1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0). Note that basic blocks are
132 included in this numbering. For example, if the entry basic block is not
133 given a label name, then it will get number 0.
135 It also shows a convention that we follow in this document. When
136 demonstrating instructions, we will follow an instruction with a comment
137 that defines the type and name of value produced.
145 LLVM programs are composed of ``Module``'s, each of which is a
146 translation unit of the input programs. Each module consists of
147 functions, global variables, and symbol table entries. Modules may be
148 combined together with the LLVM linker, which merges function (and
149 global variable) definitions, resolves forward declarations, and merges
150 symbol table entries. Here is an example of the "hello world" module:
154 ; Declare the string constant as a global constant.
155 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
157 ; External declaration of the puts function
158 declare i32 @puts(i8* nocapture) nounwind
160 ; Definition of main function
161 define i32 @main() { ; i32()*
162 ; Convert [13 x i8]* to i8 *...
163 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
165 ; Call puts function to write out the string to stdout.
166 call i32 @puts(i8* %cast210)
171 !1 = metadata !{i32 42}
174 This example is made up of a :ref:`global variable <globalvars>` named
175 "``.str``", an external declaration of the "``puts``" function, a
176 :ref:`function definition <functionstructure>` for "``main``" and
177 :ref:`named metadata <namedmetadatastructure>` "``foo``".
179 In general, a module is made up of a list of global values (where both
180 functions and global variables are global values). Global values are
181 represented by a pointer to a memory location (in this case, a pointer
182 to an array of char, and a pointer to a function), and have one of the
183 following :ref:`linkage types <linkage>`.
190 All Global Variables and Functions have one of the following types of
194 Global values with "``private``" linkage are only directly
195 accessible by objects in the current module. In particular, linking
196 code into a module with an private global value may cause the
197 private to be renamed as necessary to avoid collisions. Because the
198 symbol is private to the module, all references can be updated. This
199 doesn't show up in any symbol table in the object file.
201 Similar to private, but the value shows as a local symbol
202 (``STB_LOCAL`` in the case of ELF) in the object file. This
203 corresponds to the notion of the '``static``' keyword in C.
204 ``available_externally``
205 Globals with "``available_externally``" linkage are never emitted
206 into the object file corresponding to the LLVM module. They exist to
207 allow inlining and other optimizations to take place given knowledge
208 of the definition of the global, which is known to be somewhere
209 outside the module. Globals with ``available_externally`` linkage
210 are allowed to be discarded at will, and are otherwise the same as
211 ``linkonce_odr``. This linkage type is only allowed on definitions,
214 Globals with "``linkonce``" linkage are merged with other globals of
215 the same name when linkage occurs. This can be used to implement
216 some forms of inline functions, templates, or other code which must
217 be generated in each translation unit that uses it, but where the
218 body may be overridden with a more definitive definition later.
219 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
220 that ``linkonce`` linkage does not actually allow the optimizer to
221 inline the body of this function into callers because it doesn't
222 know if this definition of the function is the definitive definition
223 within the program or whether it will be overridden by a stronger
224 definition. To enable inlining and other optimizations, use
225 "``linkonce_odr``" linkage.
227 "``weak``" linkage has the same merging semantics as ``linkonce``
228 linkage, except that unreferenced globals with ``weak`` linkage may
229 not be discarded. This is used for globals that are declared "weak"
232 "``common``" linkage is most similar to "``weak``" linkage, but they
233 are used for tentative definitions in C, such as "``int X;``" at
234 global scope. Symbols with "``common``" linkage are merged in the
235 same way as ``weak symbols``, and they may not be deleted if
236 unreferenced. ``common`` symbols may not have an explicit section,
237 must have a zero initializer, and may not be marked
238 ':ref:`constant <globalvars>`'. Functions and aliases may not have
241 .. _linkage_appending:
244 "``appending``" linkage may only be applied to global variables of
245 pointer to array type. When two global variables with appending
246 linkage are linked together, the two global arrays are appended
247 together. This is the LLVM, typesafe, equivalent of having the
248 system linker append together "sections" with identical names when
251 The semantics of this linkage follow the ELF object file model: the
252 symbol is weak until linked, if not linked, the symbol becomes null
253 instead of being an undefined reference.
254 ``linkonce_odr``, ``weak_odr``
255 Some languages allow differing globals to be merged, such as two
256 functions with different semantics. Other languages, such as
257 ``C++``, ensure that only equivalent globals are ever merged (the
258 "one definition rule" --- "ODR"). Such languages can use the
259 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
260 global will only be merged with equivalent globals. These linkage
261 types are otherwise the same as their non-``odr`` versions.
263 If none of the above identifiers are used, the global is externally
264 visible, meaning that it participates in linkage and can be used to
265 resolve external symbol references.
267 It is illegal for a function *declaration* to have any linkage type
268 other than ``external`` or ``extern_weak``.
275 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
276 :ref:`invokes <i_invoke>` can all have an optional calling convention
277 specified for the call. The calling convention of any pair of dynamic
278 caller/callee must match, or the behavior of the program is undefined.
279 The following calling conventions are supported by LLVM, and more may be
282 "``ccc``" - The C calling convention
283 This calling convention (the default if no other calling convention
284 is specified) matches the target C calling conventions. This calling
285 convention supports varargs function calls and tolerates some
286 mismatch in the declared prototype and implemented declaration of
287 the function (as does normal C).
288 "``fastcc``" - The fast calling convention
289 This calling convention attempts to make calls as fast as possible
290 (e.g. by passing things in registers). This calling convention
291 allows the target to use whatever tricks it wants to produce fast
292 code for the target, without having to conform to an externally
293 specified ABI (Application Binary Interface). `Tail calls can only
294 be optimized when this, the GHC or the HiPE convention is
295 used. <CodeGenerator.html#id80>`_ This calling convention does not
296 support varargs and requires the prototype of all callees to exactly
297 match the prototype of the function definition.
298 "``coldcc``" - The cold calling convention
299 This calling convention attempts to make code in the caller as
300 efficient as possible under the assumption that the call is not
301 commonly executed. As such, these calls often preserve all registers
302 so that the call does not break any live ranges in the caller side.
303 This calling convention does not support varargs and requires the
304 prototype of all callees to exactly match the prototype of the
305 function definition. Furthermore the inliner doesn't consider such function
307 "``cc 10``" - GHC convention
308 This calling convention has been implemented specifically for use by
309 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
310 It passes everything in registers, going to extremes to achieve this
311 by disabling callee save registers. This calling convention should
312 not be used lightly but only for specific situations such as an
313 alternative to the *register pinning* performance technique often
314 used when implementing functional programming languages. At the
315 moment only X86 supports this convention and it has the following
318 - On *X86-32* only supports up to 4 bit type parameters. No
319 floating point types are supported.
320 - On *X86-64* only supports up to 10 bit type parameters and 6
321 floating point parameters.
323 This calling convention supports `tail call
324 optimization <CodeGenerator.html#id80>`_ but requires both the
325 caller and callee are using it.
326 "``cc 11``" - The HiPE calling convention
327 This calling convention has been implemented specifically for use by
328 the `High-Performance Erlang
329 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
330 native code compiler of the `Ericsson's Open Source Erlang/OTP
331 system <http://www.erlang.org/download.shtml>`_. It uses more
332 registers for argument passing than the ordinary C calling
333 convention and defines no callee-saved registers. The calling
334 convention properly supports `tail call
335 optimization <CodeGenerator.html#id80>`_ but requires that both the
336 caller and the callee use it. It uses a *register pinning*
337 mechanism, similar to GHC's convention, for keeping frequently
338 accessed runtime components pinned to specific hardware registers.
339 At the moment only X86 supports this convention (both 32 and 64
341 "``webkit_jscc``" - WebKit's JavaScript calling convention
342 This calling convention has been implemented for `WebKit FTL JIT
343 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
344 stack right to left (as cdecl does), and returns a value in the
345 platform's customary return register.
346 "``anyregcc``" - Dynamic calling convention for code patching
347 This is a special convention that supports patching an arbitrary code
348 sequence in place of a call site. This convention forces the call
349 arguments into registers but allows them to be dynamcially
350 allocated. This can currently only be used with calls to
351 llvm.experimental.patchpoint because only this intrinsic records
352 the location of its arguments in a side table. See :doc:`StackMaps`.
353 "``preserve_mostcc``" - The `PreserveMost` calling convention
354 This calling convention attempts to make the code in the caller as little
355 intrusive as possible. This calling convention behaves identical to the `C`
356 calling convention on how arguments and return values are passed, but it
357 uses a different set of caller/callee-saved registers. This alleviates the
358 burden of saving and recovering a large register set before and after the
359 call in the caller. If the arguments are passed in callee-saved registers,
360 then they will be preserved by the callee across the call. This doesn't
361 apply for values returned in callee-saved registers.
363 - On X86-64 the callee preserves all general purpose registers, except for
364 R11. R11 can be used as a scratch register. Floating-point registers
365 (XMMs/YMMs) are not preserved and need to be saved by the caller.
367 The idea behind this convention is to support calls to runtime functions
368 that have a hot path and a cold path. The hot path is usually a small piece
369 of code that doesn't many registers. The cold path might need to call out to
370 another function and therefore only needs to preserve the caller-saved
371 registers, which haven't already been saved by the caller. The
372 `PreserveMost` calling convention is very similar to the `cold` calling
373 convention in terms of caller/callee-saved registers, but they are used for
374 different types of function calls. `coldcc` is for function calls that are
375 rarely executed, whereas `preserve_mostcc` function calls are intended to be
376 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
377 doesn't prevent the inliner from inlining the function call.
379 This calling convention will be used by a future version of the ObjectiveC
380 runtime and should therefore still be considered experimental at this time.
381 Although this convention was created to optimize certain runtime calls to
382 the ObjectiveC runtime, it is not limited to this runtime and might be used
383 by other runtimes in the future too. The current implementation only
384 supports X86-64, but the intention is to support more architectures in the
386 "``preserve_allcc``" - The `PreserveAll` calling convention
387 This calling convention attempts to make the code in the caller even less
388 intrusive than the `PreserveMost` calling convention. This calling
389 convention also behaves identical to the `C` calling convention on how
390 arguments and return values are passed, but it uses a different set of
391 caller/callee-saved registers. This removes the burden of saving and
392 recovering a large register set before and after the call in the caller. If
393 the arguments are passed in callee-saved registers, then they will be
394 preserved by the callee across the call. This doesn't apply for values
395 returned in callee-saved registers.
397 - On X86-64 the callee preserves all general purpose registers, except for
398 R11. R11 can be used as a scratch register. Furthermore it also preserves
399 all floating-point registers (XMMs/YMMs).
401 The idea behind this convention is to support calls to runtime functions
402 that don't need to call out to any other functions.
404 This calling convention, like the `PreserveMost` calling convention, will be
405 used by a future version of the ObjectiveC runtime and should be considered
406 experimental at this time.
407 "``cc <n>``" - Numbered convention
408 Any calling convention may be specified by number, allowing
409 target-specific calling conventions to be used. Target specific
410 calling conventions start at 64.
412 More calling conventions can be added/defined on an as-needed basis, to
413 support Pascal conventions or any other well-known target-independent
416 .. _visibilitystyles:
421 All Global Variables and Functions have one of the following visibility
424 "``default``" - Default style
425 On targets that use the ELF object file format, default visibility
426 means that the declaration is visible to other modules and, in
427 shared libraries, means that the declared entity may be overridden.
428 On Darwin, default visibility means that the declaration is visible
429 to other modules. Default visibility corresponds to "external
430 linkage" in the language.
431 "``hidden``" - Hidden style
432 Two declarations of an object with hidden visibility refer to the
433 same object if they are in the same shared object. Usually, hidden
434 visibility indicates that the symbol will not be placed into the
435 dynamic symbol table, so no other module (executable or shared
436 library) can reference it directly.
437 "``protected``" - Protected style
438 On ELF, protected visibility indicates that the symbol will be
439 placed in the dynamic symbol table, but that references within the
440 defining module will bind to the local symbol. That is, the symbol
441 cannot be overridden by another module.
448 All Global Variables, Functions and Aliases can have one of the following
452 "``dllimport``" causes the compiler to reference a function or variable via
453 a global pointer to a pointer that is set up by the DLL exporting the
454 symbol. On Microsoft Windows targets, the pointer name is formed by
455 combining ``__imp_`` and the function or variable name.
457 "``dllexport``" causes the compiler to provide a global pointer to a pointer
458 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
459 Microsoft Windows targets, the pointer name is formed by combining
460 ``__imp_`` and the function or variable name. Since this storage class
461 exists for defining a dll interface, the compiler, assembler and linker know
462 it is externally referenced and must refrain from deleting the symbol.
467 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
468 types <t_struct>`. Literal types are uniqued structurally, but identified types
469 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
470 to forward declare a type which is not yet available.
472 An example of a identified structure specification is:
476 %mytype = type { %mytype*, i32 }
478 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
479 literal types are uniqued in recent versions of LLVM.
486 Global variables define regions of memory allocated at compilation time
489 Global variables definitions must be initialized, may have an explicit section
490 to be placed in, and may have an optional explicit alignment specified.
492 Global variables in other translation units can also be declared, in which
493 case they don't have an initializer.
495 A variable may be defined as ``thread_local``, which means that it will
496 not be shared by threads (each thread will have a separated copy of the
497 variable). Not all targets support thread-local variables. Optionally, a
498 TLS model may be specified:
501 For variables that are only used within the current shared library.
503 For variables in modules that will not be loaded dynamically.
505 For variables defined in the executable and only used within it.
507 The models correspond to the ELF TLS models; see `ELF Handling For
508 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
509 more information on under which circumstances the different models may
510 be used. The target may choose a different TLS model if the specified
511 model is not supported, or if a better choice of model can be made.
513 A variable may be defined as a global ``constant``, which indicates that
514 the contents of the variable will **never** be modified (enabling better
515 optimization, allowing the global data to be placed in the read-only
516 section of an executable, etc). Note that variables that need runtime
517 initialization cannot be marked ``constant`` as there is a store to the
520 LLVM explicitly allows *declarations* of global variables to be marked
521 constant, even if the final definition of the global is not. This
522 capability can be used to enable slightly better optimization of the
523 program, but requires the language definition to guarantee that
524 optimizations based on the 'constantness' are valid for the translation
525 units that do not include the definition.
527 As SSA values, global variables define pointer values that are in scope
528 (i.e. they dominate) all basic blocks in the program. Global variables
529 always define a pointer to their "content" type because they describe a
530 region of memory, and all memory objects in LLVM are accessed through
533 Global variables can be marked with ``unnamed_addr`` which indicates
534 that the address is not significant, only the content. Constants marked
535 like this can be merged with other constants if they have the same
536 initializer. Note that a constant with significant address *can* be
537 merged with a ``unnamed_addr`` constant, the result being a constant
538 whose address is significant.
540 A global variable may be declared to reside in a target-specific
541 numbered address space. For targets that support them, address spaces
542 may affect how optimizations are performed and/or what target
543 instructions are used to access the variable. The default address space
544 is zero. The address space qualifier must precede any other attributes.
546 LLVM allows an explicit section to be specified for globals. If the
547 target supports it, it will emit globals to the section specified.
549 By default, global initializers are optimized by assuming that global
550 variables defined within the module are not modified from their
551 initial values before the start of the global initializer. This is
552 true even for variables potentially accessible from outside the
553 module, including those with external linkage or appearing in
554 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
555 by marking the variable with ``externally_initialized``.
557 An explicit alignment may be specified for a global, which must be a
558 power of 2. If not present, or if the alignment is set to zero, the
559 alignment of the global is set by the target to whatever it feels
560 convenient. If an explicit alignment is specified, the global is forced
561 to have exactly that alignment. Targets and optimizers are not allowed
562 to over-align the global if the global has an assigned section. In this
563 case, the extra alignment could be observable: for example, code could
564 assume that the globals are densely packed in their section and try to
565 iterate over them as an array, alignment padding would break this
568 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
572 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
573 [AddrSpace] [unnamed_addr] [ExternallyInitialized]
574 <global | constant> <Type>
575 [, section "name"] [, align <Alignment>]
577 For example, the following defines a global in a numbered address space
578 with an initializer, section, and alignment:
582 @G = addrspace(5) constant float 1.0, section "foo", align 4
584 The following example just declares a global variable
588 @G = external global i32
590 The following example defines a thread-local global with the
591 ``initialexec`` TLS model:
595 @G = thread_local(initialexec) global i32 0, align 4
597 .. _functionstructure:
602 LLVM function definitions consist of the "``define``" keyword, an
603 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
604 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
605 an optional :ref:`calling convention <callingconv>`,
606 an optional ``unnamed_addr`` attribute, a return type, an optional
607 :ref:`parameter attribute <paramattrs>` for the return type, a function
608 name, a (possibly empty) argument list (each with optional :ref:`parameter
609 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
610 an optional section, an optional alignment, an optional :ref:`garbage
611 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
612 curly brace, a list of basic blocks, and a closing curly brace.
614 LLVM function declarations consist of the "``declare``" keyword, an
615 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
616 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
617 an optional :ref:`calling convention <callingconv>`,
618 an optional ``unnamed_addr`` attribute, a return type, an optional
619 :ref:`parameter attribute <paramattrs>` for the return type, a function
620 name, a possibly empty list of arguments, an optional alignment, an optional
621 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
623 A function definition contains a list of basic blocks, forming the CFG (Control
624 Flow Graph) for the function. Each basic block may optionally start with a label
625 (giving the basic block a symbol table entry), contains a list of instructions,
626 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
627 function return). If an explicit label is not provided, a block is assigned an
628 implicit numbered label, using the next value from the same counter as used for
629 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
630 entry block does not have an explicit label, it will be assigned label "%0",
631 then the first unnamed temporary in that block will be "%1", etc.
633 The first basic block in a function is special in two ways: it is
634 immediately executed on entrance to the function, and it is not allowed
635 to have predecessor basic blocks (i.e. there can not be any branches to
636 the entry block of a function). Because the block can have no
637 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
639 LLVM allows an explicit section to be specified for functions. If the
640 target supports it, it will emit functions to the section specified.
642 An explicit alignment may be specified for a function. If not present,
643 or if the alignment is set to zero, the alignment of the function is set
644 by the target to whatever it feels convenient. If an explicit alignment
645 is specified, the function is forced to have at least that much
646 alignment. All alignments must be a power of 2.
648 If the ``unnamed_addr`` attribute is given, the address is know to not
649 be significant and two identical functions can be merged.
653 define [linkage] [visibility] [DLLStorageClass]
655 <ResultType> @<FunctionName> ([argument list])
656 [unnamed_addr] [fn Attrs] [section "name"] [align N]
657 [gc] [prefix Constant] { ... }
664 Aliases act as "second name" for the aliasee value (which can be either
665 function, global variable, another alias or bitcast of global value).
666 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
667 :ref:`visibility style <visibility>`, and an optional :ref:`DLL storage class
672 @<Name> = [Visibility] [DLLStorageClass] alias [Linkage] <AliaseeTy> @<Aliasee>
674 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
675 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
676 might not correctly handle dropping a weak symbol that is aliased by a non-weak
679 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
682 The aliasee must be a definition.
684 .. _namedmetadatastructure:
689 Named metadata is a collection of metadata. :ref:`Metadata
690 nodes <metadata>` (but not metadata strings) are the only valid
691 operands for a named metadata.
695 ; Some unnamed metadata nodes, which are referenced by the named metadata.
696 !0 = metadata !{metadata !"zero"}
697 !1 = metadata !{metadata !"one"}
698 !2 = metadata !{metadata !"two"}
700 !name = !{!0, !1, !2}
707 The return type and each parameter of a function type may have a set of
708 *parameter attributes* associated with them. Parameter attributes are
709 used to communicate additional information about the result or
710 parameters of a function. Parameter attributes are considered to be part
711 of the function, not of the function type, so functions with different
712 parameter attributes can have the same function type.
714 Parameter attributes are simple keywords that follow the type specified.
715 If multiple parameter attributes are needed, they are space separated.
720 declare i32 @printf(i8* noalias nocapture, ...)
721 declare i32 @atoi(i8 zeroext)
722 declare signext i8 @returns_signed_char()
724 Note that any attributes for the function result (``nounwind``,
725 ``readonly``) come immediately after the argument list.
727 Currently, only the following parameter attributes are defined:
730 This indicates to the code generator that the parameter or return
731 value should be zero-extended to the extent required by the target's
732 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
733 the caller (for a parameter) or the callee (for a return value).
735 This indicates to the code generator that the parameter or return
736 value should be sign-extended to the extent required by the target's
737 ABI (which is usually 32-bits) by the caller (for a parameter) or
738 the callee (for a return value).
740 This indicates that this parameter or return value should be treated
741 in a special target-dependent fashion during while emitting code for
742 a function call or return (usually, by putting it in a register as
743 opposed to memory, though some targets use it to distinguish between
744 two different kinds of registers). Use of this attribute is
747 This indicates that the pointer parameter should really be passed by
748 value to the function. The attribute implies that a hidden copy of
749 the pointee is made between the caller and the callee, so the callee
750 is unable to modify the value in the caller. This attribute is only
751 valid on LLVM pointer arguments. It is generally used to pass
752 structs and arrays by value, but is also valid on pointers to
753 scalars. The copy is considered to belong to the caller not the
754 callee (for example, ``readonly`` functions should not write to
755 ``byval`` parameters). This is not a valid attribute for return
758 The byval attribute also supports specifying an alignment with the
759 align attribute. It indicates the alignment of the stack slot to
760 form and the known alignment of the pointer specified to the call
761 site. If the alignment is not specified, then the code generator
762 makes a target-specific assumption.
768 .. Warning:: This feature is unstable and not fully implemented.
770 The ``inalloca`` argument attribute allows the caller to take the
771 address of outgoing stack arguments. An ``inalloca`` argument must
772 be a pointer to stack memory produced by an ``alloca`` instruction.
773 The alloca, or argument allocation, must also be tagged with the
774 inalloca keyword. Only the past argument may have the ``inalloca``
775 attribute, and that argument is guaranteed to be passed in memory.
777 An argument allocation may be used by a call at most once because
778 the call may deallocate it. The ``inalloca`` attribute cannot be
779 used in conjunction with other attributes that affect argument
780 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
781 ``inalloca`` attribute also disables LLVM's implicit lowering of
782 large aggregate return values, which means that frontend authors
783 must lower them with ``sret`` pointers.
785 When the call site is reached, the argument allocation must have
786 been the most recent stack allocation that is still live, or the
787 results are undefined. It is possible to allocate additional stack
788 space after an argument allocation and before its call site, but it
789 must be cleared off with :ref:`llvm.stackrestore
792 See :doc:`InAlloca` for more information on how to use this
796 This indicates that the pointer parameter specifies the address of a
797 structure that is the return value of the function in the source
798 program. This pointer must be guaranteed by the caller to be valid:
799 loads and stores to the structure may be assumed by the callee
800 not to trap and to be properly aligned. This may only be applied to
801 the first parameter. This is not a valid attribute for return
804 This indicates that pointer values :ref:`based <pointeraliasing>` on
805 the argument or return value do not alias pointer values which are
806 not *based* on it, ignoring certain "irrelevant" dependencies. For a
807 call to the parent function, dependencies between memory references
808 from before or after the call and from those during the call are
809 "irrelevant" to the ``noalias`` keyword for the arguments and return
810 value used in that call. The caller shares the responsibility with
811 the callee for ensuring that these requirements are met. For further
812 details, please see the discussion of the NoAlias response in `alias
813 analysis <AliasAnalysis.html#MustMayNo>`_.
815 Note that this definition of ``noalias`` is intentionally similar
816 to the definition of ``restrict`` in C99 for function arguments,
817 though it is slightly weaker.
819 For function return values, C99's ``restrict`` is not meaningful,
820 while LLVM's ``noalias`` is.
822 This indicates that the callee does not make any copies of the
823 pointer that outlive the callee itself. This is not a valid
824 attribute for return values.
829 This indicates that the pointer parameter can be excised using the
830 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
831 attribute for return values and can only be applied to one parameter.
834 This indicates that the function always returns the argument as its return
835 value. This is an optimization hint to the code generator when generating
836 the caller, allowing tail call optimization and omission of register saves
837 and restores in some cases; it is not checked or enforced when generating
838 the callee. The parameter and the function return type must be valid
839 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
840 valid attribute for return values and can only be applied to one parameter.
844 Garbage Collector Names
845 -----------------------
847 Each function may specify a garbage collector name, which is simply a
852 define void @f() gc "name" { ... }
854 The compiler declares the supported values of *name*. Specifying a
855 collector which will cause the compiler to alter its output in order to
856 support the named garbage collection algorithm.
863 Prefix data is data associated with a function which the code generator
864 will emit immediately before the function body. The purpose of this feature
865 is to allow frontends to associate language-specific runtime metadata with
866 specific functions and make it available through the function pointer while
867 still allowing the function pointer to be called. To access the data for a
868 given function, a program may bitcast the function pointer to a pointer to
869 the constant's type. This implies that the IR symbol points to the start
872 To maintain the semantics of ordinary function calls, the prefix data must
873 have a particular format. Specifically, it must begin with a sequence of
874 bytes which decode to a sequence of machine instructions, valid for the
875 module's target, which transfer control to the point immediately succeeding
876 the prefix data, without performing any other visible action. This allows
877 the inliner and other passes to reason about the semantics of the function
878 definition without needing to reason about the prefix data. Obviously this
879 makes the format of the prefix data highly target dependent.
