading an ``i24`` reads at most three bytes. When loading a
4410 value of a type like ``i20`` with a size that is not an integral number
4411 of bytes, the result is undefined if the value was not originally
4412 written using a store of the same type.
4417 .. code-block:: llvm
4419 %ptr = alloca i32 ; yields {i32*}:ptr
4420 store i32 3, i32* %ptr ; yields {void}
4421 %val = load i32* %ptr ; yields {i32}:val = i32 3
4425 '``store``' Instruction
4426 ^^^^^^^^^^^^^^^^^^^^^^^
4433 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4434 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4439 The '``store``' instruction is used to write to memory.
4444 There are two arguments to the '``store``' instruction: a value to store
4445 and an address at which to store it. The type of the '``<pointer>``'
4446 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4447 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4448 then the optimizer is not allowed to modify the number or order of
4449 execution of this ``store`` with other :ref:`volatile
4450 operations <volatile>`.
4452 If the ``store`` is marked as ``atomic``, it takes an extra
4453 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4454 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4455 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4456 when they may see multiple atomic stores. The type of the pointee must
4457 be an integer type whose bit width is a power of two greater than or
4458 equal to eight and less than or equal to a target-specific size limit.
4459 ``align`` must be explicitly specified on atomic stores, and the store
4460 has undefined behavior if the alignment is not set to a value which is
4461 at least the size in bytes of the pointee. ``!nontemporal`` does not
4462 have any defined semantics for atomic stores.
4464 The optional constant "align" argument specifies the alignment of the
4465 operation (that is, the alignment of the memory address). A value of 0
4466 or an omitted "align" argument means that the operation has the abi
4467 alignment for the target. It is the responsibility of the code emitter
4468 to ensure that the alignment information is correct. Overestimating the
4469 alignment results in an undefined behavior. Underestimating the
4470 alignment may produce less efficient code. An alignment of 1 is always
4473 The optional !nontemporal metadata must reference a single metatadata
4474 name <index> corresponding to a metadata node with one i32 entry of
4475 value 1. The existence of the !nontemporal metatadata on the instruction
4476 tells the optimizer and code generator that this load is not expected to
4477 be reused in the cache. The code generator may select special
4478 instructions to save cache bandwidth, such as the MOVNT instruction on
4484 The contents of memory are updated to contain '``<value>``' at the
4485 location specified by the '``<pointer>``' operand. If '``<value>``' is
4486 of scalar type then the number of bytes written does not exceed the
4487 minimum number of bytes needed to hold all bits of the type. For
4488 example, storing an ``i24`` writes at most three bytes. When writing a
4489 value of a type like ``i20`` with a size that is not an integral number
4490 of bytes, it is unspecified what happens to the extra bits that do not
4491 belong to the type, but they will typically be overwritten.
4496 .. code-block:: llvm
4498 %ptr = alloca i32 ; yields {i32*}:ptr
4499 store i32 3, i32* %ptr ; yields {void}
4500 %val = load i32* %ptr ; yields {i32}:val = i32 3
4504 '``fence``' Instruction
4505 ^^^^^^^^^^^^^^^^^^^^^^^
4512 fence [singlethread] <ordering> ; yields {void}
4517 The '``fence``' instruction is used to introduce happens-before edges
4523 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4524 defines what *synchronizes-with* edges they add. They can only be given
4525 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4530 A fence A which has (at least) ``release`` ordering semantics
4531 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4532 semantics if and only if there exist atomic operations X and Y, both
4533 operating on some atomic object M, such that A is sequenced before X, X
4534 modifies M (either directly or through some side effect of a sequence
4535 headed by X), Y is sequenced before B, and Y observes M. This provides a
4536 *happens-before* dependency between A and B. Rather than an explicit
4537 ``fence``, one (but not both) of the atomic operations X or Y might
4538 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4539 still *synchronize-with* the explicit ``fence`` and establish the
4540 *happens-before* edge.
4542 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4543 ``acquire`` and ``release`` semantics specified above, participates in
4544 the global program order of other ``seq_cst`` operations and/or fences.
4546 The optional ":ref:`singlethread <singlethread>`" argument specifies
4547 that the fence only synchronizes with other fences in the same thread.
4548 (This is useful for interacting with signal handlers.)
4553 .. code-block:: llvm
4555 fence acquire ; yields {void}
4556 fence singlethread seq_cst ; yields {void}
4560 '``cmpxchg``' Instruction
4561 ^^^^^^^^^^^^^^^^^^^^^^^^^
4568 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4573 The '``cmpxchg``' instruction is used to atomically modify memory. It
4574 loads a value in memory and compares it to a given value. If they are
4575 equal, it stores a new value into the memory.
4580 There are three arguments to the '``cmpxchg``' instruction: an address
4581 to operate on, a value to compare to the value currently be at that
4582 address, and a new value to place at that address if the compared values
4583 are equal. The type of '<cmp>' must be an integer type whose bit width
4584 is a power of two greater than or equal to eight and less than or equal
4585 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4586 type, and the type of '<pointer>' must be a pointer to that type. If the
4587 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4588 to modify the number or order of execution of this ``cmpxchg`` with
4589 other :ref:`volatile operations <volatile>`.
4591 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4592 synchronizes with other atomic operations.
4594 The optional "``singlethread``" argument declares that the ``cmpxchg``
4595 is only atomic with respect to code (usually signal handlers) running in
4596 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4597 respect to all other code in the system.
4599 The pointer passed into cmpxchg must have alignment greater than or
4600 equal to the size in memory of the operand.
4605 The contents of memory at the location specified by the '``<pointer>``'
4606 operand is read and compared to '``<cmp>``'; if the read value is the
4607 equal, '``<new>``' is written. The original value at the location is
4610 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4611 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4612 atomic load with an ordering parameter determined by dropping any
4613 ``release`` part of the ``cmpxchg``'s ordering.
4618 .. code-block:: llvm
4621 %orig = atomic load i32* %ptr unordered ; yields {i32}
4625 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4626 %squared = mul i32 %cmp, %cmp
4627 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4628 %success = icmp eq i32 %cmp, %old
4629 br i1 %success, label %done, label %loop
4636 '``atomicrmw``' Instruction
4637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4644 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4649 The '``atomicrmw``' instruction is used to atomically modify memory.
4654 There are three arguments to the '``atomicrmw``' instruction: an
4655 operation to apply, an address whose value to modify, an argument to the
4656 operation. The operation must be one of the following keywords:
4670 The type of '<value>' must be an integer type whose bit width is a power
4671 of two greater than or equal to eight and less than or equal to a
4672 target-specific size limit. The type of the '``<pointer>``' operand must
4673 be a pointer to that type. If the ``atomicrmw`` is marked as
4674 ``volatile``, then the optimizer is not allowed to modify the number or
4675 order of execution of this ``atomicrmw`` with other :ref:`volatile
4676 operations <volatile>`.
4681 The contents of memory at the location specified by the '``<pointer>``'
4682 operand are atomically read, modified, and written back. The original
4683 value at the location is returned. The modification is specified by the
4686 - xchg: ``*ptr = val``
4687 - add: ``*ptr = *ptr + val``
4688 - sub: ``*ptr = *ptr - val``
4689 - and: ``*ptr = *ptr & val``
4690 - nand: ``*ptr = ~(*ptr & val)``
4691 - or: ``*ptr = *ptr | val``
4692 - xor: ``*ptr = *ptr ^ val``
4693 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4694 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4695 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4697 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4703 .. code-block:: llvm
4705 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4707 .. _i_getelementptr:
4709 '``getelementptr``' Instruction
4710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4717 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4718 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4719 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4724 The '``getelementptr``' instruction is used to get the address of a
4725 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4726 address calculation only and does not access memory.
4731 The first argument is always a pointer or a vector of pointers, and
4732 forms the basis of the calculation. The remaining arguments are indices
4733 that indicate which of the elements of the aggregate object are indexed.
4734 The interpretation of each index is dependent on the type being indexed
4735 into. The first index always indexes the pointer value given as the
4736 first argument, the second index indexes a value of the type pointed to
4737 (not necessarily the value directly pointed to, since the first index
4738 can be non-zero), etc. The first type indexed into must be a pointer
4739 value, subsequent types can be arrays, vectors, and structs. Note that
4740 subsequent types being indexed into can never be pointers, since that
4741 would require loading the pointer before continuing calculation.
4743 The type of each index argument depends on the type it is indexing into.
4744 When indexing into a (optionally packed) structure, only ``i32`` integer
4745 **constants** are allowed (when using a vector of indices they must all
4746 be the **same** ``i32`` integer constant). When indexing into an array,
4747 pointer or vector, integers of any width are allowed, and they are not
4748 required to be constant. These integers are treated as signed values
4751 For example, let's consider a C code fragment and how it gets compiled
4767 int *foo(struct ST *s) {
4768 return &s[1].Z.B[5][13];
4771 The LLVM code generated by Clang is:
4773 .. code-block:: llvm
4775 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4776 %struct.ST = type { i32, double, %struct.RT }
4778 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4780 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4787 In the example above, the first index is indexing into the
4788 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4789 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4790 indexes into the third element of the structure, yielding a
4791 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4792 structure. The third index indexes into the second element of the
4793 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4794 dimensions of the array are subscripted into, yielding an '``i32``'
4795 type. The '``getelementptr``' instruction returns a pointer to this
4796 element, thus computing a value of '``i32*``' type.