881 Prefix data is laid out as if it were an initializer for a global variable
882 of the prefix data's type. No padding is automatically placed between the
883 prefix data and the function body. If padding is required, it must be part
886 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
887 which encodes the ``nop`` instruction:
891 define void @f() prefix i8 144 { ... }
893 Generally prefix data can be formed by encoding a relative branch instruction
894 which skips the metadata, as in this example of valid prefix data for the
895 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
899 %0 = type <{ i8, i8, i8* }>
901 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
903 A function may have prefix data but no body. This has similar semantics
904 to the ``available_externally`` linkage in that the data may be used by the
905 optimizers but will not be emitted in the object file.
912 Attribute groups are groups of attributes that are referenced by objects within
913 the IR. They are important for keeping ``.ll`` files readable, because a lot of
914 functions will use the same set of attributes. In the degenerative case of a
915 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
916 group will capture the important command line flags used to build that file.
918 An attribute group is a module-level object. To use an attribute group, an
919 object references the attribute group's ID (e.g. ``#37``). An object may refer
920 to more than one attribute group. In that situation, the attributes from the
921 different groups are merged.
923 Here is an example of attribute groups for a function that should always be
924 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
928 ; Target-independent attributes:
929 attributes #0 = { alwaysinline alignstack=4 }
931 ; Target-dependent attributes:
932 attributes #1 = { "no-sse" }
934 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
935 define void @f() #0 #1 { ... }
942 Function attributes are set to communicate additional information about
943 a function. Function attributes are considered to be part of the
944 function, not of the function type, so functions with different function
945 attributes can have the same function type.
947 Function attributes are simple keywords that follow the type specified.
948 If multiple attributes are needed, they are space separated. For
953 define void @f() noinline { ... }
954 define void @f() alwaysinline { ... }
955 define void @f() alwaysinline optsize { ... }
956 define void @f() optsize { ... }
959 This attribute indicates that, when emitting the prologue and
960 epilogue, the backend should forcibly align the stack pointer.
961 Specify the desired alignment, which must be a power of two, in
964 This attribute indicates that the inliner should attempt to inline
965 this function into callers whenever possible, ignoring any active
966 inlining size threshold for this caller.
968 This indicates that the callee function at a call site should be
969 recognized as a built-in function, even though the function's declaration
970 uses the ``nobuiltin`` attribute. This is only valid at call sites for
971 direct calls to functions which are declared with the ``nobuiltin``
974 This attribute indicates that this function is rarely called. When
975 computing edge weights, basic blocks post-dominated by a cold
976 function call are also considered to be cold; and, thus, given low
979 This attribute indicates that the source code contained a hint that
980 inlining this function is desirable (such as the "inline" keyword in
981 C/C++). It is just a hint; it imposes no requirements on the
984 This attribute suggests that optimization passes and code generator
985 passes make choices that keep the code size of this function as small
986 as possible and perform optimizations that may sacrifice runtime
987 performance in order to minimize the size of the generated code.
989 This attribute disables prologue / epilogue emission for the
990 function. This can have very system-specific consequences.
992 This indicates that the callee function at a call site is not recognized as
993 a built-in function. LLVM will retain the original call and not replace it
994 with equivalent code based on the semantics of the built-in function, unless
995 the call site uses the ``builtin`` attribute. This is valid at call sites
996 and on function declarations and definitions.
998 This attribute indicates that calls to the function cannot be
999 duplicated. A call to a ``noduplicate`` function may be moved
1000 within its parent function, but may not be duplicated within
1001 its parent function.
1003 A function containing a ``noduplicate`` call may still
1004 be an inlining candidate, provided that the call is not
1005 duplicated by inlining. That implies that the function has
1006 internal linkage and only has one call site, so the original
1007 call is dead after inlining.
1009 This attributes disables implicit floating point instructions.
1011 This attribute indicates that the inliner should never inline this
1012 function in any situation. This attribute may not be used together
1013 with the ``alwaysinline`` attribute.
1015 This attribute suppresses lazy symbol binding for the function. This
1016 may make calls to the function faster, at the cost of extra program
1017 startup time if the function is not called during program startup.
1019 This attribute indicates that the code generator should not use a
1020 red zone, even if the target-specific ABI normally permits it.
1022 This function attribute indicates that the function never returns
1023 normally. This produces undefined behavior at runtime if the
1024 function ever does dynamically return.
1026 This function attribute indicates that the function never returns
1027 with an unwind or exceptional control flow. If the function does
1028 unwind, its runtime behavior is undefined.
1030 This function attribute indicates that the function is not optimized
1031 by any optimization or code generator passes with the
1032 exception of interprocedural optimization passes.
1033 This attribute cannot be used together with the ``alwaysinline``
1034 attribute; this attribute is also incompatible
1035 with the ``minsize`` attribute and the ``optsize`` attribute.
1037 This attribute requires the ``noinline`` attribute to be specified on
1038 the function as well, so the function is never inlined into any caller.
1039 Only functions with the ``alwaysinline`` attribute are valid
1040 candidates for inlining into the body of this function.
1042 This attribute suggests that optimization passes and code generator
1043 passes make choices that keep the code size of this function low,
1044 and otherwise do optimizations specifically to reduce code size as
1045 long as they do not significantly impact runtime performance.
1047 On a function, this attribute indicates that the function computes its
1048 result (or decides to unwind an exception) based strictly on its arguments,
1049 without dereferencing any pointer arguments or otherwise accessing
1050 any mutable state (e.g. memory, control registers, etc) visible to
1051 caller functions. It does not write through any pointer arguments
1052 (including ``byval`` arguments) and never changes any state visible
1053 to callers. This means that it cannot unwind exceptions by calling
1054 the ``C++`` exception throwing methods.
1056 On an argument, this attribute indicates that the function does not
1057 dereference that pointer argument, even though it may read or write the
1058 memory that the pointer points to if accessed through other pointers.
1060 On a function, this attribute indicates that the function does not write
1061 through any pointer arguments (including ``byval`` arguments) or otherwise
1062 modify any state (e.g. memory, control registers, etc) visible to
1063 caller functions. It may dereference pointer arguments and read
1064 state that may be set in the caller. A readonly function always
1065 returns the same value (or unwinds an exception identically) when
1066 called with the same set of arguments and global state. It cannot
1067 unwind an exception by calling the ``C++`` exception throwing
1070 On an argument, this attribute indicates that the function does not write
1071 through this pointer argument, even though it may write to the memory that
1072 the pointer points to.
1074 This attribute indicates that this function can return twice. The C
1075 ``setjmp`` is an example of such a function. The compiler disables
1076 some optimizations (like tail calls) in the caller of these
1078 ``sanitize_address``
1079 This attribute indicates that AddressSanitizer checks
1080 (dynamic address safety analysis) are enabled for this function.
1082 This attribute indicates that MemorySanitizer checks (dynamic detection
1083 of accesses to uninitialized memory) are enabled for this function.
1085 This attribute indicates that ThreadSanitizer checks
1086 (dynamic thread safety analysis) are enabled for this function.
1088 This attribute indicates that the function should emit a stack
1089 smashing protector. It is in the form of a "canary" --- a random value
1090 placed on the stack before the local variables that's checked upon
1091 return from the function to see if it has been overwritten. A
1092 heuristic is used to determine if a function needs stack protectors
1093 or not. The heuristic used will enable protectors for functions with:
1095 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1096 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1097 - Calls to alloca() with variable sizes or constant sizes greater than
1098 ``ssp-buffer-size``.
1100 Variables that are identified as requiring a protector will be arranged
1101 on the stack such that they are adjacent to the stack protector guard.
1103 If a function that has an ``ssp`` attribute is inlined into a
1104 function that doesn't have an ``ssp`` attribute, then the resulting
1105 function will have an ``ssp`` attribute.
1107 This attribute indicates that the function should *always* emit a
1108 stack smashing protector. This overrides the ``ssp`` function
1111 Variables that are identified as requiring a protector will be arranged
1112 on the stack such that they are adjacent to the stack protector guard.
1113 The specific layout rules are:
1115 #. Large arrays and structures containing large arrays
1116 (``>= ssp-buffer-size``) are closest to the stack protector.
1117 #. Small arrays and structures containing small arrays
1118 (``< ssp-buffer-size``) are 2nd closest to the protector.
1119 #. Variables that have had their address taken are 3rd closest to the
1122 If a function that has an ``sspreq`` attribute is inlined into a
1123 function that doesn't have an ``sspreq`` attribute or which has an
1124 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1125 an ``sspreq`` attribute.
1127 This attribute indicates that the function should emit a stack smashing
1128 protector. This attribute causes a strong heuristic to be used when
1129 determining if a function needs stack protectors. The strong heuristic
1130 will enable protectors for functions with:
1132 - Arrays of any size and type
1133 - Aggregates containing an array of any size and type.
1134 - Calls to alloca().
1135 - Local variables that have had their address taken.
1137 Variables that are identified as requiring a protector will be arranged
1138 on the stack such that they are adjacent to the stack protector guard.
1139 The specific layout rules are:
1141 #. Large arrays and structures containing large arrays
1142 (``>= ssp-buffer-size``) are closest to the stack protector.
1143 #. Small arrays and structures containing small arrays
1144 (``< ssp-buffer-size``) are 2nd closest to the protector.
1145 #. Variables that have had their address taken are 3rd closest to the
1148 This overrides the ``ssp`` function attribute.
1150 If a function that has an ``sspstrong`` attribute is inlined into a
1151 function that doesn't have an ``sspstrong`` attribute, then the
1152 resulting function will have an ``sspstrong`` attribute.
1154 This attribute indicates that the ABI being targeted requires that
1155 an unwind table entry be produce for this function even if we can
1156 show that no exceptions passes by it. This is normally the case for
1157 the ELF x86-64 abi, but it can be disabled for some compilation
1162 Module-Level Inline Assembly
1163 ----------------------------
1165 Modules may contain "module-level inline asm" blocks, which corresponds
1166 to the GCC "file scope inline asm" blocks. These blocks are internally
1167 concatenated by LLVM and treated as a single unit, but may be separated
1168 in the ``.ll`` file if desired. The syntax is very simple:
1170 .. code-block:: llvm
1172 module asm "inline asm code goes here"
1173 module asm "more can go here"
1175 The strings can contain any character by escaping non-printable
1176 characters. The escape sequence used is simply "\\xx" where "xx" is the
1177 two digit hex code for the number.
1179 The inline asm code is simply printed to the machine code .s file when
1180 assembly code is generated.
1182 .. _langref_datalayout:
1187 A module may specify a target specific data layout string that specifies
1188 how data is to be laid out in memory. The syntax for the data layout is
1191 .. code-block:: llvm
1193 target datalayout = "layout specification"
1195 The *layout specification* consists of a list of specifications
1196 separated by the minus sign character ('-'). Each specification starts
1197 with a letter and may include other information after the letter to
1198 define some aspect of the data layout. The specifications accepted are
1202 Specifies that the target lays out data in big-endian form. That is,
1203 the bits with the most significance have the lowest address
1206 Specifies that the target lays out data in little-endian form. That
1207 is, the bits with the least significance have the lowest address
1210 Specifies the natural alignment of the stack in bits. Alignment
1211 promotion of stack variables is limited to the natural stack
1212 alignment to avoid dynamic stack realignment. The stack alignment
1213 must be a multiple of 8-bits. If omitted, the natural stack
1214 alignment defaults to "unspecified", which does not prevent any
1215 alignment promotions.
1216 ``p[n]:<size>:<abi>:<pref>``
1217 This specifies the *size* of a pointer and its ``<abi>`` and
1218 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1219 bits. The address space, ``n`` is optional, and if not specified,
1220 denotes the default address space 0. The value of ``n`` must be
1221 in the range [1,2^23).
1222 ``i<size>:<abi>:<pref>``
1223 This specifies the alignment for an integer type of a given bit
1224 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1225 ``v<size>:<abi>:<pref>``
1226 This specifies the alignment for a vector type of a given bit
1228 ``f<size>:<abi>:<pref>``
1229 This specifies the alignment for a floating point type of a given bit
1230 ``<size>``. Only values of ``<size>`` that are supported by the target
1231 will work. 32 (float) and 64 (double) are supported on all targets; 80
1232 or 128 (different flavors of long double) are also supported on some
1235 This specifies the alignment for an object of aggregate type.
1237 If present, specifies that llvm names are mangled in the output. The
1240 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1241 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1242 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1243 symbols get a ``_`` prefix.
1244 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1245 functions also get a suffix based on the frame size.
1246 ``n<size1>:<size2>:<size3>...``
1247 This specifies a set of native integer widths for the target CPU in
1248 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1249 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1250 this set are considered to support most general arithmetic operations
1253 On every specification that takes a ``<abi>:<pref>``, specifying the
1254 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1255 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1257 When constructing the data layout for a given target, LLVM starts with a
1258 default set of specifications which are then (possibly) overridden by
1259 the specifications in the ``datalayout`` keyword. The default
1260 specifications are given in this list:
1262 - ``E`` - big endian
1263 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1264 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1265 same as the default address space.
1266 - ``S0`` - natural stack alignment is unspecified
1267 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1268 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1269 - ``i16:16:16`` - i16 is 16-bit aligned
1270 - ``i32:32:32`` - i32 is 32-bit aligned
1271 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1272 alignment of 64-bits
1273 - ``f16:16:16`` - half is 16-bit aligned
1274 - ``f32:32:32`` - float is 32-bit aligned
1275 - ``f64:64:64`` - double is 64-bit aligned
1276 - ``f128:128:128`` - quad is 128-bit aligned
1277 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1278 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1279 - ``a:0:64`` - aggregates are 64-bit aligned
1281 When LLVM is determining the alignment for a given type, it uses the
1284 #. If the type sought is an exact match for one of the specifications,
1285 that specification is used.
1286 #. If no match is found, and the type sought is an integer type, then
1287 the smallest integer type that is larger than the bitwidth of the
1288 sought type is used. If none of the specifications are larger than
1289 the bitwidth then the largest integer type is used. For example,
1290 given the default specifications above, the i7 type will use the
1291 alignment of i8 (next largest) while both i65 and i256 will use the
1292 alignment of i64 (largest specified).
1293 #. If no match is found, and the type sought is a vector type, then the
1294 largest vector type that is smaller than the sought vector type will
1295 be used as a fall back. This happens because <128 x double> can be
1296 implemented in terms of 64 <2 x double>, for example.
1298 The function of the data layout string may not be what you expect.
1299 Notably, this is not a specification from the frontend of what alignment
1300 the code generator should use.
1302 Instead, if specified, the target data layout is required to match what
1303 the ultimate *code generator* expects. This string is used by the
1304 mid-level optimizers to improve code, and this only works if it matches
1305 what the ultimate code generator uses. If you would like to generate IR
1306 that does not embed this target-specific detail into the IR, then you
1307 don't have to specify the string. This will disable some optimizations
1308 that require precise layout information, but this also prevents those
1309 optimizations from introducing target specificity into the IR.
1316 A module may specify a target triple string that describes the target
1317 host. The syntax for the target triple is simply:
1319 .. code-block:: llvm
1321 target triple = "x86_64-apple-macosx10.7.0"
1323 The *target triple* string consists of a series of identifiers delimited
1324 by the minus sign character ('-'). The canonical forms are:
1328 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1329 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1331 This information is passed along to the backend so that it generates
1332 code for the proper architecture. It's possible to override this on the
1333 command line with the ``-mtriple`` command line option.
1335 .. _pointeraliasing:
1337 Pointer Aliasing Rules
1338 ----------------------
1340 Any memory access must be done through a pointer value associated with
1341 an address range of the memory access, otherwise the behavior is
1342 undefined. Pointer values are associated with address ranges according
1343 to the following rules:
1345 - A pointer value is associated with the addresses associated with any
1346 value it is *based* on.
1347 - An address of a global variable is associated with the address range
1348 of the variable's storage.
1349 - The result value of an allocation instruction is associated with the
1350 address range of the allocated storage.
1351 - A null pointer in the default address-space is associated with no
1353 - An integer constant other than zero or a pointer value returned from
1354 a function not defined within LLVM may be associated with address
1355 ranges allocated through mechanisms other than those provided by
1356 LLVM. Such ranges shall not overlap with any ranges of addresses
1357 allocated by mechanisms provided by LLVM.
1359 A pointer value is *based* on another pointer value according to the
1362 - A pointer value formed from a ``getelementptr`` operation is *based*
1363 on the first operand of the ``getelementptr``.
1364 - The result value of a ``bitcast`` is *based* on the operand of the
1366 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1367 values that contribute (directly or indirectly) to the computation of
1368 the pointer's value.
1369 - The "*based* on" relationship is transitive.
1371 Note that this definition of *"based"* is intentionally similar to the
1372 definition of *"based"* in C99, though it is slightly weaker.
1374 LLVM IR does not associate types with memory. The result type of a
1375 ``load`` merely indicates the size and alignment of the memory from
1376 which to load, as well as the interpretation of the value. The first
1377 operand type of a ``store`` similarly only indicates the size and
1378 alignment of the store.
1380 Consequently, type-based alias analysis, aka TBAA, aka
1381 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1382 :ref:`Metadata <metadata>` may be used to encode additional information
1383 which specialized optimization passes may use to implement type-based
1388 Volatile Memory Accesses
1389 ------------------------
1391 Certain memory accesses, such as :ref:`load <i_load>`'s,
1392 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1393 marked ``volatile``. The optimizers must not change the number of
1394 volatile operations or change their order of execution relative to other
1395 volatile operations. The optimizers *may* change the order of volatile
1396 operations relative to non-volatile operations. This is not Java's
1397 "volatile" and has no cross-thread synchronization behavior.
1399 IR-level volatile loads and stores cannot safely be optimized into
1400 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1401 flagged volatile. Likewise, the backend should never split or merge
1402 target-legal volatile load/store instructions.
1404 .. admonition:: Rationale
1406 Platforms may rely on volatile loads and stores of natively supported
1407 data width to be executed as single instruction. For example, in C
1408 this holds for an l-value of volatile primitive type with native
1409 hardware support, but not necessarily for aggregate types. The
1410 frontend upholds these expectations, which are intentionally
1411 unspecified in the IR. The rules above ensure that IR transformation
1412 do not violate the frontend's contract with the language.
1416 Memory Model for Concurrent Operations
1417 --------------------------------------
1419 The LLVM IR does not define any way to start parallel threads of
1420 execution or to register signal handlers. Nonetheless, there are
1421 platform-specific ways to create them, and we define LLVM IR's behavior
1422 in their presence. This model is inspired by the C++0x memory model.
1424 For a more informal introduction to this model, see the :doc:`Atomics`.
1426 We define a *happens-before* partial order as the least partial order
1429 - Is a superset of single-thread program order, and
1430 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1431 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1432 techniques, like pthread locks, thread creation, thread joining,
1433 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1434 Constraints <ordering>`).
1436 Note that program order does not introduce *happens-before* edges
1437 between a thread and signals executing inside that thread.
1439 Every (defined) read operation (load instructions, memcpy, atomic
1440 loads/read-modify-writes, etc.) R reads a series of bytes written by
1441 (defined) write operations (store instructions, atomic
1442 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1443 section, initialized globals are considered to have a write of the
1444 initializer which is atomic and happens before any other read or write
1445 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1446 may see any write to the same byte, except:
1448 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1449 write\ :sub:`2` happens before R\ :sub:`byte`, then
1450 R\ :sub:`byte` does not see write\ :sub:`1`.
1451 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1452 R\ :sub:`byte` does not see write\ :sub:`3`.
1454 Given that definition, R\ :sub:`byte` is defined as follows:
1456 - If R is volatile, the result is target-dependent. (Volatile is
1457 supposed to give guarantees which can support ``sig_atomic_t`` in
1458 C/C++, and may be used for accesses to addresses which do not behave
1459 like normal memory. It does not generally provide cross-thread
1461 - Otherwise, if there is no write to the same byte that happens before
1462 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1463 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1464 R\ :sub:`byte` returns the value written by that write.
1465 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1466 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1467 Memory Ordering Constraints <ordering>` section for additional
1468 constraints on how the choice is made.
1469 - Otherwise R\ :sub:`byte` returns ``undef``.
1471 R returns the value composed of the series of bytes it read. This
1472 implies that some bytes within the value may be ``undef`` **without**
1473 the entire value being ``undef``. Note that this only defines the
1474 semantics of the operation; it doesn't mean that targets will emit more
1475 than one instruction to read the series of bytes.
1477 Note that in cases where none of the atomic intrinsics are used, this
1478 model places only one restriction on IR transformations on top of what
1479 is required for single-threaded execution: introducing a store to a byte
1480 which might not otherwise be stored is not allowed in general.
1481 (Specifically, in the case where another thread might write to and read
1482 from an address, introducing a store can change a load that may see
1483 exactly one write into a load that may see multiple writes.)
1487 Atomic Memory Ordering Constraints
1488 ----------------------------------
1490 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1491 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1492 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1493 ordering parameters that determine which other atomic instructions on
1494 the same address they *synchronize with*. These semantics are borrowed
1495 from Java and C++0x, but are somewhat more colloquial. If these
1496 descriptions aren't precise enough, check those specs (see spec
1497 references in the :doc:`atomics guide <Atomics>`).
1498 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1499 differently since they don't take an address. See that instruction's
1500 documentation for details.
1502 For a simpler introduction to the ordering constraints, see the
1506 The set of values that can be read is governed by the happens-before
1507 partial order. A value cannot be read unless some operation wrote
1508 it. This is intended to provide a guarantee strong enough to model
1509 Java's non-volatile shared variables. This ordering cannot be
1510 specified for read-modify-write operations; it is not strong enough
1511 to make them atomic in any interesting way.
1513 In addition to the guarantees of ``unordered``, there is a single
1514 total order for modifications by ``monotonic`` operations on each
1515 address. All modification orders must be compatible with the
1516 happens-before order. There is no guarantee that the modification
1517 orders can be combined to a global total order for the whole program
1518 (and this often will not be possible). The read in an atomic
1519 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1520 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1521 order immediately before the value it writes. If one atomic read
1522 happens before another atomic read of the same address, the later
1523 read must see the same value or a later value in the address's
1524 modification order. This disallows reordering of ``monotonic`` (or
1525 stronger) operations on the same address. If an address is written
1526 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1527 read that address repeatedly, the other threads must eventually see
1528 the write. This corresponds to the C++0x/C1x
1529 ``memory_order_relaxed``.
1531 In addition to the guarantees of ``monotonic``, a
1532 *synchronizes-with* edge may be formed with a ``release`` operation.
1533 This is intended to model C++'s ``memory_order_acquire``.
1535 In addition to the guarantees of ``monotonic``, if this operation
1536 writes a value which is subsequently read by an ``acquire``
1537 operation, it *synchronizes-with* that operation. (This isn't a
1538 complete description; see the C++0x definition of a release
1539 sequence.) This corresponds to the C++0x/C1x
1540 ``memory_order_release``.
1541 ``acq_rel`` (acquire+release)
1542 Acts as both an ``acquire`` and ``release`` operation on its
1543 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1544 ``seq_cst`` (sequentially consistent)
1545 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1546 operation which only reads, ``release`` for an operation which only
1547 writes), there is a global total order on all
1548 sequentially-consistent operations on all addresses, which is
1549 consistent with the *happens-before* partial order and with the
1550 modification orders of all the affected addresses. Each
1551 sequentially-consistent read sees the last preceding write to the
1552 same address in this global order. This corresponds to the C++0x/C1x
1553 ``memory_order_seq_cst`` and Java volatile.
1557 If an atomic operation is marked ``singlethread``, it only *synchronizes
1558 with* or participates in modification and seq\_cst total orderings with
1559 other operations running in the same thread (for example, in signal
1567 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1568 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1569 :ref:`frem <i_frem>`) have the following flags that can set to enable
1570 otherwise unsafe floating point operations
1573 No NaNs - Allow optimizations to assume the arguments and result are not
1574 NaN. Such optimizations are required to retain defined behavior over
1575 NaNs, but the value of the result is undefined.
1578 No Infs - Allow optimizations to assume the arguments and result are not
1579 +/-Inf. Such optimizations are required to retain defined behavior over
1580 +/-Inf, but the value of the result is undefined.
1583 No Signed Zeros - Allow optimizations to treat the sign of a zero
1584 argument or result as insignificant.
1587 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1588 argument rather than perform division.
1591 Fast - Allow algebraically equivalent transformations that may
1592 dramatically change results in floating point (e.g. reassociate). This
1593 flag implies all the others.
1600 The LLVM type system is one of the most important features of the
1601 intermediate representation. Being typed enables a number of
1602 optimizations to be performed on the intermediate representation
1603 directly, without having to do extra analyses on the side before the
1604 transformation. A strong type system makes it easier to read the
1605 generated code and enables novel analyses and transformations that are
1606 not feasible to perform on normal three address code representations.
1616 The void type does not represent any value and has no size.
1634 The function type can be thought of as a function signature. It consists of a
1635 return type and a list of formal parameter types. The return type of a function
1636 type is a void type or first class type --- except for :ref:`label <t_label>`
1637 and :ref:`metadata <t_metadata>` types.