4798 Note that it is perfectly legal to index partially through a structure,
4799 returning a pointer to an inner element. Because of this, the LLVM code
4800 for the given testcase is equivalent to:
4802 .. code-block:: llvm
4804 define i32* @foo(%struct.ST* %s) {
4805 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4806 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4807 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4808 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4809 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4813 If the ``inbounds`` keyword is present, the result value of the
4814 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4815 pointer is not an *in bounds* address of an allocated object, or if any
4816 of the addresses that would be formed by successive addition of the
4817 offsets implied by the indices to the base address with infinitely
4818 precise signed arithmetic are not an *in bounds* address of that
4819 allocated object. The *in bounds* addresses for an allocated object are
4820 all the addresses that point into the object, plus the address one byte
4821 past the end. In cases where the base is a vector of pointers the
4822 ``inbounds`` keyword applies to each of the computations element-wise.
4824 If the ``inbounds`` keyword is not present, the offsets are added to the
4825 base address with silently-wrapping two's complement arithmetic. If the
4826 offsets have a different width from the pointer, they are sign-extended
4827 or truncated to the width of the pointer. The result value of the
4828 ``getelementptr`` may be outside the object pointed to by the base
4829 pointer. The result value may not necessarily be used to access memory
4830 though, even if it happens to point into allocated storage. See the
4831 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4834 The getelementptr instruction is often confusing. For some more insight
4835 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
4840 .. code-block:: llvm
4842 ; yields [12 x i8]*:aptr
4843 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4845 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4847 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4849 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4851 In cases where the pointer argument is a vector of pointers, each index
4852 must be a vector with the same number of elements. For example:
4854 .. code-block:: llvm
4856 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
4858 Conversion Operations
4859 ---------------------
4861 The instructions in this category are the conversion instructions
4862 (casting) which all take a single operand and a type. They perform
4863 various bit conversions on the operand.
4865 '``trunc .. to``' Instruction
4866 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4873 <result> = trunc <ty> <value> to <ty2> ; yields ty2
4878 The '``trunc``' instruction truncates its operand to the type ``ty2``.
4883 The '``trunc``' instruction takes a value to trunc, and a type to trunc
4884 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
4885 of the same number of integers. The bit size of the ``value`` must be
4886 larger than the bit size of the destination type, ``ty2``. Equal sized
4887 types are not allowed.
4892 The '``trunc``' instruction truncates the high order bits in ``value``
4893 and converts the remaining bits to ``ty2``. Since the source size must
4894 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
4895 It will always truncate bits.
4900 .. code-block:: llvm
4902 %X = trunc i32 257 to i8 ; yields i8:1
4903 %Y = trunc i32 123 to i1 ; yields i1:true
4904 %Z = trunc i32 122 to i1 ; yields i1:false
4905 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
4907 '``zext .. to``' Instruction
4908 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4915 <result> = zext <ty> <value> to <ty2> ; yields ty2
4920 The '``zext``' instruction zero extends its operand to type ``ty2``.
4925 The '``zext``' instruction takes a value to cast, and a type to cast it
4926 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4927 the same number of integers. The bit size of the ``value`` must be
4928 smaller than the bit size of the destination type, ``ty2``.
4933 The ``zext`` fills the high order bits of the ``value`` with zero bits
4934 until it reaches the size of the destination type, ``ty2``.
4936 When zero extending from i1, the result will always be either 0 or 1.
4941 .. code-block:: llvm
4943 %X = zext i32 257 to i64 ; yields i64:257
4944 %Y = zext i1 true to i32 ; yields i32:1
4945 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4947 '``sext .. to``' Instruction
4948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4955 <result> = sext <ty> <value> to <ty2> ; yields ty2
4960 The '``sext``' sign extends ``value`` to the type ``ty2``.
4965 The '``sext``' instruction takes a value to cast, and a type to cast it
4966 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4967 the same number of integers. The bit size of the ``value`` must be
4968 smaller than the bit size of the destination type, ``ty2``.
4973 The '``sext``' instruction performs a sign extension by copying the sign
4974 bit (highest order bit) of the ``value`` until it reaches the bit size
4975 of the type ``ty2``.
4977 When sign extending from i1, the extension always results in -1 or 0.
4982 .. code-block:: llvm
4984 %X = sext i8 -1 to i16 ; yields i16 :65535
4985 %Y = sext i1 true to i32 ; yields i32:-1
4986 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4988 '``fptrunc .. to``' Instruction
4989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4996 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5001 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5006 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5007 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5008 The size of ``value`` must be larger than the size of ``ty2``. This
5009 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5014 The '``fptrunc``' instruction truncates a ``value`` from a larger
5015 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5016 point <t_floating>` type. If the value cannot fit within the
5017 destination type, ``ty2``, then the results are undefined.
5022 .. code-block:: llvm
5024 %X = fptrunc double 123.0 to float ; yields float:123.0
5025 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5027 '``fpext .. to``' Instruction
5028 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5035 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5040 The '``fpext``' extends a floating point ``value`` to a larger floating
5046 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5047 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5048 to. The source type must be smaller than the destination type.
5053 The '``fpext``' instruction extends the ``value`` from a smaller
5054 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5055 point <t_floating>` type. The ``fpext`` cannot be used to make a
5056 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5057 *no-op cast* for a floating point cast.
5062 .. code-block:: llvm
5064 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5065 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5067 '``fptoui .. to``' Instruction
5068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5075 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5080 The '``fptoui``' converts a floating point ``value`` to its unsigned
5081 integer equivalent of type ``ty2``.
5086 The '``fptoui``' instruction takes a value to cast, which must be a
5087 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5088 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5089 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5090 type with the same number of elements as ``ty``
5095 The '``fptoui``' instruction converts its :ref:`floating
5096 point <t_floating>` operand into the nearest (rounding towards zero)
5097 unsigned integer value. If the value cannot fit in ``ty2``, the results
5103 .. code-block:: llvm
5105 %X = fptoui double 123.0 to i32 ; yields i32:123
5106 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5107 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5109 '``fptosi .. to``' Instruction
5110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5117 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5122 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5123 ``value`` to type ``ty2``.
5128 The '``fptosi``' instruction takes a value to cast, which must be a
5129 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5130 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5131 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5132 type with the same number of elements as ``ty``
5137 The '``fptosi``' instruction converts its :ref:`floating
5138 point <t_floating>` operand into the nearest (rounding towards zero)
5139 signed integer value. If the value cannot fit in ``ty2``, the results
5145 .. code-block:: llvm
5147 %X = fptosi double -123.0 to i32 ; yields i32:-123
5148 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5149 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5151 '``uitofp .. to``' Instruction
5152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5159 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5164 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5165 and converts that value to the ``ty2`` type.
5170 The '``uitofp``' instruction takes a value to cast, which must be a
5171 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5172 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5173 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5174 type with the same number of elements as ``ty``
5179 The '``uitofp``' instruction interprets its operand as an unsigned
5180 integer quantity and converts it to the corresponding floating point
5181 value. If the value cannot fit in the floating point value, the results
5187 .. code-block:: llvm
5189 %X = uitofp i32 257 to float ; yields float:257.0
5190 %Y = uitofp i8 -1 to double ; yields double:255.0
5192 '``sitofp .. to``' Instruction
5193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5200 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5205 The '``sitofp``' instruction regards ``value`` as a signed integer and
5206 converts that value to the ``ty2`` type.
5211 The '``sitofp``' instruction takes a value to cast, which must be a
5212 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5213 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5214 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5215 type with the same number of elements as ``ty``
5220 The '``sitofp``' instruction interprets its operand as a signed integer
5221 quantity and converts it to the corresponding floating point value. If
5222 the value cannot fit in the floating point value, the results are
5228 .. code-block:: llvm
5230 %X = sitofp i32 257 to float ; yields float:257.0
5231 %Y = sitofp i8 -1 to double ; yields double:-1.0
5235 '``ptrtoint .. to``' Instruction
5236 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5243 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5248 The '``ptrtoint``' instruction converts the pointer or a vector of
5249 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5254 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5255 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5256 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5257 a vector of integers type.
5262 The '``ptrtoint``' instruction converts ``value`` to integer type
5263 ``ty2`` by interpreting the pointer value as an integer and either
5264 truncating or zero extending that value to the size of the integer type.
5265 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5266 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5267 the same size, then nothing is done (*no-op cast*) other than a type
5273 .. code-block:: llvm
5275 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5276 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5277 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5281 '``inttoptr .. to``' Instruction
5282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5289 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5294 The '``inttoptr``' instruction converts an integer ``value`` to a
5295 pointer type, ``ty2``.
5300 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5301 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5307 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5308 applying either a zero extension or a truncation depending on the size
5309 of the integer ``value``. If ``value`` is larger than the size of a
5310 pointer then a truncation is done. If ``value`` is smaller than the size
5311 of a pointer then a zero extension is done. If they are the same size,
5312 nothing is done (*no-op cast*).
5317 .. code-block:: llvm
5319 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5320 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5321 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5322 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5326 '``bitcast .. to``' Instruction
5327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5334 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5339 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5345 The '``bitcast``' instruction takes a value to cast, which must be a
5346 non-aggregate first class value, and a type to cast it to, which must
5347 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5348 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5349 If the source type is a pointer, the destination type must also be a
5350 pointer. This instruction supports bitwise conversion of vectors to
5351 integers and to vectors of other types (as long as they have the same
5357 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5358 always a *no-op cast* because no bits change with this conversion. The
5359 conversion is done as if the ``value`` had been stored to memory and
5360 read back as type ``ty2``. Pointer (or vector of pointers) types may
5361 only be converted to other pointer (or vector of pointers) types with
5362 this instruction. To convert pointers to other types, use the
5363 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5369 .. code-block:: llvm
5371 %X = bitcast i8 255 to i8 ; yields i8 :-1
5372 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5373 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5374 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5381 The instructions in this category are the "miscellaneous" instructions,
5382 which defy better classification.
5386 '``icmp``' Instruction
5387 ^^^^^^^^^^^^^^^^^^^^^^
5394 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5399 The '``icmp``' instruction returns a boolean value or a vector of
5400 boolean values based on comparison of its two integer, integer vector,
5401 pointer, or pointer vector operands.