1643 <returntype> (<parameter list>)
1645 ...where '``<parameter list>``' is a comma-separated list of type
1646 specifiers. Optionally, the parameter list may include a type ``...``, which
1647 indicates that the function takes a variable number of arguments. Variable
1648 argument functions can access their arguments with the :ref:`variable argument
1649 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1650 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1654 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1655 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1656 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1657 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1658 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 | ``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. |
1660 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1661 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1662 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1669 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1670 Values of these types are the only ones which can be produced by
1678 These are the types that are valid in registers from CodeGen's perspective.
1687 The integer type is a very simple type that simply specifies an
1688 arbitrary bit width for the integer type desired. Any bit width from 1
1689 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1697 The number of bits the integer will occupy is specified by the ``N``
1703 +----------------+------------------------------------------------+
1704 | ``i1`` | a single-bit integer. |
1705 +----------------+------------------------------------------------+
1706 | ``i32`` | a 32-bit integer. |
1707 +----------------+------------------------------------------------+
1708 | ``i1942652`` | a really big integer of over 1 million bits. |
1709 +----------------+------------------------------------------------+
1713 Floating Point Types
1714 """"""""""""""""""""
1723 - 16-bit floating point value
1726 - 32-bit floating point value
1729 - 64-bit floating point value
1732 - 128-bit floating point value (112-bit mantissa)
1735 - 80-bit floating point value (X87)
1738 - 128-bit floating point value (two 64-bits)
1745 The x86_mmx type represents a value held in an MMX register on an x86
1746 machine. The operations allowed on it are quite limited: parameters and
1747 return values, load and store, and bitcast. User-specified MMX
1748 instructions are represented as intrinsic or asm calls with arguments
1749 and/or results of this type. There are no arrays, vectors or constants
1766 The pointer type is used to specify memory locations. Pointers are
1767 commonly used to reference objects in memory.
1769 Pointer types may have an optional address space attribute defining the
1770 numbered address space where the pointed-to object resides. The default
1771 address space is number zero. The semantics of non-zero address spaces
1772 are target-specific.
1774 Note that LLVM does not permit pointers to void (``void*``) nor does it
1775 permit pointers to labels (``label*``). Use ``i8*`` instead.
1785 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1786 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1787 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1788 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1789 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1790 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1791 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1800 A vector type is a simple derived type that represents a vector of
1801 elements. Vector types are used when multiple primitive data are
1802 operated in parallel using a single instruction (SIMD). A vector type
1803 requires a size (number of elements) and an underlying primitive data
1804 type. Vector types are considered :ref:`first class <t_firstclass>`.
1810 < <# elements> x <elementtype> >
1812 The number of elements is a constant integer value larger than 0;
1813 elementtype may be any integer or floating point type, or a pointer to
1814 these types. Vectors of size zero are not allowed.
1818 +-------------------+--------------------------------------------------+
1819 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1820 +-------------------+--------------------------------------------------+
1821 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1822 +-------------------+--------------------------------------------------+
1823 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1824 +-------------------+--------------------------------------------------+
1825 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1826 +-------------------+--------------------------------------------------+
1835 The label type represents code labels.
1850 The metadata type represents embedded metadata. No derived types may be
1851 created from metadata except for :ref:`function <t_function>` arguments.
1864 Aggregate Types are a subset of derived types that can contain multiple
1865 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1866 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1876 The array type is a very simple derived type that arranges elements
1877 sequentially in memory. The array type requires a size (number of
1878 elements) and an underlying data type.
1884 [<# elements> x <elementtype>]
1886 The number of elements is a constant integer value; ``elementtype`` may
1887 be any type with a size.
1891 +------------------+--------------------------------------+
1892 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1893 +------------------+--------------------------------------+
1894 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1895 +------------------+--------------------------------------+
1896 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1897 +------------------+--------------------------------------+
1899 Here are some examples of multidimensional arrays:
1901 +-----------------------------+----------------------------------------------------------+
1902 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1903 +-----------------------------+----------------------------------------------------------+
1904 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1905 +-----------------------------+----------------------------------------------------------+
1906 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1907 +-----------------------------+----------------------------------------------------------+
1909 There is no restriction on indexing beyond the end of the array implied
1910 by a static type (though there are restrictions on indexing beyond the
1911 bounds of an allocated object in some cases). This means that
1912 single-dimension 'variable sized array' addressing can be implemented in
1913 LLVM with a zero length array type. An implementation of 'pascal style
1914 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1924 The structure type is used to represent a collection of data members
1925 together in memory. The elements of a structure may be any type that has
1928 Structures in memory are accessed using '``load``' and '``store``' by
1929 getting a pointer to a field with the '``getelementptr``' instruction.
1930 Structures in registers are accessed using the '``extractvalue``' and
1931 '``insertvalue``' instructions.
1933 Structures may optionally be "packed" structures, which indicate that
1934 the alignment of the struct is one byte, and that there is no padding
1935 between the elements. In non-packed structs, padding between field types
1936 is inserted as defined by the DataLayout string in the module, which is
1937 required to match what the underlying code generator expects.
1939 Structures can either be "literal" or "identified". A literal structure
1940 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1941 identified types are always defined at the top level with a name.
1942 Literal types are uniqued by their contents and can never be recursive
1943 or opaque since there is no way to write one. Identified types can be
1944 recursive, can be opaqued, and are never uniqued.
1950 %T1 = type { <type list> } ; Identified normal struct type
1951 %T2 = type <{ <type list> }> ; Identified packed struct type
1955 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1956 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1957 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1958 | ``{ 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``. |
1959 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1960 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1961 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1965 Opaque Structure Types
1966 """"""""""""""""""""""
1970 Opaque structure types are used to represent named structure types that
1971 do not have a body specified. This corresponds (for example) to the C
1972 notion of a forward declared structure.
1983 +--------------+-------------------+
1984 | ``opaque`` | An opaque type. |
1985 +--------------+-------------------+
1990 LLVM has several different basic types of constants. This section
1991 describes them all and their syntax.
1996 **Boolean constants**
1997 The two strings '``true``' and '``false``' are both valid constants
1999 **Integer constants**
2000 Standard integers (such as '4') are constants of the
2001 :ref:`integer <t_integer>` type. Negative numbers may be used with
2003 **Floating point constants**
2004 Floating point constants use standard decimal notation (e.g.
2005 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2006 hexadecimal notation (see below). The assembler requires the exact
2007 decimal value of a floating-point constant. For example, the
2008 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2009 decimal in binary. Floating point constants must have a :ref:`floating
2010 point <t_floating>` type.
2011 **Null pointer constants**
2012 The identifier '``null``' is recognized as a null pointer constant
2013 and must be of :ref:`pointer type <t_pointer>`.
2015 The one non-intuitive notation for constants is the hexadecimal form of
2016 floating point constants. For example, the form
2017 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2018 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2019 constants are required (and the only time that they are generated by the
2020 disassembler) is when a floating point constant must be emitted but it
2021 cannot be represented as a decimal floating point number in a reasonable
2022 number of digits. For example, NaN's, infinities, and other special
2023 values are represented in their IEEE hexadecimal format so that assembly
2024 and disassembly do not cause any bits to change in the constants.
2026 When using the hexadecimal form, constants of types half, float, and
2027 double are represented using the 16-digit form shown above (which
2028 matches the IEEE754 representation for double); half and float values
2029 must, however, be exactly representable as IEEE 754 half and single
2030 precision, respectively. Hexadecimal format is always used for long
2031 double, and there are three forms of long double. The 80-bit format used
2032 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2033 128-bit format used by PowerPC (two adjacent doubles) is represented by
2034 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2035 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2036 will only work if they match the long double format on your target.
2037 The IEEE 16-bit format (half precision) is represented by ``0xH``
2038 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2039 (sign bit at the left).
2041 There are no constants of type x86_mmx.
2043 .. _complexconstants:
2048 Complex constants are a (potentially recursive) combination of simple
2049 constants and smaller complex constants.
2051 **Structure constants**
2052 Structure constants are represented with notation similar to
2053 structure type definitions (a comma separated list of elements,
2054 surrounded by braces (``{}``)). For example:
2055 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2056 "``@G = external global i32``". Structure constants must have
2057 :ref:`structure type <t_struct>`, and the number and types of elements
2058 must match those specified by the type.
2060 Array constants are represented with notation similar to array type
2061 definitions (a comma separated list of elements, surrounded by
2062 square brackets (``[]``)). For example:
2063 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2064 :ref:`array type <t_array>`, and the number and types of elements must
2065 match those specified by the type.
2066 **Vector constants**
2067 Vector constants are represented with notation similar to vector
2068 type definitions (a comma separated list of elements, surrounded by
2069 less-than/greater-than's (``<>``)). For example:
2070 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2071 must have :ref:`vector type <t_vector>`, and the number and types of
2072 elements must match those specified by the type.
2073 **Zero initialization**
2074 The string '``zeroinitializer``' can be used to zero initialize a
2075 value to zero of *any* type, including scalar and
2076 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2077 having to print large zero initializers (e.g. for large arrays) and
2078 is always exactly equivalent to using explicit zero initializers.
2080 A metadata node is a structure-like constant with :ref:`metadata
2081 type <t_metadata>`. For example:
2082 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2083 constants that are meant to be interpreted as part of the
2084 instruction stream, metadata is a place to attach additional
2085 information such as debug info.
2087 Global Variable and Function Addresses
2088 --------------------------------------
2090 The addresses of :ref:`global variables <globalvars>` and
2091 :ref:`functions <functionstructure>` are always implicitly valid
2092 (link-time) constants. These constants are explicitly referenced when
2093 the :ref:`identifier for the global <identifiers>` is used and always have
2094 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2097 .. code-block:: llvm
2101 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2108 The string '``undef``' can be used anywhere a constant is expected, and
2109 indicates that the user of the value may receive an unspecified
2110 bit-pattern. Undefined values may be of any type (other than '``label``'
2111 or '``void``') and be used anywhere a constant is permitted.
2113 Undefined values are useful because they indicate to the compiler that
2114 the program is well defined no matter what value is used. This gives the
2115 compiler more freedom to optimize. Here are some examples of
2116 (potentially surprising) transformations that are valid (in pseudo IR):
2118 .. code-block:: llvm
2128 This is safe because all of the output bits are affected by the undef
2129 bits. Any output bit can have a zero or one depending on the input bits.
2131 .. code-block:: llvm
2142 These logical operations have bits that are not always affected by the
2143 input. For example, if ``%X`` has a zero bit, then the output of the
2144 '``and``' operation will always be a zero for that bit, no matter what
2145 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2146 optimize or assume that the result of the '``and``' is '``undef``'.
2147 However, it is safe to assume that all bits of the '``undef``' could be
2148 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2149 all the bits of the '``undef``' operand to the '``or``' could be set,
2150 allowing the '``or``' to be folded to -1.
2152 .. code-block:: llvm
2154 %A = select undef, %X, %Y
2155 %B = select undef, 42, %Y
2156 %C = select %X, %Y, undef
2166 This set of examples shows that undefined '``select``' (and conditional
2167 branch) conditions can go *either way*, but they have to come from one
2168 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2169 both known to have a clear low bit, then ``%A`` would have to have a
2170 cleared low bit. However, in the ``%C`` example, the optimizer is
2171 allowed to assume that the '``undef``' operand could be the same as
2172 ``%Y``, allowing the whole '``select``' to be eliminated.
2174 .. code-block:: llvm
2176 %A = xor undef, undef
2193 This example points out that two '``undef``' operands are not
2194 necessarily the same. This can be surprising to people (and also matches
2195 C semantics) where they assume that "``X^X``" is always zero, even if
2196 ``X`` is undefined. This isn't true for a number of reasons, but the
2197 short answer is that an '``undef``' "variable" can arbitrarily change
2198 its value over its "live range". This is true because the variable
2199 doesn't actually *have a live range*. Instead, the value is logically
2200 read from arbitrary registers that happen to be around when needed, so
2201 the value is not necessarily consistent over time. In fact, ``%A`` and
2202 ``%C`` need to have the same semantics or the core LLVM "replace all
2203 uses with" concept would not hold.
2205 .. code-block:: llvm
2213 These examples show the crucial difference between an *undefined value*
2214 and *undefined behavior*. An undefined value (like '``undef``') is
2215 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2216 operation can be constant folded to '``undef``', because the '``undef``'
2217 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2218 However, in the second example, we can make a more aggressive
2219 assumption: because the ``undef`` is allowed to be an arbitrary value,
2220 we are allowed to assume that it could be zero. Since a divide by zero
2221 has *undefined behavior*, we are allowed to assume that the operation
2222 does not execute at all. This allows us to delete the divide and all
2223 code after it. Because the undefined operation "can't happen", the
2224 optimizer can assume that it occurs in dead code.
2226 .. code-block:: llvm
2228 a: store undef -> %X
2229 b: store %X -> undef
2234 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2235 value can be assumed to not have any effect; we can assume that the
2236 value is overwritten with bits that happen to match what was already
2237 there. However, a store *to* an undefined location could clobber
2238 arbitrary memory, therefore, it has undefined behavior.
2245 Poison values are similar to :ref:`undef values <undefvalues>`, however
2246 they also represent the fact that an instruction or constant expression
2247 which cannot evoke side effects has nevertheless detected a condition
2248 which results in undefined behavior.
2250 There is currently no way of representing a poison value in the IR; they
2251 only exist when produced by operations such as :ref:`add <i_add>` with
2254 Poison value behavior is defined in terms of value *dependence*:
2256 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2257 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2258 their dynamic predecessor basic block.
2259 - Function arguments depend on the corresponding actual argument values
2260 in the dynamic callers of their functions.
2261 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2262 instructions that dynamically transfer control back to them.
2263 - :ref:`Invoke <i_invoke>` instructions depend on the
2264 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2265 call instructions that dynamically transfer control back to them.
2266 - Non-volatile loads and stores depend on the most recent stores to all
2267 of the referenced memory addresses, following the order in the IR
2268 (including loads and stores implied by intrinsics such as
2269 :ref:`@llvm.memcpy <int_memcpy>`.)
2270 - An instruction with externally visible side effects depends on the
2271 most recent preceding instruction with externally visible side
2272 effects, following the order in the IR. (This includes :ref:`volatile
2273 operations <volatile>`.)
2274 - An instruction *control-depends* on a :ref:`terminator
2275 instruction <terminators>` if the terminator instruction has
2276 multiple successors and the instruction is always executed when
2277 control transfers to one of the successors, and may not be executed
2278 when control is transferred to another.
2279 - Additionally, an instruction also *control-depends* on a terminator
2280 instruction if the set of instructions it otherwise depends on would
2281 be different if the terminator had transferred control to a different
2283 - Dependence is transitive.
2285 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2286 with the additional affect that any instruction which has a *dependence*
2287 on a poison value has undefined behavior.
2289 Here are some examples:
2291 .. code-block:: llvm
2294 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2295 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2296 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2297 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2299 store i32 %poison, i32* @g ; Poison value stored to memory.
2300 %poison2 = load i32* @g ; Poison value loaded back from memory.
2302 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2304 %narrowaddr = bitcast i32* @g to i16*
2305 %wideaddr = bitcast i32* @g to i64*
2306 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2307 %poison4 = load i64* %wideaddr ; Returns a poison value.
2309 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2310 br i1 %cmp, label %true, label %end ; Branch to either destination.
2313 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2314 ; it has undefined behavior.
2318 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2319 ; Both edges into this PHI are
2320 ; control-dependent on %cmp, so this
2321 ; always results in a poison value.
2323 store volatile i32 0, i32* @g ; This would depend on the store in %true
2324 ; if %cmp is true, or the store in %entry
2325 ; otherwise, so this is undefined behavior.
2327 br i1 %cmp, label %second_true, label %second_end
2328 ; The same branch again, but this time the
2329 ; true block doesn't have side effects.
2336 store volatile i32 0, i32* @g ; This time, the instruction always depends
2337 ; on the store in %end. Also, it is
2338 ; control-equivalent to %end, so this is
2339 ; well-defined (ignoring earlier undefined
2340 ; behavior in this example).
2344 Addresses of Basic Blocks
2345 -------------------------
2347 ``blockaddress(@function, %block)``
2349 The '``blockaddress``' constant computes the address of the specified
2350 basic block in the specified function, and always has an ``i8*`` type.
2351 Taking the address of the entry block is illegal.
2353 This value only has defined behavior when used as an operand to the
2354 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2355 against null. Pointer equality tests between labels addresses results in
2356 undefined behavior --- though, again, comparison against null is ok, and
2357 no label is equal to the null pointer. This may be passed around as an
2358 opaque pointer sized value as long as the bits are not inspected. This
2359 allows ``ptrtoint`` and arithmetic to be performed on these values so
2360 long as the original value is reconstituted before the ``indirectbr``
2363 Finally, some targets may provide defined semantics when using the value
2364 as the operand to an inline assembly, but that is target specific.
2368 Constant Expressions
2369 --------------------
2371 Constant expressions are used to allow expressions involving other
2372 constants to be used as constants. Constant expressions may be of any
2373 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2374 that does not have side effects (e.g. load and call are not supported).
2375 The following is the syntax for constant expressions:
2377 ``trunc (CST to TYPE)``
2378 Truncate a constant to another type. The bit size of CST must be
2379 larger than the bit size of TYPE. Both types must be integers.
2380 ``zext (CST to TYPE)``
2381 Zero extend a constant to another type. The bit size of CST must be
2382 smaller than the bit size of TYPE. Both types must be integers.
2383 ``sext (CST to TYPE)``
2384 Sign extend a constant to another type. The bit size of CST must be
2385 smaller than the bit size of TYPE. Both types must be integers.
2386 ``fptrunc (CST to TYPE)``
2387 Truncate a floating point constant to another floating point type.
2388 The size of CST must be larger than the size of TYPE. Both types
2389 must be floating point.
2390 ``fpext (CST to TYPE)``
2391 Floating point extend a constant to another type. The size of CST
2392 must be smaller or equal to the size of TYPE. Both types must be
2394 ``fptoui (CST to TYPE)``
2395 Convert a floating point constant to the corresponding unsigned
2396 integer constant. TYPE must be a scalar or vector integer type. CST
2397 must be of scalar or vector floating point type. Both CST and TYPE
2398 must be scalars, or vectors of the same number of elements. If the
2399 value won't fit in the integer type, the results are undefined.
2400 ``fptosi (CST to TYPE)``
2401 Convert a floating point constant to the corresponding signed
2402 integer constant. TYPE must be a scalar or vector integer type. CST
2403 must be of scalar or vector floating point type. Both CST and TYPE
2404 must be scalars, or vectors of the same number of elements. If the
2405 value won't fit in the integer type, the results are undefined.
2406 ``uitofp (CST to TYPE)``
2407 Convert an unsigned integer constant to the corresponding floating
2408 point constant. TYPE must be a scalar or vector floating point type.
2409 CST must be of scalar or vector integer type. Both CST and TYPE must
2410 be scalars, or vectors of the same number of elements. If the value
2411 won't fit in the floating point type, the results are undefined.
2412 ``sitofp (CST to TYPE)``
2413 Convert a signed integer constant to the corresponding floating
2414 point constant. TYPE must be a scalar or vector floating point type.
2415 CST must be of scalar or vector integer type. Both CST and TYPE must
2416 be scalars, or vectors of the same number of elements. If the value
2417 won't fit in the floating point type, the results are undefined.
2418 ``ptrtoint (CST to TYPE)``
2419 Convert a pointer typed constant to the corresponding integer
2420 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2421 pointer type. The ``CST`` value is zero extended, truncated, or
2422 unchanged to make it fit in ``TYPE``.
2423 ``inttoptr (CST to TYPE)``
2424 Convert an integer constant to a pointer constant. TYPE must be a
2425 pointer type. CST must be of integer type. The CST value is zero
2426 extended, truncated, or unchanged to make it fit in a pointer size.
2427 This one is *really* dangerous!
2428 ``bitcast (CST to TYPE)``
2429 Convert a constant, CST, to another TYPE. The constraints of the
2430 operands are the same as those for the :ref:`bitcast
2431 instruction <i_bitcast>`.
2432 ``addrspacecast (CST to TYPE)``
2433 Convert a constant pointer or constant vector of pointer, CST, to another
2434 TYPE in a different address space. The constraints of the operands are the
2435 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2436 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2437 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2438 constants. As with the :ref:`getelementptr <i_getelementptr>`
2439 instruction, the index list may have zero or more indexes, which are
2440 required to make sense for the type of "CSTPTR".
2441 ``select (COND, VAL1, VAL2)``
2442 Perform the :ref:`select operation <i_select>` on constants.
2443 ``icmp COND (VAL1, VAL2)``
2444 Performs the :ref:`icmp operation <i_icmp>` on constants.
2445 ``fcmp COND (VAL1, VAL2)``
2446 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2447 ``extractelement (VAL, IDX)``
2448 Perform the :ref:`extractelement operation <i_extractelement>` on
2450 ``insertelement (VAL, ELT, IDX)``
2451 Perform the :ref:`insertelement operation <i_insertelement>` on
2453 ``shufflevector (VEC1, VEC2, IDXMASK)``
2454 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2456 ``extractvalue (VAL, IDX0, IDX1, ...)``
2457 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2458 constants. The index list is interpreted in a similar manner as
2459 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2460 least one index value must be specified.
2461 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2462 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2463 The index list is interpreted in a similar manner as indices in a
2464 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2465 value must be specified.
2466 ``OPCODE (LHS, RHS)``
2467 Perform the specified operation of the LHS and RHS constants. OPCODE
2468 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2469 binary <bitwiseops>` operations. The constraints on operands are
2470 the same as those for the corresponding instruction (e.g. no bitwise
2471 operations on floating point values are allowed).
2478 Inline Assembler Expressions
2479 ----------------------------
2481 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2482 Inline Assembly <moduleasm>`) through the use of a special value. This
2483 value represents the inline assembler as a string (containing the
2484 instructions to emit), a list of operand constraints (stored as a
2485 string), a flag that indicates whether or not the inline asm expression
2486 has side effects, and a flag indicating whether the function containing
2487 the asm needs to align its stack conservatively. An example inline
2488 assembler expression is:
2490 .. code-block:: llvm
2492 i32 (i32) asm "bswap $0", "=r,r"
2494 Inline assembler expressions may **only** be used as the callee operand
2495 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2496 Thus, typically we have:
2498 .. code-block:: llvm
2500 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2502 Inline asms with side effects not visible in the constraint list must be
2503 marked as having side effects. This is done through the use of the
2504 '``sideeffect``' keyword, like so:
2506 .. code-block:: llvm
2508 call void asm sideeffect "eieio", ""()
2510 In some cases inline asms will contain code that will not work unless
2511 the stack is aligned in some way, such as calls or SSE instructions on
2512 x86, yet will not contain code that does that alignment within the asm.
2513 The compiler should make conservative assumptions about what the asm
2514 might contain and should generate its usual stack alignment code in the
2515 prologue if the '``alignstack``' keyword is present:
2517 .. code-block:: llvm
2519 call void asm alignstack "eieio", ""()
2521 Inline asms also support using non-standard assembly dialects. The
2522 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2523 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2524 the only supported dialects. An example is:
2526 .. code-block:: llvm
2528 call void asm inteldialect "eieio", ""()
2530 If multiple keywords appear the '``sideeffect``' keyword must come
2531 first, the '``alignstack``' keyword second and the '``inteldialect``'
2537 The call instructions that wrap inline asm nodes may have a
2538 "``!srcloc``" MDNode attached to it that contains a list of constant
2539 integers. If present, the code generator will use the integer as the
2540 location cookie value when report errors through the ``LLVMContext``
2541 error reporting mechanisms. This allows a front-end to correlate backend
2542 errors that occur with inline asm back to the source code that produced
2545 .. code-block:: llvm
2547 call void asm sideeffect "something bad", ""(), !srcloc !42
2549 !42 = !{ i32 1234567 }
2551 It is up to the front-end to make sense of the magic numbers it places
2552 in the IR. If the MDNode contains multiple constants, the code generator
2553 will use the one that corresponds to the line of the asm that the error
2558 Metadata Nodes and Metadata Strings
2559 -----------------------------------
2561 LLVM IR allows metadata to be attached to instructions in the program
2562 that can convey extra information about the code to the optimizers and
2563 code generator. One example application of metadata is source-level
2564 debug information. There are two metadata primitives: strings and nodes.
2565 All metadata has the ``metadata`` type and is identified in syntax by a
2566 preceding exclamation point ('``!``').
2568 A metadata string is a string surrounded by double quotes. It can
2569 contain any character by escaping non-printable characters with
2570 "``\xx``" where "``xx``" is the two digit hex code. For example:
2573 Metadata nodes are represented with notation similar to structure
2574 constants (a comma separated list of elements, surrounded by braces and
2575 preceded by an exclamation point). Metadata nodes can have any values as
2576 their operand. For example:
2578 .. code-block:: llvm
2580 !{ metadata !"test\00", i32 10}
2582 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2583 metadata nodes, which can be looked up in the module symbol table. For
2586 .. code-block:: llvm
2588 !foo = metadata !{!4, !3}
2590 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2591 function is using two metadata arguments:
2593 .. code-block:: llvm
2595 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2597 Metadata can be attached with an instruction. Here metadata ``!21`` is
2598 attached to the ``add`` instruction using the ``!dbg`` identifier:
2600 .. code-block:: llvm
2602 %indvar.next = add i64 %indvar, 1, !dbg !21
2604 More information about specific metadata nodes recognized by the
2605 optimizers and code generator is found below.