5406 The '``icmp``' instruction takes three operands. The first operand is
5407 the condition code indicating the kind of comparison to perform. It is
5408 not a value, just a keyword. The possible condition code are:
5411 #. ``ne``: not equal
5412 #. ``ugt``: unsigned greater than
5413 #. ``uge``: unsigned greater or equal
5414 #. ``ult``: unsigned less than
5415 #. ``ule``: unsigned less or equal
5416 #. ``sgt``: signed greater than
5417 #. ``sge``: signed greater or equal
5418 #. ``slt``: signed less than
5419 #. ``sle``: signed less or equal
5421 The remaining two arguments must be :ref:`integer <t_integer>` or
5422 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5423 must also be identical types.
5428 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5429 code given as ``cond``. The comparison performed always yields either an
5430 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5432 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5433 otherwise. No sign interpretation is necessary or performed.
5434 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5435 otherwise. No sign interpretation is necessary or performed.
5436 #. ``ugt``: interprets the operands as unsigned values and yields
5437 ``true`` if ``op1`` is greater than ``op2``.
5438 #. ``uge``: interprets the operands as unsigned values and yields
5439 ``true`` if ``op1`` is greater than or equal to ``op2``.
5440 #. ``ult``: interprets the operands as unsigned values and yields
5441 ``true`` if ``op1`` is less than ``op2``.
5442 #. ``ule``: interprets the operands as unsigned values and yields
5443 ``true`` if ``op1`` is less than or equal to ``op2``.
5444 #. ``sgt``: interprets the operands as signed values and yields ``true``
5445 if ``op1`` is greater than ``op2``.
5446 #. ``sge``: interprets the operands as signed values and yields ``true``
5447 if ``op1`` is greater than or equal to ``op2``.
5448 #. ``slt``: interprets the operands as signed values and yields ``true``
5449 if ``op1`` is less than ``op2``.
5450 #. ``sle``: interprets the operands as signed values and yields ``true``
5451 if ``op1`` is less than or equal to ``op2``.
5453 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5454 are compared as if they were integers.
5456 If the operands are integer vectors, then they are compared element by
5457 element. The result is an ``i1`` vector with the same number of elements
5458 as the values being compared. Otherwise, the result is an ``i1``.
5463 .. code-block:: llvm
5465 <result> = icmp eq i32 4, 5 ; yields: result=false
5466 <result> = icmp ne float* %X, %X ; yields: result=false
5467 <result> = icmp ult i16 4, 5 ; yields: result=true
5468 <result> = icmp sgt i16 4, 5 ; yields: result=false
5469 <result> = icmp ule i16 -4, 5 ; yields: result=false
5470 <result> = icmp sge i16 4, 5 ; yields: result=false
5472 Note that the code generator does not yet support vector types with the
5473 ``icmp`` instruction.
5477 '``fcmp``' Instruction
5478 ^^^^^^^^^^^^^^^^^^^^^^
5485 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5490 The '``fcmp``' instruction returns a boolean value or vector of boolean
5491 values based on comparison of its operands.
5493 If the operands are floating point scalars, then the result type is a
5494 boolean (:ref:`i1 <t_integer>`).
5496 If the operands are floating point vectors, then the result type is a
5497 vector of boolean with the same number of elements as the operands being
5503 The '``fcmp``' instruction takes three operands. The first operand is
5504 the condition code indicating the kind of comparison to perform. It is
5505 not a value, just a keyword. The possible condition code are:
5507 #. ``false``: no comparison, always returns false
5508 #. ``oeq``: ordered and equal
5509 #. ``ogt``: ordered and greater than
5510 #. ``oge``: ordered and greater than or equal
5511 #. ``olt``: ordered and less than
5512 #. ``ole``: ordered and less than or equal
5513 #. ``one``: ordered and not equal
5514 #. ``ord``: ordered (no nans)
5515 #. ``ueq``: unordered or equal
5516 #. ``ugt``: unordered or greater than
5517 #. ``uge``: unordered or greater than or equal
5518 #. ``ult``: unordered or less than
5519 #. ``ule``: unordered or less than or equal
5520 #. ``une``: unordered or not equal
5521 #. ``uno``: unordered (either nans)
5522 #. ``true``: no comparison, always returns true
5524 *Ordered* means that neither operand is a QNAN while *unordered* means
5525 that either operand may be a QNAN.
5527 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5528 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5529 type. They must have identical types.
5534 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5535 condition code given as ``cond``. If the operands are vectors, then the
5536 vectors are compared element by element. Each comparison performed
5537 always yields an :ref:`i1 <t_integer>` result, as follows:
5539 #. ``false``: always yields ``false``, regardless of operands.
5540 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5541 is equal to ``op2``.
5542 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5543 is greater than ``op2``.
5544 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5545 is greater than or equal to ``op2``.
5546 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5547 is less than ``op2``.
5548 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5549 is less than or equal to ``op2``.
5550 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5551 is not equal to ``op2``.
5552 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5553 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5555 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5556 greater than ``op2``.
5557 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5558 greater than or equal to ``op2``.
5559 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5561 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5562 less than or equal to ``op2``.
5563 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5564 not equal to ``op2``.
5565 #. ``uno``: yields ``true`` if either operand is a QNAN.
5566 #. ``true``: always yields ``true``, regardless of operands.
5571 .. code-block:: llvm
5573 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5574 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5575 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5576 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5578 Note that the code generator does not yet support vector types with the
5579 ``fcmp`` instruction.
5583 '``phi``' Instruction
5584 ^^^^^^^^^^^^^^^^^^^^^
5591 <result> = phi <ty> [ <val0>, <label0>], ...
5596 The '``phi``' instruction is used to implement the φ node in the SSA
5597 graph representing the function.
5602 The type of the incoming values is specified with the first type field.
5603 After this, the '``phi``' instruction takes a list of pairs as
5604 arguments, with one pair for each predecessor basic block of the current
5605 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5606 the value arguments to the PHI node. Only labels may be used as the
5609 There must be no non-phi instructions between the start of a basic block
5610 and the PHI instructions: i.e. PHI instructions must be first in a basic
5613 For the purposes of the SSA form, the use of each incoming value is
5614 deemed to occur on the edge from the corresponding predecessor block to
5615 the current block (but after any definition of an '``invoke``'
5616 instruction's return value on the same edge).
5621 At runtime, the '``phi``' instruction logically takes on the value
5622 specified by the pair corresponding to the predecessor basic block that
5623 executed just prior to the current block.
5628 .. code-block:: llvm
5630 Loop: ; Infinite loop that counts from 0 on up...
5631 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5632 %nextindvar = add i32 %indvar, 1
5637 '``select``' Instruction
5638 ^^^^^^^^^^^^^^^^^^^^^^^^
5645 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5647 selty is either i1 or {<N x i1>}
5652 The '``select``' instruction is used to choose one value based on a
5653 condition, without branching.
5658 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5659 values indicating the condition, and two values of the same :ref:`first
5660 class <t_firstclass>` type. If the val1/val2 are vectors and the
5661 condition is a scalar, then entire vectors are selected, not individual
5667 If the condition is an i1 and it evaluates to 1, the instruction returns
5668 the first value argument; otherwise, it returns the second value
5671 If the condition is a vector of i1, then the value arguments must be
5672 vectors of the same size, and the selection is done element by element.
5677 .. code-block:: llvm
5679 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5683 '``call``' Instruction
5684 ^^^^^^^^^^^^^^^^^^^^^^
5691 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5696 The '``call``' instruction represents a simple function call.
5701 This instruction requires several arguments:
5703 #. The optional "tail" marker indicates that the callee function does
5704 not access any allocas or varargs in the caller. Note that calls may
5705 be marked "tail" even if they do not occur before a
5706 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5707 function call is eligible for tail call optimization, but `might not
5708 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5709 The code generator may optimize calls marked "tail" with either 1)
5710 automatic `sibling call
5711 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5712 callee have matching signatures, or 2) forced tail call optimization
5713 when the following extra requirements are met:
5715 - Caller and callee both have the calling convention ``fastcc``.
5716 - The call is in tail position (ret immediately follows call and ret
5717 uses value of call or is void).
5718 - Option ``-tailcallopt`` is enabled, or
5719 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5720 - `Platform specific constraints are
5721 met. <CodeGenerator.html#tailcallopt>`_
5723 #. The optional "cconv" marker indicates which :ref:`calling
5724 convention <callingconv>` the call should use. If none is
5725 specified, the call defaults to using C calling conventions. The
5726 calling convention of the call must match the calling convention of
5727 the target function, or else the behavior is undefined.
5728 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5729 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5731 #. '``ty``': the type of the call instruction itself which is also the
5732 type of the return value. Functions that return no value are marked
5734 #. '``fnty``': shall be the signature of the pointer to function value
5735 being invoked. The argument types must match the types implied by
5736 this signature. This type can be omitted if the function is not
5737 varargs and if the function type does not return a pointer to a
5739 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5740 be invoked. In most cases, this is a direct function invocation, but
5741 indirect ``call``'s are just as possible, calling an arbitrary pointer
5743 #. '``function args``': argument list whose types match the function
5744 signature argument types and parameter attributes. All arguments must
5745 be of :ref:`first class <t_firstclass>` type. If the function signature
5746 indicates the function accepts a variable number of arguments, the
5747 extra arguments can be specified.
5748 #. The optional :ref:`function attributes <fnattrs>` list. Only
5749 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5750 attributes are valid here.
5755 The '``call``' instruction is used to cause control flow to transfer to
5756 a specified function, with its incoming arguments bound to the specified
5757 values. Upon a '``ret``' instruction in the called function, control
5758 flow continues with the instruction after the function call, and the
5759 return value of the function is bound to the result argument.