2610 In LLVM IR, memory does not have types, so LLVM's own type system is not
2611 suitable for doing TBAA. Instead, metadata is added to the IR to
2612 describe a type system of a higher level language. This can be used to
2613 implement typical C/C++ TBAA, but it can also be used to implement
2614 custom alias analysis behavior for other languages.
2616 The current metadata format is very simple. TBAA metadata nodes have up
2617 to three fields, e.g.:
2619 .. code-block:: llvm
2621 !0 = metadata !{ metadata !"an example type tree" }
2622 !1 = metadata !{ metadata !"int", metadata !0 }
2623 !2 = metadata !{ metadata !"float", metadata !0 }
2624 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2626 The first field is an identity field. It can be any value, usually a
2627 metadata string, which uniquely identifies the type. The most important
2628 name in the tree is the name of the root node. Two trees with different
2629 root node names are entirely disjoint, even if they have leaves with
2632 The second field identifies the type's parent node in the tree, or is
2633 null or omitted for a root node. A type is considered to alias all of
2634 its descendants and all of its ancestors in the tree. Also, a type is
2635 considered to alias all types in other trees, so that bitcode produced
2636 from multiple front-ends is handled conservatively.
2638 If the third field is present, it's an integer which if equal to 1
2639 indicates that the type is "constant" (meaning
2640 ``pointsToConstantMemory`` should return true; see `other useful
2641 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2643 '``tbaa.struct``' Metadata
2644 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2646 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2647 aggregate assignment operations in C and similar languages, however it
2648 is defined to copy a contiguous region of memory, which is more than
2649 strictly necessary for aggregate types which contain holes due to
2650 padding. Also, it doesn't contain any TBAA information about the fields
2653 ``!tbaa.struct`` metadata can describe which memory subregions in a
2654 memcpy are padding and what the TBAA tags of the struct are.
2656 The current metadata format is very simple. ``!tbaa.struct`` metadata
2657 nodes are a list of operands which are in conceptual groups of three.
2658 For each group of three, the first operand gives the byte offset of a
2659 field in bytes, the second gives its size in bytes, and the third gives
2662 .. code-block:: llvm
2664 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2666 This describes a struct with two fields. The first is at offset 0 bytes
2667 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2668 and has size 4 bytes and has tbaa tag !2.
2670 Note that the fields need not be contiguous. In this example, there is a
2671 4 byte gap between the two fields. This gap represents padding which
2672 does not carry useful data and need not be preserved.
2674 '``fpmath``' Metadata
2675 ^^^^^^^^^^^^^^^^^^^^^
2677 ``fpmath`` metadata may be attached to any instruction of floating point
2678 type. It can be used to express the maximum acceptable error in the
2679 result of that instruction, in ULPs, thus potentially allowing the
2680 compiler to use a more efficient but less accurate method of computing
2681 it. ULP is defined as follows:
2683 If ``x`` is a real number that lies between two finite consecutive
2684 floating-point numbers ``a`` and ``b``, without being equal to one
2685 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2686 distance between the two non-equal finite floating-point numbers
2687 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2689 The metadata node shall consist of a single positive floating point
2690 number representing the maximum relative error, for example:
2692 .. code-block:: llvm
2694 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2696 '``range``' Metadata
2697 ^^^^^^^^^^^^^^^^^^^^
2699 ``range`` metadata may be attached only to loads of integer types. It
2700 expresses the possible ranges the loaded value is in. The ranges are
2701 represented with a flattened list of integers. The loaded value is known
2702 to be in the union of the ranges defined by each consecutive pair. Each
2703 pair has the following properties:
2705 - The type must match the type loaded by the instruction.
2706 - The pair ``a,b`` represents the range ``[a,b)``.
2707 - Both ``a`` and ``b`` are constants.
2708 - The range is allowed to wrap.
2709 - The range should not represent the full or empty set. That is,
2712 In addition, the pairs must be in signed order of the lower bound and
2713 they must be non-contiguous.
2717 .. code-block:: llvm
2719 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2720 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2721 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2722 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2724 !0 = metadata !{ i8 0, i8 2 }
2725 !1 = metadata !{ i8 255, i8 2 }
2726 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2727 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2732 It is sometimes useful to attach information to loop constructs. Currently,
2733 loop metadata is implemented as metadata attached to the branch instruction
2734 in the loop latch block. This type of metadata refer to a metadata node that is
2735 guaranteed to be separate for each loop. The loop identifier metadata is
2736 specified with the name ``llvm.loop``.
2738 The loop identifier metadata is implemented using a metadata that refers to
2739 itself to avoid merging it with any other identifier metadata, e.g.,
2740 during module linkage or function inlining. That is, each loop should refer
2741 to their own identification metadata even if they reside in separate functions.
2742 The following example contains loop identifier metadata for two separate loop
2745 .. code-block:: llvm
2747 !0 = metadata !{ metadata !0 }
2748 !1 = metadata !{ metadata !1 }
2750 The loop identifier metadata can be used to specify additional per-loop
2751 metadata. Any operands after the first operand can be treated as user-defined
2752 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2753 by the loop vectorizer to indicate how many times to unroll the loop:
2755 .. code-block:: llvm
2757 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2759 !0 = metadata !{ metadata !0, metadata !1 }
2760 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2765 Metadata types used to annotate memory accesses with information helpful
2766 for optimizations are prefixed with ``llvm.mem``.
2768 '``llvm.mem.parallel_loop_access``' Metadata
2769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2771 For a loop to be parallel, in addition to using
2772 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2773 also all of the memory accessing instructions in the loop body need to be
2774 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2775 is at least one memory accessing instruction not marked with the metadata,
2776 the loop must be considered a sequential loop. This causes parallel loops to be
2777 converted to sequential loops due to optimization passes that are unaware of
2778 the parallel semantics and that insert new memory instructions to the loop
2781 Example of a loop that is considered parallel due to its correct use of
2782 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2783 metadata types that refer to the same loop identifier metadata.
2785 .. code-block:: llvm
2789 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
2791 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2793 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2797 !0 = metadata !{ metadata !0 }
2799 It is also possible to have nested parallel loops. In that case the
2800 memory accesses refer to a list of loop identifier metadata nodes instead of
2801 the loop identifier metadata node directly:
2803 .. code-block:: llvm
2807 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
2809 br label %inner.for.body
2813 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
2815 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
2817 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2821 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
2823 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2825 outer.for.end: ; preds = %for.body
2827 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2828 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2829 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2831 '``llvm.vectorizer``'
2832 ^^^^^^^^^^^^^^^^^^^^^
2834 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2835 vectorization parameters such as vectorization factor and unroll factor.
2837 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2838 loop identification metadata.
2840 '``llvm.vectorizer.unroll``' Metadata
2841 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2843 This metadata instructs the loop vectorizer to unroll the specified
2844 loop exactly ``N`` times.
2846 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2847 operand is an integer specifying the unroll factor. For example:
2849 .. code-block:: llvm
2851 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2853 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2856 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2857 determined automatically.
2859 '``llvm.vectorizer.width``' Metadata
2860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2862 This metadata sets the target width of the vectorizer to ``N``. Without
2863 this metadata, the vectorizer will choose a width automatically.
2864 Regardless of this metadata, the vectorizer will only vectorize loops if
2865 it believes it is valid to do so.
2867 The first operand is the string ``llvm.vectorizer.width`` and the second
2868 operand is an integer specifying the width. For example:
2870 .. code-block:: llvm
2872 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2874 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2877 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2880 Module Flags Metadata
2881 =====================
2883 Information about the module as a whole is difficult to convey to LLVM's
2884 subsystems. The LLVM IR isn't sufficient to transmit this information.
2885 The ``llvm.module.flags`` named metadata exists in order to facilitate
2886 this. These flags are in the form of key / value pairs --- much like a
2887 dictionary --- making it easy for any subsystem who cares about a flag to
2890 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2891 Each triplet has the following form:
2893 - The first element is a *behavior* flag, which specifies the behavior
2894 when two (or more) modules are merged together, and it encounters two
2895 (or more) metadata with the same ID. The supported behaviors are
2897 - The second element is a metadata string that is a unique ID for the
2898 metadata. Each module may only have one flag entry for each unique ID (not
2899 including entries with the **Require** behavior).
2900 - The third element is the value of the flag.
2902 When two (or more) modules are merged together, the resulting
2903 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2904 each unique metadata ID string, there will be exactly one entry in the merged
2905 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2906 be determined by the merge behavior flag, as described below. The only exception
2907 is that entries with the *Require* behavior are always preserved.
2909 The following behaviors are supported:
2920 Emits an error if two values disagree, otherwise the resulting value
2921 is that of the operands.
2925 Emits a warning if two values disagree. The result value will be the
2926 operand for the flag from the first module being linked.
2930 Adds a requirement that another module flag be present and have a
2931 specified value after linking is performed. The value must be a
2932 metadata pair, where the first element of the pair is the ID of the
2933 module flag to be restricted, and the second element of the pair is
2934 the value the module flag should be restricted to. This behavior can
2935 be used to restrict the allowable results (via triggering of an
2936 error) of linking IDs with the **Override** behavior.
2940 Uses the specified value, regardless of the behavior or value of the
2941 other module. If both modules specify **Override**, but the values
2942 differ, an error will be emitted.
2946 Appends the two values, which are required to be metadata nodes.
2950 Appends the two values, which are required to be metadata
2951 nodes. However, duplicate entries in the second list are dropped
2952 during the append operation.
2954 It is an error for a particular unique flag ID to have multiple behaviors,
2955 except in the case of **Require** (which adds restrictions on another metadata
2956 value) or **Override**.
2958 An example of module flags:
2960 .. code-block:: llvm
2962 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2963 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2964 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2965 !3 = metadata !{ i32 3, metadata !"qux",
2967 metadata !"foo", i32 1
2970 !llvm.module.flags = !{ !0, !1, !2, !3 }
2972 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2973 if two or more ``!"foo"`` flags are seen is to emit an error if their
2974 values are not equal.
2976 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2977 behavior if two or more ``!"bar"`` flags are seen is to use the value
2980 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2981 behavior if two or more ``!"qux"`` flags are seen is to emit a
2982 warning if their values are not equal.
2984 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2988 metadata !{ metadata !"foo", i32 1 }
2990 The behavior is to emit an error if the ``llvm.module.flags`` does not
2991 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2994 Objective-C Garbage Collection Module Flags Metadata
2995 ----------------------------------------------------
2997 On the Mach-O platform, Objective-C stores metadata about garbage
2998 collection in a special section called "image info". The metadata
2999 consists of a version number and a bitmask specifying what types of
3000 garbage collection are supported (if any) by the file. If two or more
3001 modules are linked together their garbage collection metadata needs to
3002 be merged rather than appended together.
3004 The Objective-C garbage collection module flags metadata consists of the
3005 following key-value pairs:
3014 * - ``Objective-C Version``
3015 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3017 * - ``Objective-C Image Info Version``
3018 - **[Required]** --- The version of the image info section. Currently
3021 * - ``Objective-C Image Info Section``
3022 - **[Required]** --- The section to place the metadata. Valid values are
3023 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3024 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3025 Objective-C ABI version 2.
3027 * - ``Objective-C Garbage Collection``
3028 - **[Required]** --- Specifies whether garbage collection is supported or
3029 not. Valid values are 0, for no garbage collection, and 2, for garbage
3030 collection supported.
3032 * - ``Objective-C GC Only``
3033 - **[Optional]** --- Specifies that only garbage collection is supported.
3034 If present, its value must be 6. This flag requires that the
3035 ``Objective-C Garbage Collection`` flag have the value 2.
3037 Some important flag interactions:
3039 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3040 merged with a module with ``Objective-C Garbage Collection`` set to
3041 2, then the resulting module has the
3042 ``Objective-C Garbage Collection`` flag set to 0.
3043 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3044 merged with a module with ``Objective-C GC Only`` set to 6.
3046 Automatic Linker Flags Module Flags Metadata
3047 --------------------------------------------
3049 Some targets support embedding flags to the linker inside individual object
3050 files. Typically this is used in conjunction with language extensions which
3051 allow source files to explicitly declare the libraries they depend on, and have
3052 these automatically be transmitted to the linker via object files.
3054 These flags are encoded in the IR using metadata in the module flags section,
3055 using the ``Linker Options`` key. The merge behavior for this flag is required
3056 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3057 node which should be a list of other metadata nodes, each of which should be a
3058 list of metadata strings defining linker options.
3060 For example, the following metadata section specifies two separate sets of
3061 linker options, presumably to link against ``libz`` and the ``Cocoa``
3064 !0 = metadata !{ i32 6, metadata !"Linker Options",
3066 metadata !{ metadata !"-lz" },
3067 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3068 !llvm.module.flags = !{ !0 }
3070 The metadata encoding as lists of lists of options, as opposed to a collapsed
3071 list of options, is chosen so that the IR encoding can use multiple option
3072 strings to specify e.g., a single library, while still having that specifier be
3073 preserved as an atomic element that can be recognized by a target specific
3074 assembly writer or object file emitter.
3076 Each individual option is required to be either a valid option for the target's
3077 linker, or an option that is reserved by the target specific assembly writer or
3078 object file emitter. No other aspect of these options is defined by the IR.
3080 .. _intrinsicglobalvariables:
3082 Intrinsic Global Variables
3083 ==========================
3085 LLVM has a number of "magic" global variables that contain data that
3086 affect code generation or other IR semantics. These are documented here.
3087 All globals of this sort should have a section specified as
3088 "``llvm.metadata``". This section and all globals that start with
3089 "``llvm.``" are reserved for use by LLVM.
3093 The '``llvm.used``' Global Variable
3094 -----------------------------------
3096 The ``@llvm.used`` global is an array which has
3097 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3098 pointers to named global variables, functions and aliases which may optionally
3099 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3102 .. code-block:: llvm
3107 @llvm.used = appending global [2 x i8*] [
3109 i8* bitcast (i32* @Y to i8*)
3110 ], section "llvm.metadata"
3112 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3113 and linker are required to treat the symbol as if there is a reference to the
3114 symbol that it cannot see (which is why they have to be named). For example, if
3115 a variable has internal linkage and no references other than that from the
3116 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3117 references from inline asms and other things the compiler cannot "see", and
3118 corresponds to "``attribute((used))``" in GNU C.
3120 On some targets, the code generator must emit a directive to the
3121 assembler or object file to prevent the assembler and linker from
3122 molesting the symbol.
3124 .. _gv_llvmcompilerused:
3126 The '``llvm.compiler.used``' Global Variable
3127 --------------------------------------------
3129 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3130 directive, except that it only prevents the compiler from touching the
3131 symbol. On targets that support it, this allows an intelligent linker to
3132 optimize references to the symbol without being impeded as it would be
3135 This is a rare construct that should only be used in rare circumstances,
3136 and should not be exposed to source languages.
3138 .. _gv_llvmglobalctors:
3140 The '``llvm.global_ctors``' Global Variable
3141 -------------------------------------------
3143 .. code-block:: llvm
3145 %0 = type { i32, void ()* }
3146 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3148 The ``@llvm.global_ctors`` array contains a list of constructor
3149 functions and associated priorities. The functions referenced by this
3150 array will be called in ascending order of priority (i.e. lowest first)
3151 when the module is loaded. The order of functions with the same priority
3154 .. _llvmglobaldtors:
3156 The '``llvm.global_dtors``' Global Variable
3157 -------------------------------------------
3159 .. code-block:: llvm
3161 %0 = type { i32, void ()* }
3162 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3164 The ``@llvm.global_dtors`` array contains a list of destructor functions
3165 and associated priorities. The functions referenced by this array will
3166 be called in descending order of priority (i.e. highest first) when the
3167 module is loaded. The order of functions with the same priority is not
3170 Instruction Reference
3171 =====================
3173 The LLVM instruction set consists of several different classifications
3174 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3175 instructions <binaryops>`, :ref:`bitwise binary
3176 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3177 :ref:`other instructions <otherops>`.
3181 Terminator Instructions
3182 -----------------------
3184 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3185 program ends with a "Terminator" instruction, which indicates which
3186 block should be executed after the current block is finished. These
3187 terminator instructions typically yield a '``void``' value: they produce
3188 control flow, not values (the one exception being the
3189 ':ref:`invoke <i_invoke>`' instruction).
3191 The terminator instructions are: ':ref:`ret <i_ret>`',
3192 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3193 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3194 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3198 '``ret``' Instruction
3199 ^^^^^^^^^^^^^^^^^^^^^
3206 ret <type> <value> ; Return a value from a non-void function
3207 ret void ; Return from void function
3212 The '``ret``' instruction is used to return control flow (and optionally
3213 a value) from a function back to the caller.
3215 There are two forms of the '``ret``' instruction: one that returns a
3216 value and then causes control flow, and one that just causes control
3222 The '``ret``' instruction optionally accepts a single argument, the
3223 return value. The type of the return value must be a ':ref:`first
3224 class <t_firstclass>`' type.
3226 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3227 return type and contains a '``ret``' instruction with no return value or
3228 a return value with a type that does not match its type, or if it has a
3229 void return type and contains a '``ret``' instruction with a return
3235 When the '``ret``' instruction is executed, control flow returns back to
3236 the calling function's context. If the caller is a
3237 ":ref:`call <i_call>`" instruction, execution continues at the
3238 instruction after the call. If the caller was an
3239 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3240 beginning of the "normal" destination block. If the instruction returns
3241 a value, that value shall set the call or invoke instruction's return
3247 .. code-block:: llvm
3249 ret i32 5 ; Return an integer value of 5
3250 ret void ; Return from a void function
3251 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3255 '``br``' Instruction
3256 ^^^^^^^^^^^^^^^^^^^^
3263 br i1 <cond>, label <iftrue>, label <iffalse>
3264 br label <dest> ; Unconditional branch
3269 The '``br``' instruction is used to cause control flow to transfer to a
3270 different basic block in the current function. There are two forms of
3271 this instruction, corresponding to a conditional branch and an
3272 unconditional branch.
3277 The conditional branch form of the '``br``' instruction takes a single
3278 '``i1``' value and two '``label``' values. The unconditional form of the
3279 '``br``' instruction takes a single '``label``' value as a target.
3284 Upon execution of a conditional '``br``' instruction, the '``i1``'
3285 argument is evaluated. If the value is ``true``, control flows to the
3286 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3287 to the '``iffalse``' ``label`` argument.
3292 .. code-block:: llvm
3295 %cond = icmp eq i32 %a, %b
3296 br i1 %cond, label %IfEqual, label %IfUnequal
3304 '``switch``' Instruction
3305 ^^^^^^^^^^^^^^^^^^^^^^^^
3312 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3317 The '``switch``' instruction is used to transfer control flow to one of
3318 several different places. It is a generalization of the '``br``'
3319 instruction, allowing a branch to occur to one of many possible
3325 The '``switch``' instruction uses three parameters: an integer
3326 comparison value '``value``', a default '``label``' destination, and an
3327 array of pairs of comparison value constants and '``label``'s. The table
3328 is not allowed to contain duplicate constant entries.
3333 The ``switch`` instruction specifies a table of values and destinations.
3334 When the '``switch``' instruction is executed, this table is searched
3335 for the given value. If the value is found, control flow is transferred
3336 to the corresponding destination; otherwise, control flow is transferred
3337 to the default destination.
3342 Depending on properties of the target machine and the particular
3343 ``switch`` instruction, this instruction may be code generated in
3344 different ways. For example, it could be generated as a series of
3345 chained conditional branches or with a lookup table.
3350 .. code-block:: llvm
3352 ; Emulate a conditional br instruction
3353 %Val = zext i1 %value to i32
3354 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3356 ; Emulate an unconditional br instruction
3357 switch i32 0, label %dest [ ]
3359 ; Implement a jump table:
3360 switch i32 %val, label %otherwise [ i32 0, label %onzero
3362 i32 2, label %ontwo ]
3366 '``indirectbr``' Instruction
3367 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3374 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3379 The '``indirectbr``' instruction implements an indirect branch to a
3380 label within the current function, whose address is specified by
3381 "``address``". Address must be derived from a
3382 :ref:`blockaddress <blockaddress>` constant.
3387 The '``address``' argument is the address of the label to jump to. The
3388 rest of the arguments indicate the full set of possible destinations
3389 that the address may point to. Blocks are allowed to occur multiple
3390 times in the destination list, though this isn't particularly useful.
3392 This destination list is required so that dataflow analysis has an
3393 accurate understanding of the CFG.
3398 Control transfers to the block specified in the address argument. All
3399 possible destination blocks must be listed in the label list, otherwise
3400 this instruction has undefined behavior. This implies that jumps to
3401 labels defined in other functions have undefined behavior as well.
3406 This is typically implemented with a jump through a register.
3411 .. code-block:: llvm
3413 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3417 '``invoke``' Instruction
3418 ^^^^^^^^^^^^^^^^^^^^^^^^
3425 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3426 to label <normal label> unwind label <exception label>
3431 The '``invoke``' instruction causes control to transfer to a specified
3432 function, with the possibility of control flow transfer to either the
3433 '``normal``' label or the '``exception``' label. If the callee function
3434 returns with the "``ret``" instruction, control flow will return to the
3435 "normal" label. If the callee (or any indirect callees) returns via the
3436 ":ref:`resume <i_resume>`" instruction or other exception handling
3437 mechanism, control is interrupted and continued at the dynamically
3438 nearest "exception" label.
3440 The '``exception``' label is a `landing
3441 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3442 '``exception``' label is required to have the
3443 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3444 information about the behavior of the program after unwinding happens,
3445 as its first non-PHI instruction. The restrictions on the
3446 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3447 instruction, so that the important information contained within the
3448 "``landingpad``" instruction can't be lost through normal code motion.
3453 This instruction requires several arguments:
3455 #. The optional "cconv" marker indicates which :ref:`calling
3456 convention <callingconv>` the call should use. If none is
3457 specified, the call defaults to using C calling conventions.
3458 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3459 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3461 #. '``ptr to function ty``': shall be the signature of the pointer to
3462 function value being invoked. In most cases, this is a direct
3463 function invocation, but indirect ``invoke``'s are just as possible,
3464 branching off an arbitrary pointer to function value.
3465 #. '``function ptr val``': An LLVM value containing a pointer to a
3466 function to be invoked.
3467 #. '``function args``': argument list whose types match the function
3468 signature argument types and parameter attributes. All arguments must
3469 be of :ref:`first class <t_firstclass>` type. If the function signature
3470 indicates the function accepts a variable number of arguments, the
3471 extra arguments can be specified.
3472 #. '``normal label``': the label reached when the called function
3473 executes a '``ret``' instruction.
3474 #. '``exception label``': the label reached when a callee returns via
3475 the :ref:`resume <i_resume>` instruction or other exception handling
3477 #. The optional :ref:`function attributes <fnattrs>` list. Only
3478 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3479 attributes are valid here.
3484 This instruction is designed to operate as a standard '``call``'
3485 instruction in most regards. The primary difference is that it
3486 establishes an association with a label, which is used by the runtime
3487 library to unwind the stack.
3489 This instruction is used in languages with destructors to ensure that
3490 proper cleanup is performed in the case of either a ``longjmp`` or a
3491 thrown exception. Additionally, this is important for implementation of
3492 '``catch``' clauses in high-level languages that support them.
3494 For the purposes of the SSA form, the definition of the value returned
3495 by the '``invoke``' instruction is deemed to occur on the edge from the
3496 current block to the "normal" label. If the callee unwinds then no
3497 return value is available.
3502 .. code-block:: llvm
3504 %retval = invoke i32 @Test(i32 15) to label %Continue
3505 unwind label %TestCleanup ; {i32}:retval set
3506 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3507 unwind label %TestCleanup ; {i32}:retval set
3511 '``resume``' Instruction
3512 ^^^^^^^^^^^^^^^^^^^^^^^^
3519 resume <type> <value>
3524 The '``resume``' instruction is a terminator instruction that has no
3530 The '``resume``' instruction requires one argument, which must have the
3531 same type as the result of any '``landingpad``' instruction in the same
3537 The '``resume``' instruction resumes propagation of an existing
3538 (in-flight) exception whose unwinding was interrupted with a
3539 :ref:`landingpad <i_landingpad>` instruction.
3544 .. code-block:: llvm
3546 resume { i8*, i32 } %exn
3550 '``unreachable``' Instruction
3551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3563 The '``unreachable``' instruction has no defined semantics. This
3564 instruction is used to inform the optimizer that a particular portion of
3565 the code is not reachable. This can be used to indicate that the code
3566 after a no-return function cannot be reached, and other facts.
3571 The '``unreachable``' instruction has no defined semantics.
3578 Binary operators are used to do most of the computation in a program.
3579 They require two operands of the same type, execute an operation on
3580 them, and produce a single value. The operands might represent multiple
3581 data, as is the case with the :ref:`vector <t_vector>` data type. The
3582 result value has the same type as its operands.
3584 There are several different binary operators:
3588 '``add``' Instruction
3589 ^^^^^^^^^^^^^^^^^^^^^
3596 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3597 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3598 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3599 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3604 The '``add``' instruction returns the sum of its two operands.
3609 The two arguments to the '``add``' instruction must be
3610 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3611 arguments must have identical types.