5764 .. code-block:: llvm
5766 %retval = call i32 @test(i32 %argc)
5767 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5768 %X = tail call i32 @foo() ; yields i32
5769 %Y = tail call fastcc i32 @foo() ; yields i32
5770 call void %foo(i8 97 signext)
5772 %struct.A = type { i32, i8 }
5773 %r = call %struct.A @foo() ; yields { 32, i8 }
5774 %gr = extractvalue %struct.A %r, 0 ; yields i32
5775 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5776 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5777 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5779 llvm treats calls to some functions with names and arguments that match
5780 the standard C99 library as being the C99 library functions, and may
5781 perform optimizations or generate code for them under that assumption.
5782 This is something we'd like to change in the future to provide better
5783 support for freestanding environments and non-C-based languages.
5787 '``va_arg``' Instruction
5788 ^^^^^^^^^^^^^^^^^^^^^^^^
5795 <resultval> = va_arg <va_list*> <arglist>, <argty>
5800 The '``va_arg``' instruction is used to access arguments passed through
5801 the "variable argument" area of a function call. It is used to implement
5802 the ``va_arg`` macro in C.
5807 This instruction takes a ``va_list*`` value and the type of the
5808 argument. It returns a value of the specified argument type and
5809 increments the ``va_list`` to point to the next argument. The actual
5810 type of ``va_list`` is target specific.
5815 The '``va_arg``' instruction loads an argument of the specified type
5816 from the specified ``va_list`` and causes the ``va_list`` to point to
5817 the next argument. For more information, see the variable argument
5818 handling :ref:`Intrinsic Functions <int_varargs>`.
5820 It is legal for this instruction to be called in a function which does
5821 not take a variable number of arguments, for example, the ``vfprintf``
5824 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5825 function <intrinsics>` because it takes a type as an argument.
5830 See the :ref:`variable argument processing <int_varargs>` section.
5832 Note that the code generator does not yet fully support va\_arg on many
5833 targets. Also, it does not currently support va\_arg with aggregate
5834 types on any target.
5838 '``landingpad``' Instruction
5839 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5846 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
5847 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
5849 <clause> := catch <type> <value>
5850 <clause> := filter <array constant type> <array constant>
5855 The '``landingpad``' instruction is used by `LLVM's exception handling
5856 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5857 is a landing pad --- one where the exception lands, and corresponds to the
5858 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
5859 defines values supplied by the personality function (``pers_fn``) upon
5860 re-entry to the function. The ``resultval`` has the type ``resultty``.
5865 This instruction takes a ``pers_fn`` value. This is the personality
5866 function associated with the unwinding mechanism. The optional
5867 ``cleanup`` flag indicates that the landing pad block is a cleanup.
5869 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
5870 contains the global variable representing the "type" that may be caught
5871 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
5872 clause takes an array constant as its argument. Use
5873 "``[0 x i8**] undef``" for a filter which cannot throw. The
5874 '``landingpad``' instruction must contain *at least* one ``clause`` or
5875 the ``cleanup`` flag.
5880 The '``landingpad``' instruction defines the values which are set by the
5881 personality function (``pers_fn``) upon re-entry to the function, and
5882 therefore the "result type" of the ``landingpad`` instruction. As with
5883 calling conventions, how the personality function results are
5884 represented in LLVM IR is target specific.
5886 The clauses are applied in order from top to bottom. If two
5887 ``landingpad`` instructions are merged together through inlining, the
5888 clauses from the calling function are appended to the list of clauses.
5889 When the call stack is being unwound due to an exception being thrown,
5890 the exception is compared against each ``clause`` in turn. If it doesn't
5891 match any of the clauses, and the ``cleanup`` flag is not set, then
5892 unwinding continues further up the call stack.
5894 The ``landingpad`` instruction has several restrictions:
5896 - A landing pad block is a basic block which is the unwind destination
5897 of an '``invoke``' instruction.
5898 - A landing pad block must have a '``landingpad``' instruction as its
5899 first non-PHI instruction.
5900 - There can be only one '``landingpad``' instruction within the landing
5902 - A basic block that is not a landing pad block may not include a
5903 '``landingpad``' instruction.
5904 - All '``landingpad``' instructions in a function must have the same
5905 personality function.
5910 .. code-block:: llvm
5912 ;; A landing pad which can catch an integer.
5913 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5915 ;; A landing pad that is a cleanup.
5916 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5918 ;; A landing pad which can catch an integer and can only throw a double.
5919 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5921 filter [1 x i8**] [@_ZTId]
5928 LLVM supports the notion of an "intrinsic function". These functions
5929 have well known names and semantics and are required to follow certain
5930 restrictions. Overall, these intrinsics represent an extension mechanism
5931 for the LLVM language that does not require changing all of the
5932 transformations in LLVM when adding to the language (or the bitcode
5933 reader/writer, the parser, etc...).
5935 Intrinsic function names must all start with an "``llvm.``" prefix. This
5936 prefix is reserved in LLVM for intrinsic names; thus, function names may
5937 not begin with this prefix. Intrinsic functions must always be external
5938 functions: you cannot define the body of intrinsic functions. Intrinsic
5939 functions may only be used in call or invoke instructions: it is illegal
5940 to take the address of an intrinsic function. Additionally, because
5941 intrinsic functions are part of the LLVM language, it is required if any
5942 are added that they be documented here.
5944 Some intrinsic functions can be overloaded, i.e., the intrinsic
5945 represents a family of functions that perform the same operation but on
5946 different data types. Because LLVM can represent over 8 million
5947 different integer types, overloading is used commonly to allow an
5948 intrinsic function to operate on any integer type. One or more of the
5949 argument types or the result type can be overloaded to accept any
5950 integer type. Argument types may also be defined as exactly matching a
5951 previous argument's type or the result type. This allows an intrinsic
5952 function which accepts multiple arguments, but needs all of them to be
5953 of the same type, to only be overloaded with respect to a single
5954 argument or the result.
5956 Overloaded intrinsics will have the names of its overloaded argument
5957 types encoded into its function name, each preceded by a period. Only
5958 those types which are overloaded result in a name suffix. Arguments
5959 whose type is matched against another type do not. For example, the
5960 ``llvm.ctpop`` function can take an integer of any width and returns an
5961 integer of exactly the same integer width. This leads to a family of
5962 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
5963 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
5964 overloaded, and only one type suffix is required. Because the argument's
5965 type is matched against the return type, it does not require its own
5968 To learn how to add an intrinsic function, please see the `Extending
5969 LLVM Guide <ExtendingLLVM.html>`_.
5973 Variable Argument Handling Intrinsics
5974 -------------------------------------
5976 Variable argument support is defined in LLVM with the
5977 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
5978 functions. These functions are related to the similarly named macros
5979 defined in the ``<stdarg.h>`` header file.
5981 All of these functions operate on arguments that use a target-specific
5982 value type "``va_list``". The LLVM assembly language reference manual
5983 does not define what this type is, so all transformations should be
5984 prepared to handle these functions regardless of the type used.
5986 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
5987 variable argument handling intrinsic functions are used.
5989 .. code-block:: llvm
5991 define i32 @test(i32 %X, ...) {
5992 ; Initialize variable argument processing
5994 %ap2 = bitcast i8** %ap to i8*
5995 call void @llvm.va_start(i8* %ap2)
5997 ; Read a single integer argument
5998 %tmp = va_arg i8** %ap, i32
6000 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6002 %aq2 = bitcast i8** %aq to i8*
6003 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6004 call void @llvm.va_end(i8* %aq2)
6006 ; Stop processing of arguments.
6007 call void @llvm.va_end(i8* %ap2)
6011 declare void @llvm.va_start(i8*)
6012 declare void @llvm.va_copy(i8*, i8*)
6013 declare void @llvm.va_end(i8*)
6017 '``llvm.va_start``' Intrinsic
6018 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6025 declare void %llvm.va_start(i8* <arglist>)
6030 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6031 subsequent use by ``va_arg``.
6036 The argument is a pointer to a ``va_list`` element to initialize.
6041 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6042 available in C. In a target-dependent way, it initializes the
6043 ``va_list`` element to which the argument points, so that the next call
6044 to ``va_arg`` will produce the first variable argument passed to the
6045 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6046 to know the last argument of the function as the compiler can figure
6049 '``llvm.va_end``' Intrinsic
6050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6057 declare void @llvm.va_end(i8* <arglist>)
6062 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6063 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6068 The argument is a pointer to a ``va_list`` to destroy.
6073 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6074 available in C. In a target-dependent way, it destroys the ``va_list``
6075 element to which the argument points. Calls to
6076 :ref:`llvm.va_start <int_va_start>` and
6077 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6082 '``llvm.va_copy``' Intrinsic
6083 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6090 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6095 The '``llvm.va_copy``' intrinsic copies the current argument position
6096 from the source argument list to the destination argument list.
6101 The first argument is a pointer to a ``va_list`` element to initialize.
6102 The second argument is a pointer to a ``va_list`` element to copy from.
6107 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6108 available in C. In a target-dependent way, it copies the source
6109 ``va_list`` element into the destination ``va_list`` element. This
6110 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6111 arbitrarily complex and require, for example, memory allocation.
6113 Accurate Garbage Collection Intrinsics
6114 --------------------------------------
6116 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6117 (GC) requires the implementation and generation of these intrinsics.
6118 These intrinsics allow identification of :ref:`GC roots on the
6119 stack <int_gcroot>`, as well as garbage collector implementations that
6120 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6121 Front-ends for type-safe garbage collected languages should generate
6122 these intrinsics to make use of the LLVM garbage collectors. For more
6123 details, see `Accurate Garbage Collection with
6124 LLVM <GarbageCollection.html>`_.
6126 The garbage collection intrinsics only operate on objects in the generic
6127 address space (address space zero).
6131 '``llvm.gcroot``' Intrinsic
6132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6139 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6144 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6145 the code generator, and allows some metadata to be associated with it.
6150 The first argument specifies the address of a stack object that contains
6151 the root pointer. The second pointer (which must be either a constant or
6152 a global value address) contains the meta-data to be associated with the
6158 At runtime, a call to this intrinsic stores a null pointer into the
6159 "ptrloc" location. At compile-time, the code generator generates
6160 information to allow the runtime to find the pointer at GC safe points.
6161 The '``llvm.gcroot``' intrinsic may only be used in a function which
6162 :ref:`specifies a GC algorithm <gc>`.
6166 '``llvm.gcread``' Intrinsic
6167 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6174 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6179 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6180 locations, allowing garbage collector implementations that require read
6186 The second argument is the address to read from, which should be an
6187 address allocated from the garbage collector. The first object is a
6188 pointer to the start of the referenced object, if needed by the language
6189 runtime (otherwise null).
6194 The '``llvm.gcread``' intrinsic has the same semantics as a load
6195 instruction, but may be replaced with substantially more complex code by
6196 the garbage collector runtime, as needed. The '``llvm.gcread``'
6197 intrinsic may only be used in a function which :ref:`specifies a GC
6202 '``llvm.gcwrite``' Intrinsic
6203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6210 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6215 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6216 locations, allowing garbage collector implementations that require write
6217 barriers (such as generational or reference counting collectors).
6222 The first argument is the reference to store, the second is the start of
6223 the object to store it to, and the third is the address of the field of
6224 Obj to store to. If the runtime does not require a pointer to the
6225 object, Obj may be null.
6230 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6231 instruction, but may be replaced with substantially more complex code by
6232 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6233 intrinsic may only be used in a function which :ref:`specifies a GC
6236 Code Generator Intrinsics
6237 -------------------------
6239 These intrinsics are provided by LLVM to expose special features that
6240 may only be implemented with code generator support.
6242 '``llvm.returnaddress``' Intrinsic
6243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6250 declare i8 *@llvm.returnaddress(i32 <level>)
6255 The '``llvm.returnaddress``' intrinsic attempts to compute a
6256 target-specific value indicating the return address of the current
6257 function or one of its callers.
6262 The argument to this intrinsic indicates which function to return the
6263 address for. Zero indicates the calling function, one indicates its
6264 caller, etc. The argument is **required** to be a constant integer
6270 The '``llvm.returnaddress``' intrinsic either returns a pointer
6271 indicating the return address of the specified call frame, or zero if it
6272 cannot be identified. The value returned by this intrinsic is likely to
6273 be incorrect or 0 for arguments other than zero, so it should only be
6274 used for debugging purposes.
6276 Note that calling this intrinsic does not prevent function inlining or
6277 other aggressive transformations, so the value returned may not be that
6278 of the obvious source-language caller.
6280 '``llvm.frameaddress``' Intrinsic
6281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6288 declare i8* @llvm.frameaddress(i32 <level>)
6293 The '``llvm.frameaddress``' intrinsic attempts to return the
6294 target-specific frame pointer value for the specified stack frame.
6299 The argument to this intrinsic indicates which function to return the
6300 frame pointer for. Zero indicates the calling function, one indicates
6301 its caller, etc. The argument is **required** to be a constant integer
6307 The '``llvm.frameaddress``' intrinsic either returns a pointer
6308 indicating the frame address of the specified call frame, or zero if it
6309 cannot be identified. The value returned by this intrinsic is likely to
6310 be incorrect or 0 for arguments other than zero, so it should only be
6311 used for debugging purposes.
6313 Note that calling this intrinsic does not prevent function inlining or
6314 other aggressive transformations, so the value returned may not be that
6315 of the obvious source-language caller.
6319 '``llvm.stacksave``' Intrinsic
6320 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6327 declare i8* @llvm.stacksave()
6332 The '``llvm.stacksave``' intrinsic is used to remember the current state
6333 of the function stack, for use with
6334 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6335 implementing language features like scoped automatic variable sized
6341 This intrinsic returns a opaque pointer value that can be passed to
6342 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6343 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6344 ``llvm.stacksave``, it effectively restores the state of the stack to
6345 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6346 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6347 were allocated after the ``llvm.stacksave`` was executed.
6349 .. _int_stackrestore:
6351 '``llvm.stackrestore``' Intrinsic
6352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6359 declare void @llvm.stackrestore(i8* %ptr)
6364 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6365 the function stack to the state it was in when the corresponding
6366 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6367 useful for implementing language features like scoped automatic variable
6368 sized arrays in C99.
6373 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6375 '``llvm.prefetch``' Intrinsic
6376 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6383 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6388 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6389 insert a prefetch instruction if supported; otherwise, it is a noop.
6390 Prefetches have no effect on the behavior of the program but can change
6391 its performance characteristics.
6396 ``address`` is the address to be prefetched, ``rw`` is the specifier
6397 determining if the fetch should be for a read (0) or write (1), and
6398 ``locality`` is a temporal locality specifier ranging from (0) - no
6399 locality, to (3) - extremely local keep in cache. The ``cache type``
6400 specifies whether the prefetch is performed on the data (1) or
6401 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6402 arguments must be constant integers.
6407 This intrinsic does not modify the behavior of the program. In
6408 particular, prefetches cannot trap and do not produce a value. On
6409 targets that support this intrinsic, the prefetch can provide hints to
6410 the processor cache for better performance.
6412 '``llvm.pcmarker``' Intrinsic
6413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6420 declare void @llvm.pcmarker(i32 <id>)
6425 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6426 Counter (PC) in a region of code to simulators and other tools. The
6427 method is target specific, but it is expected that the marker will use
6428 exported symbols to transmit the PC of the marker. The marker makes no
6429 guarantees that it will remain with any specific instruction after
6430 optimizations. It is possible that the presence of a marker will inhibit
6431 optimizations. The intended use is to be inserted after optimizations to
6432 allow correlations of simulation runs.
6437 ``id`` is a numerical id identifying the marker.
6442 This intrinsic does not modify the behavior of the program. Backends
6443 that do not support this intrinsic may ignore it.
6445 '``llvm.readcyclecounter``' Intrinsic
6446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6453 declare i64 @llvm.readcyclecounter()
6458 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6459 counter register (or similar low latency, high accuracy clocks) on those
6460 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6461 should map to RPCC. As the backing counters overflow quickly (on the
6462 order of 9 seconds on alpha), this should only be used for small
6468 When directly supported, reading the cycle counter should not modify any
6469 memory. Implementations are allowed to either return a application
6470 specific value or a system wide value. On backends without support, this
6471 is lowered to a constant 0.
6473 Standard C Library Intrinsics
6474 -----------------------------
6476 LLVM provides intrinsics for a few important standard C library
6477 functions. These intrinsics allow source-language front-ends to pass
6478 information about the alignment of the pointer arguments to the code
6479 generator, providing opportunity for more efficient code generation.
6483 '``llvm.memcpy``' Intrinsic
6484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6489 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6490 integer bit width and for different address spaces. Not all targets
6491 support all bit widths however.
6495 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6496 i32 <len>, i32 <align>, i1 <isvolatile>)
6497 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6498 i64 <len>, i32 <align>, i1 <isvolatile>)
6503 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6504 source location to the destination location.
6506 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6507 intrinsics do not return a value, takes extra alignment/isvolatile
6508 arguments and the pointers can be in specified address spaces.
6513 The first argument is a pointer to the destination, the second is a
6514 pointer to the source. The third argument is an integer argument
6515 specifying the number of bytes to copy, the fourth argument is the
6516 alignment of the source and destination locations, and the fifth is a
6517 boolean indicating a volatile access.
6519 If the call to this intrinsic has an alignment value that is not 0 or 1,
6520 then the caller guarantees that both the source and destination pointers
6521 are aligned to that boundary.
6523 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6524 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6525 very cleanly specified and it is unwise to depend on it.
6530 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6531 source location to the destination location, which are not allowed to
6532 overlap. It copies "len" bytes of memory over. If the argument is known
6533 to be aligned to some boundary, this can be specified as the fourth
6534 argument, otherwise it should be set to 0 or 1.
6536 '``llvm.memmove``' Intrinsic
6537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6542 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6543 bit width and for different address space. Not all targets support all
6548 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6549 i32 <len>, i32 <align>, i1 <isvolatile>)
6550 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6551 i64 <len>, i32 <align>, i1 <isvolatile>)
6556 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6557 source location to the destination location. It is similar to the
6558 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6561 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6562 intrinsics do not return a value, takes extra alignment/isvolatile
6563 arguments and the pointers can be in specified address spaces.
6568 The first argument is a pointer to the destination, the second is a
6569 pointer to the source. The third argument is an integer argument
6570 specifying the number of bytes to copy, the fourth argument is the
6571 alignment of the source and destination locations, and the fifth is a
6572 boolean indicating a volatile access.
6574 If the call to this intrinsic has an alignment value that is not 0 or 1,
6575 then the caller guarantees that the source and destination pointers are
6576 aligned to that boundary.
6578 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6579 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6580 not very cleanly specified and it is unwise to depend on it.
6585 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6586 source location to the destination location, which may overlap. It
6587 copies "len" bytes of memory over. If the argument is known to be
6588 aligned to some boundary, this can be specified as the fourth argument,
6589 otherwise it should be set to 0 or 1.
6591 '``llvm.memset.*``' Intrinsics
6592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6597 This is an overloaded intrinsic. You can use llvm.memset on any integer
6598 bit width and for different address spaces. However, not all targets
6599 support all bit widths.
6603 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6604 i32 <len>, i32 <align>, i1 <isvolatile>)
6605 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6606 i64 <len>, i32 <align>, i1 <isvolatile>)
6611 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6612 particular byte value.
6614 Note that, unlike the standard libc function, the ``llvm.memset``
6615 intrinsic does not return a value and takes extra alignment/volatile
6616 arguments. Also, the destination can be in an arbitrary address space.