3616 The value produced is the integer sum of the two operands.
3618 If the sum has unsigned overflow, the result returned is the
3619 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3622 Because LLVM integers use a two's complement representation, this
3623 instruction is appropriate for both signed and unsigned integers.
3625 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3626 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3627 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3628 unsigned and/or signed overflow, respectively, occurs.
3633 .. code-block:: llvm
3635 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3639 '``fadd``' Instruction
3640 ^^^^^^^^^^^^^^^^^^^^^^
3647 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3652 The '``fadd``' instruction returns the sum of its two operands.
3657 The two arguments to the '``fadd``' instruction must be :ref:`floating
3658 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3659 Both arguments must have identical types.
3664 The value produced is the floating point sum of the two operands. This
3665 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3666 which are optimization hints to enable otherwise unsafe floating point
3672 .. code-block:: llvm
3674 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3676 '``sub``' Instruction
3677 ^^^^^^^^^^^^^^^^^^^^^
3684 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3685 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3686 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3687 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3692 The '``sub``' instruction returns the difference of its two operands.
3694 Note that the '``sub``' instruction is used to represent the '``neg``'
3695 instruction present in most other intermediate representations.
3700 The two arguments to the '``sub``' instruction must be
3701 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3702 arguments must have identical types.
3707 The value produced is the integer difference of the two operands.
3709 If the difference has unsigned overflow, the result returned is the
3710 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3713 Because LLVM integers use a two's complement representation, this
3714 instruction is appropriate for both signed and unsigned integers.
3716 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3717 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3718 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3719 unsigned and/or signed overflow, respectively, occurs.
3724 .. code-block:: llvm
3726 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3727 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3731 '``fsub``' Instruction
3732 ^^^^^^^^^^^^^^^^^^^^^^
3739 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3744 The '``fsub``' instruction returns the difference of its two operands.
3746 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3747 instruction present in most other intermediate representations.
3752 The two arguments to the '``fsub``' instruction must be :ref:`floating
3753 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3754 Both arguments must have identical types.
3759 The value produced is the floating point difference of the two operands.
3760 This instruction can also take any number of :ref:`fast-math
3761 flags <fastmath>`, which are optimization hints to enable otherwise
3762 unsafe floating point optimizations:
3767 .. code-block:: llvm
3769 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3770 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3772 '``mul``' Instruction
3773 ^^^^^^^^^^^^^^^^^^^^^
3780 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3781 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3782 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3783 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3788 The '``mul``' instruction returns the product of its two operands.
3793 The two arguments to the '``mul``' instruction must be
3794 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3795 arguments must have identical types.
3800 The value produced is the integer product of the two operands.
3802 If the result of the multiplication has unsigned overflow, the result
3803 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3804 bit width of the result.
3806 Because LLVM integers use a two's complement representation, and the
3807 result is the same width as the operands, this instruction returns the
3808 correct result for both signed and unsigned integers. If a full product
3809 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3810 sign-extended or zero-extended as appropriate to the width of the full
3813 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3814 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3815 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3816 unsigned and/or signed overflow, respectively, occurs.
3821 .. code-block:: llvm
3823 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3827 '``fmul``' Instruction
3828 ^^^^^^^^^^^^^^^^^^^^^^
3835 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3840 The '``fmul``' instruction returns the product of its two operands.
3845 The two arguments to the '``fmul``' instruction must be :ref:`floating
3846 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3847 Both arguments must have identical types.
3852 The value produced is the floating point product of the two operands.
3853 This instruction can also take any number of :ref:`fast-math
3854 flags <fastmath>`, which are optimization hints to enable otherwise
3855 unsafe floating point optimizations:
3860 .. code-block:: llvm
3862 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3864 '``udiv``' Instruction
3865 ^^^^^^^^^^^^^^^^^^^^^^
3872 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3873 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3878 The '``udiv``' instruction returns the quotient of its two operands.
3883 The two arguments to the '``udiv``' instruction must be
3884 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3885 arguments must have identical types.
3890 The value produced is the unsigned integer quotient of the two operands.
3892 Note that unsigned integer division and signed integer division are
3893 distinct operations; for signed integer division, use '``sdiv``'.
3895 Division by zero leads to undefined behavior.
3897 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3898 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3899 such, "((a udiv exact b) mul b) == a").
3904 .. code-block:: llvm
3906 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3908 '``sdiv``' Instruction
3909 ^^^^^^^^^^^^^^^^^^^^^^
3916 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3917 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3922 The '``sdiv``' instruction returns the quotient of its two operands.
3927 The two arguments to the '``sdiv``' instruction must be
3928 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3929 arguments must have identical types.
3934 The value produced is the signed integer quotient of the two operands
3935 rounded towards zero.
3937 Note that signed integer division and unsigned integer division are
3938 distinct operations; for unsigned integer division, use '``udiv``'.
3940 Division by zero leads to undefined behavior. Overflow also leads to
3941 undefined behavior; this is a rare case, but can occur, for example, by
3942 doing a 32-bit division of -2147483648 by -1.
3944 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3945 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3950 .. code-block:: llvm
3952 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3956 '``fdiv``' Instruction
3957 ^^^^^^^^^^^^^^^^^^^^^^
3964 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3969 The '``fdiv``' instruction returns the quotient of its two operands.
3974 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3975 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3976 Both arguments must have identical types.
3981 The value produced is the floating point quotient of the two operands.
3982 This instruction can also take any number of :ref:`fast-math
3983 flags <fastmath>`, which are optimization hints to enable otherwise
3984 unsafe floating point optimizations:
3989 .. code-block:: llvm
3991 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3993 '``urem``' Instruction
3994 ^^^^^^^^^^^^^^^^^^^^^^
4001 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
4006 The '``urem``' instruction returns the remainder from the unsigned
4007 division of its two arguments.
4012 The two arguments to the '``urem``' instruction must be
4013 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4014 arguments must have identical types.
4019 This instruction returns the unsigned integer *remainder* of a division.
4020 This instruction always performs an unsigned division to get the
4023 Note that unsigned integer remainder and signed integer remainder are
4024 distinct operations; for signed integer remainder, use '``srem``'.
4026 Taking the remainder of a division by zero leads to undefined behavior.
4031 .. code-block:: llvm
4033 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
4035 '``srem``' Instruction
4036 ^^^^^^^^^^^^^^^^^^^^^^
4043 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
4048 The '``srem``' instruction returns the remainder from the signed
4049 division of its two operands. This instruction can also take
4050 :ref:`vector <t_vector>` versions of the values in which case the elements
4056 The two arguments to the '``srem``' instruction must be
4057 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4058 arguments must have identical types.
4063 This instruction returns the *remainder* of a division (where the result
4064 is either zero or has the same sign as the dividend, ``op1``), not the
4065 *modulo* operator (where the result is either zero or has the same sign
4066 as the divisor, ``op2``) of a value. For more information about the
4067 difference, see `The Math
4068 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4069 table of how this is implemented in various languages, please see
4071 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4073 Note that signed integer remainder and unsigned integer remainder are
4074 distinct operations; for unsigned integer remainder, use '``urem``'.
4076 Taking the remainder of a division by zero leads to undefined behavior.
4077 Overflow also leads to undefined behavior; this is a rare case, but can
4078 occur, for example, by taking the remainder of a 32-bit division of
4079 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4080 rule lets srem be implemented using instructions that return both the
4081 result of the division and the remainder.)
4086 .. code-block:: llvm
4088 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
4092 '``frem``' Instruction
4093 ^^^^^^^^^^^^^^^^^^^^^^
4100 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
4105 The '``frem``' instruction returns the remainder from the division of
4111 The two arguments to the '``frem``' instruction must be :ref:`floating
4112 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4113 Both arguments must have identical types.
4118 This instruction returns the *remainder* of a division. The remainder
4119 has the same sign as the dividend. This instruction can also take any
4120 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4121 to enable otherwise unsafe floating point optimizations:
4126 .. code-block:: llvm
4128 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4132 Bitwise Binary Operations
4133 -------------------------
4135 Bitwise binary operators are used to do various forms of bit-twiddling
4136 in a program. They are generally very efficient instructions and can
4137 commonly be strength reduced from other instructions. They require two
4138 operands of the same type, execute an operation on them, and produce a
4139 single value. The resulting value is the same type as its operands.
4141 '``shl``' Instruction
4142 ^^^^^^^^^^^^^^^^^^^^^
4149 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4150 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4151 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4152 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4157 The '``shl``' instruction returns the first operand shifted to the left
4158 a specified number of bits.
4163 Both arguments to the '``shl``' instruction must be the same
4164 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4165 '``op2``' is treated as an unsigned value.
4170 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4171 where ``n`` is the width of the result. If ``op2`` is (statically or
4172 dynamically) negative or equal to or larger than the number of bits in
4173 ``op1``, the result is undefined. If the arguments are vectors, each
4174 vector element of ``op1`` is shifted by the corresponding shift amount
4177 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4178 value <poisonvalues>` if it shifts out any non-zero bits. If the
4179 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4180 value <poisonvalues>` if it shifts out any bits that disagree with the
4181 resultant sign bit. As such, NUW/NSW have the same semantics as they
4182 would if the shift were expressed as a mul instruction with the same
4183 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4188 .. code-block:: llvm
4190 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4191 <result> = shl i32 4, 2 ; yields {i32}: 16
4192 <result> = shl i32 1, 10 ; yields {i32}: 1024
4193 <result> = shl i32 1, 32 ; undefined
4194 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4196 '``lshr``' Instruction
4197 ^^^^^^^^^^^^^^^^^^^^^^
4204 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4205 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4210 The '``lshr``' instruction (logical shift right) returns the first
4211 operand shifted to the right a specified number of bits with zero fill.
4216 Both arguments to the '``lshr``' instruction must be the same
4217 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4218 '``op2``' is treated as an unsigned value.
4223 This instruction always performs a logical shift right operation. The
4224 most significant bits of the result will be filled with zero bits after
4225 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4226 than the number of bits in ``op1``, the result is undefined. If the
4227 arguments are vectors, each vector element of ``op1`` is shifted by the
4228 corresponding shift amount in ``op2``.
4230 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4231 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4237 .. code-block:: llvm
4239 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4240 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4241 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4242 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4243 <result> = lshr i32 1, 32 ; undefined
4244 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4246 '``ashr``' Instruction
4247 ^^^^^^^^^^^^^^^^^^^^^^
4254 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4255 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4260 The '``ashr``' instruction (arithmetic shift right) returns the first
4261 operand shifted to the right a specified number of bits with sign
4267 Both arguments to the '``ashr``' instruction must be the same
4268 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4269 '``op2``' is treated as an unsigned value.
4274 This instruction always performs an arithmetic shift right operation,
4275 The most significant bits of the result will be filled with the sign bit
4276 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4277 than the number of bits in ``op1``, the result is undefined. If the
4278 arguments are vectors, each vector element of ``op1`` is shifted by the
4279 corresponding shift amount in ``op2``.
4281 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4282 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4288 .. code-block:: llvm
4290 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4291 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4292 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4293 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4294 <result> = ashr i32 1, 32 ; undefined
4295 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4297 '``and``' Instruction
4298 ^^^^^^^^^^^^^^^^^^^^^
4305 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4310 The '``and``' instruction returns the bitwise logical and of its two
4316 The two arguments to the '``and``' instruction must be
4317 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4318 arguments must have identical types.
4323 The truth table used for the '``and``' instruction is:
4340 .. code-block:: llvm
4342 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4343 <result> = and i32 15, 40 ; yields {i32}:result = 8
4344 <result> = and i32 4, 8 ; yields {i32}:result = 0
4346 '``or``' Instruction
4347 ^^^^^^^^^^^^^^^^^^^^
4354 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4359 The '``or``' instruction returns the bitwise logical inclusive or of its
4365 The two arguments to the '``or``' instruction must be
4366 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4367 arguments must have identical types.
4372 The truth table used for the '``or``' instruction is:
4391 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4392 <result> = or i32 15, 40 ; yields {i32}:result = 47
4393 <result> = or i32 4, 8 ; yields {i32}:result = 12
4395 '``xor``' Instruction
4396 ^^^^^^^^^^^^^^^^^^^^^
4403 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4408 The '``xor``' instruction returns the bitwise logical exclusive or of
4409 its two operands. The ``xor`` is used to implement the "one's
4410 complement" operation, which is the "~" operator in C.
4415 The two arguments to the '``xor``' instruction must be
4416 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4417 arguments must have identical types.
4422 The truth table used for the '``xor``' instruction is:
4439 .. code-block:: llvm
4441 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4442 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4443 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4444 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4449 LLVM supports several instructions to represent vector operations in a
4450 target-independent manner. These instructions cover the element-access
4451 and vector-specific operations needed to process vectors effectively.
4452 While LLVM does directly support these vector operations, many
4453 sophisticated algorithms will want to use target-specific intrinsics to
4454 take full advantage of a specific target.
4456 .. _i_extractelement:
4458 '``extractelement``' Instruction
4459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4466 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4471 The '``extractelement``' instruction extracts a single scalar element
4472 from a vector at a specified index.
4477 The first operand of an '``extractelement``' instruction is a value of
4478 :ref:`vector <t_vector>` type. The second operand is an index indicating
4479 the position from which to extract the element. The index may be a
4485 The result is a scalar of the same type as the element type of ``val``.
4486 Its value is the value at position ``idx`` of ``val``. If ``idx``
4487 exceeds the length of ``val``, the results are undefined.
4492 .. code-block:: llvm
4494 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4496 .. _i_insertelement:
4498 '``insertelement``' Instruction
4499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4506 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4511 The '``insertelement``' instruction inserts a scalar element into a
4512 vector at a specified index.
4517 The first operand of an '``insertelement``' instruction is a value of
4518 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4519 type must equal the element type of the first operand. The third operand
4520 is an index indicating the position at which to insert the value. The
4521 index may be a variable.
4526 The result is a vector of the same type as ``val``. Its element values
4527 are those of ``val`` except at position ``idx``, where it gets the value
4528 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4534 .. code-block:: llvm
4536 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4538 .. _i_shufflevector:
4540 '``shufflevector``' Instruction
4541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4548 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4553 The '``shufflevector``' instruction constructs a permutation of elements
4554 from two input vectors, returning a vector with the same element type as
4555 the input and length that is the same as the shuffle mask.
4560 The first two operands of a '``shufflevector``' instruction are vectors
4561 with the same type. The third argument is a shuffle mask whose element
4562 type is always 'i32'. The result of the instruction is a vector whose
4563 length is the same as the shuffle mask and whose element type is the
4564 same as the element type of the first two operands.
4566 The shuffle mask operand is required to be a constant vector with either
4567 constant integer or undef values.
4572 The elements of the two input vectors are numbered from left to right
4573 across both of the vectors. The shuffle mask operand specifies, for each
4574 element of the result vector, which element of the two input vectors the
4575 result element gets. The element selector may be undef (meaning "don't
4576 care") and the second operand may be undef if performing a shuffle from
4582 .. code-block:: llvm
4584 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4585 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4586 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4587 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4588 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4589 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4590 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4591 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4593 Aggregate Operations
4594 --------------------
4596 LLVM supports several instructions for working with
4597 :ref:`aggregate <t_aggregate>` values.
4601 '``extractvalue``' Instruction
4602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4609 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4614 The '``extractvalue``' instruction extracts the value of a member field
4615 from an :ref:`aggregate <t_aggregate>` value.
4620 The first operand of an '``extractvalue``' instruction is a value of
4621 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4622 constant indices to specify which value to extract in a similar manner
4623 as indices in a '``getelementptr``' instruction.
4625 The major differences to ``getelementptr`` indexing are:
4627 - Since the value being indexed is not a pointer, the first index is
4628 omitted and assumed to be zero.
4629 - At least one index must be specified.
4630 - Not only struct indices but also array indices must be in bounds.
4635 The result is the value at the position in the aggregate specified by
4641 .. code-block:: llvm
4643 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4647 '``insertvalue``' Instruction
4648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4655 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4660 The '``insertvalue``' instruction inserts a value into a member field in
4661 an :ref:`aggregate <t_aggregate>` value.
4666 The first operand of an '``insertvalue``' instruction is a value of
4667 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4668 a first-class value to insert. The following operands are constant
4669 indices indicating the position at which to insert the value in a
4670 similar manner as indices in a '``extractvalue``' instruction. The value
4671 to insert must have the same type as the value identified by the
4677 The result is an aggregate of the same type as ``val``. Its value is
4678 that of ``val`` except that the value at the position specified by the
4679 indices is that of ``elt``.
4684 .. code-block:: llvm
4686 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4687 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4688 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4692 Memory Access and Addressing Operations
4693 ---------------------------------------
4695 A key design point of an SSA-based representation is how it represents
4696 memory. In LLVM, no memory locations are in SSA form, which makes things
4697 very simple. This section describes how to read, write, and allocate
4702 '``alloca``' Instruction
4703 ^^^^^^^^^^^^^^^^^^^^^^^^
4710 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields {type*}:result
4715 The '``alloca``' instruction allocates memory on the stack frame of the
4716 currently executing function, to be automatically released when this
4717 function returns to its caller. The object is always allocated in the
4718 generic address space (address space zero).
4723 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4724 bytes of memory on the runtime stack, returning a pointer of the
4725 appropriate type to the program. If "NumElements" is specified, it is
4726 the number of elements allocated, otherwise "NumElements" is defaulted
4727 to be one. If a constant alignment is specified, the value result of the
4728 allocation is guaranteed to be aligned to at least that boundary. If not
4729 specified, or if zero, the target can choose to align the allocation on
4730 any convenient boundary compatible with the type.
4732 '``type``' may be any sized type.
4737 Memory is allocated; a pointer is returned. The operation is undefined
4738 if there is insufficient stack space for the allocation. '``alloca``'d
4739 memory is automatically released when the function returns. The
4740 '``alloca``' instruction is commonly used to represent automatic
4741 variables that must have an address available. When the function returns
4742 (either with the ``ret`` or ``resume`` instructions), the memory is
4743 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4744 The order in which memory is allocated (ie., which way the stack grows)
4750 .. code-block:: llvm
4752 %ptr = alloca i32 ; yields {i32*}:ptr
4753 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4754 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4755 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4759 '``load``' Instruction
4760 ^^^^^^^^^^^^^^^^^^^^^^
4767 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4768 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4769 !<index> = !{ i32 1 }
4774 The '``load``' instruction is used to read from memory.
4779 The argument to the ``load`` instruction specifies the memory address
4780 from which to load. The pointer must point to a :ref:`first
4781 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4782 then the optimizer is not allowed to modify the number or order of
4783 execution of this ``load`` with other :ref:`volatile
4784 operations <volatile>`.
4786 If the ``load`` is marked as ``atomic``, it takes an extra
4787 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4788 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4789 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4790 when they may see multiple atomic stores. The type of the pointee must
4791 be an integer type whose bit width is a power of two greater than or
4792 equal to eight and less than or equal to a target-specific size limit.
4793 ``align`` must be explicitly specified on atomic loads, and the load has
4794 undefined behavior if the alignment is not set to a value which is at
4795 least the size in bytes of the pointee. ``!nontemporal`` does not have
4796 any defined semantics for atomic loads.
4798 The optional constant ``align`` argument specifies the alignment of the
4799 operation (that is, the alignment of the memory address). A value of 0
4800 or an omitted ``align`` argument means that the operation has the ABI
4801 alignment for the target. It is the responsibility of the code emitter
4802 to ensure that the alignment information is correct. Overestimating the
4803 alignment results in undefined behavior. Underestimating the alignment
4804 may produce less efficient code. An alignment of 1 is always safe.
4806 The optional ``!nontemporal`` metadata must reference a single
4807 metadata name ``<index>`` corresponding to a metadata node with one
4808 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4809 metadata on the instruction tells the optimizer and code generator
4810 that this load is not expected to be reused in the cache. The code
4811 generator may select special instructions to save cache bandwidth, such
4812 as the ``MOVNT`` instruction on x86.
4814 The optional ``!invariant.load`` metadata must reference a single
4815 metadata name ``<index>`` corresponding to a metadata node with no
4816 entries. The existence of the ``!invariant.load`` metadata on the
4817 instruction tells the optimizer and code generator that this load
4818 address points to memory which does not change value during program
4819 execution. The optimizer may then move this load around, for example, by
4820 hoisting it out of loops using loop invariant code motion.
4825 The location of memory pointed to is loaded. If the value being loaded
4826 is of scalar type then the number of bytes read does not exceed the
4827 minimum number of bytes needed to hold all bits of the type. For
4828 example, loading an ``i24`` reads at most three bytes. When loading a
4829 value of a type like ``i20`` with a size that is not an integral number
4830 of bytes, the result is undefined if the value was not originally
4831 written using a store of the same type.
4836 .. code-block:: llvm
4838 %ptr = alloca i32 ; yields {i32*}:ptr
4839 store i32 3, i32* %ptr ; yields {void}
4840 %val = load i32* %ptr ; yields {i32}:val = i32 3
4844 '``store``' Instruction
4845 ^^^^^^^^^^^^^^^^^^^^^^^
4852 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4853 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4858 The '``store``' instruction is used to write to memory.
4863 There are two arguments to the ``store`` instruction: a value to store
4864 and an address at which to store it. The type of the ``<pointer>``
4865 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4866 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4867 then the optimizer is not allowed to modify the number or order of
4868 execution of this ``store`` with other :ref:`volatile
4869 operations <volatile>`.
4871 If the ``store`` is marked as ``atomic``, it takes an extra
4872 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4873 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4874 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4875 when they may see multiple atomic stores. The type of the pointee must
4876 be an integer type whose bit width is a power of two greater than or
4877 equal to eight and less than or equal to a target-specific size limit.
4878 ``align`` must be explicitly specified on atomic stores, and the store
4879 has undefined behavior if the alignment is not set to a value which is
4880 at least the size in bytes of the pointee. ``!nontemporal`` does not
4881 have any defined semantics for atomic stores.
4883 The optional constant ``align`` argument specifies the alignment of the
4884 operation (that is, the alignment of the memory address). A value of 0
4885 or an omitted ``align`` argument means that the operation has the ABI
4886 alignment for the target. It is the responsibility of the code emitter
4887 to ensure that the alignment information is correct. Overestimating the
4888 alignment results in undefined behavior. Underestimating the
4889 alignment may produce less efficient code. An alignment of 1 is always
4892 The optional ``!nontemporal`` metadata must reference a single metadata
4893 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4894 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4895 tells the optimizer and code generator that this load is not expected to
4896 be reused in the cache. The code generator may select special
4897 instructions to save cache bandwidth, such as the MOVNT instruction on
4903 The contents of memory are updated to contain ``<value>`` at the
4904 location specified by the ``<pointer>`` operand. If ``<value>`` is
4905 of scalar type then the number of bytes written does not exceed the
4906 minimum number of bytes needed to hold all bits of the type. For
4907 example, storing an ``i24`` writes at most three bytes. When writing a
4908 value of a type like ``i20`` with a size that is not an integral number
4909 of bytes, it is unspecified what happens to the extra bits that do not
4910 belong to the type, but they will typically be overwritten.
4915 .. code-block:: llvm
4917 %ptr = alloca i32 ; yields {i32*}:ptr
4918 store i32 3, i32* %ptr ; yields {void}
4919 %val = load i32* %ptr ; yields {i32}:val = i32 3
4923 '``fence``' Instruction
4924 ^^^^^^^^^^^^^^^^^^^^^^^
4931 fence [singlethread] <ordering> ; yields {void}
4936 The '``fence``' instruction is used to introduce happens-before edges
4942 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4943 defines what *synchronizes-with* edges they add. They can only be given
4944 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4949 A fence A which has (at least) ``release`` ordering semantics
4950 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4951 semantics if and only if there exist atomic operations X and Y, both
4952 operating on some atomic object M, such that A is sequenced before X, X
4953 modifies M (either directly or through some side effect of a sequence
4954 headed by X), Y is sequenced before B, and Y observes M. This provides a
4955 *happens-before* dependency between A and B. Rather than an explicit
4956 ``fence``, one (but not both) of the atomic operations X or Y might
4957 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4958 still *synchronize-with* the explicit ``fence`` and establish the
4959 *happens-before* edge.
4961 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4962 ``acquire`` and ``release`` semantics specified above, participates in
4963 the global program order of other ``seq_cst`` operations and/or fences.
4965 The optional ":ref:`singlethread <singlethread>`" argument specifies
4966 that the fence only synchronizes with other fences in the same thread.
4967 (This is useful for interacting with signal handlers.)
4972 .. code-block:: llvm
4974 fence acquire ; yields {void}
4975 fence singlethread seq_cst ; yields {void}
4979 '``cmpxchg``' Instruction
4980 ^^^^^^^^^^^^^^^^^^^^^^^^^
4987 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields {ty}
4992 The '``cmpxchg``' instruction is used to atomically modify memory. It
4993 loads a value in memory and compares it to a given value. If they are
4994 equal, it stores a new value into the memory.
4999 There are three arguments to the '``cmpxchg``' instruction: an address
5000 to operate on, a value to compare to the value currently be at that
5001 address, and a new value to place at that address if the compared values
5002 are equal. The type of '<cmp>' must be an integer type whose bit width
5003 is a power of two greater than or equal to eight and less than or equal
5004 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5005 type, and the type of '<pointer>' must be a pointer to that type. If the
5006 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5007 to modify the number or order of execution of this ``cmpxchg`` with
5008 other :ref:`volatile operations <volatile>`.