6621 The first argument is a pointer to the destination to fill, the second
6622 is the byte value with which to fill it, the third argument is an
6623 integer argument specifying the number of bytes to fill, and the fourth
6624 argument is the known alignment of the destination location.
6626 If the call to this intrinsic has an alignment value that is not 0 or 1,
6627 then the caller guarantees that the destination pointer is aligned to
6630 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6631 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6632 very cleanly specified and it is unwise to depend on it.
6637 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6638 at the destination location. If the argument is known to be aligned to
6639 some boundary, this can be specified as the fourth argument, otherwise
6640 it should be set to 0 or 1.
6642 '``llvm.sqrt.*``' Intrinsic
6643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6648 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6649 floating point or vector of floating point type. Not all targets support
6654 declare float @llvm.sqrt.f32(float %Val)
6655 declare double @llvm.sqrt.f64(double %Val)
6656 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6657 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6658 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6663 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6664 returning the same value as the libm '``sqrt``' functions would. Unlike
6665 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6666 negative numbers other than -0.0 (which allows for better optimization,
6667 because there is no need to worry about errno being set).
6668 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6673 The argument and return value are floating point numbers of the same
6679 This function returns the sqrt of the specified operand if it is a
6680 nonnegative floating point number.
6682 '``llvm.powi.*``' Intrinsic
6683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6688 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6689 floating point or vector of floating point type. Not all targets support
6694 declare float @llvm.powi.f32(float %Val, i32 %power)
6695 declare double @llvm.powi.f64(double %Val, i32 %power)
6696 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6697 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6698 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6703 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6704 specified (positive or negative) power. The order of evaluation of
6705 multiplications is not defined. When a vector of floating point type is
6706 used, the second argument remains a scalar integer value.
6711 The second argument is an integer power, and the first is a value to
6712 raise to that power.
6717 This function returns the first value raised to the second power with an
6718 unspecified sequence of rounding operations.
6720 '``llvm.sin.*``' Intrinsic
6721 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6726 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6727 floating point or vector of floating point type. Not all targets support
6732 declare float @llvm.sin.f32(float %Val)
6733 declare double @llvm.sin.f64(double %Val)
6734 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6735 declare fp128 @llvm.sin.f128(fp128 %Val)
6736 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6741 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6746 The argument and return value are floating point numbers of the same
6752 This function returns the sine of the specified operand, returning the
6753 same values as the libm ``sin`` functions would, and handles error
6754 conditions in the same way.
6756 '``llvm.cos.*``' Intrinsic
6757 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6762 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6763 floating point or vector of floating point type. Not all targets support
6768 declare float @llvm.cos.f32(float %Val)
6769 declare double @llvm.cos.f64(double %Val)
6770 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6771 declare fp128 @llvm.cos.f128(fp128 %Val)
6772 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6777 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6782 The argument and return value are floating point numbers of the same
6788 This function returns the cosine of the specified operand, returning the
6789 same values as the libm ``cos`` functions would, and handles error
6790 conditions in the same way.
6792 '``llvm.pow.*``' Intrinsic
6793 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6798 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6799 floating point or vector of floating point type. Not all targets support
6804 declare float @llvm.pow.f32(float %Val, float %Power)
6805 declare double @llvm.pow.f64(double %Val, double %Power)
6806 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6807 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6808 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6813 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6814 specified (positive or negative) power.
6819 The second argument is a floating point power, and the first is a value
6820 to raise to that power.
6825 This function returns the first value raised to the second power,
6826 returning the same values as the libm ``pow`` functions would, and
6827 handles error conditions in the same way.
6829 '``llvm.exp.*``' Intrinsic
6830 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6835 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
6836 floating point or vector of floating point type. Not all targets support
6841 declare float @llvm.exp.f32(float %Val)
6842 declare double @llvm.exp.f64(double %Val)
6843 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6844 declare fp128 @llvm.exp.f128(fp128 %Val)
6845 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6850 The '``llvm.exp.*``' intrinsics perform the exp function.
6855 The argument and return value are floating point numbers of the same
6861 This function returns the same values as the libm ``exp`` functions
6862 would, and handles error conditions in the same way.
6864 '``llvm.exp2.*``' Intrinsic
6865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6870 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
6871 floating point or vector of floating point type. Not all targets support
6876 declare float @llvm.exp2.f32(float %Val)
6877 declare double @llvm.exp2.f64(double %Val)
6878 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
6879 declare fp128 @llvm.exp2.f128(fp128 %Val)
6880 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
6885 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
6890 The argument and return value are floating point numbers of the same
6896 This function returns the same values as the libm ``exp2`` functions
6897 would, and handles error conditions in the same way.
6899 '``llvm.log.*``' Intrinsic
6900 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6905 This is an overloaded intrinsic. You can use ``llvm.log`` on any
6906 floating point or vector of floating point type. Not all targets support
6911 declare float @llvm.log.f32(float %Val)
6912 declare double @llvm.log.f64(double %Val)
6913 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6914 declare fp128 @llvm.log.f128(fp128 %Val)
6915 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6920 The '``llvm.log.*``' intrinsics perform the log function.
6925 The argument and return value are floating point numbers of the same
6931 This function returns the same values as the libm ``log`` functions
6932 would, and handles error conditions in the same way.
6934 '``llvm.log10.*``' Intrinsic
6935 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6940 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
6941 floating point or vector of floating point type. Not all targets support
6946 declare float @llvm.log10.f32(float %Val)
6947 declare double @llvm.log10.f64(double %Val)
6948 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
6949 declare fp128 @llvm.log10.f128(fp128 %Val)
6950 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
6955 The '``llvm.log10.*``' intrinsics perform the log10 function.
6960 The argument and return value are floating point numbers of the same
6966 This function returns the same values as the libm ``log10`` functions
6967 would, and handles error conditions in the same way.
6969 '``llvm.log2.*``' Intrinsic
6970 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6975 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
6976 floating point or vector of floating point type. Not all targets support
6981 declare float @llvm.log2.f32(float %Val)
6982 declare double @llvm.log2.f64(double %Val)
6983 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
6984 declare fp128 @llvm.log2.f128(fp128 %Val)
6985 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
6990 The '``llvm.log2.*``' intrinsics perform the log2 function.
6995 The argument and return value are floating point numbers of the same
7001 This function returns the same values as the libm ``log2`` functions
7002 would, and handles error conditions in the same way.
7004 '``llvm.fma.*``' Intrinsic
7005 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7010 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7011 floating point or vector of floating point type. Not all targets support
7016 declare float @llvm.fma.f32(float %a, float %b, float %c)
7017 declare double @llvm.fma.f64(double %a, double %b, double %c)
7018 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7019 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7020 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7025 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7031 The argument and return value are floating point numbers of the same
7037 This function returns the same values as the libm ``fma`` functions
7040 '``llvm.fabs.*``' Intrinsic
7041 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7046 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7047 floating point or vector of floating point type. Not all targets support
7052 declare float @llvm.fabs.f32(float %Val)
7053 declare double @llvm.fabs.f64(double %Val)
7054 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7055 declare fp128 @llvm.fabs.f128(fp128 %Val)
7056 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7061 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7067 The argument and return value are floating point numbers of the same
7073 This function returns the same values as the libm ``fabs`` functions
7074 would, and handles error conditions in the same way.
7076 '``llvm.floor.*``' Intrinsic
7077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7082 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7083 floating point or vector of floating point type. Not all targets support
7088 declare float @llvm.floor.f32(float %Val)
7089 declare double @llvm.floor.f64(double %Val)
7090 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7091 declare fp128 @llvm.floor.f128(fp128 %Val)
7092 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7097 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7102 The argument and return value are floating point numbers of the same
7108 This function returns the same values as the libm ``floor`` functions
7109 would, and handles error conditions in the same way.
7111 '``llvm.ceil.*``' Intrinsic
7112 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7117 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7118 floating point or vector of floating point type. Not all targets support
7123 declare float @llvm.ceil.f32(float %Val)
7124 declare double @llvm.ceil.f64(double %Val)
7125 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7126 declare fp128 @llvm.ceil.f128(fp128 %Val)
7127 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7132 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7137 The argument and return value are floating point numbers of the same
7143 This function returns the same values as the libm ``ceil`` functions
7144 would, and handles error conditions in the same way.
7146 '``llvm.trunc.*``' Intrinsic
7147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7152 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7153 floating point or vector of floating point type. Not all targets support
7158 declare float @llvm.trunc.f32(float %Val)
7159 declare double @llvm.trunc.f64(double %Val)
7160 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7161 declare fp128 @llvm.trunc.f128(fp128 %Val)
7162 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7167 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7168 nearest integer not larger in magnitude than the operand.
7173 The argument and return value are floating point numbers of the same
7179 This function returns the same values as the libm ``trunc`` functions
7180 would, and handles error conditions in the same way.
7182 '``llvm.rint.*``' Intrinsic
7183 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7188 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7189 floating point or vector of floating point type. Not all targets support
7194 declare float @llvm.rint.f32(float %Val)
7195 declare double @llvm.rint.f64(double %Val)
7196 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7197 declare fp128 @llvm.rint.f128(fp128 %Val)
7198 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7203 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7204 nearest integer. It may raise an inexact floating-point exception if the
7205 operand isn't an integer.
7210 The argument and return value are floating point numbers of the same
7216 This function returns the same values as the libm ``rint`` functions
7217 would, and handles error conditions in the same way.
7219 '``llvm.nearbyint.*``' Intrinsic
7220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7225 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7226 floating point or vector of floating point type. Not all targets support
7231 declare float @llvm.nearbyint.f32(float %Val)
7232 declare double @llvm.nearbyint.f64(double %Val)
7233 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7234 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7235 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7240 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7246 The argument and return value are floating point numbers of the same
7252 This function returns the same values as the libm ``nearbyint``
7253 functions would, and handles error conditions in the same way.
7255 Bit Manipulation Intrinsics
7256 ---------------------------
7258 LLVM provides intrinsics for a few important bit manipulation
7259 operations. These allow efficient code generation for some algorithms.
7261 '``llvm.bswap.*``' Intrinsics
7262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7267 This is an overloaded intrinsic function. You can use bswap on any
7268 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7272 declare i16 @llvm.bswap.i16(i16 <id>)
7273 declare i32 @llvm.bswap.i32(i32 <id>)
7274 declare i64 @llvm.bswap.i64(i64 <id>)
7279 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7280 values with an even number of bytes (positive multiple of 16 bits).
7281 These are useful for performing operations on data that is not in the
7282 target's native byte order.
7287 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7288 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7289 intrinsic returns an i32 value that has the four bytes of the input i32
7290 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7291 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7292 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7293 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7296 '``llvm.ctpop.*``' Intrinsic
7297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7302 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7303 bit width, or on any vector with integer elements. Not all targets
7304 support all bit widths or vector types, however.
7308 declare i8 @llvm.ctpop.i8(i8 <src>)
7309 declare i16 @llvm.ctpop.i16(i16 <src>)
7310 declare i32 @llvm.ctpop.i32(i32 <src>)
7311 declare i64 @llvm.ctpop.i64(i64 <src>)
7312 declare i256 @llvm.ctpop.i256(i256 <src>)
7313 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7318 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7324 The only argument is the value to be counted. The argument may be of any
7325 integer type, or a vector with integer elements. The return type must
7326 match the argument type.
7331 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7332 each element of a vector.
7334 '``llvm.ctlz.*``' Intrinsic
7335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7340 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7341 integer bit width, or any vector whose elements are integers. Not all
7342 targets support all bit widths or vector types, however.
7346 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7347 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7348 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7349 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7350 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7351 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7356 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7357 leading zeros in a variable.
7362 The first argument is the value to be counted. This argument may be of
7363 any integer type, or a vectory with integer element type. The return
7364 type must match the first argument type.
7366 The second argument must be a constant and is a flag to indicate whether
7367 the intrinsic should ensure that a zero as the first argument produces a
7368 defined result. Historically some architectures did not provide a
7369 defined result for zero values as efficiently, and many algorithms are
7370 now predicated on avoiding zero-value inputs.
7375 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7376 zeros in a variable, or within each element of the vector. If
7377 ``src == 0`` then the result is the size in bits of the type of ``src``
7378 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7379 ``llvm.ctlz(i32 2) = 30``.
7381 '``llvm.cttz.*``' Intrinsic
7382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7387 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7388 integer bit width, or any vector of integer elements. Not all targets
7389 support all bit widths or vector types, however.
7393 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7394 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7395 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7396 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7397 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7398 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7403 The '``llvm.cttz``' family of intrinsic functions counts the number of
7409 The first argument is the value to be counted. This argument may be of
7410 any integer type, or a vectory with integer element type. The return
7411 type must match the first argument type.
7413 The second argument must be a constant and is a flag to indicate whether
7414 the intrinsic should ensure that a zero as the first argument produces a
7415 defined result. Historically some architectures did not provide a
7416 defined result for zero values as efficiently, and many algorithms are
7417 now predicated on avoiding zero-value inputs.
7422 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7423 zeros in a variable, or within each element of a vector. If ``src == 0``
7424 then the result is the size in bits of the type of ``src`` if
7425 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7426 ``llvm.cttz(2) = 1``.
7428 Arithmetic with Overflow Intrinsics
7429 -----------------------------------
7431 LLVM provides intrinsics for some arithmetic with overflow operations.
7433 '``llvm.sadd.with.overflow.*``' Intrinsics
7434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7439 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7440 on any integer bit width.
7444 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7445 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7446 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7451 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7452 a signed addition of the two arguments, and indicate whether an overflow
7453 occurred during the signed summation.
7458 The arguments (%a and %b) and the first element of the result structure
7459 may be of integer types of any bit width, but they must have the same
7460 bit width. The second element of the result structure must be of type
7461 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7467 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7468 a signed addition of the two variables. They return a structure --- the
7469 first element of which is the signed summation, and the second element
7470 of which is a bit specifying if the signed summation resulted in an
7476 .. code-block:: llvm
7478 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7479 %sum = extractvalue {i32, i1} %res, 0
7480 %obit = extractvalue {i32, i1} %res, 1
7481 br i1 %obit, label %overflow, label %normal
7483 '``llvm.uadd.with.overflow.*``' Intrinsics
7484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7489 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7490 on any integer bit width.
7494 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7495 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7496 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7501 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7502 an unsigned addition of the two arguments, and indicate whether a carry
7503 occurred during the unsigned summation.
7508 The arguments (%a and %b) and the first element of the result structure
7509 may be of integer types of any bit width, but they must have the same
7510 bit width. The second element of the result structure must be of type
7511 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7517 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7518 an unsigned addition of the two arguments. They return a structure --- the
7519 first element of which is the sum, and the second element of which is a
7520 bit specifying if the unsigned summation resulted in a carry.
7525 .. code-block:: llvm
7527 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7528 %sum = extractvalue {i32, i1} %res, 0
7529 %obit = extractvalue {i32, i1} %res, 1
7530 br i1 %obit, label %carry, label %normal
7532 '``llvm.ssub.with.overflow.*``' Intrinsics
7533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7538 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7539 on any integer bit width.
7543 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7544 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7545 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7550 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7551 a signed subtraction of the two arguments, and indicate whether an
7552 overflow occurred during the signed subtraction.
7557 The arguments (%a and %b) and the first element of the result structure
7558 may be of integer types of any bit width, but they must have the same
7559 bit width. The second element of the result structure must be of type
7560 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7566 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7567 a signed subtraction of the two arguments. They return a structure --- the
7568 first element of which is the subtraction, and the second element of
7569 which is a bit specifying if the signed subtraction resulted in an
7575 .. code-block:: llvm
7577 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7578 %sum = extractvalue {i32, i1} %res, 0
7579 %obit = extractvalue {i32, i1} %res, 1
7580 br i1 %obit, label %overflow, label %normal
7582 '``llvm.usub.with.overflow.*``' Intrinsics
7583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7588 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7589 on any integer bit width.
7593 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7594 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7595 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7600 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7601 an unsigned subtraction of the two arguments, and indicate whether an
7602 overflow occurred during the unsigned subtraction.
7607 The arguments (%a and %b) and the first element of the result structure
7608 may be of integer types of any bit width, but they must have the same
7609 bit width. The second element of the result structure must be of type
7610 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7616 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7617 an unsigned subtraction of the two arguments. They return a structure ---
7618 the first element of which is the subtraction, and the second element of
7619 which is a bit specifying if the unsigned subtraction resulted in an
7625 .. code-block:: llvm
7627 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7628 %sum = extractvalue {i32, i1} %res, 0
7629 %obit = extractvalue {i32, i1} %res, 1
7630 br i1 %obit, label %overflow, label %normal
7632 '``llvm.smul.with.overflow.*``' Intrinsics
7633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7638 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7639 on any integer bit width.
7643 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7644 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7645 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7650 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7651 a signed multiplication of the two arguments, and indicate whether an
7652 overflow occurred during the signed multiplication.
7657 The arguments (%a and %b) and the first element of the result structure
7658 may be of integer types of any bit width, but they must have the same
7659 bit width. The second element of the result structure must be of type
7660 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7666 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7667 a signed multiplication of the two arguments. They return a structure ---
7668 the first element of which is the multiplication, and the second element
7669 of which is a bit specifying if the signed multiplication resulted in an
7675 .. code-block:: llvm
7677 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7678 %sum = extractvalue {i32, i1} %res, 0
7679 %obit = extractvalue {i32, i1} %res, 1
7680 br i1 %obit, label %overflow, label %normal
7682 '``llvm.umul.with.overflow.*``' Intrinsics
7683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7688 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7689 on any integer bit width.
7693 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7694 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7695 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7700 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7701 a unsigned multiplication of the two arguments, and indicate whether an
7702 overflow occurred during the unsigned multiplication.
7707 The arguments (%a and %b) and the first element of the result structure
7708 may be of integer types of any bit width, but they must have the same
7709 bit width. The second element of the result structure must be of type
7710 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7716 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7717 an unsigned multiplication of the two arguments. They return a structure ---
7718 the first element of which is the multiplication, and the second
7719 element of which is a bit specifying if the unsigned multiplication
7720 resulted in an overflow.
7725 .. code-block:: llvm
7727 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7728 %sum = extractvalue {i32, i1} %res, 0
7729 %obit = extractvalue {i32, i1} %res, 1
7730 br i1 %obit, label %overflow, label %normal
7732 Specialised Arithmetic Intrinsics
7733 ---------------------------------
7735 '``llvm.fmuladd.*``' Intrinsic
7736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7743 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7744 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7749 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7750 expressions that can be fused if the code generator determines that (a) the
7751 target instruction set has support for a fused operation, and (b) that the
7752 fused operation is more efficient than the equivalent, separate pair of mul
7753 and add instructions.
7758 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7759 multiplicands, a and b, and an addend c.
7768 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7770 is equivalent to the expression a \* b + c, except that rounding will
7771 not be performed between the multiplication and addition steps if the
7772 code generator fuses the operations. Fusion is not guaranteed, even if
7773 the target platform supports it. If a fused multiply-add is required the
7774 corresponding llvm.fma.\* intrinsic function should be used instead.
7779 .. code-block:: llvm
7781 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7783 Half Precision Floating Point Intrinsics
7784 ----------------------------------------
7786 For most target platforms, half precision floating point is a
7787 storage-only format. This means that it is a dense encoding (in memory)
7788 but does not support computation in the format.