5010 The success and failure :ref:`ordering <ordering>` arguments specify how this
5011 ``cmpxchg`` synchronizes with other atomic operations. The both ordering
5012 parameters must be at least ``monotonic``, the ordering constraint on failure
5013 must be no stronger than that on success, and the failure ordering cannot be
5014 either ``release`` or ``acq_rel``.
5016 The optional "``singlethread``" argument declares that the ``cmpxchg``
5017 is only atomic with respect to code (usually signal handlers) running in
5018 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5019 respect to all other code in the system.
5021 The pointer passed into cmpxchg must have alignment greater than or
5022 equal to the size in memory of the operand.
5027 The contents of memory at the location specified by the '``<pointer>``'
5028 operand is read and compared to '``<cmp>``'; if the read value is the
5029 equal, '``<new>``' is written. The original value at the location is
5032 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5033 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5034 load with an ordering parameter determined the second ordering parameter.
5039 .. code-block:: llvm
5042 %orig = atomic load i32* %ptr unordered ; yields {i32}
5046 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5047 %squared = mul i32 %cmp, %cmp
5048 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields {i32}
5049 %success = icmp eq i32 %cmp, %old
5050 br i1 %success, label %done, label %loop
5057 '``atomicrmw``' Instruction
5058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5065 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
5070 The '``atomicrmw``' instruction is used to atomically modify memory.
5075 There are three arguments to the '``atomicrmw``' instruction: an
5076 operation to apply, an address whose value to modify, an argument to the
5077 operation. The operation must be one of the following keywords:
5091 The type of '<value>' must be an integer type whose bit width is a power
5092 of two greater than or equal to eight and less than or equal to a
5093 target-specific size limit. The type of the '``<pointer>``' operand must
5094 be a pointer to that type. If the ``atomicrmw`` is marked as
5095 ``volatile``, then the optimizer is not allowed to modify the number or
5096 order of execution of this ``atomicrmw`` with other :ref:`volatile
5097 operations <volatile>`.
5102 The contents of memory at the location specified by the '``<pointer>``'
5103 operand are atomically read, modified, and written back. The original
5104 value at the location is returned. The modification is specified by the
5107 - xchg: ``*ptr = val``
5108 - add: ``*ptr = *ptr + val``
5109 - sub: ``*ptr = *ptr - val``
5110 - and: ``*ptr = *ptr & val``
5111 - nand: ``*ptr = ~(*ptr & val)``
5112 - or: ``*ptr = *ptr | val``
5113 - xor: ``*ptr = *ptr ^ val``
5114 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5115 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5116 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5118 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5124 .. code-block:: llvm
5126 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
5128 .. _i_getelementptr:
5130 '``getelementptr``' Instruction
5131 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5138 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5139 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5140 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5145 The '``getelementptr``' instruction is used to get the address of a
5146 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5147 address calculation only and does not access memory.
5152 The first argument is always a pointer or a vector of pointers, and
5153 forms the basis of the calculation. The remaining arguments are indices
5154 that indicate which of the elements of the aggregate object are indexed.
5155 The interpretation of each index is dependent on the type being indexed
5156 into. The first index always indexes the pointer value given as the
5157 first argument, the second index indexes a value of the type pointed to
5158 (not necessarily the value directly pointed to, since the first index
5159 can be non-zero), etc. The first type indexed into must be a pointer
5160 value, subsequent types can be arrays, vectors, and structs. Note that
5161 subsequent types being indexed into can never be pointers, since that
5162 would require loading the pointer before continuing calculation.
5164 The type of each index argument depends on the type it is indexing into.
5165 When indexing into a (optionally packed) structure, only ``i32`` integer
5166 **constants** are allowed (when using a vector of indices they must all
5167 be the **same** ``i32`` integer constant). When indexing into an array,
5168 pointer or vector, integers of any width are allowed, and they are not
5169 required to be constant. These integers are treated as signed values
5172 For example, let's consider a C code fragment and how it gets compiled
5188 int *foo(struct ST *s) {
5189 return &s[1].Z.B[5][13];
5192 The LLVM code generated by Clang is:
5194 .. code-block:: llvm
5196 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5197 %struct.ST = type { i32, double, %struct.RT }
5199 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5201 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5208 In the example above, the first index is indexing into the
5209 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5210 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5211 indexes into the third element of the structure, yielding a
5212 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5213 structure. The third index indexes into the second element of the
5214 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5215 dimensions of the array are subscripted into, yielding an '``i32``'
5216 type. The '``getelementptr``' instruction returns a pointer to this
5217 element, thus computing a value of '``i32*``' type.
5219 Note that it is perfectly legal to index partially through a structure,
5220 returning a pointer to an inner element. Because of this, the LLVM code
5221 for the given testcase is equivalent to:
5223 .. code-block:: llvm
5225 define i32* @foo(%struct.ST* %s) {
5226 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5227 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5228 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5229 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5230 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5234 If the ``inbounds`` keyword is present, the result value of the
5235 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5236 pointer is not an *in bounds* address of an allocated object, or if any
5237 of the addresses that would be formed by successive addition of the
5238 offsets implied by the indices to the base address with infinitely
5239 precise signed arithmetic are not an *in bounds* address of that
5240 allocated object. The *in bounds* addresses for an allocated object are
5241 all the addresses that point into the object, plus the address one byte
5242 past the end. In cases where the base is a vector of pointers the
5243 ``inbounds`` keyword applies to each of the computations element-wise.
5245 If the ``inbounds`` keyword is not present, the offsets are added to the
5246 base address with silently-wrapping two's complement arithmetic. If the
5247 offsets have a different width from the pointer, they are sign-extended
5248 or truncated to the width of the pointer. The result value of the
5249 ``getelementptr`` may be outside the object pointed to by the base
5250 pointer. The result value may not necessarily be used to access memory
5251 though, even if it happens to point into allocated storage. See the
5252 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5255 The getelementptr instruction is often confusing. For some more insight
5256 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5261 .. code-block:: llvm
5263 ; yields [12 x i8]*:aptr
5264 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5266 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5268 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5270 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5272 In cases where the pointer argument is a vector of pointers, each index
5273 must be a vector with the same number of elements. For example:
5275 .. code-block:: llvm
5277 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5279 Conversion Operations
5280 ---------------------
5282 The instructions in this category are the conversion instructions
5283 (casting) which all take a single operand and a type. They perform
5284 various bit conversions on the operand.
5286 '``trunc .. to``' Instruction
5287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5294 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5299 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5304 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5305 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5306 of the same number of integers. The bit size of the ``value`` must be
5307 larger than the bit size of the destination type, ``ty2``. Equal sized
5308 types are not allowed.
5313 The '``trunc``' instruction truncates the high order bits in ``value``
5314 and converts the remaining bits to ``ty2``. Since the source size must
5315 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5316 It will always truncate bits.
5321 .. code-block:: llvm
5323 %X = trunc i32 257 to i8 ; yields i8:1
5324 %Y = trunc i32 123 to i1 ; yields i1:true
5325 %Z = trunc i32 122 to i1 ; yields i1:false
5326 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5328 '``zext .. to``' Instruction
5329 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5336 <result> = zext <ty> <value> to <ty2> ; yields ty2
5341 The '``zext``' instruction zero extends its operand to type ``ty2``.
5346 The '``zext``' instruction takes a value to cast, and a type to cast it
5347 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5348 the same number of integers. The bit size of the ``value`` must be
5349 smaller than the bit size of the destination type, ``ty2``.
5354 The ``zext`` fills the high order bits of the ``value`` with zero bits
5355 until it reaches the size of the destination type, ``ty2``.
5357 When zero extending from i1, the result will always be either 0 or 1.
5362 .. code-block:: llvm
5364 %X = zext i32 257 to i64 ; yields i64:257
5365 %Y = zext i1 true to i32 ; yields i32:1
5366 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5368 '``sext .. to``' Instruction
5369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5376 <result> = sext <ty> <value> to <ty2> ; yields ty2
5381 The '``sext``' sign extends ``value`` to the type ``ty2``.
5386 The '``sext``' instruction takes a value to cast, and a type to cast it
5387 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5388 the same number of integers. The bit size of the ``value`` must be
5389 smaller than the bit size of the destination type, ``ty2``.
5394 The '``sext``' instruction performs a sign extension by copying the sign
5395 bit (highest order bit) of the ``value`` until it reaches the bit size
5396 of the type ``ty2``.
5398 When sign extending from i1, the extension always results in -1 or 0.
5403 .. code-block:: llvm
5405 %X = sext i8 -1 to i16 ; yields i16 :65535
5406 %Y = sext i1 true to i32 ; yields i32:-1
5407 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5409 '``fptrunc .. to``' Instruction
5410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5417 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5422 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5427 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5428 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5429 The size of ``value`` must be larger than the size of ``ty2``. This
5430 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5435 The '``fptrunc``' instruction truncates a ``value`` from a larger
5436 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5437 point <t_floating>` type. If the value cannot fit within the
5438 destination type, ``ty2``, then the results are undefined.
5443 .. code-block:: llvm
5445 %X = fptrunc double 123.0 to float ; yields float:123.0
5446 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5448 '``fpext .. to``' Instruction
5449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5456 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5461 The '``fpext``' extends a floating point ``value`` to a larger floating
5467 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5468 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5469 to. The source type must be smaller than the destination type.
5474 The '``fpext``' instruction extends the ``value`` from a smaller
5475 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5476 point <t_floating>` type. The ``fpext`` cannot be used to make a
5477 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5478 *no-op cast* for a floating point cast.
5483 .. code-block:: llvm
5485 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5486 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5488 '``fptoui .. to``' Instruction
5489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5496 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5501 The '``fptoui``' converts a floating point ``value`` to its unsigned
5502 integer equivalent of type ``ty2``.
5507 The '``fptoui``' instruction takes a value to cast, which must be a
5508 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5509 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5510 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5511 type with the same number of elements as ``ty``
5516 The '``fptoui``' instruction converts its :ref:`floating
5517 point <t_floating>` operand into the nearest (rounding towards zero)
5518 unsigned integer value. If the value cannot fit in ``ty2``, the results
5524 .. code-block:: llvm
5526 %X = fptoui double 123.0 to i32 ; yields i32:123
5527 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5528 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5530 '``fptosi .. to``' Instruction
5531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5538 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5543 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5544 ``value`` to type ``ty2``.
5549 The '``fptosi``' instruction takes a value to cast, which must be a
5550 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5551 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5552 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5553 type with the same number of elements as ``ty``
5558 The '``fptosi``' instruction converts its :ref:`floating
5559 point <t_floating>` operand into the nearest (rounding towards zero)
5560 signed integer value. If the value cannot fit in ``ty2``, the results
5566 .. code-block:: llvm
5568 %X = fptosi double -123.0 to i32 ; yields i32:-123
5569 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5570 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5572 '``uitofp .. to``' Instruction
5573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5580 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5585 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5586 and converts that value to the ``ty2`` type.
5591 The '``uitofp``' instruction takes a value to cast, which must be a
5592 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5593 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5594 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5595 type with the same number of elements as ``ty``
5600 The '``uitofp``' instruction interprets its operand as an unsigned
5601 integer quantity and converts it to the corresponding floating point
5602 value. If the value cannot fit in the floating point value, the results
5608 .. code-block:: llvm
5610 %X = uitofp i32 257 to float ; yields float:257.0
5611 %Y = uitofp i8 -1 to double ; yields double:255.0
5613 '``sitofp .. to``' Instruction
5614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5621 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5626 The '``sitofp``' instruction regards ``value`` as a signed integer and
5627 converts that value to the ``ty2`` type.
5632 The '``sitofp``' instruction takes a value to cast, which must be a
5633 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5634 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5635 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5636 type with the same number of elements as ``ty``
5641 The '``sitofp``' instruction interprets its operand as a signed integer
5642 quantity and converts it to the corresponding floating point value. If
5643 the value cannot fit in the floating point value, the results are
5649 .. code-block:: llvm
5651 %X = sitofp i32 257 to float ; yields float:257.0
5652 %Y = sitofp i8 -1 to double ; yields double:-1.0
5656 '``ptrtoint .. to``' Instruction
5657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5664 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5669 The '``ptrtoint``' instruction converts the pointer or a vector of
5670 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5675 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5676 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5677 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5678 a vector of integers type.
5683 The '``ptrtoint``' instruction converts ``value`` to integer type
5684 ``ty2`` by interpreting the pointer value as an integer and either
5685 truncating or zero extending that value to the size of the integer type.
5686 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5687 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5688 the same size, then nothing is done (*no-op cast*) other than a type
5694 .. code-block:: llvm
5696 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5697 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5698 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5702 '``inttoptr .. to``' Instruction
5703 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5710 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5715 The '``inttoptr``' instruction converts an integer ``value`` to a
5716 pointer type, ``ty2``.
5721 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5722 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5728 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5729 applying either a zero extension or a truncation depending on the size
5730 of the integer ``value``. If ``value`` is larger than the size of a
5731 pointer then a truncation is done. If ``value`` is smaller than the size
5732 of a pointer then a zero extension is done. If they are the same size,
5733 nothing is done (*no-op cast*).
5738 .. code-block:: llvm
5740 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5741 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5742 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5743 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5747 '``bitcast .. to``' Instruction
5748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5755 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5760 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5766 The '``bitcast``' instruction takes a value to cast, which must be a
5767 non-aggregate first class value, and a type to cast it to, which must
5768 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5769 bit sizes of ``value`` and the destination type, ``ty2``, must be
5770 identical. If the source type is a pointer, the destination type must
5771 also be a pointer of the same size. This instruction supports bitwise
5772 conversion of vectors to integers and to vectors of other types (as
5773 long as they have the same size).
5778 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5779 is always a *no-op cast* because no bits change with this
5780 conversion. The conversion is done as if the ``value`` had been stored
5781 to memory and read back as type ``ty2``. Pointer (or vector of
5782 pointers) types may only be converted to other pointer (or vector of
5783 pointers) types with the same address space through this instruction.
5784 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
5785 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5790 .. code-block:: llvm
5792 %X = bitcast i8 255 to i8 ; yields i8 :-1
5793 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5794 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5795 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5797 .. _i_addrspacecast:
5799 '``addrspacecast .. to``' Instruction
5800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5807 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
5812 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
5813 address space ``n`` to type ``pty2`` in address space ``m``.
5818 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
5819 to cast and a pointer type to cast it to, which must have a different
5825 The '``addrspacecast``' instruction converts the pointer value
5826 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
5827 value modification, depending on the target and the address space
5828 pair. Pointer conversions within the same address space must be
5829 performed with the ``bitcast`` instruction. Note that if the address space
5830 conversion is legal then both result and operand refer to the same memory
5836 .. code-block:: llvm
5838 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
5839 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
5840 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
5847 The instructions in this category are the "miscellaneous" instructions,
5848 which defy better classification.
5852 '``icmp``' Instruction
5853 ^^^^^^^^^^^^^^^^^^^^^^
5860 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5865 The '``icmp``' instruction returns a boolean value or a vector of
5866 boolean values based on comparison of its two integer, integer vector,
5867 pointer, or pointer vector operands.
5872 The '``icmp``' instruction takes three operands. The first operand is
5873 the condition code indicating the kind of comparison to perform. It is
5874 not a value, just a keyword. The possible condition code are:
5877 #. ``ne``: not equal
5878 #. ``ugt``: unsigned greater than
5879 #. ``uge``: unsigned greater or equal
5880 #. ``ult``: unsigned less than
5881 #. ``ule``: unsigned less or equal
5882 #. ``sgt``: signed greater than
5883 #. ``sge``: signed greater or equal
5884 #. ``slt``: signed less than
5885 #. ``sle``: signed less or equal
5887 The remaining two arguments must be :ref:`integer <t_integer>` or
5888 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5889 must also be identical types.
5894 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5895 code given as ``cond``. The comparison performed always yields either an
5896 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5898 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5899 otherwise. No sign interpretation is necessary or performed.
5900 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5901 otherwise. No sign interpretation is necessary or performed.
5902 #. ``ugt``: interprets the operands as unsigned values and yields
5903 ``true`` if ``op1`` is greater than ``op2``.
5904 #. ``uge``: interprets the operands as unsigned values and yields
5905 ``true`` if ``op1`` is greater than or equal to ``op2``.
5906 #. ``ult``: interprets the operands as unsigned values and yields
5907 ``true`` if ``op1`` is less than ``op2``.
5908 #. ``ule``: interprets the operands as unsigned values and yields
5909 ``true`` if ``op1`` is less than or equal to ``op2``.
5910 #. ``sgt``: interprets the operands as signed values and yields ``true``
5911 if ``op1`` is greater than ``op2``.
5912 #. ``sge``: interprets the operands as signed values and yields ``true``
5913 if ``op1`` is greater than or equal to ``op2``.
5914 #. ``slt``: interprets the operands as signed values and yields ``true``
5915 if ``op1`` is less than ``op2``.
5916 #. ``sle``: interprets the operands as signed values and yields ``true``
5917 if ``op1`` is less than or equal to ``op2``.
5919 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5920 are compared as if they were integers.
5922 If the operands are integer vectors, then they are compared element by
5923 element. The result is an ``i1`` vector with the same number of elements
5924 as the values being compared. Otherwise, the result is an ``i1``.
5929 .. code-block:: llvm
5931 <result> = icmp eq i32 4, 5 ; yields: result=false
5932 <result> = icmp ne float* %X, %X ; yields: result=false
5933 <result> = icmp ult i16 4, 5 ; yields: result=true
5934 <result> = icmp sgt i16 4, 5 ; yields: result=false
5935 <result> = icmp ule i16 -4, 5 ; yields: result=false
5936 <result> = icmp sge i16 4, 5 ; yields: result=false
5938 Note that the code generator does not yet support vector types with the
5939 ``icmp`` instruction.
5943 '``fcmp``' Instruction
5944 ^^^^^^^^^^^^^^^^^^^^^^
5951 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5956 The '``fcmp``' instruction returns a boolean value or vector of boolean
5957 values based on comparison of its operands.
5959 If the operands are floating point scalars, then the result type is a
5960 boolean (:ref:`i1 <t_integer>`).
5962 If the operands are floating point vectors, then the result type is a
5963 vector of boolean with the same number of elements as the operands being
5969 The '``fcmp``' instruction takes three operands. The first operand is
5970 the condition code indicating the kind of comparison to perform. It is
5971 not a value, just a keyword. The possible condition code are:
5973 #. ``false``: no comparison, always returns false
5974 #. ``oeq``: ordered and equal
5975 #. ``ogt``: ordered and greater than
5976 #. ``oge``: ordered and greater than or equal
5977 #. ``olt``: ordered and less than
5978 #. ``ole``: ordered and less than or equal
5979 #. ``one``: ordered and not equal
5980 #. ``ord``: ordered (no nans)
5981 #. ``ueq``: unordered or equal
5982 #. ``ugt``: unordered or greater than
5983 #. ``uge``: unordered or greater than or equal
5984 #. ``ult``: unordered or less than
5985 #. ``ule``: unordered or less than or equal
5986 #. ``une``: unordered or not equal
5987 #. ``uno``: unordered (either nans)
5988 #. ``true``: no comparison, always returns true
5990 *Ordered* means that neither operand is a QNAN while *unordered* means
5991 that either operand may be a QNAN.
5993 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5994 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5995 type. They must have identical types.
6000 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6001 condition code given as ``cond``. If the operands are vectors, then the
6002 vectors are compared element by element. Each comparison performed
6003 always yields an :ref:`i1 <t_integer>` result, as follows:
6005 #. ``false``: always yields ``false``, regardless of operands.
6006 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6007 is equal to ``op2``.
6008 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6009 is greater than ``op2``.
6010 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6011 is greater than or equal to ``op2``.
6012 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6013 is less than ``op2``.
6014 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6015 is less than or equal to ``op2``.
6016 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6017 is not equal to ``op2``.
6018 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6019 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6021 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6022 greater than ``op2``.
6023 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6024 greater than or equal to ``op2``.
6025 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6027 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6028 less than or equal to ``op2``.
6029 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6030 not equal to ``op2``.
6031 #. ``uno``: yields ``true`` if either operand is a QNAN.
6032 #. ``true``: always yields ``true``, regardless of operands.
6037 .. code-block:: llvm
6039 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6040 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6041 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6042 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6044 Note that the code generator does not yet support vector types with the
6045 ``fcmp`` instruction.
6049 '``phi``' Instruction
6050 ^^^^^^^^^^^^^^^^^^^^^
6057 <result> = phi <ty> [ <val0>, <label0>], ...
6062 The '``phi``' instruction is used to implement the φ node in the SSA
6063 graph representing the function.
6068 The type of the incoming values is specified with the first type field.
6069 After this, the '``phi``' instruction takes a list of pairs as
6070 arguments, with one pair for each predecessor basic block of the current
6071 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6072 the value arguments to the PHI node. Only labels may be used as the
6075 There must be no non-phi instructions between the start of a basic block
6076 and the PHI instructions: i.e. PHI instructions must be first in a basic
6079 For the purposes of the SSA form, the use of each incoming value is
6080 deemed to occur on the edge from the corresponding predecessor block to
6081 the current block (but after any definition of an '``invoke``'
6082 instruction's return value on the same edge).
6087 At runtime, the '``phi``' instruction logically takes on the value
6088 specified by the pair corresponding to the predecessor basic block that
6089 executed just prior to the current block.
6094 .. code-block:: llvm
6096 Loop: ; Infinite loop that counts from 0 on up...
6097 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6098 %nextindvar = add i32 %indvar, 1
6103 '``select``' Instruction
6104 ^^^^^^^^^^^^^^^^^^^^^^^^
6111 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6113 selty is either i1 or {<N x i1>}
6118 The '``select``' instruction is used to choose one value based on a
6119 condition, without branching.
6124 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6125 values indicating the condition, and two values of the same :ref:`first
6126 class <t_firstclass>` type. If the val1/val2 are vectors and the
6127 condition is a scalar, then entire vectors are selected, not individual
6133 If the condition is an i1 and it evaluates to 1, the instruction returns
6134 the first value argument; otherwise, it returns the second value
6137 If the condition is a vector of i1, then the value arguments must be
6138 vectors of the same size, and the selection is done element by element.
6143 .. code-block:: llvm
6145 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6149 '``call``' Instruction
6150 ^^^^^^^^^^^^^^^^^^^^^^
6157 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6162 The '``call``' instruction represents a simple function call.
6167 This instruction requires several arguments:
6169 #. The optional "tail" marker indicates that the callee function does
6170 not access any allocas or varargs in the caller. Note that calls may
6171 be marked "tail" even if they do not occur before a
6172 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
6173 function call is eligible for tail call optimization, but `might not
6174 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
6175 The code generator may optimize calls marked "tail" with either 1)
6176 automatic `sibling call
6177 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6178 callee have matching signatures, or 2) forced tail call optimization
6179 when the following extra requirements are met:
6181 - Caller and callee both have the calling convention ``fastcc``.
6182 - The call is in tail position (ret immediately follows call and ret
6183 uses value of call or is void).
6184 - Option ``-tailcallopt`` is enabled, or
6185 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6186 - `Platform specific constraints are
6187 met. <CodeGenerator.html#tailcallopt>`_
6189 #. The optional "cconv" marker indicates which :ref:`calling
6190 convention <callingconv>` the call should use. If none is
6191 specified, the call defaults to using C calling conventions. The
6192 calling convention of the call must match the calling convention of
6193 the target function, or else the behavior is undefined.
6194 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6195 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6197 #. '``ty``': the type of the call instruction itself which is also the
6198 type of the return value. Functions that return no value are marked
6200 #. '``fnty``': shall be the signature of the pointer to function value
6201 being invoked. The argument types must match the types implied by
6202 this signature. This type can be omitted if the function is not
6203 varargs and if the function type does not return a pointer to a
6205 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6206 be invoked. In most cases, this is a direct function invocation, but
6207 indirect ``call``'s are just as possible, calling an arbitrary pointer
6209 #. '``function args``': argument list whose types match the function
6210 signature argument types and parameter attributes. All arguments must
6211 be of :ref:`first class <t_firstclass>` type. If the function signature
6212 indicates the function accepts a variable number of arguments, the
6213 extra arguments can be specified.
6214 #. The optional :ref:`function attributes <fnattrs>` list. Only
6215 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6216 attributes are valid here.
6221 The '``call``' instruction is used to cause control flow to transfer to
6222 a specified function, with its incoming arguments bound to the specified
6223 values. Upon a '``ret``' instruction in the called function, control
6224 flow continues with the instruction after the function call, and the
6225 return value of the function is bound to the result argument.
6230 .. code-block:: llvm
6232 %retval = call i32 @test(i32 %argc)
6233 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6234 %X = tail call i32 @foo() ; yields i32
6235 %Y = tail call fastcc i32 @foo() ; yields i32
6236 call void %foo(i8 97 signext)
6238 %struct.A = type { i32, i8 }
6239 %r = call %struct.A @foo() ; yields { 32, i8 }
6240 %gr = extractvalue %struct.A %r, 0 ; yields i32
6241 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6242 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6243 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6245 llvm treats calls to some functions with names and arguments that match
6246 the standard C99 library as being the C99 library functions, and may
6247 perform optimizations or generate code for them under that assumption.
6248 This is something we'd like to change in the future to provide better
6249 support for freestanding environments and non-C-based languages.