7790 This means that code must first load the half-precision floating point
7791 value as an i16, then convert it to float with
7792 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7793 then be performed on the float value (including extending to double
7794 etc). To store the value back to memory, it is first converted to float
7795 if needed, then converted to i16 with
7796 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7799 .. _int_convert_to_fp16:
7801 '``llvm.convert.to.fp16``' Intrinsic
7802 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7809 declare i16 @llvm.convert.to.fp16(f32 %a)
7814 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7815 from single precision floating point format to half precision floating
7821 The intrinsic function contains single argument - the value to be
7827 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7828 from single precision floating point format to half precision floating
7829 point format. The return value is an ``i16`` which contains the
7835 .. code-block:: llvm
7837 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7838 store i16 %res, i16* @x, align 2
7840 .. _int_convert_from_fp16:
7842 '``llvm.convert.from.fp16``' Intrinsic
7843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7850 declare f32 @llvm.convert.from.fp16(i16 %a)
7855 The '``llvm.convert.from.fp16``' intrinsic function performs a
7856 conversion from half precision floating point format to single precision
7857 floating point format.
7862 The intrinsic function contains single argument - the value to be
7868 The '``llvm.convert.from.fp16``' intrinsic function performs a
7869 conversion from half single precision floating point format to single
7870 precision floating point format. The input half-float value is
7871 represented by an ``i16`` value.
7876 .. code-block:: llvm
7878 %a = load i16* @x, align 2
7879 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7884 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
7885 prefix), are described in the `LLVM Source Level
7886 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
7889 Exception Handling Intrinsics
7890 -----------------------------
7892 The LLVM exception handling intrinsics (which all start with
7893 ``llvm.eh.`` prefix), are described in the `LLVM Exception
7894 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
7898 Trampoline Intrinsics
7899 ---------------------
7901 These intrinsics make it possible to excise one parameter, marked with
7902 the :ref:`nest <nest>` attribute, from a function. The result is a
7903 callable function pointer lacking the nest parameter - the caller does
7904 not need to provide a value for it. Instead, the value to use is stored
7905 in advance in a "trampoline", a block of memory usually allocated on the
7906 stack, which also contains code to splice the nest value into the
7907 argument list. This is used to implement the GCC nested function address
7910 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
7911 then the resulting function pointer has signature ``i32 (i32, i32)*``.
7912 It can be created as follows:
7914 .. code-block:: llvm
7916 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7917 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7918 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7919 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7920 %fp = bitcast i8* %p to i32 (i32, i32)*
7922 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
7923 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
7927 '``llvm.init.trampoline``' Intrinsic
7928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7935 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7940 This fills the memory pointed to by ``tramp`` with executable code,
7941 turning it into a trampoline.
7946 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
7947 pointers. The ``tramp`` argument must point to a sufficiently large and
7948 sufficiently aligned block of memory; this memory is written to by the
7949 intrinsic. Note that the size and the alignment are target-specific -
7950 LLVM currently provides no portable way of determining them, so a
7951 front-end that generates this intrinsic needs to have some
7952 target-specific knowledge. The ``func`` argument must hold a function
7953 bitcast to an ``i8*``.
7958 The block of memory pointed to by ``tramp`` is filled with target
7959 dependent code, turning it into a function. Then ``tramp`` needs to be
7960 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
7961 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
7962 function's signature is the same as that of ``func`` with any arguments
7963 marked with the ``nest`` attribute removed. At most one such ``nest``
7964 argument is allowed, and it must be of pointer type. Calling the new
7965 function is equivalent to calling ``func`` with the same argument list,
7966 but with ``nval`` used for the missing ``nest`` argument. If, after
7967 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
7968 modified, then the effect of any later call to the returned function
7969 pointer is undefined.
7973 '``llvm.adjust.trampoline``' Intrinsic
7974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7981 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7986 This performs any required machine-specific adjustment to the address of
7987 a trampoline (passed as ``tramp``).
7992 ``tramp`` must point to a block of memory which already has trampoline
7993 code filled in by a previous call to
7994 :ref:`llvm.init.trampoline <int_it>`.
7999 On some architectures the address of the code to be executed needs to be
8000 different to the address where the trampoline is actually stored. This
8001 intrinsic returns the executable address corresponding to ``tramp``
8002 after performing the required machine specific adjustments. The pointer
8003 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8008 This class of intrinsics exists to information about the lifetime of
8009 memory objects and ranges where variables are immutable.
8011 '``llvm.lifetime.start``' Intrinsic
8012 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8019 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8024 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8030 The first argument is a constant integer representing the size of the
8031 object, or -1 if it is variable sized. The second argument is a pointer
8037 This intrinsic indicates that before this point in the code, the value
8038 of the memory pointed to by ``ptr`` is dead. This means that it is known
8039 to never be used and has an undefined value. A load from the pointer
8040 that precedes this intrinsic can be replaced with ``'undef'``.
8042 '``llvm.lifetime.end``' Intrinsic
8043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8050 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8055 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8061 The first argument is a constant integer representing the size of the
8062 object, or -1 if it is variable sized. The second argument is a pointer
8068 This intrinsic indicates that after this point in the code, the value of
8069 the memory pointed to by ``ptr`` is dead. This means that it is known to
8070 never be used and has an undefined value. Any stores into the memory
8071 object following this intrinsic may be removed as dead.
8073 '``llvm.invariant.start``' Intrinsic
8074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8081 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8086 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8087 a memory object will not change.
8092 The first argument is a constant integer representing the size of the
8093 object, or -1 if it is variable sized. The second argument is a pointer
8099 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8100 the return value, the referenced memory location is constant and
8103 '``llvm.invariant.end``' Intrinsic
8104 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8111 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8116 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8117 memory object are mutable.
8122 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8123 The second argument is a constant integer representing the size of the
8124 object, or -1 if it is variable sized and the third argument is a
8125 pointer to the object.
8130 This intrinsic indicates that the memory is mutable again.
8135 This class of intrinsics is designed to be generic and has no specific
8138 '``llvm.var.annotation``' Intrinsic
8139 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8146 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8151 The '``llvm.var.annotation``' intrinsic.
8156 The first argument is a pointer to a value, the second is a pointer to a
8157 global string, the third is a pointer to a global string which is the
8158 source file name, and the last argument is the line number.
8163 This intrinsic allows annotation of local variables with arbitrary
8164 strings. This can be useful for special purpose optimizations that want
8165 to look for these annotations. These have no other defined use; they are
8166 ignored by code generation and optimization.
8168 '``llvm.annotation.*``' Intrinsic
8169 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8174 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8175 any integer bit width.
8179 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8180 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8181 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8182 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8183 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8188 The '``llvm.annotation``' intrinsic.
8193 The first argument is an integer value (result of some expression), the
8194 second is a pointer to a global string, the third is a pointer to a
8195 global string which is the source file name, and the last argument is
8196 the line number. It returns the value of the first argument.
8201 This intrinsic allows annotations to be put on arbitrary expressions
8202 with arbitrary strings. This can be useful for special purpose
8203 optimizations that want to look for these annotations. These have no
8204 other defined use; they are ignored by code generation and optimization.
8206 '``llvm.trap``' Intrinsic
8207 ^^^^^^^^^^^^^^^^^^^^^^^^^
8214 declare void @llvm.trap() noreturn nounwind
8219 The '``llvm.trap``' intrinsic.
8229 This intrinsic is lowered to the target dependent trap instruction. If
8230 the target does not have a trap instruction, this intrinsic will be
8231 lowered to a call of the ``abort()`` function.
8233 '``llvm.debugtrap``' Intrinsic
8234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8241 declare void @llvm.debugtrap() nounwind
8246 The '``llvm.debugtrap``' intrinsic.
8256 This intrinsic is lowered to code which is intended to cause an
8257 execution trap with the intention of requesting the attention of a
8260 '``llvm.stackprotector``' Intrinsic
8261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8268 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8273 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8274 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8275 is placed on the stack before local variables.
8280 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8281 The first argument is the value loaded from the stack guard
8282 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8283 enough space to hold the value of the guard.
8288 This intrinsic causes the prologue/epilogue inserter to force the
8289 position of the ``AllocaInst`` stack slot to be before local variables
8290 on the stack. This is to ensure that if a local variable on the stack is
8291 overwritten, it will destroy the value of the guard. When the function
8292 exits, the guard on the stack is checked against the original guard. If
8293 they are different, then the program aborts by calling the
8294 ``__stack_chk_fail()`` function.
8296 '``llvm.objectsize``' Intrinsic
8297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8304 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8305 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8310 The ``llvm.objectsize`` intrinsic is designed to provide information to
8311 the optimizers to determine at compile time whether a) an operation
8312 (like memcpy) will overflow a buffer that corresponds to an object, or
8313 b) that a runtime check for overflow isn't necessary. An object in this
8314 context means an allocation of a specific class, structure, array, or
8320 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8321 argument is a pointer to or into the ``object``. The second argument is
8322 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8323 or -1 (if false) when the object size is unknown. The second argument
8324 only accepts constants.
8329 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8330 the size of the object concerned. If the size cannot be determined at
8331 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8332 on the ``min`` argument).
8334 '``llvm.expect``' Intrinsic
8335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8342 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8343 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8348 The ``llvm.expect`` intrinsic provides information about expected (the
8349 most probable) value of ``val``, which can be used by optimizers.
8354 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8355 a value. The second argument is an expected value, this needs to be a
8356 constant value, variables are not allowed.
8361 This intrinsic is lowered to the ``val``.
8363 '``llvm.donothing``' Intrinsic
8364 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8371 declare void @llvm.donothing() nounwind readnone
8376 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8377 only intrinsic that can be called with an invoke instruction.
8387 This intrinsic does nothing, and it's removed by optimizers and ignored