6253 '``va_arg``' Instruction
6254 ^^^^^^^^^^^^^^^^^^^^^^^^
6261 <resultval> = va_arg <va_list*> <arglist>, <argty>
6266 The '``va_arg``' instruction is used to access arguments passed through
6267 the "variable argument" area of a function call. It is used to implement
6268 the ``va_arg`` macro in C.
6273 This instruction takes a ``va_list*`` value and the type of the
6274 argument. It returns a value of the specified argument type and
6275 increments the ``va_list`` to point to the next argument. The actual
6276 type of ``va_list`` is target specific.
6281 The '``va_arg``' instruction loads an argument of the specified type
6282 from the specified ``va_list`` and causes the ``va_list`` to point to
6283 the next argument. For more information, see the variable argument
6284 handling :ref:`Intrinsic Functions <int_varargs>`.
6286 It is legal for this instruction to be called in a function which does
6287 not take a variable number of arguments, for example, the ``vfprintf``
6290 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6291 function <intrinsics>` because it takes a type as an argument.
6296 See the :ref:`variable argument processing <int_varargs>` section.
6298 Note that the code generator does not yet fully support va\_arg on many
6299 targets. Also, it does not currently support va\_arg with aggregate
6300 types on any target.
6304 '``landingpad``' Instruction
6305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6312 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6313 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6315 <clause> := catch <type> <value>
6316 <clause> := filter <array constant type> <array constant>
6321 The '``landingpad``' instruction is used by `LLVM's exception handling
6322 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6323 is a landing pad --- one where the exception lands, and corresponds to the
6324 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6325 defines values supplied by the personality function (``pers_fn``) upon
6326 re-entry to the function. The ``resultval`` has the type ``resultty``.
6331 This instruction takes a ``pers_fn`` value. This is the personality
6332 function associated with the unwinding mechanism. The optional
6333 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6335 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6336 contains the global variable representing the "type" that may be caught
6337 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6338 clause takes an array constant as its argument. Use
6339 "``[0 x i8**] undef``" for a filter which cannot throw. The
6340 '``landingpad``' instruction must contain *at least* one ``clause`` or
6341 the ``cleanup`` flag.
6346 The '``landingpad``' instruction defines the values which are set by the
6347 personality function (``pers_fn``) upon re-entry to the function, and
6348 therefore the "result type" of the ``landingpad`` instruction. As with
6349 calling conventions, how the personality function results are
6350 represented in LLVM IR is target specific.
6352 The clauses are applied in order from top to bottom. If two
6353 ``landingpad`` instructions are merged together through inlining, the
6354 clauses from the calling function are appended to the list of clauses.
6355 When the call stack is being unwound due to an exception being thrown,
6356 the exception is compared against each ``clause`` in turn. If it doesn't
6357 match any of the clauses, and the ``cleanup`` flag is not set, then
6358 unwinding continues further up the call stack.
6360 The ``landingpad`` instruction has several restrictions:
6362 - A landing pad block is a basic block which is the unwind destination
6363 of an '``invoke``' instruction.
6364 - A landing pad block must have a '``landingpad``' instruction as its
6365 first non-PHI instruction.
6366 - There can be only one '``landingpad``' instruction within the landing
6368 - A basic block that is not a landing pad block may not include a
6369 '``landingpad``' instruction.
6370 - All '``landingpad``' instructions in a function must have the same
6371 personality function.
6376 .. code-block:: llvm
6378 ;; A landing pad which can catch an integer.
6379 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6381 ;; A landing pad that is a cleanup.
6382 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6384 ;; A landing pad which can catch an integer and can only throw a double.
6385 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6387 filter [1 x i8**] [@_ZTId]
6394 LLVM supports the notion of an "intrinsic function". These functions
6395 have well known names and semantics and are required to follow certain
6396 restrictions. Overall, these intrinsics represent an extension mechanism
6397 for the LLVM language that does not require changing all of the
6398 transformations in LLVM when adding to the language (or the bitcode
6399 reader/writer, the parser, etc...).
6401 Intrinsic function names must all start with an "``llvm.``" prefix. This
6402 prefix is reserved in LLVM for intrinsic names; thus, function names may
6403 not begin with this prefix. Intrinsic functions must always be external
6404 functions: you cannot define the body of intrinsic functions. Intrinsic
6405 functions may only be used in call or invoke instructions: it is illegal
6406 to take the address of an intrinsic function. Additionally, because
6407 intrinsic functions are part of the LLVM language, it is required if any
6408 are added that they be documented here.
6410 Some intrinsic functions can be overloaded, i.e., the intrinsic
6411 represents a family of functions that perform the same operation but on
6412 different data types. Because LLVM can represent over 8 million
6413 different integer types, overloading is used commonly to allow an
6414 intrinsic function to operate on any integer type. One or more of the
6415 argument types or the result type can be overloaded to accept any
6416 integer type. Argument types may also be defined as exactly matching a
6417 previous argument's type or the result type. This allows an intrinsic
6418 function which accepts multiple arguments, but needs all of them to be
6419 of the same type, to only be overloaded with respect to a single
6420 argument or the result.
6422 Overloaded intrinsics will have the names of its overloaded argument
6423 types encoded into its function name, each preceded by a period. Only
6424 those types which are overloaded result in a name suffix. Arguments
6425 whose type is matched against another type do not. For example, the
6426 ``llvm.ctpop`` function can take an integer of any width and returns an
6427 integer of exactly the same integer width. This leads to a family of
6428 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6429 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6430 overloaded, and only one type suffix is required. Because the argument's
6431 type is matched against the return type, it does not require its own
6434 To learn how to add an intrinsic function, please see the `Extending
6435 LLVM Guide <ExtendingLLVM.html>`_.
6439 Variable Argument Handling Intrinsics
6440 -------------------------------------
6442 Variable argument support is defined in LLVM with the
6443 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6444 functions. These functions are related to the similarly named macros
6445 defined in the ``<stdarg.h>`` header file.
6447 All of these functions operate on arguments that use a target-specific
6448 value type "``va_list``". The LLVM assembly language reference manual
6449 does not define what this type is, so all transformations should be
6450 prepared to handle these functions regardless of the type used.
6452 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6453 variable argument handling intrinsic functions are used.
6455 .. code-block:: llvm
6457 define i32 @test(i32 %X, ...) {
6458 ; Initialize variable argument processing
6460 %ap2 = bitcast i8** %ap to i8*
6461 call void @llvm.va_start(i8* %ap2)
6463 ; Read a single integer argument
6464 %tmp = va_arg i8** %ap, i32
6466 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6468 %aq2 = bitcast i8** %aq to i8*
6469 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6470 call void @llvm.va_end(i8* %aq2)
6472 ; Stop processing of arguments.
6473 call void @llvm.va_end(i8* %ap2)
6477 declare void @llvm.va_start(i8*)
6478 declare void @llvm.va_copy(i8*, i8*)
6479 declare void @llvm.va_end(i8*)
6483 '``llvm.va_start``' Intrinsic
6484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6491 declare void @llvm.va_start(i8* <arglist>)
6496 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6497 subsequent use by ``va_arg``.
6502 The argument is a pointer to a ``va_list`` element to initialize.
6507 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6508 available in C. In a target-dependent way, it initializes the
6509 ``va_list`` element to which the argument points, so that the next call
6510 to ``va_arg`` will produce the first variable argument passed to the
6511 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6512 to know the last argument of the function as the compiler can figure
6515 '``llvm.va_end``' Intrinsic
6516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6523 declare void @llvm.va_end(i8* <arglist>)
6528 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6529 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6534 The argument is a pointer to a ``va_list`` to destroy.
6539 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6540 available in C. In a target-dependent way, it destroys the ``va_list``
6541 element to which the argument points. Calls to
6542 :ref:`llvm.va_start <int_va_start>` and
6543 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6548 '``llvm.va_copy``' Intrinsic
6549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6556 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6561 The '``llvm.va_copy``' intrinsic copies the current argument position
6562 from the source argument list to the destination argument list.
6567 The first argument is a pointer to a ``va_list`` element to initialize.
6568 The second argument is a pointer to a ``va_list`` element to copy from.
6573 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6574 available in C. In a target-dependent way, it copies the source
6575 ``va_list`` element into the destination ``va_list`` element. This
6576 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6577 arbitrarily complex and require, for example, memory allocation.
6579 Accurate Garbage Collection Intrinsics
6580 --------------------------------------
6582 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6583 (GC) requires the implementation and generation of these intrinsics.
6584 These intrinsics allow identification of :ref:`GC roots on the
6585 stack <int_gcroot>`, as well as garbage collector implementations that
6586 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6587 Front-ends for type-safe garbage collected languages should generate
6588 these intrinsics to make use of the LLVM garbage collectors. For more
6589 details, see `Accurate Garbage Collection with
6590 LLVM <GarbageCollection.html>`_.
6592 The garbage collection intrinsics only operate on objects in the generic
6593 address space (address space zero).
6597 '``llvm.gcroot``' Intrinsic
6598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6605 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6610 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6611 the code generator, and allows some metadata to be associated with it.
6616 The first argument specifies the address of a stack object that contains
6617 the root pointer. The second pointer (which must be either a constant or
6618 a global value address) contains the meta-data to be associated with the
6624 At runtime, a call to this intrinsic stores a null pointer into the
6625 "ptrloc" location. At compile-time, the code generator generates
6626 information to allow the runtime to find the pointer at GC safe points.
6627 The '``llvm.gcroot``' intrinsic may only be used in a function which
6628 :ref:`specifies a GC algorithm <gc>`.
6632 '``llvm.gcread``' Intrinsic
6633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6640 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6645 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6646 locations, allowing garbage collector implementations that require read
6652 The second argument is the address to read from, which should be an
6653 address allocated from the garbage collector. The first object is a
6654 pointer to the start of the referenced object, if needed by the language
6655 runtime (otherwise null).
6660 The '``llvm.gcread``' intrinsic has the same semantics as a load
6661 instruction, but may be replaced with substantially more complex code by
6662 the garbage collector runtime, as needed. The '``llvm.gcread``'
6663 intrinsic may only be used in a function which :ref:`specifies a GC
6668 '``llvm.gcwrite``' Intrinsic
6669 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6676 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6681 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6682 locations, allowing garbage collector implementations that require write
6683 barriers (such as generational or reference counting collectors).
6688 The first argument is the reference to store, the second is the start of
6689 the object to store it to, and the third is the address of the field of
6690 Obj to store to. If the runtime does not require a pointer to the
6691 object, Obj may be null.
6696 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6697 instruction, but may be replaced with substantially more complex code by
6698 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6699 intrinsic may only be used in a function which :ref:`specifies a GC
6702 Code Generator Intrinsics
6703 -------------------------
6705 These intrinsics are provided by LLVM to expose special features that
6706 may only be implemented with code generator support.
6708 '``llvm.returnaddress``' Intrinsic
6709 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6716 declare i8 *@llvm.returnaddress(i32 <level>)
6721 The '``llvm.returnaddress``' intrinsic attempts to compute a
6722 target-specific value indicating the return address of the current
6723 function or one of its callers.
6728 The argument to this intrinsic indicates which function to return the
6729 address for. Zero indicates the calling function, one indicates its
6730 caller, etc. The argument is **required** to be a constant integer
6736 The '``llvm.returnaddress``' intrinsic either returns a pointer
6737 indicating the return address of the specified call frame, or zero if it
6738 cannot be identified. The value returned by this intrinsic is likely to
6739 be incorrect or 0 for arguments other than zero, so it should only be
6740 used for debugging purposes.
6742 Note that calling this intrinsic does not prevent function inlining or
6743 other aggressive transformations, so the value returned may not be that
6744 of the obvious source-language caller.
6746 '``llvm.frameaddress``' Intrinsic
6747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6754 declare i8* @llvm.frameaddress(i32 <level>)
6759 The '``llvm.frameaddress``' intrinsic attempts to return the
6760 target-specific frame pointer value for the specified stack frame.
6765 The argument to this intrinsic indicates which function to return the
6766 frame pointer for. Zero indicates the calling function, one indicates
6767 its caller, etc. The argument is **required** to be a constant integer
6773 The '``llvm.frameaddress``' intrinsic either returns a pointer
6774 indicating the frame address of the specified call frame, or zero if it
6775 cannot be identified. The value returned by this intrinsic is likely to
6776 be incorrect or 0 for arguments other than zero, so it should only be
6777 used for debugging purposes.
6779 Note that calling this intrinsic does not prevent function inlining or
6780 other aggressive transformations, so the value returned may not be that
6781 of the obvious source-language caller.
6785 '``llvm.stacksave``' Intrinsic
6786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6793 declare i8* @llvm.stacksave()
6798 The '``llvm.stacksave``' intrinsic is used to remember the current state
6799 of the function stack, for use with
6800 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6801 implementing language features like scoped automatic variable sized
6807 This intrinsic returns a opaque pointer value that can be passed to
6808 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6809 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6810 ``llvm.stacksave``, it effectively restores the state of the stack to
6811 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6812 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6813 were allocated after the ``llvm.stacksave`` was executed.
6815 .. _int_stackrestore:
6817 '``llvm.stackrestore``' Intrinsic
6818 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6825 declare void @llvm.stackrestore(i8* %ptr)
6830 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6831 the function stack to the state it was in when the corresponding
6832 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6833 useful for implementing language features like scoped automatic variable
6834 sized arrays in C99.
6839 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6841 '``llvm.prefetch``' Intrinsic
6842 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6849 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6854 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6855 insert a prefetch instruction if supported; otherwise, it is a noop.
6856 Prefetches have no effect on the behavior of the program but can change
6857 its performance characteristics.
6862 ``address`` is the address to be prefetched, ``rw`` is the specifier
6863 determining if the fetch should be for a read (0) or write (1), and
6864 ``locality`` is a temporal locality specifier ranging from (0) - no
6865 locality, to (3) - extremely local keep in cache. The ``cache type``
6866 specifies whether the prefetch is performed on the data (1) or
6867 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6868 arguments must be constant integers.
6873 This intrinsic does not modify the behavior of the program. In
6874 particular, prefetches cannot trap and do not produce a value. On
6875 targets that support this intrinsic, the prefetch can provide hints to
6876 the processor cache for better performance.
6878 '``llvm.pcmarker``' Intrinsic
6879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6886 declare void @llvm.pcmarker(i32 <id>)
6891 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6892 Counter (PC) in a region of code to simulators and other tools. The
6893 method is target specific, but it is expected that the marker will use
6894 exported symbols to transmit the PC of the marker. The marker makes no
6895 guarantees that it will remain with any specific instruction after
6896 optimizations. It is possible that the presence of a marker will inhibit
6897 optimizations. The intended use is to be inserted after optimizations to
6898 allow correlations of simulation runs.
6903 ``id`` is a numerical id identifying the marker.
6908 This intrinsic does not modify the behavior of the program. Backends
6909 that do not support this intrinsic may ignore it.
6911 '``llvm.readcyclecounter``' Intrinsic
6912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6919 declare i64 @llvm.readcyclecounter()
6924 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6925 counter register (or similar low latency, high accuracy clocks) on those
6926 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6927 should map to RPCC. As the backing counters overflow quickly (on the
6928 order of 9 seconds on alpha), this should only be used for small
6934 When directly supported, reading the cycle counter should not modify any
6935 memory. Implementations are allowed to either return a application
6936 specific value or a system wide value. On backends without support, this
6937 is lowered to a constant 0.
6939 Note that runtime support may be conditional on the privilege-level code is
6940 running at and the host platform.
6942 '``llvm.clear_cache``' Intrinsic
6943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6950 declare void @llvm.clear_cache(i8*, i8*)
6955 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
6956 in the specified range to the execution unit of the processor. On
6957 targets with non-unified instruction and data cache, the implementation
6958 flushes the instruction cache.
6963 On platforms with coherent instruction and data caches (e.g. x86), this
6964 intrinsic is a nop. On platforms with non-coherent instruction and data
6965 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropiate
6966 instructions or a system call, if cache flushing requires special
6969 The default behavior is to emit a call to ``__clear_cache'' from the run
6972 This instrinsic does *not* empty the instruction pipeline. Modifications
6973 of the current function are outside the scope of the intrinsic.
6975 Standard C Library Intrinsics
6976 -----------------------------
6978 LLVM provides intrinsics for a few important standard C library
6979 functions. These intrinsics allow source-language front-ends to pass
6980 information about the alignment of the pointer arguments to the code
6981 generator, providing opportunity for more efficient code generation.
6985 '``llvm.memcpy``' Intrinsic
6986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6991 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6992 integer bit width and for different address spaces. Not all targets
6993 support all bit widths however.
6997 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6998 i32 <len>, i32 <align>, i1 <isvolatile>)
6999 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7000 i64 <len>, i32 <align>, i1 <isvolatile>)
7005 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7006 source location to the destination location.
7008 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7009 intrinsics do not return a value, takes extra alignment/isvolatile
7010 arguments and the pointers can be in specified address spaces.
7015 The first argument is a pointer to the destination, the second is a
7016 pointer to the source. The third argument is an integer argument
7017 specifying the number of bytes to copy, the fourth argument is the
7018 alignment of the source and destination locations, and the fifth is a
7019 boolean indicating a volatile access.
7021 If the call to this intrinsic has an alignment value that is not 0 or 1,
7022 then the caller guarantees that both the source and destination pointers
7023 are aligned to that boundary.
7025 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7026 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7027 very cleanly specified and it is unwise to depend on it.
7032 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7033 source location to the destination location, which are not allowed to
7034 overlap. It copies "len" bytes of memory over. If the argument is known
7035 to be aligned to some boundary, this can be specified as the fourth
7036 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7038 '``llvm.memmove``' Intrinsic
7039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7044 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7045 bit width and for different address space. Not all targets support all
7050 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7051 i32 <len>, i32 <align>, i1 <isvolatile>)
7052 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7053 i64 <len>, i32 <align>, i1 <isvolatile>)
7058 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7059 source location to the destination location. It is similar to the
7060 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7063 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7064 intrinsics do not return a value, takes extra alignment/isvolatile
7065 arguments and the pointers can be in specified address spaces.
7070 The first argument is a pointer to the destination, the second is a
7071 pointer to the source. The third argument is an integer argument
7072 specifying the number of bytes to copy, the fourth argument is the
7073 alignment of the source and destination locations, and the fifth is a
7074 boolean indicating a volatile access.
7076 If the call to this intrinsic has an alignment value that is not 0 or 1,
7077 then the caller guarantees that the source and destination pointers are
7078 aligned to that boundary.
7080 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7081 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7082 not very cleanly specified and it is unwise to depend on it.
7087 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7088 source location to the destination location, which may overlap. It
7089 copies "len" bytes of memory over. If the argument is known to be
7090 aligned to some boundary, this can be specified as the fourth argument,
7091 otherwise it should be set to 0 or 1 (both meaning no alignment).
7093 '``llvm.memset.*``' Intrinsics
7094 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7099 This is an overloaded intrinsic. You can use llvm.memset on any integer
7100 bit width and for different address spaces. However, not all targets
7101 support all bit widths.
7105 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7106 i32 <len>, i32 <align>, i1 <isvolatile>)
7107 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7108 i64 <len>, i32 <align>, i1 <isvolatile>)
7113 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7114 particular byte value.
7116 Note that, unlike the standard libc function, the ``llvm.memset``
7117 intrinsic does not return a value and takes extra alignment/volatile
7118 arguments. Also, the destination can be in an arbitrary address space.
7123 The first argument is a pointer to the destination to fill, the second
7124 is the byte value with which to fill it, the third argument is an
7125 integer argument specifying the number of bytes to fill, and the fourth
7126 argument is the known alignment of the destination location.
7128 If the call to this intrinsic has an alignment value that is not 0 or 1,
7129 then the caller guarantees that the destination pointer is aligned to
7132 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7133 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7134 very cleanly specified and it is unwise to depend on it.
7139 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7140 at the destination location. If the argument is known to be aligned to
7141 some boundary, this can be specified as the fourth argument, otherwise
7142 it should be set to 0 or 1 (both meaning no alignment).
7144 '``llvm.sqrt.*``' Intrinsic
7145 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7150 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7151 floating point or vector of floating point type. Not all targets support
7156 declare float @llvm.sqrt.f32(float %Val)
7157 declare double @llvm.sqrt.f64(double %Val)
7158 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7159 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7160 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7165 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7166 returning the same value as the libm '``sqrt``' functions would. Unlike
7167 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7168 negative numbers other than -0.0 (which allows for better optimization,
7169 because there is no need to worry about errno being set).
7170 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7175 The argument and return value are floating point numbers of the same
7181 This function returns the sqrt of the specified operand if it is a
7182 nonnegative floating point number.
7184 '``llvm.powi.*``' Intrinsic
7185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7190 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7191 floating point or vector of floating point type. Not all targets support
7196 declare float @llvm.powi.f32(float %Val, i32 %power)
7197 declare double @llvm.powi.f64(double %Val, i32 %power)
7198 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7199 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7200 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7205 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7206 specified (positive or negative) power. The order of evaluation of
7207 multiplications is not defined. When a vector of floating point type is
7208 used, the second argument remains a scalar integer value.
7213 The second argument is an integer power, and the first is a value to
7214 raise to that power.
7219 This function returns the first value raised to the second power with an
7220 unspecified sequence of rounding operations.
7222 '``llvm.sin.*``' Intrinsic
7223 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7228 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7229 floating point or vector of floating point type. Not all targets support
7234 declare float @llvm.sin.f32(float %Val)
7235 declare double @llvm.sin.f64(double %Val)
7236 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7237 declare fp128 @llvm.sin.f128(fp128 %Val)
7238 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7243 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7248 The argument and return value are floating point numbers of the same
7254 This function returns the sine of the specified operand, returning the
7255 same values as the libm ``sin`` functions would, and handles error
7256 conditions in the same way.
7258 '``llvm.cos.*``' Intrinsic
7259 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7264 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7265 floating point or vector of floating point type. Not all targets support
7270 declare float @llvm.cos.f32(float %Val)
7271 declare double @llvm.cos.f64(double %Val)
7272 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7273 declare fp128 @llvm.cos.f128(fp128 %Val)
7274 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7279 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7284 The argument and return value are floating point numbers of the same
7290 This function returns the cosine of the specified operand, returning the
7291 same values as the libm ``cos`` functions would, and handles error
7292 conditions in the same way.
7294 '``llvm.pow.*``' Intrinsic
7295 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7300 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7301 floating point or vector of floating point type. Not all targets support
7306 declare float @llvm.pow.f32(float %Val, float %Power)
7307 declare double @llvm.pow.f64(double %Val, double %Power)
7308 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7309 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7310 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7315 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7316 specified (positive or negative) power.
7321 The second argument is a floating point power, and the first is a value
7322 to raise to that power.
7327 This function returns the first value raised to the second power,
7328 returning the same values as the libm ``pow`` functions would, and
7329 handles error conditions in the same way.
7331 '``llvm.exp.*``' Intrinsic
7332 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7337 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7338 floating point or vector of floating point type. Not all targets support
7343 declare float @llvm.exp.f32(float %Val)
7344 declare double @llvm.exp.f64(double %Val)
7345 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7346 declare fp128 @llvm.exp.f128(fp128 %Val)
7347 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7352 The '``llvm.exp.*``' intrinsics perform the exp function.
7357 The argument and return value are floating point numbers of the same
7363 This function returns the same values as the libm ``exp`` functions
7364 would, and handles error conditions in the same way.
7366 '``llvm.exp2.*``' Intrinsic
7367 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7372 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7373 floating point or vector of floating point type. Not all targets support
7378 declare float @llvm.exp2.f32(float %Val)
7379 declare double @llvm.exp2.f64(double %Val)
7380 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7381 declare fp128 @llvm.exp2.f128(fp128 %Val)
7382 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7387 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7392 The argument and return value are floating point numbers of the same
7398 This function returns the same values as the libm ``exp2`` functions
7399 would, and handles error conditions in the same way.
7401 '``llvm.log.*``' Intrinsic
7402 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7407 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7408 floating point or vector of floating point type. Not all targets support
7413 declare float @llvm.log.f32(float %Val)
7414 declare double @llvm.log.f64(double %Val)
7415 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7416 declare fp128 @llvm.log.f128(fp128 %Val)
7417 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7422 The '``llvm.log.*``' intrinsics perform the log function.
7427 The argument and return value are floating point numbers of the same
7433 This function returns the same values as the libm ``log`` functions
7434 would, and handles error conditions in the same way.
7436 '``llvm.log10.*``' Intrinsic
7437 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7442 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7443 floating point or vector of floating point type. Not all targets support
7448 declare float @llvm.log10.f32(float %Val)
7449 declare double @llvm.log10.f64(double %Val)
7450 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7451 declare fp128 @llvm.log10.f128(fp128 %Val)
7452 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7457 The '``llvm.log10.*``' intrinsics perform the log10 function.
7462 The argument and return value are floating point numbers of the same
7468 This function returns the same values as the libm ``log10`` functions
7469 would, and handles error conditions in the same way.
7471 '``llvm.log2.*``' Intrinsic
7472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7477 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7478 floating point or vector of floating point type. Not all targets support
7483 declare float @llvm.log2.f32(float %Val)
7484 declare double @llvm.log2.f64(double %Val)
7485 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7486 declare fp128 @llvm.log2.f128(fp128 %Val)
7487 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7492 The '``llvm.log2.*``' intrinsics perform the log2 function.
7497 The argument and return value are floating point numbers of the same
7503 This function returns the same values as the libm ``log2`` functions
7504 would, and handles error conditions in the same way.
7506 '``llvm.fma.*``' Intrinsic
7507 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7512 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7513 floating point or vector of floating point type. Not all targets support
7518 declare float @llvm.fma.f32(float %a, float %b, float %c)
7519 declare double @llvm.fma.f64(double %a, double %b, double %c)
7520 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7521 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7522 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7527 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7533 The argument and return value are floating point numbers of the same
7539 This function returns the same values as the libm ``fma`` functions
7540 would, and does not set errno.
7542 '``llvm.fabs.*``' Intrinsic
7543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7548 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7549 floating point or vector of floating point type. Not all targets support
7554 declare float @llvm.fabs.f32(float %Val)
7555 declare double @llvm.fabs.f64(double %Val)
7556 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7557 declare fp128 @llvm.fabs.f128(fp128 %Val)
7558 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7563 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7569 The argument and return value are floating point numbers of the same
7575 This function returns the same values as the libm ``fabs`` functions
7576 would, and handles error conditions in the same way.
7578 '``llvm.copysign.*``' Intrinsic
7579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7584 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7585 floating point or vector of floating point type. Not all targets support
7590 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7591 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7592 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7593 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7594 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7599 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7600 first operand and the sign of the second operand.
7605 The arguments and return value are floating point numbers of the same
7611 This function returns the same values as the libm ``copysign``
7612 functions would, and handles error conditions in the same way.
7614 '``llvm.floor.*``' Intrinsic
7615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7620 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7621 floating point or vector of floating point type. Not all targets support
7626 declare float @llvm.floor.f32(float %Val)
7627 declare double @llvm.floor.f64(double %Val)
7628 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7629 declare fp128 @llvm.floor.f128(fp128 %Val)
7630 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7635 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7640 The argument and return value are floating point numbers of the same
7646 This function returns the same values as the libm ``floor`` functions
7647 would, and handles error conditions in the same way.
7649 '``llvm.ceil.*``' Intrinsic
7650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7655 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7656 floating point or vector of floating point type. Not all targets support
7661 declare float @llvm.ceil.f32(float %Val)
7662 declare double @llvm.ceil.f64(double %Val)
7663 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7664 declare fp128 @llvm.ceil.f128(fp128 %Val)
7665 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7670 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7675 The argument and return value are floating point numbers of the same
7681 This function returns the same values as the libm ``ceil`` functions
7682 would, and handles error conditions in the same way.
7684 '``llvm.trunc.*``' Intrinsic
7685 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7690 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7691 floating point or vector of floating point type. Not all targets support
7696 declare float @llvm.trunc.f32(float %Val)
7697 declare double @llvm.trunc.f64(double %Val)
7698 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7699 declare fp128 @llvm.trunc.f128(fp128 %Val)
7700 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7705 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7706 nearest integer not larger in magnitude than the operand.
7711 The argument and return value are floating point numbers of the same
7717 This function returns the same values as the libm ``trunc`` functions
7718 would, and handles error conditions in the same way.
7720 '``llvm.rint.*``' Intrinsic
7721 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7726 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7727 floating point or vector of floating point type. Not all targets support
7732 declare float @llvm.rint.f32(float %Val)
7733 declare double @llvm.rint.f64(double %Val)
7734 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7735 declare fp128 @llvm.rint.f128(fp128 %Val)
7736 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7741 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7742 nearest integer. It may raise an inexact floating-point exception if the
7743 operand isn't an integer.
7748 The argument and return value are floating point numbers of the same
7754 This function returns the same values as the libm ``rint`` functions
7755 would, and handles error conditions in the same way.
7757 '``llvm.nearbyint.*``' Intrinsic
7758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7763 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7764 floating point or vector of floating point type. Not all targets support
7769 declare float @llvm.nearbyint.f32(float %Val)
7770 declare double @llvm.nearbyint.f64(double %Val)
7771 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7772 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7773 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7778 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7784 The argument and return value are floating point numbers of the same
7790 This function returns the same values as the libm ``nearbyint``
7791 functions would, and handles error conditions in the same way.
7793 '``llvm.round.*``' Intrinsic
7794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7799 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7800 floating point or vector of floating point type. Not all targets support
7805 declare float @llvm.round.f32(float %Val)
7806 declare double @llvm.round.f64(double %Val)
7807 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7808 declare fp128 @llvm.round.f128(fp128 %Val)
7809 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7814 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7820 The argument and return value are floating point numbers of the same
7826 This function returns the same values as the libm ``round``
7827 functions would, and handles error conditions in the same way.
7829 Bit Manipulation Intrinsics
7830 ---------------------------
7832 LLVM provides intrinsics for a few important bit manipulation
7833 operations. These allow efficient code generation for some algorithms.
7835 '``llvm.bswap.*``' Intrinsics
7836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7841 This is an overloaded intrinsic function. You can use bswap on any
7842 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7846 declare i16 @llvm.bswap.i16(i16 <id>)
7847 declare i32 @llvm.bswap.i32(i32 <id>)
7848 declare i64 @llvm.bswap.i64(i64 <id>)
7853 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7854 values with an even number of bytes (positive multiple of 16 bits).
7855 These are useful for performing operations on data that is not in the
7856 target's native byte order.
7861 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7862 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7863 intrinsic returns an i32 value that has the four bytes of the input i32
7864 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7865 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7866 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7867 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7870 '``llvm.ctpop.*``' Intrinsic
7871 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7876 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7877 bit width, or on any vector with integer elements. Not all targets
7878 support all bit widths or vector types, however.
7882 declare i8 @llvm.ctpop.i8(i8 <src>)
7883 declare i16 @llvm.ctpop.i16(i16 <src>)
7884 declare i32 @llvm.ctpop.i32(i32 <src>)
7885 declare i64 @llvm.ctpop.i64(i64 <src>)
7886 declare i256 @llvm.ctpop.i256(i256 <src>)
7887 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7892 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7898 The only argument is the value to be counted. The argument may be of any
7899 integer type, or a vector with integer elements. The return type must
7900 match the argument type.
7905 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7906 each element of a vector.
7908 '``llvm.ctlz.*``' Intrinsic
7909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7914 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7915 integer bit width, or any vector whose elements are integers. Not all
7916 targets support all bit widths or vector types, however.
7920 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7921 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7922 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7923 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7924 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7925 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7930 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7931 leading zeros in a variable.
7936 The first argument is the value to be counted. This argument may be of
7937 any integer type, or a vectory with integer element type. The return
7938 type must match the first argument type.
7940 The second argument must be a constant and is a flag to indicate whether
7941 the intrinsic should ensure that a zero as the first argument produces a
7942 defined result. Historically some architectures did not provide a
7943 defined result for zero values as efficiently, and many algorithms are
7944 now predicated on avoiding zero-value inputs.
7949 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7950 zeros in a variable, or within each element of the vector. If
7951 ``src == 0`` then the result is the size in bits of the type of ``src``
7952 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7953 ``llvm.ctlz(i32 2) = 30``.
7955 '``llvm.cttz.*``' Intrinsic
7956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7961 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7962 integer bit width, or any vector of integer elements. Not all targets
7963 support all bit widths or vector types, however.
7967 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7968 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7969 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7970 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7971 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7972 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7977 The '``llvm.cttz``' family of intrinsic functions counts the number of
7983 The first argument is the value to be counted. This argument may be of
7984 any integer type, or a vectory with integer element type. The return
7985 type must match the first argument type.
7987 The second argument must be a constant and is a flag to indicate whether
7988 the intrinsic should ensure that a zero as the first argument produces a
7989 defined result. Historically some architectures did not provide a
7990 defined result for zero values as efficiently, and many algorithms are
7991 now predicated on avoiding zero-value inputs.
7996 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7997 zeros in a variable, or within each element of a vector. If ``src == 0``
7998 then the result is the size in bits of the type of ``src`` if
7999 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8000 ``llvm.cttz(2) = 1``.
8002 Arithmetic with Overflow Intrinsics
8003 -----------------------------------
8005 LLVM provides intrinsics for some arithmetic with overflow operations.
8007 '``llvm.sadd.with.overflow.*``' Intrinsics
8008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8013 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8014 on any integer bit width.
8018 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8019 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8020 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8025 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8026 a signed addition of the two arguments, and indicate whether an overflow
8027 occurred during the signed summation.
8032 The arguments (%a and %b) and the first element of the result structure
8033 may be of integer types of any bit width, but they must have the same
8034 bit width. The second element of the result structure must be of type
8035 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8041 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8042 a signed addition of the two variables. They return a structure --- the
8043 first element of which is the signed summation, and the second element
8044 of which is a bit specifying if the signed summation resulted in an
8050 .. code-block:: llvm
8052 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8053 %sum = extractvalue {i32, i1} %res, 0
8054 %obit = extractvalue {i32, i1} %res, 1
8055 br i1 %obit, label %overflow, label %normal
8057 '``llvm.uadd.with.overflow.*``' Intrinsics
8058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8063 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8064 on any integer bit width.
8068 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8069 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8070 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8075 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8076 an unsigned addition of the two arguments, and indicate whether a carry
8077 occurred during the unsigned summation.
8082 The arguments (%a and %b) and the first element of the result structure
8083 may be of integer types of any bit width, but they must have the same
8084 bit width. The second element of the result structure must be of type
8085 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8091 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8092 an unsigned addition of the two arguments. They return a structure --- the
8093 first element of which is the sum, and the second element of which is a
8094 bit specifying if the unsigned summation resulted in a carry.
8099 .. code-block:: llvm
8101 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8102 %sum = extractvalue {i32, i1} %res, 0
8103 %obit = extractvalue {i32, i1} %res, 1
8104 br i1 %obit, label %carry, label %normal
8106 '``llvm.ssub.with.overflow.*``' Intrinsics
8107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8112 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8113 on any integer bit width.
8117 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8118 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8119 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8124 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8125 a signed subtraction of the two arguments, and indicate whether an
8126 overflow occurred during the signed subtraction.
8131 The arguments (%a and %b) and the first element of the result structure
8132 may be of integer types of any bit width, but they must have the same
8133 bit width. The second element of the result structure must be of type
8134 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8140 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8141 a signed subtraction of the two arguments. They return a structure --- the
8142 first element of which is the subtraction, and the second element of
8143 which is a bit specifying if the signed subtraction resulted in an
8149 .. code-block:: llvm
8151 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8152 %sum = extractvalue {i32, i1} %res, 0
8153 %obit = extractvalue {i32, i1} %res, 1
8154 br i1 %obit, label %overflow, label %normal
8156 '``llvm.usub.with.overflow.*``' Intrinsics
8157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8162 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8163 on any integer bit width.
8167 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8168 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8169 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8174 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8175 an unsigned subtraction of the two arguments, and indicate whether an
8176 overflow occurred during the unsigned subtraction.
8181 The arguments (%a and %b) and the first element of the result structure
8182 may be of integer types of any bit width, but they must have the same
8183 bit width. The second element of the result structure must be of type
8184 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8190 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8191 an unsigned subtraction of the two arguments. They return a structure ---
8192 the first element of which is the subtraction, and the second element of
8193 which is a bit specifying if the unsigned subtraction resulted in an
8199 .. code-block:: llvm
8201 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8202 %sum = extractvalue {i32, i1} %res, 0
8203 %obit = extractvalue {i32, i1} %res, 1
8204 br i1 %obit, label %overflow, label %normal
8206 '``llvm.smul.with.overflow.*``' Intrinsics
8207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8212 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8213 on any integer bit width.
8217 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8218 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8219 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8224 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8225 a signed multiplication of the two arguments, and indicate whether an
8226 overflow occurred during the signed multiplication.
8231 The arguments (%a and %b) and the first element of the result structure
8232 may be of integer types of any bit width, but they must have the same
8233 bit width. The second element of the result structure must be of type
8234 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8240 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8241 a signed multiplication of the two arguments. They return a structure ---
8242 the first element of which is the multiplication, and the second element
8243 of which is a bit specifying if the signed multiplication resulted in an
8249 .. code-block:: llvm
8251 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8252 %sum = extractvalue {i32, i1} %res, 0
8253 %obit = extractvalue {i32, i1} %res, 1
8254 br i1 %obit, label %overflow, label %normal
8256 '``llvm.umul.with.overflow.*``' Intrinsics
8257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8262 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8263 on any integer bit width.
8267 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8268 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8269 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8274 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8275 a unsigned multiplication of the two arguments, and indicate whether an
8276 overflow occurred during the unsigned multiplication.
8281 The arguments (%a and %b) and the first element of the result structure
8282 may be of integer types of any bit width, but they must have the same
8283 bit width. The second element of the result structure must be of type
8284 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8290 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8291 an unsigned multiplication of the two arguments. They return a structure ---
8292 the first element of which is the multiplication, and the second
8293 element of which is a bit specifying if the unsigned multiplication
8294 resulted in an overflow.
8299 .. code-block:: llvm
8301 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8302 %sum = extractvalue {i32, i1} %res, 0
8303 %obit = extractvalue {i32, i1} %res, 1
8304 br i1 %obit, label %overflow, label %normal
8306 Specialised Arithmetic Intrinsics
8307 ---------------------------------
8309 '``llvm.fmuladd.*``' Intrinsic
8310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8317 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8318 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8323 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8324 expressions that can be fused if the code generator determines that (a) the
8325 target instruction set has support for a fused operation, and (b) that the
8326 fused operation is more efficient than the equivalent, separate pair of mul
8327 and add instructions.
8332 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8333 multiplicands, a and b, and an addend c.
8342 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8344 is equivalent to the expression a \* b + c, except that rounding will
8345 not be performed between the multiplication and addition steps if the
8346 code generator fuses the operations. Fusion is not guaranteed, even if
8347 the target platform supports it. If a fused multiply-add is required the
8348 corresponding llvm.fma.\* intrinsic function should be used
8349 instead. This never sets errno, just as '``llvm.fma.*``'.
8354 .. code-block:: llvm
8356 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8358 Half Precision Floating Point Intrinsics
8359 ----------------------------------------
8361 For most target platforms, half precision floating point is a
8362 storage-only format. This means that it is a dense encoding (in memory)
8363 but does not support computation in the format.
8365 This means that code must first load the half-precision floating point
8366 value as an i16, then convert it to float with
8367 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8368 then be performed on the float value (including extending to double
8369 etc). To store the value back to memory, it is first converted to float
8370 if needed, then converted to i16 with
8371 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8374 .. _int_convert_to_fp16:
8376 '``llvm.convert.to.fp16``' Intrinsic
8377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8384 declare i16 @llvm.convert.to.fp16(f32 %a)
8389 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8390 from single precision floating point format to half precision floating
8396 The intrinsic function contains single argument - the value to be
8402 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8403 from single precision floating point format to half precision floating
8404 point format. The return value is an ``i16`` which contains the
8410 .. code-block:: llvm
8412 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8413 store i16 %res, i16* @x, align 2
8415 .. _int_convert_from_fp16:
8417 '``llvm.convert.from.fp16``' Intrinsic
8418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8425 declare f32 @llvm.convert.from.fp16(i16 %a)
8430 The '``llvm.convert.from.fp16``' intrinsic function performs a
8431 conversion from half precision floating point format to single precision
8432 floating point format.
8437 The intrinsic function contains single argument - the value to be
8443 The '``llvm.convert.from.fp16``' intrinsic function performs a
8444 conversion from half single precision floating point format to single
8445 precision floating point format. The input half-float value is
8446 represented by an ``i16`` value.
8451 .. code-block:: llvm
8453 %a = load i16* @x, align 2
8454 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8459 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8460 prefix), are described in the `LLVM Source Level
8461 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8464 Exception Handling Intrinsics
8465 -----------------------------
8467 The LLVM exception handling intrinsics (which all start with
8468 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8469 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8473 Trampoline Intrinsics
8474 ---------------------
8476 These intrinsics make it possible to excise one parameter, marked with
8477 the :ref:`nest <nest>` attribute, from a function. The result is a
8478 callable function pointer lacking the nest parameter - the caller does
8479 not need to provide a value for it. Instead, the value to use is stored
8480 in advance in a "trampoline", a block of memory usually allocated on the
8481 stack, which also contains code to splice the nest value into the
8482 argument list. This is used to implement the GCC nested function address
8485 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8486 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8487 It can be created as follows:
8489 .. code-block:: llvm
8491 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8492 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8493 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8494 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8495 %fp = bitcast i8* %p to i32 (i32, i32)*
8497 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8498 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8502 '``llvm.init.trampoline``' Intrinsic
8503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8510 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8515 This fills the memory pointed to by ``tramp`` with executable code,
8516 turning it into a trampoline.
8521 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8522 pointers. The ``tramp`` argument must point to a sufficiently large and
8523 sufficiently aligned block of memory; this memory is written to by the
8524 intrinsic. Note that the size and the alignment are target-specific -
8525 LLVM currently provides no portable way of determining them, so a
8526 front-end that generates this intrinsic needs to have some
8527 target-specific knowledge. The ``func`` argument must hold a function
8528 bitcast to an ``i8*``.
8533 The block of memory pointed to by ``tramp`` is filled with target
8534 dependent code, turning it into a function. Then ``tramp`` needs to be
8535 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8536 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8537 function's signature is the same as that of ``func`` with any arguments
8538 marked with the ``nest`` attribute removed. At most one such ``nest``
8539 argument is allowed, and it must be of pointer type. Calling the new
8540 function is equivalent to calling ``func`` with the same argument list,
8541 but with ``nval`` used for the missing ``nest`` argument. If, after
8542 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8543 modified, then the effect of any later call to the returned function
8544 pointer is undefined.
8548 '``llvm.adjust.trampoline``' Intrinsic
8549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8556 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8561 This performs any required machine-specific adjustment to the address of
8562 a trampoline (passed as ``tramp``).
8567 ``tramp`` must point to a block of memory which already has trampoline
8568 code filled in by a previous call to
8569 :ref:`llvm.init.trampoline <int_it>`.
8574 On some architectures the address of the code to be executed needs to be
8575 different to the address where the trampoline is actually stored. This
8576 intrinsic returns the executable address corresponding to ``tramp``
8577 after performing the required machine specific adjustments. The pointer
8578 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8583 This class of intrinsics exists to information about the lifetime of
8584 memory objects and ranges where variables are immutable.
8588 '``llvm.lifetime.start``' Intrinsic
8589 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8596 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8601 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8607 The first argument is a constant integer representing the size of the
8608 object, or -1 if it is variable sized. The second argument is a pointer
8614 This intrinsic indicates that before this point in the code, the value
8615 of the memory pointed to by ``ptr`` is dead. This means that it is known
8616 to never be used and has an undefined value. A load from the pointer
8617 that precedes this intrinsic can be replaced with ``'undef'``.
8621 '``llvm.lifetime.end``' Intrinsic
8622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8629 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8634 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8640 The first argument is a constant integer representing the size of the
8641 object, or -1 if it is variable sized. The second argument is a pointer
8647 This intrinsic indicates that after this point in the code, the value of
8648 the memory pointed to by ``ptr`` is dead. This means that it is known to
8649 never be used and has an undefined value. Any stores into the memory
8650 object following this intrinsic may be removed as dead.
8652 '``llvm.invariant.start``' Intrinsic
8653 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8660 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8665 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8666 a memory object will not change.
8671 The first argument is a constant integer representing the size of the
8672 object, or -1 if it is variable sized. The second argument is a pointer
8678 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8679 the return value, the referenced memory location is constant and
8682 '``llvm.invariant.end``' Intrinsic
8683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8690 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8695 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8696 memory object are mutable.
8701 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8702 The second argument is a constant integer representing the size of the
8703 object, or -1 if it is variable sized and the third argument is a
8704 pointer to the object.
8709 This intrinsic indicates that the memory is mutable again.
8714 This class of intrinsics is designed to be generic and has no specific
8717 '``llvm.var.annotation``' Intrinsic
8718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8725 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8730 The '``llvm.var.annotation``' intrinsic.
8735 The first argument is a pointer to a value, the second is a pointer to a
8736 global string, the third is a pointer to a global string which is the
8737 source file name, and the last argument is the line number.
8742 This intrinsic allows annotation of local variables with arbitrary
8743 strings. This can be useful for special purpose optimizations that want
8744 to look for these annotations. These have no other defined use; they are
8745 ignored by code generation and optimization.
8747 '``llvm.ptr.annotation.*``' Intrinsic
8748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8753 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8754 pointer to an integer of any width. *NOTE* you must specify an address space for
8755 the pointer. The identifier for the default address space is the integer
8760 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8761 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8762 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8763 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8764 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8769 The '``llvm.ptr.annotation``' intrinsic.
8774 The first argument is a pointer to an integer value of arbitrary bitwidth
8775 (result of some expression), the second is a pointer to a global string, the
8776 third is a pointer to a global string which is the source file name, and the
8777 last argument is the line number. It returns the value of the first argument.
8782 This intrinsic allows annotation of a pointer to an integer with arbitrary
8783 strings. This can be useful for special purpose optimizations that want to look
8784 for these annotations. These have no other defined use; they are ignored by code
8785 generation and optimization.
8787 '``llvm.annotation.*``' Intrinsic
8788 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8793 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8794 any integer bit width.
8798 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8799 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8800 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8801 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8802 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8807 The '``llvm.annotation``' intrinsic.
8812 The first argument is an integer value (result of some expression), the
8813 second is a pointer to a global string, the third is a pointer to a
8814 global string which is the source file name, and the last argument is
8815 the line number. It returns the value of the first argument.
8820 This intrinsic allows annotations to be put on arbitrary expressions
8821 with arbitrary strings. This can be useful for special purpose
8822 optimizations that want to look for these annotations. These have no
8823 other defined use; they are ignored by code generation and optimization.
8825 '``llvm.trap``' Intrinsic
8826 ^^^^^^^^^^^^^^^^^^^^^^^^^
8833 declare void @llvm.trap() noreturn nounwind
8838 The '``llvm.trap``' intrinsic.
8848 This intrinsic is lowered to the target dependent trap instruction. If
8849 the target does not have a trap instruction, this intrinsic will be
8850 lowered to a call of the ``abort()`` function.
8852 '``llvm.debugtrap``' Intrinsic
8853 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8860 declare void @llvm.debugtrap() nounwind
8865 The '``llvm.debugtrap``' intrinsic.
8875 This intrinsic is lowered to code which is intended to cause an
8876 execution trap with the intention of requesting the attention of a
8879 '``llvm.stackprotector``' Intrinsic
8880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8887 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8892 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8893 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8894 is placed on the stack before local variables.
8899 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8900 The first argument is the value loaded from the stack guard
8901 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8902 enough space to hold the value of the guard.
8907 This intrinsic causes the prologue/epilogue inserter to force the position of
8908 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8909 to ensure that if a local variable on the stack is overwritten, it will destroy
8910 the value of the guard. When the function exits, the guard on the stack is
8911 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8912 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8913 calling the ``__stack_chk_fail()`` function.
8915 '``llvm.stackprotectorcheck``' Intrinsic
8916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8923 declare void @llvm.stackprotectorcheck(i8** <guard>)
8928 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8929 created stack protector and if they are not equal calls the
8930 ``__stack_chk_fail()`` function.
8935 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8936 the variable ``@__stack_chk_guard``.
8941 This intrinsic is provided to perform the stack protector check by comparing
8942 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8943 values do not match call the ``__stack_chk_fail()`` function.
8945 The reason to provide this as an IR level intrinsic instead of implementing it
8946 via other IR operations is that in order to perform this operation at the IR
8947 level without an intrinsic, one would need to create additional basic blocks to
8948 handle the success/failure cases. This makes it difficult to stop the stack
8949 protector check from disrupting sibling tail calls in Codegen. With this
8950 intrinsic, we are able to generate the stack protector basic blocks late in
8951 codegen after the tail call decision has occurred.
8953 '``llvm.objectsize``' Intrinsic
8954 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8961 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8962 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8967 The ``llvm.objectsize`` intrinsic is designed to provide information to
8968 the optimizers to determine at compile time whether a) an operation
8969 (like memcpy) will overflow a buffer that corresponds to an object, or
8970 b) that a runtime check for overflow isn't necessary. An object in this
8971 context means an allocation of a specific class, structure, array, or
8977 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8978 argument is a pointer to or into the ``object``. The second argument is
8979 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8980 or -1 (if false) when the object size is unknown. The second argument
8981 only accepts constants.
8986 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8987 the size of the object concerned. If the size cannot be determined at
8988 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8989 on the ``min`` argument).
8991 '``llvm.expect``' Intrinsic
8992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8997 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9002 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9003 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9004 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9009 The ``llvm.expect`` intrinsic provides information about expected (the
9010 most probable) value of ``val``, which can be used by optimizers.
9015 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9016 a value. The second argument is an expected value, this needs to be a
9017 constant value, variables are not allowed.
9022 This intrinsic is lowered to the ``val``.
9024 '``llvm.donothing``' Intrinsic
9025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9032 declare void @llvm.donothing() nounwind readnone
9037 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9038 only intrinsic that can be called with an invoke instruction.
9048 This intrinsic does nothing, and it's removed by optimizers and ignored
9051 Stack Map Intrinsics
9052 --------------------
9054 LLVM provides experimental intrinsics to support runtime patching
9055 mechanisms commonly desired in dynamic language JITs. These intrinsics
9056 are described in :doc:`StackMaps`.